diff --git a/include/crm/common/scheduler_internal.h b/include/crm/common/scheduler_internal.h
index 39e806d92c..c2ebaf329b 100644
--- a/include/crm/common/scheduler_internal.h
+++ b/include/crm/common/scheduler_internal.h
@@ -1,286 +1,289 @@
/*
* Copyright 2004-2024 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU Lesser General Public License
* version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY.
*/
#ifndef PCMK__CRM_COMMON_SCHEDULER_INTERNAL__H
#define PCMK__CRM_COMMON_SCHEDULER_INTERNAL__H
#include <crm/common/action_relation_internal.h>
#include <crm/common/actions_internal.h>
#include <crm/common/attrs_internal.h>
#include <crm/common/bundles_internal.h>
#include <crm/common/clone_internal.h>
#include <crm/common/digest_internal.h>
#include <crm/common/failcounts_internal.h>
#include <crm/common/group_internal.h>
#include <crm/common/history_internal.h>
#include <crm/common/location_internal.h>
#include <crm/common/nodes_internal.h>
#include <crm/common/primitive_internal.h>
#include <crm/common/remote_internal.h>
#include <crm/common/resources_internal.h>
#include <crm/common/roles_internal.h>
#include <crm/common/rules_internal.h>
#include <crm/common/tickets_internal.h>
#ifdef __cplusplus
extern "C" {
#endif
enum pcmk__check_parameters {
/* Clear fail count if parameters changed for un-expired start or monitor
* last_failure.
*/
pcmk__check_last_failure,
/* Clear fail count if parameters changed for start, monitor, promote, or
* migrate_from actions for active resources.
*/
pcmk__check_active,
};
// Scheduling options and conditions
enum pcmk__scheduler_flags {
// No scheduler flags set (compare with equality rather than bit set)
pcmk__sched_none = 0ULL,
/* These flags are dynamically determined conditions */
// Whether partition has quorum (via \c PCMK_XA_HAVE_QUORUM attribute)
//! \deprecated Call pcmk_has_quorum() to check quorum instead
pcmk__sched_quorate = (1ULL << 0),
// Whether cluster is symmetric (via symmetric-cluster property)
pcmk__sched_symmetric_cluster = (1ULL << 1),
// Whether scheduling encountered a non-configuration error
pcmk__sched_processing_error = (1ULL << 2),
// Whether cluster is in maintenance mode (via maintenance-mode property)
pcmk__sched_in_maintenance = (1ULL << 3),
// Whether fencing is enabled (via stonith-enabled property)
pcmk__sched_fencing_enabled = (1ULL << 4),
// Whether cluster has a fencing resource (via CIB resources)
/*! \deprecated To indicate the cluster has a fencing resource, add either a
* fencing resource configuration or the have-watchdog cluster option to the
* input CIB
*/
pcmk__sched_have_fencing = (1ULL << 5),
// Whether any resource provides or requires unfencing (via CIB resources)
pcmk__sched_enable_unfencing = (1ULL << 6),
// Whether concurrent fencing is allowed (via concurrent-fencing property)
pcmk__sched_concurrent_fencing = (1ULL << 7),
/*
* Whether resources removed from the configuration should be stopped (via
* stop-orphan-resources property)
*/
pcmk__sched_stop_removed_resources = (1ULL << 8),
/*
* Whether recurring actions removed from the configuration should be
* cancelled (via stop-orphan-actions property)
*/
pcmk__sched_cancel_removed_actions = (1ULL << 9),
// Whether to stop all resources (via stop-all-resources property)
pcmk__sched_stop_all = (1ULL << 10),
// Whether scheduler processing encountered a warning
pcmk__sched_processing_warning = (1ULL << 11),
/*
* Whether start failure should be treated as if
* \c PCMK_META_MIGRATION_THRESHOLD is 1 (via
* \c PCMK_OPT_START_FAILURE_IS_FATAL property)
*/
pcmk__sched_start_failure_fatal = (1ULL << 12),
// Whether unseen nodes should be fenced (via startup-fencing property)
pcmk__sched_startup_fencing = (1ULL << 14),
/*
* Whether resources should be left stopped when their node shuts down
* cleanly (via shutdown-lock property)
*/
pcmk__sched_shutdown_lock = (1ULL << 15),
/*
* Whether resources' current state should be probed (when unknown) before
* scheduling any other actions (via the enable-startup-probes property)
*/
pcmk__sched_probe_resources = (1ULL << 16),
// Whether the CIB status section has been parsed yet
pcmk__sched_have_status = (1ULL << 17),
// Whether the cluster includes any Pacemaker Remote nodes (via CIB)
pcmk__sched_have_remote_nodes = (1ULL << 18),
/* The remaining flags are scheduling options that must be set explicitly */
/*
* Whether to skip unpacking the CIB status section and stop the scheduling
* sequence after applying node-specific location criteria (skipping
* assignment, ordering, actions, etc.).
*/
pcmk__sched_location_only = (1ULL << 20),
// Whether sensitive resource attributes have been masked
pcmk__sched_sanitized = (1ULL << 21),
// Skip counting of total, disabled, and blocked resource instances
pcmk__sched_no_counts = (1ULL << 23),
// Whether node scores should be output instead of logged
pcmk__sched_output_scores = (1ULL << 25),
// Whether to show node and resource utilization (in log or output)
pcmk__sched_show_utilization = (1ULL << 26),
/*
* Whether to stop the scheduling sequence after unpacking the CIB,
* calculating cluster status, and applying node health (skipping
* applying node-specific location criteria, assignment, etc.)
*/
pcmk__sched_validate_only = (1ULL << 27),
};
// Implementation of pcmk__scheduler_private_t
struct pcmk__scheduler_private {
// Be careful about when each piece of information is available and final
char *local_node_name; // Name of node running scheduler (if known)
crm_time_t *now; // Time to use when evaluating rules
pcmk__output_t *out; // Output object for displaying messages
GHashTable *options; // Cluster options
const char *fence_action; // Default fencing action
guint fence_timeout_ms; // Default fencing action timeout (in ms)
guint priority_fencing_ms; // Priority-based fencing delay (in ms)
guint shutdown_lock_ms; // How long to lock resources (in ms)
guint node_pending_ms; // Pending join times out after this (in ms)
+
+ // @TODO convert to enum
const char *placement_strategy; // Value of placement-strategy property
+
xmlNode *rsc_defaults; // Configured resource defaults
xmlNode *op_defaults; // Configured operation defaults
GList *resources; // Resources in cluster
GHashTable *templates; // Key = template ID, value = resource list
GHashTable *tags; // Key = tag ID, value = element list
GList *actions; // All scheduled actions
GHashTable *singletons; // Scheduled non-resource actions
int next_action_id; // Counter used as ID for actions
xmlNode *failed; // History entries of failed actions
GList *param_check; // History entries that need to be checked
GList *stop_needed; // Containers that need stop actions
GList *location_constraints; // Location constraints
GList *colocation_constraints; // Colocation constraints
GList *ordering_constraints; // Ordering constraints
GHashTable *ticket_constraints; // Key = ticket ID, value = pcmk__ticket_t
int next_ordering_id; // Counter used as ID for orderings
int ninstances; // Total number of resource instances
int blocked_resources; // Number of blocked resources in cluster
int disabled_resources; // Number of disabled resources in cluster
time_t recheck_by; // Hint to controller when to reschedule
xmlNode *graph; // Transition graph
int synapse_count; // Number of transition graph synapses
};
// Group of enum pcmk__warnings flags for warnings we want to log once
extern uint32_t pcmk__warnings;
/*!
* \internal
* \brief Log a resource-tagged message at info severity
*
* \param[in] rsc Tag message with this resource's ID
* \param[in] fmt... printf(3)-style format and arguments
*/
#define pcmk__rsc_info(rsc, fmt, args...) \
crm_log_tag(LOG_INFO, ((rsc) == NULL)? "<NULL>" : (rsc)->id, (fmt), ##args)
/*!
* \internal
* \brief Log a resource-tagged message at debug severity
*
* \param[in] rsc Tag message with this resource's ID
* \param[in] fmt... printf(3)-style format and arguments
*/
#define pcmk__rsc_debug(rsc, fmt, args...) \
crm_log_tag(LOG_DEBUG, ((rsc) == NULL)? "<NULL>" : (rsc)->id, (fmt), ##args)
/*!
* \internal
* \brief Log a resource-tagged message at trace severity
*
* \param[in] rsc Tag message with this resource's ID
* \param[in] fmt... printf(3)-style format and arguments
*/
#define pcmk__rsc_trace(rsc, fmt, args...) \
crm_log_tag(LOG_TRACE, ((rsc) == NULL)? "<NULL>" : (rsc)->id, (fmt), ##args)
/*!
* \internal
* \brief Log an error and remember that current scheduler input has errors
*
* \param[in,out] scheduler Scheduler data
* \param[in] fmt... printf(3)-style format and arguments
*/
#define pcmk__sched_err(scheduler, fmt...) do { \
pcmk__set_scheduler_flags((scheduler), \
pcmk__sched_processing_error); \
crm_err(fmt); \
} while (0)
/*!
* \internal
* \brief Log a warning and remember that current scheduler input has warnings
*
* \param[in,out] scheduler Scheduler data
* \param[in] fmt... printf(3)-style format and arguments
*/
#define pcmk__sched_warn(scheduler, fmt...) do { \
pcmk__set_scheduler_flags((scheduler), \
pcmk__sched_processing_warning); \
crm_warn(fmt); \
} while (0)
/*!
* \internal
* \brief Set scheduler flags
*
* \param[in,out] scheduler Scheduler data
* \param[in] flags_to_set Group of enum pcmk__scheduler_flags to set
*/
#define pcmk__set_scheduler_flags(scheduler, flags_to_set) do { \
(scheduler)->flags = pcmk__set_flags_as(__func__, __LINE__, \
LOG_TRACE, "Scheduler", crm_system_name, \
(scheduler)->flags, (flags_to_set), #flags_to_set); \
} while (0)
/*!
* \internal
* \brief Clear scheduler flags
*
* \param[in,out] scheduler Scheduler data
* \param[in] flags_to_clear Group of enum pcmk__scheduler_flags to clear
*/
#define pcmk__clear_scheduler_flags(scheduler, flags_to_clear) do { \
(scheduler)->flags = pcmk__clear_flags_as(__func__, __LINE__, \
LOG_TRACE, "Scheduler", crm_system_name, \
(scheduler)->flags, (flags_to_clear), #flags_to_clear); \
} while (0)
#ifdef __cplusplus
}
#endif
#endif // PCMK__CRM_COMMON_SCHEDULER_INTERNAL__H
diff --git a/include/pcmki/pcmki_transition.h b/include/pcmki/pcmki_transition.h
index d66d7e43ce..63ac639281 100644
--- a/include/pcmki/pcmki_transition.h
+++ b/include/pcmki/pcmki_transition.h
@@ -1,181 +1,185 @@
/*
* Copyright 2004-2024 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU Lesser General Public License
* version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY.
*/
#ifndef PCMK__PCMKI_PCMKI_TRANSITION__H
#define PCMK__PCMKI_PCMKI_TRANSITION__H
#include <stdbool.h> // bool
#include <stdint.h> // uint32_t
#include <sys/types.h> // time_t
#include <glib.h> // guint, GList, GHashTable
#include <libxml/tree.h> // xmlNode
#include <crm/common/scheduler_types.h> // pcmk_scheduler_t
#include <crm/lrmd_events.h> // lrmd_event_data_t
#ifdef __cplusplus
extern "C" {
#endif
enum pcmk__graph_action_type {
pcmk__pseudo_graph_action,
pcmk__rsc_graph_action,
pcmk__cluster_graph_action,
+ /* @TODO maybe separate a new pcmk__fencing_graph_action from
+ * pcmk__cluster_graph_action to make code cleaner (for example, see
+ * initiate_action())
+ */
};
enum pcmk__synapse_flags {
pcmk__synapse_ready = (1 << 0),
pcmk__synapse_failed = (1 << 1),
pcmk__synapse_executed = (1 << 2),
pcmk__synapse_confirmed = (1 << 3),
};
typedef struct {
int id;
int priority;
uint32_t flags; // Group of pcmk__synapse_flags
GList *actions; /* pcmk__graph_action_t* */
GList *inputs; /* pcmk__graph_action_t* */
} pcmk__graph_synapse_t;
#define pcmk__set_synapse_flags(synapse, flags_to_set) do { \
(synapse)->flags = pcmk__set_flags_as(__func__, __LINE__, \
LOG_TRACE, \
"Synapse", "synapse", \
(synapse)->flags, (flags_to_set), #flags_to_set); \
} while (0)
#define pcmk__clear_synapse_flags(synapse, flags_to_clear) do { \
(synapse)->flags = pcmk__clear_flags_as(__func__, __LINE__, \
LOG_TRACE, \
"Synapse", "synapse", \
(synapse)->flags, (flags_to_clear), #flags_to_clear); \
} while (0)
enum pcmk__graph_action_flags {
pcmk__graph_action_sent_update = (1 << 0), /* sent to the CIB */
pcmk__graph_action_executed = (1 << 1), /* sent to the CRM */
pcmk__graph_action_confirmed = (1 << 2),
pcmk__graph_action_failed = (1 << 3),
};
typedef struct {
int id;
int timeout;
int timer;
guint interval_ms;
GHashTable *params;
enum pcmk__graph_action_type type;
pcmk__graph_synapse_t *synapse;
uint32_t flags; // Group of pcmk__graph_action_flags
xmlNode *xml;
} pcmk__graph_action_t;
#define pcmk__set_graph_action_flags(action, flags_to_set) do { \
(action)->flags = pcmk__set_flags_as(__func__, __LINE__, \
LOG_TRACE, \
"Action", "action", \
(action)->flags, (flags_to_set), #flags_to_set); \
} while (0)
#define pcmk__clear_graph_action_flags(action, flags_to_clear) do { \
(action)->flags = pcmk__clear_flags_as(__func__, __LINE__, \
LOG_TRACE, \
"Action", "action", \
(action)->flags, (flags_to_clear), #flags_to_clear); \
} while (0)
// What to do after finished processing a transition graph
enum pcmk__graph_next {
// Order matters: lowest priority to highest
pcmk__graph_done, // Transition complete, nothing further needed
pcmk__graph_wait, // Transition interrupted, wait for further changes
pcmk__graph_restart, // Transition interrupted, start a new one
pcmk__graph_shutdown, // Transition interrupted, local shutdown needed
};
typedef struct {
int id;
char *source;
int abort_priority;
bool complete;
const char *abort_reason;
enum pcmk__graph_next completion_action;
int num_actions;
int num_synapses;
int batch_limit;
guint network_delay;
guint stonith_timeout;
int fired;
int pending;
int skipped;
int completed;
int incomplete;
GList *synapses; /* pcmk__graph_synapse_t* */
int migration_limit;
//! Failcount after one failed stop action
char *failed_stop_offset;
//! Failcount after one failed start action
char *failed_start_offset;
//! Time (from epoch) by which the controller should re-run the scheduler
time_t recheck_by;
} pcmk__graph_t;
typedef struct {
int (*pseudo) (pcmk__graph_t *graph, pcmk__graph_action_t *action);
int (*rsc) (pcmk__graph_t *graph, pcmk__graph_action_t *action);
int (*cluster) (pcmk__graph_t *graph, pcmk__graph_action_t *action);
int (*fence) (pcmk__graph_t *graph, pcmk__graph_action_t *action);
bool (*allowed) (pcmk__graph_t *graph, pcmk__graph_action_t *action);
} pcmk__graph_functions_t;
enum pcmk__graph_status {
pcmk__graph_active, // Some actions have been performed
pcmk__graph_pending, // No actions performed yet
pcmk__graph_complete,
pcmk__graph_terminated,
};
void pcmk__set_graph_functions(pcmk__graph_functions_t *fns);
pcmk__graph_t *pcmk__unpack_graph(const xmlNode *xml_graph,
const char *reference);
enum pcmk__graph_status pcmk__execute_graph(pcmk__graph_t *graph);
void pcmk__update_graph(pcmk__graph_t *graph,
const pcmk__graph_action_t *action);
void pcmk__free_graph(pcmk__graph_t *graph);
const char *pcmk__graph_status2text(enum pcmk__graph_status state);
void pcmk__log_graph(unsigned int log_level, pcmk__graph_t *graph);
void pcmk__log_graph_action(int log_level, pcmk__graph_action_t *action);
void pcmk__log_transition_summary(const pcmk_scheduler_t *scheduler,
const char *filename);
lrmd_event_data_t *pcmk__event_from_graph_action(const xmlNode *resource,
const pcmk__graph_action_t *action,
int status, int rc,
const char *exit_reason);
#ifdef __cplusplus
}
#endif
#endif // PCMK__PCMKI_PCMKI_TRANSITION__H
diff --git a/lib/common/ipc_client.c b/lib/common/ipc_client.c
index c644b9c94c..6f2d65be2f 100644
--- a/lib/common/ipc_client.c
+++ b/lib/common/ipc_client.c
@@ -1,1678 +1,1682 @@
/*
* Copyright 2004-2024 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU Lesser General Public License
* version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY.
*/
#include <crm_internal.h>
#if defined(HAVE_UCRED) || defined(HAVE_SOCKPEERCRED)
#include <sys/socket.h>
#elif defined(HAVE_GETPEERUCRED)
#include <ucred.h>
#endif
#include <stdio.h>
#include <sys/types.h>
#include <errno.h>
#include <bzlib.h>
#include <crm/crm.h> /* indirectly: pcmk_err_generic */
#include <crm/common/xml.h>
#include <crm/common/ipc.h>
#include <crm/common/ipc_internal.h>
#include "crmcommon_private.h"
static int is_ipc_provider_expected(qb_ipcc_connection_t *qb_ipc, int sock,
uid_t refuid, gid_t refgid, pid_t *gotpid,
uid_t *gotuid, gid_t *gotgid);
/*!
* \brief Create a new object for using Pacemaker daemon IPC
*
* \param[out] api Where to store new IPC object
* \param[in] server Which Pacemaker daemon the object is for
*
* \return Standard Pacemaker result code
*
* \note The caller is responsible for freeing *api using pcmk_free_ipc_api().
* \note This is intended to supersede crm_ipc_new() but currently only
* supports the controller, pacemakerd, and schedulerd IPC API.
*/
int
pcmk_new_ipc_api(pcmk_ipc_api_t **api, enum pcmk_ipc_server server)
{
if (api == NULL) {
return EINVAL;
}
*api = calloc(1, sizeof(pcmk_ipc_api_t));
if (*api == NULL) {
return errno;
}
(*api)->server = server;
if (pcmk_ipc_name(*api, false) == NULL) {
pcmk_free_ipc_api(*api);
*api = NULL;
return EOPNOTSUPP;
}
(*api)->ipc_size_max = 0;
// Set server methods and max_size (if not default)
switch (server) {
case pcmk_ipc_attrd:
(*api)->cmds = pcmk__attrd_api_methods();
break;
case pcmk_ipc_based:
(*api)->ipc_size_max = 512 * 1024; // 512KB
break;
case pcmk_ipc_controld:
(*api)->cmds = pcmk__controld_api_methods();
break;
case pcmk_ipc_execd:
break;
case pcmk_ipc_fenced:
break;
case pcmk_ipc_pacemakerd:
(*api)->cmds = pcmk__pacemakerd_api_methods();
break;
case pcmk_ipc_schedulerd:
(*api)->cmds = pcmk__schedulerd_api_methods();
// @TODO max_size could vary by client, maybe take as argument?
(*api)->ipc_size_max = 5 * 1024 * 1024; // 5MB
break;
default: // pcmk_ipc_unknown
pcmk_free_ipc_api(*api);
*api = NULL;
return EINVAL;
}
if ((*api)->cmds == NULL) {
pcmk_free_ipc_api(*api);
*api = NULL;
return ENOMEM;
}
(*api)->ipc = crm_ipc_new(pcmk_ipc_name(*api, false),
(*api)->ipc_size_max);
if ((*api)->ipc == NULL) {
pcmk_free_ipc_api(*api);
*api = NULL;
return ENOMEM;
}
// If daemon API has its own data to track, allocate it
if ((*api)->cmds->new_data != NULL) {
if ((*api)->cmds->new_data(*api) != pcmk_rc_ok) {
pcmk_free_ipc_api(*api);
*api = NULL;
return ENOMEM;
}
}
crm_trace("Created %s API IPC object", pcmk_ipc_name(*api, true));
return pcmk_rc_ok;
}
static void
free_daemon_specific_data(pcmk_ipc_api_t *api)
{
if ((api != NULL) && (api->cmds != NULL)) {
if ((api->cmds->free_data != NULL) && (api->api_data != NULL)) {
api->cmds->free_data(api->api_data);
api->api_data = NULL;
}
free(api->cmds);
api->cmds = NULL;
}
}
/*!
* \internal
* \brief Call an IPC API event callback, if one is registed
*
* \param[in,out] api IPC API connection
* \param[in] event_type The type of event that occurred
* \param[in] status Event status
* \param[in,out] event_data Event-specific data
*/
void
pcmk__call_ipc_callback(pcmk_ipc_api_t *api, enum pcmk_ipc_event event_type,
crm_exit_t status, void *event_data)
{
if ((api != NULL) && (api->cb != NULL)) {
api->cb(api, event_type, status, event_data, api->user_data);
}
}
/*!
* \internal
* \brief Clean up after an IPC disconnect
*
* \param[in,out] user_data IPC API connection that disconnected
*
* \note This function can be used as a main loop IPC destroy callback.
*/
static void
ipc_post_disconnect(gpointer user_data)
{
pcmk_ipc_api_t *api = user_data;
crm_info("Disconnected from %s", pcmk_ipc_name(api, true));
// Perform any daemon-specific handling needed
if ((api->cmds != NULL) && (api->cmds->post_disconnect != NULL)) {
api->cmds->post_disconnect(api);
}
// Call client's registered event callback
pcmk__call_ipc_callback(api, pcmk_ipc_event_disconnect, CRM_EX_DISCONNECT,
NULL);
/* If this is being called from a running main loop, mainloop_gio_destroy()
* will free ipc and mainloop_io immediately after calling this function.
* If this is called from a stopped main loop, these will leak, so the best
* practice is to close the connection before stopping the main loop.
*/
api->ipc = NULL;
api->mainloop_io = NULL;
if (api->free_on_disconnect) {
/* pcmk_free_ipc_api() has already been called, but did not free api
* or api->cmds because this function needed them. Do that now.
*/
free_daemon_specific_data(api);
crm_trace("Freeing IPC API object after disconnect");
free(api);
}
}
/*!
* \brief Free the contents of an IPC API object
*
* \param[in,out] api IPC API object to free
*/
void
pcmk_free_ipc_api(pcmk_ipc_api_t *api)
{
bool free_on_disconnect = false;
if (api == NULL) {
return;
}
crm_debug("Releasing %s IPC API", pcmk_ipc_name(api, true));
if (api->ipc != NULL) {
if (api->mainloop_io != NULL) {
/* We need to keep the api pointer itself around, because it is the
* user data for the IPC client destroy callback. That will be
* triggered by the pcmk_disconnect_ipc() call below, but it might
* happen later in the main loop (if still running).
*
* This flag tells the destroy callback to free the object. It can't
* do that unconditionally, because the application might call this
* function after a disconnect that happened by other means.
*/
free_on_disconnect = api->free_on_disconnect = true;
}
pcmk_disconnect_ipc(api); // Frees api if free_on_disconnect is true
}
if (!free_on_disconnect) {
free_daemon_specific_data(api);
crm_trace("Freeing IPC API object");
free(api);
}
}
/*!
* \brief Get the IPC name used with an IPC API connection
*
* \param[in] api IPC API connection
* \param[in] for_log If true, return human-friendly name instead of IPC name
*
* \return IPC API's human-friendly or connection name, or if none is available,
* "Pacemaker" if for_log is true and NULL if for_log is false
*/
const char *
pcmk_ipc_name(const pcmk_ipc_api_t *api, bool for_log)
{
if (api == NULL) {
return for_log? "Pacemaker" : NULL;
}
if (for_log) {
const char *name = pcmk__server_log_name(api->server);
return pcmk__s(name, "Pacemaker");
}
switch (api->server) {
// These servers do not have pcmk_ipc_api_t implementations yet
case pcmk_ipc_based:
case pcmk_ipc_execd:
case pcmk_ipc_fenced:
return NULL;
default:
return pcmk__server_ipc_name(api->server);
}
}
/*!
* \brief Check whether an IPC API connection is active
*
* \param[in,out] api IPC API connection
*
* \return true if IPC is connected, false otherwise
*/
bool
pcmk_ipc_is_connected(pcmk_ipc_api_t *api)
{
return (api != NULL) && crm_ipc_connected(api->ipc);
}
/*!
* \internal
* \brief Call the daemon-specific API's dispatch function
*
* Perform daemon-specific handling of IPC reply dispatch. It is the daemon
* method's responsibility to call the client's registered event callback, as
* well as allocate and free any event data.
*
* \param[in,out] api IPC API connection
* \param[in,out] message IPC reply XML to dispatch
*/
static bool
call_api_dispatch(pcmk_ipc_api_t *api, xmlNode *message)
{
crm_log_xml_trace(message, "ipc-received");
if ((api->cmds != NULL) && (api->cmds->dispatch != NULL)) {
return api->cmds->dispatch(api, message);
}
return false;
}
/*!
* \internal
* \brief Dispatch previously read IPC data
*
* \param[in] buffer Data read from IPC
* \param[in,out] api IPC object
*
* \return Standard Pacemaker return code. In particular:
*
* pcmk_rc_ok: There are no more messages expected from the server. Quit
* reading.
* EINPROGRESS: There are more messages expected from the server. Keep reading.
*
* All other values indicate an error.
*/
static int
dispatch_ipc_data(const char *buffer, pcmk_ipc_api_t *api)
{
bool more = false;
xmlNode *msg;
if (buffer == NULL) {
crm_warn("Empty message received from %s IPC",
pcmk_ipc_name(api, true));
return ENOMSG;
}
msg = pcmk__xml_parse(buffer);
if (msg == NULL) {
crm_warn("Malformed message received from %s IPC",
pcmk_ipc_name(api, true));
return EPROTO;
}
more = call_api_dispatch(api, msg);
pcmk__xml_free(msg);
if (more) {
return EINPROGRESS;
} else {
return pcmk_rc_ok;
}
}
/*!
* \internal
* \brief Dispatch data read from IPC source
*
* \param[in] buffer Data read from IPC
* \param[in] length Number of bytes of data in buffer (ignored)
* \param[in,out] user_data IPC object
*
* \return Always 0 (meaning connection is still required)
*
* \note This function can be used as a main loop IPC dispatch callback.
*/
static int
dispatch_ipc_source_data(const char *buffer, ssize_t length, gpointer user_data)
{
pcmk_ipc_api_t *api = user_data;
CRM_CHECK(api != NULL, return 0);
dispatch_ipc_data(buffer, api);
return 0;
}
/*!
* \brief Check whether an IPC connection has data available (without main loop)
*
* \param[in] api IPC API connection
* \param[in] timeout_ms If less than 0, poll indefinitely; if 0, poll once
* and return immediately; otherwise, poll for up to
* this many milliseconds
*
* \return Standard Pacemaker return code
*
* \note Callers of pcmk_connect_ipc() using pcmk_ipc_dispatch_poll should call
* this function to check whether IPC data is available. Return values of
* interest include pcmk_rc_ok meaning data is available, and EAGAIN
* meaning no data is available; all other values indicate errors.
* \todo This does not allow the caller to poll multiple file descriptors at
* once. If there is demand for that, we could add a wrapper for
* pcmk__ipc_fd(api->ipc), so the caller can call poll() themselves.
*/
int
pcmk_poll_ipc(const pcmk_ipc_api_t *api, int timeout_ms)
{
int rc;
struct pollfd pollfd = { 0, };
if ((api == NULL) || (api->dispatch_type != pcmk_ipc_dispatch_poll)) {
return EINVAL;
}
rc = pcmk__ipc_fd(api->ipc, &(pollfd.fd));
if (rc != pcmk_rc_ok) {
crm_debug("Could not obtain file descriptor for %s IPC: %s",
pcmk_ipc_name(api, true), pcmk_rc_str(rc));
return rc;
}
pollfd.events = POLLIN;
rc = poll(&pollfd, 1, timeout_ms);
if (rc < 0) {
/* Some UNIX systems return negative and set EAGAIN for failure to
* allocate memory; standardize the return code in that case
*/
return (errno == EAGAIN)? ENOMEM : errno;
} else if (rc == 0) {
return EAGAIN;
}
return pcmk_rc_ok;
}
/*!
* \brief Dispatch available messages on an IPC connection (without main loop)
*
* \param[in,out] api IPC API connection
*
* \return Standard Pacemaker return code
*
* \note Callers of pcmk_connect_ipc() using pcmk_ipc_dispatch_poll should call
* this function when IPC data is available.
*/
void
pcmk_dispatch_ipc(pcmk_ipc_api_t *api)
{
if (api == NULL) {
return;
}
while (crm_ipc_ready(api->ipc) > 0) {
if (crm_ipc_read(api->ipc) > 0) {
dispatch_ipc_data(crm_ipc_buffer(api->ipc), api);
}
}
}
// \return Standard Pacemaker return code
static int
connect_with_main_loop(pcmk_ipc_api_t *api)
{
int rc;
struct ipc_client_callbacks callbacks = {
.dispatch = dispatch_ipc_source_data,
.destroy = ipc_post_disconnect,
};
rc = pcmk__add_mainloop_ipc(api->ipc, G_PRIORITY_DEFAULT, api,
&callbacks, &(api->mainloop_io));
if (rc != pcmk_rc_ok) {
return rc;
}
crm_debug("Connected to %s IPC (attached to main loop)",
pcmk_ipc_name(api, true));
/* After this point, api->mainloop_io owns api->ipc, so api->ipc
* should not be explicitly freed.
*/
return pcmk_rc_ok;
}
// \return Standard Pacemaker return code
static int
connect_without_main_loop(pcmk_ipc_api_t *api)
{
int rc = pcmk__connect_generic_ipc(api->ipc);
if (rc != pcmk_rc_ok) {
crm_ipc_close(api->ipc);
} else {
crm_debug("Connected to %s IPC (without main loop)",
pcmk_ipc_name(api, true));
}
return rc;
}
/*!
* \internal
* \brief Connect to a Pacemaker daemon via IPC (retrying after soft errors)
*
* \param[in,out] api IPC API instance
* \param[in] dispatch_type How IPC replies should be dispatched
* \param[in] attempts How many times to try (in case of soft error)
*
* \return Standard Pacemaker return code
*/
int
pcmk__connect_ipc(pcmk_ipc_api_t *api, enum pcmk_ipc_dispatch dispatch_type,
int attempts)
{
int rc = pcmk_rc_ok;
if ((api == NULL) || (attempts < 1)) {
return EINVAL;
}
if (api->ipc == NULL) {
api->ipc = crm_ipc_new(pcmk_ipc_name(api, false), api->ipc_size_max);
if (api->ipc == NULL) {
return ENOMEM;
}
}
if (crm_ipc_connected(api->ipc)) {
crm_trace("Already connected to %s", pcmk_ipc_name(api, true));
return pcmk_rc_ok;
}
api->dispatch_type = dispatch_type;
crm_debug("Attempting connection to %s (up to %d time%s)",
pcmk_ipc_name(api, true), attempts, pcmk__plural_s(attempts));
for (int remaining = attempts - 1; remaining >= 0; --remaining) {
switch (dispatch_type) {
case pcmk_ipc_dispatch_main:
rc = connect_with_main_loop(api);
break;
case pcmk_ipc_dispatch_sync:
case pcmk_ipc_dispatch_poll:
rc = connect_without_main_loop(api);
break;
}
if ((remaining == 0) || ((rc != EAGAIN) && (rc != EALREADY))) {
break; // Result is final
}
// Retry after soft error (interrupted by signal, etc.)
pcmk__sleep_ms((attempts - remaining) * 500);
crm_debug("Re-attempting connection to %s (%d attempt%s remaining)",
pcmk_ipc_name(api, true), remaining,
pcmk__plural_s(remaining));
}
if (rc != pcmk_rc_ok) {
return rc;
}
if ((api->cmds != NULL) && (api->cmds->post_connect != NULL)) {
rc = api->cmds->post_connect(api);
if (rc != pcmk_rc_ok) {
crm_ipc_close(api->ipc);
}
}
return rc;
}
/*!
* \brief Connect to a Pacemaker daemon via IPC
*
* \param[in,out] api IPC API instance
* \param[in] dispatch_type How IPC replies should be dispatched
*
* \return Standard Pacemaker return code
*/
int
pcmk_connect_ipc(pcmk_ipc_api_t *api, enum pcmk_ipc_dispatch dispatch_type)
{
int rc = pcmk__connect_ipc(api, dispatch_type, 2);
if (rc != pcmk_rc_ok) {
crm_err("Connection to %s failed: %s",
pcmk_ipc_name(api, true), pcmk_rc_str(rc));
}
return rc;
}
/*!
* \brief Disconnect an IPC API instance
*
* \param[in,out] api IPC API connection
*
* \return Standard Pacemaker return code
*
* \note If the connection is attached to a main loop, this function should be
* called before quitting the main loop, to ensure that all memory is
* freed.
*/
void
pcmk_disconnect_ipc(pcmk_ipc_api_t *api)
{
if ((api == NULL) || (api->ipc == NULL)) {
return;
}
switch (api->dispatch_type) {
case pcmk_ipc_dispatch_main:
{
mainloop_io_t *mainloop_io = api->mainloop_io;
// Make sure no code with access to api can use these again
api->mainloop_io = NULL;
api->ipc = NULL;
mainloop_del_ipc_client(mainloop_io);
// After this point api might have already been freed
}
break;
case pcmk_ipc_dispatch_poll:
case pcmk_ipc_dispatch_sync:
{
crm_ipc_t *ipc = api->ipc;
// Make sure no code with access to api can use ipc again
api->ipc = NULL;
// This should always be the case already, but to be safe
api->free_on_disconnect = false;
crm_ipc_close(ipc);
crm_ipc_destroy(ipc);
ipc_post_disconnect(api);
}
break;
}
}
/*!
* \brief Register a callback for IPC API events
*
* \param[in,out] api IPC API connection
* \param[in] callback Callback to register
* \param[in] userdata Caller data to pass to callback
*
* \note This function may be called multiple times to update the callback
* and/or user data. The caller remains responsible for freeing
* userdata in any case (after the IPC is disconnected, if the
* user data is still registered with the IPC).
*/
void
pcmk_register_ipc_callback(pcmk_ipc_api_t *api, pcmk_ipc_callback_t cb,
void *user_data)
{
if (api == NULL) {
return;
}
api->cb = cb;
api->user_data = user_data;
}
/*!
* \internal
* \brief Send an XML request across an IPC API connection
*
* \param[in,out] api IPC API connection
* \param[in] request XML request to send
*
* \return Standard Pacemaker return code
*
* \note Daemon-specific IPC API functions should call this function to send
* requests, because it handles different dispatch types appropriately.
*/
int
pcmk__send_ipc_request(pcmk_ipc_api_t *api, const xmlNode *request)
{
int rc;
xmlNode *reply = NULL;
enum crm_ipc_flags flags = crm_ipc_flags_none;
if ((api == NULL) || (api->ipc == NULL) || (request == NULL)) {
return EINVAL;
}
crm_log_xml_trace(request, "ipc-sent");
// Synchronous dispatch requires waiting for a reply
if ((api->dispatch_type == pcmk_ipc_dispatch_sync)
&& (api->cmds != NULL)
&& (api->cmds->reply_expected != NULL)
&& (api->cmds->reply_expected(api, request))) {
flags = crm_ipc_client_response;
}
- // The 0 here means a default timeout of 5 seconds
+ /* The 0 here means a default timeout of 5 seconds
+ *
+ * @TODO Maybe add a timeout_ms member to pcmk_ipc_api_t and a
+ * pcmk_set_ipc_timeout() setter for it, then use it here.
+ */
rc = crm_ipc_send(api->ipc, request, flags, 0, &reply);
if (rc < 0) {
return pcmk_legacy2rc(rc);
} else if (rc == 0) {
return ENODATA;
}
// With synchronous dispatch, we dispatch any reply now
if (reply != NULL) {
bool more = call_api_dispatch(api, reply);
pcmk__xml_free(reply);
while (more) {
rc = crm_ipc_read(api->ipc);
if (rc == -EAGAIN) {
continue;
} else if (rc == -ENOMSG || rc == pcmk_ok) {
return pcmk_rc_ok;
} else if (rc < 0) {
return -rc;
}
rc = dispatch_ipc_data(crm_ipc_buffer(api->ipc), api);
if (rc == pcmk_rc_ok) {
more = false;
} else if (rc == EINPROGRESS) {
more = true;
} else {
continue;
}
}
}
return pcmk_rc_ok;
}
/*!
* \internal
* \brief Create the XML for an IPC request to purge a node from the peer cache
*
* \param[in] api IPC API connection
* \param[in] node_name If not NULL, name of node to purge
* \param[in] nodeid If not 0, node ID of node to purge
*
* \return Newly allocated IPC request XML
*
* \note The controller, fencer, and pacemakerd use the same request syntax, but
* the attribute manager uses a different one. The CIB manager doesn't
* have any syntax for it. The executor and scheduler don't connect to the
* cluster layer and thus don't have or need any syntax for it.
*
* \todo Modify the attribute manager to accept the common syntax (as well
* as its current one, for compatibility with older clients). Modify
* the CIB manager to accept and honor the common syntax. Modify the
* executor and scheduler to accept the syntax (immediately returning
* success), just for consistency. Modify this function to use the
* common syntax with all daemons if their version supports it.
*/
static xmlNode *
create_purge_node_request(const pcmk_ipc_api_t *api, const char *node_name,
uint32_t nodeid)
{
xmlNode *request = NULL;
const char *client = crm_system_name? crm_system_name : "client";
switch (api->server) {
case pcmk_ipc_attrd:
request = pcmk__xe_create(NULL, __func__);
crm_xml_add(request, PCMK__XA_T, PCMK__VALUE_ATTRD);
crm_xml_add(request, PCMK__XA_SRC, crm_system_name);
crm_xml_add(request, PCMK_XA_TASK, PCMK__ATTRD_CMD_PEER_REMOVE);
pcmk__xe_set_bool_attr(request, PCMK__XA_REAP, true);
pcmk__xe_add_node(request, node_name, nodeid);
break;
case pcmk_ipc_controld:
case pcmk_ipc_fenced:
case pcmk_ipc_pacemakerd:
request = pcmk__new_request(api->server, client, NULL,
pcmk_ipc_name(api, false),
CRM_OP_RM_NODE_CACHE, NULL);
if (nodeid > 0) {
crm_xml_add_ll(request, PCMK_XA_ID, (long long) nodeid);
}
crm_xml_add(request, PCMK_XA_UNAME, node_name);
break;
case pcmk_ipc_based:
case pcmk_ipc_execd:
case pcmk_ipc_schedulerd:
break;
default: // pcmk_ipc_unknown (shouldn't be possible)
return NULL;
}
return request;
}
/*!
* \brief Ask a Pacemaker daemon to purge a node from its peer cache
*
* \param[in,out] api IPC API connection
* \param[in] node_name If not NULL, name of node to purge
* \param[in] nodeid If not 0, node ID of node to purge
*
* \return Standard Pacemaker return code
*
* \note At least one of node_name or nodeid must be specified.
*/
int
pcmk_ipc_purge_node(pcmk_ipc_api_t *api, const char *node_name, uint32_t nodeid)
{
int rc = 0;
xmlNode *request = NULL;
if (api == NULL) {
return EINVAL;
}
if ((node_name == NULL) && (nodeid == 0)) {
return EINVAL;
}
request = create_purge_node_request(api, node_name, nodeid);
if (request == NULL) {
return EOPNOTSUPP;
}
rc = pcmk__send_ipc_request(api, request);
pcmk__xml_free(request);
crm_debug("%s peer cache purge of node %s[%lu]: rc=%d",
pcmk_ipc_name(api, true), node_name, (unsigned long) nodeid, rc);
return rc;
}
/*
* Generic IPC API (to eventually be deprecated as public API and made internal)
*/
struct crm_ipc_s {
struct pollfd pfd;
unsigned int max_buf_size; // maximum bytes we can send or receive over IPC
unsigned int buf_size; // size of allocated buffer
int msg_size;
int need_reply;
char *buffer;
char *server_name; // server IPC name being connected to
qb_ipcc_connection_t *ipc;
};
/*!
* \brief Create a new (legacy) object for using Pacemaker daemon IPC
*
* \param[in] name IPC system name to connect to
* \param[in] max_size Use a maximum IPC buffer size of at least this size
*
* \return Newly allocated IPC object on success, NULL otherwise
*
* \note The caller is responsible for freeing the result using
* crm_ipc_destroy().
* \note This should be considered deprecated for use with daemons supported by
* pcmk_new_ipc_api().
*/
crm_ipc_t *
crm_ipc_new(const char *name, size_t max_size)
{
crm_ipc_t *client = NULL;
client = calloc(1, sizeof(crm_ipc_t));
if (client == NULL) {
crm_err("Could not create IPC connection: %s", strerror(errno));
return NULL;
}
client->server_name = strdup(name);
if (client->server_name == NULL) {
crm_err("Could not create %s IPC connection: %s",
name, strerror(errno));
free(client);
return NULL;
}
client->buf_size = pcmk__ipc_buffer_size(max_size);
client->buffer = malloc(client->buf_size);
if (client->buffer == NULL) {
crm_err("Could not create %s IPC connection: %s",
name, strerror(errno));
free(client->server_name);
free(client);
return NULL;
}
/* Clients initiating connection pick the max buf size */
client->max_buf_size = client->buf_size;
client->pfd.fd = -1;
client->pfd.events = POLLIN;
client->pfd.revents = 0;
return client;
}
/*!
* \internal
* \brief Connect a generic (not daemon-specific) IPC object
*
* \param[in,out] ipc Generic IPC object to connect
*
* \return Standard Pacemaker return code
*/
int
pcmk__connect_generic_ipc(crm_ipc_t *ipc)
{
uid_t cl_uid = 0;
gid_t cl_gid = 0;
pid_t found_pid = 0;
uid_t found_uid = 0;
gid_t found_gid = 0;
int rc = pcmk_rc_ok;
if (ipc == NULL) {
return EINVAL;
}
ipc->need_reply = FALSE;
ipc->ipc = qb_ipcc_connect(ipc->server_name, ipc->buf_size);
if (ipc->ipc == NULL) {
return errno;
}
rc = qb_ipcc_fd_get(ipc->ipc, &ipc->pfd.fd);
if (rc < 0) { // -errno
crm_ipc_close(ipc);
return -rc;
}
rc = pcmk_daemon_user(&cl_uid, &cl_gid);
rc = pcmk_legacy2rc(rc);
if (rc != pcmk_rc_ok) {
crm_ipc_close(ipc);
return rc;
}
rc = is_ipc_provider_expected(ipc->ipc, ipc->pfd.fd, cl_uid, cl_gid,
&found_pid, &found_uid, &found_gid);
if (rc != pcmk_rc_ok) {
if (rc == pcmk_rc_ipc_unauthorized) {
crm_info("%s IPC provider authentication failed: process %lld has "
"uid %lld (expected %lld) and gid %lld (expected %lld)",
ipc->server_name,
(long long) PCMK__SPECIAL_PID_AS_0(found_pid),
(long long) found_uid, (long long) cl_uid,
(long long) found_gid, (long long) cl_gid);
}
crm_ipc_close(ipc);
return rc;
}
ipc->max_buf_size = qb_ipcc_get_buffer_size(ipc->ipc);
if (ipc->max_buf_size > ipc->buf_size) {
free(ipc->buffer);
ipc->buffer = calloc(ipc->max_buf_size, sizeof(char));
if (ipc->buffer == NULL) {
rc = errno;
crm_ipc_close(ipc);
return rc;
}
ipc->buf_size = ipc->max_buf_size;
}
return pcmk_rc_ok;
}
void
crm_ipc_close(crm_ipc_t * client)
{
if (client) {
if (client->ipc) {
qb_ipcc_connection_t *ipc = client->ipc;
client->ipc = NULL;
qb_ipcc_disconnect(ipc);
}
}
}
void
crm_ipc_destroy(crm_ipc_t * client)
{
if (client) {
if (client->ipc && qb_ipcc_is_connected(client->ipc)) {
crm_notice("Destroying active %s IPC connection",
client->server_name);
/* The next line is basically unsafe
*
* If this connection was attached to mainloop and mainloop is active,
* the 'disconnected' callback will end up back here and we'll end
* up free'ing the memory twice - something that can still happen
* even without this if we destroy a connection and it closes before
* we call exit
*/
/* crm_ipc_close(client); */
} else {
crm_trace("Destroying inactive %s IPC connection",
client->server_name);
}
free(client->buffer);
free(client->server_name);
free(client);
}
}
/*!
* \internal
* \brief Get the file descriptor for a generic IPC object
*
* \param[in,out] ipc Generic IPC object to get file descriptor for
* \param[out] fd Where to store file descriptor
*
* \return Standard Pacemaker return code
*/
int
pcmk__ipc_fd(crm_ipc_t *ipc, int *fd)
{
if ((ipc == NULL) || (fd == NULL)) {
return EINVAL;
}
if ((ipc->ipc == NULL) || (ipc->pfd.fd < 0)) {
return ENOTCONN;
}
*fd = ipc->pfd.fd;
return pcmk_rc_ok;
}
int
crm_ipc_get_fd(crm_ipc_t * client)
{
int fd = -1;
if (pcmk__ipc_fd(client, &fd) != pcmk_rc_ok) {
crm_err("Could not obtain file descriptor for %s IPC",
((client == NULL)? "unspecified" : client->server_name));
errno = EINVAL;
return -EINVAL;
}
return fd;
}
bool
crm_ipc_connected(crm_ipc_t * client)
{
bool rc = FALSE;
if (client == NULL) {
crm_trace("No client");
return FALSE;
} else if (client->ipc == NULL) {
crm_trace("No connection");
return FALSE;
} else if (client->pfd.fd < 0) {
crm_trace("Bad descriptor");
return FALSE;
}
rc = qb_ipcc_is_connected(client->ipc);
if (rc == FALSE) {
client->pfd.fd = -EINVAL;
}
return rc;
}
/*!
* \brief Check whether an IPC connection is ready to be read
*
* \param[in,out] client Connection to check
*
* \return Positive value if ready to be read, 0 if not ready, -errno on error
*/
int
crm_ipc_ready(crm_ipc_t *client)
{
int rc;
pcmk__assert(client != NULL);
if (!crm_ipc_connected(client)) {
return -ENOTCONN;
}
client->pfd.revents = 0;
rc = poll(&(client->pfd), 1, 0);
return (rc < 0)? -errno : rc;
}
// \return Standard Pacemaker return code
static int
crm_ipc_decompress(crm_ipc_t * client)
{
pcmk__ipc_header_t *header = (pcmk__ipc_header_t *)(void*)client->buffer;
if (header->size_compressed) {
int rc = 0;
unsigned int size_u = 1 + header->size_uncompressed;
/* never let buf size fall below our max size required for ipc reads. */
unsigned int new_buf_size = QB_MAX((sizeof(pcmk__ipc_header_t) + size_u), client->max_buf_size);
char *uncompressed = pcmk__assert_alloc(1, new_buf_size);
crm_trace("Decompressing message data %u bytes into %u bytes",
header->size_compressed, size_u);
rc = BZ2_bzBuffToBuffDecompress(uncompressed + sizeof(pcmk__ipc_header_t), &size_u,
client->buffer + sizeof(pcmk__ipc_header_t), header->size_compressed, 1, 0);
rc = pcmk__bzlib2rc(rc);
if (rc != pcmk_rc_ok) {
crm_err("Decompression failed: %s " QB_XS " rc=%d",
pcmk_rc_str(rc), rc);
free(uncompressed);
return rc;
}
pcmk__assert(size_u == header->size_uncompressed);
memcpy(uncompressed, client->buffer, sizeof(pcmk__ipc_header_t)); /* Preserve the header */
header = (pcmk__ipc_header_t *)(void*)uncompressed;
free(client->buffer);
client->buf_size = new_buf_size;
client->buffer = uncompressed;
}
pcmk__assert(client->buffer[sizeof(pcmk__ipc_header_t)
+ header->size_uncompressed - 1] == 0);
return pcmk_rc_ok;
}
long
crm_ipc_read(crm_ipc_t * client)
{
pcmk__ipc_header_t *header = NULL;
pcmk__assert((client != NULL) && (client->ipc != NULL)
&& (client->buffer != NULL));
client->buffer[0] = 0;
client->msg_size = qb_ipcc_event_recv(client->ipc, client->buffer,
client->buf_size, 0);
if (client->msg_size >= 0) {
int rc = crm_ipc_decompress(client);
if (rc != pcmk_rc_ok) {
return pcmk_rc2legacy(rc);
}
header = (pcmk__ipc_header_t *)(void*)client->buffer;
if (!pcmk__valid_ipc_header(header)) {
return -EBADMSG;
}
crm_trace("Received %s IPC event %d size=%u rc=%d text='%.100s'",
client->server_name, header->qb.id, header->qb.size,
client->msg_size,
client->buffer + sizeof(pcmk__ipc_header_t));
} else {
crm_trace("No message received from %s IPC: %s",
client->server_name, pcmk_strerror(client->msg_size));
if (client->msg_size == -EAGAIN) {
return -EAGAIN;
}
}
if (!crm_ipc_connected(client) || client->msg_size == -ENOTCONN) {
crm_err("Connection to %s IPC failed", client->server_name);
}
if (header) {
/* Data excluding the header */
return header->size_uncompressed;
}
return -ENOMSG;
}
const char *
crm_ipc_buffer(crm_ipc_t * client)
{
pcmk__assert(client != NULL);
return client->buffer + sizeof(pcmk__ipc_header_t);
}
uint32_t
crm_ipc_buffer_flags(crm_ipc_t * client)
{
pcmk__ipc_header_t *header = NULL;
pcmk__assert(client != NULL);
if (client->buffer == NULL) {
return 0;
}
header = (pcmk__ipc_header_t *)(void*)client->buffer;
return header->flags;
}
const char *
crm_ipc_name(crm_ipc_t * client)
{
pcmk__assert(client != NULL);
return client->server_name;
}
// \return Standard Pacemaker return code
static int
internal_ipc_get_reply(crm_ipc_t *client, int request_id, int ms_timeout,
ssize_t *bytes)
{
time_t timeout = time(NULL) + 1 + pcmk__timeout_ms2s(ms_timeout);
int rc = pcmk_rc_ok;
/* get the reply */
crm_trace("Waiting on reply to %s IPC message %d",
client->server_name, request_id);
do {
*bytes = qb_ipcc_recv(client->ipc, client->buffer, client->buf_size, 1000);
if (*bytes > 0) {
pcmk__ipc_header_t *hdr = NULL;
rc = crm_ipc_decompress(client);
if (rc != pcmk_rc_ok) {
return rc;
}
hdr = (pcmk__ipc_header_t *)(void*)client->buffer;
if (hdr->qb.id == request_id) {
/* Got it */
break;
} else if (hdr->qb.id < request_id) {
xmlNode *bad = pcmk__xml_parse(crm_ipc_buffer(client));
crm_err("Discarding old reply %d (need %d)", hdr->qb.id, request_id);
crm_log_xml_notice(bad, "OldIpcReply");
} else {
xmlNode *bad = pcmk__xml_parse(crm_ipc_buffer(client));
crm_err("Discarding newer reply %d (need %d)", hdr->qb.id, request_id);
crm_log_xml_notice(bad, "ImpossibleReply");
pcmk__assert(hdr->qb.id <= request_id);
}
} else if (!crm_ipc_connected(client)) {
crm_err("%s IPC provider disconnected while waiting for message %d",
client->server_name, request_id);
break;
}
} while (time(NULL) < timeout);
if (*bytes < 0) {
rc = (int) -*bytes; // System errno
}
return rc;
}
/*!
* \brief Send an IPC XML message
*
* \param[in,out] client Connection to IPC server
* \param[in] message XML message to send
* \param[in] flags Bitmask of crm_ipc_flags
* \param[in] ms_timeout Give up if not sent within this much time
* (5 seconds if 0, or no timeout if negative)
* \param[out] reply Reply from server (or NULL if none)
*
* \return Negative errno on error, otherwise size of reply received in bytes
* if reply was needed, otherwise number of bytes sent
*/
int
crm_ipc_send(crm_ipc_t *client, const xmlNode *message,
enum crm_ipc_flags flags, int32_t ms_timeout, xmlNode **reply)
{
int rc = 0;
ssize_t qb_rc = 0;
ssize_t bytes = 0;
struct iovec *iov;
static uint32_t id = 0;
static int factor = 8;
pcmk__ipc_header_t *header;
if (client == NULL) {
crm_notice("Can't send IPC request without connection (bug?): %.100s",
message);
return -ENOTCONN;
} else if (!crm_ipc_connected(client)) {
/* Don't even bother */
crm_notice("Can't send %s IPC requests: Connection closed",
client->server_name);
return -ENOTCONN;
}
if (ms_timeout == 0) {
ms_timeout = 5000;
}
if (client->need_reply) {
qb_rc = qb_ipcc_recv(client->ipc, client->buffer, client->buf_size, ms_timeout);
if (qb_rc < 0) {
crm_warn("Sending %s IPC disabled until pending reply received",
client->server_name);
return -EALREADY;
} else {
crm_notice("Sending %s IPC re-enabled after pending reply received",
client->server_name);
client->need_reply = FALSE;
}
}
id++;
CRM_LOG_ASSERT(id != 0); /* Crude wrap-around detection */
rc = pcmk__ipc_prepare_iov(id, message, client->max_buf_size, &iov, &bytes);
if (rc != pcmk_rc_ok) {
crm_warn("Couldn't prepare %s IPC request: %s " QB_XS " rc=%d",
client->server_name, pcmk_rc_str(rc), rc);
return pcmk_rc2legacy(rc);
}
header = iov[0].iov_base;
pcmk__set_ipc_flags(header->flags, client->server_name, flags);
if (pcmk_is_set(flags, crm_ipc_proxied)) {
/* Don't look for a synchronous response */
pcmk__clear_ipc_flags(flags, "client", crm_ipc_client_response);
}
if(header->size_compressed) {
if(factor < 10 && (client->max_buf_size / 10) < (bytes / factor)) {
crm_notice("Compressed message exceeds %d0%% of configured IPC "
"limit (%u bytes); consider setting PCMK_ipc_buffer to "
"%u or higher",
factor, client->max_buf_size, 2 * client->max_buf_size);
factor++;
}
}
crm_trace("Sending %s IPC request %d of %u bytes using %dms timeout",
client->server_name, header->qb.id, header->qb.size, ms_timeout);
if ((ms_timeout > 0) || !pcmk_is_set(flags, crm_ipc_client_response)) {
time_t timeout = time(NULL) + 1 + pcmk__timeout_ms2s(ms_timeout);
do {
/* @TODO Is this check really needed? Won't qb_ipcc_sendv() return
* an error if it's not connected?
*/
if (!crm_ipc_connected(client)) {
goto send_cleanup;
}
qb_rc = qb_ipcc_sendv(client->ipc, iov, 2);
} while ((qb_rc == -EAGAIN) && (time(NULL) < timeout));
rc = (int) qb_rc; // Negative of system errno, or bytes sent
if (qb_rc <= 0) {
goto send_cleanup;
} else if (!pcmk_is_set(flags, crm_ipc_client_response)) {
crm_trace("Not waiting for reply to %s IPC request %d",
client->server_name, header->qb.id);
goto send_cleanup;
}
rc = internal_ipc_get_reply(client, header->qb.id, ms_timeout, &bytes);
if (rc != pcmk_rc_ok) {
/* We didn't get the reply in time, so disable future sends for now.
* The only alternative would be to close the connection since we
* don't know how to detect and discard out-of-sequence replies.
*
* @TODO Implement out-of-sequence detection
*/
client->need_reply = TRUE;
}
rc = (int) bytes; // Negative system errno, or size of reply received
} else {
// No timeout, and client response needed
do {
qb_rc = qb_ipcc_sendv_recv(client->ipc, iov, 2, client->buffer,
client->buf_size, -1);
} while ((qb_rc == -EAGAIN) && crm_ipc_connected(client));
rc = (int) qb_rc; // Negative system errno, or size of reply received
}
if (rc > 0) {
pcmk__ipc_header_t *hdr = (pcmk__ipc_header_t *)(void*)client->buffer;
crm_trace("Received %d-byte reply %d to %s IPC %d: %.100s",
rc, hdr->qb.id, client->server_name, header->qb.id,
crm_ipc_buffer(client));
if (reply) {
*reply = pcmk__xml_parse(crm_ipc_buffer(client));
}
} else {
crm_trace("No reply to %s IPC %d: rc=%d",
client->server_name, header->qb.id, rc);
}
send_cleanup:
if (!crm_ipc_connected(client)) {
crm_notice("Couldn't send %s IPC request %d: Connection closed "
QB_XS " rc=%d", client->server_name, header->qb.id, rc);
} else if (rc == -ETIMEDOUT) {
crm_warn("%s IPC request %d failed: %s after %dms " QB_XS " rc=%d",
client->server_name, header->qb.id, pcmk_strerror(rc),
ms_timeout, rc);
crm_write_blackbox(0, NULL);
} else if (rc <= 0) {
crm_warn("%s IPC request %d failed: %s " QB_XS " rc=%d",
client->server_name, header->qb.id,
((rc == 0)? "No bytes sent" : pcmk_strerror(rc)), rc);
}
pcmk_free_ipc_event(iov);
return rc;
}
/*!
* \brief Ensure an IPC provider has expected user or group
*
* \param[in] qb_ipc libqb client connection if available
* \param[in] sock Connected Unix socket for IPC
* \param[in] refuid Expected user ID
* \param[in] refgid Expected group ID
* \param[out] gotpid If not NULL, where to store provider's actual process ID
* (or 1 on platforms where ID is not available)
* \param[out] gotuid If not NULL, where to store provider's actual user ID
* \param[out] gotgid If not NULL, where to store provider's actual group ID
*
* \return Standard Pacemaker return code
* \note An actual user ID of 0 (root) will always be considered authorized,
* regardless of the expected values provided. The caller can use the
* output arguments to be stricter than this function.
*/
static int
is_ipc_provider_expected(qb_ipcc_connection_t *qb_ipc, int sock,
uid_t refuid, gid_t refgid,
pid_t *gotpid, uid_t *gotuid, gid_t *gotgid)
{
int rc = EOPNOTSUPP;
pid_t found_pid = 0;
uid_t found_uid = 0;
gid_t found_gid = 0;
#ifdef HAVE_QB_IPCC_AUTH_GET
if (qb_ipc != NULL) {
rc = qb_ipcc_auth_get(qb_ipc, &found_pid, &found_uid, &found_gid);
rc = -rc; // libqb returns 0 or -errno
if (rc == pcmk_rc_ok) {
goto found;
}
}
#endif
#ifdef HAVE_UCRED
{
struct ucred ucred;
socklen_t ucred_len = sizeof(ucred);
if (getsockopt(sock, SOL_SOCKET, SO_PEERCRED, &ucred, &ucred_len) < 0) {
rc = errno;
} else if (ucred_len != sizeof(ucred)) {
rc = EOPNOTSUPP;
} else {
found_pid = ucred.pid;
found_uid = ucred.uid;
found_gid = ucred.gid;
goto found;
}
}
#endif
#ifdef HAVE_SOCKPEERCRED
{
struct sockpeercred sockpeercred;
socklen_t sockpeercred_len = sizeof(sockpeercred);
if (getsockopt(sock, SOL_SOCKET, SO_PEERCRED,
&sockpeercred, &sockpeercred_len) < 0) {
rc = errno;
} else if (sockpeercred_len != sizeof(sockpeercred)) {
rc = EOPNOTSUPP;
} else {
found_pid = sockpeercred.pid;
found_uid = sockpeercred.uid;
found_gid = sockpeercred.gid;
goto found;
}
}
#endif
#ifdef HAVE_GETPEEREID // For example, FreeBSD
if (getpeereid(sock, &found_uid, &found_gid) < 0) {
rc = errno;
} else {
found_pid = PCMK__SPECIAL_PID;
goto found;
}
#endif
#ifdef HAVE_GETPEERUCRED
{
ucred_t *ucred = NULL;
if (getpeerucred(sock, &ucred) < 0) {
rc = errno;
} else {
found_pid = ucred_getpid(ucred);
found_uid = ucred_geteuid(ucred);
found_gid = ucred_getegid(ucred);
ucred_free(ucred);
goto found;
}
}
#endif
return rc; // If we get here, nothing succeeded
found:
if (gotpid != NULL) {
*gotpid = found_pid;
}
if (gotuid != NULL) {
*gotuid = found_uid;
}
if (gotgid != NULL) {
*gotgid = found_gid;
}
if ((found_uid != 0) && (found_uid != refuid) && (found_gid != refgid)) {
return pcmk_rc_ipc_unauthorized;
}
return pcmk_rc_ok;
}
int
crm_ipc_is_authentic_process(int sock, uid_t refuid, gid_t refgid,
pid_t *gotpid, uid_t *gotuid, gid_t *gotgid)
{
int ret = is_ipc_provider_expected(NULL, sock, refuid, refgid,
gotpid, gotuid, gotgid);
/* The old function had some very odd return codes*/
if (ret == 0) {
return 1;
} else if (ret == pcmk_rc_ipc_unauthorized) {
return 0;
} else {
return pcmk_rc2legacy(ret);
}
}
int
pcmk__ipc_is_authentic_process_active(const char *name, uid_t refuid,
gid_t refgid, pid_t *gotpid)
{
static char last_asked_name[PATH_MAX / 2] = ""; /* log spam prevention */
int fd;
int rc = pcmk_rc_ipc_unresponsive;
int auth_rc = 0;
int32_t qb_rc;
pid_t found_pid = 0; uid_t found_uid = 0; gid_t found_gid = 0;
qb_ipcc_connection_t *c;
#ifdef HAVE_QB_IPCC_CONNECT_ASYNC
struct pollfd pollfd = { 0, };
int poll_rc;
c = qb_ipcc_connect_async(name, 0,
&(pollfd.fd));
#else
c = qb_ipcc_connect(name, 0);
#endif
if (c == NULL) {
crm_info("Could not connect to %s IPC: %s", name, strerror(errno));
rc = pcmk_rc_ipc_unresponsive;
goto bail;
}
#ifdef HAVE_QB_IPCC_CONNECT_ASYNC
pollfd.events = POLLIN;
do {
poll_rc = poll(&pollfd, 1, 2000);
} while ((poll_rc == -1) && (errno == EINTR));
/* If poll() failed, given that disconnect function is not registered yet,
* qb_ipcc_disconnect() won't clean up the socket. In any case, call
* qb_ipcc_connect_continue() here so that it may fail and do the cleanup
* for us.
*/
if (qb_ipcc_connect_continue(c) != 0) {
crm_info("Could not connect to %s IPC: %s", name,
(poll_rc == 0)?"timeout":strerror(errno));
rc = pcmk_rc_ipc_unresponsive;
c = NULL; // qb_ipcc_connect_continue cleaned up for us
goto bail;
}
#endif
qb_rc = qb_ipcc_fd_get(c, &fd);
if (qb_rc != 0) {
rc = (int) -qb_rc; // System errno
crm_err("Could not get fd from %s IPC: %s " QB_XS " rc=%d",
name, pcmk_rc_str(rc), rc);
goto bail;
}
auth_rc = is_ipc_provider_expected(c, fd, refuid, refgid,
&found_pid, &found_uid, &found_gid);
if (auth_rc == pcmk_rc_ipc_unauthorized) {
crm_err("Daemon (IPC %s) effectively blocked with unauthorized"
" process %lld (uid: %lld, gid: %lld)",
name, (long long) PCMK__SPECIAL_PID_AS_0(found_pid),
(long long) found_uid, (long long) found_gid);
rc = pcmk_rc_ipc_unauthorized;
goto bail;
}
if (auth_rc != pcmk_rc_ok) {
rc = auth_rc;
crm_err("Could not get peer credentials from %s IPC: %s "
QB_XS " rc=%d", name, pcmk_rc_str(rc), rc);
goto bail;
}
if (gotpid != NULL) {
*gotpid = found_pid;
}
rc = pcmk_rc_ok;
if ((found_uid != refuid || found_gid != refgid)
&& strncmp(last_asked_name, name, sizeof(last_asked_name))) {
if ((found_uid == 0) && (refuid != 0)) {
crm_warn("Daemon (IPC %s) runs as root, whereas the expected"
" credentials are %lld:%lld, hazard of violating"
" the least privilege principle",
name, (long long) refuid, (long long) refgid);
} else {
crm_notice("Daemon (IPC %s) runs as %lld:%lld, whereas the"
" expected credentials are %lld:%lld, which may"
" mean a different set of privileges than expected",
name, (long long) found_uid, (long long) found_gid,
(long long) refuid, (long long) refgid);
}
memccpy(last_asked_name, name, '\0', sizeof(last_asked_name));
}
bail:
if (c != NULL) {
qb_ipcc_disconnect(c);
}
return rc;
}
// Deprecated functions kept only for backward API compatibility
// LCOV_EXCL_START
#include <crm/common/ipc_client_compat.h>
bool
crm_ipc_connect(crm_ipc_t *client)
{
int rc = pcmk__connect_generic_ipc(client);
if (rc == pcmk_rc_ok) {
return true;
}
if ((client != NULL) && (client->ipc == NULL)) {
errno = (rc > 0)? rc : ENOTCONN;
crm_debug("Could not establish %s IPC connection: %s (%d)",
client->server_name, pcmk_rc_str(errno), errno);
} else if (rc == pcmk_rc_ipc_unauthorized) {
crm_err("%s IPC provider authentication failed",
(client == NULL)? "Pacemaker" : client->server_name);
errno = ECONNABORTED;
} else {
crm_err("Could not verify authenticity of %s IPC provider",
(client == NULL)? "Pacemaker" : client->server_name);
errno = ENOTCONN;
}
return false;
}
// LCOV_EXCL_STOP
// End deprecated API
diff --git a/lib/common/iso8601.c b/lib/common/iso8601.c
index dd2e75de6c..007a64fff5 100644
--- a/lib/common/iso8601.c
+++ b/lib/common/iso8601.c
@@ -1,2251 +1,2251 @@
/*
* Copyright 2005-2024 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU Lesser General Public License
* version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY.
*/
/*
* References:
* https://en.wikipedia.org/wiki/ISO_8601
* http://www.staff.science.uu.nl/~gent0113/calendar/isocalendar.htm
*/
#include <crm_internal.h>
#include <crm/crm.h>
#include <time.h>
#include <ctype.h>
#include <inttypes.h>
#include <limits.h> // INT_MIN, INT_MAX
#include <string.h>
#include <stdbool.h>
#include <crm/common/iso8601.h>
#include <crm/common/iso8601_internal.h>
#include "crmcommon_private.h"
/*
* Andrew's code was originally written for OSes whose "struct tm" contains:
* long tm_gmtoff; :: Seconds east of UTC
* const char *tm_zone; :: Timezone abbreviation
* Some OSes lack these, instead having:
* time_t (or long) timezone;
:: "difference between UTC and local standard time"
* char *tzname[2] = { "...", "..." };
* I (David Lee) confess to not understanding the details. So my attempted
* generalisations for where their use is necessary may be flawed.
*
* 1. Does "difference between ..." subtract the same or opposite way?
* 2. Should it use "altzone" instead of "timezone"?
* 3. Should it use tzname[0] or tzname[1]? Interaction with timezone/altzone?
*/
#if defined(HAVE_STRUCT_TM_TM_GMTOFF)
# define GMTOFF(tm) ((tm)->tm_gmtoff)
#else
/* Note: extern variable; macro argument not actually used. */
# define GMTOFF(tm) (-timezone+daylight)
#endif
#define HOUR_SECONDS (60 * 60)
#define DAY_SECONDS (HOUR_SECONDS * 24)
/*!
* \internal
* \brief Validate a seconds/microseconds tuple
*
* The microseconds value must be in the correct range, and if both are nonzero
* they must have the same sign.
*
* \param[in] sec Seconds
* \param[in] usec Microseconds
*
* \return true if the seconds/microseconds tuple is valid, or false otherwise
*/
#define valid_sec_usec(sec, usec) \
((QB_ABS(usec) < QB_TIME_US_IN_SEC) \
&& (((sec) == 0) || ((usec) == 0) || (((sec) < 0) == ((usec) < 0))))
// A date/time or duration
struct crm_time_s {
int years; // Calendar year (date/time) or number of years (duration)
int months; // Number of months (duration only)
int days; // Ordinal day of year (date/time) or number of days (duration)
int seconds; // Seconds of day (date/time) or number of seconds (duration)
int offset; // Seconds offset from UTC (date/time only)
bool duration; // True if duration
};
static crm_time_t *parse_date(const char *date_str);
static crm_time_t *
crm_get_utc_time(const crm_time_t *dt)
{
crm_time_t *utc = NULL;
if (dt == NULL) {
errno = EINVAL;
return NULL;
}
utc = crm_time_new_undefined();
utc->years = dt->years;
utc->days = dt->days;
utc->seconds = dt->seconds;
utc->offset = 0;
if (dt->offset) {
crm_time_add_seconds(utc, -dt->offset);
} else {
/* Durations (which are the only things that can include months, never have a timezone */
utc->months = dt->months;
}
crm_time_log(LOG_TRACE, "utc-source", dt,
crm_time_log_date | crm_time_log_timeofday | crm_time_log_with_timezone);
crm_time_log(LOG_TRACE, "utc-target", utc,
crm_time_log_date | crm_time_log_timeofday | crm_time_log_with_timezone);
return utc;
}
crm_time_t *
crm_time_new(const char *date_time)
{
tzset();
if (date_time == NULL) {
return pcmk__copy_timet(time(NULL));
}
return parse_date(date_time);
}
/*!
* \brief Allocate memory for an uninitialized time object
*
* \return Newly allocated time object
* \note The caller is responsible for freeing the return value using
* crm_time_free().
*/
crm_time_t *
crm_time_new_undefined(void)
{
return (crm_time_t *) pcmk__assert_alloc(1, sizeof(crm_time_t));
}
/*!
* \brief Check whether a time object has been initialized yet
*
* \param[in] t Time object to check
*
* \return TRUE if time object has been initialized, FALSE otherwise
*/
bool
crm_time_is_defined(const crm_time_t *t)
{
// Any nonzero member indicates something has been done to t
return (t != NULL) && (t->years || t->months || t->days || t->seconds
|| t->offset || t->duration);
}
void
crm_time_free(crm_time_t * dt)
{
if (dt == NULL) {
return;
}
free(dt);
}
static int
year_days(int year)
{
int d = 365;
if (crm_time_leapyear(year)) {
d++;
}
return d;
}
/* From http://myweb.ecu.edu/mccartyr/ISOwdALG.txt :
*
* 5. Find the Jan1Weekday for Y (Monday=1, Sunday=7)
* YY = (Y-1) % 100
* C = (Y-1) - YY
* G = YY + YY/4
* Jan1Weekday = 1 + (((((C / 100) % 4) x 5) + G) % 7)
*/
int
crm_time_january1_weekday(int year)
{
int YY = (year - 1) % 100;
int C = (year - 1) - YY;
int G = YY + YY / 4;
int jan1 = 1 + (((((C / 100) % 4) * 5) + G) % 7);
crm_trace("YY=%d, C=%d, G=%d", YY, C, G);
crm_trace("January 1 %.4d: %d", year, jan1);
return jan1;
}
int
crm_time_weeks_in_year(int year)
{
int weeks = 52;
int jan1 = crm_time_january1_weekday(year);
/* if jan1 == thursday */
if (jan1 == 4) {
weeks++;
} else {
jan1 = crm_time_january1_weekday(year + 1);
/* if dec31 == thursday aka. jan1 of next year is a friday */
if (jan1 == 5) {
weeks++;
}
}
return weeks;
}
// Jan-Dec plus Feb of leap years
static int month_days[13] = {
31, 28, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31, 29
};
/*!
* \brief Return number of days in given month of given year
*
* \param[in] month Ordinal month (1-12)
* \param[in] year Gregorian year
*
* \return Number of days in given month (0 if given month or year is invalid)
*/
int
crm_time_days_in_month(int month, int year)
{
if ((month < 1) || (month > 12) || (year < 1)) {
return 0;
}
if ((month == 2) && crm_time_leapyear(year)) {
month = 13;
}
return month_days[month - 1];
}
bool
crm_time_leapyear(int year)
{
gboolean is_leap = FALSE;
if (year % 4 == 0) {
is_leap = TRUE;
}
if (year % 100 == 0 && year % 400 != 0) {
is_leap = FALSE;
}
return is_leap;
}
/*!
* \internal
* \brief Get ordinal day number of year corresponding to given date
*
* \param[in] y Year
* \param[in] m Month (1-12)
* \param[in] d Day of month (1-31)
*
* \return Day number of year \p y corresponding to month \p m and day \p d,
* or 0 for invalid arguments
*/
static int
get_ordinal_days(uint32_t y, uint32_t m, uint32_t d)
{
int result = 0;
CRM_CHECK((y > 0) && (y <= INT_MAX) && (m >= 1) && (m <= 12)
&& (d >= 1) && (d <= 31), return 0);
result = d;
for (int lpc = 1; lpc < m; lpc++) {
result += crm_time_days_in_month(lpc, y);
}
return result;
}
void
crm_time_log_alias(int log_level, const char *file, const char *function,
int line, const char *prefix, const crm_time_t *date_time,
int flags)
{
char *date_s = crm_time_as_string(date_time, flags);
if (log_level == LOG_STDOUT) {
printf("%s%s%s\n",
(prefix? prefix : ""), (prefix? ": " : ""), date_s);
} else {
do_crm_log_alias(log_level, file, function, line, "%s%s%s",
(prefix? prefix : ""), (prefix? ": " : ""), date_s);
}
free(date_s);
}
static void
crm_time_get_sec(int sec, uint32_t *h, uint32_t *m, uint32_t *s)
{
uint32_t hours, minutes, seconds;
seconds = QB_ABS(sec);
hours = seconds / HOUR_SECONDS;
seconds -= HOUR_SECONDS * hours;
minutes = seconds / 60;
seconds -= 60 * minutes;
crm_trace("%d == %.2" PRIu32 ":%.2" PRIu32 ":%.2" PRIu32,
sec, hours, minutes, seconds);
*h = hours;
*m = minutes;
*s = seconds;
}
int
crm_time_get_timeofday(const crm_time_t *dt, uint32_t *h, uint32_t *m,
uint32_t *s)
{
crm_time_get_sec(dt->seconds, h, m, s);
return TRUE;
}
int
crm_time_get_timezone(const crm_time_t *dt, uint32_t *h, uint32_t *m)
{
uint32_t s;
crm_time_get_sec(dt->seconds, h, m, &s);
return TRUE;
}
long long
crm_time_get_seconds(const crm_time_t *dt)
{
int lpc;
crm_time_t *utc = NULL;
long long in_seconds = 0;
if (dt == NULL) {
return 0;
}
+ // @TODO This is inefficient if dt is already in UTC
utc = crm_get_utc_time(dt);
if (utc == NULL) {
return 0;
}
+ // @TODO We should probably use <= if dt is a duration
for (lpc = 1; lpc < utc->years; lpc++) {
long long dmax = year_days(lpc);
in_seconds += DAY_SECONDS * dmax;
}
- /* utc->months is an offset that can only be set for a duration.
- * By definition, the value is variable depending on the date to
- * which it is applied.
- *
- * Force 30-day months so that something vaguely sane happens
- * for anyone that tries to use a month in this way.
+ /* utc->months can be set only for durations. By definition, the value
+ * varies depending on the (unknown) start date to which the duration will
+ * be applied. Assume 30-day months so that something vaguely sane happens
+ * in this case.
*/
if (utc->months > 0) {
in_seconds += DAY_SECONDS * 30 * (long long) (utc->months);
}
if (utc->days > 0) {
in_seconds += DAY_SECONDS * (long long) (utc->days - 1);
}
in_seconds += utc->seconds;
crm_time_free(utc);
return in_seconds;
}
#define EPOCH_SECONDS 62135596800ULL /* Calculated using crm_time_get_seconds() */
long long
crm_time_get_seconds_since_epoch(const crm_time_t *dt)
{
return (dt == NULL)? 0 : (crm_time_get_seconds(dt) - EPOCH_SECONDS);
}
int
crm_time_get_gregorian(const crm_time_t *dt, uint32_t *y, uint32_t *m,
uint32_t *d)
{
int months = 0;
int days = dt->days;
if(dt->years != 0) {
for (months = 1; months <= 12 && days > 0; months++) {
int mdays = crm_time_days_in_month(months, dt->years);
if (mdays >= days) {
break;
} else {
days -= mdays;
}
}
} else if (dt->months) {
/* This is a duration including months, don't convert the days field */
months = dt->months;
} else {
/* This is a duration not including months, still don't convert the days field */
}
*y = dt->years;
*m = months;
*d = days;
crm_trace("%.4d-%.3d -> %.4d-%.2d-%.2d", dt->years, dt->days, dt->years, months, days);
return TRUE;
}
int
crm_time_get_ordinal(const crm_time_t *dt, uint32_t *y, uint32_t *d)
{
*y = dt->years;
*d = dt->days;
return TRUE;
}
int
crm_time_get_isoweek(const crm_time_t *dt, uint32_t *y, uint32_t *w,
uint32_t *d)
{
/*
* Monday 29 December 2008 is written "2009-W01-1"
* Sunday 3 January 2010 is written "2009-W53-7"
*/
int year_num = 0;
int jan1 = crm_time_january1_weekday(dt->years);
int h = -1;
CRM_CHECK(dt->days > 0, return FALSE);
/* 6. Find the Weekday for Y M D */
h = dt->days + jan1 - 1;
*d = 1 + ((h - 1) % 7);
/* 7. Find if Y M D falls in YearNumber Y-1, WeekNumber 52 or 53 */
if (dt->days <= (8 - jan1) && jan1 > 4) {
crm_trace("year--, jan1=%d", jan1);
year_num = dt->years - 1;
*w = crm_time_weeks_in_year(year_num);
} else {
year_num = dt->years;
}
/* 8. Find if Y M D falls in YearNumber Y+1, WeekNumber 1 */
if (year_num == dt->years) {
int dmax = year_days(year_num);
int correction = 4 - *d;
if ((dmax - dt->days) < correction) {
crm_trace("year++, jan1=%d, i=%d vs. %d", jan1, dmax - dt->days, correction);
year_num = dt->years + 1;
*w = 1;
}
}
/* 9. Find if Y M D falls in YearNumber Y, WeekNumber 1 through 53 */
if (year_num == dt->years) {
int j = dt->days + (7 - *d) + (jan1 - 1);
*w = j / 7;
if (jan1 > 4) {
*w -= 1;
}
}
*y = year_num;
crm_trace("Converted %.4d-%.3d to %.4" PRIu32 "-W%.2" PRIu32 "-%" PRIu32,
dt->years, dt->days, *y, *w, *d);
return TRUE;
}
#define DATE_MAX 128
/*!
* \internal
* \brief Print "<seconds>.<microseconds>" to a buffer
*
* \param[in] sec Seconds
* \param[in] usec Microseconds (must be of same sign as \p sec and of
* absolute value less than \p QB_TIME_US_IN_SEC)
* \param[in,out] buf Result buffer
* \param[in,out] offset Current offset within \p buf
*/
static inline void
sec_usec_as_string(long long sec, int usec, char *buf, size_t *offset)
{
*offset += snprintf(buf + *offset, DATE_MAX - *offset, "%s%lld.%06d",
((sec == 0) && (usec < 0))? "-" : "",
sec, QB_ABS(usec));
}
/*!
* \internal
* \brief Get a string representation of a duration
*
* \param[in] dt Time object to interpret as a duration
* \param[in] usec Microseconds to add to \p dt
* \param[in] show_usec Whether to include microseconds in \p result
* \param[out] result Where to store the result string
*/
static void
crm_duration_as_string(const crm_time_t *dt, int usec, bool show_usec,
char *result)
{
size_t offset = 0;
pcmk__assert(valid_sec_usec(dt->seconds, usec));
if (dt->years) {
offset += snprintf(result + offset, DATE_MAX - offset, "%4d year%s ",
dt->years, pcmk__plural_s(dt->years));
}
if (dt->months) {
offset += snprintf(result + offset, DATE_MAX - offset, "%2d month%s ",
dt->months, pcmk__plural_s(dt->months));
}
if (dt->days) {
offset += snprintf(result + offset, DATE_MAX - offset, "%2d day%s ",
dt->days, pcmk__plural_s(dt->days));
}
// At least print seconds (and optionally usecs)
if ((offset == 0) || (dt->seconds != 0) || (show_usec && (usec != 0))) {
if (show_usec) {
sec_usec_as_string(dt->seconds, usec, result, &offset);
} else {
offset += snprintf(result + offset, DATE_MAX - offset, "%d",
dt->seconds);
}
offset += snprintf(result + offset, DATE_MAX - offset, " second%s",
pcmk__plural_s(dt->seconds));
}
// More than one minute, so provide a more readable breakdown into units
if (QB_ABS(dt->seconds) >= 60) {
uint32_t h = 0;
uint32_t m = 0;
uint32_t s = 0;
uint32_t u = QB_ABS(usec);
bool print_sec_component = false;
crm_time_get_sec(dt->seconds, &h, &m, &s);
print_sec_component = ((s != 0) || (show_usec && (u != 0)));
offset += snprintf(result + offset, DATE_MAX - offset, " (");
if (h) {
offset += snprintf(result + offset, DATE_MAX - offset,
"%" PRIu32 " hour%s%s", h, pcmk__plural_s(h),
((m != 0) || print_sec_component)? " " : "");
}
if (m) {
offset += snprintf(result + offset, DATE_MAX - offset,
"%" PRIu32 " minute%s%s", m, pcmk__plural_s(m),
print_sec_component? " " : "");
}
if (print_sec_component) {
if (show_usec) {
sec_usec_as_string(s, u, result, &offset);
} else {
offset += snprintf(result + offset, DATE_MAX - offset,
"%" PRIu32, s);
}
offset += snprintf(result + offset, DATE_MAX - offset, " second%s",
pcmk__plural_s(dt->seconds));
}
offset += snprintf(result + offset, DATE_MAX - offset, ")");
}
}
/*!
* \internal
* \brief Get a string representation of a time object
*
* \param[in] dt Time to convert to string
* \param[in] usec Microseconds to add to \p dt
* \param[in] flags Group of \p crm_time_* string format options
* \param[out] result Where to store the result string
*
* \note \p result must be of size \p DATE_MAX or larger.
*/
static void
time_as_string_common(const crm_time_t *dt, int usec, uint32_t flags,
char *result)
{
crm_time_t *utc = NULL;
size_t offset = 0;
if (!crm_time_is_defined(dt)) {
strcpy(result, "<undefined time>");
return;
}
pcmk__assert(valid_sec_usec(dt->seconds, usec));
/* Simple cases: as duration, seconds, or seconds since epoch.
* These never depend on time zone.
*/
if (pcmk_is_set(flags, crm_time_log_duration)) {
crm_duration_as_string(dt, usec, pcmk_is_set(flags, crm_time_usecs),
result);
return;
}
if (pcmk_any_flags_set(flags, crm_time_seconds|crm_time_epoch)) {
long long seconds = 0;
if (pcmk_is_set(flags, crm_time_seconds)) {
seconds = crm_time_get_seconds(dt);
} else {
seconds = crm_time_get_seconds_since_epoch(dt);
}
if (pcmk_is_set(flags, crm_time_usecs)) {
sec_usec_as_string(seconds, usec, result, &offset);
} else {
snprintf(result, DATE_MAX, "%lld", seconds);
}
return;
}
// Convert to UTC if local timezone was not requested
if ((dt->offset != 0) && !pcmk_is_set(flags, crm_time_log_with_timezone)) {
crm_trace("UTC conversion");
utc = crm_get_utc_time(dt);
dt = utc;
}
// As readable string
if (pcmk_is_set(flags, crm_time_log_date)) {
if (pcmk_is_set(flags, crm_time_weeks)) { // YYYY-WW-D
uint32_t y = 0;
uint32_t w = 0;
uint32_t d = 0;
if (crm_time_get_isoweek(dt, &y, &w, &d)) {
offset += snprintf(result + offset, DATE_MAX - offset,
"%" PRIu32 "-W%.2" PRIu32 "-%" PRIu32,
y, w, d);
}
} else if (pcmk_is_set(flags, crm_time_ordinal)) { // YYYY-DDD
uint32_t y = 0;
uint32_t d = 0;
if (crm_time_get_ordinal(dt, &y, &d)) {
offset += snprintf(result + offset, DATE_MAX - offset,
"%" PRIu32 "-%.3" PRIu32, y, d);
}
} else { // YYYY-MM-DD
uint32_t y = 0;
uint32_t m = 0;
uint32_t d = 0;
if (crm_time_get_gregorian(dt, &y, &m, &d)) {
offset += snprintf(result + offset, DATE_MAX - offset,
"%.4" PRIu32 "-%.2" PRIu32 "-%.2" PRIu32,
y, m, d);
}
}
}
if (pcmk_is_set(flags, crm_time_log_timeofday)) {
uint32_t h = 0, m = 0, s = 0;
if (offset > 0) {
offset += snprintf(result + offset, DATE_MAX - offset, " ");
}
if (crm_time_get_timeofday(dt, &h, &m, &s)) {
offset += snprintf(result + offset, DATE_MAX - offset,
"%.2" PRIu32 ":%.2" PRIu32 ":%.2" PRIu32,
h, m, s);
if (pcmk_is_set(flags, crm_time_usecs)) {
offset += snprintf(result + offset, DATE_MAX - offset,
".%06" PRIu32, QB_ABS(usec));
}
}
if (pcmk_is_set(flags, crm_time_log_with_timezone)
&& (dt->offset != 0)) {
crm_time_get_sec(dt->offset, &h, &m, &s);
offset += snprintf(result + offset, DATE_MAX - offset,
" %c%.2" PRIu32 ":%.2" PRIu32,
((dt->offset < 0)? '-' : '+'), h, m);
} else {
offset += snprintf(result + offset, DATE_MAX - offset, "Z");
}
}
crm_time_free(utc);
}
/*!
* \brief Get a string representation of a \p crm_time_t object
*
* \param[in] dt Time to convert to string
* \param[in] flags Group of \p crm_time_* string format options
*
* \note The caller is responsible for freeing the return value using \p free().
*/
char *
crm_time_as_string(const crm_time_t *dt, int flags)
{
char result[DATE_MAX] = { '\0', };
time_as_string_common(dt, 0, flags, result);
return pcmk__str_copy(result);
}
/*!
* \internal
* \brief Determine number of seconds from an hour:minute:second string
*
* \param[in] time_str Time specification string
* \param[out] result Number of seconds equivalent to time_str
*
* \return TRUE if specification was valid, FALSE (and set errno) otherwise
* \note This may return the number of seconds in a day (which is out of bounds
* for a time object) if given 24:00:00.
*/
static bool
crm_time_parse_sec(const char *time_str, int *result)
{
int rc;
uint32_t hour = 0;
uint32_t minute = 0;
uint32_t second = 0;
*result = 0;
// Must have at least hour, but minutes and seconds are optional
rc = sscanf(time_str, "%" SCNu32 ":%" SCNu32 ":%" SCNu32,
&hour, &minute, &second);
if (rc == 1) {
rc = sscanf(time_str, "%2" SCNu32 "%2" SCNu32 "%2" SCNu32,
&hour, &minute, &second);
}
if (rc == 0) {
crm_err("%s is not a valid ISO 8601 time specification", time_str);
errno = EINVAL;
return FALSE;
}
crm_trace("Got valid time: %.2" PRIu32 ":%.2" PRIu32 ":%.2" PRIu32,
hour, minute, second);
if ((hour == 24) && (minute == 0) && (second == 0)) {
// Equivalent to 00:00:00 of next day, return number of seconds in day
} else if (hour >= 24) {
crm_err("%s is not a valid ISO 8601 time specification "
"because %" PRIu32 " is not a valid hour", time_str, hour);
errno = EINVAL;
return FALSE;
}
if (minute >= 60) {
crm_err("%s is not a valid ISO 8601 time specification "
"because %" PRIu32 " is not a valid minute", time_str, minute);
errno = EINVAL;
return FALSE;
}
if (second >= 60) {
crm_err("%s is not a valid ISO 8601 time specification "
"because %" PRIu32 " is not a valid second", time_str, second);
errno = EINVAL;
return FALSE;
}
*result = (hour * HOUR_SECONDS) + (minute * 60) + second;
return TRUE;
}
static bool
crm_time_parse_offset(const char *offset_str, int *offset)
{
tzset();
if (offset_str == NULL) {
// Use local offset
#if defined(HAVE_STRUCT_TM_TM_GMTOFF)
time_t now = time(NULL);
struct tm *now_tm = localtime(&now);
#endif
int h_offset = GMTOFF(now_tm) / HOUR_SECONDS;
int m_offset = (GMTOFF(now_tm) - (HOUR_SECONDS * h_offset)) / 60;
if (h_offset < 0 && m_offset < 0) {
m_offset = 0 - m_offset;
}
*offset = (HOUR_SECONDS * h_offset) + (60 * m_offset);
return TRUE;
}
if (offset_str[0] == 'Z') { // @TODO invalid if anything after?
*offset = 0;
return TRUE;
}
*offset = 0;
if ((offset_str[0] == '+') || (offset_str[0] == '-')
|| isdigit((int)offset_str[0])) {
gboolean negate = FALSE;
if (offset_str[0] == '+') {
offset_str++;
} else if (offset_str[0] == '-') {
negate = TRUE;
offset_str++;
}
if (crm_time_parse_sec(offset_str, offset) == FALSE) {
return FALSE;
}
if (negate) {
*offset = 0 - *offset;
}
} // @TODO else invalid?
return TRUE;
}
/*!
* \internal
* \brief Parse the time portion of an ISO 8601 date/time string
*
* \param[in] time_str Time portion of specification (after any 'T')
* \param[in,out] a_time Time object to parse into
*
* \return TRUE if valid time was parsed, FALSE (and set errno) otherwise
* \note This may add a day to a_time (if the time is 24:00:00).
*/
static bool
crm_time_parse(const char *time_str, crm_time_t *a_time)
{
uint32_t h, m, s;
char *offset_s = NULL;
tzset();
if (time_str) {
if (crm_time_parse_sec(time_str, &(a_time->seconds)) == FALSE) {
return FALSE;
}
offset_s = strstr(time_str, "Z");
if (offset_s == NULL) {
offset_s = strstr(time_str, " ");
if (offset_s) {
while (isspace(offset_s[0])) {
offset_s++;
}
}
}
}
if (crm_time_parse_offset(offset_s, &(a_time->offset)) == FALSE) {
return FALSE;
}
crm_time_get_sec(a_time->offset, &h, &m, &s);
crm_trace("Got tz: %c%2." PRIu32 ":%.2" PRIu32,
(a_time->offset < 0)? '-' : '+', h, m);
if (a_time->seconds == DAY_SECONDS) {
// 24:00:00 == 00:00:00 of next day
a_time->seconds = 0;
crm_time_add_days(a_time, 1);
}
return TRUE;
}
/*
* \internal
* \brief Parse a time object from an ISO 8601 date/time specification
*
* \param[in] date_str ISO 8601 date/time specification (or
* \c PCMK__VALUE_EPOCH)
*
* \return New time object on success, NULL (and set errno) otherwise
*/
static crm_time_t *
parse_date(const char *date_str)
{
const char *time_s = NULL;
crm_time_t *dt = NULL;
uint32_t year = 0U;
uint32_t month = 0U;
uint32_t day = 0U;
uint32_t week = 0U;
int rc = 0;
if (pcmk__str_empty(date_str)) {
crm_err("No ISO 8601 date/time specification given");
goto invalid;
}
if ((date_str[0] == 'T')
|| ((strlen(date_str) > 2) && (date_str[2] == ':'))) {
/* Just a time supplied - Infer current date */
dt = crm_time_new(NULL);
if (date_str[0] == 'T') {
time_s = date_str + 1;
} else {
time_s = date_str;
}
goto parse_time;
}
dt = crm_time_new_undefined();
if ((strncasecmp(PCMK__VALUE_EPOCH, date_str, 5) == 0)
&& ((date_str[5] == '\0')
|| (date_str[5] == '/')
|| isspace(date_str[5]))) {
dt->days = 1;
dt->years = 1970;
crm_time_log(LOG_TRACE, "Unpacked", dt, crm_time_log_date | crm_time_log_timeofday);
return dt;
}
/* YYYY-MM-DD */
rc = sscanf(date_str, "%" SCNu32 "-%" SCNu32 "-%" SCNu32 "",
&year, &month, &day);
if (rc == 1) {
/* YYYYMMDD */
rc = sscanf(date_str, "%4" SCNu32 "%2" SCNu32 "%2" SCNu32 "",
&year, &month, &day);
}
if (rc == 3) {
if ((month < 1U) || (month > 12U)) {
crm_err("'%s' is not a valid ISO 8601 date/time specification "
"because '%" PRIu32 "' is not a valid month",
date_str, month);
goto invalid;
} else if ((year < 1U) || (year > INT_MAX)) {
crm_err("'%s' is not a valid ISO 8601 date/time specification "
"because '%" PRIu32 "' is not a valid year",
date_str, year);
goto invalid;
} else if ((day < 1) || (day > INT_MAX)
|| (day > crm_time_days_in_month(month, year))) {
crm_err("'%s' is not a valid ISO 8601 date/time specification "
"because '%" PRIu32 "' is not a valid day of the month",
date_str, day);
goto invalid;
} else {
dt->years = year;
dt->days = get_ordinal_days(year, month, day);
crm_trace("Parsed Gregorian date '%.4" PRIu32 "-%.3d' "
"from date string '%s'", year, dt->days, date_str);
}
goto parse_time;
}
/* YYYY-DDD */
rc = sscanf(date_str, "%" SCNu32 "-%" SCNu32, &year, &day);
if (rc == 2) {
if ((year < 1U) || (year > INT_MAX)) {
crm_err("'%s' is not a valid ISO 8601 date/time specification "
"because '%" PRIu32 "' is not a valid year",
date_str, year);
goto invalid;
} else if ((day < 1U) || (day > INT_MAX) || (day > year_days(year))) {
crm_err("'%s' is not a valid ISO 8601 date/time specification "
"because '%" PRIu32 "' is not a valid day of year %"
PRIu32 " (1-%d)",
date_str, day, year, year_days(year));
goto invalid;
}
crm_trace("Parsed ordinal year %d and days %d from date string '%s'",
year, day, date_str);
dt->days = day;
dt->years = year;
goto parse_time;
}
/* YYYY-Www-D */
rc = sscanf(date_str, "%" SCNu32 "-W%" SCNu32 "-%" SCNu32,
&year, &week, &day);
if (rc == 3) {
if ((week < 1U) || (week > crm_time_weeks_in_year(year))) {
crm_err("'%s' is not a valid ISO 8601 date/time specification "
"because '%" PRIu32 "' is not a valid week of year %"
PRIu32 " (1-%d)",
date_str, week, year, crm_time_weeks_in_year(year));
goto invalid;
} else if ((day < 1U) || (day > 7U)) {
crm_err("'%s' is not a valid ISO 8601 date/time specification "
"because '%" PRIu32 "' is not a valid day of the week",
date_str, day);
goto invalid;
} else {
/*
* See https://en.wikipedia.org/wiki/ISO_week_date
*
* Monday 29 December 2008 is written "2009-W01-1"
* Sunday 3 January 2010 is written "2009-W53-7"
* Saturday 27 September 2008 is written "2008-W37-6"
*
* If 1 January is on a Monday, Tuesday, Wednesday or Thursday, it
* is in week 1. If 1 January is on a Friday, Saturday or Sunday,
* it is in week 52 or 53 of the previous year.
*/
int jan1 = crm_time_january1_weekday(year);
crm_trace("Parsed year %" PRIu32 " (Jan 1 = %d), week %" PRIu32
", and day %" PRIu32 " from date string '%s'",
year, jan1, week, day, date_str);
dt->years = year;
crm_time_add_days(dt, (week - 1) * 7);
if (jan1 <= 4) {
crm_time_add_days(dt, 1 - jan1);
} else {
crm_time_add_days(dt, 8 - jan1);
}
crm_time_add_days(dt, day);
}
goto parse_time;
}
crm_err("'%s' is not a valid ISO 8601 date/time specification", date_str);
goto invalid;
parse_time:
if (time_s == NULL) {
time_s = date_str + strspn(date_str, "0123456789-W");
if ((time_s[0] == ' ') || (time_s[0] == 'T')) {
++time_s;
} else {
time_s = NULL;
}
}
if ((time_s != NULL) && (crm_time_parse(time_s, dt) == FALSE)) {
goto invalid;
}
crm_time_log(LOG_TRACE, "Unpacked", dt, crm_time_log_date | crm_time_log_timeofday);
if (crm_time_check(dt) == FALSE) {
crm_err("'%s' is not a valid ISO 8601 date/time specification",
date_str);
goto invalid;
}
return dt;
invalid:
crm_time_free(dt);
errno = EINVAL;
return NULL;
}
// Parse an ISO 8601 numeric value and return number of characters consumed
static int
parse_int(const char *str, int *result)
{
unsigned int lpc;
int offset = (str[0] == 'T')? 1 : 0;
bool negate = false;
*result = 0;
// @TODO This cannot handle combinations of these characters
switch (str[offset]) {
case '.':
case ',':
return 0; // Fractions are not supported
case '-':
negate = true;
offset++;
break;
case '+':
case ':':
offset++;
break;
default:
break;
}
for (lpc = 0; (lpc < 10) && isdigit(str[offset]); lpc++) {
const int digit = str[offset++] - '0';
if ((*result * 10LL + digit) > INT_MAX) {
return 0; // Overflow
}
*result = *result * 10 + digit;
}
if (negate) {
*result = 0 - *result;
}
return (lpc > 0)? offset : 0;
}
/*!
* \brief Parse a time duration from an ISO 8601 duration specification
*
* \param[in] period_s ISO 8601 duration specification (optionally followed by
* whitespace, after which the rest of the string will be
* ignored)
*
* \return New time object on success, NULL (and set errno) otherwise
* \note It is the caller's responsibility to return the result using
* crm_time_free().
*/
crm_time_t *
crm_time_parse_duration(const char *period_s)
{
gboolean is_time = FALSE;
crm_time_t *diff = NULL;
if (pcmk__str_empty(period_s)) {
crm_err("No ISO 8601 time duration given");
goto invalid;
}
if (period_s[0] != 'P') {
crm_err("'%s' is not a valid ISO 8601 time duration "
"because it does not start with a 'P'", period_s);
goto invalid;
}
if ((period_s[1] == '\0') || isspace(period_s[1])) {
crm_err("'%s' is not a valid ISO 8601 time duration "
"because nothing follows 'P'", period_s);
goto invalid;
}
diff = crm_time_new_undefined();
for (const char *current = period_s + 1;
current[0] && (current[0] != '/') && !isspace(current[0]);
++current) {
int an_int = 0, rc;
long long result = 0LL;
if (current[0] == 'T') {
/* A 'T' separates year/month/day from hour/minute/seconds. We don't
* require it strictly, but just use it to differentiate month from
* minutes.
*/
is_time = TRUE;
continue;
}
// An integer must be next
rc = parse_int(current, &an_int);
if (rc == 0) {
crm_err("'%s' is not a valid ISO 8601 time duration "
"because no valid integer at '%s'", period_s, current);
goto invalid;
}
current += rc;
// A time unit must be next (we're not strict about the order)
switch (current[0]) {
case 'Y':
diff->years = an_int;
break;
case 'M':
if (!is_time) { // Months
diff->months = an_int;
} else { // Minutes
result = diff->seconds + an_int * 60LL;
if ((result < INT_MIN) || (result > INT_MAX)) {
crm_err("'%s' is not a valid ISO 8601 time duration "
"because integer at '%s' is too %s",
period_s, current - rc,
((result > 0)? "large" : "small"));
goto invalid;
} else {
diff->seconds = (int) result;
}
}
break;
case 'W':
result = diff->days + an_int * 7LL;
if ((result < INT_MIN) || (result > INT_MAX)) {
crm_err("'%s' is not a valid ISO 8601 time duration "
"because integer at '%s' is too %s",
period_s, current - rc,
((result > 0)? "large" : "small"));
goto invalid;
} else {
diff->days = (int) result;
}
break;
case 'D':
result = diff->days + (long long) an_int;
if ((result < INT_MIN) || (result > INT_MAX)) {
crm_err("'%s' is not a valid ISO 8601 time duration "
"because integer at '%s' is too %s",
period_s, current - rc,
((result > 0)? "large" : "small"));
goto invalid;
} else {
diff->days = (int) result;
}
break;
case 'H':
result = diff->seconds + (long long) an_int * HOUR_SECONDS;
if ((result < INT_MIN) || (result > INT_MAX)) {
crm_err("'%s' is not a valid ISO 8601 time duration "
"because integer at '%s' is too %s",
period_s, current - rc,
((result > 0)? "large" : "small"));
goto invalid;
} else {
diff->seconds = (int) result;
}
break;
case 'S':
result = diff->seconds + (long long) an_int;
if ((result < INT_MIN) || (result > INT_MAX)) {
crm_err("'%s' is not a valid ISO 8601 time duration "
"because integer at '%s' is too %s",
period_s, current - rc,
((result > 0)? "large" : "small"));
goto invalid;
} else {
diff->seconds = (int) result;
}
break;
case '\0':
crm_err("'%s' is not a valid ISO 8601 time duration "
"because no units after %d", period_s, an_int);
goto invalid;
default:
crm_err("'%s' is not a valid ISO 8601 time duration "
"because '%c' is not a valid time unit",
period_s, current[0]);
goto invalid;
}
}
if (!crm_time_is_defined(diff)) {
crm_err("'%s' is not a valid ISO 8601 time duration "
"because no amounts and units given", period_s);
goto invalid;
}
diff->duration = TRUE;
return diff;
invalid:
crm_time_free(diff);
errno = EINVAL;
return NULL;
}
/*!
* \brief Parse a time period from an ISO 8601 interval specification
*
* \param[in] period_str ISO 8601 interval specification (start/end,
* start/duration, or duration/end)
*
* \return New time period object on success, NULL (and set errno) otherwise
* \note The caller is responsible for freeing the result using
* crm_time_free_period().
*/
crm_time_period_t *
crm_time_parse_period(const char *period_str)
{
const char *original = period_str;
crm_time_period_t *period = NULL;
if (pcmk__str_empty(period_str)) {
crm_err("No ISO 8601 time period given");
goto invalid;
}
tzset();
period = pcmk__assert_alloc(1, sizeof(crm_time_period_t));
if (period_str[0] == 'P') {
period->diff = crm_time_parse_duration(period_str);
if (period->diff == NULL) {
goto error;
}
} else {
period->start = parse_date(period_str);
if (period->start == NULL) {
goto error;
}
}
period_str = strstr(original, "/");
if (period_str) {
++period_str;
if (period_str[0] == 'P') {
if (period->diff != NULL) {
crm_err("'%s' is not a valid ISO 8601 time period "
"because it has two durations",
original);
goto invalid;
}
period->diff = crm_time_parse_duration(period_str);
if (period->diff == NULL) {
goto error;
}
} else {
period->end = parse_date(period_str);
if (period->end == NULL) {
goto error;
}
}
} else if (period->diff != NULL) {
// Only duration given, assume start is now
period->start = crm_time_new(NULL);
} else {
// Only start given
crm_err("'%s' is not a valid ISO 8601 time period "
"because it has no duration or ending time",
original);
goto invalid;
}
if (period->start == NULL) {
period->start = crm_time_subtract(period->end, period->diff);
} else if (period->end == NULL) {
period->end = crm_time_add(period->start, period->diff);
}
if (crm_time_check(period->start) == FALSE) {
crm_err("'%s' is not a valid ISO 8601 time period "
"because the start is invalid", period_str);
goto invalid;
}
if (crm_time_check(period->end) == FALSE) {
crm_err("'%s' is not a valid ISO 8601 time period "
"because the end is invalid", period_str);
goto invalid;
}
return period;
invalid:
errno = EINVAL;
error:
crm_time_free_period(period);
return NULL;
}
/*!
* \brief Free a dynamically allocated time period object
*
* \param[in,out] period Time period to free
*/
void
crm_time_free_period(crm_time_period_t *period)
{
if (period) {
crm_time_free(period->start);
crm_time_free(period->end);
crm_time_free(period->diff);
free(period);
}
}
void
crm_time_set(crm_time_t *target, const crm_time_t *source)
{
crm_trace("target=%p, source=%p", target, source);
CRM_CHECK(target != NULL && source != NULL, return);
target->years = source->years;
target->days = source->days;
target->months = source->months; /* Only for durations */
target->seconds = source->seconds;
target->offset = source->offset;
crm_time_log(LOG_TRACE, "source", source,
crm_time_log_date | crm_time_log_timeofday | crm_time_log_with_timezone);
crm_time_log(LOG_TRACE, "target", target,
crm_time_log_date | crm_time_log_timeofday | crm_time_log_with_timezone);
}
static void
ha_set_tm_time(crm_time_t *target, const struct tm *source)
{
int h_offset = 0;
int m_offset = 0;
/* Ensure target is fully initialized */
target->years = 0;
target->months = 0;
target->days = 0;
target->seconds = 0;
target->offset = 0;
target->duration = FALSE;
if (source->tm_year > 0) {
/* years since 1900 */
target->years = 1900;
crm_time_add_years(target, source->tm_year);
}
if (source->tm_yday >= 0) {
/* days since January 1 [0-365] */
target->days = 1 + source->tm_yday;
}
if (source->tm_hour >= 0) {
target->seconds += HOUR_SECONDS * source->tm_hour;
}
if (source->tm_min >= 0) {
target->seconds += 60 * source->tm_min;
}
if (source->tm_sec >= 0) {
target->seconds += source->tm_sec;
}
/* tm_gmtoff == offset from UTC in seconds */
h_offset = GMTOFF(source) / HOUR_SECONDS;
m_offset = (GMTOFF(source) - (HOUR_SECONDS * h_offset)) / 60;
crm_trace("Time offset is %lds (%.2d:%.2d)",
GMTOFF(source), h_offset, m_offset);
target->offset += HOUR_SECONDS * h_offset;
target->offset += 60 * m_offset;
}
void
crm_time_set_timet(crm_time_t *target, const time_t *source)
{
ha_set_tm_time(target, localtime(source));
}
/*!
* \internal
* \brief Set one time object to another if the other is earlier
*
* \param[in,out] target Time object to set
* \param[in] source Time object to use if earlier
*/
void
pcmk__set_time_if_earlier(crm_time_t *target, const crm_time_t *source)
{
if ((target != NULL) && (source != NULL)
&& (!crm_time_is_defined(target)
|| (crm_time_compare(source, target) < 0))) {
crm_time_set(target, source);
}
}
crm_time_t *
pcmk_copy_time(const crm_time_t *source)
{
crm_time_t *target = crm_time_new_undefined();
crm_time_set(target, source);
return target;
}
/*!
* \internal
* \brief Convert a \p time_t time to a \p crm_time_t time
*
* \param[in] source Time to convert
*
* \return A \p crm_time_t object representing \p source
*/
crm_time_t *
pcmk__copy_timet(time_t source)
{
crm_time_t *target = crm_time_new_undefined();
crm_time_set_timet(target, &source);
return target;
}
crm_time_t *
crm_time_add(const crm_time_t *dt, const crm_time_t *value)
{
crm_time_t *utc = NULL;
crm_time_t *answer = NULL;
if ((dt == NULL) || (value == NULL)) {
errno = EINVAL;
return NULL;
}
answer = pcmk_copy_time(dt);
utc = crm_get_utc_time(value);
if (utc == NULL) {
crm_time_free(answer);
return NULL;
}
crm_time_add_years(answer, utc->years);
crm_time_add_months(answer, utc->months);
crm_time_add_days(answer, utc->days);
crm_time_add_seconds(answer, utc->seconds);
crm_time_free(utc);
return answer;
}
/*!
* \internal
* \brief Return the XML attribute name corresponding to a time component
*
* \param[in] component Component to check
*
* \return XML attribute name corresponding to \p component, or NULL if
* \p component is invalid
*/
const char *
pcmk__time_component_attr(enum pcmk__time_component component)
{
switch (component) {
case pcmk__time_years:
return PCMK_XA_YEARS;
case pcmk__time_months:
return PCMK_XA_MONTHS;
case pcmk__time_weeks:
return PCMK_XA_WEEKS;
case pcmk__time_days:
return PCMK_XA_DAYS;
case pcmk__time_hours:
return PCMK_XA_HOURS;
case pcmk__time_minutes:
return PCMK_XA_MINUTES;
case pcmk__time_seconds:
return PCMK_XA_SECONDS;
default:
return NULL;
}
}
typedef void (*component_fn_t)(crm_time_t *, int);
/*!
* \internal
* \brief Get the addition function corresponding to a time component
* \param[in] component Component to check
*
* \return Addition function corresponding to \p component, or NULL if
* \p component is invalid
*/
static component_fn_t
component_fn(enum pcmk__time_component component)
{
switch (component) {
case pcmk__time_years:
return crm_time_add_years;
case pcmk__time_months:
return crm_time_add_months;
case pcmk__time_weeks:
return crm_time_add_weeks;
case pcmk__time_days:
return crm_time_add_days;
case pcmk__time_hours:
return crm_time_add_hours;
case pcmk__time_minutes:
return crm_time_add_minutes;
case pcmk__time_seconds:
return crm_time_add_seconds;
default:
return NULL;
}
}
/*!
* \internal
* \brief Add the value of an XML attribute to a time object
*
* \param[in,out] t Time object to add to
* \param[in] component Component of \p t to add to
* \param[in] xml XML with value to add
*
* \return Standard Pacemaker return code
*/
int
pcmk__add_time_from_xml(crm_time_t *t, enum pcmk__time_component component,
const xmlNode *xml)
{
long long value;
const char *attr = pcmk__time_component_attr(component);
component_fn_t add = component_fn(component);
if ((t == NULL) || (attr == NULL) || (add == NULL)) {
return EINVAL;
}
if (xml == NULL) {
return pcmk_rc_ok;
}
if (pcmk__scan_ll(crm_element_value(xml, attr), &value,
0LL) != pcmk_rc_ok) {
return pcmk_rc_unpack_error;
}
if ((value < INT_MIN) || (value > INT_MAX)) {
return ERANGE;
}
if (value != 0LL) {
add(t, (int) value);
}
return pcmk_rc_ok;
}
crm_time_t *
crm_time_calculate_duration(const crm_time_t *dt, const crm_time_t *value)
{
crm_time_t *utc = NULL;
crm_time_t *answer = NULL;
if ((dt == NULL) || (value == NULL)) {
errno = EINVAL;
return NULL;
}
utc = crm_get_utc_time(value);
if (utc == NULL) {
return NULL;
}
answer = crm_get_utc_time(dt);
if (answer == NULL) {
crm_time_free(utc);
return NULL;
}
answer->duration = TRUE;
crm_time_add_years(answer, -utc->years);
if(utc->months != 0) {
crm_time_add_months(answer, -utc->months);
}
crm_time_add_days(answer, -utc->days);
crm_time_add_seconds(answer, -utc->seconds);
crm_time_free(utc);
return answer;
}
crm_time_t *
crm_time_subtract(const crm_time_t *dt, const crm_time_t *value)
{
crm_time_t *utc = NULL;
crm_time_t *answer = NULL;
if ((dt == NULL) || (value == NULL)) {
errno = EINVAL;
return NULL;
}
utc = crm_get_utc_time(value);
if (utc == NULL) {
return NULL;
}
answer = pcmk_copy_time(dt);
crm_time_add_years(answer, -utc->years);
if(utc->months != 0) {
crm_time_add_months(answer, -utc->months);
}
crm_time_add_days(answer, -utc->days);
crm_time_add_seconds(answer, -utc->seconds);
crm_time_free(utc);
return answer;
}
/*!
* \brief Check whether a time object represents a sensible date/time
*
* \param[in] dt Date/time object to check
*
* \return \c true if years, days, and seconds are sensible, \c false otherwise
*/
bool
crm_time_check(const crm_time_t *dt)
{
return (dt != NULL)
&& (dt->days > 0) && (dt->days <= year_days(dt->years))
&& (dt->seconds >= 0) && (dt->seconds < DAY_SECONDS);
}
#define do_cmp_field(l, r, field) \
if(rc == 0) { \
if(l->field > r->field) { \
crm_trace("%s: %d > %d", \
#field, l->field, r->field); \
rc = 1; \
} else if(l->field < r->field) { \
crm_trace("%s: %d < %d", \
#field, l->field, r->field); \
rc = -1; \
} \
}
int
crm_time_compare(const crm_time_t *a, const crm_time_t *b)
{
int rc = 0;
crm_time_t *t1 = crm_get_utc_time(a);
crm_time_t *t2 = crm_get_utc_time(b);
if ((t1 == NULL) && (t2 == NULL)) {
rc = 0;
} else if (t1 == NULL) {
rc = -1;
} else if (t2 == NULL) {
rc = 1;
} else {
do_cmp_field(t1, t2, years);
do_cmp_field(t1, t2, days);
do_cmp_field(t1, t2, seconds);
}
crm_time_free(t1);
crm_time_free(t2);
return rc;
}
/*!
* \brief Add a given number of seconds to a date/time or duration
*
* \param[in,out] a_time Date/time or duration to add seconds to
* \param[in] extra Number of seconds to add
*/
void
crm_time_add_seconds(crm_time_t *a_time, int extra)
{
int days = extra / DAY_SECONDS;
pcmk__assert(a_time != NULL);
crm_trace("Adding %d seconds (including %d whole day%s) to %d",
extra, days, pcmk__plural_s(days), a_time->seconds);
a_time->seconds += extra % DAY_SECONDS;
// Check whether the addition crossed a day boundary
if (a_time->seconds > DAY_SECONDS) {
++days;
a_time->seconds -= DAY_SECONDS;
} else if (a_time->seconds < 0) {
--days;
a_time->seconds += DAY_SECONDS;
}
crm_time_add_days(a_time, days);
}
#define ydays(t) (crm_time_leapyear((t)->years)? 366 : 365)
/*!
* \brief Add days to a date/time
*
* \param[in,out] a_time Time to modify
* \param[in] extra Number of days to add (may be negative to subtract)
*/
void
crm_time_add_days(crm_time_t *a_time, int extra)
{
pcmk__assert(a_time != NULL);
crm_trace("Adding %d days to %.4d-%.3d", extra, a_time->years, a_time->days);
if (extra > 0) {
while ((a_time->days + (long long) extra) > ydays(a_time)) {
if ((a_time->years + 1LL) > INT_MAX) {
a_time->days = ydays(a_time); // Clip to latest we can handle
return;
}
extra -= ydays(a_time);
a_time->years++;
}
} else if (extra < 0) {
const int min_days = a_time->duration? 0 : 1;
while ((a_time->days + (long long) extra) < min_days) {
if ((a_time->years - 1) < 1) {
a_time->days = 1; // Clip to earliest we can handle (no BCE)
return;
}
a_time->years--;
extra += ydays(a_time);
}
}
a_time->days += extra;
}
void
crm_time_add_months(crm_time_t * a_time, int extra)
{
int lpc;
uint32_t y, m, d, dmax;
crm_time_get_gregorian(a_time, &y, &m, &d);
crm_trace("Adding %d months to %.4" PRIu32 "-%.2" PRIu32 "-%.2" PRIu32,
extra, y, m, d);
if (extra > 0) {
for (lpc = extra; lpc > 0; lpc--) {
m++;
if (m == 13) {
m = 1;
y++;
}
}
} else {
for (lpc = -extra; lpc > 0; lpc--) {
m--;
if (m == 0) {
m = 12;
y--;
}
}
}
dmax = crm_time_days_in_month(m, y);
if (dmax < d) {
/* Preserve day-of-month unless the month doesn't have enough days */
d = dmax;
}
crm_trace("Calculated %.4" PRIu32 "-%.2" PRIu32 "-%.2" PRIu32, y, m, d);
a_time->years = y;
a_time->days = get_ordinal_days(y, m, d);
crm_time_get_gregorian(a_time, &y, &m, &d);
crm_trace("Got %.4" PRIu32 "-%.2" PRIu32 "-%.2" PRIu32, y, m, d);
}
void
crm_time_add_minutes(crm_time_t * a_time, int extra)
{
crm_time_add_seconds(a_time, extra * 60);
}
void
crm_time_add_hours(crm_time_t * a_time, int extra)
{
crm_time_add_seconds(a_time, extra * HOUR_SECONDS);
}
void
crm_time_add_weeks(crm_time_t * a_time, int extra)
{
crm_time_add_days(a_time, extra * 7);
}
void
crm_time_add_years(crm_time_t * a_time, int extra)
{
pcmk__assert(a_time != NULL);
if ((extra > 0) && ((a_time->years + (long long) extra) > INT_MAX)) {
a_time->years = INT_MAX;
} else if ((extra < 0) && ((a_time->years + (long long) extra) < 1)) {
a_time->years = 1; // Clip to earliest we can handle (no BCE)
} else {
a_time->years += extra;
}
}
static void
ha_get_tm_time(struct tm *target, const crm_time_t *source)
{
*target = (struct tm) {
.tm_year = source->years - 1900,
.tm_mday = source->days,
.tm_sec = source->seconds % 60,
.tm_min = ( source->seconds / 60 ) % 60,
.tm_hour = source->seconds / HOUR_SECONDS,
.tm_isdst = -1, /* don't adjust */
#if defined(HAVE_STRUCT_TM_TM_GMTOFF)
.tm_gmtoff = source->offset
#endif
};
mktime(target);
}
/* The high-resolution variant of time object was added to meet an immediate
* need, and is kept internal API.
*
* @TODO The long-term goal is to come up with a clean, unified design for a
* time type (or types) that meets all the various needs, to replace
* crm_time_t, pcmk__time_hr_t, and struct timespec (in lrmd_cmd_t).
*/
pcmk__time_hr_t *
pcmk__time_hr_convert(pcmk__time_hr_t *target, const crm_time_t *dt)
{
pcmk__time_hr_t *hr_dt = NULL;
if (dt) {
hr_dt = target;
if (hr_dt == NULL) {
hr_dt = pcmk__assert_alloc(1, sizeof(pcmk__time_hr_t));
}
*hr_dt = (pcmk__time_hr_t) {
.years = dt->years,
.months = dt->months,
.days = dt->days,
.seconds = dt->seconds,
.offset = dt->offset,
.duration = dt->duration
};
}
return hr_dt;
}
void
pcmk__time_set_hr_dt(crm_time_t *target, const pcmk__time_hr_t *hr_dt)
{
pcmk__assert((target != NULL) && (hr_dt != NULL));
*target = (crm_time_t) {
.years = hr_dt->years,
.months = hr_dt->months,
.days = hr_dt->days,
.seconds = hr_dt->seconds,
.offset = hr_dt->offset,
.duration = hr_dt->duration
};
}
/*!
* \internal
* \brief Return the current time as a high-resolution time
*
* \param[out] epoch If not NULL, this will be set to seconds since epoch
*
* \return Newly allocated high-resolution time set to the current time
*/
pcmk__time_hr_t *
pcmk__time_hr_now(time_t *epoch)
{
struct timespec tv;
crm_time_t dt;
pcmk__time_hr_t *hr;
qb_util_timespec_from_epoch_get(&tv);
if (epoch != NULL) {
*epoch = tv.tv_sec;
}
crm_time_set_timet(&dt, &(tv.tv_sec));
hr = pcmk__time_hr_convert(NULL, &dt);
if (hr != NULL) {
hr->useconds = tv.tv_nsec / QB_TIME_NS_IN_USEC;
}
return hr;
}
pcmk__time_hr_t *
pcmk__time_hr_new(const char *date_time)
{
pcmk__time_hr_t *hr_dt = NULL;
if (date_time == NULL) {
hr_dt = pcmk__time_hr_now(NULL);
} else {
crm_time_t *dt;
dt = parse_date(date_time);
hr_dt = pcmk__time_hr_convert(NULL, dt);
crm_time_free(dt);
}
return hr_dt;
}
void
pcmk__time_hr_free(pcmk__time_hr_t * hr_dt)
{
free(hr_dt);
}
/*!
