Page MenuHomeClusterLabs Projects
Files F4624401

No OneTemporary
Actions

View Options

This file is larger than 256 KB, so syntax highlighting was skipped.
diff --git a/daemons/controld/controld_execd.c b/daemons/controld/controld_execd.c
index dfcb38f99d..8edb57e992 100644
--- a/daemons/controld/controld_execd.c
+++ b/daemons/controld/controld_execd.c
@@ -1,2441 +1,2441 @@
/*
* Copyright 2004-2023 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 <regex.h>
#include <sys/param.h>
#include <sys/types.h>
#include <sys/wait.h>
#include <crm/crm.h>
#include <crm/lrmd.h> // lrmd_event_data_t, lrmd_rsc_info_t, etc.
#include <crm/services.h>
#include <crm/msg_xml.h>
#include <crm/common/xml.h>
#include <crm/pengine/rules.h>
#include <crm/lrmd_internal.h>
#include <pacemaker-internal.h>
#include <pacemaker-controld.h>
#define START_DELAY_THRESHOLD 5 * 60 * 1000
#define MAX_LRM_REG_FAILS 30
struct delete_event_s {
int rc;
const char *rsc;
lrm_state_t *lrm_state;
};
static gboolean is_rsc_active(lrm_state_t * lrm_state, const char *rsc_id);
static gboolean build_active_RAs(lrm_state_t * lrm_state, xmlNode * rsc_list);
static gboolean stop_recurring_actions(gpointer key, gpointer value, gpointer user_data);
static lrmd_event_data_t *construct_op(const lrm_state_t *lrm_state,
const xmlNode *rsc_op,
const char *rsc_id,
const char *operation);
static void do_lrm_rsc_op(lrm_state_t *lrm_state, lrmd_rsc_info_t *rsc,
xmlNode *msg, struct ra_metadata_s *md);
static gboolean lrm_state_verify_stopped(lrm_state_t * lrm_state, enum crmd_fsa_state cur_state,
int log_level);
static void
lrm_connection_destroy(void)
{
if (pcmk_is_set(controld_globals.fsa_input_register, R_LRM_CONNECTED)) {
crm_crit("Connection to executor failed");
register_fsa_input(C_FSA_INTERNAL, I_ERROR, NULL);
controld_clear_fsa_input_flags(R_LRM_CONNECTED);
} else {
crm_info("Disconnected from executor");
}
}
static char *
make_stop_id(const char *rsc, int call_id)
{
return crm_strdup_printf("%s:%d", rsc, call_id);
}
static void
copy_instance_keys(gpointer key, gpointer value, gpointer user_data)
{
if (strstr(key, CRM_META "_") == NULL) {
g_hash_table_replace(user_data, strdup((const char *)key), strdup((const char *)value));
}
}
static void
copy_meta_keys(gpointer key, gpointer value, gpointer user_data)
{
if (strstr(key, CRM_META "_") != NULL) {
g_hash_table_replace(user_data, strdup((const char *)key), strdup((const char *)value));
}
}
/*!
* \internal
* \brief Remove a recurring operation from a resource's history
*
* \param[in,out] history Resource history to modify
* \param[in] op Operation to remove
*
* \return TRUE if the operation was found and removed, FALSE otherwise
*/
static gboolean
history_remove_recurring_op(rsc_history_t *history, const lrmd_event_data_t *op)
{
GList *iter;
for (iter = history->recurring_op_list; iter != NULL; iter = iter->next) {
lrmd_event_data_t *existing = iter->data;
if ((op->interval_ms == existing->interval_ms)
&& pcmk__str_eq(op->rsc_id, existing->rsc_id, pcmk__str_none)
&& pcmk__str_eq(op->op_type, existing->op_type, pcmk__str_casei)) {
history->recurring_op_list = g_list_delete_link(history->recurring_op_list, iter);
lrmd_free_event(existing);
return TRUE;
}
}
return FALSE;
}
/*!
* \internal
* \brief Free all recurring operations in resource history
*
* \param[in,out] history Resource history to modify
*/
static void
history_free_recurring_ops(rsc_history_t *history)
{
GList *iter;
for (iter = history->recurring_op_list; iter != NULL; iter = iter->next) {
lrmd_free_event(iter->data);
}
g_list_free(history->recurring_op_list);
history->recurring_op_list = NULL;
}
/*!
* \internal
* \brief Free resource history
*
* \param[in,out] history Resource history to free
*/
void
history_free(gpointer data)
{
rsc_history_t *history = (rsc_history_t*)data;
if (history->stop_params) {
g_hash_table_destroy(history->stop_params);
}
/* Don't need to free history->rsc.id because it's set to history->id */
free(history->rsc.type);
free(history->rsc.standard);
free(history->rsc.provider);
lrmd_free_event(history->failed);
lrmd_free_event(history->last);
free(history->id);
history_free_recurring_ops(history);
free(history);
}
static void
update_history_cache(lrm_state_t * lrm_state, lrmd_rsc_info_t * rsc, lrmd_event_data_t * op)
{
int target_rc = 0;
rsc_history_t *entry = NULL;
if (op->rsc_deleted) {
crm_debug("Purged history for '%s' after %s", op->rsc_id, op->op_type);
controld_delete_resource_history(op->rsc_id, lrm_state->node_name,
NULL, crmd_cib_smart_opt());
return;
}
if (pcmk__str_eq(op->op_type, PCMK_ACTION_NOTIFY, pcmk__str_casei)) {
return;
}
crm_debug("Updating history for '%s' with %s op", op->rsc_id, op->op_type);
entry = g_hash_table_lookup(lrm_state->resource_history, op->rsc_id);
if (entry == NULL && rsc) {
entry = calloc(1, sizeof(rsc_history_t));
entry->id = strdup(op->rsc_id);
g_hash_table_insert(lrm_state->resource_history, entry->id, entry);
entry->rsc.id = entry->id;
entry->rsc.type = strdup(rsc->type);
entry->rsc.standard = strdup(rsc->standard);
pcmk__str_update(&entry->rsc.provider, rsc->provider);
} else if (entry == NULL) {
crm_info("Resource %s no longer exists, not updating cache", op->rsc_id);
return;
}
entry->last_callid = op->call_id;
target_rc = rsc_op_expected_rc(op);
if (op->op_status == PCMK_EXEC_CANCELLED) {
if (op->interval_ms > 0) {
crm_trace("Removing cancelled recurring op: " PCMK__OP_FMT,
op->rsc_id, op->op_type, op->interval_ms);
history_remove_recurring_op(entry, op);
return;
} else {
crm_trace("Skipping " PCMK__OP_FMT " rc=%d, status=%d",
op->rsc_id, op->op_type, op->interval_ms, op->rc,
op->op_status);
}
} else if (did_rsc_op_fail(op, target_rc)) {
/* Store failed monitors here, otherwise the block below will cause them
* to be forgotten when a stop happens.
*/
if (entry->failed) {
lrmd_free_event(entry->failed);
}
entry->failed = lrmd_copy_event(op);
} else if (op->interval_ms == 0) {
if (entry->last) {
lrmd_free_event(entry->last);
}
entry->last = lrmd_copy_event(op);
if (op->params && pcmk__strcase_any_of(op->op_type, PCMK_ACTION_START,
PCMK_ACTION_RELOAD,
- CRMD_ACTION_RELOAD_AGENT,
+ PCMK_ACTION_RELOAD_AGENT,
PCMK_ACTION_MONITOR, NULL)) {
if (entry->stop_params) {
g_hash_table_destroy(entry->stop_params);
}
entry->stop_params = pcmk__strkey_table(free, free);
g_hash_table_foreach(op->params, copy_instance_keys, entry->stop_params);
}
}
if (op->interval_ms > 0) {
/* Ensure there are no duplicates */
history_remove_recurring_op(entry, op);
crm_trace("Adding recurring op: " PCMK__OP_FMT,
op->rsc_id, op->op_type, op->interval_ms);
entry->recurring_op_list = g_list_prepend(entry->recurring_op_list, lrmd_copy_event(op));
} else if ((entry->recurring_op_list != NULL)
&& !pcmk__str_eq(op->op_type, PCMK_ACTION_MONITOR,
pcmk__str_casei)) {
crm_trace("Dropping %d recurring ops because of: " PCMK__OP_FMT,
g_list_length(entry->recurring_op_list), op->rsc_id,
op->op_type, op->interval_ms);
history_free_recurring_ops(entry);
}
}
/*!
* \internal
* \brief Send a direct OK ack for a resource task
*
* \param[in] lrm_state LRM connection
* \param[in] input Input message being ack'ed
* \param[in] rsc_id ID of affected resource
* \param[in] rsc Affected resource (if available)
* \param[in] task Operation task being ack'ed
* \param[in] ack_host Name of host to send ack to
* \param[in] ack_sys IPC system name to ack
*/
static void
send_task_ok_ack(const lrm_state_t *lrm_state, const ha_msg_input_t *input,
const char *rsc_id, const lrmd_rsc_info_t *rsc,
const char *task, const char *ack_host, const char *ack_sys)
{
lrmd_event_data_t *op = construct_op(lrm_state, input->xml, rsc_id, task);
lrmd__set_result(op, PCMK_OCF_OK, PCMK_EXEC_DONE, NULL);
controld_ack_event_directly(ack_host, ack_sys, rsc, op, rsc_id);
lrmd_free_event(op);
}
static inline const char *
op_node_name(lrmd_event_data_t *op)
{
return pcmk__s(op->remote_nodename, controld_globals.our_nodename);
}
void
lrm_op_callback(lrmd_event_data_t * op)
{
CRM_CHECK(op != NULL, return);
switch (op->type) {
case lrmd_event_disconnect:
if (op->remote_nodename == NULL) {
/* If this is the local executor IPC connection, set the right
* bits in the controller when the connection goes down.
*/
lrm_connection_destroy();
}
break;
case lrmd_event_exec_complete:
{
lrm_state_t *lrm_state = lrm_state_find(op_node_name(op));
CRM_ASSERT(lrm_state != NULL);
process_lrm_event(lrm_state, op, NULL, NULL);
}
break;
default:
break;
}
}
static void
try_local_executor_connect(long long action, fsa_data_t *msg_data,
lrm_state_t *lrm_state)
{
int rc = pcmk_rc_ok;
crm_debug("Connecting to the local executor");
// If we can connect, great
rc = controld_connect_local_executor(lrm_state);
if (rc == pcmk_rc_ok) {
controld_set_fsa_input_flags(R_LRM_CONNECTED);
crm_info("Connection to the local executor established");
return;
}
// Otherwise, if we can try again, set a timer to do so
if (lrm_state->num_lrm_register_fails < MAX_LRM_REG_FAILS) {
crm_warn("Failed to connect to the local executor %d time%s "
"(%d max): %s", lrm_state->num_lrm_register_fails,
pcmk__plural_s(lrm_state->num_lrm_register_fails),
MAX_LRM_REG_FAILS, pcmk_rc_str(rc));
controld_start_wait_timer();
crmd_fsa_stall(FALSE);
return;
}
// Otherwise give up
crm_err("Failed to connect to the executor the max allowed "
"%d time%s: %s", lrm_state->num_lrm_register_fails,
pcmk__plural_s(lrm_state->num_lrm_register_fails),
pcmk_rc_str(rc));
register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL);
}
/* A_LRM_CONNECT */
void
do_lrm_control(long long action,
enum crmd_fsa_cause cause,
enum crmd_fsa_state cur_state,
enum crmd_fsa_input current_input, fsa_data_t * msg_data)
{
/* This only pertains to local executor connections. Remote connections are
* handled as resources within the scheduler. Connecting and disconnecting
* from remote executor instances is handled differently.
*/
lrm_state_t *lrm_state = NULL;
if (controld_globals.our_nodename == NULL) {
return; /* Nothing to do */
}
lrm_state = lrm_state_find_or_create(controld_globals.our_nodename);
if (lrm_state == NULL) {
register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL);
return;
}
if (action & A_LRM_DISCONNECT) {
if (lrm_state_verify_stopped(lrm_state, cur_state, LOG_INFO) == FALSE) {
if (action == A_LRM_DISCONNECT) {
crmd_fsa_stall(FALSE);
return;
}
}
controld_clear_fsa_input_flags(R_LRM_CONNECTED);
crm_info("Disconnecting from the executor");
lrm_state_disconnect(lrm_state);
lrm_state_reset_tables(lrm_state, FALSE);
crm_notice("Disconnected from the executor");
}
if (action & A_LRM_CONNECT) {
try_local_executor_connect(action, msg_data, lrm_state);
}
if (action & ~(A_LRM_CONNECT | A_LRM_DISCONNECT)) {
crm_err("Unexpected action %s in %s", fsa_action2string(action),
__func__);
}
}
static gboolean
lrm_state_verify_stopped(lrm_state_t * lrm_state, enum crmd_fsa_state cur_state, int log_level)
{
int counter = 0;
gboolean rc = TRUE;
const char *when = "lrm disconnect";
GHashTableIter gIter;
const char *key = NULL;
rsc_history_t *entry = NULL;
active_op_t *pending = NULL;
crm_debug("Checking for active resources before exit");
if (cur_state == S_TERMINATE) {
log_level = LOG_ERR;
when = "shutdown";
} else if (pcmk_is_set(controld_globals.fsa_input_register, R_SHUTDOWN)) {
when = "shutdown... waiting";
}
if ((lrm_state->active_ops != NULL) && lrm_state_is_connected(lrm_state)) {
guint removed = g_hash_table_foreach_remove(lrm_state->active_ops,
stop_recurring_actions,
lrm_state);
guint nremaining = g_hash_table_size(lrm_state->active_ops);
if (removed || nremaining) {
crm_notice("Stopped %u recurring operation%s at %s (%u remaining)",
removed, pcmk__plural_s(removed), when, nremaining);
}
}
if (lrm_state->active_ops != NULL) {
g_hash_table_iter_init(&gIter, lrm_state->active_ops);
while (g_hash_table_iter_next(&gIter, NULL, (void **)&pending)) {
/* Ignore recurring actions in the shutdown calculations */
if (pending->interval_ms == 0) {
counter++;
}
}
}
if (counter > 0) {
do_crm_log(log_level, "%d pending executor operation%s at %s",
counter, pcmk__plural_s(counter), when);
if ((cur_state == S_TERMINATE)
|| !pcmk_is_set(controld_globals.fsa_input_register,
R_SENT_RSC_STOP)) {
g_hash_table_iter_init(&gIter, lrm_state->active_ops);
while (g_hash_table_iter_next(&gIter, (gpointer*)&key, (gpointer*)&pending)) {
do_crm_log(log_level, "Pending action: %s (%s)", key, pending->op_key);
}
} else {
rc = FALSE;
}
return rc;
}
if (lrm_state->resource_history == NULL) {
return rc;
}
if (pcmk_is_set(controld_globals.fsa_input_register, R_SHUTDOWN)) {
/* At this point we're not waiting, we're just shutting down */
when = "shutdown";
}
counter = 0;
g_hash_table_iter_init(&gIter, lrm_state->resource_history);
while (g_hash_table_iter_next(&gIter, NULL, (gpointer*)&entry)) {
if (is_rsc_active(lrm_state, entry->id) == FALSE) {
continue;
}
counter++;
if (log_level == LOG_ERR) {
crm_info("Found %s active at %s", entry->id, when);
} else {
crm_trace("Found %s active at %s", entry->id, when);
}
if (lrm_state->active_ops != NULL) {
GHashTableIter hIter;
g_hash_table_iter_init(&hIter, lrm_state->active_ops);
while (g_hash_table_iter_next(&hIter, (gpointer*)&key, (gpointer*)&pending)) {
if (pcmk__str_eq(entry->id, pending->rsc_id, pcmk__str_none)) {
crm_notice("%sction %s (%s) incomplete at %s",
pending->interval_ms == 0 ? "A" : "Recurring a",
key, pending->op_key, when);
}
}
}
}
if (counter) {
crm_err("%d resource%s active at %s",
counter, (counter == 1)? " was" : "s were", when);
}
return rc;
}
static gboolean
is_rsc_active(lrm_state_t * lrm_state, const char *rsc_id)
{
rsc_history_t *entry = NULL;
entry = g_hash_table_lookup(lrm_state->resource_history, rsc_id);
if (entry == NULL || entry->last == NULL) {
return FALSE;
}
crm_trace("Processing %s: %s.%d=%d", rsc_id, entry->last->op_type,
entry->last->interval_ms, entry->last->rc);
if ((entry->last->rc == PCMK_OCF_OK)
&& pcmk__str_eq(entry->last->op_type, PCMK_ACTION_STOP,
pcmk__str_casei)) {
return FALSE;
} else if (entry->last->rc == PCMK_OCF_OK
&& pcmk__str_eq(entry->last->op_type, CRMD_ACTION_MIGRATE, pcmk__str_casei)) {
// A stricter check is too complex ... leave that to the scheduler
return FALSE;
} else if (entry->last->rc == PCMK_OCF_NOT_RUNNING) {
return FALSE;
} else if ((entry->last->interval_ms == 0)
&& (entry->last->rc == PCMK_OCF_NOT_CONFIGURED)) {
/* Badly configured resources can't be reliably stopped */
return FALSE;
}
return TRUE;
}
static gboolean
build_active_RAs(lrm_state_t * lrm_state, xmlNode * rsc_list)
{
GHashTableIter iter;
rsc_history_t *entry = NULL;
g_hash_table_iter_init(&iter, lrm_state->resource_history);
while (g_hash_table_iter_next(&iter, NULL, (void **)&entry)) {
GList *gIter = NULL;
xmlNode *xml_rsc = create_xml_node(rsc_list, XML_LRM_TAG_RESOURCE);
crm_xml_add(xml_rsc, XML_ATTR_ID, entry->id);
crm_xml_add(xml_rsc, XML_ATTR_TYPE, entry->rsc.type);
crm_xml_add(xml_rsc, XML_AGENT_ATTR_CLASS, entry->rsc.standard);
crm_xml_add(xml_rsc, XML_AGENT_ATTR_PROVIDER, entry->rsc.provider);
if (entry->last && entry->last->params) {
const char *container = g_hash_table_lookup(entry->last->params, CRM_META"_"XML_RSC_ATTR_CONTAINER);
if (container) {
crm_trace("Resource %s is a part of container resource %s", entry->id, container);
crm_xml_add(xml_rsc, XML_RSC_ATTR_CONTAINER, container);
}
}
controld_add_resource_history_xml(xml_rsc, &(entry->rsc), entry->failed,
lrm_state->node_name);
controld_add_resource_history_xml(xml_rsc, &(entry->rsc), entry->last,
lrm_state->node_name);
for (gIter = entry->recurring_op_list; gIter != NULL; gIter = gIter->next) {
controld_add_resource_history_xml(xml_rsc, &(entry->rsc), gIter->data,
lrm_state->node_name);
}
}
return FALSE;
}
xmlNode *
controld_query_executor_state(void)
{
xmlNode *xml_state = NULL;
xmlNode *xml_data = NULL;
xmlNode *rsc_list = NULL;
crm_node_t *peer = NULL;
lrm_state_t *lrm_state = lrm_state_find(controld_globals.our_nodename);
if (!lrm_state) {
crm_err("Could not find executor state for node %s",
controld_globals.our_nodename);
return NULL;
}
peer = crm_get_peer_full(0, lrm_state->node_name, CRM_GET_PEER_ANY);
CRM_CHECK(peer != NULL, return NULL);
xml_state = create_node_state_update(peer,
node_update_cluster|node_update_peer,
NULL, __func__);
if (xml_state == NULL) {
return NULL;
}
xml_data = create_xml_node(xml_state, XML_CIB_TAG_LRM);
crm_xml_add(xml_data, XML_ATTR_ID, peer->uuid);
rsc_list = create_xml_node(xml_data, XML_LRM_TAG_RESOURCES);
/* Build a list of active (not always running) resources */
build_active_RAs(lrm_state, rsc_list);
crm_log_xml_trace(xml_state, "Current executor state");
return xml_state;
}
/*!
* \internal
* \brief Map standard Pacemaker return code to operation status and OCF code
*
* \param[out] event Executor event whose status and return code should be set
* \param[in] rc Standard Pacemaker return code
*/
void
controld_rc2event(lrmd_event_data_t *event, int rc)
{
/* This is called for cleanup requests from controller peers/clients, not
* for resource actions, so no exit reason is needed.
*/
switch (rc) {
case pcmk_rc_ok:
lrmd__set_result(event, PCMK_OCF_OK, PCMK_EXEC_DONE, NULL);
break;
case EACCES:
lrmd__set_result(event, PCMK_OCF_INSUFFICIENT_PRIV,
PCMK_EXEC_ERROR, NULL);
break;
default:
lrmd__set_result(event, PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_ERROR,
NULL);
break;
}
}
/*!
* \internal
* \brief Trigger a new transition after CIB status was deleted
*
* If a CIB status delete was not expected (as part of the transition graph),
* trigger a new transition by updating the (arbitrary) "last-lrm-refresh"
* cluster property.
*
* \param[in] from_sys IPC name that requested the delete
* \param[in] rsc_id Resource whose status was deleted (for logging only)
*/
void
controld_trigger_delete_refresh(const char *from_sys, const char *rsc_id)
{
if (!pcmk__str_eq(from_sys, CRM_SYSTEM_TENGINE, pcmk__str_casei)) {
char *now_s = crm_strdup_printf("%lld", (long long) time(NULL));
crm_debug("Triggering a refresh after %s cleaned %s", from_sys, rsc_id);
cib__update_node_attr(controld_globals.logger_out,
controld_globals.cib_conn, cib_none,
XML_CIB_TAG_CRMCONFIG, NULL, NULL, NULL, NULL,
"last-lrm-refresh", now_s, NULL, NULL);
free(now_s);
}
}
static void
notify_deleted(lrm_state_t * lrm_state, ha_msg_input_t * input, const char *rsc_id, int rc)
{
lrmd_event_data_t *op = NULL;
const char *from_sys = crm_element_value(input->msg, F_CRM_SYS_FROM);
const char *from_host = crm_element_value(input->msg, F_CRM_HOST_FROM);
crm_info("Notifying %s on %s that %s was%s deleted",
from_sys, (from_host? from_host : "localhost"), rsc_id,
((rc == pcmk_ok)? "" : " not"));
op = construct_op(lrm_state, input->xml, rsc_id, CRMD_ACTION_DELETE);
controld_rc2event(op, pcmk_legacy2rc(rc));
controld_ack_event_directly(from_host, from_sys, NULL, op, rsc_id);
lrmd_free_event(op);
controld_trigger_delete_refresh(from_sys, rsc_id);
}
static gboolean
lrm_remove_deleted_rsc(gpointer key, gpointer value, gpointer user_data)
{
struct delete_event_s *event = user_data;
struct pending_deletion_op_s *op = value;
if (pcmk__str_eq(event->rsc, op->rsc, pcmk__str_none)) {
notify_deleted(event->lrm_state, op->input, event->rsc, event->rc);
return TRUE;
}
return FALSE;
}
static gboolean
lrm_remove_deleted_op(gpointer key, gpointer value, gpointer user_data)
{
const char *rsc = user_data;
active_op_t *pending = value;
if (pcmk__str_eq(rsc, pending->rsc_id, pcmk__str_none)) {
crm_info("Removing op %s:%d for deleted resource %s",
pending->op_key, pending->call_id, rsc);
return TRUE;
}
return FALSE;
}
static void
delete_rsc_entry(lrm_state_t *lrm_state, ha_msg_input_t *input,
const char *rsc_id, GHashTableIter *rsc_iter, int rc,
const char *user_name, bool from_cib)
{
struct delete_event_s event;
CRM_CHECK(rsc_id != NULL, return);
if (rc == pcmk_ok) {
char *rsc_id_copy = strdup(rsc_id);
if (rsc_iter) {
g_hash_table_iter_remove(rsc_iter);
} else {
g_hash_table_remove(lrm_state->resource_history, rsc_id_copy);
}
if (from_cib) {
controld_delete_resource_history(rsc_id_copy, lrm_state->node_name,
user_name, crmd_cib_smart_opt());
}
g_hash_table_foreach_remove(lrm_state->active_ops,
lrm_remove_deleted_op, rsc_id_copy);
free(rsc_id_copy);
}
if (input) {
notify_deleted(lrm_state, input, rsc_id, rc);
}
event.rc = rc;
event.rsc = rsc_id;
event.lrm_state = lrm_state;
g_hash_table_foreach_remove(lrm_state->deletion_ops, lrm_remove_deleted_rsc, &event);
}
static inline gboolean
last_failed_matches_op(rsc_history_t *entry, const char *op, guint interval_ms)
{
if (entry == NULL) {
return FALSE;
}
if (op == NULL) {
return TRUE;
}
return (pcmk__str_eq(op, entry->failed->op_type, pcmk__str_casei)
&& (interval_ms == entry->failed->interval_ms));
}
/*!
* \internal
* \brief Clear a resource's last failure
*
* Erase a resource's last failure on a particular node from both the
* LRM resource history in the CIB, and the resource history remembered
* for the LRM state.
*
* \param[in] rsc_id Resource name
* \param[in] node_name Node name
* \param[in] operation If specified, only clear if matching this operation
* \param[in] interval_ms If operation is specified, it has this interval
*/
void
lrm_clear_last_failure(const char *rsc_id, const char *node_name,
const char *operation, guint interval_ms)
{
lrm_state_t *lrm_state = lrm_state_find(node_name);
if (lrm_state == NULL) {
return;
}
if (lrm_state->resource_history != NULL) {
rsc_history_t *entry = g_hash_table_lookup(lrm_state->resource_history,
rsc_id);
if (last_failed_matches_op(entry, operation, interval_ms)) {
lrmd_free_event(entry->failed);
entry->failed = NULL;
}
}
}
/* Returns: gboolean - cancellation is in progress */
static gboolean
cancel_op(lrm_state_t * lrm_state, const char *rsc_id, const char *key, int op, gboolean remove)
{
int rc = pcmk_ok;
char *local_key = NULL;
active_op_t *pending = NULL;
CRM_CHECK(op != 0, return FALSE);
CRM_CHECK(rsc_id != NULL, return FALSE);
if (key == NULL) {
local_key = make_stop_id(rsc_id, op);
key = local_key;
}
pending = g_hash_table_lookup(lrm_state->active_ops, key);
if (pending) {
if (remove && !pcmk_is_set(pending->flags, active_op_remove)) {
controld_set_active_op_flags(pending, active_op_remove);
crm_debug("Scheduling %s for removal", key);
}
if (pcmk_is_set(pending->flags, active_op_cancelled)) {
crm_debug("Operation %s already cancelled", key);
free(local_key);
return FALSE;
}
controld_set_active_op_flags(pending, active_op_cancelled);
} else {
crm_info("No pending op found for %s", key);
free(local_key);
return FALSE;
}
crm_debug("Cancelling op %d for %s (%s)", op, rsc_id, key);
rc = lrm_state_cancel(lrm_state, pending->rsc_id, pending->op_type,
pending->interval_ms);
if (rc == pcmk_ok) {
crm_debug("Op %d for %s (%s): cancelled", op, rsc_id, key);
free(local_key);
return TRUE;
}
crm_debug("Op %d for %s (%s): Nothing to cancel", op, rsc_id, key);
/* The caller needs to make sure the entry is
* removed from the active operations list
*
* Usually by returning TRUE inside the worker function
* supplied to g_hash_table_foreach_remove()
*
* Not removing the entry from active operations will block
* the node from shutting down
*/
free(local_key);
return FALSE;
}
struct cancel_data {
gboolean done;
gboolean remove;
const char *key;
lrmd_rsc_info_t *rsc;
lrm_state_t *lrm_state;
};
static gboolean
cancel_action_by_key(gpointer key, gpointer value, gpointer user_data)
{
gboolean remove = FALSE;
struct cancel_data *data = user_data;
active_op_t *op = value;
if (pcmk__str_eq(op->op_key, data->key, pcmk__str_none)) {
data->done = TRUE;
remove = !cancel_op(data->lrm_state, data->rsc->id, key, op->call_id, data->remove);
}
return remove;
}
static gboolean
cancel_op_key(lrm_state_t * lrm_state, lrmd_rsc_info_t * rsc, const char *key, gboolean remove)
{
guint removed = 0;
struct cancel_data data;
CRM_CHECK(rsc != NULL, return FALSE);
CRM_CHECK(key != NULL, return FALSE);
data.key = key;
data.rsc = rsc;
data.done = FALSE;
data.remove = remove;
data.lrm_state = lrm_state;
removed = g_hash_table_foreach_remove(lrm_state->active_ops,
cancel_action_by_key, &data);
crm_trace("Removed %u op cache entries, new size: %u",
removed, g_hash_table_size(lrm_state->active_ops));
return data.done;
}
/*!
* \internal
* \brief Retrieve resource information from LRM
*
* \param[in,out] lrm_state Executor connection state to use
* \param[in] rsc_xml XML containing resource configuration
* \param[in] do_create If true, register resource if not already
* \param[out] rsc_info Where to store information obtained from executor
*
* \retval pcmk_ok Success (and rsc_info holds newly allocated result)
* \retval -EINVAL Required information is missing from arguments
* \retval -ENOTCONN No active connection to LRM
* \retval -ENODEV Resource not found
* \retval -errno Error communicating with executor when registering resource
*
* \note Caller is responsible for freeing result on success.
*/
static int
get_lrm_resource(lrm_state_t *lrm_state, const xmlNode *rsc_xml,
gboolean do_create, lrmd_rsc_info_t **rsc_info)
{
const char *id = ID(rsc_xml);
CRM_CHECK(lrm_state && rsc_xml && rsc_info, return -EINVAL);
CRM_CHECK(id, return -EINVAL);
if (lrm_state_is_connected(lrm_state) == FALSE) {
return -ENOTCONN;
}
crm_trace("Retrieving resource information for %s from the executor", id);
*rsc_info = lrm_state_get_rsc_info(lrm_state, id, 0);
// If resource isn't known by ID, try clone name, if provided
if (!*rsc_info) {
const char *long_id = crm_element_value(rsc_xml, XML_ATTR_ID_LONG);
if (long_id) {
*rsc_info = lrm_state_get_rsc_info(lrm_state, long_id, 0);
}
}
if ((*rsc_info == NULL) && do_create) {
const char *class = crm_element_value(rsc_xml, XML_AGENT_ATTR_CLASS);
const char *provider = crm_element_value(rsc_xml, XML_AGENT_ATTR_PROVIDER);
const char *type = crm_element_value(rsc_xml, XML_ATTR_TYPE);
int rc;
crm_trace("Registering resource %s with the executor", id);
rc = lrm_state_register_rsc(lrm_state, id, class, provider, type,
lrmd_opt_drop_recurring);
if (rc != pcmk_ok) {
fsa_data_t *msg_data = NULL;
crm_err("Could not register resource %s with the executor on %s: %s "
CRM_XS " rc=%d",
id, lrm_state->node_name, pcmk_strerror(rc), rc);
/* Register this as an internal error if this involves the local
* executor. Otherwise, we're likely dealing with an unresponsive
* remote node, which is not an FSA failure.
*/
if (lrm_state_is_local(lrm_state) == TRUE) {
register_fsa_error(C_FSA_INTERNAL, I_FAIL, NULL);
}
return rc;
}
*rsc_info = lrm_state_get_rsc_info(lrm_state, id, 0);
}
return *rsc_info? pcmk_ok : -ENODEV;
}
static void
delete_resource(lrm_state_t *lrm_state, const char *id, lrmd_rsc_info_t *rsc,
GHashTableIter *iter, const char *sys, const char *user,
ha_msg_input_t *request, bool unregister, bool from_cib)
{
int rc = pcmk_ok;
crm_info("Removing resource %s from executor for %s%s%s",
id, sys, (user? " as " : ""), (user? user : ""));
if (rsc && unregister) {
rc = lrm_state_unregister_rsc(lrm_state, id, 0);
}
if (rc == pcmk_ok) {
crm_trace("Resource %s deleted from executor", id);
} else if (rc == -EINPROGRESS) {
crm_info("Deletion of resource '%s' from executor is pending", id);
if (request) {
struct pending_deletion_op_s *op = NULL;
char *ref = crm_element_value_copy(request->msg, XML_ATTR_REFERENCE);
op = calloc(1, sizeof(struct pending_deletion_op_s));
op->rsc = strdup(rsc->id);
op->input = copy_ha_msg_input(request);
g_hash_table_insert(lrm_state->deletion_ops, ref, op);
}
return;
} else {
crm_warn("Could not delete '%s' from executor for %s%s%s: %s "
CRM_XS " rc=%d", id, sys, (user? " as " : ""),
(user? user : ""), pcmk_strerror(rc), rc);
}
delete_rsc_entry(lrm_state, request, id, iter, rc, user, from_cib);
}
static int
get_fake_call_id(lrm_state_t *lrm_state, const char *rsc_id)
{
int call_id = 999999999;
rsc_history_t *entry = NULL;
if(lrm_state) {
entry = g_hash_table_lookup(lrm_state->resource_history, rsc_id);
}
/* Make sure the call id is greater than the last successful operation,
* otherwise the failure will not result in a possible recovery of the resource
* as it could appear the failure occurred before the successful start */
if (entry) {
call_id = entry->last_callid + 1;
}
if (call_id < 0) {
call_id = 1;
}
return call_id;
}
static void
fake_op_status(lrm_state_t *lrm_state, lrmd_event_data_t *op, int op_status,
enum ocf_exitcode op_exitcode, const char *exit_reason)
{
op->call_id = get_fake_call_id(lrm_state, op->rsc_id);
op->t_run = time(NULL);
op->t_rcchange = op->t_run;
lrmd__set_result(op, op_exitcode, op_status, exit_reason);
}
static void
force_reprobe(lrm_state_t *lrm_state, const char *from_sys,
const char *from_host, const char *user_name,
gboolean is_remote_node, bool reprobe_all_nodes)
{
GHashTableIter gIter;
rsc_history_t *entry = NULL;
crm_info("Clearing resource history on node %s", lrm_state->node_name);
g_hash_table_iter_init(&gIter, lrm_state->resource_history);
while (g_hash_table_iter_next(&gIter, NULL, (void **)&entry)) {
/* only unregister the resource during a reprobe if it is not a remote connection
* resource. otherwise unregistering the connection will terminate remote-node
* membership */
bool unregister = true;
if (is_remote_lrmd_ra(NULL, NULL, entry->id)) {
unregister = false;
if (reprobe_all_nodes) {
lrm_state_t *remote_lrm_state = lrm_state_find(entry->id);
if (remote_lrm_state != NULL) {
/* If reprobing all nodes, be sure to reprobe the remote
* node before clearing its connection resource
*/
force_reprobe(remote_lrm_state, from_sys, from_host,
user_name, TRUE, reprobe_all_nodes);
}
}
}
/* Don't delete from the CIB, since we'll delete the whole node's LRM
* state from the CIB soon
*/
delete_resource(lrm_state, entry->id, &entry->rsc, &gIter, from_sys,
user_name, NULL, unregister, false);
}
/* Now delete the copy in the CIB */
controld_delete_node_state(lrm_state->node_name, controld_section_lrm,
cib_scope_local);
// @COMPAT DCs < 1.1.14 need this deleted (in case it was explicitly false)
update_attrd(lrm_state->node_name, CRM_OP_PROBED, NULL, user_name, is_remote_node);
}
/*!
