Page Menu
Home
ClusterLabs Projects
Search
Configure Global Search
Log In
Files
F4512108
No One
Temporary
Actions
View File
Edit File
Delete File
View Transforms
Subscribe
Mute Notifications
Flag For Later
Award Token
Size
206 KB
Referenced Files
None
Subscribers
None
View Options
diff --git a/daemons/schedulerd/pacemaker-schedulerd.c b/daemons/schedulerd/pacemaker-schedulerd.c
index cee938fecd..4fc6aaa7d1 100644
--- a/daemons/schedulerd/pacemaker-schedulerd.c
+++ b/daemons/schedulerd/pacemaker-schedulerd.c
@@ -1,359 +1,360 @@
/*
* Copyright 2004-2020 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 <stdio.h>
+#include <stdbool.h>
#include <sys/stat.h>
#include <sys/types.h>
#include <unistd.h>
#include <stdlib.h>
#include <errno.h>
#include <fcntl.h>
#include <libxml/parser.h>
#include <crm/common/ipc_internal.h>
#include <crm/common/mainloop.h>
#include <crm/pengine/internal.h>
#include <pacemaker-internal.h>
#include <crm/msg_xml.h>
#define OPTARGS "hVc"
static GMainLoop *mainloop = NULL;
static qb_ipcs_service_t *ipcs = NULL;
static pe_working_set_t *sched_data_set = NULL;
#define get_series() was_processing_error?1:was_processing_warning?2:3
typedef struct series_s {
const char *name;
const char *param;
int wrap;
} series_t;
series_t series[] = {
{"pe-unknown", "_do_not_match_anything_", -1},
{"pe-error", "pe-error-series-max", -1},
{"pe-warn", "pe-warn-series-max", 200},
{"pe-input", "pe-input-series-max", 400},
};
void pengine_shutdown(int nsig);
static gboolean
process_pe_message(xmlNode *msg, xmlNode *xml_data, pcmk__client_t *sender)
{
static char *last_digest = NULL;
static char *filename = NULL;
const char *sys_to = crm_element_value(msg, F_CRM_SYS_TO);
const char *op = crm_element_value(msg, F_CRM_TASK);
const char *ref = crm_element_value(msg, F_CRM_REFERENCE);
crm_trace("Processing %s op (ref=%s)...", op, ref);
if (op == NULL) {
/* error */
} else if (strcasecmp(op, CRM_OP_HELLO) == 0) {
/* ignore */
} else if (pcmk__str_eq(crm_element_value(msg, F_CRM_MSG_TYPE), XML_ATTR_RESPONSE, pcmk__str_casei)) {
/* ignore */
} else if (sys_to == NULL || strcasecmp(sys_to, CRM_SYSTEM_PENGINE) != 0) {
crm_trace("Bad sys-to %s", crm_str(sys_to));
return FALSE;
} else if (strcasecmp(op, CRM_OP_PECALC) == 0) {
unsigned int seq;
int series_id = 0;
int series_wrap = 0;
char *digest = NULL;
const char *value = NULL;
time_t execution_date = time(NULL);
xmlNode *converted = NULL;
xmlNode *reply = NULL;
gboolean is_repoke = FALSE;
gboolean process = TRUE;
- crm_config_error = FALSE;
- crm_config_warning = FALSE;
+ pcmk__config_error = false;
+ pcmk__config_warning = false;
was_processing_error = FALSE;
was_processing_warning = FALSE;
if (sched_data_set == NULL) {
sched_data_set = pe_new_working_set();
CRM_ASSERT(sched_data_set != NULL);
pe__set_working_set_flags(sched_data_set,
pe_flag_no_counts|pe_flag_no_compat);
}
digest = calculate_xml_versioned_digest(xml_data, FALSE, FALSE, CRM_FEATURE_SET);
converted = copy_xml(xml_data);
if (cli_config_update(&converted, NULL, TRUE) == FALSE) {
sched_data_set->graph = create_xml_node(NULL, XML_TAG_GRAPH);
crm_xml_add_int(sched_data_set->graph, "transition_id", 0);
crm_xml_add_int(sched_data_set->graph, "cluster-delay", 0);
process = FALSE;
free(digest);
} else if (pcmk__str_eq(digest, last_digest, pcmk__str_casei)) {
crm_info("Input has not changed since last time, not saving to disk");
is_repoke = TRUE;
free(digest);
} else {
free(last_digest);
last_digest = digest;
}
if (process) {
pcmk__schedule_actions(sched_data_set, converted, NULL);
}
series_id = get_series();
series_wrap = series[series_id].wrap;
value = pe_pref(sched_data_set->config_hash, series[series_id].param);
if (value != NULL) {
series_wrap = (int) crm_parse_ll(value, NULL);
if (errno != 0) {
series_wrap = series[series_id].wrap;
}
} else {
pcmk__config_warn("No value specified for cluster preference: %s",
series[series_id].param);
}
if (pcmk__read_series_sequence(PE_STATE_DIR, series[series_id].name,
&seq) != pcmk_rc_ok) {
// @TODO maybe handle errors better ...
seq = 0;
}
crm_trace("Series %s: wrap=%d, seq=%u, pref=%s",
series[series_id].name, series_wrap, seq, value);
sched_data_set->input = NULL;
reply = create_reply(msg, sched_data_set->graph);
CRM_ASSERT(reply != NULL);
if (is_repoke == FALSE) {
free(filename);
filename = pcmk__series_filename(PE_STATE_DIR,
series[series_id].name, seq, true);
}
crm_xml_add(reply, F_CRM_TGRAPH_INPUT, filename);
crm_xml_add_int(reply, "graph-errors", was_processing_error);
crm_xml_add_int(reply, "graph-warnings", was_processing_warning);
- crm_xml_add_int(reply, "config-errors", crm_config_error);
- crm_xml_add_int(reply, "config-warnings", crm_config_warning);
+ crm_xml_add_int(reply, "config-errors", pcmk__config_error);
+ crm_xml_add_int(reply, "config-warnings", pcmk__config_warning);
if (pcmk__ipc_send_xml(sender, 0, reply,
crm_ipc_server_event) != pcmk_rc_ok) {
int graph_file_fd = 0;
char *graph_file = NULL;
umask(S_IWGRP | S_IWOTH | S_IROTH);
graph_file = crm_strdup_printf("%s/pengine.graph.XXXXXX",
PE_STATE_DIR);
graph_file_fd = mkstemp(graph_file);
crm_err("Couldn't send transition graph to peer, writing to %s instead",
graph_file);
crm_xml_add(reply, F_CRM_TGRAPH, graph_file);
write_xml_fd(sched_data_set->graph, graph_file, graph_file_fd, FALSE);
free(graph_file);
free_xml(first_named_child(reply, F_CRM_DATA));
CRM_ASSERT(pcmk__ipc_send_xml(sender, 0, reply,
crm_ipc_server_event) == pcmk_rc_ok);
}
free_xml(reply);
pe_reset_working_set(sched_data_set);
pcmk__log_transition_summary(filename);
if (is_repoke == FALSE && series_wrap != 0) {
unlink(filename);
crm_xml_add_ll(xml_data, "execution-date", (long long) execution_date);
write_xml_file(xml_data, filename, TRUE);
pcmk__write_series_sequence(PE_STATE_DIR, series[series_id].name,
++seq, series_wrap);
} else {
crm_trace("Not writing out %s: %d & %d", filename, is_repoke, series_wrap);
}
free_xml(converted);
}
return TRUE;
}
static int32_t
pe_ipc_accept(qb_ipcs_connection_t * c, uid_t uid, gid_t gid)
{
crm_trace("Connection %p", c);
if (pcmk__new_client(c, uid, gid) == NULL) {
return -EIO;
}
return 0;
}
gboolean process_pe_message(xmlNode *msg, xmlNode *xml_data,
pcmk__client_t *sender);
static int32_t
pe_ipc_dispatch(qb_ipcs_connection_t * qbc, void *data, size_t size)
{
uint32_t id = 0;
uint32_t flags = 0;
pcmk__client_t *c = pcmk__find_client(qbc);
xmlNode *msg = pcmk__client_data2xml(c, data, &id, &flags);
pcmk__ipc_send_ack(c, id, flags, "ack");
if (msg != NULL) {
xmlNode *data_xml = get_message_xml(msg, F_CRM_DATA);
process_pe_message(msg, data_xml, c);
free_xml(msg);
}
return 0;
}
/* Error code means? */
static int32_t
pe_ipc_closed(qb_ipcs_connection_t * c)
{
pcmk__client_t *client = pcmk__find_client(c);
if (client == NULL) {
return 0;
}
crm_trace("Connection %p", c);
pcmk__free_client(client);
return 0;
}
static void
pe_ipc_destroy(qb_ipcs_connection_t * c)
{
crm_trace("Connection %p", c);
pe_ipc_closed(c);
}
struct qb_ipcs_service_handlers ipc_callbacks = {
.connection_accept = pe_ipc_accept,
.connection_created = NULL,
.msg_process = pe_ipc_dispatch,
.connection_closed = pe_ipc_closed,
.connection_destroyed = pe_ipc_destroy
};
static pcmk__cli_option_t long_options[] = {
// long option, argument type, storage, short option, description, flags
{
"help", no_argument, NULL, '?',
"\tThis text", pcmk__option_default
},
{
"verbose", no_argument, NULL, 'V',
"\tIncrease debug output", pcmk__option_default
},
{ 0, 0, 0, 0 }
};
int
main(int argc, char **argv)
{
int flag;
int index = 0;
int argerr = 0;
crm_log_preinit(NULL, argc, argv);
pcmk__set_cli_options(NULL, "[options]", long_options,
"daemon for calculating a Pacemaker cluster's "
"response to events");
mainloop_add_signal(SIGTERM, pengine_shutdown);
while (1) {
flag = pcmk__next_cli_option(argc, argv, &index, NULL);
if (flag == -1)
break;
switch (flag) {
case 'V':
crm_bump_log_level(argc, argv);
break;
case 'h': /* Help message */
pcmk__cli_help('?', CRM_EX_OK);
break;
default:
++argerr;
break;
}
}
if (argc - optind == 1 && pcmk__str_eq("metadata", argv[optind], pcmk__str_casei)) {
pe_metadata();
return CRM_EX_OK;
}
if (optind > argc) {
++argerr;
}
if (argerr) {
pcmk__cli_help('?', CRM_EX_USAGE);
}
crm_log_init(NULL, LOG_INFO, TRUE, FALSE, argc, argv, FALSE);
crm_notice("Starting Pacemaker scheduler");
if (pcmk__daemon_can_write(PE_STATE_DIR, NULL) == FALSE) {
crm_err("Terminating due to bad permissions on " PE_STATE_DIR);
fprintf(stderr,
"ERROR: Bad permissions on " PE_STATE_DIR " (see logs for details)\n");
fflush(stderr);
return CRM_EX_FATAL;
}
ipcs = mainloop_add_ipc_server(CRM_SYSTEM_PENGINE, QB_IPC_SHM, &ipc_callbacks);
if (ipcs == NULL) {
crm_err("Failed to create IPC server: shutting down and inhibiting respawn");
crm_exit(CRM_EX_FATAL);
}
/* Create the mainloop and run it... */
mainloop = g_main_loop_new(NULL, FALSE);
crm_notice("Pacemaker scheduler successfully started and accepting connections");
g_main_loop_run(mainloop);
pe_free_working_set(sched_data_set);
pcmk__unregister_formats();
crm_info("Exiting %s", crm_system_name);
crm_exit(CRM_EX_OK);
}
void
pengine_shutdown(int nsig)
{
mainloop_del_ipc_server(ipcs);
pe_free_working_set(sched_data_set);
crm_exit(CRM_EX_OK);
}
diff --git a/include/crm/common/internal.h b/include/crm/common/internal.h
index 54a5ca0c29..60b3e675a1 100644
--- a/include/crm/common/internal.h
+++ b/include/crm/common/internal.h
@@ -1,374 +1,388 @@
/*
* Copyright 2015-2020 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 CRM_COMMON_INTERNAL__H
#define CRM_COMMON_INTERNAL__H
#include <unistd.h> // getpid()
#include <stdbool.h> // bool
#include <stdint.h> // uint8_t, uint64_t
#include <string.h> // strcmp()
#include <fcntl.h> // open()
#include <sys/types.h> // uid_t, gid_t, pid_t
#include <glib.h> // guint, GList, GHashTable
#include <libxml/tree.h> // xmlNode
#include <crm/common/util.h> // crm_strdup_printf()
#include <crm/common/logging.h> // do_crm_log_unlikely(), etc.
#include <crm/common/mainloop.h> // mainloop_io_t, struct ipc_client_callbacks
#include <crm/common/strings_internal.h>
// Internal ACL-related utilities (from acl.c)
char *pcmk__uid2username(uid_t uid);
const char *pcmk__update_acl_user(xmlNode *request, const char *field,
const char *peer_user);
#if ENABLE_ACL
# include <string.h>
static inline bool
pcmk__is_privileged(const char *user)
{
return user && (!strcmp(user, CRM_DAEMON_USER) || !strcmp(user, "root"));
}
#endif
#if SUPPORT_CIBSECRETS
// Internal CIB utilities (from cib_secrets.c) */
int pcmk__substitute_secrets(const char *rsc_id, GHashTable *params);
#endif
/* internal digest-related utilities (from digest.c) */
bool pcmk__verify_digest(xmlNode *input, const char *expected);
/* internal I/O utilities (from io.c) */
int pcmk__real_path(const char *path, char **resolved_path);
char *pcmk__series_filename(const char *directory, const char *series,
int sequence, bool bzip);
int pcmk__read_series_sequence(const char *directory, const char *series,
unsigned int *seq);
void pcmk__write_series_sequence(const char *directory, const char *series,
unsigned int sequence, int max);
int pcmk__chown_series_sequence(const char *directory, const char *series,
uid_t uid, gid_t gid);
int pcmk__build_path(const char *path_c, mode_t mode);
bool pcmk__daemon_can_write(const char *dir, const char *file);
void pcmk__sync_directory(const char *name);
int pcmk__file_contents(const char *filename, char **contents);
int pcmk__write_sync(int fd, const char *contents);
int pcmk__set_nonblocking(int fd);
const char *pcmk__get_tmpdir(void);
void pcmk__close_fds_in_child(bool);
/*!
* \internal
* \brief Open /dev/null to consume next available file descriptor
*
* Open /dev/null, disregarding the result. This is intended when daemonizing to
* be able to null stdin, stdout, and stderr.
*
* \param[in] flags O_RDONLY (stdin) or O_WRONLY (stdout and stderr)
*/
static inline void
pcmk__open_devnull(int flags)
{
// Static analysis clutter
// cppcheck-suppress leakReturnValNotUsed
(void) open("/dev/null", flags);
}
/* internal logging utilities */
+/*!
+ * \internal
+ * \brief Log a configuration error
+ *
+ * \param[in] fmt printf(3)-style format string
+ * \param[in] ... Arguments for format string
+ */
# define pcmk__config_err(fmt...) do { \
- crm_config_error = TRUE; \
+ pcmk__config_error = true; \
crm_err(fmt); \
} while (0)
+/*!
+ * \internal
+ * \brief Log a configuration warning
+ *
+ * \param[in] fmt printf(3)-style format string
+ * \param[in] ... Arguments for format string
+ */
# define pcmk__config_warn(fmt...) do { \
- crm_config_warning = TRUE; \
+ pcmk__config_warning = true; \
crm_warn(fmt); \
} while (0)
/*!
* \internal
* \brief Execute code depending on whether message would be logged
*
* This is similar to do_crm_log_unlikely() except instead of logging, it either
* continues past this statement or executes else_action depending on whether a
* message of the given severity would be logged or not. This allows whole
* blocks of code to be skipped if tracing or debugging is turned off.
*
* \param[in] level Severity at which to continue past this statement
* \param[in] else_action Code block to execute if severity would not be logged
*
* \note else_action must not contain a break or continue statement
*/
# define pcmk__log_else(level, else_action) do { \
static struct qb_log_callsite *trace_cs = NULL; \
\
if (trace_cs == NULL) { \
trace_cs = qb_log_callsite_get(__func__, __FILE__, "log_else", \
level, __LINE__, 0); \
} \
if (!crm_is_callsite_active(trace_cs, level, 0)) { \
else_action; \
} \
} while(0)
/* internal main loop utilities (from mainloop.c) */
int pcmk__add_mainloop_ipc(crm_ipc_t *ipc, int priority, void *userdata,
struct ipc_client_callbacks *callbacks,
mainloop_io_t **source);
/* internal procfs utilities (from procfs.c) */
pid_t pcmk__procfs_pid_of(const char *name);
unsigned int pcmk__procfs_num_cores(void);
/* internal XML schema functions (from xml.c) */
void crm_schema_init(void);
void crm_schema_cleanup(void);
/* internal functions related to process IDs (from pid.c) */
/*!
* \internal
* \brief Check whether process exists (by PID and optionally executable path)
*
* \param[in] pid PID of process to check
* \param[in] daemon If not NULL, path component to match with procfs entry
*
* \return Standard Pacemaker return code
* \note Particular return codes of interest include pcmk_rc_ok for alive,
* ESRCH for process is not alive (verified by kill and/or executable path
* match), EACCES for caller unable or not allowed to check. A result of
* "alive" is less reliable when \p daemon is not provided or procfs is
* not available, since there is no guarantee that the PID has not been
* recycled for another process.
* \note This function cannot be used to verify \e authenticity of the process.
*/
int pcmk__pid_active(pid_t pid, const char *daemon);
int pcmk__read_pidfile(const char *filename, pid_t *pid);
int pcmk__pidfile_matches(const char *filename, pid_t expected_pid,
const char *expected_name, pid_t *pid);
int pcmk__lock_pidfile(const char *filename, const char *name);
/* internal functions related to resource operations (from operations.c) */
// printf-style format to create operation ID from resource, action, interval
#define PCMK__OP_FMT "%s_%s_%u"
char *pcmk__op_key(const char *rsc_id, const char *op_type, guint interval_ms);
char *pcmk__notify_key(const char *rsc_id, const char *notify_type,
const char *op_type);
char *pcmk__transition_key(int transition_id, int action_id, int target_rc,
const char *node);
void pcmk__filter_op_for_digest(xmlNode *param_set);
// bitwise arithmetic utilities
/*!
* \internal
* \brief Set specified flags in a flag group
*
* \param[in] function Function name of caller
* \param[in] line Line number of caller
* \param[in] log_level Log a message at this level
* \param[in] flag_type Label describing this flag group (for logging)
* \param[in] target Name of object whose flags these are (for logging)
* \param[in] flag_group Flag group being manipulated
* \param[in] flags Which flags in the group should be set
* \param[in] flags_str Readable equivalent of \p flags (for logging)
*
* \return Possibly modified flag group
*/
static inline uint64_t
pcmk__set_flags_as(const char *function, int line, uint8_t log_level,
const char *flag_type, const char *target,
uint64_t flag_group, uint64_t flags, const char *flags_str)
{
uint64_t result = flag_group | flags;
if (result != flag_group) {
do_crm_log_unlikely(log_level,
"%s flags 0x%.8llx (%s) for %s set by %s:%d",
((flag_type == NULL)? "Group of" : flag_type),
(unsigned long long) flags,
((flags_str == NULL)? "flags" : flags_str),
((target == NULL)? "target" : target),
function, line);
}
return result;
}
/*!
* \internal
* \brief Clear specified flags in a flag group
*
* \param[in] function Function name of caller
* \param[in] line Line number of caller
* \param[in] log_level Log a message at this level
* \param[in] flag_type Label describing this flag group (for logging)
* \param[in] target Name of object whose flags these are (for logging)
* \param[in] flag_group Flag group being manipulated
* \param[in] flags Which flags in the group should be cleared
* \param[in] flags_str Readable equivalent of \p flags (for logging)
*
* \return Possibly modified flag group
*/
static inline uint64_t
pcmk__clear_flags_as(const char *function, int line, uint8_t log_level,
const char *flag_type, const char *target,
uint64_t flag_group, uint64_t flags, const char *flags_str)
{
uint64_t result = flag_group & ~flags;
if (result != flag_group) {
do_crm_log_unlikely(log_level,
"%s flags 0x%.8llx (%s) for %s cleared by %s:%d",
((flag_type == NULL)? "Group of" : flag_type),
(unsigned long long) flags,
((flags_str == NULL)? "flags" : flags_str),
((target == NULL)? "target" : target),
function, line);
}
return result;
}
// miscellaneous utilities (from utils.c)
const char *pcmk_message_name(const char *name);
extern int pcmk__score_red;
extern int pcmk__score_green;
extern int pcmk__score_yellow;
/*!
* \internal
* \brief Resize a dynamically allocated memory block
*
* \param[in] ptr Memory block to resize (or NULL to allocate new memory)
* \param[in] size New size of memory block in bytes (must be > 0)
*
* \return Pointer to resized memory block
*
* \note This asserts on error, so the result is guaranteed to be non-NULL
* (which is the main advantage of this over directly using realloc()).
*/
static inline void *
pcmk__realloc(void *ptr, size_t size)
{
void *new_ptr;
// realloc(p, 0) can replace free(p) but this wrapper can't
CRM_ASSERT(size > 0);
new_ptr = realloc(ptr, size);
if (new_ptr == NULL) {
free(ptr);
abort();
}
return new_ptr;
}
/* Error domains for use with g_set_error (from results.c) */
GQuark pcmk__rc_error_quark(void);
GQuark pcmk__exitc_error_quark(void);
#define PCMK__RC_ERROR pcmk__rc_error_quark()
#define PCMK__EXITC_ERROR pcmk__exitc_error_quark()
static inline char *
pcmk__getpid_s(void)
{
return crm_strdup_printf("%lu", (unsigned long) getpid());
}
// More efficient than g_list_length(list) == 1
static inline bool
pcmk__list_of_1(GList *list)
{
return list && (list->next == NULL);
}
// More efficient than g_list_length(list) > 1
static inline bool
pcmk__list_of_multiple(GList *list)
{
return list && (list->next != NULL);
}
/* convenience functions for failure-related node attributes */
#define PCMK__FAIL_COUNT_PREFIX "fail-count"
#define PCMK__LAST_FAILURE_PREFIX "last-failure"
/*!
* \internal
* \brief Generate a failure-related node attribute name for a resource
*
* \param[in] prefix Start of attribute name
* \param[in] rsc_id Resource name
* \param[in] op Operation name
* \param[in] interval_ms Operation interval
*
* \return Newly allocated string with attribute name
*
* \note Failure attributes are named like PREFIX-RSC#OP_INTERVAL (for example,
* "fail-count-myrsc#monitor_30000"). The '#' is used because it is not
* a valid character in a resource ID, to reliably distinguish where the
* operation name begins. The '_' is used simply to be more comparable to
* action labels like "myrsc_monitor_30000".