* \internal
* \brief Expand a date/time format string, including %N for nanoseconds
*
* \param[in] format Date/time format string as per strftime(3) with the
* addition of %N for nanoseconds
* \param[in] hr_dt Time value to format
*
* \return Newly allocated string with formatted string
*/
char *
pcmk__time_format_hr(const char *format, const pcmk__time_hr_t *hr_dt)
{
int scanned_pos = 0; // How many characters of format have been parsed
int printed_pos = 0; // How many characters of format have been processed
size_t date_len = 0;
char nano_s[10] = { '\0', };
char date_s[128] = { '\0', };
struct tm tm = { 0, };
crm_time_t dt = { 0, };
if (format == NULL) {
return NULL;
}
pcmk__time_set_hr_dt(&dt, hr_dt);
ha_get_tm_time(&tm, &dt);
sprintf(nano_s, "%06d000", hr_dt->useconds);
while (format[scanned_pos] != '\0') {
int fmt_pos; // Index after last character to pass as-is
int nano_digits = 0; // Length of %N field width (if any)
char *tmp_fmt_s = NULL;
size_t nbytes = 0;
// Look for next format specifier
const char *mark_s = strchr(&format[scanned_pos], '%');
if (mark_s == NULL) {
// No more specifiers, so pass remaining string to strftime() as-is
scanned_pos = strlen(format);
fmt_pos = scanned_pos;
} else {
fmt_pos = mark_s - format; // Index of %
// Skip % and any field width
scanned_pos = fmt_pos + 1;
while (isdigit(format[scanned_pos])) {
scanned_pos++;
}
switch (format[scanned_pos]) {
case '\0': // Literal % and possibly digits at end of string
fmt_pos = scanned_pos; // Pass remaining string as-is
break;
case 'N': // %[width]N
scanned_pos++;
// Parse field width
nano_digits = atoi(&format[fmt_pos + 1]);
nano_digits = QB_MAX(nano_digits, 0);
nano_digits = QB_MIN(nano_digits, 6);
break;
default: // Some other specifier
if (format[++scanned_pos] != '\0') { // More to parse
continue;
}
fmt_pos = scanned_pos; // Pass remaining string as-is
break;
}
}
if (date_len >= sizeof(date_s)) {
return NULL; // No room for remaining string
}
tmp_fmt_s = strndup(&format[printed_pos], fmt_pos - printed_pos);
#ifdef HAVE_FORMAT_NONLITERAL
#pragma GCC diagnostic push
#pragma GCC diagnostic ignored "-Wformat-nonliteral"
#endif
nbytes = strftime(&date_s[date_len], sizeof(date_s) - date_len,
tmp_fmt_s, &tm);
#ifdef HAVE_FORMAT_NONLITERAL
#pragma GCC diagnostic pop
#endif
free(tmp_fmt_s);
if (nbytes == 0) { // Would overflow buffer
return NULL;
}
date_len += nbytes;
printed_pos = scanned_pos;
if (nano_digits != 0) {
int nc = 0;
if (date_len >= sizeof(date_s)) {
return NULL; // No room to add nanoseconds
}
nc = snprintf(&date_s[date_len], sizeof(date_s) - date_len,
"%.*s", nano_digits, nano_s);
if ((nc < 0) || (nc == (sizeof(date_s) - date_len))) {
return NULL; // Error or would overflow buffer
}
date_len += nc;
}
}
return (date_len == 0)? NULL : pcmk__str_copy(date_s);
}
/*!
* \internal
* \brief Return a human-friendly string corresponding to an epoch time value
*
* \param[in] source Pointer to epoch time value (or \p NULL for current time)
* \param[in] flags Group of \p crm_time_* flags controlling display format
* (0 to use \p ctime() with newline removed)
*
* \return String representation of \p source on success (may be empty depending
* on \p flags; guaranteed not to be \p NULL)
*
* \note The caller is responsible for freeing the return value using \p free().
*/
char *
pcmk__epoch2str(const time_t *source, uint32_t flags)
{
time_t epoch_time = (source == NULL)? time(NULL) : *source;
if (flags == 0) {
return pcmk__str_copy(pcmk__trim(ctime(&epoch_time)));
} else {
crm_time_t dt;
crm_time_set_timet(&dt, &epoch_time);
return crm_time_as_string(&dt, flags);
}
}
/*!
* \internal
* \brief Return a human-friendly string corresponding to seconds-and-
* nanoseconds value
*
* Time is shown with microsecond resolution if \p crm_time_usecs is in \p
* flags.
*
* \param[in] ts Time in seconds and nanoseconds (or \p NULL for current
* time)
* \param[in] flags Group of \p crm_time_* flags controlling display format
*
* \return String representation of \p ts on success (may be empty depending on
* \p flags; guaranteed not to be \p NULL)
*
* \note The caller is responsible for freeing the return value using \p free().
*/
char *
pcmk__timespec2str(const struct timespec *ts, uint32_t flags)
{
struct timespec tmp_ts;
crm_time_t dt;
char result[DATE_MAX] = { 0 };
if (ts == NULL) {
qb_util_timespec_from_epoch_get(&tmp_ts);
ts = &tmp_ts;
}
crm_time_set_timet(&dt, &ts->tv_sec);
time_as_string_common(&dt, ts->tv_nsec / QB_TIME_NS_IN_USEC, flags, result);
return pcmk__str_copy(result);
}
/*!
* \internal
* \brief Given a millisecond interval, return a log-friendly string
*
* \param[in] interval_ms Interval in milliseconds
*
* \return Readable version of \p interval_ms
*
* \note The return value is a pointer to static memory that will be
* overwritten by later calls to this function.
*/
const char *
pcmk__readable_interval(guint interval_ms)
{
#define MS_IN_S (1000)
#define MS_IN_M (MS_IN_S * 60)
#define MS_IN_H (MS_IN_M * 60)
#define MS_IN_D (MS_IN_H * 24)
#define MAXSTR sizeof("..d..h..m..s...ms")
static char str[MAXSTR];
int offset = 0;
str[0] = '\0';
if (interval_ms >= MS_IN_D) {
offset += snprintf(str + offset, MAXSTR - offset, "%ud",
interval_ms / MS_IN_D);
interval_ms -= (interval_ms / MS_IN_D) * MS_IN_D;
}
if (interval_ms >= MS_IN_H) {
offset += snprintf(str + offset, MAXSTR - offset, "%uh",
interval_ms / MS_IN_H);
interval_ms -= (interval_ms / MS_IN_H) * MS_IN_H;
}
if (interval_ms >= MS_IN_M) {
offset += snprintf(str + offset, MAXSTR - offset, "%um",
interval_ms / MS_IN_M);
interval_ms -= (interval_ms / MS_IN_M) * MS_IN_M;
}
// Ns, N.NNNs, or NNNms
if (interval_ms >= MS_IN_S) {
offset += snprintf(str + offset, MAXSTR - offset, "%u",
interval_ms / MS_IN_S);
interval_ms -= (interval_ms / MS_IN_S) * MS_IN_S;
if (interval_ms > 0) {
offset += snprintf(str + offset, MAXSTR - offset, ".%03u",
interval_ms);
}
(void) snprintf(str + offset, MAXSTR - offset, "s");
} else if (interval_ms > 0) {
(void) snprintf(str + offset, MAXSTR - offset, "%ums", interval_ms);
} else if (str[0] == '\0') {
strcpy(str, "0s");
}
return str;
}
diff --git a/lib/pacemaker/pcmk_graph_producer.c b/lib/pacemaker/pcmk_graph_producer.c
index 31ea0227d6..e6cff3e26c 100644
--- a/lib/pacemaker/pcmk_graph_producer.c
+++ b/lib/pacemaker/pcmk_graph_producer.c
@@ -1,1107 +1,1108 @@
/*
* Copyright 2004-2024 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU General Public License version 2
* or later (GPLv2+) WITHOUT ANY WARRANTY.
*/
#include <crm_internal.h>
#include <sys/param.h>
#include <crm/crm.h>
#include <crm/cib.h>
#include <crm/common/xml.h>
#include <glib.h>
#include <pacemaker-internal.h>
#include "libpacemaker_private.h"
// Convenience macros for logging action properties
#define action_type_str(flags) \
(pcmk_is_set((flags), pcmk__action_pseudo)? "pseudo-action" : "action")
#define action_optional_str(flags) \
(pcmk_is_set((flags), pcmk__action_optional)? "optional" : "required")
#define action_runnable_str(flags) \
(pcmk_is_set((flags), pcmk__action_runnable)? "runnable" : "unrunnable")
#define action_node_str(a) \
(((a)->node == NULL)? "no node" : (a)->node->priv->name)
/*!
* \internal
* \brief Add an XML node tag for a specified ID
*
* \param[in] id Node UUID to add
* \param[in,out] xml Parent XML tag to add to
*/
static xmlNode*
add_node_to_xml_by_id(const char *id, xmlNode *xml)
{
xmlNode *node_xml;
node_xml = pcmk__xe_create(xml, PCMK_XE_NODE);
crm_xml_add(node_xml, PCMK_XA_ID, id);
return node_xml;
}
/*!
* \internal
* \brief Add an XML node tag for a specified node
*
* \param[in] node Node to add
* \param[in,out] xml XML to add node to
*/
static void
add_node_to_xml(const pcmk_node_t *node, void *xml)
{
add_node_to_xml_by_id(node->priv->id, (xmlNode *) xml);
}
/*!
* \internal
* \brief Count (optionally add to XML) nodes needing maintenance state update
*
* \param[in,out] xml Parent XML tag to add to, if any
* \param[in] scheduler Scheduler data
*
* \return Count of nodes added
* \note Only Pacemaker Remote nodes are considered currently
*/
static int
add_maintenance_nodes(xmlNode *xml, const pcmk_scheduler_t *scheduler)
{
xmlNode *maintenance = NULL;
int count = 0;
if (xml != NULL) {
maintenance = pcmk__xe_create(xml, PCMK__XE_MAINTENANCE);
}
for (const GList *iter = scheduler->nodes;
iter != NULL; iter = iter->next) {
const pcmk_node_t *node = iter->data;
if (!pcmk__is_pacemaker_remote_node(node)) {
continue;
}
if ((node->details->maintenance
&& !pcmk_is_set(node->priv->flags, pcmk__node_remote_maint))
|| (!node->details->maintenance
&& pcmk_is_set(node->priv->flags, pcmk__node_remote_maint))) {
if (maintenance != NULL) {
crm_xml_add(add_node_to_xml_by_id(node->priv->id,
maintenance),
PCMK__XA_NODE_IN_MAINTENANCE,
(node->details->maintenance? "1" : "0"));
}
count++;
}
}
crm_trace("%s %d nodes in need of maintenance mode update in state",
((maintenance == NULL)? "Counted" : "Added"), count);
return count;
}
/*!
* \internal
* \brief Add pseudo action with nodes needing maintenance state update
*
* \param[in,out] scheduler Scheduler data
*/
static void
add_maintenance_update(pcmk_scheduler_t *scheduler)
{
pcmk_action_t *action = NULL;
if (add_maintenance_nodes(NULL, scheduler) != 0) {
action = get_pseudo_op(PCMK_ACTION_MAINTENANCE_NODES, scheduler);
pcmk__set_action_flags(action, pcmk__action_always_in_graph);
}
}
/*!
* \internal
* \brief Add XML with nodes that an action is expected to bring down
*
* If a specified action is expected to bring any nodes down, add an XML block
* with their UUIDs. When a node is lost, this allows the controller to
* determine whether it was expected.
*
* \param[in,out] xml Parent XML tag to add to
* \param[in] action Action to check for downed nodes
*/
static void
add_downed_nodes(xmlNode *xml, const pcmk_action_t *action)
{
CRM_CHECK((xml != NULL) && (action != NULL) && (action->node != NULL),
return);
if (pcmk__str_eq(action->task, PCMK_ACTION_DO_SHUTDOWN, pcmk__str_none)) {
/* Shutdown makes the action's node down */
xmlNode *downed = pcmk__xe_create(xml, PCMK__XE_DOWNED);
add_node_to_xml_by_id(action->node->priv->id, downed);
} else if (pcmk__str_eq(action->task, PCMK_ACTION_STONITH,
pcmk__str_none)) {
/* Fencing makes the action's node and any hosted guest nodes down */
const char *fence = g_hash_table_lookup(action->meta,
PCMK__META_STONITH_ACTION);
if (pcmk__is_fencing_action(fence)) {
xmlNode *downed = pcmk__xe_create(xml, PCMK__XE_DOWNED);
add_node_to_xml_by_id(action->node->priv->id, downed);
pe_foreach_guest_node(action->node->priv->scheduler,
action->node, add_node_to_xml, downed);
}
} else if ((action->rsc != NULL)
&& pcmk_is_set(action->rsc->flags,
pcmk__rsc_is_remote_connection)
&& pcmk__str_eq(action->task, PCMK_ACTION_STOP,
pcmk__str_none)) {
/* Stopping a remote connection resource makes connected node down,
* unless it's part of a migration
*/
GList *iter;
pcmk_action_t *input;
bool migrating = false;
for (iter = action->actions_before; iter != NULL; iter = iter->next) {
input = ((pcmk__related_action_t *) iter->data)->action;
if ((input->rsc != NULL)
&& pcmk__str_eq(action->rsc->id, input->rsc->id, pcmk__str_none)
&& pcmk__str_eq(input->task, PCMK_ACTION_MIGRATE_FROM,
pcmk__str_none)) {
migrating = true;
break;
}
}
if (!migrating) {
xmlNode *downed = pcmk__xe_create(xml, PCMK__XE_DOWNED);
add_node_to_xml_by_id(action->rsc->id, downed);
}
}
}
/*!
* \internal
* \brief Create a transition graph operation key for a clone action
*
* \param[in] action Clone action
* \param[in] interval_ms Action interval in milliseconds
*
* \return Newly allocated string with transition graph operation key
*/
static char *
clone_op_key(const pcmk_action_t *action, guint interval_ms)
{
if (pcmk__str_eq(action->task, PCMK_ACTION_NOTIFY, pcmk__str_none)) {
const char *n_type = g_hash_table_lookup(action->meta, "notify_type");
const char *n_task = g_hash_table_lookup(action->meta,
"notify_operation");
return pcmk__notify_key(action->rsc->priv->history_id, n_type,
n_task);
}
return pcmk__op_key(action->rsc->priv->history_id,
pcmk__s(action->cancel_task, action->task),
interval_ms);
}
/*!
* \internal
* \brief Add node details to transition graph action XML
*
* \param[in] action Scheduled action
* \param[in,out] xml Transition graph action XML for \p action
*/
static void
add_node_details(const pcmk_action_t *action, xmlNode *xml)
{
pcmk_node_t *router_node = pcmk__connection_host_for_action(action);
crm_xml_add(xml, PCMK__META_ON_NODE, action->node->priv->name);
crm_xml_add(xml, PCMK__META_ON_NODE_UUID, action->node->priv->id);
if (router_node != NULL) {
crm_xml_add(xml, PCMK__XA_ROUTER_NODE, router_node->priv->name);
}
}
/*!
* \internal
* \brief Add resource details to transition graph action XML
*
* \param[in] action Scheduled action
* \param[in,out] action_xml Transition graph action XML for \p action
*/
static void
add_resource_details(const pcmk_action_t *action, xmlNode *action_xml)
{
xmlNode *rsc_xml = NULL;
const char *attr_list[] = {
PCMK_XA_CLASS,
PCMK_XA_PROVIDER,
PCMK_XA_TYPE,
};
/* If a resource is locked to a node via PCMK_OPT_SHUTDOWN_LOCK, mark its
* actions so the controller can preserve the lock when the action
* completes.
*/
if (pcmk__action_locks_rsc_to_node(action)) {
crm_xml_add_ll(action_xml, PCMK_OPT_SHUTDOWN_LOCK,
(long long) action->rsc->priv->lock_time);
}
// List affected resource
rsc_xml = pcmk__xe_create(action_xml,
(const char *) action->rsc->priv->xml->name);
if (pcmk_is_set(action->rsc->flags, pcmk__rsc_removed)
&& (action->rsc->priv->history_id != NULL)) {
/* Use the numbered instance name here, because if there is more
* than one instance on a node, we need to make sure the command
* goes to the right one.
*
* This is important even for anonymous clones, because the clone's
* unique meta-attribute might have just been toggled from on to
* off.
*/
crm_debug("Using orphan clone name %s instead of history ID %s",
action->rsc->id, action->rsc->priv->history_id);
crm_xml_add(rsc_xml, PCMK_XA_ID, action->rsc->priv->history_id);
crm_xml_add(rsc_xml, PCMK__XA_LONG_ID, action->rsc->id);
} else if (!pcmk_is_set(action->rsc->flags, pcmk__rsc_unique)) {
const char *xml_id = pcmk__xe_id(action->rsc->priv->xml);
crm_debug("Using anonymous clone name %s for %s (aka %s)",
xml_id, action->rsc->id, action->rsc->priv->history_id);
/* ID is what we'd like client to use
* LONG_ID is what they might know it as instead
*
* LONG_ID is only strictly needed /here/ during the
* transition period until all nodes in the cluster
* are running the new software /and/ have rebooted
* once (meaning that they've only ever spoken to a DC
- * supporting this feature).
+ * supporting this feature). (@TODO The effect of removing this on
+ * regression tests suggests that it is still needed for unique clones)
*
* If anyone toggles the unique flag to 'on', the
* 'instance free' name will correspond to an orphan
* and fall into the clause above instead
*/
crm_xml_add(rsc_xml, PCMK_XA_ID, xml_id);
if ((action->rsc->priv->history_id != NULL)
&& !pcmk__str_eq(xml_id, action->rsc->priv->history_id,
pcmk__str_none)) {
crm_xml_add(rsc_xml, PCMK__XA_LONG_ID,
action->rsc->priv->history_id);
} else {
crm_xml_add(rsc_xml, PCMK__XA_LONG_ID, action->rsc->id);
}
} else {
pcmk__assert(action->rsc->priv->history_id == NULL);
crm_xml_add(rsc_xml, PCMK_XA_ID, action->rsc->id);
}
for (int lpc = 0; lpc < PCMK__NELEM(attr_list); lpc++) {
crm_xml_add(rsc_xml, attr_list[lpc],
g_hash_table_lookup(action->rsc->priv->meta,
attr_list[lpc]));
}
}
/*!
* \internal
* \brief Add action attributes to transition graph action XML
*
* \param[in,out] action Scheduled action
* \param[in,out] action_xml Transition graph action XML for \p action
*/
static void
add_action_attributes(pcmk_action_t *action, xmlNode *action_xml)
{
xmlNode *args_xml = NULL;
pcmk_resource_t *rsc = action->rsc;
/* We create free-standing XML to start, so we can sort the attributes
* before adding it to action_xml, which keeps the scheduler regression
* test graphs comparable.
*/
args_xml = pcmk__xe_create(action_xml, PCMK__XE_ATTRIBUTES);
crm_xml_add(args_xml, PCMK_XA_CRM_FEATURE_SET, CRM_FEATURE_SET);
g_hash_table_foreach(action->extra, hash2field, args_xml);
if ((rsc != NULL) && (action->node != NULL)) {
// Get the resource instance attributes, evaluated properly for node
GHashTable *params = pe_rsc_params(rsc, action->node,
rsc->priv->scheduler);
pcmk__substitute_remote_addr(rsc, params);
g_hash_table_foreach(params, hash2smartfield, args_xml);
} else if ((rsc != NULL)
&& (rsc->priv->variant <= pcmk__rsc_variant_primitive)) {
GHashTable *params = pe_rsc_params(rsc, NULL, rsc->priv->scheduler);
g_hash_table_foreach(params, hash2smartfield, args_xml);
}
g_hash_table_foreach(action->meta, hash2metafield, args_xml);
if (rsc != NULL) {
pcmk_resource_t *parent = rsc;
while (parent != NULL) {
parent->priv->cmds->add_graph_meta(parent, args_xml);
parent = parent->priv->parent;
}
pcmk__add_guest_meta_to_xml(args_xml, action);
}
pcmk__xe_sort_attrs(args_xml);
}
/*!
* \internal
* \brief Create the transition graph XML for a scheduled action
*
* \param[in,out] parent Parent XML element to add action to
* \param[in,out] action Scheduled action
* \param[in] skip_details If false, add action details as sub-elements
* \param[in] scheduler Scheduler data
*/
static void
create_graph_action(xmlNode *parent, pcmk_action_t *action, bool skip_details,
const pcmk_scheduler_t *scheduler)
{
bool needs_node_info = true;
bool needs_maintenance_info = false;
xmlNode *action_xml = NULL;
if ((action == NULL) || (scheduler == NULL)) {
return;
}
// Create the top-level element based on task
if (pcmk__str_eq(action->task, PCMK_ACTION_STONITH, pcmk__str_none)) {
/* All fences need node info; guest node fences are pseudo-events */
if (pcmk_is_set(action->flags, pcmk__action_pseudo)) {
action_xml = pcmk__xe_create(parent, PCMK__XE_PSEUDO_EVENT);
} else {
action_xml = pcmk__xe_create(parent, PCMK__XE_CRM_EVENT);
}
} else if (pcmk__str_any_of(action->task,
PCMK_ACTION_DO_SHUTDOWN,
PCMK_ACTION_CLEAR_FAILCOUNT, NULL)) {
action_xml = pcmk__xe_create(parent, PCMK__XE_CRM_EVENT);
} else if (pcmk__str_eq(action->task, PCMK_ACTION_LRM_DELETE,
pcmk__str_none)) {
// CIB-only clean-up for shutdown locks
action_xml = pcmk__xe_create(parent, PCMK__XE_CRM_EVENT);
crm_xml_add(action_xml, PCMK__XA_MODE, PCMK__VALUE_CIB);
} else if (pcmk_is_set(action->flags, pcmk__action_pseudo)) {
if (pcmk__str_eq(action->task, PCMK_ACTION_MAINTENANCE_NODES,
pcmk__str_none)) {
needs_maintenance_info = true;
}
action_xml = pcmk__xe_create(parent, PCMK__XE_PSEUDO_EVENT);
needs_node_info = false;
} else {
action_xml = pcmk__xe_create(parent, PCMK__XE_RSC_OP);
}
crm_xml_add_int(action_xml, PCMK_XA_ID, action->id);
crm_xml_add(action_xml, PCMK_XA_OPERATION, action->task);
if ((action->rsc != NULL) && (action->rsc->priv->history_id != NULL)) {
char *clone_key = NULL;
guint interval_ms;
if (pcmk__guint_from_hash(action->meta, PCMK_META_INTERVAL, 0,
&interval_ms) != pcmk_rc_ok) {
interval_ms = 0;
}
clone_key = clone_op_key(action, interval_ms);
crm_xml_add(action_xml, PCMK__XA_OPERATION_KEY, clone_key);
crm_xml_add(action_xml, "internal_" PCMK__XA_OPERATION_KEY,
action->uuid);
free(clone_key);
} else {
crm_xml_add(action_xml, PCMK__XA_OPERATION_KEY, action->uuid);
}
if (needs_node_info && (action->node != NULL)) {
add_node_details(action, action_xml);
pcmk__insert_dup(action->meta, PCMK__META_ON_NODE,
action->node->priv->name);
pcmk__insert_dup(action->meta, PCMK__META_ON_NODE_UUID,
action->node->priv->id);
}
if (skip_details) {
return;
}
if ((action->rsc != NULL)
&& !pcmk_is_set(action->flags, pcmk__action_pseudo)) {
// This is a real resource action, so add resource details
add_resource_details(action, action_xml);
}
/* List any attributes in effect */
add_action_attributes(action, action_xml);
/* List any nodes this action is expected to make down */
if (needs_node_info && (action->node != NULL)) {
add_downed_nodes(action_xml, action);
}
if (needs_maintenance_info) {
add_maintenance_nodes(action_xml, scheduler);
}
}
/*!
* \internal
* \brief Check whether an action should be added to the transition graph
*
* \param[in,out] action Action to check
*
* \return true if action should be added to graph, otherwise false
*/
static bool
should_add_action_to_graph(pcmk_action_t *action)
{
if (!pcmk_is_set(action->flags, pcmk__action_runnable)) {
crm_trace("Ignoring action %s (%d): unrunnable",
action->uuid, action->id);
return false;
}
if (pcmk_is_set(action->flags, pcmk__action_optional)
&& !pcmk_is_set(action->flags, pcmk__action_always_in_graph)) {
crm_trace("Ignoring action %s (%d): optional",
action->uuid, action->id);
return false;
}
/* Actions for unmanaged resources should be excluded from the graph,
* with the exception of monitors and cancellation of recurring monitors.
*/
if ((action->rsc != NULL)
&& !pcmk_is_set(action->rsc->flags, pcmk__rsc_managed)
&& !pcmk__str_eq(action->task, PCMK_ACTION_MONITOR, pcmk__str_none)) {
const char *interval_ms_s;
/* A cancellation of a recurring monitor will get here because the task
* is cancel rather than monitor, but the interval can still be used to
* recognize it. The interval has been normalized to milliseconds by
* this point, so a string comparison is sufficient.
*/
interval_ms_s = g_hash_table_lookup(action->meta, PCMK_META_INTERVAL);
if (pcmk__str_eq(interval_ms_s, "0", pcmk__str_null_matches)) {
crm_trace("Ignoring action %s (%d): for unmanaged resource (%s)",
action->uuid, action->id, action->rsc->id);
return false;
}
}
/* Always add pseudo-actions, fence actions, and shutdown actions (already
* determined to be required and runnable by this point)
*/
if (pcmk_is_set(action->flags, pcmk__action_pseudo)
|| pcmk__strcase_any_of(action->task, PCMK_ACTION_STONITH,
PCMK_ACTION_DO_SHUTDOWN, NULL)) {
return true;
}
if (action->node == NULL) {
pcmk__sched_err(action->scheduler,
"Skipping action %s (%d) "
"because it was not assigned to a node (bug?)",
action->uuid, action->id);
pcmk__log_action("Unassigned", action, false);
return false;
}
if (pcmk_is_set(action->flags, pcmk__action_on_dc)) {
crm_trace("Action %s (%d) should be dumped: "
"can run on DC instead of %s",
action->uuid, action->id, pcmk__node_name(action->node));
} else if (pcmk__is_guest_or_bundle_node(action->node)
&& !pcmk_is_set(action->node->priv->flags,
pcmk__node_remote_reset)) {
crm_trace("Action %s (%d) should be dumped: "
"assuming will be runnable on guest %s",
action->uuid, action->id, pcmk__node_name(action->node));
} else if (!action->node->details->online) {
pcmk__sched_err(action->scheduler,
"Skipping action %s (%d) "
"because it was scheduled for offline node (bug?)",
action->uuid, action->id);
pcmk__log_action("Offline node", action, false);
return false;
} else if (action->node->details->unclean) {
pcmk__sched_err(action->scheduler,
"Skipping action %s (%d) "
"because it was scheduled for unclean node (bug?)",
action->uuid, action->id);
pcmk__log_action("Unclean node", action, false);
return false;
}
return true;
}
/*!
* \internal
* \brief Check whether an ordering's flags can change an action
*
* \param[in] ordering Ordering to check
*
* \return true if ordering has flags that can change an action, false otherwise
*/
static bool
ordering_can_change_actions(const pcmk__related_action_t *ordering)
{
return pcmk_any_flags_set(ordering->flags,
~(pcmk__ar_then_implies_first_graphed
|pcmk__ar_first_implies_then_graphed
|pcmk__ar_ordered));
}
/*!
* \internal
* \brief Check whether an action input should be in the transition graph
*
* \param[in] action Action to check
* \param[in,out] input Action input to check
*
* \return true if input should be in graph, false otherwise
* \note This function may not only check an input, but disable it under certian
* circumstances (load or anti-colocation orderings that are not needed).
*/
static bool
should_add_input_to_graph(const pcmk_action_t *action,
pcmk__related_action_t *input)
{
if (input->graphed) {
return true;
}
if (input->flags == pcmk__ar_none) {
crm_trace("Ignoring %s (%d) input %s (%d): "
"ordering disabled",
action->uuid, action->id,
input->action->uuid, input->action->id);
return false;
} else if (!pcmk_is_set(input->action->flags, pcmk__action_runnable)
&& !ordering_can_change_actions(input)) {
crm_trace("Ignoring %s (%d) input %s (%d): "
"optional and input unrunnable",
action->uuid, action->id,
input->action->uuid, input->action->id);
return false;
} else if (!pcmk_is_set(input->action->flags, pcmk__action_runnable)
&& pcmk_is_set(input->flags, pcmk__ar_min_runnable)) {
crm_trace("Ignoring %s (%d) input %s (%d): "
"minimum number of instances required but input unrunnable",
action->uuid, action->id,
input->action->uuid, input->action->id);
return false;
} else if (pcmk_is_set(input->flags, pcmk__ar_unmigratable_then_blocks)
&& !pcmk_is_set(input->action->flags, pcmk__action_runnable)) {
crm_trace("Ignoring %s (%d) input %s (%d): "
"input blocked if 'then' unmigratable",
action->uuid, action->id,
input->action->uuid, input->action->id);
return false;
} else if (pcmk_is_set(input->flags, pcmk__ar_if_first_unmigratable)
&& pcmk_is_set(input->action->flags, pcmk__action_migratable)) {
crm_trace("Ignoring %s (%d) input %s (%d): ordering applies "
"only if input is unmigratable, but it is migratable",
action->uuid, action->id,
input->action->uuid, input->action->id);
return false;
} else if ((input->flags == pcmk__ar_ordered)
&& pcmk_is_set(input->action->flags, pcmk__action_migratable)
&& pcmk__ends_with(input->action->uuid, "_stop_0")) {
crm_trace("Ignoring %s (%d) input %s (%d): "
"optional but stop in migration",
action->uuid, action->id,
input->action->uuid, input->action->id);
return false;
} else if (input->flags == pcmk__ar_if_on_same_node_or_target) {
pcmk_node_t *input_node = input->action->node;
if ((action->rsc != NULL)
&& pcmk__str_eq(action->task, PCMK_ACTION_MIGRATE_TO,
pcmk__str_none)) {
pcmk_node_t *assigned = action->rsc->priv->assigned_node;
/* For load_stopped -> migrate_to orderings, we care about where
* the resource has been assigned, not where migrate_to will be
* executed.
*/
if (!pcmk__same_node(input_node, assigned)) {
crm_trace("Ignoring %s (%d) input %s (%d): "
"migration target %s is not same as input node %s",
action->uuid, action->id,
input->action->uuid, input->action->id,
(assigned? assigned->priv->name : "<none>"),
(input_node? input_node->priv->name : "<none>"));
input->flags = pcmk__ar_none;
return false;
}
} else if (!pcmk__same_node(input_node, action->node)) {
crm_trace("Ignoring %s (%d) input %s (%d): "
"not on same node (%s vs %s)",
action->uuid, action->id,
input->action->uuid, input->action->id,
(action->node? action->node->priv->name : "<none>"),
(input_node? input_node->priv->name : "<none>"));
input->flags = pcmk__ar_none;
return false;
} else if (pcmk_is_set(input->action->flags, pcmk__action_optional)) {
crm_trace("Ignoring %s (%d) input %s (%d): "
"ordering optional",
action->uuid, action->id,
input->action->uuid, input->action->id);
input->flags = pcmk__ar_none;
return false;
}
} else if (input->flags == pcmk__ar_if_required_on_same_node) {
if (input->action->node && action->node
&& !pcmk__same_node(input->action->node, action->node)) {
crm_trace("Ignoring %s (%d) input %s (%d): "
"not on same node (%s vs %s)",
action->uuid, action->id,
input->action->uuid, input->action->id,
pcmk__node_name(action->node),
pcmk__node_name(input->action->node));
input->flags = pcmk__ar_none;
return false;
} else if (pcmk_is_set(input->action->flags, pcmk__action_optional)) {
crm_trace("Ignoring %s (%d) input %s (%d): optional",
action->uuid, action->id,
input->action->uuid, input->action->id);
input->flags = pcmk__ar_none;
return false;
}
} else if (input->action->rsc
&& input->action->rsc != action->rsc
&& pcmk_is_set(input->action->rsc->flags, pcmk__rsc_failed)
&& !pcmk_is_set(input->action->rsc->flags, pcmk__rsc_managed)
&& pcmk__ends_with(input->action->uuid, "_stop_0")
&& pcmk__is_clone(action->rsc)) {
crm_warn("Ignoring requirement that %s complete before %s:"
" unmanaged failed resources cannot prevent clone shutdown",
input->action->uuid, action->uuid);
return false;
} else if (pcmk_is_set(input->action->flags, pcmk__action_optional)
&& !pcmk_any_flags_set(input->action->flags,
pcmk__action_always_in_graph
|pcmk__action_added_to_graph)
&& !should_add_action_to_graph(input->action)) {
crm_trace("Ignoring %s (%d) input %s (%d): "
"input optional",
action->uuid, action->id,
input->action->uuid, input->action->id);
return false;
}
crm_trace("%s (%d) input %s %s (%d) on %s should be dumped: %s %s %#.6x",
action->uuid, action->id, action_type_str(input->action->flags),
input->action->uuid, input->action->id,
action_node_str(input->action),
action_runnable_str(input->action->flags),
action_optional_str(input->action->flags), input->flags);
return true;
}
/*!
* \internal
* \brief Check whether an ordering creates an ordering loop
*
* \param[in] init_action "First" action in ordering
* \param[in] action Callers should always set this the same as
* \p init_action (this function may use a different
* value for recursive calls)
* \param[in,out] input Action wrapper for "then" action in ordering
*
* \return true if the ordering creates a loop, otherwise false
*/
bool
pcmk__graph_has_loop(const pcmk_action_t *init_action,
const pcmk_action_t *action, pcmk__related_action_t *input)
{
bool has_loop = false;
if (pcmk_is_set(input->action->flags, pcmk__action_detect_loop)) {
crm_trace("Breaking tracking loop: %s@%s -> %s@%s (%#.6x)",
input->action->uuid,
input->action->node? input->action->node->priv->name : "",
action->uuid,
action->node? action->node->priv->name : "",
input->flags);
return false;
}
// Don't need to check inputs that won't be used
if (!should_add_input_to_graph(action, input)) {
return false;
}
if (input->action == init_action) {
crm_debug("Input loop found in %s@%s ->...-> %s@%s",
action->uuid,
action->node? action->node->priv->name : "",
init_action->uuid,
init_action->node? init_action->node->priv->name : "");
return true;
}
pcmk__set_action_flags(input->action, pcmk__action_detect_loop);
crm_trace("Checking inputs of action %s@%s input %s@%s (%#.6x)"
"for graph loop with %s@%s ",
action->uuid,
action->node? action->node->priv->name : "",
input->action->uuid,
input->action->node? input->action->node->priv->name : "",
input->flags,
init_action->uuid,
init_action->node? init_action->node->priv->name : "");
// Recursively check input itself for loops
for (GList *iter = input->action->actions_before;
iter != NULL; iter = iter->next) {
if (pcmk__graph_has_loop(init_action, input->action,
(pcmk__related_action_t *) iter->data)) {
// Recursive call already logged a debug message
has_loop = true;
break;
}
}
pcmk__clear_action_flags(input->action, pcmk__action_detect_loop);
if (!has_loop) {
crm_trace("No input loop found in %s@%s -> %s@%s (%#.6x)",
input->action->uuid,
input->action->node? input->action->node->priv->name : "",
action->uuid,
action->node? action->node->priv->name : "",
input->flags);
}
return has_loop;
}
/*!
* \internal
* \brief Create a synapse XML element for a transition graph
*
* \param[in] action Action that synapse is for
* \param[in,out] scheduler Scheduler data containing graph
*
* \return Newly added XML element for new graph synapse
*/
static xmlNode *
create_graph_synapse(const pcmk_action_t *action, pcmk_scheduler_t *scheduler)
{
int synapse_priority = 0;
xmlNode *syn = pcmk__xe_create(scheduler->priv->graph, PCMK__XE_SYNAPSE);
crm_xml_add_int(syn, PCMK_XA_ID, scheduler->priv->synapse_count++);
if (action->rsc != NULL) {
synapse_priority = action->rsc->priv->priority;
}
if (action->priority > synapse_priority) {
synapse_priority = action->priority;
}
if (synapse_priority > 0) {
crm_xml_add_int(syn, PCMK__XA_PRIORITY, synapse_priority);
}
return syn;
}
/*!
* \internal
* \brief Add an action to the transition graph XML if appropriate
*
* \param[in,out] data Action to possibly add
* \param[in,out] user_data Scheduler data
*
* \note This will de-duplicate the action inputs, meaning that the
* pcmk__related_action_t:type flags can no longer be relied on to retain
* their original settings. That means this MUST be called after
* pcmk__apply_orderings() is complete, and nothing after this should rely
* on those type flags. (For example, some code looks for type equal to
* some flag rather than whether the flag is set, and some code looks for
* particular combinations of flags -- such code must be done before
* pcmk__create_graph().)
*/
static void
add_action_to_graph(gpointer data, gpointer user_data)
{
pcmk_action_t *action = (pcmk_action_t *) data;
pcmk_scheduler_t *scheduler = (pcmk_scheduler_t *) user_data;
xmlNode *syn = NULL;
xmlNode *set = NULL;
xmlNode *in = NULL;
/* If we haven't already, de-duplicate inputs (even if we won't be adding
* the action to the graph, so that crm_simulate's dot graphs don't have
* duplicates).
*/
if (!pcmk_is_set(action->flags, pcmk__action_inputs_deduplicated)) {
pcmk__deduplicate_action_inputs(action);
pcmk__set_action_flags(action, pcmk__action_inputs_deduplicated);
}
if (pcmk_is_set(action->flags, pcmk__action_added_to_graph)
|| !should_add_action_to_graph(action)) {
return; // Already added, or shouldn't be
}
pcmk__set_action_flags(action, pcmk__action_added_to_graph);
crm_trace("Adding action %d (%s%s%s) to graph",
action->id, action->uuid,
((action->node == NULL)? "" : " on "),
((action->node == NULL)? "" : action->node->priv->name));
syn = create_graph_synapse(action, scheduler);
set = pcmk__xe_create(syn, PCMK__XE_ACTION_SET);
in = pcmk__xe_create(syn, PCMK__XE_INPUTS);
create_graph_action(set, action, false, scheduler);
for (GList *lpc = action->actions_before; lpc != NULL; lpc = lpc->next) {
pcmk__related_action_t *input = lpc->data;
if (should_add_input_to_graph(action, input)) {
xmlNode *input_xml = pcmk__xe_create(in, PCMK__XE_TRIGGER);
input->graphed = true;
create_graph_action(input_xml, input->action, true, scheduler);
}
}
}
static int transition_id = 0;
/*!
* \internal
* \brief Log a message after calculating a transition
*
* \param[in] scheduler Scheduler data
* \param[in] filename Where transition input is stored
*/
void
pcmk__log_transition_summary(const pcmk_scheduler_t *scheduler,
const char *filename)
{
if (pcmk_is_set(scheduler->flags, pcmk__sched_processing_error)
|| pcmk__config_has_error) {
crm_err("Calculated transition %d (with errors)%s%s",
transition_id,
(filename == NULL)? "" : ", saving inputs in ",
(filename == NULL)? "" : filename);
} else if (pcmk_is_set(scheduler->flags, pcmk__sched_processing_warning)
|| pcmk__config_has_warning) {
crm_warn("Calculated transition %d (with warnings)%s%s",
transition_id,
(filename == NULL)? "" : ", saving inputs in ",
(filename == NULL)? "" : filename);
} else {
crm_notice("Calculated transition %d%s%s",
transition_id,
(filename == NULL)? "" : ", saving inputs in ",
(filename == NULL)? "" : filename);
}
if (pcmk__config_has_error) {
crm_notice("Configuration errors found during scheduler processing,"
" please run \"crm_verify -L\" to identify issues");
}
}
/*!
* \internal
* \brief Add a resource's actions to the transition graph
*
* \param[in,out] rsc Resource whose actions should be added
*/
void
pcmk__add_rsc_actions_to_graph(pcmk_resource_t *rsc)
{
GList *iter = NULL;
pcmk__assert(rsc != NULL);
pcmk__rsc_trace(rsc, "Adding actions for %s to graph", rsc->id);
// First add the resource's own actions
g_list_foreach(rsc->priv->actions, add_action_to_graph,
rsc->priv->scheduler);
// Then recursively add its children's actions (appropriate to variant)
for (iter = rsc->priv->children; iter != NULL; iter = iter->next) {
pcmk_resource_t *child_rsc = (pcmk_resource_t *) iter->data;
child_rsc->priv->cmds->add_actions_to_graph(child_rsc);
}
}
/*!
* \internal
* \brief Create a transition graph with all cluster actions needed
*
* \param[in,out] scheduler Scheduler data
*/
void
pcmk__create_graph(pcmk_scheduler_t *scheduler)
{
GList *iter = NULL;
const char *value = NULL;
long long limit = 0LL;
GHashTable *config_hash = scheduler->priv->options;
int rc = pcmk_rc_ok;
transition_id++;
crm_trace("Creating transition graph %d", transition_id);
scheduler->priv->graph = pcmk__xe_create(NULL, PCMK__XE_TRANSITION_GRAPH);
value = pcmk__cluster_option(config_hash, PCMK_OPT_CLUSTER_DELAY);
crm_xml_add(scheduler->priv->graph, PCMK_OPT_CLUSTER_DELAY, value);
value = pcmk__cluster_option(config_hash, PCMK_OPT_STONITH_TIMEOUT);
crm_xml_add(scheduler->priv->graph, PCMK_OPT_STONITH_TIMEOUT, value);
crm_xml_add(scheduler->priv->graph, PCMK__XA_FAILED_STOP_OFFSET,
PCMK_VALUE_INFINITY);
if (pcmk_is_set(scheduler->flags, pcmk__sched_start_failure_fatal)) {
crm_xml_add(scheduler->priv->graph, PCMK__XA_FAILED_START_OFFSET,
PCMK_VALUE_INFINITY);
} else {
crm_xml_add(scheduler->priv->graph, PCMK__XA_FAILED_START_OFFSET, "1");
}
value = pcmk__cluster_option(config_hash, PCMK_OPT_BATCH_LIMIT);
crm_xml_add(scheduler->priv->graph, PCMK_OPT_BATCH_LIMIT, value);
crm_xml_add_int(scheduler->priv->graph, "transition_id", transition_id);
value = pcmk__cluster_option(config_hash, PCMK_OPT_MIGRATION_LIMIT);
rc = pcmk__scan_ll(value, &limit, 0LL);
if (rc != pcmk_rc_ok) {
crm_warn("Ignoring invalid value '%s' for " PCMK_OPT_MIGRATION_LIMIT
": %s", value, pcmk_rc_str(rc));
} else if (limit > 0) {
crm_xml_add(scheduler->priv->graph, PCMK_OPT_MIGRATION_LIMIT, value);
}
if (scheduler->priv->recheck_by > 0) {
char *recheck_epoch = NULL;
recheck_epoch = crm_strdup_printf("%llu", (unsigned long long)
scheduler->priv->recheck_by);
crm_xml_add(scheduler->priv->graph, "recheck-by", recheck_epoch);
free(recheck_epoch);
}
/* The following code will de-duplicate action inputs, so nothing past this
* should rely on the action input type flags retaining their original
* values.
*/
// Add resource actions to graph
for (iter = scheduler->priv->resources; iter != NULL; iter = iter->next) {
pcmk_resource_t *rsc = (pcmk_resource_t *) iter->data;
pcmk__rsc_trace(rsc, "Processing actions for %s", rsc->id);
rsc->priv->cmds->add_actions_to_graph(rsc);
}
// Add pseudo-action for list of nodes with maintenance state update
add_maintenance_update(scheduler);
// Add non-resource (node) actions
for (iter = scheduler->priv->actions; iter != NULL; iter = iter->next) {
pcmk_action_t *action = (pcmk_action_t *) iter->data;
if ((action->rsc != NULL)
&& (action->node != NULL)
&& action->node->details->shutdown
&& !pcmk_is_set(action->rsc->flags, pcmk__rsc_maintenance)
&& !pcmk_any_flags_set(action->flags,
pcmk__action_optional|pcmk__action_runnable)
&& pcmk__str_eq(action->task, PCMK_ACTION_STOP, pcmk__str_none)) {
/* Eventually we should just ignore the 'fence' case, but for now
* it's the best way to detect (in CTS) when CIB resource updates
* are being lost.
*/
if (pcmk_is_set(scheduler->flags, pcmk__sched_quorate)
|| (scheduler->no_quorum_policy == pcmk_no_quorum_ignore)) {
const bool managed = pcmk_is_set(action->rsc->flags,
pcmk__rsc_managed);
const bool failed = pcmk_is_set(action->rsc->flags,
pcmk__rsc_failed);
crm_crit("Cannot %s %s because of %s:%s%s (%s)",
action->node->details->unclean? "fence" : "shut down",
pcmk__node_name(action->node), action->rsc->id,
(managed? " blocked" : " unmanaged"),
(failed? " failed" : ""), action->uuid);
}
}
add_action_to_graph((gpointer) action, (gpointer) scheduler);
}
crm_log_xml_trace(scheduler->priv->graph, "graph");
}
diff --git a/lib/pacemaker/pcmk_injections.c b/lib/pacemaker/pcmk_injections.c
index d728cfea1e..2f51e5f676 100644
--- a/lib/pacemaker/pcmk_injections.c
+++ b/lib/pacemaker/pcmk_injections.c
@@ -1,796 +1,797 @@
/*
* Copyright 2009-2024 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU General Public License version 2
* or later (GPLv2+) WITHOUT ANY WARRANTY.
*/
#include <crm_internal.h>
#include <stdio.h>
#include <unistd.h>
#include <stdlib.h>
#include <sys/stat.h>
#include <sys/param.h>
#include <sys/types.h>
#include <dirent.h>
#include <crm/crm.h>
#include <crm/cib.h>
#include <crm/cib/internal.h>
#include <crm/common/util.h>
#include <crm/common/iso8601.h>
#include <crm/common/xml_internal.h>
#include <crm/lrmd_events.h> // lrmd_event_data_t, etc.
#include <crm/lrmd_internal.h>
#include <crm/pengine/status.h>
#include <pacemaker-internal.h>
#include "libpacemaker_private.h"
+// @TODO Replace this with a new scheduler flag
bool pcmk__simulate_node_config = false;
#define XPATH_NODE_CONFIG "//" PCMK_XE_NODE "[@" PCMK_XA_UNAME "='%s']"
#define XPATH_NODE_STATE "//" PCMK__XE_NODE_STATE "[@" PCMK_XA_UNAME "='%s']"
#define XPATH_NODE_STATE_BY_ID "//" PCMK__XE_NODE_STATE "[@" PCMK_XA_ID "='%s']"
#define XPATH_RSC_HISTORY XPATH_NODE_STATE \
"//" PCMK__XE_LRM_RESOURCE "[@" PCMK_XA_ID "='%s']"
/*!
* \internal
* \brief Inject a fictitious transient node attribute into scheduler input
*
* \param[in,out] out Output object for displaying error messages
* \param[in,out] cib_node \c PCMK__XE_NODE_STATE XML to inject attribute into
* \param[in] name Transient node attribute name to inject
* \param[in] value Transient node attribute value to inject
*/
static void
inject_transient_attr(pcmk__output_t *out, xmlNode *cib_node,
const char *name, const char *value)
{
xmlNode *attrs = NULL;
xmlNode *instance_attrs = NULL;
const char *node_uuid = pcmk__xe_id(cib_node);
out->message(out, "inject-attr", name, value, cib_node);
attrs = pcmk__xe_first_child(cib_node, PCMK__XE_TRANSIENT_ATTRIBUTES, NULL,
NULL);
if (attrs == NULL) {
attrs = pcmk__xe_create(cib_node, PCMK__XE_TRANSIENT_ATTRIBUTES);
crm_xml_add(attrs, PCMK_XA_ID, node_uuid);
}
instance_attrs = pcmk__xe_first_child(attrs, PCMK_XE_INSTANCE_ATTRIBUTES,
NULL, NULL);
if (instance_attrs == NULL) {
instance_attrs = pcmk__xe_create(attrs, PCMK_XE_INSTANCE_ATTRIBUTES);
crm_xml_add(instance_attrs, PCMK_XA_ID, node_uuid);
}
crm_create_nvpair_xml(instance_attrs, NULL, name, value);
}
/*!
* \internal
* \brief Inject a fictitious fail count into a scheduler input
*
* \param[in,out] out Output object for displaying error messages
* \param[in,out] cib_conn CIB connection
* \param[in,out] cib_node Node state XML to inject into
* \param[in] resource ID of resource for fail count to inject
* \param[in] task Action name for fail count to inject
* \param[in] interval_ms Action interval (in milliseconds) for fail count
* \param[in] exit_status Action result for fail count to inject (if
* \c PCMK_OCF_OK, or \c PCMK_OCF_NOT_RUNNING when
* \p interval_ms is 0, inject nothing)
* \param[in] infinity If true, set fail count to "INFINITY", otherwise
* increase it by 1
*/
void
pcmk__inject_failcount(pcmk__output_t *out, cib_t *cib_conn, xmlNode *cib_node,
const char *resource, const char *task,
guint interval_ms, int exit_status, bool infinity)
{
char *name = NULL;
char *value = NULL;
int failcount = 0;
xmlNode *output = NULL;
CRM_CHECK((out != NULL) && (cib_conn != NULL) && (cib_node != NULL)
&& (resource != NULL) && (task != NULL), return);
if ((exit_status == PCMK_OCF_OK)
|| ((exit_status == PCMK_OCF_NOT_RUNNING) && (interval_ms == 0))) {
return;
}
// Get current failcount and increment it
name = pcmk__failcount_name(resource, task, interval_ms);
if (cib__get_node_attrs(out, cib_conn, PCMK_XE_STATUS,
pcmk__xe_id(cib_node), NULL, NULL, NULL, name,
NULL, &output) == pcmk_rc_ok) {
if (crm_element_value_int(output, PCMK_XA_VALUE, &failcount) != 0) {
failcount = 0;
}
}
if (infinity) {
value = pcmk__str_copy(PCMK_VALUE_INFINITY);
} else {
value = pcmk__itoa(failcount + 1);
}
inject_transient_attr(out, cib_node, name, value);
free(name);
free(value);
pcmk__xml_free(output);
name = pcmk__lastfailure_name(resource, task, interval_ms);
value = pcmk__ttoa(time(NULL));
inject_transient_attr(out, cib_node, name, value);
free(name);
free(value);
}
/*!
* \internal
* \brief Create a CIB configuration entry for a fictitious node
*
* \param[in,out] cib_conn CIB object to use
* \param[in] node Node name to use
*/
static void
create_node_entry(cib_t *cib_conn, const char *node)
{
int rc = pcmk_ok;
char *xpath = crm_strdup_printf(XPATH_NODE_CONFIG, node);
rc = cib_conn->cmds->query(cib_conn, xpath, NULL, cib_xpath|cib_sync_call);
if (rc == -ENXIO) { // Only add if not already existing
xmlNode *cib_object = pcmk__xe_create(NULL, PCMK_XE_NODE);
crm_xml_add(cib_object, PCMK_XA_ID, node); // Use node name as ID
crm_xml_add(cib_object, PCMK_XA_UNAME, node);
cib_conn->cmds->create(cib_conn, PCMK_XE_NODES, cib_object,
cib_sync_call);
/* Not bothering with subsequent query to see if it exists,
we'll bomb out later in the call to query_node_uuid()... */
pcmk__xml_free(cib_object);
}
free(xpath);
}
/*!
* \internal
* \brief Synthesize a fake executor event for an action
*
* \param[in] cib_resource XML for any existing resource action history
* \param[in] task Name of action to synthesize
* \param[in] interval_ms Interval of action to synthesize
* \param[in] outcome Result of action to synthesize
*
* \return Newly allocated executor event
* \note It is the caller's responsibility to free the result with
* lrmd_free_event().
*/
static lrmd_event_data_t *
create_op(const xmlNode *cib_resource, const char *task, guint interval_ms,
int outcome)
{
lrmd_event_data_t *op = NULL;
xmlNode *xop = NULL;
op = lrmd_new_event(pcmk__xe_id(cib_resource), task, interval_ms);
lrmd__set_result(op, outcome, PCMK_EXEC_DONE, "Simulated action result");
op->params = NULL; // Not needed for simulation purposes
op->t_run = time(NULL);
op->t_rcchange = op->t_run;
// Use a call ID higher than any existing history entries
op->call_id = 0;
for (xop = pcmk__xe_first_child(cib_resource, NULL, NULL, NULL);
xop != NULL; xop = pcmk__xe_next(xop, NULL)) {
int tmp = 0;
crm_element_value_int(xop, PCMK__XA_CALL_ID, &tmp);
if (tmp > op->call_id) {
op->call_id = tmp;
}
}
op->call_id++;
return op;
}
/*!
* \internal
* \brief Inject a fictitious resource history entry into a scheduler input
*
* \param[in,out] cib_resource Resource history XML to inject entry into
* \param[in,out] op Action result to inject
* \param[in] node Name of node where the action occurred
* \param[in] target_rc Expected result for action to inject
*
* \return XML of injected resource history entry
*/
xmlNode *
pcmk__inject_action_result(xmlNode *cib_resource, lrmd_event_data_t *op,
const char *node, int target_rc)
{
return pcmk__create_history_xml(cib_resource, op, CRM_FEATURE_SET,
target_rc, node, crm_system_name);
}
/*!
* \internal
* \brief Inject a fictitious node into a scheduler input
*
* \param[in,out] cib_conn Scheduler input CIB to inject node into
* \param[in] node Name of node to inject
* \param[in] uuid UUID of node to inject
*
* \return XML of \c PCMK__XE_NODE_STATE entry for new node
* \note If the global pcmk__simulate_node_config has been set to true, a
* node entry in the configuration section will be added, as well as a
* node state entry in the status section.
*/
xmlNode *
pcmk__inject_node(cib_t *cib_conn, const char *node, const char *uuid)
{
int rc = pcmk_ok;
xmlNode *cib_object = NULL;
char *xpath = crm_strdup_printf(XPATH_NODE_STATE, node);
bool duplicate = false;
char *found_uuid = NULL;
if (pcmk__simulate_node_config) {
create_node_entry(cib_conn, node);
}
rc = cib_conn->cmds->query(cib_conn, xpath, &cib_object,
cib_xpath|cib_sync_call);
if ((cib_object != NULL) && (pcmk__xe_id(cib_object) == NULL)) {
crm_err("Detected multiple " PCMK__XE_NODE_STATE " entries for "
"xpath=%s, bailing",
xpath);
duplicate = true;
goto done;
}
if (rc == -ENXIO) {
if (uuid == NULL) {
query_node_uuid(cib_conn, node, &found_uuid, NULL);
} else {
found_uuid = strdup(uuid);
}
if (found_uuid) {
char *xpath_by_uuid = crm_strdup_printf(XPATH_NODE_STATE_BY_ID,
found_uuid);
/* It's possible that a PCMK__XE_NODE_STATE entry doesn't have a
* PCMK_XA_UNAME yet
*/
rc = cib_conn->cmds->query(cib_conn, xpath_by_uuid, &cib_object,
cib_xpath|cib_sync_call);
if ((cib_object != NULL) && (pcmk__xe_id(cib_object) == NULL)) {
crm_err("Can't inject node state for %s because multiple "
"state entries found for ID %s", node, found_uuid);
duplicate = true;
free(xpath_by_uuid);
goto done;
} else if (cib_object != NULL) {
crm_xml_add(cib_object, PCMK_XA_UNAME, node);
rc = cib_conn->cmds->modify(cib_conn, PCMK_XE_STATUS,
cib_object, cib_sync_call);
}
free(xpath_by_uuid);
}
}
if (rc == -ENXIO) {
cib_object = pcmk__xe_create(NULL, PCMK__XE_NODE_STATE);
crm_xml_add(cib_object, PCMK_XA_ID, found_uuid);
crm_xml_add(cib_object, PCMK_XA_UNAME, node);
cib_conn->cmds->create(cib_conn, PCMK_XE_STATUS, cib_object,
cib_sync_call);
pcmk__xml_free(cib_object);
rc = cib_conn->cmds->query(cib_conn, xpath, &cib_object,
cib_xpath|cib_sync_call);
crm_trace("Injecting node state for %s (rc=%d)", node, rc);
}
done:
free(found_uuid);
free(xpath);
if (duplicate) {
crm_log_xml_warn(cib_object, "Duplicates");
crm_exit(CRM_EX_SOFTWARE);
return NULL; // not reached, but makes static analysis happy
}
pcmk__assert(rc == pcmk_ok);
return cib_object;
}
/*!
* \internal
* \brief Inject a fictitious node state change into a scheduler input
*
* \param[in,out] cib_conn Scheduler input CIB to inject into
* \param[in] node Name of node to inject change for
* \param[in] up If true, change state to online, otherwise offline
*
* \return XML of changed (or added) node state entry
*/
xmlNode *
pcmk__inject_node_state_change(cib_t *cib_conn, const char *node, bool up)
{
xmlNode *cib_node = pcmk__inject_node(cib_conn, node, NULL);
if (up) {
pcmk__xe_set_props(cib_node,
PCMK__XA_IN_CCM, PCMK_VALUE_TRUE,
PCMK_XA_CRMD, PCMK_VALUE_ONLINE,
PCMK__XA_JOIN, CRMD_JOINSTATE_MEMBER,
PCMK_XA_EXPECTED, CRMD_JOINSTATE_MEMBER,
NULL);
} else {
pcmk__xe_set_props(cib_node,
PCMK__XA_IN_CCM, PCMK_VALUE_FALSE,
PCMK_XA_CRMD, PCMK_VALUE_OFFLINE,
PCMK__XA_JOIN, CRMD_JOINSTATE_DOWN,
PCMK_XA_EXPECTED, CRMD_JOINSTATE_DOWN,
NULL);
}
crm_xml_add(cib_node, PCMK_XA_CRM_DEBUG_ORIGIN, crm_system_name);
return cib_node;
}
/*!
* \internal
* \brief Check whether a node has history for a given resource
*
* \param[in,out] cib_node Node state XML to check
* \param[in] resource Resource name to check for
*
* \return Resource's \c PCMK__XE_LRM_RESOURCE XML entry beneath \p cib_node if
* found, otherwise \c NULL
*/
static xmlNode *
find_resource_xml(xmlNode *cib_node, const char *resource)
{
const char *node = crm_element_value(cib_node, PCMK_XA_UNAME);
char *xpath = crm_strdup_printf(XPATH_RSC_HISTORY, node, resource);
xmlNode *match = get_xpath_object(xpath, cib_node, LOG_TRACE);
free(xpath);
return match;
}
/*!
* \internal
* \brief Inject a resource history element into a scheduler input
*
* \param[in,out] out Output object for displaying error messages
* \param[in,out] cib_node Node state XML to inject resource history entry into
* \param[in] resource ID (in configuration) of resource to inject
* \param[in] lrm_name ID as used in history (could be clone instance)
* \param[in] rclass Resource agent class of resource to inject
* \param[in] rtype Resource agent type of resource to inject
* \param[in] rprovider Resource agent provider of resource to inject
*
* \return XML of injected resource history element
* \note If a history element already exists under either \p resource or
* \p lrm_name, this will return it rather than injecting a new one.
*/
xmlNode *
pcmk__inject_resource_history(pcmk__output_t *out, xmlNode *cib_node,
const char *resource, const char *lrm_name,
const char *rclass, const char *rtype,
const char *rprovider)
{
xmlNode *lrm = NULL;
xmlNode *container = NULL;
xmlNode *cib_resource = NULL;
cib_resource = find_resource_xml(cib_node, resource);
if (cib_resource != NULL) {
/* If an existing LRM history entry uses the resource name,
* continue using it, even if lrm_name is different.
*/
return cib_resource;
}
// Check for history entry under preferred name
if (strcmp(resource, lrm_name) != 0) {
cib_resource = find_resource_xml(cib_node, lrm_name);
if (cib_resource != NULL) {
return cib_resource;
}
}
if ((rclass == NULL) || (rtype == NULL)) {
// @TODO query configuration for class, provider, type
out->err(out,
"Resource %s not found in the status section of %s "
"(supply class and type to continue)",
resource, pcmk__xe_id(cib_node));
return NULL;
} else if (!pcmk__strcase_any_of(rclass,
PCMK_RESOURCE_CLASS_OCF,
PCMK_RESOURCE_CLASS_STONITH,
PCMK_RESOURCE_CLASS_SERVICE,
PCMK_RESOURCE_CLASS_SYSTEMD,
PCMK_RESOURCE_CLASS_LSB, NULL)) {
out->err(out, "Invalid class for %s: %s", resource, rclass);
return NULL;
} else if (pcmk_is_set(pcmk_get_ra_caps(rclass), pcmk_ra_cap_provider)
&& (rprovider == NULL)) {
// @TODO query configuration for provider
out->err(out, "Please specify the provider for resource %s", resource);
return NULL;
}
crm_info("Injecting new resource %s into node state '%s'",
lrm_name, pcmk__xe_id(cib_node));
lrm = pcmk__xe_first_child(cib_node, PCMK__XE_LRM, NULL, NULL);
if (lrm == NULL) {
const char *node_uuid = pcmk__xe_id(cib_node);
lrm = pcmk__xe_create(cib_node, PCMK__XE_LRM);
crm_xml_add(lrm, PCMK_XA_ID, node_uuid);
}
container = pcmk__xe_first_child(lrm, PCMK__XE_LRM_RESOURCES, NULL, NULL);
if (container == NULL) {
container = pcmk__xe_create(lrm, PCMK__XE_LRM_RESOURCES);
}
cib_resource = pcmk__xe_create(container, PCMK__XE_LRM_RESOURCE);
// If we're creating a new entry, use the preferred name
crm_xml_add(cib_resource, PCMK_XA_ID, lrm_name);
crm_xml_add(cib_resource, PCMK_XA_CLASS, rclass);
crm_xml_add(cib_resource, PCMK_XA_PROVIDER, rprovider);
crm_xml_add(cib_resource, PCMK_XA_TYPE, rtype);
return cib_resource;
}
/*!
* \internal
* \brief Inject a ticket attribute into ticket state
*
* \param[in,out] out Output object for displaying error messages
* \param[in] ticket_id Ticket whose state should be changed
* \param[in] attr_name Ticket attribute name to inject
* \param[in] attr_value Boolean value of ticket attribute to inject
* \param[in,out] cib CIB object to use
*
* \return Standard Pacemaker return code
*/
static int
set_ticket_state_attr(pcmk__output_t *out, const char *ticket_id,
const char *attr_name, bool attr_value, cib_t *cib)
{
int rc = pcmk_rc_ok;
xmlNode *xml_top = NULL;
xmlNode *ticket_state_xml = NULL;
// Check for an existing ticket state entry
rc = pcmk__get_ticket_state(cib, ticket_id, &ticket_state_xml);
if (rc == pcmk_rc_duplicate_id) {
out->err(out, "Multiple " PCMK__XE_TICKET_STATE "s match ticket_id=%s",
ticket_id);
rc = pcmk_rc_ok;
}
if (rc == pcmk_rc_ok) { // Ticket state found, use it
crm_debug("Injecting attribute into existing ticket state %s",
ticket_id);
xml_top = ticket_state_xml;
} else if (rc == ENXIO) { // No ticket state, create it
xmlNode *xml_obj = NULL;
xml_top = pcmk__xe_create(NULL, PCMK_XE_STATUS);
xml_obj = pcmk__xe_create(xml_top, PCMK_XE_TICKETS);
ticket_state_xml = pcmk__xe_create(xml_obj, PCMK__XE_TICKET_STATE);
crm_xml_add(ticket_state_xml, PCMK_XA_ID, ticket_id);
} else { // Error
return rc;
}
// Add the attribute to the ticket state
pcmk__xe_set_bool_attr(ticket_state_xml, attr_name, attr_value);
crm_log_xml_debug(xml_top, "Update");
// Commit the change to the CIB
rc = cib->cmds->modify(cib, PCMK_XE_STATUS, xml_top, cib_sync_call);
rc = pcmk_legacy2rc(rc);
pcmk__xml_free(xml_top);
return rc;
}
/*!
* \internal
* \brief Inject a fictitious action into the cluster
*
* \param[in,out] out Output object for displaying error messages
* \param[in] spec Action specification to inject
* \param[in,out] cib CIB object for scheduler input
* \param[in] scheduler Scheduler data
*/
static void
inject_action(pcmk__output_t *out, const char *spec, cib_t *cib,
const pcmk_scheduler_t *scheduler)
{
int rc;
int outcome = PCMK_OCF_OK;
guint interval_ms = 0;
char *key = NULL;
char *node = NULL;
char *task = NULL;
char *resource = NULL;
const char *rtype = NULL;
const char *rclass = NULL;
const char *rprovider = NULL;
xmlNode *cib_op = NULL;
xmlNode *cib_node = NULL;
xmlNode *cib_resource = NULL;
const pcmk_resource_t *rsc = NULL;
lrmd_event_data_t *op = NULL;
bool infinity = false;
out->message(out, "inject-spec", spec);
key = pcmk__assert_alloc(1, strlen(spec) + 1);
node = pcmk__assert_alloc(1, strlen(spec) + 1);
rc = sscanf(spec, "%[^@]@%[^=]=%d", key, node, &outcome);
if (rc != 3) {
out->err(out, "Invalid operation spec: %s. Only found %d fields",
spec, rc);
goto done;
}
parse_op_key(key, &resource, &task, &interval_ms);
rsc = pe_find_resource(scheduler->priv->resources, resource);
if (rsc == NULL) {
out->err(out, "Invalid resource name: %s", resource);
goto done;
}
rclass = crm_element_value(rsc->priv->xml, PCMK_XA_CLASS);
rtype = crm_element_value(rsc->priv->xml, PCMK_XA_TYPE);
rprovider = crm_element_value(rsc->priv->xml, PCMK_XA_PROVIDER);
cib_node = pcmk__inject_node(cib, node, NULL);
pcmk__assert(cib_node != NULL);
if (pcmk__str_eq(task, PCMK_ACTION_STOP, pcmk__str_none)) {
infinity = true;
} else if (pcmk__str_eq(task, PCMK_ACTION_START, pcmk__str_none)
&& pcmk_is_set(scheduler->flags,
pcmk__sched_start_failure_fatal)) {
infinity = true;
}
pcmk__inject_failcount(out, cib, cib_node, resource, task, interval_ms,
outcome, infinity);
cib_resource = pcmk__inject_resource_history(out, cib_node,
resource, resource,
rclass, rtype, rprovider);
pcmk__assert(cib_resource != NULL);
op = create_op(cib_resource, task, interval_ms, outcome);
pcmk__assert(op != NULL);
cib_op = pcmk__inject_action_result(cib_resource, op, node, 0);
pcmk__assert(cib_op != NULL);
lrmd_free_event(op);
rc = cib->cmds->modify(cib, PCMK_XE_STATUS, cib_node, cib_sync_call);
pcmk__assert(rc == pcmk_ok);
done:
free(task);
free(node);
free(key);
}
/*!
* \internal
* \brief Inject fictitious scheduler inputs
*
* \param[in,out] scheduler Scheduler data
* \param[in,out] cib CIB object for scheduler input to modify
* \param[in] injections Injections to apply
*/
void
pcmk__inject_scheduler_input(pcmk_scheduler_t *scheduler, cib_t *cib,
const pcmk_injections_t *injections)
{
int rc = pcmk_ok;
const GList *iter = NULL;
xmlNode *cib_node = NULL;
pcmk__output_t *out = scheduler->priv->out;
out->message(out, "inject-modify-config", injections->quorum,
injections->watchdog);
if (injections->quorum != NULL) {
xmlNode *top = pcmk__xe_create(NULL, PCMK_XE_CIB);
/* crm_xml_add(top, PCMK_XA_DC_UUID, dc_uuid); */
crm_xml_add(top, PCMK_XA_HAVE_QUORUM, injections->quorum);
rc = cib->cmds->modify(cib, NULL, top, cib_sync_call);
pcmk__assert(rc == pcmk_ok);
}
if (injections->watchdog != NULL) {
rc = cib__update_node_attr(out, cib, cib_sync_call, PCMK_XE_CRM_CONFIG,
NULL, NULL, NULL, NULL,
PCMK_OPT_HAVE_WATCHDOG, injections->watchdog,
NULL, NULL);
pcmk__assert(rc == pcmk_rc_ok);
}
for (iter = injections->node_up; iter != NULL; iter = iter->next) {
const char *node = (const char *) iter->data;
out->message(out, "inject-modify-node", "Online", node);
cib_node = pcmk__inject_node_state_change(cib, node, true);
pcmk__assert(cib_node != NULL);
rc = cib->cmds->modify(cib, PCMK_XE_STATUS, cib_node, cib_sync_call);
pcmk__assert(rc == pcmk_ok);
pcmk__xml_free(cib_node);
}
for (iter = injections->node_down; iter != NULL; iter = iter->next) {
const char *node = (const char *) iter->data;
char *xpath = NULL;
out->message(out, "inject-modify-node", "Offline", node);
cib_node = pcmk__inject_node_state_change(cib, node, false);
pcmk__assert(cib_node != NULL);
rc = cib->cmds->modify(cib, PCMK_XE_STATUS, cib_node, cib_sync_call);
pcmk__assert(rc == pcmk_ok);
pcmk__xml_free(cib_node);
xpath = crm_strdup_printf("//" PCMK__XE_NODE_STATE
"[@" PCMK_XA_UNAME "='%s']"
"/" PCMK__XE_LRM,
node);
cib->cmds->remove(cib, xpath, NULL, cib_xpath|cib_sync_call);
free(xpath);
xpath = crm_strdup_printf("//" PCMK__XE_NODE_STATE
"[@" PCMK_XA_UNAME "='%s']"
"/" PCMK__XE_TRANSIENT_ATTRIBUTES,
node);
cib->cmds->remove(cib, xpath, NULL, cib_xpath|cib_sync_call);
free(xpath);
}
for (iter = injections->node_fail; iter != NULL; iter = iter->next) {
const char *node = (const char *) iter->data;
out->message(out, "inject-modify-node", "Failing", node);
cib_node = pcmk__inject_node_state_change(cib, node, true);
crm_xml_add(cib_node, PCMK__XA_IN_CCM, PCMK_VALUE_FALSE);
pcmk__assert(cib_node != NULL);
rc = cib->cmds->modify(cib, PCMK_XE_STATUS, cib_node, cib_sync_call);
pcmk__assert(rc == pcmk_ok);
pcmk__xml_free(cib_node);
}
for (iter = injections->ticket_grant; iter != NULL; iter = iter->next) {
const char *ticket_id = (const char *) iter->data;
out->message(out, "inject-modify-ticket", "Granting", ticket_id);
rc = set_ticket_state_attr(out, ticket_id, PCMK__XA_GRANTED, true, cib);
pcmk__assert(rc == pcmk_rc_ok);
}
for (iter = injections->ticket_revoke; iter != NULL; iter = iter->next) {
const char *ticket_id = (const char *) iter->data;
out->message(out, "inject-modify-ticket", "Revoking", ticket_id);
rc = set_ticket_state_attr(out, ticket_id, PCMK__XA_GRANTED, false,
cib);
pcmk__assert(rc == pcmk_rc_ok);
}
for (iter = injections->ticket_standby; iter != NULL; iter = iter->next) {
const char *ticket_id = (const char *) iter->data;
out->message(out, "inject-modify-ticket", "Standby", ticket_id);
rc = set_ticket_state_attr(out, ticket_id, PCMK_XA_STANDBY, true, cib);
pcmk__assert(rc == pcmk_rc_ok);
}
for (iter = injections->ticket_activate; iter != NULL; iter = iter->next) {
const char *ticket_id = (const char *) iter->data;
out->message(out, "inject-modify-ticket", "Activating", ticket_id);
rc = set_ticket_state_attr(out, ticket_id, PCMK_XA_STANDBY, false, cib);
pcmk__assert(rc == pcmk_rc_ok);
}
for (iter = injections->op_inject; iter != NULL; iter = iter->next) {
inject_action(out, (const char *) iter->data, cib, scheduler);
}
if (!out->is_quiet(out)) {
out->end_list(out);
}
}
void
pcmk_free_injections(pcmk_injections_t *injections)
{
if (injections == NULL) {
return;
}
g_list_free_full(injections->node_up, g_free);
g_list_free_full(injections->node_down, g_free);
g_list_free_full(injections->node_fail, g_free);
g_list_free_full(injections->op_fail, g_free);
g_list_free_full(injections->op_inject, g_free);
g_list_free_full(injections->ticket_grant, g_free);
g_list_free_full(injections->ticket_revoke, g_free);
g_list_free_full(injections->ticket_standby, g_free);
g_list_free_full(injections->ticket_activate, g_free);
free(injections->quorum);
free(injections->watchdog);
free(injections);
}
diff --git a/lib/pacemaker/pcmk_sched_actions.c b/lib/pacemaker/pcmk_sched_actions.c
index e6ed67c0d9..60f6ec236a 100644
--- a/lib/pacemaker/pcmk_sched_actions.c
+++ b/lib/pacemaker/pcmk_sched_actions.c
@@ -1,1942 +1,1956 @@
/*
* Copyright 2004-2024 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU General Public License version 2
* or later (GPLv2+) WITHOUT ANY WARRANTY.
*/
#include <crm_internal.h>
#include <stdio.h>
#include <sys/param.h>
#include <glib.h>
#include <crm/lrmd_internal.h>
#include <crm/common/scheduler_internal.h>
#include <pacemaker-internal.h>
#include "libpacemaker_private.h"
/*!
* \internal
* \brief Get the action flags relevant to ordering constraints
*
* \param[in,out] action Action to check
* \param[in] node Node that *other* action in the ordering is on
* (used only for clone resource actions)
*
* \return Action flags that should be used for orderings
*/
static uint32_t
action_flags_for_ordering(pcmk_action_t *action, const pcmk_node_t *node)
{
bool runnable = false;
uint32_t flags;
// For non-resource actions, return the action flags
if (action->rsc == NULL) {
return action->flags;
}
/* For non-clone resources, or a clone action not assigned to a node,
* return the flags as determined by the resource method without a node
* specified.
*/
flags = action->rsc->priv->cmds->action_flags(action, NULL);
if ((node == NULL) || !pcmk__is_clone(action->rsc)) {
return flags;
}
/* Otherwise (i.e., for clone resource actions on a specific node), first
* remember whether the non-node-specific action is runnable.
*/
runnable = pcmk_is_set(flags, pcmk__action_runnable);
// Then recheck the resource method with the node
flags = action->rsc->priv->cmds->action_flags(action, node);
/* For clones in ordering constraints, the node-specific "runnable" doesn't
* matter, just the non-node-specific setting (i.e., is the action runnable
* anywhere).
*
* This applies only to runnable, and only for ordering constraints. This
* function shouldn't be used for other types of constraints without
* changes. Not very satisfying, but it's logical and appears to work well.
*/
if (runnable && !pcmk_is_set(flags, pcmk__action_runnable)) {
pcmk__set_raw_action_flags(flags, action->rsc->id,
pcmk__action_runnable);
}
return flags;
}
/*!
* \internal
* \brief Get action UUID that should be used with a resource ordering
*
* When an action is ordered relative to an action for a collective resource
* (clone, group, or bundle), it actually needs to be ordered after all
* instances of the collective have completed the relevant action (for example,
* given "start CLONE then start RSC", RSC must wait until all instances of
* CLONE have started). Given the UUID and resource of the first action in an
* ordering, this returns the UUID of the action that should actually be used
* for ordering (for example, "CLONE_started_0" instead of "CLONE_start_0").
*
* \param[in] first_uuid UUID of first action in ordering
* \param[in] first_rsc Resource of first action in ordering
*
* \return Newly allocated copy of UUID to use with ordering
* \note It is the caller's responsibility to free the return value.
*/
static char *
action_uuid_for_ordering(const char *first_uuid,
const pcmk_resource_t *first_rsc)
{
guint interval_ms = 0;
char *uuid = NULL;
char *rid = NULL;
char *first_task_str = NULL;
enum pcmk__action_type first_task = pcmk__action_unspecified;
enum pcmk__action_type remapped_task = pcmk__action_unspecified;
// Only non-notify actions for collective resources need remapping
if ((strstr(first_uuid, PCMK_ACTION_NOTIFY) != NULL)
|| (first_rsc->priv->variant < pcmk__rsc_variant_group)) {
goto done;
}
// Only non-recurring actions need remapping
pcmk__assert(parse_op_key(first_uuid, &rid, &first_task_str, &interval_ms));
if (interval_ms > 0) {
goto done;
}
first_task = pcmk__parse_action(first_task_str);
switch (first_task) {
case pcmk__action_stop:
case pcmk__action_start:
case pcmk__action_notify:
case pcmk__action_promote:
case pcmk__action_demote:
remapped_task = first_task + 1;
break;
case pcmk__action_stopped:
case pcmk__action_started:
case pcmk__action_notified:
case pcmk__action_promoted:
case pcmk__action_demoted:
remapped_task = first_task;
break;
case pcmk__action_monitor:
case pcmk__action_shutdown:
case pcmk__action_fence:
break;
default:
crm_err("Unknown action '%s' in ordering", first_task_str);
break;
}
if (remapped_task != pcmk__action_unspecified) {
/* If a clone or bundle has notifications enabled, the ordering will be
* relative to when notifications have been sent for the remapped task.
*/
if (pcmk_is_set(first_rsc->flags, pcmk__rsc_notify)
&& (pcmk__is_clone(first_rsc) || pcmk__is_bundled(first_rsc))) {
uuid = pcmk__notify_key(rid, "confirmed-post",
pcmk__action_text(remapped_task));
} else {
uuid = pcmk__op_key(rid, pcmk__action_text(remapped_task), 0);
}
pcmk__rsc_trace(first_rsc,
"Remapped action UUID %s to %s for ordering purposes",
first_uuid, uuid);
}
done:
free(first_task_str);
free(rid);
return (uuid != NULL)? uuid : pcmk__str_copy(first_uuid);
}
/*!
* \internal
* \brief Get actual action that should be used with an ordering
*
* When an action is ordered relative to an action for a collective resource
* (clone, group, or bundle), it actually needs to be ordered after all
* instances of the collective have completed the relevant action (for example,
* given "start CLONE then start RSC", RSC must wait until all instances of
* CLONE have started). Given the first action in an ordering, this returns the
* the action that should actually be used for ordering (for example, the
* started action instead of the start action).
*
* \param[in] action First action in an ordering
*
* \return Actual action that should be used for the ordering
*/
static pcmk_action_t *
action_for_ordering(pcmk_action_t *action)
{
pcmk_action_t *result = action;
pcmk_resource_t *rsc = action->rsc;
if (rsc == NULL) {
return result;
}
if ((rsc->priv->variant >= pcmk__rsc_variant_group)
&& (action->uuid != NULL)) {
char *uuid = action_uuid_for_ordering(action->uuid, rsc);
result = find_first_action(rsc->priv->actions, uuid, NULL, NULL);
if (result == NULL) {
crm_warn("Not remapping %s to %s because %s does not have "
"remapped action", action->uuid, uuid, rsc->id);
result = action;
}
free(uuid);
}
return result;
}
/*!
* \internal
* \brief Wrapper for update_ordered_actions() method for readability
*
* \param[in,out] rsc Resource to call method for
* \param[in,out] first 'First' action in an ordering
* \param[in,out] then 'Then' action in an ordering
* \param[in] node If not NULL, limit scope of ordering to this
* node (only used when interleaving instances)
* \param[in] flags Action flags for \p first for ordering purposes
* \param[in] filter Action flags to limit scope of certain updates
* (may include pcmk__action_optional to affect only
* mandatory actions, and pe_action_runnable to
* affect only runnable actions)
* \param[in] type Group of enum pcmk__action_relation_flags to apply
* \param[in,out] scheduler Scheduler data
*
* \return Group of enum pcmk__updated flags indicating what was updated
*/
static inline uint32_t
update(pcmk_resource_t *rsc, pcmk_action_t *first, pcmk_action_t *then,
const pcmk_node_t *node, uint32_t flags, uint32_t filter, uint32_t type,
pcmk_scheduler_t *scheduler)
{
return rsc->priv->cmds->update_ordered_actions(first, then, node, flags,
filter, type, scheduler);
}
/*!
* \internal
* \brief Update flags for ordering's actions appropriately for ordering's flags
*
* \param[in,out] first First action in an ordering
* \param[in,out] then Then action in an ordering
* \param[in] first_flags Action flags for \p first for ordering purposes
* \param[in] then_flags Action flags for \p then for ordering purposes
* \param[in,out] order Action wrapper for \p first in ordering
* \param[in,out] scheduler Scheduler data
*
* \return Group of enum pcmk__updated flags
*/
static uint32_t
update_action_for_ordering_flags(pcmk_action_t *first, pcmk_action_t *then,
uint32_t first_flags, uint32_t then_flags,
pcmk__related_action_t *order,
pcmk_scheduler_t *scheduler)
{
uint32_t changed = pcmk__updated_none;
/* The node will only be used for clones. If interleaved, node will be NULL,
* otherwise the ordering scope will be limited to the node. Normally, the
* whole 'then' clone should restart if 'first' is restarted, so then->node
* is needed.
*/
pcmk_node_t *node = then->node;
if (pcmk_is_set(order->flags, pcmk__ar_first_implies_same_node_then)) {
/* For unfencing, only instances of 'then' on the same node as 'first'
* (the unfencing operation) should restart, so reset node to
* first->node, at which point this case is handled like a normal
* pcmk__ar_first_implies_then.
*/
pcmk__clear_relation_flags(order->flags,
pcmk__ar_first_implies_same_node_then);
pcmk__set_relation_flags(order->flags, pcmk__ar_first_implies_then);
node = first->node;
pcmk__rsc_trace(then->rsc,
"%s then %s: mapped "
"pcmk__ar_first_implies_same_node_then to "
"pcmk__ar_first_implies_then on %s",
first->uuid, then->uuid, pcmk__node_name(node));
}
if (pcmk_is_set(order->flags, pcmk__ar_first_implies_then)) {
if (then->rsc != NULL) {
changed |= update(then->rsc, first, then, node,
first_flags & pcmk__action_optional,
pcmk__action_optional,
pcmk__ar_first_implies_then, scheduler);
} else if (!pcmk_is_set(first_flags, pcmk__action_optional)
&& pcmk_is_set(then->flags, pcmk__action_optional)) {
pcmk__clear_action_flags(then, pcmk__action_optional);
pcmk__set_updated_flags(changed, first, pcmk__updated_then);
}
pcmk__rsc_trace(then->rsc,
"%s then %s: %s after pcmk__ar_first_implies_then",
first->uuid, then->uuid,
(changed? "changed" : "unchanged"));
}
if (pcmk_is_set(order->flags, pcmk__ar_intermediate_stop)
&& (then->rsc != NULL)) {
enum pcmk__action_flags restart = pcmk__action_optional
|pcmk__action_runnable;
changed |= update(then->rsc, first, then, node, first_flags, restart,
pcmk__ar_intermediate_stop, scheduler);
pcmk__rsc_trace(then->rsc,
"%s then %s: %s after pcmk__ar_intermediate_stop",
first->uuid, then->uuid,
(changed? "changed" : "unchanged"));
}
if (pcmk_is_set(order->flags, pcmk__ar_then_implies_first)) {
if (first->rsc != NULL) {
changed |= update(first->rsc, first, then, node, first_flags,
pcmk__action_optional,
pcmk__ar_then_implies_first, scheduler);
} else if (!pcmk_is_set(first_flags, pcmk__action_optional)
&& pcmk_is_set(first->flags, pcmk__action_runnable)) {
pcmk__clear_action_flags(first, pcmk__action_runnable);
pcmk__set_updated_flags(changed, first, pcmk__updated_first);
}
pcmk__rsc_trace(then->rsc,
"%s then %s: %s after pcmk__ar_then_implies_first",
first->uuid, then->uuid,
(changed? "changed" : "unchanged"));
}
if (pcmk_is_set(order->flags, pcmk__ar_promoted_then_implies_first)) {
if (then->rsc != NULL) {
changed |= update(then->rsc, first, then, node,
first_flags & pcmk__action_optional,
pcmk__action_optional,
pcmk__ar_promoted_then_implies_first, scheduler);
}
pcmk__rsc_trace(then->rsc,
"%s then %s: %s after "
"pcmk__ar_promoted_then_implies_first",
first->uuid, then->uuid,
(changed? "changed" : "unchanged"));
}
if (pcmk_is_set(order->flags, pcmk__ar_min_runnable)) {
if (then->rsc != NULL) {
changed |= update(then->rsc, first, then, node, first_flags,
pcmk__action_runnable, pcmk__ar_min_runnable,
scheduler);
} else if (pcmk_is_set(first_flags, pcmk__action_runnable)) {
// We have another runnable instance of "first"
then->runnable_before++;
/* Mark "then" as runnable if it requires a certain number of
* "before" instances to be runnable, and they now are.
*/
if ((then->runnable_before >= then->required_runnable_before)
&& !pcmk_is_set(then->flags, pcmk__action_runnable)) {
pcmk__set_action_flags(then, pcmk__action_runnable);
pcmk__set_updated_flags(changed, first, pcmk__updated_then);
}
}
pcmk__rsc_trace(then->rsc, "%s then %s: %s after pcmk__ar_min_runnable",
first->uuid, then->uuid,
(changed? "changed" : "unchanged"));
}
if (pcmk_is_set(order->flags, pcmk__ar_nested_remote_probe)
&& (then->rsc != NULL)) {
if (!pcmk_is_set(first_flags, pcmk__action_runnable)
&& (first->rsc != NULL)
&& (first->rsc->priv->active_nodes != NULL)) {
pcmk__rsc_trace(then->rsc,
"%s then %s: ignoring because first is stopping",
first->uuid, then->uuid);
order->flags = pcmk__ar_none;
} else {
changed |= update(then->rsc, first, then, node, first_flags,
pcmk__action_runnable,
pcmk__ar_unrunnable_first_blocks, scheduler);
}
pcmk__rsc_trace(then->rsc,
"%s then %s: %s after pcmk__ar_nested_remote_probe",
first->uuid, then->uuid,
(changed? "changed" : "unchanged"));
}
if (pcmk_is_set(order->flags, pcmk__ar_unrunnable_first_blocks)) {
if (then->rsc != NULL) {
changed |= update(then->rsc, first, then, node, first_flags,
pcmk__action_runnable,
pcmk__ar_unrunnable_first_blocks, scheduler);
} else if (!pcmk_is_set(first_flags, pcmk__action_runnable)
&& pcmk_is_set(then->flags, pcmk__action_runnable)) {
pcmk__clear_action_flags(then, pcmk__action_runnable);
pcmk__set_updated_flags(changed, first, pcmk__updated_then);
}
pcmk__rsc_trace(then->rsc,
"%s then %s: %s after pcmk__ar_unrunnable_first_blocks",
first->uuid, then->uuid,
(changed? "changed" : "unchanged"));
}
if (pcmk_is_set(order->flags, pcmk__ar_unmigratable_then_blocks)) {
if (then->rsc != NULL) {
changed |= update(then->rsc, first, then, node, first_flags,
pcmk__action_optional,
pcmk__ar_unmigratable_then_blocks, scheduler);
}
pcmk__rsc_trace(then->rsc,
"%s then %s: %s after "
"pcmk__ar_unmigratable_then_blocks",
first->uuid, then->uuid,
(changed? "changed" : "unchanged"));
}
if (pcmk_is_set(order->flags, pcmk__ar_first_else_then)) {
if (then->rsc != NULL) {
changed |= update(then->rsc, first, then, node, first_flags,
pcmk__action_optional, pcmk__ar_first_else_then,
scheduler);
}
pcmk__rsc_trace(then->rsc,
"%s then %s: %s after pcmk__ar_first_else_then",
first->uuid, then->uuid,
(changed? "changed" : "unchanged"));
}
if (pcmk_is_set(order->flags, pcmk__ar_ordered)) {
if (then->rsc != NULL) {
changed |= update(then->rsc, first, then, node, first_flags,
pcmk__action_runnable, pcmk__ar_ordered,
scheduler);
}
pcmk__rsc_trace(then->rsc, "%s then %s: %s after pcmk__ar_ordered",
first->uuid, then->uuid,
(changed? "changed" : "unchanged"));
}
if (pcmk_is_set(order->flags, pcmk__ar_asymmetric)) {
if (then->rsc != NULL) {
changed |= update(then->rsc, first, then, node, first_flags,
pcmk__action_runnable, pcmk__ar_asymmetric,
scheduler);
}
pcmk__rsc_trace(then->rsc, "%s then %s: %s after pcmk__ar_asymmetric",
first->uuid, then->uuid,
(changed? "changed" : "unchanged"));
}
if (pcmk_is_set(first->flags, pcmk__action_runnable)
&& pcmk_is_set(order->flags, pcmk__ar_first_implies_then_graphed)
&& !pcmk_is_set(first_flags, pcmk__action_optional)) {
pcmk__rsc_trace(then->rsc, "%s will be in graph because %s is required",
then->uuid, first->uuid);
pcmk__set_action_flags(then, pcmk__action_always_in_graph);
// Don't bother marking 'then' as changed just for this
}
if (pcmk_is_set(order->flags, pcmk__ar_then_implies_first_graphed)
&& !pcmk_is_set(then_flags, pcmk__action_optional)) {
pcmk__rsc_trace(then->rsc, "%s will be in graph because %s is required",
first->uuid, then->uuid);
pcmk__set_action_flags(first, pcmk__action_always_in_graph);
// Don't bother marking 'first' as changed just for this
}
if (pcmk_any_flags_set(order->flags, pcmk__ar_first_implies_then
|pcmk__ar_then_implies_first
|pcmk__ar_intermediate_stop)
&& (first->rsc != NULL)
&& !pcmk_is_set(first->rsc->flags, pcmk__rsc_managed)
&& pcmk_is_set(first->rsc->flags, pcmk__rsc_blocked)
&& !pcmk_is_set(first->flags, pcmk__action_runnable)
&& pcmk__str_eq(first->task, PCMK_ACTION_STOP, pcmk__str_none)) {
+ /* @TODO This seems odd; why wouldn't an unrunnable "first" already
+ * block "then" before this? Note that the unmanaged-stop-{1,2}
+ * scheduler regression tests and the test CIB for T209 have tests for
+ * "stop then stop" relations that would be good for checking any
+ * changes.
+ */
if (pcmk_is_set(then->flags, pcmk__action_runnable)) {
pcmk__clear_action_flags(then, pcmk__action_runnable);
pcmk__set_updated_flags(changed, first, pcmk__updated_then);
}
pcmk__rsc_trace(then->rsc,
"%s then %s: %s after checking whether first "
"is blocked, unmanaged, unrunnable stop",
first->uuid, then->uuid,
(changed? "changed" : "unchanged"));
}
return changed;
}
// Convenience macros for logging action properties
#define action_type_str(flags) \
(pcmk_is_set((flags), pcmk__action_pseudo)? "pseudo-action" : "action")
#define action_optional_str(flags) \
(pcmk_is_set((flags), pcmk__action_optional)? "optional" : "required")
#define action_runnable_str(flags) \
(pcmk_is_set((flags), pcmk__action_runnable)? "runnable" : "unrunnable")
#define action_node_str(a) \
(((a)->node == NULL)? "no node" : (a)->node->priv->name)
/*!
* \internal
* \brief Update an action's flags for all orderings where it is "then"
*
* \param[in,out] then Action to update
* \param[in,out] scheduler Scheduler data
*/
void
pcmk__update_action_for_orderings(pcmk_action_t *then,
pcmk_scheduler_t *scheduler)
{
GList *lpc = NULL;
uint32_t changed = pcmk__updated_none;
int last_flags = then->flags;
pcmk__rsc_trace(then->rsc, "Updating %s %s (%s %s) on %s",
action_type_str(then->flags), then->uuid,
action_optional_str(then->flags),
action_runnable_str(then->flags), action_node_str(then));
if (then->required_runnable_before > 0) {
/* Initialize current known "runnable before" actions. As
* update_action_for_ordering_flags() is called for each of then's
* before actions, this number will increment as runnable 'first'
* actions are encountered.
*/
then->runnable_before = 0;
/* The pcmk__ar_min_runnable clause of
* update_action_for_ordering_flags() (called below)
* will reset runnable if appropriate.
*/
pcmk__clear_action_flags(then, pcmk__action_runnable);
}
for (lpc = then->actions_before; lpc != NULL; lpc = lpc->next) {
pcmk__related_action_t *other = lpc->data;
pcmk_action_t *first = other->action;
pcmk_node_t *then_node = then->node;
pcmk_node_t *first_node = first->node;
const uint32_t target = pcmk__rsc_node_assigned;
if ((first->rsc != NULL)
&& pcmk__is_group(first->rsc)
&& pcmk__str_eq(first->task, PCMK_ACTION_START, pcmk__str_none)) {
first_node = first->rsc->priv->fns->location(first->rsc, NULL,
target);
if (first_node != NULL) {
pcmk__rsc_trace(first->rsc, "Found %s for 'first' %s",
pcmk__node_name(first_node), first->uuid);
}
}
if (pcmk__is_group(then->rsc)
&& pcmk__str_eq(then->task, PCMK_ACTION_START, pcmk__str_none)) {
then_node = then->rsc->priv->fns->location(then->rsc, NULL, target);
if (then_node != NULL) {
pcmk__rsc_trace(then->rsc, "Found %s for 'then' %s",
pcmk__node_name(then_node), then->uuid);
}
}
// Disable constraint if it only applies when on same node, but isn't
if (pcmk_is_set(other->flags, pcmk__ar_if_on_same_node)
&& (first_node != NULL) && (then_node != NULL)
&& !pcmk__same_node(first_node, then_node)) {
pcmk__rsc_trace(then->rsc,
"Disabled ordering %s on %s then %s on %s: "
"not same node",
other->action->uuid, pcmk__node_name(first_node),
then->uuid, pcmk__node_name(then_node));
other->flags = pcmk__ar_none;
continue;
}
pcmk__clear_updated_flags(changed, then, pcmk__updated_first);
if ((first->rsc != NULL)
&& pcmk_is_set(other->flags, pcmk__ar_then_cancels_first)
&& !pcmk_is_set(then->flags, pcmk__action_optional)) {
/* 'then' is required, so we must abandon 'first'
* (e.g. a required stop cancels any agent reload).
*/
pcmk__set_action_flags(other->action, pcmk__action_optional);
if (!strcmp(first->task, PCMK_ACTION_RELOAD_AGENT)) {
pcmk__clear_rsc_flags(first->rsc, pcmk__rsc_reload);
}
}
if ((first->rsc != NULL) && (then->rsc != NULL)
&& (first->rsc != then->rsc) && !is_parent(then->rsc, first->rsc)) {
first = action_for_ordering(first);
}
if (first != other->action) {
pcmk__rsc_trace(then->rsc, "Ordering %s after %s instead of %s",
then->uuid, first->uuid, other->action->uuid);
}
pcmk__rsc_trace(then->rsc,
"%s (%#.6x) then %s (%#.6x): type=%#.6x node=%s",
first->uuid, first->flags, then->uuid, then->flags,
other->flags, action_node_str(first));
if (first == other->action) {
/* 'first' was not remapped (e.g. from 'start' to 'running'), which
* could mean it is a non-resource action, a primitive resource
* action, or already expanded.
*/
uint32_t first_flags, then_flags;
first_flags = action_flags_for_ordering(first, then_node);
then_flags = action_flags_for_ordering(then, first_node);
changed |= update_action_for_ordering_flags(first, then,
first_flags, then_flags,
other, scheduler);
/* 'first' was for a complex resource (clone, group, etc),
* create a new dependency if necessary
*/
} else if (order_actions(first, then, other->flags)) {
/* This was the first time 'first' and 'then' were associated,
* start again to get the new actions_before list
*/
pcmk__set_updated_flags(changed, then, pcmk__updated_then);
pcmk__rsc_trace(then->rsc,
"Disabled ordering %s then %s in favor of %s "
"then %s",
other->action->uuid, then->uuid, first->uuid,
then->uuid);
other->flags = pcmk__ar_none;
}
if (pcmk_is_set(changed, pcmk__updated_first)) {
crm_trace("Re-processing %s and its 'after' actions "
"because it changed", first->uuid);
for (GList *lpc2 = first->actions_after; lpc2 != NULL;
lpc2 = lpc2->next) {
pcmk__related_action_t *other = lpc2->data;
pcmk__update_action_for_orderings(other->action, scheduler);
}
pcmk__update_action_for_orderings(first, scheduler);
}
}
if (then->required_runnable_before > 0) {
if (last_flags == then->flags) {
pcmk__clear_updated_flags(changed, then, pcmk__updated_then);
} else {
pcmk__set_updated_flags(changed, then, pcmk__updated_then);
}
}
if (pcmk_is_set(changed, pcmk__updated_then)) {
crm_trace("Re-processing %s and its 'after' actions because it changed",
then->uuid);
if (pcmk_is_set(last_flags, pcmk__action_runnable)
&& !pcmk_is_set(then->flags, pcmk__action_runnable)) {
pcmk__block_colocation_dependents(then);
}
pcmk__update_action_for_orderings(then, scheduler);
for (lpc = then->actions_after; lpc != NULL; lpc = lpc->next) {
pcmk__related_action_t *other = lpc->data;
pcmk__update_action_for_orderings(other->action, scheduler);
}
}
}
static inline bool
is_primitive_action(const pcmk_action_t *action)
{
return (action != NULL) && pcmk__is_primitive(action->rsc);
}
/*!
* \internal
* \brief Clear a single action flag and set reason text
*
* \param[in,out] action Action whose flag should be cleared
* \param[in] flag Action flag that should be cleared
* \param[in] reason Action that is the reason why flag is being cleared
*/
#define clear_action_flag_because(action, flag, reason) do { \
if (pcmk_is_set((action)->flags, (flag))) { \
pcmk__clear_action_flags(action, flag); \
if ((action)->rsc != (reason)->rsc) { \
char *reason_text = pe__action2reason((reason), (flag)); \
pe_action_set_reason((action), reason_text, false); \
free(reason_text); \
} \
} \
} while (0)
/*!
* \internal
* \brief Update actions in an asymmetric ordering
*
* If the "first" action in an asymmetric ordering is unrunnable, make the
* "second" action unrunnable as well, if appropriate.
*
* \param[in] first 'First' action in an asymmetric ordering
* \param[in,out] then 'Then' action in an asymmetric ordering
*/
static void
handle_asymmetric_ordering(const pcmk_action_t *first, pcmk_action_t *then)
{
/* Only resource actions after an unrunnable 'first' action need updates for
* asymmetric ordering.
*/
if ((then->rsc == NULL)
|| pcmk_is_set(first->flags, pcmk__action_runnable)) {
return;
}
// Certain optional 'then' actions are unaffected by unrunnable 'first'
if (pcmk_is_set(then->flags, pcmk__action_optional)) {
enum rsc_role_e then_rsc_role;
then_rsc_role = then->rsc->priv->fns->state(then->rsc, TRUE);
if ((then_rsc_role == pcmk_role_stopped)
&& pcmk__str_eq(then->task, PCMK_ACTION_STOP, pcmk__str_none)) {
/* If 'then' should stop after 'first' but is already stopped, the
* ordering is irrelevant.
*/
return;
} else if ((then_rsc_role >= pcmk_role_started)
&& pcmk__str_eq(then->task, PCMK_ACTION_START, pcmk__str_none)
&& pe__rsc_running_on_only(then->rsc, then->node)) {
/* Similarly if 'then' should start after 'first' but is already
* started on a single node.
*/
return;
}
}
// 'First' can't run, so 'then' can't either
clear_action_flag_because(then, pcmk__action_optional, first);
clear_action_flag_because(then, pcmk__action_runnable, first);
}
/*!
* \internal
* \brief Set action bits appropriately when pcmk__ar_intermediate_stop is used
*
* \param[in,out] first 'First' action in ordering
* \param[in,out] then 'Then' action in ordering
* \param[in] filter What action flags to care about
*
* \note pcmk__ar_intermediate_stop is set for "stop resource before starting
* it" and "stop later group member before stopping earlier group member"
*/
static void
handle_restart_ordering(pcmk_action_t *first, pcmk_action_t *then,
uint32_t filter)
{
const char *reason = NULL;
pcmk__assert(is_primitive_action(first) && is_primitive_action(then));
// We need to update the action in two cases:
// ... if 'then' is required
if (pcmk_is_set(filter, pcmk__action_optional)
&& !pcmk_is_set(then->flags, pcmk__action_optional)) {
reason = "restart";
}
/* ... if 'then' is unrunnable action on same resource (if a resource
* should restart but can't start, we still want to stop)
*/
if (pcmk_is_set(filter, pcmk__action_runnable)
&& !pcmk_is_set(then->flags, pcmk__action_runnable)
&& pcmk_is_set(then->rsc->flags, pcmk__rsc_managed)
&& (first->rsc == then->rsc)) {
reason = "stop";
}
if (reason == NULL) {
return;
}
pcmk__rsc_trace(first->rsc, "Handling %s -> %s for %s",
first->uuid, then->uuid, reason);
// Make 'first' required if it is runnable
if (pcmk_is_set(first->flags, pcmk__action_runnable)) {
clear_action_flag_because(first, pcmk__action_optional, then);
}
// Make 'first' required if 'then' is required
if (!pcmk_is_set(then->flags, pcmk__action_optional)) {
clear_action_flag_because(first, pcmk__action_optional, then);
}
// Make 'first' unmigratable if 'then' is unmigratable
if (!pcmk_is_set(then->flags, pcmk__action_migratable)) {
clear_action_flag_because(first, pcmk__action_migratable, then);
}
// Make 'then' unrunnable if 'first' is required but unrunnable
if (!pcmk_is_set(first->flags, pcmk__action_optional)
&& !pcmk_is_set(first->flags, pcmk__action_runnable)) {
clear_action_flag_because(then, pcmk__action_runnable, first);
}
}
/*!
* \internal
* \brief Update two actions according to an ordering between them
*
* Given information about an ordering of two actions, update the actions' flags
* (and runnable_before members if appropriate) as appropriate for the ordering.
* Effects may cascade to other orderings involving the actions as well.
*
* \param[in,out] first 'First' action in an ordering
* \param[in,out] then 'Then' action in an ordering
* \param[in] node If not NULL, limit scope of ordering to this node
* (ignored)
* \param[in] flags Action flags for \p first for ordering purposes
* \param[in] filter Action flags to limit scope of certain updates (may
* include pcmk__action_optional to affect only
* mandatory actions, and pcmk__action_runnable to
* affect only runnable actions)
* \param[in] type Group of enum pcmk__action_relation_flags to apply
* \param[in,out] scheduler Scheduler data
*
* \return Group of enum pcmk__updated flags indicating what was updated
*/
uint32_t
pcmk__update_ordered_actions(pcmk_action_t *first, pcmk_action_t *then,
const pcmk_node_t *node, uint32_t flags,
uint32_t filter, uint32_t type,
pcmk_scheduler_t *scheduler)
{
uint32_t changed = pcmk__updated_none;
uint32_t then_flags = 0U;
uint32_t first_flags = 0U;
pcmk__assert((first != NULL) && (then != NULL) && (scheduler != NULL));
then_flags = then->flags;
first_flags = first->flags;
if (pcmk_is_set(type, pcmk__ar_asymmetric)) {
handle_asymmetric_ordering(first, then);
}
if (pcmk_is_set(type, pcmk__ar_then_implies_first)
&& !pcmk_is_set(then_flags, pcmk__action_optional)) {
// Then is required, and implies first should be, too
if (pcmk_is_set(filter, pcmk__action_optional)
&& !pcmk_is_set(flags, pcmk__action_optional)
&& pcmk_is_set(first_flags, pcmk__action_optional)) {
clear_action_flag_because(first, pcmk__action_optional, then);
}
if (pcmk_is_set(flags, pcmk__action_migratable)
&& !pcmk_is_set(then->flags, pcmk__action_migratable)) {
clear_action_flag_because(first, pcmk__action_migratable, then);
}
}
if (pcmk_is_set(type, pcmk__ar_promoted_then_implies_first)
&& (then->rsc != NULL)
&& (then->rsc->priv->orig_role == pcmk_role_promoted)
&& pcmk_is_set(filter, pcmk__action_optional)
&& !pcmk_is_set(then->flags, pcmk__action_optional)) {
clear_action_flag_because(first, pcmk__action_optional, then);
if (pcmk_is_set(first->flags, pcmk__action_migratable)
&& !pcmk_is_set(then->flags, pcmk__action_migratable)) {
clear_action_flag_because(first, pcmk__action_migratable, then);
}
}
if (pcmk_is_set(type, pcmk__ar_unmigratable_then_blocks)
&& pcmk_is_set(filter, pcmk__action_optional)) {
if (!pcmk_all_flags_set(then->flags, pcmk__action_migratable
|pcmk__action_runnable)) {
clear_action_flag_because(first, pcmk__action_runnable, then);
}
if (!pcmk_is_set(then->flags, pcmk__action_optional)) {
clear_action_flag_because(first, pcmk__action_optional, then);
}
}
if (pcmk_is_set(type, pcmk__ar_first_else_then)
&& pcmk_is_set(filter, pcmk__action_optional)
&& !pcmk_is_set(first->flags, pcmk__action_runnable)) {
clear_action_flag_because(then, pcmk__action_migratable, first);
pcmk__clear_action_flags(then, pcmk__action_pseudo);
}
if (pcmk_is_set(type, pcmk__ar_unrunnable_first_blocks)
&& pcmk_is_set(filter, pcmk__action_runnable)
&& pcmk_is_set(then->flags, pcmk__action_runnable)
&& !pcmk_is_set(flags, pcmk__action_runnable)) {
clear_action_flag_because(then, pcmk__action_runnable, first);
clear_action_flag_because(then, pcmk__action_migratable, first);
}
if (pcmk_is_set(type, pcmk__ar_first_implies_then)
&& pcmk_is_set(filter, pcmk__action_optional)
&& pcmk_is_set(then->flags, pcmk__action_optional)
&& !pcmk_is_set(flags, pcmk__action_optional)
&& !pcmk_is_set(first->flags, pcmk__action_migratable)) {
clear_action_flag_because(then, pcmk__action_optional, first);
}
if (pcmk_is_set(type, pcmk__ar_intermediate_stop)) {
handle_restart_ordering(first, then, filter);
}
if (then_flags != then->flags) {
pcmk__set_updated_flags(changed, first, pcmk__updated_then);
pcmk__rsc_trace(then->rsc,
"%s on %s: flags are now %#.6x (was %#.6x) "
"because of 'first' %s (%#.6x)",
then->uuid, pcmk__node_name(then->node),
then->flags, then_flags, first->uuid, first->flags);
if ((then->rsc != NULL) && (then->rsc->priv->parent != NULL)) {
// Required to handle "X_stop then X_start" for cloned groups
pcmk__update_action_for_orderings(then, scheduler);
}
}
if (first_flags != first->flags) {
pcmk__set_updated_flags(changed, first, pcmk__updated_first);
pcmk__rsc_trace(first->rsc,
"%s on %s: flags are now %#.6x (was %#.6x) "
"because of 'then' %s (%#.6x)",
first->uuid, pcmk__node_name(first->node),
first->flags, first_flags, then->uuid, then->flags);
}
return changed;
}
/*!
* \internal
* \brief Trace-log an action (optionally with its dependent actions)
*
* \param[in] pre_text If not NULL, prefix the log with this plus ": "
* \param[in] action Action to log
* \param[in] details If true, recursively log dependent actions
*/
void
pcmk__log_action(const char *pre_text, const pcmk_action_t *action,
bool details)
{
const char *node_uname = NULL;
const char *node_uuid = NULL;
const char *desc = NULL;
CRM_CHECK(action != NULL, return);
if (!pcmk_is_set(action->flags, pcmk__action_pseudo)) {
if (action->node != NULL) {
node_uname = action->node->priv->name;
node_uuid = action->node->priv->id;
} else {
node_uname = "<none>";
}
}
switch (pcmk__parse_action(action->task)) {
case pcmk__action_fence:
case pcmk__action_shutdown:
if (pcmk_is_set(action->flags, pcmk__action_pseudo)) {
desc = "Pseudo ";
} else if (pcmk_is_set(action->flags, pcmk__action_optional)) {
desc = "Optional ";
} else if (!pcmk_is_set(action->flags, pcmk__action_runnable)) {
desc = "!!Non-Startable!! ";
} else {
desc = "(Provisional) ";
}
crm_trace("%s%s%sAction %d: %s%s%s%s%s%s",
((pre_text == NULL)? "" : pre_text),
((pre_text == NULL)? "" : ": "),
desc, action->id, action->uuid,
(node_uname? "\ton " : ""), (node_uname? node_uname : ""),
(node_uuid? "\t\t(" : ""), (node_uuid? node_uuid : ""),
(node_uuid? ")" : ""));
break;
default:
if (pcmk_is_set(action->flags, pcmk__action_optional)) {
desc = "Optional ";
} else if (pcmk_is_set(action->flags, pcmk__action_pseudo)) {
desc = "Pseudo ";
} else if (!pcmk_is_set(action->flags, pcmk__action_runnable)) {
desc = "!!Non-Startable!! ";
} else {
desc = "(Provisional) ";
}
crm_trace("%s%s%sAction %d: %s %s%s%s%s%s%s",
((pre_text == NULL)? "" : pre_text),
((pre_text == NULL)? "" : ": "),
desc, action->id, action->uuid,
(action->rsc? action->rsc->id : "<none>"),
(node_uname? "\ton " : ""), (node_uname? node_uname : ""),
(node_uuid? "\t\t(" : ""), (node_uuid? node_uuid : ""),
(node_uuid? ")" : ""));
break;
}
if (details) {
const GList *iter = NULL;
const pcmk__related_action_t *other = NULL;
crm_trace("\t\t====== Preceding Actions");
for (iter = action->actions_before; iter != NULL; iter = iter->next) {
other = (const pcmk__related_action_t *) iter->data;
pcmk__log_action("\t\t", other->action, false);
}
crm_trace("\t\t====== Subsequent Actions");
for (iter = action->actions_after; iter != NULL; iter = iter->next) {
other = (const pcmk__related_action_t *) iter->data;
pcmk__log_action("\t\t", other->action, false);
}
crm_trace("\t\t====== End");
} else {
crm_trace("\t\t(before=%d, after=%d)",
g_list_length(action->actions_before),
g_list_length(action->actions_after));
}
}
/*!
* \internal
* \brief Create a new shutdown action for a node
*
* \param[in,out] node Node being shut down
*
* \return Newly created shutdown action for \p node
*/
pcmk_action_t *
pcmk__new_shutdown_action(pcmk_node_t *node)
{
char *shutdown_id = NULL;
pcmk_action_t *shutdown_op = NULL;
pcmk__assert(node != NULL);
shutdown_id = crm_strdup_printf("%s-%s", PCMK_ACTION_DO_SHUTDOWN,
node->priv->name);
shutdown_op = custom_action(NULL, shutdown_id, PCMK_ACTION_DO_SHUTDOWN,
node, FALSE, node->priv->scheduler);
pcmk__order_stops_before_shutdown(node, shutdown_op);
pcmk__insert_meta(shutdown_op, PCMK__META_OP_NO_WAIT, PCMK_VALUE_TRUE);
return shutdown_op;
}
/*!
* \internal
* \brief Calculate and add an operation digest to XML
*
* Calculate an operation digest, which enables us to later determine when a
* restart is needed due to the resource's parameters being changed, and add it
* to given XML.
*
* \param[in] op Operation result from executor
* \param[in,out] update XML to add digest to
*/
static void
add_op_digest_to_xml(const lrmd_event_data_t *op, xmlNode *update)
{
char *digest = NULL;
xmlNode *args_xml = NULL;
if (op->params == NULL) {
return;
}
args_xml = pcmk__xe_create(NULL, PCMK_XE_PARAMETERS);
g_hash_table_foreach(op->params, hash2field, args_xml);
pcmk__filter_op_for_digest(args_xml);
digest = pcmk__digest_operation(args_xml);
crm_xml_add(update, PCMK__XA_OP_DIGEST, digest);
pcmk__xml_free(args_xml);
free(digest);
}
#define FAKE_TE_ID "xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx"
/*!
* \internal
* \brief Create XML for resource operation history update
*
* \param[in,out] parent Parent XML node to add to
* \param[in,out] op Operation event data
* \param[in] caller_version DC feature set
* \param[in] target_rc Expected result of operation
* \param[in] node Name of node on which operation was performed
* \param[in] origin Arbitrary description of update source
*
* \return Newly created XML node for history update
*/
xmlNode *
pcmk__create_history_xml(xmlNode *parent, lrmd_event_data_t *op,
const char *caller_version, int target_rc,
const char *node, const char *origin)
{
char *key = NULL;
char *magic = NULL;
char *op_id = NULL;
char *op_id_additional = NULL;
char *local_user_data = NULL;
const char *exit_reason = NULL;
xmlNode *xml_op = NULL;
const char *task = NULL;
CRM_CHECK(op != NULL, return NULL);
crm_trace("Creating history XML for %s-interval %s action for %s on %s "
"(DC version: %s, origin: %s)",
pcmk__readable_interval(op->interval_ms), op->op_type, op->rsc_id,
((node == NULL)? "no node" : node), caller_version, origin);
task = op->op_type;
/* Record a successful agent reload as a start, and a failed one as a
* monitor, to make life easier for the scheduler when determining the
* current state.
*
* @COMPAT We should check "reload" here only if the operation was for a
* pre-OCF-1.1 resource agent, but we don't know that here, and we should
* only ever get results for actions scheduled by us, so we can reasonably
* assume any "reload" is actually a pre-1.1 agent reload.
+ *
+ * @TODO This remapping can make log messages with task confusing for users
+ * (for example, an "Initiating reload ..." followed by "... start ...
+ * confirmed"). Either do this remapping in the scheduler if possible, or
+ * store the original task in a new XML attribute for later logging.
*/
if (pcmk__str_any_of(task, PCMK_ACTION_RELOAD, PCMK_ACTION_RELOAD_AGENT,
NULL)) {
if (op->op_status == PCMK_EXEC_DONE) {
task = PCMK_ACTION_START;
} else {
task = PCMK_ACTION_MONITOR;
}
}
key = pcmk__op_key(op->rsc_id, task, op->interval_ms);
if (pcmk__str_eq(task, PCMK_ACTION_NOTIFY, pcmk__str_none)) {
const char *n_type = crm_meta_value(op->params, "notify_type");
const char *n_task = crm_meta_value(op->params, "notify_operation");
CRM_LOG_ASSERT(n_type != NULL);
CRM_LOG_ASSERT(n_task != NULL);
op_id = pcmk__notify_key(op->rsc_id, n_type, n_task);
if (op->op_status != PCMK_EXEC_PENDING) {
/* Ignore notify errors.
*
* @TODO It might be better to keep the correct result here, and
* ignore it in process_graph_event().
*/
lrmd__set_result(op, PCMK_OCF_OK, PCMK_EXEC_DONE, NULL);
}
/* Migration history is preserved separately, which usually matters for
* multiple nodes and is important for future cluster transitions.
*/
} else if (pcmk__str_any_of(op->op_type, PCMK_ACTION_MIGRATE_TO,
PCMK_ACTION_MIGRATE_FROM, NULL)) {
op_id = strdup(key);
} else if (did_rsc_op_fail(op, target_rc)) {
op_id = pcmk__op_key(op->rsc_id, "last_failure", 0);
if (op->interval_ms == 0) {
/* Ensure 'last' gets updated, in case PCMK_META_RECORD_PENDING is
* true
*/
op_id_additional = pcmk__op_key(op->rsc_id, "last", 0);
}
exit_reason = op->exit_reason;
} else if (op->interval_ms > 0) {
op_id = strdup(key);
} else {
op_id = pcmk__op_key(op->rsc_id, "last", 0);
}
again:
xml_op = pcmk__xe_first_child(parent, PCMK__XE_LRM_RSC_OP, PCMK_XA_ID,
op_id);
if (xml_op == NULL) {
xml_op = pcmk__xe_create(parent, PCMK__XE_LRM_RSC_OP);
}
if (op->user_data == NULL) {
crm_debug("Generating fake transition key for: " PCMK__OP_FMT
" %d from %s", op->rsc_id, op->op_type, op->interval_ms,
op->call_id, origin);
local_user_data = pcmk__transition_key(-1, op->call_id, target_rc,
FAKE_TE_ID);
op->user_data = local_user_data;
}
if (magic == NULL) {
magic = crm_strdup_printf("%d:%d;%s", op->op_status, op->rc,
(const char *) op->user_data);
}
crm_xml_add(xml_op, PCMK_XA_ID, op_id);
crm_xml_add(xml_op, PCMK__XA_OPERATION_KEY, key);
crm_xml_add(xml_op, PCMK_XA_OPERATION, task);
crm_xml_add(xml_op, PCMK_XA_CRM_DEBUG_ORIGIN, origin);
crm_xml_add(xml_op, PCMK_XA_CRM_FEATURE_SET, caller_version);
crm_xml_add(xml_op, PCMK__XA_TRANSITION_KEY, op->user_data);
crm_xml_add(xml_op, PCMK__XA_TRANSITION_MAGIC, magic);
crm_xml_add(xml_op, PCMK_XA_EXIT_REASON, pcmk__s(exit_reason, ""));
crm_xml_add(xml_op, PCMK__META_ON_NODE, node); // For context during triage
crm_xml_add_int(xml_op, PCMK__XA_CALL_ID, op->call_id);
crm_xml_add_int(xml_op, PCMK__XA_RC_CODE, op->rc);
crm_xml_add_int(xml_op, PCMK__XA_OP_STATUS, op->op_status);
crm_xml_add_ms(xml_op, PCMK_META_INTERVAL, op->interval_ms);
if ((op->t_run > 0) || (op->t_rcchange > 0) || (op->exec_time > 0)
|| (op->queue_time > 0)) {
crm_trace("Timing data (" PCMK__OP_FMT "): "
"last=%lld change=%lld exec=%u queue=%u",
op->rsc_id, op->op_type, op->interval_ms,
(long long) op->t_run, (long long) op->t_rcchange,
op->exec_time, op->queue_time);
if ((op->interval_ms > 0) && (op->t_rcchange > 0)) {
// Recurring ops may have changed rc after initial run
crm_xml_add_ll(xml_op, PCMK_XA_LAST_RC_CHANGE,
(long long) op->t_rcchange);
} else {
crm_xml_add_ll(xml_op, PCMK_XA_LAST_RC_CHANGE,
(long long) op->t_run);
}
crm_xml_add_int(xml_op, PCMK_XA_EXEC_TIME, op->exec_time);
crm_xml_add_int(xml_op, PCMK_XA_QUEUE_TIME, op->queue_time);
}
if (pcmk__str_any_of(op->op_type, PCMK_ACTION_MIGRATE_TO,
PCMK_ACTION_MIGRATE_FROM, NULL)) {
/* Record PCMK__META_MIGRATE_SOURCE and PCMK__META_MIGRATE_TARGET always
* for migrate ops.
*/
const char *name = PCMK__META_MIGRATE_SOURCE;
crm_xml_add(xml_op, name, crm_meta_value(op->params, name));
name = PCMK__META_MIGRATE_TARGET;
crm_xml_add(xml_op, name, crm_meta_value(op->params, name));
}
add_op_digest_to_xml(op, xml_op);
if (op_id_additional) {
free(op_id);
op_id = op_id_additional;
op_id_additional = NULL;
goto again;
}
if (local_user_data) {
free(local_user_data);
op->user_data = NULL;
}
free(magic);
free(op_id);
free(key);
return xml_op;
}
/*!
* \internal
* \brief Check whether an action shutdown-locks a resource to a node
*
* If the PCMK_OPT_SHUTDOWN_LOCK cluster property is set, resources will not be
* recovered on a different node if cleanly stopped, and may start only on that
* same node. This function checks whether that applies to a given action, so
* that the transition graph can be marked appropriately.
