diff --git a/daemons/attrd/attrd_attributes.c b/daemons/attrd/attrd_attributes.c index cb4ca9b8b1..3e98a794c6 100644 --- a/daemons/attrd/attrd_attributes.c +++ b/daemons/attrd/attrd_attributes.c @@ -1,212 +1,215 @@ /* * Copyright 2013-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include #include #include "pacemaker-attrd.h" static attribute_t * attrd_create_attribute(xmlNode *xml) { int is_private = 0; - int dampen = 0; + long long dampen = 0; + const char *readable_delay = "no"; const char *name = crm_element_value(xml, PCMK__XA_ATTR_NAME); const char *set_type = crm_element_value(xml, PCMK__XA_ATTR_SET_TYPE); const char *dampen_s = crm_element_value(xml, PCMK__XA_ATTR_DAMPENING); attribute_t *a = NULL; if (set_type == NULL) { set_type = PCMK_XE_INSTANCE_ATTRIBUTES; } /* Set type is meaningful only when writing to the CIB. Private * attributes are not written. */ crm_element_value_int(xml, PCMK__XA_ATTR_IS_PRIVATE, &is_private); if ((is_private != 0) && !pcmk__str_any_of(set_type, PCMK_XE_INSTANCE_ATTRIBUTES, PCMK_XE_UTILIZATION, NULL)) { crm_warn("Ignoring attribute %s with invalid set type %s", pcmk__s(name, "(unidentified)"), set_type); return NULL; } a = calloc(1, sizeof(attribute_t)); CRM_ASSERT(a != NULL); a->is_private = is_private; pcmk__str_update(&a->id, name); pcmk__str_update(&a->set_type, set_type); a->set_id = crm_element_value_copy(xml, PCMK__XA_ATTR_SET); a->uuid = crm_element_value_copy(xml, PCMK__XA_ATTR_UUID); a->values = pcmk__strikey_table(NULL, attrd_free_attribute_value); a->user = crm_element_value_copy(xml, PCMK__XA_ATTR_USER); crm_trace("Performing all %s operations as user '%s'", a->id, a->user); if (dampen_s != NULL) { dampen = crm_get_msec(dampen_s); } - crm_trace("Created attribute %s with %s write delay", a->id, - (a->timeout_ms == 0)? "no" : pcmk__readable_interval(a->timeout_ms)); - if(dampen > 0) { - a->timeout_ms = dampen; + if (dampen > 0) { + a->timeout_ms = (int) QB_MIN(dampen, INT_MAX); a->timer = attrd_add_timer(a->id, a->timeout_ms, a); + readable_delay = pcmk__readable_interval(a->timeout_ms); } else if (dampen < 0) { crm_warn("Ignoring invalid delay %s for attribute %s", dampen_s, a->id); } + crm_trace("Created attribute %s with %s write delay", + a->id, readable_delay); + g_hash_table_replace(attributes, a->id, a); return a; } static int attrd_update_dampening(attribute_t *a, xmlNode *xml, const char *attr) { const char *dvalue = crm_element_value(xml, PCMK__XA_ATTR_DAMPENING); - int dampen = 0; + long long dampen = 0; if (dvalue == NULL) { crm_warn("Could not update %s: peer did not specify value for delay", attr); return EINVAL; } dampen = crm_get_msec(dvalue); if (dampen < 0) { crm_warn("Could not update %s: invalid delay value %dms (%s)", attr, dampen, dvalue); return EINVAL; } if (a->timeout_ms != dampen) { mainloop_timer_del(a->timer); - a->timeout_ms = dampen; + a->timeout_ms = (int) QB_MIN(dampen, INT_MAX); if (dampen > 0) { a->timer = attrd_add_timer(attr, a->timeout_ms, a); crm_info("Update attribute %s delay to %dms (%s)", attr, dampen, dvalue); } else { a->timer = NULL; crm_info("Update attribute %s to remove delay", attr); } /* If dampening changed, do an immediate write-out, * otherwise repeated dampening changes would prevent write-outs */ attrd_write_or_elect_attribute(a); } return pcmk_rc_ok; } GHashTable *attributes = NULL; /*! * \internal * \brief Create an XML representation of an attribute for use in peer messages * * \param[in,out] parent Create attribute XML as child element of this * \param[in] a Attribute to represent * \param[in] v Attribute value to represent * \param[in] force_write If true, value should be written even if unchanged * * \return XML representation of attribute */ xmlNode * attrd_add_value_xml(xmlNode *parent, const attribute_t *a, const attribute_value_t *v, bool force_write) { xmlNode *xml = create_xml_node(parent, __func__); crm_xml_add(xml, PCMK__XA_ATTR_NAME, a->id); crm_xml_add(xml, PCMK__XA_ATTR_SET, a->set_id); crm_xml_add(xml, PCMK__XA_ATTR_UUID, a->uuid); crm_xml_add(xml, PCMK__XA_ATTR_USER, a->user); pcmk__xe_add_node(xml, v->nodename, v->nodeid); if (pcmk_is_set(v->flags, attrd_value_remote)) { crm_xml_add_int(xml, PCMK__XA_ATTR_IS_REMOTE, 1); } crm_xml_add(xml, PCMK__XA_ATTR_VALUE, v->current); crm_xml_add_int(xml, PCMK__XA_ATTR_DAMPENING, a->timeout_ms / 1000); crm_xml_add_int(xml, PCMK__XA_ATTR_IS_PRIVATE, a->is_private); crm_xml_add_int(xml, PCMK__XA_ATTR_FORCE, force_write); return xml; } void attrd_clear_value_seen(void) { GHashTableIter aIter; GHashTableIter vIter; attribute_t *a; attribute_value_t *v = NULL; g_hash_table_iter_init(&aIter, attributes); while (g_hash_table_iter_next(&aIter, NULL, (gpointer *) & a)) { g_hash_table_iter_init(&vIter, a->values); while (g_hash_table_iter_next(&vIter, NULL, (gpointer *) & v)) { attrd_clear_value_flags(v, attrd_value_from_peer); } } } attribute_t * attrd_populate_attribute(xmlNode *xml, const char *attr) { attribute_t *a = NULL; bool update_both = false; const char *op = crm_element_value(xml, PCMK__XA_TASK); // NULL because PCMK__ATTRD_CMD_SYNC_RESPONSE has no PCMK__XA_TASK update_both = pcmk__str_eq(op, PCMK__ATTRD_CMD_UPDATE_BOTH, pcmk__str_null_matches); // Look up or create attribute entry a = g_hash_table_lookup(attributes, attr); if (a == NULL) { if (update_both || pcmk__str_eq(op, PCMK__ATTRD_CMD_UPDATE, pcmk__str_none)) { a = attrd_create_attribute(xml); if (a == NULL) { return NULL; } } else { crm_warn("Could not update %s: attribute not found", attr); return NULL; } } // Update attribute dampening if (update_both || pcmk__str_eq(op, PCMK__ATTRD_CMD_UPDATE_DELAY, pcmk__str_none)) { int rc = attrd_update_dampening(a, xml, attr); if (rc != pcmk_rc_ok || !update_both) { return NULL; } } return a; } diff --git a/daemons/controld/controld_execd.c b/daemons/controld/controld_execd.c index 464e3fce5a..0bd66b4147 100644 --- a/daemons/controld/controld_execd.c +++ b/daemons/controld/controld_execd.c @@ -1,2447 +1,2449 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include // lrmd_event_data_t, lrmd_rsc_info_t, etc. #include #include #include #include #include #include #include #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("Lost connection to local executor"); register_fsa_input(C_FSA_INTERNAL, I_ERROR, NULL); controld_clear_fsa_input_flags(R_LRM_CONNECTED); } } 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, 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); lrm_state_disconnect(lrm_state); lrm_state_reset_tables(lrm_state, FALSE); } 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, PCMK_ACTION_MIGRATE_TO, 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, PCMK_XA_ID, entry->id); crm_xml_add(xml_rsc, PCMK_XA_TYPE, entry->rsc.type); crm_xml_add(xml_rsc, PCMK_XA_CLASS, entry->rsc.standard); crm_xml_add(xml_rsc, PCMK_XA_PROVIDER, entry->rsc.provider); if (entry->last && entry->last->params) { static const char *name = CRM_META "_" PCMK__META_CONTAINER; const char *container = g_hash_table_lookup(entry->last->params, name); if (container) { crm_trace("Resource %s is a part of container resource %s", entry->id, container); crm_xml_add(xml_rsc, PCMK__META_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 = pcmk__get_node(0, lrm_state->node_name, NULL, pcmk__node_search_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, PCMK_XA_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, PCMK_XE_CRM_CONFIG, 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, PCMK__XA_CRM_SYS_FROM); const char *from_host = crm_element_value(input->msg, PCMK__XA_SRC); 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, PCMK_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, PCMK__XA_LONG_ID); 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, PCMK_XA_CLASS); const char *provider = crm_element_value(rsc_xml, PCMK_XA_PROVIDER); const char *type = crm_element_value(rsc_xml, PCMK_XA_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, PCMK_XA_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, PCMK_XA_OPERATION); const char *target_node = crm_element_value(action, PCMK__META_ON_NODE); xmlNode *xml_rsc = find_xml_node(action, PCMK_XE_PRIMITIVE, 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, PCMK__XA_OPERATION_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, PCMK__META_ON_NODE); } 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, PCMK_XE_PRIMITIVE, 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, PCMK__XE_ATTRIBUTES, TRUE); CRM_CHECK(params != NULL, return FALSE); meta_key = crm_meta_name(PCMK_XA_OPERATION); op_task = crm_element_value(params, meta_key); free(meta_key); CRM_CHECK(op_task != NULL, return FALSE); meta_key = crm_meta_name(PCMK_META_INTERVAL); 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(PCMK__XA_CALL_ID); 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, PCMK_XA_CRM_FEATURE_SET); 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, PCMK_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); } if (!pcmk_is_set(controld_globals.fsa_input_register, R_HA_DISCONNECTED)) { 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, PCMK__XA_CRM_USER, NULL); crm_op = crm_element_value(input->msg, PCMK__XA_CRM_TASK); from_sys = crm_element_value(input->msg, PCMK__XA_CRM_SYS_FROM); if (!pcmk__str_eq(from_sys, CRM_SYSTEM_TENGINE, pcmk__str_none)) { from_host = crm_element_value(input->msg, PCMK__XA_SRC); } if (pcmk__str_eq(crm_op, PCMK_ACTION_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 = PCMK_ACTION_DELETE; } else if (input->xml != NULL) { operation = crm_element_value(input->xml, PCMK_XA_OPERATION); } 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, PCMK_VALUE_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, PCMK__META_ON_NODE); } 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, PCMK_XE_PRIMITIVE, TRUE); gboolean create_rsc = !pcmk__str_eq(operation, PCMK_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, PCMK_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, PCMK_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(PCMK_XA_CRM_FEATURE_SET), 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 "_" PCMK__META_OP_TARGET_RC); op_delay = crm_meta_value(params, PCMK_META_START_DELAY); pcmk__scan_min_int(op_delay, &op->start_delay, 0); op_timeout = crm_meta_value(params, PCMK_META_TIMEOUT); pcmk__scan_min_int(op_timeout, &op->timeout, 0); if (pcmk__guint_from_hash(params, CRM_META "_" PCMK_META_INTERVAL, 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, PCMK_XE_PRIMITIVE, FALSE); class = crm_element_value(primitive, PCMK_XA_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); + long long timeout_ms = crm_get_msec(op_timeout); + + op->timeout = (int) QB_MIN(timeout_ms, INT_MAX); } } 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, PCMK__XA_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 = pcmk__get_node(0, controld_globals.our_nodename, NULL, pcmk__node_search_cluster); update = create_node_state_update(peer, node_update_none, NULL, __func__); iter = create_xml_node(update, XML_CIB_TAG_LRM); crm_xml_add(iter, PCMK_XA_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, PCMK_XA_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, PCMK_XA_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, PCMK_ACTION_MIGRATE_TO) == 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, PCMK_XA_OPERATION); CRM_CHECK(!pcmk__str_empty(operation), return); transition = crm_element_value(msg, PCMK__XA_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, 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 = 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, pcmk__readable_action(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__flag_text(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, PCMK_OPT_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 ", pcmk__readable_action(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, PCMK_XE_PRIMITIVE, TRUE); const char *standard = crm_element_value(xml, PCMK_XA_CLASS); const char *provider = crm_element_value(xml, PCMK_XA_PROVIDER); const char *type = crm_element_value(xml, PCMK_XA_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, PCMK__META_ON_NODE); } 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_fencing.c b/daemons/controld/controld_fencing.c index 373ecc5ba4..8736a1f96e 100644 --- a/daemons/controld/controld_fencing.c +++ b/daemons/controld/controld_fencing.c @@ -1,1115 +1,1115 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include static void tengine_stonith_history_synced(stonith_t *st, stonith_event_t *st_event); /* * stonith failure counting * * We don't want to get stuck in a permanent fencing loop. Keep track of the * number of fencing failures for each target node, and the most we'll restart a * transition for. */ struct st_fail_rec { int count; }; static bool fence_reaction_panic = false; static unsigned long int stonith_max_attempts = 10; static GHashTable *stonith_failures = NULL; /*! * \internal * \brief Update max fencing attempts before giving up * * \param[in] value New max fencing attempts */ static void update_stonith_max_attempts(const char *value) { stonith_max_attempts = char2score(value); if (stonith_max_attempts < 1UL) { stonith_max_attempts = 10UL; } } /*! * \internal * \brief Configure reaction to notification of local node being fenced * * \param[in] reaction_s Reaction type */ static void set_fence_reaction(const char *reaction_s) { if (pcmk__str_eq(reaction_s, "panic", pcmk__str_casei)) { fence_reaction_panic = true; } else { if (!pcmk__str_eq(reaction_s, "stop", pcmk__str_casei)) { crm_warn("Invalid value '%s' for %s, using 'stop'", reaction_s, PCMK_OPT_FENCE_REACTION); } fence_reaction_panic = false; } } /*! * \internal * \brief Configure fencing options based on the CIB * * \param[in,out] options Name/value pairs for configured options */ void controld_configure_fencing(GHashTable *options) { const char *value = NULL; value = g_hash_table_lookup(options, PCMK_OPT_FENCE_REACTION); set_fence_reaction(value); value = g_hash_table_lookup(options, PCMK_OPT_STONITH_MAX_ATTEMPTS); update_stonith_max_attempts(value); } static gboolean too_many_st_failures(const char *target) { GHashTableIter iter; const char *key = NULL; struct st_fail_rec *value = NULL; if (stonith_failures == NULL) { return FALSE; } if (target == NULL) { g_hash_table_iter_init(&iter, stonith_failures); while (g_hash_table_iter_next(&iter, (gpointer *) &key, (gpointer *) &value)) { if (value->count >= stonith_max_attempts) { target = (const char*)key; goto too_many; } } } else { value = g_hash_table_lookup(stonith_failures, target); if ((value != NULL) && (value->count >= stonith_max_attempts)) { goto too_many; } } return FALSE; too_many: crm_warn("Too many failures (%d) to fence %s, giving up", value->count, target); return TRUE; } /*! * \internal * \brief Reset a stonith fail count * * \param[in] target Name of node to reset, or NULL for all */ void st_fail_count_reset(const char *target) { if (stonith_failures == NULL) { return; } if (target) { struct st_fail_rec *rec = NULL; rec = g_hash_table_lookup(stonith_failures, target); if (rec) { rec->count = 0; } } else { GHashTableIter iter; const char *key = NULL; struct st_fail_rec *rec = NULL; g_hash_table_iter_init(&iter, stonith_failures); while (g_hash_table_iter_next(&iter, (gpointer *) &key, (gpointer *) &rec)) { rec->count = 0; } } } static void st_fail_count_increment(const char *target) { struct st_fail_rec *rec = NULL; if (stonith_failures == NULL) { stonith_failures = pcmk__strkey_table(free, free); } rec = g_hash_table_lookup(stonith_failures, target); if (rec) { rec->count++; } else { rec = malloc(sizeof(struct st_fail_rec)); if(rec == NULL) { return; } rec->count = 1; g_hash_table_insert(stonith_failures, strdup(target), rec); } } /* end stonith fail count functions */ static void cib_fencing_updated(xmlNode *msg, int call_id, int rc, xmlNode *output, void *user_data) { if (rc < pcmk_ok) { crm_err("Fencing update %d for %s: failed - %s (%d)", call_id, (char *)user_data, pcmk_strerror(rc), rc); crm_log_xml_warn(msg, "Failed update"); abort_transition(INFINITY, pcmk__graph_shutdown, "CIB