*/
static inline char *
pcmk__fail_attr_name(const char *prefix, const char *rsc_id, const char *op,
guint interval_ms)
{
CRM_CHECK(prefix && rsc_id && op, return NULL);
return crm_strdup_printf("%s-%s#%s_%u", prefix, rsc_id, op, interval_ms);
}
static inline char *
pcmk__failcount_name(const char *rsc_id, const char *op, guint interval_ms)
{
return pcmk__fail_attr_name(PCMK__FAIL_COUNT_PREFIX, rsc_id, op,
interval_ms);
}
static inline char *
pcmk__lastfailure_name(const char *rsc_id, const char *op, guint interval_ms)
{
return pcmk__fail_attr_name(PCMK__LAST_FAILURE_PREFIX, rsc_id, op,
interval_ms);
}
#endif /* CRM_COMMON_INTERNAL__H */
diff --git a/include/crm/common/logging.h b/include/crm/common/logging.h
index 86bb108a2a..1aa8c21f1c 100644
--- a/include/crm/common/logging.h
+++ b/include/crm/common/logging.h
@@ -1,398 +1,403 @@
/*
* Copyright 2004-2020 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.
*/
#ifdef __cplusplus
extern "C" {
#endif
/**
* \file
* \brief Wrappers for and extensions to libqb logging
* \ingroup core
*/
#ifndef CRM_LOGGING__H
# define CRM_LOGGING__H
# include <stdio.h>
# include <glib.h>
# include <qb/qblog.h>
# include <libxml/tree.h>
/* Define custom log priorities.
*
* syslog(3) uses int for priorities, but libqb's struct qb_log_callsite uses
* uint8_t, so make sure they fit in the latter.
*/
// Define something even less desired than debug
# ifndef LOG_TRACE
# define LOG_TRACE (LOG_DEBUG+1)
# endif
// Print message to stdout instead of logging it
# ifndef LOG_STDOUT
# define LOG_STDOUT 254
# endif
// Don't send message anywhere
# ifndef LOG_NEVER
# define LOG_NEVER 255
# endif
/* "Extended information" logging support */
#ifdef QB_XS
# define CRM_XS QB_XS
# define crm_extended_logging(t, e) qb_log_ctl((t), QB_LOG_CONF_EXTENDED, (e))
#else
# define CRM_XS "|"
/* A caller might want to check the return value, so we can't define this as a
* no-op, and we can't simply define it to be 0 because gcc will then complain
* when the value isn't checked.
*/
static inline int
crm_extended_logging(int t, int e)
{
return 0;
}
#endif
extern unsigned int crm_log_level;
-extern gboolean crm_config_error;
-extern gboolean crm_config_warning;
extern unsigned int crm_trace_nonlog;
+/* These are set when a configuration issue is found, and turn on extra messages
+ * at the end of processing. They are set via wrapper functions and do not need
+ * to be set directly.
+ */
+extern bool pcmk__config_error;
+extern bool pcmk__config_warning;
+
enum xml_log_options
{
xml_log_option_filtered = 0x0001,
xml_log_option_formatted = 0x0002,
xml_log_option_text = 0x0004, /* add this option to dump text into xml */
xml_log_option_full_fledged = 0x0008, /* crm_xml_dump: serialize using libxml */
xml_log_option_diff_plus = 0x0010,
xml_log_option_diff_minus = 0x0020,
xml_log_option_diff_short = 0x0040,
xml_log_option_diff_all = 0x0100,
xml_log_option_dirty_add = 0x1000,
xml_log_option_open = 0x2000,
xml_log_option_children = 0x4000,
xml_log_option_close = 0x8000,
};
void crm_enable_blackbox(int nsig);
void crm_disable_blackbox(int nsig);
void crm_write_blackbox(int nsig, struct qb_log_callsite *callsite);
void crm_update_callsites(void);
void crm_log_deinit(void);
gboolean crm_log_cli_init(const char *entity);
void crm_log_preinit(const char *entity, int argc, char **argv);
gboolean crm_log_init(const char *entity, uint8_t level, gboolean daemon,
gboolean to_stderr, int argc, char **argv, gboolean quiet);
void crm_log_args(int argc, char **argv);
void crm_log_output_fn(const char *file, const char *function, int line, int level,
const char *prefix, const char *output);
// Log a block of text line by line
# define crm_log_output(level, prefix, output) crm_log_output_fn(__FILE__, __FUNCTION__, __LINE__, level, prefix, output)
gboolean crm_add_logfile(const char *filename);
void crm_bump_log_level(int argc, char **argv);
void crm_enable_stderr(int enable);
gboolean crm_is_callsite_active(struct qb_log_callsite *cs, uint8_t level, uint32_t tags);
void log_data_element(int log_level, const char *file, const char *function, int line,
const char *prefix, xmlNode * data, int depth, gboolean formatted);
/* returns the old value */
unsigned int set_crm_log_level(unsigned int level);
unsigned int get_crm_log_level(void);
/*
* Throughout the macros below, note the leading, pre-comma, space in the
* various ' , ##args' occurrences to aid portability across versions of 'gcc'.
* https://gcc.gnu.org/onlinedocs/cpp/Variadic-Macros.html#Variadic-Macros
*/
#if defined(__clang__)
# define CRM_TRACE_INIT_DATA(name)
# else
# include <assert.h> // required by QB_LOG_INIT_DATA() macro
# define CRM_TRACE_INIT_DATA(name) QB_LOG_INIT_DATA(name)
#endif
/* Using "switch" instead of "if" in these macro definitions keeps
* static analysis from complaining about constant evaluations
*/
/*!
* \brief Log a message
*
* \param[in] level Priority at which to log the message
* \param[in] fmt printf-style format string literal for message
* \param[in] args Any arguments needed by format string
*
* \note This is a macro, and \p level may be evaluated more than once.
*/
# define do_crm_log(level, fmt, args...) do { \
switch (level) { \
case LOG_STDOUT: \
printf(fmt "\n" , ##args); \
break; \
case LOG_NEVER: \
break; \
default: \
qb_log_from_external_source(__func__, __FILE__, fmt, \
(level), __LINE__, 0 , ##args); \
break; \
} \
} while (0)
/*!
* \brief Log a message that is likely to be filtered out
*
* \param[in] level Priority at which to log the message
* \param[in] fmt printf-style format string for message
* \param[in] args Any arguments needed by format string
*
* \note This is a macro, and \p level may be evaluated more than once.
* This does nothing when level is LOG_STDOUT.
*/
# define do_crm_log_unlikely(level, fmt, args...) do { \
switch (level) { \
case LOG_STDOUT: case LOG_NEVER: \
break; \
default: { \
static struct qb_log_callsite *trace_cs = NULL; \
if (trace_cs == NULL) { \
trace_cs = qb_log_callsite_get(__func__, __FILE__, fmt, \
(level), __LINE__, 0); \
} \
if (crm_is_callsite_active(trace_cs, (level), 0)) { \
qb_log_from_external_source(__func__, __FILE__, fmt, \
(level), __LINE__, 0 , ##args); \
} \
} \
break; \
} \
} while (0)
# define CRM_LOG_ASSERT(expr) do { \
if(__unlikely((expr) == FALSE)) { \
static struct qb_log_callsite *core_cs = NULL; \
if(core_cs == NULL) { \
core_cs = qb_log_callsite_get(__func__, __FILE__, "log-assert", LOG_TRACE, __LINE__, 0); \
} \
crm_abort(__FILE__, __FUNCTION__, __LINE__, #expr, \
core_cs?core_cs->targets:FALSE, TRUE); \
} \
} while(0)
/* 'failure_action' MUST NOT be 'continue' as it will apply to the
* macro's do-while loop
*/
# define CRM_CHECK(expr, failure_action) do { \
if(__unlikely((expr) == FALSE)) { \
static struct qb_log_callsite *core_cs = NULL; \
if(core_cs == NULL) { \
core_cs = qb_log_callsite_get(__func__, __FILE__, "check-assert", LOG_TRACE, __LINE__, 0); \
} \
crm_abort(__FILE__, __FUNCTION__, __LINE__, #expr, \
core_cs?core_cs->targets:FALSE, TRUE); \
failure_action; \
} \
} while(0)
/*!
* \brief Log XML line-by-line in a formatted fashion
*
* \param[in] level Priority at which to log the messages
* \param[in] text Prefix for each line
* \param[in] xml XML to log
*
* \note This is a macro, and \p level may be evaluated more than once.
* This does nothing when level is LOG_STDOUT.
*/
# define do_crm_log_xml(level, text, xml) do { \
switch (level) { \
case LOG_STDOUT: case LOG_NEVER: \
break; \
default: { \
static struct qb_log_callsite *xml_cs = NULL; \
if (xml_cs == NULL) { \
xml_cs = qb_log_callsite_get(__func__, __FILE__, \
"xml-blob", (level), __LINE__, 0); \
} \
if (crm_is_callsite_active(xml_cs, (level), 0)) { \
log_data_element((level), __FILE__, __FUNCTION__, \
__LINE__, text, xml, 1, xml_log_option_formatted); \
} \
} \
break; \
} \
} while(0)
/*!
* \brief Log a message as if it came from a different code location
*
* \param[in] level Priority at which to log the message
* \param[in] file Source file name to use instead of __FILE__
* \param[in] function Source function name to use instead of __func__
* \param[in] line Source line number to use instead of __line__
* \param[in] fmt printf-style format string literal for message
* \param[in] args Any arguments needed by format string
*
* \note This is a macro, and \p level may be evaluated more than once.
*/
# define do_crm_log_alias(level, file, function, line, fmt, args...) do { \
switch (level) { \
case LOG_STDOUT: \
printf(fmt "\n" , ##args); \
break; \
case LOG_NEVER: \
break; \
default: \
qb_log_from_external_source(function, file, fmt, (level), \
line, 0 , ##args); \
break; \
} \
} while (0)
/*!
* \brief Send a system error message to both the log and stderr
*
* \param[in] level Priority at which to log the message
* \param[in] fmt printf-style format string for message
* \param[in] args Any arguments needed by format string
*
* \deprecated One of the other logging functions should be used with
* pcmk_strerror() instead.
* \note This is a macro, and \p level may be evaluated more than once.
* \note Because crm_perror() adds the system error message and error number
* onto the end of fmt, that information will become extended information
* if CRM_XS is used inside fmt and will not show up in syslog.
*/
# define crm_perror(level, fmt, args...) do { \
switch (level) { \
case LOG_NEVER: \
break; \
default: { \
const char *err = strerror(errno); \
/* cast to int makes coverity happy when level == 0 */ \
if ((level) <= (int) crm_log_level) { \
fprintf(stderr, fmt ": %s (%d)\n" , ##args, err, errno);\
} \
do_crm_log((level), fmt ": %s (%d)" , ##args, err, errno); \
} \
break; \
} \
} while (0)
/*!
* \brief Log a message with a tag (for use with PCMK_trace_tags)
*
* \param[in] level Priority at which to log the message
* \param[in] tag String to tag message with
* \param[in] fmt printf-style format string for message
* \param[in] args Any arguments needed by format string
*
* \note This is a macro, and \p level may be evaluated more than once.
* This does nothing when level is LOG_STDOUT.
*/
# define crm_log_tag(level, tag, fmt, args...) do { \
switch (level) { \
case LOG_STDOUT: case LOG_NEVER: \
break; \
default: { \
static struct qb_log_callsite *trace_tag_cs = NULL; \
int converted_tag = g_quark_try_string(tag); \
if (trace_tag_cs == NULL) { \
trace_tag_cs = qb_log_callsite_get(__func__, __FILE__, \
fmt, (level), __LINE__, converted_tag); \
} \
if (crm_is_callsite_active(trace_tag_cs, (level), \
converted_tag)) { \
qb_log_from_external_source(__func__, __FILE__, fmt, \
(level), __LINE__, converted_tag , ##args); \
} \
} \
} \
} while (0)
# define crm_emerg(fmt, args...) qb_log(LOG_EMERG, fmt , ##args)
# define crm_crit(fmt, args...) qb_logt(LOG_CRIT, 0, fmt , ##args)
# define crm_err(fmt, args...) qb_logt(LOG_ERR, 0, fmt , ##args)
# define crm_warn(fmt, args...) qb_logt(LOG_WARNING, 0, fmt , ##args)
# define crm_notice(fmt, args...) qb_logt(LOG_NOTICE, 0, fmt , ##args)
# define crm_info(fmt, args...) qb_logt(LOG_INFO, 0, fmt , ##args)
# define crm_debug(fmt, args...) do_crm_log_unlikely(LOG_DEBUG, fmt , ##args)
# define crm_trace(fmt, args...) do_crm_log_unlikely(LOG_TRACE, fmt , ##args)
# define crm_log_xml_crit(xml, text) do_crm_log_xml(LOG_CRIT, text, xml)
# define crm_log_xml_err(xml, text) do_crm_log_xml(LOG_ERR, text, xml)
# define crm_log_xml_warn(xml, text) do_crm_log_xml(LOG_WARNING, text, xml)
# define crm_log_xml_notice(xml, text) do_crm_log_xml(LOG_NOTICE, text, xml)
# define crm_log_xml_info(xml, text) do_crm_log_xml(LOG_INFO, text, xml)
# define crm_log_xml_debug(xml, text) do_crm_log_xml(LOG_DEBUG, text, xml)
# define crm_log_xml_trace(xml, text) do_crm_log_xml(LOG_TRACE, text, xml)
# define crm_log_xml_explicit(xml, text) do { \
static struct qb_log_callsite *digest_cs = NULL; \
digest_cs = qb_log_callsite_get( \
__func__, __FILE__, text, LOG_TRACE, __LINE__, \
crm_trace_nonlog); \
if (digest_cs && digest_cs->targets) { \
do_crm_log_xml(LOG_TRACE, text, xml); \
} \
} while(0)
# define crm_str(x) (const char*)(x?x:"<null>")
#ifndef PCMK__NO_COMPAT
/* Everything here is deprecated and kept only for public API backward
* compatibility. It will be moved to compatibility.h in a future release.
*/
/*!
* \brief Log a message using constant priority
*
* \param[in] level Priority at which to log the message
* \param[in] fmt printf-style format string literal for message
* \param[in] args Any arguments needed by format string
*
* \deprecated Use one of the other logging functions instead
* \note This is a macro, and \p level may be evaluated more than once.
* This does nothing when level is LOG_STDOUT.
*/
# define do_crm_log_always(level, fmt, args...) do { \
switch (level) { \
case LOG_STDOUT: case LOG_NEVER: \
break; \
default: \
qb_log((level), fmt , ##args); \
break; \
} \
} while (0)
#endif // PCMK__NO_COMPAT
#ifdef __cplusplus
}
#endif
#endif
diff --git a/include/crm/pengine/internal.h b/include/crm/pengine/internal.h
index c2acfa796c..c3a69f0f35 100644
--- a/include/crm/pengine/internal.h
+++ b/include/crm/pengine/internal.h
@@ -1,596 +1,604 @@
/*
* Copyright 2004-2020 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 PE_INTERNAL__H
# define PE_INTERNAL__H
# include <string.h>
# include <crm/pengine/status.h>
# include <crm/pengine/remote_internal.h>
# include <crm/common/internal.h>
# include <crm/common/output_internal.h>
# define pe_rsc_info(rsc, fmt, args...) crm_log_tag(LOG_INFO, rsc ? rsc->id : "<NULL>", fmt, ##args)
# define pe_rsc_debug(rsc, fmt, args...) crm_log_tag(LOG_DEBUG, rsc ? rsc->id : "<NULL>", fmt, ##args)
# define pe_rsc_trace(rsc, fmt, args...) crm_log_tag(LOG_TRACE, rsc ? rsc->id : "<NULL>", fmt, ##args)
-# define pe_err(fmt...) { was_processing_error = TRUE; crm_config_error = TRUE; crm_err(fmt); }
-# define pe_warn(fmt...) { was_processing_warning = TRUE; crm_config_warning = TRUE; crm_warn(fmt); }
+# define pe_err(fmt...) do { \
+ was_processing_error = TRUE; \
+ pcmk__config_err(fmt); \
+ } while (0)
+
+# define pe_warn(fmt...) do { \
+ was_processing_warning = TRUE; \
+ pcmk__config_warn(fmt); \
+ } while (0)
+
# define pe_proc_err(fmt...) { was_processing_error = TRUE; crm_err(fmt); }
# define pe_proc_warn(fmt...) { was_processing_warning = TRUE; crm_warn(fmt); }
#define pe__set_working_set_flags(working_set, flags_to_set) do { \
(working_set)->flags = pcmk__set_flags_as(__FUNCTION__, __LINE__, \
LOG_TRACE, "Working set", crm_system_name, \
(working_set)->flags, (flags_to_set), #flags_to_set); \
} while (0)
#define pe__clear_working_set_flags(working_set, flags_to_clear) do { \
(working_set)->flags = pcmk__clear_flags_as(__FUNCTION__, __LINE__, \
LOG_TRACE, "Working set", crm_system_name, \
(working_set)->flags, (flags_to_clear), #flags_to_clear); \
} while (0)
#define pe__set_resource_flags(resource, flags_to_set) do { \
(resource)->flags = pcmk__set_flags_as(__FUNCTION__, __LINE__, \
LOG_TRACE, "Resource", (resource)->id, (resource)->flags, \
(flags_to_set), #flags_to_set); \
} while (0)
#define pe__clear_resource_flags(resource, flags_to_clear) do { \
(resource)->flags = pcmk__clear_flags_as(__FUNCTION__, __LINE__, \
LOG_TRACE, "Resource", (resource)->id, (resource)->flags, \
(flags_to_clear), #flags_to_clear); \
} while (0)
#define pe__set_action_flags(action, flags_to_set) do { \
(action)->flags = pcmk__set_flags_as(__FUNCTION__, __LINE__, \
LOG_TRACE, \
"Action", (action)->uuid, \
(action)->flags, \
(flags_to_set), \
#flags_to_set); \
} while (0)
#define pe__clear_action_flags(action, flags_to_clear) do { \
(action)->flags = pcmk__clear_flags_as(__FUNCTION__, __LINE__, \
LOG_TRACE, \
"Action", (action)->uuid, \
(action)->flags, \
(flags_to_clear), \
#flags_to_clear); \
} while (0)
#define pe__set_raw_action_flags(action_flags, action_name, flags_to_set) do { \
action_flags = pcmk__set_flags_as(__FUNCTION__, __LINE__, \
LOG_TRACE, "Action", action_name, \
(action_flags), \
(flags_to_set), #flags_to_set); \
} while (0)
#define pe__clear_raw_action_flags(action_flags, action_name, flags_to_clear) do { \
action_flags = pcmk__clear_flags_as(__FUNCTION__, __LINE__, \
LOG_TRACE, \
"Action", action_name, \
(action_flags), \
(flags_to_clear), \
#flags_to_clear); \
} while (0)
#define pe__set_action_flags_as(function, line, action, flags_to_set) do { \
(action)->flags = pcmk__set_flags_as((function), (line), \
LOG_TRACE, \
"Action", (action)->uuid, \
(action)->flags, \
(flags_to_set), \
#flags_to_set); \
} while (0)
#define pe__clear_action_flags_as(function, line, action, flags_to_clear) do { \
(action)->flags = pcmk__clear_flags_as((function), (line), \
LOG_TRACE, \
"Action", (action)->uuid, \
(action)->flags, \
(flags_to_clear), \
#flags_to_clear); \
} while (0)
#define pe__set_order_flags(order_flags, flags_to_set) do { \
order_flags = pcmk__set_flags_as(__FUNCTION__, __LINE__, LOG_TRACE, \
"Ordering", "constraint", \
order_flags, (flags_to_set), \
#flags_to_set); \
} while (0)
#define pe__clear_order_flags(order_flags, flags_to_clear) do { \
order_flags = pcmk__clear_flags_as(__FUNCTION__, __LINE__, LOG_TRACE, \
"Ordering", "constraint", \
order_flags, (flags_to_clear), \
#flags_to_clear); \
} while (0)
#define pe__set_graph_flags(graph_flags, gr_action, flags_to_set) do { \
graph_flags = pcmk__set_flags_as(__FUNCTION__, __LINE__, \
LOG_TRACE, "Graph", \
(gr_action)->uuid, graph_flags, \
(flags_to_set), #flags_to_set); \
} while (0)
#define pe__clear_graph_flags(graph_flags, gr_action, flags_to_clear) do { \
graph_flags = pcmk__clear_flags_as(__FUNCTION__, __LINE__, \
LOG_TRACE, "Graph", \
(gr_action)->uuid, graph_flags, \
(flags_to_clear), #flags_to_clear); \
} while (0)
// Some warnings we don't want to print every transition
enum pe_warn_once_e {
pe_wo_blind = 0x0001,
pe_wo_restart_type = 0x0002,
pe_wo_role_after = 0x0004,
pe_wo_poweroff = 0x0008,
pe_wo_require_all = 0x0010,
pe_wo_order_score = 0x0020,
pe_wo_neg_threshold = 0x0040,
};
extern uint32_t pe_wo;
#define pe_warn_once(pe_wo_bit, fmt...) do { \
if (!pcmk_is_set(pe_wo, pe_wo_bit)) { \
if (pe_wo_bit == pe_wo_blind) { \
crm_warn(fmt); \
} else { \
pe_warn(fmt); \
} \
pe_wo = pcmk__set_flags_as(__FUNCTION__, __LINE__, LOG_TRACE, \
"Warn-once", "logging", pe_wo, \
(pe_wo_bit), #pe_wo_bit); \
} \
} while (0);
typedef struct pe__location_constraint_s {
char *id; // Constraint XML ID
pe_resource_t *rsc_lh; // Resource being located
enum rsc_role_e role_filter; // Role to locate
enum pe_discover_e discover_mode; // Resource discovery
GListPtr node_list_rh; // List of pe_node_t*
} pe__location_t;
typedef struct pe__order_constraint_s {
int id;
enum pe_ordering type;
void *lh_opaque;
pe_resource_t *lh_rsc;
pe_action_t *lh_action;
char *lh_action_task;
void *rh_opaque;
pe_resource_t *rh_rsc;
pe_action_t *rh_action;
char *rh_action_task;
} pe__ordering_t;
typedef struct notify_data_s {
GSList *keys; // Environment variable name/value pairs
const char *action;
pe_action_t *pre;
pe_action_t *post;
pe_action_t *pre_done;
pe_action_t *post_done;
GListPtr active; /* notify_entry_t* */
GListPtr inactive; /* notify_entry_t* */
GListPtr start; /* notify_entry_t* */
GListPtr stop; /* notify_entry_t* */
GListPtr demote; /* notify_entry_t* */
GListPtr promote; /* notify_entry_t* */
GListPtr master; /* notify_entry_t* */
GListPtr slave; /* notify_entry_t* */
GHashTable *allowed_nodes;
} notify_data_t;
bool pe_can_fence(pe_working_set_t *data_set, pe_node_t *node);
int pe__add_scores(int score1, int score2);
void add_hash_param(GHashTable * hash, const char *name, const char *value);
char *native_parameter(pe_resource_t * rsc, pe_node_t * node, gboolean create, const char *name,
pe_working_set_t * data_set);
pe_node_t *native_location(const pe_resource_t *rsc, GList **list, int current);
void pe_metadata(void);
void verify_pe_options(GHashTable * options);