*
* \param[in] action Action to check
*
* \return true if \p action locks its resource to the action's node,
* otherwise false
*/
bool
pcmk__action_locks_rsc_to_node(const pcmk_action_t *action)
{
// Only resource actions taking place on resource's lock node are locked
if ((action == NULL) || (action->rsc == NULL)
|| !pcmk__same_node(action->node, action->rsc->priv->lock_node)) {
return false;
}
/* During shutdown, only stops are locked (otherwise, another action such as
* a demote would cause the controller to clear the lock)
*/
if (action->node->details->shutdown && (action->task != NULL)
&& (strcmp(action->task, PCMK_ACTION_STOP) != 0)) {
return false;
}
return true;
}
/* lowest to highest */
static gint
sort_action_id(gconstpointer a, gconstpointer b)
{
const pcmk__related_action_t *action_wrapper2 = a;
const pcmk__related_action_t *action_wrapper1 = b;
if (a == NULL) {
return 1;
}
if (b == NULL) {
return -1;
}
if (action_wrapper1->action->id < action_wrapper2->action->id) {
return 1;
}
if (action_wrapper1->action->id > action_wrapper2->action->id) {
return -1;
}
return 0;
}
/*!
* \internal
* \brief Remove any duplicate action inputs, merging action flags
*
* \param[in,out] action Action whose inputs should be checked
*/
void
pcmk__deduplicate_action_inputs(pcmk_action_t *action)
{
GList *item = NULL;
GList *next = NULL;
pcmk__related_action_t *last_input = NULL;
action->actions_before = g_list_sort(action->actions_before,
sort_action_id);
for (item = action->actions_before; item != NULL; item = next) {
pcmk__related_action_t *input = item->data;
next = item->next;
if ((last_input != NULL)
&& (input->action->id == last_input->action->id)) {
crm_trace("Input %s (%d) duplicate skipped for action %s (%d)",
input->action->uuid, input->action->id,
action->uuid, action->id);
/* For the purposes of scheduling, the ordering flags no longer
* matter, but crm_simulate looks at certain ones when creating a
* dot graph. Combining the flags is sufficient for that purpose.
*/
pcmk__set_relation_flags(last_input->flags, input->flags);
if (input->graphed) {
last_input->graphed = true;
}
free(item->data);
action->actions_before = g_list_delete_link(action->actions_before,
item);
} else {
last_input = input;
input->graphed = false;
}
}
}
/*!
* \internal
* \brief Output all scheduled actions
*
* \param[in,out] scheduler Scheduler data
*/
void
pcmk__output_actions(pcmk_scheduler_t *scheduler)
{
pcmk__output_t *out = scheduler->priv->out;
// Output node (non-resource) actions
for (GList *iter = scheduler->priv->actions;
iter != NULL; iter = iter->next) {
char *node_name = NULL;
char *task = NULL;
pcmk_action_t *action = (pcmk_action_t *) iter->data;
if (action->rsc != NULL) {
continue; // Resource actions will be output later
} else if (pcmk_is_set(action->flags, pcmk__action_optional)) {
continue; // This action was not scheduled
}
if (pcmk__str_eq(action->task, PCMK_ACTION_DO_SHUTDOWN,
pcmk__str_none)) {
task = strdup("Shutdown");
} else if (pcmk__str_eq(action->task, PCMK_ACTION_STONITH,
pcmk__str_none)) {
const char *op = g_hash_table_lookup(action->meta,
PCMK__META_STONITH_ACTION);
task = crm_strdup_printf("Fence (%s)", op);
} else {
continue; // Don't display other node action types
}
if (pcmk__is_guest_or_bundle_node(action->node)) {
const pcmk_resource_t *remote = action->node->priv->remote;
node_name = crm_strdup_printf("%s (resource: %s)",
pcmk__node_name(action->node),
remote->priv->launcher->id);
} else if (action->node != NULL) {
node_name = crm_strdup_printf("%s", pcmk__node_name(action->node));
}
out->message(out, "node-action", task, node_name, action->reason);
free(node_name);
free(task);
}
// Output resource actions
for (GList *iter = scheduler->priv->resources;
iter != NULL; iter = iter->next) {
pcmk_resource_t *rsc = (pcmk_resource_t *) iter->data;
rsc->priv->cmds->output_actions(rsc);
}
}
/*!
* \internal
* \brief Get action name needed to compare digest for configuration changes
*
* \param[in] task Action name from history
* \param[in] interval_ms Action interval (in milliseconds)
*
* \return Action name whose digest should be compared
*/
static const char *
task_for_digest(const char *task, guint interval_ms)
{
/* Certain actions need to be compared against the parameters used to start
* the resource.
*/
if ((interval_ms == 0)
&& pcmk__str_any_of(task, PCMK_ACTION_MONITOR, PCMK_ACTION_MIGRATE_FROM,
PCMK_ACTION_PROMOTE, NULL)) {
task = PCMK_ACTION_START;
}
return task;
}
/*!
* \internal
* \brief Check whether only sanitized parameters to an action changed
*
* When collecting CIB files for troubleshooting, crm_report will mask
* sensitive resource parameters. If simulations were run using that, affected
* resources would appear to need a restart, which would complicate
* troubleshooting. To avoid that, we save a "secure digest" of non-sensitive
* parameters. This function used that digest to check whether only masked
* parameters are different.
*
* \param[in] xml_op Resource history entry with secure digest
* \param[in] digest_data Operation digest information being compared
* \param[in] scheduler Scheduler data
*
* \return true if only sanitized parameters changed, otherwise false
*/
static bool
only_sanitized_changed(const xmlNode *xml_op,
const pcmk__op_digest_t *digest_data,
const pcmk_scheduler_t *scheduler)
{
const char *digest_secure = NULL;
if (!pcmk_is_set(scheduler->flags, pcmk__sched_sanitized)) {
// The scheduler is not being run as a simulation
return false;
}
digest_secure = crm_element_value(xml_op, PCMK__XA_OP_SECURE_DIGEST);
return (digest_data->rc != pcmk__digest_match) && (digest_secure != NULL)
&& (digest_data->digest_secure_calc != NULL)
&& (strcmp(digest_data->digest_secure_calc, digest_secure) == 0);
}
/*!
* \internal
* \brief Force a restart due to a configuration change
*
* \param[in,out] rsc Resource that action is for
* \param[in] task Name of action whose configuration changed
* \param[in] interval_ms Action interval (in milliseconds)
* \param[in,out] node Node where resource should be restarted
*/
static void
force_restart(pcmk_resource_t *rsc, const char *task, guint interval_ms,
pcmk_node_t *node)
{
char *key = pcmk__op_key(rsc->id, task, interval_ms);
pcmk_action_t *required = custom_action(rsc, key, task, NULL, FALSE,
rsc->priv->scheduler);
pe_action_set_reason(required, "resource definition change", true);
trigger_unfencing(rsc, node, "Device parameters changed", NULL,
rsc->priv->scheduler);
}
/*!
* \internal
* \brief Schedule a reload of a resource on a node
*
* \param[in,out] data Resource to reload
* \param[in] user_data Where resource should be reloaded
*/
static void
schedule_reload(gpointer data, gpointer user_data)
{
pcmk_resource_t *rsc = data;
const pcmk_node_t *node = user_data;
pcmk_action_t *reload = NULL;
// For collective resources, just call recursively for children
if (rsc->priv->variant > pcmk__rsc_variant_primitive) {
g_list_foreach(rsc->priv->children, schedule_reload, user_data);
return;
}
// Skip the reload in certain situations
if ((node == NULL)
|| !pcmk_is_set(rsc->flags, pcmk__rsc_managed)
|| pcmk_is_set(rsc->flags, pcmk__rsc_failed)) {
pcmk__rsc_trace(rsc, "Skip reload of %s:%s%s %s",
rsc->id,
pcmk_is_set(rsc->flags, pcmk__rsc_managed)? "" : " unmanaged",
pcmk_is_set(rsc->flags, pcmk__rsc_failed)? " failed" : "",
(node == NULL)? "inactive" : node->priv->name);
return;
}
/* If a resource's configuration changed while a start was pending,
* force a full restart instead of a reload.
*/
if (pcmk_is_set(rsc->flags, pcmk__rsc_start_pending)) {
pcmk__rsc_trace(rsc,
"%s: preventing agent reload because start pending",
rsc->id);
custom_action(rsc, stop_key(rsc), PCMK_ACTION_STOP, node, FALSE,
rsc->priv->scheduler);
return;
}
// Schedule the reload
pcmk__set_rsc_flags(rsc, pcmk__rsc_reload);
reload = custom_action(rsc, reload_key(rsc), PCMK_ACTION_RELOAD_AGENT, node,
FALSE, rsc->priv->scheduler);
pe_action_set_reason(reload, "resource definition change", FALSE);
// Set orderings so that a required stop or demote cancels the reload
pcmk__new_ordering(NULL, NULL, reload, rsc, stop_key(rsc), NULL,
pcmk__ar_ordered|pcmk__ar_then_cancels_first,
rsc->priv->scheduler);
pcmk__new_ordering(NULL, NULL, reload, rsc, demote_key(rsc), NULL,
pcmk__ar_ordered|pcmk__ar_then_cancels_first,
rsc->priv->scheduler);
}
/*!
* \internal
* \brief Handle any configuration change for an action
*
* Given an action from resource history, if the resource's configuration
* changed since the action was done, schedule any actions needed (restart,
* reload, unfencing, rescheduling recurring actions, etc.).
*
* \param[in,out] rsc Resource that action is for
* \param[in,out] node Node that action was on
* \param[in] xml_op Action XML from resource history
*
* \return true if action configuration changed, otherwise false
*/
bool
pcmk__check_action_config(pcmk_resource_t *rsc, pcmk_node_t *node,
const xmlNode *xml_op)
{
guint interval_ms = 0;
const char *task = NULL;
const pcmk__op_digest_t *digest_data = NULL;
CRM_CHECK((rsc != NULL) && (node != NULL) && (xml_op != NULL),
return false);
task = crm_element_value(xml_op, PCMK_XA_OPERATION);
CRM_CHECK(task != NULL, return false);
crm_element_value_ms(xml_op, PCMK_META_INTERVAL, &interval_ms);
// If this is a recurring action, check whether it has been orphaned
if (interval_ms > 0) {
if (pcmk__find_action_config(rsc, task, interval_ms, false) != NULL) {
pcmk__rsc_trace(rsc,
"%s-interval %s for %s on %s is in configuration",
pcmk__readable_interval(interval_ms), task, rsc->id,
pcmk__node_name(node));
} else if (pcmk_is_set(rsc->priv->scheduler->flags,
pcmk__sched_cancel_removed_actions)) {
pcmk__schedule_cancel(rsc,
crm_element_value(xml_op, PCMK__XA_CALL_ID),
task, interval_ms, node, "orphan");
return true;
} else {
pcmk__rsc_debug(rsc, "%s-interval %s for %s on %s is orphaned",
pcmk__readable_interval(interval_ms), task, rsc->id,
pcmk__node_name(node));
return true;
}
}
crm_trace("Checking %s-interval %s for %s on %s for configuration changes",
pcmk__readable_interval(interval_ms), task, rsc->id,
pcmk__node_name(node));
task = task_for_digest(task, interval_ms);
digest_data = rsc_action_digest_cmp(rsc, xml_op, node,
rsc->priv->scheduler);
if (only_sanitized_changed(xml_op, digest_data, rsc->priv->scheduler)) {
if (!pcmk__is_daemon && (rsc->priv->scheduler->priv->out != NULL)) {
pcmk__output_t *out = rsc->priv->scheduler->priv->out;
out->info(out,
"Only 'private' parameters to %s-interval %s for %s "
"on %s changed: %s",
pcmk__readable_interval(interval_ms), task, rsc->id,
pcmk__node_name(node),
crm_element_value(xml_op, PCMK__XA_TRANSITION_MAGIC));
}
return false;
}
switch (digest_data->rc) {
case pcmk__digest_restart:
crm_log_xml_debug(digest_data->params_restart, "params:restart");
force_restart(rsc, task, interval_ms, node);
return true;
case pcmk__digest_unknown:
case pcmk__digest_mismatch:
// Changes that can potentially be handled by an agent reload
if (interval_ms > 0) {
/* Recurring actions aren't reloaded per se, they are just
* re-scheduled so the next run uses the new parameters.
* The old instance will be cancelled automatically.
*/
crm_log_xml_debug(digest_data->params_all, "params:reschedule");
pcmk__reschedule_recurring(rsc, task, interval_ms, node);
} else if (crm_element_value(xml_op,
PCMK__XA_OP_RESTART_DIGEST) != NULL) {
// Agent supports reload, so use it
trigger_unfencing(rsc, node,
"Device parameters changed (reload)", NULL,
rsc->priv->scheduler);
crm_log_xml_debug(digest_data->params_all, "params:reload");
schedule_reload((gpointer) rsc, (gpointer) node);
} else {
pcmk__rsc_trace(rsc,
"Restarting %s "
"because agent doesn't support reload",
rsc->id);
crm_log_xml_debug(digest_data->params_restart,
"params:restart");
force_restart(rsc, task, interval_ms, node);
}
return true;
default:
break;
}
return false;
}
/*!
* \internal
* \brief Create a list of resource's action history entries, sorted by call ID
*
* \param[in] rsc_entry Resource's \c PCMK__XE_LRM_RSC_OP status XML
* \param[out] start_index Where to store index of start-like action, if any
* \param[out] stop_index Where to store index of stop action, if any
*/
static GList *
rsc_history_as_list(const xmlNode *rsc_entry, int *start_index, int *stop_index)
{
GList *ops = NULL;
for (xmlNode *rsc_op = pcmk__xe_first_child(rsc_entry, PCMK__XE_LRM_RSC_OP,
NULL, NULL);
rsc_op != NULL; rsc_op = pcmk__xe_next(rsc_op, PCMK__XE_LRM_RSC_OP)) {
ops = g_list_prepend(ops, rsc_op);
}
ops = g_list_sort(ops, sort_op_by_callid);
calculate_active_ops(ops, start_index, stop_index);
return ops;
}
/*!
* \internal
* \brief Process a resource's action history from the CIB status
*
* Given a resource's action history, if the resource's configuration
* changed since the actions were done, schedule any actions needed (restart,
* reload, unfencing, rescheduling recurring actions, clean-up, etc.).
* (This also cancels recurring actions for maintenance mode, which is not
* entirely related but convenient to do here.)
*
* \param[in] rsc_entry Resource's \c PCMK__XE_LRM_RSC_OP status XML
* \param[in,out] rsc Resource whose history is being processed
* \param[in,out] node Node whose history is being processed
*/
static void
process_rsc_history(const xmlNode *rsc_entry, pcmk_resource_t *rsc,
pcmk_node_t *node)
{
int offset = -1;
int stop_index = 0;
int start_index = 0;
GList *sorted_op_list = NULL;
if (pcmk_is_set(rsc->flags, pcmk__rsc_removed)) {
if (pcmk__is_anonymous_clone(pe__const_top_resource(rsc, false))) {
+ /* @TODO Should this be done for bundled primitives as well? Added
+ * by 2ac43ae31
+ */
pcmk__rsc_trace(rsc,
"Skipping configuration check "
"for orphaned clone instance %s",
rsc->id);
} else {
pcmk__rsc_trace(rsc,
"Skipping configuration check and scheduling "
"clean-up for orphaned resource %s", rsc->id);
pcmk__schedule_cleanup(rsc, node, false);
}
return;
}
if (pe_find_node_id(rsc->priv->active_nodes,
node->priv->id) == NULL) {
if (pcmk__rsc_agent_changed(rsc, node, rsc_entry, false)) {
pcmk__schedule_cleanup(rsc, node, false);
}
pcmk__rsc_trace(rsc,
"Skipping configuration check for %s "
"because no longer active on %s",
rsc->id, pcmk__node_name(node));
return;
}
pcmk__rsc_trace(rsc, "Checking for configuration changes for %s on %s",
rsc->id, pcmk__node_name(node));
if (pcmk__rsc_agent_changed(rsc, node, rsc_entry, true)) {
pcmk__schedule_cleanup(rsc, node, false);
}
sorted_op_list = rsc_history_as_list(rsc_entry, &start_index, &stop_index);
if (start_index < stop_index) {
return; // Resource is stopped
}
for (GList *iter = sorted_op_list; iter != NULL; iter = iter->next) {
xmlNode *rsc_op = (xmlNode *) iter->data;
const char *task = NULL;
guint interval_ms = 0;
if (++offset < start_index) {
// Skip actions that happened before a start
continue;
}
task = crm_element_value(rsc_op, PCMK_XA_OPERATION);
crm_element_value_ms(rsc_op, PCMK_META_INTERVAL, &interval_ms);
if ((interval_ms > 0)
&& (pcmk_is_set(rsc->flags, pcmk__rsc_maintenance)
|| node->details->maintenance)) {
// Maintenance mode cancels recurring operations
pcmk__schedule_cancel(rsc,
crm_element_value(rsc_op, PCMK__XA_CALL_ID),
task, interval_ms, node, "maintenance mode");
} else if ((interval_ms > 0)
|| pcmk__strcase_any_of(task, PCMK_ACTION_MONITOR,
PCMK_ACTION_START,
PCMK_ACTION_PROMOTE,
PCMK_ACTION_MIGRATE_FROM, NULL)) {
/* If a resource operation failed, and the operation's definition
* has changed, clear any fail count so they can be retried fresh.
*/
if (pe__bundle_needs_remote_name(rsc)) {
/* We haven't assigned resources to nodes yet, so if the
* REMOTE_CONTAINER_HACK is used, we may calculate the digest
* based on the literal "#uname" value rather than the properly
* substituted value. That would mistakenly make the action
* definition appear to have been changed. Defer the check until
* later in this case.
*/
pe__add_param_check(rsc_op, rsc, node, pcmk__check_active,
rsc->priv->scheduler);
} else if (pcmk__check_action_config(rsc, node, rsc_op)
&& (pe_get_failcount(node, rsc, NULL, pcmk__fc_effective,
NULL) != 0)) {
pe__clear_failcount(rsc, node, "action definition changed",
rsc->priv->scheduler);
}
}
}
g_list_free(sorted_op_list);
}
/*!
* \internal
* \brief Process a node's action history from the CIB status
*
* Given a node's resource history, if the resource's configuration changed
* since the actions were done, schedule any actions needed (restart,
* reload, unfencing, rescheduling recurring actions, clean-up, etc.).
* (This also cancels recurring actions for maintenance mode, which is not
* entirely related but convenient to do here.)
*
* \param[in,out] node Node whose history is being processed
* \param[in] lrm_rscs Node's \c PCMK__XE_LRM_RESOURCES from CIB status XML
*/
static void
process_node_history(pcmk_node_t *node, const xmlNode *lrm_rscs)
{
crm_trace("Processing node history for %s", pcmk__node_name(node));
for (const xmlNode *rsc_entry = pcmk__xe_first_child(lrm_rscs,
PCMK__XE_LRM_RESOURCE,
NULL, NULL);
rsc_entry != NULL;
rsc_entry = pcmk__xe_next(rsc_entry, PCMK__XE_LRM_RESOURCE)) {
if (rsc_entry->children != NULL) {
GList *result = pcmk__rscs_matching_id(pcmk__xe_id(rsc_entry),
node->priv->scheduler);
for (GList *iter = result; iter != NULL; iter = iter->next) {
pcmk_resource_t *rsc = (pcmk_resource_t *) iter->data;
if (pcmk__is_primitive(rsc)) {
process_rsc_history(rsc_entry, rsc, node);
}
}
g_list_free(result);
}
}
}
// XPath to find a node's resource history
#define XPATH_NODE_HISTORY "/" PCMK_XE_CIB "/" PCMK_XE_STATUS \
"/" PCMK__XE_NODE_STATE \
"[@" PCMK_XA_UNAME "='%s']" \
"/" PCMK__XE_LRM "/" PCMK__XE_LRM_RESOURCES
/*!
* \internal
* \brief Process any resource configuration changes in the CIB status
*
* Go through all nodes' resource history, and if a resource's configuration
* changed since its actions were done, schedule any actions needed (restart,
* reload, unfencing, rescheduling recurring actions, clean-up, etc.).
* (This also cancels recurring actions for maintenance mode, which is not
* entirely related but convenient to do here.)
*
* \param[in,out] scheduler Scheduler data
*/
void
pcmk__handle_rsc_config_changes(pcmk_scheduler_t *scheduler)
{
crm_trace("Check resource and action configuration for changes");
/* Rather than iterate through the status section, iterate through the nodes
* and search for the appropriate status subsection for each. This skips
* orphaned nodes and lets us eliminate some cases before searching the XML.
*/
for (GList *iter = scheduler->nodes; iter != NULL; iter = iter->next) {
pcmk_node_t *node = (pcmk_node_t *) iter->data;
/* Don't bother checking actions for a node that can't run actions ...
* unless it's in maintenance mode, in which case we still need to
* cancel any existing recurring monitors.
*/
if (node->details->maintenance
|| pcmk__node_available(node, false, false)) {
char *xpath = NULL;
xmlNode *history = NULL;
xpath = crm_strdup_printf(XPATH_NODE_HISTORY, node->priv->name);
history = get_xpath_object(xpath, scheduler->input, LOG_NEVER);
free(xpath);
process_node_history(node, history);
}
}
}
diff --git a/lib/pacemaker/pcmk_sched_location.c b/lib/pacemaker/pcmk_sched_location.c
index a3915b5435..22b76940d2 100644
--- a/lib/pacemaker/pcmk_sched_location.c
+++ b/lib/pacemaker/pcmk_sched_location.c
@@ -1,781 +1,783 @@
/*
* Copyright 2004-2024 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU General Public License version 2
* or later (GPLv2+) WITHOUT ANY WARRANTY.
*/
#include <crm_internal.h>
#include <stdbool.h>
#include <glib.h>
#include <crm/crm.h>
#include <crm/common/rules_internal.h>
#include <crm/pengine/status.h>
#include <pacemaker-internal.h>
#include "libpacemaker_private.h"
/*!
* \internal
* \brief Parse a role configuration for a location constraint
*
* \param[in] role_spec Role specification
* \param[out] role Where to store parsed role
*
* \return true if role specification is valid, otherwise false
*/
static bool
parse_location_role(const char *role_spec, enum rsc_role_e *role)
{
if (role_spec == NULL) {
*role = pcmk_role_unknown;
return true;
}
*role = pcmk_parse_role(role_spec);
switch (*role) {
case pcmk_role_unknown:
return false;
case pcmk_role_started:
case pcmk_role_unpromoted:
/* Any promotable clone instance cannot be promoted without being in
* the unpromoted role first. Therefore, any constraint for the
* started or unpromoted role applies to every role.
*/
*role = pcmk_role_unknown;
break;
default:
break;
}
return true;
}
/*!
* \internal
* \brief Get the score attribute name (if any) used for a rule
*
* \param[in] rule_xml Rule XML
* \param[out] allocated If the score attribute name needs to be allocated,
* this will be set to the non-const equivalent of the
* return value (should be set to NULL when passed)
* \param[in] rule_input Values used to evaluate rule criteria
*
* \return Score attribute name used for rule, or NULL if none
* \note The caller is responsible for freeing \p *allocated if it is non-NULL.
*/
static const char *
score_attribute_name(const xmlNode *rule_xml, char **allocated,
const pcmk_rule_input_t *rule_input)
{
const char *name = NULL;
name = crm_element_value(rule_xml, PCMK_XA_SCORE_ATTRIBUTE);
if (name == NULL) {
return NULL;
}
/* A score attribute name may use submatches extracted from a
* resource ID regular expression. For example, if score-attribute is
* "loc-\1", rsc-pattern is "ip-(.*)", and the resource ID is "ip-db", then
* the score attribute name is "loc-db".
*/
if ((rule_input->rsc_id != NULL) && (rule_input->rsc_id_nmatches > 0)) {
*allocated = pcmk__replace_submatches(name, rule_input->rsc_id,
rule_input->rsc_id_submatches,
rule_input->rsc_id_nmatches);
if (*allocated != NULL) {
name = *allocated;
}
}
return name;
}
/*!
* \internal
* \brief Parse a score from a rule without a score attribute
*
* \param[in] rule_xml Rule XML
* \param[out] score Where to store parsed score
*
* \return Standard Pacemaker return code
*/
static int
score_from_rule(const xmlNode *rule_xml, int *score)
{
int rc = pcmk_rc_ok;
const char *score_s = crm_element_value(rule_xml, PCMK_XA_SCORE);
if (score_s == NULL) { // Not possible with schema validation enabled
pcmk__config_err("Ignoring location constraint rule %s because "
"neither " PCMK_XA_SCORE " nor "
PCMK_XA_SCORE_ATTRIBUTE " was specified",
pcmk__xe_id(rule_xml));
return pcmk_rc_unpack_error;
}
rc = pcmk_parse_score(score_s, score, 0);
if (rc != pcmk_rc_ok) { // Not possible with schema validation enabled
pcmk__config_err("Ignoring location constraint rule %s because "
"'%s' is not a valid " PCMK_XA_SCORE ": %s",
pcmk__xe_id(rule_xml), score_s, pcmk_rc_str(rc));
return pcmk_rc_unpack_error;
}
return pcmk_rc_ok;
}
/*!
* \internal
* \brief Get a rule score from a node attribute
*
* \param[in] constraint_id Location constraint ID (for logging only)
* \param[in] attr_name Name of node attribute with score
* \param[in] node Node to get attribute for
* \param[in] rsc Resource being located
* \param[out] score Where to store parsed score
*
* \return Standard Pacemaker return code (pcmk_rc_ok if a valid score was
* parsed, ENXIO if the node attribute was unset, and some other value
* if the node attribute value was invalid)
*/
static int
score_from_attr(const char *constraint_id, const char *attr_name,
const pcmk_node_t *node, const pcmk_resource_t *rsc, int *score)
{
int rc = pcmk_rc_ok;
const char *target = NULL;
const char *score_s = NULL;
target = g_hash_table_lookup(rsc->priv->meta,
PCMK_META_CONTAINER_ATTRIBUTE_TARGET);
score_s = pcmk__node_attr(node, attr_name, target, pcmk__rsc_node_current);
if (pcmk__str_empty(score_s)) {
crm_info("Ignoring location %s for %s on %s "
"because it has no node attribute %s",
constraint_id, rsc->id, pcmk__node_name(node), attr_name);
return ENXIO;
}
rc = pcmk_parse_score(score_s, score, 0);
if (rc != pcmk_rc_ok) {
crm_warn("Ignoring location %s for node %s because node "
"attribute %s value '%s' is not a valid score: %s",
constraint_id, pcmk__node_name(node), attr_name,
score_s, pcmk_rc_str(rc));
return rc;
}
return pcmk_rc_ok;
}
/*!
* \internal
* \brief Generate a location constraint from a rule
*
* \param[in,out] rsc Resource that constraint is for
* \param[in] rule_xml Rule XML (sub-element of location constraint)
* \param[in] discovery Value of \c PCMK_XA_RESOURCE_DISCOVERY for
* constraint
* \param[out] next_change Where to set when rule evaluation will change
* \param[in,out] rule_input Values used to evaluate rule criteria
* (node-specific values will be overwritten by
* this function)
* \param[in] constraint_id ID of location constraint (for logging only)
*
* \return true if rule is valid, otherwise false
*/
static bool
generate_location_rule(pcmk_resource_t *rsc, xmlNode *rule_xml,
const char *discovery, crm_time_t *next_change,
pcmk_rule_input_t *rule_input, const char *constraint_id)
{
const char *rule_id = NULL;
const char *score_attr = NULL;
const char *boolean = NULL;
const char *role_spec = NULL;
GList *iter = NULL;
int score = 0;
char *local_score_attr = NULL;
pcmk__location_t *location_rule = NULL;
enum rsc_role_e role = pcmk_role_unknown;
enum pcmk__combine combine = pcmk__combine_unknown;
rule_xml = pcmk__xe_resolve_idref(rule_xml, rsc->priv->scheduler->input);
if (rule_xml == NULL) {
return false; // Error already logged
}
rule_id = crm_element_value(rule_xml, PCMK_XA_ID);
if (rule_id == NULL) {
pcmk__config_err("Ignoring location constraint '%s' because its rule "
"has no " PCMK_XA_ID,
constraint_id);
return false;
}
boolean = crm_element_value(rule_xml, PCMK_XA_BOOLEAN_OP);
role_spec = crm_element_value(rule_xml, PCMK_XA_ROLE);
if (parse_location_role(role_spec, &role)) {
crm_trace("Setting rule %s role filter to %s", rule_id, role_spec);
} else {
pcmk__config_err("Ignoring location constraint '%s' because rule '%s' "
"has invalid " PCMK_XA_ROLE " '%s'",
constraint_id, rule_id, role_spec);
return false;
}
combine = pcmk__parse_combine(boolean);
switch (combine) {
case pcmk__combine_and:
case pcmk__combine_or:
break;
default: // Not possible with schema validation enabled
pcmk__config_err("Ignoring location constraint '%s' because rule "
"'%s' has invalid " PCMK_XA_BOOLEAN_OP " '%s'",
constraint_id, rule_id, boolean);
return false;
}
/* Users may configure the rule with either a score or the name of a
* node attribute whose value should be used as the constraint score for
* that node.
*/
score_attr = score_attribute_name(rule_xml, &local_score_attr, rule_input);
if ((score_attr == NULL)
&& (score_from_rule(rule_xml, &score) != pcmk_rc_ok)) {
return false;
}
location_rule = pcmk__new_location(rule_id, rsc, 0, discovery, NULL);
CRM_CHECK(location_rule != NULL, return NULL);
location_rule->role_filter = role;
for (iter = rsc->priv->scheduler->nodes;
iter != NULL; iter = iter->next) {
pcmk_node_t *node = iter->data;
pcmk_node_t *local = NULL;
rule_input->node_attrs = node->priv->attrs;
rule_input->rsc_params = pe_rsc_params(rsc, node,
rsc->priv->scheduler);
if (pcmk_evaluate_rule(rule_xml, rule_input,
next_change) != pcmk_rc_ok) {
continue;
}
if ((score_attr != NULL)
&& (score_from_attr(constraint_id, score_attr, node, rsc,
&score) != pcmk_rc_ok)) {
continue; // Message already logged
}
local = pe__copy_node(node);
location_rule->nodes = g_list_prepend(location_rule->nodes, local);
local->assign->score = score;
pcmk__rsc_trace(rsc,
"Location %s score for %s on %s is %s via rule %s",
constraint_id, rsc->id, pcmk__node_name(node),
pcmk_readable_score(score), rule_id);
}
free(local_score_attr);
if (location_rule->nodes == NULL) {
crm_trace("No matching nodes for location constraint rule %s", rule_id);
} else {
crm_trace("Location constraint rule %s matched %d nodes",
rule_id, g_list_length(location_rule->nodes));
}
return true;
}
static void
unpack_rsc_location(xmlNode *xml_obj, pcmk_resource_t *rsc,
const char *role_spec, const char *score,
char *rsc_id_match, int rsc_id_nmatches,
regmatch_t *rsc_id_submatches)
{
const char *rsc_id = crm_element_value(xml_obj, PCMK_XA_RSC);
const char *id = crm_element_value(xml_obj, PCMK_XA_ID);
const char *node = crm_element_value(xml_obj, PCMK_XA_NODE);
const char *discovery = crm_element_value(xml_obj,
PCMK_XA_RESOURCE_DISCOVERY);
if (rsc == NULL) {
pcmk__config_warn("Ignoring constraint '%s' because resource '%s' "
"does not exist", id, rsc_id);
return;
}
if (score == NULL) {
score = crm_element_value(xml_obj, PCMK_XA_SCORE);
}
if ((node != NULL) && (score != NULL)) {
int score_i = 0;
int rc = pcmk_rc_ok;
pcmk_node_t *match = pcmk_find_node(rsc->priv->scheduler, node);
enum rsc_role_e role = pcmk_role_unknown;
pcmk__location_t *location = NULL;
if (match == NULL) {
crm_info("Ignoring location constraint %s "
"because '%s' is not a known node",
pcmk__s(id, "without ID"), node);
return;
}
rc = pcmk_parse_score(score, &score_i, 0);
if (rc != pcmk_rc_ok) { // Not possible with schema validation enabled
pcmk__config_err("Ignoring location constraint %s "
"because '%s' is not a valid score", id, score);
return;
}
if (role_spec == NULL) {
role_spec = crm_element_value(xml_obj, PCMK_XA_ROLE);
}
if (parse_location_role(role_spec, &role)) {
crm_trace("Setting location constraint %s role filter: %s",
id, role_spec);
} else { // Not possible with schema validation enabled
pcmk__config_err("Ignoring location constraint %s "
"because '%s' is not a valid " PCMK_XA_ROLE,
id, role_spec);
return;
}
location = pcmk__new_location(id, rsc, score_i, discovery, match);
if (location == NULL) {
return; // Error already logged
}
location->role_filter = role;
} else {
crm_time_t *next_change = crm_time_new_undefined();
xmlNode *rule_xml = pcmk__xe_first_child(xml_obj, PCMK_XE_RULE, NULL,
NULL);
pcmk_rule_input_t rule_input = {
.now = rsc->priv->scheduler->priv->now,
.rsc_meta = rsc->priv->meta,
.rsc_id = rsc_id_match,
.rsc_id_submatches = rsc_id_submatches,
.rsc_id_nmatches = rsc_id_nmatches,
};
generate_location_rule(rsc, rule_xml, discovery, next_change,
&rule_input, id);
/* If there is a point in the future when the evaluation of a rule will
* change, make sure the scheduler is re-run by that time.
*/
if (crm_time_is_defined(next_change)) {
time_t t = (time_t) crm_time_get_seconds_since_epoch(next_change);
pe__update_recheck_time(t, rsc->priv->scheduler,
"location rule evaluation");
}
crm_time_free(next_change);
}
}
static void
unpack_simple_location(xmlNode *xml_obj, pcmk_scheduler_t *scheduler)
{
const char *id = crm_element_value(xml_obj, PCMK_XA_ID);
const char *value = crm_element_value(xml_obj, PCMK_XA_RSC);
if (value) {
pcmk_resource_t *rsc;
rsc = pcmk__find_constraint_resource(scheduler->priv->resources, value);
unpack_rsc_location(xml_obj, rsc, NULL, NULL, NULL, 0, NULL);
}
value = crm_element_value(xml_obj, PCMK_XA_RSC_PATTERN);
if (value) {
regex_t regex;
bool invert = false;
if (value[0] == '!') {
value++;
invert = true;
}
if (regcomp(®ex, value, REG_EXTENDED) != 0) {
pcmk__config_err("Ignoring constraint '%s' because "
PCMK_XA_RSC_PATTERN
" has invalid value '%s'", id, value);
return;
}
for (GList *iter = scheduler->priv->resources;
iter != NULL; iter = iter->next) {
pcmk_resource_t *r = iter->data;
int nregs = 0;
regmatch_t *pmatch = NULL;
int status;
if (regex.re_nsub > 0) {
nregs = regex.re_nsub + 1;
} else {
nregs = 1;
}
pmatch = pcmk__assert_alloc(nregs, sizeof(regmatch_t));
status = regexec(®ex, r->id, nregs, pmatch, 0);
if (!invert && (status == 0)) {
crm_debug("'%s' matched '%s' for %s", r->id, value, id);
unpack_rsc_location(xml_obj, r, NULL, NULL, r->id, nregs,
pmatch);
} else if (invert && (status != 0)) {
crm_debug("'%s' is an inverted match of '%s' for %s",
r->id, value, id);
unpack_rsc_location(xml_obj, r, NULL, NULL, NULL, 0, NULL);
} else {
crm_trace("'%s' does not match '%s' for %s", r->id, value, id);
}
free(pmatch);
}
+ // @TODO Maybe log a notice if we did not match any resources
+
regfree(®ex);
}
}
// \return Standard Pacemaker return code
static int
unpack_location_tags(xmlNode *xml_obj, xmlNode **expanded_xml,
pcmk_scheduler_t *scheduler)
{
const char *id = NULL;
const char *rsc_id = NULL;
const char *state = NULL;
pcmk_resource_t *rsc = NULL;
pcmk__idref_t *tag = NULL;
xmlNode *rsc_set = NULL;
*expanded_xml = NULL;
CRM_CHECK(xml_obj != NULL, return EINVAL);
id = pcmk__xe_id(xml_obj);
if (id == NULL) {
pcmk__config_err("Ignoring <%s> constraint without " PCMK_XA_ID,
xml_obj->name);
return pcmk_rc_unpack_error;
}
// Check whether there are any resource sets with template or tag references
*expanded_xml = pcmk__expand_tags_in_sets(xml_obj, scheduler);
if (*expanded_xml != NULL) {
crm_log_xml_trace(*expanded_xml, "Expanded " PCMK_XE_RSC_LOCATION);
return pcmk_rc_ok;
}
rsc_id = crm_element_value(xml_obj, PCMK_XA_RSC);
if (rsc_id == NULL) {
return pcmk_rc_ok;
}
if (!pcmk__valid_resource_or_tag(scheduler, rsc_id, &rsc, &tag)) {
pcmk__config_err("Ignoring constraint '%s' because '%s' is not a "
"valid resource or tag", id, rsc_id);
return pcmk_rc_unpack_error;
} else if (rsc != NULL) {
// No template is referenced
return pcmk_rc_ok;
}
state = crm_element_value(xml_obj, PCMK_XA_ROLE);
*expanded_xml = pcmk__xml_copy(NULL, xml_obj);
/* Convert any template or tag reference into constraint
* PCMK_XE_RESOURCE_SET
*/
if (!pcmk__tag_to_set(*expanded_xml, &rsc_set, PCMK_XA_RSC,
false, scheduler)) {
pcmk__xml_free(*expanded_xml);
*expanded_xml = NULL;
return pcmk_rc_unpack_error;
}
if (rsc_set != NULL) {
if (state != NULL) {
/* Move PCMK_XA_RSC_ROLE into converted PCMK_XE_RESOURCE_SET as
* PCMK_XA_ROLE attribute
*/
crm_xml_add(rsc_set, PCMK_XA_ROLE, state);
pcmk__xe_remove_attr(*expanded_xml, PCMK_XA_ROLE);
}
crm_log_xml_trace(*expanded_xml, "Expanded " PCMK_XE_RSC_LOCATION);
} else {
// No sets
pcmk__xml_free(*expanded_xml);
*expanded_xml = NULL;
}
return pcmk_rc_ok;
}
// \return Standard Pacemaker return code
static int
unpack_location_set(xmlNode *location, xmlNode *set,
pcmk_scheduler_t *scheduler)
{
xmlNode *xml_rsc = NULL;
pcmk_resource_t *resource = NULL;
const char *set_id;
const char *role;
const char *local_score;
CRM_CHECK(set != NULL, return EINVAL);
set_id = pcmk__xe_id(set);
if (set_id == NULL) {
pcmk__config_err("Ignoring " PCMK_XE_RESOURCE_SET " without "
PCMK_XA_ID " in constraint '%s'",
pcmk__s(pcmk__xe_id(location), "(missing ID)"));
return pcmk_rc_unpack_error;
}
role = crm_element_value(set, PCMK_XA_ROLE);
local_score = crm_element_value(set, PCMK_XA_SCORE);
for (xml_rsc = pcmk__xe_first_child(set, PCMK_XE_RESOURCE_REF, NULL, NULL);
xml_rsc != NULL;
xml_rsc = pcmk__xe_next(xml_rsc, PCMK_XE_RESOURCE_REF)) {
resource = pcmk__find_constraint_resource(scheduler->priv->resources,
pcmk__xe_id(xml_rsc));
if (resource == NULL) {
pcmk__config_err("%s: No resource found for %s",
set_id, pcmk__xe_id(xml_rsc));
return pcmk_rc_unpack_error;
}
unpack_rsc_location(location, resource, role, local_score, NULL, 0,
NULL);
}
return pcmk_rc_ok;
}
void
pcmk__unpack_location(xmlNode *xml_obj, pcmk_scheduler_t *scheduler)
{
xmlNode *set = NULL;
bool any_sets = false;
xmlNode *orig_xml = NULL;
xmlNode *expanded_xml = NULL;
if (unpack_location_tags(xml_obj, &expanded_xml, scheduler) != pcmk_rc_ok) {
return;
}
if (expanded_xml) {
orig_xml = xml_obj;
xml_obj = expanded_xml;
}
for (set = pcmk__xe_first_child(xml_obj, PCMK_XE_RESOURCE_SET, NULL, NULL);
set != NULL; set = pcmk__xe_next(set, PCMK_XE_RESOURCE_SET)) {
any_sets = true;
set = pcmk__xe_resolve_idref(set, scheduler->input);
if ((set == NULL) // Configuration error, message already logged
|| (unpack_location_set(xml_obj, set, scheduler) != pcmk_rc_ok)) {
if (expanded_xml) {
pcmk__xml_free(expanded_xml);
}
return;
}
}
if (expanded_xml) {
pcmk__xml_free(expanded_xml);
xml_obj = orig_xml;
}
if (!any_sets) {
unpack_simple_location(xml_obj, scheduler);
}
}
/*!
* \internal
* \brief Add a new location constraint to scheduler data
*
* \param[in] id XML ID of location constraint
* \param[in,out] rsc Resource in location constraint
* \param[in] node_score Constraint score
* \param[in] probe_mode When resource should be probed on node
* \param[in] node Node in constraint (or NULL if rule-based)
*
* \return Newly allocated location constraint on success, otherwise NULL
* \note The result will be added to the cluster (via \p rsc) and should not be
* freed separately.
*/
pcmk__location_t *
pcmk__new_location(const char *id, pcmk_resource_t *rsc,
int node_score, const char *probe_mode, pcmk_node_t *node)
{
pcmk__location_t *new_con = NULL;
CRM_CHECK((node != NULL) || (node_score == 0), return NULL);
if (id == NULL) {
pcmk__config_err("Invalid constraint: no ID specified");
return NULL;
}
if (rsc == NULL) {
pcmk__config_err("Invalid constraint %s: no resource specified", id);
return NULL;
}
new_con = pcmk__assert_alloc(1, sizeof(pcmk__location_t));
new_con->id = pcmk__str_copy(id);
new_con->rsc = rsc;
new_con->nodes = NULL;
new_con->role_filter = pcmk_role_unknown;
if (pcmk__str_eq(probe_mode, PCMK_VALUE_ALWAYS,
pcmk__str_null_matches|pcmk__str_casei)) {
new_con->probe_mode = pcmk__probe_always;
} else if (pcmk__str_eq(probe_mode, PCMK_VALUE_NEVER, pcmk__str_casei)) {
new_con->probe_mode = pcmk__probe_never;
} else if (pcmk__str_eq(probe_mode, PCMK_VALUE_EXCLUSIVE,
pcmk__str_casei)) {
new_con->probe_mode = pcmk__probe_exclusive;
pcmk__set_rsc_flags(rsc, pcmk__rsc_exclusive_probes);
} else {
pcmk__config_err("Invalid " PCMK_XA_RESOURCE_DISCOVERY " value %s "
"in location constraint", probe_mode);
}
if (node != NULL) {
pcmk_node_t *copy = pe__copy_node(node);
copy->assign->score = node_score;
new_con->nodes = g_list_prepend(NULL, copy);
}
rsc->priv->scheduler->priv->location_constraints =
g_list_prepend(rsc->priv->scheduler->priv->location_constraints,
new_con);
rsc->priv->location_constraints =
g_list_prepend(rsc->priv->location_constraints, new_con);
return new_con;
}
/*!
* \internal
* \brief Apply all location constraints
*
* \param[in,out] scheduler Scheduler data
*/
void
pcmk__apply_locations(pcmk_scheduler_t *scheduler)
{
for (GList *iter = scheduler->priv->location_constraints;
iter != NULL; iter = iter->next) {
pcmk__location_t *location = iter->data;
location->rsc->priv->cmds->apply_location(location->rsc, location);
}
}
/*!
* \internal
* \brief Apply a location constraint to a resource's allowed node scores
*
* \param[in,out] rsc Resource to apply constraint to
* \param[in,out] location Location constraint to apply
*
* \note This does not consider the resource's children, so the resource's
* apply_location() method should be used instead in most cases.
*/
void
pcmk__apply_location(pcmk_resource_t *rsc, pcmk__location_t *location)
{
bool need_role = false;
pcmk__assert((rsc != NULL) && (location != NULL));
// If a role was specified, ensure constraint is applicable
need_role = (location->role_filter > pcmk_role_unknown);
if (need_role && (location->role_filter != rsc->priv->next_role)) {
pcmk__rsc_trace(rsc,
"Not applying %s to %s because role will be %s not %s",
location->id, rsc->id,
pcmk_role_text(rsc->priv->next_role),
pcmk_role_text(location->role_filter));
return;
}
if (location->nodes == NULL) {
pcmk__rsc_trace(rsc, "Not applying %s to %s because no nodes match",
location->id, rsc->id);
return;
}
for (GList *iter = location->nodes; iter != NULL; iter = iter->next) {
pcmk_node_t *node = iter->data;
pcmk_node_t *allowed_node = NULL;
allowed_node = g_hash_table_lookup(rsc->priv->allowed_nodes,
node->priv->id);
pcmk__rsc_trace(rsc, "Applying %s%s%s to %s score on %s: %c %s",
location->id,
(need_role? " for role " : ""),
(need_role? pcmk_role_text(location->role_filter) : ""),
rsc->id, pcmk__node_name(node),
((allowed_node == NULL)? '=' : '+'),
pcmk_readable_score(node->assign->score));
if (allowed_node == NULL) {
allowed_node = pe__copy_node(node);
g_hash_table_insert(rsc->priv->allowed_nodes,
(gpointer) allowed_node->priv->id,
allowed_node);
} else {
allowed_node->assign->score =
pcmk__add_scores(allowed_node->assign->score,
node->assign->score);
}
if (allowed_node->assign->probe_mode < location->probe_mode) {
if (location->probe_mode == pcmk__probe_exclusive) {
pcmk__set_rsc_flags(rsc, pcmk__rsc_exclusive_probes);
}
/* exclusive > never > always... always is default */
allowed_node->assign->probe_mode = location->probe_mode;
}
}
}
diff --git a/lib/pacemaker/pcmk_sched_ordering.c b/lib/pacemaker/pcmk_sched_ordering.c
index 5bc2ffc919..2308be795e 100644
--- a/lib/pacemaker/pcmk_sched_ordering.c
+++ b/lib/pacemaker/pcmk_sched_ordering.c
@@ -1,1496 +1,1499 @@
/*
* Copyright 2004-2024 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU General Public License version 2
* or later (GPLv2+) WITHOUT ANY WARRANTY.
*/
#include <crm_internal.h>
#include <inttypes.h> // PRIx32
#include <stdbool.h>
#include <glib.h>
#include <crm/crm.h>
#include <pacemaker-internal.h>
#include "libpacemaker_private.h"
enum pe_order_kind {
pe_order_kind_optional,
pe_order_kind_mandatory,
pe_order_kind_serialize,
};
enum ordering_symmetry {
ordering_asymmetric, // the only relation in an asymmetric ordering
ordering_symmetric, // the normal relation in a symmetric ordering
ordering_symmetric_inverse, // the inverse relation in a symmetric ordering
};
+// @TODO de-functionize this for readability and possibly better log messages
#define EXPAND_CONSTRAINT_IDREF(__set, __rsc, __name) do { \
__rsc = pcmk__find_constraint_resource(scheduler->priv->resources, \
__name); \
if (__rsc == NULL) { \
pcmk__config_err("%s: No resource found for %s", __set, __name);\
return pcmk_rc_unpack_error; \
} \
} while (0)
static const char *
invert_action(const char *action)
{
if (pcmk__str_eq(action, PCMK_ACTION_START, pcmk__str_none)) {
return PCMK_ACTION_STOP;
} else if (pcmk__str_eq(action, PCMK_ACTION_STOP, pcmk__str_none)) {
return PCMK_ACTION_START;
} else if (pcmk__str_eq(action, PCMK_ACTION_PROMOTE, pcmk__str_none)) {
return PCMK_ACTION_DEMOTE;
} else if (pcmk__str_eq(action, PCMK_ACTION_DEMOTE, pcmk__str_none)) {
return PCMK_ACTION_PROMOTE;
} else if (pcmk__str_eq(action, PCMK_ACTION_PROMOTED, pcmk__str_none)) {
return PCMK_ACTION_DEMOTED;
} else if (pcmk__str_eq(action, PCMK_ACTION_DEMOTED, pcmk__str_none)) {
return PCMK_ACTION_PROMOTED;
} else if (pcmk__str_eq(action, PCMK_ACTION_RUNNING, pcmk__str_none)) {
return PCMK_ACTION_STOPPED;
} else if (pcmk__str_eq(action, PCMK_ACTION_STOPPED, pcmk__str_none)) {
return PCMK_ACTION_RUNNING;
}
pcmk__config_warn("Unknown action '%s' specified in order constraint",
action);
return NULL;
}
static enum pe_order_kind
get_ordering_type(const xmlNode *xml_obj)
{
enum pe_order_kind kind_e = pe_order_kind_mandatory;
const char *kind = crm_element_value(xml_obj, PCMK_XA_KIND);
if (kind == NULL) {
const char *score = crm_element_value(xml_obj, PCMK_XA_SCORE);
kind_e = pe_order_kind_mandatory;
if (score) {
// @COMPAT deprecated informally since 1.0.7, formally since 2.0.1
int score_i = 0;
(void) pcmk_parse_score(score, &score_i, 0);
if (score_i == 0) {
kind_e = pe_order_kind_optional;
}
pcmk__warn_once(pcmk__wo_order_score,
"Support for '" PCMK_XA_SCORE "' in "
PCMK_XE_RSC_ORDER " is deprecated and will be "
"removed in a future release "
"(use '" PCMK_XA_KIND "' instead)");
}
} else if (pcmk__str_eq(kind, PCMK_VALUE_MANDATORY, pcmk__str_none)) {
kind_e = pe_order_kind_mandatory;
} else if (pcmk__str_eq(kind, PCMK_VALUE_OPTIONAL, pcmk__str_none)) {
kind_e = pe_order_kind_optional;
} else if (pcmk__str_eq(kind, PCMK_VALUE_SERIALIZE, pcmk__str_none)) {
kind_e = pe_order_kind_serialize;
} else {
pcmk__config_err("Resetting '" PCMK_XA_KIND "' for constraint %s to "
"'" PCMK_VALUE_MANDATORY "' because '%s' is not valid",
pcmk__s(pcmk__xe_id(xml_obj), "missing ID"), kind);
}
return kind_e;
}
/*!
* \internal
* \brief Get ordering symmetry from XML
*
* \param[in] xml_obj Ordering XML
* \param[in] parent_kind Default ordering kind
* \param[in] parent_symmetrical_s Parent element's \c PCMK_XA_SYMMETRICAL
* setting, if any
*
* \retval ordering_symmetric Ordering is symmetric
* \retval ordering_asymmetric Ordering is asymmetric
*/
static enum ordering_symmetry
get_ordering_symmetry(const xmlNode *xml_obj, enum pe_order_kind parent_kind,
const char *parent_symmetrical_s)
{
int rc = pcmk_rc_ok;
bool symmetric = false;
enum pe_order_kind kind = parent_kind; // Default to parent's kind
// Check ordering XML for explicit kind
if ((crm_element_value(xml_obj, PCMK_XA_KIND) != NULL)
|| (crm_element_value(xml_obj, PCMK_XA_SCORE) != NULL)) {
kind = get_ordering_type(xml_obj);
}
// Check ordering XML (and parent) for explicit PCMK_XA_SYMMETRICAL setting
rc = pcmk__xe_get_bool_attr(xml_obj, PCMK_XA_SYMMETRICAL, &symmetric);
if (rc != pcmk_rc_ok && parent_symmetrical_s != NULL) {
symmetric = crm_is_true(parent_symmetrical_s);
rc = pcmk_rc_ok;
}
if (rc == pcmk_rc_ok) {
if (symmetric) {
if (kind == pe_order_kind_serialize) {
pcmk__config_warn("Ignoring " PCMK_XA_SYMMETRICAL
" for '%s' because not valid with "
PCMK_XA_KIND " of '" PCMK_VALUE_SERIALIZE "'",
pcmk__xe_id(xml_obj));
} else {
return ordering_symmetric;
}
}
return ordering_asymmetric;
}
// Use default symmetry
if (kind == pe_order_kind_serialize) {
return ordering_asymmetric;
}
return ordering_symmetric;
}
/*!
* \internal
* \brief Get ordering flags appropriate to ordering kind
*
* \param[in] kind Ordering kind
* \param[in] first Action name for 'first' action
* \param[in] symmetry This ordering's symmetry role
*
* \return Minimal ordering flags appropriate to \p kind
*/
static uint32_t
ordering_flags_for_kind(enum pe_order_kind kind, const char *first,
enum ordering_symmetry symmetry)
{
uint32_t flags = pcmk__ar_none; // so we trace-log all flags set
switch (kind) {
case pe_order_kind_optional:
pcmk__set_relation_flags(flags, pcmk__ar_ordered);
break;
case pe_order_kind_serialize:
/* This flag is not used anywhere directly but means the relation
* will not match an equality comparison against pcmk__ar_none or
* pcmk__ar_ordered.
*/
pcmk__set_relation_flags(flags, pcmk__ar_serialize);
break;
case pe_order_kind_mandatory:
pcmk__set_relation_flags(flags, pcmk__ar_ordered);
switch (symmetry) {
case ordering_asymmetric:
pcmk__set_relation_flags(flags, pcmk__ar_asymmetric);
break;
case ordering_symmetric:
pcmk__set_relation_flags(flags,
pcmk__ar_first_implies_then);
if (pcmk__strcase_any_of(first, PCMK_ACTION_START,
PCMK_ACTION_PROMOTE, NULL)) {
pcmk__set_relation_flags(flags,
pcmk__ar_unrunnable_first_blocks);
}
break;
case ordering_symmetric_inverse:
pcmk__set_relation_flags(flags,
pcmk__ar_then_implies_first);
break;
}
break;
}
return flags;
}
/*!
* \internal
* \brief Find resource corresponding to ID specified in ordering
*
* \param[in] xml Ordering XML
* \param[in] resource_attr XML attribute name for resource ID
* \param[in] scheduler Scheduler data
*
* \return Resource corresponding to \p id, or NULL if none
*/
static pcmk_resource_t *
get_ordering_resource(const xmlNode *xml, const char *resource_attr,
const pcmk_scheduler_t *scheduler)
{
pcmk_resource_t *rsc = NULL;
const char *rsc_id = crm_element_value(xml, resource_attr);
if (rsc_id == NULL) {
pcmk__config_err("Ignoring constraint '%s' without %s",
pcmk__xe_id(xml), resource_attr);
return NULL;
}
rsc = pcmk__find_constraint_resource(scheduler->priv->resources, rsc_id);
if (rsc == NULL) {
pcmk__config_err("Ignoring constraint '%s' because resource '%s' "
"does not exist", pcmk__xe_id(xml), rsc_id);
return NULL;
}
return rsc;
}
/*!
* \internal
* \brief Determine minimum number of 'first' instances required in ordering
*
* \param[in] rsc 'First' resource in ordering
* \param[in] xml Ordering XML
*
* \return Minimum 'first' instances required (or 0 if not applicable)
*/
static int
get_minimum_first_instances(const pcmk_resource_t *rsc, const xmlNode *xml)
{
const char *clone_min = NULL;
bool require_all = false;
if (!pcmk__is_clone(rsc)) {
return 0;
}
clone_min = g_hash_table_lookup(rsc->priv->meta, PCMK_META_CLONE_MIN);
if (clone_min != NULL) {
int clone_min_int = 0;
pcmk__scan_min_int(clone_min, &clone_min_int, 0);
return clone_min_int;
}
/* @COMPAT 1.1.13:
* PCMK_XA_REQUIRE_ALL=PCMK_VALUE_FALSE is deprecated equivalent of
* PCMK_META_CLONE_MIN=1
*/
if (pcmk__xe_get_bool_attr(xml, PCMK_XA_REQUIRE_ALL,
&require_all) != ENODATA) {
pcmk__warn_once(pcmk__wo_require_all,
"Support for " PCMK_XA_REQUIRE_ALL " in ordering "
"constraints is deprecated and will be removed in a "
"future release (use " PCMK_META_CLONE_MIN " clone "
"meta-attribute instead)");
if (!require_all) {
return 1;
}
}
return 0;
}
/*!
* \internal
* \brief Create orderings for a constraint with \c PCMK_META_CLONE_MIN > 0
*
* \param[in] id Ordering ID
* \param[in,out] rsc_first 'First' resource in ordering (a clone)
* \param[in] action_first 'First' action in ordering
* \param[in] rsc_then 'Then' resource in ordering
* \param[in] action_then 'Then' action in ordering
* \param[in] flags Ordering flags
* \param[in] clone_min Minimum required instances of 'first'
*/
static void
clone_min_ordering(const char *id,
pcmk_resource_t *rsc_first, const char *action_first,
pcmk_resource_t *rsc_then, const char *action_then,
uint32_t flags, int clone_min)
{
// Create a pseudo-action for when the minimum instances are active
char *task = crm_strdup_printf(PCMK_ACTION_CLONE_ONE_OR_MORE ":%s", id);
pcmk_action_t *clone_min_met = get_pseudo_op(task,
rsc_first->priv->scheduler);
free(task);
/* Require the pseudo-action to have the required number of actions to be
* considered runnable before allowing the pseudo-action to be runnable.
*/
clone_min_met->required_runnable_before = clone_min;
// Order the actions for each clone instance before the pseudo-action
for (GList *iter = rsc_first->priv->children;
iter != NULL; iter = iter->next) {
pcmk_resource_t *child = iter->data;
pcmk__new_ordering(child, pcmk__op_key(child->id, action_first, 0),
NULL, NULL, NULL, clone_min_met,
pcmk__ar_min_runnable
|pcmk__ar_first_implies_then_graphed,
rsc_first->priv->scheduler);
}
// Order "then" action after the pseudo-action (if runnable)
pcmk__new_ordering(NULL, NULL, clone_min_met, rsc_then,
pcmk__op_key(rsc_then->id, action_then, 0),
NULL, flags|pcmk__ar_unrunnable_first_blocks,
rsc_first->priv->scheduler);
}
/*!
* \internal
* \brief Create new ordering for inverse of symmetric constraint
*
* \param[in] id Ordering ID (for logging only)
* \param[in] kind Ordering kind
* \param[in] rsc_first 'First' resource in ordering (a clone)
* \param[in] action_first 'First' action in ordering
* \param[in,out] rsc_then 'Then' resource in ordering
* \param[in] action_then 'Then' action in ordering
*/
static void
inverse_ordering(const char *id, enum pe_order_kind kind,
pcmk_resource_t *rsc_first, const char *action_first,
pcmk_resource_t *rsc_then, const char *action_then)
{
action_then = invert_action(action_then);
action_first = invert_action(action_first);
if ((action_then == NULL) || (action_first == NULL)) {
pcmk__config_warn("Cannot invert constraint '%s' "
"(please specify inverse manually)", id);
} else {
uint32_t flags = ordering_flags_for_kind(kind, action_first,
ordering_symmetric_inverse);
pcmk__order_resource_actions(rsc_then, action_then, rsc_first,
action_first, flags);
}
}
static void
unpack_simple_rsc_order(xmlNode *xml_obj, pcmk_scheduler_t *scheduler)
{
pcmk_resource_t *rsc_then = NULL;
pcmk_resource_t *rsc_first = NULL;
int min_required_before = 0;
enum pe_order_kind kind = pe_order_kind_mandatory;
uint32_t flags = pcmk__ar_none;
enum ordering_symmetry symmetry;
const char *action_then = NULL;
const char *action_first = NULL;
const char *id = NULL;
CRM_CHECK(xml_obj != NULL, return);
id = crm_element_value(xml_obj, PCMK_XA_ID);
if (id == NULL) {
pcmk__config_err("Ignoring <%s> constraint without " PCMK_XA_ID,
xml_obj->name);
return;
}
rsc_first = get_ordering_resource(xml_obj, PCMK_XA_FIRST, scheduler);
if (rsc_first == NULL) {
return;
}
rsc_then = get_ordering_resource(xml_obj, PCMK_XA_THEN, scheduler);
if (rsc_then == NULL) {
return;
}
action_first = crm_element_value(xml_obj, PCMK_XA_FIRST_ACTION);
if (action_first == NULL) {
action_first = PCMK_ACTION_START;
}
action_then = crm_element_value(xml_obj, PCMK_XA_THEN_ACTION);
if (action_then == NULL) {
action_then = action_first;
}
kind = get_ordering_type(xml_obj);
symmetry = get_ordering_symmetry(xml_obj, kind, NULL);
flags = ordering_flags_for_kind(kind, action_first, symmetry);
/* If there is a minimum number of instances that must be runnable before
* the 'then' action is runnable, we use a pseudo-action for convenience:
* minimum number of clone instances have runnable actions ->
* pseudo-action is runnable -> dependency is runnable.
*/
min_required_before = get_minimum_first_instances(rsc_first, xml_obj);
if (min_required_before > 0) {
clone_min_ordering(id, rsc_first, action_first, rsc_then, action_then,
flags, min_required_before);
} else {
pcmk__order_resource_actions(rsc_first, action_first, rsc_then,
action_then, flags);
}
if (symmetry == ordering_symmetric) {
inverse_ordering(id, kind, rsc_first, action_first,
rsc_then, action_then);
}
}
/*!
* \internal
* \brief Create a new ordering between two actions
*
* \param[in,out] first_rsc Resource for 'first' action (if NULL and
* \p first_action is a resource action, that
* resource will be used)
* \param[in,out] first_action_task Action key for 'first' action (if NULL and
* \p first_action is not NULL, its UUID will
* be used)
* \param[in,out] first_action 'first' action (if NULL, \p first_rsc and
* \p first_action_task must be set)
*
* \param[in] then_rsc Resource for 'then' action (if NULL and
* \p then_action is a resource action, that
* resource will be used)
* \param[in,out] then_action_task Action key for 'then' action (if NULL and
* \p then_action is not NULL, its UUID will
* be used)
* \param[in] then_action 'then' action (if NULL, \p then_rsc and
* \p then_action_task must be set)
*
* \param[in] flags Group of enum pcmk__action_relation_flags
* \param[in,out] sched Scheduler data to add ordering to
*
* \note This function takes ownership of first_action_task and
* then_action_task, which do not need to be freed by the caller.
*/
void
pcmk__new_ordering(pcmk_resource_t *first_rsc, char *first_action_task,
pcmk_action_t *first_action, pcmk_resource_t *then_rsc,
char *then_action_task, pcmk_action_t *then_action,
uint32_t flags, pcmk_scheduler_t *sched)
{
pcmk__action_relation_t *order = NULL;
// One of action or resource must be specified for each side
CRM_CHECK(((first_action != NULL) || (first_rsc != NULL))
&& ((then_action != NULL) || (then_rsc != NULL)),
free(first_action_task); free(then_action_task); return);
if ((first_rsc == NULL) && (first_action != NULL)) {
first_rsc = first_action->rsc;
}
if ((then_rsc == NULL) && (then_action != NULL)) {
then_rsc = then_action->rsc;
}
order = pcmk__assert_alloc(1, sizeof(pcmk__action_relation_t));
order->id = sched->priv->next_ordering_id++;
order->flags = flags;
order->rsc1 = first_rsc;
order->rsc2 = then_rsc;
order->action1 = first_action;
order->action2 = then_action;
order->task1 = first_action_task;
order->task2 = then_action_task;
if ((order->task1 == NULL) && (first_action != NULL)) {
order->task1 = strdup(first_action->uuid);
}
if ((order->task2 == NULL) && (then_action != NULL)) {
order->task2 = strdup(then_action->uuid);
}
if ((order->rsc1 == NULL) && (first_action != NULL)) {
order->rsc1 = first_action->rsc;
}
if ((order->rsc2 == NULL) && (then_action != NULL)) {
order->rsc2 = then_action->rsc;
}
pcmk__rsc_trace(first_rsc, "Created ordering %d for %s then %s",
(sched->priv->next_ordering_id - 1),
pcmk__s(order->task1, "an underspecified action"),
pcmk__s(order->task2, "an underspecified action"));
sched->priv->ordering_constraints =
g_list_prepend(sched->priv->ordering_constraints, order);
pcmk__order_migration_equivalents(order);
}
/*!
* \brief Unpack a set in an ordering constraint
*
* \param[in] set Set XML to unpack
* \param[in] parent_kind \c PCMK_XE_RSC_ORDER XML \c PCMK_XA_KIND
* attribute
* \param[in] parent_symmetrical_s \c PCMK_XE_RSC_ORDER XML
* \c PCMK_XA_SYMMETRICAL attribute
* \param[in,out] scheduler Scheduler data
*
* \return Standard Pacemaker return code
*/
static int
unpack_order_set(const xmlNode *set, enum pe_order_kind parent_kind,
const char *parent_symmetrical_s, pcmk_scheduler_t *scheduler)
{
GList *set_iter = NULL;
GList *resources = NULL;
pcmk_resource_t *last = NULL;
pcmk_resource_t *resource = NULL;
int local_kind = parent_kind;
bool sequential = false;
uint32_t flags = pcmk__ar_ordered;
enum ordering_symmetry symmetry;
char *key = NULL;
const char *id = pcmk__xe_id(set);
const char *action = crm_element_value(set, PCMK_XA_ACTION);
const char *sequential_s = crm_element_value(set, PCMK_XA_SEQUENTIAL);
const char *kind_s = crm_element_value(set, PCMK_XA_KIND);
if (action == NULL) {
action = PCMK_ACTION_START;
}
if (kind_s) {
local_kind = get_ordering_type(set);
}
if (sequential_s == NULL) {
sequential_s = "1";
}
sequential = crm_is_true(sequential_s);
symmetry = get_ordering_symmetry(set, parent_kind, parent_symmetrical_s);
flags = ordering_flags_for_kind(local_kind, action, symmetry);
for (const xmlNode *xml_rsc = pcmk__xe_first_child(set,
PCMK_XE_RESOURCE_REF,
NULL, NULL);
xml_rsc != NULL;
xml_rsc = pcmk__xe_next(xml_rsc, PCMK_XE_RESOURCE_REF)) {
EXPAND_CONSTRAINT_IDREF(id, resource, pcmk__xe_id(xml_rsc));
resources = g_list_append(resources, resource);
}
if (pcmk__list_of_1(resources)) {
crm_trace("Single set: %s", id);
goto done;
}
set_iter = resources;
while (set_iter != NULL) {
resource = (pcmk_resource_t *) set_iter->data;
set_iter = set_iter->next;
key = pcmk__op_key(resource->id, action, 0);
if (local_kind == pe_order_kind_serialize) {
/* Serialize before everything that comes after */
for (GList *iter = set_iter; iter != NULL; iter = iter->next) {
pcmk_resource_t *then_rsc = iter->data;
char *then_key = pcmk__op_key(then_rsc->id, action, 0);
pcmk__new_ordering(resource, strdup(key), NULL, then_rsc,
then_key, NULL, flags, scheduler);
}
} else if (sequential) {
if (last != NULL) {
pcmk__order_resource_actions(last, action, resource, action,
flags);
}
last = resource;
}
free(key);
}
if (symmetry == ordering_asymmetric) {
goto done;
}
last = NULL;
action = invert_action(action);
flags = ordering_flags_for_kind(local_kind, action,
ordering_symmetric_inverse);
set_iter = resources;
while (set_iter != NULL) {
resource = (pcmk_resource_t *) set_iter->data;
set_iter = set_iter->next;
if (sequential) {
if (last != NULL) {
pcmk__order_resource_actions(resource, action, last, action,
flags);
}
last = resource;
}
}
done:
g_list_free(resources);
return pcmk_rc_ok;
}
/*!
* \brief Order two resource sets relative to each other
*
* \param[in] id Ordering ID (for logging)
* \param[in] set1 First listed set
* \param[in] set2 Second listed set
* \param[in] kind Ordering kind
* \param[in,out] scheduler Scheduler data
* \param[in] symmetry Which ordering symmetry applies to this relation
*
* \return Standard Pacemaker return code
*/
static int
order_rsc_sets(const char *id, const xmlNode *set1, const xmlNode *set2,
enum pe_order_kind kind, pcmk_scheduler_t *scheduler,
enum ordering_symmetry symmetry)
{
const xmlNode *xml_rsc = NULL;
const xmlNode *xml_rsc_2 = NULL;
pcmk_resource_t *rsc_1 = NULL;
pcmk_resource_t *rsc_2 = NULL;
const char *action_1 = crm_element_value(set1, PCMK_XA_ACTION);
const char *action_2 = crm_element_value(set2, PCMK_XA_ACTION);
uint32_t flags = pcmk__ar_none;
bool require_all = true;
(void) pcmk__xe_get_bool_attr(set1, PCMK_XA_REQUIRE_ALL, &require_all);
if (action_1 == NULL) {
action_1 = PCMK_ACTION_START;
}
if (action_2 == NULL) {
action_2 = PCMK_ACTION_START;
}
if (symmetry == ordering_symmetric_inverse) {
action_1 = invert_action(action_1);
action_2 = invert_action(action_2);
}
if (pcmk__str_eq(PCMK_ACTION_STOP, action_1, pcmk__str_none)
|| pcmk__str_eq(PCMK_ACTION_DEMOTE, action_1, pcmk__str_none)) {
/* Assuming: A -> ( B || C) -> D
* The one-or-more logic only applies during the start/promote phase.
* During shutdown neither B nor can shutdown until D is down, so simply
* turn require_all back on.
*/
require_all = true;
}
flags = ordering_flags_for_kind(kind, action_1, symmetry);
/* If we have an unordered set1, whether it is sequential or not is
* irrelevant in regards to set2.
*/
if (!require_all) {
char *task = crm_strdup_printf(PCMK_ACTION_ONE_OR_MORE ":%s",
pcmk__xe_id(set1));
pcmk_action_t *unordered_action = get_pseudo_op(task, scheduler);
free(task);
unordered_action->required_runnable_before = 1;
for (xml_rsc = pcmk__xe_first_child(set1, PCMK_XE_RESOURCE_REF, NULL,
NULL);
xml_rsc != NULL;
xml_rsc = pcmk__xe_next(xml_rsc, PCMK_XE_RESOURCE_REF)) {
EXPAND_CONSTRAINT_IDREF(id, rsc_1, pcmk__xe_id(xml_rsc));
/* Add an ordering constraint between every element in set1 and the
* pseudo action. If any action in set1 is runnable the pseudo
* action will be runnable.
*/
pcmk__new_ordering(rsc_1, pcmk__op_key(rsc_1->id, action_1, 0),
NULL, NULL, NULL, unordered_action,
pcmk__ar_min_runnable
|pcmk__ar_first_implies_then_graphed,
scheduler);
}
for (xml_rsc_2 = pcmk__xe_first_child(set2, PCMK_XE_RESOURCE_REF, NULL,
NULL);
xml_rsc_2 != NULL;
xml_rsc_2 = pcmk__xe_next(xml_rsc_2, PCMK_XE_RESOURCE_REF)) {
EXPAND_CONSTRAINT_IDREF(id, rsc_2, pcmk__xe_id(xml_rsc_2));
/* Add an ordering constraint between the pseudo-action and every
* element in set2. If the pseudo-action is runnable, every action
* in set2 will be runnable.
*/
pcmk__new_ordering(NULL, NULL, unordered_action,
rsc_2, pcmk__op_key(rsc_2->id, action_2, 0),
NULL, flags|pcmk__ar_unrunnable_first_blocks,
scheduler);
}
return pcmk_rc_ok;
}
if (pcmk__xe_attr_is_true(set1, PCMK_XA_SEQUENTIAL)) {
if (symmetry == ordering_symmetric_inverse) {
// Get the first one
xml_rsc = pcmk__xe_first_child(set1, PCMK_XE_RESOURCE_REF, NULL,
NULL);
if (xml_rsc != NULL) {
EXPAND_CONSTRAINT_IDREF(id, rsc_1, pcmk__xe_id(xml_rsc));
}
} else {
// Get the last one
const char *rid = NULL;
for (xml_rsc = pcmk__xe_first_child(set1, PCMK_XE_RESOURCE_REF,
NULL, NULL);
xml_rsc != NULL;
xml_rsc = pcmk__xe_next(xml_rsc, PCMK_XE_RESOURCE_REF)) {
rid = pcmk__xe_id(xml_rsc);
}
EXPAND_CONSTRAINT_IDREF(id, rsc_1, rid);
}
}
if (pcmk__xe_attr_is_true(set2, PCMK_XA_SEQUENTIAL)) {
if (symmetry == ordering_symmetric_inverse) {
// Get the last one
const char *rid = NULL;
for (xml_rsc = pcmk__xe_first_child(set2, PCMK_XE_RESOURCE_REF,
NULL, NULL);
xml_rsc != NULL;
xml_rsc = pcmk__xe_next(xml_rsc, PCMK_XE_RESOURCE_REF)) {
rid = pcmk__xe_id(xml_rsc);
}
EXPAND_CONSTRAINT_IDREF(id, rsc_2, rid);
} else {
// Get the first one
xml_rsc = pcmk__xe_first_child(set2, PCMK_XE_RESOURCE_REF, NULL,
NULL);
if (xml_rsc != NULL) {
EXPAND_CONSTRAINT_IDREF(id, rsc_2, pcmk__xe_id(xml_rsc));
}
}
}
if ((rsc_1 != NULL) && (rsc_2 != NULL)) {
pcmk__order_resource_actions(rsc_1, action_1, rsc_2, action_2, flags);
} else if (rsc_1 != NULL) {
for (xml_rsc = pcmk__xe_first_child(set2, PCMK_XE_RESOURCE_REF, NULL,
NULL);
xml_rsc != NULL;
xml_rsc = pcmk__xe_next(xml_rsc, PCMK_XE_RESOURCE_REF)) {
EXPAND_CONSTRAINT_IDREF(id, rsc_2, pcmk__xe_id(xml_rsc));
pcmk__order_resource_actions(rsc_1, action_1, rsc_2, action_2,
flags);
}
} else if (rsc_2 != NULL) {
for (xml_rsc = pcmk__xe_first_child(set1, PCMK_XE_RESOURCE_REF, NULL,
NULL);
xml_rsc != NULL;
xml_rsc = pcmk__xe_next(xml_rsc, PCMK_XE_RESOURCE_REF)) {
EXPAND_CONSTRAINT_IDREF(id, rsc_1, pcmk__xe_id(xml_rsc));
pcmk__order_resource_actions(rsc_1, action_1, rsc_2, action_2,
flags);
}
} else {
for (xml_rsc = pcmk__xe_first_child(set1, PCMK_XE_RESOURCE_REF, NULL,
NULL);
xml_rsc != NULL;
xml_rsc = pcmk__xe_next(xml_rsc, PCMK_XE_RESOURCE_REF)) {
EXPAND_CONSTRAINT_IDREF(id, rsc_1, pcmk__xe_id(xml_rsc));
for (xmlNode *xml_rsc_2 = pcmk__xe_first_child(set2,
PCMK_XE_RESOURCE_REF,
NULL, NULL);
xml_rsc_2 != NULL;
xml_rsc_2 = pcmk__xe_next(xml_rsc_2, PCMK_XE_RESOURCE_REF)) {
EXPAND_CONSTRAINT_IDREF(id, rsc_2, pcmk__xe_id(xml_rsc_2));
pcmk__order_resource_actions(rsc_1, action_1, rsc_2,
action_2, flags);
}
}
}
return pcmk_rc_ok;
}
/*!
* \internal
* \brief If an ordering constraint uses resource tags, expand them
*
* \param[in,out] xml_obj Ordering constraint XML
* \param[out] expanded_xml Equivalent XML with tags expanded
* \param[in] scheduler Scheduler data
*
* \return Standard Pacemaker return code (specifically, pcmk_rc_ok on success,
* and pcmk_rc_unpack_error on invalid configuration)
*/
static int
unpack_order_tags(xmlNode *xml_obj, xmlNode **expanded_xml,
const pcmk_scheduler_t *scheduler)
{
const char *id_first = NULL;
const char *id_then = NULL;
const char *action_first = NULL;
const char *action_then = NULL;
pcmk_resource_t *rsc_first = NULL;
pcmk_resource_t *rsc_then = NULL;
pcmk__idref_t *tag_first = NULL;
pcmk__idref_t *tag_then = NULL;
xmlNode *rsc_set_first = NULL;
xmlNode *rsc_set_then = NULL;
bool any_sets = false;
// Check whether there are any resource sets with template or tag references
*expanded_xml = pcmk__expand_tags_in_sets(xml_obj, scheduler);
if (*expanded_xml != NULL) {
crm_log_xml_trace(*expanded_xml, "Expanded " PCMK_XE_RSC_ORDER);
return pcmk_rc_ok;
}
id_first = crm_element_value(xml_obj, PCMK_XA_FIRST);
id_then = crm_element_value(xml_obj, PCMK_XA_THEN);
if ((id_first == NULL) || (id_then == NULL)) {
return pcmk_rc_ok;
}
if (!pcmk__valid_resource_or_tag(scheduler, id_first, &rsc_first,
&tag_first)) {
pcmk__config_err("Ignoring constraint '%s' because '%s' is not a "
"valid resource or tag",
pcmk__xe_id(xml_obj), id_first);
return pcmk_rc_unpack_error;
}
if (!pcmk__valid_resource_or_tag(scheduler, id_then, &rsc_then,
&tag_then)) {
pcmk__config_err("Ignoring constraint '%s' because '%s' is not a "
"valid resource or tag",
pcmk__xe_id(xml_obj), id_then);
return pcmk_rc_unpack_error;
}
if ((rsc_first != NULL) && (rsc_then != NULL)) {
// Neither side references a template or tag
return pcmk_rc_ok;
}
action_first = crm_element_value(xml_obj, PCMK_XA_FIRST_ACTION);
action_then = crm_element_value(xml_obj, PCMK_XA_THEN_ACTION);
*expanded_xml = pcmk__xml_copy(NULL, xml_obj);
/* Convert template/tag reference in PCMK_XA_FIRST into constraint
* PCMK_XE_RESOURCE_SET
*/
if (!pcmk__tag_to_set(*expanded_xml, &rsc_set_first, PCMK_XA_FIRST, true,
scheduler)) {
pcmk__xml_free(*expanded_xml);
*expanded_xml = NULL;
return pcmk_rc_unpack_error;
}
if (rsc_set_first != NULL) {
if (action_first != NULL) {
/* Move PCMK_XA_FIRST_ACTION into converted PCMK_XE_RESOURCE_SET as
* PCMK_XA_ACTION
*/
crm_xml_add(rsc_set_first, PCMK_XA_ACTION, action_first);
pcmk__xe_remove_attr(*expanded_xml, PCMK_XA_FIRST_ACTION);
}
any_sets = true;
}
/* Convert template/tag reference in PCMK_XA_THEN into constraint
* PCMK_XE_RESOURCE_SET
*/
if (!pcmk__tag_to_set(*expanded_xml, &rsc_set_then, PCMK_XA_THEN, true,
scheduler)) {
pcmk__xml_free(*expanded_xml);
*expanded_xml = NULL;
return pcmk_rc_unpack_error;
}
if (rsc_set_then != NULL) {
if (action_then != NULL) {
/* Move PCMK_XA_THEN_ACTION into converted PCMK_XE_RESOURCE_SET as
* PCMK_XA_ACTION
*/
crm_xml_add(rsc_set_then, PCMK_XA_ACTION, action_then);
pcmk__xe_remove_attr(*expanded_xml, PCMK_XA_THEN_ACTION);
}
any_sets = true;
}
if (any_sets) {
crm_log_xml_trace(*expanded_xml, "Expanded " PCMK_XE_RSC_ORDER);
} else {
pcmk__xml_free(*expanded_xml);
*expanded_xml = NULL;
}
return pcmk_rc_ok;
}
/*!
* \internal
* \brief Unpack ordering constraint XML
*
* \param[in,out] xml_obj Ordering constraint XML to unpack
* \param[in,out] scheduler Scheduler data
*/
void
pcmk__unpack_ordering(xmlNode *xml_obj, pcmk_scheduler_t *scheduler)
{
xmlNode *set = NULL;
xmlNode *last = NULL;
xmlNode *orig_xml = NULL;
xmlNode *expanded_xml = NULL;
const char *id = crm_element_value(xml_obj, PCMK_XA_ID);
const char *invert = crm_element_value(xml_obj, PCMK_XA_SYMMETRICAL);
enum pe_order_kind kind = get_ordering_type(xml_obj);
enum ordering_symmetry symmetry = get_ordering_symmetry(xml_obj, kind,
NULL);
// Expand any resource tags in the constraint XML
if (unpack_order_tags(xml_obj, &expanded_xml, scheduler) != pcmk_rc_ok) {
return;
}
if (expanded_xml != NULL) {
orig_xml = xml_obj;
xml_obj = expanded_xml;
}
// If the constraint has resource sets, unpack them
for (set = pcmk__xe_first_child(xml_obj, PCMK_XE_RESOURCE_SET, NULL, NULL);
set != NULL; set = pcmk__xe_next(set, PCMK_XE_RESOURCE_SET)) {
set = pcmk__xe_resolve_idref(set, scheduler->input);
if ((set == NULL) // Configuration error, message already logged
|| (unpack_order_set(set, kind, invert, scheduler) != pcmk_rc_ok)) {
if (expanded_xml != NULL) {
pcmk__xml_free(expanded_xml);
}
return;
}
if (last != NULL) {
if (order_rsc_sets(id, last, set, kind, scheduler,
symmetry) != pcmk_rc_ok) {
if (expanded_xml != NULL) {
pcmk__xml_free(expanded_xml);
}
return;
}
if ((symmetry == ordering_symmetric)
&& (order_rsc_sets(id, set, last, kind, scheduler,
ordering_symmetric_inverse) != pcmk_rc_ok)) {
if (expanded_xml != NULL) {
pcmk__xml_free(expanded_xml);
}
return;
}
}
last = set;
}
if (expanded_xml) {
pcmk__xml_free(expanded_xml);
xml_obj = orig_xml;
}
// If the constraint has no resource sets, unpack it as a simple ordering
if (last == NULL) {
return unpack_simple_rsc_order(xml_obj, scheduler);
}
}
static bool
ordering_is_invalid(pcmk_action_t *action, pcmk__related_action_t *input)
{
/* Prevent user-defined ordering constraints between resources
* running in a guest node and the resource that defines that node.
*/
if (!pcmk_is_set(input->flags, pcmk__ar_guest_allowed)
&& (input->action->rsc != NULL)
&& pcmk__rsc_corresponds_to_guest(action->rsc, input->action->node)) {
pcmk__config_warn("Invalid ordering constraint between %s and %s",
input->action->rsc->id, action->rsc->id);
return true;
}
/* If there's an order like
* "rscB_stop node2"-> "load_stopped_node2" -> "rscA_migrate_to node1"
*
* then rscA is being migrated from node1 to node2, while rscB is being
* migrated from node2 to node1. If there would be a graph loop,
* break the order "load_stopped_node2" -> "rscA_migrate_to node1".
*/
if ((input->flags == pcmk__ar_if_on_same_node_or_target)
&& (action->rsc != NULL)
&& pcmk__str_eq(action->task, PCMK_ACTION_MIGRATE_TO, pcmk__str_none)
&& pcmk__graph_has_loop(action, action, input)) {
return true;
}
return false;
}
void
pcmk__disable_invalid_orderings(pcmk_scheduler_t *scheduler)
{
for (GList *iter = scheduler->priv->actions;
iter != NULL; iter = iter->next) {
pcmk_action_t *action = (pcmk_action_t *) iter->data;
pcmk__related_action_t *input = NULL;
for (GList *input_iter = action->actions_before;
input_iter != NULL; input_iter = input_iter->next) {
input = input_iter->data;
if (ordering_is_invalid(action, input)) {
input->flags = pcmk__ar_none;
}
}
}
}
/*!
* \internal
* \brief Order stops on a node before the node's shutdown
*
* \param[in,out] node Node being shut down
* \param[in] shutdown_op Shutdown action for node
*/
void
pcmk__order_stops_before_shutdown(pcmk_node_t *node, pcmk_action_t *shutdown_op)
{
for (GList *iter = node->priv->scheduler->priv->actions;
iter != NULL; iter = iter->next) {
pcmk_action_t *action = (pcmk_action_t *) iter->data;
// Only stops on the node shutting down are relevant
if (!pcmk__same_node(action->node, node)
|| !pcmk__str_eq(action->task, PCMK_ACTION_STOP, pcmk__str_none)) {
continue;
}
// Resources and nodes in maintenance mode won't be touched
if (pcmk_is_set(action->rsc->flags, pcmk__rsc_maintenance)) {
pcmk__rsc_trace(action->rsc,
"Not ordering %s before shutdown of %s because "
"resource in maintenance mode",
action->uuid, pcmk__node_name(node));
continue;
} else if (node->details->maintenance) {
pcmk__rsc_trace(action->rsc,
"Not ordering %s before shutdown of %s because "
"node in maintenance mode",
action->uuid, pcmk__node_name(node));
continue;
}
/* Don't touch a resource that is unmanaged or blocked, to avoid
* blocking the shutdown (though if another action depends on this one,
* we may still end up blocking)
+ *
+ * @TODO This "if" looks wrong, create a regression test for these cases
*/
if (!pcmk_any_flags_set(action->rsc->flags,
pcmk__rsc_managed|pcmk__rsc_blocked)) {
pcmk__rsc_trace(action->rsc,
"Not ordering %s before shutdown of %s because "
"resource is unmanaged or blocked",
action->uuid, pcmk__node_name(node));
continue;
}
pcmk__rsc_trace(action->rsc, "Ordering %s before shutdown of %s",
action->uuid, pcmk__node_name(node));
pcmk__clear_action_flags(action, pcmk__action_optional);
pcmk__new_ordering(action->rsc, NULL, action, NULL,
strdup(PCMK_ACTION_DO_SHUTDOWN), shutdown_op,
pcmk__ar_ordered|pcmk__ar_unrunnable_first_blocks,
node->priv->scheduler);
}
}
/*!
* \brief Find resource actions matching directly or as child
*
* \param[in] rsc Resource to check
* \param[in] original_key Action key to search for (possibly referencing
* parent of \rsc)
*
* \return Newly allocated list of matching actions
* \note It is the caller's responsibility to free the result with g_list_free()
*/
static GList *
find_actions_by_task(const pcmk_resource_t *rsc, const char *original_key)
{
// Search under given task key directly
GList *list = find_actions(rsc->priv->actions, original_key, NULL);
if (list == NULL) {
// Search again using this resource's ID
char *key = NULL;
char *task = NULL;
guint interval_ms = 0;
CRM_CHECK(parse_op_key(original_key, NULL, &task, &interval_ms),
return NULL);
key = pcmk__op_key(rsc->id, task, interval_ms);
list = find_actions(rsc->priv->actions, key, NULL);
free(key);
free(task);
}
return list;
}
/*!
* \internal
* \brief Order relevant resource actions after a given action
*
* \param[in,out] first_action Action to order after (or NULL if none runnable)
* \param[in] rsc Resource whose actions should be ordered
* \param[in,out] order Ordering constraint being applied
*/
static void
order_resource_actions_after(pcmk_action_t *first_action,
const pcmk_resource_t *rsc,
pcmk__action_relation_t *order)
{
GList *then_actions = NULL;
uint32_t flags = pcmk__ar_none;
CRM_CHECK((rsc != NULL) && (order != NULL), return);
flags = order->flags;
pcmk__rsc_trace(rsc, "Applying ordering %d for 'then' resource %s",
order->id, rsc->id);
if (order->action2 != NULL) {
then_actions = g_list_prepend(NULL, order->action2);
} else {
then_actions = find_actions_by_task(rsc, order->task2);
}
if (then_actions == NULL) {
pcmk__rsc_trace(rsc, "Ignoring ordering %d: no %s actions found for %s",
order->id, order->task2, rsc->id);
return;
}
if ((first_action != NULL) && (first_action->rsc == rsc)
&& pcmk_is_set(first_action->flags, pcmk__action_migration_abort)) {
pcmk__rsc_trace(rsc,
"Detected dangling migration ordering (%s then %s %s)",
first_action->uuid, order->task2, rsc->id);
pcmk__clear_relation_flags(flags, pcmk__ar_first_implies_then);
}
if ((first_action == NULL)
&& !pcmk_is_set(flags, pcmk__ar_first_implies_then)) {
pcmk__rsc_debug(rsc,
"Ignoring ordering %d for %s: No first action found",
order->id, rsc->id);
g_list_free(then_actions);
return;
}
for (GList *iter = then_actions; iter != NULL; iter = iter->next) {
pcmk_action_t *then_action_iter = (pcmk_action_t *) iter->data;
if (first_action != NULL) {
order_actions(first_action, then_action_iter, flags);
} else {
pcmk__clear_action_flags(then_action_iter, pcmk__action_runnable);
crm_warn("%s of %s is unrunnable because there is no %s of %s "
"to order it after", then_action_iter->task, rsc->id,
order->task1, order->rsc1->id);
}
}
g_list_free(then_actions);
}
static void
rsc_order_first(pcmk_resource_t *first_rsc, pcmk__action_relation_t *order)
{
GList *first_actions = NULL;
pcmk_action_t *first_action = order->action1;
pcmk_resource_t *then_rsc = order->rsc2;
pcmk__assert(first_rsc != NULL);
pcmk__rsc_trace(first_rsc, "Applying ordering constraint %d (first: %s)",
order->id, first_rsc->id);
if (first_action != NULL) {
first_actions = g_list_prepend(NULL, first_action);
} else {
first_actions = find_actions_by_task(first_rsc, order->task1);
}
if ((first_actions == NULL) && (first_rsc == then_rsc)) {
pcmk__rsc_trace(first_rsc,
"Ignoring constraint %d: first (%s for %s) not found",
order->id, order->task1, first_rsc->id);
} else if (first_actions == NULL) {
char *key = NULL;
char *op_type = NULL;
guint interval_ms = 0;
enum rsc_role_e first_role;
parse_op_key(order->task1, NULL, &op_type, &interval_ms);
key = pcmk__op_key(first_rsc->id, op_type, interval_ms);
first_role = first_rsc->priv->fns->state(first_rsc, TRUE);
if ((first_role == pcmk_role_stopped)
&& pcmk__str_eq(op_type, PCMK_ACTION_STOP, pcmk__str_none)) {
free(key);
pcmk__rsc_trace(first_rsc,
"Ignoring constraint %d: first (%s for %s) "
"not found",
order->id, order->task1, first_rsc->id);
} else if ((first_role == pcmk_role_unpromoted)
&& pcmk__str_eq(op_type, PCMK_ACTION_DEMOTE,
pcmk__str_none)) {
free(key);
pcmk__rsc_trace(first_rsc,
"Ignoring constraint %d: first (%s for %s) "
"not found",
order->id, order->task1, first_rsc->id);
} else {
pcmk__rsc_trace(first_rsc,
"Creating first (%s for %s) for constraint %d ",
order->task1, first_rsc->id, order->id);
first_action = custom_action(first_rsc, key, op_type, NULL, TRUE,
first_rsc->priv->scheduler);
first_actions = g_list_prepend(NULL, first_action);
}
free(op_type);
}
if (then_rsc == NULL) {
if (order->action2 == NULL) {
pcmk__rsc_trace(first_rsc, "Ignoring constraint %d: then not found",
order->id);
return;
}
then_rsc = order->action2->rsc;
}
for (GList *iter = first_actions; iter != NULL; iter = iter->next) {
first_action = iter->data;
if (then_rsc == NULL) {
order_actions(first_action, order->action2, order->flags);
} else {
order_resource_actions_after(first_action, then_rsc, order);
}
}
g_list_free(first_actions);
}
// GFunc to call pcmk__block_colocation_dependents()
static void
block_colocation_dependents(gpointer data, gpointer user_data)
{
pcmk__block_colocation_dependents(data);
}
// GFunc to call pcmk__update_action_for_orderings()
static void
update_action_for_orderings(gpointer data, gpointer user_data)
{
pcmk__update_action_for_orderings((pcmk_action_t *) data,
(pcmk_scheduler_t *) user_data);
}
/*!
* \internal
* \brief Apply all ordering constraints
*
* \param[in,out] sched Scheduler data
*/
void
pcmk__apply_orderings(pcmk_scheduler_t *sched)
{
crm_trace("Applying ordering constraints");
/* Ordering constraints need to be processed in the order they were created.
* rsc_order_first() and order_resource_actions_after() require the relevant
* actions to already exist in some cases, but rsc_order_first() will create
* the 'first' action in certain cases. Thus calling rsc_order_first() can
* change the behavior of later-created orderings.
*
* Also, g_list_append() should be avoided for performance reasons, so we
* prepend orderings when creating them and reverse the list here.
*
* @TODO This is brittle and should be carefully redesigned so that the
* order of creation doesn't matter, and the reverse becomes unneeded.
*/
sched->priv->ordering_constraints =
g_list_reverse(sched->priv->ordering_constraints);
for (GList *iter = sched->priv->ordering_constraints;
iter != NULL; iter = iter->next) {
pcmk__action_relation_t *order = iter->data;
pcmk_resource_t *rsc = order->rsc1;
if (rsc != NULL) {
rsc_order_first(rsc, order);
continue;
}
rsc = order->rsc2;
if (rsc != NULL) {
order_resource_actions_after(order->action1, rsc, order);
} else {
crm_trace("Applying ordering constraint %d (non-resource actions)",
order->id);
order_actions(order->action1, order->action2, order->flags);
}
}
g_list_foreach(sched->priv->actions, block_colocation_dependents, NULL);
crm_trace("Ordering probes");
pcmk__order_probes(sched);
crm_trace("Updating %d actions", g_list_length(sched->priv->actions));
g_list_foreach(sched->priv->actions, update_action_for_orderings, sched);
pcmk__disable_invalid_orderings(sched);
}
/*!
* \internal
* \brief Order a given action after each action in a given list
*
* \param[in,out] after "After" action
* \param[in,out] list List of "before" actions
*/
void
pcmk__order_after_each(pcmk_action_t *after, GList *list)
{
const char *after_desc = (after->task == NULL)? after->uuid : after->task;
for (GList *iter = list; iter != NULL; iter = iter->next) {
pcmk_action_t *before = (pcmk_action_t *) iter->data;
const char *before_desc = before->task? before->task : before->uuid;
crm_debug("Ordering %s on %s before %s on %s",
before_desc, pcmk__node_name(before->node),
after_desc, pcmk__node_name(after->node));
order_actions(before, after, pcmk__ar_ordered);
}
}
/*!
* \internal
* \brief Order promotions and demotions for restarts of a clone or bundle
*
* \param[in,out] rsc Clone or bundle to order
*/
void
pcmk__promotable_restart_ordering(pcmk_resource_t *rsc)
{
// Order start and promote after all instances are stopped
pcmk__order_resource_actions(rsc, PCMK_ACTION_STOPPED,
rsc, PCMK_ACTION_START,
pcmk__ar_ordered);
pcmk__order_resource_actions(rsc, PCMK_ACTION_STOPPED,
rsc, PCMK_ACTION_PROMOTE,
pcmk__ar_ordered);
// Order stop, start, and promote after all instances are demoted
pcmk__order_resource_actions(rsc, PCMK_ACTION_DEMOTED,
rsc, PCMK_ACTION_STOP,
pcmk__ar_ordered);
pcmk__order_resource_actions(rsc, PCMK_ACTION_DEMOTED,
rsc, PCMK_ACTION_START,
pcmk__ar_ordered);
pcmk__order_resource_actions(rsc, PCMK_ACTION_DEMOTED,
rsc, PCMK_ACTION_PROMOTE,
pcmk__ar_ordered);
// Order promote after all instances are started
pcmk__order_resource_actions(rsc, PCMK_ACTION_RUNNING,
rsc, PCMK_ACTION_PROMOTE,
pcmk__ar_ordered);
// Order demote after all instances are demoted
pcmk__order_resource_actions(rsc, PCMK_ACTION_DEMOTE,
rsc, PCMK_ACTION_DEMOTED,
pcmk__ar_ordered);
}
diff --git a/lib/pacemaker/pcmk_sched_primitive.c b/lib/pacemaker/pcmk_sched_primitive.c
index ad68c6b49f..d95eef25a4 100644
--- a/lib/pacemaker/pcmk_sched_primitive.c
+++ b/lib/pacemaker/pcmk_sched_primitive.c
@@ -1,1716 +1,1717 @@
/*
* Copyright 2004-2024 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU General Public License version 2
* or later (GPLv2+) WITHOUT ANY WARRANTY.
*/
#include <crm_internal.h>
#include <stdbool.h>
#include <stdint.h> // uint8_t, uint32_t
#include <crm/common/xml.h>
#include <pacemaker-internal.h>
#include "libpacemaker_private.h"
static void stop_resource(pcmk_resource_t *rsc, pcmk_node_t *node,
bool optional);
static void start_resource(pcmk_resource_t *rsc, pcmk_node_t *node,
bool optional);
static void demote_resource(pcmk_resource_t *rsc, pcmk_node_t *node,
bool optional);
static void promote_resource(pcmk_resource_t *rsc, pcmk_node_t *node,
bool optional);
static void assert_role_error(pcmk_resource_t *rsc, pcmk_node_t *node,
bool optional);
#define RSC_ROLE_MAX (pcmk_role_promoted + 1)
static enum rsc_role_e rsc_state_matrix[RSC_ROLE_MAX][RSC_ROLE_MAX] = {
/* This array lists the immediate next role when transitioning from one role
* to a target role. For example, when going from Stopped to Promoted, the
* next role is Unpromoted, because the resource must be started before it
* can be promoted. The current state then becomes Started, which is fed
* into this array again, giving a next role of Promoted.
*
* Current role Immediate next role Final target role
* ------------ ------------------- -----------------
*/
/* Unknown */ { pcmk_role_unknown, /* Unknown */
pcmk_role_stopped, /* Stopped */
pcmk_role_stopped, /* Started */
pcmk_role_stopped, /* Unpromoted */
pcmk_role_stopped, /* Promoted */
},
/* Stopped */ { pcmk_role_stopped, /* Unknown */
pcmk_role_stopped, /* Stopped */
pcmk_role_started, /* Started */
pcmk_role_unpromoted, /* Unpromoted */
pcmk_role_unpromoted, /* Promoted */
},
/* Started */ { pcmk_role_stopped, /* Unknown */
pcmk_role_stopped, /* Stopped */
pcmk_role_started, /* Started */
pcmk_role_unpromoted, /* Unpromoted */
pcmk_role_promoted, /* Promoted */
},
/* Unpromoted */ { pcmk_role_stopped, /* Unknown */
pcmk_role_stopped, /* Stopped */
pcmk_role_stopped, /* Started */
pcmk_role_unpromoted, /* Unpromoted */
pcmk_role_promoted, /* Promoted */
},
/* Promoted */ { pcmk_role_stopped, /* Unknown */
pcmk_role_unpromoted, /* Stopped */
pcmk_role_unpromoted, /* Started */
pcmk_role_unpromoted, /* Unpromoted */
pcmk_role_promoted, /* Promoted */
},
};
/*!
* \internal
* \brief Function to schedule actions needed for a role change
*
* \param[in,out] rsc Resource whose role is changing
* \param[in,out] node Node where resource will be in its next role
* \param[in] optional Whether scheduled actions should be optional
*/
typedef void (*rsc_transition_fn)(pcmk_resource_t *rsc, pcmk_node_t *node,
bool optional);
static rsc_transition_fn rsc_action_matrix[RSC_ROLE_MAX][RSC_ROLE_MAX] = {
/* This array lists the function needed to transition directly from one role
* to another. NULL indicates that nothing is needed.
*
* Current role Transition function Next role
* ------------ ------------------- ----------
*/
/* Unknown */ { assert_role_error, /* Unknown */
stop_resource, /* Stopped */
assert_role_error, /* Started */
assert_role_error, /* Unpromoted */
assert_role_error, /* Promoted */
},
/* Stopped */ { assert_role_error, /* Unknown */
NULL, /* Stopped */
start_resource, /* Started */
start_resource, /* Unpromoted */
assert_role_error, /* Promoted */
},
/* Started */ { assert_role_error, /* Unknown */
stop_resource, /* Stopped */
NULL, /* Started */
NULL, /* Unpromoted */
promote_resource, /* Promoted */
},
/* Unpromoted */ { assert_role_error, /* Unknown */
stop_resource, /* Stopped */
stop_resource, /* Started */
NULL, /* Unpromoted */
promote_resource, /* Promoted */
},
/* Promoted */ { assert_role_error, /* Unknown */
demote_resource, /* Stopped */
demote_resource, /* Started */
demote_resource, /* Unpromoted */
NULL, /* Promoted */
},
};
/*!
* \internal
* \brief Get a list of a resource's allowed nodes sorted by node score
*
* \param[in] rsc Resource to check
*
* \return List of allowed nodes sorted by node score
*/
static GList *
sorted_allowed_nodes(const pcmk_resource_t *rsc)
{
if (rsc->priv->allowed_nodes != NULL) {
GList *nodes = g_hash_table_get_values(rsc->priv->allowed_nodes);
if (nodes != NULL) {
return pcmk__sort_nodes(nodes, pcmk__current_node(rsc));
}
}
return NULL;
}
/*!
* \internal
* \brief Assign a resource to its best allowed node, if possible
*
* \param[in,out] rsc Resource to choose a node for
* \param[in] prefer If not \c NULL, prefer this node when all else
* equal
* \param[in] stop_if_fail If \c true and \p rsc can't be assigned to a
* node, set next role to stopped and update
* existing actions
*
* \return true if \p rsc could be assigned to a node, otherwise false
*
* \note If \p stop_if_fail is \c false, then \c pcmk__unassign_resource() can
* completely undo the assignment. A successful assignment can be either
* undone or left alone as final. A failed assignment has the same effect
* as calling pcmk__unassign_resource(); there are no side effects on
* roles or actions.
*/
static bool
assign_best_node(pcmk_resource_t *rsc, const pcmk_node_t *prefer,
bool stop_if_fail)
{
GList *nodes = NULL;
pcmk_node_t *chosen = NULL;
pcmk_node_t *best = NULL;
const pcmk_node_t *most_free_node = pcmk__ban_insufficient_capacity(rsc);
if (prefer == NULL) {
prefer = most_free_node;
}
if (!pcmk_is_set(rsc->flags, pcmk__rsc_unassigned)) {
// We've already finished assignment of resources to nodes
return rsc->priv->assigned_node != NULL;
}
// Sort allowed nodes by score
nodes = sorted_allowed_nodes(rsc);
if (nodes != NULL) {
best = (pcmk_node_t *) nodes->data; // First node has best score
}
if ((prefer != NULL) && (nodes != NULL)) {
// Get the allowed node version of prefer
chosen = g_hash_table_lookup(rsc->priv->allowed_nodes,
prefer->priv->id);
if (chosen == NULL) {
pcmk__rsc_trace(rsc, "Preferred node %s for %s was unknown",
pcmk__node_name(prefer), rsc->id);
/* Favor the preferred node as long as its score is at least as good as
* the best allowed node's.
*
* An alternative would be to favor the preferred node even if the best
* node is better, when the best node's score is less than INFINITY.
*/
} else if (chosen->assign->score < best->assign->score) {
pcmk__rsc_trace(rsc, "Preferred node %s for %s was unsuitable",
pcmk__node_name(chosen), rsc->id);
chosen = NULL;
} else if (!pcmk__node_available(chosen, true, false)) {
pcmk__rsc_trace(rsc, "Preferred node %s for %s was unavailable",
pcmk__node_name(chosen), rsc->id);
chosen = NULL;
} else {
pcmk__rsc_trace(rsc,
"Chose preferred node %s for %s "
"(ignoring %d candidates)",
pcmk__node_name(chosen), rsc->id,
g_list_length(nodes));
}
}
if ((chosen == NULL) && (best != NULL)) {
/* Either there is no preferred node, or the preferred node is not
* suitable, but another node is allowed to run the resource.
*/
chosen = best;
if (!pcmk__is_unique_clone(rsc->priv->parent)
&& (chosen->assign->score > 0) // Zero not acceptable
&& pcmk__node_available(chosen, false, false)) {
/* If the resource is already running on a node, prefer that node if
* it is just as good as the chosen node.
*
* We don't do this for unique clone instances, because
* pcmk__assign_instances() has already assigned instances to their
* running nodes when appropriate, and if we get here, we don't want
* remaining unassigned instances to prefer a node that's already
* running another instance.
*/
pcmk_node_t *running = pcmk__current_node(rsc);
if (running == NULL) {
// Nothing to do
} else if (!pcmk__node_available(running, true, false)) {
pcmk__rsc_trace(rsc,
"Current node for %s (%s) can't run resources",
rsc->id, pcmk__node_name(running));
} else {
int nodes_with_best_score = 1;
for (GList *iter = nodes->next; iter; iter = iter->next) {
pcmk_node_t *allowed = (pcmk_node_t *) iter->data;
if (allowed->assign->score != chosen->assign->score) {
// The nodes are sorted by score, so no more are equal
break;
}
if (pcmk__same_node(allowed, running)) {
// Scores are equal, so prefer the current node
chosen = allowed;
}
nodes_with_best_score++;
}
if (nodes_with_best_score > 1) {
uint8_t log_level = LOG_INFO;
if (chosen->assign->score >= PCMK_SCORE_INFINITY) {
log_level = LOG_WARNING;
}
do_crm_log(log_level,
"Chose %s for %s from %d nodes with score %s",
pcmk__node_name(chosen), rsc->id,
nodes_with_best_score,
pcmk_readable_score(chosen->assign->score));
}
}
}
pcmk__rsc_trace(rsc, "Chose %s for %s from %d candidates",
pcmk__node_name(chosen), rsc->id, g_list_length(nodes));
}
pcmk__assign_resource(rsc, chosen, false, stop_if_fail);
g_list_free(nodes);
return rsc->priv->assigned_node != NULL;
}
/*!
* \internal
* \brief Apply a "this with" colocation to a node's allowed node scores
*
* \param[in,out] colocation Colocation to apply
* \param[in,out] rsc Resource being assigned
*/
static void
apply_this_with(pcmk__colocation_t *colocation, pcmk_resource_t *rsc)
{
GHashTable *archive = NULL;
pcmk_resource_t *other = colocation->primary;
// In certain cases, we will need to revert the node scores
if ((colocation->dependent_role >= pcmk_role_promoted)
|| ((colocation->score < 0)
&& (colocation->score > -PCMK_SCORE_INFINITY))) {
archive = pcmk__copy_node_table(rsc->priv->allowed_nodes);
}
if (pcmk_is_set(other->flags, pcmk__rsc_unassigned)) {
pcmk__rsc_trace(rsc,
"%s: Assigning colocation %s primary %s first"
"(score=%d role=%s)",
rsc->id, colocation->id, other->id,
colocation->score,
pcmk_role_text(colocation->dependent_role));
other->priv->cmds->assign(other, NULL, true);
}
// Apply the colocation score to this resource's allowed node scores
rsc->priv->cmds->apply_coloc_score(rsc, other, colocation, true);
if ((archive != NULL)
&& !pcmk__any_node_available(rsc->priv->allowed_nodes)) {
pcmk__rsc_info(rsc,
"%s: Reverting scores from colocation with %s "
"because no nodes allowed",
rsc->id, other->id);
g_hash_table_destroy(rsc->priv->allowed_nodes);
rsc->priv->allowed_nodes = archive;
archive = NULL;
}
if (archive != NULL) {
g_hash_table_destroy(archive);
}
}
/*!
* \internal
* \brief Update a Pacemaker Remote node once its connection has been assigned
*
* \param[in] connection Connection resource that has been assigned
*/
static void
remote_connection_assigned(const pcmk_resource_t *connection)
{
pcmk_node_t *remote_node = pcmk_find_node(connection->priv->scheduler,
connection->id);
CRM_CHECK(remote_node != NULL, return);
if ((connection->priv->assigned_node != NULL)
&& (connection->priv->next_role != pcmk_role_stopped)) {
crm_trace("Pacemaker Remote node %s will be online",
remote_node->priv->id);
remote_node->details->online = TRUE;
if (!pcmk_is_set(remote_node->priv->flags, pcmk__node_seen)) {
// Avoid unnecessary fence, since we will attempt connection
remote_node->details->unclean = FALSE;
}
} else {
crm_trace("Pacemaker Remote node %s will be shut down "
"(%sassigned connection's next role is %s)",
remote_node->priv->id,
((connection->priv->assigned_node == NULL)? "un" : ""),
pcmk_role_text(connection->priv->next_role));
remote_node->details->shutdown = TRUE;
}
}
/*!
* \internal
* \brief Assign a primitive resource to a node
*
* \param[in,out] rsc Resource to assign to a node
* \param[in] prefer Node to prefer, if all else is equal
* \param[in] stop_if_fail If \c true and \p rsc can't be assigned to a
* node, set next role to stopped and update
* existing actions
*
* \return Node that \p rsc is assigned to, if assigned entirely to one node
*
* \note If \p stop_if_fail is \c false, then \c pcmk__unassign_resource() can
* completely undo the assignment. A successful assignment can be either
* undone or left alone as final. A failed assignment has the same effect
* as calling pcmk__unassign_resource(); there are no side effects on
* roles or actions.
*/
pcmk_node_t *
pcmk__primitive_assign(pcmk_resource_t *rsc, const pcmk_node_t *prefer,
bool stop_if_fail)
{
GList *this_with_colocations = NULL;
GList *with_this_colocations = NULL;
GList *iter = NULL;
pcmk_resource_t *parent = NULL;
pcmk__colocation_t *colocation = NULL;
pcmk_scheduler_t *scheduler = NULL;
pcmk__assert(pcmk__is_primitive(rsc));
scheduler = rsc->priv->scheduler;
parent = rsc->priv->parent;
// Never assign a child without parent being assigned first
if ((parent != NULL) && !pcmk_is_set(parent->flags, pcmk__rsc_assigning)) {
pcmk__rsc_debug(rsc, "%s: Assigning parent %s first",
rsc->id, parent->id);
parent->priv->cmds->assign(parent, prefer, stop_if_fail);
}
if (!pcmk_is_set(rsc->flags, pcmk__rsc_unassigned)) {
// Assignment has already been done
const char *node_name = "no node";
if (rsc->priv->assigned_node != NULL) {
node_name = pcmk__node_name(rsc->priv->assigned_node);
}
pcmk__rsc_debug(rsc, "%s: pre-assigned to %s", rsc->id, node_name);
return rsc->priv->assigned_node;
}
// Ensure we detect assignment loops
if (pcmk_is_set(rsc->flags, pcmk__rsc_assigning)) {
pcmk__rsc_debug(rsc, "Breaking assignment loop involving %s", rsc->id);
return NULL;
}
pcmk__set_rsc_flags(rsc, pcmk__rsc_assigning);
pe__show_node_scores(true, rsc, "Pre-assignment",
rsc->priv->allowed_nodes, scheduler);
this_with_colocations = pcmk__this_with_colocations(rsc);
with_this_colocations = pcmk__with_this_colocations(rsc);
// Apply mandatory colocations first, to satisfy as many as possible
for (iter = this_with_colocations; iter != NULL; iter = iter->next) {
colocation = iter->data;
if ((colocation->score <= -PCMK_SCORE_INFINITY)
|| (colocation->score >= PCMK_SCORE_INFINITY)) {
apply_this_with(colocation, rsc);
}
}
for (iter = with_this_colocations; iter != NULL; iter = iter->next) {
colocation = iter->data;
if ((colocation->score <= -PCMK_SCORE_INFINITY)
|| (colocation->score >= PCMK_SCORE_INFINITY)) {
pcmk__add_dependent_scores(colocation, rsc);
}
}
pe__show_node_scores(true, rsc, "Mandatory-colocations",
rsc->priv->allowed_nodes, scheduler);
// Then apply optional colocations
for (iter = this_with_colocations; iter != NULL; iter = iter->next) {
colocation = iter->data;
if ((colocation->score > -PCMK_SCORE_INFINITY)
&& (colocation->score < PCMK_SCORE_INFINITY)) {
apply_this_with(colocation, rsc);
}
}
for (iter = with_this_colocations; iter != NULL; iter = iter->next) {
colocation = iter->data;
if ((colocation->score > -PCMK_SCORE_INFINITY)
&& (colocation->score < PCMK_SCORE_INFINITY)) {
pcmk__add_dependent_scores(colocation, rsc);
}
}
g_list_free(this_with_colocations);
g_list_free(with_this_colocations);
if (rsc->priv->next_role == pcmk_role_stopped) {
pcmk__rsc_trace(rsc,
"Banning %s from all nodes because it will be stopped",
rsc->id);
resource_location(rsc, NULL, -PCMK_SCORE_INFINITY,
PCMK_META_TARGET_ROLE, scheduler);
} else if ((rsc->priv->next_role > rsc->priv->orig_role)
&& !pcmk_is_set(scheduler->flags, pcmk__sched_quorate)
&& (scheduler->no_quorum_policy == pcmk_no_quorum_freeze)) {
crm_notice("Resource %s cannot be elevated from %s to %s due to "
PCMK_OPT_NO_QUORUM_POLICY "=" PCMK_VALUE_FREEZE,
rsc->id, pcmk_role_text(rsc->priv->orig_role),
pcmk_role_text(rsc->priv->next_role));
pe__set_next_role(rsc, rsc->priv->orig_role,
PCMK_OPT_NO_QUORUM_POLICY "=" PCMK_VALUE_FREEZE);
}
pe__show_node_scores(!pcmk_is_set(scheduler->flags,
pcmk__sched_output_scores),
rsc, __func__, rsc->priv->allowed_nodes, scheduler);
// Unmanage resource if fencing is enabled but no device is configured
if (pcmk_is_set(scheduler->flags, pcmk__sched_fencing_enabled)
&& !pcmk_is_set(scheduler->flags, pcmk__sched_have_fencing)) {
pcmk__clear_rsc_flags(rsc, pcmk__rsc_managed);
}
if (!pcmk_is_set(rsc->flags, pcmk__rsc_managed)) {
// Unmanaged resources stay on their current node
const char *reason = NULL;
pcmk_node_t *assign_to = NULL;
pe__set_next_role(rsc, rsc->priv->orig_role, "unmanaged");
assign_to = pcmk__current_node(rsc);
if (assign_to == NULL) {
reason = "inactive";
} else if (rsc->priv->orig_role == pcmk_role_promoted) {
reason = "promoted";
} else if (pcmk_is_set(rsc->flags, pcmk__rsc_failed)) {
reason = "failed";
} else {
reason = "active";
}
pcmk__rsc_info(rsc, "Unmanaged resource %s assigned to %s: %s", rsc->id,
(assign_to? assign_to->priv->name : "no node"),
reason);
pcmk__assign_resource(rsc, assign_to, true, stop_if_fail);
} else if (pcmk_is_set(scheduler->flags, pcmk__sched_stop_all)) {
// Must stop at some point, but be consistent with stop_if_fail
if (stop_if_fail) {
pcmk__rsc_debug(rsc,
"Forcing %s to stop: " PCMK_OPT_STOP_ALL_RESOURCES,
rsc->id);
}
pcmk__assign_resource(rsc, NULL, true, stop_if_fail);
} else if (!assign_best_node(rsc, prefer, stop_if_fail)) {
// Assignment failed
if (!pcmk_is_set(rsc->flags, pcmk__rsc_removed)) {
pcmk__rsc_info(rsc, "Resource %s cannot run anywhere", rsc->id);
} else if ((rsc->priv->active_nodes != NULL) && stop_if_fail) {
pcmk__rsc_info(rsc, "Stopping removed resource %s", rsc->id);
}
}
pcmk__clear_rsc_flags(rsc, pcmk__rsc_assigning);
if (pcmk_is_set(rsc->flags, pcmk__rsc_is_remote_connection)) {
remote_connection_assigned(rsc);
}
return rsc->priv->assigned_node;
}
/*!
* \internal
* \brief Schedule actions to bring resource down and back to current role
*
* \param[in,out] rsc Resource to restart
* \param[in,out] current Node that resource should be brought down on
* \param[in] need_stop Whether the resource must be stopped
* \param[in] need_promote Whether the resource must be promoted
*
* \return Role that resource would have after scheduled actions are taken
*/
static void
schedule_restart_actions(pcmk_resource_t *rsc, pcmk_node_t *current,
bool need_stop, bool need_promote)
{
enum rsc_role_e role = rsc->priv->orig_role;
enum rsc_role_e next_role;
rsc_transition_fn fn = NULL;
pcmk__set_rsc_flags(rsc, pcmk__rsc_restarting);
// Bring resource down to a stop on its current node
while (role != pcmk_role_stopped) {
next_role = rsc_state_matrix[role][pcmk_role_stopped];
pcmk__rsc_trace(rsc, "Creating %s action to take %s down from %s to %s",
(need_stop? "required" : "optional"), rsc->id,
pcmk_role_text(role), pcmk_role_text(next_role));
fn = rsc_action_matrix[role][next_role];
if (fn == NULL) {
break;
}
fn(rsc, current, !need_stop);
role = next_role;
}
// Bring resource up to its next role on its next node
while ((rsc->priv->orig_role <= rsc->priv->next_role)
&& (role != rsc->priv->orig_role)
&& !pcmk_is_set(rsc->flags, pcmk__rsc_blocked)) {
bool required = need_stop;
next_role = rsc_state_matrix[role][rsc->priv->orig_role];
if ((next_role == pcmk_role_promoted) && need_promote) {
required = true;
}
pcmk__rsc_trace(rsc, "Creating %s action to take %s up from %s to %s",
(required? "required" : "optional"), rsc->id,
pcmk_role_text(role), pcmk_role_text(next_role));
fn = rsc_action_matrix[role][next_role];
if (fn == NULL) {
break;
}
fn(rsc, rsc->priv->assigned_node, !required);
role = next_role;
}
pcmk__clear_rsc_flags(rsc, pcmk__rsc_restarting);
}
/*!
* \internal
* \brief If a resource's next role is not explicitly specified, set a default
*
* \param[in,out] rsc Resource to set next role for
*
* \return "explicit" if next role was explicitly set, otherwise "implicit"
*/
static const char *
set_default_next_role(pcmk_resource_t *rsc)
{
if (rsc->priv->next_role != pcmk_role_unknown) {
return "explicit";
}
if (rsc->priv->assigned_node == NULL) {
pe__set_next_role(rsc, pcmk_role_stopped, "assignment");
} else {
pe__set_next_role(rsc, pcmk_role_started, "assignment");
}
return "implicit";
}
/*!
* \internal
* \brief Create an action to represent an already pending start
*
* \param[in,out] rsc Resource to create start action for
*/
static void
create_pending_start(pcmk_resource_t *rsc)
{
pcmk_action_t *start = NULL;
pcmk__rsc_trace(rsc,
"Creating action for %s to represent already pending start",
rsc->id);
start = start_action(rsc, rsc->priv->assigned_node, TRUE);
pcmk__set_action_flags(start, pcmk__action_always_in_graph);
}
/*!
* \internal
* \brief Schedule actions needed to take a resource to its next role
*
* \param[in,out] rsc Resource to schedule actions for
*/
static void
schedule_role_transition_actions(pcmk_resource_t *rsc)
{
enum rsc_role_e role = rsc->priv->orig_role;
while (role != rsc->priv->next_role) {
enum rsc_role_e next_role =
rsc_state_matrix[role][rsc->priv->next_role];
rsc_transition_fn fn = NULL;
pcmk__rsc_trace(rsc,
"Creating action to take %s from %s to %s "
"(ending at %s)",
rsc->id, pcmk_role_text(role),
pcmk_role_text(next_role),
pcmk_role_text(rsc->priv->next_role));
fn = rsc_action_matrix[role][next_role];
if (fn == NULL) {
break;
}
fn(rsc, rsc->priv->assigned_node, false);
role = next_role;
}
}
/*!