update failed", NULL); } else { crm_info("Fencing update %d for %s: complete", call_id, (char *)user_data); } } static void send_stonith_update(pcmk__graph_action_t *action, const char *target, const char *uuid) { int rc = pcmk_ok; crm_node_t *peer = NULL; /* We (usually) rely on the membership layer to do node_update_cluster, * and the peer status callback to do node_update_peer, because the node * might have already rejoined before we get the stonith result here. */ int flags = node_update_join | node_update_expected; /* zero out the node-status & remove all LRM status info */ xmlNode *node_state = NULL; CRM_CHECK(target != NULL, return); CRM_CHECK(uuid != NULL, return); /* Make sure the membership and join caches are accurate. * Try getting any existing node cache entry also by node uuid in case it * doesn't have an uname yet. */ peer = pcmk__get_node(0, target, uuid, pcmk__node_search_any); CRM_CHECK(peer != NULL, return); if (peer->state == NULL) { /* Usually, we rely on the membership layer to update the cluster state * in the CIB. However, if the node has never been seen, do it here, so * the node is not considered unclean. */ flags |= node_update_cluster; } if (peer->uuid == NULL) { crm_info("Recording uuid '%s' for node '%s'", uuid, target); peer->uuid = strdup(uuid); } crmd_peer_down(peer, TRUE); /* Generate a node state update for the CIB */ node_state = create_node_state_update(peer, flags, NULL, __func__); /* we have to mark whether or not remote nodes have already been fenced */ if (peer->flags & crm_remote_node) { char *now_s = pcmk__ttoa(time(NULL)); crm_xml_add(node_state, XML_NODE_IS_FENCED, now_s); free(now_s); } /* Force our known ID */ crm_xml_add(node_state, PCMK_XA_ID, uuid); rc = controld_globals.cib_conn->cmds->modify(controld_globals.cib_conn, PCMK_XE_STATUS, node_state, cib_scope_local |cib_can_create); /* Delay processing the trigger until the update completes */ crm_debug("Sending fencing update %d for %s", rc, target); fsa_register_cib_callback(rc, strdup(target), cib_fencing_updated); // Make sure it sticks /* controld_globals.cib_conn->cmds->bump_epoch(controld_globals.cib_conn, * cib_scope_local); */ controld_delete_node_state(peer->uname, controld_section_all, cib_scope_local); free_xml(node_state); return; } /*! * \internal * \brief Abort transition due to stonith failure * * \param[in] abort_action Whether to restart or stop transition * \param[in] target Don't restart if this (NULL for any) has too many failures * \param[in] reason Log this stonith action XML as abort reason (or NULL) */ static void abort_for_stonith_failure(enum pcmk__graph_next abort_action, const char *target, const xmlNode *reason) { /* If stonith repeatedly fails, we eventually give up on starting a new * transition for that reason. */ if ((abort_action != pcmk__graph_wait) && too_many_st_failures(target)) { abort_action = pcmk__graph_wait; } abort_transition(INFINITY, abort_action, "Stonith failed", reason); } /* * stonith cleanup list * * If the DC is shot, proper notifications might not go out. * The stonith cleanup list allows the cluster to (re-)send * notifications once a new DC is elected. */ static GList *stonith_cleanup_list = NULL; /*! * \internal * \brief Add a node to the stonith cleanup list * * \param[in] target Name of node to add */ void add_stonith_cleanup(const char *target) { stonith_cleanup_list = g_list_append(stonith_cleanup_list, strdup(target)); } /*! * \internal * \brief Remove a node from the stonith cleanup list * * \param[in] Name of node to remove */ void remove_stonith_cleanup(const char *target) { GList *iter = stonith_cleanup_list; while (iter != NULL) { GList *tmp = iter; char *iter_name = tmp->data; iter = iter->next; if (pcmk__str_eq(target, iter_name, pcmk__str_casei)) { crm_trace("Removing %s from the cleanup list", iter_name); stonith_cleanup_list = g_list_delete_link(stonith_cleanup_list, tmp); free(iter_name); } } } /*! * \internal * \brief Purge all entries from the stonith cleanup list */ void purge_stonith_cleanup(void) { if (stonith_cleanup_list) { GList *iter = NULL; for (iter = stonith_cleanup_list; iter != NULL; iter = iter->next) { char *target = iter->data; crm_info("Purging %s from stonith cleanup list", target); free(target); } g_list_free(stonith_cleanup_list); stonith_cleanup_list = NULL; } } /*! * \internal * \brief Send stonith updates for all entries in cleanup list, then purge it */ void execute_stonith_cleanup(void) { GList *iter; for (iter = stonith_cleanup_list; iter != NULL; iter = iter->next) { char *target = iter->data; crm_node_t *target_node = pcmk__get_node(0, target, NULL, pcmk__node_search_cluster); const char *uuid = crm_peer_uuid(target_node); crm_notice("Marking %s, target of a previous stonith action, as clean", target); send_stonith_update(NULL, target, uuid); free(target); } g_list_free(stonith_cleanup_list); stonith_cleanup_list = NULL; } /* end stonith cleanup list functions */ /* stonith API client * * Functions that need to interact directly with the fencer via its API */ static stonith_t *stonith_api = NULL; static mainloop_timer_t *controld_fencer_connect_timer = NULL; static char *te_client_id = NULL; static gboolean fail_incompletable_stonith(pcmk__graph_t *graph) { GList *lpc = NULL; const char *task = NULL; xmlNode *last_action = NULL; if (graph == NULL) { return FALSE; } for (lpc = graph->synapses; lpc != NULL; lpc = lpc->next) { GList *lpc2 = NULL; pcmk__graph_synapse_t *synapse = (pcmk__graph_synapse_t *) lpc->data; if (pcmk_is_set(synapse->flags, pcmk__synapse_confirmed)) { continue; } for (lpc2 = synapse->actions; lpc2 != NULL; lpc2 = lpc2->next) { pcmk__graph_action_t *action = (pcmk__graph_action_t *) lpc2->data; if ((action->type != pcmk__cluster_graph_action) || pcmk_is_set(action->flags, pcmk__graph_action_confirmed)) { continue; } task = crm_element_value(action->xml, PCMK_XA_OPERATION); if (pcmk__str_eq(task, PCMK_ACTION_STONITH, pcmk__str_casei)) { pcmk__set_graph_action_flags(action, pcmk__graph_action_failed); last_action = action->xml; pcmk__update_graph(graph, action); crm_notice("Failing action %d (%s): fencer terminated", action->id, ID(action->xml)); } } } if (last_action != NULL) { crm_warn("Fencer failure resulted in unrunnable actions"); abort_for_stonith_failure(pcmk__graph_restart, NULL, last_action); return TRUE; } return FALSE; } static void tengine_stonith_connection_destroy(stonith_t *st, stonith_event_t *e) { te_cleanup_stonith_history_sync(st, FALSE); if (pcmk_is_set(controld_globals.fsa_input_register, R_ST_REQUIRED)) { crm_err("Lost fencer connection (will attempt to reconnect)"); if (!mainloop_timer_running(controld_fencer_connect_timer)) { mainloop_timer_start(controld_fencer_connect_timer); } } else { crm_info("Disconnected from fencer"); } if (stonith_api) { /* the client API won't properly reconnect notifications * if they are still in the table - so remove them */ if (stonith_api->state != stonith_disconnected) { stonith_api->cmds->disconnect(st); } stonith_api->cmds->remove_notification(stonith_api, NULL); } if (AM_I_DC) { fail_incompletable_stonith(controld_globals.transition_graph); trigger_graph(); } } /*! * \internal * \brief Handle an event notification from the fencing API * * \param[in] st Fencing API connection (ignored) * \param[in] event Fencing API event notification */ static void handle_fence_notification(stonith_t *st, stonith_event_t *event) { bool succeeded = true; const char *executioner = "the cluster"; const char *client = "a client"; const char *reason = NULL; int exec_status; if (te_client_id == NULL) { te_client_id = crm_strdup_printf("%s.%lu", crm_system_name, (unsigned long) getpid()); } if (event == NULL) { crm_err("Notify data not found"); return; } if (event->executioner != NULL) { executioner = event->executioner; } if (event->client_origin != NULL) { client = event->client_origin; } exec_status = stonith__event_execution_status(event); if ((stonith__event_exit_status(event) != CRM_EX_OK) || (exec_status != PCMK_EXEC_DONE)) { succeeded = false; if (exec_status == PCMK_EXEC_DONE) { exec_status = PCMK_EXEC_ERROR; } } reason = stonith__event_exit_reason(event); crmd_alert_fencing_op(event); if (pcmk__str_eq(PCMK_ACTION_ON, event->action, pcmk__str_none)) { // Unfencing doesn't need special handling, just a log message if (succeeded) { crm_notice("%s was unfenced by %s at the request of %s@%s", event->target, executioner, client, event->origin); } else { crm_err("Unfencing of %s by %s failed (%s%s%s) with exit status %d", event->target, executioner, pcmk_exec_status_str(exec_status), ((reason == NULL)? "" : ": "), ((reason == NULL)? "" : reason), stonith__event_exit_status(event)); } return; } if (succeeded && pcmk__str_eq(event->target, controld_globals.our_nodename, pcmk__str_casei)) { /* We were notified of our own fencing. Most likely, either fencing was * misconfigured, or fabric fencing that doesn't cut cluster * communication is in use. * * Either way, shutting down the local host is a good idea, to require * administrator intervention. Also, other nodes would otherwise likely * set our status to lost because of the fencing callback and discard * our subsequent election votes as "not part of our cluster". */ crm_crit("We were allegedly just fenced by %s for %s!", executioner, event->origin); // Dumps blackbox if enabled if (fence_reaction_panic) { pcmk__panic(__func__); } else { crm_exit(CRM_EX_FATAL); } return; // Should never get here } /* Update the count of fencing failures for this target, in case we become * DC later. The current DC has already updated its fail count in * tengine_stonith_callback(). */ if (!AM_I_DC) { if (succeeded) { st_fail_count_reset(event->target); } else { st_fail_count_increment(event->target); } } crm_notice("Peer %s was%s terminated (%s) by %s on behalf of %s@%s: " "%s%s%s%s " CRM_XS " event=%s", event->target, (succeeded? "" : " not"), event->action, executioner, client, event->origin, (succeeded? "OK" : pcmk_exec_status_str(exec_status)), ((reason == NULL)? "" : " ("), ((reason == NULL)? "" : reason), ((reason == NULL)? "" : ")"), event->id); if (succeeded) { crm_node_t *peer = pcmk__search_node_caches(0, event->target, pcmk__node_search_any |pcmk__node_search_known); const char *uuid = NULL; if (peer == NULL) { return; } uuid = crm_peer_uuid(peer); if (AM_I_DC) { /* The DC always sends updates */ send_stonith_update(NULL, event->target, uuid); /* @TODO Ideally, at this point, we'd check whether the fenced node * hosted any guest nodes, and call remote_node_down() for them. * Unfortunately, the controller doesn't have a simple, reliable way * to map hosts to guests. It might be possible to track this in the * peer cache via crm_remote_peer_cache_refresh(). For now, we rely * on the scheduler creating fence pseudo-events for the guests. */ if (!pcmk__str_eq(client, te_client_id, pcmk__str_casei)) { /* Abort the current transition if it wasn't the cluster that * initiated fencing. */ crm_info("External fencing operation from %s fenced %s", client, event->target); abort_transition(INFINITY, pcmk__graph_restart, "External Fencing Operation", NULL); } } else if (pcmk__str_eq(controld_globals.dc_name, event->target, pcmk__str_null_matches|pcmk__str_casei) && !pcmk_is_set(peer->flags, crm_remote_node)) { // Assume the target was our DC if we don't currently have one if (controld_globals.dc_name != NULL) { crm_notice("Fencing target %s was our DC", event->target); } else { crm_notice("Fencing target %s may have been our DC", event->target); } /* Given the CIB resyncing that occurs around elections, * have one node update the CIB now and, if the new DC is different, * have them do so too after the election */ if (pcmk__str_eq(event->executioner, controld_globals.our_nodename, pcmk__str_casei)) { send_stonith_update(NULL, event->target, uuid); } add_stonith_cleanup(event->target); } /* If the target is a remote node, and we host its connection, * immediately fail all monitors so it can be recovered quickly. * The connection won't necessarily drop when a remote node is fenced, * so the failure might not otherwise be detected until the next poke. */ if (pcmk_is_set(peer->flags, crm_remote_node)) { remote_ra_fail(event->target); } crmd_peer_down(peer, TRUE); } } /*! * \brief Connect to fencer * * \param[in] user_data If NULL, retry failures now, otherwise retry in mainloop timer * * \return G_SOURCE_REMOVE on success, G_SOURCE_CONTINUE to retry * \note If user_data is NULL, this will wait 2s between attempts, for up to * 30 attempts, meaning the controller could be blocked as long as 58s. */ gboolean controld_timer_fencer_connect(gpointer user_data) { int rc = pcmk_ok; if (stonith_api == NULL) { stonith_api = stonith_api_new(); if (stonith_api == NULL) { crm_err("Could not connect to fencer: API memory allocation failed"); return G_SOURCE_REMOVE; } } if (stonith_api->state != stonith_disconnected) { crm_trace("Already connected to fencer, no need to retry"); return G_SOURCE_REMOVE; } if (user_data == NULL) { // Blocking (retry failures now until successful) rc = stonith_api_connect_retry(stonith_api, crm_system_name, 30); if (rc != pcmk_ok) { crm_err("Could not connect to fencer in 30 attempts: %s " CRM_XS " rc=%d", pcmk_strerror(rc), rc); } } else { // Non-blocking (retry failures later in main loop) rc = stonith_api->cmds->connect(stonith_api, crm_system_name, NULL); if (controld_fencer_connect_timer == NULL) { controld_fencer_connect_timer = mainloop_timer_add("controld_fencer_connect", 1000, TRUE, controld_timer_fencer_connect, GINT_TO_POINTER(TRUE)); } if (rc != pcmk_ok) { if (pcmk_is_set(controld_globals.fsa_input_register, R_ST_REQUIRED)) { crm_notice("Fencer connection failed (will retry): %s " CRM_XS " rc=%d", pcmk_strerror(rc), rc); if (!mainloop_timer_running(controld_fencer_connect_timer)) { mainloop_timer_start(controld_fencer_connect_timer); } return G_SOURCE_CONTINUE; } else { crm_info("Fencer connection failed (ignoring because no longer required): %s " CRM_XS " rc=%d", pcmk_strerror(rc), rc); } return G_SOURCE_REMOVE; } } if (rc == pcmk_ok) { stonith_api->cmds->register_notification(stonith_api, T_STONITH_NOTIFY_DISCONNECT, tengine_stonith_connection_destroy); stonith_api->cmds->register_notification(stonith_api, T_STONITH_NOTIFY_FENCE, handle_fence_notification); stonith_api->cmds->register_notification(stonith_api, T_STONITH_NOTIFY_HISTORY_SYNCED, tengine_stonith_history_synced); te_trigger_stonith_history_sync(TRUE); crm_notice("Fencer successfully connected"); } return G_SOURCE_REMOVE; } void controld_disconnect_fencer(bool destroy) { if (stonith_api) { // Prevent fencer connection from coming up again controld_clear_fsa_input_flags(R_ST_REQUIRED); if (stonith_api->state != stonith_disconnected) { stonith_api->cmds->disconnect(stonith_api); } stonith_api->cmds->remove_notification(stonith_api, NULL); } if (destroy) { if (stonith_api) { stonith_api->cmds->free(stonith_api); stonith_api = NULL; } if (controld_fencer_connect_timer) { mainloop_timer_del(controld_fencer_connect_timer); controld_fencer_connect_timer = NULL; } if (te_client_id) { free(te_client_id); te_client_id = NULL; } } } static gboolean do_stonith_history_sync(gpointer user_data) { if (stonith_api && (stonith_api->state != stonith_disconnected)) { stonith_history_t *history = NULL; te_cleanup_stonith_history_sync(stonith_api, FALSE); stonith_api->cmds->history(stonith_api, st_opt_sync_call | st_opt_broadcast, NULL, &history, 5); stonith_history_free(history); return TRUE; } else { crm_info("Skip triggering stonith history-sync as stonith is disconnected"); return FALSE; } } static void tengine_stonith_callback(stonith_t *stonith, stonith_callback_data_t *data) { char *uuid = NULL; int stonith_id = -1; int transition_id = -1; pcmk__graph_action_t *action = NULL; const char *target = NULL; if ((data == NULL) || (data->userdata == NULL)) { crm_err("Ignoring fence operation %d result: " "No transition key given (bug?)", ((data == NULL)? -1 : data->call_id)); return; } if (!AM_I_DC) { const char *reason = stonith__exit_reason(data); if (reason == NULL) { reason = pcmk_exec_status_str(stonith__execution_status(data)); } crm_notice("Result of fence operation %d: %d (%s) " CRM_XS " key=%s", data->call_id, stonith__exit_status(data), reason, (const char *) data->userdata); return; } CRM_CHECK(decode_transition_key(data->userdata, &uuid, &transition_id, &stonith_id, NULL), goto bail); if (controld_globals.transition_graph->complete || (stonith_id < 0) || !pcmk__str_eq(uuid, controld_globals.te_uuid, pcmk__str_none) || (controld_globals.transition_graph->id != transition_id)) { crm_info("Ignoring fence operation %d result: " "Not from current transition " CRM_XS " complete=%s action=%d uuid=%s (vs %s) transition=%d (vs %d)", data->call_id, pcmk__btoa(controld_globals.transition_graph->complete), stonith_id, uuid, controld_globals.te_uuid, transition_id, controld_globals.transition_graph->id); goto bail; } action = controld_get_action(stonith_id); if (action == NULL) { crm_err("Ignoring fence operation %d result: " "Action %d not found in transition graph (bug?) " CRM_XS " uuid=%s transition=%d", data->call_id, stonith_id, uuid, transition_id); goto bail; } target = crm_element_value(action->xml, PCMK__META_ON_NODE); if (target == NULL) { crm_err("Ignoring fence operation %d result: No target given (bug?)", data->call_id); goto bail; } stop_te_timer(action); if (stonith__exit_status(data) == CRM_EX_OK) { const char *uuid = crm_element_value(action->xml, PCMK__META_ON_NODE_UUID); const char *op = crm_meta_value(action->params, "stonith_action"); crm_info("Fence operation %d for %s succeeded", data->call_id, target); if (!