void common_update_score(pe_resource_t * rsc, const char *id, int score);
void native_add_running(pe_resource_t * rsc, pe_node_t * node, pe_working_set_t * data_set);
gboolean native_unpack(pe_resource_t * rsc, pe_working_set_t * data_set);
gboolean group_unpack(pe_resource_t * rsc, pe_working_set_t * data_set);
gboolean clone_unpack(pe_resource_t * rsc, pe_working_set_t * data_set);
gboolean pe__unpack_bundle(pe_resource_t *rsc, pe_working_set_t *data_set);
pe_resource_t *native_find_rsc(pe_resource_t *rsc, const char *id, const pe_node_t *node,
int flags);
gboolean native_active(pe_resource_t * rsc, gboolean all);
gboolean group_active(pe_resource_t * rsc, gboolean all);
gboolean clone_active(pe_resource_t * rsc, gboolean all);
gboolean pe__bundle_active(pe_resource_t *rsc, gboolean all);
void native_print(pe_resource_t * rsc, const char *pre_text, long options, void *print_data);
void group_print(pe_resource_t * rsc, const char *pre_text, long options, void *print_data);
void clone_print(pe_resource_t * rsc, const char *pre_text, long options, void *print_data);
void pe__print_bundle(pe_resource_t *rsc, const char *pre_text, long options,
void *print_data);
int pe__name_and_nvpairs_xml(pcmk__output_t *out, bool is_list, const char *tag_name
, size_t pairs_count, ...);
char *pe__node_display_name(pe_node_t *node, bool print_detail);
static inline const char *
pe__rsc_bool_str(pe_resource_t *rsc, uint64_t rsc_flag)
{
return pcmk__btoa(pcmk_is_set(rsc->flags, rsc_flag));
}
int pe__ban_html(pcmk__output_t *out, va_list args);
int pe__ban_text(pcmk__output_t *out, va_list args);
int pe__ban_xml(pcmk__output_t *out, va_list args);
int pe__clone_xml(pcmk__output_t *out, va_list args);
int pe__clone_html(pcmk__output_t *out, va_list args);
int pe__clone_text(pcmk__output_t *out, va_list args);
int pe__cluster_counts_html(pcmk__output_t *out, va_list args);
int pe__cluster_counts_text(pcmk__output_t *out, va_list args);
int pe__cluster_counts_xml(pcmk__output_t *out, va_list args);
int pe__cluster_dc_html(pcmk__output_t *out, va_list args);
int pe__cluster_dc_text(pcmk__output_t *out, va_list args);
int pe__cluster_dc_xml(pcmk__output_t *out, va_list args);
int pe__cluster_maint_mode_html(pcmk__output_t *out, va_list args);
int pe__cluster_maint_mode_text(pcmk__output_t *out, va_list args);
int pe__cluster_options_html(pcmk__output_t *out, va_list args);
int pe__cluster_options_log(pcmk__output_t *out, va_list args);
int pe__cluster_options_text(pcmk__output_t *out, va_list args);
int pe__cluster_options_xml(pcmk__output_t *out, va_list args);
int pe__cluster_stack_html(pcmk__output_t *out, va_list args);
int pe__cluster_stack_text(pcmk__output_t *out, va_list args);
int pe__cluster_stack_xml(pcmk__output_t *out, va_list args);
int pe__cluster_summary(pcmk__output_t *out, va_list args);
int pe__cluster_summary_html(pcmk__output_t *out, va_list args);
int pe__cluster_times_html(pcmk__output_t *out, va_list args);
int pe__cluster_times_xml(pcmk__output_t *out, va_list args);
int pe__cluster_times_text(pcmk__output_t *out, va_list args);
int pe__failed_action_text(pcmk__output_t *out, va_list args);
int pe__failed_action_xml(pcmk__output_t *out, va_list args);
int pe__group_xml(pcmk__output_t *out, va_list args);
int pe__group_html(pcmk__output_t *out, va_list args);
int pe__group_text(pcmk__output_t *out, va_list args);
int pe__bundle_xml(pcmk__output_t *out, va_list args);
int pe__bundle_html(pcmk__output_t *out, va_list args);
int pe__bundle_text(pcmk__output_t *out, va_list args);
int pe__node_html(pcmk__output_t *out, va_list args);
int pe__node_text(pcmk__output_t *out, va_list args);
int pe__node_xml(pcmk__output_t *out, va_list args);
int pe__node_attribute_html(pcmk__output_t *out, va_list args);
int pe__node_attribute_text(pcmk__output_t *out, va_list args);
int pe__node_attribute_xml(pcmk__output_t *out, va_list args);
int pe__node_list_html(pcmk__output_t *out, va_list args);
int pe__node_list_text(pcmk__output_t *out, va_list args);
int pe__node_list_xml(pcmk__output_t *out, va_list args);
int pe__op_history_text(pcmk__output_t *out, va_list args);
int pe__op_history_xml(pcmk__output_t *out, va_list args);
int pe__resource_history_text(pcmk__output_t *out, va_list args);
int pe__resource_history_xml(pcmk__output_t *out, va_list args);
int pe__resource_xml(pcmk__output_t *out, va_list args);
int pe__resource_html(pcmk__output_t *out, va_list args);
int pe__resource_text(pcmk__output_t *out, va_list args);
int pe__ticket_html(pcmk__output_t *out, va_list args);
int pe__ticket_text(pcmk__output_t *out, va_list args);
int pe__ticket_xml(pcmk__output_t *out, va_list args);
void native_free(pe_resource_t * rsc);
void group_free(pe_resource_t * rsc);
void clone_free(pe_resource_t * rsc);
void pe__free_bundle(pe_resource_t *rsc);
enum rsc_role_e native_resource_state(const pe_resource_t * rsc, gboolean current);
enum rsc_role_e group_resource_state(const pe_resource_t * rsc, gboolean current);
enum rsc_role_e clone_resource_state(const pe_resource_t * rsc, gboolean current);
enum rsc_role_e pe__bundle_resource_state(const pe_resource_t *rsc,
gboolean current);
void pe__count_common(pe_resource_t *rsc);
void pe__count_bundle(pe_resource_t *rsc);
gboolean common_unpack(xmlNode * xml_obj, pe_resource_t ** rsc, pe_resource_t * parent,
pe_working_set_t * data_set);
void common_free(pe_resource_t * rsc);
pe_node_t *pe__copy_node(const pe_node_t *this_node);
extern time_t get_effective_time(pe_working_set_t * data_set);
/* Failure handling utilities (from failcounts.c) */
// bit flags for fail count handling options
enum pe_fc_flags_e {
pe_fc_default = 0x00,
pe_fc_effective = 0x01, // don't count expired failures
pe_fc_fillers = 0x02, // if container, include filler failures in count
};
int pe_get_failcount(pe_node_t *node, pe_resource_t *rsc, time_t *last_failure,
uint32_t flags, xmlNode *xml_op,
pe_working_set_t *data_set);
pe_action_t *pe__clear_failcount(pe_resource_t *rsc, pe_node_t *node,
const char *reason,
pe_working_set_t *data_set);
/* Functions for finding/counting a resource's active nodes */
pe_node_t *pe__find_active_on(const pe_resource_t *rsc,
unsigned int *count_all,
unsigned int *count_clean);
pe_node_t *pe__find_active_requires(const pe_resource_t *rsc,
unsigned int *count);
static inline pe_node_t *
pe__current_node(const pe_resource_t *rsc)
{
return pe__find_active_on(rsc, NULL, NULL);
}
/* Binary like operators for lists of nodes */
extern void node_list_exclude(GHashTable * list, GListPtr list2, gboolean merge_scores);
GHashTable *pe__node_list2table(GList *list);
static inline gpointer
pe_hash_table_lookup(GHashTable * hash, gconstpointer key)
{
if (hash) {
return g_hash_table_lookup(hash, key);
}
return NULL;
}
extern pe_action_t *get_pseudo_op(const char *name, pe_working_set_t * data_set);
extern gboolean order_actions(pe_action_t * lh_action, pe_action_t * rh_action, enum pe_ordering order);
/* Printing functions for debug */
extern void print_node(const char *pre_text, pe_node_t * node, gboolean details);
extern void print_str_str(gpointer key, gpointer value, gpointer user_data);
extern void pe__output_node(pe_node_t * node, gboolean details, pcmk__output_t *out);
extern void dump_node_capacity(int level, const char *comment, pe_node_t * node);
extern void dump_rsc_utilization(int level, const char *comment, pe_resource_t * rsc, pe_node_t * node);
void pe__show_node_weights_as(const char *file, const char *function,
int line, bool to_log, pe_resource_t *rsc,
const char *comment, GHashTable *nodes);
#define pe__show_node_weights(level, rsc, text, nodes) \
pe__show_node_weights_as(__FILE__, __FUNCTION__, __LINE__, \
(level), (rsc), (text), (nodes))
/* Sorting functions */
extern gint sort_rsc_priority(gconstpointer a, gconstpointer b);
extern gint sort_rsc_index(gconstpointer a, gconstpointer b);
extern xmlNode *find_rsc_op_entry(pe_resource_t * rsc, const char *key);
extern pe_action_t *custom_action(pe_resource_t * rsc, char *key, const char *task, pe_node_t * on_node,
gboolean optional, gboolean foo, pe_working_set_t * data_set);
# define delete_key(rsc) pcmk__op_key(rsc->id, CRMD_ACTION_DELETE, 0)
# define delete_action(rsc, node, optional) custom_action( \
rsc, delete_key(rsc), CRMD_ACTION_DELETE, node, \
optional, TRUE, data_set);
# define stopped_key(rsc) pcmk__op_key(rsc->id, CRMD_ACTION_STOPPED, 0)
# define stopped_action(rsc, node, optional) custom_action( \
rsc, stopped_key(rsc), CRMD_ACTION_STOPPED, node, \
optional, TRUE, data_set);
# define stop_key(rsc) pcmk__op_key(rsc->id, CRMD_ACTION_STOP, 0)
# define stop_action(rsc, node, optional) custom_action( \
rsc, stop_key(rsc), CRMD_ACTION_STOP, node, \
optional, TRUE, data_set);
# define reload_key(rsc) pcmk__op_key(rsc->id, CRMD_ACTION_RELOAD, 0)
# define start_key(rsc) pcmk__op_key(rsc->id, CRMD_ACTION_START, 0)
# define start_action(rsc, node, optional) custom_action( \
rsc, start_key(rsc), CRMD_ACTION_START, node, \
optional, TRUE, data_set)
# define started_key(rsc) pcmk__op_key(rsc->id, CRMD_ACTION_STARTED, 0)
# define started_action(rsc, node, optional) custom_action( \
rsc, started_key(rsc), CRMD_ACTION_STARTED, node, \
optional, TRUE, data_set)
# define promote_key(rsc) pcmk__op_key(rsc->id, CRMD_ACTION_PROMOTE, 0)
# define promote_action(rsc, node, optional) custom_action( \
rsc, promote_key(rsc), CRMD_ACTION_PROMOTE, node, \
optional, TRUE, data_set)
# define promoted_key(rsc) pcmk__op_key(rsc->id, CRMD_ACTION_PROMOTED, 0)
# define promoted_action(rsc, node, optional) custom_action( \
rsc, promoted_key(rsc), CRMD_ACTION_PROMOTED, node, \
optional, TRUE, data_set)
# define demote_key(rsc) pcmk__op_key(rsc->id, CRMD_ACTION_DEMOTE, 0)
# define demote_action(rsc, node, optional) custom_action( \
rsc, demote_key(rsc), CRMD_ACTION_DEMOTE, node, \
optional, TRUE, data_set)
# define demoted_key(rsc) pcmk__op_key(rsc->id, CRMD_ACTION_DEMOTED, 0)
# define demoted_action(rsc, node, optional) custom_action( \
rsc, demoted_key(rsc), CRMD_ACTION_DEMOTED, node, \
optional, TRUE, data_set)
extern int pe_get_configured_timeout(pe_resource_t *rsc, const char *action,
pe_working_set_t *data_set);
extern pe_action_t *find_first_action(GListPtr input, const char *uuid, const char *task,
pe_node_t * on_node);
extern enum action_tasks get_complex_task(pe_resource_t * rsc, const char *name,
gboolean allow_non_atomic);
extern GListPtr find_actions(GListPtr input, const char *key, const pe_node_t *on_node);
GList *find_actions_exact(GList *input, const char *key,
const pe_node_t *on_node);
extern GListPtr find_recurring_actions(GListPtr input, pe_node_t * not_on_node);
GList *pe__resource_actions(const pe_resource_t *rsc, const pe_node_t *node,
const char *task, bool require_node);
extern void pe_free_action(pe_action_t * action);
extern void resource_location(pe_resource_t * rsc, pe_node_t * node, int score, const char *tag,
pe_working_set_t * data_set);
extern gint sort_op_by_callid(gconstpointer a, gconstpointer b);
extern gboolean get_target_role(pe_resource_t * rsc, enum rsc_role_e *role);
extern pe_resource_t *find_clone_instance(pe_resource_t * rsc, const char *sub_id,
pe_working_set_t * data_set);
extern void destroy_ticket(gpointer data);
extern pe_ticket_t *ticket_new(const char *ticket_id, pe_working_set_t * data_set);
// Resources for manipulating resource names
const char *pe_base_name_end(const char *id);
char *clone_strip(const char *last_rsc_id);
char *clone_zero(const char *last_rsc_id);
static inline bool
pe_base_name_eq(pe_resource_t *rsc, const char *id)
{
if (id && rsc && rsc->id) {
// Number of characters in rsc->id before any clone suffix
size_t base_len = pe_base_name_end(rsc->id) - rsc->id + 1;
return (strlen(id) == base_len) && !strncmp(id, rsc->id, base_len);
}
return FALSE;
}
int pe__target_rc_from_xml(xmlNode *xml_op);
gint sort_node_uname(gconstpointer a, gconstpointer b);
bool is_set_recursive(pe_resource_t * rsc, long long flag, bool any);
enum rsc_digest_cmp_val {
/*! Digests are the same */
RSC_DIGEST_MATCH = 0,
/*! Params that require a restart changed */
RSC_DIGEST_RESTART,
/*! Some parameter changed. */
RSC_DIGEST_ALL,
/*! rsc op didn't have a digest associated with it, so
* it is unknown if parameters changed or not. */
RSC_DIGEST_UNKNOWN,
};
typedef struct op_digest_cache_s {
enum rsc_digest_cmp_val rc;
xmlNode *params_all;
xmlNode *params_secure;
xmlNode *params_restart;
char *digest_all_calc;
char *digest_secure_calc;
char *digest_restart_calc;
} op_digest_cache_t;
op_digest_cache_t *rsc_action_digest_cmp(pe_resource_t * rsc, xmlNode * xml_op, pe_node_t * node,
pe_working_set_t * data_set);
pe_action_t *pe_fence_op(pe_node_t * node, const char *op, bool optional, const char *reason, bool priority_delay, pe_working_set_t * data_set);
void trigger_unfencing(
pe_resource_t * rsc, pe_node_t *node, const char *reason, pe_action_t *dependency, pe_working_set_t * data_set);
void pe_action_set_reason(pe_action_t *action, const char *reason, bool overwrite);
void pe_action_set_flag_reason(const char *function, long line, pe_action_t *action, pe_action_t *reason, const char *text, enum pe_action_flags flags, bool overwrite);
#define pe_action_required(action, reason, text) pe_action_set_flag_reason(__FUNCTION__, __LINE__, action, reason, text, pe_action_optional, FALSE)
#define pe_action_implies(action, reason, flag) pe_action_set_flag_reason(__FUNCTION__, __LINE__, action, reason, NULL, flag, FALSE)
void pe__set_resource_flags_recursive(pe_resource_t *rsc, uint64_t flags);
void pe__clear_resource_flags_recursive(pe_resource_t *rsc, uint64_t flags);
gboolean add_tag_ref(GHashTable * tags, const char * tag_name, const char * obj_ref);
void print_rscs_brief(GListPtr rsc_list, const char * pre_text, long options,
void * print_data, gboolean print_all);
int pe__rscs_brief_output(pcmk__output_t *out, GListPtr rsc_list, long options, gboolean print_all);
void pe_fence_node(pe_working_set_t * data_set, pe_node_t * node, const char *reason, bool priority_delay);
pe_node_t *pe_create_node(const char *id, const char *uname, const char *type,
const char *score, pe_working_set_t * data_set);
void common_print(pe_resource_t * rsc, const char *pre_text, const char *name, pe_node_t *node, long options, void *print_data);
int pe__common_output_text(pcmk__output_t *out, pe_resource_t * rsc, const char *name, pe_node_t *node, long options);
int pe__common_output_html(pcmk__output_t *out, pe_resource_t * rsc, const char *name, pe_node_t *node, long options);
pe_resource_t *pe__find_bundle_replica(const pe_resource_t *bundle,
const pe_node_t *node);
bool pe__bundle_needs_remote_name(pe_resource_t *rsc);
const char *pe__add_bundle_remote_name(pe_resource_t *rsc, xmlNode *xml,
const char *field);
const char *pe_node_attribute_calculated(const pe_node_t *node,
const char *name,
const pe_resource_t *rsc);
const char *pe_node_attribute_raw(pe_node_t *node, const char *name);
bool pe__is_universal_clone(pe_resource_t *rsc,
pe_working_set_t *data_set);
void pe__add_param_check(xmlNode *rsc_op, pe_resource_t *rsc, pe_node_t *node,
enum pe_check_parameters, pe_working_set_t *data_set);
void pe__foreach_param_check(pe_working_set_t *data_set,
void (*cb)(pe_resource_t*, pe_node_t*, xmlNode*,
enum pe_check_parameters,
pe_working_set_t*));
void pe__free_param_checks(pe_working_set_t *data_set);
bool pe__shutdown_requested(pe_node_t *node);
void pe__update_recheck_time(time_t recheck, pe_working_set_t *data_set);
/*!
* \internal
* \brief Register xml formatting message functions.
*/
void pe__register_messages(pcmk__output_t *out);
void pe__unpack_dataset_nvpairs(xmlNode *xml_obj, const char *set_name,
pe_rule_eval_data_t *rule_data, GHashTable *hash,
const char *always_first, gboolean overwrite,
pe_working_set_t *data_set);
bool pe__resource_is_disabled(pe_resource_t *rsc);
pe_action_t *pe__clear_resource_history(pe_resource_t *rsc, pe_node_t *node,
pe_working_set_t *data_set);
GListPtr pe__rscs_with_tag(pe_working_set_t *data_set, const char *tag_name);
GListPtr pe__unames_with_tag(pe_working_set_t *data_set, const char *tag_name);
bool pe__rsc_has_tag(pe_working_set_t *data_set, const char *rsc, const char *tag);
bool pe__uname_has_tag(pe_working_set_t *data_set, const char *node, const char *tag);
bool pe__rsc_running_on_any_node_in_list(pe_resource_t *rsc, GListPtr node_list);
GListPtr pe__filter_rsc_list(GListPtr rscs, GListPtr filter);
bool pcmk__rsc_filtered_by_node(pe_resource_t *rsc, GListPtr only_node);
gboolean pe__bundle_is_filtered(pe_resource_t *rsc, GListPtr only_rsc, gboolean check_parent);
gboolean pe__clone_is_filtered(pe_resource_t *rsc, GListPtr only_rsc, gboolean check_parent);
gboolean pe__group_is_filtered(pe_resource_t *rsc, GListPtr only_rsc, gboolean check_parent);
gboolean pe__native_is_filtered(pe_resource_t *rsc, GListPtr only_rsc, gboolean check_parent);
#endif
diff --git a/lib/common/utils.c b/lib/common/utils.c
index 35e168dd5b..c53c9fcc2d 100644
--- a/lib/common/utils.c
+++ b/lib/common/utils.c
@@ -1,616 +1,616 @@
/*
* Copyright 2004-2020 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>
#ifndef _GNU_SOURCE
# define _GNU_SOURCE
#endif
#include <sys/types.h>
#include <sys/wait.h>
#include <sys/stat.h>
#include <sys/utsname.h>
#include <stdio.h>
#include <unistd.h>
#include <string.h>
#include <stdlib.h>
#include <limits.h>
#include <pwd.h>
#include <time.h>
#include <libgen.h>
#include <signal.h>
#include <qb/qbdefs.h>
#include <crm/crm.h>
#include <crm/services.h>
#include <crm/msg_xml.h>
#include <crm/cib/internal.h>
#include <crm/common/xml.h>
#include <crm/common/util.h>
#include <crm/common/ipc.h>
#include <crm/common/iso8601.h>
#include <crm/common/mainloop.h>
#include <libxml2/libxml/relaxng.h>
#ifndef PW_BUFFER_LEN
# define PW_BUFFER_LEN 500
#endif
CRM_TRACE_INIT_DATA(common);
-gboolean crm_config_error = FALSE;
-gboolean crm_config_warning = FALSE;
+bool pcmk__config_error = false;
+bool pcmk__config_warning = false;
char *crm_system_name = NULL;
int pcmk__score_red = 0;
int pcmk__score_green = 0;
int pcmk__score_yellow = 0;
int
char2score(const char *score)
{
int score_f = 0;
if (score == NULL) {
} else if (pcmk_str_is_minus_infinity(score)) {
score_f = -CRM_SCORE_INFINITY;
} else if (pcmk_str_is_infinity(score)) {
score_f = CRM_SCORE_INFINITY;
} else if (pcmk__str_eq(score, "red", pcmk__str_casei)) {
score_f = pcmk__score_red;
} else if (pcmk__str_eq(score, "yellow", pcmk__str_casei)) {
score_f = pcmk__score_yellow;
} else if (pcmk__str_eq(score, "green", pcmk__str_casei)) {
score_f = pcmk__score_green;
} else {
score_f = crm_parse_int(score, NULL);
if (score_f > 0 && score_f > CRM_SCORE_INFINITY) {
score_f = CRM_SCORE_INFINITY;
} else if (score_f < 0 && score_f < -CRM_SCORE_INFINITY) {
score_f = -CRM_SCORE_INFINITY;
}
}
return score_f;
}
char *
score2char_stack(int score, char *buf, size_t len)
{
if (score >= CRM_SCORE_INFINITY) {
strncpy(buf, CRM_INFINITY_S, 9);
} else if (score <= -CRM_SCORE_INFINITY) {
strncpy(buf, CRM_MINUS_INFINITY_S , 10);
} else {
return crm_itoa_stack(score, buf, len);
}
return buf;
}
char *
score2char(int score)
{
if (score >= CRM_SCORE_INFINITY) {
return strdup(CRM_INFINITY_S);
} else if (score <= -CRM_SCORE_INFINITY) {
return strdup(CRM_MINUS_INFINITY_S);
}
return crm_itoa(score);
}
int
crm_user_lookup(const char *name, uid_t * uid, gid_t * gid)
{
int rc = pcmk_ok;
char *buffer = NULL;
struct passwd pwd;
struct passwd *pwentry = NULL;
buffer = calloc(1, PW_BUFFER_LEN);
if (buffer == NULL) {
return -ENOMEM;
}
rc = getpwnam_r(name, &pwd, buffer, PW_BUFFER_LEN, &pwentry);
if (pwentry) {
if (uid) {
*uid = pwentry->pw_uid;
}
if (gid) {
*gid = pwentry->pw_gid;
}
crm_trace("User %s has uid=%d gid=%d", name, pwentry->pw_uid, pwentry->pw_gid);
} else {
rc = rc? -rc : -EINVAL;
crm_info("User %s lookup: %s", name, pcmk_strerror(rc));
}
free(buffer);
return rc;
}
/*!