* \internal
* \brief Create all actions needed for a given primitive resource
*
* \param[in,out] rsc Primitive resource to create actions for
*/
void
pcmk__primitive_create_actions(pcmk_resource_t *rsc)
{
bool need_stop = false;
bool need_promote = false;
bool is_moving = false;
bool allow_migrate = false;
bool multiply_active = false;
pcmk_node_t *current = NULL;
pcmk_node_t *migration_target = NULL;
unsigned int num_all_active = 0;
unsigned int num_clean_active = 0;
const char *next_role_source = NULL;
pcmk__assert(pcmk__is_primitive(rsc));
next_role_source = set_default_next_role(rsc);
pcmk__rsc_trace(rsc,
"Creating all actions for %s transition from %s to %s "
"(%s) on %s",
rsc->id, pcmk_role_text(rsc->priv->orig_role),
pcmk_role_text(rsc->priv->next_role), next_role_source,
pcmk__node_name(rsc->priv->assigned_node));
current = rsc->priv->fns->active_node(rsc, &num_all_active,
&num_clean_active);
g_list_foreach(rsc->priv->dangling_migration_sources,
pcmk__abort_dangling_migration, rsc);
if ((current != NULL) && (rsc->priv->assigned_node != NULL)
&& !pcmk__same_node(current, rsc->priv->assigned_node)
&& (rsc->priv->next_role >= pcmk_role_started)) {
pcmk__rsc_trace(rsc, "Moving %s from %s to %s",
rsc->id, pcmk__node_name(current),
pcmk__node_name(rsc->priv->assigned_node));
is_moving = true;
allow_migrate = pcmk__rsc_can_migrate(rsc, current);
// This is needed even if migrating (though I'm not sure why ...)
need_stop = true;
}
// Check whether resource is partially migrated and/or multiply active
migration_target = rsc->priv->partial_migration_target;
if ((rsc->priv->partial_migration_source != NULL)
&& (migration_target != NULL) && allow_migrate && (num_all_active == 2)
&& pcmk__same_node(current, rsc->priv->partial_migration_source)
&& pcmk__same_node(rsc->priv->assigned_node, migration_target)) {
/* A partial migration is in progress, and the migration target remains
* the same as when the migration began.
*/
pcmk__rsc_trace(rsc,
"Partial migration of %s from %s to %s will continue",
rsc->id,
pcmk__node_name(rsc->priv->partial_migration_source),
pcmk__node_name(migration_target));
} else if ((rsc->priv->partial_migration_source != NULL)
|| (migration_target != NULL)) {
// A partial migration is in progress but can't be continued
if (num_all_active > 2) {
// The resource is migrating *and* multiply active!
crm_notice("Forcing recovery of %s because it is migrating "
"from %s to %s and possibly active elsewhere",
rsc->id,
pcmk__node_name(rsc->priv->partial_migration_source),
pcmk__node_name(migration_target));
} else {
// The migration source or target isn't available
crm_notice("Forcing recovery of %s because it can no longer "
"migrate from %s to %s",
rsc->id,
pcmk__node_name(rsc->priv->partial_migration_source),
pcmk__node_name(migration_target));
}
need_stop = true;
rsc->priv->partial_migration_source = NULL;
rsc->priv->partial_migration_target = NULL;
allow_migrate = false;
} else if (pcmk_is_set(rsc->flags, pcmk__rsc_needs_fencing)) {
multiply_active = (num_all_active > 1);
} else {
/* If a resource has PCMK_META_REQUIRES set to PCMK_VALUE_NOTHING or
* PCMK_VALUE_QUORUM, don't consider it active on unclean nodes (similar
* to how all resources behave when PCMK_OPT_STONITH_ENABLED is false).
* We can start such resources elsewhere before fencing completes, and
* if we considered the resource active on the failed node, we would
* attempt recovery for being active on multiple nodes.
*/
multiply_active = (num_clean_active > 1);
}
if (multiply_active) {
const char *class = crm_element_value(rsc->priv->xml, PCMK_XA_CLASS);
// Resource was (possibly) incorrectly multiply active
pcmk__sched_err(rsc->priv->scheduler,
"%s resource %s might be active on %u nodes (%s)",
pcmk__s(class, "Untyped"), rsc->id, num_all_active,
pcmk__multiply_active_text(rsc));
crm_notice("For more information, see \"What are multiply active "
"resources?\" at "
"https://projects.clusterlabs.org/w/clusterlabs/faq/");
switch (rsc->priv->multiply_active_policy) {
case pcmk__multiply_active_restart:
need_stop = true;
break;
case pcmk__multiply_active_unexpected:
need_stop = true; // stop_resource() will skip expected node
pcmk__set_rsc_flags(rsc, pcmk__rsc_stop_unexpected);
break;
default:
break;
}
} else {
pcmk__clear_rsc_flags(rsc, pcmk__rsc_stop_unexpected);
}
if (pcmk_is_set(rsc->flags, pcmk__rsc_start_pending)) {
create_pending_start(rsc);
}
if (is_moving) {
// Remaining tests are only for resources staying where they are
} else if (pcmk_is_set(rsc->flags, pcmk__rsc_failed)) {
if (pcmk_is_set(rsc->flags, pcmk__rsc_stop_if_failed)) {
need_stop = true;
pcmk__rsc_trace(rsc, "Recovering %s", rsc->id);
} else {
pcmk__rsc_trace(rsc, "Recovering %s by demotion", rsc->id);
if (rsc->priv->next_role == pcmk_role_promoted) {
need_promote = true;
}
}
} else if (pcmk_is_set(rsc->flags, pcmk__rsc_blocked)) {
pcmk__rsc_trace(rsc, "Blocking further actions on %s", rsc->id);
need_stop = true;
} else if ((rsc->priv->orig_role > pcmk_role_started)
&& (current != NULL)
&& (rsc->priv->assigned_node != NULL)) {
pcmk_action_t *start = NULL;
pcmk__rsc_trace(rsc, "Creating start action for promoted resource %s",
rsc->id);
start = start_action(rsc, rsc->priv->assigned_node, TRUE);
if (!pcmk_is_set(start->flags, pcmk__action_optional)) {
// Recovery of a promoted resource
pcmk__rsc_trace(rsc, "%s restart is required for recovery", rsc->id);
need_stop = true;
}
}
// Create any actions needed to bring resource down and back up to same role
schedule_restart_actions(rsc, current, need_stop, need_promote);
// Create any actions needed to take resource from this role to the next
schedule_role_transition_actions(rsc);
pcmk__create_recurring_actions(rsc);
if (allow_migrate) {
pcmk__create_migration_actions(rsc, current);
}
}
/*!
* \internal
* \brief Ban a resource from any allowed nodes that are Pacemaker Remote nodes
*
* \param[in] rsc Resource to check
*/
static void
rsc_avoids_remote_nodes(const pcmk_resource_t *rsc)
{
GHashTableIter iter;
pcmk_node_t *node = NULL;
g_hash_table_iter_init(&iter, rsc->priv->allowed_nodes);
while (g_hash_table_iter_next(&iter, NULL, (void **) &node)) {
if (node->priv->remote != NULL) {
node->assign->score = -PCMK_SCORE_INFINITY;
}
}
}
/*!
* \internal
* \brief Return allowed nodes as (possibly sorted) list
*
* Convert a resource's hash table of allowed nodes to a list. If printing to
* stdout, sort the list, to keep action ID numbers consistent for regression
* test output (while avoiding the performance hit on a live cluster).
*
* \param[in] rsc Resource to check for allowed nodes
*
* \return List of resource's allowed nodes
* \note Callers should take care not to rely on the list being sorted.
*/
static GList *
allowed_nodes_as_list(const pcmk_resource_t *rsc)
{
GList *allowed_nodes = NULL;
if (rsc->priv->allowed_nodes != NULL) {
allowed_nodes = g_hash_table_get_values(rsc->priv->allowed_nodes);
}
if (!pcmk__is_daemon) {
allowed_nodes = g_list_sort(allowed_nodes, pe__cmp_node_name);
}
return allowed_nodes;
}
/*!
* \internal
* \brief Create implicit constraints needed for a primitive resource
*
* \param[in,out] rsc Primitive resource to create implicit constraints for
*/
void
pcmk__primitive_internal_constraints(pcmk_resource_t *rsc)
{
GList *allowed_nodes = NULL;
bool check_unfencing = false;
bool check_utilization = false;
pcmk_scheduler_t *scheduler = NULL;
pcmk__assert(pcmk__is_primitive(rsc));
scheduler = rsc->priv->scheduler;
if (!pcmk_is_set(rsc->flags, pcmk__rsc_managed)) {
pcmk__rsc_trace(rsc,
"Skipping implicit constraints for unmanaged resource "
"%s", rsc->id);
return;
}
// Whether resource requires unfencing
check_unfencing = !pcmk_is_set(rsc->flags, pcmk__rsc_fence_device)
&& pcmk_is_set(scheduler->flags,
pcmk__sched_enable_unfencing)
&& pcmk_is_set(rsc->flags, pcmk__rsc_needs_unfencing);
// Whether a non-default placement strategy is used
check_utilization = (g_hash_table_size(rsc->priv->utilization) > 0)
&& !pcmk__str_eq(scheduler->priv->placement_strategy,
PCMK_VALUE_DEFAULT, pcmk__str_casei);
// Order stops before starts (i.e. restart)
pcmk__new_ordering(rsc, pcmk__op_key(rsc->id, PCMK_ACTION_STOP, 0), NULL,
rsc, pcmk__op_key(rsc->id, PCMK_ACTION_START, 0), NULL,
pcmk__ar_ordered
|pcmk__ar_first_implies_then
|pcmk__ar_intermediate_stop, scheduler);
// Promotable ordering: demote before stop, start before promote
if (pcmk_is_set(pe__const_top_resource(rsc, false)->flags,
pcmk__rsc_promotable)
|| (rsc->priv->orig_role > pcmk_role_unpromoted)) {
pcmk__new_ordering(rsc, pcmk__op_key(rsc->id, PCMK_ACTION_DEMOTE, 0),
NULL,
rsc, pcmk__op_key(rsc->id, PCMK_ACTION_STOP, 0),
NULL,
pcmk__ar_promoted_then_implies_first, scheduler);
pcmk__new_ordering(rsc, pcmk__op_key(rsc->id, PCMK_ACTION_START, 0),
NULL,
rsc, pcmk__op_key(rsc->id, PCMK_ACTION_PROMOTE, 0),
NULL,
pcmk__ar_unrunnable_first_blocks, scheduler);
}
// Don't clear resource history if probing on same node
pcmk__new_ordering(rsc, pcmk__op_key(rsc->id, PCMK_ACTION_LRM_DELETE, 0),
NULL, rsc,
pcmk__op_key(rsc->id, PCMK_ACTION_MONITOR, 0),
NULL,
pcmk__ar_if_on_same_node|pcmk__ar_then_cancels_first,
scheduler);
// Certain checks need allowed nodes
if (check_unfencing || check_utilization
|| (rsc->priv->launcher != NULL)) {
allowed_nodes = allowed_nodes_as_list(rsc);
}
if (check_unfencing) {
g_list_foreach(allowed_nodes, pcmk__order_restart_vs_unfence, rsc);
}
if (check_utilization) {
pcmk__create_utilization_constraints(rsc, allowed_nodes);
}
if (rsc->priv->launcher != NULL) {
pcmk_resource_t *remote_rsc = NULL;
if (pcmk_is_set(rsc->flags, pcmk__rsc_is_remote_connection)) {
// rsc is the implicit remote connection for a guest or bundle node
/* Guest resources are not allowed to run on Pacemaker Remote nodes,
* to avoid nesting remotes. However, bundles are allowed.
*/
if (!pcmk_is_set(rsc->flags, pcmk__rsc_remote_nesting_allowed)) {
rsc_avoids_remote_nodes(rsc->priv->launcher);
}
/* If someone cleans up a guest or bundle node's launcher, we will
* likely schedule a (re-)probe of the launcher and recovery of the
* connection. Order the connection stop after the launcher probe,
* so that if we detect the launcher running, we will trigger a new
* transition and avoid the unnecessary recovery.
*/
pcmk__order_resource_actions(rsc->priv->launcher,
PCMK_ACTION_MONITOR,
rsc, PCMK_ACTION_STOP,
pcmk__ar_ordered);
/* A user can specify that a resource must start on a Pacemaker Remote
* node by explicitly configuring it with the PCMK__META_CONTAINER
* meta-attribute. This is of questionable merit, since location
* constraints can accomplish the same thing. But we support it, so here
* we check whether a resource (that is not itself a remote connection)
* has PCMK__META_CONTAINER set to a remote node or guest node resource.
*/
} else if (pcmk_is_set(rsc->priv->launcher->flags,
pcmk__rsc_is_remote_connection)) {
remote_rsc = rsc->priv->launcher;
} else {
remote_rsc =
pe__resource_contains_guest_node(scheduler,
rsc->priv->launcher);
}
if (remote_rsc != NULL) {
/* Force the resource on the Pacemaker Remote node instead of
* colocating the resource with the launcher.
*/
for (GList *item = allowed_nodes; item; item = item->next) {
pcmk_node_t *node = item->data;
if (node->priv->remote != remote_rsc) {
node->assign->score = -PCMK_SCORE_INFINITY;
}
}
} else {
/* This resource is either launched by a resource that does NOT
* represent a Pacemaker Remote node, or a Pacemaker Remote
* connection resource for a guest node or bundle.
*/
int score;
crm_trace("Order and colocate %s relative to its launcher %s",
rsc->id, rsc->priv->launcher->id);
pcmk__new_ordering(rsc->priv->launcher,
pcmk__op_key(rsc->priv->launcher->id,
PCMK_ACTION_START, 0),
NULL, rsc,
pcmk__op_key(rsc->id, PCMK_ACTION_START, 0),
NULL,
pcmk__ar_first_implies_then
|pcmk__ar_unrunnable_first_blocks, scheduler);
pcmk__new_ordering(rsc,
pcmk__op_key(rsc->id, PCMK_ACTION_STOP, 0),
NULL,
rsc->priv->launcher,
pcmk__op_key(rsc->priv->launcher->id,
PCMK_ACTION_STOP, 0),
NULL, pcmk__ar_then_implies_first, scheduler);
- if (pcmk_is_set(rsc->flags, pcmk__rsc_remote_nesting_allowed)) {
+ if (pcmk_is_set(rsc->flags, pcmk__rsc_remote_nesting_allowed)
+ /* @TODO: && non-bundle Pacemaker Remote nodes exist */) {
score = 10000; /* Highly preferred but not essential */
} else {
score = PCMK_SCORE_INFINITY; // Force to run on same host
}
pcmk__new_colocation("#resource-with-container", NULL, score, rsc,
rsc->priv->launcher, NULL, NULL,
pcmk__coloc_influence);
}
}
if (pcmk_is_set(rsc->flags, pcmk__rsc_is_remote_connection)
|| pcmk_is_set(rsc->flags, pcmk__rsc_fence_device)) {
/* Remote connections and fencing devices are not allowed to run on
* Pacemaker Remote nodes
*/
rsc_avoids_remote_nodes(rsc);
}
g_list_free(allowed_nodes);
}
/*!
* \internal
* \brief Apply a colocation's score to node scores or resource priority
*
* Given a colocation constraint, apply its score to the dependent's
* allowed node scores (if we are still placing resources) or priority (if
* we are choosing promotable clone instance roles).
*
* \param[in,out] dependent Dependent resource in colocation
* \param[in] primary Primary resource in colocation
* \param[in] colocation Colocation constraint to apply
* \param[in] for_dependent true if called on behalf of dependent
*
* \return The score added to the dependent's priority
*/
int
pcmk__primitive_apply_coloc_score(pcmk_resource_t *dependent,
const pcmk_resource_t *primary,
const pcmk__colocation_t *colocation,
bool for_dependent)
{
enum pcmk__coloc_affects filter_results;
pcmk__assert((dependent != NULL) && (primary != NULL)
&& (colocation != NULL));
if (for_dependent) {
// Always process on behalf of primary resource
return primary->priv->cmds->apply_coloc_score(dependent, primary,
colocation, false);
}
filter_results = pcmk__colocation_affects(dependent, primary, colocation,
false);
pcmk__rsc_trace(dependent, "%s %s with %s (%s, score=%d, filter=%d)",
((colocation->score > 0)? "Colocating" : "Anti-colocating"),
dependent->id, primary->id, colocation->id,
colocation->score,
filter_results);
switch (filter_results) {
case pcmk__coloc_affects_role:
return pcmk__apply_coloc_to_priority(dependent, primary,
colocation);
case pcmk__coloc_affects_location:
pcmk__apply_coloc_to_scores(dependent, primary, colocation);
return 0;
default: // pcmk__coloc_affects_nothing
return 0;
}
}
/* Primitive implementation of
* pcmk__assignment_methods_t:with_this_colocations()
*/
void
pcmk__with_primitive_colocations(const pcmk_resource_t *rsc,
const pcmk_resource_t *orig_rsc, GList **list)
{
const pcmk_resource_t *parent = NULL;
pcmk__assert(pcmk__is_primitive(rsc) && (list != NULL));
parent = rsc->priv->parent;
if (rsc == orig_rsc) {
/* For the resource itself, add all of its own colocations and relevant
* colocations from its parent (if any).
*/
pcmk__add_with_this_list(list, rsc->priv->with_this_colocations,
orig_rsc);
if (parent != NULL) {
parent->priv->cmds->with_this_colocations(parent, orig_rsc, list);
}
} else {
// For an ancestor, add only explicitly configured constraints
for (GList *iter = rsc->priv->with_this_colocations;
iter != NULL; iter = iter->next) {
pcmk__colocation_t *colocation = iter->data;
if (pcmk_is_set(colocation->flags, pcmk__coloc_explicit)) {
pcmk__add_with_this(list, colocation, orig_rsc);
}
}
}
}
/* Primitive implementation of
* pcmk__assignment_methods_t:this_with_colocations()
*/
void
pcmk__primitive_with_colocations(const pcmk_resource_t *rsc,
const pcmk_resource_t *orig_rsc, GList **list)
{
const pcmk_resource_t *parent = NULL;
pcmk__assert(pcmk__is_primitive(rsc) && (list != NULL));
parent = rsc->priv->parent;
if (rsc == orig_rsc) {
/* For the resource itself, add all of its own colocations and relevant
* colocations from its parent (if any).
*/
pcmk__add_this_with_list(list, rsc->priv->this_with_colocations,
orig_rsc);
if (parent != NULL) {
parent->priv->cmds->this_with_colocations(parent, orig_rsc, list);
}
} else {
// For an ancestor, add only explicitly configured constraints
for (GList *iter = rsc->priv->this_with_colocations;
iter != NULL; iter = iter->next) {
pcmk__colocation_t *colocation = iter->data;
if (pcmk_is_set(colocation->flags, pcmk__coloc_explicit)) {
pcmk__add_this_with(list, colocation, orig_rsc);
}
}
}
}
/*!
* \internal
* \brief Return action flags for a given primitive resource action
*
* \param[in,out] action Action to get flags for
* \param[in] node If not NULL, limit effects to this node (ignored)
*
* \return Flags appropriate to \p action on \p node
*/
uint32_t
pcmk__primitive_action_flags(pcmk_action_t *action, const pcmk_node_t *node)
{
pcmk__assert(action != NULL);
return (uint32_t) action->flags;
}
/*!
* \internal
* \brief Check whether a node is a multiply active resource's expected node
*
* \param[in] rsc Resource to check
* \param[in] node Node to check
*
* \return \c true if \p rsc is multiply active with
* \c PCMK_META_MULTIPLE_ACTIVE set to \c PCMK_VALUE_STOP_UNEXPECTED,
* and \p node is the node where it will remain active
* \note This assumes that the resource's next role cannot be changed to stopped
* after this is called, which should be reasonable if status has already
* been unpacked and resources have been assigned to nodes.
*/
static bool
is_expected_node(const pcmk_resource_t *rsc, const pcmk_node_t *node)
{
return pcmk_all_flags_set(rsc->flags,
pcmk__rsc_stop_unexpected|pcmk__rsc_restarting)
&& (rsc->priv->next_role > pcmk_role_stopped)
&& pcmk__same_node(rsc->priv->assigned_node, node);
}
/*!
* \internal
* \brief Schedule actions needed to stop a resource wherever it is active
*
* \param[in,out] rsc Resource being stopped
* \param[in] node Node where resource is being stopped (ignored)
* \param[in] optional Whether actions should be optional
*/
static void
stop_resource(pcmk_resource_t *rsc, pcmk_node_t *node, bool optional)
{
for (GList *iter = rsc->priv->active_nodes;
iter != NULL; iter = iter->next) {
pcmk_node_t *current = (pcmk_node_t *) iter->data;
pcmk_action_t *stop = NULL;
if (is_expected_node(rsc, current)) {
/* We are scheduling restart actions for a multiply active resource
* with PCMK_META_MULTIPLE_ACTIVE=PCMK_VALUE_STOP_UNEXPECTED, and
* this is where it should not be stopped.
*/
pcmk__rsc_trace(rsc,
"Skipping stop of multiply active resource %s "
"on expected node %s",
rsc->id, pcmk__node_name(current));
continue;
}
if (rsc->priv->partial_migration_target != NULL) {
// Continue migration if node originally was and remains target
if (pcmk__same_node(current, rsc->priv->partial_migration_target)
&& pcmk__same_node(current, rsc->priv->assigned_node)) {
pcmk__rsc_trace(rsc,
"Skipping stop of %s on %s "
"because partial migration there will continue",
rsc->id, pcmk__node_name(current));
continue;
} else {
pcmk__rsc_trace(rsc,
"Forcing stop of %s on %s "
"because migration target changed",
rsc->id, pcmk__node_name(current));
optional = false;
}
}
pcmk__rsc_trace(rsc, "Scheduling stop of %s on %s",
rsc->id, pcmk__node_name(current));
stop = stop_action(rsc, current, optional);
if (rsc->priv->assigned_node == NULL) {
pe_action_set_reason(stop, "node availability", true);
} else if (pcmk_all_flags_set(rsc->flags, pcmk__rsc_restarting
|pcmk__rsc_stop_unexpected)) {
/* We are stopping a multiply active resource on a node that is
* not its expected node, and we are still scheduling restart
* actions, so the stop is for being multiply active.
*/
pe_action_set_reason(stop, "being multiply active", true);
}
if (!pcmk_is_set(rsc->flags, pcmk__rsc_managed)) {
pcmk__clear_action_flags(stop, pcmk__action_runnable);
}
if (pcmk_is_set(rsc->flags, pcmk__rsc_needs_unfencing)) {
pcmk_action_t *unfence = pe_fence_op(current, PCMK_ACTION_ON, true,
NULL, false,
rsc->priv->scheduler);
order_actions(stop, unfence, pcmk__ar_then_implies_first);
if (!pcmk__node_unfenced(current)) {
pcmk__sched_err(rsc->priv->scheduler,
"Stopping %s until %s can be unfenced",
rsc->id, pcmk__node_name(current));
}
}
}
}
/*!
* \internal
* \brief Schedule actions needed to start a resource on a node
*
* \param[in,out] rsc Resource being started
* \param[in,out] node Node where resource should be started
* \param[in] optional Whether actions should be optional
*/
static void
start_resource(pcmk_resource_t *rsc, pcmk_node_t *node, bool optional)
{
pcmk_action_t *start = NULL;
pcmk__assert(node != NULL);
pcmk__rsc_trace(rsc, "Scheduling %s start of %s on %s (score %d)",
(optional? "optional" : "required"), rsc->id,
pcmk__node_name(node), node->assign->score);
start = start_action(rsc, node, TRUE);
pcmk__order_vs_unfence(rsc, node, start, pcmk__ar_first_implies_then);
if (pcmk_is_set(start->flags, pcmk__action_runnable) && !optional) {
pcmk__clear_action_flags(start, pcmk__action_optional);
}
if (is_expected_node(rsc, node)) {
/* This could be a problem if the start becomes necessary for other
* reasons later.
*/
pcmk__rsc_trace(rsc,
"Start of multiply active resouce %s "
"on expected node %s will be a pseudo-action",
rsc->id, pcmk__node_name(node));
pcmk__set_action_flags(start, pcmk__action_pseudo);
}
}
/*!
* \internal
* \brief Schedule actions needed to promote a resource on a node
*
* \param[in,out] rsc Resource being promoted
* \param[in] node Node where resource should be promoted
* \param[in] optional Whether actions should be optional
*/
static void
promote_resource(pcmk_resource_t *rsc, pcmk_node_t *node, bool optional)
{
GList *iter = NULL;
GList *action_list = NULL;
bool runnable = true;
pcmk__assert(node != NULL);
// Any start must be runnable for promotion to be runnable
action_list = pe__resource_actions(rsc, node, PCMK_ACTION_START, true);
for (iter = action_list; iter != NULL; iter = iter->next) {
pcmk_action_t *start = (pcmk_action_t *) iter->data;
if (!pcmk_is_set(start->flags, pcmk__action_runnable)) {
runnable = false;
}
}
g_list_free(action_list);
if (runnable) {
pcmk_action_t *promote = promote_action(rsc, node, optional);
pcmk__rsc_trace(rsc, "Scheduling %s promotion of %s on %s",
(optional? "optional" : "required"), rsc->id,
pcmk__node_name(node));
if (is_expected_node(rsc, node)) {
/* This could be a problem if the promote becomes necessary for
* other reasons later.
*/
pcmk__rsc_trace(rsc,
"Promotion of multiply active resouce %s "
"on expected node %s will be a pseudo-action",
rsc->id, pcmk__node_name(node));
pcmk__set_action_flags(promote, pcmk__action_pseudo);
}
} else {
pcmk__rsc_trace(rsc, "Not promoting %s on %s: start unrunnable",
rsc->id, pcmk__node_name(node));
action_list = pe__resource_actions(rsc, node, PCMK_ACTION_PROMOTE,
true);
for (iter = action_list; iter != NULL; iter = iter->next) {
pcmk_action_t *promote = (pcmk_action_t *) iter->data;
pcmk__clear_action_flags(promote, pcmk__action_runnable);
}
g_list_free(action_list);
}
}
/*!
* \internal
* \brief Schedule actions needed to demote a resource wherever it is active
*
* \param[in,out] rsc Resource being demoted
* \param[in] node Node where resource should be demoted (ignored)
* \param[in] optional Whether actions should be optional
*/
static void
demote_resource(pcmk_resource_t *rsc, pcmk_node_t *node, bool optional)
{
/* Since this will only be called for a primitive (possibly as an instance
* of a collective resource), the resource is multiply active if it is
* running on more than one node, so we want to demote on all of them as
* part of recovery, regardless of which one is the desired node.
*/
for (GList *iter = rsc->priv->active_nodes;
iter != NULL; iter = iter->next) {
pcmk_node_t *current = (pcmk_node_t *) iter->data;
if (is_expected_node(rsc, current)) {
pcmk__rsc_trace(rsc,
"Skipping demote of multiply active resource %s "
"on expected node %s",
rsc->id, pcmk__node_name(current));
} else {
pcmk__rsc_trace(rsc, "Scheduling %s demotion of %s on %s",
(optional? "optional" : "required"), rsc->id,
pcmk__node_name(current));
demote_action(rsc, current, optional);
}
}
}
static void
assert_role_error(pcmk_resource_t *rsc, pcmk_node_t *node, bool optional)
{
pcmk__assert(false);
}
/*!
* \internal
* \brief Schedule cleanup of a resource
*
* \param[in,out] rsc Resource to clean up
* \param[in] node Node to clean up on
* \param[in] optional Whether clean-up should be optional
*/
void
pcmk__schedule_cleanup(pcmk_resource_t *rsc, const pcmk_node_t *node,
bool optional)
{
/* If the cleanup is required, its orderings are optional, because they're
* relevant only if both actions are required. Conversely, if the cleanup is
* optional, the orderings make the then action required if the first action
* becomes required.
*/
uint32_t flag = optional? pcmk__ar_first_implies_then : pcmk__ar_ordered;
CRM_CHECK((rsc != NULL) && (node != NULL), return);
if (pcmk_is_set(rsc->flags, pcmk__rsc_failed)) {
pcmk__rsc_trace(rsc, "Skipping clean-up of %s on %s: resource failed",
rsc->id, pcmk__node_name(node));
return;
}
if (node->details->unclean || !node->details->online) {
pcmk__rsc_trace(rsc, "Skipping clean-up of %s on %s: node unavailable",
rsc->id, pcmk__node_name(node));
return;
}
crm_notice("Scheduling clean-up of %s on %s",
rsc->id, pcmk__node_name(node));
delete_action(rsc, node, optional);
// stop -> clean-up -> start
pcmk__order_resource_actions(rsc, PCMK_ACTION_STOP,
rsc, PCMK_ACTION_DELETE, flag);
pcmk__order_resource_actions(rsc, PCMK_ACTION_DELETE,
rsc, PCMK_ACTION_START, flag);
}
/*!
* \internal
* \brief Add primitive meta-attributes relevant to graph actions to XML
*
* \param[in] rsc Primitive resource whose meta-attributes should be added
* \param[in,out] xml Transition graph action attributes XML to add to
*/
void
pcmk__primitive_add_graph_meta(const pcmk_resource_t *rsc, xmlNode *xml)
{
char *name = NULL;
char *value = NULL;
const pcmk_resource_t *parent = NULL;
pcmk__assert(pcmk__is_primitive(rsc) && (xml != NULL));
/* Clone instance numbers get set internally as meta-attributes, and are
* needed in the transition graph (for example, to tell unique clone
* instances apart).
*/
value = g_hash_table_lookup(rsc->priv->meta, PCMK__META_CLONE);
if (value != NULL) {
name = crm_meta_name(PCMK__META_CLONE);
crm_xml_add(xml, name, value);
free(name);
}
// Not sure if this one is really needed ...
value = g_hash_table_lookup(rsc->priv->meta, PCMK_META_REMOTE_NODE);
if (value != NULL) {
name = crm_meta_name(PCMK_META_REMOTE_NODE);
crm_xml_add(xml, name, value);
free(name);
}
/* The PCMK__META_CONTAINER meta-attribute can be set on the primitive
* itself or one of its ancestors, so check them all and keep the highest.
*/
for (parent = rsc; parent != NULL; parent = parent->priv->parent) {
if (parent->priv->launcher != NULL) {
crm_xml_add(xml, CRM_META "_" PCMK__META_CONTAINER,
parent->priv->launcher->id);
}
}
/* Bundle replica children will get their external-ip set internally as a
* meta-attribute. The graph action needs it, but under a different naming
* convention than other meta-attributes.
*/
value = g_hash_table_lookup(rsc->priv->meta, "external-ip");
if (value != NULL) {
crm_xml_add(xml, "pcmk_external_ip", value);
}
}
// Primitive implementation of pcmk__assignment_methods_t:add_utilization()
void
pcmk__primitive_add_utilization(const pcmk_resource_t *rsc,
const pcmk_resource_t *orig_rsc,
GList *all_rscs, GHashTable *utilization)
{
pcmk__assert(pcmk__is_primitive(rsc) && (orig_rsc != NULL)
&& (utilization != NULL));
if (!pcmk_is_set(rsc->flags, pcmk__rsc_unassigned)) {
return;
}
pcmk__rsc_trace(orig_rsc,
"%s: Adding primitive %s as colocated utilization",
orig_rsc->id, rsc->id);
pcmk__release_node_capacity(utilization, rsc);
}
/*!
* \internal
* \brief Get epoch time of node's shutdown attribute (or now if none)
*
* \param[in,out] node Node to check
*
* \return Epoch time corresponding to shutdown attribute if set or now if not
*/
static time_t
shutdown_time(pcmk_node_t *node)
{
const char *shutdown = pcmk__node_attr(node, PCMK__NODE_ATTR_SHUTDOWN, NULL,
pcmk__rsc_node_current);
time_t result = 0;
if (shutdown != NULL) {
long long result_ll;
int rc = pcmk__scan_ll(shutdown, &result_ll, 0LL);
if (rc == pcmk_rc_ok) {
result = (time_t) result_ll;
} else {
crm_warn("Ignoring invalid value '%s' for %s "
PCMK__NODE_ATTR_SHUTDOWN " attribute: %s",
shutdown, pcmk__node_name(node), pcmk_rc_str(rc));
}
}
return (result == 0)? get_effective_time(node->priv->scheduler) : result;
}
/*!
* \internal
* \brief Ban a resource from a node if it's not locked to the node
*
* \param[in] data Node to check
* \param[in,out] user_data Resource to check
*/
static void
ban_if_not_locked(gpointer data, gpointer user_data)
{
const pcmk_node_t *node = (const pcmk_node_t *) data;
pcmk_resource_t *rsc = (pcmk_resource_t *) user_data;
if (!pcmk__same_node(node, rsc->priv->lock_node)) {
resource_location(rsc, node, -PCMK_SCORE_INFINITY,
PCMK_OPT_SHUTDOWN_LOCK, rsc->priv->scheduler);
}
}
// Primitive implementation of pcmk__assignment_methods_t:shutdown_lock()
void
pcmk__primitive_shutdown_lock(pcmk_resource_t *rsc)
{
pcmk_scheduler_t *scheduler = NULL;
pcmk__assert(pcmk__is_primitive(rsc));
scheduler = rsc->priv->scheduler;
// Fence devices and remote connections can't be locked
if (pcmk_any_flags_set(rsc->flags, pcmk__rsc_fence_device
|pcmk__rsc_is_remote_connection)) {
return;
}
if (rsc->priv->lock_node != NULL) {
// The lock was obtained from resource history
if (rsc->priv->active_nodes != NULL) {
/* The resource was started elsewhere even though it is now
* considered locked. This shouldn't be possible, but as a
* failsafe, we don't want to disturb the resource now.
*/
pcmk__rsc_info(rsc,
"Cancelling shutdown lock "
"because %s is already active", rsc->id);
pe__clear_resource_history(rsc, rsc->priv->lock_node);
rsc->priv->lock_node = NULL;
rsc->priv->lock_time = 0;
}
// Only a resource active on exactly one node can be locked
} else if (pcmk__list_of_1(rsc->priv->active_nodes)) {
pcmk_node_t *node = rsc->priv->active_nodes->data;
if (node->details->shutdown) {
if (node->details->unclean) {
pcmk__rsc_debug(rsc,
"Not locking %s to unclean %s for shutdown",
rsc->id, pcmk__node_name(node));
} else {
rsc->priv->lock_node = node;
rsc->priv->lock_time = shutdown_time(node);
}
}
}
if (rsc->priv->lock_node == NULL) {
// No lock needed
return;
}
if (scheduler->priv->shutdown_lock_ms > 0U) {
time_t lock_expiration = rsc->priv->lock_time
+ pcmk__timeout_ms2s(scheduler->priv->shutdown_lock_ms);
pcmk__rsc_info(rsc, "Locking %s to %s due to shutdown (expires @%lld)",
rsc->id, pcmk__node_name(rsc->priv->lock_node),
(long long) lock_expiration);
pe__update_recheck_time(++lock_expiration, scheduler,
"shutdown lock expiration");
} else {
pcmk__rsc_info(rsc, "Locking %s to %s due to shutdown",
rsc->id, pcmk__node_name(rsc->priv->lock_node));
}
// If resource is locked to one node, ban it from all other nodes
g_list_foreach(scheduler->nodes, ban_if_not_locked, rsc);
}
diff --git a/lib/pacemaker/pcmk_scheduler.c b/lib/pacemaker/pcmk_scheduler.c
index 8960c298f5..de30f4d8f1 100644
--- a/lib/pacemaker/pcmk_scheduler.c
+++ b/lib/pacemaker/pcmk_scheduler.c
@@ -1,880 +1,883 @@
/*
* Copyright 2004-2024 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU General Public License version 2
* or later (GPLv2+) WITHOUT ANY WARRANTY.
*/
#include <crm_internal.h>
#include <crm/crm.h>
#include <crm/cib.h>
#include <crm/cib/internal.h>
#include <crm/common/xml.h>
#include <crm/common/xml_internal.h>
#include <crm/common/scheduler_internal.h>
#include <glib.h>
#include <crm/pengine/status.h>
#include <pacemaker-internal.h>
#include "libpacemaker_private.h"
CRM_TRACE_INIT_DATA(pacemaker);
/*!
* \internal
* \brief Do deferred action checks after assignment
*
* When unpacking the resource history, the scheduler checks for resource
* configurations that have changed since an action was run. However, at that
* time, bundles using the REMOTE_CONTAINER_HACK don't have their final
* parameter information, so instead they add a deferred check to a list. This
* function processes one entry in that list.
*
* \param[in,out] rsc Resource that action history is for
* \param[in,out] node Node that action history is for
* \param[in] rsc_op Action history entry
* \param[in] check Type of deferred check to do
*/
static void
check_params(pcmk_resource_t *rsc, pcmk_node_t *node, const xmlNode *rsc_op,
enum pcmk__check_parameters check)
{
const char *reason = NULL;
pcmk__op_digest_t *digest_data = NULL;
switch (check) {
case pcmk__check_active:
if (pcmk__check_action_config(rsc, node, rsc_op)
&& pe_get_failcount(node, rsc, NULL, pcmk__fc_effective,
NULL)) {
reason = "action definition changed";
}
break;
case pcmk__check_last_failure:
digest_data = rsc_action_digest_cmp(rsc, rsc_op, node,
rsc->priv->scheduler);
switch (digest_data->rc) {
case pcmk__digest_unknown:
crm_trace("Resource %s history entry %s on %s has "
"no digest to compare",
rsc->id, pcmk__xe_id(rsc_op), node->priv->id);
break;
case pcmk__digest_match:
break;
default:
reason = "resource parameters have changed";
break;
}
break;
}
if (reason != NULL) {
pe__clear_failcount(rsc, node, reason, rsc->priv->scheduler);
}
}
/*!
* \internal
* \brief Check whether a resource has failcount clearing scheduled on a node
*
* \param[in] node Node to check
* \param[in] rsc Resource to check
*
* \return true if \p rsc has failcount clearing scheduled on \p node,
* otherwise false
*/
static bool
failcount_clear_action_exists(const pcmk_node_t *node,
const pcmk_resource_t *rsc)
{
GList *list = pe__resource_actions(rsc, node, PCMK_ACTION_CLEAR_FAILCOUNT,
TRUE);
if (list != NULL) {
g_list_free(list);
return true;
}
return false;
}
/*!
* \internal
* \brief Ban a resource from a node if it reached its failure threshold there
*
* \param[in,out] data Resource to check failure threshold for
* \param[in] user_data Node to check resource on
*/
static void
check_failure_threshold(gpointer data, gpointer user_data)
{
pcmk_resource_t *rsc = data;
const pcmk_node_t *node = user_data;
// If this is a collective resource, apply recursively to children instead
if (rsc->priv->children != NULL) {
g_list_foreach(rsc->priv->children, check_failure_threshold,
user_data);
return;
}
if (!failcount_clear_action_exists(node, rsc)) {
/* Don't force the resource away from this node due to a failcount
* that's going to be cleared.
*
* @TODO Failcount clearing can be scheduled in
* pcmk__handle_rsc_config_changes() via process_rsc_history(), or in
* schedule_resource_actions() via check_params(). This runs well before
* then, so it cannot detect those, meaning we might check the migration
* threshold when we shouldn't. Worst case, we stop or move the
* resource, then move it back in the next transition.
*/
pcmk_resource_t *failed = NULL;
if (pcmk__threshold_reached(rsc, node, &failed)) {
resource_location(failed, node, -PCMK_SCORE_INFINITY,
"__fail_limit__", rsc->priv->scheduler);
}
}
}
/*!
* \internal
* \brief If resource has exclusive discovery, ban node if not allowed
*
* Location constraints have a PCMK_XA_RESOURCE_DISCOVERY option that allows
* users to specify where probes are done for the affected resource. If this is
* set to \c exclusive, probes will only be done on nodes listed in exclusive
* constraints. This function bans the resource from the node if the node is not
* listed.
*
* \param[in,out] data Resource to check
* \param[in] user_data Node to check resource on
*/
static void
apply_exclusive_discovery(gpointer data, gpointer user_data)
{
pcmk_resource_t *rsc = data;
const pcmk_node_t *node = user_data;
+ /* @TODO This checks rsc and the top rsc, but should probably check all
+ * ancestors (a cloned group could have it set on the group)
+ */
if (pcmk_is_set(rsc->flags, pcmk__rsc_exclusive_probes)
|| pcmk_is_set(pe__const_top_resource(rsc, false)->flags,
pcmk__rsc_exclusive_probes)) {
pcmk_node_t *match = NULL;
// If this is a collective resource, apply recursively to children
g_list_foreach(rsc->priv->children, apply_exclusive_discovery,
user_data);
match = g_hash_table_lookup(rsc->priv->allowed_nodes,
node->priv->id);
if ((match != NULL)
&& (match->assign->probe_mode != pcmk__probe_exclusive)) {
match->assign->score = -PCMK_SCORE_INFINITY;
}
}
}
/*!
* \internal
* \brief Apply stickiness to a resource if appropriate
*
* \param[in,out] data Resource to check for stickiness
* \param[in] user_data Ignored
*/
static void
apply_stickiness(gpointer data, gpointer user_data)
{
pcmk_resource_t *rsc = data;
pcmk_node_t *node = NULL;
// If this is a collective resource, apply recursively to children instead
if (rsc->priv->children != NULL) {
g_list_foreach(rsc->priv->children, apply_stickiness, NULL);
return;
}
/* A resource is sticky if it is managed, has stickiness configured, and is
* active on a single node.
*/
if (!pcmk_is_set(rsc->flags, pcmk__rsc_managed)
|| (rsc->priv->stickiness < 1)
|| !pcmk__list_of_1(rsc->priv->active_nodes)) {
return;
}
node = rsc->priv->active_nodes->data;
/* In a symmetric cluster, stickiness can always be used. In an
* asymmetric cluster, we have to check whether the resource is still
* allowed on the node, so we don't keep the resource somewhere it is no
* longer explicitly enabled.
*/
if (!pcmk_is_set(rsc->priv->scheduler->flags,
pcmk__sched_symmetric_cluster)
&& (g_hash_table_lookup(rsc->priv->allowed_nodes,
node->priv->id) == NULL)) {
pcmk__rsc_debug(rsc,
"Ignoring %s stickiness because the cluster is "
"asymmetric and %s is not explicitly allowed",
rsc->id, pcmk__node_name(node));
return;
}
pcmk__rsc_debug(rsc, "Resource %s has %d stickiness on %s",
rsc->id, rsc->priv->stickiness, pcmk__node_name(node));
resource_location(rsc, node, rsc->priv->stickiness, "stickiness",
rsc->priv->scheduler);
}
/*!
* \internal
* \brief Apply shutdown locks for all resources as appropriate
*
* \param[in,out] scheduler Scheduler data
*/
static void
apply_shutdown_locks(pcmk_scheduler_t *scheduler)
{
if (!pcmk_is_set(scheduler->flags, pcmk__sched_shutdown_lock)) {
return;
}
for (GList *iter = scheduler->priv->resources;
iter != NULL; iter = iter->next) {
pcmk_resource_t *rsc = (pcmk_resource_t *) iter->data;
rsc->priv->cmds->shutdown_lock(rsc);
}
}
/*
* \internal
* \brief Apply node-specific scheduling criteria
*
* After the CIB has been unpacked, process node-specific scheduling criteria
* including shutdown locks, location constraints, resource stickiness,
* migration thresholds, and exclusive resource discovery.
*/
static void
apply_node_criteria(pcmk_scheduler_t *scheduler)
{
crm_trace("Applying node-specific scheduling criteria");
apply_shutdown_locks(scheduler);
pcmk__apply_locations(scheduler);
g_list_foreach(scheduler->priv->resources, apply_stickiness, NULL);
for (GList *node_iter = scheduler->nodes; node_iter != NULL;
node_iter = node_iter->next) {
for (GList *rsc_iter = scheduler->priv->resources;
rsc_iter != NULL; rsc_iter = rsc_iter->next) {
check_failure_threshold(rsc_iter->data, node_iter->data);
apply_exclusive_discovery(rsc_iter->data, node_iter->data);
}
}
}
/*!
* \internal
* \brief Assign resources to nodes
*
* \param[in,out] scheduler Scheduler data
*/
static void
assign_resources(pcmk_scheduler_t *scheduler)
{
GList *iter = NULL;
crm_trace("Assigning resources to nodes");
if (!pcmk__str_eq(scheduler->priv->placement_strategy, PCMK_VALUE_DEFAULT,
pcmk__str_casei)) {
pcmk__sort_resources(scheduler);
}
pcmk__show_node_capacities("Original", scheduler);
if (pcmk_is_set(scheduler->flags, pcmk__sched_have_remote_nodes)) {
/* Assign remote connection resources first (which will also assign any
* colocation dependencies). If the connection is migrating, always
* prefer the partial migration target.
*/
for (iter = scheduler->priv->resources;
iter != NULL; iter = iter->next) {
pcmk_resource_t *rsc = (pcmk_resource_t *) iter->data;
const pcmk_node_t *target = rsc->priv->partial_migration_target;
if (pcmk_is_set(rsc->flags, pcmk__rsc_is_remote_connection)) {
pcmk__rsc_trace(rsc, "Assigning remote connection resource '%s'",
rsc->id);
rsc->priv->cmds->assign(rsc, target, true);
}
}
}
/* now do the rest of the resources */
for (iter = scheduler->priv->resources; iter != NULL; iter = iter->next) {
pcmk_resource_t *rsc = (pcmk_resource_t *) iter->data;
if (!pcmk_is_set(rsc->flags, pcmk__rsc_is_remote_connection)) {
pcmk__rsc_trace(rsc, "Assigning %s resource '%s'",
rsc->priv->xml->name, rsc->id);
rsc->priv->cmds->assign(rsc, NULL, true);
}
}
pcmk__show_node_capacities("Remaining", scheduler);
}
/*!
* \internal
* \brief Schedule fail count clearing on online nodes if resource is orphaned
*
* \param[in,out] data Resource to check
* \param[in] user_data Ignored
*/
static void
clear_failcounts_if_orphaned(gpointer data, gpointer user_data)
{
pcmk_resource_t *rsc = data;
if (!pcmk_is_set(rsc->flags, pcmk__rsc_removed)) {
return;
}
crm_trace("Clear fail counts for orphaned resource %s", rsc->id);
/* There's no need to recurse into rsc->private->children because those
* should just be unassigned clone instances.
*/
for (GList *iter = rsc->priv->scheduler->nodes;
iter != NULL; iter = iter->next) {
pcmk_node_t *node = (pcmk_node_t *) iter->data;
pcmk_action_t *clear_op = NULL;
if (!node->details->online) {
continue;
}
if (pe_get_failcount(node, rsc, NULL, pcmk__fc_effective, NULL) == 0) {
continue;
}
clear_op = pe__clear_failcount(rsc, node, "it is orphaned",
rsc->priv->scheduler);
/* We can't use order_action_then_stop() here because its
* pcmk__ar_guest_allowed breaks things
*/
pcmk__new_ordering(clear_op->rsc, NULL, clear_op, rsc, stop_key(rsc),
NULL, pcmk__ar_ordered, rsc->priv->scheduler);
}
}
/*!
* \internal
* \brief Schedule any resource actions needed
*
* \param[in,out] scheduler Scheduler data
*/
static void
schedule_resource_actions(pcmk_scheduler_t *scheduler)
{
// Process deferred action checks
pe__foreach_param_check(scheduler, check_params);
pe__free_param_checks(scheduler);
if (pcmk_is_set(scheduler->flags, pcmk__sched_probe_resources)) {
crm_trace("Scheduling probes");
pcmk__schedule_probes(scheduler);
}
if (pcmk_is_set(scheduler->flags, pcmk__sched_stop_removed_resources)) {
g_list_foreach(scheduler->priv->resources, clear_failcounts_if_orphaned,
NULL);
}
crm_trace("Scheduling resource actions");
for (GList *iter = scheduler->priv->resources;
iter != NULL; iter = iter->next) {
pcmk_resource_t *rsc = (pcmk_resource_t *) iter->data;
rsc->priv->cmds->create_actions(rsc);
}
}
/*!
* \internal
* \brief Check whether a resource or any of its descendants are managed
*
* \param[in] rsc Resource to check
*
* \return true if resource or any descendant is managed, otherwise false
*/
static bool
is_managed(const pcmk_resource_t *rsc)
{
if (pcmk_is_set(rsc->flags, pcmk__rsc_managed)) {
return true;
}
for (GList *iter = rsc->priv->children;
iter != NULL; iter = iter->next) {
if (is_managed((pcmk_resource_t *) iter->data)) {
return true;
}
}
return false;
}
/*!
* \internal
* \brief Check whether any resources in the cluster are managed
*
* \param[in] scheduler Scheduler data
*
* \return true if any resource is managed, otherwise false
*/
static bool
any_managed_resources(const pcmk_scheduler_t *scheduler)
{
for (const GList *iter = scheduler->priv->resources;
iter != NULL; iter = iter->next) {
if (is_managed((const pcmk_resource_t *) iter->data)) {
return true;
}
}
return false;
}
/*!
* \internal
* \brief Check whether a node requires fencing
*
* \param[in] node Node to check
* \param[in] have_managed Whether any resource in cluster is managed
*
* \return true if \p node should be fenced, otherwise false
*/
static bool
needs_fencing(const pcmk_node_t *node, bool have_managed)
{
return have_managed && node->details->unclean
&& pe_can_fence(node->priv->scheduler, node);
}
/*!
* \internal
* \brief Check whether a node requires shutdown
*
* \param[in] node Node to check
*
* \return true if \p node should be shut down, otherwise false
*/
static bool
needs_shutdown(const pcmk_node_t *node)
{
if (pcmk__is_pacemaker_remote_node(node)) {
/* Do not send shutdown actions for Pacemaker Remote nodes.
* @TODO We might come up with a good use for this in the future.
*/
return false;
}
return node->details->online && node->details->shutdown;
}
/*!
* \internal
* \brief Track and order non-DC fencing
*
* \param[in,out] list List of existing non-DC fencing actions
* \param[in,out] action Fencing action to prepend to \p list
* \param[in] scheduler Scheduler data
*
* \return (Possibly new) head of \p list
*/
static GList *
add_nondc_fencing(GList *list, pcmk_action_t *action,
const pcmk_scheduler_t *scheduler)
{
if (!pcmk_is_set(scheduler->flags, pcmk__sched_concurrent_fencing)
&& (list != NULL)) {
/* Concurrent fencing is disabled, so order each non-DC
* fencing in a chain. If there is any DC fencing or
* shutdown, it will be ordered after the last action in the
* chain later.
*/
order_actions((pcmk_action_t *) list->data, action, pcmk__ar_ordered);
}
return g_list_prepend(list, action);
}
/*!
* \internal
* \brief Schedule a node for fencing
*
* \param[in,out] node Node that requires fencing
*/
static pcmk_action_t *
schedule_fencing(pcmk_node_t *node)
{
pcmk_action_t *fencing = pe_fence_op(node, NULL, FALSE, "node is unclean",
FALSE, node->priv->scheduler);
pcmk__sched_warn(node->priv->scheduler, "Scheduling node %s for fencing",
pcmk__node_name(node));
pcmk__order_vs_fence(fencing, node->priv->scheduler);
return fencing;
}
/*!
* \internal
* \brief Create and order node fencing and shutdown actions
*
* \param[in,out] scheduler Scheduler data
*/
static void
schedule_fencing_and_shutdowns(pcmk_scheduler_t *scheduler)
{
pcmk_action_t *dc_down = NULL;
bool integrity_lost = false;
bool have_managed = any_managed_resources(scheduler);
GList *fencing_ops = NULL;
GList *shutdown_ops = NULL;
crm_trace("Scheduling fencing and shutdowns as needed");
if (!have_managed) {
crm_notice("No fencing will be done until there are resources "
"to manage");
}
// Check each node for whether it needs fencing or shutdown
for (GList *iter = scheduler->nodes; iter != NULL; iter = iter->next) {
pcmk_node_t *node = (pcmk_node_t *) iter->data;
pcmk_action_t *fencing = NULL;
const bool is_dc = pcmk__same_node(node, scheduler->dc_node);
/* Guest nodes are "fenced" by recovering their container resource,
* so handle them separately.
*/
if (pcmk__is_guest_or_bundle_node(node)) {
if (pcmk_is_set(node->priv->flags, pcmk__node_remote_reset)
&& have_managed && pe_can_fence(scheduler, node)) {
pcmk__fence_guest(node);
}
continue;
}
if (needs_fencing(node, have_managed)) {
fencing = schedule_fencing(node);
// Track DC and non-DC fence actions separately
if (is_dc) {
dc_down = fencing;
} else {
fencing_ops = add_nondc_fencing(fencing_ops, fencing,
scheduler);
}
} else if (needs_shutdown(node)) {
pcmk_action_t *down_op = pcmk__new_shutdown_action(node);
// Track DC and non-DC shutdown actions separately
if (is_dc) {
dc_down = down_op;
} else {
shutdown_ops = g_list_prepend(shutdown_ops, down_op);
}
}
if ((fencing == NULL) && node->details->unclean) {
integrity_lost = true;
pcmk__config_warn("Node %s is unclean but cannot be fenced",
pcmk__node_name(node));
}
}
if (integrity_lost) {
if (!pcmk_is_set(scheduler->flags, pcmk__sched_fencing_enabled)) {
pcmk__config_warn("Resource functionality and data integrity "
"cannot be guaranteed (configure, enable, "
"and test fencing to correct this)");
} else if (!pcmk_is_set(scheduler->flags, pcmk__sched_quorate)) {
crm_notice("Unclean nodes will not be fenced until quorum is "
"attained or " PCMK_OPT_NO_QUORUM_POLICY " is set to "
PCMK_VALUE_IGNORE);
}
}
if (dc_down != NULL) {
/* Order any non-DC shutdowns before any DC shutdown, to avoid repeated
* DC elections. However, we don't want to order non-DC shutdowns before
* a DC *fencing*, because even though we don't want a node that's
* shutting down to become DC, the DC fencing could be ordered before a
* clone stop that's also ordered before the shutdowns, thus leading to
* a graph loop.
*/
if (pcmk__str_eq(dc_down->task, PCMK_ACTION_DO_SHUTDOWN,
pcmk__str_none)) {
pcmk__order_after_each(dc_down, shutdown_ops);
}
// Order any non-DC fencing before any DC fencing or shutdown
if (pcmk_is_set(scheduler->flags, pcmk__sched_concurrent_fencing)) {
/* With concurrent fencing, order each non-DC fencing action
* separately before any DC fencing or shutdown.
*/
pcmk__order_after_each(dc_down, fencing_ops);
} else if (fencing_ops != NULL) {
/* Without concurrent fencing, the non-DC fencing actions are
* already ordered relative to each other, so we just need to order
* the DC fencing after the last action in the chain (which is the
* first item in the list).
*/
order_actions((pcmk_action_t *) fencing_ops->data, dc_down,
pcmk__ar_ordered);
}
}
g_list_free(fencing_ops);
g_list_free(shutdown_ops);
}
static void
log_resource_details(pcmk_scheduler_t *scheduler)
{
pcmk__output_t *out = scheduler->priv->out;
GList *all = NULL;
/* Due to the `crm_mon --node=` feature, out->message() for all the
* resource-related messages expects a list of nodes that we are allowed to
* output information for. Here, we create a wildcard to match all nodes.
*/
all = g_list_prepend(all, (gpointer) "*");
for (GList *item = scheduler->priv->resources;
item != NULL; item = item->next) {
pcmk_resource_t *rsc = (pcmk_resource_t *) item->data;
// Log all resources except inactive orphans
if (!pcmk_is_set(rsc->flags, pcmk__rsc_removed)
|| (rsc->priv->orig_role != pcmk_role_stopped)) {
out->message(out, (const char *) rsc->priv->xml->name, 0UL,
rsc, all, all);
}
}
g_list_free(all);
}
static void
log_all_actions(pcmk_scheduler_t *scheduler)
{
/* This only ever outputs to the log, so ignore whatever output object was
* previously set and just log instead.
*/
pcmk__output_t *prev_out = scheduler->priv->out;
pcmk__output_t *out = NULL;
if (pcmk__log_output_new(&out) != pcmk_rc_ok) {
return;
}
pe__register_messages(out);
pcmk__register_lib_messages(out);
pcmk__output_set_log_level(out, LOG_NOTICE);
scheduler->priv->out = out;
out->begin_list(out, NULL, NULL, "Actions");
pcmk__output_actions(scheduler);
out->end_list(out);
out->finish(out, CRM_EX_OK, true, NULL);
pcmk__output_free(out);
scheduler->priv->out = prev_out;
}
/*!
* \internal
* \brief Log all required but unrunnable actions at trace level
*
* \param[in] scheduler Scheduler data
*/
static void
log_unrunnable_actions(const pcmk_scheduler_t *scheduler)
{
const uint64_t flags = pcmk__action_optional
|pcmk__action_runnable
|pcmk__action_pseudo;
crm_trace("Required but unrunnable actions:");
for (const GList *iter = scheduler->priv->actions;
iter != NULL; iter = iter->next) {
const pcmk_action_t *action = (const pcmk_action_t *) iter->data;
if (!pcmk_any_flags_set(action->flags, flags)) {
pcmk__log_action("\t", action, true);
}
}
}
/*!
* \internal
* \brief Unpack the CIB for scheduling
*
* \param[in,out] cib CIB XML to unpack (may be NULL if already unpacked)
* \param[in] flags Scheduler flags to set in addition to defaults
* \param[in,out] scheduler Scheduler data
*/
static void
unpack_cib(xmlNode *cib, unsigned long long flags, pcmk_scheduler_t *scheduler)
{
if (pcmk_is_set(scheduler->flags, pcmk__sched_have_status)) {
crm_trace("Reusing previously calculated cluster status");
pcmk__set_scheduler_flags(scheduler, flags);
return;
}
pcmk__assert(cib != NULL);
crm_trace("Calculating cluster status");
/* This will zero the entire struct without freeing anything first, so
* callers should never call pcmk__schedule_actions() with a populated data
* set unless pcmk__sched_have_status is set (i.e. cluster_status() was
* previously called, whether directly or via pcmk__schedule_actions()).
*/
set_working_set_defaults(scheduler);
pcmk__set_scheduler_flags(scheduler, flags);
scheduler->input = cib;
cluster_status(scheduler); // Sets pcmk__sched_have_status
}
/*!
* \internal
* \brief Run the scheduler for a given CIB
*
* \param[in,out] cib CIB XML to use as scheduler input
* \param[in] flags Scheduler flags to set in addition to defaults
* \param[in,out] scheduler Scheduler data
*/
void
pcmk__schedule_actions(xmlNode *cib, unsigned long long flags,
pcmk_scheduler_t *scheduler)
{
unpack_cib(cib, flags, scheduler);
pcmk__set_assignment_methods(scheduler);
pcmk__apply_node_health(scheduler);
pcmk__unpack_constraints(scheduler);
if (pcmk_is_set(scheduler->flags, pcmk__sched_validate_only)) {
return;
}
if (!pcmk_is_set(scheduler->flags, pcmk__sched_location_only)
&& pcmk__is_daemon) {
log_resource_details(scheduler);
}
apply_node_criteria(scheduler);
if (pcmk_is_set(scheduler->flags, pcmk__sched_location_only)) {
return;
}
pcmk__create_internal_constraints(scheduler);
pcmk__handle_rsc_config_changes(scheduler);
assign_resources(scheduler);
schedule_resource_actions(scheduler);
/* Remote ordering constraints need to happen prior to calculating fencing
* because it is one more place we can mark nodes as needing fencing.
*/
pcmk__order_remote_connection_actions(scheduler);
schedule_fencing_and_shutdowns(scheduler);
pcmk__apply_orderings(scheduler);
log_all_actions(scheduler);
pcmk__create_graph(scheduler);
if (get_crm_log_level() == LOG_TRACE) {
log_unrunnable_actions(scheduler);
}
}
/*!
* \internal
* \brief Initialize scheduler data
*
* Make our own copies of the CIB XML and date/time object, if they're not
* \c NULL. This way we don't have to take ownership of the objects passed via
* the API.
*
* This function is most useful for public API functions that want the caller
* to retain ownership of the CIB object
*
* \param[in,out] out Output object
* \param[in] input The CIB XML to check (if \c NULL, use current CIB)
* \param[in] date Date and time to use in the scheduler (if \c NULL,
* use current date and time). This can be used for
* checking whether a rule is in effect at a certa
* date and time.
* \param[out] scheduler Where to store initialized scheduler data
*
* \return Standard Pacemaker return code
*/
int
pcmk__init_scheduler(pcmk__output_t *out, xmlNodePtr input, const crm_time_t *date,
pcmk_scheduler_t **scheduler)
{
// Allows for cleaner syntax than dereferencing the scheduler argument
pcmk_scheduler_t *new_scheduler = NULL;
new_scheduler = pe_new_working_set();
if (new_scheduler == NULL) {
return ENOMEM;
}
pcmk__set_scheduler_flags(new_scheduler, pcmk__sched_no_counts);
// Populate the scheduler data
// Make our own copy of the given input or fetch the CIB and use that
if (input != NULL) {
new_scheduler->input = pcmk__xml_copy(NULL, input);
if (new_scheduler->input == NULL) {
out->err(out, "Failed to copy input XML");
pe_free_working_set(new_scheduler);
return ENOMEM;
}
} else {
int rc = cib__signon_query(out, NULL, &(new_scheduler->input));
if (rc != pcmk_rc_ok) {
pe_free_working_set(new_scheduler);
return rc;
}
}
// Make our own copy of the given crm_time_t object; otherwise
// cluster_status() populates with the current time
if (date != NULL) {
// pcmk_copy_time() guarantees non-NULL
new_scheduler->priv->now = pcmk_copy_time(date);
}
// Unpack everything
cluster_status(new_scheduler);
*scheduler = new_scheduler;
return pcmk_rc_ok;
}
diff --git a/lib/pengine/bundle.c b/lib/pengine/bundle.c
index f19fbd11c0..fc2d1a3d80 100644
--- a/lib/pengine/bundle.c
+++ b/lib/pengine/bundle.c
@@ -1,2091 +1,2097 @@
/*
* Copyright 2004-2024 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU Lesser General Public License
* version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY.
*/
#include <crm_internal.h>
#include <ctype.h>
#include <stdint.h>
#include <crm/pengine/status.h>
#include <crm/pengine/internal.h>
#include <crm/common/xml.h>
#include <crm/common/output.h>
#include <crm/common/xml_internal.h>
#include <pe_status_private.h>
enum pe__bundle_mount_flags {
pe__bundle_mount_none = 0x00,
// mount instance-specific subdirectory rather than source directly
pe__bundle_mount_subdir = 0x01
};
typedef struct {
char *source;
char *target;
char *options;
uint32_t flags; // bitmask of pe__bundle_mount_flags
} pe__bundle_mount_t;
typedef struct {
char *source;
char *target;
} pe__bundle_port_t;
enum pe__container_agent {
PE__CONTAINER_AGENT_UNKNOWN,
PE__CONTAINER_AGENT_DOCKER,
PE__CONTAINER_AGENT_PODMAN,
};
#define PE__CONTAINER_AGENT_UNKNOWN_S "unknown"
#define PE__CONTAINER_AGENT_DOCKER_S "docker"
#define PE__CONTAINER_AGENT_PODMAN_S "podman"
typedef struct pe__bundle_variant_data_s {
int promoted_max;
int nreplicas;
int nreplicas_per_host;
char *prefix;
char *image;
const char *ip_last;
char *host_network;
char *host_netmask;
char *control_port;
char *container_network;
char *ip_range_start;
gboolean add_host;
gchar *container_host_options;
char *container_command;
char *launcher_options;
const char *attribute_target;
pcmk_resource_t *child;
GList *replicas; // pcmk__bundle_replica_t *
GList *ports; // pe__bundle_port_t *
GList *mounts; // pe__bundle_mount_t *
+ /* @TODO Maybe use a more object-oriented design instead, with a set of
+ * methods that are different per type rather than switching on this
+ */
enum pe__container_agent agent_type;
} pe__bundle_variant_data_t;
#define get_bundle_variant_data(data, rsc) do { \
pcmk__assert(pcmk__is_bundle(rsc)); \
data = rsc->priv->variant_opaque; \
} while (0)
/*!
* \internal
* \brief Get maximum number of bundle replicas allowed to run
*
* \param[in] rsc Bundle or bundled resource to check
*
* \return Maximum replicas for bundle corresponding to \p rsc
*/
int
pe__bundle_max(const pcmk_resource_t *rsc)
{
const pe__bundle_variant_data_t *bundle_data = NULL;
get_bundle_variant_data(bundle_data, pe__const_top_resource(rsc, true));
return bundle_data->nreplicas;
}
/*!
* \internal
* \brief Get the resource inside a bundle
*
* \param[in] bundle Bundle to check
*
* \return Resource inside \p bundle if any, otherwise NULL
*/
pcmk_resource_t *
pe__bundled_resource(const pcmk_resource_t *rsc)
{
const pe__bundle_variant_data_t *bundle_data = NULL;
get_bundle_variant_data(bundle_data, pe__const_top_resource(rsc, true));
return bundle_data->child;
}
/*!
* \internal
* \brief Get containerized resource corresponding to a given bundle container
*
* \param[in] instance Collective instance that might be a bundle container
*
* \return Bundled resource instance inside \p instance if it is a bundle
* container instance, otherwise NULL
*/
const pcmk_resource_t *
pe__get_rsc_in_container(const pcmk_resource_t *instance)
{
const pe__bundle_variant_data_t *data = NULL;
const pcmk_resource_t *top = pe__const_top_resource(instance, true);
if (!pcmk__is_bundle(top)) {
return NULL;
}
get_bundle_variant_data(data, top);
for (const GList *iter = data->replicas; iter != NULL; iter = iter->next) {
const pcmk__bundle_replica_t *replica = iter->data;
if (instance == replica->container) {
return replica->child;
}
}
return NULL;
}
/*!
* \internal
* \brief Check whether a given node is created by a bundle
*
* \param[in] bundle Bundle resource to check
* \param[in] node Node to check
*
* \return true if \p node is an instance of \p bundle, otherwise false
*/
bool
pe__node_is_bundle_instance(const pcmk_resource_t *bundle,
const pcmk_node_t *node)
{
pe__bundle_variant_data_t *bundle_data = NULL;
get_bundle_variant_data(bundle_data, bundle);
for (GList *iter = bundle_data->replicas; iter != NULL; iter = iter->next) {
pcmk__bundle_replica_t *replica = iter->data;
if (pcmk__same_node(node, replica->node)) {
return true;
}
}
return false;
}
/*!
* \internal
* \brief Get the container of a bundle's first replica
*
* \param[in] bundle Bundle resource to get container for
*
* \return Container resource from first replica of \p bundle if any,
* otherwise NULL
*/
pcmk_resource_t *
pe__first_container(const pcmk_resource_t *bundle)
{
const pe__bundle_variant_data_t *bundle_data = NULL;
const pcmk__bundle_replica_t *replica = NULL;
get_bundle_variant_data(bundle_data, bundle);
if (bundle_data->replicas == NULL) {
return NULL;
}
replica = bundle_data->replicas->data;
return replica->container;
}
/*!
* \internal
* \brief Iterate over bundle replicas
*
* \param[in,out] bundle Bundle to iterate over
* \param[in] fn Function to call for each replica (its return value
* indicates whether to continue iterating)
* \param[in,out] user_data Pointer to pass to \p fn
*/
void
pe__foreach_bundle_replica(pcmk_resource_t *bundle,
bool (*fn)(pcmk__bundle_replica_t *, void *),
void *user_data)
{
const pe__bundle_variant_data_t *bundle_data = NULL;
get_bundle_variant_data(bundle_data, bundle);
for (GList *iter = bundle_data->replicas; iter != NULL; iter = iter->next) {
if (!fn((pcmk__bundle_replica_t *) iter->data, user_data)) {
break;
}
}
}
/*!
* \internal
* \brief Iterate over const bundle replicas
*
* \param[in] bundle Bundle to iterate over
* \param[in] fn Function to call for each replica (its return value
* indicates whether to continue iterating)
* \param[in,out] user_data Pointer to pass to \p fn
*/
void
pe__foreach_const_bundle_replica(const pcmk_resource_t *bundle,
bool (*fn)(const pcmk__bundle_replica_t *,
void *),
void *user_data)
{
const pe__bundle_variant_data_t *bundle_data = NULL;
get_bundle_variant_data(bundle_data, bundle);
for (const GList *iter = bundle_data->replicas; iter != NULL;
iter = iter->next) {
if (!fn((const pcmk__bundle_replica_t *) iter->data, user_data)) {
break;
}
}
}
static char *
next_ip(const char *last_ip)
{
unsigned int oct1 = 0;
unsigned int oct2 = 0;
unsigned int oct3 = 0;
unsigned int oct4 = 0;
int rc = sscanf(last_ip, "%u.%u.%u.%u", &oct1, &oct2, &oct3, &oct4);
if (rc != 4) {
/*@ TODO check for IPv6 */
return NULL;
} else if (oct3 > 253) {
return NULL;
} else if (oct4 > 253) {
++oct3;
oct4 = 1;
} else {
++oct4;
}
return crm_strdup_printf("%u.%u.%u.%u", oct1, oct2, oct3, oct4);
}
static void
allocate_ip(pe__bundle_variant_data_t *data, pcmk__bundle_replica_t *replica,
GString *buffer)
{
if(data->ip_range_start == NULL) {
return;
} else if(data->ip_last) {
replica->ipaddr = next_ip(data->ip_last);
} else {
replica->ipaddr = strdup(data->ip_range_start);
}
data->ip_last = replica->ipaddr;
switch (data->agent_type) {
case PE__CONTAINER_AGENT_DOCKER:
case PE__CONTAINER_AGENT_PODMAN:
if (data->add_host) {
g_string_append_printf(buffer, " --add-host=%s-%d:%s",
data->prefix, replica->offset,
replica->ipaddr);
} else {
g_string_append_printf(buffer, " --hosts-entry=%s=%s-%d",
replica->ipaddr, data->prefix,
replica->offset);
}
break;
default: // PE__CONTAINER_AGENT_UNKNOWN
break;
}
}
static xmlNode *
create_resource(const char *name, const char *provider, const char *kind)
{
xmlNode *rsc = pcmk__xe_create(NULL, PCMK_XE_PRIMITIVE);
crm_xml_add(rsc, PCMK_XA_ID, name);
crm_xml_add(rsc, PCMK_XA_CLASS, PCMK_RESOURCE_CLASS_OCF);
crm_xml_add(rsc, PCMK_XA_PROVIDER, provider);
crm_xml_add(rsc, PCMK_XA_TYPE, kind);
return rsc;
}
/*!
* \internal
* \brief Check whether cluster can manage resource inside container
*
* \param[in,out] data Container variant data
*
* \return TRUE if networking configuration is acceptable, FALSE otherwise
*
* \note The resource is manageable if an IP range or control port has been
* specified. If a control port is used without an IP range, replicas per
* host must be 1.
*/
static bool
valid_network(pe__bundle_variant_data_t *data)
{
if(data->ip_range_start) {
return TRUE;
}
if(data->control_port) {
if(data->nreplicas_per_host > 1) {
pcmk__config_err("Specifying the '" PCMK_XA_CONTROL_PORT "' for %s "
"requires '" PCMK_XA_REPLICAS_PER_HOST "=1'",
data->prefix);
data->nreplicas_per_host = 1;
// @TODO to be sure:
// pcmk__clear_rsc_flags(rsc, pcmk__rsc_unique);
}
return TRUE;
}
return FALSE;
}
static int
create_ip_resource(pcmk_resource_t *parent, pe__bundle_variant_data_t *data,
pcmk__bundle_replica_t *replica)
{
if(data->ip_range_start) {
char *id = NULL;
xmlNode *xml_ip = NULL;
xmlNode *xml_obj = NULL;
id = crm_strdup_printf("%s-ip-%s", data->prefix, replica->ipaddr);
pcmk__xml_sanitize_id(id);
xml_ip = create_resource(id, "heartbeat", "IPaddr2");
free(id);
xml_obj = pcmk__xe_create(xml_ip, PCMK_XE_INSTANCE_ATTRIBUTES);
pcmk__xe_set_id(xml_obj, "%s-attributes-%d",
data->prefix, replica->offset);
crm_create_nvpair_xml(xml_obj, NULL, "ip", replica->ipaddr);
if(data->host_network) {
crm_create_nvpair_xml(xml_obj, NULL, "nic", data->host_network);
}
if(data->host_netmask) {
crm_create_nvpair_xml(xml_obj, NULL,
"cidr_netmask", data->host_netmask);
} else {
crm_create_nvpair_xml(xml_obj, NULL, "cidr_netmask", "32");
}
xml_obj = pcmk__xe_create(xml_ip, PCMK_XE_OPERATIONS);
crm_create_op_xml(xml_obj, pcmk__xe_id(xml_ip), PCMK_ACTION_MONITOR,
"60s", NULL);
// TODO: Other ops? Timeouts and intervals from underlying resource?
if (pe__unpack_resource(xml_ip, &replica->ip, parent,
parent->priv->scheduler) != pcmk_rc_ok) {
return pcmk_rc_unpack_error;
}
parent->priv->children = g_list_append(parent->priv->children,
replica->ip);
}
return pcmk_rc_ok;
}
static const char*
container_agent_str(enum pe__container_agent t)
{
switch (t) {
case PE__CONTAINER_AGENT_DOCKER: return PE__CONTAINER_AGENT_DOCKER_S;
case PE__CONTAINER_AGENT_PODMAN: return PE__CONTAINER_AGENT_PODMAN_S;
default: // PE__CONTAINER_AGENT_UNKNOWN
break;
}
return PE__CONTAINER_AGENT_UNKNOWN_S;
}
static int
create_container_resource(pcmk_resource_t *parent,
const pe__bundle_variant_data_t *data,
pcmk__bundle_replica_t *replica)
{
char *id = NULL;
xmlNode *xml_container = NULL;
xmlNode *xml_obj = NULL;
// Agent-specific
const char *hostname_opt = NULL;
const char *env_opt = NULL;
const char *agent_str = NULL;
GString *buffer = NULL;
GString *dbuffer = NULL;
// Where syntax differences are drop-in replacements, set them now
switch (data->agent_type) {
case PE__CONTAINER_AGENT_DOCKER:
case PE__CONTAINER_AGENT_PODMAN:
hostname_opt = "-h ";
env_opt = "-e ";
break;
default: // PE__CONTAINER_AGENT_UNKNOWN
return pcmk_rc_unpack_error;
}
agent_str = container_agent_str(data->agent_type);
buffer = g_string_sized_new(4096);
id = crm_strdup_printf("%s-%s-%d", data->prefix, agent_str,
replica->offset);
pcmk__xml_sanitize_id(id);
xml_container = create_resource(id, "heartbeat", agent_str);
free(id);
xml_obj = pcmk__xe_create(xml_container, PCMK_XE_INSTANCE_ATTRIBUTES);
pcmk__xe_set_id(xml_obj, "%s-attributes-%d", data->prefix, replica->offset);
crm_create_nvpair_xml(xml_obj, NULL, "image", data->image);
crm_create_nvpair_xml(xml_obj, NULL, "allow_pull", PCMK_VALUE_TRUE);
crm_create_nvpair_xml(xml_obj, NULL, "force_kill", PCMK_VALUE_FALSE);
crm_create_nvpair_xml(xml_obj, NULL, "reuse", PCMK_VALUE_FALSE);
if (data->agent_type == PE__CONTAINER_AGENT_DOCKER) {
g_string_append(buffer, " --restart=no");
}
/* Set a container hostname only if we have an IP to map it to. The user can
* set -h or --uts=host themselves if they want a nicer name for logs, but
* this makes applications happy who need their hostname to match the IP
* they bind to.
*/
if (data->ip_range_start != NULL) {
g_string_append_printf(buffer, " %s%s-%d", hostname_opt, data->prefix,
replica->offset);
}
pcmk__g_strcat(buffer, " ", env_opt, "PCMK_stderr=1", NULL);
if (data->container_network != NULL) {
pcmk__g_strcat(buffer, " --net=", data->container_network, NULL);
}
if (data->control_port != NULL) {
pcmk__g_strcat(buffer, " ", env_opt, "PCMK_" PCMK__ENV_REMOTE_PORT "=",
data->control_port, NULL);
} else {
g_string_append_printf(buffer, " %sPCMK_" PCMK__ENV_REMOTE_PORT "=%d",
env_opt, DEFAULT_REMOTE_PORT);
}
for (GList *iter = data->mounts; iter != NULL; iter = iter->next) {
pe__bundle_mount_t *mount = (pe__bundle_mount_t *) iter->data;
char *source = NULL;
if (pcmk_is_set(mount->flags, pe__bundle_mount_subdir)) {
source = crm_strdup_printf("%s/%s-%d", mount->source, data->prefix,
replica->offset);
pcmk__add_separated_word(&dbuffer, 1024, source, ",");
}
switch (data->agent_type) {
case PE__CONTAINER_AGENT_DOCKER:
case PE__CONTAINER_AGENT_PODMAN:
pcmk__g_strcat(buffer,
" -v ", pcmk__s(source, mount->source),
":", mount->target, NULL);
if (mount->options != NULL) {
pcmk__g_strcat(buffer, ":", mount->options, NULL);
}
break;
default:
break;
}
free(source);
}
for (GList *iter = data->ports; iter != NULL; iter = iter->next) {
pe__bundle_port_t *port = (pe__bundle_port_t *) iter->data;
switch (data->agent_type) {
case PE__CONTAINER_AGENT_DOCKER:
case PE__CONTAINER_AGENT_PODMAN:
if (replica->ipaddr != NULL) {
pcmk__g_strcat(buffer,
" -p ", replica->ipaddr, ":", port->source,
":", port->target, NULL);
} else if (!pcmk__str_eq(data->container_network,
PCMK_VALUE_HOST, pcmk__str_none)) {
// No need to do port mapping if net == host
pcmk__g_strcat(buffer,
" -p ", port->source, ":", port->target,
NULL);
}
break;
default:
break;
}
}
/* @COMPAT: We should use pcmk__add_word() here, but we can't yet, because
* it would cause restarts during rolling upgrades.
*
* In a previous version of the container resource creation logic, if
* data->launcher_options is not NULL, we append
* (" %s", data->launcher_options) even if data->launcher_options is an
* empty string. Likewise for data->container_host_options. Using
*
* pcmk__add_word(buffer, 0, data->launcher_options)
*
* removes that extra trailing space, causing a resource definition change.
*/
if (data->launcher_options != NULL) {
pcmk__g_strcat(buffer, " ", data->launcher_options, NULL);
}
if (data->container_host_options != NULL) {
pcmk__g_strcat(buffer, " ", data->container_host_options, NULL);
}
crm_create_nvpair_xml(xml_obj, NULL, "run_opts",
(const char *) buffer->str);
g_string_free(buffer, TRUE);
crm_create_nvpair_xml(xml_obj, NULL, "mount_points",
(dbuffer != NULL)? (const char *) dbuffer->str : "");
if (dbuffer != NULL) {
g_string_free(dbuffer, TRUE);
}
if (replica->child != NULL) {
if (data->container_command != NULL) {
crm_create_nvpair_xml(xml_obj, NULL, "run_cmd",
data->container_command);
} else {
crm_create_nvpair_xml(xml_obj, NULL, "run_cmd",
SBIN_DIR "/" PCMK__SERVER_REMOTED);
}
/* TODO: Allow users to specify their own?
*
* We just want to know if the container is alive; we'll monitor the
* child independently.
*/
crm_create_nvpair_xml(xml_obj, NULL, "monitor_cmd", "/bin/true");
#if 0
/* @TODO Consider supporting the use case where we can start and stop
* resources, but not proxy local commands (such as setting node
* attributes), by running the local executor in stand-alone mode.
* However, this would probably be better done via ACLs as with other
* Pacemaker Remote nodes.
*/
} else if ((child != NULL) && data->untrusted) {
crm_create_nvpair_xml(xml_obj, NULL, "run_cmd",
CRM_DAEMON_DIR "/" PCMK__SERVER_EXECD);
crm_create_nvpair_xml(xml_obj, NULL, "monitor_cmd",
CRM_DAEMON_DIR "/pacemaker/cts-exec-helper -c poke");
#endif
} else {
if (data->container_command != NULL) {
crm_create_nvpair_xml(xml_obj, NULL, "run_cmd",
data->container_command);
}
/* TODO: Allow users to specify their own?
*
* We don't know what's in the container, so we just want to know if it
* is alive.
*/
crm_create_nvpair_xml(xml_obj, NULL, "monitor_cmd", "/bin/true");
}
xml_obj = pcmk__xe_create(xml_container, PCMK_XE_OPERATIONS);
crm_create_op_xml(xml_obj, pcmk__xe_id(xml_container), PCMK_ACTION_MONITOR,
"60s", NULL);
// TODO: Other ops? Timeouts and intervals from underlying resource?
if (pe__unpack_resource(xml_container, &replica->container, parent,
parent->priv->scheduler) != pcmk_rc_ok) {
return pcmk_rc_unpack_error;
}
pcmk__set_rsc_flags(replica->container, pcmk__rsc_replica_container);
parent->priv->children = g_list_append(parent->priv->children,
replica->container);
return pcmk_rc_ok;
}
/*!
* \brief Ban a node from a resource's (and its children's) allowed nodes list
*
* \param[in,out] rsc Resource to modify
* \param[in] uname Name of node to ban
*/
static void
disallow_node(pcmk_resource_t *rsc, const char *uname)
{
gpointer match = g_hash_table_lookup(rsc->priv->allowed_nodes, uname);
if (match) {
((pcmk_node_t *) match)->assign->score = -PCMK_SCORE_INFINITY;
((pcmk_node_t *) match)->assign->probe_mode = pcmk__probe_never;
}
g_list_foreach(rsc->priv->children, (GFunc) disallow_node,
(gpointer) uname);
}
static int
create_remote_resource(pcmk_resource_t *parent, pe__bundle_variant_data_t *data,
pcmk__bundle_replica_t *replica)
{
if (replica->child && valid_network(data)) {
GHashTableIter gIter;
pcmk_node_t *node = NULL;
xmlNode *xml_remote = NULL;
char *id = crm_strdup_printf("%s-%d", data->prefix, replica->offset);
char *port_s = NULL;
const char *uname = NULL;
const char *connect_name = NULL;
pcmk_scheduler_t *scheduler = parent->priv->scheduler;
if (pe_find_resource(scheduler->priv->resources, id) != NULL) {
free(id);
// The biggest hammer we have
id = crm_strdup_printf("pcmk-internal-%s-remote-%d",
replica->child->id, replica->offset);
//@TODO return error instead of asserting?
pcmk__assert(pe_find_resource(scheduler->priv->resources,
id) == NULL);
}
/* REMOTE_CONTAINER_HACK: Using "#uname" as the server name when the
* connection does not have its own IP is a magic string that we use to
* support nested remotes (i.e. a bundle running on a remote node).
*/
connect_name = (replica->ipaddr? replica->ipaddr : "#uname");
if (data->control_port == NULL) {
port_s = pcmk__itoa(DEFAULT_REMOTE_PORT);
}
/* This sets replica->container as replica->remote's container, which is
* similar to what happens with guest nodes. This is how the scheduler
* knows that the bundle node is fenced by recovering the container, and
* that remote should be ordered relative to the container.
*/
xml_remote = pe_create_remote_xml(NULL, id, replica->container->id,
NULL, NULL, NULL,
connect_name, (data->control_port?
data->control_port : port_s));
free(port_s);
/* Abandon our created ID, and pull the copy from the XML, because we
* need something that will get freed during scheduler data cleanup to
* use as the node ID and uname.
*/
free(id);
id = NULL;
uname = pcmk__xe_id(xml_remote);
/* Ensure a node has been created for the guest (it may have already
* been, if it has a permanent node attribute), and ensure its weight is
* -INFINITY so no other resources can run on it.
*/
node = pcmk_find_node(scheduler, uname);
if (node == NULL) {
node = pe_create_node(uname, uname, PCMK_VALUE_REMOTE,
-PCMK_SCORE_INFINITY, scheduler);
} else {
node->assign->score = -PCMK_SCORE_INFINITY;
}
node->assign->probe_mode = pcmk__probe_never;
/* unpack_remote_nodes() ensures that each remote node and guest node
* has a pcmk_node_t entry. Ideally, it would do the same for bundle
* nodes. Unfortunately, a bundle has to be mostly unpacked before it's
* obvious what nodes will be needed, so we do it just above.
*
* Worse, that means that the node may have been utilized while
* unpacking other resources, without our weight correction. The most
* likely place for this to happen is when pe__unpack_resource() calls
* resource_location() to set a default score in symmetric clusters.
* This adds a node *copy* to each resource's allowed nodes, and these
* copies will have the wrong weight.
*
* As a hacky workaround, fix those copies here.
*
* @TODO Possible alternative: ensure bundles are unpacked before other
* resources, so the weight is correct before any copies are made.
*/
g_list_foreach(scheduler->priv->resources,
(GFunc) disallow_node, (gpointer) uname);
replica->node = pe__copy_node(node);
replica->node->assign->score = 500;
replica->node->assign->probe_mode = pcmk__probe_exclusive;
/* Ensure the node shows up as allowed and with the correct discovery set */
if (replica->child->priv->allowed_nodes != NULL) {
g_hash_table_destroy(replica->child->priv->allowed_nodes);
}
replica->child->priv->allowed_nodes =
pcmk__strkey_table(NULL, pcmk__free_node_copy);
g_hash_table_insert(replica->child->priv->allowed_nodes,
(gpointer) replica->node->priv->id,
pe__copy_node(replica->node));
{
const pcmk_resource_t *parent = replica->child->priv->parent;
pcmk_node_t *copy = pe__copy_node(replica->node);
copy->assign->score = -PCMK_SCORE_INFINITY;
g_hash_table_insert(parent->priv->allowed_nodes,
(gpointer) replica->node->priv->id, copy);
}
if (pe__unpack_resource(xml_remote, &replica->remote, parent,
scheduler) != pcmk_rc_ok) {
return pcmk_rc_unpack_error;
}
g_hash_table_iter_init(&gIter, replica->remote->priv->allowed_nodes);
while (g_hash_table_iter_next(&gIter, NULL, (void **)&node)) {
if (pcmk__is_pacemaker_remote_node(node)) {
/* Remote resources can only run on 'normal' cluster node */
node->assign->score = -PCMK_SCORE_INFINITY;
}
}
replica->node->priv->remote = replica->remote;
// Ensure pcmk__is_guest_or_bundle_node() functions correctly
replica->remote->priv->launcher = replica->container;
/* A bundle's #kind is closer to "container" (guest node) than the
* "remote" set by pe_create_node().
*/
pcmk__insert_dup(replica->node->priv->attrs,
CRM_ATTR_KIND, "container");
/* One effect of this is that unpack_launcher() will add
* replica->remote to replica->container's launched resources, which
* will make pe__resource_contains_guest_node() true for
* replica->container.
*
* replica->child does NOT get added to replica->container's launched
* resources. The only noticeable effect if it did would be for its
* fail count to be taken into account when checking
* replica->container's migration threshold.
*/
parent->priv->children = g_list_append(parent->priv->children,
replica->remote);
}
return pcmk_rc_ok;
}
static int
create_replica_resources(pcmk_resource_t *parent,
pe__bundle_variant_data_t *data,
pcmk__bundle_replica_t *replica)
{
int rc = pcmk_rc_ok;
rc = create_container_resource(parent, data, replica);
if (rc != pcmk_rc_ok) {
return rc;
}
rc = create_ip_resource(parent, data, replica);
if (rc != pcmk_rc_ok) {
return rc;
}
rc = create_remote_resource(parent, data, replica);
if (rc != pcmk_rc_ok) {
return rc;
}
if ((replica->child != NULL) && (replica->ipaddr != NULL)) {
pcmk__insert_meta(replica->child->priv, "external-ip", replica->ipaddr);
}
if (replica->remote != NULL) {
/*
* Allow the remote connection resource to be allocated to a
* different node than the one on which the container is active.
*
* This makes it possible to have Pacemaker Remote nodes running
* containers with the remote executor inside in order to start
* services inside those containers.
*/
pcmk__set_rsc_flags(replica->remote, pcmk__rsc_remote_nesting_allowed);
}
return rc;
}
static void
mount_add(pe__bundle_variant_data_t *bundle_data, const char *source,
const char *target, const char *options, uint32_t flags)
{
pe__bundle_mount_t *mount = pcmk__assert_alloc(1,
sizeof(pe__bundle_mount_t));
mount->source = pcmk__str_copy(source);
mount->target = pcmk__str_copy(target);
mount->options = pcmk__str_copy(options);
mount->flags = flags;
bundle_data->mounts = g_list_append(bundle_data->mounts, mount);
}
static void
mount_free(pe__bundle_mount_t *mount)
{
free(mount->source);
free(mount->target);
free(mount->options);
free(mount);
}
static void
port_free(pe__bundle_port_t *port)
{
free(port->source);
free(port->target);
free(port);
}
static pcmk__bundle_replica_t *
replica_for_remote(pcmk_resource_t *remote)
{
pcmk_resource_t *top = remote;
pe__bundle_variant_data_t *bundle_data = NULL;
if (top == NULL) {
return NULL;
}
while (top->priv->parent != NULL) {
top = top->priv->parent;
}
get_bundle_variant_data(bundle_data, top);
for (GList *gIter = bundle_data->replicas; gIter != NULL;
gIter = gIter->next) {
pcmk__bundle_replica_t *replica = gIter->data;
if (replica->remote == remote) {
return replica;
}
}
CRM_LOG_ASSERT(FALSE);
return NULL;
}
bool
pe__bundle_needs_remote_name(pcmk_resource_t *rsc)
{
const char *value;
GHashTable *params = NULL;
if (rsc == NULL) {
return false;
}
// Use NULL node since pcmk__bundle_expand() uses that to set value
params = pe_rsc_params(rsc, NULL, rsc->priv->scheduler);
value = g_hash_table_lookup(params, PCMK_REMOTE_RA_ADDR);
return pcmk__str_eq(value, "#uname", pcmk__str_casei)
&& xml_contains_remote_node(rsc->priv->xml);
}
const char *
pe__add_bundle_remote_name(pcmk_resource_t *rsc, xmlNode *xml,
const char *field)
{
// REMOTE_CONTAINER_HACK: Allow remote nodes that start containers with pacemaker remote inside
pcmk_node_t *node = NULL;
pcmk__bundle_replica_t *replica = NULL;
if (!pe__bundle_needs_remote_name(rsc)) {
return NULL;
}
replica = replica_for_remote(rsc);
if (replica == NULL) {
return NULL;
}
node = replica->container->priv->assigned_node;
if (node == NULL) {
/* If it won't be running anywhere after the
* transition, go with where it's running now.
*/
node = pcmk__current_node(replica->container);
}
if(node == NULL) {
crm_trace("Cannot determine address for bundle connection %s", rsc->id);
return NULL;
}
crm_trace("Setting address for bundle connection %s to bundle host %s",
rsc->id, pcmk__node_name(node));
if(xml != NULL && field != NULL) {
crm_xml_add(xml, field, node->priv->name);
}
return node->priv->name;
}
#define pe__set_bundle_mount_flags(mount_xml, flags, flags_to_set) do { \
flags = pcmk__set_flags_as(__func__, __LINE__, LOG_TRACE, \
"Bundle mount", pcmk__xe_id(mount_xml), \
flags, (flags_to_set), #flags_to_set); \
} while (0)
gboolean
pe__unpack_bundle(pcmk_resource_t *rsc, pcmk_scheduler_t *scheduler)
{
const char *value = NULL;
xmlNode *xml_obj = NULL;
const xmlNode *xml_child = NULL;
xmlNode *xml_resource = NULL;
pe__bundle_variant_data_t *bundle_data = NULL;
bool need_log_mount = TRUE;
pcmk__assert(rsc != NULL);
pcmk__rsc_trace(rsc, "Processing resource %s...", rsc->id);
bundle_data = pcmk__assert_alloc(1, sizeof(pe__bundle_variant_data_t));
rsc->priv->variant_opaque = bundle_data;
bundle_data->prefix = strdup(rsc->id);
xml_obj = pcmk__xe_first_child(rsc->priv->xml, PCMK_XE_DOCKER, NULL,
NULL);
if (xml_obj != NULL) {
bundle_data->agent_type = PE__CONTAINER_AGENT_DOCKER;
}
if (xml_obj == NULL) {
xml_obj = pcmk__xe_first_child(rsc->priv->xml, PCMK_XE_PODMAN, NULL,
NULL);
if (xml_obj != NULL) {
bundle_data->agent_type = PE__CONTAINER_AGENT_PODMAN;
}
}
if (xml_obj == NULL) {
return FALSE;
}
// Use 0 for default, minimum, and invalid PCMK_XA_PROMOTED_MAX
value = crm_element_value(xml_obj, PCMK_XA_PROMOTED_MAX);
pcmk__scan_min_int(value, &bundle_data->promoted_max, 0);
/* Default replicas to PCMK_XA_PROMOTED_MAX if it was specified and 1
* otherwise
*/
value = crm_element_value(xml_obj, PCMK_XA_REPLICAS);
if ((value == NULL) && (bundle_data->promoted_max > 0)) {
bundle_data->nreplicas = bundle_data->promoted_max;
} else {
pcmk__scan_min_int(value, &bundle_data->nreplicas, 1);
}
/*
* Communication between containers on the same host via the
* floating IPs only works if the container is started with:
* --userland-proxy=false --ip-masq=false
*/
value = crm_element_value(xml_obj, PCMK_XA_REPLICAS_PER_HOST);
pcmk__scan_min_int(value, &bundle_data->nreplicas_per_host, 1);
if (bundle_data->nreplicas_per_host == 1) {
pcmk__clear_rsc_flags(rsc, pcmk__rsc_unique);
}
bundle_data->container_command =
crm_element_value_copy(xml_obj, PCMK_XA_RUN_COMMAND);
bundle_data->launcher_options = crm_element_value_copy(xml_obj,
PCMK_XA_OPTIONS);
bundle_data->image = crm_element_value_copy(xml_obj, PCMK_XA_IMAGE);
bundle_data->container_network = crm_element_value_copy(xml_obj,
PCMK_XA_NETWORK);
xml_obj = pcmk__xe_first_child(rsc->priv->xml, PCMK_XE_NETWORK, NULL,
NULL);
if(xml_obj) {
bundle_data->ip_range_start =
crm_element_value_copy(xml_obj, PCMK_XA_IP_RANGE_START);
bundle_data->host_netmask =
crm_element_value_copy(xml_obj, PCMK_XA_HOST_NETMASK);
bundle_data->host_network =
crm_element_value_copy(xml_obj, PCMK_XA_HOST_INTERFACE);
bundle_data->control_port =
crm_element_value_copy(xml_obj, PCMK_XA_CONTROL_PORT);
value = crm_element_value(xml_obj, PCMK_XA_ADD_HOST);
if (crm_str_to_boolean(value, &bundle_data->add_host) != 1) {
bundle_data->add_host = TRUE;
}
for (xml_child = pcmk__xe_first_child(xml_obj, PCMK_XE_PORT_MAPPING,
NULL, NULL);
xml_child != NULL;
xml_child = pcmk__xe_next(xml_child, PCMK_XE_PORT_MAPPING)) {
pe__bundle_port_t *port =
pcmk__assert_alloc(1, sizeof(pe__bundle_port_t));
port->source = crm_element_value_copy(xml_child, PCMK_XA_PORT);
if(port->source == NULL) {
port->source = crm_element_value_copy(xml_child, PCMK_XA_RANGE);
} else {
port->target = crm_element_value_copy(xml_child,
PCMK_XA_INTERNAL_PORT);
}
if(port->source != NULL && strlen(port->source) > 0) {
if(port->target == NULL) {
port->target = strdup(port->source);
}
bundle_data->ports = g_list_append(bundle_data->ports, port);
} else {
pcmk__config_err("Invalid " PCMK_XA_PORT " directive %s",
pcmk__xe_id(xml_child));
port_free(port);
}
}
}
xml_obj = pcmk__xe_first_child(rsc->priv->xml, PCMK_XE_STORAGE, NULL,
NULL);
for (xml_child = pcmk__xe_first_child(xml_obj, PCMK_XE_STORAGE_MAPPING,
NULL, NULL);
xml_child != NULL;
xml_child = pcmk__xe_next(xml_child, PCMK_XE_STORAGE_MAPPING)) {
const char *source = crm_element_value(xml_child, PCMK_XA_SOURCE_DIR);
const char *target = crm_element_value(xml_child, PCMK_XA_TARGET_DIR);
const char *options = crm_element_value(xml_child, PCMK_XA_OPTIONS);
int flags = pe__bundle_mount_none;
if (source == NULL) {
source = crm_element_value(xml_child, PCMK_XA_SOURCE_DIR_ROOT);
pe__set_bundle_mount_flags(xml_child, flags,
pe__bundle_mount_subdir);
}
if (source && target) {
mount_add(bundle_data, source, target, options, flags);
if (strcmp(target, "/var/log") == 0) {
need_log_mount = FALSE;
}
} else {
pcmk__config_err("Invalid mount directive %s",
pcmk__xe_id(xml_child));
}
}
xml_obj = pcmk__xe_first_child(rsc->priv->xml, PCMK_XE_PRIMITIVE, NULL,
NULL);
if (xml_obj && valid_network(bundle_data)) {
const char *suffix = NULL;
char *value = NULL;
xmlNode *xml_set = NULL;
xml_resource = pcmk__xe_create(NULL, PCMK_XE_CLONE);
/* @COMPAT We no longer use the <master> tag, but we need to keep it as
* part of the resource name, so that bundles don't restart in a rolling
* upgrade. (It also avoids needing to change regression tests.)
*/
suffix = (const char *) xml_resource->name;
if (bundle_data->promoted_max > 0) {
suffix = "master";
}
pcmk__xe_set_id(xml_resource, "%s-%s", bundle_data->prefix, suffix);
xml_set = pcmk__xe_create(xml_resource, PCMK_XE_META_ATTRIBUTES);
pcmk__xe_set_id(xml_set, "%s-%s-meta",
bundle_data->prefix, xml_resource->name);
crm_create_nvpair_xml(xml_set, NULL,
PCMK_META_ORDERED, PCMK_VALUE_TRUE);
value = pcmk__itoa(bundle_data->nreplicas);
crm_create_nvpair_xml(xml_set, NULL, PCMK_META_CLONE_MAX, value);
free(value);
value = pcmk__itoa(bundle_data->nreplicas_per_host);
crm_create_nvpair_xml(xml_set, NULL, PCMK_META_CLONE_NODE_MAX, value);
free(value);
crm_create_nvpair_xml(xml_set, NULL, PCMK_META_GLOBALLY_UNIQUE,
pcmk__btoa(bundle_data->nreplicas_per_host > 1));
if (bundle_data->promoted_max) {
crm_create_nvpair_xml(xml_set, NULL,
PCMK_META_PROMOTABLE, PCMK_VALUE_TRUE);
value = pcmk__itoa(bundle_data->promoted_max);
crm_create_nvpair_xml(xml_set, NULL, PCMK_META_PROMOTED_MAX, value);
free(value);
}
//crm_xml_add(xml_obj, PCMK_XA_ID, bundle_data->prefix);
pcmk__xml_copy(xml_resource, xml_obj);
} else if(xml_obj) {
pcmk__config_err("Cannot control %s inside %s without either "
PCMK_XA_IP_RANGE_START " or " PCMK_XA_CONTROL_PORT,
rsc->id, pcmk__xe_id(xml_obj));
return FALSE;
}
if(xml_resource) {
int lpc = 0;
GList *childIter = NULL;
pe__bundle_port_t *port = NULL;
GString *buffer = NULL;
if (pe__unpack_resource(xml_resource, &(bundle_data->child), rsc,
scheduler) != pcmk_rc_ok) {
return FALSE;
}
/* Currently, we always map the default authentication key location
* into the same location inside the container.
*
* Ideally, we would respect the host's PCMK_authkey_location, but:
* - it may be different on different nodes;
* - the actual connection will do extra checking to make sure the key
* file exists and is readable, that we can't do here on the DC
* - tools such as crm_resource and crm_simulate may not have the same
* environment variables as the cluster, causing operation digests to
* differ
*
* Always using the default location inside the container is fine,
* because we control the pacemaker_remote environment, and it avoids
* having to pass another environment variable to the container.
*
* @TODO A better solution may be to have only pacemaker_remote use the
* environment variable, and have the cluster nodes use a new
* cluster option for key location. This would introduce the limitation
* of the location being the same on all cluster nodes, but that's
* reasonable.
*/
mount_add(bundle_data, DEFAULT_REMOTE_KEY_LOCATION,
DEFAULT_REMOTE_KEY_LOCATION, NULL, pe__bundle_mount_none);
if (need_log_mount) {
mount_add(bundle_data, CRM_BUNDLE_DIR, "/var/log", NULL,
pe__bundle_mount_subdir);
}
port = pcmk__assert_alloc(1, sizeof(pe__bundle_port_t));
if(bundle_data->control_port) {
port->source = strdup(bundle_data->control_port);
} else {
/* If we wanted to respect PCMK_remote_port, we could use
* crm_default_remote_port() here and elsewhere in this file instead
* of DEFAULT_REMOTE_PORT.
*
* However, it gains nothing, since we control both the container
* environment and the connection resource parameters, and the user
* can use a different port if desired by setting
* PCMK_XA_CONTROL_PORT.
*/
port->source = pcmk__itoa(DEFAULT_REMOTE_PORT);
}
port->target = strdup(port->source);
bundle_data->ports = g_list_append(bundle_data->ports, port);
buffer = g_string_sized_new(1024);
for (childIter = bundle_data->child->priv->children;
childIter != NULL; childIter = childIter->next) {
pcmk__bundle_replica_t *replica = NULL;
replica = pcmk__assert_alloc(1, sizeof(pcmk__bundle_replica_t));
replica->child = childIter->data;
pcmk__set_rsc_flags(replica->child, pcmk__rsc_exclusive_probes);
replica->offset = lpc++;
// Ensure the child's notify gets set based on the underlying primitive's value
if (pcmk_is_set(replica->child->flags, pcmk__rsc_notify)) {
pcmk__set_rsc_flags(bundle_data->child, pcmk__rsc_notify);
}
allocate_ip(bundle_data, replica, buffer);
bundle_data->replicas = g_list_append(bundle_data->replicas,
replica);
bundle_data->attribute_target =
g_hash_table_lookup(replica->child->priv->meta,
PCMK_META_CONTAINER_ATTRIBUTE_TARGET);
}
bundle_data->container_host_options = g_string_free(buffer, FALSE);
if (bundle_data->attribute_target) {
pcmk__insert_dup(rsc->priv->meta,
PCMK_META_CONTAINER_ATTRIBUTE_TARGET,
bundle_data->attribute_target);
pcmk__insert_dup(bundle_data->child->priv->meta,
PCMK_META_CONTAINER_ATTRIBUTE_TARGET,
bundle_data->attribute_target);
}
} else {
// Just a naked container, no pacemaker-remote
GString *buffer = g_string_sized_new(1024);
for (int lpc = 0; lpc < bundle_data->nreplicas; lpc++) {
pcmk__bundle_replica_t *replica = NULL;
replica = pcmk__assert_alloc(1, sizeof(pcmk__bundle_replica_t));
replica->offset = lpc;
allocate_ip(bundle_data, replica, buffer);
bundle_data->replicas = g_list_append(bundle_data->replicas,
replica);
}
bundle_data->container_host_options = g_string_free(buffer, FALSE);
}
for (GList *gIter = bundle_data->replicas; gIter != NULL;
gIter = gIter->next) {
pcmk__bundle_replica_t *replica = gIter->data;
if (create_replica_resources(rsc, bundle_data, replica) != pcmk_rc_ok) {
pcmk__config_err("Failed unpacking resource %s", rsc->id);
rsc->priv->fns->free(rsc);
return FALSE;
}
/* Utilization needs special handling for bundles. It makes no sense for
* the inner primitive to have utilization, because it is tied
* one-to-one to the guest node created by the container resource -- and
* there's no way to set capacities for that guest node anyway.
*
* What the user really wants is to configure utilization for the
* container. However, the schema only allows utilization for
* primitives, and the container resource is implicit anyway, so the
* user can *only* configure utilization for the inner primitive. If
* they do, move the primitive's utilization values to the container.
*
* @TODO This means that bundles without an inner primitive can't have
* utilization. An alternative might be to allow utilization values in
* the top-level bundle XML in the schema, and copy those to each
* container.
*/
if (replica->child != NULL) {
GHashTable *empty = replica->container->priv->utilization;
replica->container->priv->utilization =
replica->child->priv->utilization;
replica->child->priv->utilization = empty;
}
}
if (bundle_data->child) {
rsc->priv->children = g_list_append(rsc->priv->children,
bundle_data->child);
}
return TRUE;
}
static int
replica_resource_active(pcmk_resource_t *rsc, gboolean all)
{
if (rsc) {
gboolean child_active = rsc->priv->fns->active(rsc, all);
if (child_active && !all) {
return TRUE;
} else if (!child_active && all) {
return FALSE;
}
}
return -1;
}
gboolean
pe__bundle_active(pcmk_resource_t *rsc, gboolean all)
{
pe__bundle_variant_data_t *bundle_data = NULL;
GList *iter = NULL;
get_bundle_variant_data(bundle_data, rsc);
for (iter = bundle_data->replicas; iter != NULL; iter = iter->next) {
pcmk__bundle_replica_t *replica = iter->data;
int rsc_active;
rsc_active = replica_resource_active(replica->ip, all);
if (rsc_active >= 0) {
return (gboolean) rsc_active;
}
rsc_active = replica_resource_active(replica->child, all);
if (rsc_active >= 0) {
return (gboolean) rsc_active;
}
rsc_active = replica_resource_active(replica->container, all);
if (rsc_active >= 0) {
return (gboolean) rsc_active;
}
rsc_active = replica_resource_active(replica->remote, all);
if (rsc_active >= 0) {
return (gboolean) rsc_active;
}
}
/* If "all" is TRUE, we've already checked that no resources were inactive,
* so return TRUE; if "all" is FALSE, we didn't find any active resources,
* so return FALSE.
*/
return all;
}
/*!
* \internal
* \brief Find the bundle replica corresponding to a given node
*
* \param[in] bundle Top-level bundle resource
* \param[in] node Node to search for
*
* \return Bundle replica if found, NULL otherwise
*/
pcmk_resource_t *
pe__find_bundle_replica(const pcmk_resource_t *bundle, const pcmk_node_t *node)
{
pe__bundle_variant_data_t *bundle_data = NULL;
pcmk__assert((bundle != NULL) && (node != NULL));
get_bundle_variant_data(bundle_data, bundle);
for (GList *gIter = bundle_data->replicas; gIter != NULL;
gIter = gIter->next) {
pcmk__bundle_replica_t *replica = gIter->data;
pcmk__assert((replica != NULL) && (replica->node != NULL));
if (pcmk__same_node(replica->node, node)) {
return replica->child;
}
}
return NULL;
}
PCMK__OUTPUT_ARGS("bundle", "uint32_t", "pcmk_resource_t *", "GList *",
"GList *")
int
pe__bundle_xml(pcmk__output_t *out, va_list args)
{
uint32_t show_opts = va_arg(args, uint32_t);
pcmk_resource_t *rsc = va_arg(args, pcmk_resource_t *);
GList *only_node = va_arg(args, GList *);
GList *only_rsc = va_arg(args, GList *);
pe__bundle_variant_data_t *bundle_data = NULL;
int rc = pcmk_rc_no_output;
gboolean printed_header = FALSE;
gboolean print_everything = TRUE;
const char *desc = NULL;
pcmk__assert(rsc != NULL);
get_bundle_variant_data(bundle_data, rsc);
if (rsc->priv->fns->is_filtered(rsc, only_rsc, TRUE)) {
return rc;
}
print_everything = pcmk__str_in_list(rsc->id, only_rsc, pcmk__str_star_matches);
for (GList *gIter = bundle_data->replicas; gIter != NULL;
gIter = gIter->next) {
pcmk__bundle_replica_t *replica = gIter->data;
pcmk_resource_t *ip = replica->ip;
pcmk_resource_t *child = replica->child;
pcmk_resource_t *container = replica->container;
pcmk_resource_t *remote = replica->remote;
char *id = NULL;
gboolean print_ip, print_child, print_ctnr, print_remote;
pcmk__assert(replica != NULL);
if (pcmk__rsc_filtered_by_node(container, only_node)) {
continue;
}
print_ip = (ip != NULL)
&& !ip->priv->fns->is_filtered(ip, only_rsc,
print_everything);
print_child = (child != NULL)
&& !child->priv->fns->is_filtered(child, only_rsc,
print_everything);
print_ctnr = !container->priv->fns->is_filtered(container, only_rsc,
print_everything);
print_remote = (remote != NULL)
&& !remote->priv->fns->is_filtered(remote, only_rsc,
print_everything);
if (!print_everything && !print_ip && !print_child && !print_ctnr && !print_remote) {
continue;
}
if (!printed_header) {
const char *type = container_agent_str(bundle_data->agent_type);
const char *unique = pcmk__flag_text(rsc->flags, pcmk__rsc_unique);
const char *maintenance = pcmk__flag_text(rsc->flags,
pcmk__rsc_maintenance);
const char *managed = pcmk__flag_text(rsc->flags,
pcmk__rsc_managed);
const char *failed = pcmk__flag_text(rsc->flags, pcmk__rsc_failed);
printed_header = TRUE;
desc = pe__resource_description(rsc, show_opts);
rc = pe__name_and_nvpairs_xml(out, true, PCMK_XE_BUNDLE,
PCMK_XA_ID, rsc->id,
PCMK_XA_TYPE, type,
PCMK_XA_IMAGE, bundle_data->image,
PCMK_XA_UNIQUE, unique,
PCMK_XA_MAINTENANCE, maintenance,
PCMK_XA_MANAGED, managed,
PCMK_XA_FAILED, failed,
PCMK_XA_DESCRIPTION, desc,
NULL);
pcmk__assert(rc == pcmk_rc_ok);
}
id = pcmk__itoa(replica->offset);
rc = pe__name_and_nvpairs_xml(out, true, PCMK_XE_REPLICA,
PCMK_XA_ID, id,
NULL);
free(id);
pcmk__assert(rc == pcmk_rc_ok);
if (print_ip) {
out->message(out, (const char *) ip->priv->xml->name, show_opts,
ip, only_node, only_rsc);
}
if (print_child) {
out->message(out, (const char *) child->priv->xml->name,
show_opts, child, only_node, only_rsc);
}
if (print_ctnr) {
out->message(out, (const char *) container->priv->xml->name,
show_opts, container, only_node, only_rsc);
}
if (print_remote) {
out->message(out, (const char *) remote->priv->xml->name,
show_opts, remote, only_node, only_rsc);
}
pcmk__output_xml_pop_parent(out); // replica
}
if (printed_header) {
pcmk__output_xml_pop_parent(out); // bundle
}
return rc;
}
static void
pe__bundle_replica_output_html(pcmk__output_t *out,
pcmk__bundle_replica_t *replica,
pcmk_node_t *node, uint32_t show_opts)
{
pcmk_resource_t *rsc = replica->child;
int offset = 0;
char buffer[LINE_MAX];
if(rsc == NULL) {
rsc = replica->container;
}
if (replica->remote) {
offset += snprintf(buffer + offset, LINE_MAX - offset, "%s",
rsc_printable_id(replica->remote));
} else {
offset += snprintf(buffer + offset, LINE_MAX - offset, "%s",
rsc_printable_id(replica->container));
}
if (replica->ipaddr) {
offset += snprintf(buffer + offset, LINE_MAX - offset, " (%s)",
replica->ipaddr);
}
pe__common_output_html(out, rsc, buffer, node, show_opts);
}
/*!
* \internal
* \brief Get a string describing a resource's unmanaged state or lack thereof
*
* \param[in] rsc Resource to describe
*
* \return A string indicating that a resource is in maintenance mode or
* otherwise unmanaged, or an empty string otherwise
*/
static const char *
get_unmanaged_str(const pcmk_resource_t *rsc)
{
if (pcmk_is_set(rsc->flags, pcmk__rsc_maintenance)) {
return " (maintenance)";
}
if (!pcmk_is_set(rsc->flags, pcmk__rsc_managed)) {
return " (unmanaged)";
}
return "";
}
PCMK__OUTPUT_ARGS("bundle", "uint32_t", "pcmk_resource_t *", "GList *",
"GList *")
int
pe__bundle_html(pcmk__output_t *out, va_list args)
{
uint32_t show_opts = va_arg(args, uint32_t);
pcmk_resource_t *rsc = va_arg(args, pcmk_resource_t *);
GList *only_node = va_arg(args, GList *);
GList *only_rsc = va_arg(args, GList *);
const char *desc = NULL;
pe__bundle_variant_data_t *bundle_data = NULL;
int rc = pcmk_rc_no_output;
gboolean print_everything = TRUE;
pcmk__assert(rsc != NULL);
get_bundle_variant_data(bundle_data, rsc);
desc = pe__resource_description(rsc, show_opts);
if (rsc->priv->fns->is_filtered(rsc, only_rsc, TRUE)) {
return rc;
}
print_everything = pcmk__str_in_list(rsc->id, only_rsc, pcmk__str_star_matches);
for (GList *gIter = bundle_data->replicas; gIter != NULL;
gIter = gIter->next) {
pcmk__bundle_replica_t *replica = gIter->data;
pcmk_resource_t *ip = replica->ip;
pcmk_resource_t *child = replica->child;
pcmk_resource_t *container = replica->container;
pcmk_resource_t *remote = replica->remote;
gboolean print_ip, print_child, print_ctnr, print_remote;
pcmk__assert(replica != NULL);
if (pcmk__rsc_filtered_by_node(container, only_node)) {
continue;
}
print_ip = (ip != NULL)
&& !ip->priv->fns->is_filtered(ip, only_rsc,
print_everything);
print_child = (child != NULL)
&& !child->priv->fns->is_filtered(child, only_rsc,
print_everything);
print_ctnr = !container->priv->fns->is_filtered(container, only_rsc,
print_everything);
print_remote = (remote != NULL)
&& !remote->priv->fns->is_filtered(remote, only_rsc,
print_everything);
if (pcmk_is_set(show_opts, pcmk_show_implicit_rscs) ||
(print_everything == FALSE && (print_ip || print_child || print_ctnr || print_remote))) {
/* The text output messages used below require pe_print_implicit to
* be set to do anything.
*/
uint32_t new_show_opts = show_opts | pcmk_show_implicit_rscs;
PCMK__OUTPUT_LIST_HEADER(out, FALSE, rc, "Container bundle%s: %s [%s]%s%s%s%s%s",
(bundle_data->nreplicas > 1)? " set" : "",
rsc->id, bundle_data->image,
pcmk_is_set(rsc->flags, pcmk__rsc_unique)? " (unique)" : "",
desc ? " (" : "", desc ? desc : "", desc ? ")" : "",
get_unmanaged_str(rsc));
if (pcmk__list_of_multiple(bundle_data->replicas)) {
out->begin_list(out, NULL, NULL, "Replica[%d]", replica->offset);
}
if (print_ip) {
out->message(out, (const char *) ip->priv->xml->name,
new_show_opts, ip, only_node, only_rsc);
}
if (print_child) {
out->message(out, (const char *) child->priv->xml->name,
new_show_opts, child, only_node, only_rsc);
}
if (print_ctnr) {
out->message(out, (const char *) container->priv->xml->name,
new_show_opts, container, only_node, only_rsc);
}
if (print_remote) {
out->message(out, (const char *) remote->priv->xml->name,
new_show_opts, remote, only_node, only_rsc);
}
if (pcmk__list_of_multiple(bundle_data->replicas)) {
out->end_list(out);
}
} else if (print_everything == FALSE && !(print_ip || print_child || print_ctnr || print_remote)) {
continue;
} else {
PCMK__OUTPUT_LIST_HEADER(out, FALSE, rc, "Container bundle%s: %s [%s]%s%s%s%s%s",
(bundle_data->nreplicas > 1)? " set" : "",
rsc->id, bundle_data->image,
pcmk_is_set(rsc->flags, pcmk__rsc_unique)? " (unique)" : "",
desc ? " (" : "", desc ? desc : "", desc ? ")" : "",
get_unmanaged_str(rsc));
pe__bundle_replica_output_html(out, replica,
pcmk__current_node(container),
show_opts);
}
}
PCMK__OUTPUT_LIST_FOOTER(out, rc);
return rc;
}
static void
pe__bundle_replica_output_text(pcmk__output_t *out,
pcmk__bundle_replica_t *replica,
pcmk_node_t *node, uint32_t show_opts)
{
const pcmk_resource_t *rsc = replica->child;
int offset = 0;
char buffer[LINE_MAX];
if(rsc == NULL) {
rsc = replica->container;
}
if (replica->remote) {
offset += snprintf(buffer + offset, LINE_MAX - offset, "%s",
rsc_printable_id(replica->remote));
} else {
offset += snprintf(buffer + offset, LINE_MAX - offset, "%s",
rsc_printable_id(replica->container));
}
if (replica->ipaddr) {
offset += snprintf(buffer + offset, LINE_MAX - offset, " (%s)",
replica->ipaddr);
}
pe__common_output_text(out, rsc, buffer, node, show_opts);
}
PCMK__OUTPUT_ARGS("bundle", "uint32_t", "pcmk_resource_t *", "GList *",
"GList *")
int
pe__bundle_text(pcmk__output_t *out, va_list args)
{
uint32_t show_opts = va_arg(args, uint32_t);
pcmk_resource_t *rsc = va_arg(args, pcmk_resource_t *);
GList *only_node = va_arg(args, GList *);
GList *only_rsc = va_arg(args, GList *);
const char *desc = NULL;
pe__bundle_variant_data_t *bundle_data = NULL;
int rc = pcmk_rc_no_output;
gboolean print_everything = TRUE;
desc = pe__resource_description(rsc, show_opts);
pcmk__assert(rsc != NULL);
get_bundle_variant_data(bundle_data, rsc);
if (rsc->priv->fns->is_filtered(rsc, only_rsc, TRUE)) {
return rc;
}
print_everything = pcmk__str_in_list(rsc->id, only_rsc, pcmk__str_star_matches);
for (GList *gIter = bundle_data->replicas; gIter != NULL;
gIter = gIter->next) {
pcmk__bundle_replica_t *replica = gIter->data;
pcmk_resource_t *ip = replica->ip;
pcmk_resource_t *child = replica->child;
pcmk_resource_t *container = replica->container;
pcmk_resource_t *remote = replica->remote;
gboolean print_ip, print_child, print_ctnr, print_remote;
pcmk__assert(replica != NULL);
if (pcmk__rsc_filtered_by_node(container, only_node)) {
continue;
}
print_ip = (ip != NULL)
&& !ip->priv->fns->is_filtered(ip, only_rsc,
print_everything);
print_child = (child != NULL)
&& !child->priv->fns->is_filtered(child, only_rsc,
print_everything);
print_ctnr = !container->priv->fns->is_filtered(container, only_rsc,
print_everything);
print_remote = (remote != NULL)
&& !remote->priv->fns->is_filtered(remote, only_rsc,
print_everything);
if (pcmk_is_set(show_opts, pcmk_show_implicit_rscs) ||
(print_everything == FALSE && (print_ip || print_child || print_ctnr || print_remote))) {
/* The text output messages used below require pe_print_implicit to
* be set to do anything.