(pcmk_is_set(action->flags, pcmk__graph_action_confirmed))) { te_action_confirmed(action, NULL); if (pcmk__str_eq(PCMK_ACTION_ON, op, pcmk__str_casei)) { const char *value = NULL; char *now = pcmk__ttoa(time(NULL)); gboolean is_remote_node = FALSE; /* This check is not 100% reliable, since this node is not * guaranteed to have the remote node cached. However, it * doesn't have to be reliable, since the attribute manager can * learn a node's "remoteness" by other means sooner or later. * This allows it to learn more quickly if this node does have * the information. */ if (g_hash_table_lookup(crm_remote_peer_cache, uuid) != NULL) { is_remote_node = TRUE; } update_attrd(target, CRM_ATTR_UNFENCED, now, NULL, is_remote_node); free(now); value = crm_meta_value(action->params, PCMK__META_DIGESTS_ALL); update_attrd(target, CRM_ATTR_DIGESTS_ALL, value, NULL, is_remote_node); value = crm_meta_value(action->params, PCMK__META_DIGESTS_SECURE); update_attrd(target, CRM_ATTR_DIGESTS_SECURE, value, NULL, is_remote_node); } else if (!(pcmk_is_set(action->flags, pcmk__graph_action_sent_update))) { send_stonith_update(action, target, uuid); pcmk__set_graph_action_flags(action, pcmk__graph_action_sent_update); } } st_fail_count_reset(target); } else { enum pcmk__graph_next abort_action = pcmk__graph_restart; int status = stonith__execution_status(data); const char *reason = stonith__exit_reason(data); if (reason == NULL) { if (status == PCMK_EXEC_DONE) { reason = "Agent returned error"; } else { reason = pcmk_exec_status_str(status); } } pcmk__set_graph_action_flags(action, pcmk__graph_action_failed); /* If no fence devices were available, there's no use in immediately * checking again, so don't start a new transition in that case. */ if (status == PCMK_EXEC_NO_FENCE_DEVICE) { crm_warn("Fence operation %d for %s failed: %s " "(aborting transition and giving up for now)", data->call_id, target, reason); abort_action = pcmk__graph_wait; } else { crm_notice("Fence operation %d for %s failed: %s " "(aborting transition)", data->call_id, target, reason); } /* Increment the fail count now, so abort_for_stonith_failure() can * check it. Non-DC nodes will increment it in * handle_fence_notification(). */ st_fail_count_increment(target); abort_for_stonith_failure(abort_action, target, NULL); } pcmk__update_graph(controld_globals.transition_graph, action); trigger_graph(); bail: free(data->userdata); free(uuid); return; } static int fence_with_delay(const char *target, const char *type, int delay) { uint32_t options = st_opt_none; // Group of enum stonith_call_options int timeout_sec = (int) (controld_globals.transition_graph->stonith_timeout / 1000); if (crmd_join_phase_count(crm_join_confirmed) == 1) { stonith__set_call_options(options, target, st_opt_allow_suicide); } return stonith_api->cmds->fence_with_delay(stonith_api, options, target, type, timeout_sec, 0, delay); } /*! * \internal * \brief Execute a fencing action from a transition graph * * \param[in] graph Transition graph being executed (ignored) * \param[in] action Fencing action to execute * * \return Standard Pacemaker return code */ int controld_execute_fence_action(pcmk__graph_t *graph, pcmk__graph_action_t *action) { int rc = 0; const char *id = ID(action->xml); const char *uuid = crm_element_value(action->xml, PCMK__META_ON_NODE_UUID); const char *target = crm_element_value(action->xml, PCMK__META_ON_NODE); const char *type = crm_meta_value(action->params, "stonith_action"); char *transition_key = NULL; const char *priority_delay = NULL; int delay_i = 0; gboolean invalid_action = FALSE; int stonith_timeout = (int) (controld_globals.transition_graph->stonith_timeout / 1000); CRM_CHECK(id != NULL, invalid_action = TRUE); CRM_CHECK(uuid != NULL, invalid_action = TRUE); CRM_CHECK(type != NULL, invalid_action = TRUE); CRM_CHECK(target != NULL, invalid_action = TRUE); if (invalid_action) { crm_log_xml_warn(action->xml, "BadAction"); return EPROTO; } priority_delay = crm_meta_value(action->params, PCMK_OPT_PRIORITY_FENCING_DELAY); crm_notice("Requesting fencing (%s) targeting node %s " CRM_XS " action=%s timeout=%i%s%s", type, target, id, stonith_timeout, priority_delay ? " priority_delay=" : "", priority_delay ? priority_delay : ""); /* Passing NULL means block until we can connect... */ controld_timer_fencer_connect(NULL); pcmk__scan_min_int(priority_delay, &delay_i, 0); rc = fence_with_delay(target, type, delay_i); transition_key = pcmk__transition_key(controld_globals.transition_graph->id, action->id, 0, controld_globals.te_uuid), stonith_api->cmds->register_callback(stonith_api, rc, (stonith_timeout + (delay_i > 0 ? delay_i : 0)), st_opt_timeout_updates, transition_key, "tengine_stonith_callback", tengine_stonith_callback); return pcmk_rc_ok; } bool controld_verify_stonith_watchdog_timeout(const char *value) { - long st_timeout = value? crm_get_msec(value) : 0; + long long st_timeout = (value != NULL)? crm_get_msec(value) : 0; const char *our_nodename = controld_globals.our_nodename; if (st_timeout == 0 || (stonith_api && (stonith_api->state != stonith_disconnected) && stonith__watchdog_fencing_enabled_for_node_api(stonith_api, our_nodename))) { return pcmk__valid_stonith_watchdog_timeout(value); } return true; } /* end stonith API client functions */ /* * stonith history synchronization * * Each node's fencer keeps track of a cluster-wide fencing history. When a node * joins or leaves, we need to synchronize the history across all nodes. */ static crm_trigger_t *stonith_history_sync_trigger = NULL; static mainloop_timer_t *stonith_history_sync_timer_short = NULL; static mainloop_timer_t *stonith_history_sync_timer_long = NULL; void te_cleanup_stonith_history_sync(stonith_t *st, bool free_timers) { if (free_timers) { mainloop_timer_del(stonith_history_sync_timer_short); stonith_history_sync_timer_short = NULL; mainloop_timer_del(stonith_history_sync_timer_long); stonith_history_sync_timer_long = NULL; } else { mainloop_timer_stop(stonith_history_sync_timer_short); mainloop_timer_stop(stonith_history_sync_timer_long); } if (st) { st->cmds->remove_notification(st, T_STONITH_NOTIFY_HISTORY_SYNCED); } } static void tengine_stonith_history_synced(stonith_t *st, stonith_event_t *st_event) { te_cleanup_stonith_history_sync(st, FALSE); crm_debug("Fence-history synced - cancel all timers"); } static gboolean stonith_history_sync_set_trigger(gpointer user_data) { mainloop_set_trigger(stonith_history_sync_trigger); return FALSE; } void te_trigger_stonith_history_sync(bool long_timeout) { /* trigger a sync in 5s to give more nodes the * chance to show up so that we don't create * unnecessary stonith-history-sync traffic * * the long timeout of 30s is there as a fallback * so that after a successful connection to fenced * we will wait for 30s for the DC to trigger a * history-sync * if this doesn't happen we trigger a sync locally * (e.g. fenced segfaults and is restarted by pacemakerd) */ /* as we are finally checking the stonith-connection * in do_stonith_history_sync we should be fine * leaving stonith_history_sync_time & stonith_history_sync_trigger * around */ if (stonith_history_sync_trigger == NULL) { stonith_history_sync_trigger = mainloop_add_trigger(G_PRIORITY_LOW, do_stonith_history_sync, NULL); } if (long_timeout) { if(stonith_history_sync_timer_long == NULL) { stonith_history_sync_timer_long = mainloop_timer_add("history_sync_long", 30000, FALSE, stonith_history_sync_set_trigger, NULL); } crm_info("Fence history will be synchronized cluster-wide within 30 seconds"); mainloop_timer_start(stonith_history_sync_timer_long); } else { if(stonith_history_sync_timer_short == NULL) { stonith_history_sync_timer_short = mainloop_timer_add("history_sync_short", 5000, FALSE, stonith_history_sync_set_trigger, NULL); } crm_info("Fence history will be synchronized cluster-wide within 5 seconds"); mainloop_timer_start(stonith_history_sync_timer_short); } } /* end stonith history synchronization functions */ diff --git a/daemons/fenced/fenced_cib.c b/daemons/fenced/fenced_cib.c index 19e9b16869..676d615cf7 100644 --- a/daemons/fenced/fenced_cib.c +++ b/daemons/fenced/fenced_cib.c @@ -1,749 +1,749 @@ /* * Copyright 2009-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include #include #include #include static xmlNode *local_cib = NULL; static cib_t *cib_api = NULL; static bool have_cib_devices = FALSE; /*! * \internal * \brief Check whether a node has a specific attribute name/value * * \param[in] node Name of node to check * \param[in] name Name of an attribute to look for * \param[in] value The value the named attribute needs to be set to in order to be considered a match * * \return TRUE if the locally cached CIB has the specified node attribute */ gboolean node_has_attr(const char *node, const char *name, const char *value) { GString *xpath = NULL; xmlNode *match; CRM_CHECK((local_cib != NULL) && (node != NULL) && (name != NULL) && (value != NULL), return FALSE); /* Search for the node's attributes in the CIB. While the schema allows * multiple sets of instance attributes, and allows instance attributes to * use id-ref to reference values elsewhere, that is intended for resources, * so we ignore that here. */ xpath = g_string_sized_new(256); pcmk__g_strcat(xpath, "//" PCMK_XE_NODES "/" PCMK_XE_NODE "[@" PCMK_XA_UNAME "='", node, "']" "/" PCMK_XE_INSTANCE_ATTRIBUTES "/" PCMK_XE_NVPAIR "[@" PCMK_XA_NAME "='", name, "' " "and @" PCMK_XA_VALUE "='", value, "']", NULL); match = get_xpath_object((const char *) xpath->str, local_cib, LOG_NEVER); g_string_free(xpath, TRUE); return (match != NULL); } static void add_topology_level(xmlNode *match) { char *desc = NULL; pcmk__action_result_t result = PCMK__UNKNOWN_RESULT; CRM_CHECK(match != NULL, return); fenced_register_level(match, &desc, &result); fenced_send_level_notification(STONITH_OP_LEVEL_ADD, &result, desc); pcmk__reset_result(&result); free(desc); } static void topology_remove_helper(const char *node, int level) { char *desc = NULL; pcmk__action_result_t result = PCMK__UNKNOWN_RESULT; xmlNode *data = create_xml_node(NULL, XML_TAG_FENCING_LEVEL); crm_xml_add(data, F_STONITH_ORIGIN, __func__); crm_xml_add_int(data, PCMK_XA_INDEX, level); crm_xml_add(data, PCMK_XA_TARGET, node); fenced_unregister_level(data, &desc, &result); fenced_send_level_notification(STONITH_OP_LEVEL_DEL, &result, desc); pcmk__reset_result(&result); free_xml(data); free(desc); } static void remove_topology_level(xmlNode *match) { int index = 0; char *key = NULL; CRM_CHECK(match != NULL, return); key = stonith_level_key(match, fenced_target_by_unknown); crm_element_value_int(match, PCMK_XA_INDEX, &index); topology_remove_helper(key, index); free(key); } static void register_fencing_topology(xmlXPathObjectPtr xpathObj) { int max = numXpathResults(xpathObj), lpc = 0; for (lpc = 0; lpc < max; lpc++) { xmlNode *match = getXpathResult(xpathObj, lpc); remove_topology_level(match); add_topology_level(match); } } /* Fencing */ void fencing_topology_init(void) { xmlXPathObjectPtr xpathObj = NULL; const char *xpath = "//" XML_TAG_FENCING_LEVEL; crm_trace("Full topology refresh"); free_topology_list(); init_topology_list(); /* Grab everything */ xpathObj = xpath_search(local_cib, xpath); register_fencing_topology(xpathObj); freeXpathObject(xpathObj); } static void remove_cib_device(xmlXPathObjectPtr xpathObj) { int max = numXpathResults(xpathObj), lpc = 0; for (lpc = 0; lpc < max; lpc++) { const char *rsc_id = NULL; const char *standard = NULL; xmlNode *match = getXpathResult(xpathObj, lpc); CRM_LOG_ASSERT(match != NULL); if(match != NULL) { standard = crm_element_value(match, PCMK_XA_CLASS); } if (!pcmk__str_eq(standard, PCMK_RESOURCE_CLASS_STONITH, pcmk__str_casei)) { continue; } rsc_id = crm_element_value(match, PCMK_XA_ID); stonith_device_remove(rsc_id, true); } } #define XPATH_WATCHDOG_TIMEOUT "//" PCMK_XE_NVPAIR \ "[@" PCMK_XA_NAME "='" \ PCMK_OPT_STONITH_WATCHDOG_TIMEOUT "']" static void update_stonith_watchdog_timeout_ms(xmlNode *cib) { - long timeout_ms = 0; + long long timeout_ms = 0; xmlNode *stonith_watchdog_xml = NULL; const char *value = NULL; stonith_watchdog_xml = get_xpath_object(XPATH_WATCHDOG_TIMEOUT, cib, LOG_NEVER); if (stonith_watchdog_xml) { value = crm_element_value(stonith_watchdog_xml, PCMK_XA_VALUE); } if (value) { timeout_ms = crm_get_msec(value); } if (timeout_ms < 0) { timeout_ms = pcmk__auto_stonith_watchdog_timeout(); } - stonith_watchdog_timeout_ms = timeout_ms; + stonith_watchdog_timeout_ms = (long) QB_MIN(timeout_ms, LONG_MAX); } /*! * \internal * \brief Update all STONITH device definitions based on current CIB */ static void cib_devices_update(void) { GHashTableIter iter; stonith_device_t *device = NULL; crm_info("Updating devices to version %s.%s.