* \brief Get user and group IDs of pacemaker daemon user
*
* \param[out] uid If non-NULL, where to store daemon user ID
* \param[out] gid If non-NULL, where to store daemon group ID
*
* \return pcmk_ok on success, -errno otherwise
*/
int
pcmk_daemon_user(uid_t *uid, gid_t *gid)
{
static uid_t daemon_uid;
static gid_t daemon_gid;
static bool found = false;
int rc = pcmk_err_generic;
if (!found) {
rc = crm_user_lookup(CRM_DAEMON_USER, &daemon_uid, &daemon_gid);
if (rc == pcmk_ok) {
found = true;
}
}
if (found) {
if (uid) {
*uid = daemon_uid;
}
if (gid) {
*gid = daemon_gid;
}
}
return rc;
}
static int
crm_version_helper(const char *text, const char **end_text)
{
int atoi_result = -1;
CRM_ASSERT(end_text != NULL);
errno = 0;
if (text != NULL && text[0] != 0) {
/* seemingly sacrificing const-correctness -- because while strtol
doesn't modify the input, it doesn't want to artificially taint the
"end_text" pointer-to-pointer-to-first-char-in-string with constness
in case the input wasn't actually constant -- by semantic definition
not a single character will get modified so it shall be perfectly
safe to make compiler happy with dropping "const" qualifier here */
atoi_result = (int) strtol(text, (char **) end_text, 10);
if (errno == EINVAL) {
crm_err("Conversion of '%s' %c failed", text, text[0]);
atoi_result = -1;
}
}
return atoi_result;
}
/*
* version1 < version2 : -1
* version1 = version2 : 0
* version1 > version2 : 1
*/
int
compare_version(const char *version1, const char *version2)
{
int rc = 0;
int lpc = 0;
const char *ver1_iter, *ver2_iter;
if (version1 == NULL && version2 == NULL) {
return 0;
} else if (version1 == NULL) {
return -1;
} else if (version2 == NULL) {
return 1;
}
ver1_iter = version1;
ver2_iter = version2;
while (1) {
int digit1 = 0;
int digit2 = 0;
lpc++;
if (ver1_iter == ver2_iter) {
break;
}
if (ver1_iter != NULL) {
digit1 = crm_version_helper(ver1_iter, &ver1_iter);
}
if (ver2_iter != NULL) {
digit2 = crm_version_helper(ver2_iter, &ver2_iter);
}
if (digit1 < digit2) {
rc = -1;
break;
} else if (digit1 > digit2) {
rc = 1;
break;
}
if (ver1_iter != NULL && *ver1_iter == '.') {
ver1_iter++;
}
if (ver1_iter != NULL && *ver1_iter == '\0') {
ver1_iter = NULL;
}
if (ver2_iter != NULL && *ver2_iter == '.') {
ver2_iter++;
}
if (ver2_iter != NULL && *ver2_iter == 0) {
ver2_iter = NULL;
}
}
if (rc == 0) {
crm_trace("%s == %s (%d)", version1, version2, lpc);
} else if (rc < 0) {
crm_trace("%s < %s (%d)", version1, version2, lpc);
} else if (rc > 0) {
crm_trace("%s > %s (%d)", version1, version2, lpc);
}
return rc;
}
/*!
* \brief Parse milliseconds from a Pacemaker interval specification
*
* \param[in] input Pacemaker time interval specification (a bare number of
* seconds, a number with a unit optionally with whitespace
* before and/or after the number, or an ISO 8601 duration)
*
* \return Milliseconds equivalent of given specification on success (limited
* to the range of an unsigned integer), 0 if input is NULL,
* or 0 (and set errno to EINVAL) on error
*/
guint
crm_parse_interval_spec(const char *input)
{
long long msec = -1;
errno = 0;
if (input == NULL) {
return 0;
} else if (input[0] == 'P') {
crm_time_t *period_s = crm_time_parse_duration(input);
if (period_s) {
msec = 1000 * crm_time_get_seconds(period_s);
crm_time_free(period_s);
}
} else {
msec = crm_get_msec(input);
}
if (msec < 0) {
crm_warn("Using 0 instead of '%s'", input);
errno = EINVAL;
return 0;
}
return (msec >= G_MAXUINT)? G_MAXUINT : (guint) msec;
}
extern bool crm_is_daemon;
/* coverity[+kill] */
void
crm_abort(const char *file, const char *function, int line,
const char *assert_condition, gboolean do_core, gboolean do_fork)
{
int rc = 0;
int pid = 0;
int status = 0;
/* Implied by the parent's error logging below */
/* crm_write_blackbox(0); */
if(crm_is_daemon == FALSE) {
/* This is a command line tool - do not fork */
/* crm_add_logfile(NULL); * Record it to a file? */
crm_enable_stderr(TRUE); /* Make sure stderr is enabled so we can tell the caller */
do_fork = FALSE; /* Just crash if needed */
}
if (do_core == FALSE) {
crm_err("%s: Triggered assert at %s:%d : %s", function, file, line, assert_condition);
return;
} else if (do_fork) {
pid = fork();
} else {
crm_err("%s: Triggered fatal assert at %s:%d : %s", function, file, line, assert_condition);
}
if (pid == -1) {
crm_crit("%s: Cannot create core for non-fatal assert at %s:%d : %s",
function, file, line, assert_condition);
return;
} else if(pid == 0) {
/* Child process */
abort();
return;
}
/* Parent process */
crm_err("%s: Forked child %d to record non-fatal assert at %s:%d : %s",
function, pid, file, line, assert_condition);
crm_write_blackbox(SIGTRAP, NULL);
do {
rc = waitpid(pid, &status, 0);
if(rc == pid) {
return; /* Job done */
}
} while(errno == EINTR);
if (errno == ECHILD) {
/* crm_mon does this */
crm_trace("Cannot wait on forked child %d - SIGCHLD is probably set to SIG_IGN", pid);
return;
}
crm_perror(LOG_ERR, "Cannot wait on forked child %d", pid);
}
void
crm_make_daemon(const char *name, gboolean daemonize, const char *pidfile)
{
int rc;
pid_t pid;
if (daemonize == FALSE) {
return;
}
/* Check before we even try... */
rc = pcmk__pidfile_matches(pidfile, 1, name, &pid);
if ((rc != pcmk_rc_ok) && (rc != ENOENT)) {
crm_err("%s: already running [pid %lld in %s]",
name, (long long) pid, pidfile);
printf("%s: already running [pid %lld in %s]\n",
name, (long long) pid, pidfile);
crm_exit(CRM_EX_ERROR);
}
pid = fork();
if (pid < 0) {
fprintf(stderr, "%s: could not start daemon\n", name);
crm_perror(LOG_ERR, "fork");
crm_exit(CRM_EX_OSERR);
} else if (pid > 0) {
crm_exit(CRM_EX_OK);
}
rc = pcmk__lock_pidfile(pidfile, name);
if (rc != pcmk_rc_ok) {
crm_err("Could not lock '%s' for %s: %s " CRM_XS " rc=%d",
pidfile, name, pcmk_rc_str(rc), rc);
printf("Could not lock '%s' for %s: %s (%d)\n",
pidfile, name, pcmk_rc_str(rc), rc);
crm_exit(CRM_EX_ERROR);
}
umask(S_IWGRP | S_IWOTH | S_IROTH);
close(STDIN_FILENO);
pcmk__open_devnull(O_RDONLY); // stdin (fd 0)
close(STDOUT_FILENO);
pcmk__open_devnull(O_WRONLY); // stdout (fd 1)
close(STDERR_FILENO);
pcmk__open_devnull(O_WRONLY); // stderr (fd 2)
}
char *
crm_meta_name(const char *field)
{
int lpc = 0;
int max = 0;
char *crm_name = NULL;
CRM_CHECK(field != NULL, return NULL);
crm_name = crm_strdup_printf(CRM_META "_%s", field);
/* Massage the names so they can be used as shell variables */
max = strlen(crm_name);
for (; lpc < max; lpc++) {
switch (crm_name[lpc]) {
case '-':
crm_name[lpc] = '_';
break;
}
}
return crm_name;
}
const char *
crm_meta_value(GHashTable * hash, const char *field)
{
char *key = NULL;
const char *value = NULL;
key = crm_meta_name(field);
if (key) {
value = g_hash_table_lookup(hash, key);
free(key);
}
return value;
}
#ifdef HAVE_UUID_UUID_H
# include <uuid/uuid.h>
#endif
char *
crm_generate_uuid(void)
{
unsigned char uuid[16];
char *buffer = malloc(37); /* Including NUL byte */
uuid_generate(uuid);
uuid_unparse(uuid, buffer);
return buffer;
}
/*!
* \brief Get name to be used as identifier for cluster messages
*
* \param[in] name Actual system name to check
*
* \return Non-NULL cluster message identifier corresponding to name
*
* \note The Pacemaker daemons were renamed in version 2.0.0, but the old names
* must continue to be used as the identifier for cluster messages, so
* that mixed-version clusters are possible during a rolling upgrade.
*/
const char *
pcmk_message_name(const char *name)
{
if (name == NULL) {
return "unknown";
} else if (!strcmp(name, "pacemaker-attrd")) {
return "attrd";
} else if (!strcmp(name, "pacemaker-based")) {
return CRM_SYSTEM_CIB;
} else if (!strcmp(name, "pacemaker-controld")) {
return CRM_SYSTEM_CRMD;
} else if (!strcmp(name, "pacemaker-execd")) {
return CRM_SYSTEM_LRMD;
} else if (!strcmp(name, "pacemaker-fenced")) {
return "stonith-ng";
} else if (!strcmp(name, "pacemaker-schedulerd")) {
return CRM_SYSTEM_PENGINE;
} else {
return name;
}
}
/*!
* \brief Check whether a string represents a cluster daemon name
*
* \param[in] name String to check
*
* \return TRUE if name is standard client name used by daemons, FALSE otherwise
*/
bool
crm_is_daemon_name(const char *name)
{
name = pcmk_message_name(name);
return (!strcmp(name, CRM_SYSTEM_CRMD)
|| !strcmp(name, CRM_SYSTEM_STONITHD)
|| !strcmp(name, "stonith-ng")
|| !strcmp(name, "attrd")
|| !strcmp(name, CRM_SYSTEM_CIB)
|| !strcmp(name, CRM_SYSTEM_MCP)
|| !strcmp(name, CRM_SYSTEM_DC)
|| !strcmp(name, CRM_SYSTEM_TENGINE)
|| !strcmp(name, CRM_SYSTEM_LRMD));
}
#include <md5.h>
char *
crm_md5sum(const char *buffer)
{
int lpc = 0, len = 0;
char *digest = NULL;
unsigned char raw_digest[MD5_DIGEST_SIZE];
if (buffer == NULL) {
buffer = "";
}
len = strlen(buffer);
crm_trace("Beginning digest of %d bytes", len);
digest = malloc(2 * MD5_DIGEST_SIZE + 1);
if(digest) {
md5_buffer(buffer, len, raw_digest);
for (lpc = 0; lpc < MD5_DIGEST_SIZE; lpc++) {
sprintf(digest + (2 * lpc), "%02x", raw_digest[lpc]);
}
digest[(2 * MD5_DIGEST_SIZE)] = 0;
crm_trace("Digest %s.", digest);
} else {
crm_err("Could not create digest");
}
return digest;
}
#ifdef HAVE_GNUTLS_GNUTLS_H
void
crm_gnutls_global_init(void)
{
signal(SIGPIPE, SIG_IGN);
gnutls_global_init();
}
#endif
/*!
* \brief Get the local hostname
*
* \return Newly allocated string with name, or NULL (and set errno) on error
*/
char *
pcmk_hostname()
{
struct utsname hostinfo;
return (uname(&hostinfo) < 0)? NULL : strdup(hostinfo.nodename);
}
bool
pcmk_str_is_infinity(const char *s) {
return pcmk__str_any_of(s, CRM_INFINITY_S, CRM_PLUS_INFINITY_S, NULL);
}
bool
pcmk_str_is_minus_infinity(const char *s) {
return pcmk__str_eq(s, CRM_MINUS_INFINITY_S, pcmk__str_none);
}
diff --git a/lib/pacemaker/pcmk_sched_allocate.c b/lib/pacemaker/pcmk_sched_allocate.c
index eeedb7611b..f3877f87a6 100644
--- a/lib/pacemaker/pcmk_sched_allocate.c
+++ b/lib/pacemaker/pcmk_sched_allocate.c
@@ -1,3059 +1,3059 @@
/*
* Copyright 2004-2020 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/msg_xml.h>
#include <crm/common/xml.h>
#include <glib.h>
#include <crm/pengine/status.h>
#include <pacemaker-internal.h>
CRM_TRACE_INIT_DATA(pacemaker);
void set_alloc_actions(pe_working_set_t * data_set);
extern void ReloadRsc(pe_resource_t * rsc, pe_node_t *node, pe_working_set_t * data_set);
extern gboolean DeleteRsc(pe_resource_t * rsc, pe_node_t * node, gboolean optional, pe_working_set_t * data_set);
static void apply_remote_node_ordering(pe_working_set_t *data_set);
static enum remote_connection_state get_remote_node_state(pe_node_t *node);
enum remote_connection_state {
remote_state_unknown = 0,
remote_state_alive = 1,
remote_state_resting = 2,
remote_state_failed = 3,
remote_state_stopped = 4
};
static const char *
state2text(enum remote_connection_state state)
{
switch (state) {
case remote_state_unknown:
return "unknown";
case remote_state_alive:
return "alive";
case remote_state_resting:
return "resting";
case remote_state_failed:
return "failed";
case remote_state_stopped:
return "stopped";
}
return "impossible";
}
resource_alloc_functions_t resource_class_alloc_functions[] = {
{
pcmk__native_merge_weights,
pcmk__native_allocate,
native_create_actions,
native_create_probe,
native_internal_constraints,
native_rsc_colocation_lh,
native_rsc_colocation_rh,
native_rsc_location,
native_action_flags,
native_update_actions,
native_expand,
native_append_meta,
},
{
pcmk__group_merge_weights,
pcmk__group_allocate,
group_create_actions,
native_create_probe,
group_internal_constraints,
group_rsc_colocation_lh,
group_rsc_colocation_rh,
group_rsc_location,
group_action_flags,
group_update_actions,
group_expand,
group_append_meta,
},
{
pcmk__native_merge_weights,
pcmk__clone_allocate,
clone_create_actions,
clone_create_probe,
clone_internal_constraints,
clone_rsc_colocation_lh,
clone_rsc_colocation_rh,
clone_rsc_location,
clone_action_flags,
pcmk__multi_update_actions,
clone_expand,
clone_append_meta,
},
{
pcmk__native_merge_weights,
pcmk__bundle_allocate,
pcmk__bundle_create_actions,
pcmk__bundle_create_probe,
pcmk__bundle_internal_constraints,
pcmk__bundle_rsc_colocation_lh,
pcmk__bundle_rsc_colocation_rh,
pcmk__bundle_rsc_location,
pcmk__bundle_action_flags,
pcmk__multi_update_actions,
pcmk__bundle_expand,
pcmk__bundle_append_meta,
}
};
gboolean
update_action_flags(pe_action_t * action, enum pe_action_flags flags, const char *source, int line)
{
static unsigned long calls = 0;
gboolean changed = FALSE;
gboolean clear = pcmk_is_set(flags, pe_action_clear);
enum pe_action_flags last = action->flags;
if (clear) {
pe__clear_action_flags_as(source, line, action, flags);
} else {
pe__set_action_flags_as(source, line, action, flags);
}
if (last != action->flags) {
calls++;
changed = TRUE;
/* Useful for tracking down _who_ changed a specific flag */
/* CRM_ASSERT(calls != 534); */
pe__clear_raw_action_flags(flags, "action update", pe_action_clear);
crm_trace("%s on %s: %sset flags 0x%.6x (was 0x%.6x, now 0x%.6x, %lu, %s)",
action->uuid, action->node ? action->node->details->uname : "[none]",
clear ? "un-" : "", flags, last, action->flags, calls, source);
}
return changed;
}
static gboolean
check_rsc_parameters(pe_resource_t * rsc, pe_node_t * node, xmlNode * rsc_entry,
gboolean active_here, pe_working_set_t * data_set)
{
int attr_lpc = 0;
gboolean force_restart = FALSE;
gboolean delete_resource = FALSE;
gboolean changed = FALSE;
const char *value = NULL;
const char *old_value = NULL;
const char *attr_list[] = {
XML_ATTR_TYPE,
XML_AGENT_ATTR_CLASS,
XML_AGENT_ATTR_PROVIDER
};
for (; attr_lpc < DIMOF(attr_list); attr_lpc++) {
value = crm_element_value(rsc->xml, attr_list[attr_lpc]);
old_value = crm_element_value(rsc_entry, attr_list[attr_lpc]);
if (value == old_value /* i.e. NULL */
|| pcmk__str_eq(value, old_value, pcmk__str_none)) {
continue;
}
changed = TRUE;
trigger_unfencing(rsc, node, "Device definition changed", NULL, data_set);
if (active_here) {
force_restart = TRUE;
crm_notice("Forcing restart of %s on %s, %s changed: %s -> %s",
rsc->id, node->details->uname, attr_list[attr_lpc],
crm_str(old_value), crm_str(value));
}
}
if (force_restart) {
/* make sure the restart happens */
stop_action(rsc, node, FALSE);
pe__set_resource_flags(rsc, pe_rsc_start_pending);
delete_resource = TRUE;
} else if (changed) {
delete_resource = TRUE;
}
return delete_resource;
}
static void
CancelXmlOp(pe_resource_t * rsc, xmlNode * xml_op, pe_node_t * active_node,
const char *reason, pe_working_set_t * data_set)
{
guint interval_ms = 0;
pe_action_t *cancel = NULL;
const char *task = NULL;
const char *call_id = NULL;
CRM_CHECK(xml_op != NULL, return);
CRM_CHECK(active_node != NULL, return);
task = crm_element_value(xml_op, XML_LRM_ATTR_TASK);
call_id = crm_element_value(xml_op, XML_LRM_ATTR_CALLID);
crm_element_value_ms(xml_op, XML_LRM_ATTR_INTERVAL_MS, &interval_ms);
crm_info("Action " PCMK__OP_FMT " on %s will be stopped: %s",
rsc->id, task, interval_ms,
active_node->details->uname, (reason? reason : "unknown"));
cancel = pe_cancel_op(rsc, task, interval_ms, active_node, data_set);
add_hash_param(cancel->meta, XML_LRM_ATTR_CALLID, call_id);
custom_action_order(rsc, stop_key(rsc), NULL, rsc, NULL, cancel, pe_order_optional, data_set);
}
static gboolean
check_action_definition(pe_resource_t * rsc, pe_node_t * active_node, xmlNode * xml_op,
pe_working_set_t * data_set)
{
char *key = NULL;
guint interval_ms = 0;
const op_digest_cache_t *digest_data = NULL;
gboolean did_change = FALSE;
const char *task = crm_element_value(xml_op, XML_LRM_ATTR_TASK);
const char *digest_secure = NULL;
CRM_CHECK(active_node != NULL, return FALSE);
crm_element_value_ms(xml_op, XML_LRM_ATTR_INTERVAL_MS, &interval_ms);
if (interval_ms > 0) {
xmlNode *op_match = NULL;
/* we need to reconstruct the key because of the way we used to construct resource IDs */
key = pcmk__op_key(rsc->id, task, interval_ms);
pe_rsc_trace(rsc, "Checking parameters for %s", key);
op_match = find_rsc_op_entry(rsc, key);
if ((op_match == NULL)
&& pcmk_is_set(data_set->flags, pe_flag_stop_action_orphans)) {
CancelXmlOp(rsc, xml_op, active_node, "orphan", data_set);
free(key);
return TRUE;
} else if (op_match == NULL) {
pe_rsc_debug(rsc, "Orphan action detected: %s on %s", key, active_node->details->uname);
free(key);
return TRUE;
}
free(key);
key = NULL;
}
crm_trace("Testing " PCMK__OP_FMT " on %s",
rsc->id, task, interval_ms, active_node->details->uname);
if ((interval_ms == 0) && pcmk__str_eq(task, RSC_STATUS, pcmk__str_casei)) {
/* Reload based on the start action not a probe */
task = RSC_START;
} else if ((interval_ms == 0) && pcmk__str_eq(task, RSC_MIGRATED, pcmk__str_casei)) {
/* Reload based on the start action not a migrate */
task = RSC_START;
} else if ((interval_ms == 0) && pcmk__str_eq(task, RSC_PROMOTE, pcmk__str_casei)) {
/* Reload based on the start action not a promote */
task = RSC_START;
}
digest_data = rsc_action_digest_cmp(rsc, xml_op, active_node, data_set);
if (pcmk_is_set(data_set->flags, pe_flag_sanitized)) {
digest_secure = crm_element_value(xml_op, XML_LRM_ATTR_SECURE_DIGEST);
}
if(digest_data->rc != RSC_DIGEST_MATCH
&& digest_secure
&& digest_data->digest_secure_calc
&& strcmp(digest_data->digest_secure_calc, digest_secure) == 0) {
if (pcmk_is_set(data_set->flags, pe_flag_stdout)) {
printf("Only 'private' parameters to " PCMK__OP_FMT
" on %s changed: %s\n",
rsc->id, task, interval_ms, active_node->details->uname,
crm_element_value(xml_op, XML_ATTR_TRANSITION_MAGIC));
}
} else if (digest_data->rc == RSC_DIGEST_RESTART) {
/* Changes that force a restart */
pe_action_t *required = NULL;
did_change = TRUE;
key = pcmk__op_key(rsc->id, task, interval_ms);
crm_log_xml_info(digest_data->params_restart, "params:restart");
required = custom_action(rsc, key, task, NULL, TRUE, TRUE, data_set);
pe_action_set_flag_reason(__FUNCTION__, __LINE__, required, NULL,
"resource definition change", pe_action_optional, TRUE);
trigger_unfencing(rsc, active_node, "Device parameters changed", NULL, data_set);
} else if ((digest_data->rc == RSC_DIGEST_ALL) || (digest_data->rc == RSC_DIGEST_UNKNOWN)) {
/* Changes that can potentially be handled by a reload */
const char *digest_restart = crm_element_value(xml_op, XML_LRM_ATTR_RESTART_DIGEST);
did_change = TRUE;
trigger_unfencing(rsc, active_node, "Device parameters changed (reload)", NULL, data_set);
crm_log_xml_info(digest_data->params_all, "params:reload");
key = pcmk__op_key(rsc->id, task, interval_ms);
if (interval_ms > 0) {
pe_action_t *op = NULL;
#if 0
/* Always reload/restart the entire resource */
ReloadRsc(rsc, active_node, data_set);
#else
/* Re-sending the recurring op is sufficient - the old one will be cancelled automatically */
op = custom_action(rsc, key, task, active_node, TRUE, TRUE, data_set);
pe__set_action_flags(op, pe_action_reschedule);
#endif
} else if (digest_restart) {
pe_rsc_trace(rsc, "Reloading '%s' action for resource %s", task, rsc->id);
/* Reload this resource */
ReloadRsc(rsc, active_node, data_set);
free(key);
} else {
pe_action_t *required = NULL;
pe_rsc_trace(rsc, "Resource %s doesn't know how to reload", rsc->id);
/* Re-send the start/demote/promote op
* Recurring ops will be detected independently
*/
required = custom_action(rsc, key, task, NULL, TRUE, TRUE, data_set);
pe_action_set_flag_reason(__FUNCTION__, __LINE__, required, NULL,
"resource definition change", pe_action_optional, TRUE);
}
}
return did_change;
}
/*!