*/
uint32_t new_show_opts = show_opts | pcmk_show_implicit_rscs;
PCMK__OUTPUT_LIST_HEADER(out, FALSE, rc, "Container bundle%s: %s [%s]%s%s%s%s%s",
(bundle_data->nreplicas > 1)? " set" : "",
rsc->id, bundle_data->image,
pcmk_is_set(rsc->flags, pcmk__rsc_unique)? " (unique)" : "",
desc ? " (" : "", desc ? desc : "", desc ? ")" : "",
get_unmanaged_str(rsc));
if (pcmk__list_of_multiple(bundle_data->replicas)) {
out->list_item(out, NULL, "Replica[%d]", replica->offset);
}
out->begin_list(out, NULL, NULL, NULL);
if (print_ip) {
out->message(out, (const char *) ip->priv->xml->name,
new_show_opts, ip, only_node, only_rsc);
}
if (print_child) {
out->message(out, (const char *) child->priv->xml->name,
new_show_opts, child, only_node, only_rsc);
}
if (print_ctnr) {
out->message(out, (const char *) container->priv->xml->name,
new_show_opts, container, only_node, only_rsc);
}
if (print_remote) {
out->message(out, (const char *) remote->priv->xml->name,
new_show_opts, remote, only_node, only_rsc);
}
out->end_list(out);
} else if (print_everything == FALSE && !(print_ip || print_child || print_ctnr || print_remote)) {
continue;
} else {
PCMK__OUTPUT_LIST_HEADER(out, FALSE, rc, "Container bundle%s: %s [%s]%s%s%s%s%s",
(bundle_data->nreplicas > 1)? " set" : "",
rsc->id, bundle_data->image,
pcmk_is_set(rsc->flags, pcmk__rsc_unique)? " (unique)" : "",
desc ? " (" : "", desc ? desc : "", desc ? ")" : "",
get_unmanaged_str(rsc));
pe__bundle_replica_output_text(out, replica,
pcmk__current_node(container),
show_opts);
}
}
PCMK__OUTPUT_LIST_FOOTER(out, rc);
return rc;
}
static void
free_bundle_replica(pcmk__bundle_replica_t *replica)
{
if (replica == NULL) {
return;
}
pcmk__free_node_copy(replica->node);
replica->node = NULL;
if (replica->ip) {
pcmk__xml_free(replica->ip->priv->xml);
replica->ip->priv->xml = NULL;
replica->ip->priv->fns->free(replica->ip);
}
if (replica->container) {
pcmk__xml_free(replica->container->priv->xml);
replica->container->priv->xml = NULL;
replica->container->priv->fns->free(replica->container);
}
if (replica->remote) {
pcmk__xml_free(replica->remote->priv->xml);
replica->remote->priv->xml = NULL;
replica->remote->priv->fns->free(replica->remote);
}
free(replica->ipaddr);
free(replica);
}
void
pe__free_bundle(pcmk_resource_t *rsc)
{
pe__bundle_variant_data_t *bundle_data = NULL;
CRM_CHECK(rsc != NULL, return);
get_bundle_variant_data(bundle_data, rsc);
pcmk__rsc_trace(rsc, "Freeing %s", rsc->id);
free(bundle_data->prefix);
free(bundle_data->image);
free(bundle_data->control_port);
free(bundle_data->host_network);
free(bundle_data->host_netmask);
free(bundle_data->ip_range_start);
free(bundle_data->container_network);
free(bundle_data->launcher_options);
free(bundle_data->container_command);
g_free(bundle_data->container_host_options);
g_list_free_full(bundle_data->replicas,
(GDestroyNotify) free_bundle_replica);
g_list_free_full(bundle_data->mounts, (GDestroyNotify)mount_free);
g_list_free_full(bundle_data->ports, (GDestroyNotify)port_free);
g_list_free(rsc->priv->children);
if(bundle_data->child) {
pcmk__xml_free(bundle_data->child->priv->xml);
bundle_data->child->priv->xml = NULL;
bundle_data->child->priv->fns->free(bundle_data->child);
}
common_free(rsc);
}
enum rsc_role_e
pe__bundle_resource_state(const pcmk_resource_t *rsc, gboolean current)
{
enum rsc_role_e container_role = pcmk_role_unknown;
return container_role;
}
/*!
* \brief Get the number of configured replicas in a bundle
*
* \param[in] rsc Bundle resource
*
* \return Number of configured replicas, or 0 on error
*/
int
pe_bundle_replicas(const pcmk_resource_t *rsc)
{
if (pcmk__is_bundle(rsc)) {
pe__bundle_variant_data_t *bundle_data = NULL;
get_bundle_variant_data(bundle_data, rsc);
return bundle_data->nreplicas;
}
return 0;
}
void
pe__count_bundle(pcmk_resource_t *rsc)
{
pe__bundle_variant_data_t *bundle_data = NULL;
get_bundle_variant_data(bundle_data, rsc);
for (GList *item = bundle_data->replicas; item != NULL; item = item->next) {
pcmk__bundle_replica_t *replica = item->data;
if (replica->ip) {
replica->ip->priv->fns->count(replica->ip);
}
if (replica->child) {
replica->child->priv->fns->count(replica->child);
}
if (replica->container) {
replica->container->priv->fns->count(replica->container);
}
if (replica->remote) {
replica->remote->priv->fns->count(replica->remote);
}
}
}
gboolean
pe__bundle_is_filtered(const pcmk_resource_t *rsc, GList *only_rsc,
gboolean check_parent)
{
gboolean passes = FALSE;
pe__bundle_variant_data_t *bundle_data = NULL;
if (pcmk__str_in_list(rsc_printable_id(rsc), only_rsc, pcmk__str_star_matches)) {
passes = TRUE;
} else {
get_bundle_variant_data(bundle_data, rsc);
for (GList *gIter = bundle_data->replicas; gIter != NULL; gIter = gIter->next) {
pcmk__bundle_replica_t *replica = gIter->data;
pcmk_resource_t *ip = replica->ip;
pcmk_resource_t *child = replica->child;
pcmk_resource_t *container = replica->container;
pcmk_resource_t *remote = replica->remote;
if ((ip != NULL)
&& !ip->priv->fns->is_filtered(ip, only_rsc, FALSE)) {
passes = TRUE;
break;
}
if ((child != NULL)
&& !child->priv->fns->is_filtered(child, only_rsc, FALSE)) {
passes = TRUE;
break;
}
if (!container->priv->fns->is_filtered(container, only_rsc,
FALSE)) {
passes = TRUE;
break;
}
if ((remote != NULL)
&& !remote->priv->fns->is_filtered(remote, only_rsc, FALSE)) {
passes = TRUE;
break;
}
}
}
return !passes;
}
/*!
* \internal
* \brief Get a list of a bundle's containers
*
* \param[in] bundle Bundle resource
*
* \return Newly created list of \p bundle's containers
* \note It is the caller's responsibility to free the result with
* g_list_free().
*/
GList *
pe__bundle_containers(const pcmk_resource_t *bundle)
{
+ /* @TODO It would be more efficient to do this once when unpacking the
+ * bundle, creating a new GList* in the variant data
+ */
GList *containers = NULL;
const pe__bundle_variant_data_t *data = NULL;
get_bundle_variant_data(data, bundle);
for (GList *iter = data->replicas; iter != NULL; iter = iter->next) {
pcmk__bundle_replica_t *replica = iter->data;
containers = g_list_append(containers, replica->container);
}
return containers;
}
// Bundle implementation of pcmk__rsc_methods_t:active_node()
pcmk_node_t *
pe__bundle_active_node(const pcmk_resource_t *rsc, unsigned int *count_all,
unsigned int *count_clean)
{
pcmk_node_t *active = NULL;
pcmk_node_t *node = NULL;
pcmk_resource_t *container = NULL;
GList *containers = NULL;
GList *iter = NULL;
GHashTable *nodes = NULL;
const pe__bundle_variant_data_t *data = NULL;
if (count_all != NULL) {
*count_all = 0;
}
if (count_clean != NULL) {
*count_clean = 0;
}
if (rsc == NULL) {
return NULL;
}
/* For the purposes of this method, we only care about where the bundle's
* containers are active, so build a list of active containers.
*/
get_bundle_variant_data(data, rsc);
for (iter = data->replicas; iter != NULL; iter = iter->next) {
pcmk__bundle_replica_t *replica = iter->data;
if (replica->container->priv->active_nodes != NULL) {
containers = g_list_append(containers, replica->container);
}
}
if (containers == NULL) {
return NULL;
}
/* If the bundle has only a single active container, just use that
* container's method. If live migration is ever supported for bundle
* containers, this will allow us to prefer the migration source when there
* is only one container and it is migrating. For now, this just lets us
* avoid creating the nodes table.
*/
if (pcmk__list_of_1(containers)) {
container = containers->data;
node = container->priv->fns->active_node(container, count_all,
count_clean);
g_list_free(containers);
return node;
}
// Add all containers' active nodes to a hash table (for uniqueness)
nodes = g_hash_table_new(NULL, NULL);
for (iter = containers; iter != NULL; iter = iter->next) {
container = iter->data;
for (GList *node_iter = container->priv->active_nodes;
node_iter != NULL; node_iter = node_iter->next) {
node = node_iter->data;
// If insert returns true, we haven't counted this node yet
if (g_hash_table_insert(nodes, (gpointer) node->details,
(gpointer) node)
&& !pe__count_active_node(rsc, node, &active, count_all,
count_clean)) {
goto done;
}
}
}
done:
g_list_free(containers);
g_hash_table_destroy(nodes);
return active;
}
/*!
* \internal
* \brief Get maximum bundle resource instances per node
*
* \param[in] rsc Bundle resource to check
*
* \return Maximum number of \p rsc instances that can be active on one node
*/
unsigned int
pe__bundle_max_per_node(const pcmk_resource_t *rsc)
{
pe__bundle_variant_data_t *bundle_data = NULL;
get_bundle_variant_data(bundle_data, rsc);
pcmk__assert(bundle_data->nreplicas_per_host >= 0);
return (unsigned int) bundle_data->nreplicas_per_host;
}
diff --git a/lib/pengine/clone.c b/lib/pengine/clone.c
index 24953cd131..f8db1db221 100644
--- a/lib/pengine/clone.c
+++ b/lib/pengine/clone.c
@@ -1,1256 +1,1258 @@
/*
* Copyright 2004-2024 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU Lesser General Public License
* version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY.
*/
#include <crm_internal.h>
#include <stdint.h>
#include <crm/pengine/status.h>
#include <crm/pengine/internal.h>
#include <pe_status_private.h>
#include <crm/common/xml.h>
#include <crm/common/output.h>
#include <crm/common/xml_internal.h>
#include <crm/common/scheduler_internal.h>
typedef struct clone_variant_data_s {
int clone_max;
int clone_node_max;
int promoted_max;
int promoted_node_max;
int total_clones;
uint32_t flags; // Group of enum pcmk__clone_flags
notify_data_t *stop_notify;
notify_data_t *start_notify;
notify_data_t *demote_notify;
notify_data_t *promote_notify;
xmlNode *xml_obj_child;
} clone_variant_data_t;
#define get_clone_variant_data(data, rsc) do { \
pcmk__assert(pcmk__is_clone(rsc)); \
data = rsc->priv->variant_opaque; \
} while (0)
/*!
* \internal
* \brief Return the maximum number of clone instances allowed to be run
*
* \param[in] clone Clone or clone instance to check
*
* \return Maximum instances for \p clone
*/
int
pe__clone_max(const pcmk_resource_t *clone)
{
const clone_variant_data_t *clone_data = NULL;
get_clone_variant_data(clone_data, pe__const_top_resource(clone, false));
return clone_data->clone_max;
}
/*!
* \internal
* \brief Return the maximum number of clone instances allowed per node
*
* \param[in] clone Promotable clone or clone instance to check
*
* \return Maximum allowed instances per node for \p clone
*/
int
pe__clone_node_max(const pcmk_resource_t *clone)
{
const clone_variant_data_t *clone_data = NULL;
get_clone_variant_data(clone_data, pe__const_top_resource(clone, false));
return clone_data->clone_node_max;
}
/*!
* \internal
* \brief Return the maximum number of clone instances allowed to be promoted
*
* \param[in] clone Promotable clone or clone instance to check
*
* \return Maximum promoted instances for \p clone
*/
int
pe__clone_promoted_max(const pcmk_resource_t *clone)
{
clone_variant_data_t *clone_data = NULL;
get_clone_variant_data(clone_data, pe__const_top_resource(clone, false));
return clone_data->promoted_max;
}
/*!
* \internal
* \brief Return the maximum number of clone instances allowed to be promoted
*
* \param[in] clone Promotable clone or clone instance to check
*
* \return Maximum promoted instances for \p clone
*/
int
pe__clone_promoted_node_max(const pcmk_resource_t *clone)
{
clone_variant_data_t *clone_data = NULL;
get_clone_variant_data(clone_data, pe__const_top_resource(clone, false));
return clone_data->promoted_node_max;
}
static GList *
sorted_hash_table_values(GHashTable *table)
{
GList *retval = NULL;
GHashTableIter iter;
gpointer key, value;
g_hash_table_iter_init(&iter, table);
while (g_hash_table_iter_next(&iter, &key, &value)) {
if (!g_list_find_custom(retval, value, (GCompareFunc) strcmp)) {
retval = g_list_prepend(retval, (char *) value);
}
}
retval = g_list_sort(retval, (GCompareFunc) strcmp);
return retval;
}
static GList *
nodes_with_status(GHashTable *table, const char *status)
{
GList *retval = NULL;
GHashTableIter iter;
gpointer key, value;
g_hash_table_iter_init(&iter, table);
while (g_hash_table_iter_next(&iter, &key, &value)) {
if (!strcmp((char *) value, status)) {
retval = g_list_prepend(retval, key);
}
}
retval = g_list_sort(retval, (GCompareFunc) pcmk__numeric_strcasecmp);
return retval;
}
static GString *
node_list_to_str(const GList *list)
{
GString *retval = NULL;
for (const GList *iter = list; iter != NULL; iter = iter->next) {
pcmk__add_word(&retval, 1024, (const char *) iter->data);
}
return retval;
}
static void
clone_header(pcmk__output_t *out, int *rc, const pcmk_resource_t *rsc,
clone_variant_data_t *clone_data, const char *desc)
{
GString *attrs = NULL;
if (pcmk_is_set(rsc->flags, pcmk__rsc_promotable)) {
pcmk__add_separated_word(&attrs, 64, "promotable", ", ");
}
if (pcmk_is_set(rsc->flags, pcmk__rsc_unique)) {
pcmk__add_separated_word(&attrs, 64, "unique", ", ");
}
if (pe__resource_is_disabled(rsc)) {
pcmk__add_separated_word(&attrs, 64, "disabled", ", ");
}
if (pcmk_is_set(rsc->flags, pcmk__rsc_maintenance)) {
pcmk__add_separated_word(&attrs, 64, "maintenance", ", ");
} else if (!pcmk_is_set(rsc->flags, pcmk__rsc_managed)) {
pcmk__add_separated_word(&attrs, 64, "unmanaged", ", ");
}
if (attrs != NULL) {
PCMK__OUTPUT_LIST_HEADER(out, FALSE, *rc, "Clone Set: %s [%s] (%s)%s%s%s",
rsc->id,
pcmk__xe_id(clone_data->xml_obj_child),
(const char *) attrs->str, desc ? " (" : "",
desc ? desc : "", desc ? ")" : "");
g_string_free(attrs, TRUE);
} else {
PCMK__OUTPUT_LIST_HEADER(out, FALSE, *rc, "Clone Set: %s [%s]%s%s%s",
rsc->id,
pcmk__xe_id(clone_data->xml_obj_child),
desc ? " (" : "", desc ? desc : "",
desc ? ")" : "");
}
}
void
pe__force_anon(const char *standard, pcmk_resource_t *rsc, const char *rid,
pcmk_scheduler_t *scheduler)
{
if (pcmk__is_clone(rsc)) {
clone_variant_data_t *clone_data = rsc->priv->variant_opaque;
pcmk__config_warn("Ignoring " PCMK_META_GLOBALLY_UNIQUE " for %s "
"because %s resources such as %s can be used only as "
"anonymous clones", rsc->id, standard, rid);
clone_data->clone_node_max = 1;
clone_data->clone_max = QB_MIN(clone_data->clone_max,
g_list_length(scheduler->nodes));
}
}
pcmk_resource_t *
pe__create_clone_child(pcmk_resource_t *rsc, pcmk_scheduler_t *scheduler)
{
gboolean as_orphan = FALSE;
char *inc_num = NULL;
char *inc_max = NULL;
pcmk_resource_t *child_rsc = NULL;
xmlNode *child_copy = NULL;
clone_variant_data_t *clone_data = NULL;
get_clone_variant_data(clone_data, rsc);
CRM_CHECK(clone_data->xml_obj_child != NULL, return FALSE);
if (clone_data->total_clones >= clone_data->clone_max) {
// If we've already used all available instances, this is an orphan
as_orphan = TRUE;
}
// Allocate instance numbers in numerical order (starting at 0)
inc_num = pcmk__itoa(clone_data->total_clones);
inc_max = pcmk__itoa(clone_data->clone_max);
child_copy = pcmk__xml_copy(NULL, clone_data->xml_obj_child);
crm_xml_add(child_copy, PCMK__META_CLONE, inc_num);
if (pe__unpack_resource(child_copy, &child_rsc, rsc,
scheduler) != pcmk_rc_ok) {
goto bail;
}
/* child_rsc->globally_unique = rsc->globally_unique; */
pcmk__assert(child_rsc != NULL);
clone_data->total_clones += 1;
pcmk__rsc_trace(child_rsc, "Setting clone attributes for: %s",
child_rsc->id);
rsc->priv->children = g_list_append(rsc->priv->children, child_rsc);
if (as_orphan) {
pe__set_resource_flags_recursive(child_rsc, pcmk__rsc_removed);
}
pcmk__insert_meta(child_rsc->priv, PCMK_META_CLONE_MAX, inc_max);
pcmk__rsc_trace(rsc, "Added %s instance %s", rsc->id, child_rsc->id);
bail:
free(inc_num);
free(inc_max);
return child_rsc;
}
/*!
* \internal
* \brief Unpack a nonnegative integer value from a resource meta-attribute
*
* \param[in] rsc Resource with meta-attribute
* \param[in] meta_name Name of meta-attribute to unpack
* \param[in] deprecated_name If not NULL, try unpacking this
* if \p meta_name is unset
* \param[in] default_value Value to use if unset
*
* \return Integer parsed from resource's specified meta-attribute if a valid
* nonnegative integer, \p default_value if unset, or 0 if invalid
*/
static int
unpack_meta_int(const pcmk_resource_t *rsc, const char *meta_name,
const char *deprecated_name, int default_value)
{
int integer = default_value;
const char *value = g_hash_table_lookup(rsc->priv->meta, meta_name);
if ((value == NULL) && (deprecated_name != NULL)) {
value = g_hash_table_lookup(rsc->priv->meta, deprecated_name);
if (value != NULL) {
if (pcmk__str_eq(deprecated_name, PCMK__META_PROMOTED_MAX_LEGACY,
pcmk__str_none)) {
pcmk__warn_once(pcmk__wo_clone_master_max,
"Support for the " PCMK__META_PROMOTED_MAX_LEGACY
" meta-attribute (such as in %s) is deprecated "
"and will be removed in a future release. Use the "
PCMK_META_PROMOTED_MAX " meta-attribute instead.",
rsc->id);
} else if (pcmk__str_eq(deprecated_name, PCMK__META_PROMOTED_NODE_MAX_LEGACY,
pcmk__str_none)) {
pcmk__warn_once(pcmk__wo_clone_master_node_max,
"Support for the " PCMK__META_PROMOTED_NODE_MAX_LEGACY
" meta-attribute (such as in %s) is deprecated "
"and will be removed in a future release. Use the "
PCMK_META_PROMOTED_NODE_MAX " meta-attribute instead.",
rsc->id);
}
}
}
if (value != NULL) {
pcmk__scan_min_int(value, &integer, 0);
}
return integer;
}
gboolean
clone_unpack(pcmk_resource_t *rsc, pcmk_scheduler_t *scheduler)
{
int lpc = 0;
xmlNode *a_child = NULL;
xmlNode *xml_obj = rsc->priv->xml;
clone_variant_data_t *clone_data = NULL;
pcmk__rsc_trace(rsc, "Processing resource %s...", rsc->id);
clone_data = pcmk__assert_alloc(1, sizeof(clone_variant_data_t));
rsc->priv->variant_opaque = clone_data;
if (pcmk_is_set(rsc->flags, pcmk__rsc_promotable)) {
// Use 1 as default but 0 for minimum and invalid
// @COMPAT PCMK__META_PROMOTED_MAX_LEGACY deprecated since 2.0.0
clone_data->promoted_max =
unpack_meta_int(rsc, PCMK_META_PROMOTED_MAX,
PCMK__META_PROMOTED_MAX_LEGACY, 1);
// Use 1 as default but 0 for minimum and invalid
// @COMPAT PCMK__META_PROMOTED_NODE_MAX_LEGACY deprecated since 2.0.0
clone_data->promoted_node_max =
unpack_meta_int(rsc, PCMK_META_PROMOTED_NODE_MAX,
PCMK__META_PROMOTED_NODE_MAX_LEGACY, 1);
}
// Use 1 as default but 0 for minimum and invalid
clone_data->clone_node_max = unpack_meta_int(rsc, PCMK_META_CLONE_NODE_MAX,
NULL, 1);
/* Use number of nodes (but always at least 1, which is handy for crm_verify
* for a CIB without nodes) as default, but 0 for minimum and invalid
+ *
+ * @TODO Exclude bundle nodes when counting
*/
clone_data->clone_max = unpack_meta_int(rsc, PCMK_META_CLONE_MAX, NULL,
QB_MAX(1, g_list_length(scheduler->nodes)));
if (crm_is_true(g_hash_table_lookup(rsc->priv->meta,
PCMK_META_ORDERED))) {
clone_data->flags = pcmk__set_flags_as(__func__, __LINE__, LOG_TRACE,
"Clone", rsc->id,
clone_data->flags,
pcmk__clone_ordered,
"pcmk__clone_ordered");
}
if (!pcmk_is_set(rsc->flags, pcmk__rsc_unique)
&& (clone_data->clone_node_max > 1)) {
pcmk__config_err("Ignoring " PCMK_META_CLONE_NODE_MAX " of %d for %s "
"because anonymous clones support only one instance "
"per node", clone_data->clone_node_max, rsc->id);
clone_data->clone_node_max = 1;
}
pcmk__rsc_trace(rsc, "Options for %s", rsc->id);
pcmk__rsc_trace(rsc, "\tClone max: %d", clone_data->clone_max);
pcmk__rsc_trace(rsc, "\tClone node max: %d", clone_data->clone_node_max);
pcmk__rsc_trace(rsc, "\tClone is unique: %s",
pcmk__flag_text(rsc->flags, pcmk__rsc_unique));
pcmk__rsc_trace(rsc, "\tClone is promotable: %s",
pcmk__flag_text(rsc->flags, pcmk__rsc_promotable));
// Clones may contain a single group or primitive
for (a_child = pcmk__xe_first_child(xml_obj, NULL, NULL, NULL);
a_child != NULL; a_child = pcmk__xe_next(a_child, NULL)) {
if (pcmk__str_any_of((const char *) a_child->name,
PCMK_XE_PRIMITIVE, PCMK_XE_GROUP, NULL)) {
clone_data->xml_obj_child = a_child;
break;
}
}
if (clone_data->xml_obj_child == NULL) {
pcmk__config_err("%s has nothing to clone", rsc->id);
return FALSE;
}
/*
* Make clones ever so slightly sticky by default
*
* This helps ensure clone instances are not shuffled around the cluster
* for no benefit in situations when pre-allocation is not appropriate
*/
if (g_hash_table_lookup(rsc->priv->meta,
PCMK_META_RESOURCE_STICKINESS) == NULL) {
pcmk__insert_meta(rsc->priv, PCMK_META_RESOURCE_STICKINESS, "1");
}
/* This ensures that the PCMK_META_GLOBALLY_UNIQUE value always exists for
* children to inherit when being unpacked, as well as in resource agents'
* environment.
*/
pcmk__insert_meta(rsc->priv, PCMK_META_GLOBALLY_UNIQUE,
pcmk__flag_text(rsc->flags, pcmk__rsc_unique));
if (clone_data->clone_max <= 0) {
/* Create one child instance so that unpack_find_resource() will hook up
* any orphans up to the parent correctly.
*/
if (pe__create_clone_child(rsc, scheduler) == NULL) {
return FALSE;
}
} else {
// Create a child instance for each available instance number
for (lpc = 0; lpc < clone_data->clone_max; lpc++) {
if (pe__create_clone_child(rsc, scheduler) == NULL) {
return FALSE;
}
}
}
pcmk__rsc_trace(rsc, "Added %d children to resource %s...",
clone_data->clone_max, rsc->id);
return TRUE;
}
gboolean
clone_active(pcmk_resource_t * rsc, gboolean all)
{
for (GList *gIter = rsc->priv->children;
gIter != NULL; gIter = gIter->next) {
pcmk_resource_t *child_rsc = (pcmk_resource_t *) gIter->data;
gboolean child_active = child_rsc->priv->fns->active(child_rsc, all);
if (all == FALSE && child_active) {
return TRUE;
} else if (all && child_active == FALSE) {
return FALSE;
}
}
if (all) {
return TRUE;
} else {
return FALSE;
}
}
static const char *
configured_role_str(pcmk_resource_t * rsc)
{
const char *target_role = g_hash_table_lookup(rsc->priv->meta,
PCMK_META_TARGET_ROLE);
if ((target_role == NULL) && (rsc->priv->children != NULL)) {
// Any instance will do
pcmk_resource_t *instance = rsc->priv->children->data;
target_role = g_hash_table_lookup(instance->priv->meta,
PCMK_META_TARGET_ROLE);
}
return target_role;
}
static enum rsc_role_e
configured_role(pcmk_resource_t *rsc)
{
enum rsc_role_e role = pcmk_role_unknown;
const char *target_role = configured_role_str(rsc);
if (target_role != NULL) {
role = pcmk_parse_role(target_role);
if (role == pcmk_role_unknown) {
pcmk__config_err("Invalid " PCMK_META_TARGET_ROLE
" for resource %s", rsc->id);
}
}
return role;
}
bool
is_set_recursive(const pcmk_resource_t *rsc, long long flag, bool any)
{
bool all = !any;
if (pcmk_is_set(rsc->flags, flag)) {
if(any) {
return TRUE;
}
} else if(all) {
return FALSE;
}
for (GList *gIter = rsc->priv->children;
gIter != NULL; gIter = gIter->next) {
if(is_set_recursive(gIter->data, flag, any)) {
if(any) {
return TRUE;
}
} else if(all) {
return FALSE;
}
}
if(all) {
return TRUE;
}
return FALSE;
}
PCMK__OUTPUT_ARGS("clone", "uint32_t", "pcmk_resource_t *", "GList *",
"GList *")
int
pe__clone_xml(pcmk__output_t *out, va_list args)
{
uint32_t show_opts = va_arg(args, uint32_t);
pcmk_resource_t *rsc = va_arg(args, pcmk_resource_t *);
GList *only_node = va_arg(args, GList *);
GList *only_rsc = va_arg(args, GList *);
GList *all = NULL;
int rc = pcmk_rc_no_output;
gboolean printed_header = FALSE;
gboolean print_everything = TRUE;
if (rsc->priv->fns->is_filtered(rsc, only_rsc, TRUE)) {
return rc;
}
print_everything = pcmk__str_in_list(rsc_printable_id(rsc), only_rsc, pcmk__str_star_matches) ||
(strstr(rsc->id, ":") != NULL && pcmk__str_in_list(rsc->id, only_rsc, pcmk__str_star_matches));
all = g_list_prepend(all, (gpointer) "*");
for (GList *gIter = rsc->priv->children;
gIter != NULL; gIter = gIter->next) {
pcmk_resource_t *child_rsc = (pcmk_resource_t *) gIter->data;
if (pcmk__rsc_filtered_by_node(child_rsc, only_node)) {
continue;
}
if (child_rsc->priv->fns->is_filtered(child_rsc, only_rsc,
print_everything)) {
continue;
}
if (!printed_header) {
const char *multi_state = pcmk__flag_text(rsc->flags,
pcmk__rsc_promotable);
const char *unique = pcmk__flag_text(rsc->flags, pcmk__rsc_unique);
const char *maintenance = pcmk__flag_text(rsc->flags,
pcmk__rsc_maintenance);
const char *managed = pcmk__flag_text(rsc->flags,
pcmk__rsc_managed);
const char *disabled = pcmk__btoa(pe__resource_is_disabled(rsc));
const char *failed = pcmk__flag_text(rsc->flags, pcmk__rsc_failed);
const char *ignored = pcmk__flag_text(rsc->flags,
pcmk__rsc_ignore_failure);
const char *target_role = configured_role_str(rsc);
const char *desc = pe__resource_description(rsc, show_opts);
printed_header = TRUE;
rc = pe__name_and_nvpairs_xml(out, true, PCMK_XE_CLONE,
PCMK_XA_ID, rsc->id,
PCMK_XA_MULTI_STATE, multi_state,
PCMK_XA_UNIQUE, unique,
PCMK_XA_MAINTENANCE, maintenance,
PCMK_XA_MANAGED, managed,
PCMK_XA_DISABLED, disabled,
PCMK_XA_FAILED, failed,
PCMK_XA_FAILURE_IGNORED, ignored,
PCMK_XA_TARGET_ROLE, target_role,
PCMK_XA_DESCRIPTION, desc,
NULL);
pcmk__assert(rc == pcmk_rc_ok);
}
out->message(out, (const char *) child_rsc->priv->xml->name,
show_opts, child_rsc, only_node, all);
}
if (printed_header) {
pcmk__output_xml_pop_parent(out);
}
g_list_free(all);
return rc;
}
PCMK__OUTPUT_ARGS("clone", "uint32_t", "pcmk_resource_t *", "GList *",
"GList *")
int
pe__clone_default(pcmk__output_t *out, va_list args)
{
uint32_t show_opts = va_arg(args, uint32_t);
pcmk_resource_t *rsc = va_arg(args, pcmk_resource_t *);
GList *only_node = va_arg(args, GList *);
GList *only_rsc = va_arg(args, GList *);
GHashTable *stopped = NULL;
GString *list_text = NULL;
GList *promoted_list = NULL;
GList *started_list = NULL;
GList *gIter = NULL;
const char *desc = NULL;
clone_variant_data_t *clone_data = NULL;
int active_instances = 0;
int rc = pcmk_rc_no_output;
gboolean print_everything = TRUE;
desc = pe__resource_description(rsc, show_opts);
get_clone_variant_data(clone_data, rsc);
if (rsc->priv->fns->is_filtered(rsc, only_rsc, TRUE)) {
return rc;
}
print_everything = pcmk__str_in_list(rsc_printable_id(rsc), only_rsc, pcmk__str_star_matches) ||
(strstr(rsc->id, ":") != NULL && pcmk__str_in_list(rsc->id, only_rsc, pcmk__str_star_matches));
for (gIter = rsc->priv->children; gIter != NULL; gIter = gIter->next) {
gboolean print_full = FALSE;
pcmk_resource_t *child_rsc = (pcmk_resource_t *) gIter->data;
gboolean partially_active = child_rsc->priv->fns->active(child_rsc,
FALSE);
if (pcmk__rsc_filtered_by_node(child_rsc, only_node)) {
continue;
}
if (child_rsc->priv->fns->is_filtered(child_rsc, only_rsc,
print_everything)) {
continue;
}
if (pcmk_is_set(show_opts, pcmk_show_clone_detail)) {
print_full = TRUE;
}
if (pcmk_is_set(rsc->flags, pcmk__rsc_unique)) {
// Print individual instance when unique (except stopped orphans)
if (partially_active
|| !pcmk_is_set(rsc->flags, pcmk__rsc_removed)) {
print_full = TRUE;
}
// Everything else in this block is for anonymous clones
} else if (pcmk_is_set(show_opts, pcmk_show_pending)
&& (child_rsc->priv->pending_action != NULL)
&& (strcmp(child_rsc->priv->pending_action,
"probe") != 0)) {
// Print individual instance when non-probe action is pending
print_full = TRUE;
} else if (partially_active == FALSE) {
// List stopped instances when requested (except orphans)
if (!pcmk_is_set(child_rsc->flags, pcmk__rsc_removed)
&& !pcmk_is_set(show_opts, pcmk_show_clone_detail)
&& pcmk_is_set(show_opts, pcmk_show_inactive_rscs)) {
if (stopped == NULL) {
stopped = pcmk__strkey_table(free, free);
}
pcmk__insert_dup(stopped, child_rsc->id, "Stopped");
}
} else if (is_set_recursive(child_rsc, pcmk__rsc_removed, TRUE)
|| !is_set_recursive(child_rsc, pcmk__rsc_managed, FALSE)
|| is_set_recursive(child_rsc, pcmk__rsc_failed, TRUE)) {
// Print individual instance when active orphaned/unmanaged/failed
print_full = TRUE;
} else if (child_rsc->priv->fns->active(child_rsc, TRUE)) {
// Instance of fully active anonymous clone
pcmk_node_t *location = NULL;
location = child_rsc->priv->fns->location(child_rsc, NULL,
pcmk__rsc_node_current);
if (location) {
// Instance is active on a single node
enum rsc_role_e a_role;
a_role = child_rsc->priv->fns->state(child_rsc, TRUE);
if (location->details->online == FALSE && location->details->unclean) {
print_full = TRUE;
} else if (a_role > pcmk_role_unpromoted) {
promoted_list = g_list_append(promoted_list, location);
} else {
started_list = g_list_append(started_list, location);
}
} else {
/* uncolocated group - bleh */
print_full = TRUE;
}
} else {
// Instance of partially active anonymous clone
print_full = TRUE;
}
if (print_full) {
GList *all = NULL;
clone_header(out, &rc, rsc, clone_data, desc);
/* Print every resource that's a child of this clone. */
all = g_list_prepend(all, (gpointer) "*");
out->message(out, (const char *) child_rsc->priv->xml->name,
show_opts, child_rsc, only_node, all);
g_list_free(all);
}
}
if (pcmk_is_set(show_opts, pcmk_show_clone_detail)) {
PCMK__OUTPUT_LIST_FOOTER(out, rc);
return pcmk_rc_ok;
}
/* Promoted */
promoted_list = g_list_sort(promoted_list, pe__cmp_node_name);
for (gIter = promoted_list; gIter; gIter = gIter->next) {
pcmk_node_t *host = gIter->data;
if (!pcmk__str_in_list(host->priv->name, only_node,
pcmk__str_star_matches|pcmk__str_casei)) {
continue;
}
pcmk__add_word(&list_text, 1024, host->priv->name);
active_instances++;
}
g_list_free(promoted_list);
if ((list_text != NULL) && (list_text->len > 0)) {
clone_header(out, &rc, rsc, clone_data, desc);
out->list_item(out, NULL, PCMK_ROLE_PROMOTED ": [ %s ]",
(const char *) list_text->str);
g_string_truncate(list_text, 0);
}
/* Started/Unpromoted */
started_list = g_list_sort(started_list, pe__cmp_node_name);
for (gIter = started_list; gIter; gIter = gIter->next) {
pcmk_node_t *host = gIter->data;
if (!pcmk__str_in_list(host->priv->name, only_node,
pcmk__str_star_matches|pcmk__str_casei)) {
continue;
}
pcmk__add_word(&list_text, 1024, host->priv->name);
active_instances++;
}
g_list_free(started_list);
if ((list_text != NULL) && (list_text->len > 0)) {
clone_header(out, &rc, rsc, clone_data, desc);
if (pcmk_is_set(rsc->flags, pcmk__rsc_promotable)) {
enum rsc_role_e role = configured_role(rsc);
if (role == pcmk_role_unpromoted) {
out->list_item(out, NULL,
PCMK_ROLE_UNPROMOTED
" (" PCMK_META_TARGET_ROLE "): [ %s ]",
(const char *) list_text->str);
} else {
out->list_item(out, NULL, PCMK_ROLE_UNPROMOTED ": [ %s ]",
(const char *) list_text->str);
}
} else {
out->list_item(out, NULL, "Started: [ %s ]",
(const char *) list_text->str);
}
}
if (list_text != NULL) {
g_string_free(list_text, TRUE);
}
if (pcmk_is_set(show_opts, pcmk_show_inactive_rscs)) {
if (!pcmk_is_set(rsc->flags, pcmk__rsc_unique)
&& (clone_data->clone_max > active_instances)) {
GList *nIter;
GList *list = g_hash_table_get_values(rsc->priv->allowed_nodes);
/* Custom stopped table for non-unique clones */
if (stopped != NULL) {
g_hash_table_destroy(stopped);
stopped = NULL;
}
if (list == NULL) {
/* Clusters with PCMK_OPT_SYMMETRIC_CLUSTER=false haven't
* calculated allowed nodes yet. If we've not probed for them
* yet, the Stopped list will be empty.
*/
list = g_hash_table_get_values(rsc->priv->probed_nodes);
}
list = g_list_sort(list, pe__cmp_node_name);
for (nIter = list; nIter != NULL; nIter = nIter->next) {
pcmk_node_t *node = (pcmk_node_t *) nIter->data;
if ((pcmk__find_node_in_list(rsc->priv->active_nodes,
node->priv->name) == NULL)
&& pcmk__str_in_list(node->priv->name, only_node,
pcmk__str_star_matches|pcmk__str_casei)) {
xmlNode *probe_op = NULL;
const char *state = "Stopped";
if (configured_role(rsc) == pcmk_role_stopped) {
state = "Stopped (disabled)";
}
if (stopped == NULL) {
stopped = pcmk__strkey_table(free, free);
}
probe_op = pe__failed_probe_for_rsc(rsc,
node->priv->name);
if (probe_op != NULL) {
int rc;
pcmk__scan_min_int(crm_element_value(probe_op,
PCMK__XA_RC_CODE),
&rc, 0);
g_hash_table_insert(stopped, strdup(node->priv->name),
crm_strdup_printf("Stopped (%s)",
crm_exit_str(rc)));
} else {
pcmk__insert_dup(stopped, node->priv->name, state);
}
}
}
g_list_free(list);
}
if (stopped != NULL) {
GList *list = sorted_hash_table_values(stopped);
clone_header(out, &rc, rsc, clone_data, desc);
for (GList *status_iter = list; status_iter != NULL; status_iter = status_iter->next) {
const char *status = status_iter->data;
GList *nodes = nodes_with_status(stopped, status);
GString *nodes_str = node_list_to_str(nodes);
if (nodes_str != NULL) {
if (nodes_str->len > 0) {
out->list_item(out, NULL, "%s: [ %s ]", status,
(const char *) nodes_str->str);
}
g_string_free(nodes_str, TRUE);
}
g_list_free(nodes);
}
g_list_free(list);
g_hash_table_destroy(stopped);
/* If there are no instances of this clone (perhaps because there are no
* nodes configured), simply output the clone header by itself. This can
* come up in PCS testing.
*/
} else if (active_instances == 0) {
clone_header(out, &rc, rsc, clone_data, desc);
PCMK__OUTPUT_LIST_FOOTER(out, rc);
return rc;
}
}
PCMK__OUTPUT_LIST_FOOTER(out, rc);
return rc;
}
void
clone_free(pcmk_resource_t * rsc)
{
clone_variant_data_t *clone_data = NULL;
get_clone_variant_data(clone_data, rsc);
pcmk__rsc_trace(rsc, "Freeing %s", rsc->id);
for (GList *gIter = rsc->priv->children;
gIter != NULL; gIter = gIter->next) {
pcmk_resource_t *child_rsc = (pcmk_resource_t *) gIter->data;
pcmk__assert(child_rsc != NULL);
pcmk__rsc_trace(child_rsc, "Freeing child %s", child_rsc->id);
pcmk__xml_free(child_rsc->priv->xml);
child_rsc->priv->xml = NULL;
/* There could be a saved unexpanded xml */
pcmk__xml_free(child_rsc->priv->orig_xml);
child_rsc->priv->orig_xml = NULL;
child_rsc->priv->fns->free(child_rsc);
}
g_list_free(rsc->priv->children);
if (clone_data) {
pcmk__assert((clone_data->demote_notify == NULL)
&& (clone_data->stop_notify == NULL)
&& (clone_data->start_notify == NULL)
&& (clone_data->promote_notify == NULL));
}
common_free(rsc);
}
enum rsc_role_e
clone_resource_state(const pcmk_resource_t * rsc, gboolean current)
{
enum rsc_role_e clone_role = pcmk_role_unknown;
for (GList *gIter = rsc->priv->children;
gIter != NULL; gIter = gIter->next) {
pcmk_resource_t *child_rsc = (pcmk_resource_t *) gIter->data;
enum rsc_role_e a_role = child_rsc->priv->fns->state(child_rsc,
current);
if (a_role > clone_role) {
clone_role = a_role;
}
}
pcmk__rsc_trace(rsc, "%s role: %s", rsc->id, pcmk_role_text(clone_role));
return clone_role;
}
/*!
* \internal
* \brief Check whether a clone has an instance for every node
*
* \param[in] rsc Clone to check
* \param[in] scheduler Scheduler data
*/
bool
pe__is_universal_clone(const pcmk_resource_t *rsc,
const pcmk_scheduler_t *scheduler)
{
if (pcmk__is_clone(rsc)) {
clone_variant_data_t *clone_data = rsc->priv->variant_opaque;
if (clone_data->clone_max == g_list_length(scheduler->nodes)) {
return TRUE;
}
}
return FALSE;
}
gboolean
pe__clone_is_filtered(const pcmk_resource_t *rsc, GList *only_rsc,
gboolean check_parent)
{
gboolean passes = FALSE;
clone_variant_data_t *clone_data = NULL;
if (pcmk__str_in_list(rsc_printable_id(rsc), only_rsc, pcmk__str_star_matches)) {
passes = TRUE;
} else {
get_clone_variant_data(clone_data, rsc);
passes = pcmk__str_in_list(pcmk__xe_id(clone_data->xml_obj_child),
only_rsc, pcmk__str_star_matches);
if (!passes) {
for (const GList *iter = rsc->priv->children;
iter != NULL; iter = iter->next) {
const pcmk_resource_t *child_rsc = NULL;
child_rsc = (const pcmk_resource_t *) iter->data;
if (!child_rsc->priv->fns->is_filtered(child_rsc, only_rsc,
FALSE)) {
passes = TRUE;
break;
}
}
}
}
return !passes;
}
const char *
pe__clone_child_id(const pcmk_resource_t *rsc)
{
clone_variant_data_t *clone_data = NULL;
get_clone_variant_data(clone_data, rsc);
return pcmk__xe_id(clone_data->xml_obj_child);
}
/*!
* \internal
* \brief Check whether a clone is ordered
*
* \param[in] clone Clone resource to check
*
* \return true if clone is ordered, otherwise false
*/
bool
pe__clone_is_ordered(const pcmk_resource_t *clone)
{
clone_variant_data_t *clone_data = NULL;
get_clone_variant_data(clone_data, clone);
return pcmk_is_set(clone_data->flags, pcmk__clone_ordered);
}
/*!
* \internal
* \brief Set a clone flag
*
* \param[in,out] clone Clone resource to set flag for
* \param[in] flag Clone flag to set
*
* \return Standard Pacemaker return code (either pcmk_rc_ok if flag was not
* already set or pcmk_rc_already if it was)
*/
int
pe__set_clone_flag(pcmk_resource_t *clone, enum pcmk__clone_flags flag)
{
clone_variant_data_t *clone_data = NULL;
get_clone_variant_data(clone_data, clone);
if (pcmk_is_set(clone_data->flags, flag)) {
return pcmk_rc_already;
}
clone_data->flags = pcmk__set_flags_as(__func__, __LINE__, LOG_TRACE,
"Clone", clone->id,
clone_data->flags, flag, "flag");
return pcmk_rc_ok;
}
/*!
* \internal
* \brief Check whether a clone flag is set
*
* \param[in] group Clone resource to check
* \param[in] flags Flag or flags to check
*
* \return \c true if all \p flags are set for \p clone, otherwise \c false
*/
bool
pe__clone_flag_is_set(const pcmk_resource_t *clone, uint32_t flags)
{
clone_variant_data_t *clone_data = NULL;
get_clone_variant_data(clone_data, clone);
pcmk__assert(clone_data != NULL);
return pcmk_all_flags_set(clone_data->flags, flags);
}
/*!
* \internal
* \brief Create pseudo-actions needed for promotable clones
*
* \param[in,out] clone Promotable clone to create actions for
* \param[in] any_promoting Whether any instances will be promoted
* \param[in] any_demoting Whether any instance will be demoted
*/
void
pe__create_promotable_pseudo_ops(pcmk_resource_t *clone, bool any_promoting,
bool any_demoting)
{
pcmk_action_t *action = NULL;
pcmk_action_t *action_complete = NULL;
clone_variant_data_t *clone_data = NULL;
get_clone_variant_data(clone_data, clone);
// Create a "promote" action for the clone itself
action = pe__new_rsc_pseudo_action(clone, PCMK_ACTION_PROMOTE,
!any_promoting, true);
// Create a "promoted" action for when all promotions are done
action_complete = pe__new_rsc_pseudo_action(clone, PCMK_ACTION_PROMOTED,
!any_promoting, true);
action_complete->priority = PCMK_SCORE_INFINITY;
// Create notification pseudo-actions for promotion
if (clone_data->promote_notify == NULL) {
clone_data->promote_notify = pe__action_notif_pseudo_ops(clone,
PCMK_ACTION_PROMOTE,
action,
action_complete);
}
// Create a "demote" action for the clone itself
action = pe__new_rsc_pseudo_action(clone, PCMK_ACTION_DEMOTE,
!any_demoting, true);
// Create a "demoted" action for when all demotions are done
action_complete = pe__new_rsc_pseudo_action(clone, PCMK_ACTION_DEMOTED,
!any_demoting, true);
action_complete->priority = PCMK_SCORE_INFINITY;
// Create notification pseudo-actions for demotion
if (clone_data->demote_notify == NULL) {
clone_data->demote_notify = pe__action_notif_pseudo_ops(clone,
PCMK_ACTION_DEMOTE,
action,
action_complete);
if (clone_data->promote_notify != NULL) {
order_actions(clone_data->stop_notify->post_done,
clone_data->promote_notify->pre, pcmk__ar_ordered);
order_actions(clone_data->start_notify->post_done,
clone_data->promote_notify->pre, pcmk__ar_ordered);
order_actions(clone_data->demote_notify->post_done,
clone_data->promote_notify->pre, pcmk__ar_ordered);
order_actions(clone_data->demote_notify->post_done,
clone_data->start_notify->pre, pcmk__ar_ordered);
order_actions(clone_data->demote_notify->post_done,
clone_data->stop_notify->pre, pcmk__ar_ordered);
}
}
}
/*!
* \internal
* \brief Create all notification data and actions for a clone
*
* \param[in,out] clone Clone to create notifications for
*/
void
pe__create_clone_notifications(pcmk_resource_t *clone)
{
clone_variant_data_t *clone_data = NULL;
get_clone_variant_data(clone_data, clone);
pe__create_action_notifications(clone, clone_data->start_notify);
pe__create_action_notifications(clone, clone_data->stop_notify);
pe__create_action_notifications(clone, clone_data->promote_notify);
pe__create_action_notifications(clone, clone_data->demote_notify);
}
/*!
* \internal
* \brief Free all notification data for a clone
*
* \param[in,out] clone Clone to free notification data for
*/
void
pe__free_clone_notification_data(pcmk_resource_t *clone)
{
clone_variant_data_t *clone_data = NULL;
get_clone_variant_data(clone_data, clone);
pe__free_action_notification_data(clone_data->demote_notify);
clone_data->demote_notify = NULL;
pe__free_action_notification_data(clone_data->stop_notify);
clone_data->stop_notify = NULL;
pe__free_action_notification_data(clone_data->start_notify);
clone_data->start_notify = NULL;
pe__free_action_notification_data(clone_data->promote_notify);
clone_data->promote_notify = NULL;
}
/*!
* \internal
* \brief Create pseudo-actions for clone start/stop notifications
*
* \param[in,out] clone Clone to create pseudo-actions for
* \param[in,out] start Start action for \p clone
* \param[in,out] stop Stop action for \p clone
* \param[in,out] started Started action for \p clone
* \param[in,out] stopped Stopped action for \p clone
*/
void
pe__create_clone_notif_pseudo_ops(pcmk_resource_t *clone,
pcmk_action_t *start, pcmk_action_t *started,
pcmk_action_t *stop, pcmk_action_t *stopped)
{
clone_variant_data_t *clone_data = NULL;
get_clone_variant_data(clone_data, clone);
if (clone_data->start_notify == NULL) {
clone_data->start_notify = pe__action_notif_pseudo_ops(clone,
PCMK_ACTION_START,
start, started);
}
if (clone_data->stop_notify == NULL) {
clone_data->stop_notify = pe__action_notif_pseudo_ops(clone,
PCMK_ACTION_STOP,
stop, stopped);
if ((clone_data->start_notify != NULL)
&& (clone_data->stop_notify != NULL)) {
order_actions(clone_data->stop_notify->post_done,
clone_data->start_notify->pre, pcmk__ar_ordered);
}
}
}
/*!
* \internal
* \brief Get maximum clone resource instances per node
*
* \param[in] rsc Clone resource to check
*
* \return Maximum number of \p rsc instances that can be active on one node
*/
unsigned int
pe__clone_max_per_node(const pcmk_resource_t *rsc)
{
const clone_variant_data_t *clone_data = NULL;
get_clone_variant_data(clone_data, rsc);
return clone_data->clone_node_max;
}
diff --git a/lib/pengine/pe_actions.c b/lib/pengine/pe_actions.c
index 7bc65dcad1..3497e56779 100644
--- a/lib/pengine/pe_actions.c
+++ b/lib/pengine/pe_actions.c
@@ -1,1770 +1,1774 @@
/*
* Copyright 2004-2024 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU Lesser General Public License
* version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY.
*/
#include <crm_internal.h>
#include <glib.h>
#include <stdbool.h>
#include <crm/crm.h>
#include <crm/common/xml.h>
#include <crm/common/scheduler_internal.h>
#include <crm/pengine/internal.h>
#include <crm/common/xml_internal.h>
#include "pe_status_private.h"
static void unpack_operation(pcmk_action_t *action, const xmlNode *xml_obj,
guint interval_ms);
static void
add_singleton(pcmk_scheduler_t *scheduler, pcmk_action_t *action)
{
if (scheduler->priv->singletons == NULL) {
scheduler->priv->singletons = pcmk__strkey_table(NULL, NULL);
}
g_hash_table_insert(scheduler->priv->singletons, action->uuid, action);
}
static pcmk_action_t *
lookup_singleton(pcmk_scheduler_t *scheduler, const char *action_uuid)
{
/* @TODO This is the only use of the pcmk_scheduler_t:singletons hash table.
* Compare the performance of this approach to keeping the
* pcmk_scheduler_t:actions list sorted by action key and just searching
* that instead.
*/
if (scheduler->priv->singletons == NULL) {
return NULL;
}
return g_hash_table_lookup(scheduler->priv->singletons, action_uuid);
}
/*!
* \internal
* \brief Find an existing action that matches arguments
*
* \param[in] key Action key to match
* \param[in] rsc Resource to match (if any)
* \param[in] node Node to match (if any)
* \param[in] scheduler Scheduler data
*
* \return Existing action that matches arguments (or NULL if none)
*/
static pcmk_action_t *
find_existing_action(const char *key, const pcmk_resource_t *rsc,
const pcmk_node_t *node, const pcmk_scheduler_t *scheduler)
{
/* When rsc is NULL, it would be quicker to check
* scheduler->priv->singletons, but checking all scheduler->priv->actions
* takes the node into account.
*/
GList *actions = (rsc == NULL)? scheduler->priv->actions : rsc->priv->actions;
GList *matches = find_actions(actions, key, node);
pcmk_action_t *action = NULL;
if (matches == NULL) {
return NULL;
}
CRM_LOG_ASSERT(!pcmk__list_of_multiple(matches));
action = matches->data;
g_list_free(matches);
return action;
}
/*!
* \internal
* \brief Find the XML configuration corresponding to a specific action key
*
* \param[in] rsc Resource to find action configuration for
* \param[in] key "RSC_ACTION_INTERVAL" of action to find
* \param[in] include_disabled If false, do not return disabled actions
*
* \return XML configuration of desired action if any, otherwise NULL
*/
static xmlNode *
find_exact_action_config(const pcmk_resource_t *rsc, const char *action_name,
guint interval_ms, bool include_disabled)
{
for (xmlNode *operation = pcmk__xe_first_child(rsc->priv->ops_xml,
PCMK_XE_OP, NULL, NULL);
operation != NULL; operation = pcmk__xe_next(operation, PCMK_XE_OP)) {
bool enabled = false;
const char *config_name = NULL;
const char *interval_spec = NULL;
guint tmp_ms = 0U;
// @TODO This does not consider meta-attributes, rules, defaults, etc.
if (!include_disabled
&& (pcmk__xe_get_bool_attr(operation, PCMK_META_ENABLED,
&enabled) == pcmk_rc_ok) && !enabled) {
continue;
}
interval_spec = crm_element_value(operation, PCMK_META_INTERVAL);
pcmk_parse_interval_spec(interval_spec, &tmp_ms);
if (tmp_ms != interval_ms) {
continue;
}
config_name = crm_element_value(operation, PCMK_XA_NAME);
if (pcmk__str_eq(action_name, config_name, pcmk__str_none)) {
return operation;
}
}
return NULL;
}
/*!
* \internal
* \brief Find the XML configuration of a resource action
*
* \param[in] rsc Resource to find action configuration for
* \param[in] action_name Action name to search for
* \param[in] interval_ms Action interval (in milliseconds) to search for
* \param[in] include_disabled If false, do not return disabled actions
*
* \return XML configuration of desired action if any, otherwise NULL
*/
xmlNode *
pcmk__find_action_config(const pcmk_resource_t *rsc, const char *action_name,
guint interval_ms, bool include_disabled)
{
xmlNode *action_config = NULL;
// Try requested action first
action_config = find_exact_action_config(rsc, action_name, interval_ms,
include_disabled);
// For migrate_to and migrate_from actions, retry with "migrate"
// @TODO This should be either documented or deprecated
if ((action_config == NULL)
&& pcmk__str_any_of(action_name, PCMK_ACTION_MIGRATE_TO,
PCMK_ACTION_MIGRATE_FROM, NULL)) {
action_config = find_exact_action_config(rsc, "migrate", 0,
include_disabled);
}
return action_config;
}
/*!
* \internal
* \brief Create a new action object
*
* \param[in] key Action key
* \param[in] task Action name
* \param[in,out] rsc Resource that action is for (if any)
* \param[in] node Node that action is on (if any)
* \param[in] optional Whether action should be considered optional
* \param[in,out] scheduler Scheduler data
*
* \return Newly allocated action
* \note This function takes ownership of \p key. It is the caller's
* responsibility to free the return value with pe_free_action().
*/
static pcmk_action_t *
new_action(char *key, const char *task, pcmk_resource_t *rsc,
const pcmk_node_t *node, bool optional, pcmk_scheduler_t *scheduler)
{
pcmk_action_t *action = pcmk__assert_alloc(1, sizeof(pcmk_action_t));
action->rsc = rsc;
action->task = pcmk__str_copy(task);
action->uuid = key;
action->scheduler = scheduler;
if (node) {
action->node = pe__copy_node(node);
}
if (pcmk__str_eq(task, PCMK_ACTION_LRM_DELETE, pcmk__str_casei)) {
// Resource history deletion for a node can be done on the DC
pcmk__set_action_flags(action, pcmk__action_on_dc);
}
pcmk__set_action_flags(action, pcmk__action_runnable);
if (optional) {
pcmk__set_action_flags(action, pcmk__action_optional);
} else {
pcmk__clear_action_flags(action, pcmk__action_optional);
}
if (rsc == NULL) {
action->meta = pcmk__strkey_table(free, free);
} else {
guint interval_ms = 0;
parse_op_key(key, NULL, NULL, &interval_ms);
action->op_entry = pcmk__find_action_config(rsc, task, interval_ms,
true);
/* If the given key is for one of the many notification pseudo-actions
* (pre_notify_promote, etc.), the actual action name is "notify"
*/
if ((action->op_entry == NULL) && (strstr(key, "_notify_") != NULL)) {
action->op_entry = find_exact_action_config(rsc, PCMK_ACTION_NOTIFY,
0, true);
}
unpack_operation(action, action->op_entry, interval_ms);
}
pcmk__rsc_trace(rsc, "Created %s action %d (%s): %s for %s on %s",
(optional? "optional" : "required"),
scheduler->priv->next_action_id, key, task,
((rsc == NULL)? "no resource" : rsc->id),
pcmk__node_name(node));
action->id = scheduler->priv->next_action_id++;
scheduler->priv->actions = g_list_prepend(scheduler->priv->actions, action);
if (rsc == NULL) {
add_singleton(scheduler, action);
} else {
rsc->priv->actions = g_list_prepend(rsc->priv->actions, action);
}
return action;
}
/*!
* \internal
* \brief Unpack a resource's action-specific instance parameters
*
* \param[in] action_xml XML of action's configuration in CIB (if any)
* \param[in,out] node_attrs Table of node attributes (for rule evaluation)
* \param[in,out] scheduler Cluster working set (for rule evaluation)
*
* \return Newly allocated hash table of action-specific instance parameters
*/
GHashTable *
pcmk__unpack_action_rsc_params(const xmlNode *action_xml,
GHashTable *node_attrs,
pcmk_scheduler_t *scheduler)
{
GHashTable *params = pcmk__strkey_table(free, free);
const pcmk_rule_input_t rule_input = {
.now = scheduler->priv->now,
.node_attrs = node_attrs,
};
pe__unpack_dataset_nvpairs(action_xml, PCMK_XE_INSTANCE_ATTRIBUTES,
&rule_input, params, NULL, scheduler);
return params;
}
/*!
* \internal
* \brief Update an action's optional flag
*
* \param[in,out] action Action to update
* \param[in] optional Requested optional status
*/
static void
update_action_optional(pcmk_action_t *action, gboolean optional)
{
// Force a non-recurring action to be optional if its resource is unmanaged
if ((action->rsc != NULL) && (action->node != NULL)
&& !pcmk_is_set(action->flags, pcmk__action_pseudo)
&& !pcmk_is_set(action->rsc->flags, pcmk__rsc_managed)
&& (g_hash_table_lookup(action->meta, PCMK_META_INTERVAL) == NULL)) {
pcmk__rsc_debug(action->rsc,
"%s on %s is optional (%s is unmanaged)",
action->uuid, pcmk__node_name(action->node),
action->rsc->id);
pcmk__set_action_flags(action, pcmk__action_optional);
// We shouldn't clear runnable here because ... something
// Otherwise require the action if requested
} else if (!optional) {
pcmk__clear_action_flags(action, pcmk__action_optional);
}
}
static enum pe_quorum_policy
effective_quorum_policy(pcmk_resource_t *rsc, pcmk_scheduler_t *scheduler)
{
enum pe_quorum_policy policy = scheduler->no_quorum_policy;
if (pcmk_is_set(scheduler->flags, pcmk__sched_quorate)) {
policy = pcmk_no_quorum_ignore;
} else if (scheduler->no_quorum_policy == pcmk_no_quorum_demote) {
switch (rsc->priv->orig_role) {
case pcmk_role_promoted:
case pcmk_role_unpromoted:
if (rsc->priv->next_role > pcmk_role_unpromoted) {
pe__set_next_role(rsc, pcmk_role_unpromoted,
PCMK_OPT_NO_QUORUM_POLICY "=demote");
}
policy = pcmk_no_quorum_ignore;
break;
default:
policy = pcmk_no_quorum_stop;
break;
}
}
return policy;
}
/*!
* \internal
* \brief Update a resource action's runnable flag
*
* \param[in,out] action Action to update
* \param[in,out] scheduler Scheduler data
*
* \note This may also schedule fencing if a stop is unrunnable.
*/
static void
update_resource_action_runnable(pcmk_action_t *action,
pcmk_scheduler_t *scheduler)
{
pcmk_resource_t *rsc = action->rsc;
if (pcmk_is_set(action->flags, pcmk__action_pseudo)) {
return;
}
if (action->node == NULL) {
pcmk__rsc_trace(rsc, "%s is unrunnable (unallocated)", action->uuid);
pcmk__clear_action_flags(action, pcmk__action_runnable);
} else if (!pcmk_is_set(action->flags, pcmk__action_on_dc)
&& !(action->node->details->online)
&& (!pcmk__is_guest_or_bundle_node(action->node)
|| pcmk_is_set(action->node->priv->flags,
pcmk__node_remote_reset))) {
pcmk__clear_action_flags(action, pcmk__action_runnable);
do_crm_log(LOG_WARNING, "%s on %s is unrunnable (node is offline)",
action->uuid, pcmk__node_name(action->node));
if (pcmk_is_set(rsc->flags, pcmk__rsc_managed)
&& pcmk__str_eq(action->task, PCMK_ACTION_STOP, pcmk__str_casei)
&& !(action->node->details->unclean)) {
pe_fence_node(scheduler, action->node, "stop is unrunnable", false);
}
} else if (!pcmk_is_set(action->flags, pcmk__action_on_dc)
&& action->node->details->pending) {
pcmk__clear_action_flags(action, pcmk__action_runnable);
do_crm_log(LOG_WARNING,
"Action %s on %s is unrunnable (node is pending)",
action->uuid, pcmk__node_name(action->node));
} else if (action->needs == pcmk__requires_nothing) {
pe_action_set_reason(action, NULL, TRUE);
if (pcmk__is_guest_or_bundle_node(action->node)
&& !pe_can_fence(scheduler, action->node)) {
/* An action that requires nothing usually does not require any
* fencing in order to be runnable. However, there is an exception:
* such an action cannot be completed if it is on a guest node whose
* host is unclean and cannot be fenced.
*/
pcmk__rsc_debug(rsc,
"%s on %s is unrunnable "
"(node's host cannot be fenced)",
action->uuid, pcmk__node_name(action->node));
pcmk__clear_action_flags(action, pcmk__action_runnable);
} else {
pcmk__rsc_trace(rsc,
"%s on %s does not require fencing or quorum",
action->uuid, pcmk__node_name(action->node));
pcmk__set_action_flags(action, pcmk__action_runnable);
}
} else {
switch (effective_quorum_policy(rsc, scheduler)) {
case pcmk_no_quorum_stop:
pcmk__rsc_debug(rsc, "%s on %s is unrunnable (no quorum)",
action->uuid, pcmk__node_name(action->node));
pcmk__clear_action_flags(action, pcmk__action_runnable);
pe_action_set_reason(action, "no quorum", true);
break;
case pcmk_no_quorum_freeze:
if (!rsc->priv->fns->active(rsc, TRUE)
|| (rsc->priv->next_role > rsc->priv->orig_role)) {
pcmk__rsc_debug(rsc, "%s on %s is unrunnable (no quorum)",
action->uuid,
pcmk__node_name(action->node));
pcmk__clear_action_flags(action, pcmk__action_runnable);
pe_action_set_reason(action, "quorum freeze", true);
}
break;
default:
//pe_action_set_reason(action, NULL, TRUE);
pcmk__set_action_flags(action, pcmk__action_runnable);
break;
}
}
}
static bool
valid_stop_on_fail(const char *value)
{
return !pcmk__strcase_any_of(value,
PCMK_VALUE_STANDBY, PCMK_VALUE_DEMOTE,
PCMK_VALUE_STOP, NULL);
}
/*!
* \internal
* \brief Validate (and possibly reset) resource action's on_fail meta-attribute
*
* \param[in] rsc Resource that action is for
* \param[in] action_name Action name
* \param[in] action_config Action configuration XML from CIB (if any)
* \param[in,out] meta Table of action meta-attributes
*/
static void
validate_on_fail(const pcmk_resource_t *rsc, const char *action_name,
const xmlNode *action_config, GHashTable *meta)
{
const char *name = NULL;
const char *role = NULL;
const char *interval_spec = NULL;
const char *value = g_hash_table_lookup(meta, PCMK_META_ON_FAIL);
guint interval_ms = 0U;
// Stop actions can only use certain on-fail values
if (pcmk__str_eq(action_name, PCMK_ACTION_STOP, pcmk__str_none)
&& !valid_stop_on_fail(value)) {
pcmk__config_err("Resetting '" PCMK_META_ON_FAIL "' for %s stop "
"action to default value because '%s' is not "
"allowed for stop", rsc->id, value);
g_hash_table_remove(meta, PCMK_META_ON_FAIL);
return;
}
/* Demote actions default on-fail to the on-fail value for the first
* recurring monitor for the promoted role (if any).
*/
if (pcmk__str_eq(action_name, PCMK_ACTION_DEMOTE, pcmk__str_none)
&& (value == NULL)) {
/* @TODO This does not consider promote options set in a meta-attribute
* block (which may have rules that need to be evaluated) rather than
* XML properties.
*/
for (xmlNode *operation = pcmk__xe_first_child(rsc->priv->ops_xml,
PCMK_XE_OP, NULL, NULL);
operation != NULL;
operation = pcmk__xe_next(operation, PCMK_XE_OP)) {
bool enabled = false;
const char *promote_on_fail = NULL;
/* We only care about explicit on-fail (if promote uses default, so
* can demote)
*/
promote_on_fail = crm_element_value(operation, PCMK_META_ON_FAIL);
if (promote_on_fail == NULL) {
continue;
}
// We only care about recurring monitors for the promoted role
name = crm_element_value(operation, PCMK_XA_NAME);
role = crm_element_value(operation, PCMK_XA_ROLE);
if (!pcmk__str_eq(name, PCMK_ACTION_MONITOR, pcmk__str_none)
|| !pcmk__strcase_any_of(role, PCMK_ROLE_PROMOTED,
PCMK__ROLE_PROMOTED_LEGACY, NULL)) {
continue;
}
interval_spec = crm_element_value(operation, PCMK_META_INTERVAL);
pcmk_parse_interval_spec(interval_spec, &interval_ms);
if (interval_ms == 0U) {
continue;
}
// We only care about enabled monitors
if ((pcmk__xe_get_bool_attr(operation, PCMK_META_ENABLED,
&enabled) == pcmk_rc_ok) && !enabled) {
continue;
}
/* Demote actions can't default to
* PCMK_META_ON_FAIL=PCMK_VALUE_DEMOTE
*/
if (pcmk__str_eq(promote_on_fail, PCMK_VALUE_DEMOTE,
pcmk__str_casei)) {
continue;
}
// Use value from first applicable promote action found
pcmk__insert_dup(meta, PCMK_META_ON_FAIL, promote_on_fail);
}
return;
}
if (pcmk__str_eq(action_name, PCMK_ACTION_LRM_DELETE, pcmk__str_none)
&& !pcmk__str_eq(value, PCMK_VALUE_IGNORE, pcmk__str_casei)) {
pcmk__insert_dup(meta, PCMK_META_ON_FAIL, PCMK_VALUE_IGNORE);
return;
}
// PCMK_META_ON_FAIL=PCMK_VALUE_DEMOTE is allowed only for certain actions
if (pcmk__str_eq(value, PCMK_VALUE_DEMOTE, pcmk__str_casei)) {
name = crm_element_value(action_config, PCMK_XA_NAME);
role = crm_element_value(action_config, PCMK_XA_ROLE);
interval_spec = crm_element_value(action_config, PCMK_META_INTERVAL);
pcmk_parse_interval_spec(interval_spec, &interval_ms);
if (!pcmk__str_eq(name, PCMK_ACTION_PROMOTE, pcmk__str_none)
&& ((interval_ms == 0U)
|| !pcmk__str_eq(name, PCMK_ACTION_MONITOR, pcmk__str_none)
|| !pcmk__strcase_any_of(role, PCMK_ROLE_PROMOTED,
PCMK__ROLE_PROMOTED_LEGACY, NULL))) {
pcmk__config_err("Resetting '" PCMK_META_ON_FAIL "' for %s %s "
"action to default value because 'demote' is not "
"allowed for it", rsc->id, name);
g_hash_table_remove(meta, PCMK_META_ON_FAIL);
return;
}
}
}
static int
unpack_timeout(const char *value)
{
long long timeout_ms = crm_get_msec(value);
if (timeout_ms <= 0) {
timeout_ms = PCMK_DEFAULT_ACTION_TIMEOUT_MS;
}
return (int) QB_MIN(timeout_ms, INT_MAX);
}
// true if value contains valid, non-NULL interval origin for recurring op
static bool
unpack_interval_origin(const char *value, const xmlNode *xml_obj,
guint interval_ms, const crm_time_t *now,
long long *start_delay)
{
long long result = 0;
guint interval_sec = pcmk__timeout_ms2s(interval_ms);
crm_time_t *origin = NULL;
// Ignore unspecified values and non-recurring operations
if ((value == NULL) || (interval_ms == 0) || (now == NULL)) {
return false;
}
// Parse interval origin from text
origin = crm_time_new(value);
if (origin == NULL) {
pcmk__config_err("Ignoring '" PCMK_META_INTERVAL_ORIGIN "' for "
"operation '%s' because '%s' is not valid",
pcmk__s(pcmk__xe_id(xml_obj), "(missing ID)"), value);
return false;
}
// Get seconds since origin (negative if origin is in the future)
result = crm_time_get_seconds(now) - crm_time_get_seconds(origin);
crm_time_free(origin);
// Calculate seconds from closest interval to now
result = result % interval_sec;
// Calculate seconds remaining until next interval
result = ((result <= 0)? 0 : interval_sec) - result;
crm_info("Calculated a start delay of %llds for operation '%s'",
result, pcmk__s(pcmk__xe_id(xml_obj), "(unspecified)"));
if (start_delay != NULL) {
*start_delay = result * 1000; // milliseconds
}
return true;
}
static int
unpack_start_delay(const char *value, GHashTable *meta)
{
long long start_delay_ms = 0;
if (value == NULL) {
return 0;
}
start_delay_ms = crm_get_msec(value);
start_delay_ms = QB_MIN(start_delay_ms, INT_MAX);
if (start_delay_ms < 0) {
start_delay_ms = 0;
}
if (meta != NULL) {
g_hash_table_replace(meta, strdup(PCMK_META_START_DELAY),
pcmk__itoa(start_delay_ms));
}
return (int) start_delay_ms;
}
/*!
* \internal
* \brief Find a resource's most frequent recurring monitor
*
* \param[in] rsc Resource to check
*
* \return Operation XML configured for most frequent recurring monitor for
* \p rsc (if any)
*/
static xmlNode *
most_frequent_monitor(const pcmk_resource_t *rsc)
{
guint min_interval_ms = G_MAXUINT;
xmlNode *op = NULL;
for (xmlNode *operation = pcmk__xe_first_child(rsc->priv->ops_xml,
PCMK_XE_OP, NULL, NULL);
operation != NULL; operation = pcmk__xe_next(operation, PCMK_XE_OP)) {
bool enabled = false;
guint interval_ms = 0U;
const char *interval_spec = crm_element_value(operation,
PCMK_META_INTERVAL);
// We only care about enabled recurring monitors
if (!pcmk__str_eq(crm_element_value(operation, PCMK_XA_NAME),
PCMK_ACTION_MONITOR, pcmk__str_none)) {
continue;
}
pcmk_parse_interval_spec(interval_spec, &interval_ms);
if (interval_ms == 0U) {
continue;
}
// @TODO This does not consider meta-attributes, rules, defaults, etc.
if ((pcmk__xe_get_bool_attr(operation, PCMK_META_ENABLED,
&enabled) == pcmk_rc_ok) && !enabled) {
continue;
}
if (interval_ms < min_interval_ms) {
min_interval_ms = interval_ms;
op = operation;
}
}
return op;
}
/*!
* \internal
* \brief Unpack action meta-attributes
*
* \param[in,out] rsc Resource that action is for
* \param[in] node Node that action is on
* \param[in] action_name Action name
* \param[in] interval_ms Action interval (in milliseconds)
* \param[in] action_config Action XML configuration from CIB (if any)
*
* Unpack a resource action's meta-attributes (normalizing the interval,
* timeout, and start delay values as integer milliseconds) from its CIB XML
* configuration (including defaults).
*
* \return Newly allocated hash table with normalized action meta-attributes
*/
GHashTable *
pcmk__unpack_action_meta(pcmk_resource_t *rsc, const pcmk_node_t *node,
const char *action_name, guint interval_ms,
const xmlNode *action_config)
{
GHashTable *meta = NULL;
const char *timeout_spec = NULL;
const char *str = NULL;
const pcmk_rule_input_t rule_input = {
/* Node attributes are not set because node expressions are not allowed
* for meta-attributes
*/
.now = rsc->priv->scheduler->priv->now,
.rsc_standard = crm_element_value(rsc->priv->xml, PCMK_XA_CLASS),
.rsc_provider = crm_element_value(rsc->priv->xml, PCMK_XA_PROVIDER),
.rsc_agent = crm_element_value(rsc->priv->xml, PCMK_XA_TYPE),
.op_name = action_name,
.op_interval_ms = interval_ms,
};
meta = pcmk__strkey_table(free, free);
if (action_config != NULL) {
// <op> <meta_attributes> take precedence over defaults
pe__unpack_dataset_nvpairs(action_config, PCMK_XE_META_ATTRIBUTES,
&rule_input, meta, NULL,
rsc->priv->scheduler);
/* Anything set as an <op> XML property has highest precedence.
* This ensures we use the name and interval from the <op> tag.
* (See below for the only exception, fence device start/probe timeout.)
*/
for (xmlAttrPtr attr = action_config->properties;
attr != NULL; attr = attr->next) {
pcmk__insert_dup(meta, (const char *) attr->name,
pcmk__xml_attr_value(attr));
}
}
// Derive default timeout for probes from recurring monitor timeouts
if (pcmk_is_probe(action_name, interval_ms)
&& (g_hash_table_lookup(meta, PCMK_META_TIMEOUT) == NULL)) {
xmlNode *min_interval_mon = most_frequent_monitor(rsc);
if (min_interval_mon != NULL) {
/* @TODO This does not consider timeouts set in
* PCMK_XE_META_ATTRIBUTES blocks (which may also have rules that
* need to be evaluated).
*/
timeout_spec = crm_element_value(min_interval_mon,
PCMK_META_TIMEOUT);
if (timeout_spec != NULL) {
pcmk__rsc_trace(rsc,
"Setting default timeout for %s probe to "
"most frequent monitor's timeout '%s'",
rsc->id, timeout_spec);
pcmk__insert_dup(meta, PCMK_META_TIMEOUT, timeout_spec);
}
}
}
// Cluster-wide <op_defaults> <meta_attributes>
pe__unpack_dataset_nvpairs(rsc->priv->scheduler->priv->op_defaults,
PCMK_XE_META_ATTRIBUTES, &rule_input, meta, NULL,
rsc->priv->scheduler);
g_hash_table_remove(meta, PCMK_XA_ID);
// Normalize interval to milliseconds
if (interval_ms > 0) {
g_hash_table_insert(meta, pcmk__str_copy(PCMK_META_INTERVAL),
crm_strdup_printf("%u", interval_ms));
} else {
g_hash_table_remove(meta, PCMK_META_INTERVAL);
}
/* Timeout order of precedence (highest to lowest):
* 1. pcmk_monitor_timeout resource parameter (only for starts and probes
* when rsc has pcmk_ra_cap_fence_params; this gets used for recurring
* monitors via the executor instead)
* 2. timeout configured in <op> (with <op timeout> taking precedence over
* <op> <meta_attributes>)
* 3. timeout configured in <op_defaults> <meta_attributes>
* 4. PCMK_DEFAULT_ACTION_TIMEOUT_MS
*/
// Check for pcmk_monitor_timeout
if (pcmk_is_set(pcmk_get_ra_caps(rule_input.rsc_standard),
pcmk_ra_cap_fence_params)
&& (pcmk__str_eq(action_name, PCMK_ACTION_START, pcmk__str_none)
|| pcmk_is_probe(action_name, interval_ms))) {
GHashTable *params = pe_rsc_params(rsc, node, rsc->priv->scheduler);
timeout_spec = g_hash_table_lookup(params, "pcmk_monitor_timeout");
if (timeout_spec != NULL) {
pcmk__rsc_trace(rsc,
"Setting timeout for %s %s to "
"pcmk_monitor_timeout (%s)",
rsc->id, action_name, timeout_spec);
pcmk__insert_dup(meta, PCMK_META_TIMEOUT, timeout_spec);
}
}
// Normalize timeout to positive milliseconds
timeout_spec = g_hash_table_lookup(meta, PCMK_META_TIMEOUT);
g_hash_table_insert(meta, pcmk__str_copy(PCMK_META_TIMEOUT),
pcmk__itoa(unpack_timeout(timeout_spec)));
// Ensure on-fail has a valid value
validate_on_fail(rsc, action_name, action_config, meta);
// Normalize PCMK_META_START_DELAY
str = g_hash_table_lookup(meta, PCMK_META_START_DELAY);
if (str != NULL) {
unpack_start_delay(str, meta);
} else {
long long start_delay = 0;
str = g_hash_table_lookup(meta, PCMK_META_INTERVAL_ORIGIN);
if (unpack_interval_origin(str, action_config, interval_ms,
rsc->priv->scheduler->priv->now,
&start_delay)) {
g_hash_table_insert(meta, pcmk__str_copy(PCMK_META_START_DELAY),
crm_strdup_printf("%lld", start_delay));
}
}
return meta;
}
/*!
* \internal
* \brief Determine an action's quorum and fencing dependency
*
* \param[in] rsc Resource that action is for
* \param[in] action_name Name of action being unpacked
*
* \return Quorum and fencing dependency appropriate to action
*/
enum pcmk__requires
pcmk__action_requires(const pcmk_resource_t *rsc, const char *action_name)
{
const char *value = NULL;
enum pcmk__requires requires = pcmk__requires_nothing;
CRM_CHECK((rsc != NULL) && (action_name != NULL), return requires);
if (!pcmk__strcase_any_of(action_name, PCMK_ACTION_START,
PCMK_ACTION_PROMOTE, NULL)) {
value = "nothing (not start or promote)";
} else if (pcmk_is_set(rsc->flags, pcmk__rsc_needs_fencing)) {
requires = pcmk__requires_fencing;
value = "fencing";
} else if (pcmk_is_set(rsc->flags, pcmk__rsc_needs_quorum)) {
requires = pcmk__requires_quorum;
value = "quorum";
} else {
value = "nothing";
}
pcmk__rsc_trace(rsc, "%s of %s requires %s", action_name, rsc->id, value);
return requires;
}
/*!
* \internal
* \brief Parse action failure response from a user-provided string
*
* \param[in] rsc Resource that action is for
* \param[in] action_name Name of action
* \param[in] interval_ms Action interval (in milliseconds)
* \param[in] value User-provided configuration value for on-fail
*
* \return Action failure response parsed from \p text
*/
enum pcmk__on_fail
pcmk__parse_on_fail(const pcmk_resource_t *rsc, const char *action_name,
guint interval_ms, const char *value)
{
const char *desc = NULL;
bool needs_remote_reset = false;
enum pcmk__on_fail on_fail = pcmk__on_fail_ignore;
const pcmk_scheduler_t *scheduler = NULL;
// There's no enum value for unknown or invalid, so assert
pcmk__assert((rsc != NULL) && (action_name != NULL));
scheduler = rsc->priv->scheduler;
if (value == NULL) {
// Use default
} else if (pcmk__str_eq(value, PCMK_VALUE_BLOCK, pcmk__str_casei)) {
on_fail = pcmk__on_fail_block;
desc = "block";
} else if (pcmk__str_eq(value, PCMK_VALUE_FENCE, pcmk__str_casei)) {
if (pcmk_is_set(scheduler->flags, pcmk__sched_fencing_enabled)) {
on_fail = pcmk__on_fail_fence_node;
desc = "node fencing";
} else {
pcmk__config_err("Resetting '" PCMK_META_ON_FAIL "' for "
"%s of %s to 'stop' because 'fence' is not "
"valid when fencing is disabled",
action_name, rsc->id);
+ /* @TODO This should probably do
+ g_hash_table_remove(meta, PCMK_META_ON_FAIL);
+ like the other "Resetting" spots, to avoid repeating the message
+ */
on_fail = pcmk__on_fail_stop;
desc = "stop resource";
}
} else if (pcmk__str_eq(value, PCMK_VALUE_STANDBY, pcmk__str_casei)) {
on_fail = pcmk__on_fail_standby_node;
desc = "node standby";
} else if (pcmk__strcase_any_of(value,
PCMK_VALUE_IGNORE, PCMK_VALUE_NOTHING,
NULL)) {
desc = "ignore";
} else if (pcmk__str_eq(value, "migrate", pcmk__str_casei)) {
on_fail = pcmk__on_fail_ban;
desc = "force migration";
} else if (pcmk__str_eq(value, PCMK_VALUE_STOP, pcmk__str_casei)) {
on_fail = pcmk__on_fail_stop;
desc = "stop resource";
} else if (pcmk__str_eq(value, PCMK_VALUE_RESTART, pcmk__str_casei)) {
on_fail = pcmk__on_fail_restart;
desc = "restart (and possibly migrate)";
} else if (pcmk__str_eq(value, PCMK_VALUE_RESTART_CONTAINER,
pcmk__str_casei)) {
if (rsc->priv->launcher == NULL) {
pcmk__rsc_debug(rsc,
"Using default " PCMK_META_ON_FAIL " for %s "
"of %s because it does not have a launcher",
action_name, rsc->id);
} else {
on_fail = pcmk__on_fail_restart_container;
desc = "restart container (and possibly migrate)";
}
} else if (pcmk__str_eq(value, PCMK_VALUE_DEMOTE, pcmk__str_casei)) {
on_fail = pcmk__on_fail_demote;
desc = "demote instance";
} else {
pcmk__config_err("Using default '" PCMK_META_ON_FAIL "' for "
"%s of %s because '%s' is not valid",
action_name, rsc->id, value);
}
/* Remote node connections are handled specially. Failures that result
* in dropping an active connection must result in fencing. The only
* failures that don't are probes and starts. The user can explicitly set
* PCMK_META_ON_FAIL=PCMK_VALUE_FENCE to fence after start failures.
*/
if (pcmk_is_set(rsc->flags, pcmk__rsc_is_remote_connection)
&& pcmk__is_remote_node(pcmk_find_node(scheduler, rsc->id))
&& !pcmk_is_probe(action_name, interval_ms)
&& !pcmk__str_eq(action_name, PCMK_ACTION_START, pcmk__str_none)) {
needs_remote_reset = true;
if (!pcmk_is_set(rsc->flags, pcmk__rsc_managed)) {
desc = NULL; // Force default for unmanaged connections
}
}
if (desc != NULL) {
// Explicit value used, default not needed
} else if (rsc->priv->launcher != NULL) {
on_fail = pcmk__on_fail_restart_container;
desc = "restart container (and possibly migrate) (default)";
} else if (needs_remote_reset) {
if (pcmk_is_set(rsc->flags, pcmk__rsc_managed)) {
if (pcmk_is_set(scheduler->flags, pcmk__sched_fencing_enabled)) {
desc = "fence remote node (default)";
} else {
desc = "recover remote node connection (default)";
}
on_fail = pcmk__on_fail_reset_remote;
} else {
on_fail = pcmk__on_fail_stop;
desc = "stop unmanaged remote node (enforcing default)";
}
} else if (pcmk__str_eq(action_name, PCMK_ACTION_STOP, pcmk__str_none)) {
if (pcmk_is_set(scheduler->flags, pcmk__sched_fencing_enabled)) {
on_fail = pcmk__on_fail_fence_node;
desc = "resource fence (default)";
} else {
on_fail = pcmk__on_fail_block;
desc = "resource block (default)";
}
} else {
on_fail = pcmk__on_fail_restart;
desc = "restart (and possibly migrate) (default)";
}
pcmk__rsc_trace(rsc, "Failure handling for %s-interval %s of %s: %s",
pcmk__readable_interval(interval_ms), action_name,
rsc->id, desc);
return on_fail;
}
/*!
* \internal
* \brief Determine a resource's role after failure of an action
*
* \param[in] rsc Resource that action is for
* \param[in] action_name Action name
* \param[in] on_fail Failure handling for action
* \param[in] meta Unpacked action meta-attributes
*
* \return Resource role that results from failure of action
*/
enum rsc_role_e
pcmk__role_after_failure(const pcmk_resource_t *rsc, const char *action_name,
enum pcmk__on_fail on_fail, GHashTable *meta)
{
enum rsc_role_e role = pcmk_role_unknown;
// Set default for role after failure specially in certain circumstances
switch (on_fail) {
case pcmk__on_fail_stop:
role = pcmk_role_stopped;
break;
case pcmk__on_fail_reset_remote:
if (rsc->priv->remote_reconnect_ms != 0U) {
role = pcmk_role_stopped;
}
break;
default:
break;
}
if (role == pcmk_role_unknown) {
// Use default
if (pcmk__str_eq(action_name, PCMK_ACTION_PROMOTE, pcmk__str_none)) {
role = pcmk_role_unpromoted;
} else {
role = pcmk_role_started;
}
}
pcmk__rsc_trace(rsc, "Role after %s %s failure is: %s",
rsc->id, action_name, pcmk_role_text(role));
return role;
}
/*!
* \internal
* \brief Unpack action configuration
*
* Unpack a resource action's meta-attributes (normalizing the interval,
* timeout, and start delay values as integer milliseconds), requirements, and
* failure policy from its CIB XML configuration (including defaults).
*
* \param[in,out] action Resource action to unpack into
* \param[in] xml_obj Action configuration XML (NULL for defaults only)
* \param[in] interval_ms How frequently to perform the operation
*/
static void
unpack_operation(pcmk_action_t *action, const xmlNode *xml_obj,
guint interval_ms)
{
const char *value = NULL;
action->meta = pcmk__unpack_action_meta(action->rsc, action->node,
action->task, interval_ms, xml_obj);
action->needs = pcmk__action_requires(action->rsc, action->task);
value = g_hash_table_lookup(action->meta, PCMK_META_ON_FAIL);
action->on_fail = pcmk__parse_on_fail(action->rsc, action->task,
interval_ms, value);
action->fail_role = pcmk__role_after_failure(action->rsc, action->task,
action->on_fail, action->meta);
}
/*!
* \brief Create or update an action object
*
* \param[in,out] rsc Resource that action is for (if any)
* \param[in,out] key Action key (must be non-NULL)
* \param[in] task Action name (must be non-NULL)
* \param[in] on_node Node that action is on (if any)
* \param[in] optional Whether action should be considered optional
* \param[in,out] scheduler Scheduler data
*
* \return Action object corresponding to arguments (guaranteed not to be
* \c NULL)
* \note This function takes ownership of (and might free) \p key, and
* \p scheduler takes ownership of the returned action (the caller should
* not free it).
*/
pcmk_action_t *
custom_action(pcmk_resource_t *rsc, char *key, const char *task,
const pcmk_node_t *on_node, gboolean optional,
pcmk_scheduler_t *scheduler)
{
pcmk_action_t *action = NULL;
pcmk__assert((key != NULL) && (task != NULL) && (scheduler != NULL));
action = find_existing_action(key, rsc, on_node, scheduler);
if (action == NULL) {
action = new_action(key, task, rsc, on_node, optional, scheduler);
} else {
free(key);
}
update_action_optional(action, optional);
if (rsc != NULL) {
/* An action can be initially created with a NULL node, and later have
* the node added via find_existing_action() (above) -> find_actions().
* That is why the extra parameters are unpacked here rather than in
* new_action().
*/
if ((action->node != NULL) && (action->op_entry != NULL)
&& !pcmk_is_set(action->flags, pcmk__action_attrs_evaluated)) {
GHashTable *attrs = action->node->priv->attrs;
if (action->extra != NULL) {
g_hash_table_destroy(action->extra);
}
action->extra = pcmk__unpack_action_rsc_params(action->op_entry,
attrs, scheduler);
pcmk__set_action_flags(action, pcmk__action_attrs_evaluated);
}
update_resource_action_runnable(action, scheduler);
}
if (action->extra == NULL) {
action->extra = pcmk__strkey_table(free, free);
}
return action;
}
pcmk_action_t *
get_pseudo_op(const char *name, pcmk_scheduler_t *scheduler)
{
pcmk_action_t *op = lookup_singleton(scheduler, name);
if (op == NULL) {
op = custom_action(NULL, strdup(name), name, NULL, TRUE, scheduler);
pcmk__set_action_flags(op, pcmk__action_pseudo|pcmk__action_runnable);
}
return op;
}
static GList *
find_unfencing_devices(GList *candidates, GList *matches)
{
for (GList *gIter = candidates; gIter != NULL; gIter = gIter->next) {
pcmk_resource_t *candidate = gIter->data;
if (candidate->priv->children != NULL) {
matches = find_unfencing_devices(candidate->priv->children,
matches);
} else if (!pcmk_is_set(candidate->flags, pcmk__rsc_fence_device)) {
continue;
} else if (pcmk_is_set(candidate->flags, pcmk__rsc_needs_unfencing)) {
matches = g_list_prepend(matches, candidate);
} else if (pcmk__str_eq(g_hash_table_lookup(candidate->priv->meta,
PCMK_STONITH_PROVIDES),
PCMK_VALUE_UNFENCING, pcmk__str_casei)) {
matches = g_list_prepend(matches, candidate);
}
}
return matches;
}
static int
node_priority_fencing_delay(const pcmk_node_t *node,
const pcmk_scheduler_t *scheduler)
{
int member_count = 0;
int online_count = 0;
int top_priority = 0;
int lowest_priority = 0;
GList *gIter = NULL;
// PCMK_OPT_PRIORITY_FENCING_DELAY is disabled
if (scheduler->priv->priority_fencing_ms == 0U) {
return 0;
}
/* No need to request a delay if the fencing target is not a normal cluster
* member, for example if it's a remote node or a guest node. */
if (node->priv->variant != pcmk__node_variant_cluster) {
return 0;
}
// No need to request a delay if the fencing target is in our partition
if (node->details->online) {
return 0;
}
for (gIter = scheduler->nodes; gIter != NULL; gIter = gIter->next) {
pcmk_node_t *n = gIter->data;
if (n->priv->variant != pcmk__node_variant_cluster) {
continue;
}
member_count ++;
if (n->details->online) {
online_count++;
}
if (member_count == 1
|| n->priv->priority > top_priority) {
top_priority = n->priv->priority;
}
if (member_count == 1
|| n->priv->priority < lowest_priority) {
lowest_priority = n->priv->priority;
}
}
// No need to delay if we have more than half of the cluster members
if (online_count > member_count / 2) {
return 0;
}
/* All the nodes have equal priority.
* Any configured corresponding `pcmk_delay_base/max` will be applied. */
if (lowest_priority == top_priority) {
return 0;
}
if (node->priv->priority < top_priority) {
return 0;
}
return pcmk__timeout_ms2s(scheduler->priv->priority_fencing_ms);
}
pcmk_action_t *
pe_fence_op(pcmk_node_t *node, const char *op, bool optional,
const char *reason, bool priority_delay,
pcmk_scheduler_t *scheduler)
{
char *op_key = NULL;
pcmk_action_t *stonith_op = NULL;
if(op == NULL) {
op = scheduler->priv->fence_action;
}
op_key = crm_strdup_printf("%s-%s-%s",
PCMK_ACTION_STONITH, node->priv->name, op);
stonith_op = lookup_singleton(scheduler, op_key);
if(stonith_op == NULL) {
stonith_op = custom_action(NULL, op_key, PCMK_ACTION_STONITH, node,
TRUE, scheduler);
pcmk__insert_meta(stonith_op, PCMK__META_ON_NODE, node->priv->name);
pcmk__insert_meta(stonith_op, PCMK__META_ON_NODE_UUID,
node->priv->id);
pcmk__insert_meta(stonith_op, PCMK__META_STONITH_ACTION, op);
if (pcmk_is_set(scheduler->flags, pcmk__sched_enable_unfencing)) {
/* Extra work to detect device changes
*/
GString *digests_all = g_string_sized_new(1024);
GString *digests_secure = g_string_sized_new(1024);
GList *matches = find_unfencing_devices(scheduler->priv->resources,
NULL);
for (GList *gIter = matches; gIter != NULL; gIter = gIter->next) {
pcmk_resource_t *match = gIter->data;
const char *agent = g_hash_table_lookup(match->priv->meta,
PCMK_XA_TYPE);
pcmk__op_digest_t *data = NULL;
data = pe__compare_fencing_digest(match, agent, node,
scheduler);
if (data->rc == pcmk__digest_mismatch) {
optional = FALSE;
crm_notice("Unfencing node %s because the definition of "
"%s changed", pcmk__node_name(node), match->id);
if (!pcmk__is_daemon && (scheduler->priv->out != NULL)) {
pcmk__output_t *out = scheduler->priv->out;
out->info(out,
"notice: Unfencing node %s because the "
"definition of %s changed",
pcmk__node_name(node), match->id);
}
}
pcmk__g_strcat(digests_all,
match->id, ":", agent, ":",
data->digest_all_calc, ",", NULL);
pcmk__g_strcat(digests_secure,
match->id, ":", agent, ":",
data->digest_secure_calc, ",", NULL);
}
pcmk__insert_dup(stonith_op->meta, PCMK__META_DIGESTS_ALL,
digests_all->str);
g_string_free(digests_all, TRUE);
pcmk__insert_dup(stonith_op->meta, PCMK__META_DIGESTS_SECURE,
digests_secure->str);
g_string_free(digests_secure, TRUE);
g_list_free(matches);
}
} else {
free(op_key);
}
if ((scheduler->priv->priority_fencing_ms > 0U)
/* It's a suitable case where PCMK_OPT_PRIORITY_FENCING_DELAY
* applies. At least add PCMK_OPT_PRIORITY_FENCING_DELAY field as
* an indicator.
*/
&& (priority_delay
/* The priority delay needs to be recalculated if this function has
* been called by schedule_fencing_and_shutdowns() after node
* priority has already been calculated by native_add_running().
*/
|| g_hash_table_lookup(stonith_op->meta,
PCMK_OPT_PRIORITY_FENCING_DELAY) != NULL)) {
/* Add PCMK_OPT_PRIORITY_FENCING_DELAY to the fencing op even if
* it's 0 for the targeting node. So that it takes precedence over
* any possible `pcmk_delay_base/max`.
*/
char *delay_s = pcmk__itoa(node_priority_fencing_delay(node,
scheduler));
g_hash_table_insert(stonith_op->meta,
strdup(PCMK_OPT_PRIORITY_FENCING_DELAY),
delay_s);
}
if(optional == FALSE && pe_can_fence(scheduler, node)) {
pcmk__clear_action_flags(stonith_op, pcmk__action_optional);
pe_action_set_reason(stonith_op, reason, false);
} else if(reason && stonith_op->reason == NULL) {
stonith_op->reason = strdup(reason);
}
return stonith_op;
}
void
pe_free_action(pcmk_action_t *action)
{
if (action == NULL) {
return;
}
g_list_free_full(action->actions_before, free);
g_list_free_full(action->actions_after, free);
if (action->extra) {
g_hash_table_destroy(action->extra);
}
if (action->meta) {
g_hash_table_destroy(action->meta);
}
pcmk__free_node_copy(action->node);
free(action->cancel_task);
free(action->reason);
free(action->task);
free(action->uuid);
free(action);
}
enum pcmk__action_type
get_complex_task(const pcmk_resource_t *rsc, const char *name)
{
enum pcmk__action_type task = pcmk__parse_action(name);
if (pcmk__is_primitive(rsc)) {
switch (task) {
case pcmk__action_stopped:
case pcmk__action_started:
case pcmk__action_demoted:
case pcmk__action_promoted:
crm_trace("Folding %s back into its atomic counterpart for %s",
name, rsc->id);
--task;
break;
default:
break;
}
}
return task;
}
/*!
* \internal
* \brief Find first matching action in a list
*
* \param[in] input List of actions to search
* \param[in] uuid If not NULL, action must have this UUID
* \param[in] task If not NULL, action must have this action name
* \param[in] on_node If not NULL, action must be on this node
*
* \return First action in list that matches criteria, or NULL if none
*/
pcmk_action_t *
find_first_action(const GList *input, const char *uuid, const char *task,
const pcmk_node_t *on_node)
{
CRM_CHECK(uuid || task, return NULL);
for (const GList *gIter = input; gIter != NULL; gIter = gIter->next) {
pcmk_action_t *action = (pcmk_action_t *) gIter->data;
if (uuid != NULL && !pcmk__str_eq(uuid, action->uuid, pcmk__str_casei)) {
continue;
} else if (task != NULL && !pcmk__str_eq(task, action->task, pcmk__str_casei)) {
continue;
} else if (on_node == NULL) {
return action;
} else if (action->node == NULL) {
continue;
} else if (pcmk__same_node(on_node, action->node)) {
return action;
}
}
return NULL;
}
GList *
find_actions(GList *input, const char *key, const pcmk_node_t *on_node)
{
GList *gIter = input;
GList *result = NULL;
CRM_CHECK(key != NULL, return NULL);
for (; gIter != NULL; gIter = gIter->next) {
pcmk_action_t *action = (pcmk_action_t *) gIter->data;
if (!pcmk__str_eq(key, action->uuid, pcmk__str_casei)) {
continue;
} else if (on_node == NULL) {
crm_trace("Action %s matches (ignoring node)", key);
result = g_list_prepend(result, action);
} else if (action->node == NULL) {
crm_trace("Action %s matches (unallocated, assigning to %s)",
key, pcmk__node_name(on_node));
action->node = pe__copy_node(on_node);
result = g_list_prepend(result, action);
} else if (pcmk__same_node(on_node, action->node)) {
crm_trace("Action %s on %s matches", key, pcmk__node_name(on_node));
result = g_list_prepend(result, action);
}
}
return result;
}
GList *
find_actions_exact(GList *input, const char *key, const pcmk_node_t *on_node)
{
GList *result = NULL;
CRM_CHECK(key != NULL, return NULL);
if (on_node == NULL) {
return NULL;
}
for (GList *gIter = input; gIter != NULL; gIter = gIter->next) {
pcmk_action_t *action = (pcmk_action_t *) gIter->data;
if ((action->node != NULL)
&& pcmk__str_eq(key, action->uuid, pcmk__str_casei)
&& pcmk__same_node(on_node, action->node)) {
crm_trace("Action %s on %s matches", key, pcmk__node_name(on_node));
result = g_list_prepend(result, action);
}
}
return result;
}
/*!
* \brief Find all actions of given type for a resource
*
* \param[in] rsc Resource to search
* \param[in] node Find only actions scheduled on this node
* \param[in] task Action name to search for
* \param[in] require_node If TRUE, NULL node or action node will not match
*
* \return List of actions found (or NULL if none)
* \note If node is not NULL and require_node is FALSE, matching actions
* without a node will be assigned to node.
*/
GList *
pe__resource_actions(const pcmk_resource_t *rsc, const pcmk_node_t *node,
const char *task, bool require_node)
{
GList *result = NULL;
char *key = pcmk__op_key(rsc->id, task, 0);
if (require_node) {
result = find_actions_exact(rsc->priv->actions, key, node);
} else {
result = find_actions(rsc->priv->actions, key, node);
}
free(key);
return result;
}
/*!
* \internal
* \brief Create an action reason string based on the action itself
*
* \param[in] action Action to create reason string for
* \param[in] flag Action flag that was cleared
*
* \return Newly allocated string suitable for use as action reason
* \note It is the caller's responsibility to free() the result.
*/
char *
pe__action2reason(const pcmk_action_t *action, enum pcmk__action_flags flag)
{
const char *change = NULL;
switch (flag) {
case pcmk__action_runnable:
change = "unrunnable";
break;
case pcmk__action_migratable:
change = "unmigrateable";
break;
case pcmk__action_optional:
change = "required";
break;
default:
// Bug: caller passed unsupported flag
CRM_CHECK(change != NULL, change = "");
break;
}
return crm_strdup_printf("%s%s%s %s", change,
(action->rsc == NULL)? "" : " ",
(action->rsc == NULL)? "" : action->rsc->id,
action->task);
}
void pe_action_set_reason(pcmk_action_t *action, const char *reason,
bool overwrite)
{
if (action->reason != NULL && overwrite) {
pcmk__rsc_trace(action->rsc, "Changing %s reason from '%s' to '%s'",
action->uuid, action->reason,
pcmk__s(reason, "(none)"));
} else if (action->reason == NULL) {
pcmk__rsc_trace(action->rsc, "Set %s reason to '%s'",
action->uuid, pcmk__s(reason, "(none)"));
} else {
// crm_assert(action->reason != NULL && !overwrite);
return;
}
pcmk__str_update(&action->reason, reason);
}
/*!
* \internal
* \brief Create an action to clear a resource's history from CIB
*
* \param[in,out] rsc Resource to clear
* \param[in] node Node to clear history on
*/
void
pe__clear_resource_history(pcmk_resource_t *rsc, const pcmk_node_t *node)
{
pcmk__assert((rsc != NULL) && (node != NULL));
custom_action(rsc, pcmk__op_key(rsc->id, PCMK_ACTION_LRM_DELETE, 0),
PCMK_ACTION_LRM_DELETE, node, FALSE, rsc->priv->scheduler);
}
#define sort_return(an_int, why) do { \
free(a_uuid); \
free(b_uuid); \
crm_trace("%s (%d) %c %s (%d) : %s", \
a_xml_id, a_call_id, an_int>0?'>':an_int<0?'<':'=', \
b_xml_id, b_call_id, why); \
return an_int; \
} while(0)
int
pe__is_newer_op(const xmlNode *xml_a, const xmlNode *xml_b)
{
int a_call_id = -1;
int b_call_id = -1;
char *a_uuid = NULL;
char *b_uuid = NULL;
const char *a_xml_id = crm_element_value(xml_a, PCMK_XA_ID);
const char *b_xml_id = crm_element_value(xml_b, PCMK_XA_ID);
const char *a_node = crm_element_value(xml_a, PCMK__META_ON_NODE);
const char *b_node = crm_element_value(xml_b, PCMK__META_ON_NODE);
bool same_node = pcmk__str_eq(a_node, b_node, pcmk__str_casei);
if (same_node && pcmk__str_eq(a_xml_id, b_xml_id, pcmk__str_none)) {
/* We have duplicate PCMK__XE_LRM_RSC_OP entries in the status
* section which is unlikely to be a good thing
* - we can handle it easily enough, but we need to get
* to the bottom of why it's happening.
*/
pcmk__config_err("Duplicate " PCMK__XE_LRM_RSC_OP " entries named %s",
a_xml_id);
sort_return(0, "duplicate");
}
crm_element_value_int(xml_a, PCMK__XA_CALL_ID, &a_call_id);
crm_element_value_int(xml_b, PCMK__XA_CALL_ID, &b_call_id);
if (a_call_id == -1 && b_call_id == -1) {
/* both are pending ops so it doesn't matter since
* stops are never pending
*/
sort_return(0, "pending");
} else if (same_node && a_call_id >= 0 && a_call_id < b_call_id) {
sort_return(-1, "call id");
} else if (same_node && b_call_id >= 0 && a_call_id > b_call_id) {
sort_return(1, "call id");
} else if (a_call_id >= 0 && b_call_id >= 0
&& (!same_node || a_call_id == b_call_id)) {
/* The op and last_failed_op are the same. Order on
* PCMK_XA_LAST_RC_CHANGE.
*/
time_t last_a = -1;
time_t last_b = -1;
crm_element_value_epoch(xml_a, PCMK_XA_LAST_RC_CHANGE, &last_a);
crm_element_value_epoch(xml_b, PCMK_XA_LAST_RC_CHANGE, &last_b);
crm_trace("rc-change: %lld vs %lld",
(long long) last_a, (long long) last_b);
if (last_a >= 0 && last_a < last_b) {
sort_return(-1, "rc-change");
} else if (last_b >= 0 && last_a > last_b) {
sort_return(1, "rc-change");
}
sort_return(0, "rc-change");
} else {
/* One of the inputs is a pending operation.
* Attempt to use PCMK__XA_TRANSITION_MAGIC to determine its age relative
* to the other.
*/
int a_id = -1;
int b_id = -1;
const char *a_magic = crm_element_value(xml_a,
PCMK__XA_TRANSITION_MAGIC);
const char *b_magic = crm_element_value(xml_b,
PCMK__XA_TRANSITION_MAGIC);
CRM_CHECK(a_magic != NULL && b_magic != NULL, sort_return(0, "No magic"));
if (!decode_transition_magic(a_magic, &a_uuid, &a_id, NULL, NULL, NULL,
NULL)) {
sort_return(0, "bad magic a");
}
if (!decode_transition_magic(b_magic, &b_uuid, &b_id, NULL, NULL, NULL,
NULL)) {
sort_return(0, "bad magic b");
}
/* try to determine the relative age of the operation...
* some pending operations (e.g. a start) may have been superseded
* by a subsequent stop
*
* [a|b]_id == -1 means it's a shutdown operation and _always_ comes last
*/
if (!pcmk__str_eq(a_uuid, b_uuid, pcmk__str_casei) || a_id == b_id) {
/*
* some of the logic in here may be redundant...
*
* if the UUID from the TE doesn't match then one better
* be a pending operation.
* pending operations don't survive between elections and joins
* because we query the LRM directly
*/
if (b_call_id == -1) {
sort_return(-1, "transition + call");
} else if (a_call_id == -1) {
sort_return(1, "transition + call");
}
} else if ((a_id >= 0 && a_id < b_id) || b_id == -1) {
sort_return(-1, "transition");
} else if ((b_id >= 0 && a_id > b_id) || a_id == -1) {
sort_return(1, "transition");
}
}
/* we should never end up here */
CRM_CHECK(FALSE, sort_return(0, "default"));
}
gint
sort_op_by_callid(gconstpointer a, gconstpointer b)
{
return pe__is_newer_op((const xmlNode *) a, (const xmlNode *) b);
}
/*!
* \internal
* \brief Create a new pseudo-action for a resource
*
* \param[in,out] rsc Resource to create action for
* \param[in] task Action name
* \param[in] optional Whether action should be considered optional
* \param[in] runnable Whethe action should be considered runnable
*
* \return New action object corresponding to arguments
*/
pcmk_action_t *
pe__new_rsc_pseudo_action(pcmk_resource_t *rsc, const char *task, bool optional,
bool runnable)
{
pcmk_action_t *action = NULL;
pcmk__assert((rsc != NULL) && (task != NULL));
action = custom_action(rsc, pcmk__op_key(rsc->id, task, 0), task, NULL,
optional, rsc->priv->scheduler);
pcmk__set_action_flags(action, pcmk__action_pseudo);
if (runnable) {
pcmk__set_action_flags(action, pcmk__action_runnable);
}
return action;
}
/*!
* \internal
* \brief Add the expected result to an action
*
* \param[in,out] action Action to add expected result to
* \param[in] expected_result Expected result to add
*
* \note This is more efficient than calling pcmk__insert_meta().
*/
void
pe__add_action_expected_result(pcmk_action_t *action, int expected_result)
{
pcmk__assert((action != NULL) && (action->meta != NULL));
g_hash_table_insert(action->meta, pcmk__str_copy(PCMK__META_OP_TARGET_RC),
pcmk__itoa(expected_result));
}
diff --git a/lib/pengine/pe_notif.c b/lib/pengine/pe_notif.c
index 4e17cdd5c2..89d678a712 100644
--- a/lib/pengine/pe_notif.c
+++ b/lib/pengine/pe_notif.c
@@ -1,1016 +1,1027 @@
/*
* Copyright 2004-2024 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU General Public License version 2
* or later (GPLv2+) WITHOUT ANY WARRANTY.
*/
#include <crm_internal.h>
#include <crm/common/xml.h>
#include <crm/pengine/internal.h>
#include <pacemaker-internal.h>
#include "pe_status_private.h"
typedef struct notify_entry_s {
const pcmk_resource_t *rsc;
const pcmk_node_t *node;
} notify_entry_t;
/*!
* \internal
* \brief Compare two notification entries
*
* Compare two notification entries, where the one with the alphabetically first
* resource name (or if equal, node ID) sorts as first, with NULL sorting as
* less than non-NULL.
*
* \param[in] a First notification entry to compare
* \param[in] b Second notification entry to compare
*
* \return -1 if \p a sorts before \p b, 0 if they are equal, otherwise 1
*/
static gint
compare_notify_entries(gconstpointer a, gconstpointer b)
{
int tmp;
const notify_entry_t *entry_a = a;
const notify_entry_t *entry_b = b;
// NULL a or b is not actually possible
if ((entry_a == NULL) && (entry_b == NULL)) {
return 0;
}
if (entry_a == NULL) {
return 1;
}
if (entry_b == NULL) {
return -1;
}
// NULL resources sort first
if ((entry_a->rsc == NULL) && (entry_b->rsc == NULL)) {
return 0;
}
if (entry_a->rsc == NULL) {
return 1;
}
if (entry_b->rsc == NULL) {
return -1;
}
// Compare resource names
tmp = strcmp(entry_a->rsc->id, entry_b->rsc->id);
if (tmp != 0) {
return tmp;
}
// Otherwise NULL nodes sort first
if ((entry_a->node == NULL) && (entry_b->node == NULL)) {
return 0;
}
if (entry_a->node == NULL) {
return 1;
}
if (entry_b->node == NULL) {
return -1;
}
// Finally, compare node IDs
return strcmp(entry_a->node->priv->id, entry_b->node->priv->id);
}
/*!
* \internal
* \brief Duplicate a notification entry
*
* \param[in] entry Entry to duplicate
*
* \return Newly allocated duplicate of \p entry
* \note It is the caller's responsibility to free the return value.
*/
static notify_entry_t *
dup_notify_entry(const notify_entry_t *entry)
{
notify_entry_t *dup = pcmk__assert_alloc(1, sizeof(notify_entry_t));
dup->rsc = entry->rsc;
dup->node = entry->node;
return dup;
}
/*!
* \internal
* \brief Given a list of nodes, create strings with node names
*
* \param[in] list List of nodes (as pcmk_node_t *)
* \param[out] all_node_names If not NULL, will be set to space-separated list
* of the names of all nodes in \p list
* \param[out] host_node_names Same as \p all_node_names, except active
* guest nodes will list the name of their host
*
* \note The caller is responsible for freeing the output argument values using
* \p g_string_free().
*/
static void
get_node_names(const GList *list, GString **all_node_names,
GString **host_node_names)
{
if (all_node_names != NULL) {
*all_node_names = NULL;
}
if (host_node_names != NULL) {
*host_node_names = NULL;
}
for (const GList *iter = list; iter != NULL; iter = iter->next) {
const pcmk_node_t *node = (const pcmk_node_t *) iter->data;
if (node->priv->name == NULL) {
+ /* @TODO This breaks the comparability of the various notification
+ * variables and thus any agent relying on it. Maybe add "UNKNOWN"
+ * or something like that.
+ */
continue;
}
// Always add to list of all node names
if (all_node_names != NULL) {
pcmk__add_word(all_node_names, 1024, node->priv->name);
}
// Add to host node name list if appropriate
if (host_node_names != NULL) {
if (pcmk__is_guest_or_bundle_node(node)) {
const pcmk_resource_t *launcher = NULL;
launcher = node->priv->remote->priv->launcher;
if (launcher->priv->active_nodes != NULL) {
node = pcmk__current_node(launcher);
if (node->priv->name == NULL) {
continue;
}
}
}
pcmk__add_word(host_node_names, 1024, node->priv->name);
}
}
if ((all_node_names != NULL) && (*all_node_names == NULL)) {
*all_node_names = g_string_new(" ");
}
if ((host_node_names != NULL) && (*host_node_names == NULL)) {
*host_node_names = g_string_new(" ");
}
}
/*!
* \internal
* \brief Create strings of instance and node names from notification entries
*
* \param[in,out] list List of notification entries (will be sorted here)
* \param[out] rsc_names If not NULL, will be set to space-separated list
* of clone instances from \p list
* \param[out] node_names If not NULL, will be set to space-separated list
* of node names from \p list
*
* \return (Possibly new) head of sorted \p list
* \note The caller is responsible for freeing the output argument values using
* \p g_list_free_full() and \p g_string_free().
*/
static GList *
notify_entries_to_strings(GList *list, GString **rsc_names,
GString **node_names)
{
const char *last_rsc_id = NULL;
// Initialize output lists to NULL
if (rsc_names != NULL) {
*rsc_names = NULL;
}
if (node_names != NULL) {
*node_names = NULL;
}
// Sort input list for user-friendliness (and ease of filtering duplicates)
list = g_list_sort(list, compare_notify_entries);
for (GList *gIter = list; gIter != NULL; gIter = gIter->next) {
notify_entry_t *entry = (notify_entry_t *) gIter->data;
// Entry must have a resource (with ID)
CRM_LOG_ASSERT((entry != NULL) && (entry->rsc != NULL)
&& (entry->rsc->id != NULL));
if ((entry == NULL) || (entry->rsc == NULL)
|| (entry->rsc->id == NULL)) {
continue;
}
// Entry must have a node unless listing inactive resources
CRM_LOG_ASSERT((node_names == NULL) || (entry->node != NULL));
if ((node_names != NULL) && (entry->node == NULL)) {
continue;
}
// Don't add duplicates of a particular clone instance
if (pcmk__str_eq(entry->rsc->id, last_rsc_id, pcmk__str_none)) {
continue;
}
last_rsc_id = entry->rsc->id;
if (rsc_names != NULL) {
pcmk__add_word(rsc_names, 1024, entry->rsc->id);
}
if ((node_names != NULL) && (entry->node->priv->name != NULL)) {
pcmk__add_word(node_names, 1024, entry->node->priv->name);
}
}
// If there are no entries, return "empty" lists
if ((rsc_names != NULL) && (*rsc_names == NULL)) {
*rsc_names = g_string_new(" ");
}
if ((node_names != NULL) && (*node_names == NULL)) {
*node_names = g_string_new(" ");
}
return list;
}
/*!
* \internal
* \brief Copy a meta-attribute into a notify action
*
* \param[in] key Name of meta-attribute to copy
* \param[in] value Value of meta-attribute to copy
* \param[in,out] user_data Notify action to copy into
*/
static void
copy_meta_to_notify(gpointer key, gpointer value, gpointer user_data)
{
pcmk_action_t *notify = (pcmk_action_t *) user_data;
/* Any existing meta-attributes (for example, the action timeout) are for
* the notify action itself, so don't override those.
*/
if (g_hash_table_lookup(notify->meta, (const char *) key) != NULL) {
return;
}
pcmk__insert_dup(notify->meta, (const char *) key, (const char *) value);
}
static void
add_notify_data_to_action_meta(const notify_data_t *n_data,
pcmk_action_t *action)
{
for (const GSList *item = n_data->keys; item; item = item->next) {
const pcmk_nvpair_t *nvpair = (const pcmk_nvpair_t *) item->data;
pcmk__insert_meta(action, nvpair->name, nvpair->value);
}
}
/*!
* \internal
* \brief Create a new notify pseudo-action for a clone resource
*
* \param[in,out] rsc Clone resource that notification is for
* \param[in] action Action to use in notify action key
* \param[in] notif_action PCMK_ACTION_NOTIFY or PCMK_ACTION_NOTIFIED
* \param[in] notif_type "pre", "post", "confirmed-pre", "confirmed-post"
*
* \return Newly created notify pseudo-action
*/
static pcmk_action_t *
new_notify_pseudo_action(pcmk_resource_t *rsc, const pcmk_action_t *action,
const char *notif_action, const char *notif_type)
{
pcmk_action_t *notify = NULL;
notify = custom_action(rsc,
pcmk__notify_key(rsc->id, notif_type, action->task),
notif_action, NULL,
pcmk_is_set(action->flags, pcmk__action_optional),
rsc->priv->scheduler);
pcmk__set_action_flags(notify, pcmk__action_pseudo);
pcmk__insert_meta(notify, "notify_key_type", notif_type);
pcmk__insert_meta(notify, "notify_key_operation", action->task);
return notify;
}
/*!
* \internal
* \brief Create a new notify action for a clone instance
*
* \param[in,out] rsc Clone instance that notification is for
* \param[in] node Node that notification is for
* \param[in,out] op Action that notification is for
* \param[in,out] notify_done Parent pseudo-action for notifications complete
* \param[in] n_data Notification values to add to action meta-data
*
* \return Newly created notify action
*/
static pcmk_action_t *
new_notify_action(pcmk_resource_t *rsc, const pcmk_node_t *node,
pcmk_action_t *op, pcmk_action_t *notify_done,
const notify_data_t *n_data)
{
char *key = NULL;
pcmk_action_t *notify_action = NULL;
const char *value = NULL;
const char *task = NULL;
const char *skip_reason = NULL;
CRM_CHECK((rsc != NULL) && (node != NULL), return NULL);
// Ensure we have all the info we need
if (op == NULL) {
skip_reason = "no action";
} else if (notify_done == NULL) {
skip_reason = "no parent notification";
} else if (!node->details->online) {
skip_reason = "node offline";
} else if (!pcmk_is_set(op->flags, pcmk__action_runnable)) {
skip_reason = "original action not runnable";
}
if (skip_reason != NULL) {
pcmk__rsc_trace(rsc, "Skipping notify action for %s on %s: %s",
rsc->id, pcmk__node_name(node), skip_reason);
return NULL;
}
value = g_hash_table_lookup(op->meta, "notify_type"); // "pre" or "post"
task = g_hash_table_lookup(op->meta, "notify_operation"); // original action
pcmk__rsc_trace(rsc, "Creating notify action for %s on %s (%s-%s)",
rsc->id, pcmk__node_name(node), value, task);
// Create the notify action
key = pcmk__notify_key(rsc->id, value, task);
notify_action = custom_action(rsc, key, op->task, node,
pcmk_is_set(op->flags, pcmk__action_optional),
rsc->priv->scheduler);
// Add meta-data to notify action
g_hash_table_foreach(op->meta, copy_meta_to_notify, notify_action);
add_notify_data_to_action_meta(n_data, notify_action);
// Order notify after original action and before parent notification
order_actions(op, notify_action, pcmk__ar_ordered);
order_actions(notify_action, notify_done, pcmk__ar_ordered);
return notify_action;
}
/*!
* \internal
* \brief Create a new "post-" notify action for a clone instance
*
* \param[in,out] rsc Clone instance that notification is for
* \param[in] node Node that notification is for
* \param[in,out] n_data Notification values to add to action meta-data
*/
static void
new_post_notify_action(pcmk_resource_t *rsc, const pcmk_node_t *node,
notify_data_t *n_data)
{
pcmk_action_t *notify = NULL;
pcmk__assert(n_data != NULL);
// Create the "post-" notify action for specified instance
notify = new_notify_action(rsc, node, n_data->post, n_data->post_done,
n_data);
if (notify != NULL) {
notify->priority = PCMK_SCORE_INFINITY;
}
// Order recurring monitors after all "post-" notifications complete
if (n_data->post_done == NULL) {
return;
}
for (GList *iter = rsc->priv->actions; iter != NULL; iter = iter->next) {
pcmk_action_t *mon = (pcmk_action_t *) iter->data;
const char *interval_ms_s = NULL;
interval_ms_s = g_hash_table_lookup(mon->meta, PCMK_META_INTERVAL);
if (pcmk__str_eq(interval_ms_s, "0", pcmk__str_null_matches)
|| pcmk__str_eq(mon->task, PCMK_ACTION_CANCEL, pcmk__str_none)) {
continue; // Not a recurring monitor
}
order_actions(n_data->post_done, mon, pcmk__ar_ordered);
}
}
/*!
* \internal
* \brief Create and order notification pseudo-actions for a clone action
*
* In addition to the actual notify actions needed for each clone instance,
* clone notifications also require pseudo-actions to provide ordering points
* in the notification process. This creates the notification data, along with
* appropriate pseudo-actions and their orderings.