%s", crm_element_value(local_cib, PCMK_XA_ADMIN_EPOCH), crm_element_value(local_cib, PCMK_XA_EPOCH), crm_element_value(local_cib, PCMK_XA_NUM_UPDATES)); g_hash_table_iter_init(&iter, device_list); while (g_hash_table_iter_next(&iter, NULL, (void **)&device)) { if (device->cib_registered) { device->dirty = TRUE; } } /* have list repopulated if cib has a watchdog-fencing-resource TODO: keep a cached list for queries happening while we are refreshing */ g_list_free_full(stonith_watchdog_targets, free); stonith_watchdog_targets = NULL; fenced_scheduler_run(local_cib); g_hash_table_iter_init(&iter, device_list); while (g_hash_table_iter_next(&iter, NULL, (void **)&device)) { if (device->dirty) { g_hash_table_iter_remove(&iter); } } } static void update_cib_stonith_devices_v1(const char *event, xmlNode * msg) { const char *reason = "none"; gboolean needs_update = FALSE; xmlXPathObjectPtr xpath_obj = NULL; /* process new constraints */ xpath_obj = xpath_search(msg, "//" F_CIB_UPDATE_RESULT "//" XML_CONS_TAG_RSC_LOCATION); if (numXpathResults(xpath_obj) > 0) { int max = numXpathResults(xpath_obj), lpc = 0; /* Safest and simplest to always recompute */ needs_update = TRUE; reason = "new location constraint"; for (lpc = 0; lpc < max; lpc++) { xmlNode *match = getXpathResult(xpath_obj, lpc); crm_log_xml_trace(match, "new constraint"); } } freeXpathObject(xpath_obj); /* process deletions */ xpath_obj = xpath_search(msg, "//" F_CIB_UPDATE_RESULT "//" PCMK__XE_DIFF_REMOVED "//" PCMK_XE_PRIMITIVE); if (numXpathResults(xpath_obj) > 0) { remove_cib_device(xpath_obj); } freeXpathObject(xpath_obj); /* process additions */ xpath_obj = xpath_search(msg, "//" F_CIB_UPDATE_RESULT "//" PCMK__XE_DIFF_ADDED "//" PCMK_XE_PRIMITIVE); if (numXpathResults(xpath_obj) > 0) { int max = numXpathResults(xpath_obj), lpc = 0; for (lpc = 0; lpc < max; lpc++) { const char *rsc_id = NULL; const char *standard = NULL; xmlNode *match = getXpathResult(xpath_obj, lpc); rsc_id = crm_element_value(match, PCMK_XA_ID); standard = crm_element_value(match, PCMK_XA_CLASS); if (!pcmk__str_eq(standard, PCMK_RESOURCE_CLASS_STONITH, pcmk__str_casei)) { continue; } crm_trace("Fencing resource %s was added or modified", rsc_id); reason = "new resource"; needs_update = TRUE; } } freeXpathObject(xpath_obj); if(needs_update) { crm_info("Updating device list from CIB: %s", reason); cib_devices_update(); } } static void update_cib_stonith_devices_v2(const char *event, xmlNode * msg) { xmlNode *change = NULL; char *reason = NULL; bool needs_update = FALSE; xmlNode *patchset = get_message_xml(msg, F_CIB_UPDATE_RESULT); for (change = pcmk__xml_first_child(patchset); change != NULL; change = pcmk__xml_next(change)) { const char *op = crm_element_value(change, PCMK_XA_OPERATION); const char *xpath = crm_element_value(change, PCMK_XA_PATH); const char *shortpath = NULL; if (pcmk__str_eq(op, "move", pcmk__str_null_matches) || (strstr(xpath, "/" PCMK_XE_STATUS) != NULL)) { continue; } else if (pcmk__str_eq(op, "delete", pcmk__str_casei) && (strstr(xpath, "/" PCMK_XE_PRIMITIVE) != NULL)) { const char *rsc_id = NULL; char *search = NULL; char *mutable = NULL; if ((strstr(xpath, PCMK_XE_INSTANCE_ATTRIBUTES) != NULL) || (strstr(xpath, PCMK_XE_META_ATTRIBUTES) != NULL)) { needs_update = TRUE; pcmk__str_update(&reason, "(meta) attribute deleted from resource"); break; } pcmk__str_update(&mutable, xpath); rsc_id = strstr(mutable, "primitive[@" PCMK_XA_ID "=\'"); if (rsc_id != NULL) { rsc_id += strlen("primitive[@" PCMK_XA_ID "=\'"); search = strchr(rsc_id, '\''); } if (search != NULL) { *search = 0; stonith_device_remove(rsc_id, true); /* watchdog_device_update called afterwards to fall back to implicit definition if needed */ } else { crm_warn("Ignoring malformed CIB update (resource deletion)"); } free(mutable); } else if (strstr(xpath, "/" PCMK_XE_RESOURCES) || strstr(xpath, "/" PCMK_XE_CONSTRAINTS) || strstr(xpath, "/" PCMK_XE_RSC_DEFAULTS)) { shortpath = strrchr(xpath, '/'); CRM_ASSERT(shortpath); reason = crm_strdup_printf("%s %s", op, shortpath+1); needs_update = TRUE; break; } } if(needs_update) { crm_info("Updating device list from CIB: %s", reason); cib_devices_update(); } else { crm_trace("No updates for device list found in CIB"); } free(reason); } static void update_cib_stonith_devices(const char *event, xmlNode * msg) { int format = 1; xmlNode *patchset = get_message_xml(msg, F_CIB_UPDATE_RESULT); CRM_ASSERT(patchset); crm_element_value_int(patchset, PCMK_XA_FORMAT, &format); switch(format) { case 1: update_cib_stonith_devices_v1(event, msg); break; case 2: update_cib_stonith_devices_v2(event, msg); break; default: crm_warn("Unknown patch format: %d", format); } } static void watchdog_device_update(void) { if (stonith_watchdog_timeout_ms > 0) { if (!g_hash_table_lookup(device_list, STONITH_WATCHDOG_ID) && !stonith_watchdog_targets) { /* getting here watchdog-fencing enabled, no device there yet and reason isn't stonith_watchdog_targets preventing that */ int rc; xmlNode *xml; xml = create_device_registration_xml( STONITH_WATCHDOG_ID, st_namespace_internal, STONITH_WATCHDOG_AGENT, NULL, /* stonith_device_register will add our own name as PCMK_STONITH_HOST_LIST param so we can skip that here */ NULL); rc = stonith_device_register(xml, TRUE); free_xml(xml); if (rc != pcmk_ok) { rc = pcmk_legacy2rc(rc); exit_code = CRM_EX_FATAL; crm_crit("Cannot register watchdog pseudo fence agent: %s", pcmk_rc_str(rc)); stonith_shutdown(0); } } } else if (g_hash_table_lookup(device_list, STONITH_WATCHDOG_ID) != NULL) { /* be silent if no device - todo parameter to stonith_device_remove */ stonith_device_remove(STONITH_WATCHDOG_ID, true); } } /*! * \internal * \brief Query the full CIB * * \return Standard Pacemaker return code */ static int fenced_query_cib(void) { int rc = pcmk_ok; crm_trace("Re-requesting full CIB"); rc = cib_api->cmds->query(cib_api, NULL, &local_cib, cib_scope_local|cib_sync_call); rc = pcmk_legacy2rc(rc); if (rc == pcmk_rc_ok) { CRM_ASSERT(local_cib != NULL); } else { crm_err("Couldn't retrieve the CIB: %s " CRM_XS " rc=%d", pcmk_rc_str(rc), rc); } return rc; } static void remove_fencing_topology(xmlXPathObjectPtr xpathObj) { int max = numXpathResults(xpathObj), lpc = 0; for (lpc = 0; lpc < max; lpc++) { xmlNode *match = getXpathResult(xpathObj, lpc); CRM_LOG_ASSERT(match != NULL); if (match && crm_element_value(match, XML_DIFF_MARKER)) { /* Deletion */ int index = 0; char *target = stonith_level_key(match, fenced_target_by_unknown); crm_element_value_int(match, PCMK_XA_INDEX, &index); if (target == NULL) { crm_err("Invalid fencing target in element %s", ID(match)); } else if (index <= 0) { crm_err("Invalid level for %s in element %s", target, ID(match)); } else { topology_remove_helper(target, index); } /* } else { Deal with modifications during the 'addition' stage */ } } } static void update_fencing_topology(const char *event, xmlNode * msg) { int format = 1; const char *xpath; xmlXPathObjectPtr xpathObj = NULL; xmlNode *patchset = get_message_xml(msg, F_CIB_UPDATE_RESULT); CRM_ASSERT(patchset); crm_element_value_int(patchset, PCMK_XA_FORMAT, &format); if(format == 1) { /* Process deletions (only) */ xpath = "//" F_CIB_UPDATE_RESULT "//" PCMK__XE_DIFF_REMOVED "//" XML_TAG_FENCING_LEVEL; xpathObj = xpath_search(msg, xpath); remove_fencing_topology(xpathObj); freeXpathObject(xpathObj); /* Process additions and changes */ xpath = "//" F_CIB_UPDATE_RESULT "//" PCMK__XE_DIFF_ADDED "//" XML_TAG_FENCING_LEVEL; xpathObj = xpath_search(msg, xpath); register_fencing_topology(xpathObj); freeXpathObject(xpathObj); } else if(format == 2) { xmlNode *change = NULL; int add[] = { 0, 0, 0 }; int del[] = { 0, 0, 0 }; xml_patch_versions(patchset, add, del); for (change = pcmk__xml_first_child(patchset); change != NULL; change = pcmk__xml_next(change)) { const char *op = crm_element_value(change, PCMK_XA_OPERATION); const char *xpath = crm_element_value(change, PCMK_XA_PATH); if(op == NULL) { continue; } else if(strstr(xpath, "/" XML_TAG_FENCING_LEVEL) != NULL) { /* Change to a specific entry */ crm_trace("Handling %s operation %d.%d.%d for %s", op, add[0], add[1], add[2], xpath); if(strcmp(op, "move") == 0) { continue; } else if(strcmp(op, "create") == 0) { add_topology_level(change->children); } else if(strcmp(op, "modify") == 0) { xmlNode *match = first_named_child(change, XML_DIFF_RESULT); if(match) { remove_topology_level(match->children); add_topology_level(match->children); } } else if(strcmp(op, "delete") == 0) { /* Nuclear option, all we have is the path and an id... not enough to remove a specific entry */ crm_info("Re-initializing fencing topology after %s operation %d.%d.%d for %s", op, add[0], add[1], add[2], xpath); fencing_topology_init(); return; } } else if (strstr(xpath, "/" XML_TAG_FENCING_TOPOLOGY) != NULL) { /* Change to the topology in general */ crm_info("Re-initializing fencing topology after top-level %s operation %d.%d.%d for %s", op, add[0], add[1], add[2], xpath); fencing_topology_init(); return; } else if (strstr(xpath, "/" PCMK_XE_CONFIGURATION)) { /* Changes to the whole config section, possibly including the topology as a whild */ if(first_named_child(change, XML_TAG_FENCING_TOPOLOGY) == NULL) { crm_trace("Nothing for us in %s operation %d.%d.%d for %s.", op, add[0], add[1], add[2], xpath); } else if(strcmp(op, "delete") == 0 || strcmp(op, "create") == 0) { crm_info("Re-initializing fencing topology after top-level %s operation %d.%d.%d for %s.", op, add[0], add[1], add[2], xpath); fencing_topology_init(); return; } } else { crm_trace("Nothing for us in %s operation %d.%d.%d for %s", op, add[0], add[1], add[2], xpath); } } } else { crm_warn("Unknown patch format: %d", format); } } static void update_cib_cache_cb(const char *event, xmlNode * msg) { long timeout_ms_saved = stonith_watchdog_timeout_ms; bool need_full_refresh = false; if(!have_cib_devices) { crm_trace("Skipping updates until we get a full dump"); return; } else if(msg == NULL) { crm_trace("Missing %s update", event); return; } /* Maintain a local copy of the CIB so that we have full access * to device definitions, location constraints, and node attributes */ if (local_cib != NULL) { int rc = pcmk_ok; xmlNode *patchset = NULL; crm_element_value_int(msg, F_CIB_RC, &rc); if (rc != pcmk_ok) { return; } patchset = get_message_xml(msg, F_CIB_UPDATE_RESULT); rc = xml_apply_patchset(local_cib, patchset, TRUE); switch (rc) { case pcmk_ok: case -pcmk_err_old_data: break; case -pcmk_err_diff_resync: case -pcmk_err_diff_failed: crm_notice("[%s] Patch aborted: %s (%d)", event, pcmk_strerror(rc), rc); free_xml(local_cib); local_cib = NULL; break; default: crm_warn("[%s] ABORTED: %s (%d)", event, pcmk_strerror(rc), rc); free_xml(local_cib); local_cib = NULL; } } if (local_cib == NULL) { if (fenced_query_cib() != pcmk_rc_ok) { return; } need_full_refresh = true; } pcmk__refresh_node_caches_from_cib(local_cib); update_stonith_watchdog_timeout_ms(local_cib); if (timeout_ms_saved != stonith_watchdog_timeout_ms) { need_full_refresh = true; } if (need_full_refresh) { fencing_topology_init(); cib_devices_update(); } else { // Partial refresh update_fencing_topology(event, msg); update_cib_stonith_devices(event, msg); } watchdog_device_update(); } static void init_cib_cache_cb(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data) { crm_info("Updating device list from CIB"); have_cib_devices = TRUE; local_cib = copy_xml(output); pcmk__refresh_node_caches_from_cib(local_cib); update_stonith_watchdog_timeout_ms(local_cib); fencing_topology_init(); cib_devices_update(); watchdog_device_update(); } static void cib_connection_destroy(gpointer user_data) { if (stonith_shutdown_flag) { crm_info("Connection to the CIB manager closed"); return; } else { crm_crit("Lost connection to the CIB manager, shutting down"); } if (cib_api) { cib_api->cmds->signoff(cib_api); } stonith_shutdown(0); } /*! * \internal * \brief Disconnect from CIB manager */ void fenced_cib_cleanup(void) { if (cib_api != NULL) { cib_api->cmds->del_notify_callback(cib_api, T_CIB_DIFF_NOTIFY, update_cib_cache_cb); cib__clean_up_connection(&cib_api); } free_xml(local_cib); local_cib = NULL; } void setup_cib(void) { int rc, retries = 0; cib_api = cib_new(); if (cib_api == NULL) { crm_err("No connection to the CIB manager"); return; } do { sleep(retries); rc = cib_api->cmds->signon(cib_api, CRM_SYSTEM_STONITHD, cib_command); } while (rc == -ENOTCONN && ++retries < 5); if (rc != pcmk_ok) { crm_err("Could not connect to the CIB manager: %s (%d)", pcmk_strerror(rc), rc); } else if (pcmk_ok != cib_api->cmds->add_notify_callback(cib_api, T_CIB_DIFF_NOTIFY, update_cib_cache_cb)) { crm_err("Could not set CIB notification callback"); } else { rc = cib_api->cmds->query(cib_api, NULL, NULL, cib_scope_local); cib_api->cmds->register_callback(cib_api, rc, 120, FALSE, NULL, "init_cib_cache_cb", init_cib_cache_cb); cib_api->cmds->set_connection_dnotify(cib_api, cib_connection_destroy); crm_info("Watching for fencing topology changes"); } } diff --git a/daemons/pacemakerd/pcmkd_subdaemons.c b/daemons/pacemakerd/pcmkd_subdaemons.c index aefe8fa9d1..f8e305bbde 100644 --- a/daemons/pacemakerd/pcmkd_subdaemons.c +++ b/daemons/pacemakerd/pcmkd_subdaemons.c @@ -1,898 +1,902 @@ /* * Copyright 2010-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include "pacemakerd.h" #if SUPPORT_COROSYNC #include "pcmkd_corosync.h" #endif #include #include #include #include #include #include #include #include #include #include #include enum child_daemon_flags { child_none = 0, child_respawn = 1 << 0, child_needs_cluster = 1 << 1, child_needs_retry = 1 << 2, child_active_before_startup = 1 << 3, }; typedef struct pcmk_child_s { pid_t pid; int respawn_count; const char *name; const char *uid; const char *command; const char *endpoint; /* IPC server name */ int check_count; uint32_t flags; } pcmk_child_t; #define PCMK_PROCESS_CHECK_INTERVAL 1 #define PCMK_PROCESS_CHECK_RETRIES 5 #define SHUTDOWN_ESCALATION_PERIOD 180000 /* 3m */ /* Index into the array below */ #define PCMK_CHILD_CONTROLD 5 static pcmk_child_t pcmk_children[] = { { 0, 0, "pacemaker-based", CRM_DAEMON_USER, CRM_DAEMON_DIR "/pacemaker-based", PCMK__SERVER_BASED_RO, 0, child_respawn | child_needs_cluster }, { 0, 0, "pacemaker-fenced", NULL, CRM_DAEMON_DIR "/pacemaker-fenced", "stonith-ng", 0, child_respawn | child_needs_cluster }, { 0, 0, "pacemaker-execd", NULL, CRM_DAEMON_DIR "/pacemaker-execd", CRM_SYSTEM_LRMD, 0, child_respawn }, { 0, 0, "pacemaker-attrd", CRM_DAEMON_USER, CRM_DAEMON_DIR "/pacemaker-attrd", T_ATTRD, 0, child_respawn | child_needs_cluster }, { 0, 0, "pacemaker-schedulerd", CRM_DAEMON_USER, CRM_DAEMON_DIR "/pacemaker-schedulerd", CRM_SYSTEM_PENGINE, 0, child_respawn }, { 0, 0, "pacemaker-controld", CRM_DAEMON_USER, CRM_DAEMON_DIR "/pacemaker-controld", CRM_SYSTEM_CRMD, 0, child_respawn | child_needs_cluster }, }; static char *opts_default[] = { NULL, NULL }; static char *opts_vgrind[] = { NULL, NULL, NULL, NULL, NULL }; crm_trigger_t *shutdown_trigger = NULL; crm_trigger_t *startup_trigger = NULL; time_t subdaemon_check_progress = 0; // Whether we need root group access to talk to cluster layer static bool need_root_group = true; /* When contacted via pacemakerd-api by a client having sbd in * the name we assume it is sbd-daemon which wants to know * if pacemakerd shutdown gracefully. * Thus when everything is shutdown properly pacemakerd * waits till it has reported the graceful completion of * shutdown to sbd and just when sbd-client closes the * connection we can assume that the report has arrived * properly so that pacemakerd can finally exit. * Following two variables are used to track that handshake. */ unsigned int shutdown_complete_state_reported_to = 0; gboolean shutdown_complete_state_reported_client_closed = FALSE; /* state we report when asked via pacemakerd-api status-ping */ const char *pacemakerd_state = XML_PING_ATTR_PACEMAKERDSTATE_INIT; gboolean running_with_sbd = FALSE; /* local copy */ GMainLoop *mainloop = NULL; static gboolean fatal_error = FALSE; static int child_liveness(pcmk_child_t *child); static gboolean escalate_shutdown(gpointer data); static int start_child(pcmk_child_t * child); static void pcmk_child_exit(mainloop_child_t * p, pid_t pid, int core, int signo, int exitcode); static void pcmk_process_exit(pcmk_child_t * child); static gboolean pcmk_shutdown_worker(gpointer user_data); static gboolean stop_child(pcmk_child_t * child, int signal); static bool pcmkd_cluster_connected(void) { #if SUPPORT_COROSYNC return pcmkd_corosync_connected(); #else return true; #endif } static gboolean check_next_subdaemon(gpointer user_data) { static int next_child = 0; int rc = child_liveness(&pcmk_children[next_child]); crm_trace("Checked %s[%lld]: %s (%d)", pcmk_children[next_child].name, (long long) PCMK__SPECIAL_PID_AS_0(pcmk_children[next_child].pid), pcmk_rc_str(rc), rc); switch (rc) { case pcmk_rc_ok: pcmk_children[next_child].check_count = 0; subdaemon_check_progress = time(NULL); break; case pcmk_rc_ipc_pid_only: // This case: it was previously OK pcmk_children[next_child].check_count++; if (pcmk_children[next_child].check_count >= PCMK_PROCESS_CHECK_RETRIES) { crm_err("%s[%lld] is unresponsive to ipc after %d tries but " "we found the pid so have it killed that we can restart", pcmk_children[next_child].name, (long long) PCMK__SPECIAL_PID_AS_0( pcmk_children[next_child].pid), pcmk_children[next_child].check_count); stop_child(&pcmk_children[next_child], SIGKILL); if (pcmk_is_set(pcmk_children[next_child].flags, child_respawn)) { /* as long as the respawn-limit isn't reached give it another round of check retries */ pcmk_children[next_child].check_count = 0; } } else { crm_notice("%s[%lld] is unresponsive to ipc after %d tries", pcmk_children[next_child].name, (long long) PCMK__SPECIAL_PID_AS_0( pcmk_children[next_child].pid), pcmk_children[next_child].check_count); if (pcmk_is_set(pcmk_children[next_child].flags, child_respawn)) { /* as long as the respawn-limit isn't reached and we haven't