* \internal
* \brief Do deferred action checks after allocation
*
* \param[in] data_set Working set for cluster
*/
static void
check_params(pe_resource_t *rsc, pe_node_t *node, xmlNode *rsc_op,
enum pe_check_parameters check, pe_working_set_t *data_set)
{
const char *reason = NULL;
op_digest_cache_t *digest_data = NULL;
switch (check) {
case pe_check_active:
if (check_action_definition(rsc, node, rsc_op, data_set)
&& pe_get_failcount(node, rsc, NULL, pe_fc_effective, NULL,
data_set)) {
reason = "action definition changed";
}
break;
case pe_check_last_failure:
digest_data = rsc_action_digest_cmp(rsc, rsc_op, node, data_set);
switch (digest_data->rc) {
case RSC_DIGEST_UNKNOWN:
crm_trace("Resource %s history entry %s on %s has no digest to compare",
rsc->id, ID(rsc_op), node->details->id);
break;
case RSC_DIGEST_MATCH:
break;
default:
reason = "resource parameters have changed";
break;
}
break;
}
if (reason) {
pe__clear_failcount(rsc, node, reason, data_set);
}
}
static void
check_actions_for(xmlNode * rsc_entry, pe_resource_t * rsc, pe_node_t * node, pe_working_set_t * data_set)
{
GListPtr gIter = NULL;
int offset = -1;
int stop_index = 0;
int start_index = 0;
const char *task = NULL;
xmlNode *rsc_op = NULL;
GListPtr op_list = NULL;
GListPtr sorted_op_list = NULL;
CRM_CHECK(node != NULL, return);
if (pcmk_is_set(rsc->flags, pe_rsc_orphan)) {
pe_resource_t *parent = uber_parent(rsc);
if(parent == NULL
|| pe_rsc_is_clone(parent) == FALSE
|| pcmk_is_set(parent->flags, pe_rsc_unique)) {
pe_rsc_trace(rsc, "Skipping param check for %s and deleting: orphan", rsc->id);
DeleteRsc(rsc, node, FALSE, data_set);
} else {
pe_rsc_trace(rsc, "Skipping param check for %s (orphan clone)", rsc->id);
}
return;
} else if (pe_find_node_id(rsc->running_on, node->details->id) == NULL) {
if (check_rsc_parameters(rsc, node, rsc_entry, FALSE, data_set)) {
DeleteRsc(rsc, node, FALSE, data_set);
}
pe_rsc_trace(rsc, "Skipping param check for %s: no longer active on %s",
rsc->id, node->details->uname);
return;
}
pe_rsc_trace(rsc, "Processing %s on %s", rsc->id, node->details->uname);
if (check_rsc_parameters(rsc, node, rsc_entry, TRUE, data_set)) {
DeleteRsc(rsc, node, FALSE, data_set);
}
for (rsc_op = __xml_first_child_element(rsc_entry); rsc_op != NULL;
rsc_op = __xml_next_element(rsc_op)) {
if (pcmk__str_eq((const char *)rsc_op->name, XML_LRM_TAG_RSC_OP, pcmk__str_none)) {
op_list = g_list_prepend(op_list, rsc_op);
}
}
sorted_op_list = g_list_sort(op_list, sort_op_by_callid);
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;
guint interval_ms = 0;
offset++;
if (start_index < stop_index) {
/* stopped */
continue;
} else if (offset < start_index) {
/* action occurred prior to a start */
continue;
}
task = crm_element_value(rsc_op, XML_LRM_ATTR_TASK);
crm_element_value_ms(rsc_op, XML_LRM_ATTR_INTERVAL_MS, &interval_ms);
if ((interval_ms > 0) &&
(pcmk_is_set(rsc->flags, pe_rsc_maintenance) || node->details->maintenance)) {
// Maintenance mode cancels recurring operations
CancelXmlOp(rsc, rsc_op, node, "maintenance mode", data_set);
} else if ((interval_ms > 0) || pcmk__strcase_any_of(task, RSC_STATUS, RSC_START,
RSC_PROMOTE, RSC_MIGRATED, 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 allocated 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, pe_check_active,
data_set);
} else if (check_action_definition(rsc, node, rsc_op, data_set)
&& pe_get_failcount(node, rsc, NULL, pe_fc_effective, NULL,
data_set)) {
pe__clear_failcount(rsc, node, "action definition changed",
data_set);
}
}
}
g_list_free(sorted_op_list);
}
static GListPtr
find_rsc_list(GListPtr result, pe_resource_t * rsc, const char *id, gboolean renamed_clones,
gboolean partial, pe_working_set_t * data_set)
{
GListPtr gIter = NULL;
gboolean match = FALSE;
if (id == NULL) {
return NULL;
}
if (rsc == NULL) {
if (data_set == NULL) {
return NULL;
}
for (gIter = data_set->resources; gIter != NULL; gIter = gIter->next) {
pe_resource_t *child = (pe_resource_t *) gIter->data;
result = find_rsc_list(result, child, id, renamed_clones, partial,
NULL);
}
return result;
}
if (partial) {
if (strstr(rsc->id, id)) {
match = TRUE;
} else if (renamed_clones && rsc->clone_name && strstr(rsc->clone_name, id)) {
match = TRUE;
}
} else {
if (strcmp(rsc->id, id) == 0) {
match = TRUE;
} else if (renamed_clones && rsc->clone_name && strcmp(rsc->clone_name, id) == 0) {
match = TRUE;
}
}
if (match) {
result = g_list_prepend(result, rsc);
}
if (rsc->children) {
gIter = rsc->children;
for (; gIter != NULL; gIter = gIter->next) {
pe_resource_t *child = (pe_resource_t *) gIter->data;
result = find_rsc_list(result, child, id, renamed_clones, partial, NULL);
}
}
return result;
}
static void
check_actions(pe_working_set_t * data_set)
{
const char *id = NULL;
pe_node_t *node = NULL;
xmlNode *lrm_rscs = NULL;
xmlNode *status = get_object_root(XML_CIB_TAG_STATUS, data_set->input);
xmlNode *node_state = NULL;
for (node_state = __xml_first_child_element(status); node_state != NULL;
node_state = __xml_next_element(node_state)) {
if (pcmk__str_eq((const char *)node_state->name, XML_CIB_TAG_STATE, pcmk__str_none)) {
id = crm_element_value(node_state, XML_ATTR_ID);
lrm_rscs = find_xml_node(node_state, XML_CIB_TAG_LRM, FALSE);
lrm_rscs = find_xml_node(lrm_rscs, XML_LRM_TAG_RESOURCES, FALSE);
node = pe_find_node_id(data_set->nodes, id);
if (node == NULL) {
continue;
/* Still need to check actions for a maintenance node to cancel existing monitor operations */
} else if (can_run_resources(node) == FALSE && node->details->maintenance == FALSE) {
crm_trace("Skipping param check for %s: can't run resources",
node->details->uname);
continue;
}
crm_trace("Processing node %s", node->details->uname);
if (node->details->online
|| pcmk_is_set(data_set->flags, pe_flag_stonith_enabled)) {
xmlNode *rsc_entry = NULL;
for (rsc_entry = __xml_first_child_element(lrm_rscs);
rsc_entry != NULL;
rsc_entry = __xml_next_element(rsc_entry)) {
if (pcmk__str_eq((const char *)rsc_entry->name, XML_LRM_TAG_RESOURCE, pcmk__str_none)) {
if (xml_has_children(rsc_entry)) {
GListPtr gIter = NULL;
GListPtr result = NULL;
const char *rsc_id = ID(rsc_entry);
CRM_CHECK(rsc_id != NULL, return);
result = find_rsc_list(NULL, NULL, rsc_id, TRUE, FALSE, data_set);
for (gIter = result; gIter != NULL; gIter = gIter->next) {
pe_resource_t *rsc = (pe_resource_t *) gIter->data;
if (rsc->variant != pe_native) {
continue;
}
check_actions_for(rsc_entry, rsc, node, data_set);
}
g_list_free(result);
}
}
}
}
}
}
}
static void
apply_placement_constraints(pe_working_set_t * data_set)
{
for (GList *gIter = data_set->placement_constraints;
gIter != NULL; gIter = gIter->next) {
pe__location_t *cons = gIter->data;
cons->rsc_lh->cmds->rsc_location(cons->rsc_lh, cons);
}
}
static gboolean
failcount_clear_action_exists(pe_node_t * node, pe_resource_t * rsc)
{
gboolean rc = FALSE;
GList *list = pe__resource_actions(rsc, node, CRM_OP_CLEAR_FAILCOUNT, TRUE);
if (list) {
rc = TRUE;
}
g_list_free(list);
return rc;
}
/*!
* \internal
* \brief Force resource away if failures hit migration threshold
*
* \param[in,out] rsc Resource to check for failures
* \param[in,out] node Node to check for failures
* \param[in,out] data_set Cluster working set to update
*/
static void
check_migration_threshold(pe_resource_t *rsc, pe_node_t *node,
pe_working_set_t *data_set)
{
int fail_count, countdown;
pe_resource_t *failed;
/* Migration threshold of 0 means never force away */
if (rsc->migration_threshold == 0) {
return;
}
// If we're ignoring failures, also ignore the migration threshold
if (pcmk_is_set(rsc->flags, pe_rsc_failure_ignored)) {
return;
}
/* If there are no failures, there's no need to force away */
fail_count = pe_get_failcount(node, rsc, NULL,
pe_fc_effective|pe_fc_fillers, NULL,
data_set);
if (fail_count <= 0) {
return;
}
/* How many more times recovery will be tried on this node */
countdown = QB_MAX(rsc->migration_threshold - fail_count, 0);
/* If failed resource has a parent, we'll force the parent away */
failed = rsc;
if (!pcmk_is_set(rsc->flags, pe_rsc_unique)) {
failed = uber_parent(rsc);
}
if (countdown == 0) {
resource_location(failed, node, -INFINITY, "__fail_limit__", data_set);
crm_warn("Forcing %s away from %s after %d failures (max=%d)",
failed->id, node->details->uname, fail_count,
rsc->migration_threshold);
} else {
crm_info("%s can fail %d more times on %s before being forced off",
failed->id, countdown, node->details->uname);
}
}
static void
common_apply_stickiness(pe_resource_t * rsc, pe_node_t * node, pe_working_set_t * data_set)
{
if (rsc->children) {
GListPtr gIter = rsc->children;
for (; gIter != NULL; gIter = gIter->next) {
pe_resource_t *child_rsc = (pe_resource_t *) gIter->data;
common_apply_stickiness(child_rsc, node, data_set);
}
return;
}
if (pcmk_is_set(rsc->flags, pe_rsc_managed)
&& rsc->stickiness != 0 && pcmk__list_of_1(rsc->running_on)) {
pe_node_t *current = pe_find_node_id(rsc->running_on, node->details->id);
pe_node_t *match = pe_hash_table_lookup(rsc->allowed_nodes, node->details->id);
if (current == NULL) {
} else if ((match != NULL)
|| pcmk_is_set(data_set->flags, pe_flag_symmetric_cluster)) {
pe_resource_t *sticky_rsc = rsc;
resource_location(sticky_rsc, node, rsc->stickiness, "stickiness", data_set);
pe_rsc_debug(sticky_rsc, "Resource %s: preferring current location"
" (node=%s, weight=%d)", sticky_rsc->id,
node->details->uname, rsc->stickiness);
} else {
GHashTableIter iter;
pe_node_t *nIter = NULL;
pe_rsc_debug(rsc, "Ignoring stickiness for %s: the cluster is asymmetric"
" and node %s is not explicitly allowed", rsc->id, node->details->uname);
g_hash_table_iter_init(&iter, rsc->allowed_nodes);
while (g_hash_table_iter_next(&iter, NULL, (void **)&nIter)) {
crm_err("%s[%s] = %d", rsc->id, nIter->details->uname, nIter->weight);
}
}
}
/* Check the migration threshold only if a failcount clear action
* has not already been placed for this resource on the node.
* There is no sense in potentially forcing the resource from this
* node if the failcount is being reset anyway.
*
* @TODO A clear_failcount operation can be scheduled in stage4() via
* check_actions_for(), or in stage5() via check_params(). This runs in
* stage2(), 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 next transition.
*/
if (failcount_clear_action_exists(node, rsc) == FALSE) {
check_migration_threshold(rsc, node, data_set);
}
}
void
complex_set_cmds(pe_resource_t * rsc)
{
GListPtr gIter = rsc->children;
rsc->cmds = &resource_class_alloc_functions[rsc->variant];
for (; gIter != NULL; gIter = gIter->next) {
pe_resource_t *child_rsc = (pe_resource_t *) gIter->data;
complex_set_cmds(child_rsc);
}
}
void
set_alloc_actions(pe_working_set_t * data_set)
{
GListPtr gIter = data_set->resources;
for (; gIter != NULL; gIter = gIter->next) {
pe_resource_t *rsc = (pe_resource_t *) gIter->data;
complex_set_cmds(rsc);
}
}
static void
calculate_system_health(gpointer gKey, gpointer gValue, gpointer user_data)
{
const char *key = (const char *)gKey;
const char *value = (const char *)gValue;
int *system_health = (int *)user_data;
if (!gKey || !gValue || !user_data) {
return;
}
if (pcmk__starts_with(key, "#health")) {
int score;
/* Convert the value into an integer */
score = char2score(value);
/* Add it to the running total */
*system_health = pe__add_scores(score, *system_health);
}
}
static gboolean
apply_system_health(pe_working_set_t * data_set)
{
GListPtr gIter = NULL;
const char *health_strategy = pe_pref(data_set->config_hash, "node-health-strategy");
int base_health = 0;
if (pcmk__str_eq(health_strategy, "none", pcmk__str_null_matches | pcmk__str_casei)) {
/* Prevent any accidental health -> score translation */
pcmk__score_red = 0;
pcmk__score_yellow = 0;
pcmk__score_green = 0;
return TRUE;
} else if (pcmk__str_eq(health_strategy, "migrate-on-red", pcmk__str_casei)) {
/* Resources on nodes which have health values of red are
* weighted away from that node.
*/
pcmk__score_red = -INFINITY;
pcmk__score_yellow = 0;
pcmk__score_green = 0;
} else if (pcmk__str_eq(health_strategy, "only-green", pcmk__str_casei)) {
/* Resources on nodes which have health values of red or yellow
* are forced away from that node.
*/
pcmk__score_red = -INFINITY;
pcmk__score_yellow = -INFINITY;
pcmk__score_green = 0;
} else if (pcmk__str_eq(health_strategy, "progressive", pcmk__str_casei)) {
/* Same as the above, but use the r/y/g scores provided by the user
* Defaults are provided by the pe_prefs table
* Also, custom health "base score" can be used
*/
base_health = crm_parse_int(pe_pref(data_set->config_hash, "node-health-base"), "0");
} else if (pcmk__str_eq(health_strategy, "custom", pcmk__str_casei)) {
/* Requires the admin to configure the rsc_location constaints for
* processing the stored health scores
*/
/* TODO: Check for the existence of appropriate node health constraints */
return TRUE;
} else {
crm_err("Unknown node health strategy: %s", health_strategy);
return FALSE;
}
crm_info("Applying automated node health strategy: %s", health_strategy);
for (gIter = data_set->nodes; gIter != NULL; gIter = gIter->next) {
int system_health = base_health;
pe_node_t *node = (pe_node_t *) gIter->data;
/* Search through the node hash table for system health entries. */
g_hash_table_foreach(node->details->attrs, calculate_system_health, &system_health);
crm_info(" Node %s has an combined system health of %d",
node->details->uname, system_health);
/* If the health is non-zero, then create a new rsc2node so that the
* weight will be added later on.
*/
if (system_health != 0) {
GListPtr gIter2 = data_set->resources;
for (; gIter2 != NULL; gIter2 = gIter2->next) {
pe_resource_t *rsc = (pe_resource_t *) gIter2->data;
rsc2node_new(health_strategy, rsc, system_health, NULL, node, data_set);
}
}
}
return TRUE;
}
gboolean
stage0(pe_working_set_t * data_set)
{
xmlNode *cib_constraints = get_object_root(XML_CIB_TAG_CONSTRAINTS, data_set->input);
if (data_set->input == NULL) {
return FALSE;
}
if (!pcmk_is_set(data_set->flags, pe_flag_have_status)) {
crm_trace("Calculating status");
cluster_status(data_set);
}
set_alloc_actions(data_set);
apply_system_health(data_set);
unpack_constraints(cib_constraints, data_set);
return TRUE;
}
/*
* Check nodes for resources started outside of the LRM
*/
gboolean
probe_resources(pe_working_set_t * data_set)
{
pe_action_t *probe_node_complete = NULL;
for (GListPtr gIter = data_set->nodes; gIter != NULL; gIter = gIter->next) {
pe_node_t *node = (pe_node_t *) gIter->data;
const char *probed = pe_node_attribute_raw(node, CRM_OP_PROBED);
if (node->details->online == FALSE) {
if (pe__is_remote_node(node) && node->details->remote_rsc
&& (get_remote_node_state(node) == remote_state_failed)) {
pe_fence_node(data_set, node, "the connection is unrecoverable", FALSE);
}
continue;
} else if (node->details->unclean) {
continue;
} else if (node->details->rsc_discovery_enabled == FALSE) {
/* resource discovery is disabled for this node */
continue;
}
if (probed != NULL && crm_is_true(probed) == FALSE) {
pe_action_t *probe_op = custom_action(NULL, crm_strdup_printf("%s-%s", CRM_OP_REPROBE, node->details->uname),
CRM_OP_REPROBE, node, FALSE, TRUE, data_set);
add_hash_param(probe_op->meta, XML_ATTR_TE_NOWAIT, XML_BOOLEAN_TRUE);
continue;
}
for (GListPtr gIter2 = data_set->resources; gIter2 != NULL; gIter2 = gIter2->next) {
pe_resource_t *rsc = (pe_resource_t *) gIter2->data;
rsc->cmds->create_probe(rsc, node, probe_node_complete, FALSE, data_set);
}
}
return TRUE;
}
static void
rsc_discover_filter(pe_resource_t *rsc, pe_node_t *node)
{
GListPtr gIter = rsc->children;
pe_resource_t *top = uber_parent(rsc);
pe_node_t *match;
if (rsc->exclusive_discover == FALSE && top->exclusive_discover == FALSE) {
return;
}
for (; gIter != NULL; gIter = gIter->next) {
pe_resource_t *child_rsc = (pe_resource_t *) gIter->data;
rsc_discover_filter(child_rsc, node);
}
match = g_hash_table_lookup(rsc->allowed_nodes, node->details->id);
if (match && match->rsc_discover_mode != pe_discover_exclusive) {
match->weight = -INFINITY;
}
}
static time_t
shutdown_time(pe_node_t *node, pe_working_set_t *data_set)
{
const char *shutdown = pe_node_attribute_raw(node, XML_CIB_ATTR_SHUTDOWN);
time_t result = 0;
if (shutdown) {
errno = 0;
result = (time_t) crm_parse_ll(shutdown, NULL);
if (errno != 0) {
result = 0;
}
}
return result? result : get_effective_time(data_set);
}
static void
apply_shutdown_lock(pe_resource_t *rsc, pe_working_set_t *data_set)
{
const char *class;
// Only primitives and (uncloned) groups may be locked
if (rsc->variant == pe_group) {
for (GList *item = rsc->children; item != NULL;
item = item->next) {
apply_shutdown_lock((pe_resource_t *) item->data, data_set);
}
} else if (rsc->variant != pe_native) {
return;
}
// Fence devices and remote connections can't be locked
class = crm_element_value(rsc->xml, XML_AGENT_ATTR_CLASS);
if (pcmk__str_eq(class, PCMK_RESOURCE_CLASS_STONITH, pcmk__str_null_matches)
|| pe__resource_is_remote_conn(rsc, data_set)) {
return;
}
if (rsc->lock_node != NULL) {
// The lock was obtained from resource history
if (rsc->running_on != 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.
*/
pe_rsc_info(rsc,
"Cancelling shutdown lock because %s is already active",
rsc->id);
pe__clear_resource_history(rsc, rsc->lock_node, data_set);
rsc->lock_node = NULL;
rsc->lock_time = 0;
}
// Only a resource active on exactly one node can be locked
} else if (pcmk__list_of_1(rsc->running_on)) {
pe_node_t *node = rsc->running_on->data;
if (node->details->shutdown) {
if (node->details->unclean) {
pe_rsc_debug(rsc, "Not locking %s to unclean %s for shutdown",
rsc->id, node->details->uname);
} else {
rsc->lock_node = node;
rsc->lock_time = shutdown_time(node, data_set);
}
}
}
if (rsc->lock_node == NULL) {
// No lock needed
return;
}
if (data_set->shutdown_lock > 0) {
time_t lock_expiration = rsc->lock_time + data_set->shutdown_lock;
pe_rsc_info(rsc, "Locking %s to %s due to shutdown (expires @%lld)",
rsc->id, rsc->lock_node->details->uname,
(long long) lock_expiration);
pe__update_recheck_time(++lock_expiration, data_set);
} else {
pe_rsc_info(rsc, "Locking %s to %s due to shutdown",
rsc->id, rsc->lock_node->details->uname);
}
// If resource is locked to one node, ban it from all other nodes
for (GList *item = data_set->nodes; item != NULL; item = item->next) {
pe_node_t *node = item->data;
if (strcmp(node->details->uname, rsc->lock_node->details->uname)) {
resource_location(rsc, node, -CRM_SCORE_INFINITY,
XML_CONFIG_ATTR_SHUTDOWN_LOCK, data_set);
}
}
}
/*
* \internal
* \brief Stage 2 of cluster status: apply node-specific criteria
*
* Count known nodes, and apply location constraints, stickiness, and exclusive
* resource discovery.