* \internal
* \brief Fail a requested action without actually executing it
*
* For an action that can't be executed, process it similarly to an actual
* execution result, with specified error status (except for notify actions,
* which will always be treated as successful).
*
* \param[in,out] lrm_state Executor connection that action is for
* \param[in] action Action XML from request
* \param[in] rc Desired return code to use
* \param[in] op_status Desired operation status to use
* \param[in] exit_reason Human-friendly detail, if error
*/
static void
synthesize_lrmd_failure(lrm_state_t *lrm_state, const xmlNode *action,
int op_status, enum ocf_exitcode rc,
const char *exit_reason)
{
lrmd_event_data_t *op = NULL;
const char *operation = crm_element_value(action, XML_LRM_ATTR_TASK);
const char *target_node = crm_element_value(action, XML_LRM_ATTR_TARGET);
xmlNode *xml_rsc = find_xml_node(action, XML_CIB_TAG_RESOURCE, TRUE);
if ((xml_rsc == NULL) || (ID(xml_rsc) == NULL)) {
/* @TODO Should we do something else, like direct ack? */
crm_info("Can't fake %s failure (%d) on %s without resource configuration",
crm_element_value(action, XML_LRM_ATTR_TASK_KEY), rc,
target_node);
return;
} else if(operation == NULL) {
/* This probably came from crm_resource -C, nothing to do */
crm_info("Can't fake %s failure (%d) on %s without operation",
ID(xml_rsc), rc, target_node);
return;
}
op = construct_op(lrm_state, action, ID(xml_rsc), operation);
if (pcmk__str_eq(operation, PCMK_ACTION_NOTIFY, pcmk__str_casei)) {
// Notifications can't fail
fake_op_status(lrm_state, op, PCMK_EXEC_DONE, PCMK_OCF_OK, NULL);
} else {
fake_op_status(lrm_state, op, op_status, rc, exit_reason);
}
crm_info("Faking " PCMK__OP_FMT " result (%d) on %s",
op->rsc_id, op->op_type, op->interval_ms, op->rc, target_node);
// Process the result as if it came from the LRM
process_lrm_event(lrm_state, op, NULL, action);
lrmd_free_event(op);
}
/*!
* \internal
* \brief Get target of an LRM operation (replacing \p NULL with local node
* name)
*
* \param[in] xml LRM operation data XML
*
* \return LRM operation target node name (local node or Pacemaker Remote node)
*/
static const char *
lrm_op_target(const xmlNode *xml)
{
const char *target = NULL;
if (xml) {
target = crm_element_value(xml, XML_LRM_ATTR_TARGET);
}
if (target == NULL) {
target = controld_globals.our_nodename;
}
return target;
}
static void
fail_lrm_resource(xmlNode *xml, lrm_state_t *lrm_state, const char *user_name,
const char *from_host, const char *from_sys)
{
lrmd_event_data_t *op = NULL;
lrmd_rsc_info_t *rsc = NULL;
xmlNode *xml_rsc = find_xml_node(xml, XML_CIB_TAG_RESOURCE, TRUE);
CRM_CHECK(xml_rsc != NULL, return);
/* The executor simply executes operations and reports the results, without
* any concept of success or failure, so to fail a resource, we must fake
* what a failure looks like.
*
* To do this, we create a fake executor operation event for the resource,
* and pass that event to the executor client callback so it will be
* processed as if it came from the executor.
*/
op = construct_op(lrm_state, xml, ID(xml_rsc), "asyncmon");
free((char*) op->user_data);
op->user_data = NULL;
op->interval_ms = 0;
if (user_name && !pcmk__is_privileged(user_name)) {
crm_err("%s does not have permission to fail %s", user_name, ID(xml_rsc));
fake_op_status(lrm_state, op, PCMK_EXEC_ERROR,
PCMK_OCF_INSUFFICIENT_PRIV,
"Unprivileged user cannot fail resources");
controld_ack_event_directly(from_host, from_sys, NULL, op, ID(xml_rsc));
lrmd_free_event(op);
return;
}
if (get_lrm_resource(lrm_state, xml_rsc, TRUE, &rsc) == pcmk_ok) {
crm_info("Failing resource %s...", rsc->id);
fake_op_status(lrm_state, op, PCMK_EXEC_DONE, PCMK_OCF_UNKNOWN_ERROR,
"Simulated failure");
process_lrm_event(lrm_state, op, NULL, xml);
op->rc = PCMK_OCF_OK; // The request to fail the resource succeeded
lrmd_free_rsc_info(rsc);
} else {
crm_info("Cannot find/create resource in order to fail it...");
crm_log_xml_warn(xml, "bad input");
fake_op_status(lrm_state, op, PCMK_EXEC_ERROR, PCMK_OCF_UNKNOWN_ERROR,
"Cannot fail unknown resource");
}
controld_ack_event_directly(from_host, from_sys, NULL, op, ID(xml_rsc));
lrmd_free_event(op);
}
static void
handle_reprobe_op(lrm_state_t *lrm_state, const char *from_sys,
const char *from_host, const char *user_name,
gboolean is_remote_node, bool reprobe_all_nodes)
{
crm_notice("Forcing the status of all resources to be redetected");
force_reprobe(lrm_state, from_sys, from_host, user_name, is_remote_node,
reprobe_all_nodes);
if (!pcmk__strcase_any_of(from_sys, CRM_SYSTEM_PENGINE, CRM_SYSTEM_TENGINE, NULL)) {
xmlNode *reply = create_request(CRM_OP_INVOKE_LRM, NULL, from_host,
from_sys, CRM_SYSTEM_LRMD,
controld_globals.our_uuid);
crm_debug("ACK'ing re-probe from %s (%s)", from_sys, from_host);
if (relay_message(reply, TRUE) == FALSE) {
crm_log_xml_err(reply, "Unable to route reply");
}
free_xml(reply);
}
}
static bool do_lrm_cancel(ha_msg_input_t *input, lrm_state_t *lrm_state,
lrmd_rsc_info_t *rsc, const char *from_host, const char *from_sys)
{
char *op_key = NULL;
char *meta_key = NULL;
int call = 0;
const char *call_id = NULL;
const char *op_task = NULL;
guint interval_ms = 0;
gboolean in_progress = FALSE;
xmlNode *params = find_xml_node(input->xml, XML_TAG_ATTRS, TRUE);
CRM_CHECK(params != NULL, return FALSE);
meta_key = crm_meta_name(XML_LRM_ATTR_TASK);
op_task = crm_element_value(params, meta_key);
free(meta_key);
CRM_CHECK(op_task != NULL, return FALSE);
meta_key = crm_meta_name(XML_LRM_ATTR_INTERVAL_MS);
if (crm_element_value_ms(params, meta_key, &interval_ms) != pcmk_ok) {
free(meta_key);
return FALSE;
}
free(meta_key);
op_key = pcmk__op_key(rsc->id, op_task, interval_ms);
meta_key = crm_meta_name(XML_LRM_ATTR_CALLID);
call_id = crm_element_value(params, meta_key);
free(meta_key);
crm_debug("Scheduler requested op %s (call=%s) be cancelled",
op_key, (call_id? call_id : "NA"));
pcmk__scan_min_int(call_id, &call, 0);
if (call == 0) {
// Normal case when the scheduler cancels a recurring op
in_progress = cancel_op_key(lrm_state, rsc, op_key, TRUE);
} else {
// Normal case when the scheduler cancels an orphan op
in_progress = cancel_op(lrm_state, rsc->id, NULL, call, TRUE);
}
// Acknowledge cancellation operation if for a remote connection resource
if (!in_progress || is_remote_lrmd_ra(NULL, NULL, rsc->id)) {
char *op_id = make_stop_id(rsc->id, call);
if (is_remote_lrmd_ra(NULL, NULL, rsc->id) == FALSE) {
crm_info("Nothing known about operation %d for %s", call, op_key);
}
controld_delete_action_history_by_key(rsc->id, lrm_state->node_name,
op_key, call);
send_task_ok_ack(lrm_state, input, rsc->id, rsc, op_task,
from_host, from_sys);
/* needed at least for cancellation of a remote operation */
if (lrm_state->active_ops != NULL) {
g_hash_table_remove(lrm_state->active_ops, op_id);
}
free(op_id);
} else {
/* No ack is needed since abcdaa8, but peers with older versions
* in a rolling upgrade need one. We didn't bump the feature set
* at that commit, so we can only compare against the previous
* CRM version (3.0.8). If any peers have feature set 3.0.9 but
* not abcdaa8, they will time out waiting for the ack (no
* released versions of Pacemaker are affected).
*/
const char *peer_version = crm_element_value(params, XML_ATTR_CRM_VERSION);
if (compare_version(peer_version, "3.0.8") <= 0) {
crm_info("Sending compatibility ack for %s cancellation to %s (CRM version %s)",
op_key, from_host, peer_version);
send_task_ok_ack(lrm_state, input, rsc->id, rsc, op_task,
from_host, from_sys);
}
}
free(op_key);
return TRUE;
}
static void
do_lrm_delete(ha_msg_input_t *input, lrm_state_t *lrm_state,
lrmd_rsc_info_t *rsc, const char *from_sys, const char *from_host,
bool crm_rsc_delete, const char *user_name)
{
bool unregister = true;
int cib_rc = controld_delete_resource_history(rsc->id, lrm_state->node_name,
user_name,
cib_dryrun|cib_sync_call);
if (cib_rc != pcmk_rc_ok) {
lrmd_event_data_t *op = NULL;
op = construct_op(lrm_state, input->xml, rsc->id, CRMD_ACTION_DELETE);
/* These are resource clean-ups, not actions, so no exit reason is
* needed.
*/
lrmd__set_result(op, pcmk_rc2ocf(cib_rc), PCMK_EXEC_ERROR, NULL);
controld_ack_event_directly(from_host, from_sys, NULL, op, rsc->id);
lrmd_free_event(op);
return;
}
if (crm_rsc_delete && is_remote_lrmd_ra(NULL, NULL, rsc->id)) {
unregister = false;
}
delete_resource(lrm_state, rsc->id, rsc, NULL, from_sys,
user_name, input, unregister, true);
}
// User data for asynchronous metadata execution
struct metadata_cb_data {
lrmd_rsc_info_t *rsc; // Copy of resource information
xmlNode *input_xml; // Copy of FSA input XML
};
static struct metadata_cb_data *
new_metadata_cb_data(lrmd_rsc_info_t *rsc, xmlNode *input_xml)
{
struct metadata_cb_data *data = NULL;
data = calloc(1, sizeof(struct metadata_cb_data));
CRM_ASSERT(data != NULL);
data->input_xml = copy_xml(input_xml);
data->rsc = lrmd_copy_rsc_info(rsc);
return data;
}
static void
free_metadata_cb_data(struct metadata_cb_data *data)
{
lrmd_free_rsc_info(data->rsc);
free_xml(data->input_xml);
free(data);
}
/*!
* \internal
* \brief Execute an action after metadata has been retrieved
*
* \param[in] pid Ignored
* \param[in] result Result of metadata action
* \param[in] user_data Metadata callback data
*/
static void
metadata_complete(int pid, const pcmk__action_result_t *result, void *user_data)
{
struct metadata_cb_data *data = (struct metadata_cb_data *) user_data;
struct ra_metadata_s *md = NULL;
lrm_state_t *lrm_state = lrm_state_find(lrm_op_target(data->input_xml));
if ((lrm_state != NULL) && pcmk__result_ok(result)) {
md = controld_cache_metadata(lrm_state->metadata_cache, data->rsc,
result->action_stdout);
}
do_lrm_rsc_op(lrm_state, data->rsc, data->input_xml, md);
free_metadata_cb_data(data);
}
/* A_LRM_INVOKE */
void
do_lrm_invoke(long long action,
enum crmd_fsa_cause cause,
enum crmd_fsa_state cur_state,
enum crmd_fsa_input current_input, fsa_data_t * msg_data)
{
lrm_state_t *lrm_state = NULL;
const char *crm_op = NULL;
const char *from_sys = NULL;
const char *from_host = NULL;
const char *operation = NULL;
ha_msg_input_t *input = fsa_typed_data(fsa_dt_ha_msg);
const char *user_name = NULL;
const char *target_node = lrm_op_target(input->xml);
gboolean is_remote_node = FALSE;
bool crm_rsc_delete = FALSE;
// Message routed to the local node is targeting a specific, non-local node
is_remote_node = !pcmk__str_eq(target_node, controld_globals.our_nodename,
pcmk__str_casei);
lrm_state = lrm_state_find(target_node);
if ((lrm_state == NULL) && is_remote_node) {
crm_err("Failing action because local node has never had connection to remote node %s",
target_node);
synthesize_lrmd_failure(NULL, input->xml, PCMK_EXEC_NOT_CONNECTED,
PCMK_OCF_UNKNOWN_ERROR,
"Local node has no connection to remote");
return;
}
CRM_ASSERT(lrm_state != NULL);
user_name = pcmk__update_acl_user(input->msg, F_CRM_USER, NULL);
crm_op = crm_element_value(input->msg, F_CRM_TASK);
from_sys = crm_element_value(input->msg, F_CRM_SYS_FROM);
if (!pcmk__str_eq(from_sys, CRM_SYSTEM_TENGINE, pcmk__str_none)) {
from_host = crm_element_value(input->msg, F_CRM_HOST_FROM);
}
if (pcmk__str_eq(crm_op, CRM_OP_LRM_DELETE, pcmk__str_none)) {
if (!pcmk__str_eq(from_sys, CRM_SYSTEM_TENGINE, pcmk__str_none)) {
crm_rsc_delete = TRUE; // from crm_resource
}
operation = CRMD_ACTION_DELETE;
} else if (input->xml != NULL) {
operation = crm_element_value(input->xml, XML_LRM_ATTR_TASK);
}
CRM_CHECK(!pcmk__str_empty(crm_op) || !pcmk__str_empty(operation), return);
crm_trace("'%s' execution request from %s as %s user",
pcmk__s(crm_op, operation),
pcmk__s(from_sys, "unknown subsystem"),
pcmk__s(user_name, "current"));
if (pcmk__str_eq(crm_op, CRM_OP_LRM_FAIL, pcmk__str_none)) {
fail_lrm_resource(input->xml, lrm_state, user_name, from_host,
from_sys);
} else if (pcmk__str_eq(crm_op, CRM_OP_LRM_REFRESH, pcmk__str_none)) {
/* @COMPAT This can only be sent by crm_resource --refresh on a
* Pacemaker Remote node running Pacemaker 1.1.9, which is extremely
* unlikely. It previously would cause the controller to re-write its
* resource history to the CIB. Just ignore it.
*/
crm_notice("Ignoring refresh request from Pacemaker Remote 1.1.9 node");
// @COMPAT DCs <1.1.14 in a rolling upgrade might schedule this op
} else if (pcmk__str_eq(operation, CRM_OP_PROBED, pcmk__str_none)) {
update_attrd(lrm_state->node_name, CRM_OP_PROBED, XML_BOOLEAN_TRUE,
user_name, is_remote_node);
} else if (pcmk__str_eq(crm_op, CRM_OP_REPROBE, pcmk__str_none)
|| pcmk__str_eq(operation, CRM_OP_REPROBE, pcmk__str_none)) {
const char *raw_target = NULL;
if (input->xml != NULL) {
// For CRM_OP_REPROBE, a NULL target means we're targeting all nodes
raw_target = crm_element_value(input->xml, XML_LRM_ATTR_TARGET);
}
handle_reprobe_op(lrm_state, from_sys, from_host, user_name,
is_remote_node, (raw_target == NULL));
} else if (operation != NULL) {
lrmd_rsc_info_t *rsc = NULL;
xmlNode *xml_rsc = find_xml_node(input->xml, XML_CIB_TAG_RESOURCE, TRUE);
gboolean create_rsc = !pcmk__str_eq(operation, CRMD_ACTION_DELETE,
pcmk__str_none);
int rc;
// We can't return anything meaningful without a resource ID
CRM_CHECK(xml_rsc && ID(xml_rsc), return);
rc = get_lrm_resource(lrm_state, xml_rsc, create_rsc, &rsc);
if (rc == -ENOTCONN) {
synthesize_lrmd_failure(lrm_state, input->xml,
PCMK_EXEC_NOT_CONNECTED,
PCMK_OCF_UNKNOWN_ERROR,
"Not connected to remote executor");
return;
} else if ((rc < 0) && !create_rsc) {
/* Delete of malformed or nonexistent resource
* (deleting something that does not exist is a success)
*/
crm_notice("Not registering resource '%s' for a %s event "
CRM_XS " get-rc=%d (%s) transition-key=%s",
ID(xml_rsc), operation,
rc, pcmk_strerror(rc), ID(input->xml));
delete_rsc_entry(lrm_state, input, ID(xml_rsc), NULL, pcmk_ok,
user_name, true);
return;
} else if (rc == -EINVAL) {
// Resource operation on malformed resource
crm_err("Invalid resource definition for %s", ID(xml_rsc));
crm_log_xml_warn(input->msg, "invalid resource");
synthesize_lrmd_failure(lrm_state, input->xml, PCMK_EXEC_ERROR,
PCMK_OCF_NOT_CONFIGURED, // fatal error
"Invalid resource definition");
return;
} else if (rc < 0) {
// Error communicating with the executor
crm_err("Could not register resource '%s' with executor: %s "
CRM_XS " rc=%d",
ID(xml_rsc), pcmk_strerror(rc), rc);
crm_log_xml_warn(input->msg, "failed registration");
synthesize_lrmd_failure(lrm_state, input->xml, PCMK_EXEC_ERROR,
PCMK_OCF_INVALID_PARAM, // hard error
"Could not register resource with executor");
return;
}
if (pcmk__str_eq(operation, CRMD_ACTION_CANCEL, pcmk__str_none)) {
if (!do_lrm_cancel(input, lrm_state, rsc, from_host, from_sys)) {
crm_log_xml_warn(input->xml, "Bad command");
}
} else if (pcmk__str_eq(operation, CRMD_ACTION_DELETE, pcmk__str_none)) {
do_lrm_delete(input, lrm_state, rsc, from_sys, from_host,
crm_rsc_delete, user_name);
} else {
struct ra_metadata_s *md = NULL;
/* Getting metadata from cache is OK except for start actions --
* always refresh from the agent for those, in case the resource
* agent was updated.
*
* @TODO Only refresh metadata for starts if the agent actually
* changed (using something like inotify, or a hash or modification
* time of the agent executable).
*/
if (strcmp(operation, PCMK_ACTION_START) != 0) {
md = controld_get_rsc_metadata(lrm_state, rsc,
controld_metadata_from_cache);
}
if ((md == NULL) && crm_op_needs_metadata(rsc->standard,
operation)) {
/* Most likely, we'll need the agent metadata to record the
* pending operation and the operation result. Get it now rather
* than wait until then, so the metadata action doesn't eat into
* the real action's timeout.
*
* @TODO Metadata is retrieved via direct execution of the
* agent, which has a couple of related issues: the executor
* should execute agents, not the controller; and metadata for
* Pacemaker Remote nodes should be collected on those nodes,
* not locally.
*/
struct metadata_cb_data *data = NULL;
data = new_metadata_cb_data(rsc, input->xml);
crm_info("Retrieving metadata for %s (%s%s%s:%s) asynchronously",
rsc->id, rsc->standard,
((rsc->provider == NULL)? "" : ":"),
((rsc->provider == NULL)? "" : rsc->provider),
rsc->type);
(void) lrmd__metadata_async(rsc, metadata_complete,
(void *) data);
} else {
do_lrm_rsc_op(lrm_state, rsc, input->xml, md);
}
}
lrmd_free_rsc_info(rsc);
} else {
crm_err("Invalid execution request: unknown command '%s' (bug?)",
crm_op);
register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL);
}
}
static lrmd_event_data_t *
construct_op(const lrm_state_t *lrm_state, const xmlNode *rsc_op,
const char *rsc_id, const char *operation)
{
lrmd_event_data_t *op = NULL;
const char *op_delay = NULL;
const char *op_timeout = NULL;
GHashTable *params = NULL;
xmlNode *primitive = NULL;
const char *class = NULL;
const char *transition = NULL;
CRM_ASSERT(rsc_id && operation);
op = lrmd_new_event(rsc_id, operation, 0);
op->type = lrmd_event_exec_complete;
op->timeout = 0;
op->start_delay = 0;
lrmd__set_result(op, PCMK_OCF_UNKNOWN, PCMK_EXEC_PENDING, NULL);
if (rsc_op == NULL) {
CRM_LOG_ASSERT(pcmk__str_eq(operation, PCMK_ACTION_STOP,
pcmk__str_casei));
op->user_data = NULL;
/* the stop_all_resources() case
* by definition there is no DC (or they'd be shutting
* us down).
* So we should put our version here.
*/
op->params = pcmk__strkey_table(free, free);
g_hash_table_insert(op->params, strdup(XML_ATTR_CRM_VERSION), strdup(CRM_FEATURE_SET));
crm_trace("Constructed %s op for %s", operation, rsc_id);
return op;
}
params = xml2list(rsc_op);
g_hash_table_remove(params, CRM_META "_op_target_rc");
op_delay = crm_meta_value(params, XML_OP_ATTR_START_DELAY);
pcmk__scan_min_int(op_delay, &op->start_delay, 0);
op_timeout = crm_meta_value(params, XML_ATTR_TIMEOUT);
pcmk__scan_min_int(op_timeout, &op->timeout, 0);
if (pcmk__guint_from_hash(params, CRM_META "_" XML_LRM_ATTR_INTERVAL_MS, 0,
&(op->interval_ms)) != pcmk_rc_ok) {
op->interval_ms = 0;
}
/* Use pcmk_monitor_timeout instead of meta timeout for stonith
recurring monitor, if set */
primitive = find_xml_node(rsc_op, XML_CIB_TAG_RESOURCE, FALSE);
class = crm_element_value(primitive, XML_AGENT_ATTR_CLASS);
if (pcmk_is_set(pcmk_get_ra_caps(class), pcmk_ra_cap_fence_params)
&& pcmk__str_eq(operation, PCMK_ACTION_MONITOR, pcmk__str_casei)
&& (op->interval_ms > 0)) {
op_timeout = g_hash_table_lookup(params, "pcmk_monitor_timeout");
if (op_timeout != NULL) {
op->timeout = crm_get_msec(op_timeout);
}
}
if (!pcmk__str_eq(operation, PCMK_ACTION_STOP, pcmk__str_casei)) {
op->params = params;
} else {
rsc_history_t *entry = NULL;
if (lrm_state) {
entry = g_hash_table_lookup(lrm_state->resource_history, rsc_id);
}
/* If we do not have stop parameters cached, use
* whatever we are given */
if (!entry || !entry->stop_params) {
op->params = params;
} else {
/* Copy the cached parameter list so that we stop the resource
* with the old attributes, not the new ones */
op->params = pcmk__strkey_table(free, free);
g_hash_table_foreach(params, copy_meta_keys, op->params);
g_hash_table_foreach(entry->stop_params, copy_instance_keys, op->params);
g_hash_table_destroy(params);
params = NULL;
}
}
/* sanity */
if (op->timeout <= 0) {
op->timeout = op->interval_ms;
}
if (op->start_delay < 0) {
op->start_delay = 0;
}
transition = crm_element_value(rsc_op, XML_ATTR_TRANSITION_KEY);
CRM_CHECK(transition != NULL, return op);
op->user_data = strdup(transition);
if (op->interval_ms != 0) {
if (pcmk__strcase_any_of(operation, PCMK_ACTION_START, PCMK_ACTION_STOP,
NULL)) {
crm_err("Start and Stop actions cannot have an interval: %u",
op->interval_ms);
op->interval_ms = 0;
}
}
crm_trace("Constructed %s op for %s: interval=%u",
operation, rsc_id, op->interval_ms);
return op;
}
/*!
* \internal
* \brief Send a (synthesized) event result
*
* Reply with a synthesized event result directly, as opposed to going through
* the executor.
*
* \param[in] to_host Host to send result to
* \param[in] to_sys IPC name to send result (NULL for transition engine)
* \param[in] rsc Type information about resource the result is for
* \param[in,out] op Event with result to send
* \param[in] rsc_id ID of resource the result is for
*/
void
controld_ack_event_directly(const char *to_host, const char *to_sys,
const lrmd_rsc_info_t *rsc, lrmd_event_data_t *op,
const char *rsc_id)
{
xmlNode *reply = NULL;
xmlNode *update, *iter;
crm_node_t *peer = NULL;
CRM_CHECK(op != NULL, return);
if (op->rsc_id == NULL) {
CRM_ASSERT(rsc_id != NULL);
op->rsc_id = strdup(rsc_id);
}
if (to_sys == NULL) {
to_sys = CRM_SYSTEM_TENGINE;
}
peer = crm_get_peer(0, controld_globals.our_nodename);
update = create_node_state_update(peer, node_update_none, NULL,
__func__);
iter = create_xml_node(update, XML_CIB_TAG_LRM);
crm_xml_add(iter, XML_ATTR_ID, controld_globals.our_uuid);
iter = create_xml_node(iter, XML_LRM_TAG_RESOURCES);
iter = create_xml_node(iter, XML_LRM_TAG_RESOURCE);
crm_xml_add(iter, XML_ATTR_ID, op->rsc_id);
controld_add_resource_history_xml(iter, rsc, op,
controld_globals.our_nodename);
reply = create_request(CRM_OP_INVOKE_LRM, update, to_host, to_sys, CRM_SYSTEM_LRMD, NULL);
crm_log_xml_trace(update, "[direct ACK]");
crm_debug("ACK'ing resource op " PCMK__OP_FMT " from %s: %s",
op->rsc_id, op->op_type, op->interval_ms, op->user_data,
crm_element_value(reply, XML_ATTR_REFERENCE));
if (relay_message(reply, TRUE) == FALSE) {
crm_log_xml_err(reply, "Unable to route reply");
}
free_xml(update);
free_xml(reply);
}
gboolean
verify_stopped(enum crmd_fsa_state cur_state, int log_level)
{
gboolean res = TRUE;
GList *lrm_state_list = lrm_state_get_list();
GList *state_entry;
for (state_entry = lrm_state_list; state_entry != NULL; state_entry = state_entry->next) {
lrm_state_t *lrm_state = state_entry->data;
if (!lrm_state_verify_stopped(lrm_state, cur_state, log_level)) {
/* keep iterating through all even when false is returned */
res = FALSE;
}
}
controld_set_fsa_input_flags(R_SENT_RSC_STOP);
g_list_free(lrm_state_list); lrm_state_list = NULL;
return res;
}
struct stop_recurring_action_s {
lrmd_rsc_info_t *rsc;
lrm_state_t *lrm_state;
};
static gboolean
stop_recurring_action_by_rsc(gpointer key, gpointer value, gpointer user_data)
{
gboolean remove = FALSE;
struct stop_recurring_action_s *event = user_data;
active_op_t *op = value;
if ((op->interval_ms != 0)
&& pcmk__str_eq(op->rsc_id, event->rsc->id, pcmk__str_none)) {
crm_debug("Cancelling op %d for %s (%s)", op->call_id, op->rsc_id, (char*)key);
remove = !cancel_op(event->lrm_state, event->rsc->id, key, op->call_id, FALSE);
}
return remove;
}
static gboolean
stop_recurring_actions(gpointer key, gpointer value, gpointer user_data)
{
gboolean remove = FALSE;
lrm_state_t *lrm_state = user_data;
active_op_t *op = value;
if (op->interval_ms != 0) {
crm_info("Cancelling op %d for %s (%s)", op->call_id, op->rsc_id,
(const char *) key);
remove = !cancel_op(lrm_state, op->rsc_id, key, op->call_id, FALSE);
}
return remove;
}
/*!
* \internal
* \brief Check whether recurring actions should be cancelled before an action
*
* \param[in] rsc_id Resource that action is for
* \param[in] action Action being performed
* \param[in] interval_ms Operation interval of \p action (in milliseconds)
*
* \return true if recurring actions should be cancelled, otherwise false
*/
static bool
should_cancel_recurring(const char *rsc_id, const char *action, guint interval_ms)
{
if (is_remote_lrmd_ra(NULL, NULL, rsc_id) && (interval_ms == 0)
&& (strcmp(action, CRMD_ACTION_MIGRATE) == 0)) {
/* Don't stop monitoring a migrating Pacemaker Remote connection
* resource until the entire migration has completed. We must detect if
* the connection is unexpectedly severed, even during a migration.
*/
return false;
}
// Cancel recurring actions before changing resource state
return (interval_ms == 0)
&& !pcmk__str_any_of(action, PCMK_ACTION_MONITOR,
PCMK_ACTION_NOTIFY, NULL);
}
/*!
* \internal
* \brief Check whether an action should not be performed at this time
*
* \param[in] operation Action to be performed
*
* \return Readable description of why action should not be performed,
* or NULL if it should be performed
*/
static const char *
should_nack_action(const char *action)
{
if (pcmk_is_set(controld_globals.fsa_input_register, R_SHUTDOWN)
&& pcmk__str_eq(action, PCMK_ACTION_START, pcmk__str_none)) {
register_fsa_input(C_SHUTDOWN, I_SHUTDOWN, NULL);
return "Not attempting start due to shutdown in progress";
}
switch (controld_globals.fsa_state) {
case S_NOT_DC:
case S_POLICY_ENGINE: // Recalculating
case S_TRANSITION_ENGINE:
break;
default:
if (!pcmk__str_eq(action, PCMK_ACTION_STOP, pcmk__str_none)) {
return "Controller cannot attempt actions at this time";
}
break;
}
return NULL;
}
static void
do_lrm_rsc_op(lrm_state_t *lrm_state, lrmd_rsc_info_t *rsc, xmlNode *msg,
struct ra_metadata_s *md)
{
int rc;
int call_id = 0;
char *op_id = NULL;
lrmd_event_data_t *op = NULL;
fsa_data_t *msg_data = NULL;
const char *transition = NULL;
const char *operation = NULL;
const char *nack_reason = NULL;
CRM_CHECK((rsc != NULL) && (msg != NULL), return);
operation = crm_element_value(msg, XML_LRM_ATTR_TASK);
CRM_CHECK(!pcmk__str_empty(operation), return);
transition = crm_element_value(msg, XML_ATTR_TRANSITION_KEY);
if (pcmk__str_empty(transition)) {
crm_log_xml_err(msg, "Missing transition number");
}
if (lrm_state == NULL) {
// This shouldn't be possible, but provide a failsafe just in case
crm_err("Cannot execute %s of %s: No executor connection "
CRM_XS " transition_key=%s",
operation, rsc->id, pcmk__s(transition, ""));
synthesize_lrmd_failure(NULL, msg, PCMK_EXEC_INVALID,
PCMK_OCF_UNKNOWN_ERROR,
"No executor connection");
return;
}
if (pcmk__str_any_of(operation, PCMK_ACTION_RELOAD,
- CRMD_ACTION_RELOAD_AGENT, NULL)) {
+ PCMK_ACTION_RELOAD_AGENT, NULL)) {
/* Pre-2.1.0 DCs will schedule reload actions only, and 2.1.0+ DCs
* will schedule reload-agent actions only. In either case, we need
* to map that to whatever the resource agent actually supports.
* Default to the OCF 1.1 name.