run out of connect retries we account this as progress we are willing to tell to sbd */ subdaemon_check_progress = time(NULL); } } /* go to the next child and see if we can make progress there */ break; case pcmk_rc_ipc_unresponsive: if (!pcmk_is_set(pcmk_children[next_child].flags, child_respawn)) { /* if a subdaemon is down and we don't want it to be restarted this is a success during shutdown. if it isn't restarted anymore due to MAX_RESPAWN it is rather no success. */ if (pcmk_children[next_child].respawn_count <= MAX_RESPAWN) { subdaemon_check_progress = time(NULL); } } if (!pcmk_is_set(pcmk_children[next_child].flags, child_active_before_startup)) { crm_trace("found %s[%lld] missing - signal-handler " "will take care of it", pcmk_children[next_child].name, (long long) PCMK__SPECIAL_PID_AS_0( pcmk_children[next_child].pid)); break; } if (pcmk_is_set(pcmk_children[next_child].flags, child_respawn)) { crm_err("%s[%lld] terminated", pcmk_children[next_child].name, (long long) PCMK__SPECIAL_PID_AS_0( pcmk_children[next_child].pid)); } else { /* orderly shutdown */ crm_notice("%s[%lld] terminated", pcmk_children[next_child].name, (long long) PCMK__SPECIAL_PID_AS_0( pcmk_children[next_child].pid)); } pcmk_process_exit(&(pcmk_children[next_child])); break; default: crm_exit(CRM_EX_FATAL); break; /* static analysis/noreturn */ } next_child++; if (next_child >= PCMK__NELEM(pcmk_children)) { next_child = 0; } return G_SOURCE_CONTINUE; } static gboolean escalate_shutdown(gpointer data) { pcmk_child_t *child = data; if (child->pid == PCMK__SPECIAL_PID) { pcmk_process_exit(child); } else if (child->pid != 0) { /* Use SIGSEGV instead of SIGKILL to create a core so we can see what it was up to */ crm_err("Child %s not terminating in a timely manner, forcing", child->name); stop_child(child, SIGSEGV); } return FALSE; } static void pcmk_child_exit(mainloop_child_t * p, pid_t pid, int core, int signo, int exitcode) { pcmk_child_t *child = mainloop_child_userdata(p); const char *name = mainloop_child_name(p); if (signo) { do_crm_log(((signo == SIGKILL)? LOG_WARNING : LOG_ERR), "%s[%d] terminated with signal %d (%s)%s", name, pid, signo, strsignal(signo), (core? " and dumped core" : "")); } else { switch(exitcode) { case CRM_EX_OK: crm_info("%s[%d] exited with status %d (%s)", name, pid, exitcode, crm_exit_str(exitcode)); break; case CRM_EX_FATAL: crm_warn("Shutting cluster down because %s[%d] had fatal failure", name, pid); child->flags &= ~child_respawn; fatal_error = TRUE; pcmk_shutdown(SIGTERM); break; case CRM_EX_PANIC: crm_emerg("%s[%d] instructed the machine to reset", name, pid); child->flags &= ~child_respawn; fatal_error = TRUE; pcmk__panic(__func__); pcmk_shutdown(SIGTERM); break; default: crm_err("%s[%d] exited with status %d (%s)", name, pid, exitcode, crm_exit_str(exitcode)); break; } } pcmk_process_exit(child); } static void pcmk_process_exit(pcmk_child_t * child) { child->pid = 0; child->flags &= ~child_active_before_startup; child->check_count = 0; child->respawn_count += 1; if (child->respawn_count > MAX_RESPAWN) { crm_err("Child respawn count exceeded by %s", child->name); child->flags &= ~child_respawn; } if (shutdown_trigger) { /* resume step-wise shutdown (returned TRUE yields no parallelizing) */ mainloop_set_trigger(shutdown_trigger); } else if (!pcmk_is_set(child->flags, child_respawn)) { /* nothing to do */ } else if (crm_is_true(pcmk__env_option(PCMK__ENV_FAIL_FAST))) { crm_err("Rebooting system because of %s", child->name); pcmk__panic(__func__); } else if (child_liveness(child) == pcmk_rc_ok) { crm_warn("One-off suppressing strict respawning of a child process %s," " appears alright per %s IPC end-point", child->name, child->endpoint); } else if (pcmk_is_set(child->flags, child_needs_cluster) && !pcmkd_cluster_connected()) { crm_notice("Not respawning %s subdaemon until cluster returns", child->name); child->flags |= child_needs_retry; } else { crm_notice("Respawning %s subdaemon after unexpected exit", child->name); start_child(child); } } static gboolean pcmk_shutdown_worker(gpointer user_data) { static int phase = PCMK__NELEM(pcmk_children) - 1; static time_t next_log = 0; if (phase == PCMK__NELEM(pcmk_children) - 1) { crm_notice("Shutting down Pacemaker"); pacemakerd_state = XML_PING_ATTR_PACEMAKERDSTATE_SHUTTINGDOWN; } for (; phase >= 0; phase--) { pcmk_child_t *child = &(pcmk_children[phase]); if (child->pid != 0) { time_t now = time(NULL); if (pcmk_is_set(child->flags, child_respawn)) { if (child->pid == PCMK__SPECIAL_PID) { crm_warn("The process behind %s IPC cannot be" " terminated, so either wait the graceful" " period of %ld s for its native termination" " if it vitally depends on some other daemons" " going down in a controlled way already," " or locate and kill the correct %s process" " on your own; set PCMK_" PCMK__ENV_FAIL_FAST "=1" " to avoid this altogether next time around", child->name, (long) SHUTDOWN_ESCALATION_PERIOD, child->command); } next_log = now + 30; child->flags &= ~child_respawn; stop_child(child, SIGTERM); if (phase < PCMK_CHILD_CONTROLD) { g_timeout_add(SHUTDOWN_ESCALATION_PERIOD, escalate_shutdown, child); } } else if (now >= next_log) { next_log = now + 30; crm_notice("Still waiting for %s to terminate " CRM_XS " pid=%lld", child->name, (long long) child->pid); } return TRUE; } /* cleanup */ crm_debug("%s confirmed stopped", child->name); child->pid = 0; } crm_notice("Shutdown complete"); pacemakerd_state = XML_PING_ATTR_PACEMAKERDSTATE_SHUTDOWNCOMPLETE; if (!fatal_error && running_with_sbd && pcmk__get_sbd_sync_resource_startup() && !shutdown_complete_state_reported_client_closed) { crm_notice("Waiting for SBD to pick up shutdown-complete-state."); return TRUE; } // @COMPAT Drop shutdown delay at 3.0.0 { const char *delay = pcmk__env_option(PCMK__ENV_SHUTDOWN_DELAY); if(delay) { + long long delay_ms = crm_get_msec(delay); + sync(); - pcmk__sleep_ms(crm_get_msec(delay)); + if (delay_ms > 0) { + pcmk__sleep_ms((unsigned int) QB_MIN(delay_ms, UINT_MAX)); + } } } g_main_loop_quit(mainloop); if (fatal_error) { crm_notice("Shutting down and staying down after fatal error"); #ifdef SUPPORT_COROSYNC pcmkd_shutdown_corosync(); #endif crm_exit(CRM_EX_FATAL); } return TRUE; } /* TODO once libqb is taught to juggle with IPC end-points carried over as bare file descriptor (https://github.com/ClusterLabs/libqb/issues/325) it shall hand over these descriptors here if/once they are successfully pre-opened in (presumably) child_liveness(), to avoid any remaining room for races */ // \return Standard Pacemaker return code static int start_child(pcmk_child_t * child) { uid_t uid = 0; gid_t gid = 0; gboolean use_valgrind = FALSE; gboolean use_callgrind = FALSE; const char *env_valgrind = pcmk__env_option(PCMK__ENV_VALGRIND_ENABLED); const char *env_callgrind = pcmk__env_option(PCMK__ENV_CALLGRIND_ENABLED); child->flags &= ~child_active_before_startup; child->check_count = 0; if (child->command == NULL) { crm_info("Nothing to do for child \"%s\"", child->name); return pcmk_rc_ok; } if (env_callgrind != NULL && crm_is_true(env_callgrind)) { use_callgrind = TRUE; use_valgrind = TRUE; } else if (env_callgrind != NULL && strstr(env_callgrind, child->name)) { use_callgrind = TRUE; use_valgrind = TRUE; } else if (env_valgrind != NULL && crm_is_true(env_valgrind)) { use_valgrind = TRUE; } else if (env_valgrind != NULL && strstr(env_valgrind, child->name)) { use_valgrind = TRUE; } if (use_valgrind && strlen(VALGRIND_BIN) == 0) { crm_warn("Cannot enable valgrind for %s:" " The location of the valgrind binary is unknown", child->name); use_valgrind = FALSE; } if (child->uid) { if (crm_user_lookup(child->uid, &uid, &gid) < 0) { crm_err("Invalid user (%s) for %s: not found", child->uid, child->name); return EACCES; } crm_info("Using uid=%u and group=%u for process %s", uid, gid, child->name); } child->pid = fork(); CRM_ASSERT(child->pid != -1); if (child->pid > 0) { /* parent */ mainloop_child_add(child->pid, 0, child->name, child, pcmk_child_exit); crm_info("Forked child %lld for process %s%s", (long long) child->pid, child->name, use_valgrind ? " (valgrind enabled: " VALGRIND_BIN ")" : ""); return pcmk_rc_ok; } else { /* Start a new session */ (void)setsid(); /* Setup the two alternate arg arrays */ opts_vgrind[0] = strdup(VALGRIND_BIN); if (use_callgrind) { opts_vgrind[1] = strdup("--tool=callgrind"); opts_vgrind[2] = strdup("--callgrind-out-file=" CRM_STATE_DIR "/callgrind.out.%p"); opts_vgrind[3] = strdup(child->command); opts_vgrind[4] = NULL; } else { opts_vgrind[1] = strdup(child->command); opts_vgrind[2] = NULL; opts_vgrind[3] = NULL; opts_vgrind[4] = NULL; } opts_default[0] = strdup(child->command); if(gid) { // Drop root group access if not needed if (!need_root_group && (setgid(gid) < 0)) { crm_warn("Could not set group to %d: %s", gid, strerror(errno)); } /* Initialize supplementary groups to only those always granted to * the user, plus haclient (so we can access IPC). */ if (initgroups(child->uid, gid) < 0) { crm_err("Cannot initialize groups for %s: %s (%d)", child->uid, pcmk_rc_str(errno), errno); } } if (uid && setuid(uid) < 0) { crm_warn("Could not set user to %s (id %d): %s", child->uid, uid, strerror(errno)); } pcmk__close_fds_in_child(true); pcmk__open_devnull(O_RDONLY); // stdin (fd 0) pcmk__open_devnull(O_WRONLY); // stdout (fd 1) pcmk__open_devnull(O_WRONLY); // stderr (fd 2) if (use_valgrind) { (void)execvp(VALGRIND_BIN, opts_vgrind); } else { (void)execvp(child->command, opts_default); } crm_crit("Could not execute %s: %s", child->command, strerror(errno)); crm_exit(CRM_EX_FATAL); } return pcmk_rc_ok; /* never reached */ } /*! * \internal * \brief Check the liveness of the child based on IPC name and PID if tracked * * \param[in,out] child Child tracked data * * \return Standard Pacemaker return code * * \note Return codes of particular interest include pcmk_rc_ipc_unresponsive * indicating that no trace of IPC liveness was detected, * pcmk_rc_ipc_unauthorized indicating that the IPC endpoint is blocked by * an unauthorized process, and pcmk_rc_ipc_pid_only indicating that * the child is up by PID but not IPC end-point (possibly starting). * \note This function doesn't modify any of \p child members but \c pid, * and is not actively toying with processes as such but invoking * \c stop_child in one particular case (there's for some reason * a different authentic holder of the IPC end-point). */ static int child_liveness(pcmk_child_t *child) { uid_t cl_uid = 0; gid_t cl_gid = 0; const uid_t root_uid = 0; const gid_t root_gid = 0; const uid_t *ref_uid; const gid_t *ref_gid; int rc = pcmk_rc_ipc_unresponsive; pid_t ipc_pid = 0; if (child->endpoint == NULL && (child->pid <= 0 || child->pid == PCMK__SPECIAL_PID)) { crm_err("Cannot track child %s for missing both API end-point and PID", child->name); rc = EINVAL; // Misuse of function when child is not trackable } else if (child->endpoint != NULL) { int legacy_rc = pcmk_ok; if (child->uid == NULL) { ref_uid = &root_uid; ref_gid = &root_gid; } else { ref_uid = &cl_uid; ref_gid = &cl_gid; legacy_rc = pcmk_daemon_user(&cl_uid, &cl_gid); } if (legacy_rc < 0) { rc = pcmk_legacy2rc(legacy_rc); crm_err("Could not find user and group IDs for user %s: %s " CRM_XS " rc=%d", CRM_DAEMON_USER, pcmk_rc_str(rc), rc); } else { rc = pcmk__ipc_is_authentic_process_active(child->endpoint, *ref_uid, *ref_gid, &ipc_pid); if ((rc == pcmk_rc_ok) || (rc == pcmk_rc_ipc_unresponsive)) { if (child->pid <= 0) { /* If rc is pcmk_rc_ok, ipc_pid is nonzero and this * initializes a new child. If rc is * pcmk_rc_ipc_unresponsive, ipc_pid is zero, and we will * investigate further. */ child->pid = ipc_pid; } else if ((ipc_pid != 0) && (child->pid != ipc_pid)) { /* An unexpected (but authorized) process is responding to * IPC. Investigate further. */ rc = pcmk_rc_ipc_unresponsive; } } } } if (rc == pcmk_rc_ipc_unresponsive) { /* If we get here, a child without IPC is being tracked, no IPC liveness * has been detected, or IPC liveness has been detected with an * unexpected (but authorized) process. This is safe on FreeBSD since * the only change possible from a proper child's PID into "special" PID * of 1 behind more loosely related process. */ int ret = pcmk__pid_active(child->pid, child->name); if (ipc_pid && ((ret != pcmk_rc_ok) || ipc_pid == PCMK__SPECIAL_PID || (pcmk__pid_active(ipc_pid, child->name) == pcmk_rc_ok))) { /* An unexpected (but authorized) process was detected at the IPC * endpoint, and either it is active, or the child we're tracking is * not. */ if (ret == pcmk_rc_ok) { /* The child we're tracking is active. Kill it, and adopt the * detected process. This assumes that our children don't fork * (thus getting a different PID owning the IPC), but rather the * tracking got out of sync because of some means external to * Pacemaker, and adopting the detected process is better than * killing it and possibly having to spawn a new child. */ /* not possessing IPC, afterall (what about corosync CPG?) */ stop_child(child, SIGKILL); } rc = pcmk_rc_ok; child->pid = ipc_pid; } else if (ret == pcmk_rc_ok) { // Our tracked child's PID was found active, but not its IPC rc = pcmk_rc_ipc_pid_only; } else if ((child->pid == 0) && (ret == EINVAL)) { // FreeBSD can return EINVAL rc = pcmk_rc_ipc_unresponsive; } else { switch (ret) { case EACCES: rc = pcmk_rc_ipc_unauthorized; break; case ESRCH: rc = pcmk_rc_ipc_unresponsive; break; default: rc = ret; break; } } } return rc; } /*! * \internal * \brief Initial one-off check of the pre-existing "child" processes * * With "child" process, we mean the subdaemon that defines an API end-point * (all of them do as of the comment) -- the possible complement is skipped * as it is deemed it has no such shared resources to cause conflicts about, * hence it can presumably be started anew without hesitation. * If that won't hold true in the future, the concept of a shared resource * will have to be generalized beyond the API end-point. * * For boundary cases that the "child" is still starting (IPC end-point is yet * to be witnessed), or more rarely (practically FreeBSD only), when there's * a pre-existing "untrackable" authentic process, we give the situation some * time to possibly unfold in the right direction, meaning that said socket * will appear or the unattainable process will disappear per the observable * IPC, respectively. * * \return Standard Pacemaker return code * * \note Since this gets run at the very start, \c respawn_count fields * for particular children get temporarily overloaded with "rounds * of waiting" tracking, restored once we are about to finish with * success (i.e. returning value >=0) and will remain unrestored * otherwise. One way to suppress liveness detection logic for * particular child is to set the said value to a negative number. */ #define WAIT_TRIES 4 /* together with interleaved sleeps, worst case ~ 1s */ int find_and_track_existing_processes(void) { bool wait_in_progress; int rc; size_t i, rounds; for (rounds = 1; rounds <= WAIT_TRIES; rounds++) { wait_in_progress = false; for (i = 0; i < PCMK__NELEM(pcmk_children); i++) { if ((pcmk_children[i].endpoint == NULL) || (pcmk_children[i].respawn_count < 0)) { continue; } rc = child_liveness(&pcmk_children[i]); if (rc == pcmk_rc_ipc_unresponsive) { /* As a speculation, don't give up if there are more rounds to * come for other reasons, but don't artificially wait just * because of this, since we would preferably start ASAP. */ continue; } // @TODO Functionize more of this to reduce nesting pcmk_children[i].respawn_count = rounds; switch (rc) { case pcmk_rc_ok: if (pcmk_children[i].pid == PCMK__SPECIAL_PID) { if (crm_is_true(pcmk__env_option(PCMK__ENV_FAIL_FAST))) { crm_crit("Cannot reliably track pre-existing" " authentic process behind %s IPC on this" " platform and PCMK_" PCMK__ENV_FAIL_FAST " requested", pcmk_children[i].endpoint); return EOPNOTSUPP; } else if (pcmk_children[i].respawn_count == WAIT_TRIES) { crm_notice("Assuming pre-existing authentic, though" " on this platform untrackable, process" " behind %s IPC is stable (was in %d" " previous samples) so rather than" " bailing out (PCMK_" PCMK__ENV_FAIL_FAST " not requested), we just switch to a" " less optimal IPC liveness monitoring" " (not very suitable for heavy load)", pcmk_children[i].name, WAIT_TRIES - 1); crm_warn("The process behind %s IPC cannot be" " terminated, so the overall shutdown" " will get delayed implicitly (%ld s)," " which serves as a graceful period for" " its native termination if it vitally" " depends on some other daemons going" " down in a controlled way already", pcmk_children[i].name, (long) SHUTDOWN_ESCALATION_PERIOD); } else { wait_in_progress = true; crm_warn("Cannot reliably track pre-existing" " authentic process behind %s IPC on this" " platform, can still disappear in %d" " attempt(s)", pcmk_children[i].endpoint, WAIT_TRIES - pcmk_children[i].respawn_count); continue; } } crm_notice("Tracking existing %s process (pid=%lld)", pcmk_children[i].name, (long long) PCMK__SPECIAL_PID_AS_0( pcmk_children[i].pid)); pcmk_children[i].respawn_count = -1; /* 0~keep watching */ pcmk_children[i].flags |= child_active_before_startup; break; case pcmk_rc_ipc_pid_only: if (pcmk_children[i].respawn_count == WAIT_TRIES) { crm_crit("%s IPC end-point for existing authentic" " process %lld did not (re)appear", pcmk_children[i].endpoint, (long long) PCMK__SPECIAL_PID_AS_0( pcmk_children[i].pid)); return rc; } wait_in_progress = true; crm_warn("Cannot find %s IPC end-point for existing" " authentic process %lld, can still (re)appear" " in %d attempts (?)", pcmk_children[i].endpoint, (long long) PCMK__SPECIAL_PID_AS_0( pcmk_children[i].pid), WAIT_TRIES - pcmk_children[i].respawn_count); continue; default: crm_crit("Checked liveness of %s: %s " CRM_XS " rc=%d", pcmk_children[i].name, pcmk_rc_str(rc), rc); return rc; } } if (!wait_in_progress) { break; } pcmk__sleep_ms(250); // Wait a bit for changes to possibly happen } for (i = 0; i < PCMK__NELEM(pcmk_children); i++) { pcmk_children[i].respawn_count = 0; /* restore pristine state */ } g_timeout_add_seconds(PCMK_PROCESS_CHECK_INTERVAL, check_next_subdaemon, NULL); return pcmk_rc_ok; } gboolean init_children_processes(void *user_data) { if (is_corosync_cluster()) { /* Corosync clusters can drop root group access, because we set * uidgid.gid.