*
* For example, the ordering sequence for starting a clone is:
*
* "pre-" notify pseudo-action for clone
* -> "pre-" notify actions for each clone instance
* -> "pre-" notifications complete pseudo-action for clone
* -> start actions for each clone instance
* -> "started" pseudo-action for clone
* -> "post-" notify pseudo-action for clone
* -> "post-" notify actions for each clone instance
* -> "post-" notifications complete pseudo-action for clone
*
* \param[in,out] rsc Clone that notifications are for
* \param[in] task Name of action that notifications are for
* \param[in,out] action If not NULL, create a "pre-" pseudo-action ordered
* before a "pre-" complete pseudo-action, ordered
* before this action
* \param[in,out] complete If not NULL, create a "post-" pseudo-action ordered
* after this action, and a "post-" complete
* pseudo-action ordered after that
*
* \return Newly created notification data
*/
notify_data_t *
pe__action_notif_pseudo_ops(pcmk_resource_t *rsc, const char *task,
pcmk_action_t *action, pcmk_action_t *complete)
{
notify_data_t *n_data = NULL;
if (!pcmk_is_set(rsc->flags, pcmk__rsc_notify)) {
return NULL;
}
n_data = pcmk__assert_alloc(1, sizeof(notify_data_t));
n_data->action = task;
if (action != NULL) { // Need "pre-" pseudo-actions
// Create "pre-" notify pseudo-action for clone
n_data->pre = new_notify_pseudo_action(rsc, action, PCMK_ACTION_NOTIFY,
"pre");
pcmk__set_action_flags(n_data->pre, pcmk__action_runnable);
pcmk__insert_meta(n_data->pre, "notify_type", "pre");
pcmk__insert_meta(n_data->pre, "notify_operation", n_data->action);
// Create "pre-" notifications complete pseudo-action for clone
n_data->pre_done = new_notify_pseudo_action(rsc, action,
PCMK_ACTION_NOTIFIED,
"confirmed-pre");
pcmk__set_action_flags(n_data->pre_done, pcmk__action_runnable);
pcmk__insert_meta(n_data->pre_done, "notify_type", "pre");
pcmk__insert_meta(n_data->pre_done, "notify_operation", n_data->action);
// Order "pre-" -> "pre-" complete -> original action
order_actions(n_data->pre, n_data->pre_done, pcmk__ar_ordered);
order_actions(n_data->pre_done, action, pcmk__ar_ordered);
}
if (complete != NULL) { // Need "post-" pseudo-actions
// Create "post-" notify pseudo-action for clone
n_data->post = new_notify_pseudo_action(rsc, complete,
PCMK_ACTION_NOTIFY, "post");
n_data->post->priority = PCMK_SCORE_INFINITY;
if (pcmk_is_set(complete->flags, pcmk__action_runnable)) {
pcmk__set_action_flags(n_data->post, pcmk__action_runnable);
} else {
pcmk__clear_action_flags(n_data->post, pcmk__action_runnable);
}
pcmk__insert_meta(n_data->post, "notify_type", "post");
pcmk__insert_meta(n_data->post, "notify_operation", n_data->action);
// Create "post-" notifications complete pseudo-action for clone
n_data->post_done = new_notify_pseudo_action(rsc, complete,
PCMK_ACTION_NOTIFIED,
"confirmed-post");
n_data->post_done->priority = PCMK_SCORE_INFINITY;
if (pcmk_is_set(complete->flags, pcmk__action_runnable)) {
pcmk__set_action_flags(n_data->post_done, pcmk__action_runnable);
} else {
pcmk__clear_action_flags(n_data->post_done, pcmk__action_runnable);
}
pcmk__insert_meta(n_data->post_done, "notify_type", "post");
pcmk__insert_meta(n_data->post_done,
"notify_operation", n_data->action);
- // Order original action complete -> "post-" -> "post-" complete
+ /* Order original action complete -> "post-" -> "post-" complete
+ *
+ * @TODO Should we add |pcmk__ar_unrunnable_first_blocks to these?
+ * Otherwise we might get an invalid transition due to unresolved
+ * dependencies when "complete" is a fencing op (which can happen at
+ * least for bundles) but that op is unrunnable (due to lack of quorum,
+ * for example).
+ */
order_actions(complete, n_data->post, pcmk__ar_first_implies_then);
order_actions(n_data->post, n_data->post_done,
pcmk__ar_first_implies_then);
}
// If we created both, order "pre-" complete -> "post-"
if ((action != NULL) && (complete != NULL)) {
order_actions(n_data->pre_done, n_data->post, pcmk__ar_ordered);
}
return n_data;
}
/*!
* \internal
* \brief Create a new notification entry
*
* \param[in] rsc Resource for notification
* \param[in] node Node for notification
*
* \return Newly allocated notification entry
* \note The caller is responsible for freeing the return value.
*/
static notify_entry_t *
new_notify_entry(const pcmk_resource_t *rsc, const pcmk_node_t *node)
{
notify_entry_t *entry = pcmk__assert_alloc(1, sizeof(notify_entry_t));
entry->rsc = rsc;
entry->node = node;
return entry;
}
/*!
* \internal
* \brief Add notification data for resource state and optionally actions
*
* \param[in,out] rsc Clone or clone instance being notified
* \param[in] activity Whether to add notification entries for actions
* \param[in,out] n_data Notification data for clone
*/
static void
collect_resource_data(pcmk_resource_t *rsc, bool activity,
notify_data_t *n_data)
{
const GList *iter = NULL;
notify_entry_t *entry = NULL;
const pcmk_node_t *node = NULL;
if (n_data == NULL) {
return;
}
if (n_data->allowed_nodes == NULL) {
n_data->allowed_nodes = rsc->priv->allowed_nodes;
}
// If this is a clone, call recursively for each instance
if (rsc->priv->children != NULL) {
for (iter = rsc->priv->children; iter != NULL; iter = iter->next) {
pcmk_resource_t *child = iter->data;
collect_resource_data(child, activity, n_data);
}
return;
}
// This is a notification for a single clone instance
if (rsc->priv->active_nodes != NULL) {
node = rsc->priv->active_nodes->data; // First is sufficient
}
entry = new_notify_entry(rsc, node);
// Add notification indicating the resource state
switch (rsc->priv->orig_role) {
case pcmk_role_stopped:
n_data->inactive = g_list_prepend(n_data->inactive, entry);
break;
case pcmk_role_started:
n_data->active = g_list_prepend(n_data->active, entry);
break;
case pcmk_role_unpromoted:
n_data->unpromoted = g_list_prepend(n_data->unpromoted, entry);
n_data->active = g_list_prepend(n_data->active,
dup_notify_entry(entry));
break;
case pcmk_role_promoted:
n_data->promoted = g_list_prepend(n_data->promoted, entry);
n_data->active = g_list_prepend(n_data->active,
dup_notify_entry(entry));
break;
default:
pcmk__sched_err(rsc->priv->scheduler,
"Resource %s role on %s (%s) is not supported for "
"notifications (bug?)",
rsc->id, pcmk__node_name(node),
pcmk_role_text(rsc->priv->orig_role));
free(entry);
break;
}
if (!activity) {
return;
}
// Add notification entries for each of the resource's actions
for (iter = rsc->priv->actions; iter != NULL; iter = iter->next) {
const pcmk_action_t *op = (const pcmk_action_t *) iter->data;
if (!pcmk_is_set(op->flags, pcmk__action_optional)
&& (op->node != NULL)) {
enum pcmk__action_type task = pcmk__parse_action(op->task);
if ((task == pcmk__action_stop) && op->node->details->unclean) {
// Create anyway (additional noise if node can't be fenced)
} else if (!pcmk_is_set(op->flags, pcmk__action_runnable)) {
continue;
}
entry = new_notify_entry(rsc, op->node);
switch (task) {
case pcmk__action_start:
n_data->start = g_list_prepend(n_data->start, entry);
break;
case pcmk__action_stop:
n_data->stop = g_list_prepend(n_data->stop, entry);
break;
case pcmk__action_promote:
n_data->promote = g_list_prepend(n_data->promote, entry);
break;
case pcmk__action_demote:
n_data->demote = g_list_prepend(n_data->demote, entry);
break;
default:
free(entry);
break;
}
}
}
}
// For (char *) value
#define add_notify_env(n_data, key, value) do { \
n_data->keys = pcmk_prepend_nvpair(n_data->keys, key, value); \
} while (0)
// For (GString *) value
#define add_notify_env_gs(n_data, key, value) do { \
n_data->keys = pcmk_prepend_nvpair(n_data->keys, key, \
(const char *) value->str); \
} while (0)
// For (GString *) value
#define add_notify_env_free_gs(n_data, key, value) do { \
n_data->keys = pcmk_prepend_nvpair(n_data->keys, key, \
(const char *) value->str); \
g_string_free(value, TRUE); value = NULL; \
} while (0)
/*!
* \internal
* \brief Create notification name/value pairs from structured data
*
* \param[in] rsc Resource that notification is for
* \param[in,out] n_data Notification data
*/
static void
add_notif_keys(const pcmk_resource_t *rsc, notify_data_t *n_data)
{
bool required = false; // Whether to make notify actions required
GString *rsc_list = NULL;
GString *node_list = NULL;
GString *metal_list = NULL;
const char *source = NULL;
GList *nodes = NULL;
n_data->stop = notify_entries_to_strings(n_data->stop,
&rsc_list, &node_list);
if ((strcmp(" ", (const char *) rsc_list->str) != 0)
&& pcmk__str_eq(n_data->action, PCMK_ACTION_STOP, pcmk__str_none)) {
required = true;
}
add_notify_env_free_gs(n_data, "notify_stop_resource", rsc_list);
add_notify_env_free_gs(n_data, "notify_stop_uname", node_list);
if ((n_data->start != NULL)
&& pcmk__str_eq(n_data->action, PCMK_ACTION_START, pcmk__str_none)) {
required = true;
}
n_data->start = notify_entries_to_strings(n_data->start,
&rsc_list, &node_list);
add_notify_env_free_gs(n_data, "notify_start_resource", rsc_list);
add_notify_env_free_gs(n_data, "notify_start_uname", node_list);
if ((n_data->demote != NULL)
&& pcmk__str_eq(n_data->action, PCMK_ACTION_DEMOTE, pcmk__str_none)) {
required = true;
}
n_data->demote = notify_entries_to_strings(n_data->demote,
&rsc_list, &node_list);
add_notify_env_free_gs(n_data, "notify_demote_resource", rsc_list);
add_notify_env_free_gs(n_data, "notify_demote_uname", node_list);
if ((n_data->promote != NULL)
&& pcmk__str_eq(n_data->action, PCMK_ACTION_PROMOTE, pcmk__str_none)) {
required = true;
}
n_data->promote = notify_entries_to_strings(n_data->promote,
&rsc_list, &node_list);
add_notify_env_free_gs(n_data, "notify_promote_resource", rsc_list);
add_notify_env_free_gs(n_data, "notify_promote_uname", node_list);
n_data->active = notify_entries_to_strings(n_data->active,
&rsc_list, &node_list);
add_notify_env_free_gs(n_data, "notify_active_resource", rsc_list);
add_notify_env_free_gs(n_data, "notify_active_uname", node_list);
n_data->unpromoted = notify_entries_to_strings(n_data->unpromoted,
&rsc_list, &node_list);
add_notify_env_gs(n_data, "notify_unpromoted_resource", rsc_list);
add_notify_env_gs(n_data, "notify_unpromoted_uname", node_list);
// Deprecated: kept for backward compatibility with older resource agents
add_notify_env_free_gs(n_data, "notify_slave_resource", rsc_list);
add_notify_env_free_gs(n_data, "notify_slave_uname", node_list);
n_data->promoted = notify_entries_to_strings(n_data->promoted,
&rsc_list, &node_list);
add_notify_env_gs(n_data, "notify_promoted_resource", rsc_list);
add_notify_env_gs(n_data, "notify_promoted_uname", node_list);
// Deprecated: kept for backward compatibility with older resource agents
add_notify_env_free_gs(n_data, "notify_master_resource", rsc_list);
add_notify_env_free_gs(n_data, "notify_master_uname", node_list);
n_data->inactive = notify_entries_to_strings(n_data->inactive,
&rsc_list, NULL);
add_notify_env_free_gs(n_data, "notify_inactive_resource", rsc_list);
nodes = g_hash_table_get_values(n_data->allowed_nodes);
if (!pcmk__is_daemon) {
/* For display purposes, sort the node list, for consistent
* regression test output (while avoiding the performance hit
* for the live cluster).
*/
nodes = g_list_sort(nodes, pe__cmp_node_name);
}
get_node_names(nodes, &node_list, NULL);
add_notify_env_free_gs(n_data, "notify_available_uname", node_list);
g_list_free(nodes);
source = g_hash_table_lookup(rsc->priv->meta,
PCMK_META_CONTAINER_ATTRIBUTE_TARGET);
if (pcmk__str_eq(PCMK_VALUE_HOST, source, pcmk__str_none)) {
get_node_names(rsc->priv->scheduler->nodes, &node_list, &metal_list);
add_notify_env_free_gs(n_data, "notify_all_hosts", metal_list);
} else {
get_node_names(rsc->priv->scheduler->nodes, &node_list, NULL);
}
add_notify_env_free_gs(n_data, "notify_all_uname", node_list);
if (required && (n_data->pre != NULL)) {
pcmk__clear_action_flags(n_data->pre, pcmk__action_optional);
pcmk__clear_action_flags(n_data->pre_done, pcmk__action_optional);
}
if (required && (n_data->post != NULL)) {
pcmk__clear_action_flags(n_data->post, pcmk__action_optional);
pcmk__clear_action_flags(n_data->post_done, pcmk__action_optional);
}
}
/*
* \internal
* \brief Find any remote connection start relevant to an action
*
* \param[in] action Action to check
*
* \return If action is behind a remote connection, connection's start
*/
static pcmk_action_t *
find_remote_start(pcmk_action_t *action)
{
if ((action != NULL) && (action->node != NULL)) {
pcmk_resource_t *remote_rsc = action->node->priv->remote;
if (remote_rsc != NULL) {
return find_first_action(remote_rsc->priv->actions, NULL,
PCMK_ACTION_START,
NULL);
}
}
return NULL;
}
/*!
* \internal
* \brief Create notify actions, and add notify data to original actions
*
* \param[in,out] rsc Clone or clone instance that notification is for
* \param[in,out] n_data Clone notification data for some action
*/
static void
create_notify_actions(pcmk_resource_t *rsc, notify_data_t *n_data)
{
GList *iter = NULL;
pcmk_action_t *stop = NULL;
pcmk_action_t *start = NULL;
enum pcmk__action_type task = pcmk__parse_action(n_data->action);
// If this is a clone, call recursively for each instance
if (rsc->priv->children != NULL) {
g_list_foreach(rsc->priv->children, (GFunc) create_notify_actions,
n_data);
return;
}
// Add notification meta-attributes to original actions
for (iter = rsc->priv->actions; iter != NULL; iter = iter->next) {
pcmk_action_t *op = (pcmk_action_t *) iter->data;
if (!pcmk_is_set(op->flags, pcmk__action_optional)
&& (op->node != NULL)) {
switch (pcmk__parse_action(op->task)) {
case pcmk__action_start:
case pcmk__action_stop:
case pcmk__action_promote:
case pcmk__action_demote:
add_notify_data_to_action_meta(n_data, op);
break;
default:
break;
}
}
}
// Skip notify action itself if original action was not needed
switch (task) {
case pcmk__action_start:
if (n_data->start == NULL) {
pcmk__rsc_trace(rsc, "No notify action needed for %s %s",
rsc->id, n_data->action);
return;
}
break;
case pcmk__action_promote:
if (n_data->promote == NULL) {
pcmk__rsc_trace(rsc, "No notify action needed for %s %s",
rsc->id, n_data->action);
return;
}
break;
case pcmk__action_demote:
if (n_data->demote == NULL) {
pcmk__rsc_trace(rsc, "No notify action needed for %s %s",
rsc->id, n_data->action);
return;
}
break;
default:
// We cannot do same for stop because it might be implied by fencing
break;
}
pcmk__rsc_trace(rsc, "Creating notify actions for %s %s",
rsc->id, n_data->action);
// Create notify actions for stop or demote
if ((rsc->priv->orig_role != pcmk_role_stopped)
&& ((task == pcmk__action_stop) || (task == pcmk__action_demote))) {
stop = find_first_action(rsc->priv->actions, NULL, PCMK_ACTION_STOP,
NULL);
for (iter = rsc->priv->active_nodes;
iter != NULL; iter = iter->next) {
pcmk_node_t *current_node = (pcmk_node_t *) iter->data;
/* If a stop is a pseudo-action implied by fencing, don't try to
* notify the node getting fenced.
*/
if ((stop != NULL)
&& pcmk_is_set(stop->flags, pcmk__action_pseudo)
&& (current_node->details->unclean
|| pcmk_is_set(current_node->priv->flags,
pcmk__node_remote_reset))) {
continue;
}
new_notify_action(rsc, current_node, n_data->pre,
n_data->pre_done, n_data);
if ((task == pcmk__action_demote) || (stop == NULL)
|| pcmk_is_set(stop->flags, pcmk__action_optional)) {
new_post_notify_action(rsc, current_node, n_data);
}
}
}
// Create notify actions for start or promote
if ((rsc->priv->next_role != pcmk_role_stopped)
&& ((task == pcmk__action_start) || (task == pcmk__action_promote))) {
start = find_first_action(rsc->priv->actions, NULL,
PCMK_ACTION_START, NULL);
if (start != NULL) {
pcmk_action_t *remote_start = find_remote_start(start);
if ((remote_start != NULL)
&& !pcmk_is_set(remote_start->flags, pcmk__action_runnable)) {
/* Start and promote actions for a clone instance behind
* a Pacemaker Remote connection happen after the
* connection starts. If the connection start is blocked, do
* not schedule notifications for these actions.
*/
return;
}
}
if (rsc->priv->assigned_node == NULL) {
pcmk__sched_err(rsc->priv->scheduler,
"Next role '%s' but %s is not allocated",
pcmk_role_text(rsc->priv->next_role), rsc->id);
return;
}
if ((task != pcmk__action_start) || (start == NULL)
|| pcmk_is_set(start->flags, pcmk__action_optional)) {
new_notify_action(rsc, rsc->priv->assigned_node, n_data->pre,
n_data->pre_done, n_data);
}
new_post_notify_action(rsc, rsc->priv->assigned_node, n_data);
}
}
/*!
* \internal
* \brief Create notification data and actions for one clone action
*
* \param[in,out] rsc Clone resource that notification is for
* \param[in,out] n_data Clone notification data for some action
*/
void
pe__create_action_notifications(pcmk_resource_t *rsc, notify_data_t *n_data)
{
if ((rsc == NULL) || (n_data == NULL)) {
return;
}
collect_resource_data(rsc, true, n_data);
add_notif_keys(rsc, n_data);
create_notify_actions(rsc, n_data);
}
/*!
* \internal
* \brief Free notification data for one action
*
* \param[in,out] n_data Notification data to free
*/
void
pe__free_action_notification_data(notify_data_t *n_data)
{
if (n_data == NULL) {
return;
}
g_list_free_full(n_data->stop, free);
g_list_free_full(n_data->start, free);
g_list_free_full(n_data->demote, free);
g_list_free_full(n_data->promote, free);
g_list_free_full(n_data->promoted, free);
g_list_free_full(n_data->unpromoted, free);
g_list_free_full(n_data->active, free);
g_list_free_full(n_data->inactive, free);
pcmk_free_nvpairs(n_data->keys);
free(n_data);
}
/*!
* \internal
* \brief Order clone "notifications complete" pseudo-action after fencing
*
* If a stop action is implied by fencing, the usual notification pseudo-actions
* will not be sufficient to order things properly, or even create all needed
* notifications if the clone is also stopping on another node, and another
* clone is ordered after it. This function creates new notification
* pseudo-actions relative to the fencing to ensure everything works properly.
*
* \param[in] stop Stop action implied by fencing
* \param[in,out] rsc Clone resource that notification is for
* \param[in,out] stonith_op Fencing action that implies \p stop
*/
void
pe__order_notifs_after_fencing(const pcmk_action_t *stop, pcmk_resource_t *rsc,
pcmk_action_t *stonith_op)
{
notify_data_t *n_data;
crm_info("Ordering notifications for implied %s after fencing", stop->uuid);
n_data = pe__action_notif_pseudo_ops(rsc, PCMK_ACTION_STOP, NULL,
stonith_op);
if (n_data != NULL) {
collect_resource_data(rsc, false, n_data);
add_notify_env(n_data, "notify_stop_resource", rsc->id);
add_notify_env(n_data, "notify_stop_uname", stop->node->priv->name);
create_notify_actions(uber_parent(rsc), n_data);
pe__free_action_notification_data(n_data);
}
}
diff --git a/lib/pengine/pe_output.c b/lib/pengine/pe_output.c
index 3d87bdbca4..04857dfba5 100644
--- a/lib/pengine/pe_output.c
+++ b/lib/pengine/pe_output.c
@@ -1,3472 +1,3473 @@
/*
* Copyright 2019-2024 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU Lesser General Public License
* version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY.
*/
#include <crm_internal.h>
#include <stdint.h>
#include <crm/common/xml_internal.h>
#include <crm/common/output.h>
#include <crm/common/scheduler_internal.h>
#include <crm/cib/util.h>
#include <crm/common/xml.h>
#include <crm/pengine/internal.h>
const char *
pe__resource_description(const pcmk_resource_t *rsc, uint32_t show_opts)
{
const char * desc = NULL;
// User-supplied description
if (pcmk_any_flags_set(show_opts, pcmk_show_rsc_only|pcmk_show_description)) {
desc = crm_element_value(rsc->priv->xml, PCMK_XA_DESCRIPTION);
}
return desc;
}
/* Never display node attributes whose name starts with one of these prefixes */
#define FILTER_STR { PCMK__FAIL_COUNT_PREFIX, PCMK__LAST_FAILURE_PREFIX, \
PCMK__NODE_ATTR_SHUTDOWN, PCMK_NODE_ATTR_TERMINATE, \
PCMK_NODE_ATTR_STANDBY, "#", NULL }
static int
compare_attribute(gconstpointer a, gconstpointer b)
{
int rc;
rc = strcmp((const char *)a, (const char *)b);
return rc;
}
/*!
* \internal
* \brief Determine whether extended information about an attribute should be added.
*
* \param[in] node Node that ran this resource
* \param[in,out] rsc_list List of resources for this node
* \param[in,out] scheduler Scheduler data
* \param[in] attrname Attribute to find
* \param[out] expected_score Expected value for this attribute
*
* \return true if extended information should be printed, false otherwise
* \note Currently, extended information is only supported for ping/pingd
* resources, for which a message will be printed if connectivity is lost
* or degraded.
*/
static bool
add_extra_info(const pcmk_node_t *node, GList *rsc_list,
pcmk_scheduler_t *scheduler, const char *attrname,
int *expected_score)
{
GList *gIter = NULL;
for (gIter = rsc_list; gIter != NULL; gIter = gIter->next) {
pcmk_resource_t *rsc = (pcmk_resource_t *) gIter->data;
const char *type = g_hash_table_lookup(rsc->priv->meta,
PCMK_XA_TYPE);
const char *name = NULL;
GHashTable *params = NULL;
if (rsc->priv->children != NULL) {
if (add_extra_info(node, rsc->priv->children, scheduler,
attrname, expected_score)) {
return true;
}
}
if (!pcmk__strcase_any_of(type, "ping", "pingd", NULL)) {
continue;
}
params = pe_rsc_params(rsc, node, scheduler);
name = g_hash_table_lookup(params, PCMK_XA_NAME);
if (name == NULL) {
name = "pingd";
}
/* To identify the resource with the attribute name. */
if (pcmk__str_eq(name, attrname, pcmk__str_casei)) {
int host_list_num = 0;
const char *hosts = g_hash_table_lookup(params, "host_list");
const char *multiplier = g_hash_table_lookup(params, "multiplier");
int multiplier_i;
if (hosts) {
char **host_list = g_strsplit(hosts, " ", 0);
host_list_num = g_strv_length(host_list);
g_strfreev(host_list);
}
if ((multiplier == NULL)
|| (pcmk__scan_min_int(multiplier, &multiplier_i,
INT_MIN) != pcmk_rc_ok)) {
/* The ocf:pacemaker:ping resource agent defaults multiplier to
* 1. The agent currently does not handle invalid text, but it
* should, and this would be a reasonable choice ...
*/
multiplier_i = 1;
}
*expected_score = host_list_num * multiplier_i;
return true;
}
}
return false;
}
static GList *
filter_attr_list(GList *attr_list, char *name)
{
int i;
const char *filt_str[] = FILTER_STR;
CRM_CHECK(name != NULL, return attr_list);
/* filtering automatic attributes */
for (i = 0; filt_str[i] != NULL; i++) {
if (g_str_has_prefix(name, filt_str[i])) {
return attr_list;
}
}
return g_list_insert_sorted(attr_list, name, compare_attribute);
}
static GList *
get_operation_list(xmlNode *rsc_entry) {
GList *op_list = NULL;
xmlNode *rsc_op = NULL;
for (rsc_op = pcmk__xe_first_child(rsc_entry, PCMK__XE_LRM_RSC_OP, NULL,
NULL);
rsc_op != NULL; rsc_op = pcmk__xe_next(rsc_op, PCMK__XE_LRM_RSC_OP)) {
const char *task = crm_element_value(rsc_op, PCMK_XA_OPERATION);
if (pcmk__str_eq(task, PCMK_ACTION_NOTIFY, pcmk__str_none)) {
continue; // Ignore notify actions
} else {
int exit_status;
pcmk__scan_min_int(crm_element_value(rsc_op, PCMK__XA_RC_CODE),
&exit_status, 0);
if ((exit_status == CRM_EX_NOT_RUNNING)
&& pcmk__str_eq(task, PCMK_ACTION_MONITOR, pcmk__str_none)
&& pcmk__str_eq(crm_element_value(rsc_op, PCMK_META_INTERVAL),
"0", pcmk__str_null_matches)) {
continue; // Ignore probes that found the resource not running
}
}
op_list = g_list_append(op_list, rsc_op);
}
op_list = g_list_sort(op_list, sort_op_by_callid);
return op_list;
}
static void
add_dump_node(gpointer key, gpointer value, gpointer user_data)
{
xmlNodePtr node = user_data;
node = pcmk__xe_create(node, (const char *) key);
pcmk__xe_set_content(node, "%s", (const char *) value);
}
static void
append_dump_text(gpointer key, gpointer value, gpointer user_data)
{
char **dump_text = user_data;
char *new_text = crm_strdup_printf("%s %s=%s",
*dump_text, (char *)key, (char *)value);
free(*dump_text);
*dump_text = new_text;
}
#define XPATH_STACK "//" PCMK_XE_NVPAIR \
"[@" PCMK_XA_NAME "='" \
PCMK_OPT_CLUSTER_INFRASTRUCTURE "']"
static const char *
get_cluster_stack(pcmk_scheduler_t *scheduler)
{
xmlNode *stack = get_xpath_object(XPATH_STACK, scheduler->input, LOG_DEBUG);
if (stack != NULL) {
return crm_element_value(stack, PCMK_XA_VALUE);
}
return PCMK_VALUE_UNKNOWN;
}
static char *
last_changed_string(const char *last_written, const char *user,
const char *client, const char *origin) {
if (last_written != NULL || user != NULL || client != NULL || origin != NULL) {
return crm_strdup_printf("%s%s%s%s%s%s%s",
last_written ? last_written : "",
user ? " by " : "",
user ? user : "",
client ? " via " : "",
client ? client : "",
origin ? " on " : "",
origin ? origin : "");
} else {
return strdup("");
}
}
static char *
op_history_string(xmlNode *xml_op, const char *task, const char *interval_ms_s,
int rc, bool print_timing) {
const char *call = crm_element_value(xml_op, PCMK__XA_CALL_ID);
char *interval_str = NULL;
char *buf = NULL;
if (interval_ms_s && !pcmk__str_eq(interval_ms_s, "0", pcmk__str_casei)) {
char *pair = pcmk__format_nvpair(PCMK_XA_INTERVAL, interval_ms_s, "ms");
interval_str = crm_strdup_printf(" %s", pair);
free(pair);
}
if (print_timing) {
char *last_change_str = NULL;
char *exec_str = NULL;
char *queue_str = NULL;
const char *value = NULL;
time_t epoch = 0;
if ((crm_element_value_epoch(xml_op, PCMK_XA_LAST_RC_CHANGE,
&epoch) == pcmk_ok)
&& (epoch > 0)) {
char *epoch_str = pcmk__epoch2str(&epoch, 0);
last_change_str = crm_strdup_printf(" %s=\"%s\"",
PCMK_XA_LAST_RC_CHANGE,
pcmk__s(epoch_str, ""));
free(epoch_str);
}
value = crm_element_value(xml_op, PCMK_XA_EXEC_TIME);
if (value) {
char *pair = pcmk__format_nvpair(PCMK_XA_EXEC_TIME, value, "ms");
exec_str = crm_strdup_printf(" %s", pair);
free(pair);
}
value = crm_element_value(xml_op, PCMK_XA_QUEUE_TIME);
if (value) {
char *pair = pcmk__format_nvpair(PCMK_XA_QUEUE_TIME, value, "ms");
queue_str = crm_strdup_printf(" %s", pair);
free(pair);
}
buf = crm_strdup_printf("(%s) %s:%s%s%s%s rc=%d (%s)", call, task,
interval_str ? interval_str : "",
last_change_str ? last_change_str : "",
exec_str ? exec_str : "",
queue_str ? queue_str : "",
rc, crm_exit_str(rc));
if (last_change_str) {
free(last_change_str);
}
if (exec_str) {
free(exec_str);
}
if (queue_str) {
free(queue_str);
}
} else {
buf = crm_strdup_printf("(%s) %s%s%s", call, task,
interval_str ? ":" : "",
interval_str ? interval_str : "");
}
if (interval_str) {
free(interval_str);
}
return buf;
}
static char *
resource_history_string(pcmk_resource_t *rsc, const char *rsc_id, bool all,
int failcount, time_t last_failure) {
char *buf = NULL;
if (rsc == NULL) {
buf = crm_strdup_printf("%s: orphan", rsc_id);
} else if (all || failcount || last_failure > 0) {
char *failcount_s = NULL;
char *lastfail_s = NULL;
if (failcount > 0) {
failcount_s = crm_strdup_printf(" %s=%d",
PCMK_XA_FAIL_COUNT, failcount);
} else {
failcount_s = strdup("");
}
if (last_failure > 0) {
buf = pcmk__epoch2str(&last_failure, 0);
lastfail_s = crm_strdup_printf(" %s='%s'",
PCMK_XA_LAST_FAILURE, buf);
free(buf);
}
buf = crm_strdup_printf("%s: " PCMK_META_MIGRATION_THRESHOLD "=%d%s%s",
rsc_id, rsc->priv->ban_after_failures,
failcount_s, pcmk__s(lastfail_s, ""));
free(failcount_s);
free(lastfail_s);
} else {
buf = crm_strdup_printf("%s:", rsc_id);
}
return buf;
}
/*!
* \internal
* \brief Get a node's feature set for status display purposes
*
* \param[in] node Node to check
*
* \return String representation of feature set if the node is fully up (using
* "<3.15.1" for older nodes that don't set the #feature-set attribute),
* otherwise NULL
*/
static const char *
get_node_feature_set(const pcmk_node_t *node)
{
if (node->details->online
&& pcmk_is_set(node->priv->flags, pcmk__node_expected_up)
&& !pcmk__is_pacemaker_remote_node(node)) {
const char *feature_set = g_hash_table_lookup(node->priv->attrs,
CRM_ATTR_FEATURE_SET);
/* The feature set attribute is present since 3.15.1. If it is missing,
* then the node must be running an earlier version.
*/
return pcmk__s(feature_set, "<3.15.1");
}
return NULL;
}
static bool
is_mixed_version(pcmk_scheduler_t *scheduler)
{
const char *feature_set = NULL;
for (GList *gIter = scheduler->nodes; gIter != NULL; gIter = gIter->next) {
pcmk_node_t *node = gIter->data;
const char *node_feature_set = get_node_feature_set(node);
if (node_feature_set != NULL) {
if (feature_set == NULL) {
feature_set = node_feature_set;
} else if (strcmp(feature_set, node_feature_set) != 0) {
return true;
}
}
}
return false;
}
static void
formatted_xml_buf(const pcmk_resource_t *rsc, GString *xml_buf, bool raw)
{
if (raw && (rsc->priv->orig_xml != NULL)) {
pcmk__xml_string(rsc->priv->orig_xml, pcmk__xml_fmt_pretty, xml_buf,
0);
} else {
pcmk__xml_string(rsc->priv->xml, pcmk__xml_fmt_pretty, xml_buf, 0);
}
}
#define XPATH_DC_VERSION "//" PCMK_XE_NVPAIR \
"[@" PCMK_XA_NAME "='" PCMK_OPT_DC_VERSION "']"
PCMK__OUTPUT_ARGS("cluster-summary", "pcmk_scheduler_t *",
"enum pcmk_pacemakerd_state", "uint32_t", "uint32_t")
static int
cluster_summary(pcmk__output_t *out, va_list args) {
pcmk_scheduler_t *scheduler = va_arg(args, pcmk_scheduler_t *);
enum pcmk_pacemakerd_state pcmkd_state =
(enum pcmk_pacemakerd_state) va_arg(args, int);
uint32_t section_opts = va_arg(args, uint32_t);
uint32_t show_opts = va_arg(args, uint32_t);
int rc = pcmk_rc_no_output;
const char *stack_s = get_cluster_stack(scheduler);
if (pcmk_is_set(section_opts, pcmk_section_stack)) {
PCMK__OUTPUT_LIST_HEADER(out, false, rc, "Cluster Summary");
out->message(out, "cluster-stack", stack_s, pcmkd_state);
}
if (pcmk_is_set(section_opts, pcmk_section_dc)) {
xmlNode *dc_version = get_xpath_object(XPATH_DC_VERSION,
scheduler->input, LOG_DEBUG);
const char *dc_version_s = dc_version?
crm_element_value(dc_version, PCMK_XA_VALUE)
: NULL;
const char *quorum = crm_element_value(scheduler->input,
PCMK_XA_HAVE_QUORUM);
char *dc_name = scheduler->dc_node? pe__node_display_name(scheduler->dc_node, pcmk_is_set(show_opts, pcmk_show_node_id)) : NULL;
bool mixed_version = is_mixed_version(scheduler);
PCMK__OUTPUT_LIST_HEADER(out, false, rc, "Cluster Summary");
out->message(out, "cluster-dc", scheduler->dc_node, quorum,
dc_version_s, dc_name, mixed_version);
free(dc_name);
}
if (pcmk_is_set(section_opts, pcmk_section_times)) {
const char *last_written = crm_element_value(scheduler->input,
PCMK_XA_CIB_LAST_WRITTEN);
const char *user = crm_element_value(scheduler->input,
PCMK_XA_UPDATE_USER);
const char *client = crm_element_value(scheduler->input,
PCMK_XA_UPDATE_CLIENT);
const char *origin = crm_element_value(scheduler->input,
PCMK_XA_UPDATE_ORIGIN);
PCMK__OUTPUT_LIST_HEADER(out, false, rc, "Cluster Summary");
out->message(out, "cluster-times", scheduler->priv->local_node_name,
last_written, user, client, origin);
}
if (pcmk_is_set(section_opts, pcmk_section_counts)) {
PCMK__OUTPUT_LIST_HEADER(out, false, rc, "Cluster Summary");
out->message(out, "cluster-counts", g_list_length(scheduler->nodes),
scheduler->priv->ninstances,
scheduler->priv->disabled_resources,
scheduler->priv->blocked_resources);
}
if (pcmk_is_set(section_opts, pcmk_section_options)) {
PCMK__OUTPUT_LIST_HEADER(out, false, rc, "Cluster Summary");
out->message(out, "cluster-options", scheduler);
}
PCMK__OUTPUT_LIST_FOOTER(out, rc);
if (pcmk_is_set(section_opts, pcmk_section_maint_mode)) {
if (out->message(out, "maint-mode", scheduler->flags) == pcmk_rc_ok) {
rc = pcmk_rc_ok;
}
}
return rc;
}
PCMK__OUTPUT_ARGS("cluster-summary", "pcmk_scheduler_t *",
"enum pcmk_pacemakerd_state", "uint32_t", "uint32_t")
static int
cluster_summary_html(pcmk__output_t *out, va_list args) {
pcmk_scheduler_t *scheduler = va_arg(args, pcmk_scheduler_t *);
enum pcmk_pacemakerd_state pcmkd_state =
(enum pcmk_pacemakerd_state) va_arg(args, int);
uint32_t section_opts = va_arg(args, uint32_t);
uint32_t show_opts = va_arg(args, uint32_t);
int rc = pcmk_rc_no_output;
const char *stack_s = get_cluster_stack(scheduler);
if (pcmk_is_set(section_opts, pcmk_section_stack)) {
PCMK__OUTPUT_LIST_HEADER(out, false, rc, "Cluster Summary");
out->message(out, "cluster-stack", stack_s, pcmkd_state);
}
/* Always print DC if none, even if not requested */
if ((scheduler->dc_node == NULL)
|| pcmk_is_set(section_opts, pcmk_section_dc)) {
xmlNode *dc_version = get_xpath_object(XPATH_DC_VERSION,
scheduler->input, LOG_DEBUG);
const char *dc_version_s = dc_version?
crm_element_value(dc_version, PCMK_XA_VALUE)
: NULL;
const char *quorum = crm_element_value(scheduler->input,
PCMK_XA_HAVE_QUORUM);
char *dc_name = scheduler->dc_node? pe__node_display_name(scheduler->dc_node, pcmk_is_set(show_opts, pcmk_show_node_id)) : NULL;
bool mixed_version = is_mixed_version(scheduler);
PCMK__OUTPUT_LIST_HEADER(out, false, rc, "Cluster Summary");
out->message(out, "cluster-dc", scheduler->dc_node, quorum,
dc_version_s, dc_name, mixed_version);
free(dc_name);
}
if (pcmk_is_set(section_opts, pcmk_section_times)) {
const char *last_written = crm_element_value(scheduler->input,
PCMK_XA_CIB_LAST_WRITTEN);
const char *user = crm_element_value(scheduler->input,
PCMK_XA_UPDATE_USER);
const char *client = crm_element_value(scheduler->input,
PCMK_XA_UPDATE_CLIENT);
const char *origin = crm_element_value(scheduler->input,
PCMK_XA_UPDATE_ORIGIN);
PCMK__OUTPUT_LIST_HEADER(out, false, rc, "Cluster Summary");
out->message(out, "cluster-times", scheduler->priv->local_node_name,
last_written, user, client, origin);
}
if (pcmk_is_set(section_opts, pcmk_section_counts)) {
PCMK__OUTPUT_LIST_HEADER(out, false, rc, "Cluster Summary");
out->message(out, "cluster-counts", g_list_length(scheduler->nodes),
scheduler->priv->ninstances,
scheduler->priv->disabled_resources,
scheduler->priv->blocked_resources);
}
if (pcmk_is_set(section_opts, pcmk_section_options)) {
/* Kind of a hack - close the list we may have opened earlier in this
* function so we can put all the options into their own list. We
* only want to do this on HTML output, though.
*/
PCMK__OUTPUT_LIST_FOOTER(out, rc);
out->begin_list(out, NULL, NULL, "Config Options");
out->message(out, "cluster-options", scheduler);
}
PCMK__OUTPUT_LIST_FOOTER(out, rc);
if (pcmk_is_set(section_opts, pcmk_section_maint_mode)) {
if (out->message(out, "maint-mode", scheduler->flags) == pcmk_rc_ok) {
rc = pcmk_rc_ok;
}
}
return rc;
}
char *
pe__node_display_name(pcmk_node_t *node, bool print_detail)
{
char *node_name;
const char *node_host = NULL;
const char *node_id = NULL;
int name_len;
pcmk__assert((node != NULL) && (node->priv->name != NULL));
/* Host is displayed only if this is a guest node and detail is requested */
if (print_detail && pcmk__is_guest_or_bundle_node(node)) {
const pcmk_resource_t *launcher = NULL;
const pcmk_node_t *host_node = NULL;
launcher = node->priv->remote->priv->launcher;
host_node = pcmk__current_node(launcher);
if (host_node && host_node->details) {
node_host = host_node->priv->name;
}
if (node_host == NULL) {
node_host = ""; /* so we at least get "uname@" to indicate guest */
}
}
/* Node ID is displayed if different from uname and detail is requested */
if (print_detail
&& !pcmk__str_eq(node->priv->name, node->priv->id,
pcmk__str_casei)) {
node_id = node->priv->id;
}
/* Determine name length */
name_len = strlen(node->priv->name) + 1;
if (node_host) {
name_len += strlen(node_host) + 1; /* "@node_host" */
}
if (node_id) {
name_len += strlen(node_id) + 3; /* + " (node_id)" */
}
/* Allocate and populate display name */
node_name = pcmk__assert_alloc(name_len, sizeof(char));
strcpy(node_name, node->priv->name);
if (node_host) {
strcat(node_name, "@");
strcat(node_name, node_host);
}
if (node_id) {
strcat(node_name, " (");
strcat(node_name, node_id);
strcat(node_name, ")");
}
return node_name;
}
int
pe__name_and_nvpairs_xml(pcmk__output_t *out, bool is_list, const char *tag_name,
...)
{
xmlNodePtr xml_node = NULL;
va_list pairs;
pcmk__assert(tag_name != NULL);
xml_node = pcmk__output_xml_peek_parent(out);
pcmk__assert(xml_node != NULL);
xml_node = pcmk__xe_create(xml_node, tag_name);
va_start(pairs, tag_name);
pcmk__xe_set_propv(xml_node, pairs);
va_end(pairs);
if (is_list) {
pcmk__output_xml_push_parent(out, xml_node);
}
return pcmk_rc_ok;
}
static const char *
role_desc(enum rsc_role_e role)
{
if (role == pcmk_role_promoted) {
return "in " PCMK_ROLE_PROMOTED " role ";
}
return "";
}
PCMK__OUTPUT_ARGS("ban", "pcmk_node_t *", "pcmk__location_t *", "uint32_t")
static int
ban_html(pcmk__output_t *out, va_list args) {
pcmk_node_t *pe_node = va_arg(args, pcmk_node_t *);
pcmk__location_t *location = va_arg(args, pcmk__location_t *);
uint32_t show_opts = va_arg(args, uint32_t);
char *node_name = pe__node_display_name(pe_node,
pcmk_is_set(show_opts, pcmk_show_node_id));
char *buf = crm_strdup_printf("%s\tprevents %s from running %son %s",
location->id, location->rsc->id,
role_desc(location->role_filter), node_name);
pcmk__output_create_html_node(out, "li", NULL, NULL, buf);
free(node_name);
free(buf);
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("ban", "pcmk_node_t *", "pcmk__location_t *", "uint32_t")
static int
ban_text(pcmk__output_t *out, va_list args) {
pcmk_node_t *pe_node = va_arg(args, pcmk_node_t *);
pcmk__location_t *location = va_arg(args, pcmk__location_t *);
uint32_t show_opts = va_arg(args, uint32_t);
char *node_name = pe__node_display_name(pe_node,
pcmk_is_set(show_opts, pcmk_show_node_id));
out->list_item(out, NULL, "%s\tprevents %s from running %son %s",
location->id, location->rsc->id,
role_desc(location->role_filter), node_name);
free(node_name);
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("ban", "pcmk_node_t *", "pcmk__location_t *", "uint32_t")
static int
ban_xml(pcmk__output_t *out, va_list args) {
pcmk_node_t *pe_node = va_arg(args, pcmk_node_t *);
pcmk__location_t *location = va_arg(args, pcmk__location_t *);
uint32_t show_opts G_GNUC_UNUSED = va_arg(args, uint32_t);
const char *promoted_only = pcmk__btoa(location->role_filter == pcmk_role_promoted);
char *weight_s = pcmk__itoa(pe_node->assign->score);
pcmk__output_create_xml_node(out, PCMK_XE_BAN,
PCMK_XA_ID, location->id,
PCMK_XA_RESOURCE, location->rsc->id,
PCMK_XA_NODE, pe_node->priv->name,
PCMK_XA_WEIGHT, weight_s,
PCMK_XA_PROMOTED_ONLY, promoted_only,
/* This is a deprecated alias for
* promoted_only. Removing it will break
* backward compatibility of the API schema,
* which will require an API schema major
* version bump.
*/
PCMK__XA_PROMOTED_ONLY_LEGACY, promoted_only,
NULL);
free(weight_s);
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("ban-list", "pcmk_scheduler_t *", "const char *", "GList *",
"uint32_t", "bool")
static int
ban_list(pcmk__output_t *out, va_list args) {
pcmk_scheduler_t *scheduler = va_arg(args, pcmk_scheduler_t *);
const char *prefix = va_arg(args, const char *);
GList *only_rsc = va_arg(args, GList *);
uint32_t show_opts = va_arg(args, uint32_t);
bool print_spacer = va_arg(args, int);
GList *gIter, *gIter2;
int rc = pcmk_rc_no_output;
/* Print each ban */
for (gIter = scheduler->priv->location_constraints;
gIter != NULL; gIter = gIter->next) {
pcmk__location_t *location = gIter->data;
const pcmk_resource_t *rsc = location->rsc;
if (prefix != NULL && !g_str_has_prefix(location->id, prefix)) {
continue;
}
if (!pcmk__str_in_list(rsc_printable_id(rsc), only_rsc,
pcmk__str_star_matches)
&& !pcmk__str_in_list(rsc_printable_id(pe__const_top_resource(rsc, false)),
only_rsc, pcmk__str_star_matches)) {
continue;
}
for (gIter2 = location->nodes; gIter2 != NULL; gIter2 = gIter2->next) {
pcmk_node_t *node = (pcmk_node_t *) gIter2->data;
if (node->assign->score < 0) {
PCMK__OUTPUT_LIST_HEADER(out, print_spacer, rc, "Negative Location Constraints");
out->message(out, "ban", node, location, show_opts);
}
}
}
PCMK__OUTPUT_LIST_FOOTER(out, rc);
return rc;
}
PCMK__OUTPUT_ARGS("cluster-counts", "unsigned int", "int", "int", "int")
static int
cluster_counts_html(pcmk__output_t *out, va_list args) {
unsigned int nnodes = va_arg(args, unsigned int);
int nresources = va_arg(args, int);
int ndisabled = va_arg(args, int);
int nblocked = va_arg(args, int);
xmlNodePtr nodes_node = pcmk__output_create_xml_node(out, "li", NULL);
xmlNodePtr resources_node = pcmk__output_create_xml_node(out, "li", NULL);
xmlNode *child = NULL;
child = pcmk__html_create(nodes_node, PCMK__XE_SPAN, NULL, NULL);
pcmk__xe_set_content(child, "%d node%s configured",
nnodes, pcmk__plural_s(nnodes));
if (ndisabled && nblocked) {
child = pcmk__html_create(resources_node, PCMK__XE_SPAN, NULL, NULL);
pcmk__xe_set_content(child, "%d resource instance%s configured (%d ",
nresources, pcmk__plural_s(nresources), ndisabled);
child = pcmk__html_create(resources_node, PCMK__XE_SPAN, NULL,
PCMK__VALUE_BOLD);
pcmk__xe_set_content(child, "DISABLED");
child = pcmk__html_create(resources_node, PCMK__XE_SPAN, NULL, NULL);
pcmk__xe_set_content(child, ", %d ", nblocked);
child = pcmk__html_create(resources_node, PCMK__XE_SPAN, NULL,
PCMK__VALUE_BOLD);
pcmk__xe_set_content(child, "BLOCKED");
child = pcmk__html_create(resources_node, PCMK__XE_SPAN, NULL, NULL);
pcmk__xe_set_content(child, " from further action due to failure)");
} else if (ndisabled && !nblocked) {
child = pcmk__html_create(resources_node, PCMK__XE_SPAN, NULL, NULL);
pcmk__xe_set_content(child, "%d resource instance%s configured (%d ",
nresources, pcmk__plural_s(nresources),
ndisabled);
child = pcmk__html_create(resources_node, PCMK__XE_SPAN, NULL,
PCMK__VALUE_BOLD);
pcmk__xe_set_content(child, "DISABLED");
child = pcmk__html_create(resources_node, PCMK__XE_SPAN, NULL, NULL);
pcmk__xe_set_content(child, ")");
} else if (!ndisabled && nblocked) {
child = pcmk__html_create(resources_node, PCMK__XE_SPAN, NULL, NULL);
pcmk__xe_set_content(child, "%d resource instance%s configured (%d ",
nresources, pcmk__plural_s(nresources),
nblocked);
child = pcmk__html_create(resources_node, PCMK__XE_SPAN, NULL,
PCMK__VALUE_BOLD);
pcmk__xe_set_content(child, "BLOCKED");
child = pcmk__html_create(resources_node, PCMK__XE_SPAN, NULL, NULL);
pcmk__xe_set_content(child, " from further action due to failure)");
} else {
child = pcmk__html_create(resources_node, PCMK__XE_SPAN, NULL, NULL);
pcmk__xe_set_content(child, "%d resource instance%s configured",
nresources, pcmk__plural_s(nresources));
}
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("cluster-counts", "unsigned int", "int", "int", "int")
static int
cluster_counts_text(pcmk__output_t *out, va_list args) {
unsigned int nnodes = va_arg(args, unsigned int);
int nresources = va_arg(args, int);
int ndisabled = va_arg(args, int);
int nblocked = va_arg(args, int);
out->list_item(out, NULL, "%d node%s configured",
nnodes, pcmk__plural_s(nnodes));
if (ndisabled && nblocked) {
out->list_item(out, NULL, "%d resource instance%s configured "
"(%d DISABLED, %d BLOCKED from "
"further action due to failure)",
nresources, pcmk__plural_s(nresources), ndisabled,
nblocked);
} else if (ndisabled && !nblocked) {
out->list_item(out, NULL, "%d resource instance%s configured "
"(%d DISABLED)",
nresources, pcmk__plural_s(nresources), ndisabled);
} else if (!ndisabled && nblocked) {
out->list_item(out, NULL, "%d resource instance%s configured "
"(%d BLOCKED from further action "
"due to failure)",
nresources, pcmk__plural_s(nresources), nblocked);
} else {
out->list_item(out, NULL, "%d resource instance%s configured",
nresources, pcmk__plural_s(nresources));
}
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("cluster-counts", "unsigned int", "int", "int", "int")
static int
cluster_counts_xml(pcmk__output_t *out, va_list args) {
unsigned int nnodes = va_arg(args, unsigned int);
int nresources = va_arg(args, int);
int ndisabled = va_arg(args, int);
int nblocked = va_arg(args, int);
xmlNodePtr nodes_node = NULL;
xmlNodePtr resources_node = NULL;
char *s = NULL;
nodes_node = pcmk__output_create_xml_node(out, PCMK_XE_NODES_CONFIGURED,
NULL);
resources_node = pcmk__output_create_xml_node(out,
PCMK_XE_RESOURCES_CONFIGURED,
NULL);
s = pcmk__itoa(nnodes);
crm_xml_add(nodes_node, PCMK_XA_NUMBER, s);
free(s);
s = pcmk__itoa(nresources);
crm_xml_add(resources_node, PCMK_XA_NUMBER, s);
free(s);
s = pcmk__itoa(ndisabled);
crm_xml_add(resources_node, PCMK_XA_DISABLED, s);
free(s);
s = pcmk__itoa(nblocked);
crm_xml_add(resources_node, PCMK_XA_BLOCKED, s);
free(s);
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("cluster-dc", "pcmk_node_t *", "const char *", "const char *",
"char *", "int")
static int
cluster_dc_html(pcmk__output_t *out, va_list args) {
pcmk_node_t *dc = va_arg(args, pcmk_node_t *);
const char *quorum = va_arg(args, const char *);
const char *dc_version_s = va_arg(args, const char *);
char *dc_name = va_arg(args, char *);
bool mixed_version = va_arg(args, int);
xmlNodePtr node = pcmk__output_create_xml_node(out, "li", NULL);
xmlNode *child = NULL;
child = pcmk__html_create(node, PCMK__XE_SPAN, NULL, PCMK__VALUE_BOLD);
pcmk__xe_set_content(child, "Current DC: ");
if (dc) {
child = pcmk__html_create(node, PCMK__XE_SPAN, NULL, NULL);
pcmk__xe_set_content(child, "%s (version %s) -",
dc_name, pcmk__s(dc_version_s, "unknown"));
if (mixed_version) {
child = pcmk__html_create(node, PCMK__XE_SPAN, NULL,
PCMK__VALUE_WARNING);
pcmk__xe_set_content(child, " MIXED-VERSION");
}
child = pcmk__html_create(node, PCMK__XE_SPAN, NULL, NULL);
pcmk__xe_set_content(child, " partition");
if (crm_is_true(quorum)) {
child = pcmk__html_create(node, PCMK__XE_SPAN, NULL, NULL);
pcmk__xe_set_content(child, " with");
} else {
child = pcmk__html_create(node, PCMK__XE_SPAN, NULL,
PCMK__VALUE_WARNING);
pcmk__xe_set_content(child, " WITHOUT");
}
child = pcmk__html_create(node, PCMK__XE_SPAN, NULL, NULL);
pcmk__xe_set_content(child, " quorum");
} else {
child = pcmk__html_create(node, PCMK__XE_SPAN, NULL,
PCMK__VALUE_WARNING);
pcmk__xe_set_content(child, "NONE");
}
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("cluster-dc", "pcmk_node_t *", "const char *", "const char *",
"char *", "int")
static int
cluster_dc_text(pcmk__output_t *out, va_list args) {
pcmk_node_t *dc = va_arg(args, pcmk_node_t *);
const char *quorum = va_arg(args, const char *);
const char *dc_version_s = va_arg(args, const char *);
char *dc_name = va_arg(args, char *);
bool mixed_version = va_arg(args, int);
if (dc) {
out->list_item(out, "Current DC",
"%s (version %s) - %spartition %s quorum",
dc_name, dc_version_s ? dc_version_s : "unknown",
mixed_version ? "MIXED-VERSION " : "",
crm_is_true(quorum) ? "with" : "WITHOUT");
} else {
out->list_item(out, "Current DC", "NONE");
}
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("cluster-dc", "pcmk_node_t *", "const char *", "const char *",
"char *", "int")
static int
cluster_dc_xml(pcmk__output_t *out, va_list args) {
pcmk_node_t *dc = va_arg(args, pcmk_node_t *);
const char *quorum = va_arg(args, const char *);
const char *dc_version_s = va_arg(args, const char *);
char *dc_name G_GNUC_UNUSED = va_arg(args, char *);
bool mixed_version = va_arg(args, int);
if (dc) {
const char *with_quorum = pcmk__btoa(crm_is_true(quorum));
const char *mixed_version_s = pcmk__btoa(mixed_version);
pcmk__output_create_xml_node(out, PCMK_XE_CURRENT_DC,
PCMK_XA_PRESENT, PCMK_VALUE_TRUE,
PCMK_XA_VERSION, pcmk__s(dc_version_s, ""),
PCMK_XA_NAME, dc->priv->name,
PCMK_XA_ID, dc->priv->id,
PCMK_XA_WITH_QUORUM, with_quorum,
PCMK_XA_MIXED_VERSION, mixed_version_s,
NULL);
} else {
pcmk__output_create_xml_node(out, PCMK_XE_CURRENT_DC,
PCMK_XA_PRESENT, PCMK_VALUE_FALSE,
NULL);
}
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("maint-mode", "uint64_t")
static int
cluster_maint_mode_text(pcmk__output_t *out, va_list args) {
uint64_t flags = va_arg(args, uint64_t);
if (pcmk_is_set(flags, pcmk__sched_in_maintenance)) {
pcmk__formatted_printf(out, "\n *** Resource management is DISABLED ***\n");
pcmk__formatted_printf(out, " The cluster will not attempt to start, stop or recover services\n");
return pcmk_rc_ok;
} else if (pcmk_is_set(flags, pcmk__sched_stop_all)) {
pcmk__formatted_printf(out, "\n *** Resource management is DISABLED ***\n");
pcmk__formatted_printf(out, " The cluster will keep all resources stopped\n");
return pcmk_rc_ok;
} else {
return pcmk_rc_no_output;
}
}
PCMK__OUTPUT_ARGS("cluster-options", "pcmk_scheduler_t *")
static int
cluster_options_html(pcmk__output_t *out, va_list args) {
pcmk_scheduler_t *scheduler = va_arg(args, pcmk_scheduler_t *);
if (pcmk_is_set(scheduler->flags, pcmk__sched_fencing_enabled)) {
out->list_item(out, NULL, "STONITH of failed nodes enabled");
} else {
out->list_item(out, NULL, "STONITH of failed nodes disabled");
}
if (pcmk_is_set(scheduler->flags, pcmk__sched_symmetric_cluster)) {
out->list_item(out, NULL, "Cluster is symmetric");
} else {
out->list_item(out, NULL, "Cluster is asymmetric");
}
switch (scheduler->no_quorum_policy) {
case pcmk_no_quorum_freeze:
out->list_item(out, NULL, "No quorum policy: Freeze resources");
break;
case pcmk_no_quorum_stop:
out->list_item(out, NULL, "No quorum policy: Stop ALL resources");
break;
case pcmk_no_quorum_demote:
out->list_item(out, NULL, "No quorum policy: Demote promotable "
"resources and stop all other resources");
break;
case pcmk_no_quorum_ignore:
out->list_item(out, NULL, "No quorum policy: Ignore");
break;
case pcmk_no_quorum_fence:
out->list_item(out, NULL,
"No quorum policy: Fence nodes in partition");
break;
}
if (pcmk_is_set(scheduler->flags, pcmk__sched_in_maintenance)) {
xmlNodePtr node = pcmk__output_create_xml_node(out, "li", NULL);
xmlNode *child = NULL;
child = pcmk__html_create(node, PCMK__XE_SPAN, NULL, NULL);
pcmk__xe_set_content(child, "Resource management: ");
child = pcmk__html_create(node, PCMK__XE_SPAN, NULL, PCMK__VALUE_BOLD);
pcmk__xe_set_content(child, "DISABLED");
child = pcmk__html_create(node, PCMK__XE_SPAN, NULL, NULL);
pcmk__xe_set_content(child,
" (the cluster will not attempt to start, stop,"
" or recover services)");
} else if (pcmk_is_set(scheduler->flags, pcmk__sched_stop_all)) {
xmlNodePtr node = pcmk__output_create_xml_node(out, "li", NULL);
xmlNode *child = NULL;
child = pcmk__html_create(node, PCMK__XE_SPAN, NULL, NULL);
pcmk__xe_set_content(child, "Resource management: ");
child = pcmk__html_create(node, PCMK__XE_SPAN, NULL, PCMK__VALUE_BOLD);
pcmk__xe_set_content(child, "STOPPED");
child = pcmk__html_create(node, PCMK__XE_SPAN, NULL, NULL);
pcmk__xe_set_content(child,
" (the cluster will keep all resources stopped)");
} else {
out->list_item(out, NULL, "Resource management: enabled");
}
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("cluster-options", "pcmk_scheduler_t *")
static int
cluster_options_log(pcmk__output_t *out, va_list args) {
pcmk_scheduler_t *scheduler = va_arg(args, pcmk_scheduler_t *);
if (pcmk_is_set(scheduler->flags, pcmk__sched_in_maintenance)) {
return out->info(out, "Resource management is DISABLED. The cluster will not attempt to start, stop or recover services.");
} else if (pcmk_is_set(scheduler->flags, pcmk__sched_stop_all)) {
return out->info(out, "Resource management is DISABLED. The cluster has stopped all resources.");
} else {
return pcmk_rc_no_output;
}
}
PCMK__OUTPUT_ARGS("cluster-options", "pcmk_scheduler_t *")
static int
cluster_options_text(pcmk__output_t *out, va_list args) {
pcmk_scheduler_t *scheduler = va_arg(args, pcmk_scheduler_t *);
if (pcmk_is_set(scheduler->flags, pcmk__sched_fencing_enabled)) {
out->list_item(out, NULL, "STONITH of failed nodes enabled");
} else {
out->list_item(out, NULL, "STONITH of failed nodes disabled");
}
if (pcmk_is_set(scheduler->flags, pcmk__sched_symmetric_cluster)) {
out->list_item(out, NULL, "Cluster is symmetric");
} else {
out->list_item(out, NULL, "Cluster is asymmetric");
}
switch (scheduler->no_quorum_policy) {
case pcmk_no_quorum_freeze:
out->list_item(out, NULL, "No quorum policy: Freeze resources");
break;
case pcmk_no_quorum_stop:
out->list_item(out, NULL, "No quorum policy: Stop ALL resources");
break;
case pcmk_no_quorum_demote:
out->list_item(out, NULL, "No quorum policy: Demote promotable "
"resources and stop all other resources");
break;
case pcmk_no_quorum_ignore:
out->list_item(out, NULL, "No quorum policy: Ignore");
break;
case pcmk_no_quorum_fence:
out->list_item(out, NULL,
"No quorum policy: Fence nodes in partition");
break;
}
return pcmk_rc_ok;
}
/*!
* \internal
* \brief Get readable string representation of a no-quorum policy
*
* \param[in] policy No-quorum policy
*
* \return String representation of \p policy
*/
static const char *
no_quorum_policy_text(enum pe_quorum_policy policy)
{
switch (policy) {
case pcmk_no_quorum_freeze:
return PCMK_VALUE_FREEZE;
case pcmk_no_quorum_stop:
return PCMK_VALUE_STOP;
case pcmk_no_quorum_demote:
return PCMK_VALUE_DEMOTE;
case pcmk_no_quorum_ignore:
return PCMK_VALUE_IGNORE;
case pcmk_no_quorum_fence:
return PCMK_VALUE_FENCE;
default:
return PCMK_VALUE_UNKNOWN;
}
}
PCMK__OUTPUT_ARGS("cluster-options", "pcmk_scheduler_t *")
static int
cluster_options_xml(pcmk__output_t *out, va_list args) {
pcmk_scheduler_t *scheduler = va_arg(args, pcmk_scheduler_t *);
const char *stonith_enabled = pcmk__flag_text(scheduler->flags,
pcmk__sched_fencing_enabled);
const char *symmetric_cluster =
pcmk__flag_text(scheduler->flags, pcmk__sched_symmetric_cluster);
const char *no_quorum_policy =
no_quorum_policy_text(scheduler->no_quorum_policy);
const char *maintenance_mode = pcmk__flag_text(scheduler->flags,
pcmk__sched_in_maintenance);
const char *stop_all_resources = pcmk__flag_text(scheduler->flags,
pcmk__sched_stop_all);
char *stonith_timeout_ms_s =
crm_strdup_printf("%u", scheduler->priv->fence_timeout_ms);
char *priority_fencing_delay_ms_s =
crm_strdup_printf("%u", scheduler->priv->priority_fencing_ms);
pcmk__output_create_xml_node(out, PCMK_XE_CLUSTER_OPTIONS,
PCMK_XA_STONITH_ENABLED, stonith_enabled,
PCMK_XA_SYMMETRIC_CLUSTER, symmetric_cluster,
PCMK_XA_NO_QUORUM_POLICY, no_quorum_policy,
PCMK_XA_MAINTENANCE_MODE, maintenance_mode,
PCMK_XA_STOP_ALL_RESOURCES, stop_all_resources,
PCMK_XA_STONITH_TIMEOUT_MS,
stonith_timeout_ms_s,
PCMK_XA_PRIORITY_FENCING_DELAY_MS,
priority_fencing_delay_ms_s,
NULL);
free(stonith_timeout_ms_s);
free(priority_fencing_delay_ms_s);
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("cluster-stack", "const char *", "enum pcmk_pacemakerd_state")
static int
cluster_stack_html(pcmk__output_t *out, va_list args) {
const char *stack_s = va_arg(args, const char *);
enum pcmk_pacemakerd_state pcmkd_state =
(enum pcmk_pacemakerd_state) va_arg(args, int);
xmlNodePtr node = pcmk__output_create_xml_node(out, "li", NULL);
xmlNode *child = NULL;
child = pcmk__html_create(node, PCMK__XE_SPAN, NULL, PCMK__VALUE_BOLD);
pcmk__xe_set_content(child, "Stack: ");
child = pcmk__html_create(node, PCMK__XE_SPAN, NULL, NULL);
pcmk__xe_set_content(child, "%s", stack_s);
if (pcmkd_state != pcmk_pacemakerd_state_invalid) {
child = pcmk__html_create(node, PCMK__XE_SPAN, NULL, NULL);
pcmk__xe_set_content(child, " (");
child = pcmk__html_create(node, PCMK__XE_SPAN, NULL, NULL);
pcmk__xe_set_content(child, "%s",
pcmk__pcmkd_state_enum2friendly(pcmkd_state));
child = pcmk__html_create(node, PCMK__XE_SPAN, NULL, NULL);
pcmk__xe_set_content(child, ")");
}
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("cluster-stack", "const char *", "enum pcmk_pacemakerd_state")
static int
cluster_stack_text(pcmk__output_t *out, va_list args) {
const char *stack_s = va_arg(args, const char *);
enum pcmk_pacemakerd_state pcmkd_state =
(enum pcmk_pacemakerd_state) va_arg(args, int);
if (pcmkd_state != pcmk_pacemakerd_state_invalid) {
out->list_item(out, "Stack", "%s (%s)",
stack_s, pcmk__pcmkd_state_enum2friendly(pcmkd_state));
} else {
out->list_item(out, "Stack", "%s", stack_s);
}
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("cluster-stack", "const char *", "enum pcmk_pacemakerd_state")
static int
cluster_stack_xml(pcmk__output_t *out, va_list args) {
const char *stack_s = va_arg(args, const char *);
enum pcmk_pacemakerd_state pcmkd_state =
(enum pcmk_pacemakerd_state) va_arg(args, int);
const char *state_s = NULL;
if (pcmkd_state != pcmk_pacemakerd_state_invalid) {
state_s = pcmk_pacemakerd_api_daemon_state_enum2text(pcmkd_state);
}
pcmk__output_create_xml_node(out, PCMK_XE_STACK,
PCMK_XA_TYPE, stack_s,
PCMK_XA_PACEMAKERD_STATE, state_s,
NULL);
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("cluster-times", "const char *", "const char *",
"const char *", "const char *", "const char *")
static int
cluster_times_html(pcmk__output_t *out, va_list args) {
const char *our_nodename = va_arg(args, const char *);
const char *last_written = va_arg(args, const char *);
const char *user = va_arg(args, const char *);
const char *client = va_arg(args, const char *);
const char *origin = va_arg(args, const char *);
xmlNodePtr updated_node = pcmk__output_create_xml_node(out, "li", NULL);
xmlNodePtr changed_node = pcmk__output_create_xml_node(out, "li", NULL);
xmlNode *child = NULL;
char *time_s = NULL;
child = pcmk__html_create(updated_node, PCMK__XE_SPAN, NULL,
PCMK__VALUE_BOLD);
pcmk__xe_set_content(child, "Last updated: ");
child = pcmk__html_create(updated_node, PCMK__XE_SPAN, NULL, NULL);
time_s = pcmk__epoch2str(NULL, 0);
pcmk__xe_set_content(child, "%s", time_s);
free(time_s);
if (our_nodename != NULL) {
child = pcmk__html_create(updated_node, PCMK__XE_SPAN, NULL, NULL);
pcmk__xe_set_content(child, " on ");
child = pcmk__html_create(updated_node, PCMK__XE_SPAN, NULL, NULL);
pcmk__xe_set_content(child, "%s", our_nodename);
}
child = pcmk__html_create(changed_node, PCMK__XE_SPAN, NULL,
PCMK__VALUE_BOLD);
pcmk__xe_set_content(child, "Last change: ");
child = pcmk__html_create(changed_node, PCMK__XE_SPAN, NULL, NULL);
time_s = last_changed_string(last_written, user, client, origin);
pcmk__xe_set_content(child, "%s", time_s);
free(time_s);
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("cluster-times", "const char *", "const char *",
"const char *", "const char *", "const char *")
static int
cluster_times_xml(pcmk__output_t *out, va_list args) {
const char *our_nodename = va_arg(args, const char *);
const char *last_written = va_arg(args, const char *);
const char *user = va_arg(args, const char *);
const char *client = va_arg(args, const char *);
const char *origin = va_arg(args, const char *);
char *time_s = pcmk__epoch2str(NULL, 0);
pcmk__output_create_xml_node(out, PCMK_XE_LAST_UPDATE,
PCMK_XA_TIME, time_s,
PCMK_XA_ORIGIN, our_nodename,
NULL);
pcmk__output_create_xml_node(out, PCMK_XE_LAST_CHANGE,
PCMK_XA_TIME, pcmk__s(last_written, ""),
PCMK_XA_USER, pcmk__s(user, ""),
PCMK_XA_CLIENT, pcmk__s(client, ""),
PCMK_XA_ORIGIN, pcmk__s(origin, ""),
NULL);
free(time_s);
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("cluster-times", "const char *", "const char *",
"const char *", "const char *", "const char *")
static int
cluster_times_text(pcmk__output_t *out, va_list args) {
const char *our_nodename = va_arg(args, const char *);
const char *last_written = va_arg(args, const char *);
const char *user = va_arg(args, const char *);
const char *client = va_arg(args, const char *);
const char *origin = va_arg(args, const char *);
char *time_s = pcmk__epoch2str(NULL, 0);
out->list_item(out, "Last updated", "%s%s%s",
time_s, (our_nodename != NULL)? " on " : "",
pcmk__s(our_nodename, ""));
free(time_s);
time_s = last_changed_string(last_written, user, client, origin);
out->list_item(out, "Last change", " %s", time_s);
free(time_s);
return pcmk_rc_ok;
}
/*!
* \internal
* \brief Display a failed action in less-technical natural language
*
* \param[in,out] out Output object to use for display
* \param[in] xml_op XML containing failed action
* \param[in] op_key Operation key of failed action
* \param[in] node_name Where failed action occurred
* \param[in] rc OCF exit code of failed action
* \param[in] status Execution status of failed action
* \param[in] exit_reason Exit reason given for failed action
* \param[in] exec_time String containing execution time in milliseconds
*/
static void
failed_action_friendly(pcmk__output_t *out, const xmlNode *xml_op,
const char *op_key, const char *node_name, int rc,
int status, const char *exit_reason,
const char *exec_time)
{
char *rsc_id = NULL;
char *task = NULL;
guint interval_ms = 0;
time_t last_change_epoch = 0;
GString *str = NULL;
if (pcmk__str_empty(op_key)
|| !parse_op_key(op_key, &rsc_id, &task, &interval_ms)) {
pcmk__str_update(&rsc_id, "unknown resource");
pcmk__str_update(&task, "unknown action");
interval_ms = 0;
}
pcmk__assert((rsc_id != NULL) && (task != NULL));
str = g_string_sized_new(256); // Should be sufficient for most messages
pcmk__g_strcat(str, rsc_id, " ", NULL);
if (interval_ms != 0) {
pcmk__g_strcat(str, pcmk__readable_interval(interval_ms), "-interval ",
NULL);
}
pcmk__g_strcat(str, pcmk__readable_action(task, interval_ms), " on ",
node_name, NULL);
if (status == PCMK_EXEC_DONE) {
pcmk__g_strcat(str, " returned '", crm_exit_str(rc), "'", NULL);
if (!pcmk__str_empty(exit_reason)) {
pcmk__g_strcat(str, " (", exit_reason, ")", NULL);
}
} else {
pcmk__g_strcat(str, " could not be executed (",
pcmk_exec_status_str(status), NULL);
if (!pcmk__str_empty(exit_reason)) {
pcmk__g_strcat(str, ": ", exit_reason, NULL);
}
g_string_append_c(str, ')');
}
if (crm_element_value_epoch(xml_op, PCMK_XA_LAST_RC_CHANGE,
&last_change_epoch) == pcmk_ok) {
char *s = pcmk__epoch2str(&last_change_epoch, 0);
pcmk__g_strcat(str, " at ", s, NULL);
free(s);
}
if (!pcmk__str_empty(exec_time)) {
int exec_time_ms = 0;
if ((pcmk__scan_min_int(exec_time, &exec_time_ms, 0) == pcmk_rc_ok)
&& (exec_time_ms > 0)) {
pcmk__g_strcat(str, " after ",
pcmk__readable_interval(exec_time_ms), NULL);
}
}
out->list_item(out, NULL, "%s", str->str);
g_string_free(str, TRUE);
free(rsc_id);
free(task);
}
/*!
* \internal
* \brief Display a failed action with technical details
*
* \param[in,out] out Output object to use for display
* \param[in] xml_op XML containing failed action
* \param[in] op_key Operation key of failed action
* \param[in] node_name Where failed action occurred
* \param[in] rc OCF exit code of failed action
* \param[in] status Execution status of failed action
* \param[in] exit_reason Exit reason given for failed action
* \param[in] exec_time String containing execution time in milliseconds
*/
static void
failed_action_technical(pcmk__output_t *out, const xmlNode *xml_op,
const char *op_key, const char *node_name, int rc,
int status, const char *exit_reason,
const char *exec_time)
{
const char *call_id = crm_element_value(xml_op, PCMK__XA_CALL_ID);
const char *queue_time = crm_element_value(xml_op, PCMK_XA_QUEUE_TIME);
const char *exit_status = crm_exit_str(rc);
const char *lrm_status = pcmk_exec_status_str(status);
time_t last_change_epoch = 0;
GString *str = NULL;
if (pcmk__str_empty(op_key)) {
op_key = "unknown operation";
}
if (pcmk__str_empty(exit_status)) {
exit_status = "unknown exit status";
}
if (pcmk__str_empty(call_id)) {
call_id = "unknown";
}
str = g_string_sized_new(256);
g_string_append_printf(str, "%s on %s '%s' (%d): call=%s, status='%s'",
op_key, node_name, exit_status, rc, call_id,
lrm_status);
if (!pcmk__str_empty(exit_reason)) {
pcmk__g_strcat(str, ", exitreason='", exit_reason, "'", NULL);
}
if (crm_element_value_epoch(xml_op, PCMK_XA_LAST_RC_CHANGE,
&last_change_epoch) == pcmk_ok) {
char *last_change_str = pcmk__epoch2str(&last_change_epoch, 0);
pcmk__g_strcat(str,
", " PCMK_XA_LAST_RC_CHANGE "="
"'", last_change_str, "'", NULL);
free(last_change_str);
}
if (!pcmk__str_empty(queue_time)) {
pcmk__g_strcat(str, ", queued=", queue_time, "ms", NULL);
}
if (!pcmk__str_empty(exec_time)) {
pcmk__g_strcat(str, ", exec=", exec_time, "ms", NULL);
}
out->list_item(out, NULL, "%s", str->str);
g_string_free(str, TRUE);
}
PCMK__OUTPUT_ARGS("failed-action", "xmlNode *", "uint32_t")
static int
failed_action_default(pcmk__output_t *out, va_list args)
{
xmlNodePtr xml_op = va_arg(args, xmlNodePtr);
uint32_t show_opts = va_arg(args, uint32_t);
const char *op_key = pcmk__xe_history_key(xml_op);
const char *node_name = crm_element_value(xml_op, PCMK_XA_UNAME);
const char *exit_reason = crm_element_value(xml_op, PCMK_XA_EXIT_REASON);
const char *exec_time = crm_element_value(xml_op, PCMK_XA_EXEC_TIME);
int rc;
int status;
pcmk__scan_min_int(crm_element_value(xml_op, PCMK__XA_RC_CODE), &rc, 0);
pcmk__scan_min_int(crm_element_value(xml_op, PCMK__XA_OP_STATUS), &status,
0);
if (pcmk__str_empty(node_name)) {
node_name = "unknown node";
}
if (pcmk_is_set(show_opts, pcmk_show_failed_detail)) {
failed_action_technical(out, xml_op, op_key, node_name, rc, status,
exit_reason, exec_time);
} else {
failed_action_friendly(out, xml_op, op_key, node_name, rc, status,
exit_reason, exec_time);
}
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("failed-action", "xmlNode *", "uint32_t")
static int
failed_action_xml(pcmk__output_t *out, va_list args) {
xmlNodePtr xml_op = va_arg(args, xmlNodePtr);
uint32_t show_opts G_GNUC_UNUSED = va_arg(args, uint32_t);
const char *op_key = pcmk__xe_history_key(xml_op);
const char *op_key_name = PCMK_XA_OP_KEY;
int rc;
int status;
const char *uname = crm_element_value(xml_op, PCMK_XA_UNAME);
const char *call_id = crm_element_value(xml_op, PCMK__XA_CALL_ID);
const char *exitstatus = NULL;
const char *exit_reason = pcmk__s(crm_element_value(xml_op,
PCMK_XA_EXIT_REASON),
"none");
const char *status_s = NULL;
time_t epoch = 0;
gchar *exit_reason_esc = NULL;
char *rc_s = NULL;
xmlNodePtr node = NULL;
if (pcmk__xml_needs_escape(exit_reason, pcmk__xml_escape_attr)) {
exit_reason_esc = pcmk__xml_escape(exit_reason, pcmk__xml_escape_attr);
exit_reason = exit_reason_esc;
}
pcmk__scan_min_int(crm_element_value(xml_op, PCMK__XA_RC_CODE), &rc, 0);
pcmk__scan_min_int(crm_element_value(xml_op, PCMK__XA_OP_STATUS), &status,
0);
if (crm_element_value(xml_op, PCMK__XA_OPERATION_KEY) == NULL) {
op_key_name = PCMK_XA_ID;
}
exitstatus = crm_exit_str(rc);
rc_s = pcmk__itoa(rc);
status_s = pcmk_exec_status_str(status);
node = pcmk__output_create_xml_node(out, PCMK_XE_FAILURE,
op_key_name, op_key,
PCMK_XA_NODE, uname,
PCMK_XA_EXITSTATUS, exitstatus,
PCMK_XA_EXITREASON, exit_reason,
PCMK_XA_EXITCODE, rc_s,
PCMK_XA_CALL, call_id,
PCMK_XA_STATUS, status_s,
NULL);
free(rc_s);
if ((crm_element_value_epoch(xml_op, PCMK_XA_LAST_RC_CHANGE,
&epoch) == pcmk_ok) && (epoch > 0)) {
const char *queue_time = crm_element_value(xml_op, PCMK_XA_QUEUE_TIME);
const char *exec = crm_element_value(xml_op, PCMK_XA_EXEC_TIME);
const char *task = crm_element_value(xml_op, PCMK_XA_OPERATION);
guint interval_ms = 0;
char *interval_ms_s = NULL;
char *rc_change = pcmk__epoch2str(&epoch,
crm_time_log_date
|crm_time_log_timeofday
|crm_time_log_with_timezone);
crm_element_value_ms(xml_op, PCMK_META_INTERVAL, &interval_ms);
interval_ms_s = crm_strdup_printf("%u", interval_ms);
pcmk__xe_set_props(node,
PCMK_XA_LAST_RC_CHANGE, rc_change,
PCMK_XA_QUEUED, queue_time,
PCMK_XA_EXEC, exec,
PCMK_XA_INTERVAL, interval_ms_s,
PCMK_XA_TASK, task,
NULL);
free(interval_ms_s);
free(rc_change);
}
g_free(exit_reason_esc);
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("failed-action-list", "pcmk_scheduler_t *", "GList *",
"GList *", "uint32_t", "bool")
static int
failed_action_list(pcmk__output_t *out, va_list args) {
pcmk_scheduler_t *scheduler = va_arg(args, pcmk_scheduler_t *);
GList *only_node = va_arg(args, GList *);
GList *only_rsc = va_arg(args, GList *);
uint32_t show_opts = va_arg(args, uint32_t);
bool print_spacer = va_arg(args, int);
xmlNode *xml_op = NULL;
int rc = pcmk_rc_no_output;
if (xmlChildElementCount(scheduler->priv->failed) == 0) {
return rc;
}
for (xml_op = pcmk__xe_first_child(scheduler->priv->failed, NULL, NULL,
NULL);
xml_op != NULL; xml_op = pcmk__xe_next(xml_op, NULL)) {
char *rsc = NULL;
if (!pcmk__str_in_list(crm_element_value(xml_op, PCMK_XA_UNAME),
only_node,
pcmk__str_star_matches|pcmk__str_casei)) {
continue;
}
if (pcmk_xe_mask_probe_failure(xml_op)) {
continue;
}
if (!parse_op_key(pcmk__xe_history_key(xml_op), &rsc, NULL, NULL)) {
continue;
}
if (!pcmk__str_in_list(rsc, only_rsc, pcmk__str_star_matches)) {
free(rsc);
continue;
}
free(rsc);
PCMK__OUTPUT_LIST_HEADER(out, print_spacer, rc, "Failed Resource Actions");
out->message(out, "failed-action", xml_op, show_opts);
}
PCMK__OUTPUT_LIST_FOOTER(out, rc);
return rc;
}
static void
status_node(pcmk_node_t *node, xmlNodePtr parent, uint32_t show_opts)
{
int health = pe__node_health(node);
xmlNode *child = NULL;
// Cluster membership
if (node->details->online) {
child = pcmk__html_create(parent, PCMK__XE_SPAN, NULL,
PCMK_VALUE_ONLINE);
pcmk__xe_set_content(child, " online");
} else {
child = pcmk__html_create(parent, PCMK__XE_SPAN, NULL,
PCMK_VALUE_OFFLINE);
pcmk__xe_set_content(child, " OFFLINE");
}
// Standby mode
if (pcmk_is_set(node->priv->flags, pcmk__node_fail_standby)) {
child = pcmk__html_create(parent, PCMK__XE_SPAN, NULL,
PCMK_VALUE_STANDBY);
if (node->details->running_rsc == NULL) {
pcmk__xe_set_content(child,
" (in standby due to " PCMK_META_ON_FAIL ")");
} else {
pcmk__xe_set_content(child,
" (in standby due to " PCMK_META_ON_FAIL ","
" with active resources)");
}
} else if (pcmk_is_set(node->priv->flags, pcmk__node_standby)) {
child = pcmk__html_create(parent, PCMK__XE_SPAN, NULL,
PCMK_VALUE_STANDBY);
if (node->details->running_rsc == NULL) {
pcmk__xe_set_content(child, " (in standby)");
} else {
pcmk__xe_set_content(child, " (in standby, with active resources)");
}
}
// Maintenance mode
if (node->details->maintenance) {
child = pcmk__html_create(parent, PCMK__XE_SPAN, NULL,
PCMK__VALUE_MAINT);
pcmk__xe_set_content(child, " (in maintenance mode)");
}
// Node health
if (health < 0) {
child = pcmk__html_create(parent, PCMK__XE_SPAN, NULL,
PCMK__VALUE_HEALTH_RED);
pcmk__xe_set_content(child, " (health is RED)");
} else if (health == 0) {
child = pcmk__html_create(parent, PCMK__XE_SPAN, NULL,
PCMK__VALUE_HEALTH_YELLOW);
pcmk__xe_set_content(child, " (health is YELLOW)");
}
// Feature set
if (pcmk_is_set(show_opts, pcmk_show_feature_set)) {
const char *feature_set = get_node_feature_set(node);
if (feature_set != NULL) {
child = pcmk__html_create(parent, PCMK__XE_SPAN, NULL, NULL);
pcmk__xe_set_content(child, ", feature set %s", feature_set);
}
}
}
PCMK__OUTPUT_ARGS("node", "pcmk_node_t *", "uint32_t", "bool",
"GList *", "GList *")
static int
node_html(pcmk__output_t *out, va_list args) {
pcmk_node_t *node = va_arg(args, pcmk_node_t *);
uint32_t show_opts = va_arg(args, uint32_t);
bool full = va_arg(args, int);
GList *only_node = va_arg(args, GList *);
GList *only_rsc = va_arg(args, GList *);
char *node_name = pe__node_display_name(node, pcmk_is_set(show_opts, pcmk_show_node_id));
if (full) {
xmlNode *item_node = NULL;
xmlNode *child = NULL;
if (pcmk_all_flags_set(show_opts, pcmk_show_brief | pcmk_show_rscs_by_node)) {
GList *rscs = pe__filter_rsc_list(node->details->running_rsc, only_rsc);
out->begin_list(out, NULL, NULL, "%s:", node_name);
item_node = pcmk__output_xml_create_parent(out, "li", NULL);
child = pcmk__html_create(item_node, PCMK__XE_SPAN, NULL, NULL);
pcmk__xe_set_content(child, "Status:");
status_node(node, item_node, show_opts);
if (rscs != NULL) {
uint32_t new_show_opts = (show_opts | pcmk_show_rsc_only) & ~pcmk_show_inactive_rscs;
out->begin_list(out, NULL, NULL, "Resources");
pe__rscs_brief_output(out, rscs, new_show_opts);
out->end_list(out);
}
pcmk__output_xml_pop_parent(out);
out->end_list(out);
} else if (pcmk_is_set(show_opts, pcmk_show_rscs_by_node)) {
GList *lpc2 = NULL;
int rc = pcmk_rc_no_output;
out->begin_list(out, NULL, NULL, "%s:", node_name);
item_node = pcmk__output_xml_create_parent(out, "li", NULL);
child = pcmk__html_create(item_node, PCMK__XE_SPAN, NULL, NULL);
pcmk__xe_set_content(child, "Status:");
status_node(node, item_node, show_opts);
for (lpc2 = node->details->running_rsc; lpc2 != NULL; lpc2 = lpc2->next) {
pcmk_resource_t *rsc = (pcmk_resource_t *) lpc2->data;
PCMK__OUTPUT_LIST_HEADER(out, false, rc, "Resources");
show_opts |= pcmk_show_rsc_only;
out->message(out, (const char *) rsc->priv->xml->name,
show_opts, rsc, only_node, only_rsc);
}
PCMK__OUTPUT_LIST_FOOTER(out, rc);
pcmk__output_xml_pop_parent(out);
out->end_list(out);
} else {
item_node = pcmk__output_create_xml_node(out, "li", NULL);
child = pcmk__html_create(item_node, PCMK__XE_SPAN, NULL,
PCMK__VALUE_BOLD);
pcmk__xe_set_content(child, "%s:", node_name);
status_node(node, item_node, show_opts);
}
} else {
out->begin_list(out, NULL, NULL, "%s:", node_name);
}
free(node_name);
return pcmk_rc_ok;
}
/*!
* \internal
* \brief Get a human-friendly textual description of a node's status
*
* \param[in] node Node to check
*
* \return String representation of node's status
*/
static const char *
node_text_status(const pcmk_node_t *node)
{
if (node->details->unclean) {
if (node->details->online) {
return "UNCLEAN (online)";
} else if (node->details->pending) {
return "UNCLEAN (pending)";
} else {
return "UNCLEAN (offline)";
}
} else if (node->details->pending) {
return "pending";
} else if (pcmk_is_set(node->priv->flags, pcmk__node_fail_standby)
&& node->details->online) {
return "standby (" PCMK_META_ON_FAIL ")";
} else if (pcmk_is_set(node->priv->flags, pcmk__node_standby)) {
if (!node->details->online) {
return "OFFLINE (standby)";
} else if (node->details->running_rsc == NULL) {
return "standby";
} else {
return "standby (with active resources)";
}
} else if (node->details->maintenance) {
if (node->details->online) {
return "maintenance";
} else {
return "OFFLINE (maintenance)";
}
} else if (node->details->online) {
return "online";
}
return "OFFLINE";
}
PCMK__OUTPUT_ARGS("node", "pcmk_node_t *", "uint32_t", "bool", "GList *",
"GList *")
static int
node_text(pcmk__output_t *out, va_list args) {
pcmk_node_t *node = va_arg(args, pcmk_node_t *);
uint32_t show_opts = va_arg(args, uint32_t);
bool full = va_arg(args, int);
GList *only_node = va_arg(args, GList *);
GList *only_rsc = va_arg(args, GList *);
if (full) {
char *node_name = pe__node_display_name(node, pcmk_is_set(show_opts, pcmk_show_node_id));
GString *str = g_string_sized_new(64);
int health = pe__node_health(node);
// Create a summary line with node type, name, and status
if (pcmk__is_guest_or_bundle_node(node)) {
g_string_append(str, "GuestNode");
} else if (pcmk__is_remote_node(node)) {
g_string_append(str, "RemoteNode");
} else {
g_string_append(str, "Node");
}
pcmk__g_strcat(str, " ", node_name, ": ", node_text_status(node), NULL);
if (health < 0) {
g_string_append(str, " (health is RED)");
} else if (health == 0) {
g_string_append(str, " (health is YELLOW)");
}
if (pcmk_is_set(show_opts, pcmk_show_feature_set)) {
const char *feature_set = get_node_feature_set(node);
if (feature_set != NULL) {
pcmk__g_strcat(str, ", feature set ", feature_set, NULL);
}
}
/* If we're grouping by node, print its resources */
if (pcmk_is_set(show_opts, pcmk_show_rscs_by_node)) {
if (pcmk_is_set(show_opts, pcmk_show_brief)) {
GList *rscs = pe__filter_rsc_list(node->details->running_rsc, only_rsc);
if (rscs != NULL) {
uint32_t new_show_opts = (show_opts | pcmk_show_rsc_only) & ~pcmk_show_inactive_rscs;
out->begin_list(out, NULL, NULL, "%s", str->str);
out->begin_list(out, NULL, NULL, "Resources");
pe__rscs_brief_output(out, rscs, new_show_opts);
out->end_list(out);
out->end_list(out);
g_list_free(rscs);
}
} else {
GList *gIter2 = NULL;
out->begin_list(out, NULL, NULL, "%s", str->str);
out->begin_list(out, NULL, NULL, "Resources");
for (gIter2 = node->details->running_rsc; gIter2 != NULL; gIter2 = gIter2->next) {
pcmk_resource_t *rsc = (pcmk_resource_t *) gIter2->data;
show_opts |= pcmk_show_rsc_only;
out->message(out, (const char *) rsc->priv->xml->name,
show_opts, rsc, only_node, only_rsc);
}
out->end_list(out);
out->end_list(out);
}
} else {
out->list_item(out, NULL, "%s", str->str);
}
g_string_free(str, TRUE);
free(node_name);
} else {
char *node_name = pe__node_display_name(node, pcmk_is_set(show_opts, pcmk_show_node_id));
out->begin_list(out, NULL, NULL, "Node: %s", node_name);
free(node_name);
}
return pcmk_rc_ok;
}
/*!
* \internal
* \brief Convert an integer health value to a string representation
*
* \param[in] health Integer health value
*
* \retval \c PCMK_VALUE_RED if \p health is less than 0
* \retval \c PCMK_VALUE_YELLOW if \p health is equal to 0
* \retval \c PCMK_VALUE_GREEN if \p health is greater than 0
*/
static const char *
health_text(int health)
{
if (health < 0) {
return PCMK_VALUE_RED;
} else if (health == 0) {
return PCMK_VALUE_YELLOW;
} else {
return PCMK_VALUE_GREEN;
}
}
/*!
* \internal
* \brief Convert a node variant to a string representation
*
* \param[in] variant Node variant
*
* \retval \c PCMK_VALUE_MEMBER if \p node_type is \c pcmk__node_variant_cluster
* \retval \c PCMK_VALUE_REMOTE if \p node_type is \c pcmk__node_variant_remote
* \retval \c PCMK_VALUE_UNKNOWN otherwise
*/
static const char *
node_variant_text(enum pcmk__node_variant variant)
{
switch (variant) {
case pcmk__node_variant_cluster:
return PCMK_VALUE_MEMBER;
case pcmk__node_variant_remote:
return PCMK_VALUE_REMOTE;
default:
return PCMK_VALUE_UNKNOWN;
}
}
PCMK__OUTPUT_ARGS("node", "pcmk_node_t *", "uint32_t", "bool", "GList *",
"GList *")
static int
node_xml(pcmk__output_t *out, va_list args) {
pcmk_node_t *node = va_arg(args, pcmk_node_t *);
uint32_t show_opts G_GNUC_UNUSED = va_arg(args, uint32_t);
bool full = va_arg(args, int);
GList *only_node = va_arg(args, GList *);
GList *only_rsc = va_arg(args, GList *);
if (full) {
const char *online = pcmk__btoa(node->details->online);
const char *standby = pcmk__flag_text(node->priv->flags,
pcmk__node_standby);
const char *standby_onfail = pcmk__flag_text(node->priv->flags,
pcmk__node_fail_standby);
const char *maintenance = pcmk__btoa(node->details->maintenance);
const char *pending = pcmk__btoa(node->details->pending);
const char *unclean = pcmk__btoa(node->details->unclean);
const char *health = health_text(pe__node_health(node));
const char *feature_set = get_node_feature_set(node);
const char *shutdown = pcmk__btoa(node->details->shutdown);
const char *expected_up = pcmk__flag_text(node->priv->flags,
pcmk__node_expected_up);
const bool is_dc = pcmk__same_node(node,
node->priv->scheduler->dc_node);
int length = g_list_length(node->details->running_rsc);
char *resources_running = pcmk__itoa(length);
const char *node_type = node_variant_text(node->priv->variant);
int rc = pcmk_rc_ok;
rc = pe__name_and_nvpairs_xml(out, true, PCMK_XE_NODE,
PCMK_XA_NAME, node->priv->name,
PCMK_XA_ID, node->priv->id,
PCMK_XA_ONLINE, online,
PCMK_XA_STANDBY, standby,
PCMK_XA_STANDBY_ONFAIL, standby_onfail,
PCMK_XA_MAINTENANCE, maintenance,
PCMK_XA_PENDING, pending,
PCMK_XA_UNCLEAN, unclean,
PCMK_XA_HEALTH, health,
PCMK_XA_FEATURE_SET, feature_set,
PCMK_XA_SHUTDOWN, shutdown,
PCMK_XA_EXPECTED_UP, expected_up,
PCMK_XA_IS_DC, pcmk__btoa(is_dc),
PCMK_XA_RESOURCES_RUNNING, resources_running,
PCMK_XA_TYPE, node_type,
NULL);
free(resources_running);
pcmk__assert(rc == pcmk_rc_ok);
if (pcmk__is_guest_or_bundle_node(node)) {
xmlNodePtr xml_node = pcmk__output_xml_peek_parent(out);
crm_xml_add(xml_node, PCMK_XA_ID_AS_RESOURCE,
node->priv->remote->priv->launcher->id);
}
if (pcmk_is_set(show_opts, pcmk_show_rscs_by_node)) {
GList *lpc = NULL;
for (lpc = node->details->running_rsc; lpc != NULL; lpc = lpc->next) {
pcmk_resource_t *rsc = (pcmk_resource_t *) lpc->data;
show_opts |= pcmk_show_rsc_only;
out->message(out, (const char *) rsc->priv->xml->name,
show_opts, rsc, only_node, only_rsc);
}
}
out->end_list(out);
} else {
pcmk__output_xml_create_parent(out, PCMK_XE_NODE,
PCMK_XA_NAME, node->priv->name,
NULL);
}
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("node-attribute", "const char *", "const char *", "bool", "int")
static int
node_attribute_text(pcmk__output_t *out, va_list args) {
const char *name = va_arg(args, const char *);
const char *value = va_arg(args, const char *);
bool add_extra = va_arg(args, int);
int expected_score = va_arg(args, int);
if (add_extra) {
int v;
if (value == NULL) {
v = 0;
} else {
pcmk__scan_min_int(value, &v, INT_MIN);
}
if (v <= 0) {
out->list_item(out, NULL, "%-32s\t: %-10s\t: Connectivity is lost", name, value);
} else if (v < expected_score) {
out->list_item(out, NULL, "%-32s\t: %-10s\t: Connectivity is degraded (Expected=%d)", name, value, expected_score);
} else {
out->list_item(out, NULL, "%-32s\t: %-10s", name, value);
}
} else {
out->list_item(out, NULL, "%-32s\t: %-10s", name, value);
}
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("node-attribute", "const char *", "const char *", "bool", "int")
static int
node_attribute_html(pcmk__output_t *out, va_list args) {
const char *name = va_arg(args, const char *);
const char *value = va_arg(args, const char *);
bool add_extra = va_arg(args, int);
int expected_score = va_arg(args, int);
if (add_extra) {
int v = 0;
xmlNodePtr item_node = pcmk__output_create_xml_node(out, "li", NULL);
xmlNode *child = NULL;
if (value != NULL) {
pcmk__scan_min_int(value, &v, INT_MIN);
}
child = pcmk__html_create(item_node, PCMK__XE_SPAN, NULL, NULL);
pcmk__xe_set_content(child, "%s: %s", name, value);
if (v <= 0) {
child = pcmk__html_create(item_node, PCMK__XE_SPAN, NULL,
PCMK__VALUE_BOLD);
pcmk__xe_set_content(child, "(connectivity is lost)");
} else if (v < expected_score) {
child = pcmk__html_create(item_node, PCMK__XE_SPAN, NULL,
PCMK__VALUE_BOLD);
pcmk__xe_set_content(child,
"(connectivity is degraded -- expected %d)",
expected_score);
}
} else {
out->list_item(out, NULL, "%s: %s", name, value);
}
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("node-and-op", "pcmk_scheduler_t *", "xmlNode *")
static int
node_and_op(pcmk__output_t *out, va_list args) {
pcmk_scheduler_t *scheduler = va_arg(args, pcmk_scheduler_t *);
xmlNodePtr xml_op = va_arg(args, xmlNodePtr);
pcmk_resource_t *rsc = NULL;
gchar *node_str = NULL;
char *last_change_str = NULL;
const char *op_rsc = crm_element_value(xml_op, PCMK_XA_RESOURCE);
int status;
time_t last_change = 0;
pcmk__scan_min_int(crm_element_value(xml_op, PCMK__XA_OP_STATUS), &status,
PCMK_EXEC_UNKNOWN);
rsc = pe_find_resource(scheduler->priv->resources, op_rsc);
if (rsc) {
const pcmk_node_t *node = pcmk__current_node(rsc);
const char *target_role = g_hash_table_lookup(rsc->priv->meta,
PCMK_META_TARGET_ROLE);
uint32_t show_opts = pcmk_show_rsc_only | pcmk_show_pending;
if (node == NULL) {
node = rsc->priv->pending_node;
}
node_str = pcmk__native_output_string(rsc, rsc_printable_id(rsc), node,
show_opts, target_role, false);
} else {
node_str = crm_strdup_printf("Unknown resource %s", op_rsc);
}
if (crm_element_value_epoch(xml_op, PCMK_XA_LAST_RC_CHANGE,
&last_change) == pcmk_ok) {
const char *exec_time = crm_element_value(xml_op, PCMK_XA_EXEC_TIME);
last_change_str = crm_strdup_printf(", %s='%s', exec=%sms",
PCMK_XA_LAST_RC_CHANGE,
pcmk__trim(ctime(&last_change)),
exec_time);
}
out->list_item(out, NULL, "%s: %s (node=%s, call=%s, rc=%s%s): %s",
node_str, pcmk__xe_history_key(xml_op),
crm_element_value(xml_op, PCMK_XA_UNAME),
crm_element_value(xml_op, PCMK__XA_CALL_ID),
crm_element_value(xml_op, PCMK__XA_RC_CODE),
last_change_str ? last_change_str : "",
pcmk_exec_status_str(status));
g_free(node_str);
free(last_change_str);
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("node-and-op", "pcmk_scheduler_t *", "xmlNode *")
static int
node_and_op_xml(pcmk__output_t *out, va_list args) {
pcmk_scheduler_t *scheduler = va_arg(args, pcmk_scheduler_t *);
xmlNodePtr xml_op = va_arg(args, xmlNodePtr);
pcmk_resource_t *rsc = NULL;
const char *uname = crm_element_value(xml_op, PCMK_XA_UNAME);
const char *call_id = crm_element_value(xml_op, PCMK__XA_CALL_ID);
const char *rc_s = crm_element_value(xml_op, PCMK__XA_RC_CODE);
const char *status_s = NULL;
const char *op_rsc = crm_element_value(xml_op, PCMK_XA_RESOURCE);
int status;
time_t last_change = 0;
xmlNode *node = NULL;
pcmk__scan_min_int(crm_element_value(xml_op, PCMK__XA_OP_STATUS),
&status, PCMK_EXEC_UNKNOWN);
status_s = pcmk_exec_status_str(status);
node = pcmk__output_create_xml_node(out, PCMK_XE_OPERATION,
PCMK_XA_OP, pcmk__xe_history_key(xml_op),
PCMK_XA_NODE, uname,
PCMK_XA_CALL, call_id,
PCMK_XA_RC, rc_s,
PCMK_XA_STATUS, status_s,
NULL);
rsc = pe_find_resource(scheduler->priv->resources, op_rsc);
if (rsc) {
const char *class = crm_element_value(rsc->priv->xml, PCMK_XA_CLASS);
const char *provider = crm_element_value(rsc->priv->xml,
PCMK_XA_PROVIDER);
const char *kind = crm_element_value(rsc->priv->xml, PCMK_XA_TYPE);
bool has_provider = pcmk_is_set(pcmk_get_ra_caps(class),
pcmk_ra_cap_provider);
char *agent_tuple = crm_strdup_printf("%s:%s:%s",
class,
(has_provider? provider : ""),
kind);
pcmk__xe_set_props(node,
PCMK_XA_RSC, rsc_printable_id(rsc),
PCMK_XA_AGENT, agent_tuple,
NULL);
free(agent_tuple);
}
if (crm_element_value_epoch(xml_op, PCMK_XA_LAST_RC_CHANGE,
&last_change) == pcmk_ok) {
const char *last_rc_change = pcmk__trim(ctime(&last_change));
const char *exec_time = crm_element_value(xml_op, PCMK_XA_EXEC_TIME);
pcmk__xe_set_props(node,
PCMK_XA_LAST_RC_CHANGE, last_rc_change,
PCMK_XA_EXEC_TIME, exec_time,
NULL);
}
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("node-attribute", "const char *", "const char *", "bool", "int")
static int
node_attribute_xml(pcmk__output_t *out, va_list args) {
const char *name = va_arg(args, const char *);
const char *value = va_arg(args, const char *);
bool add_extra = va_arg(args, int);
int expected_score = va_arg(args, int);
xmlNodePtr node = pcmk__output_create_xml_node(out, PCMK_XE_ATTRIBUTE,
PCMK_XA_NAME, name,
PCMK_XA_VALUE, value,
NULL);
if (add_extra) {
char *buf = pcmk__itoa(expected_score);
crm_xml_add(node, PCMK_XA_EXPECTED, buf);
free(buf);
}
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("node-attribute-list", "pcmk_scheduler_t *", "uint32_t",
"bool", "GList *", "GList *")
static int
node_attribute_list(pcmk__output_t *out, va_list args) {
pcmk_scheduler_t *scheduler = va_arg(args, pcmk_scheduler_t *);
uint32_t show_opts = va_arg(args, uint32_t);
bool print_spacer = va_arg(args, int);
GList *only_node = va_arg(args, GList *);
GList *only_rsc = va_arg(args, GList *);
int rc = pcmk_rc_no_output;
/* Display each node's attributes */
for (GList *gIter = scheduler->nodes; gIter != NULL; gIter = gIter->next) {
pcmk_node_t *node = gIter->data;
GList *attr_list = NULL;
GHashTableIter iter;
gpointer key;
if (!node || !node->details || !node->details->online) {
continue;
}
+ // @TODO Maybe skip filtering for XML output
g_hash_table_iter_init(&iter, node->priv->attrs);
while (g_hash_table_iter_next (&iter, &key, NULL)) {
attr_list = filter_attr_list(attr_list, key);
}
if (attr_list == NULL) {
continue;
}
if (!pcmk__str_in_list(node->priv->name, only_node,
pcmk__str_star_matches|pcmk__str_casei)) {
g_list_free(attr_list);
continue;
}
PCMK__OUTPUT_LIST_HEADER(out, print_spacer, rc, "Node Attributes");
out->message(out, "node", node, show_opts, false, only_node, only_rsc);
for (GList *aIter = attr_list; aIter != NULL; aIter = aIter->next) {
const char *name = aIter->data;
const char *value = NULL;
int expected_score = 0;
bool add_extra = false;
value = pcmk__node_attr(node, name, NULL, pcmk__rsc_node_current);
add_extra = add_extra_info(node, node->details->running_rsc,
scheduler, name, &expected_score);
/* Print attribute name and value */
out->message(out, "node-attribute", name, value, add_extra,
expected_score);
}
g_list_free(attr_list);
out->end_list(out);
}
PCMK__OUTPUT_LIST_FOOTER(out, rc);
return rc;
}
PCMK__OUTPUT_ARGS("node-capacity", "const pcmk_node_t *", "const char *")
static int
node_capacity(pcmk__output_t *out, va_list args)
{
const pcmk_node_t *node = va_arg(args, pcmk_node_t *);
const char *comment = va_arg(args, const char *);
char *dump_text = crm_strdup_printf("%s: %s capacity:",
comment, pcmk__node_name(node));
g_hash_table_foreach(node->priv->utilization, append_dump_text,
&dump_text);
out->list_item(out, NULL, "%s", dump_text);
free(dump_text);
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("node-capacity", "const pcmk_node_t *", "const char *")
static int
node_capacity_xml(pcmk__output_t *out, va_list args)
{
const pcmk_node_t *node = va_arg(args, pcmk_node_t *);
const char *uname = node->priv->name;
const char *comment = va_arg(args, const char *);
xmlNodePtr xml_node = pcmk__output_create_xml_node(out, PCMK_XE_CAPACITY,
PCMK_XA_NODE, uname,
PCMK_XA_COMMENT, comment,
NULL);
g_hash_table_foreach(node->priv->utilization, add_dump_node, xml_node);
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("node-history-list", "pcmk_scheduler_t *", "pcmk_node_t *",
"xmlNode *", "GList *", "GList *", "uint32_t", "uint32_t")
static int
node_history_list(pcmk__output_t *out, va_list args) {
pcmk_scheduler_t *scheduler = va_arg(args, pcmk_scheduler_t *);
pcmk_node_t *node = va_arg(args, pcmk_node_t *);
xmlNode *node_state = va_arg(args, xmlNode *);
GList *only_node = va_arg(args, GList *);
GList *only_rsc = va_arg(args, GList *);
uint32_t section_opts = va_arg(args, uint32_t);
uint32_t show_opts = va_arg(args, uint32_t);
xmlNode *lrm_rsc = NULL;
xmlNode *rsc_entry = NULL;
int rc = pcmk_rc_no_output;
lrm_rsc = pcmk__xe_first_child(node_state, PCMK__XE_LRM, NULL, NULL);
lrm_rsc = pcmk__xe_first_child(lrm_rsc, PCMK__XE_LRM_RESOURCES, NULL, NULL);
/* Print history of each of the node's resources */
for (rsc_entry = pcmk__xe_first_child(lrm_rsc, PCMK__XE_LRM_RESOURCE, NULL,
NULL);
rsc_entry != NULL;
rsc_entry = pcmk__xe_next(rsc_entry, PCMK__XE_LRM_RESOURCE)) {
const char *rsc_id = crm_element_value(rsc_entry, PCMK_XA_ID);
pcmk_resource_t *rsc = NULL;
const pcmk_resource_t *parent = NULL;
if (rsc_id == NULL) {
continue; // Malformed entry
}
rsc = pe_find_resource(scheduler->priv->resources, rsc_id);
if (rsc == NULL) {
continue; // Resource was removed from configuration
}
/* We can't use is_filtered here to filter group resources. For is_filtered,
* we have to decide whether to check the parent or not. If we check the
* parent, all elements of a group will always be printed because that's how
* is_filtered works for groups. If we do not check the parent, sometimes
* this will filter everything out.
*
* For other resource types, is_filtered is okay.
*/
parent = pe__const_top_resource(rsc, false);
if (pcmk__is_group(parent)) {
if (!pcmk__str_in_list(rsc_printable_id(rsc), only_rsc,
pcmk__str_star_matches)
&& !pcmk__str_in_list(rsc_printable_id(parent), only_rsc,
pcmk__str_star_matches)) {
continue;
}
} else if (rsc->priv->fns->is_filtered(rsc, only_rsc, TRUE)) {
continue;
}
if (!pcmk_is_set(section_opts, pcmk_section_operations)) {
time_t last_failure = 0;
int failcount = pe_get_failcount(node, rsc, &last_failure,
pcmk__fc_default, NULL);
if (failcount <= 0) {
continue;
}
if (rc == pcmk_rc_no_output) {
rc = pcmk_rc_ok;
out->message(out, "node", node, show_opts, false, only_node,
only_rsc);
}
out->message(out, "resource-history", rsc, rsc_id, false,
failcount, last_failure, false);
} else {
GList *op_list = get_operation_list(rsc_entry);
pcmk_resource_t *rsc = NULL;
if (op_list == NULL) {
continue;
}
rsc = pe_find_resource(scheduler->priv->resources,
crm_element_value(rsc_entry, PCMK_XA_ID));
if (rc == pcmk_rc_no_output) {
rc = pcmk_rc_ok;
out->message(out, "node", node, show_opts, false, only_node,
only_rsc);
}
out->message(out, "resource-operation-list", scheduler, rsc, node,
op_list, show_opts);
}
}
PCMK__OUTPUT_LIST_FOOTER(out, rc);
return rc;
}
PCMK__OUTPUT_ARGS("node-list", "GList *", "GList *", "GList *", "uint32_t", "bool")
static int
node_list_html(pcmk__output_t *out, va_list args) {
GList *nodes = va_arg(args, GList *);
GList *only_node = va_arg(args, GList *);
GList *only_rsc = va_arg(args, GList *);
uint32_t show_opts = va_arg(args, uint32_t);
bool print_spacer G_GNUC_UNUSED = va_arg(args, int);
int rc = pcmk_rc_no_output;
for (GList *gIter = nodes; gIter != NULL; gIter = gIter->next) {
pcmk_node_t *node = (pcmk_node_t *) gIter->data;
if (!pcmk__str_in_list(node->priv->name, only_node,
pcmk__str_star_matches|pcmk__str_casei)) {
continue;
}
PCMK__OUTPUT_LIST_HEADER(out, false, rc, "Node List");
out->message(out, "node", node, show_opts, true, only_node, only_rsc);
}
PCMK__OUTPUT_LIST_FOOTER(out, rc);
return rc;
}
PCMK__OUTPUT_ARGS("node-list", "GList *", "GList *", "GList *", "uint32_t", "bool")
static int
node_list_text(pcmk__output_t *out, va_list args) {
GList *nodes = va_arg(args, GList *);
GList *only_node = va_arg(args, GList *);
GList *only_rsc = va_arg(args, GList *);
uint32_t show_opts = va_arg(args, uint32_t);
bool print_spacer = va_arg(args, int);
/* space-separated lists of node names */
GString *online_nodes = NULL;
GString *online_remote_nodes = NULL;
GString *online_guest_nodes = NULL;
GString *offline_nodes = NULL;
GString *offline_remote_nodes = NULL;
int rc = pcmk_rc_no_output;
for (GList *gIter = nodes; gIter != NULL; gIter = gIter->next) {
pcmk_node_t *node = (pcmk_node_t *) gIter->data;
char *node_name = pe__node_display_name(node, pcmk_is_set(show_opts, pcmk_show_node_id));
if (!pcmk__str_in_list(node->priv->name, only_node,
pcmk__str_star_matches|pcmk__str_casei)) {
free(node_name);
continue;
}
PCMK__OUTPUT_LIST_HEADER(out, print_spacer, rc, "Node List");
// Determine whether to display node individually or in a list
if (node->details->unclean || node->details->pending
|| (pcmk_is_set(node->priv->flags, pcmk__node_fail_standby)
&& node->details->online)
|| pcmk_is_set(node->priv->flags, pcmk__node_standby)
|| node->details->maintenance
|| pcmk_is_set(show_opts, pcmk_show_rscs_by_node)
|| pcmk_is_set(show_opts, pcmk_show_feature_set)
|| (pe__node_health(node) <= 0)) {
// Display node individually
} else if (node->details->online) {
// Display online node in a list
if (pcmk__is_guest_or_bundle_node(node)) {
pcmk__add_word(&online_guest_nodes, 1024, node_name);
} else if (pcmk__is_remote_node(node)) {
pcmk__add_word(&online_remote_nodes, 1024, node_name);
} else {
pcmk__add_word(&online_nodes, 1024, node_name);
}
free(node_name);
continue;
} else {
// Display offline node in a list
if (pcmk__is_remote_node(node)) {
pcmk__add_word(&offline_remote_nodes, 1024, node_name);
} else if (pcmk__is_guest_or_bundle_node(node)) {
/* ignore offline guest nodes */
} else {
pcmk__add_word(&offline_nodes, 1024, node_name);
}
free(node_name);
continue;
}
/* If we get here, node is in bad state, or we're grouping by node */
out->message(out, "node", node, show_opts, true, only_node, only_rsc);
free(node_name);
}
/* If we're not grouping by node, summarize nodes by status */
if (online_nodes != NULL) {
out->list_item(out, "Online", "[ %s ]",
(const char *) online_nodes->str);
g_string_free(online_nodes, TRUE);
}
if (offline_nodes != NULL) {
out->list_item(out, "OFFLINE", "[ %s ]",
(const char *) offline_nodes->str);
g_string_free(offline_nodes, TRUE);
}
if (online_remote_nodes) {
out->list_item(out, "RemoteOnline", "[ %s ]",
(const char *) online_remote_nodes->str);
g_string_free(online_remote_nodes, TRUE);
}
if (offline_remote_nodes) {
out->list_item(out, "RemoteOFFLINE", "[ %s ]",
(const char *) offline_remote_nodes->str);
g_string_free(offline_remote_nodes, TRUE);
}
if (online_guest_nodes != NULL) {
out->list_item(out, "GuestOnline", "[ %s ]",
(const char *) online_guest_nodes->str);
g_string_free(online_guest_nodes, TRUE);
}
PCMK__OUTPUT_LIST_FOOTER(out, rc);
return rc;
}
PCMK__OUTPUT_ARGS("node-list", "GList *", "GList *", "GList *", "uint32_t", "bool")
static int
node_list_xml(pcmk__output_t *out, va_list args) {
GList *nodes = va_arg(args, GList *);
GList *only_node = va_arg(args, GList *);
GList *only_rsc = va_arg(args, GList *);
uint32_t show_opts = va_arg(args, uint32_t);
bool print_spacer G_GNUC_UNUSED = va_arg(args, int);
/* PCMK_XE_NODES acts as the list's element name for CLI tools that use
* pcmk__output_enable_list_element. Otherwise PCMK_XE_NODES is the
* value of the list's PCMK_XA_NAME attribute.