*/
if ((md != NULL)
&& pcmk_is_set(md->ra_flags, ra_supports_legacy_reload)) {
operation = PCMK_ACTION_RELOAD;
} else {
- operation = CRMD_ACTION_RELOAD_AGENT;
+ operation = PCMK_ACTION_RELOAD_AGENT;
}
}
op = construct_op(lrm_state, msg, rsc->id, operation);
CRM_CHECK(op != NULL, return);
if (should_cancel_recurring(rsc->id, operation, op->interval_ms)) {
guint removed = 0;
struct stop_recurring_action_s data;
data.rsc = rsc;
data.lrm_state = lrm_state;
removed = g_hash_table_foreach_remove(lrm_state->active_ops,
stop_recurring_action_by_rsc,
&data);
if (removed) {
crm_debug("Stopped %u recurring operation%s in preparation for "
PCMK__OP_FMT, removed, pcmk__plural_s(removed),
rsc->id, operation, op->interval_ms);
}
}
/* now do the op */
crm_notice("Requesting local execution of %s operation for %s on %s "
CRM_XS " transition_key=%s op_key=" PCMK__OP_FMT,
crm_action_str(op->op_type, op->interval_ms), rsc->id, lrm_state->node_name,
pcmk__s(transition, ""), rsc->id, operation, op->interval_ms);
nack_reason = should_nack_action(operation);
if (nack_reason != NULL) {
crm_notice("Discarding attempt to perform action %s on %s in state %s "
"(shutdown=%s)", operation, rsc->id,
fsa_state2string(controld_globals.fsa_state),
pcmk__btoa(pcmk_is_set(controld_globals.fsa_input_register,
R_SHUTDOWN)));
lrmd__set_result(op, PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_INVALID,
nack_reason);
controld_ack_event_directly(NULL, NULL, rsc, op, rsc->id);
lrmd_free_event(op);
free(op_id);
return;
}
controld_record_pending_op(lrm_state->node_name, rsc, op);
op_id = pcmk__op_key(rsc->id, op->op_type, op->interval_ms);
if (op->interval_ms > 0) {
/* cancel it so we can then restart it without conflict */
cancel_op_key(lrm_state, rsc, op_id, FALSE);
}
rc = controld_execute_resource_agent(lrm_state, rsc->id, op->op_type,
op->user_data, op->interval_ms,
op->timeout, op->start_delay,
op->params, &call_id);
if (rc == pcmk_rc_ok) {
/* record all operations so we can wait
* for them to complete during shutdown
*/
char *call_id_s = make_stop_id(rsc->id, call_id);
active_op_t *pending = NULL;
pending = calloc(1, sizeof(active_op_t));
crm_trace("Recording pending op: %d - %s %s", call_id, op_id, call_id_s);
pending->call_id = call_id;
pending->interval_ms = op->interval_ms;
pending->op_type = strdup(operation);
pending->op_key = strdup(op_id);
pending->rsc_id = strdup(rsc->id);
pending->start_time = time(NULL);
pcmk__str_update(&pending->user_data, op->user_data);
if (crm_element_value_epoch(msg, XML_CONFIG_ATTR_SHUTDOWN_LOCK,
&(pending->lock_time)) != pcmk_ok) {
pending->lock_time = 0;
}
g_hash_table_replace(lrm_state->active_ops, call_id_s, pending);
if ((op->interval_ms > 0)
&& (op->start_delay > START_DELAY_THRESHOLD)) {
int target_rc = PCMK_OCF_OK;
crm_info("Faking confirmation of %s: execution postponed for over 5 minutes", op_id);
decode_transition_key(op->user_data, NULL, NULL, NULL, &target_rc);
lrmd__set_result(op, target_rc, PCMK_EXEC_DONE, NULL);
controld_ack_event_directly(NULL, NULL, rsc, op, rsc->id);
}
pending->params = op->params;
op->params = NULL;
} else if (lrm_state_is_local(lrm_state)) {
crm_err("Could not initiate %s action for resource %s locally: %s "
CRM_XS " rc=%d", operation, rsc->id, pcmk_rc_str(rc), rc);
fake_op_status(lrm_state, op, PCMK_EXEC_NOT_CONNECTED,
PCMK_OCF_UNKNOWN_ERROR, pcmk_rc_str(rc));
process_lrm_event(lrm_state, op, NULL, NULL);
register_fsa_error(C_FSA_INTERNAL, I_FAIL, NULL);
} else {
crm_err("Could not initiate %s action for resource %s remotely on %s: "
"%s " CRM_XS " rc=%d",
operation, rsc->id, lrm_state->node_name, pcmk_rc_str(rc), rc);
fake_op_status(lrm_state, op, PCMK_EXEC_NOT_CONNECTED,
PCMK_OCF_UNKNOWN_ERROR, pcmk_rc_str(rc));
process_lrm_event(lrm_state, op, NULL, NULL);
}
free(op_id);
lrmd_free_event(op);
}
void
do_lrm_event(long long action,
enum crmd_fsa_cause cause,
enum crmd_fsa_state cur_state, enum crmd_fsa_input cur_input, fsa_data_t * msg_data)
{
CRM_CHECK(FALSE, return);
}
static char *
unescape_newlines(const char *string)
{
char *pch = NULL;
char *ret = NULL;
static const char *escaped_newline = "\\n";
if (!string) {
return NULL;
}
ret = strdup(string);
pch = strstr(ret, escaped_newline);
while (pch != NULL) {
/* Replace newline escape pattern with actual newline (and a space so we
* don't have to shuffle the rest of the buffer)
*/
pch[0] = '\n';
pch[1] = ' ';
pch = strstr(pch, escaped_newline);
}
return ret;
}
static bool
did_lrm_rsc_op_fail(lrm_state_t *lrm_state, const char * rsc_id,
const char * op_type, guint interval_ms)
{
rsc_history_t *entry = NULL;
CRM_CHECK(lrm_state != NULL, return FALSE);
CRM_CHECK(rsc_id != NULL, return FALSE);
CRM_CHECK(op_type != NULL, return FALSE);
entry = g_hash_table_lookup(lrm_state->resource_history, rsc_id);
if (entry == NULL || entry->failed == NULL) {
return FALSE;
}
if (pcmk__str_eq(entry->failed->rsc_id, rsc_id, pcmk__str_none)
&& pcmk__str_eq(entry->failed->op_type, op_type, pcmk__str_casei)
&& entry->failed->interval_ms == interval_ms) {
return TRUE;
}
return FALSE;
}
/*!
* \internal
* \brief Log the result of an executor action (actual or synthesized)
*
* \param[in] op Executor action to log result for
* \param[in] op_key Operation key for action
* \param[in] node_name Name of node action was performed on, if known
* \param[in] confirmed Whether to log that graph action was confirmed
*/
static void
log_executor_event(const lrmd_event_data_t *op, const char *op_key,
const char *node_name, gboolean confirmed)
{
int log_level = LOG_ERR;
GString *str = g_string_sized_new(100); // reasonable starting size
pcmk__g_strcat(str,
"Result of ", crm_action_str(op->op_type, op->interval_ms),
" operation for ", op->rsc_id, NULL);
if (node_name != NULL) {
pcmk__g_strcat(str, " on ", node_name, NULL);
}
switch (op->op_status) {
case PCMK_EXEC_DONE:
log_level = LOG_NOTICE;
pcmk__g_strcat(str, ": ", services_ocf_exitcode_str(op->rc), NULL);
break;
case PCMK_EXEC_TIMEOUT:
pcmk__g_strcat(str,
": ", pcmk_exec_status_str(op->op_status), " after ",
pcmk__readable_interval(op->timeout), NULL);
break;
case PCMK_EXEC_CANCELLED:
log_level = LOG_INFO;
/* order of __attribute__ and Fall through comment is IMPORTANT!
* do not change it without proper testing with both clang and gcc
* in multiple versions.
* the clang check allows to build with all versions of clang.
* the has_c_attribute check is to workaround a bug in clang version
* in rhel7. has_attribute would happily return "YES SIR WE GOT IT"
* and fail the build the next line.
*/
#ifdef __clang__
#ifdef __has_c_attribute
#if __has_attribute(fallthrough)
__attribute__((fallthrough));
#endif
#endif
#endif
// Fall through
default:
pcmk__g_strcat(str, ": ", pcmk_exec_status_str(op->op_status),
NULL);
}
if ((op->exit_reason != NULL)
&& ((op->op_status != PCMK_EXEC_DONE) || (op->rc != PCMK_OCF_OK))) {
pcmk__g_strcat(str, " (", op->exit_reason, ")", NULL);
}
g_string_append(str, " " CRM_XS);
g_string_append_printf(str, " graph action %sconfirmed; call=%d key=%s",
(confirmed? "" : "un"), op->call_id, op_key);
if (op->op_status == PCMK_EXEC_DONE) {
g_string_append_printf(str, " rc=%d", op->rc);
}
do_crm_log(log_level, "%s", str->str);
g_string_free(str, TRUE);
/* The services library has already logged the output at info or debug
* level, so just raise to notice if it looks like a failure.
*/
if ((op->output != NULL) && (op->rc != PCMK_OCF_OK)) {
char *prefix = crm_strdup_printf(PCMK__OP_FMT "@%s output",
op->rsc_id, op->op_type,
op->interval_ms, node_name);
crm_log_output(LOG_NOTICE, prefix, op->output);
free(prefix);
}
}
void
process_lrm_event(lrm_state_t *lrm_state, lrmd_event_data_t *op,
active_op_t *pending, const xmlNode *action_xml)
{
char *op_id = NULL;
char *op_key = NULL;
gboolean remove = FALSE;
gboolean removed = FALSE;
bool need_direct_ack = FALSE;
lrmd_rsc_info_t *rsc = NULL;
const char *node_name = NULL;
CRM_CHECK(op != NULL, return);
CRM_CHECK(op->rsc_id != NULL, return);
// Remap new status codes for older DCs
if (compare_version(controld_globals.dc_version, "3.2.0") < 0) {
switch (op->op_status) {
case PCMK_EXEC_NOT_CONNECTED:
lrmd__set_result(op, PCMK_OCF_CONNECTION_DIED,
PCMK_EXEC_ERROR, op->exit_reason);
break;
case PCMK_EXEC_INVALID:
lrmd__set_result(op, CRM_DIRECT_NACK_RC, PCMK_EXEC_ERROR,
op->exit_reason);
break;
default:
break;
}
}
op_id = make_stop_id(op->rsc_id, op->call_id);
op_key = pcmk__op_key(op->rsc_id, op->op_type, op->interval_ms);
// Get resource info if available (from executor state or action XML)
if (lrm_state) {
rsc = lrm_state_get_rsc_info(lrm_state, op->rsc_id, 0);
}
if ((rsc == NULL) && action_xml) {
xmlNode *xml = find_xml_node(action_xml, XML_CIB_TAG_RESOURCE, TRUE);
const char *standard = crm_element_value(xml, XML_AGENT_ATTR_CLASS);
const char *provider = crm_element_value(xml, XML_AGENT_ATTR_PROVIDER);
const char *type = crm_element_value(xml, XML_ATTR_TYPE);
if (standard && type) {
crm_info("%s agent information not cached, using %s%s%s:%s from action XML",
op->rsc_id, standard,
(provider? ":" : ""), (provider? provider : ""), type);
rsc = lrmd_new_rsc_info(op->rsc_id, standard, provider, type);
} else {
crm_err("Can't process %s result because %s agent information not cached or in XML",
op_key, op->rsc_id);
}
}
// Get node name if available (from executor state or action XML)
if (lrm_state) {
node_name = lrm_state->node_name;
} else if (action_xml) {
node_name = crm_element_value(action_xml, XML_LRM_ATTR_TARGET);
}
if(pending == NULL) {
remove = TRUE;
if (lrm_state) {
pending = g_hash_table_lookup(lrm_state->active_ops, op_id);
}
}
if (op->op_status == PCMK_EXEC_ERROR) {
switch(op->rc) {
case PCMK_OCF_NOT_RUNNING:
case PCMK_OCF_RUNNING_PROMOTED:
case PCMK_OCF_DEGRADED:
case PCMK_OCF_DEGRADED_PROMOTED:
// Leave it to the TE/scheduler to decide if this is an error
op->op_status = PCMK_EXEC_DONE;
break;
default:
/* Nothing to do */
break;
}
}
if (op->op_status != PCMK_EXEC_CANCELLED) {
/* We might not record the result, so directly acknowledge it to the
* originator instead, so it doesn't time out waiting for the result
* (especially important if part of a transition).
*/
need_direct_ack = TRUE;
if (controld_action_is_recordable(op->op_type)) {
if (node_name && rsc) {
// We should record the result, and happily, we can
time_t lock_time = (pending == NULL)? 0 : pending->lock_time;
controld_update_resource_history(node_name, rsc, op, lock_time);
need_direct_ack = FALSE;
} else if (op->rsc_deleted) {
/* We shouldn't record the result (likely the resource was
* refreshed, cleaned, or removed while this operation was
* in flight).
*/
crm_notice("Not recording %s result in CIB because "
"resource information was removed since it was initiated",
op_key);
} else {
/* This shouldn't be possible; the executor didn't consider the
* resource deleted, but we couldn't find resource or node
* information.
*/
crm_err("Unable to record %s result in CIB: %s", op_key,
(node_name? "No resource information" : "No node name"));
}
}
} else if (op->interval_ms == 0) {
/* A non-recurring operation was cancelled. Most likely, the
* never-initiated action was removed from the executor's pending
* operations list upon resource removal.
*/
need_direct_ack = TRUE;
} else if (pending == NULL) {
/* This recurring operation was cancelled, but was not pending. No
* transition actions are waiting on it, nothing needs to be done.
*/
} else if (op->user_data == NULL) {
/* This recurring operation was cancelled and pending, but we don't
* have a transition key. This should never happen.
*/
crm_err("Recurring operation %s was cancelled without transition information",
op_key);
} else if (pcmk_is_set(pending->flags, active_op_remove)) {
/* This recurring operation was cancelled (by us) and pending, and we
* have been waiting for it to finish.
*/
if (lrm_state) {
controld_delete_action_history(op);
}
/* Directly acknowledge failed recurring actions here. The above call to
* controld_delete_action_history() will not erase any corresponding
* last_failure entry, which means that the DC won't confirm the
* cancellation via process_op_deletion(), and the transition would
* otherwise wait for the action timer to pop.
*/
if (did_lrm_rsc_op_fail(lrm_state, pending->rsc_id,
pending->op_type, pending->interval_ms)) {
need_direct_ack = TRUE;
}
} else if (op->rsc_deleted) {
/* This recurring operation was cancelled (but not by us, and the
* executor does not have resource information, likely due to resource
* cleanup, refresh, or removal) and pending.
*/
crm_debug("Recurring op %s was cancelled due to resource deletion",
op_key);
need_direct_ack = TRUE;
} else {
/* This recurring operation was cancelled (but not by us, likely by the
* executor before stopping the resource) and pending. We don't need to
* do anything special.
*/
}
if (need_direct_ack) {
controld_ack_event_directly(NULL, NULL, NULL, op, op->rsc_id);
}
if(remove == FALSE) {
/* The caller will do this afterwards, but keep the logging consistent */
removed = TRUE;
} else if (lrm_state && ((op->interval_ms == 0)
|| (op->op_status == PCMK_EXEC_CANCELLED))) {
gboolean found = g_hash_table_remove(lrm_state->active_ops, op_id);
if (op->interval_ms != 0) {
removed = TRUE;
} else if (found) {
removed = TRUE;
crm_trace("Op %s (call=%d, stop-id=%s, remaining=%u): Confirmed",
op_key, op->call_id, op_id,
g_hash_table_size(lrm_state->active_ops));
}
}
log_executor_event(op, op_key, node_name, removed);
if (lrm_state) {
if (!pcmk__str_eq(op->op_type, PCMK_ACTION_META_DATA,
pcmk__str_casei)) {
crmd_alert_resource_op(lrm_state->node_name, op);
} else if (rsc && (op->rc == PCMK_OCF_OK)) {
char *metadata = unescape_newlines(op->output);
controld_cache_metadata(lrm_state->metadata_cache, rsc, metadata);
free(metadata);
}
}
if (op->rsc_deleted) {
crm_info("Deletion of resource '%s' complete after %s", op->rsc_id, op_key);
if (lrm_state) {
delete_rsc_entry(lrm_state, NULL, op->rsc_id, NULL, pcmk_ok, NULL,
true);
}
}
/* If a shutdown was escalated while operations were pending,
* then the FSA will be stalled right now... allow it to continue
*/
controld_trigger_fsa();
if (lrm_state && rsc) {
update_history_cache(lrm_state, rsc, op);
}
lrmd_free_rsc_info(rsc);
free(op_key);
free(op_id);
}
diff --git a/daemons/controld/controld_metadata.c b/daemons/controld/controld_metadata.c
index d71b7bec05..c813cebeff 100644
--- a/daemons/controld/controld_metadata.c
+++ b/daemons/controld/controld_metadata.c
@@ -1,320 +1,320 @@
/*
* Copyright 2017-2023 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU General Public License version 2
* or later (GPLv2+) WITHOUT ANY WARRANTY.
*/
#include <crm_internal.h>
#include <stdio.h>
#include <glib.h>
#include <regex.h>
#include <crm/crm.h>
#include <crm/lrmd.h>
#include <pacemaker-controld.h>
static void
ra_param_free(void *param)
{
if (param) {
struct ra_param_s *p = (struct ra_param_s *) param;
if (p->rap_name) {
free(p->rap_name);
}
free(param);
}
}
static void
metadata_free(void *metadata)
{
if (metadata) {
struct ra_metadata_s *md = (struct ra_metadata_s *) metadata;
g_list_free_full(md->ra_params, ra_param_free);
free(metadata);
}
}
GHashTable *
metadata_cache_new(void)
{
return pcmk__strkey_table(free, metadata_free);
}
void
metadata_cache_free(GHashTable *mdc)
{
if (mdc) {
crm_trace("Destroying metadata cache with %d members", g_hash_table_size(mdc));
g_hash_table_destroy(mdc);
}
}
void
metadata_cache_reset(GHashTable *mdc)
{
if (mdc) {
crm_trace("Resetting metadata cache with %d members",
g_hash_table_size(mdc));
g_hash_table_remove_all(mdc);
}
}
static struct ra_param_s *
ra_param_from_xml(xmlNode *param_xml)
{
const char *param_name = crm_element_value(param_xml, "name");
struct ra_param_s *p;
p = calloc(1, sizeof(struct ra_param_s));
if (p == NULL) {
return NULL;
}
p->rap_name = strdup(param_name);
if (p->rap_name == NULL) {
free(p);
return NULL;
}
if (pcmk__xe_attr_is_true(param_xml, "reloadable")) {
controld_set_ra_param_flags(p, ra_param_reloadable);
}
if (pcmk__xe_attr_is_true(param_xml, "unique")) {
controld_set_ra_param_flags(p, ra_param_unique);
}
if (pcmk__xe_attr_is_true(param_xml, "private")) {
controld_set_ra_param_flags(p, ra_param_private);
}
return p;
}
static void
log_ra_ocf_version(const char *ra_key, const char *ra_ocf_version)
{
if (pcmk__str_empty(ra_ocf_version)) {
crm_warn("%s does not advertise OCF version supported", ra_key);
} else if (compare_version(ra_ocf_version, "2") >= 0) {
crm_warn("%s supports OCF version %s (this Pacemaker version supports "
PCMK_OCF_VERSION " and might not work properly with agent)",
ra_key, ra_ocf_version);
} else if (compare_version(ra_ocf_version, PCMK_OCF_VERSION) > 0) {
crm_info("%s supports OCF version %s (this Pacemaker version supports "
PCMK_OCF_VERSION " and might not use all agent features)",
ra_key, ra_ocf_version);
} else {
crm_debug("%s supports OCF version %s", ra_key, ra_ocf_version);
}
}
struct ra_metadata_s *
controld_cache_metadata(GHashTable *mdc, const lrmd_rsc_info_t *rsc,
const char *metadata_str)
{
char *key = NULL;
const char *reason = NULL;
xmlNode *metadata = NULL;
xmlNode *match = NULL;
struct ra_metadata_s *md = NULL;
bool any_private_params = false;
bool ocf1_1 = false;
CRM_CHECK(mdc && rsc && metadata_str, return NULL);
key = crm_generate_ra_key(rsc->standard, rsc->provider, rsc->type);
if (!key) {
reason = "Invalid resource agent standard or type";
goto err;
}
metadata = string2xml(metadata_str);
if (!metadata) {
reason = "Metadata is not valid XML";
goto err;
}
md = calloc(1, sizeof(struct ra_metadata_s));
if (md == NULL) {
reason = "Could not allocate memory";
goto err;
}
if (strcmp(rsc->standard, PCMK_RESOURCE_CLASS_OCF) == 0) {
xmlChar *content = NULL;
xmlNode *version_element = first_named_child(metadata, "version");
if (version_element != NULL) {
content = xmlNodeGetContent(version_element);
}
log_ra_ocf_version(key, (const char *) content);
if (content != NULL) {
ocf1_1 = (compare_version((const char *) content, "1.1") >= 0);
xmlFree(content);
}
}
// Check supported actions
match = first_named_child(metadata, "actions");
for (match = first_named_child(match, "action"); match != NULL;
match = crm_next_same_xml(match)) {
const char *action_name = crm_element_value(match, "name");
- if (pcmk__str_eq(action_name, CRMD_ACTION_RELOAD_AGENT,
+ if (pcmk__str_eq(action_name, PCMK_ACTION_RELOAD_AGENT,
pcmk__str_none)) {
if (ocf1_1) {
controld_set_ra_flags(md, key, ra_supports_reload_agent);
} else {
crm_notice("reload-agent action will not be used with %s "
"because it does not support OCF 1.1 or later", key);
}
} else if (!ocf1_1 && pcmk__str_eq(action_name, PCMK_ACTION_RELOAD,
pcmk__str_casei)) {
controld_set_ra_flags(md, key, ra_supports_legacy_reload);
}
}
// Build a parameter list
match = first_named_child(metadata, "parameters");
for (match = first_named_child(match, "parameter"); match != NULL;
match = crm_next_same_xml(match)) {
const char *param_name = crm_element_value(match, "name");
if (param_name == NULL) {
crm_warn("Metadata for %s:%s:%s has parameter without a name",
rsc->standard, rsc->provider, rsc->type);
} else {
struct ra_param_s *p = ra_param_from_xml(match);
if (p == NULL) {
reason = "Could not allocate memory";
goto err;
}
if (pcmk_is_set(p->rap_flags, ra_param_private)) {
any_private_params = true;
}
md->ra_params = g_list_prepend(md->ra_params, p);
}
}
/* Newer resource agents support the "private" parameter attribute to
* indicate sensitive parameters. For backward compatibility with older
* agents, implicitly treat a few common names as private when the agent
* doesn't specify any explicitly.
*/
if (!any_private_params) {
for (GList *iter = md->ra_params; iter != NULL; iter = iter->next) {
struct ra_param_s *p = iter->data;
if (pcmk__str_any_of(p->rap_name, "password", "passwd", "user",
NULL)) {
controld_set_ra_param_flags(p, ra_param_private);
}
}
}
g_hash_table_replace(mdc, key, md);
free_xml(metadata);
return md;
err:
crm_warn("Unable to update metadata for %s (%s%s%s:%s): %s",
rsc->id, rsc->standard, ((rsc->provider == NULL)? "" : ":"),
pcmk__s(rsc->provider, ""), rsc->type, reason);
free(key);
free_xml(metadata);
metadata_free(md);
return NULL;
}
/*!
* \internal
* \brief Get meta-data for a resource
*
* \param[in,out] lrm_state Use meta-data cache from this executor connection
* \param[in] rsc Resource to get meta-data for
* \param[in] source Allowed meta-data sources (bitmask of
* enum controld_metadata_source_e values)
*
* \return Meta-data cache entry for given resource, or NULL if not available
*/
struct ra_metadata_s *
controld_get_rsc_metadata(lrm_state_t *lrm_state, const lrmd_rsc_info_t *rsc,
uint32_t source)
{
struct ra_metadata_s *metadata = NULL;
char *metadata_str = NULL;
char *key = NULL;
int rc = pcmk_ok;
CRM_CHECK((lrm_state != NULL) && (rsc != NULL), return NULL);
if (pcmk_is_set(source, controld_metadata_from_cache)) {
key = crm_generate_ra_key(rsc->standard, rsc->provider, rsc->type);
if (key != NULL) {
metadata = g_hash_table_lookup(lrm_state->metadata_cache, key);
free(key);
}
if (metadata != NULL) {
crm_debug("Retrieved metadata for %s (%s%s%s:%s) from cache",
rsc->id, rsc->standard,
((rsc->provider == NULL)? "" : ":"),
((rsc->provider == NULL)? "" : rsc->provider),
rsc->type);
return metadata;
}
}
if (!pcmk_is_set(source, controld_metadata_from_agent)) {
return NULL;
}
/* For most actions, metadata was cached asynchronously before action
* execution (via metadata_complete()).
*
* However if that failed, and for other actions, retrieve the metadata now
* via a local, synchronous, direct execution of the agent.
*
* This has multiple issues, which is why this is just a fallback: the
* executor should execute agents, not the controller; metadata for
* Pacemaker Remote nodes should be collected on those nodes, not locally;
* the metadata call shouldn't eat into the timeout of the real action being
* performed; and the synchronous call blocks the controller (which also
* means that if the metadata action tries to contact the controller,
* everything will hang until the timeout).
*/
crm_debug("Retrieving metadata for %s (%s%s%s:%s) synchronously",
rsc->id, rsc->standard,
((rsc->provider == NULL)? "" : ":"),
((rsc->provider == NULL)? "" : rsc->provider),
rsc->type);
rc = lrm_state_get_metadata(lrm_state, rsc->standard, rsc->provider,
rsc->type, &metadata_str, 0);
if (rc != pcmk_ok) {
crm_warn("Failed to get metadata for %s (%s%s%s:%s): %s",
rsc->id, rsc->standard,
((rsc->provider == NULL)? "" : ":"),
((rsc->provider == NULL)? "" : rsc->provider),
rsc->type, pcmk_strerror(rc));
return NULL;
}
metadata = controld_cache_metadata(lrm_state->metadata_cache, rsc,
metadata_str);
free(metadata_str);
return metadata;
}
diff --git a/daemons/controld/controld_remote_ra.c b/daemons/controld/controld_remote_ra.c
index 91a9730b83..ccb66514e9 100644
--- a/daemons/controld/controld_remote_ra.c
+++ b/daemons/controld/controld_remote_ra.c
@@ -1,1470 +1,1470 @@
/*
* Copyright 2013-2023 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU General Public License version 2
* or later (GPLv2+) WITHOUT ANY WARRANTY.
*/
#include <crm_internal.h>
#include <crm/crm.h>
#include <crm/msg_xml.h>
#include <crm/common/xml_internal.h>
#include <crm/lrmd.h>
#include <crm/lrmd_internal.h>
#include <crm/services.h>
#include <pacemaker-controld.h>
#define REMOTE_LRMD_RA "remote"
/* The max start timeout before cmd retry */
#define MAX_START_TIMEOUT_MS 10000
#define cmd_set_flags(cmd, flags_to_set) do { \
(cmd)->status = pcmk__set_flags_as(__func__, __LINE__, LOG_TRACE, \
"Remote command", (cmd)->rsc_id, (cmd)->status, \
(flags_to_set), #flags_to_set); \
} while (0)
#define cmd_clear_flags(cmd, flags_to_clear) do { \
(cmd)->status = pcmk__clear_flags_as(__func__, __LINE__, LOG_TRACE, \
"Remote command", (cmd)->rsc_id, (cmd)->status, \
(flags_to_clear), #flags_to_clear); \
} while (0)
enum remote_cmd_status {
cmd_reported_success = (1 << 0),
cmd_cancel = (1 << 1),
};
typedef struct remote_ra_cmd_s {
/*! the local node the cmd is issued from */
char *owner;
/*! the remote node the cmd is executed on */
char *rsc_id;
/*! the action to execute */
char *action;
/*! some string the client wants us to give it back */
char *userdata;
/*! start delay in ms */
int start_delay;
/*! timer id used for start delay. */
int delay_id;
/*! timeout in ms for cmd */
int timeout;
int remaining_timeout;
/*! recurring interval in ms */
guint interval_ms;
/*! interval timer id */
int interval_id;
int monitor_timeout_id;
int takeover_timeout_id;
/*! action parameters */
lrmd_key_value_t *params;
pcmk__action_result_t result;
int call_id;
time_t start_time;
uint32_t status;
} remote_ra_cmd_t;
#define lrm_remote_set_flags(lrm_state, flags_to_set) do { \
lrm_state_t *lrm = (lrm_state); \
remote_ra_data_t *ra = lrm->remote_ra_data; \
ra->status = pcmk__set_flags_as(__func__, __LINE__, LOG_TRACE, "Remote", \
lrm->node_name, ra->status, \
(flags_to_set), #flags_to_set); \
} while (0)
#define lrm_remote_clear_flags(lrm_state, flags_to_clear) do { \
lrm_state_t *lrm = (lrm_state); \
remote_ra_data_t *ra = lrm->remote_ra_data; \
ra->status = pcmk__clear_flags_as(__func__, __LINE__, LOG_TRACE, "Remote", \
lrm->node_name, ra->status, \
(flags_to_clear), #flags_to_clear); \
} while (0)
enum remote_status {
expect_takeover = (1 << 0),
takeover_complete = (1 << 1),
remote_active = (1 << 2),
/* Maintenance mode is difficult to determine from the controller's context,
* so we have it signalled back with the transition from the scheduler.
*/
remote_in_maint = (1 << 3),
/* Similar for whether we are controlling a guest node or remote node.
* Fortunately there is a meta-attribute in the transition already and
* as the situation doesn't change over time we can use the
* resource start for noting down the information for later use when
* the attributes aren't at hand.
*/
controlling_guest = (1 << 4),
};
typedef struct remote_ra_data_s {
crm_trigger_t *work;
remote_ra_cmd_t *cur_cmd;
GList *cmds;
GList *recurring_cmds;
uint32_t status;
} remote_ra_data_t;
static int handle_remote_ra_start(lrm_state_t * lrm_state, remote_ra_cmd_t * cmd, int timeout_ms);
static void handle_remote_ra_stop(lrm_state_t * lrm_state, remote_ra_cmd_t * cmd);
static GList *fail_all_monitor_cmds(GList * list);
static void
free_cmd(gpointer user_data)
{
remote_ra_cmd_t *cmd = user_data;
if (!cmd) {
return;
}
if (cmd->delay_id) {
g_source_remove(cmd->delay_id);
}
if (cmd->interval_id) {
g_source_remove(cmd->interval_id);
}
if (cmd->monitor_timeout_id) {
g_source_remove(cmd->monitor_timeout_id);
}
if (cmd->takeover_timeout_id) {
g_source_remove(cmd->takeover_timeout_id);
}
free(cmd->owner);
free(cmd->rsc_id);
free(cmd->action);
free(cmd->userdata);
pcmk__reset_result(&(cmd->result));
lrmd_key_value_freeall(cmd->params);
free(cmd);
}
static int
generate_callid(void)
{
static int remote_ra_callid = 0;
remote_ra_callid++;
if (remote_ra_callid <= 0) {
remote_ra_callid = 1;
}
return remote_ra_callid;
}
static gboolean
recurring_helper(gpointer data)
{
remote_ra_cmd_t *cmd = data;
lrm_state_t *connection_rsc = NULL;
cmd->interval_id = 0;
connection_rsc = lrm_state_find(cmd->rsc_id);
if (connection_rsc && connection_rsc->remote_ra_data) {
remote_ra_data_t *ra_data = connection_rsc->remote_ra_data;
ra_data->recurring_cmds = g_list_remove(ra_data->recurring_cmds, cmd);
ra_data->cmds = g_list_append(ra_data->cmds, cmd);
mainloop_set_trigger(ra_data->work);
}
return FALSE;
}
static gboolean
start_delay_helper(gpointer data)
{
remote_ra_cmd_t *cmd = data;
lrm_state_t *connection_rsc = NULL;
cmd->delay_id = 0;
connection_rsc = lrm_state_find(cmd->rsc_id);
if (connection_rsc && connection_rsc->remote_ra_data) {
remote_ra_data_t *ra_data = connection_rsc->remote_ra_data;
mainloop_set_trigger(ra_data->work);
}
return FALSE;
}
static bool
should_purge_attributes(crm_node_t *node)
{
bool purge = true;
crm_node_t *conn_node = NULL;
lrm_state_t *connection_rsc = NULL;
if (!node->conn_host) {
return purge;
}
/* Get the node that was hosting the remote connection resource from the
* peer cache. That's the one we really care about here.
*/
conn_node = crm_get_peer(0, node->conn_host);
if (conn_node == NULL) {
return purge;
}
/* Check the uptime of connection_rsc. If it hasn't been running long
* enough, set purge=true. "Long enough" means it started running earlier
* than the timestamp when we noticed it went away in the first place.
*/
connection_rsc = lrm_state_find(node->uname);
if (connection_rsc != NULL) {
lrmd_t *lrm = connection_rsc->conn;
time_t uptime = lrmd__uptime(lrm);
time_t now = time(NULL);
/* Add 20s of fuzziness to give corosync a while to notice the remote
* host is gone. On various error conditions (failure to get uptime,
* peer_lost isn't set) we default to purging.
*/
if (uptime > 0 &&
conn_node->peer_lost > 0 &&
uptime + 20 >= now - conn_node->peer_lost) {
purge = false;
}
}
return purge;
}
static enum controld_section_e
section_to_delete(bool purge)
{
if (pcmk_is_set(controld_globals.flags, controld_shutdown_lock_enabled)) {
if (purge) {
return controld_section_all_unlocked;
} else {
return controld_section_lrm_unlocked;
}
} else {
if (purge) {
return controld_section_all;
} else {
return controld_section_lrm;
}
}
}
static void
purge_remote_node_attrs(int call_opt, crm_node_t *node)
{
bool purge = should_purge_attributes(node);
enum controld_section_e section = section_to_delete(purge);
/* Purge node from attrd's memory */
if (purge) {
update_attrd_remote_node_removed(node->uname, NULL);
}
controld_delete_node_state(node->uname, section, call_opt);
}
/*!
* \internal
* \brief Handle cluster communication related to pacemaker_remote node joining
*
* \param[in] node_name Name of newly integrated pacemaker_remote node
*/
static void
remote_node_up(const char *node_name)
{
int call_opt;
xmlNode *update, *state;
crm_node_t *node;
lrm_state_t *connection_rsc = NULL;
CRM_CHECK(node_name != NULL, return);
crm_info("Announcing Pacemaker Remote node %s", node_name);
call_opt = crmd_cib_smart_opt();
/* Delete node's probe_complete attribute. This serves two purposes:
*
* - @COMPAT DCs < 1.1.14 in a rolling upgrade might use it
* - deleting it (or any attribute for that matter) here ensures the
* attribute manager learns the node is remote
*/
update_attrd(node_name, CRM_OP_PROBED, NULL, NULL, TRUE);
/* Ensure node is in the remote peer cache with member status */
node = crm_remote_peer_get(node_name);
CRM_CHECK(node != NULL, return);
purge_remote_node_attrs(call_opt, node);
pcmk__update_peer_state(__func__, node, CRM_NODE_MEMBER, 0);
/* Apply any start state that we were given from the environment on the
* remote node.
*/
connection_rsc = lrm_state_find(node->uname);
if (connection_rsc != NULL) {
lrmd_t *lrm = connection_rsc->conn;
const char *start_state = lrmd__node_start_state(lrm);
if (start_state) {
set_join_state(start_state, node->uname, node->uuid, true);
}
}
/* pacemaker_remote nodes don't participate in the membership layer,
* so cluster nodes don't automatically get notified when they come and go.
* We send a cluster message to the DC, and update the CIB node state entry,
* so the DC will get it sooner (via message) or later (via CIB refresh),
* and any other interested parties can query the CIB.