${gid}=1 via CMAP, which allows these processes to connect * to corosync. */ need_root_group = false; } /* start any children that have not been detected */ for (int i = 0; i < PCMK__NELEM(pcmk_children); i++) { if (pcmk_children[i].pid != 0) { /* we are already tracking it */ continue; } start_child(&(pcmk_children[i])); } /* From this point on, any daemons being started will be due to * respawning rather than node start. * * This may be useful for the daemons to know */ pcmk__set_env_option(PCMK__ENV_RESPAWNED, PCMK_VALUE_TRUE, false); pacemakerd_state = XML_PING_ATTR_PACEMAKERDSTATE_RUNNING; return TRUE; } void pcmk_shutdown(int nsig) { if (shutdown_trigger == NULL) { shutdown_trigger = mainloop_add_trigger(G_PRIORITY_HIGH, pcmk_shutdown_worker, NULL); } mainloop_set_trigger(shutdown_trigger); } void restart_cluster_subdaemons(void) { for (int i = 0; i < PCMK__NELEM(pcmk_children); i++) { if (!pcmk_is_set(pcmk_children[i].flags, child_needs_retry) || pcmk_children[i].pid != 0) { continue; } crm_notice("Respawning cluster-based subdaemon: %s", pcmk_children[i].name); if (start_child(&pcmk_children[i])) { pcmk_children[i].flags &= ~child_needs_retry; } } } static gboolean stop_child(pcmk_child_t * child, int signal) { if (signal == 0) { signal = SIGTERM; } /* why to skip PID of 1? - FreeBSD ~ how untrackable process behind IPC is masqueraded as - elsewhere: how "init" task is designated; in particular, in systemd arrangement of socket-based activation, this is pretty real */ if (child->command == NULL || child->pid == PCMK__SPECIAL_PID) { crm_debug("Nothing to do for child \"%s\" (process %lld)", child->name, (long long) PCMK__SPECIAL_PID_AS_0(child->pid)); return TRUE; } if (child->pid <= 0) { crm_trace("Client %s not running", child->name); return TRUE; } errno = 0; if (kill(child->pid, signal) == 0) { crm_notice("Stopping %s "CRM_XS" sent signal %d to process %lld", child->name, signal, (long long) child->pid); } else { crm_err("Could not stop %s (process %lld) with signal %d: %s", child->name, (long long) child->pid, signal, strerror(errno)); } return TRUE; } diff --git a/lib/pengine/pe_actions.c b/lib/pengine/pe_actions.c index e9238e0dd7..eca0ee74c6 100644 --- a/lib/pengine/pe_actions.c +++ b/lib/pengine/pe_actions.c @@ -1,1894 +1,1896 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include "pe_status_private.h" static void unpack_operation(pcmk_action_t *action, const xmlNode *xml_obj, guint interval_ms); static void add_singleton(pcmk_scheduler_t *scheduler, pcmk_action_t *action) { if (scheduler->singletons == NULL) { scheduler->singletons = pcmk__strkey_table(NULL, NULL); } g_hash_table_insert(scheduler->singletons, action->uuid, action); } static pcmk_action_t * lookup_singleton(pcmk_scheduler_t *scheduler, const char *action_uuid) { if (scheduler->singletons == NULL) { return NULL; } return g_hash_table_lookup(scheduler->singletons, action_uuid); } /*! * \internal * \brief Find an existing action that matches arguments * * \param[in] key Action key to match * \param[in] rsc Resource to match (if any) * \param[in] node Node to match (if any) * \param[in] scheduler Scheduler data * * \return Existing action that matches arguments (or NULL if none) */ static pcmk_action_t * find_existing_action(const char *key, const pcmk_resource_t *rsc, const pcmk_node_t *node, const pcmk_scheduler_t *scheduler) { GList *matches = NULL; pcmk_action_t *action = NULL; /* When rsc is NULL, it would be quicker to check scheduler->singletons, * but checking all scheduler->actions takes the node into account. */ matches = find_actions(((rsc == NULL)? scheduler->actions : rsc->actions), key, node); if (matches == NULL) { return NULL; } CRM_LOG_ASSERT(!pcmk__list_of_multiple(matches)); action = matches->data; g_list_free(matches); return action; } /*! * \internal * \brief Find the XML configuration corresponding to a specific action key * * \param[in] rsc Resource to find action configuration for * \param[in] key "RSC_ACTION_INTERVAL" of action to find * \param[in] include_disabled If false, do not return disabled actions * * \return XML configuration of desired action if any, otherwise NULL */ static xmlNode * find_exact_action_config(const pcmk_resource_t *rsc, const char *action_name, guint interval_ms, bool include_disabled) { for (xmlNode *operation = first_named_child(rsc->ops_xml, PCMK_XE_OP); operation != NULL; operation = crm_next_same_xml(operation)) { bool enabled = false; const char *config_name = NULL; const char *interval_spec = NULL; guint tmp_ms = 0U; // @TODO This does not consider meta-attributes, rules, defaults, etc. if (!include_disabled && (pcmk__xe_get_bool_attr(operation, PCMK_META_ENABLED, &enabled) == pcmk_rc_ok) && !enabled) { continue; } interval_spec = crm_element_value(operation, PCMK_META_INTERVAL); pcmk_parse_interval_spec(interval_spec, &tmp_ms); if (tmp_ms != interval_ms) { continue; } config_name = crm_element_value(operation, PCMK_XA_NAME); if (pcmk__str_eq(action_name, config_name, pcmk__str_none)) { return operation; } } return NULL; } /*! * \internal * \brief Find the XML configuration of a resource action * * \param[in] rsc Resource to find action configuration for * \param[in] action_name Action name to search for * \param[in] interval_ms Action interval (in milliseconds) to search for * \param[in] include_disabled If false, do not return disabled actions * * \return XML configuration of desired action if any, otherwise NULL */ xmlNode * pcmk__find_action_config(const pcmk_resource_t *rsc, const char *action_name, guint interval_ms, bool include_disabled) { xmlNode *action_config = NULL; // Try requested action first action_config = find_exact_action_config(rsc, action_name, interval_ms, include_disabled); // For migrate_to and migrate_from actions, retry with "migrate" // @TODO This should be either documented or deprecated if ((action_config == NULL) && pcmk__str_any_of(action_name, PCMK_ACTION_MIGRATE_TO, PCMK_ACTION_MIGRATE_FROM, NULL)) { action_config = find_exact_action_config(rsc, "migrate", 0, include_disabled); } return action_config; } /*! * \internal * \brief Create a new action object * * \param[in] key Action key * \param[in] task Action name * \param[in,out] rsc Resource that action is for (if any) * \param[in] node Node that action is on (if any) * \param[in] optional Whether action should be considered optional * \param[in,out] scheduler Scheduler data * * \return Newly allocated action * \note This function takes ownership of \p key. It is the caller's * responsibility to free the return value with pe_free_action(). */ static pcmk_action_t * new_action(char *key, const char *task, pcmk_resource_t *rsc, const pcmk_node_t *node, bool optional, pcmk_scheduler_t *scheduler) { pcmk_action_t *action = calloc(1, sizeof(pcmk_action_t)); CRM_ASSERT(action != NULL); action->rsc = rsc; action->task = strdup(task); CRM_ASSERT(action->task != NULL); action->uuid = key; if (node) { action->node = pe__copy_node(node); } if (pcmk__str_eq(task, PCMK_ACTION_LRM_DELETE, pcmk__str_casei)) { // Resource history deletion for a node can be done on the DC pcmk__set_action_flags(action, pcmk_action_on_dc); } pcmk__set_action_flags(action, pcmk_action_runnable); if (optional) { pcmk__set_action_flags(action, pcmk_action_optional); } else { pcmk__clear_action_flags(action, pcmk_action_optional); } if (rsc == NULL) { action->meta = pcmk__strkey_table(free, free); } else { guint interval_ms = 0; parse_op_key(key, NULL, NULL, &interval_ms); action->op_entry = pcmk__find_action_config(rsc, task, interval_ms, true); /* If the given key is for one of the many notification pseudo-actions * (pre_notify_promote, etc.), the actual action name is "notify" */ if ((action->op_entry == NULL) && (strstr(key, "_notify_") != NULL)) { action->op_entry = find_exact_action_config(rsc, PCMK_ACTION_NOTIFY, 0, true); } unpack_operation(action, action->op_entry, interval_ms); } pcmk__rsc_trace(rsc, "Created %s action %d (%s): %s for %s on %s", (optional? "optional" : "required"), scheduler->action_id, key, task, ((rsc == NULL)? "no resource" : rsc->id), pcmk__node_name(node)); action->id = scheduler->action_id++; scheduler->actions = g_list_prepend(scheduler->actions, action); if (rsc == NULL) { add_singleton(scheduler, action); } else { rsc->actions = g_list_prepend(rsc->actions, action); } return action; } /*! * \internal * \brief Unpack a resource's action-specific instance parameters * * \param[in] action_xml XML of action's configuration in CIB (if any) * \param[in,out] node_attrs Table of node attributes (for rule evaluation) * \param[in,out] scheduler Cluster working set (for rule evaluation) * * \return Newly allocated hash table of action-specific instance parameters */ GHashTable * pcmk__unpack_action_rsc_params(const xmlNode *action_xml, GHashTable *node_attrs, pcmk_scheduler_t *scheduler) { GHashTable *params = pcmk__strkey_table(free, free); pe_rule_eval_data_t rule_data = { .node_hash = node_attrs, .role = pcmk_role_unknown, .now = scheduler->now, .match_data = NULL, .rsc_data = NULL, .op_data = NULL }; pe__unpack_dataset_nvpairs(action_xml, PCMK_XE_INSTANCE_ATTRIBUTES, &rule_data, params, NULL, FALSE, scheduler); return params; } /*! * \internal * \brief Update an action's optional flag * * \param[in,out] action Action to update * \param[in] optional Requested optional status */ static void update_action_optional(pcmk_action_t *action, gboolean optional) { // Force a non-recurring action to be optional if its resource is unmanaged if ((action->rsc != NULL) && (action->node != NULL) && !pcmk_is_set(action->flags, pcmk_action_pseudo) && !pcmk_is_set(action->rsc->flags, pcmk_rsc_managed) && (g_hash_table_lookup(action->meta, PCMK_META_INTERVAL) == NULL)) { pcmk__rsc_debug(action->rsc, "%s on %s is optional (%s is unmanaged)", action->uuid, pcmk__node_name(action->node), action->rsc->id); pcmk__set_action_flags(action, pcmk_action_optional); // We shouldn't clear runnable here because ... something // Otherwise require the action if requested } else if (!optional) { pcmk__clear_action_flags(action, pcmk_action_optional); } } static enum pe_quorum_policy effective_quorum_policy(pcmk_resource_t *rsc, pcmk_scheduler_t *scheduler) { enum pe_quorum_policy policy = scheduler->no_quorum_policy; if (pcmk_is_set(scheduler->flags, pcmk_sched_quorate)) { policy = pcmk_no_quorum_ignore; } else if (scheduler->no_quorum_policy == pcmk_no_quorum_demote) { switch (rsc->role) { case pcmk_role_promoted: case pcmk_role_unpromoted: if (rsc->next_role > pcmk_role_unpromoted) { pe__set_next_role(rsc, pcmk_role_unpromoted, PCMK_OPT_NO_QUORUM_POLICY "=demote"); } policy = pcmk_no_quorum_ignore; break; default: policy = pcmk_no_quorum_stop; break; } } return policy; } /*! * \internal * \brief Update a resource action's runnable flag * * \param[in,out] action Action to update * \param[in,out] scheduler Scheduler data * * \note This may also schedule fencing if a stop is unrunnable. */ static void update_resource_action_runnable(pcmk_action_t *action, pcmk_scheduler_t *scheduler) { if (pcmk_is_set(action->flags, pcmk_action_pseudo)) { return; } if (action->node == NULL) { pcmk__rsc_trace(action->rsc, "%s is unrunnable (unallocated)", action->uuid); pcmk__clear_action_flags(action, pcmk_action_runnable); } else if (!pcmk_is_set(action->flags, pcmk_action_on_dc) && !(action->node->details->online) && (!pe__is_guest_node(action->node) || action->node->details->remote_requires_reset)) { pcmk__clear_action_flags(action, pcmk_action_runnable); do_crm_log(LOG_WARNING, "%s on %s is unrunnable (node is offline)", action->uuid, pcmk__node_name(action->node)); if (pcmk_is_set(action->rsc->flags, pcmk_rsc_managed) && pcmk__str_eq(action->task, PCMK_ACTION_STOP, pcmk__str_casei) && !