*/
out->begin_list(out, NULL, NULL, PCMK_XE_NODES);
for (GList *gIter = nodes; gIter != NULL; gIter = gIter->next) {
pcmk_node_t *node = (pcmk_node_t *) gIter->data;
if (!pcmk__str_in_list(node->priv->name, only_node,
pcmk__str_star_matches|pcmk__str_casei)) {
continue;
}
out->message(out, "node", node, show_opts, true, only_node, only_rsc);
}
out->end_list(out);
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("node-summary", "pcmk_scheduler_t *", "GList *", "GList *",
"uint32_t", "uint32_t", "bool")
static int
node_summary(pcmk__output_t *out, va_list args) {
pcmk_scheduler_t *scheduler = va_arg(args, pcmk_scheduler_t *);
GList *only_node = va_arg(args, GList *);
GList *only_rsc = va_arg(args, GList *);
uint32_t section_opts = va_arg(args, uint32_t);
uint32_t show_opts = va_arg(args, uint32_t);
bool print_spacer = va_arg(args, int);
xmlNode *node_state = NULL;
xmlNode *cib_status = pcmk_find_cib_element(scheduler->input,
PCMK_XE_STATUS);
int rc = pcmk_rc_no_output;
if (xmlChildElementCount(cib_status) == 0) {
return rc;
}
for (node_state = pcmk__xe_first_child(cib_status, PCMK__XE_NODE_STATE,
NULL, NULL);
node_state != NULL;
node_state = pcmk__xe_next(node_state, PCMK__XE_NODE_STATE)) {
pcmk_node_t *node = pe_find_node_id(scheduler->nodes,
pcmk__xe_id(node_state));
if (!node || !node->details || !node->details->online) {
continue;
}
if (!pcmk__str_in_list(node->priv->name, only_node,
pcmk__str_star_matches|pcmk__str_casei)) {
continue;
}
PCMK__OUTPUT_LIST_HEADER(out, print_spacer, rc,
pcmk_is_set(section_opts, pcmk_section_operations) ? "Operations" : "Migration Summary");
out->message(out, "node-history-list", scheduler, node, node_state,
only_node, only_rsc, section_opts, show_opts);
}
PCMK__OUTPUT_LIST_FOOTER(out, rc);
return rc;
}
PCMK__OUTPUT_ARGS("node-weight", "const pcmk_resource_t *", "const char *",
"const char *", "const char *")
static int
node_weight(pcmk__output_t *out, va_list args)
{
const pcmk_resource_t *rsc = va_arg(args, const pcmk_resource_t *);
const char *prefix = va_arg(args, const char *);
const char *uname = va_arg(args, const char *);
const char *score = va_arg(args, const char *);
if (rsc) {
out->list_item(out, NULL, "%s: %s allocation score on %s: %s",
prefix, rsc->id, uname, score);
} else {
out->list_item(out, NULL, "%s: %s = %s", prefix, uname, score);
}
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("node-weight", "const pcmk_resource_t *", "const char *",
"const char *", "const char *")
static int
node_weight_xml(pcmk__output_t *out, va_list args)
{
const pcmk_resource_t *rsc = va_arg(args, const pcmk_resource_t *);
const char *prefix = va_arg(args, const char *);
const char *uname = va_arg(args, const char *);
const char *score = va_arg(args, const char *);
xmlNodePtr node = pcmk__output_create_xml_node(out, PCMK_XE_NODE_WEIGHT,
PCMK_XA_FUNCTION, prefix,
PCMK_XA_NODE, uname,
PCMK_XA_SCORE, score,
NULL);
if (rsc) {
crm_xml_add(node, PCMK_XA_ID, rsc->id);
}
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("op-history", "xmlNode *", "const char *", "const char *", "int", "uint32_t")
static int
op_history_text(pcmk__output_t *out, va_list args) {
xmlNodePtr xml_op = va_arg(args, xmlNodePtr);
const char *task = va_arg(args, const char *);
const char *interval_ms_s = va_arg(args, const char *);
int rc = va_arg(args, int);
uint32_t show_opts = va_arg(args, uint32_t);
char *buf = op_history_string(xml_op, task, interval_ms_s, rc,
pcmk_is_set(show_opts, pcmk_show_timing));
out->list_item(out, NULL, "%s", buf);
free(buf);
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("op-history", "xmlNode *", "const char *", "const char *", "int", "uint32_t")
static int
op_history_xml(pcmk__output_t *out, va_list args) {
xmlNodePtr xml_op = va_arg(args, xmlNodePtr);
const char *task = va_arg(args, const char *);
const char *interval_ms_s = va_arg(args, const char *);
int rc = va_arg(args, int);
uint32_t show_opts = va_arg(args, uint32_t);
const char *call_id = crm_element_value(xml_op, PCMK__XA_CALL_ID);
char *rc_s = pcmk__itoa(rc);
const char *rc_text = crm_exit_str(rc);
xmlNodePtr node = NULL;
node = pcmk__output_create_xml_node(out, PCMK_XE_OPERATION_HISTORY,
PCMK_XA_CALL, call_id,
PCMK_XA_TASK, task,
PCMK_XA_RC, rc_s,
PCMK_XA_RC_TEXT, rc_text,
NULL);
free(rc_s);
if (interval_ms_s && !pcmk__str_eq(interval_ms_s, "0", pcmk__str_casei)) {
char *s = crm_strdup_printf("%sms", interval_ms_s);
crm_xml_add(node, PCMK_XA_INTERVAL, s);
free(s);
}
if (pcmk_is_set(show_opts, pcmk_show_timing)) {
const char *value = NULL;
time_t epoch = 0;
if ((crm_element_value_epoch(xml_op, PCMK_XA_LAST_RC_CHANGE,
&epoch) == pcmk_ok) && (epoch > 0)) {
char *s = pcmk__epoch2str(&epoch, 0);
crm_xml_add(node, PCMK_XA_LAST_RC_CHANGE, s);
free(s);
}
value = crm_element_value(xml_op, PCMK_XA_EXEC_TIME);
if (value) {
char *s = crm_strdup_printf("%sms", value);
crm_xml_add(node, PCMK_XA_EXEC_TIME, s);
free(s);
}
value = crm_element_value(xml_op, PCMK_XA_QUEUE_TIME);
if (value) {
char *s = crm_strdup_printf("%sms", value);
crm_xml_add(node, PCMK_XA_QUEUE_TIME, s);
free(s);
}
}
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("promotion-score", "pcmk_resource_t *", "pcmk_node_t *",
"const char *")
static int
promotion_score(pcmk__output_t *out, va_list args)
{
pcmk_resource_t *child_rsc = va_arg(args, pcmk_resource_t *);
pcmk_node_t *chosen = va_arg(args, pcmk_node_t *);
const char *score = va_arg(args, const char *);
if (chosen == NULL) {
out->list_item(out, NULL, "%s promotion score (inactive): %s",
child_rsc->id, score);
} else {
out->list_item(out, NULL, "%s promotion score on %s: %s",
child_rsc->id, pcmk__node_name(chosen), score);
}
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("promotion-score", "pcmk_resource_t *", "pcmk_node_t *",
"const char *")
static int
promotion_score_xml(pcmk__output_t *out, va_list args)
{
pcmk_resource_t *child_rsc = va_arg(args, pcmk_resource_t *);
pcmk_node_t *chosen = va_arg(args, pcmk_node_t *);
const char *score = va_arg(args, const char *);
xmlNodePtr node = pcmk__output_create_xml_node(out, PCMK_XE_PROMOTION_SCORE,
PCMK_XA_ID, child_rsc->id,
PCMK_XA_SCORE, score,
NULL);
if (chosen) {
crm_xml_add(node, PCMK_XA_NODE, chosen->priv->name);
}
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("resource-config", "const pcmk_resource_t *", "bool")
static int
resource_config(pcmk__output_t *out, va_list args) {
const pcmk_resource_t *rsc = va_arg(args, const pcmk_resource_t *);
GString *xml_buf = g_string_sized_new(1024);
bool raw = va_arg(args, int);
formatted_xml_buf(rsc, xml_buf, raw);
out->output_xml(out, PCMK_XE_XML, xml_buf->str);
g_string_free(xml_buf, TRUE);
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("resource-config", "const pcmk_resource_t *", "bool")
static int
resource_config_text(pcmk__output_t *out, va_list args) {
pcmk__formatted_printf(out, "Resource XML:\n");
return resource_config(out, args);
}
PCMK__OUTPUT_ARGS("resource-history", "pcmk_resource_t *", "const char *",
"bool", "int", "time_t", "bool")
static int
resource_history_text(pcmk__output_t *out, va_list args) {
pcmk_resource_t *rsc = va_arg(args, pcmk_resource_t *);
const char *rsc_id = va_arg(args, const char *);
bool all = va_arg(args, int);
int failcount = va_arg(args, int);
time_t last_failure = va_arg(args, time_t);
bool as_header = va_arg(args, int);
char *buf = resource_history_string(rsc, rsc_id, all, failcount, last_failure);
if (as_header) {
out->begin_list(out, NULL, NULL, "%s", buf);
} else {
out->list_item(out, NULL, "%s", buf);
}
free(buf);
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("resource-history", "pcmk_resource_t *", "const char *",
"bool", "int", "time_t", "bool")
static int
resource_history_xml(pcmk__output_t *out, va_list args) {
pcmk_resource_t *rsc = va_arg(args, pcmk_resource_t *);
const char *rsc_id = va_arg(args, const char *);
bool all = va_arg(args, int);
int failcount = va_arg(args, int);
time_t last_failure = va_arg(args, time_t);
bool as_header = va_arg(args, int);
xmlNodePtr node = pcmk__output_xml_create_parent(out,
PCMK_XE_RESOURCE_HISTORY,
PCMK_XA_ID, rsc_id,
NULL);
if (rsc == NULL) {
pcmk__xe_set_bool_attr(node, PCMK_XA_ORPHAN, true);
} else if (all || failcount || last_failure > 0) {
char *migration_s = pcmk__itoa(rsc->priv->ban_after_failures);
pcmk__xe_set_props(node,
PCMK_XA_ORPHAN, PCMK_VALUE_FALSE,
PCMK_META_MIGRATION_THRESHOLD, migration_s,
NULL);
free(migration_s);
if (failcount > 0) {
char *s = pcmk__itoa(failcount);
crm_xml_add(node, PCMK_XA_FAIL_COUNT, s);
free(s);
}
if (last_failure > 0) {
char *s = pcmk__epoch2str(&last_failure, 0);
crm_xml_add(node, PCMK_XA_LAST_FAILURE, s);
free(s);
}
}
if (!as_header) {
pcmk__output_xml_pop_parent(out);
}
return pcmk_rc_ok;
}
static void
print_resource_header(pcmk__output_t *out, uint32_t show_opts)
{
if (pcmk_is_set(show_opts, pcmk_show_rscs_by_node)) {
/* Active resources have already been printed by node */
out->begin_list(out, NULL, NULL, "Inactive Resources");
} else if (pcmk_is_set(show_opts, pcmk_show_inactive_rscs)) {
out->begin_list(out, NULL, NULL, "Full List of Resources");
} else {
out->begin_list(out, NULL, NULL, "Active Resources");
}
}
PCMK__OUTPUT_ARGS("resource-list", "pcmk_scheduler_t *", "uint32_t", "bool",
"GList *", "GList *", "bool")
static int
resource_list(pcmk__output_t *out, va_list args)
{
pcmk_scheduler_t *scheduler = va_arg(args, pcmk_scheduler_t *);
uint32_t show_opts = va_arg(args, uint32_t);
bool print_summary = va_arg(args, int);
GList *only_node = va_arg(args, GList *);
GList *only_rsc = va_arg(args, GList *);
bool print_spacer = va_arg(args, int);
GList *rsc_iter;
int rc = pcmk_rc_no_output;
bool printed_header = false;
/* If we already showed active resources by node, and
* we're not showing inactive resources, we have nothing to do
*/
if (pcmk_is_set(show_opts, pcmk_show_rscs_by_node) &&
!pcmk_is_set(show_opts, pcmk_show_inactive_rscs)) {
return rc;
}
/* If we haven't already printed resources grouped by node,
* and brief output was requested, print resource summary */
if (pcmk_is_set(show_opts, pcmk_show_brief)
&& !pcmk_is_set(show_opts, pcmk_show_rscs_by_node)) {
GList *rscs = pe__filter_rsc_list(scheduler->priv->resources, only_rsc);
PCMK__OUTPUT_SPACER_IF(out, print_spacer);
print_resource_header(out, show_opts);
printed_header = true;
rc = pe__rscs_brief_output(out, rscs, show_opts);
g_list_free(rscs);
}
/* For each resource, display it if appropriate */
for (rsc_iter = scheduler->priv->resources;
rsc_iter != NULL; rsc_iter = rsc_iter->next) {
pcmk_resource_t *rsc = (pcmk_resource_t *) rsc_iter->data;
int x;
/* Complex resources may have some sub-resources active and some inactive */
gboolean is_active = rsc->priv->fns->active(rsc, TRUE);
gboolean partially_active = rsc->priv->fns->active(rsc, FALSE);
/* Skip inactive orphans (deleted but still in CIB) */
if (pcmk_is_set(rsc->flags, pcmk__rsc_removed) && !is_active) {
continue;
/* Skip active resources if we already displayed them by node */
} else if (pcmk_is_set(show_opts, pcmk_show_rscs_by_node)) {
if (is_active) {
continue;
}
/* Skip primitives already counted in a brief summary */
} else if (pcmk_is_set(show_opts, pcmk_show_brief)
&& pcmk__is_primitive(rsc)) {
continue;
/* Skip resources that aren't at least partially active,
* unless we're displaying inactive resources
*/
} else if (!partially_active && !pcmk_is_set(show_opts, pcmk_show_inactive_rscs)) {
continue;
} else if (partially_active && !pe__rsc_running_on_any(rsc, only_node)) {
continue;
}
if (!printed_header) {
PCMK__OUTPUT_SPACER_IF(out, print_spacer);
print_resource_header(out, show_opts);
printed_header = true;
}
/* Print this resource */
x = out->message(out, (const char *) rsc->priv->xml->name,
show_opts, rsc, only_node, only_rsc);
if (x == pcmk_rc_ok) {
rc = pcmk_rc_ok;
}
}
if (print_summary && rc != pcmk_rc_ok) {
if (!printed_header) {
PCMK__OUTPUT_SPACER_IF(out, print_spacer);
print_resource_header(out, show_opts);
printed_header = true;
}
if (pcmk_is_set(show_opts, pcmk_show_rscs_by_node)) {
out->list_item(out, NULL, "No inactive resources");
} else if (pcmk_is_set(show_opts, pcmk_show_inactive_rscs)) {
out->list_item(out, NULL, "No resources");
} else {
out->list_item(out, NULL, "No active resources");
}
}
if (printed_header) {
out->end_list(out);
}
return rc;
}
PCMK__OUTPUT_ARGS("resource-operation-list", "pcmk_scheduler_t *",
"pcmk_resource_t *", "pcmk_node_t *", "GList *", "uint32_t")
static int
resource_operation_list(pcmk__output_t *out, va_list args)
{
pcmk_scheduler_t *scheduler G_GNUC_UNUSED = va_arg(args,
pcmk_scheduler_t *);
pcmk_resource_t *rsc = va_arg(args, pcmk_resource_t *);
pcmk_node_t *node = va_arg(args, pcmk_node_t *);
GList *op_list = va_arg(args, GList *);
uint32_t show_opts = va_arg(args, uint32_t);
GList *gIter = NULL;
int rc = pcmk_rc_no_output;
/* Print each operation */
for (gIter = op_list; gIter != NULL; gIter = gIter->next) {
xmlNode *xml_op = (xmlNode *) gIter->data;
const char *task = crm_element_value(xml_op, PCMK_XA_OPERATION);
const char *interval_ms_s = crm_element_value(xml_op,
PCMK_META_INTERVAL);
const char *op_rc = crm_element_value(xml_op, PCMK__XA_RC_CODE);
int op_rc_i;
pcmk__scan_min_int(op_rc, &op_rc_i, 0);
/* Display 0-interval monitors as "probe" */
if (pcmk__str_eq(task, PCMK_ACTION_MONITOR, pcmk__str_casei)
&& pcmk__str_eq(interval_ms_s, "0", pcmk__str_null_matches | pcmk__str_casei)) {
task = "probe";
}
/* If this is the first printed operation, print heading for resource */
if (rc == pcmk_rc_no_output) {
time_t last_failure = 0;
int failcount = pe_get_failcount(node, rsc, &last_failure,
pcmk__fc_default, NULL);
out->message(out, "resource-history", rsc, rsc_printable_id(rsc), true,
failcount, last_failure, true);
rc = pcmk_rc_ok;
}
/* Print the operation */
out->message(out, "op-history", xml_op, task, interval_ms_s,
op_rc_i, show_opts);
}
/* Free the list we created (no need to free the individual items) */
g_list_free(op_list);
PCMK__OUTPUT_LIST_FOOTER(out, rc);
return rc;
}
PCMK__OUTPUT_ARGS("resource-util", "pcmk_resource_t *", "pcmk_node_t *",
"const char *")
static int
resource_util(pcmk__output_t *out, va_list args)
{
pcmk_resource_t *rsc = va_arg(args, pcmk_resource_t *);
pcmk_node_t *node = va_arg(args, pcmk_node_t *);
const char *fn = va_arg(args, const char *);
char *dump_text = crm_strdup_printf("%s: %s utilization on %s:",
fn, rsc->id, pcmk__node_name(node));
g_hash_table_foreach(rsc->priv->utilization, append_dump_text,
&dump_text);
out->list_item(out, NULL, "%s", dump_text);
free(dump_text);
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("resource-util", "pcmk_resource_t *", "pcmk_node_t *",
"const char *")
static int
resource_util_xml(pcmk__output_t *out, va_list args)
{
pcmk_resource_t *rsc = va_arg(args, pcmk_resource_t *);
pcmk_node_t *node = va_arg(args, pcmk_node_t *);
const char *uname = node->priv->name;
const char *fn = va_arg(args, const char *);
xmlNodePtr xml_node = NULL;
xml_node = pcmk__output_create_xml_node(out, PCMK_XE_UTILIZATION,
PCMK_XA_RESOURCE, rsc->id,
PCMK_XA_NODE, uname,
PCMK_XA_FUNCTION, fn,
NULL);
g_hash_table_foreach(rsc->priv->utilization, add_dump_node, xml_node);
return pcmk_rc_ok;
}
static inline const char *
ticket_status(pcmk__ticket_t *ticket)
{
if (pcmk_is_set(ticket->flags, pcmk__ticket_granted)) {
return PCMK_VALUE_GRANTED;
}
return PCMK_VALUE_REVOKED;
}
static inline const char *
ticket_standby_text(pcmk__ticket_t *ticket)
{
return pcmk_is_set(ticket->flags, pcmk__ticket_standby)? " [standby]" : "";
}
PCMK__OUTPUT_ARGS("ticket", "pcmk__ticket_t *", "bool", "bool")
static int
ticket_default(pcmk__output_t *out, va_list args) {
pcmk__ticket_t *ticket = va_arg(args, pcmk__ticket_t *);
bool raw = va_arg(args, int);
bool details = va_arg(args, int);
GString *detail_str = NULL;
if (raw) {
out->list_item(out, ticket->id, "%s", ticket->id);
return pcmk_rc_ok;
}
if (details && g_hash_table_size(ticket->state) > 0) {
GHashTableIter iter;
const char *name = NULL;
const char *value = NULL;
bool already_added = false;
detail_str = g_string_sized_new(100);
pcmk__g_strcat(detail_str, "\t(", NULL);
g_hash_table_iter_init(&iter, ticket->state);
while (g_hash_table_iter_next(&iter, (void **) &name, (void **) &value)) {
if (already_added) {
g_string_append_printf(detail_str, ", %s=", name);
} else {
g_string_append_printf(detail_str, "%s=", name);
already_added = true;
}
if (pcmk__str_any_of(name, PCMK_XA_LAST_GRANTED, "expires", NULL)) {
char *epoch_str = NULL;
long long time_ll;
(void) pcmk__scan_ll(value, &time_ll, 0);
epoch_str = pcmk__epoch2str((const time_t *) &time_ll, 0);
pcmk__g_strcat(detail_str, epoch_str, NULL);
free(epoch_str);
} else {
pcmk__g_strcat(detail_str, value, NULL);
}
}
pcmk__g_strcat(detail_str, ")", NULL);
}
if (ticket->last_granted > -1) {
/* Prior to the introduction of the details & raw arguments to this
* function, last-granted would always be added in this block. We need
* to preserve that behavior. At the same time, we also need to preserve
* the existing behavior from crm_ticket, which would include last-granted
* as part of the (...) detail string.
*
* Luckily we can check detail_str - if it's NULL, either there were no
* details, or we are preserving the previous behavior of this function.
* If it's not NULL, we are either preserving the previous behavior of
* crm_ticket or we were given details=true as an argument.
*/
if (detail_str == NULL) {
char *epoch_str = pcmk__epoch2str(&(ticket->last_granted), 0);
out->list_item(out, NULL, "%s\t%s%s last-granted=\"%s\"",
ticket->id, ticket_status(ticket),
ticket_standby_text(ticket), pcmk__s(epoch_str, ""));
free(epoch_str);
} else {
out->list_item(out, NULL, "%s\t%s%s %s",
ticket->id, ticket_status(ticket),
ticket_standby_text(ticket), detail_str->str);
}
} else {
out->list_item(out, NULL, "%s\t%s%s%s", ticket->id,
ticket_status(ticket),
ticket_standby_text(ticket),
detail_str != NULL ? detail_str->str : "");
}
if (detail_str != NULL) {
g_string_free(detail_str, TRUE);
}
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("ticket", "pcmk__ticket_t *", "bool", "bool")
static int
ticket_xml(pcmk__output_t *out, va_list args) {
pcmk__ticket_t *ticket = va_arg(args, pcmk__ticket_t *);
bool raw G_GNUC_UNUSED = va_arg(args, int);
bool details G_GNUC_UNUSED = va_arg(args, int);
const char *standby = pcmk__flag_text(ticket->flags, pcmk__ticket_standby);
xmlNodePtr node = NULL;
GHashTableIter iter;
const char *name = NULL;
const char *value = NULL;
node = pcmk__output_create_xml_node(out, PCMK_XE_TICKET,
PCMK_XA_ID, ticket->id,
PCMK_XA_STATUS, ticket_status(ticket),
PCMK_XA_STANDBY, standby,
NULL);
if (ticket->last_granted > -1) {
char *buf = pcmk__epoch2str(&ticket->last_granted, 0);
crm_xml_add(node, PCMK_XA_LAST_GRANTED, buf);
free(buf);
}
g_hash_table_iter_init(&iter, ticket->state);
while (g_hash_table_iter_next(&iter, (void **) &name, (void **) &value)) {
/* PCMK_XA_LAST_GRANTED and "expires" are already added by the check
* for ticket->last_granted above.
*/
if (pcmk__str_any_of(name, PCMK_XA_LAST_GRANTED, PCMK_XA_EXPIRES,
NULL)) {
continue;
}
crm_xml_add(node, name, value);
}
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("ticket-list", "GHashTable *", "bool", "bool", "bool")
static int
ticket_list(pcmk__output_t *out, va_list args) {
GHashTable *tickets = va_arg(args, GHashTable *);
bool print_spacer = va_arg(args, int);
bool raw = va_arg(args, int);
bool details = va_arg(args, int);
GHashTableIter iter;
gpointer value;
if (g_hash_table_size(tickets) == 0) {
return pcmk_rc_no_output;
}
PCMK__OUTPUT_SPACER_IF(out, print_spacer);
/* Print section heading */
out->begin_list(out, NULL, NULL, "Tickets");
/* Print each ticket */
g_hash_table_iter_init(&iter, tickets);
while (g_hash_table_iter_next(&iter, NULL, &value)) {
pcmk__ticket_t *ticket = (pcmk__ticket_t *) value;
out->message(out, "ticket", ticket, raw, details);
}
/* Close section */
out->end_list(out);
return pcmk_rc_ok;
}
static pcmk__message_entry_t fmt_functions[] = {
{ "ban", "default", ban_text },
{ "ban", "html", ban_html },
{ "ban", "xml", ban_xml },
{ "ban-list", "default", ban_list },
{ "bundle", "default", pe__bundle_text },
{ "bundle", "xml", pe__bundle_xml },
{ "bundle", "html", pe__bundle_html },
{ "clone", "default", pe__clone_default },
{ "clone", "xml", pe__clone_xml },
{ "cluster-counts", "default", cluster_counts_text },
{ "cluster-counts", "html", cluster_counts_html },
{ "cluster-counts", "xml", cluster_counts_xml },
{ "cluster-dc", "default", cluster_dc_text },
{ "cluster-dc", "html", cluster_dc_html },
{ "cluster-dc", "xml", cluster_dc_xml },
{ "cluster-options", "default", cluster_options_text },
{ "cluster-options", "html", cluster_options_html },
{ "cluster-options", "log", cluster_options_log },
{ "cluster-options", "xml", cluster_options_xml },
{ "cluster-summary", "default", cluster_summary },
{ "cluster-summary", "html", cluster_summary_html },
{ "cluster-stack", "default", cluster_stack_text },
{ "cluster-stack", "html", cluster_stack_html },
{ "cluster-stack", "xml", cluster_stack_xml },
{ "cluster-times", "default", cluster_times_text },
{ "cluster-times", "html", cluster_times_html },
{ "cluster-times", "xml", cluster_times_xml },
{ "failed-action", "default", failed_action_default },
{ "failed-action", "xml", failed_action_xml },
{ "failed-action-list", "default", failed_action_list },
{ "group", "default", pe__group_default},
{ "group", "xml", pe__group_xml },
{ "maint-mode", "text", cluster_maint_mode_text },
{ "node", "default", node_text },
{ "node", "html", node_html },
{ "node", "xml", node_xml },
{ "node-and-op", "default", node_and_op },
{ "node-and-op", "xml", node_and_op_xml },
{ "node-capacity", "default", node_capacity },
{ "node-capacity", "xml", node_capacity_xml },
{ "node-history-list", "default", node_history_list },
{ "node-list", "default", node_list_text },
{ "node-list", "html", node_list_html },
{ "node-list", "xml", node_list_xml },
{ "node-weight", "default", node_weight },
{ "node-weight", "xml", node_weight_xml },
{ "node-attribute", "default", node_attribute_text },
{ "node-attribute", "html", node_attribute_html },
{ "node-attribute", "xml", node_attribute_xml },
{ "node-attribute-list", "default", node_attribute_list },
{ "node-summary", "default", node_summary },
{ "op-history", "default", op_history_text },
{ "op-history", "xml", op_history_xml },
{ "primitive", "default", pe__resource_text },
{ "primitive", "xml", pe__resource_xml },
{ "primitive", "html", pe__resource_html },
{ "promotion-score", "default", promotion_score },
{ "promotion-score", "xml", promotion_score_xml },
{ "resource-config", "default", resource_config },
{ "resource-config", "text", resource_config_text },
{ "resource-history", "default", resource_history_text },
{ "resource-history", "xml", resource_history_xml },
{ "resource-list", "default", resource_list },
{ "resource-operation-list", "default", resource_operation_list },
{ "resource-util", "default", resource_util },
{ "resource-util", "xml", resource_util_xml },
{ "ticket", "default", ticket_default },
{ "ticket", "xml", ticket_xml },
{ "ticket-list", "default", ticket_list },
{ NULL, NULL, NULL }
};
void
pe__register_messages(pcmk__output_t *out) {
pcmk__register_messages(out, fmt_functions);
}
diff --git a/lib/pengine/unpack.c b/lib/pengine/unpack.c
index 1fe58ab9c2..973e94bdc0 100644
--- a/lib/pengine/unpack.c
+++ b/lib/pengine/unpack.c
@@ -1,5099 +1,5109 @@
/*
* Copyright 2004-2024 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU Lesser General Public License
* version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY.
*/
#include <crm_internal.h>
#include <stdio.h>
#include <string.h>
#include <glib.h>
#include <time.h>
#include <crm/crm.h>
#include <crm/services.h>
#include <crm/common/xml.h>
#include <crm/common/xml_internal.h>
#include <crm/common/util.h>
#include <crm/pengine/internal.h>
#include <pe_status_private.h>
CRM_TRACE_INIT_DATA(pe_status);
// A (parsed) resource action history entry
struct action_history {
pcmk_resource_t *rsc; // Resource that history is for
pcmk_node_t *node; // Node that history is for
xmlNode *xml; // History entry XML
// Parsed from entry XML
const char *id; // XML ID of history entry
const char *key; // Operation key of action
const char *task; // Action name
const char *exit_reason; // Exit reason given for result
guint interval_ms; // Action interval
int call_id; // Call ID of action
int expected_exit_status; // Expected exit status of action
int exit_status; // Actual exit status of action
int execution_status; // Execution status of action
};
/* This uses pcmk__set_flags_as()/pcmk__clear_flags_as() directly rather than
* use pcmk__set_scheduler_flags()/pcmk__clear_scheduler_flags() so that the
* flag is stringified more readably in log messages.
*/
#define set_config_flag(scheduler, option, flag) do { \
GHashTable *config_hash = (scheduler)->priv->options; \
const char *scf_value = pcmk__cluster_option(config_hash, (option)); \
\
if (scf_value != NULL) { \
if (crm_is_true(scf_value)) { \
(scheduler)->flags = pcmk__set_flags_as(__func__, __LINE__, \
LOG_TRACE, "Scheduler", \
crm_system_name, (scheduler)->flags, \
(flag), #flag); \
} else { \
(scheduler)->flags = pcmk__clear_flags_as(__func__, __LINE__, \
LOG_TRACE, "Scheduler", \
crm_system_name, (scheduler)->flags, \
(flag), #flag); \
} \
} \
} while(0)
static void unpack_rsc_op(pcmk_resource_t *rsc, pcmk_node_t *node,
xmlNode *xml_op, xmlNode **last_failure,
enum pcmk__on_fail *failed);
static void determine_remote_online_status(pcmk_scheduler_t *scheduler,
pcmk_node_t *this_node);
static void add_node_attrs(const xmlNode *xml_obj, pcmk_node_t *node,
bool overwrite, pcmk_scheduler_t *scheduler);
static void determine_online_status(const xmlNode *node_state,
pcmk_node_t *this_node,
pcmk_scheduler_t *scheduler);
static void unpack_node_lrm(pcmk_node_t *node, const xmlNode *xml,
pcmk_scheduler_t *scheduler);
/*!
* \internal
* \brief Check whether a node is a dangling guest node
*
* \param[in] node Node to check
*
* \return true if \p node had a Pacemaker Remote connection resource with a
* launcher that was removed from the CIB, otherwise false.
*/
static bool
is_dangling_guest_node(pcmk_node_t *node)
{
return pcmk__is_pacemaker_remote_node(node)
&& (node->priv->remote != NULL)
&& (node->priv->remote->priv->launcher == NULL)
&& pcmk_is_set(node->priv->remote->flags,
pcmk__rsc_removed_launched);
}
/*!
* \brief Schedule a fence action for a node
*
* \param[in,out] scheduler Scheduler data
* \param[in,out] node Node to fence
* \param[in] reason Text description of why fencing is needed
* \param[in] priority_delay Whether to consider
* \c PCMK_OPT_PRIORITY_FENCING_DELAY
*/
void
pe_fence_node(pcmk_scheduler_t *scheduler, pcmk_node_t *node,
const char *reason, bool priority_delay)
{
CRM_CHECK(node, return);
if (pcmk__is_guest_or_bundle_node(node)) {
// Fence a guest or bundle node by marking its launcher as failed
pcmk_resource_t *rsc = node->priv->remote->priv->launcher;
if (!pcmk_is_set(rsc->flags, pcmk__rsc_failed)) {
if (!pcmk_is_set(rsc->flags, pcmk__rsc_managed)) {
crm_notice("Not fencing guest node %s "
"(otherwise would because %s): "
"its guest resource %s is unmanaged",
pcmk__node_name(node), reason, rsc->id);
} else {
pcmk__sched_warn(scheduler,
"Guest node %s will be fenced "
"(by recovering its guest resource %s): %s",
pcmk__node_name(node), rsc->id, reason);
/* We don't mark the node as unclean because that would prevent the
* node from running resources. We want to allow it to run resources
* in this transition if the recovery succeeds.
*/
pcmk__set_node_flags(node, pcmk__node_remote_reset);
pcmk__set_rsc_flags(rsc,
pcmk__rsc_failed|pcmk__rsc_stop_if_failed);
}
}
} else if (is_dangling_guest_node(node)) {
crm_info("Cleaning up dangling connection for guest node %s: "
"fencing was already done because %s, "
"and guest resource no longer exists",
pcmk__node_name(node), reason);
pcmk__set_rsc_flags(node->priv->remote,
pcmk__rsc_failed|pcmk__rsc_stop_if_failed);
} else if (pcmk__is_remote_node(node)) {
pcmk_resource_t *rsc = node->priv->remote;
if ((rsc != NULL) && !pcmk_is_set(rsc->flags, pcmk__rsc_managed)) {
crm_notice("Not fencing remote node %s "
"(otherwise would because %s): connection is unmanaged",
pcmk__node_name(node), reason);
} else if (!pcmk_is_set(node->priv->flags, pcmk__node_remote_reset)) {
pcmk__set_node_flags(node, pcmk__node_remote_reset);
pcmk__sched_warn(scheduler, "Remote node %s %s: %s",
pcmk__node_name(node),
pe_can_fence(scheduler, node)? "will be fenced" : "is unclean",
reason);
}
node->details->unclean = TRUE;
// No need to apply PCMK_OPT_PRIORITY_FENCING_DELAY for remote nodes
pe_fence_op(node, NULL, TRUE, reason, FALSE, scheduler);
} else if (node->details->unclean) {
crm_trace("Cluster node %s %s because %s",
pcmk__node_name(node),
pe_can_fence(scheduler, node)? "would also be fenced" : "also is unclean",
reason);
} else {
pcmk__sched_warn(scheduler, "Cluster node %s %s: %s",
pcmk__node_name(node),
pe_can_fence(scheduler, node)? "will be fenced" : "is unclean",
reason);
node->details->unclean = TRUE;
pe_fence_op(node, NULL, TRUE, reason, priority_delay, scheduler);
}
}
// @TODO xpaths can't handle templates, rules, or id-refs
// nvpair with provides or requires set to unfencing
#define XPATH_UNFENCING_NVPAIR PCMK_XE_NVPAIR \
"[(@" PCMK_XA_NAME "='" PCMK_STONITH_PROVIDES "'" \
"or @" PCMK_XA_NAME "='" PCMK_META_REQUIRES "') " \
"and @" PCMK_XA_VALUE "='" PCMK_VALUE_UNFENCING "']"
// unfencing in rsc_defaults or any resource
#define XPATH_ENABLE_UNFENCING \
"/" PCMK_XE_CIB "/" PCMK_XE_CONFIGURATION "/" PCMK_XE_RESOURCES \
"//" PCMK_XE_META_ATTRIBUTES "/" XPATH_UNFENCING_NVPAIR \
"|/" PCMK_XE_CIB "/" PCMK_XE_CONFIGURATION "/" PCMK_XE_RSC_DEFAULTS \
"/" PCMK_XE_META_ATTRIBUTES "/" XPATH_UNFENCING_NVPAIR
static void
set_if_xpath(uint64_t flag, const char *xpath, pcmk_scheduler_t *scheduler)
{
xmlXPathObjectPtr result = NULL;
if (!pcmk_is_set(scheduler->flags, flag)) {
result = xpath_search(scheduler->input, xpath);
if (result && (numXpathResults(result) > 0)) {
pcmk__set_scheduler_flags(scheduler, flag);
}
freeXpathObject(result);
}
}
gboolean
unpack_config(xmlNode *config, pcmk_scheduler_t *scheduler)
{
const char *value = NULL;
GHashTable *config_hash = pcmk__strkey_table(free, free);
const pcmk_rule_input_t rule_input = {
.now = scheduler->priv->now,
};
scheduler->priv->options = config_hash;
pe__unpack_dataset_nvpairs(config, PCMK_XE_CLUSTER_PROPERTY_SET,
&rule_input, config_hash,
PCMK_VALUE_CIB_BOOTSTRAP_OPTIONS, scheduler);
pcmk__validate_cluster_options(config_hash);
set_config_flag(scheduler, PCMK_OPT_ENABLE_STARTUP_PROBES,
pcmk__sched_probe_resources);
if (!pcmk_is_set(scheduler->flags, pcmk__sched_probe_resources)) {
crm_info("Startup probes: disabled (dangerous)");
}
value = pcmk__cluster_option(config_hash, PCMK_OPT_HAVE_WATCHDOG);
if (value && crm_is_true(value)) {
crm_info("Watchdog-based self-fencing will be performed via SBD if "
"fencing is required and " PCMK_OPT_STONITH_WATCHDOG_TIMEOUT
" is nonzero");
pcmk__set_scheduler_flags(scheduler, pcmk__sched_have_fencing);
}
/* Set certain flags via xpath here, so they can be used before the relevant
* configuration sections are unpacked.
*/
set_if_xpath(pcmk__sched_enable_unfencing, XPATH_ENABLE_UNFENCING,
scheduler);
value = pcmk__cluster_option(config_hash, PCMK_OPT_STONITH_TIMEOUT);
pcmk_parse_interval_spec(value, &(scheduler->priv->fence_timeout_ms));
crm_debug("Default fencing action timeout: %s",
pcmk__readable_interval(scheduler->priv->fence_timeout_ms));
set_config_flag(scheduler, PCMK_OPT_STONITH_ENABLED,
pcmk__sched_fencing_enabled);
if (pcmk_is_set(scheduler->flags, pcmk__sched_fencing_enabled)) {
crm_debug("STONITH of failed nodes is enabled");
} else {
crm_debug("STONITH of failed nodes is disabled");
}
scheduler->priv->fence_action =
pcmk__cluster_option(config_hash, PCMK_OPT_STONITH_ACTION);
crm_trace("STONITH will %s nodes", scheduler->priv->fence_action);
set_config_flag(scheduler, PCMK_OPT_CONCURRENT_FENCING,
pcmk__sched_concurrent_fencing);
if (pcmk_is_set(scheduler->flags, pcmk__sched_concurrent_fencing)) {
crm_debug("Concurrent fencing is enabled");
} else {
crm_debug("Concurrent fencing is disabled");
}
value = pcmk__cluster_option(config_hash, PCMK_OPT_PRIORITY_FENCING_DELAY);
if (value) {
pcmk_parse_interval_spec(value,
&(scheduler->priv->priority_fencing_ms));
crm_trace("Priority fencing delay is %s",
pcmk__readable_interval(scheduler->priv->priority_fencing_ms));
}
set_config_flag(scheduler, PCMK_OPT_STOP_ALL_RESOURCES,
pcmk__sched_stop_all);
crm_debug("Stop all active resources: %s",
pcmk__flag_text(scheduler->flags, pcmk__sched_stop_all));
set_config_flag(scheduler, PCMK_OPT_SYMMETRIC_CLUSTER,
pcmk__sched_symmetric_cluster);
if (pcmk_is_set(scheduler->flags, pcmk__sched_symmetric_cluster)) {
crm_debug("Cluster is symmetric" " - resources can run anywhere by default");
}
value = pcmk__cluster_option(config_hash, PCMK_OPT_NO_QUORUM_POLICY);
if (pcmk__str_eq(value, PCMK_VALUE_IGNORE, pcmk__str_casei)) {
scheduler->no_quorum_policy = pcmk_no_quorum_ignore;
} else if (pcmk__str_eq(value, PCMK_VALUE_FREEZE, pcmk__str_casei)) {
scheduler->no_quorum_policy = pcmk_no_quorum_freeze;
} else if (pcmk__str_eq(value, PCMK_VALUE_DEMOTE, pcmk__str_casei)) {
scheduler->no_quorum_policy = pcmk_no_quorum_demote;
} else if (pcmk__strcase_any_of(value, PCMK_VALUE_FENCE,
PCMK_VALUE_FENCE_LEGACY, NULL)) {
if (pcmk_is_set(scheduler->flags, pcmk__sched_fencing_enabled)) {
int do_panic = 0;
crm_element_value_int(scheduler->input, PCMK_XA_NO_QUORUM_PANIC,
&do_panic);
if (do_panic
|| pcmk_is_set(scheduler->flags, pcmk__sched_quorate)) {
scheduler->no_quorum_policy = pcmk_no_quorum_fence;
} else {
crm_notice("Resetting " PCMK_OPT_NO_QUORUM_POLICY
" to 'stop': cluster has never had quorum");
scheduler->no_quorum_policy = pcmk_no_quorum_stop;
}
} else {
pcmk__config_err("Resetting " PCMK_OPT_NO_QUORUM_POLICY
" to 'stop' because fencing is disabled");
scheduler->no_quorum_policy = pcmk_no_quorum_stop;
}
} else {
scheduler->no_quorum_policy = pcmk_no_quorum_stop;
}
switch (scheduler->no_quorum_policy) {
case pcmk_no_quorum_freeze:
crm_debug("On loss of quorum: Freeze resources");
break;
case pcmk_no_quorum_stop:
crm_debug("On loss of quorum: Stop ALL resources");
break;
case pcmk_no_quorum_demote:
crm_debug("On loss of quorum: "
"Demote promotable resources and stop other resources");
break;
case pcmk_no_quorum_fence:
crm_notice("On loss of quorum: Fence all remaining nodes");
break;
case pcmk_no_quorum_ignore:
crm_notice("On loss of quorum: Ignore");
break;
}
set_config_flag(scheduler, PCMK_OPT_STOP_ORPHAN_RESOURCES,
pcmk__sched_stop_removed_resources);
if (pcmk_is_set(scheduler->flags, pcmk__sched_stop_removed_resources)) {
crm_trace("Orphan resources are stopped");
} else {
crm_trace("Orphan resources are ignored");
}
set_config_flag(scheduler, PCMK_OPT_STOP_ORPHAN_ACTIONS,
pcmk__sched_cancel_removed_actions);
if (pcmk_is_set(scheduler->flags, pcmk__sched_cancel_removed_actions)) {
crm_trace("Orphan resource actions are stopped");
} else {
crm_trace("Orphan resource actions are ignored");
}
set_config_flag(scheduler, PCMK_OPT_MAINTENANCE_MODE,
pcmk__sched_in_maintenance);
crm_trace("Maintenance mode: %s",
pcmk__flag_text(scheduler->flags, pcmk__sched_in_maintenance));
set_config_flag(scheduler, PCMK_OPT_START_FAILURE_IS_FATAL,
pcmk__sched_start_failure_fatal);
if (pcmk_is_set(scheduler->flags, pcmk__sched_start_failure_fatal)) {
crm_trace("Start failures are always fatal");
} else {
crm_trace("Start failures are handled by failcount");
}
if (pcmk_is_set(scheduler->flags, pcmk__sched_fencing_enabled)) {
set_config_flag(scheduler, PCMK_OPT_STARTUP_FENCING,
pcmk__sched_startup_fencing);
}
if (pcmk_is_set(scheduler->flags, pcmk__sched_startup_fencing)) {
crm_trace("Unseen nodes will be fenced");
} else {
pcmk__warn_once(pcmk__wo_blind,
"Blind faith: not fencing unseen nodes");
}
pe__unpack_node_health_scores(scheduler);
scheduler->priv->placement_strategy =
pcmk__cluster_option(config_hash, PCMK_OPT_PLACEMENT_STRATEGY);
crm_trace("Placement strategy: %s", scheduler->priv->placement_strategy);
set_config_flag(scheduler, PCMK_OPT_SHUTDOWN_LOCK,
pcmk__sched_shutdown_lock);
if (pcmk_is_set(scheduler->flags, pcmk__sched_shutdown_lock)) {
value = pcmk__cluster_option(config_hash, PCMK_OPT_SHUTDOWN_LOCK_LIMIT);
pcmk_parse_interval_spec(value, &(scheduler->priv->shutdown_lock_ms));
crm_trace("Resources will be locked to nodes that were cleanly "
"shut down (locks expire after %s)",
pcmk__readable_interval(scheduler->priv->shutdown_lock_ms));
} else {
crm_trace("Resources will not be locked to nodes that were cleanly "
"shut down");
}
value = pcmk__cluster_option(config_hash, PCMK_OPT_NODE_PENDING_TIMEOUT);
pcmk_parse_interval_spec(value, &(scheduler->priv->node_pending_ms));
if (scheduler->priv->node_pending_ms == 0U) {
crm_trace("Do not fence pending nodes");
} else {
crm_trace("Fence pending nodes after %s",
pcmk__readable_interval(scheduler->priv->node_pending_ms));
}
return TRUE;
}
/*!
* \internal
* \brief Create a new node object in scheduler data
*
* \param[in] id ID of new node
* \param[in] uname Name of new node
* \param[in] type Type of new node
* \param[in] score Score of new node
* \param[in,out] scheduler Scheduler data
*
* \return Newly created node object
* \note The returned object is part of the scheduler data and should not be
* freed separately.
*/
pcmk_node_t *
pe_create_node(const char *id, const char *uname, const char *type,
int score, pcmk_scheduler_t *scheduler)
{
enum pcmk__node_variant variant = pcmk__node_variant_cluster;
pcmk_node_t *new_node = NULL;
if (pcmk_find_node(scheduler, uname) != NULL) {
pcmk__config_warn("More than one node entry has name '%s'", uname);
}
if (pcmk__str_eq(type, PCMK_VALUE_MEMBER,
pcmk__str_null_matches|pcmk__str_casei)) {
variant = pcmk__node_variant_cluster;
} else if (pcmk__str_eq(type, PCMK_VALUE_REMOTE, pcmk__str_casei)) {
variant = pcmk__node_variant_remote;
} else {
pcmk__config_err("Ignoring node %s with unrecognized type '%s'",
pcmk__s(uname, "without name"), type);
return NULL;
}
new_node = calloc(1, sizeof(pcmk_node_t));
if (new_node == NULL) {
pcmk__sched_err(scheduler, "Could not allocate memory for node %s",
uname);
return NULL;
}
new_node->assign = calloc(1, sizeof(struct pcmk__node_assignment));
new_node->details = calloc(1, sizeof(struct pcmk__node_details));
new_node->priv = calloc(1, sizeof(pcmk__node_private_t));
if ((new_node->assign == NULL) || (new_node->details == NULL)
|| (new_node->priv == NULL)) {
free(new_node->assign);
free(new_node->details);
free(new_node->priv);
free(new_node);
pcmk__sched_err(scheduler, "Could not allocate memory for node %s",
uname);
return NULL;
}
crm_trace("Creating node for entry %s/%s", uname, id);
new_node->assign->score = score;
new_node->priv->id = id;
new_node->priv->name = uname;
new_node->priv->flags = pcmk__node_probes_allowed;
new_node->details->online = FALSE;
new_node->details->shutdown = FALSE;
new_node->details->running_rsc = NULL;
new_node->priv->scheduler = scheduler;
new_node->priv->variant = variant;
new_node->priv->attrs = pcmk__strkey_table(free, free);
new_node->priv->utilization = pcmk__strkey_table(free, free);
new_node->priv->digest_cache = pcmk__strkey_table(free, pe__free_digests);
if (pcmk__is_pacemaker_remote_node(new_node)) {
pcmk__insert_dup(new_node->priv->attrs, CRM_ATTR_KIND, "remote");
pcmk__set_scheduler_flags(scheduler, pcmk__sched_have_remote_nodes);
} else {
pcmk__insert_dup(new_node->priv->attrs, CRM_ATTR_KIND, "cluster");
}
scheduler->nodes = g_list_insert_sorted(scheduler->nodes, new_node,
pe__cmp_node_name);
return new_node;
}
static const char *
expand_remote_rsc_meta(xmlNode *xml_obj, xmlNode *parent, pcmk_scheduler_t *data)
{
xmlNode *attr_set = NULL;
xmlNode *attr = NULL;
const char *container_id = pcmk__xe_id(xml_obj);
const char *remote_name = NULL;
const char *remote_server = NULL;
const char *remote_port = NULL;
const char *connect_timeout = "60s";
const char *remote_allow_migrate=NULL;
const char *is_managed = NULL;
+ // @TODO This doesn't handle rules or id-ref
for (attr_set = pcmk__xe_first_child(xml_obj, PCMK_XE_META_ATTRIBUTES,
NULL, NULL);
attr_set != NULL;
attr_set = pcmk__xe_next(attr_set, PCMK_XE_META_ATTRIBUTES)) {
for (attr = pcmk__xe_first_child(attr_set, NULL, NULL, NULL);
attr != NULL; attr = pcmk__xe_next(attr, NULL)) {
const char *value = crm_element_value(attr, PCMK_XA_VALUE);
const char *name = crm_element_value(attr, PCMK_XA_NAME);
if (name == NULL) { // Sanity
continue;
}
if (strcmp(name, PCMK_META_REMOTE_NODE) == 0) {
remote_name = value;
} else if (strcmp(name, PCMK_META_REMOTE_ADDR) == 0) {
remote_server = value;
} else if (strcmp(name, PCMK_META_REMOTE_PORT) == 0) {
remote_port = value;
} else if (strcmp(name, PCMK_META_REMOTE_CONNECT_TIMEOUT) == 0) {
connect_timeout = value;
} else if (strcmp(name, PCMK_META_REMOTE_ALLOW_MIGRATE) == 0) {
remote_allow_migrate = value;
} else if (strcmp(name, PCMK_META_IS_MANAGED) == 0) {
is_managed = value;
}
}
}
if (remote_name == NULL) {
return NULL;
}
if (pe_find_resource(data->priv->resources, remote_name) != NULL) {
return NULL;
}
pe_create_remote_xml(parent, remote_name, container_id,
remote_allow_migrate, is_managed,
connect_timeout, remote_server, remote_port);
return remote_name;
}
static void
handle_startup_fencing(pcmk_scheduler_t *scheduler, pcmk_node_t *new_node)
{
if ((new_node->priv->variant == pcmk__node_variant_remote)
&& (new_node->priv->remote == NULL)) {
/* Ignore fencing for remote nodes that don't have a connection resource
* associated with them. This happens when remote node entries get left
* in the nodes section after the connection resource is removed.
*/
return;
}
if (pcmk_is_set(scheduler->flags, pcmk__sched_startup_fencing)) {
// All nodes are unclean until we've seen their status entry
new_node->details->unclean = TRUE;
} else {
// Blind faith ...
new_node->details->unclean = FALSE;
}
}
gboolean
unpack_nodes(xmlNode *xml_nodes, pcmk_scheduler_t *scheduler)
{
xmlNode *xml_obj = NULL;
pcmk_node_t *new_node = NULL;
const char *id = NULL;
const char *uname = NULL;
const char *type = NULL;
for (xml_obj = pcmk__xe_first_child(xml_nodes, PCMK_XE_NODE, NULL, NULL);
xml_obj != NULL; xml_obj = pcmk__xe_next(xml_obj, PCMK_XE_NODE)) {
int score = 0;
int rc = pcmk__xe_get_score(xml_obj, PCMK_XA_SCORE, &score, 0);
new_node = NULL;
id = crm_element_value(xml_obj, PCMK_XA_ID);
uname = crm_element_value(xml_obj, PCMK_XA_UNAME);
type = crm_element_value(xml_obj, PCMK_XA_TYPE);
crm_trace("Processing node %s/%s", uname, id);
if (id == NULL) {
pcmk__config_err("Ignoring <" PCMK_XE_NODE
"> entry in configuration without id");
continue;
}
if (rc != pcmk_rc_ok) {
// Not possible with schema validation enabled
pcmk__config_warn("Using 0 as score for node %s "
"because '%s' is not a valid score: %s",
pcmk__s(uname, "without name"),
crm_element_value(xml_obj, PCMK_XA_SCORE),
pcmk_rc_str(rc));
}
new_node = pe_create_node(id, uname, type, score, scheduler);
if (new_node == NULL) {
return FALSE;
}
handle_startup_fencing(scheduler, new_node);
add_node_attrs(xml_obj, new_node, FALSE, scheduler);
crm_trace("Done with node %s",
crm_element_value(xml_obj, PCMK_XA_UNAME));
}
return TRUE;
}
static void
unpack_launcher(pcmk_resource_t *rsc, pcmk_scheduler_t *scheduler)
{
const char *launcher_id = NULL;
if (rsc->priv->children != NULL) {
g_list_foreach(rsc->priv->children, (GFunc) unpack_launcher,
scheduler);
return;
}
launcher_id = g_hash_table_lookup(rsc->priv->meta, PCMK__META_CONTAINER);
if ((launcher_id != NULL)
&& !pcmk__str_eq(launcher_id, rsc->id, pcmk__str_none)) {
pcmk_resource_t *launcher = pe_find_resource(scheduler->priv->resources,
launcher_id);
if (launcher != NULL) {
rsc->priv->launcher = launcher;
launcher->priv->launched =
g_list_append(launcher->priv->launched, rsc);
pcmk__rsc_trace(rsc, "Resource %s's launcher is %s",
rsc->id, launcher_id);
} else {
pcmk__config_err("Resource %s: Unknown " PCMK__META_CONTAINER " %s",
rsc->id, launcher_id);
}
}
}
gboolean
unpack_remote_nodes(xmlNode *xml_resources, pcmk_scheduler_t *scheduler)
{
xmlNode *xml_obj = NULL;
/* Create remote nodes and guest nodes from the resource configuration
* before unpacking resources.
*/
for (xml_obj = pcmk__xe_first_child(xml_resources, NULL, NULL, NULL);
xml_obj != NULL; xml_obj = pcmk__xe_next(xml_obj, NULL)) {
const char *new_node_id = NULL;
/* Check for remote nodes, which are defined by ocf:pacemaker:remote
* primitives.
*/
if (xml_contains_remote_node(xml_obj)) {
new_node_id = pcmk__xe_id(xml_obj);
/* The pcmk_find_node() check ensures we don't iterate over an
* expanded node that has already been added to the node list
*/
if (new_node_id
&& (pcmk_find_node(scheduler, new_node_id) == NULL)) {
crm_trace("Found remote node %s defined by resource %s",
new_node_id, pcmk__xe_id(xml_obj));
pe_create_node(new_node_id, new_node_id, PCMK_VALUE_REMOTE,
0, scheduler);
}
continue;
}
/* Check for guest nodes, which are defined by special meta-attributes
* of a primitive of any type (for example, VirtualDomain or Xen).
*/
if (pcmk__xe_is(xml_obj, PCMK_XE_PRIMITIVE)) {
/* This will add an ocf:pacemaker:remote primitive to the
* configuration for the guest node's connection, to be unpacked
* later.
*/
new_node_id = expand_remote_rsc_meta(xml_obj, xml_resources,
scheduler);
if (new_node_id
&& (pcmk_find_node(scheduler, new_node_id) == NULL)) {
crm_trace("Found guest node %s in resource %s",
new_node_id, pcmk__xe_id(xml_obj));
pe_create_node(new_node_id, new_node_id, PCMK_VALUE_REMOTE,
0, scheduler);
}
continue;
}
/* Check for guest nodes inside a group. Clones are currently not
* supported as guest nodes.
*/
if (pcmk__xe_is(xml_obj, PCMK_XE_GROUP)) {
xmlNode *xml_obj2 = NULL;
for (xml_obj2 = pcmk__xe_first_child(xml_obj, NULL, NULL, NULL);
xml_obj2 != NULL; xml_obj2 = pcmk__xe_next(xml_obj2, NULL)) {
new_node_id = expand_remote_rsc_meta(xml_obj2, xml_resources,
scheduler);
if (new_node_id
&& (pcmk_find_node(scheduler, new_node_id) == NULL)) {
crm_trace("Found guest node %s in resource %s inside group %s",
new_node_id, pcmk__xe_id(xml_obj2),
pcmk__xe_id(xml_obj));
pe_create_node(new_node_id, new_node_id, PCMK_VALUE_REMOTE,
0, scheduler);
}
}
}
}
return TRUE;
}
/* Call this after all the nodes and resources have been
* unpacked, but before the status section is read.
*
* A remote node's online status is reflected by the state
* of the remote node's connection resource. We need to link
* the remote node to this connection resource so we can have
* easy access to the connection resource during the scheduler calculations.
*/
static void
link_rsc2remotenode(pcmk_scheduler_t *scheduler, pcmk_resource_t *new_rsc)
{
pcmk_node_t *remote_node = NULL;
if (!pcmk_is_set(new_rsc->flags, pcmk__rsc_is_remote_connection)) {
return;
}
if (pcmk_is_set(scheduler->flags, pcmk__sched_location_only)) {
/* remote_nodes and remote_resources are not linked in quick location calculations */
return;
}
remote_node = pcmk_find_node(scheduler, new_rsc->id);
CRM_CHECK(remote_node != NULL, return);
pcmk__rsc_trace(new_rsc, "Linking remote connection resource %s to %s",
new_rsc->id, pcmk__node_name(remote_node));
remote_node->priv->remote = new_rsc;
if (new_rsc->priv->launcher == NULL) {
/* Handle start-up fencing for remote nodes (as opposed to guest nodes)
* the same as is done for cluster nodes.
*/
handle_startup_fencing(scheduler, remote_node);
} else {
/* pe_create_node() marks the new node as "remote" or "cluster"; now
* that we know the node is a guest node, update it correctly.
*/
pcmk__insert_dup(remote_node->priv->attrs,
CRM_ATTR_KIND, "container");
}
}
/*!
* \internal
* \brief Parse configuration XML for resource information
*
* \param[in] xml_resources Top of resource configuration XML
* \param[in,out] scheduler Scheduler data
*
* \return TRUE
*
* \note unpack_remote_nodes() MUST be called before this, so that the nodes can
* be used when pe__unpack_resource() calls resource_location()
*/
gboolean
unpack_resources(const xmlNode *xml_resources, pcmk_scheduler_t *scheduler)
{
xmlNode *xml_obj = NULL;
GList *gIter = NULL;
scheduler->priv->templates = pcmk__strkey_table(free, pcmk__free_idref);
for (xml_obj = pcmk__xe_first_child(xml_resources, NULL, NULL, NULL);
xml_obj != NULL; xml_obj = pcmk__xe_next(xml_obj, NULL)) {
pcmk_resource_t *new_rsc = NULL;
const char *id = pcmk__xe_id(xml_obj);
if (pcmk__str_empty(id)) {
pcmk__config_err("Ignoring <%s> resource without ID",
xml_obj->name);
continue;
}
if (pcmk__xe_is(xml_obj, PCMK_XE_TEMPLATE)) {
if (g_hash_table_lookup_extended(scheduler->priv->templates, id,
NULL, NULL) == FALSE) {
/* Record the template's ID for the knowledge of its existence anyway. */
pcmk__insert_dup(scheduler->priv->templates, id, NULL);
}
continue;
}
crm_trace("Unpacking <%s " PCMK_XA_ID "='%s'>", xml_obj->name, id);
if (pe__unpack_resource(xml_obj, &new_rsc, NULL,
scheduler) == pcmk_rc_ok) {
scheduler->priv->resources =
g_list_append(scheduler->priv->resources, new_rsc);
pcmk__rsc_trace(new_rsc, "Added resource %s", new_rsc->id);
} else {
pcmk__config_err("Ignoring <%s> resource '%s' "
"because configuration is invalid",
xml_obj->name, id);
}
}
for (gIter = scheduler->priv->resources;
gIter != NULL; gIter = gIter->next) {
pcmk_resource_t *rsc = (pcmk_resource_t *) gIter->data;
unpack_launcher(rsc, scheduler);
link_rsc2remotenode(scheduler, rsc);
}
scheduler->priv->resources = g_list_sort(scheduler->priv->resources,
pe__cmp_rsc_priority);
if (pcmk_is_set(scheduler->flags, pcmk__sched_location_only)) {
/* Ignore */
} else if (pcmk_is_set(scheduler->flags, pcmk__sched_fencing_enabled)
&& !pcmk_is_set(scheduler->flags, pcmk__sched_have_fencing)) {
pcmk__config_err("Resource start-up disabled since no STONITH resources have been defined");
pcmk__config_err("Either configure some or disable STONITH with the "
PCMK_OPT_STONITH_ENABLED " option");
pcmk__config_err("NOTE: Clusters with shared data need STONITH to ensure data integrity");
}
return TRUE;
}
/*!
* \internal
* \brief Validate the levels in a fencing topology
*
* \param[in] xml \c PCMK_XE_FENCING_TOPOLOGY element
*/
void
pcmk__validate_fencing_topology(const xmlNode *xml)
{
if (xml == NULL) {
return;
}
CRM_CHECK(pcmk__xe_is(xml, PCMK_XE_FENCING_TOPOLOGY), return);
for (const xmlNode *level = pcmk__xe_first_child(xml, PCMK_XE_FENCING_LEVEL,
NULL, NULL);
level != NULL; level = pcmk__xe_next(level, PCMK_XE_FENCING_LEVEL)) {
const char *id = pcmk__xe_id(level);
int index = 0;
if (pcmk__str_empty(id)) {
pcmk__config_err("Ignoring fencing level without ID");
continue;
}
if (crm_element_value_int(level, PCMK_XA_INDEX, &index) != 0) {
pcmk__config_err("Ignoring fencing level %s with invalid index",
id);
continue;
}
if ((index < ST__LEVEL_MIN) || (index > ST__LEVEL_MAX)) {
pcmk__config_err("Ignoring fencing level %s with out-of-range "
"index %d",
id, index);
}
}
}
gboolean
unpack_tags(xmlNode *xml_tags, pcmk_scheduler_t *scheduler)
{
xmlNode *xml_tag = NULL;
scheduler->priv->tags = pcmk__strkey_table(free, pcmk__free_idref);
for (xml_tag = pcmk__xe_first_child(xml_tags, PCMK_XE_TAG, NULL, NULL);
xml_tag != NULL; xml_tag = pcmk__xe_next(xml_tag, PCMK_XE_TAG)) {
xmlNode *xml_obj_ref = NULL;
const char *tag_id = pcmk__xe_id(xml_tag);
if (tag_id == NULL) {
pcmk__config_err("Ignoring <%s> without " PCMK_XA_ID,
(const char *) xml_tag->name);
continue;
}
for (xml_obj_ref = pcmk__xe_first_child(xml_tag, PCMK_XE_OBJ_REF,
NULL, NULL);
xml_obj_ref != NULL;
xml_obj_ref = pcmk__xe_next(xml_obj_ref, PCMK_XE_OBJ_REF)) {
const char *obj_ref = pcmk__xe_id(xml_obj_ref);
if (obj_ref == NULL) {
pcmk__config_err("Ignoring <%s> for tag '%s' without " PCMK_XA_ID,
xml_obj_ref->name, tag_id);
continue;
}
pcmk__add_idref(scheduler->priv->tags, tag_id, obj_ref);
}
}
return TRUE;
}
/*!
* \internal
* \brief Unpack a ticket state entry
*
* \param[in] xml_ticket XML ticket state to unpack
* \param[in,out] userdata Scheduler data
*
* \return pcmk_rc_ok (to always continue unpacking further entries)
*/
static int
unpack_ticket_state(xmlNode *xml_ticket, void *userdata)
{
pcmk_scheduler_t *scheduler = userdata;
const char *ticket_id = NULL;
const char *granted = NULL;
const char *last_granted = NULL;
const char *standby = NULL;
xmlAttrPtr xIter = NULL;
pcmk__ticket_t *ticket = NULL;
ticket_id = pcmk__xe_id(xml_ticket);
if (pcmk__str_empty(ticket_id)) {
return pcmk_rc_ok;
}
crm_trace("Processing ticket state for %s", ticket_id);
ticket = g_hash_table_lookup(scheduler->priv->ticket_constraints,
ticket_id);
if (ticket == NULL) {
ticket = ticket_new(ticket_id, scheduler);
if (ticket == NULL) {
return pcmk_rc_ok;
}
}
for (xIter = xml_ticket->properties; xIter; xIter = xIter->next) {
const char *prop_name = (const char *)xIter->name;
const char *prop_value = pcmk__xml_attr_value(xIter);
if (pcmk__str_eq(prop_name, PCMK_XA_ID, pcmk__str_none)) {
continue;
}
pcmk__insert_dup(ticket->state, prop_name, prop_value);
}
granted = g_hash_table_lookup(ticket->state, PCMK__XA_GRANTED);
if (granted && crm_is_true(granted)) {
pcmk__set_ticket_flags(ticket, pcmk__ticket_granted);
crm_info("We have ticket '%s'", ticket->id);
} else {
pcmk__clear_ticket_flags(ticket, pcmk__ticket_granted);
crm_info("We do not have ticket '%s'", ticket->id);
}
last_granted = g_hash_table_lookup(ticket->state, PCMK_XA_LAST_GRANTED);
if (last_granted) {
long long last_granted_ll = 0LL;
int rc = pcmk__scan_ll(last_granted, &last_granted_ll, 0LL);
if (rc != pcmk_rc_ok) {
crm_warn("Using %lld instead of invalid " PCMK_XA_LAST_GRANTED
" value '%s' in state for ticket %s: %s",
last_granted_ll, last_granted, ticket->id,
pcmk_rc_str(rc));
}
ticket->last_granted = (time_t) last_granted_ll;
}
standby = g_hash_table_lookup(ticket->state, PCMK_XA_STANDBY);
if (standby && crm_is_true(standby)) {
pcmk__set_ticket_flags(ticket, pcmk__ticket_standby);
if (pcmk_is_set(ticket->flags, pcmk__ticket_granted)) {
crm_info("Granted ticket '%s' is in standby-mode", ticket->id);
}
} else {
pcmk__clear_ticket_flags(ticket, pcmk__ticket_standby);
}
crm_trace("Done with ticket state for %s", ticket_id);
return pcmk_rc_ok;
}
static void
unpack_handle_remote_attrs(pcmk_node_t *this_node, const xmlNode *state,
pcmk_scheduler_t *scheduler)
{
const char *discovery = NULL;
const xmlNode *attrs = NULL;
pcmk_resource_t *rsc = NULL;
int maint = 0;
if (!pcmk__xe_is(state, PCMK__XE_NODE_STATE)) {
return;
}
if ((this_node == NULL) || !pcmk__is_pacemaker_remote_node(this_node)) {
return;
}
crm_trace("Processing Pacemaker Remote node %s",
pcmk__node_name(this_node));
pcmk__scan_min_int(crm_element_value(state, PCMK__XA_NODE_IN_MAINTENANCE),
&maint, 0);
if (maint) {
pcmk__set_node_flags(this_node, pcmk__node_remote_maint);
} else {
pcmk__clear_node_flags(this_node, pcmk__node_remote_maint);
}
rsc = this_node->priv->remote;
if (!pcmk_is_set(this_node->priv->flags, pcmk__node_remote_reset)) {
this_node->details->unclean = FALSE;
pcmk__set_node_flags(this_node, pcmk__node_seen);
}
attrs = pcmk__xe_first_child(state, PCMK__XE_TRANSIENT_ATTRIBUTES, NULL,
NULL);
add_node_attrs(attrs, this_node, TRUE, scheduler);
if (pe__shutdown_requested(this_node)) {
crm_info("%s is shutting down", pcmk__node_name(this_node));
this_node->details->shutdown = TRUE;
}
if (crm_is_true(pcmk__node_attr(this_node, PCMK_NODE_ATTR_STANDBY, NULL,
pcmk__rsc_node_current))) {
crm_info("%s is in standby mode", pcmk__node_name(this_node));
pcmk__set_node_flags(this_node, pcmk__node_standby);
}
if (crm_is_true(pcmk__node_attr(this_node, PCMK_NODE_ATTR_MAINTENANCE, NULL,
pcmk__rsc_node_current))
|| ((rsc != NULL) && !pcmk_is_set(rsc->flags, pcmk__rsc_managed))) {
crm_info("%s is in maintenance mode", pcmk__node_name(this_node));
this_node->details->maintenance = TRUE;
}
discovery = pcmk__node_attr(this_node,
PCMK__NODE_ATTR_RESOURCE_DISCOVERY_ENABLED,
NULL, pcmk__rsc_node_current);
if ((discovery != NULL) && !crm_is_true(discovery)) {
pcmk__warn_once(pcmk__wo_rdisc_enabled,
"Support for the "
PCMK__NODE_ATTR_RESOURCE_DISCOVERY_ENABLED
" node attribute is deprecated and will be removed"
" (and behave as 'true') in a future release.");
if (pcmk__is_remote_node(this_node)
&& !pcmk_is_set(scheduler->flags, pcmk__sched_fencing_enabled)) {
pcmk__config_warn("Ignoring "
PCMK__NODE_ATTR_RESOURCE_DISCOVERY_ENABLED
" attribute on Pacemaker Remote node %s"
" because fencing is disabled",
pcmk__node_name(this_node));
} else {
/* This is either a remote node with fencing enabled, or a guest
* node. We don't care whether fencing is enabled when fencing guest
* nodes, because they are "fenced" by recovering their containing
* resource.
*/
crm_info("%s has resource discovery disabled",
pcmk__node_name(this_node));
pcmk__clear_node_flags(this_node, pcmk__node_probes_allowed);
}
}
}
/*!
* \internal
* \brief Unpack a cluster node's transient attributes
*
* \param[in] state CIB node state XML
* \param[in,out] node Cluster node whose attributes are being unpacked
* \param[in,out] scheduler Scheduler data
*/
static void
unpack_transient_attributes(const xmlNode *state, pcmk_node_t *node,
pcmk_scheduler_t *scheduler)
{
const char *discovery = NULL;
const xmlNode *attrs = pcmk__xe_first_child(state,
PCMK__XE_TRANSIENT_ATTRIBUTES,
NULL, NULL);
add_node_attrs(attrs, node, TRUE, scheduler);
if (crm_is_true(pcmk__node_attr(node, PCMK_NODE_ATTR_STANDBY, NULL,
pcmk__rsc_node_current))) {
crm_info("%s is in standby mode", pcmk__node_name(node));
pcmk__set_node_flags(node, pcmk__node_standby);
}
if (crm_is_true(pcmk__node_attr(node, PCMK_NODE_ATTR_MAINTENANCE, NULL,
pcmk__rsc_node_current))) {
crm_info("%s is in maintenance mode", pcmk__node_name(node));
node->details->maintenance = TRUE;
}
discovery = pcmk__node_attr(node,
PCMK__NODE_ATTR_RESOURCE_DISCOVERY_ENABLED,
NULL, pcmk__rsc_node_current);
if ((discovery != NULL) && !crm_is_true(discovery)) {
pcmk__config_warn("Ignoring "
PCMK__NODE_ATTR_RESOURCE_DISCOVERY_ENABLED
" attribute for %s because disabling resource"
" discovery is not allowed for cluster nodes",
pcmk__node_name(node));
}
}
/*!
* \internal
* \brief Unpack a node state entry (first pass)
*
* Unpack one node state entry from status. This unpacks information from the
* \C PCMK__XE_NODE_STATE element itself and node attributes inside it, but not
* the resource history inside it. Multiple passes through the status are needed
* to fully unpack everything.
*
* \param[in] state CIB node state XML
* \param[in,out] scheduler Scheduler data
*/
static void
unpack_node_state(const xmlNode *state, pcmk_scheduler_t *scheduler)
{
const char *id = NULL;
const char *uname = NULL;
pcmk_node_t *this_node = NULL;
id = crm_element_value(state, PCMK_XA_ID);
if (id == NULL) {
pcmk__config_err("Ignoring invalid " PCMK__XE_NODE_STATE " entry without "
PCMK_XA_ID);
crm_log_xml_info(state, "missing-id");
return;
}
uname = crm_element_value(state, PCMK_XA_UNAME);
if (uname == NULL) {
/* If a joining peer makes the cluster acquire the quorum from Corosync
* but has not joined the controller CPG membership yet, it's possible
* that the created PCMK__XE_NODE_STATE entry doesn't have a
* PCMK_XA_UNAME yet. Recognize the node as pending and wait for it to
* join CPG.
*/
crm_trace("Handling " PCMK__XE_NODE_STATE " entry with id=\"%s\" "
"without " PCMK_XA_UNAME,
id);
}
this_node = pe_find_node_any(scheduler->nodes, id, uname);
if (this_node == NULL) {
crm_notice("Ignoring recorded state for removed node with name %s and "
PCMK_XA_ID " %s", pcmk__s(uname, "unknown"), id);
return;
}
if (pcmk__is_pacemaker_remote_node(this_node)) {
int remote_fenced = 0;
/* We can't determine the online status of Pacemaker Remote nodes until
* after all resource history has been unpacked. In this first pass, we
* do need to mark whether the node has been fenced, as this plays a
* role during unpacking cluster node resource state.
*/
pcmk__scan_min_int(crm_element_value(state, PCMK__XA_NODE_FENCED),
&remote_fenced, 0);
if (remote_fenced) {
pcmk__set_node_flags(this_node, pcmk__node_remote_fenced);
} else {
pcmk__clear_node_flags(this_node, pcmk__node_remote_fenced);
}
return;
}
unpack_transient_attributes(state, this_node, scheduler);
/* Provisionally mark this cluster node as clean. We have at least seen it
* in the current cluster's lifetime.
*/
this_node->details->unclean = FALSE;
pcmk__set_node_flags(this_node, pcmk__node_seen);
crm_trace("Determining online status of cluster node %s (id %s)",
pcmk__node_name(this_node), id);
determine_online_status(state, this_node, scheduler);
if (!pcmk_is_set(scheduler->flags, pcmk__sched_quorate)
&& this_node->details->online
&& (scheduler->no_quorum_policy == pcmk_no_quorum_fence)) {
/* Everything else should flow from this automatically
* (at least until the scheduler becomes able to migrate off
* healthy resources)
*/
pe_fence_node(scheduler, this_node, "cluster does not have quorum",
FALSE);
}
}
/*!
* \internal
* \brief Unpack nodes' resource history as much as possible
*
* Unpack as many nodes' resource history as possible in one pass through the
* status. We need to process Pacemaker Remote nodes' connections/containers
* before unpacking their history; the connection/container history will be
* in another node's history, so it might take multiple passes to unpack
* everything.
*
* \param[in] status CIB XML status section
* \param[in] fence If true, treat any not-yet-unpacked nodes as unseen
* \param[in,out] scheduler Scheduler data
*
* \return Standard Pacemaker return code (specifically pcmk_rc_ok if done,
* or EAGAIN if more unpacking remains to be done)
*/
static int
unpack_node_history(const xmlNode *status, bool fence,
pcmk_scheduler_t *scheduler)
{
int rc = pcmk_rc_ok;
// Loop through all PCMK__XE_NODE_STATE entries in CIB status
for (const xmlNode *state = pcmk__xe_first_child(status,
PCMK__XE_NODE_STATE, NULL,
NULL);
state != NULL; state = pcmk__xe_next(state, PCMK__XE_NODE_STATE)) {
const char *id = pcmk__xe_id(state);
const char *uname = crm_element_value(state, PCMK_XA_UNAME);
pcmk_node_t *this_node = NULL;
if ((id == NULL) || (uname == NULL)) {
// Warning already logged in first pass through status section
crm_trace("Not unpacking resource history from malformed "
PCMK__XE_NODE_STATE " without id and/or uname");
continue;
}
this_node = pe_find_node_any(scheduler->nodes, id, uname);
if (this_node == NULL) {
// Warning already logged in first pass through status section
crm_trace("Not unpacking resource history for node %s because "
"no longer in configuration", id);
continue;
}
if (pcmk_is_set(this_node->priv->flags, pcmk__node_unpacked)) {
crm_trace("Not unpacking resource history for node %s because "
"already unpacked", id);
continue;
}
if (fence) {
// We're processing all remaining nodes
} else if (pcmk__is_guest_or_bundle_node(this_node)) {
/* We can unpack a guest node's history only after we've unpacked
* other resource history to the point that we know that the node's
* connection and containing resource are both up.
*/
const pcmk_resource_t *remote = this_node->priv->remote;
const pcmk_resource_t *launcher = remote->priv->launcher;
if ((remote->priv->orig_role != pcmk_role_started)
|| (launcher->priv->orig_role != pcmk_role_started)) {
crm_trace("Not unpacking resource history for guest node %s "
"because launcher and connection are not known to "
"be up", id);
continue;
}
} else if (pcmk__is_remote_node(this_node)) {
/* We can unpack a remote node's history only after we've unpacked
* other resource history to the point that we know that the node's
* connection is up, with the exception of when shutdown locks are
* in use.
*/
pcmk_resource_t *rsc = this_node->priv->remote;
if ((rsc == NULL)
|| (!pcmk_is_set(scheduler->flags, pcmk__sched_shutdown_lock)
&& (rsc->priv->orig_role != pcmk_role_started))) {
crm_trace("Not unpacking resource history for remote node %s "
"because connection is not known to be up", id);
continue;
}
/* If fencing and shutdown locks are disabled and we're not processing
* unseen nodes, then we don't want to unpack offline nodes until online
* nodes have been unpacked. This allows us to number active clone
* instances first.
*/
} else if (!pcmk_any_flags_set(scheduler->flags,
pcmk__sched_fencing_enabled
|pcmk__sched_shutdown_lock)
&& !this_node->details->online) {
crm_trace("Not unpacking resource history for offline "
"cluster node %s", id);
continue;
}
if (pcmk__is_pacemaker_remote_node(this_node)) {
determine_remote_online_status(scheduler, this_node);
unpack_handle_remote_attrs(this_node, state, scheduler);
}
crm_trace("Unpacking resource history for %snode %s",
(fence? "unseen " : ""), id);
pcmk__set_node_flags(this_node, pcmk__node_unpacked);
unpack_node_lrm(this_node, state, scheduler);
rc = EAGAIN; // Other node histories might depend on this one
}
return rc;
}
/* remove nodes that are down, stopping */
/* create positive rsc_to_node constraints between resources and the nodes they are running on */
/* anything else? */
gboolean
unpack_status(xmlNode *status, pcmk_scheduler_t *scheduler)
{
xmlNode *state = NULL;
crm_trace("Beginning unpack");
if (scheduler->priv->ticket_constraints == NULL) {
scheduler->priv->ticket_constraints =
pcmk__strkey_table(free, destroy_ticket);
}
for (state = pcmk__xe_first_child(status, NULL, NULL, NULL); state != NULL;
state = pcmk__xe_next(state, NULL)) {
if (pcmk__xe_is(state, PCMK_XE_TICKETS)) {
pcmk__xe_foreach_child(state, PCMK__XE_TICKET_STATE,
unpack_ticket_state, scheduler);
} else if (pcmk__xe_is(state, PCMK__XE_NODE_STATE)) {
unpack_node_state(state, scheduler);
}
}
while (unpack_node_history(status, FALSE, scheduler) == EAGAIN) {
crm_trace("Another pass through node resource histories is needed");
}
// Now catch any nodes we didn't see
unpack_node_history(status,
pcmk_is_set(scheduler->flags,
pcmk__sched_fencing_enabled),
scheduler);
/* Now that we know where resources are, we can schedule stops of containers
* with failed bundle connections
*/
if (scheduler->priv->stop_needed != NULL) {
for (GList *item = scheduler->priv->stop_needed;
item != NULL; item = item->next) {
pcmk_resource_t *container = item->data;
pcmk_node_t *node = pcmk__current_node(container);
if (node) {
stop_action(container, node, FALSE);
}
}
g_list_free(scheduler->priv->stop_needed);
scheduler->priv->stop_needed = NULL;
}
/* Now that we know status of all Pacemaker Remote connections and nodes,
* we can stop connections for node shutdowns, and check the online status
* of remote/guest nodes that didn't have any node history to unpack.
*/
for (GList *gIter = scheduler->nodes; gIter != NULL; gIter = gIter->next) {
pcmk_node_t *this_node = gIter->data;
if (!pcmk__is_pacemaker_remote_node(this_node)) {
continue;
}
if (this_node->details->shutdown
&& (this_node->priv->remote != NULL)) {
pe__set_next_role(this_node->priv->remote, pcmk_role_stopped,
"remote shutdown");
}
if (!pcmk_is_set(this_node->priv->flags, pcmk__node_unpacked)) {
determine_remote_online_status(scheduler, this_node);
}
}
return TRUE;
}
/*!
* \internal
* \brief Unpack node's time when it became a member at the cluster layer
*
* \param[in] node_state Node's \c PCMK__XE_NODE_STATE entry
* \param[in,out] scheduler Scheduler data
*
* \return Epoch time when node became a cluster member
* (or scheduler effective time for legacy entries) if a member,
* 0 if not a member, or -1 if no valid information available
*/
static long long
unpack_node_member(const xmlNode *node_state, pcmk_scheduler_t *scheduler)
{
const char *member_time = crm_element_value(node_state, PCMK__XA_IN_CCM);
int member = 0;
if (member_time == NULL) {
return -1LL;
} else if (crm_str_to_boolean(member_time, &member) == 1) {
/* If in_ccm=0, we'll return 0 here. If in_ccm=1, either the entry was
* recorded as a boolean for a DC < 2.1.7, or the node is pending
* shutdown and has left the CPG, in which case it was set to 1 to avoid
* fencing for PCMK_OPT_NODE_PENDING_TIMEOUT.
*
* We return the effective time for in_ccm=1 because what's important to
* avoid fencing is that effective time minus this value is less than
* the pending node timeout.
*/
return member? (long long) get_effective_time(scheduler) : 0LL;
} else {
long long when_member = 0LL;
if ((pcmk__scan_ll(member_time, &when_member,
0LL) != pcmk_rc_ok) || (when_member < 0LL)) {
crm_warn("Unrecognized value '%s' for " PCMK__XA_IN_CCM
" in " PCMK__XE_NODE_STATE " entry", member_time);
return -1LL;
}
return when_member;
}
}
/*!
* \internal
* \brief Unpack node's time when it became online in process group
*
* \param[in] node_state Node's \c PCMK__XE_NODE_STATE entry
*
* \return Epoch time when node became online in process group (or 0 if not
* online, or 1 for legacy online entries)
*/
static long long
unpack_node_online(const xmlNode *node_state)
{
const char *peer_time = crm_element_value(node_state, PCMK_XA_CRMD);
// @COMPAT Entries recorded for DCs < 2.1.7 have "online" or "offline"
if (pcmk__str_eq(peer_time, PCMK_VALUE_OFFLINE,
pcmk__str_casei|pcmk__str_null_matches)) {
return 0LL;
} else if (pcmk__str_eq(peer_time, PCMK_VALUE_ONLINE, pcmk__str_casei)) {
return 1LL;
} else {
long long when_online = 0LL;
if ((pcmk__scan_ll(peer_time, &when_online, 0LL) != pcmk_rc_ok)
|| (when_online < 0)) {
crm_warn("Unrecognized value '%s' for " PCMK_XA_CRMD " in "
PCMK__XE_NODE_STATE " entry, assuming offline", peer_time);
return 0LL;
}
return when_online;
}
}
/*!
* \internal
* \brief Unpack node attribute for user-requested fencing
*
* \param[in] node Node to check
* \param[in] node_state Node's \c PCMK__XE_NODE_STATE entry in CIB status
*
* \return \c true if fencing has been requested for \p node, otherwise \c false
*/
static bool
unpack_node_terminate(const pcmk_node_t *node, const xmlNode *node_state)
{
long long value = 0LL;
int value_i = 0;
int rc = pcmk_rc_ok;
const char *value_s = pcmk__node_attr(node, PCMK_NODE_ATTR_TERMINATE,
NULL, pcmk__rsc_node_current);
// Value may be boolean or an epoch time
if (crm_str_to_boolean(value_s, &value_i) == 1) {
return (value_i != 0);
}
rc = pcmk__scan_ll(value_s, &value, 0LL);
if (rc == pcmk_rc_ok) {
return (value > 0);
}
crm_warn("Ignoring unrecognized value '%s' for " PCMK_NODE_ATTR_TERMINATE
"node attribute for %s: %s",
value_s, pcmk__node_name(node), pcmk_rc_str(rc));
return false;
}
static gboolean
determine_online_status_no_fencing(pcmk_scheduler_t *scheduler,
const xmlNode *node_state,
pcmk_node_t *this_node)
{
gboolean online = FALSE;
const char *join = crm_element_value(node_state, PCMK__XA_JOIN);
const char *exp_state = crm_element_value(node_state, PCMK_XA_EXPECTED);
long long when_member = unpack_node_member(node_state, scheduler);
long long when_online = unpack_node_online(node_state);
if (when_member <= 0) {
crm_trace("Node %s is %sdown", pcmk__node_name(this_node),
((when_member < 0)? "presumed " : ""));
} else if (when_online > 0) {
if (pcmk__str_eq(join, CRMD_JOINSTATE_MEMBER, pcmk__str_casei)) {
online = TRUE;
} else {
crm_debug("Node %s is not ready to run resources: %s",
pcmk__node_name(this_node), join);
}
} else if (!pcmk_is_set(this_node->priv->flags,
pcmk__node_expected_up)) {
crm_trace("Node %s controller is down: "
"member@%lld online@%lld join=%s expected=%s",
pcmk__node_name(this_node), when_member, when_online,
pcmk__s(join, "<null>"), pcmk__s(exp_state, "<null>"));
} else {
/* mark it unclean */
pe_fence_node(scheduler, this_node, "peer is unexpectedly down", FALSE);
crm_info("Node %s member@%lld online@%lld join=%s expected=%s",
pcmk__node_name(this_node), when_member, when_online,
pcmk__s(join, "<null>"), pcmk__s(exp_state, "<null>"));
}
return online;
}
/*!
* \internal
* \brief Check whether a node has taken too long to join controller group
*
* \param[in,out] scheduler Scheduler data
* \param[in] node Node to check
* \param[in] when_member Epoch time when node became a cluster member
* \param[in] when_online Epoch time when node joined controller group
*
* \return true if node has been pending (on the way up) longer than
* \c PCMK_OPT_NODE_PENDING_TIMEOUT, otherwise false
* \note This will also update the cluster's recheck time if appropriate.
*/
static inline bool
pending_too_long(pcmk_scheduler_t *scheduler, const pcmk_node_t *node,
long long when_member, long long when_online)
{
if ((scheduler->priv->node_pending_ms > 0U)
&& (when_member > 0) && (when_online <= 0)) {
// There is a timeout on pending nodes, and node is pending
time_t timeout = when_member
+ pcmk__timeout_ms2s(scheduler->priv->node_pending_ms);
if (get_effective_time(node->priv->scheduler) >= timeout) {
return true; // Node has timed out
}
// Node is pending, but still has time
pe__update_recheck_time(timeout, scheduler, "pending node timeout");
}
return false;
}
static bool
determine_online_status_fencing(pcmk_scheduler_t *scheduler,
const xmlNode *node_state,
pcmk_node_t *this_node)
{
bool termination_requested = unpack_node_terminate(this_node, node_state);
const char *join = crm_element_value(node_state, PCMK__XA_JOIN);
const char *exp_state = crm_element_value(node_state, PCMK_XA_EXPECTED);
long long when_member = unpack_node_member(node_state, scheduler);
long long when_online = unpack_node_online(node_state);
/*
- PCMK__XA_JOIN ::= member|down|pending|banned
- PCMK_XA_EXPECTED ::= member|down
@COMPAT with entries recorded for DCs < 2.1.7
- PCMK__XA_IN_CCM ::= true|false
- PCMK_XA_CRMD ::= online|offline
Since crm_feature_set 3.18.0 (pacemaker-2.1.7):
- PCMK__XA_IN_CCM ::= <timestamp>|0
Since when node has been a cluster member. A value 0 of means the node is not
a cluster member.
- PCMK_XA_CRMD ::= <timestamp>|0
Since when peer has been online in CPG. A value 0 means the peer is offline
in CPG.
*/
crm_trace("Node %s member@%lld online@%lld join=%s expected=%s%s",
pcmk__node_name(this_node), when_member, when_online,
pcmk__s(join, "<null>"), pcmk__s(exp_state, "<null>"),
(termination_requested? " (termination requested)" : ""));
if (this_node->details->shutdown) {
crm_debug("%s is shutting down", pcmk__node_name(this_node));
/* Slightly different criteria since we can't shut down a dead peer */
return (when_online > 0);
}
if (when_member < 0) {
pe_fence_node(scheduler, this_node,
"peer has not been seen by the cluster", FALSE);
return false;
}
if (pcmk__str_eq(join, CRMD_JOINSTATE_NACK, pcmk__str_none)) {
pe_fence_node(scheduler, this_node,
"peer failed Pacemaker membership criteria", FALSE);
} else if (termination_requested) {
if ((when_member <= 0) && (when_online <= 0)
&& pcmk__str_eq(join, CRMD_JOINSTATE_DOWN, pcmk__str_none)) {
crm_info("%s was fenced as requested", pcmk__node_name(this_node));
return false;
}
pe_fence_node(scheduler, this_node, "fencing was requested", false);
} else if (pcmk__str_eq(exp_state, CRMD_JOINSTATE_DOWN,
pcmk__str_null_matches)) {
if (pending_too_long(scheduler, this_node, when_member, when_online)) {
pe_fence_node(scheduler, this_node,
"peer pending timed out on joining the process group",
FALSE);
} else if ((when_member > 0) || (when_online > 0)) {
crm_info("- %s is not ready to run resources",
pcmk__node_name(this_node));
pcmk__set_node_flags(this_node, pcmk__node_standby);
this_node->details->pending = TRUE;
} else {
crm_trace("%s is down or still coming up",
pcmk__node_name(this_node));
}
} else if (when_member <= 0) {
// Consider PCMK_OPT_PRIORITY_FENCING_DELAY for lost nodes
pe_fence_node(scheduler, this_node,
"peer is no longer part of the cluster", TRUE);
} else if (when_online <= 0) {
pe_fence_node(scheduler, this_node,
"peer process is no longer available", FALSE);
/* Everything is running at this point, now check join state */
} else if (pcmk__str_eq(join, CRMD_JOINSTATE_MEMBER, pcmk__str_none)) {
crm_info("%s is active", pcmk__node_name(this_node));
} else if (pcmk__str_any_of(join, CRMD_JOINSTATE_PENDING,
CRMD_JOINSTATE_DOWN, NULL)) {
crm_info("%s is not ready to run resources",
pcmk__node_name(this_node));
pcmk__set_node_flags(this_node, pcmk__node_standby);
this_node->details->pending = TRUE;
} else {
pe_fence_node(scheduler, this_node, "peer was in an unknown state",
FALSE);
}
return (when_member > 0);
}
static void
determine_remote_online_status(pcmk_scheduler_t *scheduler,
pcmk_node_t *this_node)
{
pcmk_resource_t *rsc = this_node->priv->remote;
pcmk_resource_t *launcher = NULL;
pcmk_node_t *host = NULL;
const char *node_type = "Remote";
if (rsc == NULL) {
/* This is a leftover node state entry for a former Pacemaker Remote
* node whose connection resource was removed. Consider it offline.
*/
crm_trace("Pacemaker Remote node %s is considered OFFLINE because "
"its connection resource has been removed from the CIB",
this_node->priv->id);
this_node->details->online = FALSE;
return;
}
launcher = rsc->priv->launcher;
if (launcher != NULL) {
node_type = "Guest";
if (pcmk__list_of_1(rsc->priv->active_nodes)) {
host = rsc->priv->active_nodes->data;
}
}
/* If the resource is currently started, mark it online. */
if (rsc->priv->orig_role == pcmk_role_started) {
this_node->details->online = TRUE;
}
/* consider this node shutting down if transitioning start->stop */
if ((rsc->priv->orig_role == pcmk_role_started)
&& (rsc->priv->next_role == pcmk_role_stopped)) {
crm_trace("%s node %s shutting down because connection resource is stopping",
node_type, this_node->priv->id);
this_node->details->shutdown = TRUE;
}
/* Now check all the failure conditions. */
if ((launcher != NULL) && pcmk_is_set(launcher->flags, pcmk__rsc_failed)) {
crm_trace("Guest node %s UNCLEAN because guest resource failed",
this_node->priv->id);
this_node->details->online = FALSE;
pcmk__set_node_flags(this_node, pcmk__node_remote_reset);
} else if (pcmk_is_set(rsc->flags, pcmk__rsc_failed)) {
crm_trace("%s node %s OFFLINE because connection resource failed",
node_type, this_node->priv->id);
this_node->details->online = FALSE;
} else if ((rsc->priv->orig_role == pcmk_role_stopped)
|| ((launcher != NULL)
&& (launcher->priv->orig_role == pcmk_role_stopped))) {
crm_trace("%s node %s OFFLINE because its resource is stopped",
node_type, this_node->priv->id);
this_node->details->online = FALSE;
pcmk__clear_node_flags(this_node, pcmk__node_remote_reset);
} else if (host && (host->details->online == FALSE)
&& host->details->unclean) {
crm_trace("Guest node %s UNCLEAN because host is unclean",
this_node->priv->id);
this_node->details->online = FALSE;
pcmk__set_node_flags(this_node, pcmk__node_remote_reset);
} else {
crm_trace("%s node %s is %s",
node_type, this_node->priv->id,
this_node->details->online? "ONLINE" : "OFFLINE");
}
}
static void
determine_online_status(const xmlNode *node_state, pcmk_node_t *this_node,
pcmk_scheduler_t *scheduler)
{
gboolean online = FALSE;
const char *exp_state = crm_element_value(node_state, PCMK_XA_EXPECTED);
CRM_CHECK(this_node != NULL, return);
this_node->details->shutdown = FALSE;
if (pe__shutdown_requested(this_node)) {
this_node->details->shutdown = TRUE;
} else if (pcmk__str_eq(exp_state, CRMD_JOINSTATE_MEMBER, pcmk__str_casei)) {
pcmk__set_node_flags(this_node, pcmk__node_expected_up);
}
if (!pcmk_is_set(scheduler->flags, pcmk__sched_fencing_enabled)) {
online = determine_online_status_no_fencing(scheduler, node_state,
this_node);
} else {
online = determine_online_status_fencing(scheduler, node_state,
this_node);
}
if (online) {
this_node->details->online = TRUE;
} else {
/* remove node from contention */
this_node->assign->score = -PCMK_SCORE_INFINITY;
}
if (online && this_node->details->shutdown) {
/* don't run resources here */
this_node->assign->score = -PCMK_SCORE_INFINITY;
}
if (this_node->details->unclean) {
pcmk__sched_warn(scheduler, "%s is unclean",
pcmk__node_name(this_node));
} else if (!this_node->details->online) {
crm_trace("%s is offline", pcmk__node_name(this_node));
} else if (this_node->details->shutdown) {
crm_info("%s is shutting down", pcmk__node_name(this_node));
} else if (this_node->details->pending) {
crm_info("%s is pending", pcmk__node_name(this_node));
} else if (pcmk_is_set(this_node->priv->flags, pcmk__node_standby)) {
crm_info("%s is in standby", pcmk__node_name(this_node));
} else if (this_node->details->maintenance) {
crm_info("%s is in maintenance", pcmk__node_name(this_node));
} else {
crm_info("%s is online", pcmk__node_name(this_node));
}
}
/*!
* \internal
* \brief Find the end of a resource's name, excluding any clone suffix
*
* \param[in] id Resource ID to check
*
* \return Pointer to last character of resource's base name
*/
const char *
pe_base_name_end(const char *id)
{
if (!pcmk__str_empty(id)) {
const char *end = id + strlen(id) - 1;
for (const char *s = end; s > id; --s) {
switch (*s) {
case '0':
case '1':
case '2':
case '3':
case '4':
case '5':
case '6':
case '7':
case '8':
case '9':
break;
case ':':
return (s == end)? s : (s - 1);
default:
return end;
}
}
return end;
}
return NULL;
}
/*!
* \internal
* \brief Get a resource name excluding any clone suffix
*
* \param[in] last_rsc_id Resource ID to check
*
* \return Pointer to newly allocated string with resource's base name
* \note It is the caller's responsibility to free() the result.
* This asserts on error, so callers can assume result is not NULL.
*/
char *
clone_strip(const char *last_rsc_id)
{
const char *end = pe_base_name_end(last_rsc_id);
char *basename = NULL;
pcmk__assert(end != NULL);
basename = strndup(last_rsc_id, end - last_rsc_id + 1);
pcmk__assert(basename != NULL);
return basename;
}
/*!
* \internal
* \brief Get the name of the first instance of a cloned resource
*
* \param[in] last_rsc_id Resource ID to check
*
* \return Pointer to newly allocated string with resource's base name plus :0
* \note It is the caller's responsibility to free() the result.
* This asserts on error, so callers can assume result is not NULL.
*/
char *
clone_zero(const char *last_rsc_id)
{
const char *end = pe_base_name_end(last_rsc_id);
size_t base_name_len = end - last_rsc_id + 1;
char *zero = NULL;
pcmk__assert(end != NULL);
zero = pcmk__assert_alloc(base_name_len + 3, sizeof(char));
memcpy(zero, last_rsc_id, base_name_len);
zero[base_name_len] = ':';
zero[base_name_len + 1] = '0';
return zero;
}
static pcmk_resource_t *
create_fake_resource(const char *rsc_id, const xmlNode *rsc_entry,
pcmk_scheduler_t *scheduler)
{
pcmk_resource_t *rsc = NULL;
xmlNode *xml_rsc = pcmk__xe_create(NULL, PCMK_XE_PRIMITIVE);
pcmk__xe_copy_attrs(xml_rsc, rsc_entry, pcmk__xaf_none);
crm_xml_add(xml_rsc, PCMK_XA_ID, rsc_id);
crm_log_xml_debug(xml_rsc, "Orphan resource");
if (pe__unpack_resource(xml_rsc, &rsc, NULL, scheduler) != pcmk_rc_ok) {
return NULL;
}
if (xml_contains_remote_node(xml_rsc)) {
pcmk_node_t *node;
crm_debug("Detected orphaned remote node %s", rsc_id);
node = pcmk_find_node(scheduler, rsc_id);
if (node == NULL) {
node = pe_create_node(rsc_id, rsc_id, PCMK_VALUE_REMOTE, 0,
scheduler);
}
link_rsc2remotenode(scheduler, rsc);
if (node) {
crm_trace("Setting node %s as shutting down due to orphaned connection resource", rsc_id);
node->details->shutdown = TRUE;
}
}
if (crm_element_value(rsc_entry, PCMK__META_CONTAINER)) {
// This removed resource needs to be mapped to a launcher
crm_trace("Launched resource %s was removed from the configuration",
rsc_id);
pcmk__set_rsc_flags(rsc, pcmk__rsc_removed_launched);
}
pcmk__set_rsc_flags(rsc, pcmk__rsc_removed);
scheduler->priv->resources = g_list_append(scheduler->priv->resources, rsc);
return rsc;
}
/*!
* \internal
* \brief Create orphan instance for anonymous clone resource history
*
* \param[in,out] parent Clone resource that orphan will be added to
* \param[in] rsc_id Orphan's resource ID
* \param[in] node Where orphan is active (for logging only)
* \param[in,out] scheduler Scheduler data
*
* \return Newly added orphaned instance of \p parent
*/
static pcmk_resource_t *
create_anonymous_orphan(pcmk_resource_t *parent, const char *rsc_id,
const pcmk_node_t *node, pcmk_scheduler_t *scheduler)
{
pcmk_resource_t *top = pe__create_clone_child(parent, scheduler);
pcmk_resource_t *orphan = NULL;
// find_rsc() because we might be a cloned group
orphan = top->priv->fns->find_rsc(top, rsc_id, NULL,
pcmk_rsc_match_clone_only);
pcmk__rsc_debug(parent, "Created orphan %s for %s: %s on %s",
top->id, parent->id, rsc_id, pcmk__node_name(node));
return orphan;
}
/*!
* \internal
* \brief Check a node for an instance of an anonymous clone
*
* Return a child instance of the specified anonymous clone, in order of
* preference: (1) the instance running on the specified node, if any;
* (2) an inactive instance (i.e. within the total of \c PCMK_META_CLONE_MAX
* instances); (3) a newly created orphan (that is, \c PCMK_META_CLONE_MAX
* instances are already active).
*
* \param[in,out] scheduler Scheduler data
* \param[in] node Node on which to check for instance
* \param[in,out] parent Clone to check
* \param[in] rsc_id Name of cloned resource in history (no instance)
*/
static pcmk_resource_t *
find_anonymous_clone(pcmk_scheduler_t *scheduler, const pcmk_node_t *node,
pcmk_resource_t *parent, const char *rsc_id)
{
GList *rIter = NULL;
pcmk_resource_t *rsc = NULL;
pcmk_resource_t *inactive_instance = NULL;
gboolean skip_inactive = FALSE;
pcmk__assert(pcmk__is_anonymous_clone(parent));
// Check for active (or partially active, for cloned groups) instance
pcmk__rsc_trace(parent, "Looking for %s on %s in %s",
rsc_id, pcmk__node_name(node), parent->id);
for (rIter = parent->priv->children;
(rIter != NULL) && (rsc == NULL); rIter = rIter->next) {
GList *locations = NULL;
pcmk_resource_t *child = rIter->data;
/* Check whether this instance is already known to be active or pending
* anywhere, at this stage of unpacking. Because this function is called
* for a resource before the resource's individual operation history
* entries are unpacked, locations will generally not contain the
* desired node.
*
* However, there are three exceptions:
* (1) when child is a cloned group and we have already unpacked the
* history of another member of the group on the same node;
* (2) when we've already unpacked the history of another numbered
* instance on the same node (which can happen if
* PCMK_META_GLOBALLY_UNIQUE was flipped from true to false); and
* (3) when we re-run calculations on the same scheduler data as part of
* a simulation.
*/
child->priv->fns->location(child, &locations, pcmk__rsc_node_current
|pcmk__rsc_node_pending);
if (locations) {
/* We should never associate the same numbered anonymous clone
* instance with multiple nodes, and clone instances can't migrate,
* so there must be only one location, regardless of history.
*/
CRM_LOG_ASSERT(locations->next == NULL);
if (pcmk__same_node((pcmk_node_t *) locations->data, node)) {
/* This child instance is active on the requested node, so check
* for a corresponding configured resource. We use find_rsc()
* instead of child because child may be a cloned group, and we
* need the particular member corresponding to rsc_id.
*
* If the history entry is orphaned, rsc will be NULL.
*/
rsc = parent->priv->fns->find_rsc(child, rsc_id, NULL,
pcmk_rsc_match_clone_only);
if (rsc) {
/* If there are multiple instance history entries for an
* anonymous clone in a single node's history (which can
* happen if PCMK_META_GLOBALLY_UNIQUE is switched from true
* to false), we want to consider the instances beyond the
* first as orphans, even if there are inactive instance
* numbers available.
*/
if (rsc->priv->active_nodes != NULL) {
crm_notice("Active (now-)anonymous clone %s has "
"multiple (orphan) instance histories on %s",
parent->id, pcmk__node_name(node));
skip_inactive = TRUE;
rsc = NULL;
} else {
pcmk__rsc_trace(parent, "Resource %s, active", rsc->id);
}
}
}
g_list_free(locations);
} else {
pcmk__rsc_trace(parent, "Resource %s, skip inactive", child->id);
if (!skip_inactive && !inactive_instance
&& !pcmk_is_set(child->flags, pcmk__rsc_blocked)) {
// Remember one inactive instance in case we don't find active
inactive_instance =
parent->priv->fns->find_rsc(child, rsc_id, NULL,
pcmk_rsc_match_clone_only);
/* ... but don't use it if it was already associated with a
* pending action on another node
*/
if (inactive_instance != NULL) {
const pcmk_node_t *pending_node = NULL;
pending_node = inactive_instance->priv->pending_node;
if ((pending_node != NULL)
&& !pcmk__same_node(pending_node, node)) {
inactive_instance = NULL;
}
}
}
}
}
if ((rsc == NULL) && !skip_inactive && (inactive_instance != NULL)) {
pcmk__rsc_trace(parent, "Resource %s, empty slot",
inactive_instance->id);
rsc = inactive_instance;
}
/* If the resource has PCMK_META_REQUIRES set to PCMK_VALUE_QUORUM or
* PCMK_VALUE_NOTHING, and we don't have a clone instance for every node, we
* don't want to consume a valid instance number for unclean nodes. Such
* instances may appear to be active according to the history, but should be
* considered inactive, so we can start an instance elsewhere. Treat such
* instances as orphans.
*
* An exception is instances running on guest nodes -- since guest node
* "fencing" is actually just a resource stop, requires shouldn't apply.
*
* @TODO Ideally, we'd use an inactive instance number if it is not needed
* for any clean instances. However, we don't know that at this point.
*/
if ((rsc != NULL) && !pcmk_is_set(rsc->flags, pcmk__rsc_needs_fencing)
&& (!node->details->online || node->details->unclean)
&& !pcmk__is_guest_or_bundle_node(node)
&& !pe__is_universal_clone(parent, scheduler)) {
rsc = NULL;
}
if (rsc == NULL) {
rsc = create_anonymous_orphan(parent, rsc_id, node, scheduler);
pcmk__rsc_trace(parent, "Resource %s, orphan", rsc->id);
}
return rsc;
}
static pcmk_resource_t *
unpack_find_resource(pcmk_scheduler_t *scheduler, const pcmk_node_t *node,
const char *rsc_id)
{
pcmk_resource_t *rsc = NULL;
pcmk_resource_t *parent = NULL;
crm_trace("looking for %s", rsc_id);
rsc = pe_find_resource(scheduler->priv->resources, rsc_id);
if (rsc == NULL) {
/* If we didn't find the resource by its name in the operation history,
* check it again as a clone instance. Even when PCMK_META_CLONE_MAX=0,
* we create a single :0 orphan to match against here.
*/
char *clone0_id = clone_zero(rsc_id);
pcmk_resource_t *clone0 = pe_find_resource(scheduler->priv->resources,
clone0_id);
if (clone0 && !pcmk_is_set(clone0->flags, pcmk__rsc_unique)) {
rsc = clone0;
parent = uber_parent(clone0);
crm_trace("%s found as %s (%s)", rsc_id, clone0_id, parent->id);
} else {
crm_trace("%s is not known as %s either (orphan)",
rsc_id, clone0_id);
}
free(clone0_id);
} else if (rsc->priv->variant > pcmk__rsc_variant_primitive) {
crm_trace("Resource history for %s is orphaned "
"because it is no longer primitive", rsc_id);
return NULL;
} else {
parent = uber_parent(rsc);
}
if (pcmk__is_anonymous_clone(parent)) {
if (pcmk__is_bundled(parent)) {
rsc = pe__find_bundle_replica(parent->priv->parent, node);
} else {
char *base = clone_strip(rsc_id);
rsc = find_anonymous_clone(scheduler, node, parent, base);
free(base);
pcmk__assert(rsc != NULL);
}
}
if (rsc && !pcmk__str_eq(rsc_id, rsc->id, pcmk__str_none)
&& !pcmk__str_eq(rsc_id, rsc->priv->history_id, pcmk__str_none)) {
pcmk__str_update(&(rsc->priv->history_id), rsc_id);
pcmk__rsc_debug(rsc, "Internally renamed %s on %s to %s%s",
rsc_id, pcmk__node_name(node), rsc->id,
pcmk_is_set(rsc->flags, pcmk__rsc_removed)? " (ORPHAN)" : "");
}
return rsc;
}
static pcmk_resource_t *
process_orphan_resource(const xmlNode *rsc_entry, const pcmk_node_t *node,
pcmk_scheduler_t *scheduler)
{
pcmk_resource_t *rsc = NULL;
const char *rsc_id = crm_element_value(rsc_entry, PCMK_XA_ID);
crm_debug("Detected orphan resource %s on %s",
rsc_id, pcmk__node_name(node));
rsc = create_fake_resource(rsc_id, rsc_entry, scheduler);
if (rsc == NULL) {
return NULL;
}
if (!pcmk_is_set(scheduler->flags, pcmk__sched_stop_removed_resources)) {
pcmk__clear_rsc_flags(rsc, pcmk__rsc_managed);
} else {
CRM_CHECK(rsc != NULL, return NULL);
pcmk__rsc_trace(rsc, "Added orphan %s", rsc->id);
resource_location(rsc, NULL, -PCMK_SCORE_INFINITY,
"__orphan_do_not_run__", scheduler);
}
return rsc;
}
static void
process_rsc_state(pcmk_resource_t *rsc, pcmk_node_t *node,
enum pcmk__on_fail on_fail)
{
pcmk_node_t *tmpnode = NULL;
char *reason = NULL;
enum pcmk__on_fail save_on_fail = pcmk__on_fail_ignore;
pcmk_scheduler_t *scheduler = NULL;
bool known_active = false;
pcmk__assert(rsc != NULL);
scheduler = rsc->priv->scheduler;
known_active = (rsc->priv->orig_role > pcmk_role_stopped);
pcmk__rsc_trace(rsc, "Resource %s is %s on %s: on_fail=%s",
rsc->id, pcmk_role_text(rsc->priv->orig_role),
pcmk__node_name(node), pcmk__on_fail_text(on_fail));
/* process current state */
if (rsc->priv->orig_role != pcmk_role_unknown) {
pcmk_resource_t *iter = rsc;
while (iter) {
if (g_hash_table_lookup(iter->priv->probed_nodes,
node->priv->id) == NULL) {
pcmk_node_t *n = pe__copy_node(node);
pcmk__rsc_trace(rsc, "%s (%s in history) known on %s",
rsc->id,
pcmk__s(rsc->priv->history_id, "the same"),
pcmk__node_name(n));
g_hash_table_insert(iter->priv->probed_nodes,
(gpointer) n->priv->id, n);
}
if (pcmk_is_set(iter->flags, pcmk__rsc_unique)) {
break;
}
iter = iter->priv->parent;
}
}
/* If a managed resource is believed to be running, but node is down ... */
if (known_active && !node->details->online && !node->details->maintenance
&& pcmk_is_set(rsc->flags, pcmk__rsc_managed)) {
gboolean should_fence = FALSE;
/* If this is a guest node, fence it (regardless of whether fencing is
* enabled, because guest node fencing is done by recovery of the
* container resource rather than by the fencer). Mark the resource
* we're processing as failed. When the guest comes back up, its
* operation history in the CIB will be cleared, freeing the affected
* resource to run again once we are sure we know its state.
*/
if (pcmk__is_guest_or_bundle_node(node)) {
pcmk__set_rsc_flags(rsc, pcmk__rsc_failed|pcmk__rsc_stop_if_failed);
should_fence = TRUE;
} else if (pcmk_is_set(scheduler->flags, pcmk__sched_fencing_enabled)) {
if (pcmk__is_remote_node(node)
&& (node->priv->remote != NULL)
&& !pcmk_is_set(node->priv->remote->flags,
pcmk__rsc_failed)) {
/* Setting unseen means that fencing of the remote node will
* occur only if the connection resource is not going to start
* somewhere. This allows connection resources on a failed
* cluster node to move to another node without requiring the
* remote nodes to be fenced as well.
*/
pcmk__clear_node_flags(node, pcmk__node_seen);
reason = crm_strdup_printf("%s is active there (fencing will be"
" revoked if remote connection can "
"be re-established elsewhere)",
rsc->id);
}
should_fence = TRUE;
}
if (should_fence) {
if (reason == NULL) {
reason = crm_strdup_printf("%s is thought to be active there", rsc->id);
}
pe_fence_node(scheduler, node, reason, FALSE);
}
free(reason);
}
/* In order to calculate priority_fencing_delay correctly, save the failure information and pass it to native_add_running(). */
save_on_fail = on_fail;
if (node->details->unclean) {
/* No extra processing needed
* Also allows resources to be started again after a node is shot
*/
on_fail = pcmk__on_fail_ignore;
}
switch (on_fail) {
case pcmk__on_fail_ignore:
/* nothing to do */
break;
case pcmk__on_fail_demote:
pcmk__set_rsc_flags(rsc, pcmk__rsc_failed);
demote_action(rsc, node, FALSE);
break;
case pcmk__on_fail_fence_node:
/* treat it as if it is still running
* but also mark the node as unclean
*/
reason = crm_strdup_printf("%s failed there", rsc->id);
pe_fence_node(scheduler, node, reason, FALSE);
free(reason);
break;
case pcmk__on_fail_standby_node:
pcmk__set_node_flags(node,
pcmk__node_standby|pcmk__node_fail_standby);
break;
case pcmk__on_fail_block:
/* is_managed == FALSE will prevent any
* actions being sent for the resource
*/
pcmk__clear_rsc_flags(rsc, pcmk__rsc_managed);
pcmk__set_rsc_flags(rsc, pcmk__rsc_blocked);
break;
case pcmk__on_fail_ban:
/* make sure it comes up somewhere else
* or not at all
*/
resource_location(rsc, node, -PCMK_SCORE_INFINITY,
"__action_migration_auto__", scheduler);
break;
case pcmk__on_fail_stop:
pe__set_next_role(rsc, pcmk_role_stopped,
PCMK_META_ON_FAIL "=" PCMK_VALUE_STOP);
break;
case pcmk__on_fail_restart:
if (known_active) {
pcmk__set_rsc_flags(rsc,
pcmk__rsc_failed|pcmk__rsc_stop_if_failed);
stop_action(rsc, node, FALSE);
}
break;
case pcmk__on_fail_restart_container:
pcmk__set_rsc_flags(rsc, pcmk__rsc_failed|pcmk__rsc_stop_if_failed);
if ((rsc->priv->launcher != NULL) && pcmk__is_bundled(rsc)) {
/* A bundle's remote connection can run on a different node than
* the bundle's container. We don't necessarily know where the
* container is running yet, so remember it and add a stop
* action for it later.
*/
scheduler->priv->stop_needed =
g_list_prepend(scheduler->priv->stop_needed,
rsc->priv->launcher);
} else if (rsc->priv->launcher != NULL) {
stop_action(rsc->priv->launcher, node, FALSE);
} else if (known_active) {
stop_action(rsc, node, FALSE);
}
break;
case pcmk__on_fail_reset_remote:
pcmk__set_rsc_flags(rsc, pcmk__rsc_failed|pcmk__rsc_stop_if_failed);
if (pcmk_is_set(scheduler->flags, pcmk__sched_fencing_enabled)) {
tmpnode = NULL;
if (pcmk_is_set(rsc->flags, pcmk__rsc_is_remote_connection)) {
tmpnode = pcmk_find_node(scheduler, rsc->id);
}
if (pcmk__is_remote_node(tmpnode)
&& !pcmk_is_set(tmpnode->priv->flags,
pcmk__node_remote_fenced)) {
/* The remote connection resource failed in a way that
* should result in fencing the remote node.
*/
pe_fence_node(scheduler, tmpnode,
"remote connection is unrecoverable", FALSE);
}
}
/* require the stop action regardless if fencing is occurring or not. */
if (known_active) {
stop_action(rsc, node, FALSE);
}
/* if reconnect delay is in use, prevent the connection from exiting the
* "STOPPED" role until the failure is cleared by the delay timeout. */
if (rsc->priv->remote_reconnect_ms > 0U) {
pe__set_next_role(rsc, pcmk_role_stopped, "remote reset");
}
break;
}
/* Ensure a remote connection failure forces an unclean Pacemaker Remote
* node to be fenced. By marking the node as seen, the failure will result
* in a fencing operation regardless if we're going to attempt to reconnect
* in this transition.
*/
if (pcmk_all_flags_set(rsc->flags,
pcmk__rsc_failed|pcmk__rsc_is_remote_connection)) {
tmpnode = pcmk_find_node(scheduler, rsc->id);
if (tmpnode && tmpnode->details->unclean) {
pcmk__set_node_flags(tmpnode, pcmk__node_seen);
}
}
if (known_active) {
if (pcmk_is_set(rsc->flags, pcmk__rsc_removed)) {
if (pcmk_is_set(rsc->flags, pcmk__rsc_managed)) {
crm_notice("Removed resource %s is active on %s and will be "
"stopped when possible",
rsc->id, pcmk__node_name(node));
} else {
crm_notice("Removed resource %s must be stopped manually on %s "
"because " PCMK_OPT_STOP_ORPHAN_RESOURCES
" is set to false", rsc->id, pcmk__node_name(node));
}
}
native_add_running(rsc, node, scheduler,
(save_on_fail != pcmk__on_fail_ignore));
switch (on_fail) {
case pcmk__on_fail_ignore:
break;
case pcmk__on_fail_demote:
case pcmk__on_fail_block:
pcmk__set_rsc_flags(rsc, pcmk__rsc_failed);
break;
default:
pcmk__set_rsc_flags(rsc,
pcmk__rsc_failed|pcmk__rsc_stop_if_failed);
break;
}
} else if ((rsc->priv->history_id != NULL)
&& (strchr(rsc->priv->history_id, ':') != NULL)) {
- /* Only do this for older status sections that included instance numbers
- * Otherwise stopped instances will appear as orphans
+ /* @COMPAT This is for older (<1.1.8) status sections that included
+ * instance numbers, otherwise stopped instances are considered orphans.
+ *
+ * @TODO We should be able to drop this, but some old regression tests
+ * will need to be updated. Double-check that this is not still needed
+ * for unique clones (which may have been later converted to anonymous).
*/
pcmk__rsc_trace(rsc, "Clearing history ID %s for %s (stopped)",
rsc->priv->history_id, rsc->id);
free(rsc->priv->history_id);
rsc->priv->history_id = NULL;
} else {
GList *possible_matches = pe__resource_actions(rsc, node,
PCMK_ACTION_STOP, FALSE);
GList *gIter = possible_matches;
for (; gIter != NULL; gIter = gIter->next) {
pcmk_action_t *stop = (pcmk_action_t *) gIter->data;
pcmk__set_action_flags(stop, pcmk__action_optional);
}
g_list_free(possible_matches);
}
/* A successful stop after migrate_to on the migration source doesn't make
* the partially migrated resource stopped on the migration target.
*/
if ((rsc->priv->orig_role == pcmk_role_stopped)
&& (rsc->priv->active_nodes != NULL)
&& (rsc->priv->partial_migration_target != NULL)
&& pcmk__same_node(rsc->priv->partial_migration_source, node)) {
rsc->priv->orig_role = pcmk_role_started;
}
}
/* create active recurring operations as optional */
static void
process_recurring(pcmk_node_t *node, pcmk_resource_t *rsc,
int start_index, int stop_index,
GList *sorted_op_list, pcmk_scheduler_t *scheduler)
{
int counter = -1;
const char *task = NULL;
const char *status = NULL;
GList *gIter = sorted_op_list;
pcmk__assert(rsc != NULL);
pcmk__rsc_trace(rsc, "%s: Start index %d, stop index = %d",
rsc->id, start_index, stop_index);
for (; gIter != NULL; gIter = gIter->next) {
xmlNode *rsc_op = (xmlNode *) gIter->data;
guint interval_ms = 0;
char *key = NULL;
const char *id = pcmk__xe_id(rsc_op);
counter++;
if (node->details->online == FALSE) {
pcmk__rsc_trace(rsc, "Skipping %s on %s: node is offline",
rsc->id, pcmk__node_name(node));
break;
/* Need to check if there's a monitor for role="Stopped" */
} else if (start_index < stop_index && counter <= stop_index) {
pcmk__rsc_trace(rsc, "Skipping %s on %s: resource is not active",
id, pcmk__node_name(node));
continue;
} else if (counter < start_index) {
pcmk__rsc_trace(rsc, "Skipping %s on %s: old %d",
id, pcmk__node_name(node), counter);
continue;
}
crm_element_value_ms(rsc_op, PCMK_META_INTERVAL, &interval_ms);
if (interval_ms == 0) {
pcmk__rsc_trace(rsc, "Skipping %s on %s: non-recurring",
id, pcmk__node_name(node));
continue;
}
status = crm_element_value(rsc_op, PCMK__XA_OP_STATUS);
if (pcmk__str_eq(status, "-1", pcmk__str_casei)) {
pcmk__rsc_trace(rsc, "Skipping %s on %s: status",
id, pcmk__node_name(node));
continue;
}
task = crm_element_value(rsc_op, PCMK_XA_OPERATION);
/* create the action */
key = pcmk__op_key(rsc->id, task, interval_ms);
pcmk__rsc_trace(rsc, "Creating %s on %s", key, pcmk__node_name(node));
custom_action(rsc, key, task, node, TRUE, scheduler);
}
}
void
calculate_active_ops(const GList *sorted_op_list, int *start_index,
int *stop_index)
{
int counter = -1;
int implied_monitor_start = -1;
int implied_clone_start = -1;
const char *task = NULL;
const char *status = NULL;
*stop_index = -1;
*start_index = -1;
for (const GList *iter = sorted_op_list; iter != NULL; iter = iter->next) {
const xmlNode *rsc_op = (const xmlNode *) iter->data;
counter++;
task = crm_element_value(rsc_op, PCMK_XA_OPERATION);
status = crm_element_value(rsc_op, PCMK__XA_OP_STATUS);
if (pcmk__str_eq(task, PCMK_ACTION_STOP, pcmk__str_casei)
&& pcmk__str_eq(status, "0", pcmk__str_casei)) {
*stop_index = counter;
} else if (pcmk__strcase_any_of(task, PCMK_ACTION_START,
PCMK_ACTION_MIGRATE_FROM, NULL)) {
*start_index = counter;
} else if ((implied_monitor_start <= *stop_index)
&& pcmk__str_eq(task, PCMK_ACTION_MONITOR,
pcmk__str_casei)) {
const char *rc = crm_element_value(rsc_op, PCMK__XA_RC_CODE);
if (pcmk__strcase_any_of(rc, "0", "8", NULL)) {
implied_monitor_start = counter;
}
} else if (pcmk__strcase_any_of(task, PCMK_ACTION_PROMOTE,
PCMK_ACTION_DEMOTE, NULL)) {
implied_clone_start = counter;
}
}
if (*start_index == -1) {
if (implied_clone_start != -1) {
*start_index = implied_clone_start;
} else if (implied_monitor_start != -1) {
*start_index = implied_monitor_start;
}
}
}
// If resource history entry has shutdown lock, remember lock node and time
static void
unpack_shutdown_lock(const xmlNode *rsc_entry, pcmk_resource_t *rsc,
const pcmk_node_t *node, pcmk_scheduler_t *scheduler)
{
time_t lock_time = 0; // When lock started (i.e. node shutdown time)
if ((crm_element_value_epoch(rsc_entry, PCMK_OPT_SHUTDOWN_LOCK,
&lock_time) == pcmk_ok) && (lock_time != 0)) {
if ((scheduler->priv->shutdown_lock_ms > 0U)
&& (get_effective_time(scheduler)
> (lock_time + pcmk__timeout_ms2s(scheduler->priv->shutdown_lock_ms)))) {
pcmk__rsc_info(rsc, "Shutdown lock for %s on %s expired",
rsc->id, pcmk__node_name(node));
pe__clear_resource_history(rsc, node);
} else {
rsc->priv->lock_node = node;
rsc->priv->lock_time = lock_time;
}
}
}
/*!
* \internal
* \brief Unpack one \c PCMK__XE_LRM_RESOURCE entry from a node's CIB status
*
* \param[in,out] node Node whose status is being unpacked
* \param[in] rsc_entry \c PCMK__XE_LRM_RESOURCE XML being unpacked
* \param[in,out] scheduler Scheduler data
*
* \return Resource corresponding to the entry, or NULL if no operation history
*/
static pcmk_resource_t *
unpack_lrm_resource(pcmk_node_t *node, const xmlNode *lrm_resource,
pcmk_scheduler_t *scheduler)
{
GList *gIter = NULL;
int stop_index = -1;
int start_index = -1;
enum rsc_role_e req_role = pcmk_role_unknown;
const char *rsc_id = pcmk__xe_id(lrm_resource);
pcmk_resource_t *rsc = NULL;
GList *op_list = NULL;
GList *sorted_op_list = NULL;
xmlNode *rsc_op = NULL;
xmlNode *last_failure = NULL;
enum pcmk__on_fail on_fail = pcmk__on_fail_ignore;
enum rsc_role_e saved_role = pcmk_role_unknown;
if (rsc_id == NULL) {
pcmk__config_err("Ignoring invalid " PCMK__XE_LRM_RESOURCE
" entry: No " PCMK_XA_ID);
crm_log_xml_info(lrm_resource, "missing-id");
return NULL;
}
crm_trace("Unpacking " PCMK__XE_LRM_RESOURCE " for %s on %s",
rsc_id, pcmk__node_name(node));
/* Build a list of individual PCMK__XE_LRM_RSC_OP entries, so we can sort
* them
*/
for (rsc_op = pcmk__xe_first_child(lrm_resource, PCMK__XE_LRM_RSC_OP, NULL,
NULL);
rsc_op != NULL; rsc_op = pcmk__xe_next(rsc_op, PCMK__XE_LRM_RSC_OP)) {
op_list = g_list_prepend(op_list, rsc_op);
}
if (!pcmk_is_set(scheduler->flags, pcmk__sched_shutdown_lock)) {
if (op_list == NULL) {
// If there are no operations, there is nothing to do
return NULL;
}
}
/* find the resource */
rsc = unpack_find_resource(scheduler, node, rsc_id);
if (rsc == NULL) {
if (op_list == NULL) {
// If there are no operations, there is nothing to do
return NULL;
} else {
rsc = process_orphan_resource(lrm_resource, node, scheduler);
}
}
pcmk__assert(rsc != NULL);
// Check whether the resource is "shutdown-locked" to this node
if (pcmk_is_set(scheduler->flags, pcmk__sched_shutdown_lock)) {
unpack_shutdown_lock(lrm_resource, rsc, node, scheduler);
}
/* process operations */
saved_role = rsc->priv->orig_role;
rsc->priv->orig_role = pcmk_role_unknown;
sorted_op_list = g_list_sort(op_list, sort_op_by_callid);
for (gIter = sorted_op_list; gIter != NULL; gIter = gIter->next) {
xmlNode *rsc_op = (xmlNode *) gIter->data;
unpack_rsc_op(rsc, node, rsc_op, &last_failure, &on_fail);
}
/* create active recurring operations as optional */
calculate_active_ops(sorted_op_list, &start_index, &stop_index);
process_recurring(node, rsc, start_index, stop_index, sorted_op_list,
scheduler);
/* no need to free the contents */
g_list_free(sorted_op_list);
process_rsc_state(rsc, node, on_fail);
if (get_target_role(rsc, &req_role)) {
if ((rsc->priv->next_role == pcmk_role_unknown)
|| (req_role < rsc->priv->next_role)) {
pe__set_next_role(rsc, req_role, PCMK_META_TARGET_ROLE);
} else if (req_role > rsc->priv->next_role) {
pcmk__rsc_info(rsc,
"%s: Not overwriting calculated next role %s"
" with requested next role %s",
rsc->id, pcmk_role_text(rsc->priv->next_role),
pcmk_role_text(req_role));
}
}
if (saved_role > rsc->priv->orig_role) {
rsc->priv->orig_role = saved_role;
}
return rsc;
}
static void
handle_removed_launched_resources(const xmlNode *lrm_rsc_list,
pcmk_scheduler_t *scheduler)
{
for (const xmlNode *rsc_entry = pcmk__xe_first_child(lrm_rsc_list,
PCMK__XE_LRM_RESOURCE,
NULL, NULL);
rsc_entry != NULL;
rsc_entry = pcmk__xe_next(rsc_entry, PCMK__XE_LRM_RESOURCE)) {
pcmk_resource_t *rsc;
pcmk_resource_t *launcher = NULL;
const char *rsc_id;
const char *launcher_id = NULL;
launcher_id = crm_element_value(rsc_entry, PCMK__META_CONTAINER);
rsc_id = crm_element_value(rsc_entry, PCMK_XA_ID);
if ((launcher_id == NULL) || (rsc_id == NULL)) {
continue;
}
launcher = pe_find_resource(scheduler->priv->resources, launcher_id);
if (launcher == NULL) {
continue;
}
rsc = pe_find_resource(scheduler->priv->resources, rsc_id);
if ((rsc == NULL) || (rsc->priv->launcher != NULL)
|| !pcmk_is_set(rsc->flags, pcmk__rsc_removed_launched)) {
continue;
}
pcmk__rsc_trace(rsc, "Mapped launcher of removed resource %s to %s",
rsc->id, launcher_id);
rsc->priv->launcher = launcher;
launcher->priv->launched = g_list_append(launcher->priv->launched,
rsc);
}
}
/*!