*/
broadcast_remote_state_message(node_name, true);
update = create_xml_node(NULL, XML_CIB_TAG_STATUS);
state = create_node_state_update(node, node_update_cluster, update,
__func__);
/* Clear the XML_NODE_IS_FENCED flag in the node state. If the node ever
* needs to be fenced, this flag will allow various actions to determine
* whether the fencing has happened yet.
*/
crm_xml_add(state, XML_NODE_IS_FENCED, "0");
/* TODO: If the remote connection drops, and this (async) CIB update either
* failed or has not yet completed, later actions could mistakenly think the
* node has already been fenced (if the XML_NODE_IS_FENCED attribute was
* previously set, because it won't have been cleared). This could prevent
* actual fencing or allow recurring monitor failures to be cleared too
* soon. Ideally, we wouldn't rely on the CIB for the fenced status.
*/
controld_update_cib(XML_CIB_TAG_STATUS, update, call_opt, NULL, NULL);
free_xml(update);
}
enum down_opts {
DOWN_KEEP_LRM,
DOWN_ERASE_LRM
};
/*!
* \internal
* \brief Handle cluster communication related to pacemaker_remote node leaving
*
* \param[in] node_name Name of lost node
* \param[in] opts Whether to keep or erase LRM history
*/
static void
remote_node_down(const char *node_name, const enum down_opts opts)
{
xmlNode *update;
int call_opt = crmd_cib_smart_opt();
crm_node_t *node;
/* Purge node from attrd's memory */
update_attrd_remote_node_removed(node_name, NULL);
/* Normally, only node attributes should be erased, and the resource history
* should be kept until the node comes back up. However, after a successful
* fence, we want to clear the history as well, so we don't think resources
* are still running on the node.
*/
if (opts == DOWN_ERASE_LRM) {
controld_delete_node_state(node_name, controld_section_all, call_opt);
} else {
controld_delete_node_state(node_name, controld_section_attrs, call_opt);
}
/* Ensure node is in the remote peer cache with lost state */
node = crm_remote_peer_get(node_name);
CRM_CHECK(node != NULL, return);
pcmk__update_peer_state(__func__, node, CRM_NODE_LOST, 0);
/* Notify DC */
broadcast_remote_state_message(node_name, false);
/* Update CIB node state */
update = create_xml_node(NULL, XML_CIB_TAG_STATUS);
create_node_state_update(node, node_update_cluster, update, __func__);
controld_update_cib(XML_CIB_TAG_STATUS, update, call_opt, NULL, NULL);
free_xml(update);
}
/*!
* \internal
* \brief Handle effects of a remote RA command on node state
*
* \param[in] cmd Completed remote RA command
*/
static void
check_remote_node_state(const remote_ra_cmd_t *cmd)
{
/* Only successful actions can change node state */
if (!pcmk__result_ok(&(cmd->result))) {
return;
}
if (pcmk__str_eq(cmd->action, PCMK_ACTION_START, pcmk__str_casei)) {
remote_node_up(cmd->rsc_id);
} else if (pcmk__str_eq(cmd->action, "migrate_from", pcmk__str_casei)) {
/* After a successful migration, we don't need to do remote_node_up()
* because the DC already knows the node is up, and we don't want to
* clear LRM history etc. We do need to add the remote node to this
* host's remote peer cache, because (unless it happens to be DC)
* it hasn't been tracking the remote node, and other code relies on
* the cache to distinguish remote nodes from unseen cluster nodes.
*/
crm_node_t *node = crm_remote_peer_get(cmd->rsc_id);
CRM_CHECK(node != NULL, return);
pcmk__update_peer_state(__func__, node, CRM_NODE_MEMBER, 0);
} else if (pcmk__str_eq(cmd->action, PCMK_ACTION_STOP, pcmk__str_casei)) {
lrm_state_t *lrm_state = lrm_state_find(cmd->rsc_id);
remote_ra_data_t *ra_data = lrm_state? lrm_state->remote_ra_data : NULL;
if (ra_data) {
if (!pcmk_is_set(ra_data->status, takeover_complete)) {
/* Stop means down if we didn't successfully migrate elsewhere */
remote_node_down(cmd->rsc_id, DOWN_KEEP_LRM);
} else if (AM_I_DC == FALSE) {
/* Only the connection host and DC track node state,
* so if the connection migrated elsewhere and we aren't DC,
* un-cache the node, so we don't have stale info
*/
crm_remote_peer_cache_remove(cmd->rsc_id);
}
}
}
/* We don't do anything for successful monitors, which is correct for
* routine recurring monitors, and for monitors on nodes where the
* connection isn't supposed to be (the cluster will stop the connection in
* that case). However, if the initial probe finds the connection already
* active on the node where we want it, we probably should do
* remote_node_up(). Unfortunately, we can't distinguish that case here.
* Given that connections have to be initiated by the cluster, the chance of
* that should be close to zero.
*/
}
static void
report_remote_ra_result(remote_ra_cmd_t * cmd)
{
lrmd_event_data_t op = { 0, };
check_remote_node_state(cmd);
op.type = lrmd_event_exec_complete;
op.rsc_id = cmd->rsc_id;
op.op_type = cmd->action;
op.user_data = cmd->userdata;
op.timeout = cmd->timeout;
op.interval_ms = cmd->interval_ms;
op.t_run = (unsigned int) cmd->start_time;
op.t_rcchange = (unsigned int) cmd->start_time;
lrmd__set_result(&op, cmd->result.exit_status, cmd->result.execution_status,
cmd->result.exit_reason);
if (pcmk_is_set(cmd->status, cmd_reported_success) && !pcmk__result_ok(&(cmd->result))) {
op.t_rcchange = (unsigned int) time(NULL);
/* This edge case will likely never ever occur, but if it does the
* result is that a failure will not be processed correctly. This is only
* remotely possible because we are able to detect a connection resource's tcp
* connection has failed at any moment after start has completed. The actual
* recurring operation is just a connectivity ping.
*
* basically, we are not guaranteed that the first successful monitor op and
* a subsequent failed monitor op will not occur in the same timestamp. We have to
* make it look like the operations occurred at separate times though. */
if (op.t_rcchange == op.t_run) {
op.t_rcchange++;
}
}
if (cmd->params) {
lrmd_key_value_t *tmp;
op.params = pcmk__strkey_table(free, free);
for (tmp = cmd->params; tmp; tmp = tmp->next) {
g_hash_table_insert(op.params, strdup(tmp->key), strdup(tmp->value));
}
}
op.call_id = cmd->call_id;
op.remote_nodename = cmd->owner;
lrm_op_callback(&op);
if (op.params) {
g_hash_table_destroy(op.params);
}
lrmd__reset_result(&op);
}
static void
update_remaining_timeout(remote_ra_cmd_t * cmd)
{
cmd->remaining_timeout = ((cmd->timeout / 1000) - (time(NULL) - cmd->start_time)) * 1000;
}
static gboolean
retry_start_cmd_cb(gpointer data)
{
lrm_state_t *lrm_state = data;
remote_ra_data_t *ra_data = lrm_state->remote_ra_data;
remote_ra_cmd_t *cmd = NULL;
int rc = ETIME;
if (!ra_data || !ra_data->cur_cmd) {
return FALSE;
}
cmd = ra_data->cur_cmd;
if (!pcmk__strcase_any_of(cmd->action, PCMK_ACTION_START, "migrate_from",
NULL)) {
return FALSE;
}
update_remaining_timeout(cmd);
if (cmd->remaining_timeout > 0) {
rc = handle_remote_ra_start(lrm_state, cmd, cmd->remaining_timeout);
} else {
pcmk__set_result(&(cmd->result), PCMK_OCF_UNKNOWN_ERROR,
PCMK_EXEC_TIMEOUT,
"Not enough time remains to retry remote connection");
}
if (rc != pcmk_rc_ok) {
report_remote_ra_result(cmd);
if (ra_data->cmds) {
mainloop_set_trigger(ra_data->work);
}
ra_data->cur_cmd = NULL;
free_cmd(cmd);
} else {
/* wait for connection event */
}
return FALSE;
}
static gboolean
connection_takeover_timeout_cb(gpointer data)
{
lrm_state_t *lrm_state = NULL;
remote_ra_cmd_t *cmd = data;
crm_info("takeover event timed out for node %s", cmd->rsc_id);
cmd->takeover_timeout_id = 0;
lrm_state = lrm_state_find(cmd->rsc_id);
handle_remote_ra_stop(lrm_state, cmd);
free_cmd(cmd);
return FALSE;
}
static gboolean
monitor_timeout_cb(gpointer data)
{
lrm_state_t *lrm_state = NULL;
remote_ra_cmd_t *cmd = data;
lrm_state = lrm_state_find(cmd->rsc_id);
crm_info("Timed out waiting for remote poke response from %s%s",
cmd->rsc_id, (lrm_state? "" : " (no LRM state)"));
cmd->monitor_timeout_id = 0;
pcmk__set_result(&(cmd->result), PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_TIMEOUT,
"Remote executor did not respond");
if (lrm_state && lrm_state->remote_ra_data) {
remote_ra_data_t *ra_data = lrm_state->remote_ra_data;
if (ra_data->cur_cmd == cmd) {
ra_data->cur_cmd = NULL;
}
if (ra_data->cmds) {
mainloop_set_trigger(ra_data->work);
}
}
report_remote_ra_result(cmd);
free_cmd(cmd);
if(lrm_state) {
lrm_state_disconnect(lrm_state);
}
return FALSE;
}
static void
synthesize_lrmd_success(lrm_state_t *lrm_state, const char *rsc_id, const char *op_type)
{
lrmd_event_data_t op = { 0, };
if (lrm_state == NULL) {
/* if lrm_state not given assume local */
lrm_state = lrm_state_find(controld_globals.our_nodename);
}
CRM_ASSERT(lrm_state != NULL);
op.type = lrmd_event_exec_complete;
op.rsc_id = rsc_id;
op.op_type = op_type;
op.t_run = (unsigned int) time(NULL);
op.t_rcchange = op.t_run;
op.call_id = generate_callid();
lrmd__set_result(&op, PCMK_OCF_OK, PCMK_EXEC_DONE, NULL);
process_lrm_event(lrm_state, &op, NULL, NULL);
}
void
remote_lrm_op_callback(lrmd_event_data_t * op)
{
gboolean cmd_handled = FALSE;
lrm_state_t *lrm_state = NULL;
remote_ra_data_t *ra_data = NULL;
remote_ra_cmd_t *cmd = NULL;
crm_debug("Processing '%s%s%s' event on remote connection to %s: %s "
"(%d) status=%s (%d)",
(op->op_type? op->op_type : ""), (op->op_type? " " : ""),
lrmd_event_type2str(op->type), op->remote_nodename,
services_ocf_exitcode_str(op->rc), op->rc,
pcmk_exec_status_str(op->op_status), op->op_status);
lrm_state = lrm_state_find(op->remote_nodename);
if (!lrm_state || !lrm_state->remote_ra_data) {
crm_debug("No state information found for remote connection event");
return;
}
ra_data = lrm_state->remote_ra_data;
if (op->type == lrmd_event_new_client) {
// Another client has connected to the remote daemon
if (pcmk_is_set(ra_data->status, expect_takeover)) {
// Great, we knew this was coming
lrm_remote_clear_flags(lrm_state, expect_takeover);
lrm_remote_set_flags(lrm_state, takeover_complete);
} else {
crm_err("Disconnecting from Pacemaker Remote node %s due to "
"unexpected client takeover", op->remote_nodename);
/* In this case, lrmd_tls_connection_destroy() will be called under the control of mainloop. */
/* Do not free lrm_state->conn yet. */
/* It'll be freed in the following stop action. */
lrm_state_disconnect_only(lrm_state);
}
return;
}
/* filter all EXEC events up */
if (op->type == lrmd_event_exec_complete) {
if (pcmk_is_set(ra_data->status, takeover_complete)) {
crm_debug("ignoring event, this connection is taken over by another node");
} else {
lrm_op_callback(op);
}
return;
}
if ((op->type == lrmd_event_disconnect) && (ra_data->cur_cmd == NULL)) {
if (!pcmk_is_set(ra_data->status, remote_active)) {
crm_debug("Disconnection from Pacemaker Remote node %s complete",
lrm_state->node_name);
} else if (!remote_ra_is_in_maintenance(lrm_state)) {
crm_err("Lost connection to Pacemaker Remote node %s",
lrm_state->node_name);
ra_data->recurring_cmds = fail_all_monitor_cmds(ra_data->recurring_cmds);
ra_data->cmds = fail_all_monitor_cmds(ra_data->cmds);
} else {
crm_notice("Unmanaged Pacemaker Remote node %s disconnected",
lrm_state->node_name);
/* Do roughly what a 'stop' on the remote-resource would do */
handle_remote_ra_stop(lrm_state, NULL);
remote_node_down(lrm_state->node_name, DOWN_KEEP_LRM);
/* now fake the reply of a successful 'stop' */
synthesize_lrmd_success(NULL, lrm_state->node_name,
PCMK_ACTION_STOP);
}
return;
}
if (!ra_data->cur_cmd) {
crm_debug("no event to match");
return;
}
cmd = ra_data->cur_cmd;
/* Start actions and migrate from actions complete after connection
* comes back to us. */
if ((op->type == lrmd_event_connect)
&& pcmk__strcase_any_of(cmd->action, PCMK_ACTION_START, "migrate_from",
NULL)) {
if (op->connection_rc < 0) {
update_remaining_timeout(cmd);
if ((op->connection_rc == -ENOKEY)
|| (op->connection_rc == -EKEYREJECTED)) {
// Hard error, don't retry
pcmk__set_result(&(cmd->result), PCMK_OCF_INVALID_PARAM,
PCMK_EXEC_ERROR,
pcmk_strerror(op->connection_rc));
} else if (cmd->remaining_timeout > 3000) {
crm_trace("rescheduling start, remaining timeout %d", cmd->remaining_timeout);
g_timeout_add(1000, retry_start_cmd_cb, lrm_state);
return;
} else {
crm_trace("can't reschedule start, remaining timeout too small %d",
cmd->remaining_timeout);
pcmk__format_result(&(cmd->result), PCMK_OCF_UNKNOWN_ERROR,
PCMK_EXEC_TIMEOUT,
"%s without enough time to retry",
pcmk_strerror(op->connection_rc));
}
} else {
lrm_state_reset_tables(lrm_state, TRUE);
pcmk__set_result(&(cmd->result), PCMK_OCF_OK, PCMK_EXEC_DONE, NULL);
lrm_remote_set_flags(lrm_state, remote_active);
}
crm_debug("Remote connection event matched %s action", cmd->action);
report_remote_ra_result(cmd);
cmd_handled = TRUE;
} else if ((op->type == lrmd_event_poke)
&& pcmk__str_eq(cmd->action, PCMK_ACTION_MONITOR,
pcmk__str_casei)) {
if (cmd->monitor_timeout_id) {
g_source_remove(cmd->monitor_timeout_id);
cmd->monitor_timeout_id = 0;
}
/* Only report success the first time, after that only worry about failures.
* For this function, if we get the poke pack, it is always a success. Pokes
* only fail if the send fails, or the response times out. */
if (!pcmk_is_set(cmd->status, cmd_reported_success)) {
pcmk__set_result(&(cmd->result), PCMK_OCF_OK, PCMK_EXEC_DONE, NULL);
report_remote_ra_result(cmd);
cmd_set_flags(cmd, cmd_reported_success);
}
crm_debug("Remote poke event matched %s action", cmd->action);
/* success, keep rescheduling if interval is present. */
if (cmd->interval_ms && !pcmk_is_set(cmd->status, cmd_cancel)) {
ra_data->recurring_cmds = g_list_append(ra_data->recurring_cmds, cmd);
cmd->interval_id = g_timeout_add(cmd->interval_ms,
recurring_helper, cmd);
cmd = NULL; /* prevent free */
}
cmd_handled = TRUE;
} else if ((op->type == lrmd_event_disconnect)
&& pcmk__str_eq(cmd->action, PCMK_ACTION_MONITOR,
pcmk__str_casei)) {
if (pcmk_is_set(ra_data->status, remote_active) &&
!pcmk_is_set(cmd->status, cmd_cancel)) {
pcmk__set_result(&(cmd->result), PCMK_OCF_UNKNOWN_ERROR,
PCMK_EXEC_ERROR,
"Remote connection unexpectedly dropped "
"during monitor");
report_remote_ra_result(cmd);
crm_err("Remote connection to %s unexpectedly dropped during monitor",
lrm_state->node_name);
}
cmd_handled = TRUE;
} else if ((op->type == lrmd_event_new_client)
&& pcmk__str_eq(cmd->action, PCMK_ACTION_STOP,
pcmk__str_casei)) {
handle_remote_ra_stop(lrm_state, cmd);
cmd_handled = TRUE;
} else {
crm_debug("Event did not match %s action", ra_data->cur_cmd->action);
}
if (cmd_handled) {
ra_data->cur_cmd = NULL;
if (ra_data->cmds) {
mainloop_set_trigger(ra_data->work);
}
free_cmd(cmd);
}
}
static void
handle_remote_ra_stop(lrm_state_t * lrm_state, remote_ra_cmd_t * cmd)
{
remote_ra_data_t *ra_data = NULL;
CRM_ASSERT(lrm_state);
ra_data = lrm_state->remote_ra_data;
if (!pcmk_is_set(ra_data->status, takeover_complete)) {
/* delete pending ops when ever the remote connection is intentionally stopped */
g_hash_table_remove_all(lrm_state->active_ops);
} else {
/* we no longer hold the history if this connection has been migrated,
* however, we keep metadata cache for future use */
lrm_state_reset_tables(lrm_state, FALSE);
}
lrm_remote_clear_flags(lrm_state, remote_active);
lrm_state_disconnect(lrm_state);
if (ra_data->cmds) {
g_list_free_full(ra_data->cmds, free_cmd);
}
if (ra_data->recurring_cmds) {
g_list_free_full(ra_data->recurring_cmds, free_cmd);
}
ra_data->cmds = NULL;
ra_data->recurring_cmds = NULL;
ra_data->cur_cmd = NULL;
if (cmd) {
pcmk__set_result(&(cmd->result), PCMK_OCF_OK, PCMK_EXEC_DONE, NULL);
report_remote_ra_result(cmd);
}
}
// \return Standard Pacemaker return code
static int
handle_remote_ra_start(lrm_state_t * lrm_state, remote_ra_cmd_t * cmd, int timeout_ms)
{
const char *server = NULL;
lrmd_key_value_t *tmp = NULL;
int port = 0;
int timeout_used = timeout_ms > MAX_START_TIMEOUT_MS ? MAX_START_TIMEOUT_MS : timeout_ms;
int rc = pcmk_rc_ok;
for (tmp = cmd->params; tmp; tmp = tmp->next) {
if (pcmk__strcase_any_of(tmp->key, XML_RSC_ATTR_REMOTE_RA_ADDR,
XML_RSC_ATTR_REMOTE_RA_SERVER, NULL)) {
server = tmp->value;
} else if (pcmk__str_eq(tmp->key, XML_RSC_ATTR_REMOTE_RA_PORT, pcmk__str_casei)) {
port = atoi(tmp->value);
} else if (pcmk__str_eq(tmp->key, CRM_META "_" XML_RSC_ATTR_CONTAINER, pcmk__str_casei)) {
lrm_remote_set_flags(lrm_state, controlling_guest);
}
}
rc = controld_connect_remote_executor(lrm_state, server, port,
timeout_used);
if (rc != pcmk_rc_ok) {
pcmk__format_result(&(cmd->result), PCMK_OCF_UNKNOWN_ERROR,
PCMK_EXEC_ERROR,
"Could not connect to Pacemaker Remote node %s: %s",
lrm_state->node_name, pcmk_rc_str(rc));
}
return rc;
}
static gboolean
handle_remote_ra_exec(gpointer user_data)
{
int rc = 0;
lrm_state_t *lrm_state = user_data;
remote_ra_data_t *ra_data = lrm_state->remote_ra_data;
remote_ra_cmd_t *cmd;
GList *first = NULL;
if (ra_data->cur_cmd) {
/* still waiting on previous cmd */
return TRUE;
}
while (ra_data->cmds) {
first = ra_data->cmds;
cmd = first->data;
if (cmd->delay_id) {
/* still waiting for start delay timer to trip */
return TRUE;
}
ra_data->cmds = g_list_remove_link(ra_data->cmds, first);
g_list_free_1(first);
if (pcmk__str_any_of(cmd->action, PCMK_ACTION_START, "migrate_from",
NULL)) {
lrm_remote_clear_flags(lrm_state, expect_takeover | takeover_complete);
if (handle_remote_ra_start(lrm_state, cmd,
cmd->timeout) == pcmk_rc_ok) {
/* take care of this later when we get async connection result */
crm_debug("Initiated async remote connection, %s action will complete after connect event",
cmd->action);
ra_data->cur_cmd = cmd;
return TRUE;
}
report_remote_ra_result(cmd);
} else if (!strcmp(cmd->action, PCMK_ACTION_MONITOR)) {
if (lrm_state_is_connected(lrm_state) == TRUE) {
rc = lrm_state_poke_connection(lrm_state);
if (rc < 0) {
pcmk__set_result(&(cmd->result), PCMK_OCF_UNKNOWN_ERROR,
PCMK_EXEC_ERROR, pcmk_strerror(rc));
}
} else {
rc = -1;
pcmk__set_result(&(cmd->result), PCMK_OCF_NOT_RUNNING,
PCMK_EXEC_DONE, "Remote connection inactive");
}
if (rc == 0) {
crm_debug("Poked Pacemaker Remote at node %s, waiting for async response",
cmd->rsc_id);
ra_data->cur_cmd = cmd;
cmd->monitor_timeout_id = g_timeout_add(cmd->timeout, monitor_timeout_cb, cmd);
return TRUE;
}
report_remote_ra_result(cmd);
} else if (!strcmp(cmd->action, PCMK_ACTION_STOP)) {
if (pcmk_is_set(ra_data->status, expect_takeover)) {
/* briefly wait on stop for the takeover event to occur. If the
* takeover event does not occur during the wait period, that's fine.
* It just means that the remote-node's lrm_status section is going to get
* cleared which will require all the resources running in the remote-node
* to be explicitly re-detected via probe actions. If the takeover does occur
* successfully, then we can leave the status section intact. */
cmd->takeover_timeout_id = g_timeout_add((cmd->timeout/2), connection_takeover_timeout_cb, cmd);
ra_data->cur_cmd = cmd;
return TRUE;
}
handle_remote_ra_stop(lrm_state, cmd);
} else if (!strcmp(cmd->action, "migrate_to")) {
lrm_remote_clear_flags(lrm_state, takeover_complete);
lrm_remote_set_flags(lrm_state, expect_takeover);
pcmk__set_result(&(cmd->result), PCMK_OCF_OK, PCMK_EXEC_DONE, NULL);
report_remote_ra_result(cmd);
} else if (pcmk__str_any_of(cmd->action, PCMK_ACTION_RELOAD,
- CRMD_ACTION_RELOAD_AGENT, NULL)) {
+ PCMK_ACTION_RELOAD_AGENT, NULL)) {
/* Currently the only reloadable parameter is reconnect_interval,
* which is only used by the scheduler via the CIB, so reloads are a
* no-op.
*
* @COMPAT DC <2.1.0: We only need to check for "reload" in case
* we're in a rolling upgrade with a DC scheduling "reload" instead
* of "reload-agent". An OCF 1.1 "reload" would be a no-op anyway,
* so this would work for that purpose as well.
*/
pcmk__set_result(&(cmd->result), PCMK_OCF_OK, PCMK_EXEC_DONE, NULL);
report_remote_ra_result(cmd);
}
free_cmd(cmd);
}
return TRUE;
}
static void
remote_ra_data_init(lrm_state_t * lrm_state)
{
remote_ra_data_t *ra_data = NULL;
if (lrm_state->remote_ra_data) {
return;
}
ra_data = calloc(1, sizeof(remote_ra_data_t));
ra_data->work = mainloop_add_trigger(G_PRIORITY_HIGH, handle_remote_ra_exec, lrm_state);
lrm_state->remote_ra_data = ra_data;
}
void
remote_ra_cleanup(lrm_state_t * lrm_state)
{
remote_ra_data_t *ra_data = lrm_state->remote_ra_data;
if (!ra_data) {
return;
}
if (ra_data->cmds) {
g_list_free_full(ra_data->cmds, free_cmd);
}
if (ra_data->recurring_cmds) {
g_list_free_full(ra_data->recurring_cmds, free_cmd);
}
mainloop_destroy_trigger(ra_data->work);
free(ra_data);
lrm_state->remote_ra_data = NULL;
}
gboolean
is_remote_lrmd_ra(const char *agent, const char *provider, const char *id)
{
if (agent && provider && !strcmp(agent, REMOTE_LRMD_RA) && !strcmp(provider, "pacemaker")) {
return TRUE;
}
if ((id != NULL) && (lrm_state_find(id) != NULL)
&& !pcmk__str_eq(id, controld_globals.our_nodename, pcmk__str_casei)) {
return TRUE;
}
return FALSE;
}
lrmd_rsc_info_t *
remote_ra_get_rsc_info(lrm_state_t * lrm_state, const char *rsc_id)
{
lrmd_rsc_info_t *info = NULL;
if ((lrm_state_find(rsc_id))) {
info = calloc(1, sizeof(lrmd_rsc_info_t));
info->id = strdup(rsc_id);
info->type = strdup(REMOTE_LRMD_RA);
info->standard = strdup(PCMK_RESOURCE_CLASS_OCF);
info->provider = strdup("pacemaker");
}
return info;
}
static gboolean
is_remote_ra_supported_action(const char *action)
{
return pcmk__str_any_of(action,
PCMK_ACTION_START,
PCMK_ACTION_STOP,
PCMK_ACTION_MONITOR,
CRMD_ACTION_MIGRATE,
CRMD_ACTION_MIGRATED,
- CRMD_ACTION_RELOAD_AGENT,
+ PCMK_ACTION_RELOAD_AGENT,
PCMK_ACTION_RELOAD,
NULL);
}
static GList *
fail_all_monitor_cmds(GList * list)
{
GList *rm_list = NULL;
remote_ra_cmd_t *cmd = NULL;
GList *gIter = NULL;
for (gIter = list; gIter != NULL; gIter = gIter->next) {
cmd = gIter->data;
if ((cmd->interval_ms > 0)
&& pcmk__str_eq(cmd->action, PCMK_ACTION_MONITOR,
pcmk__str_casei)) {
rm_list = g_list_append(rm_list, cmd);
}
}
for (gIter = rm_list; gIter != NULL; gIter = gIter->next) {
cmd = gIter->data;
pcmk__set_result(&(cmd->result), PCMK_OCF_UNKNOWN_ERROR,
PCMK_EXEC_ERROR, "Lost connection to remote executor");
crm_trace("Pre-emptively failing %s %s (interval=%u, %s)",
cmd->action, cmd->rsc_id, cmd->interval_ms, cmd->userdata);
report_remote_ra_result(cmd);
list = g_list_remove(list, cmd);
free_cmd(cmd);
}
/* frees only the list data, not the cmds */
g_list_free(rm_list);
return list;
}
static GList *
remove_cmd(GList * list, const char *action, guint interval_ms)
{
remote_ra_cmd_t *cmd = NULL;
GList *gIter = NULL;
for (gIter = list; gIter != NULL; gIter = gIter->next) {
cmd = gIter->data;
if ((cmd->interval_ms == interval_ms)
&& pcmk__str_eq(cmd->action, action, pcmk__str_casei)) {
break;
}
cmd = NULL;
}
if (cmd) {
list = g_list_remove(list, cmd);
free_cmd(cmd);
}
return list;
}
int
remote_ra_cancel(lrm_state_t *lrm_state, const char *rsc_id,
const char *action, guint interval_ms)
{
lrm_state_t *connection_rsc = NULL;
remote_ra_data_t *ra_data = NULL;
connection_rsc = lrm_state_find(rsc_id);
if (!connection_rsc || !connection_rsc->remote_ra_data) {
return -EINVAL;
}
ra_data = connection_rsc->remote_ra_data;
ra_data->cmds = remove_cmd(ra_data->cmds, action, interval_ms);
ra_data->recurring_cmds = remove_cmd(ra_data->recurring_cmds, action,
interval_ms);
if (ra_data->cur_cmd &&
(ra_data->cur_cmd->interval_ms == interval_ms) &&
(pcmk__str_eq(ra_data->cur_cmd->action, action, pcmk__str_casei))) {
cmd_set_flags(ra_data->cur_cmd, cmd_cancel);
}
return 0;
}
static remote_ra_cmd_t *
handle_dup_monitor(remote_ra_data_t *ra_data, guint interval_ms,
const char *userdata)
{
GList *gIter = NULL;
remote_ra_cmd_t *cmd = NULL;
/* there are 3 places a potential duplicate monitor operation
* could exist.
* 1. recurring_cmds list. where the op is waiting for its next interval
* 2. cmds list, where the op is queued to get executed immediately
* 3. cur_cmd, which means the monitor op is in flight right now.
*/
if (interval_ms == 0) {
return NULL;
}
if (ra_data->cur_cmd &&
!pcmk_is_set(ra_data->cur_cmd->status, cmd_cancel) &&
(ra_data->cur_cmd->interval_ms == interval_ms)
&& pcmk__str_eq(ra_data->cur_cmd->action, PCMK_ACTION_MONITOR,
pcmk__str_casei)) {
cmd = ra_data->cur_cmd;
goto handle_dup;
}
for (gIter = ra_data->recurring_cmds; gIter != NULL; gIter = gIter->next) {
cmd = gIter->data;
if ((cmd->interval_ms == interval_ms)
&& pcmk__str_eq(cmd->action, PCMK_ACTION_MONITOR,
pcmk__str_casei)) {
goto handle_dup;
}
}
for (gIter = ra_data->cmds; gIter != NULL; gIter = gIter->next) {
cmd = gIter->data;
if ((cmd->interval_ms == interval_ms)
&& pcmk__str_eq(cmd->action, PCMK_ACTION_MONITOR,
pcmk__str_casei)) {
goto handle_dup;
}
}
return NULL;
handle_dup:
crm_trace("merging duplicate monitor cmd " PCMK__OP_FMT,
cmd->rsc_id, PCMK_ACTION_MONITOR, interval_ms);
/* update the userdata */
if (userdata) {
free(cmd->userdata);
cmd->userdata = strdup(userdata);
}
/* if we've already reported success, generate a new call id */
if (pcmk_is_set(cmd->status, cmd_reported_success)) {
cmd->start_time = time(NULL);
cmd->call_id = generate_callid();
cmd_clear_flags(cmd, cmd_reported_success);
}
/* if we have an interval_id set, that means we are in the process of
* waiting for this cmd's next interval. instead of waiting, cancel
* the timer and execute the action immediately */
if (cmd->interval_id) {
g_source_remove(cmd->interval_id);
cmd->interval_id = 0;
recurring_helper(cmd);
}
return cmd;
}
/*!
* \internal
* \brief Execute an action using the (internal) ocf:pacemaker:remote agent
*
* \param[in] lrm_state Executor state object for remote connection
* \param[in] rsc_id Connection resource ID
* \param[in] action Action to execute
* \param[in] userdata String to copy and pass to execution callback
* \param[in] interval_ms Action interval (in milliseconds)
* \param[in] timeout_ms Action timeout (in milliseconds)
* \param[in] start_delay_ms Delay (in milliseconds) before executing action
* \param[in,out] params Connection resource parameters
* \param[out] call_id Where to store call ID on success
*
* \return Standard Pacemaker return code
* \note This takes ownership of \p params, which should not be used or freed
* after calling this function.
*/
int
controld_execute_remote_agent(const lrm_state_t *lrm_state, const char *rsc_id,
const char *action, const char *userdata,
guint interval_ms, int timeout_ms,
int start_delay_ms, lrmd_key_value_t *params,
int *call_id)
{
lrm_state_t *connection_rsc = NULL;
remote_ra_cmd_t *cmd = NULL;
remote_ra_data_t *ra_data = NULL;
*call_id = 0;
CRM_CHECK((lrm_state != NULL) && (rsc_id != NULL) && (action != NULL)
&& (userdata != NULL) && (call_id != NULL),
lrmd_key_value_freeall(params); return EINVAL);
if (!is_remote_ra_supported_action(action)) {
lrmd_key_value_freeall(params);
return EOPNOTSUPP;
}
connection_rsc = lrm_state_find(rsc_id);
if (connection_rsc == NULL) {
lrmd_key_value_freeall(params);
return ENOTCONN;
}
remote_ra_data_init(connection_rsc);
ra_data = connection_rsc->remote_ra_data;
cmd = handle_dup_monitor(ra_data, interval_ms, userdata);
if (cmd) {
*call_id = cmd->call_id;
lrmd_key_value_freeall(params);
return pcmk_rc_ok;
}
cmd = calloc(1, sizeof(remote_ra_cmd_t));
if (cmd == NULL) {
lrmd_key_value_freeall(params);
return ENOMEM;
}
cmd->owner = strdup(lrm_state->node_name);
cmd->rsc_id = strdup(rsc_id);
cmd->action = strdup(action);
cmd->userdata = strdup(userdata);
if ((cmd->owner == NULL) || (cmd->rsc_id == NULL) || (cmd->action == NULL)
|| (cmd->userdata == NULL)) {
free_cmd(cmd);
lrmd_key_value_freeall(params);
return ENOMEM;
}
cmd->interval_ms = interval_ms;
cmd->timeout = timeout_ms;
cmd->start_delay = start_delay_ms;
cmd->params = params;
cmd->start_time = time(NULL);
cmd->call_id = generate_callid();
if (cmd->start_delay) {
cmd->delay_id = g_timeout_add(cmd->start_delay, start_delay_helper, cmd);
}
ra_data->cmds = g_list_append(ra_data->cmds, cmd);
mainloop_set_trigger(ra_data->work);
*call_id = cmd->call_id;
return pcmk_rc_ok;
}
/*!