(action->node->details->unclean)) { pe_fence_node(scheduler, action->node, "stop is unrunnable", false); } } else if (!pcmk_is_set(action->flags, pcmk_action_on_dc) && action->node->details->pending) { pcmk__clear_action_flags(action, pcmk_action_runnable); do_crm_log(LOG_WARNING, "Action %s on %s is unrunnable (node is pending)", action->uuid, pcmk__node_name(action->node)); } else if (action->needs == pcmk_requires_nothing) { pe_action_set_reason(action, NULL, TRUE); if (pe__is_guest_node(action->node) && !pe_can_fence(scheduler, action->node)) { /* An action that requires nothing usually does not require any * fencing in order to be runnable. However, there is an exception: * such an action cannot be completed if it is on a guest node whose * host is unclean and cannot be fenced. */ pcmk__rsc_debug(action->rsc, "%s on %s is unrunnable " "(node's host cannot be fenced)", action->uuid, pcmk__node_name(action->node)); pcmk__clear_action_flags(action, pcmk_action_runnable); } else { pcmk__rsc_trace(action->rsc, "%s on %s does not require fencing or quorum", action->uuid, pcmk__node_name(action->node)); pcmk__set_action_flags(action, pcmk_action_runnable); } } else { switch (effective_quorum_policy(action->rsc, scheduler)) { case pcmk_no_quorum_stop: pcmk__rsc_debug(action->rsc, "%s on %s is unrunnable (no quorum)", action->uuid, pcmk__node_name(action->node)); pcmk__clear_action_flags(action, pcmk_action_runnable); pe_action_set_reason(action, "no quorum", true); break; case pcmk_no_quorum_freeze: if (!action->rsc->fns->active(action->rsc, TRUE) || (action->rsc->next_role > action->rsc->role)) { pcmk__rsc_debug(action->rsc, "%s on %s is unrunnable (no quorum)", action->uuid, pcmk__node_name(action->node)); pcmk__clear_action_flags(action, pcmk_action_runnable); pe_action_set_reason(action, "quorum freeze", true); } break; default: //pe_action_set_reason(action, NULL, TRUE); pcmk__set_action_flags(action, pcmk_action_runnable); break; } } } /*! * \internal * \brief Update a resource object's flags for a new action on it * * \param[in,out] rsc Resource that action is for (if any) * \param[in] action New action */ static void update_resource_flags_for_action(pcmk_resource_t *rsc, const pcmk_action_t *action) { /* @COMPAT pcmk_rsc_starting and pcmk_rsc_stopping are deprecated and unused * within Pacemaker, and will eventually be removed */ if (pcmk__str_eq(action->task, PCMK_ACTION_STOP, pcmk__str_casei)) { pcmk__set_rsc_flags(rsc, pcmk_rsc_stopping); } else if (pcmk__str_eq(action->task, PCMK_ACTION_START, pcmk__str_casei)) { if (pcmk_is_set(action->flags, pcmk_action_runnable)) { pcmk__set_rsc_flags(rsc, pcmk_rsc_starting); } else { pcmk__clear_rsc_flags(rsc, pcmk_rsc_starting); } } } static bool valid_stop_on_fail(const char *value) { return !pcmk__strcase_any_of(value, "standby", "demote", "stop", NULL); } /*! * \internal * \brief Validate (and possibly reset) resource action's on_fail meta-attribute * * \param[in] rsc Resource that action is for * \param[in] action_name Action name * \param[in] action_config Action configuration XML from CIB (if any) * \param[in,out] meta Table of action meta-attributes */ static void validate_on_fail(const pcmk_resource_t *rsc, const char *action_name, const xmlNode *action_config, GHashTable *meta) { const char *name = NULL; const char *role = NULL; const char *interval_spec = NULL; const char *value = g_hash_table_lookup(meta, PCMK_META_ON_FAIL); char *key = NULL; char *new_value = NULL; guint interval_ms = 0U; // Stop actions can only use certain on-fail values if (pcmk__str_eq(action_name, PCMK_ACTION_STOP, pcmk__str_none) && !valid_stop_on_fail(value)) { pcmk__config_err("Resetting '" PCMK_META_ON_FAIL "' for %s stop " "action to default value because '%s' is not " "allowed for stop", rsc->id, value); g_hash_table_remove(meta, PCMK_META_ON_FAIL); return; } /* Demote actions default on-fail to the on-fail value for the first * recurring monitor for the promoted role (if any). */ if (pcmk__str_eq(action_name, PCMK_ACTION_DEMOTE, pcmk__str_none) && (value == NULL)) { /* @TODO This does not consider promote options set in a meta-attribute * block (which may have rules that need to be evaluated) rather than * XML properties. */ for (xmlNode *operation = first_named_child(rsc->ops_xml, PCMK_XE_OP); operation != NULL; operation = crm_next_same_xml(operation)) { bool enabled = false; const char *promote_on_fail = NULL; /* We only care about explicit on-fail (if promote uses default, so * can demote) */ promote_on_fail = crm_element_value(operation, PCMK_META_ON_FAIL); if (promote_on_fail == NULL) { continue; } // We only care about recurring monitors for the promoted role name = crm_element_value(operation, PCMK_XA_NAME); role = crm_element_value(operation, PCMK_XA_ROLE); if (!pcmk__str_eq(name, PCMK_ACTION_MONITOR, pcmk__str_none) || !pcmk__strcase_any_of(role, PCMK__ROLE_PROMOTED, PCMK__ROLE_PROMOTED_LEGACY, NULL)) { continue; } interval_spec = crm_element_value(operation, PCMK_META_INTERVAL); pcmk_parse_interval_spec(interval_spec, &interval_ms); if (interval_ms == 0U) { continue; } // We only care about enabled monitors if ((pcmk__xe_get_bool_attr(operation, PCMK_META_ENABLED, &enabled) == pcmk_rc_ok) && !enabled) { continue; } // Demote actions can't default to on-fail="demote" if (pcmk__str_eq(promote_on_fail, "demote", pcmk__str_casei)) { continue; } // Use value from first applicable promote action found key = strdup(PCMK_META_ON_FAIL); new_value = strdup(promote_on_fail); CRM_ASSERT((key != NULL) && (new_value != NULL)); g_hash_table_insert(meta, key, new_value); } return; } if (pcmk__str_eq(action_name, PCMK_ACTION_LRM_DELETE, pcmk__str_none) && !pcmk__str_eq(value, "ignore", pcmk__str_casei)) { key = strdup(PCMK_META_ON_FAIL); new_value = strdup("ignore"); CRM_ASSERT((key != NULL) && (new_value != NULL)); g_hash_table_insert(meta, key, new_value); return; } // on-fail="demote" is allowed only for certain actions if (pcmk__str_eq(value, "demote", pcmk__str_casei)) { name = crm_element_value(action_config, PCMK_XA_NAME); role = crm_element_value(action_config, PCMK_XA_ROLE); interval_spec = crm_element_value(action_config, PCMK_META_INTERVAL); pcmk_parse_interval_spec(interval_spec, &interval_ms); if (!pcmk__str_eq(name, PCMK_ACTION_PROMOTE, pcmk__str_none) && ((interval_ms == 0U) || !pcmk__str_eq(name, PCMK_ACTION_MONITOR, pcmk__str_none) || !pcmk__strcase_any_of(role, PCMK__ROLE_PROMOTED, PCMK__ROLE_PROMOTED_LEGACY, NULL))) { pcmk__config_err("Resetting '" PCMK_META_ON_FAIL "' for %s %s " "action to default value because 'demote' is not " "allowed for it", rsc->id, name); g_hash_table_remove(meta, PCMK_META_ON_FAIL); return; } } } static int unpack_timeout(const char *value) { - int timeout_ms = crm_get_msec(value); + long long timeout_ms = crm_get_msec(value); if (timeout_ms < 0) { timeout_ms = PCMK_DEFAULT_ACTION_TIMEOUT_MS; } - return timeout_ms; + return (int) QB_MIN(timeout_ms, INT_MAX); } // true if value contains valid, non-NULL interval origin for recurring op static bool unpack_interval_origin(const char *value, const xmlNode *xml_obj, guint interval_ms, const crm_time_t *now, long long *start_delay) { long long result = 0; guint interval_sec = interval_ms / 1000; crm_time_t *origin = NULL; // Ignore unspecified values and non-recurring operations if ((value == NULL) || (interval_ms == 0) || (now == NULL)) { return false; } // Parse interval origin from text origin = crm_time_new(value); if (origin == NULL) { pcmk__config_err("Ignoring '" PCMK_META_INTERVAL_ORIGIN "' for " "operation '%s' because '%s' is not valid", (ID(xml_obj)? ID(xml_obj) : "(missing ID)"), value); return false; } // Get seconds since origin (negative if origin is in the future) result = crm_time_get_seconds(now) - crm_time_get_seconds(origin); crm_time_free(origin); // Calculate seconds from closest interval to now result = result % interval_sec; // Calculate seconds remaining until next interval result = ((result <= 0)? 0 : interval_sec) - result; crm_info("Calculated a start delay of %llds for operation '%s'", result, (ID(xml_obj)? ID(xml_obj) : "(unspecified)")); if (start_delay != NULL) { *start_delay = result * 1000; // milliseconds } return true; } static int unpack_start_delay(const char *value, GHashTable *meta) { - int start_delay = 0; + long long start_delay_ms = 0; - if (value != NULL) { - start_delay = crm_get_msec(value); + if (value == NULL) { + return 0; + } - if (start_delay < 0) { - start_delay = 0; - } + start_delay_ms = crm_get_msec(value); + start_delay_ms = QB_MIN(start_delay_ms, INT_MAX); + if (start_delay_ms < 0) { + start_delay_ms = 0; + } - if (meta) { - g_hash_table_replace(meta, strdup(PCMK_META_START_DELAY), - pcmk__itoa(start_delay)); - } + if (meta != NULL) { + g_hash_table_replace(meta, strdup(PCMK_META_START_DELAY), + pcmk__itoa(start_delay_ms)); } - return start_delay; + return (int) start_delay_ms; } /*! * \internal * \brief Find a resource's most frequent recurring monitor * * \param[in] rsc Resource to check * * \return Operation XML configured for most frequent recurring monitor for * \p rsc (if any) */ static xmlNode * most_frequent_monitor(const pcmk_resource_t *rsc) { guint min_interval_ms = G_MAXUINT; xmlNode *op = NULL; for (xmlNode *operation = first_named_child(rsc->ops_xml, PCMK_XE_OP); operation != NULL; operation = crm_next_same_xml(operation)) { bool enabled = false; guint interval_ms = 0U; const char *interval_spec = crm_element_value(operation, PCMK_META_INTERVAL); // We only care about enabled recurring monitors if (!pcmk__str_eq(crm_element_value(operation, PCMK_XA_NAME), PCMK_ACTION_MONITOR, pcmk__str_none)) { continue; } pcmk_parse_interval_spec(interval_spec, &interval_ms); if (interval_ms == 0U) { continue; } // @TODO This does not consider meta-attributes, rules, defaults, etc. if ((pcmk__xe_get_bool_attr(operation, PCMK_META_ENABLED, &enabled) == pcmk_rc_ok) && !enabled) { continue; } if (interval_ms < min_interval_ms) { min_interval_ms = interval_ms; op = operation; } } return op; } /*! * \internal * \brief Unpack action meta-attributes * * \param[in,out] rsc Resource that action is for * \param[in] node Node that action is on * \param[in] action_name Action name * \param[in] interval_ms Action interval (in milliseconds) * \param[in] action_config Action XML configuration from CIB (if any) * * Unpack a resource action's meta-attributes (normalizing the interval, * timeout, and start delay values as integer milliseconds) from its CIB XML * configuration (including defaults). * * \return Newly allocated hash table with normalized action meta-attributes */ GHashTable * pcmk__unpack_action_meta(pcmk_resource_t *rsc, const pcmk_node_t *node, const char *action_name, guint interval_ms, const xmlNode *action_config) { GHashTable *meta = NULL; char *name = NULL; char *value = NULL; const char *timeout_spec = NULL; const char *str = NULL; pe_rsc_eval_data_t rsc_rule_data = { .standard = crm_element_value(rsc->xml, PCMK_XA_CLASS), .provider = crm_element_value(rsc->xml, PCMK_XA_PROVIDER), .agent = crm_element_value(rsc->xml, PCMK_XA_TYPE), }; pe_op_eval_data_t op_rule_data = { .op_name = action_name, .interval = interval_ms, }; pe_rule_eval_data_t rule_data = { .node_hash = (node == NULL)? NULL : node->details->attrs, .role = pcmk_role_unknown, .now = rsc->cluster->now, .match_data = NULL, .rsc_data = &rsc_rule_data, .op_data = &op_rule_data, }; meta = pcmk__strkey_table(free, free); // Cluster-wide pe__unpack_dataset_nvpairs(rsc->cluster->op_defaults, PCMK_XE_META_ATTRIBUTES, &rule_data, meta, NULL, FALSE, rsc->cluster); // Derive default timeout for probes from recurring monitor timeouts if (pcmk_is_probe(action_name, interval_ms)) { xmlNode *min_interval_mon = most_frequent_monitor(rsc); if (min_interval_mon != NULL) { /* @TODO This does not consider timeouts set in * PCMK_XE_META_ATTRIBUTES blocks (which may also have rules that * need to be evaluated). */ timeout_spec = crm_element_value(min_interval_mon, PCMK_META_TIMEOUT); if (timeout_spec != NULL) { pcmk__rsc_trace(rsc, "Setting default timeout for %s probe to " "most frequent monitor's timeout '%s'", rsc->id, timeout_spec); name = strdup(PCMK_META_TIMEOUT); value = strdup(timeout_spec); CRM_ASSERT((name != NULL) && (value != NULL)); g_hash_table_insert(meta, name, value); } } } if (action_config != NULL) { // take precedence over defaults pe__unpack_dataset_nvpairs(action_config, PCMK_XE_META_ATTRIBUTES, &rule_data, meta, NULL, TRUE, rsc->cluster); /* Anything set as an XML property has highest precedence. * This ensures we use the name and interval from the tag. * (See below for the only exception, fence device start/probe timeout.) */ for (xmlAttrPtr attr = action_config->properties; attr != NULL; attr = attr->next) { name = strdup((const char *) attr->name); value = strdup(pcmk__xml_attr_value(attr)); CRM_ASSERT((name != NULL) && (value != NULL)); g_hash_table_insert(meta, name, value); } } g_hash_table_remove(meta, PCMK_XA_ID); // Normalize interval to milliseconds if (interval_ms > 0) { name = strdup(PCMK_META_INTERVAL); CRM_ASSERT(name != NULL); value = crm_strdup_printf("%u", interval_ms); g_hash_table_insert(meta, name, value); } else { g_hash_table_remove(meta, PCMK_META_INTERVAL); } /* Timeout order of precedence (highest to lowest): * 1. pcmk_monitor_timeout resource parameter (only for starts and probes * when rsc has pcmk_ra_cap_fence_params; this gets used for recurring * monitors via the executor instead) * 2. timeout configured in (with taking precedence over * ) * 3. timeout configured in * 4. PCMK_DEFAULT_ACTION_TIMEOUT_MS */ // Check for pcmk_monitor_timeout if (pcmk_is_set(pcmk_get_ra_caps(rsc_rule_data.standard), pcmk_ra_cap_fence_params) && (pcmk__str_eq(action_name, PCMK_ACTION_START, pcmk__str_none) || pcmk_is_probe(action_name, interval_ms))) { GHashTable *params = pe_rsc_params(rsc, node, rsc->cluster); timeout_spec = g_hash_table_lookup(params, "pcmk_monitor_timeout"); if (timeout_spec != NULL) { pcmk__rsc_trace(rsc, "Setting timeout for %s %s to " "pcmk_monitor_timeout (%s)", rsc->id, action_name, timeout_spec); name = strdup(PCMK_META_TIMEOUT); value = strdup(timeout_spec); CRM_ASSERT((name != NULL) && (value != NULL)); g_hash_table_insert(meta, name, value); } } // Normalize timeout to positive milliseconds name = strdup(PCMK_META_TIMEOUT); CRM_ASSERT(name != NULL); timeout_spec = g_hash_table_lookup(meta, PCMK_META_TIMEOUT); g_hash_table_insert(meta, name, pcmk__itoa(unpack_timeout(timeout_spec))); // Ensure on-fail has a valid value validate_on_fail(rsc, action_name, action_config, meta); // Normalize PCMK_META_START_DELAY str = g_hash_table_lookup(meta, PCMK_META_START_DELAY); if (str != NULL) { unpack_start_delay(str, meta); } else { long long start_delay = 0; str = g_hash_table_lookup(meta, PCMK_META_INTERVAL_ORIGIN); if (unpack_interval_origin(str, action_config, interval_ms, rsc->cluster->now, &start_delay)) { name = strdup(PCMK_META_START_DELAY); CRM_ASSERT(name != NULL); g_hash_table_insert(meta, name, crm_strdup_printf("%lld", start_delay)); } } return meta; } /*! * \internal * \brief Determine an action's quorum and fencing dependency * * \param[in] rsc Resource that action is for * \param[in] action_name Name of action being unpacked * * \return Quorum and fencing dependency appropriate to action */ enum rsc_start_requirement pcmk__action_requires(const pcmk_resource_t *rsc, const char *action_name) { const char *value = NULL; enum rsc_start_requirement requires = pcmk_requires_nothing; CRM_CHECK((rsc != NULL) && (action_name != NULL), return requires); if (!pcmk__strcase_any_of(action_name, PCMK_ACTION_START, PCMK_ACTION_PROMOTE, NULL)) { value = "nothing (not start or promote)"; } else if (pcmk_is_set(rsc->flags, pcmk_rsc_needs_fencing)) { requires = pcmk_requires_fencing; value = "fencing"; } else if (pcmk_is_set(rsc->flags, pcmk_rsc_needs_quorum)) { requires = pcmk_requires_quorum; value = "quorum"; } else { value = "nothing"; } pcmk__rsc_trace(rsc, "%s of %s requires %s", action_name, rsc->id, value); return requires; } /*! * \internal * \brief Parse action failure response from a user-provided string * * \param[in] rsc Resource that action is for * \param[in] action_name Name of action * \param[in] interval_ms Action interval (in milliseconds) * \param[in] value User-provided configuration value for on-fail * * \return Action failure response parsed from \p text */ enum action_fail_response pcmk__parse_on_fail(const pcmk_resource_t *rsc, const char *action_name, guint interval_ms, const char *value) { const char *desc = NULL; bool needs_remote_reset = false; enum action_fail_response on_fail = pcmk_on_fail_ignore; if (value == NULL) { // Use default } else if (pcmk__str_eq(value, "block", pcmk__str_casei)) { on_fail = pcmk_on_fail_block; desc = "block"; } else if (pcmk__str_eq(value, "fence", pcmk__str_casei)) { if (pcmk_is_set(rsc->cluster->flags, pcmk_sched_fencing_enabled)) { on_fail = pcmk_on_fail_fence_node; desc = "node fencing"; } else { pcmk__config_err("Resetting '" PCMK_META_ON_FAIL "' for " "%s of %s to 'stop' because 'fence' is not " "valid when fencing is disabled", action_name, rsc->id); on_fail = pcmk_on_fail_stop; desc = "stop resource"; } } else if (pcmk__str_eq(value, "standby", pcmk__str_casei)) { on_fail = pcmk_on_fail_standby_node; desc = "node standby"; } else if (pcmk__strcase_any_of(value, "ignore", PCMK__VALUE_NOTHING, NULL)) { desc = "ignore"; } else if (pcmk__str_eq(value, "migrate", pcmk__str_casei)) { on_fail = pcmk_on_fail_ban; desc = "force migration"; } else if (pcmk__str_eq(value, "stop", pcmk__str_casei)) { on_fail = pcmk_on_fail_stop; desc = "stop resource"; } else if (pcmk__str_eq(value, "restart", pcmk__str_casei)) { on_fail = pcmk_on_fail_restart; desc = "restart (and possibly migrate)"; } else if (pcmk__str_eq(value, "restart-container", pcmk__str_casei)) { if (rsc->container == NULL) { pcmk__rsc_debug(rsc, "Using default " PCMK_META_ON_FAIL " for %s " "of %s because it does not have a container", action_name, rsc->id); } else { on_fail = pcmk_on_fail_restart_container; desc = "restart container (and possibly migrate)"; } } else if (pcmk__str_eq(value, "demote", pcmk__str_casei)) { on_fail = pcmk_on_fail_demote; desc = "demote instance"; } else { pcmk__config_err("Using default '" PCMK_META_ON_FAIL "' for " "%s of %s