*/
gboolean
stage2(pe_working_set_t * data_set)
{
GListPtr gIter = NULL;
if (pcmk_is_set(data_set->flags, pe_flag_shutdown_lock)) {
for (gIter = data_set->resources; gIter != NULL; gIter = gIter->next) {
apply_shutdown_lock((pe_resource_t *) gIter->data, data_set);
}
}
if (!pcmk_is_set(data_set->flags, pe_flag_no_compat)) {
// @COMPAT API backward compatibility
for (gIter = data_set->nodes; gIter != NULL; gIter = gIter->next) {
pe_node_t *node = (pe_node_t *) gIter->data;
if (node && (node->weight >= 0) && node->details->online
&& (node->details->type != node_ping)) {
data_set->max_valid_nodes++;
}
}
}
crm_trace("Applying placement constraints");
apply_placement_constraints(data_set);
gIter = data_set->nodes;
for (; gIter != NULL; gIter = gIter->next) {
GListPtr gIter2 = NULL;
pe_node_t *node = (pe_node_t *) gIter->data;
gIter2 = data_set->resources;
for (; gIter2 != NULL; gIter2 = gIter2->next) {
pe_resource_t *rsc = (pe_resource_t *) gIter2->data;
common_apply_stickiness(rsc, node, data_set);
rsc_discover_filter(rsc, node);
}
}
return TRUE;
}
/*
* Create internal resource constraints before allocation
*/
gboolean
stage3(pe_working_set_t * data_set)
{
GListPtr gIter = data_set->resources;
for (; gIter != NULL; gIter = gIter->next) {
pe_resource_t *rsc = (pe_resource_t *) gIter->data;
rsc->cmds->internal_constraints(rsc, data_set);
}
return TRUE;
}
/*
* Check for orphaned or redefined actions
*/
gboolean
stage4(pe_working_set_t * data_set)
{
check_actions(data_set);
return TRUE;
}
static void *
convert_const_pointer(const void *ptr)
{
/* Worst function ever */
return (void *)ptr;
}
static gint
sort_rsc_process_order(gconstpointer a, gconstpointer b, gpointer data)
{
int rc = 0;
int r1_weight = -INFINITY;
int r2_weight = -INFINITY;
const char *reason = "existence";
const GListPtr nodes = (GListPtr) data;
const pe_resource_t *resource1 = a;
const pe_resource_t *resource2 = b;
pe_node_t *r1_node = NULL;
pe_node_t *r2_node = NULL;
GListPtr gIter = NULL;
GHashTable *r1_nodes = NULL;
GHashTable *r2_nodes = NULL;
reason = "priority";
r1_weight = resource1->priority;
r2_weight = resource2->priority;
if (r1_weight > r2_weight) {
rc = -1;
goto done;
}
if (r1_weight < r2_weight) {
rc = 1;
goto done;
}
reason = "no node list";
if (nodes == NULL) {
goto done;
}
r1_nodes = pcmk__native_merge_weights(convert_const_pointer(resource1),
resource1->id, NULL, NULL, 1,
pe_weights_forward | pe_weights_init);
pe__show_node_weights(true, NULL, resource1->id, r1_nodes);
r2_nodes = pcmk__native_merge_weights(convert_const_pointer(resource2),
resource2->id, NULL, NULL, 1,
pe_weights_forward | pe_weights_init);
pe__show_node_weights(true, NULL, resource2->id, r2_nodes);
/* Current location score */
reason = "current location";
r1_weight = -INFINITY;
r2_weight = -INFINITY;
if (resource1->running_on) {
r1_node = pe__current_node(resource1);
r1_node = g_hash_table_lookup(r1_nodes, r1_node->details->id);
if (r1_node != NULL) {
r1_weight = r1_node->weight;
}
}
if (resource2->running_on) {
r2_node = pe__current_node(resource2);
r2_node = g_hash_table_lookup(r2_nodes, r2_node->details->id);
if (r2_node != NULL) {
r2_weight = r2_node->weight;
}
}
if (r1_weight > r2_weight) {
rc = -1;
goto done;
}
if (r1_weight < r2_weight) {
rc = 1;
goto done;
}
reason = "score";
for (gIter = nodes; gIter != NULL; gIter = gIter->next) {
pe_node_t *node = (pe_node_t *) gIter->data;
r1_node = NULL;
r2_node = NULL;
r1_weight = -INFINITY;
if (r1_nodes) {
r1_node = g_hash_table_lookup(r1_nodes, node->details->id);
}
if (r1_node) {
r1_weight = r1_node->weight;
}
r2_weight = -INFINITY;
if (r2_nodes) {
r2_node = g_hash_table_lookup(r2_nodes, node->details->id);
}
if (r2_node) {
r2_weight = r2_node->weight;
}
if (r1_weight > r2_weight) {
rc = -1;
goto done;
}
if (r1_weight < r2_weight) {
rc = 1;
goto done;
}
}
done:
crm_trace("%s (%d) on %s %c %s (%d) on %s: %s",
resource1->id, r1_weight, r1_node ? r1_node->details->id : "n/a",
rc < 0 ? '>' : rc > 0 ? '<' : '=',
resource2->id, r2_weight, r2_node ? r2_node->details->id : "n/a", reason);
if (r1_nodes) {
g_hash_table_destroy(r1_nodes);
}
if (r2_nodes) {
g_hash_table_destroy(r2_nodes);
}
return rc;
}
static void
allocate_resources(pe_working_set_t * data_set)
{
GListPtr gIter = NULL;
if (pcmk_is_set(data_set->flags, pe_flag_have_remote_nodes)) {
/* Allocate remote connection resources first (which will also allocate
* any colocation dependencies). If the connection is migrating, always
* prefer the partial migration target.
*/
for (gIter = data_set->resources; gIter != NULL; gIter = gIter->next) {
pe_resource_t *rsc = (pe_resource_t *) gIter->data;
if (rsc->is_remote_node == FALSE) {
continue;
}
pe_rsc_trace(rsc, "Allocating remote connection resource '%s'",
rsc->id);
rsc->cmds->allocate(rsc, rsc->partial_migration_target, data_set);
}
}
/* now do the rest of the resources */
for (gIter = data_set->resources; gIter != NULL; gIter = gIter->next) {
pe_resource_t *rsc = (pe_resource_t *) gIter->data;
if (rsc->is_remote_node == TRUE) {
continue;
}
pe_rsc_trace(rsc, "Allocating %s resource '%s'",
crm_element_name(rsc->xml), rsc->id);
rsc->cmds->allocate(rsc, NULL, data_set);
}
}
/* We always use pe_order_preserve with these convenience functions to exempt
* internally generated constraints from the prohibition of user constraints
* involving remote connection resources.
*
* The start ordering additionally uses pe_order_runnable_left so that the
* specified action is not runnable if the start is not runnable.
*/
static inline void
order_start_then_action(pe_resource_t *lh_rsc, pe_action_t *rh_action,
enum pe_ordering extra, pe_working_set_t *data_set)
{
if (lh_rsc && rh_action && data_set) {
custom_action_order(lh_rsc, start_key(lh_rsc), NULL,
rh_action->rsc, NULL, rh_action,
pe_order_preserve | pe_order_runnable_left | extra,
data_set);
}
}
static inline void
order_action_then_stop(pe_action_t *lh_action, pe_resource_t *rh_rsc,
enum pe_ordering extra, pe_working_set_t *data_set)
{
if (lh_action && rh_rsc && data_set) {
custom_action_order(lh_action->rsc, NULL, lh_action,
rh_rsc, stop_key(rh_rsc), NULL,
pe_order_preserve | extra, data_set);
}
}
// Clear fail counts for orphaned rsc on all online nodes
static void
cleanup_orphans(pe_resource_t * rsc, pe_working_set_t * data_set)
{
GListPtr gIter = NULL;
for (gIter = data_set->nodes; gIter != NULL; gIter = gIter->next) {
pe_node_t *node = (pe_node_t *) gIter->data;
if (node->details->online
&& pe_get_failcount(node, rsc, NULL, pe_fc_effective, NULL,
data_set)) {
pe_action_t *clear_op = NULL;
clear_op = pe__clear_failcount(rsc, node, "it is orphaned",
data_set);
/* We can't use order_action_then_stop() here because its
* pe_order_preserve breaks things
*/
custom_action_order(clear_op->rsc, NULL, clear_op,
rsc, stop_key(rsc), NULL,
pe_order_optional, data_set);
}
}
}
gboolean
stage5(pe_working_set_t * data_set)
{
GListPtr gIter = NULL;
int log_prio = show_utilization? LOG_STDOUT : LOG_TRACE;
if (!pcmk__str_eq(data_set->placement_strategy, "default", pcmk__str_casei)) {
GListPtr nodes = g_list_copy(data_set->nodes);
nodes = sort_nodes_by_weight(nodes, NULL, data_set);
data_set->resources =
g_list_sort_with_data(data_set->resources, sort_rsc_process_order, nodes);
g_list_free(nodes);
}
gIter = data_set->nodes;
for (; gIter != NULL; gIter = gIter->next) {
pe_node_t *node = (pe_node_t *) gIter->data;
dump_node_capacity(log_prio, "Original", node);
}
crm_trace("Allocating services");
/* Take (next) highest resource, assign it and create its actions */
allocate_resources(data_set);
gIter = data_set->nodes;
for (; gIter != NULL; gIter = gIter->next) {
pe_node_t *node = (pe_node_t *) gIter->data;
dump_node_capacity(log_prio, "Remaining", node);
}
// Process deferred action checks
pe__foreach_param_check(data_set, check_params);
pe__free_param_checks(data_set);
if (pcmk_is_set(data_set->flags, pe_flag_startup_probes)) {
crm_trace("Calculating needed probes");
/* This code probably needs optimization
* ptest -x with 100 nodes, 100 clones and clone-max=100:
With probes:
ptest[14781]: 2010/09/27_17:56:46 notice: TRACE: do_calculations: pengine.c:258 Calculate cluster status
ptest[14781]: 2010/09/27_17:56:46 notice: TRACE: do_calculations: pengine.c:278 Applying placement constraints
ptest[14781]: 2010/09/27_17:56:47 notice: TRACE: do_calculations: pengine.c:285 Create internal constraints
ptest[14781]: 2010/09/27_17:56:47 notice: TRACE: do_calculations: pengine.c:292 Check actions
ptest[14781]: 2010/09/27_17:56:48 notice: TRACE: do_calculations: pengine.c:299 Allocate resources
ptest[14781]: 2010/09/27_17:56:48 notice: TRACE: stage5: allocate.c:881 Allocating services
ptest[14781]: 2010/09/27_17:56:49 notice: TRACE: stage5: allocate.c:894 Calculating needed probes
ptest[14781]: 2010/09/27_17:56:51 notice: TRACE: stage5: allocate.c:899 Creating actions
ptest[14781]: 2010/09/27_17:56:52 notice: TRACE: stage5: allocate.c:905 Creating done
ptest[14781]: 2010/09/27_17:56:52 notice: TRACE: do_calculations: pengine.c:306 Processing fencing and shutdown cases
ptest[14781]: 2010/09/27_17:56:52 notice: TRACE: do_calculations: pengine.c:313 Applying ordering constraints
36s
ptest[14781]: 2010/09/27_17:57:28 notice: TRACE: do_calculations: pengine.c:320 Create transition graph
Without probes:
ptest[14637]: 2010/09/27_17:56:21 notice: TRACE: do_calculations: pengine.c:258 Calculate cluster status
ptest[14637]: 2010/09/27_17:56:22 notice: TRACE: do_calculations: pengine.c:278 Applying placement constraints
ptest[14637]: 2010/09/27_17:56:22 notice: TRACE: do_calculations: pengine.c:285 Create internal constraints
ptest[14637]: 2010/09/27_17:56:22 notice: TRACE: do_calculations: pengine.c:292 Check actions
ptest[14637]: 2010/09/27_17:56:23 notice: TRACE: do_calculations: pengine.c:299 Allocate resources
ptest[14637]: 2010/09/27_17:56:23 notice: TRACE: stage5: allocate.c:881 Allocating services
ptest[14637]: 2010/09/27_17:56:24 notice: TRACE: stage5: allocate.c:899 Creating actions
ptest[14637]: 2010/09/27_17:56:25 notice: TRACE: stage5: allocate.c:905 Creating done
ptest[14637]: 2010/09/27_17:56:25 notice: TRACE: do_calculations: pengine.c:306 Processing fencing and shutdown cases
ptest[14637]: 2010/09/27_17:56:25 notice: TRACE: do_calculations: pengine.c:313 Applying ordering constraints
ptest[14637]: 2010/09/27_17:56:25 notice: TRACE: do_calculations: pengine.c:320 Create transition graph
*/
probe_resources(data_set);
}
crm_trace("Handle orphans");
if (pcmk_is_set(data_set->flags, pe_flag_stop_rsc_orphans)) {
for (gIter = data_set->resources; gIter != NULL; gIter = gIter->next) {
pe_resource_t *rsc = (pe_resource_t *) gIter->data;
/* There's no need to recurse into rsc->children because those
* should just be unallocated clone instances.
*/
if (pcmk_is_set(rsc->flags, pe_rsc_orphan)) {
cleanup_orphans(rsc, data_set);
}
}
}
crm_trace("Creating actions");
for (gIter = data_set->resources; gIter != NULL; gIter = gIter->next) {
pe_resource_t *rsc = (pe_resource_t *) gIter->data;
rsc->cmds->create_actions(rsc, data_set);
}
crm_trace("Creating done");
return TRUE;
}
static gboolean
is_managed(const pe_resource_t * rsc)
{
GListPtr gIter = rsc->children;
if (pcmk_is_set(rsc->flags, pe_rsc_managed)) {
return TRUE;
}
for (; gIter != NULL; gIter = gIter->next) {
pe_resource_t *child_rsc = (pe_resource_t *) gIter->data;
if (is_managed(child_rsc)) {
return TRUE;
}
}
return FALSE;
}
static gboolean
any_managed_resources(pe_working_set_t * data_set)
{
GListPtr gIter = data_set->resources;
for (; gIter != NULL; gIter = gIter->next) {
pe_resource_t *rsc = (pe_resource_t *) gIter->data;
if (is_managed(rsc)) {
return TRUE;
}
}
return FALSE;
}
/*!
* \internal
* \brief Create pseudo-op for guest node fence, and order relative to it
*
* \param[in] node Guest node to fence
* \param[in] data_set Working set of CIB state
*/
static void
fence_guest(pe_node_t *node, pe_working_set_t *data_set)
{
pe_resource_t *container = node->details->remote_rsc->container;
pe_action_t *stop = NULL;
pe_action_t *stonith_op = NULL;
/* The fence action is just a label; we don't do anything differently for
* off vs. reboot. We specify it explicitly, rather than let it default to
* cluster's default action, because we are not _initiating_ fencing -- we
* are creating a pseudo-event to describe fencing that is already occurring
* by other means (container recovery).
*/
const char *fence_action = "off";
/* Check whether guest's container resource has any explicit stop or
* start (the stop may be implied by fencing of the guest's host).
*/
if (container) {
stop = find_first_action(container->actions, NULL, CRMD_ACTION_STOP, NULL);
if (find_first_action(container->actions, NULL, CRMD_ACTION_START, NULL)) {
fence_action = "reboot";
}
}
/* Create a fence pseudo-event, so we have an event to order actions
* against, and the controller can always detect it.
*/
stonith_op = pe_fence_op(node, fence_action, FALSE, "guest is unclean", FALSE, data_set);
update_action_flags(stonith_op, pe_action_pseudo | pe_action_runnable,
__FUNCTION__, __LINE__);
/* We want to imply stops/demotes after the guest is stopped, not wait until
* it is restarted, so we always order pseudo-fencing after stop, not start
* (even though start might be closer to what is done for a real reboot).
*/
if ((stop != NULL) && pcmk_is_set(stop->flags, pe_action_pseudo)) {
pe_action_t *parent_stonith_op = pe_fence_op(stop->node, NULL, FALSE, NULL, FALSE, data_set);
crm_info("Implying guest node %s is down (action %d) after %s fencing",
node->details->uname, stonith_op->id, stop->node->details->uname);
order_actions(parent_stonith_op, stonith_op,
pe_order_runnable_left|pe_order_implies_then);
} else if (stop) {
order_actions(stop, stonith_op,
pe_order_runnable_left|pe_order_implies_then);
crm_info("Implying guest node %s is down (action %d) "
"after container %s is stopped (action %d)",
node->details->uname, stonith_op->id,
container->id, stop->id);
} else {
/* If we're fencing the guest node but there's no stop for the guest
* resource, we must think the guest is already stopped. However, we may
* think so because its resource history was just cleaned. To avoid
* unnecessarily considering the guest node down if it's really up,
* order the pseudo-fencing after any stop of the connection resource,
* which will be ordered after any container (re-)probe.
*/
stop = find_first_action(node->details->remote_rsc->actions, NULL,
RSC_STOP, NULL);
if (stop) {
order_actions(stop, stonith_op, pe_order_optional);
crm_info("Implying guest node %s is down (action %d) "
"after connection is stopped (action %d)",
node->details->uname, stonith_op->id, stop->id);
} else {
/* Not sure why we're fencing, but everything must already be
* cleanly stopped.
*/
crm_info("Implying guest node %s is down (action %d) ",
node->details->uname, stonith_op->id);
}
}
/* Order/imply other actions relative to pseudo-fence as with real fence */
pcmk__order_vs_fence(stonith_op, data_set);
}
/*
* Create dependencies for stonith and shutdown operations
*/
gboolean
stage6(pe_working_set_t * data_set)
{
pe_action_t *dc_down = NULL;
pe_action_t *stonith_op = NULL;
gboolean integrity_lost = FALSE;
gboolean need_stonith = TRUE;
GListPtr gIter;
GListPtr stonith_ops = NULL;
GList *shutdown_ops = NULL;
/* Remote ordering constraints need to happen prior to calculating fencing
* because it is one more place we will mark the node as dirty.
*
* A nice side effect of doing them early is that apply_*_ordering() can be
* simpler because pe_fence_node() has already done some of the work.
*/
crm_trace("Creating remote ordering constraints");
apply_remote_node_ordering(data_set);
crm_trace("Processing fencing and shutdown cases");
if (any_managed_resources(data_set) == FALSE) {
crm_notice("Delaying fencing operations until there are resources to manage");
need_stonith = FALSE;
}
/* Check each node for stonith/shutdown */
for (gIter = data_set->nodes; gIter != NULL; gIter = gIter->next) {
pe_node_t *node = (pe_node_t *) gIter->data;
/* Guest nodes are "fenced" by recovering their container resource,
* so handle them separately.
*/
if (pe__is_guest_node(node)) {
if (node->details->remote_requires_reset && need_stonith
&& pe_can_fence(data_set, node)) {
fence_guest(node, data_set);
}
continue;
}
stonith_op = NULL;
if (node->details->unclean
&& need_stonith && pe_can_fence(data_set, node)) {
stonith_op = pe_fence_op(node, NULL, FALSE, "node is unclean", FALSE, data_set);
pe_warn("Scheduling Node %s for STONITH", node->details->uname);
pcmk__order_vs_fence(stonith_op, data_set);
if (node->details->is_dc) {
// Remember if the DC is being fenced
dc_down = stonith_op;
} else {
if (!pcmk_is_set(data_set->flags, pe_flag_concurrent_fencing)
&& (stonith_ops != 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((pe_action_t *) stonith_ops->data,
stonith_op, pe_order_optional);
}
// Remember all non-DC fencing actions in a separate list
stonith_ops = g_list_prepend(stonith_ops, stonith_op);
}
} else if (node->details->online && node->details->shutdown &&
/* TODO define what a shutdown op means for a remote node.
* For now we do not send shutdown operations for remote nodes, but
* if we can come up with a good use for this in the future, we will. */
pe__is_guest_or_remote_node(node) == FALSE) {
pe_action_t *down_op = sched_shutdown_op(node, data_set);
if (node->details->is_dc) {
// Remember if the DC is being shut down
dc_down = down_op;
} else {
// Remember non-DC shutdowns for later ordering
shutdown_ops = g_list_prepend(shutdown_ops, down_op);
}
}
if (node->details->unclean && stonith_op == NULL) {
integrity_lost = TRUE;
pe_warn("Node %s is unclean!", node->details->uname);
}
}
if (integrity_lost) {
if (!pcmk_is_set(data_set->flags, pe_flag_stonith_enabled)) {
pe_warn("YOUR RESOURCES ARE NOW LIKELY COMPROMISED");
pe_err("ENABLE STONITH TO KEEP YOUR RESOURCES SAFE");
} else if (!pcmk_is_set(data_set->flags, pe_flag_have_quorum)) {
crm_notice("Cannot fence unclean nodes until quorum is"
" attained (or no-quorum-policy is set to 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, CRM_OP_SHUTDOWN, pcmk__str_casei)) {
for (gIter = shutdown_ops; gIter != NULL; gIter = gIter->next) {
pe_action_t *node_stop = (pe_action_t *) gIter->data;
crm_debug("Ordering shutdown on %s before %s on DC %s",
node_stop->node->details->uname,
dc_down->task, dc_down->node->details->uname);
order_actions(node_stop, dc_down, pe_order_optional);
}
}
// Order any non-DC fencing before any DC fencing or shutdown
if (pcmk_is_set(data_set->flags, pe_flag_concurrent_fencing)) {
/* With concurrent fencing, order each non-DC fencing action
* separately before any DC fencing or shutdown.
*/
for (gIter = stonith_ops; gIter != NULL; gIter = gIter->next) {
order_actions((pe_action_t *) gIter->data, dc_down,
pe_order_optional);
}
} else if (stonith_ops) {
/* 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((pe_action_t *) stonith_ops->data, dc_down,
pe_order_optional);
}
}
g_list_free(stonith_ops);
g_list_free(shutdown_ops);
return TRUE;
}
/*
* Determine the sets of independent actions and the correct order for the
* actions in each set.
*
* Mark dependencies of un-runnable actions un-runnable
*
*/
static GListPtr
find_actions_by_task(GListPtr actions, pe_resource_t * rsc, const char *original_key)
{
GListPtr list = NULL;
list = find_actions(actions, original_key, NULL);
if (list == NULL) {
/* we're potentially searching a child of the original resource */
char *key = NULL;
char *task = NULL;
guint interval_ms = 0;
if (parse_op_key(original_key, NULL, &task, &interval_ms)) {
key = pcmk__op_key(rsc->id, task, interval_ms);
list = find_actions(actions, key, NULL);
} else {
crm_err("search key: %s", original_key);
}
free(key);
free(task);
}
return list;
}
static void
rsc_order_then(pe_action_t *lh_action, pe_resource_t *rsc,
pe__ordering_t *order)
{
GListPtr gIter = NULL;
GListPtr rh_actions = NULL;
pe_action_t *rh_action = NULL;
enum pe_ordering type;
CRM_CHECK(rsc != NULL, return);
CRM_CHECK(order != NULL, return);
type = order->type;
rh_action = order->rh_action;
crm_trace("Processing RH of ordering constraint %d", order->id);
if (rh_action != NULL) {
rh_actions = g_list_prepend(NULL, rh_action);
} else if (rsc != NULL) {
rh_actions = find_actions_by_task(rsc->actions, rsc, order->rh_action_task);
}
if (rh_actions == NULL) {
pe_rsc_trace(rsc, "No RH-Side (%s/%s) found for constraint..."