* \internal
* \brief Immediately fail all monitors of a remote node, if proxied here
*
* \param[in] node_name Name of pacemaker_remote node
*/
void
remote_ra_fail(const char *node_name)
{
lrm_state_t *lrm_state = lrm_state_find(node_name);
if (lrm_state && lrm_state_is_connected(lrm_state)) {
remote_ra_data_t *ra_data = lrm_state->remote_ra_data;
crm_info("Failing monitors on Pacemaker Remote node %s", node_name);
ra_data->recurring_cmds = fail_all_monitor_cmds(ra_data->recurring_cmds);
ra_data->cmds = fail_all_monitor_cmds(ra_data->cmds);
}
}
/* A guest node fencing implied by host fencing looks like:
*
* <pseudo_event id="103" operation="stonith" operation_key="stonith-lxc1-off"
* on_node="lxc1" on_node_uuid="lxc1">
* <attributes CRM_meta_on_node="lxc1" CRM_meta_on_node_uuid="lxc1"
* CRM_meta_stonith_action="off" crm_feature_set="3.0.12"/>
* <downed>
* <node id="lxc1"/>
* </downed>
* </pseudo_event>
*/
#define XPATH_PSEUDO_FENCE "/" XML_GRAPH_TAG_PSEUDO_EVENT \
"[@" XML_LRM_ATTR_TASK "='stonith']/" XML_GRAPH_TAG_DOWNED \
"/" XML_CIB_TAG_NODE
/*!
* \internal
* \brief Check a pseudo-action for Pacemaker Remote node side effects
*
* \param[in,out] xml XML of pseudo-action to check
*/
void
remote_ra_process_pseudo(xmlNode *xml)
{
xmlXPathObjectPtr search = xpath_search(xml, XPATH_PSEUDO_FENCE);
if (numXpathResults(search) == 1) {
xmlNode *result = getXpathResult(search, 0);
/* Normally, we handle the necessary side effects of a guest node stop
* action when reporting the remote agent's result. However, if the stop
* is implied due to fencing, it will be a fencing pseudo-event, and
* there won't be a result to report. Handle that case here.
*
* This will result in a duplicate call to remote_node_down() if the
* guest stop was real instead of implied, but that shouldn't hurt.
*
* There is still one corner case that isn't handled: if a guest node
* isn't running any resources when its host is fenced, it will appear
* to be cleanly stopped, so there will be no pseudo-fence, and our
* peer cache state will be incorrect unless and until the guest is
* recovered.
*/
if (result) {
const char *remote = ID(result);
if (remote) {
remote_node_down(remote, DOWN_ERASE_LRM);
}
}
}
freeXpathObject(search);
}
static void
remote_ra_maintenance(lrm_state_t * lrm_state, gboolean maintenance)
{
xmlNode *update, *state;
int call_opt;
crm_node_t *node;
call_opt = crmd_cib_smart_opt();
node = crm_remote_peer_get(lrm_state->node_name);
CRM_CHECK(node != NULL, return);
update = create_xml_node(NULL, XML_CIB_TAG_STATUS);
state = create_node_state_update(node, node_update_none, update,
__func__);
crm_xml_add(state, XML_NODE_IS_MAINTENANCE, maintenance?"1":"0");
if (controld_update_cib(XML_CIB_TAG_STATUS, update, call_opt, NULL,
NULL) == pcmk_rc_ok) {
/* TODO: still not 100% sure that async update will succeed ... */
if (maintenance) {
lrm_remote_set_flags(lrm_state, remote_in_maint);
} else {
lrm_remote_clear_flags(lrm_state, remote_in_maint);
}
}
free_xml(update);
}
#define XPATH_PSEUDO_MAINTENANCE "//" XML_GRAPH_TAG_PSEUDO_EVENT \
"[@" XML_LRM_ATTR_TASK "='" CRM_OP_MAINTENANCE_NODES "']/" \
XML_GRAPH_TAG_MAINTENANCE
/*!
* \internal
* \brief Check a pseudo-action holding updates for maintenance state
*
* \param[in,out] xml XML of pseudo-action to check
*/
void
remote_ra_process_maintenance_nodes(xmlNode *xml)
{
xmlXPathObjectPtr search = xpath_search(xml, XPATH_PSEUDO_MAINTENANCE);
if (numXpathResults(search) == 1) {
xmlNode *node;
int cnt = 0, cnt_remote = 0;
for (node =
first_named_child(getXpathResult(search, 0), XML_CIB_TAG_NODE);
node != NULL; node = pcmk__xml_next(node)) {
lrm_state_t *lrm_state = lrm_state_find(ID(node));
cnt++;
if (lrm_state && lrm_state->remote_ra_data &&
pcmk_is_set(((remote_ra_data_t *) lrm_state->remote_ra_data)->status, remote_active)) {
int is_maint;
cnt_remote++;
pcmk__scan_min_int(crm_element_value(node, XML_NODE_IS_MAINTENANCE),
&is_maint, 0);
remote_ra_maintenance(lrm_state, is_maint);
}
}
crm_trace("Action holds %d nodes (%d remotes found) "
"adjusting maintenance-mode", cnt, cnt_remote);
}
freeXpathObject(search);
}
gboolean
remote_ra_is_in_maintenance(lrm_state_t * lrm_state)
{
remote_ra_data_t *ra_data = lrm_state->remote_ra_data;
return pcmk_is_set(ra_data->status, remote_in_maint);
}
gboolean
remote_ra_controlling_guest(lrm_state_t * lrm_state)
{
remote_ra_data_t *ra_data = lrm_state->remote_ra_data;
return pcmk_is_set(ra_data->status, controlling_guest);
}
diff --git a/include/crm/common/actions.h b/include/crm/common/actions.h
index 01ddd8427a..1b391839a1 100644
--- a/include/crm/common/actions.h
+++ b/include/crm/common/actions.h
@@ -1,37 +1,38 @@
/*
* Copyright 2023 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU Lesser General Public License
* version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY.
*/
#ifndef PCMK__CRM_COMMON_ACTIONS__H
#define PCMK__CRM_COMMON_ACTIONS__H
#ifdef __cplusplus
extern "C" {
#endif
/*!
* \file
* \brief APIs related to actions
* \ingroup core
*/
// Action names as strings
#define PCMK_ACTION_DEMOTE "demote"
#define PCMK_ACTION_META_DATA "meta-data"
#define PCMK_ACTION_MONITOR "monitor"
#define PCMK_ACTION_NOTIFY "notify"
#define PCMK_ACTION_PROMOTE "promote"
#define PCMK_ACTION_RELOAD "reload"
+#define PCMK_ACTION_RELOAD_AGENT "reload-agent"
#define PCMK_ACTION_START "start"
#define PCMK_ACTION_STOP "stop"
#ifdef __cplusplus
}
#endif
#endif // PCMK__CRM_COMMON_ACTIONS__H
diff --git a/include/crm/crm.h b/include/crm/crm.h
index 74b2b698dc..258f7939c5 100644
--- a/include/crm/crm.h
+++ b/include/crm/crm.h
@@ -1,223 +1,223 @@
/*
* Copyright 2004-2023 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU Lesser General Public License
* version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY.
*/
#ifndef PCMK__CRM_CRM__H
# define PCMK__CRM_CRM__H
# include <crm_config.h>
# include <stdlib.h>
# include <glib.h>
# include <stdbool.h>
# include <string.h>
# include <libxml/tree.h>
#ifdef __cplusplus
extern "C" {
#endif
/**
* \file
* \brief A dumping ground
* \ingroup core
*/
#ifndef PCMK_ALLOW_DEPRECATED
/*!
* \brief Allow use of deprecated Pacemaker APIs
*
* By default, external code using Pacemaker headers is allowed to use
* deprecated Pacemaker APIs. If PCMK_ALLOW_DEPRECATED is defined to 0 before
* including any Pacemaker headers, deprecated APIs will be unusable. It is
* strongly recommended to leave this unchanged for production and release
* builds, to avoid breakage when users upgrade to new Pacemaker releases that
* deprecate more APIs. This should be defined to 0 only for development and
* testing builds when desiring to check for usage of currently deprecated APIs.
*/
#define PCMK_ALLOW_DEPRECATED 1
#endif
/*!
* The CRM feature set assists with compatibility in mixed-version clusters.
* The major version number increases when nodes with different versions
* would not work (rolling upgrades are not allowed). The minor version
* number increases when mixed-version clusters are allowed only during
* rolling upgrades (a node with the oldest feature set will be elected DC). The
* minor-minor version number is ignored, but allows resource agents to detect
* cluster support for various features.
*
* The feature set also affects the processing of old saved CIBs (such as for
* many scheduler regression tests).
*
* Particular feature points currently tested by Pacemaker code:
*
* >2.1: Operation updates include timing data
* >=3.0.5: XML v2 digests are created
* >=3.0.8: Peers do not need acks for cancellations
* >=3.0.9: DC will send its own shutdown request to all peers
* XML v2 patchsets are created by default
* >=3.0.13: Fail counts include operation name and interval
* >=3.2.0: DC supports PCMK_EXEC_INVALID and PCMK_EXEC_NOT_CONNECTED
*/
# define CRM_FEATURE_SET "3.18.0"
/* Pacemaker's CPG protocols use fixed-width binary fields for the sender and
* recipient of a CPG message. This imposes an arbitrary limit on cluster node
* names.
*/
//! \brief Maximum length of a Corosync cluster node name (in bytes)
#define MAX_NAME 256
# define CRM_META "CRM_meta"
extern char *crm_system_name;
/* *INDENT-OFF* */
// How we represent "infinite" scores
# define CRM_SCORE_INFINITY 1000000
# define CRM_INFINITY_S "INFINITY"
# define CRM_PLUS_INFINITY_S "+" CRM_INFINITY_S
# define CRM_MINUS_INFINITY_S "-" CRM_INFINITY_S
/* @COMPAT API < 2.0.0 Deprecated "infinity" aliases
*
* INFINITY might be defined elsewhere (e.g. math.h), so undefine it first.
* This, of course, complicates any attempt to use the other definition in any
* code that includes this header.
*/
# undef INFINITY
# define INFINITY_S "INFINITY"
# define MINUS_INFINITY_S "-INFINITY"
# define INFINITY 1000000
/* Sub-systems */
# define CRM_SYSTEM_DC "dc"
#define CRM_SYSTEM_DCIB "dcib" // Primary instance of CIB manager
# define CRM_SYSTEM_CIB "cib"
# define CRM_SYSTEM_CRMD "crmd"
# define CRM_SYSTEM_LRMD "lrmd"
# define CRM_SYSTEM_PENGINE "pengine"
# define CRM_SYSTEM_TENGINE "tengine"
# define CRM_SYSTEM_STONITHD "stonithd"
# define CRM_SYSTEM_MCP "pacemakerd"
// Names of internally generated node attributes
# define CRM_ATTR_UNAME "#uname"
# define CRM_ATTR_ID "#id"
# define CRM_ATTR_KIND "#kind"
# define CRM_ATTR_ROLE "#role"
# define CRM_ATTR_IS_DC "#is_dc"
# define CRM_ATTR_CLUSTER_NAME "#cluster-name"
# define CRM_ATTR_SITE_NAME "#site-name"
# define CRM_ATTR_UNFENCED "#node-unfenced"
# define CRM_ATTR_DIGESTS_ALL "#digests-all"
# define CRM_ATTR_DIGESTS_SECURE "#digests-secure"
# define CRM_ATTR_PROTOCOL "#attrd-protocol"
# define CRM_ATTR_FEATURE_SET "#feature-set"
/* Valid operations */
# define CRM_OP_NOOP "noop"
# define CRM_OP_JOIN_ANNOUNCE "join_announce"
# define CRM_OP_JOIN_OFFER "join_offer"
# define CRM_OP_JOIN_REQUEST "join_request"
# define CRM_OP_JOIN_ACKNAK "join_ack_nack"
# define CRM_OP_JOIN_CONFIRM "join_confirm"
# define CRM_OP_PING "ping"
# define CRM_OP_NODE_INFO "node-info"
# define CRM_OP_THROTTLE "throttle"
# define CRM_OP_VOTE "vote"
# define CRM_OP_NOVOTE "no-vote"
# define CRM_OP_HELLO "hello"
# define CRM_OP_PECALC "pe_calc"
# define CRM_OP_QUIT "quit"
# define CRM_OP_LOCAL_SHUTDOWN "start_shutdown"
# define CRM_OP_SHUTDOWN_REQ "req_shutdown"
# define CRM_OP_SHUTDOWN "do_shutdown"
# define CRM_OP_FENCE "stonith"
# define CRM_OP_REGISTER "register"
# define CRM_OP_IPC_FWD "ipc_fwd"
# define CRM_OP_INVOKE_LRM "lrm_invoke"
# define CRM_OP_LRM_REFRESH "lrm_refresh" //!< Deprecated since 1.1.10
# define CRM_OP_LRM_DELETE "lrm_delete"
# define CRM_OP_LRM_FAIL "lrm_fail"
# define CRM_OP_PROBED "probe_complete"
# define CRM_OP_REPROBE "probe_again"
# define CRM_OP_CLEAR_FAILCOUNT "clear_failcount"
# define CRM_OP_REMOTE_STATE "remote_state"
# define CRM_OP_RELAXED_SET "one-or-more"
# define CRM_OP_RELAXED_CLONE "clone-one-or-more"
# define CRM_OP_RM_NODE_CACHE "rm_node_cache"
# define CRM_OP_MAINTENANCE_NODES "maintenance_nodes"
/* Possible cluster membership states */
# define CRMD_JOINSTATE_DOWN "down"
# define CRMD_JOINSTATE_PENDING "pending"
# define CRMD_JOINSTATE_MEMBER "member"
# define CRMD_JOINSTATE_NACK "banned"
# define CRMD_ACTION_DELETE "delete"
# define CRMD_ACTION_CANCEL "cancel"
-# define CRMD_ACTION_RELOAD_AGENT "reload-agent"
+# define CRMD_ACTION_RELOAD_AGENT PCMK_ACTION_RELOAD_AGENT
# define CRMD_ACTION_MIGRATE "migrate_to"
# define CRMD_ACTION_MIGRATED "migrate_from"
# define CRMD_ACTION_STARTED "running"
# define CRMD_ACTION_STOPPED "stopped"
# define CRMD_ACTION_PROMOTED "promoted"
# define CRMD_ACTION_DEMOTED "demoted"
# define CRMD_ACTION_NOTIFIED "notified"
# define CRMD_METADATA_CALL_TIMEOUT 30000
/* short names */
# define RSC_DELETE CRMD_ACTION_DELETE
# define RSC_CANCEL CRMD_ACTION_CANCEL
# define RSC_MIGRATE CRMD_ACTION_MIGRATE
# define RSC_MIGRATED CRMD_ACTION_MIGRATED
# define RSC_STARTED CRMD_ACTION_STARTED
# define RSC_STOPPED CRMD_ACTION_STOPPED
# define RSC_PROMOTED CRMD_ACTION_PROMOTED
# define RSC_DEMOTED CRMD_ACTION_DEMOTED
# define RSC_NOTIFIED CRMD_ACTION_NOTIFIED
/* *INDENT-ON* */
# include <crm/common/actions.h>
# include <crm/common/cib.h>
# include <crm/common/logging.h>
# include <crm/common/util.h>
static inline const char *
crm_action_str(const char *task, guint interval_ms) {
if ((task != NULL) && (interval_ms == 0)
&& (strcasecmp(task, PCMK_ACTION_MONITOR) == 0)) {
return "probe";
}
return task;
}
#if !defined(PCMK_ALLOW_DEPRECATED) || (PCMK_ALLOW_DEPRECATED == 1)
#include <crm/crm_compat.h>
#endif
#ifdef __cplusplus
}
#endif
#endif
diff --git a/include/crm/pengine/internal.h b/include/crm/pengine/internal.h
index a32fbfdae8..ec291ec91e 100644
--- a/include/crm/pengine/internal.h
+++ b/include/crm/pengine/internal.h
@@ -1,753 +1,753 @@
/*
* Copyright 2004-2023 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 <stdbool.h>
# include <stdint.h>
# include <string.h>
# include <crm/msg_xml.h>
# include <crm/pengine/status.h>
# include <crm/pengine/remote_internal.h>
# include <crm/common/internal.h>
# include <crm/common/options_internal.h>
# include <crm/common/output_internal.h>
const char *pe__resource_description(const pe_resource_t *rsc, uint32_t show_opts);
enum pe__clone_flags {
// Whether instances should be started sequentially
pe__clone_ordered = (1 << 0),
// Whether promotion scores have been added
pe__clone_promotion_added = (1 << 1),
// Whether promotion constraints have been added
pe__clone_promotion_constrained = (1 << 2),
};
bool pe__clone_is_ordered(const pe_resource_t *clone);
int pe__set_clone_flag(pe_resource_t *clone, enum pe__clone_flags flag);
bool pe__clone_flag_is_set(const pe_resource_t *clone, uint32_t flags);
enum pe__group_flags {
pe__group_ordered = (1 << 0), // Members start sequentially
pe__group_colocated = (1 << 1), // Members must be on same node
};
bool pe__group_flag_is_set(const pe_resource_t *group, uint32_t flags);
pe_resource_t *pe__last_group_member(const pe_resource_t *group);
# 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...) 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(__func__, __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(__func__, __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(__func__, __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(__func__, __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(__func__, __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(__func__, __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(__func__, __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(__func__, __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(__func__, __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(__func__, __LINE__, LOG_TRACE, \
"Ordering", "constraint", \
order_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 = (1 << 0),
pe_wo_restart_type = (1 << 1),
pe_wo_role_after = (1 << 2),
pe_wo_poweroff = (1 << 3),
pe_wo_require_all = (1 << 4),
pe_wo_order_score = (1 << 5),
pe_wo_neg_threshold = (1 << 6),
pe_wo_remove_after = (1 << 7),
pe_wo_ping_node = (1 << 8),
pe_wo_order_inst = (1 << 9),
pe_wo_coloc_inst = (1 << 10),
pe_wo_group_order = (1 << 11),
pe_wo_group_coloc = (1 << 12),
pe_wo_upstart = (1 << 13),
pe_wo_nagios = (1 << 14),
pe_wo_set_ordering = (1 << 15),
};
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(__func__, __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
GList *node_list_rh; // List of pe_node_t*
} pe__location_t;
typedef struct pe__order_constraint_s {
int id;
uint32_t flags; // Group of enum pe_ordering flags
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;
const pe_resource_t *pe__const_top_resource(const pe_resource_t *rsc,
bool include_bundle);
int pe__clone_max(const pe_resource_t *clone);
int pe__clone_node_max(const pe_resource_t *clone);
int pe__clone_promoted_max(const pe_resource_t *clone);
int pe__clone_promoted_node_max(const pe_resource_t *clone);
void pe__create_clone_notifications(pe_resource_t *clone);
void pe__free_clone_notification_data(pe_resource_t *clone);
void pe__create_clone_notif_pseudo_ops(pe_resource_t *clone,
pe_action_t *start, pe_action_t *started,
pe_action_t *stop, pe_action_t *stopped);
pe_action_t *pe__new_rsc_pseudo_action(pe_resource_t *rsc, const char *task,
bool optional, bool runnable);
void pe__create_promotable_pseudo_ops(pe_resource_t *clone, bool any_promoting,
bool any_demoting);
bool pe_can_fence(const pe_working_set_t *data_set, const pe_node_t *node);
void add_hash_param(GHashTable * hash, const char *name, const char *value);
/*!
* \internal
* \enum pe__rsc_node
* \brief Type of resource location lookup to perform
*/
enum pe__rsc_node {
pe__rsc_node_assigned = 0, //!< Where resource is assigned
pe__rsc_node_current = 1, //!< Where resource is running
// @COMPAT: Use in native_location() at a compatibility break
pe__rsc_node_pending = 2, //!< Where resource is pending
};
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(pcmk__output_t *out);
void verify_pe_options(GHashTable * options);
void native_add_running(pe_resource_t * rsc, pe_node_t * node, pe_working_set_t * data_set, gboolean failed);
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);
//! \deprecated This function will be removed in a future release
void native_print(pe_resource_t *rsc, const char *pre_text, long options,
void *print_data);
//! \deprecated This function will be removed in a future release
void group_print(pe_resource_t *rsc, const char *pre_text, long options,
void *print_data);
//! \deprecated This function will be removed in a future release
void clone_print(pe_resource_t *rsc, const char *pre_text, long options,
void *print_data);
//! \deprecated This function will be removed in a future release
void pe__print_bundle(pe_resource_t *rsc, const char *pre_text, long options,
void *print_data);
gchar *pcmk__native_output_string(const pe_resource_t *rsc, const char *name,
const pe_node_t *node, uint32_t show_opts,
const char *target_role, bool show_nodes);
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);
// Clone notifications (pe_notif.c)
void pe__order_notifs_after_fencing(const pe_action_t *action,
pe_resource_t *rsc,
pe_action_t *stonith_op);
static inline const char *
pe__rsc_bool_str(const pe_resource_t *rsc, uint64_t rsc_flag)
{
return pcmk__btoa(pcmk_is_set(rsc->flags, rsc_flag));
}
int pe__clone_xml(pcmk__output_t *out, va_list args);
int pe__clone_default(pcmk__output_t *out, va_list args);
int pe__group_xml(pcmk__output_t *out, va_list args);
int pe__group_default(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__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);
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);
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 = (1 << 0),
pe_fc_effective = (1 << 1), // don't count expired failures
pe_fc_fillers = (1 << 2), // if container, include filler failures in count
};
int pe_get_failcount(const pe_node_t *node, pe_resource_t *rsc,
time_t *last_failure, uint32_t flags,
const xmlNode *xml_op);
pe_action_t *pe__clear_failcount(pe_resource_t *rsc, const pe_node_t *node,
const char *reason,
pe_working_set_t *data_set);
/* Functions for finding/counting a resource's active nodes */
bool pe__count_active_node(const pe_resource_t *rsc, pe_node_t *node,
pe_node_t **active, 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 (rsc == NULL)? NULL : rsc->fns->active_node(rsc, NULL, NULL);
}
/* Binary like operators for lists of nodes */
GHashTable *pe__node_list2table(const GList *list);
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);
void pe__show_node_scores_as(const char *file, const char *function,
int line, bool to_log, const pe_resource_t *rsc,
const char *comment, GHashTable *nodes,
pe_working_set_t *data_set);
#define pe__show_node_scores(level, rsc, text, nodes, data_set) \
pe__show_node_scores_as(__FILE__, __func__, __LINE__, \
(level), (rsc), (text), (nodes), (data_set))
xmlNode *find_rsc_op_entry(const pe_resource_t *rsc, const char *key);
pe_action_t *custom_action(pe_resource_t *rsc, char *key, const char *task,
const 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, rsc->cluster);
# 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, rsc->cluster);
# define stop_key(rsc) pcmk__op_key(rsc->id, PCMK_ACTION_STOP, 0)
# define stop_action(rsc, node, optional) custom_action( \
rsc, stop_key(rsc), PCMK_ACTION_STOP, node, \
optional, TRUE, rsc->cluster);
-# define reload_key(rsc) pcmk__op_key(rsc->id, CRMD_ACTION_RELOAD_AGENT, 0)
+# define reload_key(rsc) pcmk__op_key(rsc->id, PCMK_ACTION_RELOAD_AGENT, 0)
# define start_key(rsc) pcmk__op_key(rsc->id, PCMK_ACTION_START, 0)
# define start_action(rsc, node, optional) custom_action( \
rsc, start_key(rsc), PCMK_ACTION_START, node, \
optional, TRUE, rsc->cluster)
# 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, rsc->cluster)
# define promote_key(rsc) pcmk__op_key(rsc->id, PCMK_ACTION_PROMOTE, 0)
# define promote_action(rsc, node, optional) custom_action( \
rsc, promote_key(rsc), PCMK_ACTION_PROMOTE, node, \
optional, TRUE, rsc->cluster)
# 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, rsc->cluster)
# define demote_key(rsc) pcmk__op_key(rsc->id, PCMK_ACTION_DEMOTE, 0)
# define demote_action(rsc, node, optional) custom_action( \
rsc, demote_key(rsc), PCMK_ACTION_DEMOTE, node, \
optional, TRUE, rsc->cluster)
# 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, rsc->cluster)
extern int pe_get_configured_timeout(pe_resource_t *rsc, const char *action,
pe_working_set_t *data_set);
pe_action_t *find_first_action(const GList *input, const char *uuid,
const char *task, const pe_node_t *on_node);
enum action_tasks get_complex_task(const pe_resource_t *rsc, const char *name);
extern GList *find_actions(GList *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);
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);
void resource_location(pe_resource_t *rsc, const pe_node_t *node, int score,
const char *tag, pe_working_set_t *data_set);
extern int pe__is_newer_op(const xmlNode *xml_a, const xmlNode *xml_b,
bool same_node_default);
extern gint sort_op_by_callid(gconstpointer a, gconstpointer b);
gboolean get_target_role(const pe_resource_t *rsc, enum rsc_role_e *role);
void pe__set_next_role(pe_resource_t *rsc, enum rsc_role_e role,
const char *why);
pe_resource_t *find_clone_instance(const pe_resource_t *rsc,
const char *sub_id);
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(const 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(const xmlNode *xml_op);
gint pe__cmp_node_name(gconstpointer a, gconstpointer b);
bool is_set_recursive(const 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 *pe__calculate_digests(pe_resource_t *rsc, const char *task,
guint *interval_ms,
const pe_node_t *node,
const xmlNode *xml_op,
GHashTable *overrides,
bool calc_secure,
pe_working_set_t *data_set);
void pe__free_digests(gpointer ptr);
op_digest_cache_t *rsc_action_digest_cmp(pe_resource_t *rsc,
const 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);
char *pe__action2reason(const pe_action_t *action, enum pe_action_flags flag);
void pe_action_set_reason(pe_action_t *action, const char *reason, bool overwrite);
void pe__add_action_expected_result(pe_action_t *action, int expected_result);
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);
void pe__clear_resource_flags_on_all(pe_working_set_t *data_set, uint64_t flag);
gboolean add_tag_ref(GHashTable * tags, const char * tag_name, const char * obj_ref);
//! \deprecated This function will be removed in a future release
void print_rscs_brief(GList *rsc_list, const char * pre_text, long options,
void * print_data, gboolean print_all);
int pe__rscs_brief_output(pcmk__output_t *out, GList *rsc_list, unsigned int options);
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);
//! \deprecated This function will be removed in a future release
void common_print(pe_resource_t *rsc, const char *pre_text, const char *name,
const pe_node_t *node, long options, void *print_data);
int pe__common_output_text(pcmk__output_t *out, const pe_resource_t *rsc,
const char *name, const pe_node_t *node,
unsigned int options);
int pe__common_output_html(pcmk__output_t *out, const pe_resource_t *rsc,
const char *name, const pe_node_t *node,
unsigned int options);
//! A single instance of a bundle
typedef struct {
int offset; //!< 0-origin index of this instance in bundle
char *ipaddr; //!< IP address associated with this instance
pe_node_t *node; //!< Node created for this instance
pe_resource_t *ip; //!< IP address resource for ipaddr
pe_resource_t *child; //!< Instance of bundled resource
pe_resource_t *container; //!< Container associated with this instance
pe_resource_t *remote; //!< Pacemaker Remote connection into container
} pe__bundle_replica_t;
GList *pe__bundle_containers(const pe_resource_t *bundle);
int pe__bundle_max(const pe_resource_t *rsc);
bool pe__node_is_bundle_instance(const pe_resource_t *bundle,
const pe_node_t *node);
pe_resource_t *pe__bundled_resource(const pe_resource_t *rsc);
const pe_resource_t *pe__get_rsc_in_container(const pe_resource_t *instance);
pe_resource_t *pe__first_container(const pe_resource_t *bundle);
void pe__foreach_bundle_replica(pe_resource_t *bundle,
bool (*fn)(pe__bundle_replica_t *, void *),
void *user_data);
void pe__foreach_const_bundle_replica(const pe_resource_t *bundle,
bool (*fn)(const pe__bundle_replica_t *,
void *),
void *user_data);
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,
pe_working_set_t *data_set,
xmlNode *xml, const char *field);
const char *pe__node_attribute_calculated(const pe_node_t *node,
const char *name,
const pe_resource_t *rsc,
enum pe__rsc_node node_type,
bool force_host);
const char *pe_node_attribute_raw(const pe_node_t *node, const char *name);
bool pe__is_universal_clone(const pe_resource_t *rsc,
const pe_working_set_t *data_set);
void pe__add_param_check(const 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*,
const xmlNode*,
enum pe_check_parameters));
void pe__free_param_checks(pe_working_set_t *data_set);
bool pe__shutdown_requested(const 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.
*
* \param[in,out] out Output object to register messages with
*/
void pe__register_messages(pcmk__output_t *out);
void pe__unpack_dataset_nvpairs(const xmlNode *xml_obj, const char *set_name,
const 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(const pe_resource_t *rsc);
pe_action_t *pe__clear_resource_history(pe_resource_t *rsc,
const pe_node_t *node,
pe_working_set_t *data_set);
GList *pe__rscs_with_tag(pe_working_set_t *data_set, const char *tag_name);
GList *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_only(const pe_resource_t *rsc, const pe_node_t *node);
bool pe__rsc_running_on_any(pe_resource_t *rsc, GList *node_list);
GList *pe__filter_rsc_list(GList *rscs, GList *filter);
GList * pe__build_node_name_list(pe_working_set_t *data_set, const char *s);
GList * pe__build_rsc_list(pe_working_set_t *data_set, const char *s);
bool pcmk__rsc_filtered_by_node(pe_resource_t *rsc, GList *only_node);
gboolean pe__bundle_is_filtered(const pe_resource_t *rsc, GList *only_rsc,
gboolean check_parent);
gboolean pe__clone_is_filtered(const pe_resource_t *rsc, GList *only_rsc,
gboolean check_parent);
gboolean pe__group_is_filtered(const pe_resource_t *rsc, GList *only_rsc,
gboolean check_parent);
gboolean pe__native_is_filtered(const pe_resource_t *rsc, GList *only_rsc,
gboolean check_parent);
xmlNode *pe__failed_probe_for_rsc(const pe_resource_t *rsc, const char *name);
const char *pe__clone_child_id(const pe_resource_t *rsc);
int pe__sum_node_health_scores(const pe_node_t *node, int base_health);
int pe__node_health(pe_node_t *node);
static inline enum pcmk__health_strategy
pe__health_strategy(pe_working_set_t *data_set)
{
return pcmk__parse_health_strategy(pe_pref(data_set->config_hash,
PCMK__OPT_NODE_HEALTH_STRATEGY));
}
static inline int
pe__health_score(const char *option, pe_working_set_t *data_set)
{
return char2score(pe_pref(data_set->config_hash, option));
}
/*!
* \internal
* \brief Return a string suitable for logging as a node name
*
* \param[in] node Node to return a node name string for
*
* \return Node name if available, otherwise node ID if available,
* otherwise "unspecified node" if node is NULL or "unidentified node"
* if node has neither a name nor ID.
*/
static inline const char *
pe__node_name(const pe_node_t *node)
{
if (node == NULL) {
return "unspecified node";
} else if (node->details->uname != NULL) {
return node->details->uname;
} else if (node->details->id != NULL) {
return node->details->id;
} else {
return "unidentified node";
}
}
/*!
* \internal
* \brief Check whether two node objects refer to the same node
*
* \param[in] node1 First node object to compare
* \param[in] node2 Second node object to compare
*
* \return true if \p node1 and \p node2 refer to the same node
*/
static inline bool
pe__same_node(const pe_node_t *node1, const pe_node_t *node2)
{
return (node1 != NULL) && (node2 != NULL)
&& (node1->details == node2->details);
}
/*!
* \internal
* \brief Get the operation key from an action history entry
*
* \param[in] xml Action history entry
*
* \return Entry's operation key
*/
static inline const char *
pe__xe_history_key(const xmlNode *xml)
{
if (xml == NULL) {
return NULL;
} else {
/* @COMPAT Pacemaker <= 1.1.5 did not add the key, and used the ID
* instead. Checking for that allows us to process old saved CIBs,
* including some regression tests.
*/
const char *key = crm_element_value(xml, XML_LRM_ATTR_TASK_KEY);
return pcmk__str_empty(key)? ID(xml) : key;
}
}
#endif
diff --git a/lib/common/operations.c b/lib/common/operations.c
index 342f3bd69e..67d28c1cd4 100644
--- a/lib/common/operations.c
+++ b/lib/common/operations.c
@@ -1,531 +1,531 @@
/*
* Copyright 2004-2023 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 <stdio.h>
#include <string.h>
#include <stdlib.h>
#include <sys/types.h>
#include <ctype.h>
#include <crm/crm.h>
#include <crm/lrmd.h>
#include <crm/msg_xml.h>
#include <crm/common/xml.h>
#include <crm/common/xml_internal.h>
#include <crm/common/util.h>
/*!
* \brief Generate an operation key (RESOURCE_ACTION_INTERVAL)
*
* \param[in] rsc_id ID of resource being operated on
* \param[in] op_type Operation name
* \param[in] interval_ms Operation interval
*
* \return Newly allocated memory containing operation key as string
*
* \note This function asserts on errors, so it will never return NULL.
* The caller is responsible for freeing the result with free().
*/
char *
pcmk__op_key(const char *rsc_id, const char *op_type, guint interval_ms)
{
CRM_ASSERT(rsc_id != NULL);
CRM_ASSERT(op_type != NULL);
return crm_strdup_printf(PCMK__OP_FMT, rsc_id, op_type, interval_ms);
}
static inline gboolean
convert_interval(const char *s, guint *interval_ms)
{
unsigned long l;
errno = 0;
l = strtoul(s, NULL, 10);
if (errno != 0) {
return FALSE;
}
*interval_ms = (guint) l;
return TRUE;
}
/*!