* \internal
* \brief Unpack one node's lrm status section
*
* \param[in,out] node Node whose status is being unpacked
* \param[in] xml CIB node state XML
* \param[in,out] scheduler Scheduler data
*/
static void
unpack_node_lrm(pcmk_node_t *node, const xmlNode *xml,
pcmk_scheduler_t *scheduler)
{
bool found_removed_launched_resource = false;
// Drill down to PCMK__XE_LRM_RESOURCES section
xml = pcmk__xe_first_child(xml, PCMK__XE_LRM, NULL, NULL);
if (xml == NULL) {
return;
}
xml = pcmk__xe_first_child(xml, PCMK__XE_LRM_RESOURCES, NULL, NULL);
if (xml == NULL) {
return;
}
// Unpack each PCMK__XE_LRM_RESOURCE entry
for (const xmlNode *rsc_entry = pcmk__xe_first_child(xml,
PCMK__XE_LRM_RESOURCE,
NULL, NULL);
rsc_entry != NULL;
rsc_entry = pcmk__xe_next(rsc_entry, PCMK__XE_LRM_RESOURCE)) {
pcmk_resource_t *rsc = unpack_lrm_resource(node, rsc_entry, scheduler);
if ((rsc != NULL)
&& pcmk_is_set(rsc->flags, pcmk__rsc_removed_launched)) {
found_removed_launched_resource = true;
}
}
/* Now that all resource state has been unpacked for this node, map any
* removed launched resources to their launchers.
*/
if (found_removed_launched_resource) {
handle_removed_launched_resources(xml, scheduler);
}
}
static void
set_active(pcmk_resource_t *rsc)
{
const pcmk_resource_t *top = pe__const_top_resource(rsc, false);
if (top && pcmk_is_set(top->flags, pcmk__rsc_promotable)) {
rsc->priv->orig_role = pcmk_role_unpromoted;
} else {
rsc->priv->orig_role = pcmk_role_started;
}
}
static void
set_node_score(gpointer key, gpointer value, gpointer user_data)
{
pcmk_node_t *node = value;
int *score = user_data;
node->assign->score = *score;
}
#define XPATH_NODE_STATE "/" PCMK_XE_CIB "/" PCMK_XE_STATUS \
"/" PCMK__XE_NODE_STATE
#define SUB_XPATH_LRM_RESOURCE "/" PCMK__XE_LRM \
"/" PCMK__XE_LRM_RESOURCES \
"/" PCMK__XE_LRM_RESOURCE
#define SUB_XPATH_LRM_RSC_OP "/" PCMK__XE_LRM_RSC_OP
static xmlNode *
find_lrm_op(const char *resource, const char *op, const char *node, const char *source,
int target_rc, pcmk_scheduler_t *scheduler)
{
GString *xpath = NULL;
xmlNode *xml = NULL;
CRM_CHECK((resource != NULL) && (op != NULL) && (node != NULL),
return NULL);
xpath = g_string_sized_new(256);
pcmk__g_strcat(xpath,
XPATH_NODE_STATE "[@" PCMK_XA_UNAME "='", node, "']"
SUB_XPATH_LRM_RESOURCE "[@" PCMK_XA_ID "='", resource, "']"
SUB_XPATH_LRM_RSC_OP "[@" PCMK_XA_OPERATION "='", op, "'",
NULL);
/* Need to check against transition_magic too? */
if ((source != NULL) && (strcmp(op, PCMK_ACTION_MIGRATE_TO) == 0)) {
pcmk__g_strcat(xpath,
" and @" PCMK__META_MIGRATE_TARGET "='", source, "']",
NULL);
} else if ((source != NULL)
&& (strcmp(op, PCMK_ACTION_MIGRATE_FROM) == 0)) {
pcmk__g_strcat(xpath,
" and @" PCMK__META_MIGRATE_SOURCE "='", source, "']",
NULL);
} else {
g_string_append_c(xpath, ']');
}
xml = get_xpath_object((const char *) xpath->str, scheduler->input,
LOG_DEBUG);
g_string_free(xpath, TRUE);
if (xml && target_rc >= 0) {
int rc = PCMK_OCF_UNKNOWN_ERROR;
int status = PCMK_EXEC_ERROR;
crm_element_value_int(xml, PCMK__XA_RC_CODE, &rc);
crm_element_value_int(xml, PCMK__XA_OP_STATUS, &status);
if ((rc != target_rc) || (status != PCMK_EXEC_DONE)) {
return NULL;
}
}
return xml;
}
static xmlNode *
find_lrm_resource(const char *rsc_id, const char *node_name,
pcmk_scheduler_t *scheduler)
{
GString *xpath = NULL;
xmlNode *xml = NULL;
CRM_CHECK((rsc_id != NULL) && (node_name != NULL), return NULL);
xpath = g_string_sized_new(256);
pcmk__g_strcat(xpath,
XPATH_NODE_STATE "[@" PCMK_XA_UNAME "='", node_name, "']"
SUB_XPATH_LRM_RESOURCE "[@" PCMK_XA_ID "='", rsc_id, "']",
NULL);
xml = get_xpath_object((const char *) xpath->str, scheduler->input,
LOG_DEBUG);
g_string_free(xpath, TRUE);
return xml;
}
/*!
* \internal
* \brief Check whether a resource has no completed action history on a node
*
* \param[in,out] rsc Resource to check
* \param[in] node_name Node to check
*
* \return true if \p rsc_id is unknown on \p node_name, otherwise false
*/
static bool
unknown_on_node(pcmk_resource_t *rsc, const char *node_name)
{
bool result = false;
xmlXPathObjectPtr search;
char *xpath = NULL;
xpath = crm_strdup_printf(XPATH_NODE_STATE "[@" PCMK_XA_UNAME "='%s']"
SUB_XPATH_LRM_RESOURCE "[@" PCMK_XA_ID "='%s']"
SUB_XPATH_LRM_RSC_OP
"[@" PCMK__XA_RC_CODE "!='%d']",
node_name, rsc->id, PCMK_OCF_UNKNOWN);
search = xpath_search(rsc->priv->scheduler->input, xpath);
result = (numXpathResults(search) == 0);
freeXpathObject(search);
free(xpath);
return result;
}
/*!
* \internal
* \brief Check whether a probe/monitor indicating the resource was not running
* on a node happened after some event
*
* \param[in] rsc_id Resource being checked
* \param[in] node_name Node being checked
* \param[in] xml_op Event that monitor is being compared to
* \param[in,out] scheduler Scheduler data
*
* \return true if such a monitor happened after event, false otherwise
*/
static bool
monitor_not_running_after(const char *rsc_id, const char *node_name,
const xmlNode *xml_op, pcmk_scheduler_t *scheduler)
{
/* Any probe/monitor operation on the node indicating it was not running
* there
*/
xmlNode *monitor = find_lrm_op(rsc_id, PCMK_ACTION_MONITOR, node_name,
NULL, PCMK_OCF_NOT_RUNNING, scheduler);
return (monitor != NULL) && (pe__is_newer_op(monitor, xml_op) > 0);
}
/*!
* \internal
* \brief Check whether any non-monitor operation on a node happened after some
* event
*
* \param[in] rsc_id Resource being checked
* \param[in] node_name Node being checked
* \param[in] xml_op Event that non-monitor is being compared to
* \param[in,out] scheduler Scheduler data
*
* \return true if such a operation happened after event, false otherwise
*/
static bool
non_monitor_after(const char *rsc_id, const char *node_name,
const xmlNode *xml_op, pcmk_scheduler_t *scheduler)
{
xmlNode *lrm_resource = NULL;
lrm_resource = find_lrm_resource(rsc_id, node_name, scheduler);
if (lrm_resource == NULL) {
return false;
}
for (xmlNode *op = pcmk__xe_first_child(lrm_resource, PCMK__XE_LRM_RSC_OP,
NULL, NULL);
op != NULL; op = pcmk__xe_next(op, PCMK__XE_LRM_RSC_OP)) {
const char * task = NULL;
if (op == xml_op) {
continue;
}
task = crm_element_value(op, PCMK_XA_OPERATION);
if (pcmk__str_any_of(task, PCMK_ACTION_START, PCMK_ACTION_STOP,
PCMK_ACTION_MIGRATE_TO, PCMK_ACTION_MIGRATE_FROM,
NULL)
&& pe__is_newer_op(op, xml_op) > 0) {
return true;
}
}
return false;
}
/*!
* \internal
* \brief Check whether the resource has newer state on a node after a migration
* attempt
*
* \param[in] rsc_id Resource being checked
* \param[in] node_name Node being checked
* \param[in] migrate_to Any migrate_to event that is being compared to
* \param[in] migrate_from Any migrate_from event that is being compared to
* \param[in,out] scheduler Scheduler data
*
* \return true if such a operation happened after event, false otherwise
*/
static bool
newer_state_after_migrate(const char *rsc_id, const char *node_name,
const xmlNode *migrate_to,
const xmlNode *migrate_from,
pcmk_scheduler_t *scheduler)
{
const xmlNode *xml_op = (migrate_from != NULL)? migrate_from : migrate_to;
const char *source = crm_element_value(xml_op, PCMK__META_MIGRATE_SOURCE);
/* It's preferred to compare to the migrate event on the same node if
* existing, since call ids are more reliable.
*/
if ((xml_op != migrate_to) && (migrate_to != NULL)
&& pcmk__str_eq(node_name, source, pcmk__str_casei)) {
xml_op = migrate_to;
}
/* If there's any newer non-monitor operation on the node, or any newer
* probe/monitor operation on the node indicating it was not running there,
* the migration events potentially no longer matter for the node.
*/
return non_monitor_after(rsc_id, node_name, xml_op, scheduler)
|| monitor_not_running_after(rsc_id, node_name, xml_op, scheduler);
}
/*!
* \internal
* \brief Parse migration source and target node names from history entry
*
* \param[in] entry Resource history entry for a migration action
* \param[in] source_node If not NULL, source must match this node
* \param[in] target_node If not NULL, target must match this node
* \param[out] source_name Where to store migration source node name
* \param[out] target_name Where to store migration target node name
*
* \return Standard Pacemaker return code
*/
static int
get_migration_node_names(const xmlNode *entry, const pcmk_node_t *source_node,
const pcmk_node_t *target_node,
const char **source_name, const char **target_name)
{
*source_name = crm_element_value(entry, PCMK__META_MIGRATE_SOURCE);
*target_name = crm_element_value(entry, PCMK__META_MIGRATE_TARGET);
if ((*source_name == NULL) || (*target_name == NULL)) {
pcmk__config_err("Ignoring resource history entry %s without "
PCMK__META_MIGRATE_SOURCE " and "
PCMK__META_MIGRATE_TARGET, pcmk__xe_id(entry));
return pcmk_rc_unpack_error;
}
if ((source_node != NULL)
&& !pcmk__str_eq(*source_name, source_node->priv->name,
pcmk__str_casei|pcmk__str_null_matches)) {
pcmk__config_err("Ignoring resource history entry %s because "
PCMK__META_MIGRATE_SOURCE "='%s' does not match %s",
pcmk__xe_id(entry), *source_name,
pcmk__node_name(source_node));
return pcmk_rc_unpack_error;
}
if ((target_node != NULL)
&& !pcmk__str_eq(*target_name, target_node->priv->name,
pcmk__str_casei|pcmk__str_null_matches)) {
pcmk__config_err("Ignoring resource history entry %s because "
PCMK__META_MIGRATE_TARGET "='%s' does not match %s",
pcmk__xe_id(entry), *target_name,
pcmk__node_name(target_node));
return pcmk_rc_unpack_error;
}
return pcmk_rc_ok;
}
/*
* \internal
* \brief Add a migration source to a resource's list of dangling migrations
*
* If the migrate_to and migrate_from actions in a live migration both
* succeeded, but there is no stop on the source, the migration is considered
* "dangling." Add the source to the resource's dangling migration list, which
* will be used to schedule a stop on the source without affecting the target.
*
* \param[in,out] rsc Resource involved in migration
* \param[in] node Migration source
*/
static void
add_dangling_migration(pcmk_resource_t *rsc, const pcmk_node_t *node)
{
pcmk__rsc_trace(rsc, "Dangling migration of %s requires stop on %s",
rsc->id, pcmk__node_name(node));
rsc->priv->orig_role = pcmk_role_stopped;
rsc->priv->dangling_migration_sources =
g_list_prepend(rsc->priv->dangling_migration_sources,
(gpointer) node);
}
/*!
* \internal
* \brief Update resource role etc. after a successful migrate_to action
*
* \param[in,out] history Parsed action result history
*/
static void
unpack_migrate_to_success(struct action_history *history)
{
/* A complete migration sequence is:
* 1. migrate_to on source node (which succeeded if we get to this function)
* 2. migrate_from on target node
* 3. stop on source node
*
* If no migrate_from has happened, the migration is considered to be
* "partial". If the migrate_from succeeded but no stop has happened, the
* migration is considered to be "dangling".
*
* If a successful migrate_to and stop have happened on the source node, we
* still need to check for a partial migration, due to scenarios (easier to
* produce with batch-limit=1) like:
*
* - A resource is migrating from node1 to node2, and a migrate_to is
* initiated for it on node1.
*
* - node2 goes into standby mode while the migrate_to is pending, which
* aborts the transition.
*
* - Upon completion of the migrate_to, a new transition schedules a stop
* on both nodes and a start on node1.
*
* - If the new transition is aborted for any reason while the resource is
* stopping on node1, the transition after that stop completes will see
* the migrate_to and stop on the source, but it's still a partial
* migration, and the resource must be stopped on node2 because it is
* potentially active there due to the migrate_to.
*
* We also need to take into account that either node's history may be
* cleared at any point in the migration process.
*/
int from_rc = PCMK_OCF_OK;
int from_status = PCMK_EXEC_PENDING;
pcmk_node_t *target_node = NULL;
xmlNode *migrate_from = NULL;
const char *source = NULL;
const char *target = NULL;
bool source_newer_op = false;
bool target_newer_state = false;
bool active_on_target = false;
pcmk_scheduler_t *scheduler = history->rsc->priv->scheduler;
// Get source and target node names from XML
if (get_migration_node_names(history->xml, history->node, NULL, &source,
&target) != pcmk_rc_ok) {
return;
}
// Check for newer state on the source
source_newer_op = non_monitor_after(history->rsc->id, source, history->xml,
scheduler);
// Check for a migrate_from action from this source on the target
migrate_from = find_lrm_op(history->rsc->id, PCMK_ACTION_MIGRATE_FROM,
target, source, -1, scheduler);
if (migrate_from != NULL) {
if (source_newer_op) {
/* There's a newer non-monitor operation on the source and a
* migrate_from on the target, so this migrate_to is irrelevant to
* the resource's state.
*/
return;
}
crm_element_value_int(migrate_from, PCMK__XA_RC_CODE, &from_rc);
crm_element_value_int(migrate_from, PCMK__XA_OP_STATUS, &from_status);
}
/* If the resource has newer state on both the source and target after the
* migration events, this migrate_to is irrelevant to the resource's state.
*/
target_newer_state = newer_state_after_migrate(history->rsc->id, target,
history->xml, migrate_from,
scheduler);
if (source_newer_op && target_newer_state) {
return;
}
/* Check for dangling migration (migrate_from succeeded but stop not done).
* We know there's no stop because we already returned if the target has a
* migrate_from and the source has any newer non-monitor operation.
*/
if ((from_rc == PCMK_OCF_OK) && (from_status == PCMK_EXEC_DONE)) {
add_dangling_migration(history->rsc, history->node);
return;
}
/* Without newer state, this migrate_to implies the resource is active.
* (Clones are not allowed to migrate, so role can't be promoted.)
*/
history->rsc->priv->orig_role = pcmk_role_started;
target_node = pcmk_find_node(scheduler, target);
active_on_target = !target_newer_state && (target_node != NULL)
&& target_node->details->online;
if (from_status != PCMK_EXEC_PENDING) { // migrate_from failed on target
if (active_on_target) {
native_add_running(history->rsc, target_node, scheduler, TRUE);
} else {
// Mark resource as failed, require recovery, and prevent migration
pcmk__set_rsc_flags(history->rsc,
pcmk__rsc_failed|pcmk__rsc_stop_if_failed);
pcmk__clear_rsc_flags(history->rsc, pcmk__rsc_migratable);
}
return;
}
// The migrate_from is pending, complete but erased, or to be scheduled
/* If there is no history at all for the resource on an online target, then
* it was likely cleaned. Just return, and we'll schedule a probe. Once we
* have the probe result, it will be reflected in target_newer_state.
*/
if ((target_node != NULL) && target_node->details->online
&& unknown_on_node(history->rsc, target)) {
return;
}
if (active_on_target) {
pcmk_node_t *source_node = pcmk_find_node(scheduler, source);
native_add_running(history->rsc, target_node, scheduler, FALSE);
if ((source_node != NULL) && source_node->details->online) {
/* This is a partial migration: the migrate_to completed
* successfully on the source, but the migrate_from has not
* completed. Remember the source and target; if the newly
* chosen target remains the same when we schedule actions
* later, we may continue with the migration.
*/
history->rsc->priv->partial_migration_target = target_node;
history->rsc->priv->partial_migration_source = source_node;
}
} else if (!source_newer_op) {
// Mark resource as failed, require recovery, and prevent migration
pcmk__set_rsc_flags(history->rsc,
pcmk__rsc_failed|pcmk__rsc_stop_if_failed);
pcmk__clear_rsc_flags(history->rsc, pcmk__rsc_migratable);
}
}
/*!
* \internal
* \brief Update resource role etc. after a failed migrate_to action
*
* \param[in,out] history Parsed action result history
*/
static void
unpack_migrate_to_failure(struct action_history *history)
{
xmlNode *target_migrate_from = NULL;
const char *source = NULL;
const char *target = NULL;
pcmk_scheduler_t *scheduler = history->rsc->priv->scheduler;
// Get source and target node names from XML
if (get_migration_node_names(history->xml, history->node, NULL, &source,
&target) != pcmk_rc_ok) {
return;
}
/* If a migration failed, we have to assume the resource is active. Clones
* are not allowed to migrate, so role can't be promoted.
*/
history->rsc->priv->orig_role = pcmk_role_started;
// Check for migrate_from on the target
target_migrate_from = find_lrm_op(history->rsc->id,
PCMK_ACTION_MIGRATE_FROM, target, source,
PCMK_OCF_OK, scheduler);
if (/* If the resource state is unknown on the target, it will likely be
* probed there.
* Don't just consider it running there. We will get back here anyway in
* case the probe detects it's running there.
*/
!unknown_on_node(history->rsc, target)
/* If the resource has newer state on the target after the migration
* events, this migrate_to no longer matters for the target.
*/
&& !newer_state_after_migrate(history->rsc->id, target, history->xml,
target_migrate_from, scheduler)) {
/* The resource has no newer state on the target, so assume it's still
* active there.
* (if it is up).
*/
pcmk_node_t *target_node = pcmk_find_node(scheduler, target);
if (target_node && target_node->details->online) {
native_add_running(history->rsc, target_node, scheduler, FALSE);
}
} else if (!non_monitor_after(history->rsc->id, source, history->xml,
scheduler)) {
/* We know the resource has newer state on the target, but this
* migrate_to still matters for the source as long as there's no newer
* non-monitor operation there.
*/
// Mark node as having dangling migration so we can force a stop later
history->rsc->priv->dangling_migration_sources =
g_list_prepend(history->rsc->priv->dangling_migration_sources,
(gpointer) history->node);
}
}
/*!
* \internal
* \brief Update resource role etc. after a failed migrate_from action
*
* \param[in,out] history Parsed action result history
*/
static void
unpack_migrate_from_failure(struct action_history *history)
{
xmlNode *source_migrate_to = NULL;
const char *source = NULL;
const char *target = NULL;
pcmk_scheduler_t *scheduler = history->rsc->priv->scheduler;
// Get source and target node names from XML
if (get_migration_node_names(history->xml, NULL, history->node, &source,
&target) != pcmk_rc_ok) {
return;
}
/* If a migration failed, we have to assume the resource is active. Clones
* are not allowed to migrate, so role can't be promoted.
*/
history->rsc->priv->orig_role = pcmk_role_started;
// Check for a migrate_to on the source
source_migrate_to = find_lrm_op(history->rsc->id, PCMK_ACTION_MIGRATE_TO,
source, target, PCMK_OCF_OK, scheduler);
if (/* If the resource state is unknown on the source, it will likely be
* probed there.
* Don't just consider it running there. We will get back here anyway in
* case the probe detects it's running there.
*/
!unknown_on_node(history->rsc, source)
/* If the resource has newer state on the source after the migration
* events, this migrate_from no longer matters for the source.
*/
&& !newer_state_after_migrate(history->rsc->id, source,
source_migrate_to, history->xml,
scheduler)) {
/* The resource has no newer state on the source, so assume it's still
* active there (if it is up).
*/
pcmk_node_t *source_node = pcmk_find_node(scheduler, source);
if (source_node && source_node->details->online) {
native_add_running(history->rsc, source_node, scheduler, TRUE);
}
}
}
/*!
* \internal
* \brief Add an action to cluster's list of failed actions
*
* \param[in,out] history Parsed action result history
*/
static void
record_failed_op(struct action_history *history)
{
const pcmk_scheduler_t *scheduler = history->rsc->priv->scheduler;
if (!(history->node->details->online)) {
return;
}
for (const xmlNode *xIter = scheduler->priv->failed->children;
xIter != NULL; xIter = xIter->next) {
const char *key = pcmk__xe_history_key(xIter);
const char *uname = crm_element_value(xIter, PCMK_XA_UNAME);
if (pcmk__str_eq(history->key, key, pcmk__str_none)
&& pcmk__str_eq(uname, history->node->priv->name,
pcmk__str_casei)) {
crm_trace("Skipping duplicate entry %s on %s",
history->key, pcmk__node_name(history->node));
return;
}
}
crm_trace("Adding entry for %s on %s to failed action list",
history->key, pcmk__node_name(history->node));
crm_xml_add(history->xml, PCMK_XA_UNAME, history->node->priv->name);
crm_xml_add(history->xml, PCMK__XA_RSC_ID, history->rsc->id);
pcmk__xml_copy(scheduler->priv->failed, history->xml);
}
static char *
last_change_str(const xmlNode *xml_op)
{
time_t when;
char *result = NULL;
if (crm_element_value_epoch(xml_op, PCMK_XA_LAST_RC_CHANGE,
&when) == pcmk_ok) {
char *when_s = pcmk__epoch2str(&when, 0);
const char *p = strchr(when_s, ' ');
// Skip day of week to make message shorter
if ((p != NULL) && (*(++p) != '\0')) {
result = pcmk__str_copy(p);
}
free(when_s);
}
if (result == NULL) {
result = pcmk__str_copy("unknown_time");
}
return result;
}
/*!
* \internal
* \brief Ban a resource (or its clone if an anonymous instance) from all nodes
*
* \param[in,out] rsc Resource to ban
*/
static void
ban_from_all_nodes(pcmk_resource_t *rsc)
{
int score = -PCMK_SCORE_INFINITY;
const pcmk_scheduler_t *scheduler = rsc->priv->scheduler;
if (rsc->priv->parent != NULL) {
pcmk_resource_t *parent = uber_parent(rsc);
if (pcmk__is_anonymous_clone(parent)) {
/* For anonymous clones, if an operation with
* PCMK_META_ON_FAIL=PCMK_VALUE_STOP fails for any instance, the
* entire clone must stop.
*/
rsc = parent;
}
}
// Ban the resource from all nodes
crm_notice("%s will not be started under current conditions", rsc->id);
if (rsc->priv->allowed_nodes != NULL) {
g_hash_table_destroy(rsc->priv->allowed_nodes);
}
rsc->priv->allowed_nodes = pe__node_list2table(scheduler->nodes);
g_hash_table_foreach(rsc->priv->allowed_nodes, set_node_score, &score);
}
/*!
* \internal
* \brief Get configured failure handling and role after failure for an action
*
* \param[in,out] history Unpacked action history entry
* \param[out] on_fail Where to set configured failure handling
* \param[out] fail_role Where to set to role after failure
*/
static void
unpack_failure_handling(struct action_history *history,
enum pcmk__on_fail *on_fail,
enum rsc_role_e *fail_role)
{
xmlNode *config = pcmk__find_action_config(history->rsc, history->task,
history->interval_ms, true);
GHashTable *meta = pcmk__unpack_action_meta(history->rsc, history->node,
history->task,
history->interval_ms, config);
const char *on_fail_str = g_hash_table_lookup(meta, PCMK_META_ON_FAIL);
*on_fail = pcmk__parse_on_fail(history->rsc, history->task,
history->interval_ms, on_fail_str);
*fail_role = pcmk__role_after_failure(history->rsc, history->task, *on_fail,
meta);
g_hash_table_destroy(meta);
}
/*!
* \internal
* \brief Update resource role, failure handling, etc., after a failed action
*
* \param[in,out] history Parsed action result history
* \param[in] config_on_fail Action failure handling from configuration
* \param[in] fail_role Resource's role after failure of this action
* \param[out] last_failure This will be set to the history XML
* \param[in,out] on_fail Actual handling of action result
*/
static void
unpack_rsc_op_failure(struct action_history *history,
enum pcmk__on_fail config_on_fail,
enum rsc_role_e fail_role, xmlNode **last_failure,
enum pcmk__on_fail *on_fail)
{
bool is_probe = false;
char *last_change_s = NULL;
pcmk_scheduler_t *scheduler = history->rsc->priv->scheduler;
*last_failure = history->xml;
is_probe = pcmk_xe_is_probe(history->xml);
last_change_s = last_change_str(history->xml);
if (!pcmk_is_set(scheduler->flags, pcmk__sched_symmetric_cluster)
&& (history->exit_status == PCMK_OCF_NOT_INSTALLED)) {
crm_trace("Unexpected result (%s%s%s) was recorded for "
"%s of %s on %s at %s " QB_XS " exit-status=%d id=%s",
crm_exit_str(history->exit_status),
(pcmk__str_empty(history->exit_reason)? "" : ": "),
pcmk__s(history->exit_reason, ""),
(is_probe? "probe" : history->task), history->rsc->id,
pcmk__node_name(history->node), last_change_s,
history->exit_status, history->id);
} else {
pcmk__sched_warn(scheduler,
"Unexpected result (%s%s%s) was recorded for %s of "
"%s on %s at %s " QB_XS " exit-status=%d id=%s",
crm_exit_str(history->exit_status),
(pcmk__str_empty(history->exit_reason)? "" : ": "),
pcmk__s(history->exit_reason, ""),
(is_probe? "probe" : history->task), history->rsc->id,
pcmk__node_name(history->node), last_change_s,
history->exit_status, history->id);
if (is_probe && (history->exit_status != PCMK_OCF_OK)
&& (history->exit_status != PCMK_OCF_NOT_RUNNING)
&& (history->exit_status != PCMK_OCF_RUNNING_PROMOTED)) {
/* A failed (not just unexpected) probe result could mean the user
* didn't know resources will be probed even where they can't run.
*/
crm_notice("If it is not possible for %s to run on %s, see "
"the " PCMK_XA_RESOURCE_DISCOVERY " option for location "
"constraints",
history->rsc->id, pcmk__node_name(history->node));
}
record_failed_op(history);
}
free(last_change_s);
if (*on_fail < config_on_fail) {
pcmk__rsc_trace(history->rsc, "on-fail %s -> %s for %s",
pcmk__on_fail_text(*on_fail),
pcmk__on_fail_text(config_on_fail), history->key);
*on_fail = config_on_fail;
}
if (strcmp(history->task, PCMK_ACTION_STOP) == 0) {
resource_location(history->rsc, history->node, -PCMK_SCORE_INFINITY,
"__stop_fail__", scheduler);
} else if (strcmp(history->task, PCMK_ACTION_MIGRATE_TO) == 0) {
unpack_migrate_to_failure(history);
} else if (strcmp(history->task, PCMK_ACTION_MIGRATE_FROM) == 0) {
unpack_migrate_from_failure(history);
} else if (strcmp(history->task, PCMK_ACTION_PROMOTE) == 0) {
history->rsc->priv->orig_role = pcmk_role_promoted;
} else if (strcmp(history->task, PCMK_ACTION_DEMOTE) == 0) {
if (config_on_fail == pcmk__on_fail_block) {
history->rsc->priv->orig_role = pcmk_role_promoted;
pe__set_next_role(history->rsc, pcmk_role_stopped,
"demote with " PCMK_META_ON_FAIL "=block");
} else if (history->exit_status == PCMK_OCF_NOT_RUNNING) {
history->rsc->priv->orig_role = pcmk_role_stopped;
} else {
/* Staying in the promoted role would put the scheduler and
* controller into a loop. Setting the role to unpromoted is not
* dangerous because the resource will be stopped as part of
* recovery, and any promotion will be ordered after that stop.
*/
history->rsc->priv->orig_role = pcmk_role_unpromoted;
}
}
if (is_probe && (history->exit_status == PCMK_OCF_NOT_INSTALLED)) {
/* leave stopped */
pcmk__rsc_trace(history->rsc, "Leaving %s stopped", history->rsc->id);
history->rsc->priv->orig_role = pcmk_role_stopped;
} else if (history->rsc->priv->orig_role < pcmk_role_started) {
pcmk__rsc_trace(history->rsc, "Setting %s active", history->rsc->id);
set_active(history->rsc);
}
pcmk__rsc_trace(history->rsc,
"Resource %s: role=%s unclean=%s on_fail=%s fail_role=%s",
history->rsc->id,
pcmk_role_text(history->rsc->priv->orig_role),
pcmk__btoa(history->node->details->unclean),
pcmk__on_fail_text(config_on_fail),
pcmk_role_text(fail_role));
if ((fail_role != pcmk_role_started)
&& (history->rsc->priv->next_role < fail_role)) {
pe__set_next_role(history->rsc, fail_role, "failure");
}
if (fail_role == pcmk_role_stopped) {
ban_from_all_nodes(history->rsc);
}
}
/*!
* \internal
* \brief Block a resource with a failed action if it cannot be recovered
*
* If resource action is a failed stop and fencing is not possible, mark the
* resource as unmanaged and blocked, since recovery cannot be done.
*
* \param[in,out] history Parsed action history entry
*/
static void
block_if_unrecoverable(struct action_history *history)
{
char *last_change_s = NULL;
if (strcmp(history->task, PCMK_ACTION_STOP) != 0) {
return; // All actions besides stop are always recoverable
}
if (pe_can_fence(history->node->priv->scheduler, history->node)) {
return; // Failed stops are recoverable via fencing
}
last_change_s = last_change_str(history->xml);
pcmk__sched_err(history->node->priv->scheduler,
"No further recovery can be attempted for %s "
"because %s on %s failed (%s%s%s) at %s "
QB_XS " rc=%d id=%s",
history->rsc->id, history->task,
pcmk__node_name(history->node),
crm_exit_str(history->exit_status),
(pcmk__str_empty(history->exit_reason)? "" : ": "),
pcmk__s(history->exit_reason, ""),
last_change_s, history->exit_status, history->id);
free(last_change_s);
pcmk__clear_rsc_flags(history->rsc, pcmk__rsc_managed);
pcmk__set_rsc_flags(history->rsc, pcmk__rsc_blocked);
}
/*!
* \internal
* \brief Update action history's execution status and why
*
* \param[in,out] history Parsed action history entry
* \param[out] why Where to store reason for update
* \param[in] value New value
* \param[in] reason Description of why value was changed
*/
static inline void
remap_because(struct action_history *history, const char **why, int value,
const char *reason)
{
if (history->execution_status != value) {
history->execution_status = value;
*why = reason;
}
}
/*!
* \internal
* \brief Remap informational monitor results and operation status
*
* For the monitor results, certain OCF codes are for providing extended information
* to the user about services that aren't yet failed but not entirely healthy either.
* These must be treated as the "normal" result by Pacemaker.
*
* For operation status, the action result can be used to determine an appropriate
* status for the purposes of responding to the action. The status provided by the
* executor is not directly usable since the executor does not know what was expected.
*
* \param[in,out] history Parsed action history entry
* \param[in,out] on_fail What should be done about the result
* \param[in] expired Whether result is expired
*
* \note If the result is remapped and the node is not shutting down or failed,
* the operation will be recorded in the scheduler data's list of failed
* operations to highlight it for the user.
*
* \note This may update the resource's current and next role.
*/
static void
remap_operation(struct action_history *history,
enum pcmk__on_fail *on_fail, bool expired)
{
+ /* @TODO It would probably also be a good idea to map an exit status of
+ * CRM_EX_PROMOTED or CRM_EX_DEGRADED_PROMOTED to CRM_EX_OK for promote
+ * actions
+ */
+
bool is_probe = false;
int orig_exit_status = history->exit_status;
int orig_exec_status = history->execution_status;
const char *why = NULL;
const char *task = history->task;
// Remap degraded results to their successful counterparts
history->exit_status = pcmk__effective_rc(history->exit_status);
if (history->exit_status != orig_exit_status) {
why = "degraded result";
if (!expired && (!history->node->details->shutdown
|| history->node->details->online)) {
record_failed_op(history);
}
}
if (!pcmk__is_bundled(history->rsc)
&& pcmk_xe_mask_probe_failure(history->xml)
&& ((history->execution_status != PCMK_EXEC_DONE)
|| (history->exit_status != PCMK_OCF_NOT_RUNNING))) {
history->execution_status = PCMK_EXEC_DONE;
history->exit_status = PCMK_OCF_NOT_RUNNING;
why = "equivalent probe result";
}
/* If the executor reported an execution status of anything but done or
* error, consider that final. But for done or error, we know better whether
* it should be treated as a failure or not, because we know the expected
* result.
*/
switch (history->execution_status) {
case PCMK_EXEC_DONE:
case PCMK_EXEC_ERROR:
break;
// These should be treated as node-fatal
case PCMK_EXEC_NO_FENCE_DEVICE:
case PCMK_EXEC_NO_SECRETS:
remap_because(history, &why, PCMK_EXEC_ERROR_HARD,
"node-fatal error");
goto remap_done;
default:
goto remap_done;
}
is_probe = pcmk_xe_is_probe(history->xml);
if (is_probe) {
task = "probe";
}
if (history->expected_exit_status < 0) {
/* Pre-1.0 Pacemaker versions, and Pacemaker 1.1.6 or earlier with
* Heartbeat 2.0.7 or earlier as the cluster layer, did not include the
* expected exit status in the transition key, which (along with the
* similar case of a corrupted transition key in the CIB) will be
* reported to this function as -1. Pacemaker 2.0+ does not support
* rolling upgrades from those versions or processing of saved CIB files
* from those versions, so we do not need to care much about this case.
*/
remap_because(history, &why, PCMK_EXEC_ERROR,
"obsolete history format");
pcmk__config_warn("Expected result not found for %s on %s "
"(corrupt or obsolete CIB?)",
history->key, pcmk__node_name(history->node));
} else if (history->exit_status == history->expected_exit_status) {
remap_because(history, &why, PCMK_EXEC_DONE, "expected result");
} else {
remap_because(history, &why, PCMK_EXEC_ERROR, "unexpected result");
pcmk__rsc_debug(history->rsc,
"%s on %s: expected %d (%s), got %d (%s%s%s)",
history->key, pcmk__node_name(history->node),
history->expected_exit_status,
crm_exit_str(history->expected_exit_status),
history->exit_status,
crm_exit_str(history->exit_status),
(pcmk__str_empty(history->exit_reason)? "" : ": "),
pcmk__s(history->exit_reason, ""));
}
switch (history->exit_status) {
case PCMK_OCF_OK:
if (is_probe
&& (history->expected_exit_status == PCMK_OCF_NOT_RUNNING)) {
char *last_change_s = last_change_str(history->xml);
remap_because(history, &why, PCMK_EXEC_DONE, "probe");
pcmk__rsc_info(history->rsc,
"Probe found %s active on %s at %s",
history->rsc->id, pcmk__node_name(history->node),
last_change_s);
free(last_change_s);
}
break;
case PCMK_OCF_NOT_RUNNING:
if (is_probe
|| (history->expected_exit_status == history->exit_status)
|| !pcmk_is_set(history->rsc->flags, pcmk__rsc_managed)) {
/* For probes, recurring monitors for the Stopped role, and
* unmanaged resources, "not running" is not considered a
* failure.
*/
remap_because(history, &why, PCMK_EXEC_DONE, "exit status");
history->rsc->priv->orig_role = pcmk_role_stopped;
*on_fail = pcmk__on_fail_ignore;
pe__set_next_role(history->rsc, pcmk_role_unknown,
"not running");
}
break;
case PCMK_OCF_RUNNING_PROMOTED:
if (is_probe
&& (history->exit_status != history->expected_exit_status)) {
char *last_change_s = last_change_str(history->xml);
remap_because(history, &why, PCMK_EXEC_DONE, "probe");
pcmk__rsc_info(history->rsc,
"Probe found %s active and promoted on %s at %s",
history->rsc->id,
pcmk__node_name(history->node), last_change_s);
free(last_change_s);
}
if (!expired
|| (history->exit_status == history->expected_exit_status)) {
history->rsc->priv->orig_role = pcmk_role_promoted;
}
break;
case PCMK_OCF_FAILED_PROMOTED:
if (!expired) {
history->rsc->priv->orig_role = pcmk_role_promoted;
}
remap_because(history, &why, PCMK_EXEC_ERROR, "exit status");
break;
case PCMK_OCF_NOT_CONFIGURED:
remap_because(history, &why, PCMK_EXEC_ERROR_FATAL, "exit status");
break;
case PCMK_OCF_UNIMPLEMENT_FEATURE:
{
guint interval_ms = 0;
crm_element_value_ms(history->xml, PCMK_META_INTERVAL,
&interval_ms);
if (interval_ms == 0) {
if (!expired) {
block_if_unrecoverable(history);
}
remap_because(history, &why, PCMK_EXEC_ERROR_HARD,
"exit status");
} else {
remap_because(history, &why, PCMK_EXEC_NOT_SUPPORTED,
"exit status");
}
}
break;
case PCMK_OCF_NOT_INSTALLED:
case PCMK_OCF_INVALID_PARAM:
case PCMK_OCF_INSUFFICIENT_PRIV:
if (!expired) {
block_if_unrecoverable(history);
}
remap_because(history, &why, PCMK_EXEC_ERROR_HARD, "exit status");
break;
default:
if (history->execution_status == PCMK_EXEC_DONE) {
char *last_change_s = last_change_str(history->xml);
crm_info("Treating unknown exit status %d from %s of %s "
"on %s at %s as failure",
history->exit_status, task, history->rsc->id,
pcmk__node_name(history->node), last_change_s);
remap_because(history, &why, PCMK_EXEC_ERROR,
"unknown exit status");
free(last_change_s);
}
break;
}
remap_done:
if (why != NULL) {
pcmk__rsc_trace(history->rsc,
"Remapped %s result from [%s: %s] to [%s: %s] "
"because of %s",
history->key, pcmk_exec_status_str(orig_exec_status),
crm_exit_str(orig_exit_status),
pcmk_exec_status_str(history->execution_status),
crm_exit_str(history->exit_status), why);
}
}
// return TRUE if start or monitor last failure but parameters changed
static bool
should_clear_for_param_change(const xmlNode *xml_op, const char *task,
pcmk_resource_t *rsc, pcmk_node_t *node)
{
if (pcmk__str_any_of(task, PCMK_ACTION_START, PCMK_ACTION_MONITOR, NULL)) {
if (pe__bundle_needs_remote_name(rsc)) {
/* We haven't allocated resources yet, so we can't reliably
* substitute addr parameters for the REMOTE_CONTAINER_HACK.
* When that's needed, defer the check until later.
*/
pe__add_param_check(xml_op, rsc, node, pcmk__check_last_failure,
rsc->priv->scheduler);
} else {
pcmk__op_digest_t *digest_data = NULL;
digest_data = rsc_action_digest_cmp(rsc, xml_op, node,
rsc->priv->scheduler);
switch (digest_data->rc) {
case pcmk__digest_unknown:
crm_trace("Resource %s history entry %s on %s"
" has no digest to compare",
rsc->id, pcmk__xe_history_key(xml_op),
node->priv->id);
break;
case pcmk__digest_match:
break;
default:
return TRUE;
}
}
}
return FALSE;
}
// Order action after fencing of remote node, given connection rsc
static void
order_after_remote_fencing(pcmk_action_t *action, pcmk_resource_t *remote_conn,
pcmk_scheduler_t *scheduler)
{
pcmk_node_t *remote_node = pcmk_find_node(scheduler, remote_conn->id);
if (remote_node) {
pcmk_action_t *fence = pe_fence_op(remote_node, NULL, TRUE, NULL,
FALSE, scheduler);
order_actions(fence, action, pcmk__ar_first_implies_then);
}
}
static bool
should_ignore_failure_timeout(const pcmk_resource_t *rsc, const char *task,
guint interval_ms, bool is_last_failure)
{
/* Clearing failures of recurring monitors has special concerns. The
* executor reports only changes in the monitor result, so if the
* monitor is still active and still getting the same failure result,
* that will go undetected after the failure is cleared.
*
* Also, the operation history will have the time when the recurring
* monitor result changed to the given code, not the time when the
* result last happened.
*
* @TODO We probably should clear such failures only when the failure
* timeout has passed since the last occurrence of the failed result.
* However we don't record that information. We could maybe approximate
* that by clearing only if there is a more recent successful monitor or
* stop result, but we don't even have that information at this point
* since we are still unpacking the resource's operation history.
*
* This is especially important for remote connection resources with a
* reconnect interval, so in that case, we skip clearing failures
* if the remote node hasn't been fenced.
*/
if ((rsc->priv->remote_reconnect_ms > 0U)
&& pcmk_is_set(rsc->priv->scheduler->flags,
pcmk__sched_fencing_enabled)
&& (interval_ms != 0)
&& pcmk__str_eq(task, PCMK_ACTION_MONITOR, pcmk__str_casei)) {
pcmk_node_t *remote_node = pcmk_find_node(rsc->priv->scheduler,
rsc->id);
if (remote_node && !pcmk_is_set(remote_node->priv->flags,
pcmk__node_remote_fenced)) {
if (is_last_failure) {
crm_info("Waiting to clear monitor failure for remote node %s"
" until fencing has occurred", rsc->id);
}
return TRUE;
}
}
return FALSE;
}
/*!
* \internal
* \brief Check operation age and schedule failure clearing when appropriate
*
* This function has two distinct purposes. The first is to check whether an
* operation history entry is expired (i.e. the resource has a failure timeout,
* the entry is older than the timeout, and the resource either has no fail
* count or its fail count is entirely older than the timeout). The second is to
* schedule fail count clearing when appropriate (i.e. the operation is expired
* and either the resource has an expired fail count or the operation is a
* last_failure for a remote connection resource with a reconnect interval,
* or the operation is a last_failure for a start or monitor operation and the
* resource's parameters have changed since the operation).
*
* \param[in,out] history Parsed action result history
*
* \return true if operation history entry is expired, otherwise false
*/
static bool
check_operation_expiry(struct action_history *history)
{
bool expired = false;
bool is_last_failure = pcmk__ends_with(history->id, "_last_failure_0");
time_t last_run = 0;
int unexpired_fail_count = 0;
const char *clear_reason = NULL;
const guint expiration_sec =
pcmk__timeout_ms2s(history->rsc->priv->failure_expiration_ms);
pcmk_scheduler_t *scheduler = history->rsc->priv->scheduler;
if (history->execution_status == PCMK_EXEC_NOT_INSTALLED) {
pcmk__rsc_trace(history->rsc,
"Resource history entry %s on %s is not expired: "
"Not Installed does not expire",
history->id, pcmk__node_name(history->node));
return false; // "Not installed" must always be cleared manually
}
if ((expiration_sec > 0)
&& (crm_element_value_epoch(history->xml, PCMK_XA_LAST_RC_CHANGE,
&last_run) == 0)) {
/* Resource has a PCMK_META_FAILURE_TIMEOUT and history entry has a
* timestamp
*/
time_t now = get_effective_time(scheduler);
time_t last_failure = 0;
// Is this particular operation history older than the failure timeout?
if ((now >= (last_run + expiration_sec))
&& !should_ignore_failure_timeout(history->rsc, history->task,
history->interval_ms,
is_last_failure)) {
expired = true;
}
// Does the resource as a whole have an unexpired fail count?
unexpired_fail_count = pe_get_failcount(history->node, history->rsc,
&last_failure,
pcmk__fc_effective,
history->xml);
// Update scheduler recheck time according to *last* failure
crm_trace("%s@%lld is %sexpired @%lld with unexpired_failures=%d "
"expiration=%s last-failure@%lld",
history->id, (long long) last_run, (expired? "" : "not "),
(long long) now, unexpired_fail_count,
pcmk__readable_interval(expiration_sec * 1000),
(long long) last_failure);
last_failure += expiration_sec + 1;
if (unexpired_fail_count && (now < last_failure)) {
pe__update_recheck_time(last_failure, scheduler,
"fail count expiration");
}
}
if (expired) {
if (pe_get_failcount(history->node, history->rsc, NULL,
pcmk__fc_default, history->xml)) {
// There is a fail count ignoring timeout
if (unexpired_fail_count == 0) {
// There is no fail count considering timeout
clear_reason = "it expired";
} else {
/* This operation is old, but there is an unexpired fail count.
* In a properly functioning cluster, this should only be
* possible if this operation is not a failure (otherwise the
* fail count should be expired too), so this is really just a
* failsafe.
*/
pcmk__rsc_trace(history->rsc,
"Resource history entry %s on %s is not "
"expired: Unexpired fail count",
history->id, pcmk__node_name(history->node));
expired = false;
}
} else if (is_last_failure
&& (history->rsc->priv->remote_reconnect_ms > 0U)) {
/* Clear any expired last failure when reconnect interval is set,
* even if there is no fail count.
*/
clear_reason = "reconnect interval is set";
}
}
if (!expired && is_last_failure
&& should_clear_for_param_change(history->xml, history->task,
history->rsc, history->node)) {
clear_reason = "resource parameters have changed";
}
if (clear_reason != NULL) {
pcmk_action_t *clear_op = NULL;
// Schedule clearing of the fail count
clear_op = pe__clear_failcount(history->rsc, history->node,
clear_reason, scheduler);
if (pcmk_is_set(scheduler->flags, pcmk__sched_fencing_enabled)
&& (history->rsc->priv->remote_reconnect_ms > 0)) {
/* If we're clearing a remote connection due to a reconnect
* interval, we want to wait until any scheduled fencing
* completes.
*
* We could limit this to remote_node->details->unclean, but at
* this point, that's always true (it won't be reliable until
* after unpack_node_history() is done).
*/
crm_info("Clearing %s failure will wait until any scheduled "
"fencing of %s completes",
history->task, history->rsc->id);
order_after_remote_fencing(clear_op, history->rsc, scheduler);
}
}
if (expired && (history->interval_ms == 0)
&& pcmk__str_eq(history->task, PCMK_ACTION_MONITOR, pcmk__str_none)) {
switch (history->exit_status) {
case PCMK_OCF_OK:
case PCMK_OCF_NOT_RUNNING:
case PCMK_OCF_RUNNING_PROMOTED:
case PCMK_OCF_DEGRADED:
case PCMK_OCF_DEGRADED_PROMOTED:
// Don't expire probes that return these values
pcmk__rsc_trace(history->rsc,
"Resource history entry %s on %s is not "
"expired: Probe result",
history->id, pcmk__node_name(history->node));
expired = false;
break;
}
}
return expired;
}
int
pe__target_rc_from_xml(const xmlNode *xml_op)
{
int target_rc = 0;
const char *key = crm_element_value(xml_op, PCMK__XA_TRANSITION_KEY);
if (key == NULL) {
return -1;
}
decode_transition_key(key, NULL, NULL, NULL, &target_rc);
return target_rc;
}
/*!
* \internal
* \brief Update a resource's state for an action result
*
* \param[in,out] history Parsed action history entry
* \param[in] exit_status Exit status to base new state on
* \param[in] last_failure Resource's last_failure entry, if known
* \param[in,out] on_fail Resource's current failure handling
*/
static void
update_resource_state(struct action_history *history, int exit_status,
const xmlNode *last_failure,
enum pcmk__on_fail *on_fail)
{
bool clear_past_failure = false;
if ((exit_status == PCMK_OCF_NOT_INSTALLED)
|| (!pcmk__is_bundled(history->rsc)
&& pcmk_xe_mask_probe_failure(history->xml))) {
history->rsc->priv->orig_role = pcmk_role_stopped;
} else if (exit_status == PCMK_OCF_NOT_RUNNING) {
clear_past_failure = true;
} else if (pcmk__str_eq(history->task, PCMK_ACTION_MONITOR,
pcmk__str_none)) {
if ((last_failure != NULL)
&& pcmk__str_eq(history->key, pcmk__xe_history_key(last_failure),
pcmk__str_none)) {
clear_past_failure = true;
}
if (history->rsc->priv->orig_role < pcmk_role_started) {
set_active(history->rsc);
}
} else if (pcmk__str_eq(history->task, PCMK_ACTION_START, pcmk__str_none)) {
history->rsc->priv->orig_role = pcmk_role_started;
clear_past_failure = true;
} else if (pcmk__str_eq(history->task, PCMK_ACTION_STOP, pcmk__str_none)) {
history->rsc->priv->orig_role = pcmk_role_stopped;
clear_past_failure = true;
} else if (pcmk__str_eq(history->task, PCMK_ACTION_PROMOTE,
pcmk__str_none)) {
history->rsc->priv->orig_role = pcmk_role_promoted;
clear_past_failure = true;
} else if (pcmk__str_eq(history->task, PCMK_ACTION_DEMOTE,
pcmk__str_none)) {
if (*on_fail == pcmk__on_fail_demote) {
/* Demote clears an error only if
* PCMK_META_ON_FAIL=PCMK_VALUE_DEMOTE
*/
clear_past_failure = true;
}
history->rsc->priv->orig_role = pcmk_role_unpromoted;
} else if (pcmk__str_eq(history->task, PCMK_ACTION_MIGRATE_FROM,
pcmk__str_none)) {
history->rsc->priv->orig_role = pcmk_role_started;
clear_past_failure = true;
} else if (pcmk__str_eq(history->task, PCMK_ACTION_MIGRATE_TO,
pcmk__str_none)) {
unpack_migrate_to_success(history);
} else if (history->rsc->priv->orig_role < pcmk_role_started) {
pcmk__rsc_trace(history->rsc, "%s active on %s",
history->rsc->id, pcmk__node_name(history->node));
set_active(history->rsc);
}
if (!clear_past_failure) {
return;
}
switch (*on_fail) {
case pcmk__on_fail_stop:
case pcmk__on_fail_ban:
case pcmk__on_fail_standby_node:
case pcmk__on_fail_fence_node:
pcmk__rsc_trace(history->rsc,
"%s (%s) is not cleared by a completed %s",
history->rsc->id, pcmk__on_fail_text(*on_fail),
history->task);
break;
case pcmk__on_fail_block:
case pcmk__on_fail_ignore:
case pcmk__on_fail_demote:
case pcmk__on_fail_restart:
case pcmk__on_fail_restart_container:
*on_fail = pcmk__on_fail_ignore;
pe__set_next_role(history->rsc, pcmk_role_unknown,
"clear past failures");
break;
case pcmk__on_fail_reset_remote:
if (history->rsc->priv->remote_reconnect_ms == 0U) {
/* With no reconnect interval, the connection is allowed to
* start again after the remote node is fenced and
* completely stopped. (With a reconnect interval, we wait
* for the failure to be cleared entirely before attempting
* to reconnect.)
*/
*on_fail = pcmk__on_fail_ignore;
pe__set_next_role(history->rsc, pcmk_role_unknown,
"clear past failures and reset remote");
}
break;
}
}
/*!
* \internal
* \brief Check whether a given history entry matters for resource state
*
* \param[in] history Parsed action history entry
*
* \return true if action can affect resource state, otherwise false
*/
static inline bool
can_affect_state(struct action_history *history)
{
return pcmk__str_any_of(history->task, PCMK_ACTION_MONITOR,
PCMK_ACTION_START, PCMK_ACTION_STOP,
PCMK_ACTION_PROMOTE, PCMK_ACTION_DEMOTE,
PCMK_ACTION_MIGRATE_TO, PCMK_ACTION_MIGRATE_FROM,
"asyncmon", NULL);
}
/*!
* \internal
* \brief Unpack execution/exit status and exit reason from a history entry
*
* \param[in,out] history Action history entry to unpack
*
* \return Standard Pacemaker return code
*/
static int
unpack_action_result(struct action_history *history)
{
if ((crm_element_value_int(history->xml, PCMK__XA_OP_STATUS,
&(history->execution_status)) < 0)
|| (history->execution_status < PCMK_EXEC_PENDING)
|| (history->execution_status > PCMK_EXEC_MAX)
|| (history->execution_status == PCMK_EXEC_CANCELLED)) {
pcmk__config_err("Ignoring resource history entry %s for %s on %s "
"with invalid " PCMK__XA_OP_STATUS " '%s'",
history->id, history->rsc->id,
pcmk__node_name(history->node),
pcmk__s(crm_element_value(history->xml,
PCMK__XA_OP_STATUS),
""));
return pcmk_rc_unpack_error;
}
if ((crm_element_value_int(history->xml, PCMK__XA_RC_CODE,
&(history->exit_status)) < 0)
|| (history->exit_status < 0) || (history->exit_status > CRM_EX_MAX)) {
pcmk__config_err("Ignoring resource history entry %s for %s on %s "
"with invalid " PCMK__XA_RC_CODE " '%s'",
history->id, history->rsc->id,
pcmk__node_name(history->node),
pcmk__s(crm_element_value(history->xml,
PCMK__XA_RC_CODE),
""));
return pcmk_rc_unpack_error;
}
history->exit_reason = crm_element_value(history->xml, PCMK_XA_EXIT_REASON);
return pcmk_rc_ok;
}
/*!
* \internal
* \brief Process an action history entry whose result expired
*
* \param[in,out] history Parsed action history entry
* \param[in] orig_exit_status Action exit status before remapping
*
* \return Standard Pacemaker return code (in particular, pcmk_rc_ok means the
* entry needs no further processing)
*/
static int
process_expired_result(struct action_history *history, int orig_exit_status)
{
if (!pcmk__is_bundled(history->rsc)
&& pcmk_xe_mask_probe_failure(history->xml)
&& (orig_exit_status != history->expected_exit_status)) {
if (history->rsc->priv->orig_role <= pcmk_role_stopped) {
history->rsc->priv->orig_role = pcmk_role_unknown;
}
crm_trace("Ignoring resource history entry %s for probe of %s on %s: "
"Masked failure expired",
history->id, history->rsc->id,
pcmk__node_name(history->node));
return pcmk_rc_ok;
}
if (history->exit_status == history->expected_exit_status) {
return pcmk_rc_undetermined; // Only failures expire
}
if (history->interval_ms == 0) {
crm_notice("Ignoring resource history entry %s for %s of %s on %s: "
"Expired failure",
history->id, history->task, history->rsc->id,
pcmk__node_name(history->node));
return pcmk_rc_ok;
}
if (history->node->details->online && !history->node->details->unclean) {
/* Reschedule the recurring action. schedule_cancel() won't work at
* this stage, so as a hacky workaround, forcibly change the restart
* digest so pcmk__check_action_config() does what we want later.
*
* @TODO We should skip this if there is a newer successful monitor.
* Also, this causes rescheduling only if the history entry
* has a PCMK__XA_OP_DIGEST (which the expire-non-blocked-failure
* scheduler regression test doesn't, but that may not be a
* realistic scenario in production).
*/
crm_notice("Rescheduling %s-interval %s of %s on %s "
"after failure expired",
pcmk__readable_interval(history->interval_ms), history->task,
history->rsc->id, pcmk__node_name(history->node));
crm_xml_add(history->xml, PCMK__XA_OP_RESTART_DIGEST,
"calculated-failure-timeout");
return pcmk_rc_ok;
}
return pcmk_rc_undetermined;
}
/*!
* \internal
* \brief Process a masked probe failure
*
* \param[in,out] history Parsed action history entry
* \param[in] orig_exit_status Action exit status before remapping
* \param[in] last_failure Resource's last_failure entry, if known
* \param[in,out] on_fail Resource's current failure handling
*/
static void
mask_probe_failure(struct action_history *history, int orig_exit_status,
const xmlNode *last_failure,
enum pcmk__on_fail *on_fail)
{
pcmk_resource_t *ban_rsc = history->rsc;
if (!pcmk_is_set(history->rsc->flags, pcmk__rsc_unique)) {
ban_rsc = uber_parent(history->rsc);
}
crm_notice("Treating probe result '%s' for %s on %s as 'not running'",
crm_exit_str(orig_exit_status), history->rsc->id,
pcmk__node_name(history->node));
update_resource_state(history, history->expected_exit_status, last_failure,
on_fail);
crm_xml_add(history->xml, PCMK_XA_UNAME, history->node->priv->name);
record_failed_op(history);
resource_location(ban_rsc, history->node, -PCMK_SCORE_INFINITY,
"masked-probe-failure", ban_rsc->priv->scheduler);
}
/*!
* \internal Check whether a given failure is for a given pending action
*
* \param[in] history Parsed history entry for pending action
* \param[in] last_failure Resource's last_failure entry, if known
*
* \return true if \p last_failure is failure of pending action in \p history,
* otherwise false
* \note Both \p history and \p last_failure must come from the same
* \c PCMK__XE_LRM_RESOURCE block, as node and resource are assumed to be
* the same.
*/
static bool
failure_is_newer(const struct action_history *history,
const xmlNode *last_failure)
{
guint failure_interval_ms = 0U;
long long failure_change = 0LL;
long long this_change = 0LL;
if (last_failure == NULL) {
return false; // Resource has no last_failure entry
}
if (!pcmk__str_eq(history->task,
crm_element_value(last_failure, PCMK_XA_OPERATION),
pcmk__str_none)) {
return false; // last_failure is for different action
}
if ((crm_element_value_ms(last_failure, PCMK_META_INTERVAL,
&failure_interval_ms) != pcmk_ok)
|| (history->interval_ms != failure_interval_ms)) {
return false; // last_failure is for action with different interval
}
if ((pcmk__scan_ll(crm_element_value(history->xml, PCMK_XA_LAST_RC_CHANGE),
&this_change, 0LL) != pcmk_rc_ok)
|| (pcmk__scan_ll(crm_element_value(last_failure,
PCMK_XA_LAST_RC_CHANGE),
&failure_change, 0LL) != pcmk_rc_ok)
|| (failure_change < this_change)) {
return false; // Failure is not known to be newer
}
return true;
}
/*!
* \internal
* \brief Update a resource's role etc. for a pending action
*
* \param[in,out] history Parsed history entry for pending action
* \param[in] last_failure Resource's last_failure entry, if known
*/
static void
process_pending_action(struct action_history *history,
const xmlNode *last_failure)
{
/* For recurring monitors, a failure is recorded only in RSC_last_failure_0,
* and there might be a RSC_monitor_INTERVAL entry with the last successful
* or pending result.
*
* If last_failure contains the failure of the pending recurring monitor
* we're processing here, and is newer, the action is no longer pending.
* (Pending results have call ID -1, which sorts last, so the last failure
* if any should be known.)
*/
if (failure_is_newer(history, last_failure)) {
return;
}
if (strcmp(history->task, PCMK_ACTION_START) == 0) {
pcmk__set_rsc_flags(history->rsc, pcmk__rsc_start_pending);
set_active(history->rsc);
} else if (strcmp(history->task, PCMK_ACTION_PROMOTE) == 0) {
history->rsc->priv->orig_role = pcmk_role_promoted;
} else if ((strcmp(history->task, PCMK_ACTION_MIGRATE_TO) == 0)
&& history->node->details->unclean) {
/* A migrate_to action is pending on a unclean source, so force a stop
* on the target.
*/
const char *migrate_target = NULL;
pcmk_node_t *target = NULL;
migrate_target = crm_element_value(history->xml,
PCMK__META_MIGRATE_TARGET);
target = pcmk_find_node(history->rsc->priv->scheduler,
migrate_target);
if (target != NULL) {
stop_action(history->rsc, target, FALSE);
}
}
if (history->rsc->priv->pending_action != NULL) {
/* There should never be multiple pending actions, but as a failsafe,
* just remember the first one processed for display purposes.
*/
return;
}
if (pcmk_is_probe(history->task, history->interval_ms)) {
/* Pending probes are currently never displayed, even if pending
* operations are requested. If we ever want to change that,
* enable the below and the corresponding part of
* native.c:native_pending_action().
*/
#if 0
history->rsc->private->pending_action = strdup("probe");
history->rsc->private->pending_node = history->node;
#endif
} else {
history->rsc->priv->pending_action = strdup(history->task);
history->rsc->priv->pending_node = history->node;
}
}
static void
unpack_rsc_op(pcmk_resource_t *rsc, pcmk_node_t *node, xmlNode *xml_op,
xmlNode **last_failure, enum pcmk__on_fail *on_fail)
{
int old_rc = 0;
bool expired = false;
pcmk_resource_t *parent = rsc;
enum rsc_role_e fail_role = pcmk_role_unknown;
enum pcmk__on_fail failure_strategy = pcmk__on_fail_restart;
struct action_history history = {
.rsc = rsc,
.node = node,
.xml = xml_op,
.execution_status = PCMK_EXEC_UNKNOWN,
};
CRM_CHECK(rsc && node && xml_op, return);
history.id = pcmk__xe_id(xml_op);
if (history.id == NULL) {
pcmk__config_err("Ignoring resource history entry for %s on %s "
"without ID", rsc->id, pcmk__node_name(node));
return;
}
// Task and interval
history.task = crm_element_value(xml_op, PCMK_XA_OPERATION);
if (history.task == NULL) {
pcmk__config_err("Ignoring resource history entry %s for %s on %s "
"without " PCMK_XA_OPERATION,
history.id, rsc->id, pcmk__node_name(node));
return;
}
crm_element_value_ms(xml_op, PCMK_META_INTERVAL, &(history.interval_ms));
if (!can_affect_state(&history)) {
pcmk__rsc_trace(rsc,
"Ignoring resource history entry %s for %s on %s "
"with irrelevant action '%s'",
history.id, rsc->id, pcmk__node_name(node),
history.task);
return;
}
if (unpack_action_result(&history) != pcmk_rc_ok) {
return; // Error already logged
}
history.expected_exit_status = pe__target_rc_from_xml(xml_op);
history.key = pcmk__xe_history_key(xml_op);
crm_element_value_int(xml_op, PCMK__XA_CALL_ID, &(history.call_id));
pcmk__rsc_trace(rsc, "Unpacking %s (%s call %d on %s): %s (%s)",
history.id, history.task, history.call_id,
pcmk__node_name(node),
pcmk_exec_status_str(history.execution_status),
crm_exit_str(history.exit_status));
if (node->details->unclean) {
pcmk__rsc_trace(rsc,
"%s is running on %s, which is unclean (further action "
"depends on value of stop's on-fail attribute)",
rsc->id, pcmk__node_name(node));
}
expired = check_operation_expiry(&history);
old_rc = history.exit_status;
remap_operation(&history, on_fail, expired);
if (expired && (process_expired_result(&history, old_rc) == pcmk_rc_ok)) {
goto done;
}
if (!pcmk__is_bundled(rsc) && pcmk_xe_mask_probe_failure(xml_op)) {
mask_probe_failure(&history, old_rc, *last_failure, on_fail);
goto done;
}
if (!pcmk_is_set(rsc->flags, pcmk__rsc_unique)) {
parent = uber_parent(rsc);
}
switch (history.execution_status) {
case PCMK_EXEC_PENDING:
process_pending_action(&history, *last_failure);
goto done;
case PCMK_EXEC_DONE:
update_resource_state(&history, history.exit_status, *last_failure,
on_fail);
goto done;
case PCMK_EXEC_NOT_INSTALLED:
unpack_failure_handling(&history, &failure_strategy, &fail_role);
if (failure_strategy == pcmk__on_fail_ignore) {
crm_warn("Cannot ignore failed %s of %s on %s: "
"Resource agent doesn't exist "
QB_XS " status=%d rc=%d id=%s",
history.task, rsc->id, pcmk__node_name(node),
history.execution_status, history.exit_status,
history.id);
/* Also for printing it as "FAILED" by marking it as
* pcmk__rsc_failed later
*/
*on_fail = pcmk__on_fail_ban;
}
resource_location(parent, node, -PCMK_SCORE_INFINITY,
"hard-error", rsc->priv->scheduler);
unpack_rsc_op_failure(&history, failure_strategy, fail_role,
last_failure, on_fail);
goto done;
case PCMK_EXEC_NOT_CONNECTED:
if (pcmk__is_pacemaker_remote_node(node)
&& pcmk_is_set(node->priv->remote->flags,
pcmk__rsc_managed)) {
/* We should never get into a situation where a managed remote
* connection resource is considered OK but a resource action
* behind the connection gets a "not connected" status. But as a
* fail-safe in case a bug or unusual circumstances do lead to
* that, ensure the remote connection is considered failed.
*/
pcmk__set_rsc_flags(node->priv->remote,
pcmk__rsc_failed|pcmk__rsc_stop_if_failed);
}
break; // Not done, do error handling
case PCMK_EXEC_ERROR:
case PCMK_EXEC_ERROR_HARD:
case PCMK_EXEC_ERROR_FATAL:
case PCMK_EXEC_TIMEOUT:
case PCMK_EXEC_NOT_SUPPORTED:
case PCMK_EXEC_INVALID:
break; // Not done, do error handling
default: // No other value should be possible at this point
break;
}
unpack_failure_handling(&history, &failure_strategy, &fail_role);
if ((failure_strategy == pcmk__on_fail_ignore)
|| ((failure_strategy == pcmk__on_fail_restart_container)
&& (strcmp(history.task, PCMK_ACTION_STOP) == 0))) {
char *last_change_s = last_change_str(xml_op);
crm_warn("Pretending failed %s (%s%s%s) of %s on %s at %s succeeded "
QB_XS " %s",
history.task, crm_exit_str(history.exit_status),
(pcmk__str_empty(history.exit_reason)? "" : ": "),
pcmk__s(history.exit_reason, ""), rsc->id,
pcmk__node_name(node), last_change_s, history.id);
free(last_change_s);
update_resource_state(&history, history.expected_exit_status,
*last_failure, on_fail);
crm_xml_add(xml_op, PCMK_XA_UNAME, node->priv->name);
pcmk__set_rsc_flags(rsc, pcmk__rsc_ignore_failure);
record_failed_op(&history);
if ((failure_strategy == pcmk__on_fail_restart_container)
&& (*on_fail <= pcmk__on_fail_restart)) {
*on_fail = failure_strategy;
}
} else {
unpack_rsc_op_failure(&history, failure_strategy, fail_role,
last_failure, on_fail);
if (history.execution_status == PCMK_EXEC_ERROR_HARD) {
uint8_t log_level = LOG_ERR;
if (history.exit_status == PCMK_OCF_NOT_INSTALLED) {
log_level = LOG_NOTICE;
}
do_crm_log(log_level,
"Preventing %s from restarting on %s because "
"of hard failure (%s%s%s) " QB_XS " %s",
parent->id, pcmk__node_name(node),
crm_exit_str(history.exit_status),
(pcmk__str_empty(history.exit_reason)? "" : ": "),
pcmk__s(history.exit_reason, ""), history.id);
resource_location(parent, node, -PCMK_SCORE_INFINITY,
"hard-error", rsc->priv->scheduler);
} else if (history.execution_status == PCMK_EXEC_ERROR_FATAL) {
pcmk__sched_err(rsc->priv->scheduler,
"Preventing %s from restarting anywhere because "
"of fatal failure (%s%s%s) " QB_XS " %s",
parent->id, crm_exit_str(history.exit_status),
(pcmk__str_empty(history.exit_reason)? "" : ": "),
pcmk__s(history.exit_reason, ""), history.id);
resource_location(parent, NULL, -PCMK_SCORE_INFINITY,
"fatal-error", rsc->priv->scheduler);
}
}
done:
pcmk__rsc_trace(rsc, "%s role on %s after %s is %s (next %s)",
rsc->id, pcmk__node_name(node), history.id,
pcmk_role_text(rsc->priv->orig_role),
pcmk_role_text(rsc->priv->next_role));
}
/*!
* \internal
* \brief Insert a node attribute with value into a \c GHashTable
*
* \param[in,out] key Key to insert (either freed or owned by
* \p user_data upon return)
* \param[in] value Value to insert (owned by \p user_data upon return)
* \param[in] user_data \c GHashTable to insert into
*/
static gboolean
insert_attr(gpointer key, gpointer value, gpointer user_data)
{
GHashTable *table = user_data;
g_hash_table_insert(table, key, value);
return TRUE;
}
static void
add_node_attrs(const xmlNode *xml_obj, pcmk_node_t *node, bool overwrite,
pcmk_scheduler_t *scheduler)
{
const char *cluster_name = NULL;
const char *dc_id = crm_element_value(scheduler->input, PCMK_XA_DC_UUID);
const pcmk_rule_input_t rule_input = {
.now = scheduler->priv->now,
};
pcmk__insert_dup(node->priv->attrs,
CRM_ATTR_UNAME, node->priv->name);
pcmk__insert_dup(node->priv->attrs, CRM_ATTR_ID, node->priv->id);
if ((scheduler->dc_node == NULL)
&& pcmk__str_eq(node->priv->id, dc_id, pcmk__str_casei)) {
scheduler->dc_node = node;
pcmk__insert_dup(node->priv->attrs,
CRM_ATTR_IS_DC, PCMK_VALUE_TRUE);
} else if (!pcmk__same_node(node, scheduler->dc_node)) {
pcmk__insert_dup(node->priv->attrs,
CRM_ATTR_IS_DC, PCMK_VALUE_FALSE);
}
cluster_name = g_hash_table_lookup(scheduler->priv->options,
PCMK_OPT_CLUSTER_NAME);
if (cluster_name) {
pcmk__insert_dup(node->priv->attrs, CRM_ATTR_CLUSTER_NAME,
cluster_name);
}
if (overwrite) {
/* @TODO Try to reorder some unpacking so that we don't need the
* overwrite argument or to unpack into a temporary table
*/
GHashTable *unpacked = pcmk__strkey_table(free, free);
pe__unpack_dataset_nvpairs(xml_obj, PCMK_XE_INSTANCE_ATTRIBUTES,
&rule_input, unpacked, NULL, scheduler);
g_hash_table_foreach_steal(unpacked, insert_attr, node->priv->attrs);
g_hash_table_destroy(unpacked);
} else {
pe__unpack_dataset_nvpairs(xml_obj, PCMK_XE_INSTANCE_ATTRIBUTES,
&rule_input, node->priv->attrs, NULL,
scheduler);
}
pe__unpack_dataset_nvpairs(xml_obj, PCMK_XE_UTILIZATION, &rule_input,
node->priv->utilization, NULL, scheduler);
if (pcmk__node_attr(node, CRM_ATTR_SITE_NAME, NULL,
pcmk__rsc_node_current) == NULL) {
const char *site_name = pcmk__node_attr(node, "site-name", NULL,
pcmk__rsc_node_current);
if (site_name) {
pcmk__insert_dup(node->priv->attrs,
CRM_ATTR_SITE_NAME, site_name);
} else if (cluster_name) {
/* Default to cluster-name if unset */
pcmk__insert_dup(node->priv->attrs,
CRM_ATTR_SITE_NAME, cluster_name);
}
}
}
static GList *
extract_operations(const char *node, const char *rsc, xmlNode * rsc_entry, gboolean active_filter)
{
int counter = -1;
int stop_index = -1;
int start_index = -1;
xmlNode *rsc_op = NULL;
GList *gIter = NULL;
GList *op_list = NULL;
GList *sorted_op_list = NULL;
/* extract operations */
op_list = NULL;
sorted_op_list = NULL;
for (rsc_op = pcmk__xe_first_child(rsc_entry, PCMK__XE_LRM_RSC_OP, NULL,
NULL);
rsc_op != NULL; rsc_op = pcmk__xe_next(rsc_op, PCMK__XE_LRM_RSC_OP)) {
crm_xml_add(rsc_op, PCMK_XA_RESOURCE, rsc);
crm_xml_add(rsc_op, PCMK_XA_UNAME, node);
op_list = g_list_prepend(op_list, rsc_op);
}
if (op_list == NULL) {
/* if there are no operations, there is nothing to do */
return NULL;
}
sorted_op_list = g_list_sort(op_list, sort_op_by_callid);
/* create active recurring operations as optional */
if (active_filter == FALSE) {
return sorted_op_list;
}
op_list = NULL;
calculate_active_ops(sorted_op_list, &start_index, &stop_index);
for (gIter = sorted_op_list; gIter != NULL; gIter = gIter->next) {
xmlNode *rsc_op = (xmlNode *) gIter->data;
counter++;
if (start_index < stop_index) {
crm_trace("Skipping %s: not active", pcmk__xe_id(rsc_entry));
break;
} else if (counter < start_index) {
crm_trace("Skipping %s: old", pcmk__xe_id(rsc_op));
continue;
}
op_list = g_list_append(op_list, rsc_op);
}
g_list_free(sorted_op_list);
return op_list;
}
GList *
find_operations(const char *rsc, const char *node, gboolean active_filter,
pcmk_scheduler_t *scheduler)
{
GList *output = NULL;
GList *intermediate = NULL;
xmlNode *tmp = NULL;
xmlNode *status = pcmk__xe_first_child(scheduler->input, PCMK_XE_STATUS,
NULL, NULL);
pcmk_node_t *this_node = NULL;
xmlNode *node_state = NULL;
CRM_CHECK(status != NULL, return NULL);
for (node_state = pcmk__xe_first_child(status, PCMK__XE_NODE_STATE, NULL,
NULL);
node_state != NULL;
node_state = pcmk__xe_next(node_state, PCMK__XE_NODE_STATE)) {
const char *uname = crm_element_value(node_state, PCMK_XA_UNAME);
if (node != NULL && !pcmk__str_eq(uname, node, pcmk__str_casei)) {
continue;
}
this_node = pcmk_find_node(scheduler, uname);
if(this_node == NULL) {
CRM_LOG_ASSERT(this_node != NULL);
continue;
} else if (pcmk__is_pacemaker_remote_node(this_node)) {
determine_remote_online_status(scheduler, this_node);
} else {
determine_online_status(node_state, this_node, scheduler);
}
if (this_node->details->online
|| pcmk_is_set(scheduler->flags, pcmk__sched_fencing_enabled)) {
/* offline nodes run no resources...
* unless stonith is enabled in which case we need to
* make sure rsc start events happen after the stonith
*/
xmlNode *lrm_rsc = NULL;
tmp = pcmk__xe_first_child(node_state, PCMK__XE_LRM, NULL,
NULL);
tmp = pcmk__xe_first_child(tmp, PCMK__XE_LRM_RESOURCES, NULL,
NULL);
for (lrm_rsc = pcmk__xe_first_child(tmp, PCMK__XE_LRM_RESOURCE,
NULL, NULL);
lrm_rsc != NULL;
lrm_rsc = pcmk__xe_next(lrm_rsc, PCMK__XE_LRM_RESOURCE)) {
const char *rsc_id = crm_element_value(lrm_rsc, PCMK_XA_ID);
if ((rsc != NULL)
&& !pcmk__str_eq(rsc_id, rsc, pcmk__str_none)) {
continue;
}
intermediate = extract_operations(uname, rsc_id, lrm_rsc, active_filter);
output = g_list_concat(output, intermediate);
}
}
}
return output;
}
diff --git a/lib/services/services_linux.c b/lib/services/services_linux.c
index ff4763a20c..2f8a46d195 100644
--- a/lib/services/services_linux.c
+++ b/lib/services/services_linux.c
@@ -1,1478 +1,1482 @@
/*
* Copyright 2010-2024 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU Lesser General Public License
* version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY.
*/
#include <crm_internal.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <sys/wait.h>
#include <errno.h>
#include <unistd.h>
#include <dirent.h>
#include <grp.h>
#include <string.h>
#include <sys/time.h>
#include <sys/resource.h>
#include "crm/crm.h"
#include "crm/common/mainloop.h"
#include "crm/services.h"
#include "crm/services_internal.h"
#include "services_private.h"
static void close_pipe(int fildes[]);
/* We have two alternative ways of handling SIGCHLD when synchronously waiting
* for spawned processes to complete. Both rely on polling a file descriptor to
* discover SIGCHLD events.
*
* If sys/signalfd.h is available (e.g. on Linux), we call signalfd() to
* generate the file descriptor. Otherwise, we use the "self-pipe trick"
* (opening a pipe and writing a byte to it when SIGCHLD is received).
*/
#ifdef HAVE_SYS_SIGNALFD_H
// signalfd() implementation
#include <sys/signalfd.h>
// Everything needed to manage SIGCHLD handling
struct sigchld_data_s {
sigset_t mask; // Signals to block now (including SIGCHLD)
sigset_t old_mask; // Previous set of blocked signals
bool ignored; // If SIGCHLD for another child has been ignored
};
// Initialize SIGCHLD data and prepare for use
static bool
sigchld_setup(struct sigchld_data_s *data)
{
sigemptyset(&(data->mask));
sigaddset(&(data->mask), SIGCHLD);
sigemptyset(&(data->old_mask));
// Block SIGCHLD (saving previous set of blocked signals to restore later)
if (sigprocmask(SIG_BLOCK, &(data->mask), &(data->old_mask)) < 0) {
crm_info("Wait for child process completion failed: %s "
QB_XS " source=sigprocmask", pcmk_rc_str(errno));
return false;
}
data->ignored = false;
return true;
}
// Get a file descriptor suitable for polling for SIGCHLD events
static int
sigchld_open(struct sigchld_data_s *data)
{
int fd;
CRM_CHECK(data != NULL, return -1);
fd = signalfd(-1, &(data->mask), SFD_NONBLOCK);
if (fd < 0) {
crm_info("Wait for child process completion failed: %s "
QB_XS " source=signalfd", pcmk_rc_str(errno));
}
return fd;
}
// Close a file descriptor returned by sigchld_open()
static void
sigchld_close(int fd)
{
if (fd > 0) {
close(fd);
}
}
// Return true if SIGCHLD was received from polled fd
static bool
sigchld_received(int fd, int pid, struct sigchld_data_s *data)
{
struct signalfd_siginfo fdsi;
ssize_t s;
if (fd < 0) {
return false;
}
s = read(fd, &fdsi, sizeof(struct signalfd_siginfo));
if (s != sizeof(struct signalfd_siginfo)) {
crm_info("Wait for child process completion failed: %s "
QB_XS " source=read", pcmk_rc_str(errno));
} else if (fdsi.ssi_signo == SIGCHLD) {
if (fdsi.ssi_pid == pid) {
return true;
} else {
/* This SIGCHLD is for another child. We have to ignore it here but
* will still need to resend it after this synchronous action has
* completed and SIGCHLD has been restored to be handled by the
* previous SIGCHLD handler, so that it will be handled.
*/
data->ignored = true;
return false;
}
}
return false;
}
// Do anything needed after done waiting for SIGCHLD
static void
sigchld_cleanup(struct sigchld_data_s *data)
{
// Restore the original set of blocked signals
if ((sigismember(&(data->old_mask), SIGCHLD) == 0)
&& (sigprocmask(SIG_UNBLOCK, &(data->mask), NULL) < 0)) {
crm_warn("Could not clean up after child process completion: %s",
pcmk_rc_str(errno));
}
// Resend any ignored SIGCHLD for other children so that they'll be handled.
if (data->ignored && kill(getpid(), SIGCHLD) != 0) {
crm_warn("Could not resend ignored SIGCHLD to ourselves: %s",
pcmk_rc_str(errno));
}
}
#else // HAVE_SYS_SIGNALFD_H not defined
// Self-pipe implementation (see above for function descriptions)
struct sigchld_data_s {
int pipe_fd[2]; // Pipe file descriptors
struct sigaction sa; // Signal handling info (with SIGCHLD)
struct sigaction old_sa; // Previous signal handling info
bool ignored; // If SIGCHLD for another child has been ignored
};
// We need a global to use in the signal handler
volatile struct sigchld_data_s *last_sigchld_data = NULL;
static void
sigchld_handler(void)
{
// We received a SIGCHLD, so trigger pipe polling
if ((last_sigchld_data != NULL)
&& (last_sigchld_data->pipe_fd[1] >= 0)
&& (write(last_sigchld_data->pipe_fd[1], "", 1) == -1)) {
crm_info("Wait for child process completion failed: %s "
QB_XS " source=write", pcmk_rc_str(errno));
}
}
static bool
sigchld_setup(struct sigchld_data_s *data)
{
int rc;
data->pipe_fd[0] = data->pipe_fd[1] = -1;
if (pipe(data->pipe_fd) == -1) {
crm_info("Wait for child process completion failed: %s "
QB_XS " source=pipe", pcmk_rc_str(errno));
return false;
}
rc = pcmk__set_nonblocking(data->pipe_fd[0]);
if (rc != pcmk_rc_ok) {
crm_info("Could not set pipe input non-blocking: %s " QB_XS " rc=%d",
pcmk_rc_str(rc), rc);
}
rc = pcmk__set_nonblocking(data->pipe_fd[1]);
if (rc != pcmk_rc_ok) {
crm_info("Could not set pipe output non-blocking: %s " QB_XS " rc=%d",
pcmk_rc_str(rc), rc);
}
// Set SIGCHLD handler
data->sa.sa_handler = (sighandler_t) sigchld_handler;
data->sa.sa_flags = 0;
sigemptyset(&(data->sa.sa_mask));
if (sigaction(SIGCHLD, &(data->sa), &(data->old_sa)) < 0) {
crm_info("Wait for child process completion failed: %s "
QB_XS " source=sigaction", pcmk_rc_str(errno));
}
data->ignored = false;
// Remember data for use in signal handler
last_sigchld_data = data;
return true;
}
static int
sigchld_open(struct sigchld_data_s *data)
{
CRM_CHECK(data != NULL, return -1);
return data->pipe_fd[0];
}
static void
sigchld_close(int fd)
{
// Pipe will be closed in sigchld_cleanup()
return;
}
static bool
sigchld_received(int fd, int pid, struct sigchld_data_s *data)
{
char ch;
if (fd < 0) {
return false;
}
// Clear out the self-pipe
while (read(fd, &ch, 1) == 1) /*omit*/;
return true;
}
static void
sigchld_cleanup(struct sigchld_data_s *data)
{
// Restore the previous SIGCHLD handler
if (sigaction(SIGCHLD, &(data->old_sa), NULL) < 0) {
crm_warn("Could not clean up after child process completion: %s",
pcmk_rc_str(errno));
}
close_pipe(data->pipe_fd);
// Resend any ignored SIGCHLD for other children so that they'll be handled.
if (data->ignored && kill(getpid(), SIGCHLD) != 0) {
crm_warn("Could not resend ignored SIGCHLD to ourselves: %s",
pcmk_rc_str(errno));
}
}
#endif
/*!
* \internal
* \brief Close the two file descriptors of a pipe
*
* \param[in,out] fildes Array of file descriptors opened by pipe()
*/
static void
close_pipe(int fildes[])
{
if (fildes[0] >= 0) {
close(fildes[0]);
fildes[0] = -1;
}
if (fildes[1] >= 0) {
close(fildes[1]);
fildes[1] = -1;
}
}
#define out_type(is_stderr) ((is_stderr)? "stderr" : "stdout")
// Maximum number of bytes of stdout or stderr we'll accept
#define MAX_OUTPUT (10 * 1024 * 1024)
static gboolean
svc_read_output(int fd, svc_action_t * op, bool is_stderr)
{
char *data = NULL;
ssize_t rc = 0;
size_t len = 0;
size_t discarded = 0;
char buf[500];
static const size_t buf_read_len = sizeof(buf) - 1;
if (fd < 0) {
crm_trace("No fd for %s", op->id);
return FALSE;
}
if (is_stderr && op->stderr_data) {
len = strlen(op->stderr_data);
data = op->stderr_data;
crm_trace("Reading %s stderr into offset %lld",
op->id, (long long) len);
} else if (is_stderr == FALSE && op->stdout_data) {
len = strlen(op->stdout_data);
data = op->stdout_data;
crm_trace("Reading %s stdout into offset %lld",
op->id, (long long) len);
} else {
crm_trace("Reading %s %s", op->id, out_type(is_stderr));
}
do {
errno = 0;
rc = read(fd, buf, buf_read_len);
if (rc > 0) {
if (len < MAX_OUTPUT) {
buf[rc] = 0;
crm_trace("Received %lld bytes of %s %s: %.80s",
(long long) rc, op->id, out_type(is_stderr), buf);
data = pcmk__realloc(data, len + rc + 1);
strcpy(data + len, buf);
len += rc;
} else {
discarded += rc;
}
} else if (errno != EINTR) { // Fatal error or EOF
rc = 0;
break;
}
} while ((rc == buf_read_len) || (rc < 0));
if (discarded > 0) {
crm_warn("Truncated %s %s to %lld bytes (discarded %lld)",
op->id, out_type(is_stderr), (long long) len,
(long long) discarded);
}
if (is_stderr) {
op->stderr_data = data;
} else {
op->stdout_data = data;
}
return rc != 0;
}
static int
dispatch_stdout(gpointer userdata)
{
svc_action_t *op = (svc_action_t *) userdata;
return svc_read_output(op->opaque->stdout_fd, op, FALSE);
}
static int
dispatch_stderr(gpointer userdata)
{
svc_action_t *op = (svc_action_t *) userdata;
return svc_read_output(op->opaque->stderr_fd, op, TRUE);
}
static void
pipe_out_done(gpointer user_data)
{
svc_action_t *op = (svc_action_t *) user_data;
crm_trace("%p", op);
op->opaque->stdout_gsource = NULL;
if (op->opaque->stdout_fd > STDOUT_FILENO) {
close(op->opaque->stdout_fd);
}
op->opaque->stdout_fd = -1;
}
static void
pipe_err_done(gpointer user_data)
{
svc_action_t *op = (svc_action_t *) user_data;
op->opaque->stderr_gsource = NULL;
if (op->opaque->stderr_fd > STDERR_FILENO) {
close(op->opaque->stderr_fd);
}
op->opaque->stderr_fd = -1;
}
static struct mainloop_fd_callbacks stdout_callbacks = {
.dispatch = dispatch_stdout,
.destroy = pipe_out_done,
};
static struct mainloop_fd_callbacks stderr_callbacks = {
.dispatch = dispatch_stderr,
.destroy = pipe_err_done,
};
static void
set_ocf_env(const char *key, const char *value, gpointer user_data)
{
if (setenv(key, value, 1) != 0) {
crm_perror(LOG_ERR, "setenv failed for key:%s and value:%s", key, value);
}
}
static void
set_ocf_env_with_prefix(gpointer key, gpointer value, gpointer user_data)
{
char buffer[500];
snprintf(buffer, sizeof(buffer), strcmp(key, "OCF_CHECK_LEVEL") != 0 ? "OCF_RESKEY_%s" : "%s", (char *)key);
set_ocf_env(buffer, value, user_data);
}
static void
set_alert_env(gpointer key, gpointer value, gpointer user_data)
{
int rc;
if (value != NULL) {
rc = setenv(key, value, 1);
} else {
rc = unsetenv(key);
}
if (rc < 0) {
crm_perror(LOG_ERR, "setenv %s=%s",
(char*)key, (value? (char*)value : ""));
} else {
crm_trace("setenv %s=%s", (char*)key, (value? (char*)value : ""));
}
}
/*!
* \internal
* \brief Add environment variables suitable for an action
*
* \param[in] op Action to use
*/
static void
add_action_env_vars(const svc_action_t *op)
{
void (*env_setter)(gpointer, gpointer, gpointer) = NULL;
if (op->agent == NULL) {
env_setter = set_alert_env; /* we deal with alert handler */
} else if (pcmk__str_eq(op->standard, PCMK_RESOURCE_CLASS_OCF, pcmk__str_casei)) {
env_setter = set_ocf_env_with_prefix;
}
if (env_setter != NULL && op->params != NULL) {
g_hash_table_foreach(op->params, env_setter, NULL);
}
if (env_setter == NULL || env_setter == set_alert_env) {
return;
}
set_ocf_env("OCF_RA_VERSION_MAJOR", PCMK_OCF_MAJOR_VERSION, NULL);
set_ocf_env("OCF_RA_VERSION_MINOR", PCMK_OCF_MINOR_VERSION, NULL);
set_ocf_env("OCF_ROOT", PCMK_OCF_ROOT, NULL);
set_ocf_env("OCF_EXIT_REASON_PREFIX", PCMK_OCF_REASON_PREFIX, NULL);
if (op->rsc) {
set_ocf_env("OCF_RESOURCE_INSTANCE", op->rsc, NULL);
}
if (op->agent != NULL) {
set_ocf_env("OCF_RESOURCE_TYPE", op->agent, NULL);
}
/* Notes: this is not added to specification yet. Sept 10,2004 */
if (op->provider != NULL) {
set_ocf_env("OCF_RESOURCE_PROVIDER", op->provider, NULL);
}
}
static void
pipe_in_single_parameter(gpointer key, gpointer value, gpointer user_data)
{
svc_action_t *op = user_data;
char *buffer = crm_strdup_printf("%s=%s\n", (char *)key, (char *) value);
size_t len = strlen(buffer);
size_t total = 0;
ssize_t ret = 0;
do {
errno = 0;
ret = write(op->opaque->stdin_fd, buffer + total, len - total);
if (ret > 0) {
total += ret;
}
} while ((errno == EINTR) && (total < len));
free(buffer);
}
/*!
* \internal
* \brief Pipe parameters in via stdin for action
*
* \param[in] op Action to use
*/
static void
pipe_in_action_stdin_parameters(const svc_action_t *op)
{
if (op->params) {
g_hash_table_foreach(op->params, pipe_in_single_parameter, (gpointer) op);
}
}
gboolean
recurring_action_timer(gpointer data)
{
svc_action_t *op = data;
crm_debug("Scheduling another invocation of %s", op->id);
/* Clean out the old result */
free(op->stdout_data);
op->stdout_data = NULL;
free(op->stderr_data);
op->stderr_data = NULL;
op->opaque->repeat_timer = 0;
services_action_async(op, NULL);
return FALSE;
}
/*!
* \internal
* \brief Finalize handling of an asynchronous operation
*
* Given a completed asynchronous operation, cancel or reschedule it as
* appropriate if recurring, call its callback if registered, stop tracking it,
* and clean it up.
*
* \param[in,out] op Operation to finalize
*
* \return Standard Pacemaker return code
* \retval EINVAL Caller supplied NULL or invalid \p op
* \retval EBUSY Uncanceled recurring action has only been cleaned up
* \retval pcmk_rc_ok Action has been freed
*
* \note If the return value is not pcmk_rc_ok, the caller is responsible for
* freeing the action.
*/
int
services__finalize_async_op(svc_action_t *op)
{
CRM_CHECK((op != NULL) && !(op->synchronous), return EINVAL);
if (op->interval_ms != 0) {
// Recurring operations must be either cancelled or rescheduled
if (op->cancel) {
services__set_cancelled(op);
cancel_recurring_action(op);
} else {
op->opaque->repeat_timer = pcmk__create_timer(op->interval_ms,
recurring_action_timer,
op);
}
}
if (op->opaque->callback != NULL) {
op->opaque->callback(op);
}
// Stop tracking the operation (as in-flight or blocked)
op->pid = 0;
services_untrack_op(op);
if ((op->interval_ms != 0) && !(op->cancel)) {
// Do not free recurring actions (they will get freed when cancelled)
services_action_cleanup(op);
return EBUSY;
}
services_action_free(op);
return pcmk_rc_ok;
}
static void
close_op_input(svc_action_t *op)
{
if (op->opaque->stdin_fd >= 0) {
close(op->opaque->stdin_fd);
}
}
static void
finish_op_output(svc_action_t *op, bool is_stderr)
{
mainloop_io_t **source;
int fd;
if (is_stderr) {
source = &(op->opaque->stderr_gsource);
fd = op->opaque->stderr_fd;
} else {
source = &(op->opaque->stdout_gsource);
fd = op->opaque->stdout_fd;
}
if (op->synchronous || *source) {
crm_trace("Finish reading %s[%d] %s",
op->id, op->pid, (is_stderr? "stderr" : "stdout"));
svc_read_output(fd, op, is_stderr);
if (op->synchronous) {
close(fd);
} else {
mainloop_del_fd(*source);
*source = NULL;
}
}
}
// Log an operation's stdout and stderr
static void
log_op_output(svc_action_t *op)
{
char *prefix = crm_strdup_printf("%s[%d] error output", op->id, op->pid);
/* The library caller has better context to know how important the output
* is, so log it at info and debug severity here. They can log it again at
* higher severity if appropriate.
*/
crm_log_output(LOG_INFO, prefix, op->stderr_data);
strcpy(prefix + strlen(prefix) - strlen("error output"), "output");
crm_log_output(LOG_DEBUG, prefix, op->stdout_data);
free(prefix);
}
// Truncate exit reasons at this many characters
#define EXIT_REASON_MAX_LEN 128
static void
parse_exit_reason_from_stderr(svc_action_t *op)
{
const char *reason_start = NULL;
const char *reason_end = NULL;
const int prefix_len = strlen(PCMK_OCF_REASON_PREFIX);
if ((op->stderr_data == NULL) ||
// Only OCF agents have exit reasons in stderr
!pcmk__str_eq(op->standard, PCMK_RESOURCE_CLASS_OCF, pcmk__str_none)) {
return;
}
// Find the last occurrence of the magic string indicating an exit reason
for (const char *cur = strstr(op->stderr_data, PCMK_OCF_REASON_PREFIX);
cur != NULL; cur = strstr(cur, PCMK_OCF_REASON_PREFIX)) {
cur += prefix_len; // Skip over magic string
reason_start = cur;
}
if ((reason_start == NULL) || (reason_start[0] == '\n')
|| (reason_start[0] == '\0')) {
return; // No or empty exit reason
}
// Exit reason goes to end of line (or end of output)
reason_end = strchr(reason_start, '\n');
if (reason_end == NULL) {
reason_end = reason_start + strlen(reason_start);
}
// Limit size of exit reason to something reasonable
if (reason_end > (reason_start + EXIT_REASON_MAX_LEN)) {
reason_end = reason_start + EXIT_REASON_MAX_LEN;
}
free(op->opaque->exit_reason);
op->opaque->exit_reason = strndup(reason_start, reason_end - reason_start);
}
/*!
* \internal
* \brief Process the completion of an asynchronous child process
*
* \param[in,out] p Child process that completed
* \param[in] pid Process ID of child
* \param[in] core (Unused)
* \param[in] signo Signal that interrupted child, if any
* \param[in] exitcode Exit status of child process
*/
static void
async_action_complete(mainloop_child_t *p, pid_t pid, int core, int signo,
int exitcode)
{
svc_action_t *op = mainloop_child_userdata(p);
mainloop_clear_child_userdata(p);
CRM_CHECK(op->pid == pid,
services__set_result(op, services__generic_error(op),
PCMK_EXEC_ERROR, "Bug in mainloop handling");
return);
/* Depending on the priority the mainloop gives the stdout and stderr
* file descriptors, this function could be called before everything has
* been read from them, so force a final read now.
*/
finish_op_output(op, true);
finish_op_output(op, false);
close_op_input(op);
if (signo == 0) {
crm_debug("%s[%d] exited with status %d", op->id, op->pid, exitcode);
services__set_result(op, exitcode, PCMK_EXEC_DONE, NULL);
log_op_output(op);
parse_exit_reason_from_stderr(op);
} else if (mainloop_child_timeout(p)) {
const char *kind = services__action_kind(op);
crm_info("%s %s[%d] timed out after %s",
kind, op->id, op->pid, pcmk__readable_interval(op->timeout));
services__format_result(op, services__generic_error(op),
PCMK_EXEC_TIMEOUT,
"%s did not complete within %s",
kind, pcmk__readable_interval(op->timeout));
} else if (op->cancel) {
/* If an in-flight recurring operation was killed because it was
* cancelled, don't treat that as a failure.
*/
crm_info("%s[%d] terminated with signal %d (%s)",
op->id, op->pid, signo, strsignal(signo));
services__set_result(op, PCMK_OCF_OK, PCMK_EXEC_CANCELLED, NULL);
} else {
crm_info("%s[%d] terminated with signal %d (%s)",
op->id, op->pid, signo, strsignal(signo));
services__format_result(op, PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_ERROR,
"%s interrupted by %s signal",
services__action_kind(op), strsignal(signo));
}
services__finalize_async_op(op);
}
/*!
* \internal
* \brief Return agent standard's exit status for "generic error"
*
* When returning an internal error for an action, a value that is appropriate
* to the action's agent standard must be used. This function returns a value
* appropriate for errors in general.
*
* \param[in] op Action that error is for
*
* \return Exit status appropriate to agent standard
* \note Actions without a standard will get PCMK_OCF_UNKNOWN_ERROR.
*/
int
services__generic_error(const svc_action_t *op)
{
if ((op == NULL) || (op->standard == NULL)) {
return PCMK_OCF_UNKNOWN_ERROR;
}
#if PCMK__ENABLE_LSB
if (pcmk__str_eq(op->standard, PCMK_RESOURCE_CLASS_LSB, pcmk__str_casei)
&& pcmk__str_eq(op->action, PCMK_ACTION_STATUS, pcmk__str_casei)) {
return PCMK_LSB_STATUS_UNKNOWN;
}
#endif
return PCMK_OCF_UNKNOWN_ERROR;
}
/*!
* \internal
* \brief Return agent standard's exit status for "not installed"
*
* When returning an internal error for an action, a value that is appropriate
* to the action's agent standard must be used. This function returns a value
* appropriate for "not installed" errors.
*
* \param[in] op Action that error is for
*
* \return Exit status appropriate to agent standard
* \note Actions without a standard will get PCMK_OCF_UNKNOWN_ERROR.
*/
int
services__not_installed_error(const svc_action_t *op)
{
if ((op == NULL) || (op->standard == NULL)) {
return PCMK_OCF_UNKNOWN_ERROR;
}
#if PCMK__ENABLE_LSB
if (pcmk__str_eq(op->standard, PCMK_RESOURCE_CLASS_LSB, pcmk__str_casei)
&& pcmk__str_eq(op->action, PCMK_ACTION_STATUS, pcmk__str_casei)) {
return PCMK_LSB_STATUS_NOT_INSTALLED;
}
#endif
return PCMK_OCF_NOT_INSTALLED;
}
/*!
* \internal
* \brief Return agent standard's exit status for "insufficient privileges"
*
* When returning an internal error for an action, a value that is appropriate
* to the action's agent standard must be used. This function returns a value
* appropriate for "insufficient privileges" errors.
*
* \param[in] op Action that error is for
*
* \return Exit status appropriate to agent standard
* \note Actions without a standard will get PCMK_OCF_UNKNOWN_ERROR.
*/
int
services__authorization_error(const svc_action_t *op)
{
if ((op == NULL) || (op->standard == NULL)) {
return PCMK_OCF_UNKNOWN_ERROR;
}
#if PCMK__ENABLE_LSB
if (pcmk__str_eq(op->standard, PCMK_RESOURCE_CLASS_LSB, pcmk__str_casei)
&& pcmk__str_eq(op->action, PCMK_ACTION_STATUS, pcmk__str_casei)) {
return PCMK_LSB_STATUS_INSUFFICIENT_PRIV;
}
#endif
return PCMK_OCF_INSUFFICIENT_PRIV;
}
/*!
* \internal
* \brief Return agent standard's exit status for "not configured"
*
* When returning an internal error for an action, a value that is appropriate
* to the action's agent standard must be used. This function returns a value
* appropriate for "not configured" errors.
*
* \param[in] op Action that error is for
* \param[in] is_fatal Whether problem is cluster-wide instead of only local
*
* \return Exit status appropriate to agent standard
* \note Actions without a standard will get PCMK_OCF_UNKNOWN_ERROR.
*/
int
services__configuration_error(const svc_action_t *op, bool is_fatal)
{
if ((op == NULL) || (op->standard == NULL)) {
return PCMK_OCF_UNKNOWN_ERROR;
}
#if PCMK__ENABLE_LSB
if (pcmk__str_eq(op->standard, PCMK_RESOURCE_CLASS_LSB, pcmk__str_casei)
&& pcmk__str_eq(op->action, PCMK_ACTION_STATUS, pcmk__str_casei)) {
return PCMK_LSB_NOT_CONFIGURED;
}
#endif
return is_fatal? PCMK_OCF_NOT_CONFIGURED : PCMK_OCF_INVALID_PARAM;
}
/*!
* \internal
* \brief Set operation rc and status per errno from stat(), fork() or execvp()
*
* \param[in,out] op Operation to set rc and status for
* \param[in] error Value of errno after system call
*
* \return void
*/
void
services__handle_exec_error(svc_action_t * op, int error)
{
const char *name = op->opaque->exec;
if (name == NULL) {
name = op->agent;
if (name == NULL) {
name = op->id;
}
}
switch (error) { /* see execve(2), stat(2) and fork(2) */
case ENOENT: /* No such file or directory */
case EISDIR: /* Is a directory */
case ENOTDIR: /* Path component is not a directory */
case EINVAL: /* Invalid executable format */
case ENOEXEC: /* Invalid executable format */
services__format_result(op, services__not_installed_error(op),
PCMK_EXEC_NOT_INSTALLED, "%s: %s",
name, pcmk_rc_str(error));
break;
case EACCES: /* permission denied (various errors) */
case EPERM: /* permission denied (various errors) */
services__format_result(op, services__authorization_error(op),
PCMK_EXEC_ERROR, "%s: %s",
name, pcmk_rc_str(error));
break;
default:
services__set_result(op, services__generic_error(op),
PCMK_EXEC_ERROR, pcmk_rc_str(error));
}
}
/*!
* \internal
* \brief Exit a child process that failed before executing agent
*
* \param[in] op Action that failed
* \param[in] exit_status Exit status code to use
* \param[in] exit_reason Exit reason to output if for OCF agent
*/
static void
exit_child(const svc_action_t *op, int exit_status, const char *exit_reason)
{
if ((op != NULL) && (exit_reason != NULL)
&& pcmk__str_eq(op->standard, PCMK_RESOURCE_CLASS_OCF,
pcmk__str_none)) {
fprintf(stderr, PCMK_OCF_REASON_PREFIX "%s\n", exit_reason);
}
pcmk_common_cleanup();
_exit(exit_status);
}
static void
action_launch_child(svc_action_t *op)
{
int rc;
/* SIGPIPE is ignored (which is different from signal blocking) by the gnutls library.
* Depending on the libqb version in use, libqb may set SIGPIPE to be ignored as well.
* We do not want this to be inherited by the child process. By resetting this the signal
* to the default behavior, we avoid some potential odd problems that occur during OCF
* scripts when SIGPIPE is ignored by the environment. */
signal(SIGPIPE, SIG_DFL);
if (sched_getscheduler(0) != SCHED_OTHER) {
struct sched_param sp;
memset(&sp, 0, sizeof(sp));
sp.sched_priority = 0;
if (sched_setscheduler(0, SCHED_OTHER, &sp) == -1) {
crm_info("Could not reset scheduling policy for %s", op->id);
}
}
if (setpriority(PRIO_PROCESS, 0, 0) == -1) {
crm_info("Could not reset process priority for %s", op->id);
}
/* Man: The call setpgrp() is equivalent to setpgid(0,0)
* _and_ compiles on BSD variants too
* need to investigate if it works the same too.
*/
setpgid(0, 0);
pcmk__close_fds_in_child(false);
/* It would be nice if errors in this function could be reported as
* execution status (for example, PCMK_EXEC_NO_SECRETS for the secrets error
* below) instead of exit status. However, we've already forked, so
* exit status is all we have. At least for OCF actions, we can output an
* exit reason for the parent to parse.
+ *
+ * @TODO It might be better to substitute secrets in the parent before
+ * forking, so that if it fails, we can give a better message and result,
+ * and avoid the fork.
*/
#if PCMK__ENABLE_CIBSECRETS
rc = pcmk__substitute_secrets(op->rsc, op->params);
if (rc != pcmk_rc_ok) {
if (pcmk__str_eq(op->action, PCMK_ACTION_STOP, pcmk__str_casei)) {
crm_info("Proceeding with stop operation for %s "
"despite being unable to load CIB secrets (%s)",
op->rsc, pcmk_rc_str(rc));
} else {
crm_err("Considering %s unconfigured "
"because unable to load CIB secrets: %s",
op->rsc, pcmk_rc_str(rc));
exit_child(op, services__configuration_error(op, false),
"Unable to load CIB secrets");
}
}
#endif
add_action_env_vars(op);
/* Become the desired user */
if (op->opaque->uid && (geteuid() == 0)) {
// If requested, set effective group
if (op->opaque->gid && (setgid(op->opaque->gid) < 0)) {
crm_err("Considering %s unauthorized because could not set "
"child group to %d: %s",
op->id, op->opaque->gid, strerror(errno));
exit_child(op, services__authorization_error(op),
"Could not set group for child process");
}
// Erase supplementary group list
// (We could do initgroups() if we kept a copy of the username)
if (setgroups(0, NULL) < 0) {
crm_err("Considering %s unauthorized because could not "
"clear supplementary groups: %s", op->id, strerror(errno));
exit_child(op, services__authorization_error(op),
"Could not clear supplementary groups for child process");
}
// Set effective user
if (setuid(op->opaque->uid) < 0) {
crm_err("Considering %s unauthorized because could not set user "
"to %d: %s", op->id, op->opaque->uid, strerror(errno));
exit_child(op, services__authorization_error(op),
"Could not set user for child process");
}
}
// Execute the agent (doesn't return if successful)
execvp(op->opaque->exec, op->opaque->args);
// An earlier stat() should have avoided most possible errors
rc = errno;
services__handle_exec_error(op, rc);
crm_err("Unable to execute %s: %s", op->id, strerror(rc));
exit_child(op, op->rc, "Child process was unable to execute file");
}
/*!
* \internal
* \brief Wait for synchronous action to complete, and set its result
*
* \param[in,out] op Action to wait for
* \param[in,out] data Child signal data
*/
static void
wait_for_sync_result(svc_action_t *op, struct sigchld_data_s *data)
{
int status = 0;
int timeout = op->timeout;
time_t start = time(NULL);
struct pollfd fds[3];
int wait_rc = 0;
const char *wait_reason = NULL;
fds[0].fd = op->opaque->stdout_fd;
fds[0].events = POLLIN;
fds[0].revents = 0;
fds[1].fd = op->opaque->stderr_fd;
fds[1].events = POLLIN;
fds[1].revents = 0;
fds[2].fd = sigchld_open(data);
fds[2].events = POLLIN;
fds[2].revents = 0;
crm_trace("Waiting for %s[%d]", op->id, op->pid);
do {
int poll_rc = poll(fds, 3, timeout);
wait_reason = NULL;
if (poll_rc > 0) {
if (fds[0].revents & POLLIN) {
svc_read_output(op->opaque->stdout_fd, op, FALSE);
}
if (fds[1].revents & POLLIN) {
svc_read_output(op->opaque->stderr_fd, op, TRUE);
}
if ((fds[2].revents & POLLIN)
&& sigchld_received(fds[2].fd, op->pid, data)) {
wait_rc = waitpid(op->pid, &status, WNOHANG);
if ((wait_rc > 0) || ((wait_rc < 0) && (errno == ECHILD))) {
// Child process exited or doesn't exist
break;
} else if (wait_rc < 0) {
wait_reason = pcmk_rc_str(errno);
crm_info("Wait for completion of %s[%d] failed: %s "
QB_XS " source=waitpid",
op->id, op->pid, wait_reason);
wait_rc = 0; // Act as if process is still running
#ifndef HAVE_SYS_SIGNALFD_H
} else {
/* The child hasn't exited, so this SIGCHLD could be for
* another child. We have to ignore it here but will still
* need to resend it after this synchronous action has
* completed and SIGCHLD has been restored to be handled by
* the previous handler, so that it will be handled.
*/
data->ignored = true;
#endif
}
}
} else if (poll_rc == 0) {
// Poll timed out with no descriptors ready
timeout = 0;
break;
} else if ((poll_rc < 0) && (errno != EINTR)) {
wait_reason = pcmk_rc_str(errno);
crm_info("Wait for completion of %s[%d] failed: %s "
QB_XS " source=poll", op->id, op->pid, wait_reason);
break;
}
timeout = op->timeout - (time(NULL) - start) * 1000;
} while ((op->timeout < 0 || timeout > 0));
crm_trace("Stopped waiting for %s[%d]", op->id, op->pid);
finish_op_output(op, true);
finish_op_output(op, false);
close_op_input(op);
sigchld_close(fds[2].fd);
if (wait_rc <= 0) {
if ((op->timeout > 0) && (timeout <= 0)) {
services__format_result(op, services__generic_error(op),
PCMK_EXEC_TIMEOUT,
"%s did not exit within specified timeout",
services__action_kind(op));
crm_info("%s[%d] timed out after %dms",
op->id, op->pid, op->timeout);
} else {
services__set_result(op, services__generic_error(op),
PCMK_EXEC_ERROR, wait_reason);
}
/* If only child hasn't been successfully waited for, yet.
This is to limit killing wrong target a bit more. */
if ((wait_rc == 0) && (waitpid(op->pid, &status, WNOHANG) == 0)) {
if (kill(op->pid, SIGKILL)) {
crm_warn("Could not kill rogue child %s[%d]: %s",
op->id, op->pid, pcmk_rc_str(errno));
}
/* Safe to skip WNOHANG here as we sent non-ignorable signal. */
while ((waitpid(op->pid, &status, 0) == (pid_t) -1)
&& (errno == EINTR)) {
/* keep waiting */;
}
}
} else if (WIFEXITED(status)) {
services__set_result(op, WEXITSTATUS(status), PCMK_EXEC_DONE, NULL);
parse_exit_reason_from_stderr(op);
crm_info("%s[%d] exited with status %d", op->id, op->pid, op->rc);
} else if (WIFSIGNALED(status)) {
int signo = WTERMSIG(status);
services__format_result(op, services__generic_error(op),
PCMK_EXEC_ERROR, "%s interrupted by %s signal",
services__action_kind(op), strsignal(signo));
crm_info("%s[%d] terminated with signal %d (%s)",
op->id, op->pid, signo, strsignal(signo));
#ifdef WCOREDUMP
if (WCOREDUMP(status)) {
crm_warn("%s[%d] dumped core", op->id, op->pid);
}
#endif
} else {
// Shouldn't be possible to get here
services__set_result(op, services__generic_error(op), PCMK_EXEC_ERROR,
"Unable to wait for child to complete");
}
}
/*!
* \internal
* \brief Execute an action whose standard uses executable files
*
* \param[in,out] op Action to execute
*
* \return Standard Pacemaker return value
* \retval EBUSY Recurring operation could not be initiated
* \retval pcmk_rc_error Synchronous action failed
* \retval pcmk_rc_ok Synchronous action succeeded, or asynchronous action
* should not be freed (because it's pending or because
* it failed to execute and was already freed)
*
* \note If the return value for an asynchronous action is not pcmk_rc_ok, the
* caller is responsible for freeing the action.
*/
int
services__execute_file(svc_action_t *op)
{
int stdout_fd[2];
int stderr_fd[2];
int stdin_fd[2] = {-1, -1};
int rc;
struct stat st;
struct sigchld_data_s data = { .ignored = false };
// Catch common failure conditions early
if (stat(op->opaque->exec, &st) != 0) {
rc = errno;
crm_info("Cannot execute '%s': %s " QB_XS " stat rc=%d",
op->opaque->exec, pcmk_rc_str(rc), rc);
services__handle_exec_error(op, rc);
goto done;
}
if (pipe(stdout_fd) < 0) {
rc = errno;
crm_info("Cannot execute '%s': %s " QB_XS " pipe(stdout) rc=%d",
op->opaque->exec, pcmk_rc_str(rc), rc);
services__handle_exec_error(op, rc);
goto done;
}
if (pipe(stderr_fd) < 0) {
rc = errno;
close_pipe(stdout_fd);
crm_info("Cannot execute '%s': %s " QB_XS " pipe(stderr) rc=%d",
op->opaque->exec, pcmk_rc_str(rc), rc);
services__handle_exec_error(op, rc);
goto done;
}
if (pcmk_is_set(pcmk_get_ra_caps(op->standard), pcmk_ra_cap_stdin)) {
if (pipe(stdin_fd) < 0) {
rc = errno;
close_pipe(stdout_fd);
close_pipe(stderr_fd);
crm_info("Cannot execute '%s': %s " QB_XS " pipe(stdin) rc=%d",
op->opaque->exec, pcmk_rc_str(rc), rc);
services__handle_exec_error(op, rc);
goto done;
}
}
if (op->synchronous && !sigchld_setup(&data)) {
close_pipe(stdin_fd);
close_pipe(stdout_fd);
close_pipe(stderr_fd);
sigchld_cleanup(&data);
services__set_result(op, services__generic_error(op), PCMK_EXEC_ERROR,
"Could not manage signals for child process");
goto done;
}
op->pid = fork();
switch (op->pid) {
case -1:
rc = errno;
close_pipe(stdin_fd);
close_pipe(stdout_fd);
close_pipe(stderr_fd);
crm_info("Cannot execute '%s': %s " QB_XS " fork rc=%d",
op->opaque->exec, pcmk_rc_str(rc), rc);
services__handle_exec_error(op, rc);
if (op->synchronous) {
sigchld_cleanup(&data);
}
goto done;
break;
case 0: /* Child */
close(stdout_fd[0]);
close(stderr_fd[0]);
if (stdin_fd[1] >= 0) {
close(stdin_fd[1]);
}
if (STDOUT_FILENO != stdout_fd[1]) {
if (dup2(stdout_fd[1], STDOUT_FILENO) != STDOUT_FILENO) {
crm_warn("Can't redirect output from '%s': %s "
QB_XS " errno=%d",
op->opaque->exec, pcmk_rc_str(errno), errno);
}
close(stdout_fd[1]);
}
if (STDERR_FILENO != stderr_fd[1]) {
if (dup2(stderr_fd[1], STDERR_FILENO) != STDERR_FILENO) {
crm_warn("Can't redirect error output from '%s': %s "
QB_XS " errno=%d",
op->opaque->exec, pcmk_rc_str(errno), errno);
}
close(stderr_fd[1]);
}
if ((stdin_fd[0] >= 0) &&
(STDIN_FILENO != stdin_fd[0])) {
if (dup2(stdin_fd[0], STDIN_FILENO) != STDIN_FILENO) {
crm_warn("Can't redirect input to '%s': %s "
QB_XS " errno=%d",
op->opaque->exec, pcmk_rc_str(errno), errno);
}
close(stdin_fd[0]);
}
if (op->synchronous) {
sigchld_cleanup(&data);
}
action_launch_child(op);
pcmk__assert(false); // action_launch_child() should not return
}
/* Only the parent reaches here */
close(stdout_fd[1]);
close(stderr_fd[1]);
if (stdin_fd[0] >= 0) {
close(stdin_fd[0]);
}
op->opaque->stdout_fd = stdout_fd[0];
rc = pcmk__set_nonblocking(op->opaque->stdout_fd);
if (rc != pcmk_rc_ok) {
crm_info("Could not set '%s' output non-blocking: %s "
QB_XS " rc=%d",
op->opaque->exec, pcmk_rc_str(rc), rc);
}
op->opaque->stderr_fd = stderr_fd[0];
rc = pcmk__set_nonblocking(op->opaque->stderr_fd);
if (rc != pcmk_rc_ok) {
crm_info("Could not set '%s' error output non-blocking: %s "
QB_XS " rc=%d",
op->opaque->exec, pcmk_rc_str(rc), rc);
}
op->opaque->stdin_fd = stdin_fd[1];
if (op->opaque->stdin_fd >= 0) {
// using buffer behind non-blocking-fd here - that could be improved
// as long as no other standard uses stdin_fd assume stonith
rc = pcmk__set_nonblocking(op->opaque->stdin_fd);
if (rc != pcmk_rc_ok) {
crm_info("Could not set '%s' input non-blocking: %s "
QB_XS " fd=%d,rc=%d", op->opaque->exec,
pcmk_rc_str(rc), op->opaque->stdin_fd, rc);
}
pipe_in_action_stdin_parameters(op);
// as long as we are handling parameters directly in here just close
close(op->opaque->stdin_fd);
op->opaque->stdin_fd = -1;
}
// after fds are setup properly and before we plug anything into mainloop
if (op->opaque->fork_callback) {
op->opaque->fork_callback(op);
}
if (op->synchronous) {
wait_for_sync_result(op, &data);
sigchld_cleanup(&data);
goto done;
}
crm_trace("Waiting async for '%s'[%d]", op->opaque->exec, op->pid);
mainloop_child_add_with_flags(op->pid, op->timeout, op->id, op,
pcmk_is_set(op->flags, SVC_ACTION_LEAVE_GROUP)? mainloop_leave_pid_group : 0,
async_action_complete);
op->opaque->stdout_gsource = mainloop_add_fd(op->id,
G_PRIORITY_LOW,
op->opaque->stdout_fd, op,
&stdout_callbacks);
op->opaque->stderr_gsource = mainloop_add_fd(op->id,
G_PRIORITY_LOW,
op->opaque->stderr_fd, op,
&stderr_callbacks);
services_add_inflight_op(op);
return pcmk_rc_ok;
done:
if (op->synchronous) {
return (op->rc == PCMK_OCF_OK)? pcmk_rc_ok : pcmk_rc_error;
} else {
return services__finalize_async_op(op);
}
}
GList *
services_os_get_single_directory_list(const char *root, gboolean files, gboolean executable)
{
GList *list = NULL;
struct dirent **namelist;
int entries = 0, lpc = 0;
char buffer[PATH_MAX];
entries = scandir(root, &namelist, NULL, alphasort);
if (entries <= 0) {
return list;
}
for (lpc = 0; lpc < entries; lpc++) {
struct stat sb;
if ('.' == namelist[lpc]->d_name[0]) {
free(namelist[lpc]);
continue;
}
snprintf(buffer, sizeof(buffer), "%s/%s", root, namelist[lpc]->d_name);
if (stat(buffer, &sb)) {
continue;
}
if (S_ISDIR(sb.st_mode)) {
if (files) {
free(namelist[lpc]);
continue;
}
} else if (S_ISREG(sb.st_mode)) {
if (files == FALSE) {
free(namelist[lpc]);
continue;
} else if (executable
&& (sb.st_mode & S_IXUSR) == 0
&& (sb.st_mode & S_IXGRP) == 0 && (sb.st_mode & S_IXOTH) == 0) {
free(namelist[lpc]);
continue;
}
}
list = g_list_append(list, strdup(namelist[lpc]->d_name));
free(namelist[lpc]);
}
free(namelist);
return list;
}
GList *
services_os_get_directory_list(const char *root, gboolean files, gboolean executable)
{
GList *result = NULL;
char *dirs = strdup(root);
char *dir = NULL;
if (pcmk__str_empty(dirs)) {
free(dirs);
return result;
}
for (dir = strtok(dirs, ":"); dir != NULL; dir = strtok(NULL, ":")) {
GList *tmp = services_os_get_single_directory_list(dir, files, executable);
if (tmp) {
result = g_list_concat(result, tmp);
}
}
free(dirs);
return result;
}