" ignoring", rsc->id, order->rh_action_task);
if (lh_action) {
pe_rsc_trace(rsc, "LH-Side was: %s", lh_action->uuid);
}
return;
}
if ((lh_action != NULL) && (lh_action->rsc == rsc)
&& pcmk_is_set(lh_action->flags, pe_action_dangle)) {
pe_rsc_trace(rsc, "Detected dangling operation %s -> %s", lh_action->uuid,
order->rh_action_task);
pe__clear_order_flags(type, pe_order_implies_then);
}
gIter = rh_actions;
for (; gIter != NULL; gIter = gIter->next) {
pe_action_t *rh_action_iter = (pe_action_t *) gIter->data;
if (lh_action) {
order_actions(lh_action, rh_action_iter, type);
} else if (type & pe_order_implies_then) {
update_action_flags(rh_action_iter, pe_action_runnable | pe_action_clear, __FUNCTION__, __LINE__);
crm_warn("Unrunnable %s 0x%.6x", rh_action_iter->uuid, type);
} else {
crm_warn("neither %s 0x%.6x", rh_action_iter->uuid, type);
}
}
g_list_free(rh_actions);
}
static void
rsc_order_first(pe_resource_t *lh_rsc, pe__ordering_t *order,
pe_working_set_t *data_set)
{
GListPtr gIter = NULL;
GListPtr lh_actions = NULL;
pe_action_t *lh_action = order->lh_action;
pe_resource_t *rh_rsc = order->rh_rsc;
crm_trace("Processing LH of ordering constraint %d", order->id);
CRM_ASSERT(lh_rsc != NULL);
if (lh_action != NULL) {
lh_actions = g_list_prepend(NULL, lh_action);
} else {
lh_actions = find_actions_by_task(lh_rsc->actions, lh_rsc, order->lh_action_task);
}
if (lh_actions == NULL && lh_rsc != rh_rsc) {
char *key = NULL;
char *op_type = NULL;
guint interval_ms = 0;
parse_op_key(order->lh_action_task, NULL, &op_type, &interval_ms);
key = pcmk__op_key(lh_rsc->id, op_type, interval_ms);
if (lh_rsc->fns->state(lh_rsc, TRUE) == RSC_ROLE_STOPPED && pcmk__str_eq(op_type, RSC_STOP, pcmk__str_casei)) {
free(key);
pe_rsc_trace(lh_rsc, "No LH-Side (%s/%s) found for constraint %d with %s - ignoring",
lh_rsc->id, order->lh_action_task, order->id, order->rh_action_task);
} else if (lh_rsc->fns->state(lh_rsc, TRUE) == RSC_ROLE_SLAVE && pcmk__str_eq(op_type, RSC_DEMOTE, pcmk__str_casei)) {
free(key);
pe_rsc_trace(lh_rsc, "No LH-Side (%s/%s) found for constraint %d with %s - ignoring",
lh_rsc->id, order->lh_action_task, order->id, order->rh_action_task);
} else {
pe_rsc_trace(lh_rsc, "No LH-Side (%s/%s) found for constraint %d with %s - creating",
lh_rsc->id, order->lh_action_task, order->id, order->rh_action_task);
lh_action = custom_action(lh_rsc, key, op_type, NULL, TRUE, TRUE, data_set);
lh_actions = g_list_prepend(NULL, lh_action);
}
free(op_type);
}
gIter = lh_actions;
for (; gIter != NULL; gIter = gIter->next) {
pe_action_t *lh_action_iter = (pe_action_t *) gIter->data;
if (rh_rsc == NULL && order->rh_action) {
rh_rsc = order->rh_action->rsc;
}
if (rh_rsc) {
rsc_order_then(lh_action_iter, rh_rsc, order);
} else if (order->rh_action) {
order_actions(lh_action_iter, order->rh_action, order->type);
}
}
g_list_free(lh_actions);
}
extern void update_colo_start_chain(pe_action_t *action,
pe_working_set_t *data_set);
static int
is_recurring_action(pe_action_t *action)
{
guint interval_ms;
if (pcmk__guint_from_hash(action->meta,
XML_LRM_ATTR_INTERVAL_MS, 0,
&interval_ms) != pcmk_rc_ok) {
return 0;
}
return (interval_ms > 0);
}
static void
apply_container_ordering(pe_action_t *action, pe_working_set_t *data_set)
{
/* VMs are also classified as containers for these purposes... in
* that they both involve a 'thing' running on a real or remote
* cluster node.
*
* This allows us to be smarter about the type and extent of
* recovery actions required in various scenarios
*/
pe_resource_t *remote_rsc = NULL;
pe_resource_t *container = NULL;
enum action_tasks task = text2task(action->task);
CRM_ASSERT(action->rsc);
CRM_ASSERT(action->node);
CRM_ASSERT(pe__is_guest_or_remote_node(action->node));
remote_rsc = action->node->details->remote_rsc;
CRM_ASSERT(remote_rsc);
container = remote_rsc->container;
CRM_ASSERT(container);
if (pcmk_is_set(container->flags, pe_rsc_failed)) {
pe_fence_node(data_set, action->node, "container failed", FALSE);
}
crm_trace("Order %s action %s relative to %s%s for %s%s",
action->task, action->uuid,
pcmk_is_set(remote_rsc->flags, pe_rsc_failed)? "failed " : "",
remote_rsc->id,
pcmk_is_set(container->flags, pe_rsc_failed)? "failed " : "",
container->id);
if (pcmk__strcase_any_of(action->task, CRMD_ACTION_MIGRATE, CRMD_ACTION_MIGRATED, NULL)) {
/* Migration ops map to "no_action", but we need to apply the same
* ordering as for stop or demote (see get_router_node()).
*/
task = stop_rsc;
}
switch (task) {
case start_rsc:
case action_promote:
/* Force resource recovery if the container is recovered */
order_start_then_action(container, action, pe_order_implies_then,
data_set);
/* Wait for the connection resource to be up too */
order_start_then_action(remote_rsc, action, pe_order_none,
data_set);
break;
case stop_rsc:
case action_demote:
if (pcmk_is_set(container->flags, pe_rsc_failed)) {
/* When the container representing a guest node fails, any stop
* or demote actions for resources running on the guest node
* are implied by the container stopping. This is similar to
* how fencing operations work for cluster nodes and remote
* nodes.
*/
} else {
/* Ensure the operation happens before the connection is brought
* down.
*
* If we really wanted to, we could order these after the
* connection start, IFF the container's current role was
* stopped (otherwise we re-introduce an ordering loop when the
* connection is restarting).
*/
order_action_then_stop(action, remote_rsc, pe_order_none,
data_set);
}
break;
default:
/* Wait for the connection resource to be up */
if (is_recurring_action(action)) {
/* In case we ever get the recovery logic wrong, force
* recurring monitors to be restarted, even if just
* the connection was re-established
*/
if(task != no_action) {
order_start_then_action(remote_rsc, action,
pe_order_implies_then, data_set);
}
} else {
order_start_then_action(remote_rsc, action, pe_order_none,
data_set);
}
break;
}
}
static enum remote_connection_state
get_remote_node_state(pe_node_t *node)
{
pe_resource_t *remote_rsc = NULL;
pe_node_t *cluster_node = NULL;
CRM_ASSERT(node);
remote_rsc = node->details->remote_rsc;
CRM_ASSERT(remote_rsc);
cluster_node = pe__current_node(remote_rsc);
/* If the cluster node the remote connection resource resides on
* is unclean or went offline, we can't process any operations
* on that remote node until after it starts elsewhere.
*/
if(remote_rsc->next_role == RSC_ROLE_STOPPED || remote_rsc->allocated_to == NULL) {
/* The connection resource is not going to run anywhere */
if (cluster_node && cluster_node->details->unclean) {
/* The remote connection is failed because its resource is on a
* failed node and can't be recovered elsewhere, so we must fence.
*/
return remote_state_failed;
}
if (!pcmk_is_set(remote_rsc->flags, pe_rsc_failed)) {
/* Connection resource is cleanly stopped */
return remote_state_stopped;
}
/* Connection resource is failed */
if ((remote_rsc->next_role == RSC_ROLE_STOPPED)
&& remote_rsc->remote_reconnect_ms
&& node->details->remote_was_fenced
&& !pe__shutdown_requested(node)) {
/* We won't know whether the connection is recoverable until the
* reconnect interval expires and we reattempt connection.
*/
return remote_state_unknown;
}
/* The remote connection is in a failed state. If there are any
* resources known to be active on it (stop) or in an unknown state
* (probe), we must assume the worst and fence it.
*/
return remote_state_failed;
} else if (cluster_node == NULL) {
/* Connection is recoverable but not currently running anywhere, see if we can recover it first */
return remote_state_unknown;
} else if(cluster_node->details->unclean == TRUE
|| cluster_node->details->online == FALSE) {
/* Connection is running on a dead node, see if we can recover it first */
return remote_state_resting;
} else if (pcmk__list_of_multiple(remote_rsc->running_on)
&& remote_rsc->partial_migration_source
&& remote_rsc->partial_migration_target) {
/* We're in the middle of migrating a connection resource,
* wait until after the resource migrates before performing
* any actions.
*/
return remote_state_resting;
}
return remote_state_alive;
}
/*!
* \internal
* \brief Order actions on remote node relative to actions for the connection
*/
static void
apply_remote_ordering(pe_action_t *action, pe_working_set_t *data_set)
{
pe_resource_t *remote_rsc = NULL;
enum action_tasks task = text2task(action->task);
enum remote_connection_state state = get_remote_node_state(action->node);
enum pe_ordering order_opts = pe_order_none;
if (action->rsc == NULL) {
return;
}
CRM_ASSERT(action->node);
CRM_ASSERT(pe__is_guest_or_remote_node(action->node));
remote_rsc = action->node->details->remote_rsc;
CRM_ASSERT(remote_rsc);
crm_trace("Order %s action %s relative to %s%s (state: %s)",
action->task, action->uuid,
pcmk_is_set(remote_rsc->flags, pe_rsc_failed)? "failed " : "",
remote_rsc->id, state2text(state));
if (pcmk__strcase_any_of(action->task, CRMD_ACTION_MIGRATE, CRMD_ACTION_MIGRATED, NULL)) {
/* Migration ops map to "no_action", but we need to apply the same
* ordering as for stop or demote (see get_router_node()).
*/
task = stop_rsc;
}
switch (task) {
case start_rsc:
case action_promote:
order_opts = pe_order_none;
if (state == remote_state_failed) {
/* Force recovery, by making this action required */
pe__set_order_flags(order_opts, pe_order_implies_then);
}
/* Ensure connection is up before running this action */
order_start_then_action(remote_rsc, action, order_opts, data_set);
break;
case stop_rsc:
if(state == remote_state_alive) {
order_action_then_stop(action, remote_rsc,
pe_order_implies_first, data_set);
} else if(state == remote_state_failed) {
/* The resource is active on the node, but since we don't have a
* valid connection, the only way to stop the resource is by
* fencing the node. There is no need to order the stop relative
* to the remote connection, since the stop will become implied
* by the fencing.
*/
pe_fence_node(data_set, action->node, "resources are active and the connection is unrecoverable", FALSE);
} else if(remote_rsc->next_role == RSC_ROLE_STOPPED) {
/* State must be remote_state_unknown or remote_state_stopped.
* Since the connection is not coming back up in this
* transition, stop this resource first.
*/
order_action_then_stop(action, remote_rsc,
pe_order_implies_first, data_set);
} else {
/* The connection is going to be started somewhere else, so
* stop this resource after that completes.
*/
order_start_then_action(remote_rsc, action, pe_order_none, data_set);
}
break;
case action_demote:
/* Only order this demote relative to the connection start if the
* connection isn't being torn down. Otherwise, the demote would be
* blocked because the connection start would not be allowed.
*/
if(state == remote_state_resting || state == remote_state_unknown) {
order_start_then_action(remote_rsc, action, pe_order_none,
data_set);
} /* Otherwise we can rely on the stop ordering */
break;
default:
/* Wait for the connection resource to be up */
if (is_recurring_action(action)) {
/* In case we ever get the recovery logic wrong, force
* recurring monitors to be restarted, even if just
* the connection was re-established
*/
order_start_then_action(remote_rsc, action,
pe_order_implies_then, data_set);
} else {
pe_node_t *cluster_node = pe__current_node(remote_rsc);
if(task == monitor_rsc && state == remote_state_failed) {
/* We would only be here if we do not know the
* state of the resource on the remote node.
* Since we have no way to find out, it is
* necessary to fence the node.
*/
pe_fence_node(data_set, action->node, "resources are in an unknown state and the connection is unrecoverable", FALSE);
}
if(cluster_node && state == remote_state_stopped) {
/* The connection is currently up, but is going
* down permanently.
*
* Make sure we check services are actually
* stopped _before_ we let the connection get
* closed
*/
order_action_then_stop(action, remote_rsc,
pe_order_runnable_left, data_set);
} else {
order_start_then_action(remote_rsc, action, pe_order_none,
data_set);
}
}
break;
}
}
static void
apply_remote_node_ordering(pe_working_set_t *data_set)
{
if (!pcmk_is_set(data_set->flags, pe_flag_have_remote_nodes)) {
return;
}
for (GListPtr gIter = data_set->actions; gIter != NULL; gIter = gIter->next) {
pe_action_t *action = (pe_action_t *) gIter->data;
pe_resource_t *remote = NULL;
// We are only interested in resource actions
if (action->rsc == NULL) {
continue;
}
/* Special case: If we are clearing the failcount of an actual
* remote connection resource, then make sure this happens before
* any start of the resource in this transition.
*/
if (action->rsc->is_remote_node &&
pcmk__str_eq(action->task, CRM_OP_CLEAR_FAILCOUNT, pcmk__str_casei)) {
custom_action_order(action->rsc,
NULL,
action,
action->rsc,
pcmk__op_key(action->rsc->id, RSC_START, 0),
NULL,
pe_order_optional,
data_set);
continue;
}
// We are only interested in actions allocated to a node
if (action->node == NULL) {
continue;
}
if (!pe__is_guest_or_remote_node(action->node)) {
continue;
}
/* We are only interested in real actions.
*
* @TODO This is probably wrong; pseudo-actions might be converted to
* real actions and vice versa later in update_actions() at the end of
* stage7().
*/
if (pcmk_is_set(action->flags, pe_action_pseudo)) {
continue;
}
remote = action->node->details->remote_rsc;
if (remote == NULL) {
// Orphaned
continue;
}
/* Another special case: if a resource is moving to a Pacemaker Remote
* node, order the stop on the original node after any start of the
* remote connection. This ensures that if the connection fails to
* start, we leave the resource running on the original node.
*/
if (pcmk__str_eq(action->task, RSC_START, pcmk__str_casei)) {
for (GList *item = action->rsc->actions; item != NULL;
item = item->next) {
pe_action_t *rsc_action = item->data;
if ((rsc_action->node->details != action->node->details)
&& pcmk__str_eq(rsc_action->task, RSC_STOP, pcmk__str_casei)) {
custom_action_order(remote, start_key(remote), NULL,
action->rsc, NULL, rsc_action,
pe_order_optional, data_set);
}
}
}
/* The action occurs across a remote connection, so create
* ordering constraints that guarantee the action occurs while the node
* is active (after start, before stop ... things like that).
*
* This is somewhat brittle in that we need to make sure the results of
* this ordering are compatible with the result of get_router_node().
* It would probably be better to add XML_LRM_ATTR_ROUTER_NODE as part
* of this logic rather than action2xml().
*/
if (remote->container) {
crm_trace("Container ordering for %s", action->uuid);
apply_container_ordering(action, data_set);
} else {
crm_trace("Remote ordering for %s", action->uuid);
apply_remote_ordering(action, data_set);
}
}
}
static gboolean
order_first_probe_unneeded(pe_action_t * probe, pe_action_t * rh_action)
{
/* No need to probe the resource on the node that is being
* unfenced. Otherwise it might introduce transition loop
* since probe will be performed after the node is
* unfenced.
*/
if (pcmk__str_eq(rh_action->task, CRM_OP_FENCE, pcmk__str_casei)
&& probe->node && rh_action->node
&& probe->node->details == rh_action->node->details) {
const char *op = g_hash_table_lookup(rh_action->meta, "stonith_action");
if (pcmk__str_eq(op, "on", pcmk__str_casei)) {
return TRUE;
}
}
// Shutdown waits for probe to complete only if it's on the same node
if ((pcmk__str_eq(rh_action->task, CRM_OP_SHUTDOWN, pcmk__str_casei))
&& probe->node && rh_action->node
&& probe->node->details != rh_action->node->details) {
return TRUE;
}
return FALSE;
}
static void
order_first_probes_imply_stops(pe_working_set_t * data_set)
{
GListPtr gIter = NULL;
for (gIter = data_set->ordering_constraints; gIter != NULL; gIter = gIter->next) {
pe__ordering_t *order = gIter->data;
enum pe_ordering order_type = pe_order_optional;
pe_resource_t *lh_rsc = order->lh_rsc;
pe_resource_t *rh_rsc = order->rh_rsc;
pe_action_t *lh_action = order->lh_action;
pe_action_t *rh_action = order->rh_action;
const char *lh_action_task = order->lh_action_task;
const char *rh_action_task = order->rh_action_task;
GListPtr probes = NULL;
GListPtr rh_actions = NULL;
GListPtr pIter = NULL;
if (lh_rsc == NULL) {
continue;
} else if (rh_rsc && lh_rsc == rh_rsc) {
continue;
}
if (lh_action == NULL && lh_action_task == NULL) {
continue;
}
if (rh_action == NULL && rh_action_task == NULL) {
continue;
}
/* Technically probe is expected to return "not running", which could be
* the alternative of stop action if the status of the resource is
* unknown yet.
*/
if (lh_action && !pcmk__str_eq(lh_action->task, RSC_STOP, pcmk__str_casei)) {
continue;
} else if (lh_action == NULL
&& lh_action_task
&& !pcmk__ends_with(lh_action_task, "_" RSC_STOP "_0")) {
continue;
}
/* Do not probe the resource inside of a stopping container. Otherwise
* it might introduce transition loop since probe will be performed
* after the container starts again.
*/
if (rh_rsc && lh_rsc->container == rh_rsc) {
if (rh_action && pcmk__str_eq(rh_action->task, RSC_STOP, pcmk__str_casei)) {
continue;
} else if (rh_action == NULL && rh_action_task
&& pcmk__ends_with(rh_action_task,"_" RSC_STOP "_0")) {
continue;
}
}
if (order->type == pe_order_none) {
continue;
}
// Preserve the order options for future filtering
if (pcmk_is_set(order->type, pe_order_apply_first_non_migratable)) {
pe__set_order_flags(order_type,
pe_order_apply_first_non_migratable);
}
if (pcmk_is_set(order->type, pe_order_same_node)) {
pe__set_order_flags(order_type, pe_order_same_node);
}
// Keep the order types for future filtering
if (order->type == pe_order_anti_colocation
|| order->type == pe_order_load) {
order_type = order->type;
}
probes = pe__resource_actions(lh_rsc, NULL, RSC_STATUS, FALSE);
if (probes == NULL) {
continue;
}
if (rh_action) {
rh_actions = g_list_prepend(rh_actions, rh_action);
} else if (rh_rsc && rh_action_task) {
rh_actions = find_actions(rh_rsc->actions, rh_action_task, NULL);
}
if (rh_actions == NULL) {
g_list_free(probes);
continue;
}
crm_trace("Processing for LH probe based on ordering constraint %s -> %s"
" (id=%d, type=%.6x)",
lh_action ? lh_action->uuid : lh_action_task,
rh_action ? rh_action->uuid : rh_action_task,
order->id, order->type);
for (pIter = probes; pIter != NULL; pIter = pIter->next) {
pe_action_t *probe = (pe_action_t *) pIter->data;
GListPtr rIter = NULL;
for (rIter = rh_actions; rIter != NULL; rIter = rIter->next) {
pe_action_t *rh_action_iter = (pe_action_t *) rIter->data;
if (order_first_probe_unneeded(probe, rh_action_iter)) {
continue;
}
order_actions(probe, rh_action_iter, order_type);
}
}
g_list_free(rh_actions);
g_list_free(probes);
}
}
static void
order_first_probe_then_restart_repromote(pe_action_t * probe,
pe_action_t * after,
pe_working_set_t * data_set)
{
GListPtr gIter = NULL;
bool interleave = FALSE;
pe_resource_t *compatible_rsc = NULL;
if (probe == NULL
|| probe->rsc == NULL
|| probe->rsc->variant != pe_native) {
return;
}
if (after == NULL
// Avoid running into any possible loop
|| pcmk_is_set(after->flags, pe_action_tracking)) {
return;
}
if (!pcmk__str_eq(probe->task, RSC_STATUS, pcmk__str_casei)) {
return;
}
pe__set_action_flags(after, pe_action_tracking);
crm_trace("Processing based on %s %s -> %s %s",
probe->uuid,
probe->node ? probe->node->details->uname: "",
after->uuid,
after->node ? after->node->details->uname : "");
if (after->rsc
/* Better not build a dependency directly with a clone/group.
* We are going to proceed through the ordering chain and build
* dependencies with its children.
*/
&& after->rsc->variant == pe_native
&& probe->rsc != after->rsc) {
GListPtr then_actions = NULL;
enum pe_ordering probe_order_type = pe_order_optional;
if (pcmk__str_eq(after->task, RSC_START, pcmk__str_casei)) {
then_actions = pe__resource_actions(after->rsc, NULL, RSC_STOP, FALSE);
} else if (pcmk__str_eq(after->task, RSC_PROMOTE, pcmk__str_casei)) {
then_actions = pe__resource_actions(after->rsc, NULL, RSC_DEMOTE, FALSE);
}
for (gIter = then_actions; gIter != NULL; gIter = gIter->next) {
pe_action_t *then = (pe_action_t *) gIter->data;
// Skip any pseudo action which for example is implied by fencing
if (pcmk_is_set(then->flags, pe_action_pseudo)) {
continue;
}
order_actions(probe, then, probe_order_type);
}
g_list_free(then_actions);
}
if (after->rsc
&& after->rsc->variant > pe_group) {
const char *interleave_s = g_hash_table_lookup(after->rsc->meta,
XML_RSC_ATTR_INTERLEAVE);
interleave = crm_is_true(interleave_s);
if (interleave) {
/* For an interleaved clone, we should build a dependency only
* with the relevant clone child.
*/
compatible_rsc = find_compatible_child(probe->rsc,
after->rsc,
RSC_ROLE_UNKNOWN,
FALSE, data_set);
}
}
for (gIter = after->actions_after; gIter != NULL; gIter = gIter->next) {
pe_action_wrapper_t *after_wrapper = (pe_action_wrapper_t *) gIter->data;
/* pe_order_implies_then is the reason why a required A.start
* implies/enforces B.start to be required too, which is the cause of
* B.restart/re-promote.
*
* Not sure about pe_order_implies_then_on_node though. It's now only
* used for unfencing case, which tends to introduce transition
* loops...
*/
if (!pcmk_is_set(after_wrapper->type, pe_order_implies_then)) {
/* The order type between a group/clone and its child such as
* B.start-> B_child.start is:
* pe_order_implies_first_printed | pe_order_runnable_left
*
* Proceed through the ordering chain and build dependencies with
* its children.
*/
if (after->rsc == NULL
|| after->rsc->variant < pe_group
|| probe->rsc->parent == after->rsc
|| after_wrapper->action->rsc == NULL
|| after_wrapper->action->rsc->variant > pe_group
|| after->rsc != after_wrapper->action->rsc->parent) {
continue;
}
/* Proceed to the children of a group or a non-interleaved clone.
* For an interleaved clone, proceed only to the relevant child.