* \internal
* \brief Check for underbar-separated substring match
*
* \param[in] key Overall string being checked
* \param[in] position Match before underbar at this \p key index
* \param[in] matches Substrings to match (may contain underbars)
*
* \return \p key index of underbar before any matching substring,
* or 0 if none
*/
static size_t
match_before(const char *key, size_t position, const char **matches)
{
for (int i = 0; matches[i] != NULL; ++i) {
const size_t match_len = strlen(matches[i]);
// Must have at least X_MATCH before position
if (position > (match_len + 1)) {
const size_t possible = position - match_len - 1;
if ((key[possible] == '_')
&& (strncmp(key + possible + 1, matches[i], match_len) == 0)) {
return possible;
}
}
}
return 0;
}
gboolean
parse_op_key(const char *key, char **rsc_id, char **op_type, guint *interval_ms)
{
guint local_interval_ms = 0;
const size_t key_len = (key == NULL)? 0 : strlen(key);
// Operation keys must be formatted as RSC_ACTION_INTERVAL
size_t action_underbar = 0; // Index in key of underbar before ACTION
size_t interval_underbar = 0; // Index in key of underbar before INTERVAL
size_t possible = 0;
/* Underbar was a poor choice of separator since both RSC and ACTION can
* contain underbars. Here, list action names and name prefixes that can.
*/
const char *actions_with_underbars[] = {
CRMD_ACTION_MIGRATED,
CRMD_ACTION_MIGRATE,
NULL
};
const char *action_prefixes_with_underbars[] = {
"pre_" PCMK_ACTION_NOTIFY,
"post_" PCMK_ACTION_NOTIFY,
"confirmed-pre_" PCMK_ACTION_NOTIFY,
"confirmed-post_" PCMK_ACTION_NOTIFY,
NULL,
};
// Initialize output variables in case of early return
if (rsc_id) {
*rsc_id = NULL;
}
if (op_type) {
*op_type = NULL;
}
if (interval_ms) {
*interval_ms = 0;
}
// RSC_ACTION_INTERVAL implies a minimum of 5 characters
if (key_len < 5) {
return FALSE;
}
// Find, parse, and validate interval
interval_underbar = key_len - 2;
while ((interval_underbar > 2) && (key[interval_underbar] != '_')) {
--interval_underbar;
}
if ((interval_underbar == 2)
|| !convert_interval(key + interval_underbar + 1, &local_interval_ms)) {
return FALSE;
}
// Find the base (OCF) action name, disregarding prefixes
action_underbar = match_before(key, interval_underbar,
actions_with_underbars);
if (action_underbar == 0) {
action_underbar = interval_underbar - 2;
while ((action_underbar > 0) && (key[action_underbar] != '_')) {
--action_underbar;
}
if (action_underbar == 0) {
return FALSE;
}
}
possible = match_before(key, action_underbar,
action_prefixes_with_underbars);
if (possible != 0) {
action_underbar = possible;
}
// Set output variables
if (rsc_id != NULL) {
*rsc_id = strndup(key, action_underbar);
CRM_ASSERT(*rsc_id != NULL);
}
if (op_type != NULL) {
*op_type = strndup(key + action_underbar + 1,
interval_underbar - action_underbar - 1);
CRM_ASSERT(*op_type != NULL);
}
if (interval_ms != NULL) {
*interval_ms = local_interval_ms;
}
return TRUE;
}
char *
pcmk__notify_key(const char *rsc_id, const char *notify_type,
const char *op_type)
{
CRM_CHECK(rsc_id != NULL, return NULL);
CRM_CHECK(op_type != NULL, return NULL);
CRM_CHECK(notify_type != NULL, return NULL);
return crm_strdup_printf("%s_%s_notify_%s_0",
rsc_id, notify_type, op_type);
}
/*!
* \brief Parse a transition magic string into its constituent parts
*
* \param[in] magic Magic string to parse (must be non-NULL)
* \param[out] uuid If non-NULL, where to store copy of parsed UUID
* \param[out] transition_id If non-NULL, where to store parsed transition ID
* \param[out] action_id If non-NULL, where to store parsed action ID
* \param[out] op_status If non-NULL, where to store parsed result status
* \param[out] op_rc If non-NULL, where to store parsed actual rc
* \param[out] target_rc If non-NULL, where to stored parsed target rc
*
* \return TRUE if key was valid, FALSE otherwise
* \note If uuid is supplied and this returns TRUE, the caller is responsible
* for freeing the memory for *uuid using free().
*/
gboolean
decode_transition_magic(const char *magic, char **uuid, int *transition_id, int *action_id,
int *op_status, int *op_rc, int *target_rc)
{
int res = 0;
char *key = NULL;
gboolean result = TRUE;
int local_op_status = -1;
int local_op_rc = -1;
CRM_CHECK(magic != NULL, return FALSE);
#ifdef HAVE_SSCANF_M
res = sscanf(magic, "%d:%d;%ms", &local_op_status, &local_op_rc, &key);
#else
key = calloc(1, strlen(magic) - 3); // magic must have >=4 other characters
CRM_ASSERT(key);
res = sscanf(magic, "%d:%d;%s", &local_op_status, &local_op_rc, key);
#endif
if (res == EOF) {
crm_err("Could not decode transition information '%s': %s",
magic, pcmk_rc_str(errno));
result = FALSE;
} else if (res < 3) {
crm_warn("Transition information '%s' incomplete (%d of 3 expected items)",
magic, res);
result = FALSE;
} else {
if (op_status) {
*op_status = local_op_status;
}
if (op_rc) {
*op_rc = local_op_rc;
}
result = decode_transition_key(key, uuid, transition_id, action_id,
target_rc);
}
free(key);
return result;
}
char *
pcmk__transition_key(int transition_id, int action_id, int target_rc,
const char *node)
{
CRM_CHECK(node != NULL, return NULL);
return crm_strdup_printf("%d:%d:%d:%-*s",
action_id, transition_id, target_rc, 36, node);
}
/*!
* \brief Parse a transition key into its constituent parts
*
* \param[in] key Transition key to parse (must be non-NULL)
* \param[out] uuid If non-NULL, where to store copy of parsed UUID
* \param[out] transition_id If non-NULL, where to store parsed transition ID
* \param[out] action_id If non-NULL, where to store parsed action ID
* \param[out] target_rc If non-NULL, where to stored parsed target rc
*
* \return TRUE if key was valid, FALSE otherwise
* \note If uuid is supplied and this returns TRUE, the caller is responsible
* for freeing the memory for *uuid using free().
*/
gboolean
decode_transition_key(const char *key, char **uuid, int *transition_id, int *action_id,
int *target_rc)
{
int local_transition_id = -1;
int local_action_id = -1;
int local_target_rc = -1;
char local_uuid[37] = { '\0' };
// Initialize any supplied output arguments
if (uuid) {
*uuid = NULL;
}
if (transition_id) {
*transition_id = -1;
}
if (action_id) {
*action_id = -1;
}
if (target_rc) {
*target_rc = -1;
}
CRM_CHECK(key != NULL, return FALSE);
if (sscanf(key, "%d:%d:%d:%36s", &local_action_id, &local_transition_id,
&local_target_rc, local_uuid) != 4) {
crm_err("Invalid transition key '%s'", key);
return FALSE;
}
if (strlen(local_uuid) != 36) {
crm_warn("Invalid UUID '%s' in transition key '%s'", local_uuid, key);
}
if (uuid) {
*uuid = strdup(local_uuid);
CRM_ASSERT(*uuid);
}
if (transition_id) {
*transition_id = local_transition_id;
}
if (action_id) {
*action_id = local_action_id;
}
if (target_rc) {
*target_rc = local_target_rc;
}
return TRUE;
}
// Return true if a is an attribute that should be filtered
static bool
should_filter_for_digest(xmlAttrPtr a, void *user_data)
{
if (strncmp((const char *) a->name, CRM_META "_",
sizeof(CRM_META " ") - 1) == 0) {
return true;
}
return pcmk__str_any_of((const char *) a->name,
XML_ATTR_ID,
XML_ATTR_CRM_VERSION,
XML_LRM_ATTR_OP_DIGEST,
XML_LRM_ATTR_TARGET,
XML_LRM_ATTR_TARGET_UUID,
"pcmk_external_ip",
NULL);
}
/*!
* \internal
* \brief Remove XML attributes not needed for operation digest
*
* \param[in,out] param_set XML with operation parameters
*/
void
pcmk__filter_op_for_digest(xmlNode *param_set)
{
char *key = NULL;
char *timeout = NULL;
guint interval_ms = 0;
if (param_set == NULL) {
return;
}
/* Timeout is useful for recurring operation digests, so grab it before
* removing meta-attributes
*/
key = crm_meta_name(XML_LRM_ATTR_INTERVAL_MS);
if (crm_element_value_ms(param_set, key, &interval_ms) != pcmk_ok) {
interval_ms = 0;
}
free(key);
key = NULL;
if (interval_ms != 0) {
key = crm_meta_name(XML_ATTR_TIMEOUT);
timeout = crm_element_value_copy(param_set, key);
}
// Remove all CRM_meta_* attributes and certain other attributes
pcmk__xe_remove_matching_attrs(param_set, should_filter_for_digest, NULL);
// Add timeout back for recurring operation digests
if (timeout != NULL) {
crm_xml_add(param_set, key, timeout);
}
free(timeout);
free(key);
}
int
rsc_op_expected_rc(const lrmd_event_data_t *op)
{
int rc = 0;
if (op && op->user_data) {
decode_transition_key(op->user_data, NULL, NULL, NULL, &rc);
}
return rc;
}
gboolean
did_rsc_op_fail(lrmd_event_data_t * op, int target_rc)
{
switch (op->op_status) {
case PCMK_EXEC_CANCELLED:
case PCMK_EXEC_PENDING:
return FALSE;
case PCMK_EXEC_NOT_SUPPORTED:
case PCMK_EXEC_TIMEOUT:
case PCMK_EXEC_ERROR:
case PCMK_EXEC_NOT_CONNECTED:
case PCMK_EXEC_NO_FENCE_DEVICE:
case PCMK_EXEC_NO_SECRETS:
case PCMK_EXEC_INVALID:
return TRUE;
default:
if (target_rc != op->rc) {
return TRUE;
}
}
return FALSE;
}
/*!
* \brief Create a CIB XML element for an operation
*
* \param[in,out] parent If not NULL, make new XML node a child of this
* \param[in] prefix Generate an ID using this prefix
* \param[in] task Operation task to set
* \param[in] interval_spec Operation interval to set
* \param[in] timeout If not NULL, operation timeout to set
*
* \return New XML object on success, NULL otherwise
*/
xmlNode *
crm_create_op_xml(xmlNode *parent, const char *prefix, const char *task,
const char *interval_spec, const char *timeout)
{
xmlNode *xml_op;
CRM_CHECK(prefix && task && interval_spec, return NULL);
xml_op = create_xml_node(parent, XML_ATTR_OP);
crm_xml_set_id(xml_op, "%s-%s-%s", prefix, task, interval_spec);
crm_xml_add(xml_op, XML_LRM_ATTR_INTERVAL, interval_spec);
crm_xml_add(xml_op, "name", task);
if (timeout) {
crm_xml_add(xml_op, XML_ATTR_TIMEOUT, timeout);
}
return xml_op;
}
/*!
* \brief Check whether an operation requires resource agent meta-data
*
* \param[in] rsc_class Resource agent class (or NULL to skip class check)
* \param[in] op Operation action (or NULL to skip op check)
*
* \return true if operation needs meta-data, false otherwise
* \note At least one of rsc_class and op must be specified.
*/
bool
crm_op_needs_metadata(const char *rsc_class, const char *op)
{
/* Agent metadata is used to determine whether an agent reload is possible,
* so if this op is not relevant to that feature, we don't need metadata.
*/
CRM_CHECK((rsc_class != NULL) || (op != NULL), return false);
if ((rsc_class != NULL)
&& !pcmk_is_set(pcmk_get_ra_caps(rsc_class), pcmk_ra_cap_params)) {
// Metadata is needed only for resource classes that use parameters
return false;
}
if (op == NULL) {
return true;
}
// Metadata is needed only for these actions
return pcmk__str_any_of(op, PCMK_ACTION_START, PCMK_ACTION_MONITOR,
PCMK_ACTION_PROMOTE, PCMK_ACTION_DEMOTE,
- PCMK_ACTION_RELOAD, CRMD_ACTION_RELOAD_AGENT,
+ PCMK_ACTION_RELOAD, PCMK_ACTION_RELOAD_AGENT,
CRMD_ACTION_MIGRATE, CRMD_ACTION_MIGRATED,
PCMK_ACTION_NOTIFY, NULL);
}
/*!
* \internal
* \brief Check whether an action name is for a fencing action
*
* \param[in] action Action name to check
*
* \return true if \p action is "off", "reboot", or "poweroff", otherwise false
*/
bool
pcmk__is_fencing_action(const char *action)
{
return pcmk__str_any_of(action, "off", "reboot", "poweroff", NULL);
}
bool
pcmk_is_probe(const char *task, guint interval)
{
if (task == NULL) {
return false;
}
return (interval == 0)
&& pcmk__str_eq(task, PCMK_ACTION_MONITOR, pcmk__str_none);
}
bool
pcmk_xe_is_probe(const xmlNode *xml_op)
{
const char *task = crm_element_value(xml_op, XML_LRM_ATTR_TASK);
const char *interval_ms_s = crm_element_value(xml_op, XML_LRM_ATTR_INTERVAL_MS);
int interval_ms;
pcmk__scan_min_int(interval_ms_s, &interval_ms, 0);
return pcmk_is_probe(task, interval_ms);
}
bool
pcmk_xe_mask_probe_failure(const xmlNode *xml_op)
{
int status = PCMK_EXEC_UNKNOWN;
int rc = PCMK_OCF_OK;
if (!pcmk_xe_is_probe(xml_op)) {
return false;
}
crm_element_value_int(xml_op, XML_LRM_ATTR_OPSTATUS, &status);
crm_element_value_int(xml_op, XML_LRM_ATTR_RC, &rc);
return rc == PCMK_OCF_NOT_INSTALLED || rc == PCMK_OCF_INVALID_PARAM ||
status == PCMK_EXEC_NOT_INSTALLED;
}
diff --git a/lib/pacemaker/pcmk_sched_actions.c b/lib/pacemaker/pcmk_sched_actions.c
index 978af4a05f..da787c16e6 100644
--- a/lib/pacemaker/pcmk_sched_actions.c
+++ b/lib/pacemaker/pcmk_sched_actions.c
@@ -1,1929 +1,1929 @@
/*
* Copyright 2004-2023 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU General Public License version 2
* or later (GPLv2+) WITHOUT ANY WARRANTY.
*/
#include <crm_internal.h>
#include <stdio.h>
#include <sys/param.h>
#include <glib.h>
#include <crm/lrmd_internal.h>
#include <pacemaker-internal.h>
#include "libpacemaker_private.h"
/*!
* \internal
* \brief Get the action flags relevant to ordering constraints
*
* \param[in,out] action Action to check
* \param[in] node Node that *other* action in the ordering is on
* (used only for clone resource actions)
*
* \return Action flags that should be used for orderings
*/
static uint32_t
action_flags_for_ordering(pe_action_t *action, const pe_node_t *node)
{
bool runnable = false;
uint32_t flags;
// For non-resource actions, return the action flags
if (action->rsc == NULL) {
return action->flags;
}
/* For non-clone resources, or a clone action not assigned to a node,
* return the flags as determined by the resource method without a node
* specified.
*/
flags = action->rsc->cmds->action_flags(action, NULL);
if ((node == NULL) || !pe_rsc_is_clone(action->rsc)) {
return flags;
}
/* Otherwise (i.e., for clone resource actions on a specific node), first
* remember whether the non-node-specific action is runnable.
*/
runnable = pcmk_is_set(flags, pe_action_runnable);
// Then recheck the resource method with the node
flags = action->rsc->cmds->action_flags(action, node);
/* For clones in ordering constraints, the node-specific "runnable" doesn't
* matter, just the non-node-specific setting (i.e., is the action runnable
* anywhere).
*
* This applies only to runnable, and only for ordering constraints. This
* function shouldn't be used for other types of constraints without
* changes. Not very satisfying, but it's logical and appears to work well.
*/
if (runnable && !pcmk_is_set(flags, pe_action_runnable)) {
pe__set_raw_action_flags(flags, action->rsc->id,
pe_action_runnable);
}
return flags;
}
/*!
* \internal
* \brief Get action UUID that should be used with a resource ordering
*
* When an action is ordered relative to an action for a collective resource
* (clone, group, or bundle), it actually needs to be ordered after all
* instances of the collective have completed the relevant action (for example,
* given "start CLONE then start RSC", RSC must wait until all instances of
* CLONE have started). Given the UUID and resource of the first action in an
* ordering, this returns the UUID of the action that should actually be used
* for ordering (for example, "CLONE_started_0" instead of "CLONE_start_0").
*
* \param[in] first_uuid UUID of first action in ordering
* \param[in] first_rsc Resource of first action in ordering
*
* \return Newly allocated copy of UUID to use with ordering
* \note It is the caller's responsibility to free the return value.
*/
static char *
action_uuid_for_ordering(const char *first_uuid, const pe_resource_t *first_rsc)
{
guint interval_ms = 0;
char *uuid = NULL;
char *rid = NULL;
char *first_task_str = NULL;
enum action_tasks first_task = no_action;
enum action_tasks remapped_task = no_action;
// Only non-notify actions for collective resources need remapping
if ((strstr(first_uuid, PCMK_ACTION_NOTIFY) != NULL)
|| (first_rsc->variant < pe_group)) {
goto done;
}
// Only non-recurring actions need remapping
CRM_ASSERT(parse_op_key(first_uuid, &rid, &first_task_str, &interval_ms));
if (interval_ms > 0) {
goto done;
}
first_task = text2task(first_task_str);
switch (first_task) {
case stop_rsc:
case start_rsc:
case action_notify:
case action_promote:
case action_demote:
remapped_task = first_task + 1;
break;
case stopped_rsc:
case started_rsc:
case action_notified:
case action_promoted:
case action_demoted:
remapped_task = first_task;
break;
case monitor_rsc:
case shutdown_crm:
case stonith_node:
break;
default:
crm_err("Unknown action '%s' in ordering", first_task_str);
break;
}
if (remapped_task != no_action) {
/* If a (clone) resource has notifications enabled, we want to order
* relative to when all notifications have been sent for the remapped
* task. Only outermost resources or those in bundles have
* notifications.
*/
if (pcmk_is_set(first_rsc->flags, pe_rsc_notify)
&& ((first_rsc->parent == NULL)
|| (pe_rsc_is_clone(first_rsc)
&& (first_rsc->parent->variant == pe_container)))) {
uuid = pcmk__notify_key(rid, "confirmed-post",
task2text(remapped_task));
} else {
uuid = pcmk__op_key(rid, task2text(remapped_task), 0);
}
pe_rsc_trace(first_rsc,
"Remapped action UUID %s to %s for ordering purposes",
first_uuid, uuid);
}
done:
if (uuid == NULL) {
uuid = strdup(first_uuid);
CRM_ASSERT(uuid != NULL);
}
free(first_task_str);
free(rid);
return uuid;
}
/*!
* \internal
* \brief Get actual action that should be used with an ordering
*
* When an action is ordered relative to an action for a collective resource
* (clone, group, or bundle), it actually needs to be ordered after all
* instances of the collective have completed the relevant action (for example,
* given "start CLONE then start RSC", RSC must wait until all instances of
* CLONE have started). Given the first action in an ordering, this returns the
* the action that should actually be used for ordering (for example, the
* started action instead of the start action).
*
* \param[in] action First action in an ordering
*
* \return Actual action that should be used for the ordering
*/
static pe_action_t *
action_for_ordering(pe_action_t *action)
{
pe_action_t *result = action;
pe_resource_t *rsc = action->rsc;
if ((rsc != NULL) && (rsc->variant >= pe_group) && (action->uuid != NULL)) {
char *uuid = action_uuid_for_ordering(action->uuid, rsc);
result = find_first_action(rsc->actions, uuid, NULL, NULL);
if (result == NULL) {
crm_warn("Not remapping %s to %s because %s does not have "
"remapped action", action->uuid, uuid, rsc->id);
result = action;
}
free(uuid);
}
return result;
}
/*!
* \internal
* \brief Wrapper for update_ordered_actions() method for readability
*
* \param[in,out] rsc Resource to call method for
* \param[in,out] first 'First' action in an ordering
* \param[in,out] then 'Then' action in an ordering
* \param[in] node If not NULL, limit scope of ordering to this
* node (only used when interleaving instances)
* \param[in] flags Action flags for \p first for ordering purposes
* \param[in] filter Action flags to limit scope of certain updates
* (may include pe_action_optional to affect only
* mandatory actions, and pe_action_runnable to
* affect only runnable actions)
* \param[in] type Group of enum pe_ordering flags to apply
* \param[in,out] data_set Cluster working set
*
* \return Group of enum pcmk__updated flags indicating what was updated
*/
static inline uint32_t
update(pe_resource_t *rsc, pe_action_t *first, pe_action_t *then,
const pe_node_t *node, uint32_t flags, uint32_t filter, uint32_t type,
pe_working_set_t *data_set)
{
return rsc->cmds->update_ordered_actions(first, then, node, flags, filter,
type, data_set);
}
/*!
* \internal
* \brief Update flags for ordering's actions appropriately for ordering's flags
*
* \param[in,out] first First action in an ordering
* \param[in,out] then Then action in an ordering
* \param[in] first_flags Action flags for \p first for ordering purposes
* \param[in] then_flags Action flags for \p then for ordering purposes
* \param[in,out] order Action wrapper for \p first in ordering
* \param[in,out] data_set Cluster working set
*
* \return Group of enum pcmk__updated flags
*/
static uint32_t
update_action_for_ordering_flags(pe_action_t *first, pe_action_t *then,
uint32_t first_flags, uint32_t then_flags,
pe_action_wrapper_t *order,
pe_working_set_t *data_set)
{
uint32_t changed = pcmk__updated_none;
/* The node will only be used for clones. If interleaved, node will be NULL,
* otherwise the ordering scope will be limited to the node. Normally, the
* whole 'then' clone should restart if 'first' is restarted, so then->node
* is needed.
*/
pe_node_t *node = then->node;
if (pcmk_is_set(order->type, pe_order_implies_then_on_node)) {
/* For unfencing, only instances of 'then' on the same node as 'first'
* (the unfencing operation) should restart, so reset node to
* first->node, at which point this case is handled like a normal
* pe_order_implies_then.
*/
pe__clear_order_flags(order->type, pe_order_implies_then_on_node);
pe__set_order_flags(order->type, pe_order_implies_then);
node = first->node;
pe_rsc_trace(then->rsc,
"%s then %s: mapped pe_order_implies_then_on_node to "
"pe_order_implies_then on %s",
first->uuid, then->uuid, pe__node_name(node));
}
if (pcmk_is_set(order->type, pe_order_implies_then)) {
if (then->rsc != NULL) {
changed |= update(then->rsc, first, then, node,
first_flags & pe_action_optional,
pe_action_optional, pe_order_implies_then,
data_set);
} else if (!pcmk_is_set(first_flags, pe_action_optional)
&& pcmk_is_set(then->flags, pe_action_optional)) {
pe__clear_action_flags(then, pe_action_optional);
pcmk__set_updated_flags(changed, first, pcmk__updated_then);
}
pe_rsc_trace(then->rsc, "%s then %s: %s after pe_order_implies_then",
first->uuid, then->uuid,
(changed? "changed" : "unchanged"));
}
if (pcmk_is_set(order->type, pe_order_restart) && (then->rsc != NULL)) {
enum pe_action_flags restart = pe_action_optional|pe_action_runnable;
changed |= update(then->rsc, first, then, node, first_flags, restart,
pe_order_restart, data_set);
pe_rsc_trace(then->rsc, "%s then %s: %s after pe_order_restart",
first->uuid, then->uuid,
(changed? "changed" : "unchanged"));
}
if (pcmk_is_set(order->type, pe_order_implies_first)) {
if (first->rsc != NULL) {
changed |= update(first->rsc, first, then, node, first_flags,
pe_action_optional, pe_order_implies_first,
data_set);
} else if (!pcmk_is_set(first_flags, pe_action_optional)
&& pcmk_is_set(first->flags, pe_action_runnable)) {
pe__clear_action_flags(first, pe_action_runnable);
pcmk__set_updated_flags(changed, first, pcmk__updated_first);
}
pe_rsc_trace(then->rsc, "%s then %s: %s after pe_order_implies_first",
first->uuid, then->uuid,
(changed? "changed" : "unchanged"));
}
if (pcmk_is_set(order->type, pe_order_promoted_implies_first)) {
if (then->rsc != NULL) {
changed |= update(then->rsc, first, then, node,
first_flags & pe_action_optional,
pe_action_optional,
pe_order_promoted_implies_first, data_set);
}
pe_rsc_trace(then->rsc,
"%s then %s: %s after pe_order_promoted_implies_first",
first->uuid, then->uuid,
(changed? "changed" : "unchanged"));
}
if (pcmk_is_set(order->type, pe_order_one_or_more)) {
if (then->rsc != NULL) {
changed |= update(then->rsc, first, then, node, first_flags,
pe_action_runnable, pe_order_one_or_more,
data_set);
} else if (pcmk_is_set(first_flags, pe_action_runnable)) {
// We have another runnable instance of "first"
then->runnable_before++;
/* Mark "then" as runnable if it requires a certain number of
* "before" instances to be runnable, and they now are.
*/
if ((then->runnable_before >= then->required_runnable_before)
&& !pcmk_is_set(then->flags, pe_action_runnable)) {
pe__set_action_flags(then, pe_action_runnable);
pcmk__set_updated_flags(changed, first, pcmk__updated_then);
}
}
pe_rsc_trace(then->rsc, "%s then %s: %s after pe_order_one_or_more",
first->uuid, then->uuid,
(changed? "changed" : "unchanged"));
}
if (pcmk_is_set(order->type, pe_order_probe) && (then->rsc != NULL)) {
if (!pcmk_is_set(first_flags, pe_action_runnable)
&& (first->rsc->running_on != NULL)) {
pe_rsc_trace(then->rsc,
"%s then %s: ignoring because first is stopping",
first->uuid, then->uuid);
order->type = pe_order_none;
} else {
changed |= update(then->rsc, first, then, node, first_flags,
pe_action_runnable, pe_order_runnable_left,
data_set);
}
pe_rsc_trace(then->rsc, "%s then %s: %s after pe_order_probe",
first->uuid, then->uuid,
(changed? "changed" : "unchanged"));
}
if (pcmk_is_set(order->type, pe_order_runnable_left)) {
if (then->rsc != NULL) {
changed |= update(then->rsc, first, then, node, first_flags,
pe_action_runnable, pe_order_runnable_left,
data_set);
} else if (!pcmk_is_set(first_flags, pe_action_runnable)
&& pcmk_is_set(then->flags, pe_action_runnable)) {
pe__clear_action_flags(then, pe_action_runnable);
pcmk__set_updated_flags(changed, first, pcmk__updated_then);
}
pe_rsc_trace(then->rsc, "%s then %s: %s after pe_order_runnable_left",
first->uuid, then->uuid,
(changed? "changed" : "unchanged"));
}
if (pcmk_is_set(order->type, pe_order_implies_first_migratable)) {
if (then->rsc != NULL) {
changed |= update(then->rsc, first, then, node, first_flags,
pe_action_optional,
pe_order_implies_first_migratable, data_set);
}
pe_rsc_trace(then->rsc, "%s then %s: %s after "
"pe_order_implies_first_migratable",
first->uuid, then->uuid,
(changed? "changed" : "unchanged"));
}
if (pcmk_is_set(order->type, pe_order_pseudo_left)) {
if (then->rsc != NULL) {
changed |= update(then->rsc, first, then, node, first_flags,
pe_action_optional, pe_order_pseudo_left,
data_set);
}
pe_rsc_trace(then->rsc, "%s then %s: %s after pe_order_pseudo_left",
first->uuid, then->uuid,
(changed? "changed" : "unchanged"));
}
if (pcmk_is_set(order->type, pe_order_optional)) {
if (then->rsc != NULL) {
changed |= update(then->rsc, first, then, node, first_flags,
pe_action_runnable, pe_order_optional, data_set);
}
pe_rsc_trace(then->rsc, "%s then %s: %s after pe_order_optional",
first->uuid, then->uuid,
(changed? "changed" : "unchanged"));
}
if (pcmk_is_set(order->type, pe_order_asymmetrical)) {
if (then->rsc != NULL) {
changed |= update(then->rsc, first, then, node, first_flags,
pe_action_runnable, pe_order_asymmetrical,
data_set);
}
pe_rsc_trace(then->rsc, "%s then %s: %s after pe_order_asymmetrical",
first->uuid, then->uuid,
(changed? "changed" : "unchanged"));
}
if (pcmk_is_set(first->flags, pe_action_runnable)
&& pcmk_is_set(order->type, pe_order_implies_then_printed)
&& !pcmk_is_set(first_flags, pe_action_optional)) {
pe_rsc_trace(then->rsc, "%s will be in graph because %s is required",
then->uuid, first->uuid);
pe__set_action_flags(then, pe_action_print_always);
// Don't bother marking 'then' as changed just for this
}
if (pcmk_is_set(order->type, pe_order_implies_first_printed)
&& !pcmk_is_set(then_flags, pe_action_optional)) {
pe_rsc_trace(then->rsc, "%s will be in graph because %s is required",
first->uuid, then->uuid);
pe__set_action_flags(first, pe_action_print_always);
// Don't bother marking 'first' as changed just for this
}
if (pcmk_any_flags_set(order->type, pe_order_implies_then
|pe_order_implies_first
|pe_order_restart)
&& (first->rsc != NULL)
&& !pcmk_is_set(first->rsc->flags, pe_rsc_managed)
&& pcmk_is_set(first->rsc->flags, pe_rsc_block)
&& !pcmk_is_set(first->flags, pe_action_runnable)
&& pcmk__str_eq(first->task, PCMK_ACTION_STOP, pcmk__str_none)) {
if (pcmk_is_set(then->flags, pe_action_runnable)) {
pe__clear_action_flags(then, pe_action_runnable);
pcmk__set_updated_flags(changed, first, pcmk__updated_then);
}
pe_rsc_trace(then->rsc, "%s then %s: %s after checking whether first "
"is blocked, unmanaged, unrunnable stop",
first->uuid, then->uuid,
(changed? "changed" : "unchanged"));
}
return changed;
}
// Convenience macros for logging action properties
#define action_type_str(flags) \
(pcmk_is_set((flags), pe_action_pseudo)? "pseudo-action" : "action")
#define action_optional_str(flags) \
(pcmk_is_set((flags), pe_action_optional)? "optional" : "required")
#define action_runnable_str(flags) \
(pcmk_is_set((flags), pe_action_runnable)? "runnable" : "unrunnable")
#define action_node_str(a) \
(((a)->node == NULL)? "no node" : (a)->node->details->uname)
/*!
* \internal
* \brief Update an action's flags for all orderings where it is "then"
*
* \param[in,out] then Action to update
* \param[in,out] data_set Cluster working set
*/
void
pcmk__update_action_for_orderings(pe_action_t *then, pe_working_set_t *data_set)
{
GList *lpc = NULL;
uint32_t changed = pcmk__updated_none;
int last_flags = then->flags;
pe_rsc_trace(then->rsc, "Updating %s %s (%s %s) on %s",
action_type_str(then->flags), then->uuid,
action_optional_str(then->flags),
action_runnable_str(then->flags), action_node_str(then));
if (pcmk_is_set(then->flags, pe_action_requires_any)) {
/* Initialize current known "runnable before" actions. As
* update_action_for_ordering_flags() is called for each of then's
* before actions, this number will increment as runnable 'first'
* actions are encountered.
*/
then->runnable_before = 0;
if (then->required_runnable_before == 0) {
/* @COMPAT This ordering constraint uses the deprecated
* "require-all=false" attribute. Treat it like "clone-min=1".
*/
then->required_runnable_before = 1;
}
/* The pe_order_one_or_more clause of update_action_for_ordering_flags()
* (called below) will reset runnable if appropriate.
*/
pe__clear_action_flags(then, pe_action_runnable);
}
for (lpc = then->actions_before; lpc != NULL; lpc = lpc->next) {
pe_action_wrapper_t *other = (pe_action_wrapper_t *) lpc->data;
pe_action_t *first = other->action;
pe_node_t *then_node = then->node;
pe_node_t *first_node = first->node;
if ((first->rsc != NULL)
&& (first->rsc->variant == pe_group)
&& pcmk__str_eq(first->task, PCMK_ACTION_START, pcmk__str_none)) {
first_node = first->rsc->fns->location(first->rsc, NULL, FALSE);
if (first_node != NULL) {
pe_rsc_trace(first->rsc, "Found %s for 'first' %s",
pe__node_name(first_node), first->uuid);
}
}
if ((then->rsc != NULL)
&& (then->rsc->variant == pe_group)
&& pcmk__str_eq(then->task, PCMK_ACTION_START, pcmk__str_none)) {
then_node = then->rsc->fns->location(then->rsc, NULL, FALSE);
if (then_node != NULL) {
pe_rsc_trace(then->rsc, "Found %s for 'then' %s",
pe__node_name(then_node), then->uuid);
}
}
// Disable constraint if it only applies when on same node, but isn't
if (pcmk_is_set(other->type, pe_order_same_node)
&& (first_node != NULL) && (then_node != NULL)
&& !pe__same_node(first_node, then_node)) {
pe_rsc_trace(then->rsc,
"Disabled ordering %s on %s then %s on %s: "
"not same node",
other->action->uuid, pe__node_name(first_node),
then->uuid, pe__node_name(then_node));
other->type = pe_order_none;
continue;
}
pcmk__clear_updated_flags(changed, then, pcmk__updated_first);
if ((first->rsc != NULL)
&& pcmk_is_set(other->type, pe_order_then_cancels_first)
&& !pcmk_is_set(then->flags, pe_action_optional)) {
/* 'then' is required, so we must abandon 'first'
* (e.g. a required stop cancels any agent reload).
*/
pe__set_action_flags(other->action, pe_action_optional);
- if (!strcmp(first->task, CRMD_ACTION_RELOAD_AGENT)) {
+ if (!strcmp(first->task, PCMK_ACTION_RELOAD_AGENT)) {
pe__clear_resource_flags(first->rsc, pe_rsc_reload);
}
}
if ((first->rsc != NULL) && (then->rsc != NULL)
&& (first->rsc != then->rsc) && !is_parent(then->rsc, first->rsc)) {
first = action_for_ordering(first);
}
if (first != other->action) {
pe_rsc_trace(then->rsc, "Ordering %s after %s instead of %s",
then->uuid, first->uuid, other->action->uuid);
}
pe_rsc_trace(then->rsc,
"%s (%#.6x) then %s (%#.6x): type=%#.6x node=%s",
first->uuid, first->flags, then->uuid, then->flags,
other->type, action_node_str(first));
if (first == other->action) {
/* 'first' was not remapped (e.g. from 'start' to 'running'), which
* could mean it is a non-resource action, a primitive resource
* action, or already expanded.