*/
if (after->rsc->variant > pe_group
&& interleave == TRUE
&& (compatible_rsc == NULL
|| compatible_rsc != after_wrapper->action->rsc)) {
continue;
}
}
crm_trace("Proceeding through %s %s -> %s %s (type=0x%.6x)",
after->uuid,
after->node ? after->node->details->uname: "",
after_wrapper->action->uuid,
after_wrapper->action->node ? after_wrapper->action->node->details->uname : "",
after_wrapper->type);
order_first_probe_then_restart_repromote(probe, after_wrapper->action, data_set);
}
}
static void clear_actions_tracking_flag(pe_working_set_t * data_set)
{
GListPtr gIter = NULL;
for (gIter = data_set->actions; gIter != NULL; gIter = gIter->next) {
pe_action_t *action = (pe_action_t *) gIter->data;
if (pcmk_is_set(action->flags, pe_action_tracking)) {
pe__clear_action_flags(action, pe_action_tracking);
}
}
}
static void
order_first_rsc_probes(pe_resource_t * rsc, pe_working_set_t * data_set)
{
GListPtr gIter = NULL;
GListPtr probes = NULL;
for (gIter = rsc->children; gIter != NULL; gIter = gIter->next) {
pe_resource_t * child = (pe_resource_t *) gIter->data;
order_first_rsc_probes(child, data_set);
}
if (rsc->variant != pe_native) {
return;
}
probes = pe__resource_actions(rsc, NULL, RSC_STATUS, FALSE);
for (gIter = probes; gIter != NULL; gIter= gIter->next) {
pe_action_t *probe = (pe_action_t *) gIter->data;
GListPtr aIter = NULL;
for (aIter = probe->actions_after; aIter != NULL; aIter = aIter->next) {
pe_action_wrapper_t *after_wrapper = (pe_action_wrapper_t *) aIter->data;
order_first_probe_then_restart_repromote(probe, after_wrapper->action, data_set);
clear_actions_tracking_flag(data_set);
}
}
g_list_free(probes);
}
static void
order_first_probes(pe_working_set_t * data_set)
{
GListPtr gIter = NULL;
for (gIter = data_set->resources; gIter != NULL; gIter = gIter->next) {
pe_resource_t *rsc = (pe_resource_t *) gIter->data;
order_first_rsc_probes(rsc, data_set);
}
order_first_probes_imply_stops(data_set);
}
static void
order_then_probes(pe_working_set_t * data_set)
{
#if 0
GListPtr gIter = NULL;
for (gIter = data_set->resources; gIter != NULL; gIter = gIter->next) {
pe_resource_t *rsc = (pe_resource_t *) gIter->data;
/* Given "A then B", we would prefer to wait for A to be
* started before probing B.
*
* If A was a filesystem on which the binaries and data for B
* lived, it would have been useful if the author of B's agent
* could assume that A is running before B.monitor will be
* called.
*
* However we can't _only_ probe once A is running, otherwise
* we'd not detect the state of B if A could not be started
* for some reason.
*
* In practice however, we cannot even do an opportunistic
* version of this because B may be moving:
*
* B.probe -> B.start
* B.probe -> B.stop
* B.stop -> B.start
* A.stop -> A.start
* A.start -> B.probe
*
* So far so good, but if we add the result of this code:
*
* B.stop -> A.stop
*
* Then we get a loop:
*
* B.probe -> B.stop -> A.stop -> A.start -> B.probe
*
* We could kill the 'B.probe -> B.stop' dependency, but that
* could mean stopping B "too" soon, because B.start must wait
* for the probes to complete.
*
* Another option is to allow it only if A is a non-unique
* clone with clone-max == node-max (since we'll never be
* moving it). However, we could still be stopping one
* instance at the same time as starting another.
* The complexity of checking for allowed conditions combined
* with the ever narrowing usecase suggests that this code
* should remain disabled until someone gets smarter.
*/
pe_action_t *start = NULL;
GListPtr actions = NULL;
GListPtr probes = NULL;
actions = pe__resource_actions(rsc, NULL, RSC_START, FALSE);
if (actions) {
start = actions->data;
g_list_free(actions);
}
if(start == NULL) {
crm_err("No start action for %s", rsc->id);
continue;
}
probes = pe__resource_actions(rsc, NULL, RSC_STATUS, FALSE);
for (actions = start->actions_before; actions != NULL; actions = actions->next) {
pe_action_wrapper_t *before = (pe_action_wrapper_t *) actions->data;
GListPtr pIter = NULL;
pe_action_t *first = before->action;
pe_resource_t *first_rsc = first->rsc;
if(first->required_runnable_before) {
GListPtr clone_actions = NULL;
for (clone_actions = first->actions_before; clone_actions != NULL; clone_actions = clone_actions->next) {
before = (pe_action_wrapper_t *) clone_actions->data;
crm_trace("Testing %s -> %s (%p) for %s", first->uuid, before->action->uuid, before->action->rsc, start->uuid);
CRM_ASSERT(before->action->rsc);
first_rsc = before->action->rsc;
break;
}
} else if(!pcmk__str_eq(first->task, RSC_START, pcmk__str_casei)) {
crm_trace("Not a start op %s for %s", first->uuid, start->uuid);
}
if(first_rsc == NULL) {
continue;
} else if(uber_parent(first_rsc) == uber_parent(start->rsc)) {
crm_trace("Same parent %s for %s", first_rsc->id, start->uuid);
continue;
} else if(FALSE && pe_rsc_is_clone(uber_parent(first_rsc)) == FALSE) {
crm_trace("Not a clone %s for %s", first_rsc->id, start->uuid);
continue;
}
crm_err("Applying %s before %s %d", first->uuid, start->uuid, uber_parent(first_rsc)->variant);
for (pIter = probes; pIter != NULL; pIter = pIter->next) {
pe_action_t *probe = (pe_action_t *) pIter->data;
crm_err("Ordering %s before %s", first->uuid, probe->uuid);
order_actions(first, probe, pe_order_optional);
}
}
}
#endif
}
static void
order_probes(pe_working_set_t * data_set)
{
order_first_probes(data_set);
order_then_probes(data_set);
}
gboolean
stage7(pe_working_set_t * data_set)
{
GList *gIter = NULL;
crm_trace("Applying ordering constraints");
/* Don't ask me why, but apparently they need to be processed in
* the order they were created in... go figure
*
* Also g_list_append() has horrendous performance characteristics
* So we need to use g_list_prepend() and then reverse the list here
*/
data_set->ordering_constraints = g_list_reverse(data_set->ordering_constraints);
for (gIter = data_set->ordering_constraints; gIter != NULL; gIter = gIter->next) {
pe__ordering_t *order = gIter->data;
pe_resource_t *rsc = order->lh_rsc;
crm_trace("Applying ordering constraint: %d", order->id);
if (rsc != NULL) {
crm_trace("rsc_action-to-*");
rsc_order_first(rsc, order, data_set);
continue;
}
rsc = order->rh_rsc;
if (rsc != NULL) {
crm_trace("action-to-rsc_action");
rsc_order_then(order->lh_action, rsc, order);
} else {
crm_trace("action-to-action");
order_actions(order->lh_action, order->rh_action, order->type);
}
}
for (gIter = data_set->actions; gIter != NULL; gIter = gIter->next) {
pe_action_t *action = (pe_action_t *) gIter->data;
update_colo_start_chain(action, data_set);
}
crm_trace("Ordering probes");
order_probes(data_set);
crm_trace("Updating %d actions", g_list_length(data_set->actions));
for (gIter = data_set->actions; gIter != NULL; gIter = gIter->next) {
pe_action_t *action = (pe_action_t *) gIter->data;
update_action(action, data_set);
}
// Check for invalid orderings
for (gIter = data_set->actions; gIter != NULL; gIter = gIter->next) {
pe_action_t *action = (pe_action_t *) gIter->data;
pe_action_wrapper_t *input = NULL;
for (GList *input_iter = action->actions_before;
input_iter != NULL; input_iter = input_iter->next) {
input = (pe_action_wrapper_t *) input_iter->data;
if (pcmk__ordering_is_invalid(action, input)) {
input->type = pe_order_none;
}
}
}
LogNodeActions(data_set, FALSE);
for (gIter = data_set->resources; gIter != NULL; gIter = gIter->next) {
pe_resource_t *rsc = (pe_resource_t *) gIter->data;
LogActions(rsc, data_set, FALSE);
}
return TRUE;
}
static int transition_id = -1;
/*!
* \internal
* \brief Log a message after calculating a transition
*
* \param[in] filename Where transition input is stored
*/
void
pcmk__log_transition_summary(const char *filename)
{
if (was_processing_error) {
crm_err("Calculated transition %d (with errors), saving inputs in %s",
transition_id, filename);
} else if (was_processing_warning) {
crm_warn("Calculated transition %d (with warnings), saving inputs in %s",
transition_id, filename);
} else {
crm_notice("Calculated transition %d, saving inputs in %s",
transition_id, filename);
}
- if (crm_config_error) {
+ if (pcmk__config_error) {
crm_notice("Configuration errors found during scheduler processing,"
" please run \"crm_verify -L\" to identify issues");
}
}
/*
* Create a dependency graph to send to the transitioner (via the controller)
*/
gboolean
stage8(pe_working_set_t * data_set)
{
GListPtr gIter = NULL;
const char *value = NULL;
transition_id++;
crm_trace("Creating transition graph %d.", transition_id);
data_set->graph = create_xml_node(NULL, XML_TAG_GRAPH);
value = pe_pref(data_set->config_hash, "cluster-delay");
crm_xml_add(data_set->graph, "cluster-delay", value);
value = pe_pref(data_set->config_hash, "stonith-timeout");
crm_xml_add(data_set->graph, "stonith-timeout", value);
crm_xml_add(data_set->graph, "failed-stop-offset", "INFINITY");
if (pcmk_is_set(data_set->flags, pe_flag_start_failure_fatal)) {
crm_xml_add(data_set->graph, "failed-start-offset", "INFINITY");
} else {
crm_xml_add(data_set->graph, "failed-start-offset", "1");
}
value = pe_pref(data_set->config_hash, "batch-limit");
crm_xml_add(data_set->graph, "batch-limit", value);
crm_xml_add_int(data_set->graph, "transition_id", transition_id);
value = pe_pref(data_set->config_hash, "migration-limit");
if (crm_parse_ll(value, NULL) > 0) {
crm_xml_add(data_set->graph, "migration-limit", value);
}
if (data_set->recheck_by > 0) {
char *recheck_epoch = NULL;
recheck_epoch = crm_strdup_printf("%llu",
(long long) data_set->recheck_by);
crm_xml_add(data_set->graph, "recheck-by", recheck_epoch);
free(recheck_epoch);
}
/* errors...
slist_iter(action, pe_action_t, action_list, lpc,
if(action->optional == FALSE && action->runnable == FALSE) {
print_action("Ignoring", action, TRUE);
}
);
*/
/* The following code will de-duplicate action inputs, so nothing past this
* should rely on the action input type flags retaining their original
* values.
*/
gIter = data_set->resources;
for (; gIter != NULL; gIter = gIter->next) {
pe_resource_t *rsc = (pe_resource_t *) gIter->data;
pe_rsc_trace(rsc, "processing actions for rsc=%s", rsc->id);
rsc->cmds->expand(rsc, data_set);
}
crm_log_xml_trace(data_set->graph, "created resource-driven action list");
/* pseudo action to distribute list of nodes with maintenance state update */
add_maintenance_update(data_set);
/* catch any non-resource specific actions */
crm_trace("processing non-resource actions");
gIter = data_set->actions;
for (; gIter != NULL; gIter = gIter->next) {
pe_action_t *action = (pe_action_t *) gIter->data;
if (action->rsc
&& action->node
&& action->node->details->shutdown
&& !pcmk_is_set(action->rsc->flags, pe_rsc_maintenance)
&& !pcmk_any_flags_set(action->flags,
pe_action_optional|pe_action_runnable)
&& pcmk__str_eq(action->task, RSC_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(data_set->flags, pe_flag_have_quorum)
|| data_set->no_quorum_policy == no_quorum_ignore) {
crm_crit("Cannot %s node '%s' because of %s:%s%s (%s)",
action->node->details->unclean ? "fence" : "shut down",
action->node->details->uname, action->rsc->id,
pcmk_is_set(action->rsc->flags, pe_rsc_managed)? " blocked" : " unmanaged",
pcmk_is_set(action->rsc->flags, pe_rsc_failed)? " failed" : "",
action->uuid);
}
}
graph_element_from_action(action, data_set);
}
crm_log_xml_trace(data_set->graph, "created generic action list");
crm_trace("Created transition graph %d.", transition_id);
return TRUE;
}
void
LogNodeActions(pe_working_set_t * data_set, gboolean terminal)
{
GListPtr gIter = NULL;
for (gIter = data_set->actions; gIter != NULL; gIter = gIter->next) {
char *node_name = NULL;
char *task = NULL;
pe_action_t *action = (pe_action_t *) gIter->data;
if (action->rsc != NULL) {
continue;
} else if (pcmk_is_set(action->flags, pe_action_optional)) {
continue;
}
if (pe__is_guest_node(action->node)) {
node_name = crm_strdup_printf("%s (resource: %s)", action->node->details->uname, action->node->details->remote_rsc->container->id);
} else if(action->node) {
node_name = crm_strdup_printf("%s", action->node->details->uname);
}
if (pcmk__str_eq(action->task, CRM_OP_SHUTDOWN, pcmk__str_casei)) {
task = strdup("Shutdown");
} else if (pcmk__str_eq(action->task, CRM_OP_FENCE, pcmk__str_casei)) {
const char *op = g_hash_table_lookup(action->meta, "stonith_action");
task = crm_strdup_printf("Fence (%s)", op);
}
if(task == NULL) {
/* Nothing to report */
} else if(terminal && action->reason) {
printf(" * %s %s '%s'\n", task, node_name, action->reason);
} else if(terminal) {
printf(" * %s %s\n", task, node_name);
} else if(action->reason) {
crm_notice(" * %s %s '%s'\n", task, node_name, action->reason);
} else {
crm_notice(" * %s %s\n", task, node_name);
}
free(node_name);
free(task);
}
}
diff --git a/tools/crm_verify.c b/tools/crm_verify.c
index e6ab9aa75e..1df27ec5a7 100644
--- a/tools/crm_verify.c
+++ b/tools/crm_verify.c
@@ -1,318 +1,318 @@
/*
* Copyright 2004-2020 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 <stdio.h>
#include <sys/types.h>
#include <unistd.h>
#include <stdlib.h>
#include <errno.h>
#include <fcntl.h>
#include <libgen.h>
#include <glib.h>
#include <crm/common/xml.h>
#include <crm/common/util.h>
#include <crm/msg_xml.h>
#include <crm/cib.h>
#include <crm/pengine/status.h>
#include <pacemaker-internal.h>
gboolean USE_LIVE_CIB = FALSE;
char *cib_save = NULL;
extern gboolean stage0(pe_working_set_t * data_set);
static pcmk__cli_option_t long_options[] = {
// long option, argument type, storage, short option, description, flags
{
"help", no_argument, NULL, '?',
"\tThis text", pcmk__option_default
},
{
"version", no_argument, NULL, '$',
"\tVersion information", pcmk__option_default
},
{
"verbose", no_argument, NULL, 'V',
"\tSpecify multiple times to increase debug output\n",
pcmk__option_default
},
{
"-spacer-", no_argument, NULL, '-',
"\nData sources:", pcmk__option_default
},
{
"live-check", no_argument, NULL, 'L',
"Check the configuration used by the running cluster\n",
pcmk__option_default
},
{
"xml-file", required_argument, NULL, 'x',
"Check the configuration in the named file", pcmk__option_default
},
{
"xml-text", required_argument, NULL, 'X',
"Check the configuration in the supplied string", pcmk__option_default
},
{
"xml-pipe", no_argument, NULL, 'p',
"Check the configuration piped in via stdin", pcmk__option_default
},
{
"-spacer-", no_argument, NULL, '-',
"\nAdditional Options:", pcmk__option_default
},
{
"save-xml", required_argument, 0, 'S',
"Save verified XML to named file (most useful with -L)",
pcmk__option_default
},
{
"-spacer-", no_argument, NULL, '-',
"\nExamples:", pcmk__option_paragraph
},
{
"-spacer-", no_argument, NULL, '-',
"Check the consistency of the configuration in the running cluster:",
pcmk__option_paragraph
},
{
"-spacer-", no_argument, NULL, '-',
" crm_verify --live-check", pcmk__option_example
},
{
"-spacer-", no_argument, NULL, '-',
"Check the consistency of the configuration in a given file and "
"produce verbose output:",
pcmk__option_paragraph
},
{
"-spacer-", no_argument, NULL, '-',
" crm_verify --xml-file file.xml --verbose", pcmk__option_example
},
{0, 0, 0, 0}
};
int
main(int argc, char **argv)
{
xmlNode *cib_object = NULL;
xmlNode *status = NULL;
int argerr = 0;
int flag;
int option_index = 0;
pe_working_set_t *data_set = NULL;
cib_t *cib_conn = NULL;
int rc = pcmk_ok;
bool verbose = FALSE;
gboolean xml_stdin = FALSE;
const char *xml_tag = NULL;
const char *xml_file = NULL;
const char *xml_string = NULL;
crm_log_cli_init("crm_verify");
pcmk__set_cli_options(NULL, "[options]", long_options,
"check a Pacemaker configuration for errors\n\n"
"Check the well-formedness of a complete Pacemaker "
"XML configuration,\nits conformance to the "
"configured schema, and the presence of common\n"
"misconfigurations. Problems reported as errors "
"must be fixed before the\ncluster will work "
"properly. It is left to the administrator to decide"
"\nwhether to fix problems reported as warnings.");
while (1) {
flag = pcmk__next_cli_option(argc, argv, &option_index, NULL);
if (flag == -1)
break;
switch (flag) {
case 'X':
crm_trace("Option %c => %s", flag, optarg);
xml_string = optarg;
break;
case 'x':
crm_trace("Option %c => %s", flag, optarg);
xml_file = optarg;
break;
case 'p':
xml_stdin = TRUE;
break;
case 'S':
cib_save = optarg;
break;
case 'V':
verbose = TRUE;
crm_bump_log_level(argc, argv);
break;
case 'L':
USE_LIVE_CIB = TRUE;
break;
case '$':
case '?':
pcmk__cli_help(flag, CRM_EX_OK);
break;
default:
fprintf(stderr, "Option -%c is not yet supported\n", flag);
++argerr;
break;
}
}
if (optind < argc) {
printf("non-option ARGV-elements: ");
while (optind < argc) {
printf("%s ", argv[optind++]);
}
printf("\n");
}
if (optind > argc) {
++argerr;
}
if (argerr) {
crm_err("%d errors in option parsing", argerr);
pcmk__cli_help(flag, CRM_EX_USAGE);
}
crm_info("=#=#=#=#= Getting XML =#=#=#=#=");
if (USE_LIVE_CIB) {
cib_conn = cib_new();
rc = cib_conn->cmds->signon(cib_conn, crm_system_name, cib_command);
}
if (USE_LIVE_CIB) {
if (rc == pcmk_ok) {
int options = cib_scope_local | cib_sync_call;
crm_info("Reading XML from: live cluster");
rc = cib_conn->cmds->query(cib_conn, NULL, &cib_object, options);
}
if (rc != pcmk_ok) {
fprintf(stderr, "Live CIB query failed: %s\n", pcmk_strerror(rc));
goto done;
}
if (cib_object == NULL) {
fprintf(stderr, "Live CIB query failed: empty result\n");
rc = -ENOMSG;
goto done;
}
} else if (xml_file != NULL) {
cib_object = filename2xml(xml_file);
if (cib_object == NULL) {
fprintf(stderr, "Couldn't parse input file: %s\n", xml_file);
rc = -ENODATA;
goto done;
}
} else if (xml_string != NULL) {
cib_object = string2xml(xml_string);
if (cib_object == NULL) {
fprintf(stderr, "Couldn't parse input string: %s\n", xml_string);
rc = -ENODATA;
goto done;
}
} else if (xml_stdin) {
cib_object = stdin2xml();
if (cib_object == NULL) {
fprintf(stderr, "Couldn't parse input from STDIN.\n");
rc = -ENODATA;
goto done;
}
} else {
fprintf(stderr, "No configuration source specified."
" Use --help for usage information.\n");
rc = -ENODATA;
goto done;
}
xml_tag = crm_element_name(cib_object);
if (!pcmk__str_eq(xml_tag, XML_TAG_CIB, pcmk__str_casei)) {
fprintf(stderr,
"This tool can only check complete configurations (i.e. those starting with <cib>).\n");
rc = -EBADMSG;
goto done;
}
if (cib_save != NULL) {
write_xml_file(cib_object, cib_save, FALSE);
}
status = get_object_root(XML_CIB_TAG_STATUS, cib_object);
if (status == NULL) {
create_xml_node(cib_object, XML_CIB_TAG_STATUS);
}
if (validate_xml(cib_object, NULL, FALSE) == FALSE) {
pcmk__config_err("CIB did not pass schema validation");
free_xml(cib_object);
cib_object = NULL;
} else if (cli_config_update(&cib_object, NULL, FALSE) == FALSE) {
- crm_config_error = TRUE;
+ pcmk__config_error = true;
free_xml(cib_object);
cib_object = NULL;
fprintf(stderr, "The cluster will NOT be able to use this configuration.\n");
fprintf(stderr, "Please manually update the configuration to conform to the %s syntax.\n",
xml_latest_schema());
}
data_set = pe_new_working_set();
if (data_set == NULL) {
rc = -errno;
crm_perror(LOG_CRIT, "Unable to allocate working set");
goto done;
}
pe__set_working_set_flags(data_set, pe_flag_no_counts|pe_flag_no_compat);
if (cib_object == NULL) {
} else if (status != NULL || USE_LIVE_CIB) {
/* live queries will always have a status section and can do a full simulation */
pcmk__schedule_actions(data_set, cib_object, NULL);
} else {
data_set->now = crm_time_new(NULL);
data_set->input = cib_object;
stage0(data_set);
}
pe_free_working_set(data_set);
- if (crm_config_error) {
+ if (pcmk__config_error) {
fprintf(stderr, "Errors found during check: config not valid\n");
if (verbose == FALSE) {
fprintf(stderr, " -V may provide more details\n");
}
rc = -pcmk_err_schema_validation;
- } else if (crm_config_warning) {
+ } else if (pcmk__config_warning) {
fprintf(stderr, "Warnings found during check: config may not be valid\n");
if (verbose == FALSE) {
fprintf(stderr, " Use -V -V for more details\n");
}
rc = -pcmk_err_schema_validation;
}
if (USE_LIVE_CIB && cib_conn) {
cib_conn->cmds->signoff(cib_conn);
cib_delete(cib_conn);
}
done:
crm_exit(crm_errno2exit(rc));
}
File Metadata
Details
Attached
Mime Type
text/x-diff
Expires
Wed, Jun 25, 4:10 AM (1 d, 10 h)
Storage Engine
blob
Storage Format
Raw Data
Storage Handle
1952101
Default Alt Text
(206 KB)
Attached To
Mode
rP Pacemaker
Attached
Detach File
Event Timeline
Log In to Comment