*/
uint32_t first_flags, then_flags;
first_flags = action_flags_for_ordering(first, then_node);
then_flags = action_flags_for_ordering(then, first_node);
changed |= update_action_for_ordering_flags(first, then,
first_flags, then_flags,
other, data_set);
/* 'first' was for a complex resource (clone, group, etc),
* create a new dependency if necessary
*/
} else if (order_actions(first, then, other->type)) {
/* This was the first time 'first' and 'then' were associated,
* start again to get the new actions_before list
*/
pcmk__set_updated_flags(changed, then, pcmk__updated_then);
pe_rsc_trace(then->rsc,
"Disabled ordering %s then %s in favor of %s then %s",
other->action->uuid, then->uuid, first->uuid,
then->uuid);
other->type = pe_order_none;
}
if (pcmk_is_set(changed, pcmk__updated_first)) {
crm_trace("Re-processing %s and its 'after' actions "
"because it changed", first->uuid);
for (GList *lpc2 = first->actions_after; lpc2 != NULL;
lpc2 = lpc2->next) {
pe_action_wrapper_t *other = (pe_action_wrapper_t *) lpc2->data;
pcmk__update_action_for_orderings(other->action, data_set);
}
pcmk__update_action_for_orderings(first, data_set);
}
}
if (pcmk_is_set(then->flags, pe_action_requires_any)) {
if (last_flags == then->flags) {
pcmk__clear_updated_flags(changed, then, pcmk__updated_then);
} else {
pcmk__set_updated_flags(changed, then, pcmk__updated_then);
}
}
if (pcmk_is_set(changed, pcmk__updated_then)) {
crm_trace("Re-processing %s and its 'after' actions because it changed",
then->uuid);
if (pcmk_is_set(last_flags, pe_action_runnable)
&& !pcmk_is_set(then->flags, pe_action_runnable)) {
pcmk__block_colocation_dependents(then);
}
pcmk__update_action_for_orderings(then, data_set);
for (lpc = then->actions_after; lpc != NULL; lpc = lpc->next) {
pe_action_wrapper_t *other = (pe_action_wrapper_t *) lpc->data;
pcmk__update_action_for_orderings(other->action, data_set);
}
}
}
static inline bool
is_primitive_action(const pe_action_t *action)
{
return action && action->rsc && (action->rsc->variant == pe_native);
}
/*!
* \internal
* \brief Clear a single action flag and set reason text
*
* \param[in,out] action Action whose flag should be cleared
* \param[in] flag Action flag that should be cleared
* \param[in] reason Action that is the reason why flag is being cleared
*/
#define clear_action_flag_because(action, flag, reason) do { \
if (pcmk_is_set((action)->flags, (flag))) { \
pe__clear_action_flags(action, flag); \
if ((action)->rsc != (reason)->rsc) { \
char *reason_text = pe__action2reason((reason), (flag)); \
pe_action_set_reason((action), reason_text, false); \
free(reason_text); \
} \
} \
} while (0)
/*!
* \internal
* \brief Update actions in an asymmetric ordering
*
* If the "first" action in an asymmetric ordering is unrunnable, make the
* "second" action unrunnable as well, if appropriate.
*
* \param[in] first 'First' action in an asymmetric ordering
* \param[in,out] then 'Then' action in an asymmetric ordering
*/
static void
handle_asymmetric_ordering(const pe_action_t *first, pe_action_t *then)
{
/* Only resource actions after an unrunnable 'first' action need updates for
* asymmetric ordering.
*/
if ((then->rsc == NULL) || pcmk_is_set(first->flags, pe_action_runnable)) {
return;
}
// Certain optional 'then' actions are unaffected by unrunnable 'first'
if (pcmk_is_set(then->flags, pe_action_optional)) {
enum rsc_role_e then_rsc_role = then->rsc->fns->state(then->rsc, TRUE);
if ((then_rsc_role == RSC_ROLE_STOPPED)
&& pcmk__str_eq(then->task, PCMK_ACTION_STOP, pcmk__str_none)) {
/* If 'then' should stop after 'first' but is already stopped, the
* ordering is irrelevant.
*/
return;
} else if ((then_rsc_role >= RSC_ROLE_STARTED)
&& pcmk__str_eq(then->task, PCMK_ACTION_START, pcmk__str_none)
&& pe__rsc_running_on_only(then->rsc, then->node)) {
/* Similarly if 'then' should start after 'first' but is already
* started on a single node.
*/
return;
}
}
// 'First' can't run, so 'then' can't either
clear_action_flag_because(then, pe_action_optional, first);
clear_action_flag_because(then, pe_action_runnable, first);
}
/*!
* \internal
* \brief Set action bits appropriately when pe_restart_order is used
*
* \param[in,out] first 'First' action in an ordering with pe_restart_order
* \param[in,out] then 'Then' action in an ordering with pe_restart_order
* \param[in] filter What action flags to care about
*
* \note pe_restart_order is set for "stop resource before starting it" and
* "stop later group member before stopping earlier group member"
*/
static void
handle_restart_ordering(pe_action_t *first, pe_action_t *then, uint32_t filter)
{
const char *reason = NULL;
CRM_ASSERT(is_primitive_action(first));
CRM_ASSERT(is_primitive_action(then));
// We need to update the action in two cases:
// ... if 'then' is required
if (pcmk_is_set(filter, pe_action_optional)
&& !pcmk_is_set(then->flags, pe_action_optional)) {
reason = "restart";
}
/* ... if 'then' is unrunnable action on same resource (if a resource
* should restart but can't start, we still want to stop)
*/
if (pcmk_is_set(filter, pe_action_runnable)
&& !pcmk_is_set(then->flags, pe_action_runnable)
&& pcmk_is_set(then->rsc->flags, pe_rsc_managed)
&& (first->rsc == then->rsc)) {
reason = "stop";
}
if (reason == NULL) {
return;
}
pe_rsc_trace(first->rsc, "Handling %s -> %s for %s",
first->uuid, then->uuid, reason);
// Make 'first' required if it is runnable
if (pcmk_is_set(first->flags, pe_action_runnable)) {
clear_action_flag_because(first, pe_action_optional, then);
}
// Make 'first' required if 'then' is required
if (!pcmk_is_set(then->flags, pe_action_optional)) {
clear_action_flag_because(first, pe_action_optional, then);
}
// Make 'first' unmigratable if 'then' is unmigratable
if (!pcmk_is_set(then->flags, pe_action_migrate_runnable)) {
clear_action_flag_because(first, pe_action_migrate_runnable, then);
}
// Make 'then' unrunnable if 'first' is required but unrunnable
if (!pcmk_is_set(first->flags, pe_action_optional)
&& !pcmk_is_set(first->flags, pe_action_runnable)) {
clear_action_flag_because(then, pe_action_runnable, first);
}
}
/*!
* \internal
* \brief Update two actions according to an ordering between them
*
* Given information about an ordering of two actions, update the actions' flags
* (and runnable_before members if appropriate) as appropriate for the ordering.
* Effects may cascade to other orderings involving the actions as well.
*
* \param[in,out] first 'First' action in an ordering
* \param[in,out] then 'Then' action in an ordering
* \param[in] node If not NULL, limit scope of ordering to this node
* (ignored)
* \param[in] flags Action flags for \p first for ordering purposes
* \param[in] filter Action flags to limit scope of certain updates (may
* include pe_action_optional to affect only mandatory
* actions, and pe_action_runnable to affect only
* runnable actions)
* \param[in] type Group of enum pe_ordering flags to apply
* \param[in,out] data_set Cluster working set
*
* \return Group of enum pcmk__updated flags indicating what was updated
*/
uint32_t
pcmk__update_ordered_actions(pe_action_t *first, pe_action_t *then,
const pe_node_t *node, uint32_t flags,
uint32_t filter, uint32_t type,
pe_working_set_t *data_set)
{
uint32_t changed = pcmk__updated_none;
uint32_t then_flags = 0U;
uint32_t first_flags = 0U;
CRM_ASSERT((first != NULL) && (then != NULL) && (data_set != NULL));
then_flags = then->flags;
first_flags = first->flags;
if (pcmk_is_set(type, pe_order_asymmetrical)) {
handle_asymmetric_ordering(first, then);
}
if (pcmk_is_set(type, pe_order_implies_first)
&& !pcmk_is_set(then_flags, pe_action_optional)) {
// Then is required, and implies first should be, too
if (pcmk_is_set(filter, pe_action_optional)
&& !pcmk_is_set(flags, pe_action_optional)
&& pcmk_is_set(first_flags, pe_action_optional)) {
clear_action_flag_because(first, pe_action_optional, then);
}
if (pcmk_is_set(flags, pe_action_migrate_runnable)
&& !pcmk_is_set(then->flags, pe_action_migrate_runnable)) {
clear_action_flag_because(first, pe_action_migrate_runnable, then);
}
}
if (pcmk_is_set(type, pe_order_promoted_implies_first)
&& (then->rsc != NULL) && (then->rsc->role == RSC_ROLE_PROMOTED)
&& pcmk_is_set(filter, pe_action_optional)
&& !pcmk_is_set(then->flags, pe_action_optional)) {
clear_action_flag_because(first, pe_action_optional, then);
if (pcmk_is_set(first->flags, pe_action_migrate_runnable)
&& !pcmk_is_set(then->flags, pe_action_migrate_runnable)) {
clear_action_flag_because(first, pe_action_migrate_runnable,
then);
}
}
if (pcmk_is_set(type, pe_order_implies_first_migratable)
&& pcmk_is_set(filter, pe_action_optional)) {
if (!pcmk_all_flags_set(then->flags, pe_action_migrate_runnable
|pe_action_runnable)) {
clear_action_flag_because(first, pe_action_runnable, then);
}
if (!pcmk_is_set(then->flags, pe_action_optional)) {
clear_action_flag_because(first, pe_action_optional, then);
}
}
if (pcmk_is_set(type, pe_order_pseudo_left)
&& pcmk_is_set(filter, pe_action_optional)
&& !pcmk_is_set(first->flags, pe_action_runnable)) {
clear_action_flag_because(then, pe_action_migrate_runnable, first);
pe__clear_action_flags(then, pe_action_pseudo);
}
if (pcmk_is_set(type, pe_order_runnable_left)
&& pcmk_is_set(filter, pe_action_runnable)
&& pcmk_is_set(then->flags, pe_action_runnable)
&& !pcmk_is_set(flags, pe_action_runnable)) {
clear_action_flag_because(then, pe_action_runnable, first);
clear_action_flag_because(then, pe_action_migrate_runnable, first);
}
if (pcmk_is_set(type, pe_order_implies_then)
&& pcmk_is_set(filter, pe_action_optional)
&& pcmk_is_set(then->flags, pe_action_optional)
&& !pcmk_is_set(flags, pe_action_optional)
&& !pcmk_is_set(first->flags, pe_action_migrate_runnable)) {
clear_action_flag_because(then, pe_action_optional, first);
}
if (pcmk_is_set(type, pe_order_restart)) {
handle_restart_ordering(first, then, filter);
}
if (then_flags != then->flags) {
pcmk__set_updated_flags(changed, first, pcmk__updated_then);
pe_rsc_trace(then->rsc,
"%s on %s: flags are now %#.6x (was %#.6x) "
"because of 'first' %s (%#.6x)",
then->uuid, pe__node_name(then->node),
then->flags, then_flags, first->uuid, first->flags);
if ((then->rsc != NULL) && (then->rsc->parent != NULL)) {
// Required to handle "X_stop then X_start" for cloned groups
pcmk__update_action_for_orderings(then, data_set);
}
}
if (first_flags != first->flags) {
pcmk__set_updated_flags(changed, first, pcmk__updated_first);
pe_rsc_trace(first->rsc,
"%s on %s: flags are now %#.6x (was %#.6x) "
"because of 'then' %s (%#.6x)",
first->uuid, pe__node_name(first->node),
first->flags, first_flags, then->uuid, then->flags);
}
return changed;
}
/*!
* \internal
* \brief Trace-log an action (optionally with its dependent actions)
*
* \param[in] pre_text If not NULL, prefix the log with this plus ": "
* \param[in] action Action to log
* \param[in] details If true, recursively log dependent actions
*/
void
pcmk__log_action(const char *pre_text, const pe_action_t *action, bool details)
{
const char *node_uname = NULL;
const char *node_uuid = NULL;
const char *desc = NULL;
CRM_CHECK(action != NULL, return);
if (!pcmk_is_set(action->flags, pe_action_pseudo)) {
if (action->node != NULL) {
node_uname = action->node->details->uname;
node_uuid = action->node->details->id;
} else {
node_uname = "<none>";
}
}
switch (text2task(action->task)) {
case stonith_node:
case shutdown_crm:
if (pcmk_is_set(action->flags, pe_action_pseudo)) {
desc = "Pseudo ";
} else if (pcmk_is_set(action->flags, pe_action_optional)) {
desc = "Optional ";
} else if (!pcmk_is_set(action->flags, pe_action_runnable)) {
desc = "!!Non-Startable!! ";
} else if (pcmk_is_set(action->flags, pe_action_processed)) {
desc = "";
} else {
desc = "(Provisional) ";
}
crm_trace("%s%s%sAction %d: %s%s%s%s%s%s",
((pre_text == NULL)? "" : pre_text),
((pre_text == NULL)? "" : ": "),
desc, action->id, action->uuid,
(node_uname? "\ton " : ""), (node_uname? node_uname : ""),
(node_uuid? "\t\t(" : ""), (node_uuid? node_uuid : ""),
(node_uuid? ")" : ""));
break;
default:
if (pcmk_is_set(action->flags, pe_action_optional)) {
desc = "Optional ";
} else if (pcmk_is_set(action->flags, pe_action_pseudo)) {
desc = "Pseudo ";
} else if (!pcmk_is_set(action->flags, pe_action_runnable)) {
desc = "!!Non-Startable!! ";
} else if (pcmk_is_set(action->flags, pe_action_processed)) {
desc = "";
} else {
desc = "(Provisional) ";
}
crm_trace("%s%s%sAction %d: %s %s%s%s%s%s%s",
((pre_text == NULL)? "" : pre_text),
((pre_text == NULL)? "" : ": "),
desc, action->id, action->uuid,
(action->rsc? action->rsc->id : "<none>"),
(node_uname? "\ton " : ""), (node_uname? node_uname : ""),
(node_uuid? "\t\t(" : ""), (node_uuid? node_uuid : ""),
(node_uuid? ")" : ""));
break;
}
if (details) {
const GList *iter = NULL;
const pe_action_wrapper_t *other = NULL;
crm_trace("\t\t====== Preceding Actions");
for (iter = action->actions_before; iter != NULL; iter = iter->next) {
other = (const pe_action_wrapper_t *) iter->data;
pcmk__log_action("\t\t", other->action, false);
}
crm_trace("\t\t====== Subsequent Actions");
for (iter = action->actions_after; iter != NULL; iter = iter->next) {
other = (const pe_action_wrapper_t *) iter->data;
pcmk__log_action("\t\t", other->action, false);
}
crm_trace("\t\t====== End");
} else {
crm_trace("\t\t(before=%d, after=%d)",
g_list_length(action->actions_before),
g_list_length(action->actions_after));
}
}
/*!
* \internal
* \brief Create a new shutdown action for a node
*
* \param[in,out] node Node being shut down
*
* \return Newly created shutdown action for \p node
*/
pe_action_t *
pcmk__new_shutdown_action(pe_node_t *node)
{
char *shutdown_id = NULL;
pe_action_t *shutdown_op = NULL;
CRM_ASSERT(node != NULL);
shutdown_id = crm_strdup_printf("%s-%s", CRM_OP_SHUTDOWN,
node->details->uname);
shutdown_op = custom_action(NULL, shutdown_id, CRM_OP_SHUTDOWN, node, FALSE,
TRUE, node->details->data_set);
pcmk__order_stops_before_shutdown(node, shutdown_op);
add_hash_param(shutdown_op->meta, XML_ATTR_TE_NOWAIT, XML_BOOLEAN_TRUE);
return shutdown_op;
}
/*!
* \internal
* \brief Calculate and add an operation digest to XML
*
* Calculate an operation digest, which enables us to later determine when a
* restart is needed due to the resource's parameters being changed, and add it
* to given XML.
*
* \param[in] op Operation result from executor
* \param[in,out] update XML to add digest to
*/
static void
add_op_digest_to_xml(const lrmd_event_data_t *op, xmlNode *update)
{
char *digest = NULL;
xmlNode *args_xml = NULL;
if (op->params == NULL) {
return;
}
args_xml = create_xml_node(NULL, XML_TAG_PARAMS);
g_hash_table_foreach(op->params, hash2field, args_xml);
pcmk__filter_op_for_digest(args_xml);
digest = calculate_operation_digest(args_xml, NULL);
crm_xml_add(update, XML_LRM_ATTR_OP_DIGEST, digest);
free_xml(args_xml);
free(digest);
}
#define FAKE_TE_ID "xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx"
/*!
* \internal
* \brief Create XML for resource operation history update
*
* \param[in,out] parent Parent XML node to add to
* \param[in,out] op Operation event data
* \param[in] caller_version DC feature set
* \param[in] target_rc Expected result of operation
* \param[in] node Name of node on which operation was performed
* \param[in] origin Arbitrary description of update source
*
* \return Newly created XML node for history update
*/
xmlNode *
pcmk__create_history_xml(xmlNode *parent, lrmd_event_data_t *op,
const char *caller_version, int target_rc,
const char *node, const char *origin)
{
char *key = NULL;
char *magic = NULL;
char *op_id = NULL;
char *op_id_additional = NULL;
char *local_user_data = NULL;
const char *exit_reason = NULL;
xmlNode *xml_op = NULL;
const char *task = NULL;
CRM_CHECK(op != NULL, return NULL);
crm_trace("Creating history XML for %s-interval %s action for %s on %s "
"(DC version: %s, origin: %s)",
pcmk__readable_interval(op->interval_ms), op->op_type, op->rsc_id,
((node == NULL)? "no node" : node), caller_version, origin);
task = op->op_type;
/* Record a successful agent reload as a start, and a failed one as a
* monitor, to make life easier for the scheduler when determining the
* current state.
*
* @COMPAT We should check "reload" here only if the operation was for a
* pre-OCF-1.1 resource agent, but we don't know that here, and we should
* only ever get results for actions scheduled by us, so we can reasonably
* assume any "reload" is actually a pre-1.1 agent reload.
*/
- if (pcmk__str_any_of(task, PCMK_ACTION_RELOAD, CRMD_ACTION_RELOAD_AGENT,
+ if (pcmk__str_any_of(task, PCMK_ACTION_RELOAD, PCMK_ACTION_RELOAD_AGENT,
NULL)) {
if (op->op_status == PCMK_EXEC_DONE) {
task = PCMK_ACTION_START;
} else {
task = PCMK_ACTION_MONITOR;
}
}
key = pcmk__op_key(op->rsc_id, task, op->interval_ms);
if (pcmk__str_eq(task, PCMK_ACTION_NOTIFY, pcmk__str_none)) {
const char *n_type = crm_meta_value(op->params, "notify_type");
const char *n_task = crm_meta_value(op->params, "notify_operation");
CRM_LOG_ASSERT(n_type != NULL);
CRM_LOG_ASSERT(n_task != NULL);
op_id = pcmk__notify_key(op->rsc_id, n_type, n_task);
if (op->op_status != PCMK_EXEC_PENDING) {
/* Ignore notify errors.
*
* @TODO It might be better to keep the correct result here, and
* ignore it in process_graph_event().
*/
lrmd__set_result(op, PCMK_OCF_OK, PCMK_EXEC_DONE, NULL);
}
/* Migration history is preserved separately, which usually matters for
* multiple nodes and is important for future cluster transitions.
*/
} else if (pcmk__str_any_of(op->op_type, CRMD_ACTION_MIGRATE,
CRMD_ACTION_MIGRATED, NULL)) {
op_id = strdup(key);
} else if (did_rsc_op_fail(op, target_rc)) {
op_id = pcmk__op_key(op->rsc_id, "last_failure", 0);
if (op->interval_ms == 0) {
// Ensure 'last' gets updated, in case record-pending is true
op_id_additional = pcmk__op_key(op->rsc_id, "last", 0);
}
exit_reason = op->exit_reason;
} else if (op->interval_ms > 0) {
op_id = strdup(key);
} else {
op_id = pcmk__op_key(op->rsc_id, "last", 0);
}
again:
xml_op = pcmk__xe_match(parent, XML_LRM_TAG_RSC_OP, XML_ATTR_ID, op_id);
if (xml_op == NULL) {
xml_op = create_xml_node(parent, XML_LRM_TAG_RSC_OP);
}
if (op->user_data == NULL) {
crm_debug("Generating fake transition key for: " PCMK__OP_FMT
" %d from %s", op->rsc_id, op->op_type, op->interval_ms,
op->call_id, origin);
local_user_data = pcmk__transition_key(-1, op->call_id, target_rc,
FAKE_TE_ID);
op->user_data = local_user_data;
}
if (magic == NULL) {
magic = crm_strdup_printf("%d:%d;%s", op->op_status, op->rc,
(const char *) op->user_data);
}
crm_xml_add(xml_op, XML_ATTR_ID, op_id);
crm_xml_add(xml_op, XML_LRM_ATTR_TASK_KEY, key);
crm_xml_add(xml_op, XML_LRM_ATTR_TASK, task);
crm_xml_add(xml_op, XML_ATTR_ORIGIN, origin);
crm_xml_add(xml_op, XML_ATTR_CRM_VERSION, caller_version);
crm_xml_add(xml_op, XML_ATTR_TRANSITION_KEY, op->user_data);
crm_xml_add(xml_op, XML_ATTR_TRANSITION_MAGIC, magic);
crm_xml_add(xml_op, XML_LRM_ATTR_EXIT_REASON, pcmk__s(exit_reason, ""));
crm_xml_add(xml_op, XML_LRM_ATTR_TARGET, node); // For context during triage
crm_xml_add_int(xml_op, XML_LRM_ATTR_CALLID, op->call_id);
crm_xml_add_int(xml_op, XML_LRM_ATTR_RC, op->rc);
crm_xml_add_int(xml_op, XML_LRM_ATTR_OPSTATUS, op->op_status);
crm_xml_add_ms(xml_op, XML_LRM_ATTR_INTERVAL_MS, op->interval_ms);
if (compare_version("2.1", caller_version) <= 0) {
if (op->t_run || op->t_rcchange || op->exec_time || op->queue_time) {
crm_trace("Timing data (" PCMK__OP_FMT
"): last=%u change=%u exec=%u queue=%u",
op->rsc_id, op->op_type, op->interval_ms,
op->t_run, op->t_rcchange, op->exec_time, op->queue_time);
if ((op->interval_ms != 0) && (op->t_rcchange != 0)) {
// Recurring ops may have changed rc after initial run
crm_xml_add_ll(xml_op, XML_RSC_OP_LAST_CHANGE,
(long long) op->t_rcchange);
} else {
crm_xml_add_ll(xml_op, XML_RSC_OP_LAST_CHANGE,
(long long) op->t_run);
}
crm_xml_add_int(xml_op, XML_RSC_OP_T_EXEC, op->exec_time);
crm_xml_add_int(xml_op, XML_RSC_OP_T_QUEUE, op->queue_time);
}
}
if (pcmk__str_any_of(op->op_type, CRMD_ACTION_MIGRATE, CRMD_ACTION_MIGRATED,
NULL)) {
/*
* Record migrate_source and migrate_target always for migrate ops.
*/
const char *name = XML_LRM_ATTR_MIGRATE_SOURCE;
crm_xml_add(xml_op, name, crm_meta_value(op->params, name));
name = XML_LRM_ATTR_MIGRATE_TARGET;
crm_xml_add(xml_op, name, crm_meta_value(op->params, name));
}
add_op_digest_to_xml(op, xml_op);
if (op_id_additional) {
free(op_id);
op_id = op_id_additional;
op_id_additional = NULL;
goto again;
}
if (local_user_data) {
free(local_user_data);
op->user_data = NULL;
}
free(magic);
free(op_id);
free(key);
return xml_op;
}
/*!
* \internal
* \brief Check whether an action shutdown-locks a resource to a node
*
* If the shutdown-lock cluster property is set, resources will not be recovered
* on a different node if cleanly stopped, and may start only on that same node.
* This function checks whether that applies to a given action, so that the
* transition graph can be marked appropriately.
*
* \param[in] action Action to check
*
* \return true if \p action locks its resource to the action's node,
* otherwise false
*/
bool
pcmk__action_locks_rsc_to_node(const pe_action_t *action)
{
// Only resource actions taking place on resource's lock node are locked
if ((action == NULL) || (action->rsc == NULL)
|| !pe__same_node(action->node, action->rsc->lock_node)) {
return false;
}
/* During shutdown, only stops are locked (otherwise, another action such as
* a demote would cause the controller to clear the lock)
*/
if (action->node->details->shutdown && (action->task != NULL)
&& (strcmp(action->task, PCMK_ACTION_STOP) != 0)) {
return false;
}
return true;
}
/* lowest to highest */
static gint
sort_action_id(gconstpointer a, gconstpointer b)
{
const pe_action_wrapper_t *action_wrapper2 = (const pe_action_wrapper_t *)a;
const pe_action_wrapper_t *action_wrapper1 = (const pe_action_wrapper_t *)b;
if (a == NULL) {
return 1;
}
if (b == NULL) {
return -1;
}
if (action_wrapper1->action->id < action_wrapper2->action->id) {
return 1;
}
if (action_wrapper1->action->id > action_wrapper2->action->id) {
return -1;
}
return 0;
}
/*!
* \internal
* \brief Remove any duplicate action inputs, merging action flags
*
* \param[in,out] action Action whose inputs should be checked
*/
void
pcmk__deduplicate_action_inputs(pe_action_t *action)
{
GList *item = NULL;
GList *next = NULL;
pe_action_wrapper_t *last_input = NULL;
action->actions_before = g_list_sort(action->actions_before,
sort_action_id);
for (item = action->actions_before; item != NULL; item = next) {
pe_action_wrapper_t *input = (pe_action_wrapper_t *) item->data;
next = item->next;
if ((last_input != NULL)
&& (input->action->id == last_input->action->id)) {
crm_trace("Input %s (%d) duplicate skipped for action %s (%d)",
input->action->uuid, input->action->id,
action->uuid, action->id);
/* For the purposes of scheduling, the ordering flags no longer
* matter, but crm_simulate looks at certain ones when creating a
* dot graph. Combining the flags is sufficient for that purpose.
*/
last_input->type |= input->type;
if (input->state == pe_link_dumped) {
last_input->state = pe_link_dumped;
}
free(item->data);
action->actions_before = g_list_delete_link(action->actions_before,
item);
} else {
last_input = input;
input->state = pe_link_not_dumped;
}
}
}
/*!
* \internal
* \brief Output all scheduled actions
*
* \param[in,out] data_set Cluster working set
*/
void
pcmk__output_actions(pe_working_set_t *data_set)
{
pcmk__output_t *out = data_set->priv;
// Output node (non-resource) actions
for (GList *iter = data_set->actions; iter != NULL; iter = iter->next) {
char *node_name = NULL;
char *task = NULL;
pe_action_t *action = (pe_action_t *) iter->data;
if (action->rsc != NULL) {
continue; // Resource actions will be output later
} else if (pcmk_is_set(action->flags, pe_action_optional)) {
continue; // This action was not scheduled
}
if (pcmk__str_eq(action->task, CRM_OP_SHUTDOWN, pcmk__str_none)) {
task = strdup("Shutdown");
} else if (pcmk__str_eq(action->task, CRM_OP_FENCE, pcmk__str_none)) {
const char *op = g_hash_table_lookup(action->meta,
"stonith_action");
task = crm_strdup_printf("Fence (%s)", op);
} else {
continue; // Don't display other node action types
}
if (pe__is_guest_node(action->node)) {
const pe_resource_t *remote = action->node->details->remote_rsc;
node_name = crm_strdup_printf("%s (resource: %s)",
pe__node_name(action->node),
remote->container->id);
} else if (action->node != NULL) {
node_name = crm_strdup_printf("%s", pe__node_name(action->node));
}
out->message(out, "node-action", task, node_name, action->reason);
free(node_name);
free(task);
}
// Output resource actions
for (GList *iter = data_set->resources; iter != NULL; iter = iter->next) {
pe_resource_t *rsc = (pe_resource_t *) iter->data;
rsc->cmds->output_actions(rsc);
}
}
/*!
* \internal
* \brief Check whether action from resource history is still in configuration
*
* \param[in] rsc Resource that action is for
* \param[in] task Action's name
* \param[in] interval_ms Action's interval (in milliseconds)
*
* \return true if action is still in resource configuration, otherwise false
*/
static bool
action_in_config(const pe_resource_t *rsc, const char *task, guint interval_ms)
{
char *key = pcmk__op_key(rsc->id, task, interval_ms);
bool config = (find_rsc_op_entry(rsc, key) != NULL);
free(key);
return config;
}
/*!
* \internal
* \brief Get action name needed to compare digest for configuration changes
*
* \param[in] task Action name from history
* \param[in] interval_ms Action interval (in milliseconds)
*
* \return Action name whose digest should be compared
*/
static const char *
task_for_digest(const char *task, guint interval_ms)
{
/* Certain actions need to be compared against the parameters used to start
* the resource.
*/
if ((interval_ms == 0)
&& pcmk__str_any_of(task, PCMK_ACTION_MONITOR, RSC_MIGRATED,
PCMK_ACTION_PROMOTE, NULL)) {
task = PCMK_ACTION_START;
}
return task;
}
/*!
* \internal
* \brief Check whether only sanitized parameters to an action changed
*
* When collecting CIB files for troubleshooting, crm_report will mask
* sensitive resource parameters. If simulations were run using that, affected
* resources would appear to need a restart, which would complicate
* troubleshooting. To avoid that, we save a "secure digest" of non-sensitive
* parameters. This function used that digest to check whether only masked
* parameters are different.
*
* \param[in] xml_op Resource history entry with secure digest
* \param[in] digest_data Operation digest information being compared
* \param[in] data_set Cluster working set
*
* \return true if only sanitized parameters changed, otherwise false
*/
static bool
only_sanitized_changed(const xmlNode *xml_op,
const op_digest_cache_t *digest_data,
const pe_working_set_t *data_set)
{
const char *digest_secure = NULL;
if (!pcmk_is_set(data_set->flags, pe_flag_sanitized)) {
// The scheduler is not being run as a simulation
return false;
}
digest_secure = crm_element_value(xml_op, XML_LRM_ATTR_SECURE_DIGEST);
return (digest_data->rc != RSC_DIGEST_MATCH) && (digest_secure != NULL)
&& (digest_data->digest_secure_calc != NULL)
&& (strcmp(digest_data->digest_secure_calc, digest_secure) == 0);
}
/*!
* \internal
* \brief Force a restart due to a configuration change
*
* \param[in,out] rsc Resource that action is for
* \param[in] task Name of action whose configuration changed
* \param[in] interval_ms Action interval (in milliseconds)
* \param[in,out] node Node where resource should be restarted
*/
static void
force_restart(pe_resource_t *rsc, const char *task, guint interval_ms,
pe_node_t *node)
{
char *key = pcmk__op_key(rsc->id, task, interval_ms);
pe_action_t *required = custom_action(rsc, key, task, NULL, FALSE, TRUE,
rsc->cluster);
pe_action_set_reason(required, "resource definition change", true);
trigger_unfencing(rsc, node, "Device parameters changed", NULL,
rsc->cluster);
}
/*!
* \internal
* \brief Schedule a reload of a resource on a node
*
* \param[in,out] data Resource to reload
* \param[in] user_data Where resource should be reloaded
*/
static void
schedule_reload(gpointer data, gpointer user_data)
{
pe_resource_t *rsc = data;
const pe_node_t *node = user_data;
pe_action_t *reload = NULL;
// For collective resources, just call recursively for children
if (rsc->variant > pe_native) {
g_list_foreach(rsc->children, schedule_reload, user_data);
return;
}
// Skip the reload in certain situations
if ((node == NULL)
|| !pcmk_is_set(rsc->flags, pe_rsc_managed)
|| pcmk_is_set(rsc->flags, pe_rsc_failed)) {
pe_rsc_trace(rsc, "Skip reload of %s:%s%s %s",
rsc->id,
pcmk_is_set(rsc->flags, pe_rsc_managed)? "" : " unmanaged",
pcmk_is_set(rsc->flags, pe_rsc_failed)? " failed" : "",
(node == NULL)? "inactive" : node->details->uname);
return;
}
/* If a resource's configuration changed while a start was pending,
* force a full restart instead of a reload.
*/
if (pcmk_is_set(rsc->flags, pe_rsc_start_pending)) {
pe_rsc_trace(rsc, "%s: preventing agent reload because start pending",
rsc->id);
custom_action(rsc, stop_key(rsc), PCMK_ACTION_STOP, node, FALSE, TRUE,
rsc->cluster);
return;
}
// Schedule the reload
pe__set_resource_flags(rsc, pe_rsc_reload);
- reload = custom_action(rsc, reload_key(rsc), CRMD_ACTION_RELOAD_AGENT, node,
+ reload = custom_action(rsc, reload_key(rsc), PCMK_ACTION_RELOAD_AGENT, node,
FALSE, TRUE, rsc->cluster);
pe_action_set_reason(reload, "resource definition change", FALSE);
// Set orderings so that a required stop or demote cancels the reload
pcmk__new_ordering(NULL, NULL, reload, rsc, stop_key(rsc), NULL,
pe_order_optional|pe_order_then_cancels_first,
rsc->cluster);
pcmk__new_ordering(NULL, NULL, reload, rsc, demote_key(rsc), NULL,
pe_order_optional|pe_order_then_cancels_first,
rsc->cluster);
}
/*!
* \internal
* \brief Handle any configuration change for an action
*
* Given an action from resource history, if the resource's configuration
* changed since the action was done, schedule any actions needed (restart,
* reload, unfencing, rescheduling recurring actions, etc.).
*
* \param[in,out] rsc Resource that action is for
* \param[in,out] node Node that action was on
* \param[in] xml_op Action XML from resource history
*
* \return true if action configuration changed, otherwise false
*/
bool
pcmk__check_action_config(pe_resource_t *rsc, pe_node_t *node,
const xmlNode *xml_op)
{
guint interval_ms = 0;
const char *task = NULL;
const op_digest_cache_t *digest_data = NULL;
CRM_CHECK((rsc != NULL) && (node != NULL) && (xml_op != NULL),
return false);
task = crm_element_value(xml_op, XML_LRM_ATTR_TASK);
CRM_CHECK(task != NULL, return false);
crm_element_value_ms(xml_op, XML_LRM_ATTR_INTERVAL_MS, &interval_ms);
// If this is a recurring action, check whether it has been orphaned
if (interval_ms > 0) {
if (action_in_config(rsc, task, interval_ms)) {
pe_rsc_trace(rsc, "%s-interval %s for %s on %s is in configuration",
pcmk__readable_interval(interval_ms), task, rsc->id,
pe__node_name(node));
} else if (pcmk_is_set(rsc->cluster->flags,
pe_flag_stop_action_orphans)) {
pcmk__schedule_cancel(rsc,
crm_element_value(xml_op,
XML_LRM_ATTR_CALLID),
task, interval_ms, node, "orphan");
return true;
} else {
pe_rsc_debug(rsc, "%s-interval %s for %s on %s is orphaned",
pcmk__readable_interval(interval_ms), task, rsc->id,
pe__node_name(node));
return true;
}
}
crm_trace("Checking %s-interval %s for %s on %s for configuration changes",
pcmk__readable_interval(interval_ms), task, rsc->id,
pe__node_name(node));
task = task_for_digest(task, interval_ms);
digest_data = rsc_action_digest_cmp(rsc, xml_op, node, rsc->cluster);
if (only_sanitized_changed(xml_op, digest_data, rsc->cluster)) {
if (!pcmk__is_daemon && (rsc->cluster->priv != NULL)) {
pcmk__output_t *out = rsc->cluster->priv;
out->info(out,
"Only 'private' parameters to %s-interval %s for %s "
"on %s changed: %s",
pcmk__readable_interval(interval_ms), task, rsc->id,
pe__node_name(node),
crm_element_value(xml_op, XML_ATTR_TRANSITION_MAGIC));
}
return false;
}
switch (digest_data->rc) {
case RSC_DIGEST_RESTART:
crm_log_xml_debug(digest_data->params_restart, "params:restart");
force_restart(rsc, task, interval_ms, node);
return true;
case RSC_DIGEST_ALL:
case RSC_DIGEST_UNKNOWN:
// Changes that can potentially be handled by an agent reload
if (interval_ms > 0) {
/* Recurring actions aren't reloaded per se, they are just
* re-scheduled so the next run uses the new parameters.
* The old instance will be cancelled automatically.
*/
crm_log_xml_debug(digest_data->params_all, "params:reschedule");
pcmk__reschedule_recurring(rsc, task, interval_ms, node);
} else if (crm_element_value(xml_op,
XML_LRM_ATTR_RESTART_DIGEST) != NULL) {
// Agent supports reload, so use it
trigger_unfencing(rsc, node,
"Device parameters changed (reload)", NULL,
rsc->cluster);
crm_log_xml_debug(digest_data->params_all, "params:reload");
schedule_reload((gpointer) rsc, (gpointer) node);
} else {
pe_rsc_trace(rsc,
"Restarting %s "
"because agent doesn't support reload", rsc->id);
crm_log_xml_debug(digest_data->params_restart,
"params:restart");
force_restart(rsc, task, interval_ms, node);
}
return true;
default:
break;
}
return false;
}
/*!
* \internal
* \brief Create a list of resource's action history entries, sorted by call ID
*
* \param[in] rsc_entry Resource's <lrm_rsc_op> status XML
* \param[out] start_index Where to store index of start-like action, if any
* \param[out] stop_index Where to store index of stop action, if any
*/
static GList *
rsc_history_as_list(const xmlNode *rsc_entry, int *start_index, int *stop_index)
{
GList *ops = NULL;
for (xmlNode *rsc_op = first_named_child(rsc_entry, XML_LRM_TAG_RSC_OP);
rsc_op != NULL; rsc_op = crm_next_same_xml(rsc_op)) {
ops = g_list_prepend(ops, rsc_op);
}
ops = g_list_sort(ops, sort_op_by_callid);
calculate_active_ops(ops, start_index, stop_index);
return ops;
}
/*!
* \internal
* \brief Process a resource's action history from the CIB status
*
* Given a resource's action history, if the resource's configuration
* changed since the actions were done, schedule any actions needed (restart,
* reload, unfencing, rescheduling recurring actions, clean-up, etc.).
* (This also cancels recurring actions for maintenance mode, which is not
* entirely related but convenient to do here.)
*
* \param[in] rsc_entry Resource's <lrm_rsc_op> status XML
* \param[in,out] rsc Resource whose history is being processed
* \param[in,out] node Node whose history is being processed
*/
static void
process_rsc_history(const xmlNode *rsc_entry, pe_resource_t *rsc,
pe_node_t *node)
{
int offset = -1;
int stop_index = 0;
int start_index = 0;
GList *sorted_op_list = NULL;
if (pcmk_is_set(rsc->flags, pe_rsc_orphan)) {
if (pe_rsc_is_anon_clone(pe__const_top_resource(rsc, false))) {
pe_rsc_trace(rsc,
"Skipping configuration check "
"for orphaned clone instance %s",
rsc->id);
} else {
pe_rsc_trace(rsc,
"Skipping configuration check and scheduling clean-up "
"for orphaned resource %s", rsc->id);
pcmk__schedule_cleanup(rsc, node, false);
}
return;
}
if (pe_find_node_id(rsc->running_on, node->details->id) == NULL) {
if (pcmk__rsc_agent_changed(rsc, node, rsc_entry, false)) {
pcmk__schedule_cleanup(rsc, node, false);
}
pe_rsc_trace(rsc,
"Skipping configuration check for %s "
"because no longer active on %s",
rsc->id, pe__node_name(node));
return;
}
pe_rsc_trace(rsc, "Checking for configuration changes for %s on %s",
rsc->id, pe__node_name(node));
if (pcmk__rsc_agent_changed(rsc, node, rsc_entry, true)) {
pcmk__schedule_cleanup(rsc, node, false);
}
sorted_op_list = rsc_history_as_list(rsc_entry, &start_index, &stop_index);
if (start_index < stop_index) {
return; // Resource is stopped
}
for (GList *iter = sorted_op_list; iter != NULL; iter = iter->next) {
xmlNode *rsc_op = (xmlNode *) iter->data;
const char *task = NULL;
guint interval_ms = 0;
if (++offset < start_index) {
// Skip actions that happened before a start
continue;
}
task = crm_element_value(rsc_op, 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
pcmk__schedule_cancel(rsc,
crm_element_value(rsc_op,
XML_LRM_ATTR_CALLID),
task, interval_ms, node, "maintenance mode");
} else if ((interval_ms > 0)
|| pcmk__strcase_any_of(task, PCMK_ACTION_MONITOR,
PCMK_ACTION_START,
PCMK_ACTION_PROMOTE,
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 assigned resources to nodes yet, so if the
* REMOTE_CONTAINER_HACK is used, we may calculate the digest
* based on the literal "#uname" value rather than the properly
* substituted value. That would mistakenly make the action
* definition appear to have been changed. Defer the check until
* later in this case.
*/
pe__add_param_check(rsc_op, rsc, node, pe_check_active,
rsc->cluster);
} else if (pcmk__check_action_config(rsc, node, rsc_op)
&& (pe_get_failcount(node, rsc, NULL, pe_fc_effective,
NULL) != 0)) {
pe__clear_failcount(rsc, node, "action definition changed",
rsc->cluster);
}
}
}
g_list_free(sorted_op_list);
}
/*!
* \internal
* \brief Process a node's action history from the CIB status
*
* Given a node's resource history, if the resource's configuration changed
* since the actions were done, schedule any actions needed (restart,
* reload, unfencing, rescheduling recurring actions, clean-up, etc.).
* (This also cancels recurring actions for maintenance mode, which is not
* entirely related but convenient to do here.)
*
* \param[in,out] node Node whose history is being processed
* \param[in] lrm_rscs Node's <lrm_resources> from CIB status XML
*/
static void
process_node_history(pe_node_t *node, const xmlNode *lrm_rscs)
{
crm_trace("Processing node history for %s", pe__node_name(node));
for (const xmlNode *rsc_entry = first_named_child(lrm_rscs,
XML_LRM_TAG_RESOURCE);
rsc_entry != NULL; rsc_entry = crm_next_same_xml(rsc_entry)) {
if (xml_has_children(rsc_entry)) {
GList *result = pcmk__rscs_matching_id(ID(rsc_entry),
node->details->data_set);
for (GList *iter = result; iter != NULL; iter = iter->next) {
pe_resource_t *rsc = (pe_resource_t *) iter->data;
if (rsc->variant == pe_native) {
process_rsc_history(rsc_entry, rsc, node);
}
}
g_list_free(result);
}
}
}
// XPath to find a node's resource history
#define XPATH_NODE_HISTORY "/" XML_TAG_CIB "/" XML_CIB_TAG_STATUS \
"/" XML_CIB_TAG_STATE "[@" XML_ATTR_UNAME "='%s']" \
"/" XML_CIB_TAG_LRM "/" XML_LRM_TAG_RESOURCES
/*!
* \internal
* \brief Process any resource configuration changes in the CIB status
*
* Go through all nodes' resource history, and if a resource's configuration
* changed since its actions were done, schedule any actions needed (restart,
* reload, unfencing, rescheduling recurring actions, clean-up, etc.).
* (This also cancels recurring actions for maintenance mode, which is not
* entirely related but convenient to do here.)
*
* \param[in,out] data_set Cluster working set
*/
void
pcmk__handle_rsc_config_changes(pe_working_set_t *data_set)
{
crm_trace("Check resource and action configuration for changes");
/* Rather than iterate through the status section, iterate through the nodes
* and search for the appropriate status subsection for each. This skips
* orphaned nodes and lets us eliminate some cases before searching the XML.
*/
for (GList *iter = data_set->nodes; iter != NULL; iter = iter->next) {
pe_node_t *node = (pe_node_t *) iter->data;
/* Don't bother checking actions for a node that can't run actions ...
* unless it's in maintenance mode, in which case we still need to
* cancel any existing recurring monitors.
*/
if (node->details->maintenance
|| pcmk__node_available(node, false, false)) {
char *xpath = NULL;
xmlNode *history = NULL;
xpath = crm_strdup_printf(XPATH_NODE_HISTORY, node->details->uname);
history = get_xpath_object(xpath, data_set->input, LOG_NEVER);
free(xpath);
process_node_history(node, history);
}
}
}
diff --git a/lib/pacemaker/pcmk_sched_recurring.c b/lib/pacemaker/pcmk_sched_recurring.c
index 5b5b9758fd..43d274fe91 100644
--- a/lib/pacemaker/pcmk_sched_recurring.c
+++ b/lib/pacemaker/pcmk_sched_recurring.c
@@ -1,716 +1,716 @@
/*
* Copyright 2004-2023 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU General Public License version 2
* or later (GPLv2+) WITHOUT ANY WARRANTY.
*/
#include <crm_internal.h>
#include <stdbool.h>
#include <crm/msg_xml.h>
#include <pacemaker-internal.h>
#include "libpacemaker_private.h"
// Information parsed from an operation history entry in the CIB
struct op_history {
// XML attributes
const char *id; // ID of history entry
const char *name; // Action name
// Parsed information
char *key; // Operation key for action
enum rsc_role_e role; // Action role (or RSC_ROLE_UNKNOWN for default)
guint interval_ms; // Action interval
};
/*!
* \internal
* \brief Parse an interval from XML
*
* \param[in] xml XML containing an interval attribute
*
* \return Interval parsed from XML (or 0 as default)
*/
static guint
xe_interval(const xmlNode *xml)
{
return crm_parse_interval_spec(crm_element_value(xml,
XML_LRM_ATTR_INTERVAL));
}
/*!
* \internal
* \brief Check whether an operation exists multiple times in resource history
*
* \param[in] rsc Resource with history to search
* \param[in] name Name of action to search for
* \param[in] interval_ms Interval (in milliseconds) of action to search for
*
* \return true if an operation with \p name and \p interval_ms exists more than
* once in the operation history of \p rsc, otherwise false
*/
static bool
is_op_dup(const pe_resource_t *rsc, const char *name, guint interval_ms)
{
const char *id = NULL;
for (xmlNode *op = first_named_child(rsc->ops_xml, "op");
op != NULL; op = crm_next_same_xml(op)) {
// Check whether action name and interval match
if (!pcmk__str_eq(crm_element_value(op, "name"), name, pcmk__str_none)
|| (xe_interval(op) != interval_ms)) {
continue;
}
if (ID(op) == NULL) {
continue; // Shouldn't be possible
}
if (id == NULL) {
id = ID(op); // First matching op
} else {
pcmk__config_err("Operation %s is duplicate of %s (do not use "
"same name and interval combination more "
"than once per resource)", ID(op), id);
return true;
}
}
return false;
}
/*!
* \internal
* \brief Check whether an action name is one that can be recurring
*
* \param[in] name Action name to check
*
* \return true if \p name is an action known to be unsuitable as a recurring
* operation, otherwise false
*
* \note Pacemaker's current philosophy is to allow users to configure recurring
* operations except for a short list of actions known not to be suitable
* for that (as opposed to allowing only actions known to be suitable,
* which includes only monitor). Among other things, this approach allows
* users to define their own custom operations and make them recurring,
* though that use case is not well tested.
*/
static bool
op_cannot_recur(const char *name)
{
return pcmk__str_any_of(name, PCMK_ACTION_STOP, PCMK_ACTION_START,
PCMK_ACTION_DEMOTE, PCMK_ACTION_PROMOTE,
- CRMD_ACTION_RELOAD_AGENT,
+ PCMK_ACTION_RELOAD_AGENT,
CRMD_ACTION_MIGRATE, CRMD_ACTION_MIGRATED, NULL);
}
/*!
* \internal
* \brief Check whether a resource history entry is for a recurring action
*
* \param[in] rsc Resource that history entry is for
* \param[in] xml XML of resource history entry to check
* \param[out] op Where to store parsed info if recurring
*
* \return true if \p xml is for a recurring action, otherwise false
*/
static bool
is_recurring_history(const pe_resource_t *rsc, const xmlNode *xml,
struct op_history *op)
{
const char *role = NULL;
op->interval_ms = xe_interval(xml);
if (op->interval_ms == 0) {
return false; // Not recurring
}
op->id = ID(xml);
if (pcmk__str_empty(op->id)) {
pcmk__config_err("Ignoring resource history entry without ID");
return false; // Shouldn't be possible (unless CIB was manually edited)
}
op->name = crm_element_value(xml, "name");
if (op_cannot_recur(op->name)) {
pcmk__config_err("Ignoring %s because %s action cannot be recurring",
op->id, pcmk__s(op->name, "unnamed"));
return false;
}
// There should only be one recurring operation per action/interval
if (is_op_dup(rsc, op->name, op->interval_ms)) {
return false;
}
// Ensure role is valid if specified
role = crm_element_value(xml, "role");
if (role == NULL) {
op->role = RSC_ROLE_UNKNOWN;
} else {
op->role = text2role(role);
if (op->role == RSC_ROLE_UNKNOWN) {
pcmk__config_err("Ignoring %s because %s is not a valid role",
op->id, role);
}
}
// Disabled resources don't get monitored
op->key = pcmk__op_key(rsc->id, op->name, op->interval_ms);
if (find_rsc_op_entry(rsc, op->key) == NULL) {
crm_trace("Not creating recurring action %s for disabled resource %s",
op->id, rsc->id);
free(op->key);
return false;
}
return true;
}
/*!
* \internal
* \brief Check whether a recurring action for an active role should be optional
*
* \param[in] rsc Resource that recurring action is for
* \param[in] node Node that \p rsc will be active on (if any)
* \param[in] key Operation key for recurring action to check
* \param[in,out] start Start action for \p rsc
*
* \return true if recurring action should be optional, otherwise false
*/
static bool
active_recurring_should_be_optional(const pe_resource_t *rsc,
const pe_node_t *node, const char *key,
pe_action_t *start)
{
GList *possible_matches = NULL;
if (node == NULL) { // Should only be possible if unmanaged and stopped
pe_rsc_trace(rsc, "%s will be mandatory because resource is unmanaged",
key);
return false;
}
if (!pcmk_is_set(rsc->cmds->action_flags(start, NULL),
pe_action_optional)) {
pe_rsc_trace(rsc, "%s will be mandatory because %s is",
key, start->uuid);
return false;
}
possible_matches = find_actions_exact(rsc->actions, key, node);
if (possible_matches == NULL) {
pe_rsc_trace(rsc, "%s will be mandatory because it is not active on %s",
key, pe__node_name(node));
return false;
}
for (const GList *iter = possible_matches;
iter != NULL; iter = iter->next) {
const pe_action_t *op = (const pe_action_t *) iter->data;
if (pcmk_is_set(op->flags, pe_action_reschedule)) {
pe_rsc_trace(rsc,
"%s will be mandatory because "
"it needs to be rescheduled", key);
g_list_free(possible_matches);
return false;
}
}
g_list_free(possible_matches);
return true;
}
/*!
* \internal
* \brief Create recurring action from resource history entry for an active role
*
* \param[in,out] rsc Resource that resource history is for
* \param[in,out] start Start action for \p rsc on \p node
* \param[in] node Node that resource will be active on (if any)
* \param[in] op Resource history entry
*/
static void
recurring_op_for_active(pe_resource_t *rsc, pe_action_t *start,
const pe_node_t *node, const struct op_history *op)
{
pe_action_t *mon = NULL;
bool is_optional = true;
const bool is_default_role = (op->role == RSC_ROLE_UNKNOWN);
// We're only interested in recurring actions for active roles
if (op->role == RSC_ROLE_STOPPED) {
return;
}
is_optional = active_recurring_should_be_optional(rsc, node, op->key,
start);
if ((!is_default_role && (rsc->next_role != op->role))
|| (is_default_role && (rsc->next_role == RSC_ROLE_PROMOTED))) {
// Configured monitor role doesn't match role resource will have
if (is_optional) { // It's running, so cancel it
char *after_key = NULL;
pe_action_t *cancel_op = pcmk__new_cancel_action(rsc, op->name,
op->interval_ms,
node);
switch (rsc->role) {
case RSC_ROLE_UNPROMOTED:
case RSC_ROLE_STARTED:
if (rsc->next_role == RSC_ROLE_PROMOTED) {
after_key = promote_key(rsc);
} else if (rsc->next_role == RSC_ROLE_STOPPED) {
after_key = stop_key(rsc);
}
break;
case RSC_ROLE_PROMOTED:
after_key = demote_key(rsc);
break;
default:
break;
}
if (after_key) {
pcmk__new_ordering(rsc, NULL, cancel_op, rsc, after_key, NULL,
pe_order_runnable_left, rsc->cluster);
}
}
do_crm_log((is_optional? LOG_INFO : LOG_TRACE),
"%s recurring action %s because %s configured for %s role "
"(not %s)",
(is_optional? "Cancelling" : "Ignoring"), op->key, op->id,
role2text(is_default_role? RSC_ROLE_UNPROMOTED : op->role),
role2text(rsc->next_role));
return;
}
pe_rsc_trace(rsc,
"Creating %s recurring action %s for %s (%s %s on %s)",
(is_optional? "optional" : "mandatory"), op->key,
op->id, rsc->id, role2text(rsc->next_role),
pe__node_name(node));
mon = custom_action(rsc, strdup(op->key), op->name, node, is_optional, TRUE,
rsc->cluster);
if (!pcmk_is_set(start->flags, pe_action_runnable)) {
pe_rsc_trace(rsc, "%s is unrunnable because start is", mon->uuid);
pe__clear_action_flags(mon, pe_action_runnable);
} else if ((node == NULL) || !node->details->online
|| node->details->unclean) {
pe_rsc_trace(rsc, "%s is unrunnable because no node is available",
mon->uuid);
pe__clear_action_flags(mon, pe_action_runnable);
} else if (!pcmk_is_set(mon->flags, pe_action_optional)) {
pe_rsc_info(rsc, "Start %s-interval %s for %s on %s",
pcmk__readable_interval(op->interval_ms), mon->task,
rsc->id, pe__node_name(node));
}
if (rsc->next_role == RSC_ROLE_PROMOTED) {
pe__add_action_expected_result(mon, CRM_EX_PROMOTED);
}
// Order monitor relative to other actions
if ((node == NULL) || pcmk_is_set(rsc->flags, pe_rsc_managed)) {
pcmk__new_ordering(rsc, start_key(rsc), NULL,
NULL, strdup(mon->uuid), mon,
pe_order_implies_then|pe_order_runnable_left,
rsc->cluster);
pcmk__new_ordering(rsc, reload_key(rsc), NULL,
NULL, strdup(mon->uuid), mon,
pe_order_implies_then|pe_order_runnable_left,
rsc->cluster);
if (rsc->next_role == RSC_ROLE_PROMOTED) {
pcmk__new_ordering(rsc, promote_key(rsc), NULL,
rsc, NULL, mon,
pe_order_optional|pe_order_runnable_left,
rsc->cluster);
} else if (rsc->role == RSC_ROLE_PROMOTED) {
pcmk__new_ordering(rsc, demote_key(rsc), NULL,
rsc, NULL, mon,
pe_order_optional|pe_order_runnable_left,
rsc->cluster);
}
}
}
/*!
* \internal
* \brief Cancel a recurring action if running on a node
*
* \param[in,out] rsc Resource that action is for
* \param[in] node Node to cancel action on
* \param[in] key Operation key for action
* \param[in] name Action name
* \param[in] interval_ms Action interval (in milliseconds)
*/
static void
cancel_if_running(pe_resource_t *rsc, const pe_node_t *node, const char *key,
const char *name, guint interval_ms)
{
GList *possible_matches = find_actions_exact(rsc->actions, key, node);
pe_action_t *cancel_op = NULL;
if (possible_matches == NULL) {
return; // Recurring action isn't running on this node
}
g_list_free(possible_matches);
cancel_op = pcmk__new_cancel_action(rsc, name, interval_ms, node);
switch (rsc->next_role) {
case RSC_ROLE_STARTED:
case RSC_ROLE_UNPROMOTED:
/* Order starts after cancel. If the current role is
* stopped, this cancels the monitor before the resource
* starts; if the current role is started, then this cancels
* the monitor on a migration target before starting there.
*/
pcmk__new_ordering(rsc, NULL, cancel_op,
rsc, start_key(rsc), NULL,
pe_order_runnable_left, rsc->cluster);
break;
default:
break;
}
pe_rsc_info(rsc,
"Cancelling %s-interval %s action for %s on %s because "
"configured for " RSC_ROLE_STOPPED_S " role (not %s)",
pcmk__readable_interval(interval_ms), name, rsc->id,
pe__node_name(node), role2text(rsc->next_role));
}
/*!
* \internal
* \brief Order an action after all probes of a resource on a node
*
* \param[in,out] rsc Resource to check for probes
* \param[in] node Node to check for probes of \p rsc
* \param[in,out] action Action to order after probes of \p rsc on \p node
*/
static void
order_after_probes(pe_resource_t *rsc, const pe_node_t *node,
pe_action_t *action)
{
GList *probes = pe__resource_actions(rsc, node, PCMK_ACTION_MONITOR, FALSE);
for (GList *iter = probes; iter != NULL; iter = iter->next) {
order_actions((pe_action_t *) iter->data, action,
pe_order_runnable_left);
}
g_list_free(probes);
}
/*!
* \internal
* \brief Order an action after all stops of a resource on a node
*
* \param[in,out] rsc Resource to check for stops
* \param[in] node Node to check for stops of \p rsc
* \param[in,out] action Action to order after stops of \p rsc on \p node
*/
static void
order_after_stops(pe_resource_t *rsc, const pe_node_t *node,
pe_action_t *action)
{
GList *stop_ops = pe__resource_actions(rsc, node, PCMK_ACTION_STOP, TRUE);
for (GList *iter = stop_ops; iter != NULL; iter = iter->next) {
pe_action_t *stop = (pe_action_t *) iter->data;
if (!pcmk_is_set(stop->flags, pe_action_optional)
&& !pcmk_is_set(action->flags, pe_action_optional)
&& !pcmk_is_set(rsc->flags, pe_rsc_managed)) {
pe_rsc_trace(rsc, "%s optional on %s: unmanaged",
action->uuid, pe__node_name(node));
pe__set_action_flags(action, pe_action_optional);
}
if (!pcmk_is_set(stop->flags, pe_action_runnable)) {
crm_debug("%s unrunnable on %s: stop is unrunnable",
action->uuid, pe__node_name(node));
pe__clear_action_flags(action, pe_action_runnable);
}
if (pcmk_is_set(rsc->flags, pe_rsc_managed)) {
pcmk__new_ordering(rsc, stop_key(rsc), stop,
NULL, NULL, action,
pe_order_implies_then|pe_order_runnable_left,
rsc->cluster);
}
}
g_list_free(stop_ops);
}
/*!
* \internal
* \brief Create recurring action from resource history entry for inactive role
*
* \param[in,out] rsc Resource that resource history is for
* \param[in] node Node that resource will be active on (if any)
* \param[in] op Resource history entry
*/
static void
recurring_op_for_inactive(pe_resource_t *rsc, const pe_node_t *node,
const struct op_history *op)
{
GList *possible_matches = NULL;
// We're only interested in recurring actions for the inactive role
if (op->role != RSC_ROLE_STOPPED) {
return;
}
if (!pcmk_is_set(rsc->flags, pe_rsc_unique)) {
crm_notice("Ignoring %s (recurring monitors for " RSC_ROLE_STOPPED_S
" role are not supported for anonymous clones)", op->id);
return; // @TODO add support
}
pe_rsc_trace(rsc, "Creating recurring action %s for %s on nodes "
"where it should not be running", op->id, rsc->id);
for (GList *iter = rsc->cluster->nodes; iter != NULL; iter = iter->next) {
pe_node_t *stop_node = (pe_node_t *) iter->data;
bool is_optional = true;
pe_action_t *stopped_mon = NULL;
// Cancel action on node where resource will be active
if ((node != NULL)
&& pcmk__str_eq(stop_node->details->uname, node->details->uname,
pcmk__str_casei)) {
cancel_if_running(rsc, node, op->key, op->name, op->interval_ms);
continue;
}
// Recurring action on this node is optional if it's already active here
possible_matches = find_actions_exact(rsc->actions, op->key, stop_node);
is_optional = (possible_matches != NULL);
g_list_free(possible_matches);
pe_rsc_trace(rsc,
"Creating %s recurring action %s for %s (%s "
RSC_ROLE_STOPPED_S " on %s)",
(is_optional? "optional" : "mandatory"),
op->key, op->id, rsc->id, pe__node_name(stop_node));
stopped_mon = custom_action(rsc, strdup(op->key), op->name, stop_node,
is_optional, TRUE, rsc->cluster);
pe__add_action_expected_result(stopped_mon, CRM_EX_NOT_RUNNING);
if (pcmk_is_set(rsc->flags, pe_rsc_managed)) {
order_after_probes(rsc, stop_node, stopped_mon);
}
/* The recurring action is for the inactive role, so it shouldn't be
* performed until the resource is inactive.
*/
order_after_stops(rsc, stop_node, stopped_mon);
if (!stop_node->details->online || stop_node->details->unclean) {
pe_rsc_debug(rsc, "%s unrunnable on %s: node unavailable)",
stopped_mon->uuid, pe__node_name(stop_node));
pe__clear_action_flags(stopped_mon, pe_action_runnable);
}
if (pcmk_is_set(stopped_mon->flags, pe_action_runnable)
&& !pcmk_is_set(stopped_mon->flags, pe_action_optional)) {
crm_notice("Start recurring %s-interval %s for "
RSC_ROLE_STOPPED_S " %s on %s",
pcmk__readable_interval(op->interval_ms),
stopped_mon->task, rsc->id, pe__node_name(stop_node));
}
}
}
/*!
* \internal
* \brief Create recurring actions for a resource
*
* \param[in,out] rsc Resource to create recurring actions for
*/
void
pcmk__create_recurring_actions(pe_resource_t *rsc)
{
pe_action_t *start = NULL;
if (pcmk_is_set(rsc->flags, pe_rsc_block)) {
pe_rsc_trace(rsc, "Skipping recurring actions for blocked resource %s",
rsc->id);
return;
}
if (pcmk_is_set(rsc->flags, pe_rsc_maintenance)) {
pe_rsc_trace(rsc, "Skipping recurring actions for %s "
"in maintenance mode", rsc->id);
return;
}
if (rsc->allocated_to == NULL) {
// Recurring actions for active roles not needed
} else if (rsc->allocated_to->details->maintenance) {
pe_rsc_trace(rsc,
"Skipping recurring actions for %s on %s "
"in maintenance mode",
rsc->id, pe__node_name(rsc->allocated_to));
} else if ((rsc->next_role != RSC_ROLE_STOPPED)
|| !pcmk_is_set(rsc->flags, pe_rsc_managed)) {
// Recurring actions for active roles needed
start = start_action(rsc, rsc->allocated_to, TRUE);
}
pe_rsc_trace(rsc, "Creating any recurring actions needed for %s", rsc->id);
for (xmlNode *op = first_named_child(rsc->ops_xml, "op");
op != NULL; op = crm_next_same_xml(op)) {
struct op_history op_history = { NULL, };
if (!is_recurring_history(rsc, op, &op_history)) {
continue;
}
if (start != NULL) {
recurring_op_for_active(rsc, start, rsc->allocated_to, &op_history);
}
recurring_op_for_inactive(rsc, rsc->allocated_to, &op_history);
free(op_history.key);
}
}
/*!
* \internal
* \brief Create an executor cancel action
*
* \param[in,out] rsc Resource of action to cancel
* \param[in] task Name of action to cancel
* \param[in] interval_ms Interval of action to cancel
* \param[in] node Node of action to cancel
*
* \return Created op
*/
pe_action_t *
pcmk__new_cancel_action(pe_resource_t *rsc, const char *task, guint interval_ms,
const pe_node_t *node)
{
pe_action_t *cancel_op = NULL;
char *key = NULL;
char *interval_ms_s = NULL;
CRM_ASSERT((rsc != NULL) && (task != NULL) && (node != NULL));
// @TODO dangerous if possible to schedule another action with this key
key = pcmk__op_key(rsc->id, task, interval_ms);
cancel_op = custom_action(rsc, key, RSC_CANCEL, node, FALSE, TRUE,
rsc->cluster);
pcmk__str_update(&cancel_op->task, RSC_CANCEL);
pcmk__str_update(&cancel_op->cancel_task, task);
interval_ms_s = crm_strdup_printf("%u", interval_ms);
add_hash_param(cancel_op->meta, XML_LRM_ATTR_TASK, task);
add_hash_param(cancel_op->meta, XML_LRM_ATTR_INTERVAL_MS, interval_ms_s);
free(interval_ms_s);
return cancel_op;
}
/*!
* \internal
* \brief Schedule cancellation of a recurring action
*
* \param[in,out] rsc Resource that action is for
* \param[in] call_id Action's call ID from history
* \param[in] task Action name
* \param[in] interval_ms Action interval
* \param[in] node Node that history entry is for
* \param[in] reason Short description of why action is cancelled
*/
void
pcmk__schedule_cancel(pe_resource_t *rsc, const char *call_id, const char *task,
guint interval_ms, const pe_node_t *node,
const char *reason)
{
pe_action_t *cancel = NULL;
CRM_CHECK((rsc != NULL) && (task != NULL)
&& (node != NULL) && (reason != NULL),
return);
crm_info("Recurring %s-interval %s for %s will be stopped on %s: %s",
pcmk__readable_interval(interval_ms), task, rsc->id,
pe__node_name(node), reason);
cancel = pcmk__new_cancel_action(rsc, task, interval_ms, node);
add_hash_param(cancel->meta, XML_LRM_ATTR_CALLID, call_id);
// Cancellations happen after stops
pcmk__new_ordering(rsc, stop_key(rsc), NULL, rsc, NULL, cancel,
pe_order_optional, rsc->cluster);
}
/*!
* \internal
* \brief Reschedule a recurring action
*
* \param[in,out] rsc Resource that action is for
* \param[in] task Name of action being rescheduled
* \param[in] interval_ms Action interval (in milliseconds)
* \param[in,out] node Node where action should be rescheduled
*/
void
pcmk__reschedule_recurring(pe_resource_t *rsc, const char *task,
guint interval_ms, pe_node_t *node)
{
pe_action_t *op = NULL;
trigger_unfencing(rsc, node, "Device parameters changed (reschedule)",
NULL, rsc->cluster);
op = custom_action(rsc, pcmk__op_key(rsc->id, task, interval_ms),
task, node, TRUE, TRUE, rsc->cluster);
pe__set_action_flags(op, pe_action_reschedule);
}
/*!
* \internal
* \brief Check whether an action is recurring
*
* \param[in] action Action to check
*
* \return true if \p action has a nonzero interval, otherwise false
*/
bool
pcmk__action_is_recurring(const pe_action_t *action)
{
guint interval_ms = 0;
if (pcmk__guint_from_hash(action->meta,
XML_LRM_ATTR_INTERVAL_MS, 0,
&interval_ms) != pcmk_rc_ok) {
return false;
}
return (interval_ms > 0);
}

File Metadata

Mime Type
text/x-diff
Expires
Tue, Jul 8, 6:19 PM (16 h, 34 m)
Storage Engine
blob
Storage Format
Raw Data
Storage Handle
2002581
Default Alt Text
(306 KB)

Event Timeline