because '%s' is not valid", action_name, rsc->id, value); } /* Remote node connections are handled specially. Failures that result * in dropping an active connection must result in fencing. The only * failures that don't are probes and starts. The user can explicitly set * on-fail="fence" to fence after start failures. */ if (pe__resource_is_remote_conn(rsc) && !pcmk_is_probe(action_name, interval_ms) && !pcmk__str_eq(action_name, PCMK_ACTION_START, pcmk__str_none)) { needs_remote_reset = true; if (!pcmk_is_set(rsc->flags, pcmk_rsc_managed)) { desc = NULL; // Force default for unmanaged connections } } if (desc != NULL) { // Explicit value used, default not needed } else if (rsc->container != NULL) { on_fail = pcmk_on_fail_restart_container; desc = "restart container (and possibly migrate) (default)"; } else if (needs_remote_reset) { if (pcmk_is_set(rsc->flags, pcmk_rsc_managed)) { if (pcmk_is_set(rsc->cluster->flags, pcmk_sched_fencing_enabled)) { desc = "fence remote node (default)"; } else { desc = "recover remote node connection (default)"; } on_fail = pcmk_on_fail_reset_remote; } else { on_fail = pcmk_on_fail_stop; desc = "stop unmanaged remote node (enforcing default)"; } } else if (pcmk__str_eq(action_name, PCMK_ACTION_STOP, pcmk__str_none)) { if (pcmk_is_set(rsc->cluster->flags, pcmk_sched_fencing_enabled)) { on_fail = pcmk_on_fail_fence_node; desc = "resource fence (default)"; } else { on_fail = pcmk_on_fail_block; desc = "resource block (default)"; } } else { on_fail = pcmk_on_fail_restart; desc = "restart (and possibly migrate) (default)"; } pcmk__rsc_trace(rsc, "Failure handling for %s-interval %s of %s: %s", pcmk__readable_interval(interval_ms), action_name, rsc->id, desc); return on_fail; } /*! * \internal * \brief Determine a resource's role after failure of an action * * \param[in] rsc Resource that action is for * \param[in] action_name Action name * \param[in] on_fail Failure handling for action * \param[in] meta Unpacked action meta-attributes * * \return Resource role that results from failure of action */ enum rsc_role_e pcmk__role_after_failure(const pcmk_resource_t *rsc, const char *action_name, enum action_fail_response on_fail, GHashTable *meta) { const char *value = NULL; enum rsc_role_e role = pcmk_role_unknown; // Set default for role after failure specially in certain circumstances switch (on_fail) { case pcmk_on_fail_stop: role = pcmk_role_stopped; break; case pcmk_on_fail_reset_remote: if (rsc->remote_reconnect_ms != 0) { role = pcmk_role_stopped; } break; default: break; } // @COMPAT Check for explicitly configured role (deprecated) value = g_hash_table_lookup(meta, PCMK__META_ROLE_AFTER_FAILURE); if (value != NULL) { pcmk__warn_once(pcmk__wo_role_after, "Support for " PCMK__META_ROLE_AFTER_FAILURE " is " "deprecated and will be removed in a future release"); if (role == pcmk_role_unknown) { role = pcmk_parse_role(value); if (role == pcmk_role_unknown) { pcmk__config_err("Ignoring invalid value %s for " PCMK__META_ROLE_AFTER_FAILURE, value); } } } if (role == pcmk_role_unknown) { // Use default if (pcmk__str_eq(action_name, PCMK_ACTION_PROMOTE, pcmk__str_none)) { role = pcmk_role_unpromoted; } else { role = pcmk_role_started; } } pcmk__rsc_trace(rsc, "Role after %s %s failure is: %s", rsc->id, action_name, pcmk_role_text(role)); return role; } /*! * \internal * \brief Unpack action configuration * * Unpack a resource action's meta-attributes (normalizing the interval, * timeout, and start delay values as integer milliseconds), requirements, and * failure policy from its CIB XML configuration (including defaults). * * \param[in,out] action Resource action to unpack into * \param[in] xml_obj Action configuration XML (NULL for defaults only) * \param[in] interval_ms How frequently to perform the operation */ static void unpack_operation(pcmk_action_t *action, const xmlNode *xml_obj, guint interval_ms) { const char *value = NULL; action->meta = pcmk__unpack_action_meta(action->rsc, action->node, action->task, interval_ms, xml_obj); action->needs = pcmk__action_requires(action->rsc, action->task); value = g_hash_table_lookup(action->meta, PCMK_META_ON_FAIL); action->on_fail = pcmk__parse_on_fail(action->rsc, action->task, interval_ms, value); action->fail_role = pcmk__role_after_failure(action->rsc, action->task, action->on_fail, action->meta); } /*! * \brief Create or update an action object * * \param[in,out] rsc Resource that action is for (if any) * \param[in,out] key Action key (must be non-NULL) * \param[in] task Action name (must be non-NULL) * \param[in] on_node Node that action is on (if any) * \param[in] optional Whether action should be considered optional * \param[in,out] scheduler Scheduler data * * \return Action object corresponding to arguments (guaranteed not to be * \c NULL) * \note This function takes ownership of (and might free) \p key, and * \p scheduler takes ownership of the returned action (the caller should * not free it). */ pcmk_action_t * custom_action(pcmk_resource_t *rsc, char *key, const char *task, const pcmk_node_t *on_node, gboolean optional, pcmk_scheduler_t *scheduler) { pcmk_action_t *action = NULL; CRM_ASSERT((key != NULL) && (task != NULL) && (scheduler != NULL)); action = find_existing_action(key, rsc, on_node, scheduler); if (action == NULL) { action = new_action(key, task, rsc, on_node, optional, scheduler); } else { free(key); } update_action_optional(action, optional); if (rsc != NULL) { if ((action->node != NULL) && (action->op_entry != NULL) && !pcmk_is_set(action->flags, pcmk_action_attrs_evaluated)) { GHashTable *attrs = action->node->details->attrs; if (action->extra != NULL) { g_hash_table_destroy(action->extra); } action->extra = pcmk__unpack_action_rsc_params(action->op_entry, attrs, scheduler); pcmk__set_action_flags(action, pcmk_action_attrs_evaluated); } update_resource_action_runnable(action, scheduler); update_resource_flags_for_action(rsc, action); } if (action->extra == NULL) { action->extra = pcmk__strkey_table(free, free); } return action; } pcmk_action_t * get_pseudo_op(const char *name, pcmk_scheduler_t *scheduler) { pcmk_action_t *op = lookup_singleton(scheduler, name); if (op == NULL) { op = custom_action(NULL, strdup(name), name, NULL, TRUE, scheduler); pcmk__set_action_flags(op, pcmk_action_pseudo|pcmk_action_runnable); } return op; } static GList * find_unfencing_devices(GList *candidates, GList *matches) { for (GList *gIter = candidates; gIter != NULL; gIter = gIter->next) { pcmk_resource_t *candidate = gIter->data; if (candidate->children != NULL) { matches = find_unfencing_devices(candidate->children, matches); } else if (!pcmk_is_set(candidate->flags, pcmk_rsc_fence_device)) { continue; } else if (pcmk_is_set(candidate->flags, pcmk_rsc_needs_unfencing)) { matches = g_list_prepend(matches, candidate); } else if (pcmk__str_eq(g_hash_table_lookup(candidate->meta, PCMK_STONITH_PROVIDES), PCMK__VALUE_UNFENCING, pcmk__str_casei)) { matches = g_list_prepend(matches, candidate); } } return matches; } static int node_priority_fencing_delay(const pcmk_node_t *node, const pcmk_scheduler_t *scheduler) { int member_count = 0; int online_count = 0; int top_priority = 0; int lowest_priority = 0; GList *gIter = NULL; // PCMK_OPT_PRIORITY_FENCING_DELAY is disabled if (scheduler->priority_fencing_delay <= 0) { return 0; } /* No need to request a delay if the fencing target is not a normal cluster * member, for example if it's a remote node or a guest node. */ if (node->details->type != pcmk_node_variant_cluster) { return 0; } // No need to request a delay if the fencing target is in our partition if (node->details->online) { return 0; } for (gIter = scheduler->nodes; gIter != NULL; gIter = gIter->next) { pcmk_node_t *n = gIter->data; if (n->details->type != pcmk_node_variant_cluster) { continue; } member_count ++; if (n->details->online) { online_count++; } if (member_count == 1 || n->details->priority > top_priority) { top_priority = n->details->priority; } if (member_count == 1 || n->details->priority < lowest_priority) { lowest_priority = n->details->priority; } } // No need to delay if we have more than half of the cluster members if (online_count > member_count / 2) { return 0; } /* All the nodes have equal priority. * Any configured corresponding `pcmk_delay_base/max` will be applied. */ if (lowest_priority == top_priority) { return 0; } if (node->details->priority < top_priority) { return 0; } return scheduler->priority_fencing_delay; } pcmk_action_t * pe_fence_op(pcmk_node_t *node, const char *op, bool optional, const char *reason, bool priority_delay, pcmk_scheduler_t *scheduler) { char *op_key = NULL; pcmk_action_t *stonith_op = NULL; if(op == NULL) { op = scheduler->stonith_action; } op_key = crm_strdup_printf("%s-%s-%s", PCMK_ACTION_STONITH, node->details->uname, op); stonith_op = lookup_singleton(scheduler, op_key); if(stonith_op == NULL) { stonith_op = custom_action(NULL, op_key, PCMK_ACTION_STONITH, node, TRUE, scheduler); add_hash_param(stonith_op->meta, PCMK__META_ON_NODE, node->details->uname); add_hash_param(stonith_op->meta, PCMK__META_ON_NODE_UUID, node->details->id); add_hash_param(stonith_op->meta, "stonith_action", op); if (pcmk_is_set(scheduler->flags, pcmk_sched_enable_unfencing)) { /* Extra work to detect device changes */ GString *digests_all = g_string_sized_new(1024); GString *digests_secure = g_string_sized_new(1024); GList *matches = find_unfencing_devices(scheduler->resources, NULL); char *key = NULL; char *value = NULL; for (GList *gIter = matches; gIter != NULL; gIter = gIter->next) { pcmk_resource_t *match = gIter->data; const char *agent = g_hash_table_lookup(match->meta, PCMK_XA_TYPE); pcmk__op_digest_t *data = NULL; data = pe__compare_fencing_digest(match, agent, node, scheduler); if (data->rc == pcmk__digest_mismatch) { optional = FALSE; crm_notice("Unfencing node %s because the definition of " "%s changed", pcmk__node_name(node), match->id); if (!pcmk__is_daemon && scheduler->priv != NULL) { pcmk__output_t *out = scheduler->priv; out->info(out, "notice: Unfencing node %s because the " "definition of %s changed", pcmk__node_name(node), match->id); } } pcmk__g_strcat(digests_all, match->id, ":", agent, ":", data->digest_all_calc, ",", NULL); pcmk__g_strcat(digests_secure, match->id, ":", agent, ":", data->digest_secure_calc, ",", NULL); } key = strdup(PCMK__META_DIGESTS_ALL); value = strdup((const char *) digests_all->str); CRM_ASSERT((key != NULL) && (value != NULL)); g_hash_table_insert(stonith_op->meta, key, value); g_string_free(digests_all, TRUE); key = strdup(PCMK__META_DIGESTS_SECURE); value = strdup((const char *) digests_secure->str); CRM_ASSERT((key != NULL) && (value != NULL)); g_hash_table_insert(stonith_op->meta, key, value); g_string_free(digests_secure, TRUE); } } else { free(op_key); } if (scheduler->priority_fencing_delay > 0 /* It's a suitable case where PCMK_OPT_PRIORITY_FENCING_DELAY * applies. At least add PCMK_OPT_PRIORITY_FENCING_DELAY field as * an indicator. */ && (priority_delay /* The priority delay needs to be recalculated if this function has * been called by schedule_fencing_and_shutdowns() after node * priority has already been calculated by native_add_running(). */ || g_hash_table_lookup(stonith_op->meta, PCMK_OPT_PRIORITY_FENCING_DELAY) != NULL)) { /* Add PCMK_OPT_PRIORITY_FENCING_DELAY to the fencing op even if * it's 0 for the targeting node. So that it takes precedence over * any possible `pcmk_delay_base/max`. */ char *delay_s = pcmk__itoa(node_priority_fencing_delay(node, scheduler)); g_hash_table_insert(stonith_op->meta, strdup(PCMK_OPT_PRIORITY_FENCING_DELAY), delay_s); } if(optional == FALSE && pe_can_fence(scheduler, node)) { pcmk__clear_action_flags(stonith_op, pcmk_action_optional); pe_action_set_reason(stonith_op, reason, false); } else if(reason && stonith_op->reason == NULL) { stonith_op->reason = strdup(reason); } return stonith_op; } void pe_free_action(pcmk_action_t *action) { if (action == NULL) { return; } g_list_free_full(action->actions_before, free); g_list_free_full(action->actions_after, free); if (action->extra) { g_hash_table_destroy(action->extra); } if (action->meta) { g_hash_table_destroy(action->meta); } free(action->cancel_task); free(action->reason); free(action->task); free(action->uuid); free(action->node); free(action); } int pe_get_configured_timeout(pcmk_resource_t *rsc, const char *action, pcmk_scheduler_t *scheduler) { xmlNode *child = NULL; GHashTable *action_meta = NULL; const char *timeout_spec = NULL; - int timeout_ms = 0; + long long timeout_ms = 0; pe_rule_eval_data_t rule_data = { .node_hash = NULL, .role = pcmk_role_unknown, .now = scheduler->now, .match_data = NULL, .rsc_data = NULL, .op_data = NULL }; for (child = first_named_child(rsc->ops_xml, PCMK_XE_OP); child != NULL; child = crm_next_same_xml(child)) { if (pcmk__str_eq(action, crm_element_value(child, PCMK_XA_NAME), pcmk__str_casei)) { timeout_spec = crm_element_value(child, PCMK_META_TIMEOUT); break; } } if (timeout_spec == NULL && scheduler->op_defaults) { action_meta = pcmk__strkey_table(free, free); pe__unpack_dataset_nvpairs(scheduler->op_defaults, PCMK_XE_META_ATTRIBUTES, &rule_data, action_meta, NULL, FALSE, scheduler); timeout_spec = g_hash_table_lookup(action_meta, PCMK_META_TIMEOUT); } // @TODO check meta-attributes // @TODO maybe use min-interval monitor timeout as default for monitors timeout_ms = crm_get_msec(timeout_spec); if (timeout_ms < 0) { timeout_ms = PCMK_DEFAULT_ACTION_TIMEOUT_MS; } if (action_meta != NULL) { g_hash_table_destroy(action_meta); } - return timeout_ms; + return (int) QB_MIN(timeout_ms, INT_MAX); } enum action_tasks get_complex_task(const pcmk_resource_t *rsc, const char *name) { enum action_tasks task = text2task(name); if ((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_primitive)) { switch (task) { case pcmk_action_stopped: case pcmk_action_started: case pcmk_action_demoted: case pcmk_action_promoted: crm_trace("Folding %s back into its atomic counterpart for %s", name, rsc->id); --task; break; default: break; } } return task; } /*! * \internal * \brief Find first matching action in a list * * \param[in] input List of actions to search * \param[in] uuid If not NULL, action must have this UUID * \param[in] task If not NULL, action must have this action name * \param[in] on_node If not NULL, action must be on this node * * \return First action in list that matches criteria, or NULL if none */ pcmk_action_t * find_first_action(const GList *input, const char *uuid, const char *task, const pcmk_node_t *on_node) { CRM_CHECK(uuid || task, return NULL); for (const GList *gIter = input; gIter != NULL; gIter = gIter->next) { pcmk_action_t *action = (pcmk_action_t *) gIter->data; if (uuid != NULL && !pcmk__str_eq(uuid, action->uuid, pcmk__str_casei)) { continue; } else if (task != NULL && !pcmk__str_eq(task, action->task, pcmk__str_casei)) { continue; } else if (on_node == NULL) { return action; } else if (action->node == NULL) { continue; } else if (pcmk__same_node(on_node, action->node)) { return action; } } return NULL; } GList * find_actions(GList *input, const char *key, const pcmk_node_t *on_node) { GList *gIter = input; GList *result = NULL; CRM_CHECK(key != NULL, return NULL); for (; gIter != NULL; gIter = gIter->next) { pcmk_action_t *action = (pcmk_action_t *) gIter->data; if (!pcmk__str_eq(key, action->uuid, pcmk__str_casei)) { continue; } else if (on_node == NULL) { crm_trace("Action %s matches (ignoring node)", key); result = g_list_prepend(result, action); } else if (action->node == NULL) { crm_trace("Action %s matches (unallocated, assigning to %s)", key, pcmk__node_name(on_node)); action->node = pe__copy_node(on_node); result = g_list_prepend(result, action); } else if (pcmk__same_node(on_node, action->node)) { crm_trace("Action %s on %s matches", key, pcmk__node_name(on_node)); result = g_list_prepend(result, action); } } return result; } GList * find_actions_exact(GList *input, const char *key, const pcmk_node_t *on_node) { GList *result = NULL; CRM_CHECK(key != NULL, return NULL); if (on_node == NULL) { return NULL; } for (GList *gIter = input; gIter != NULL; gIter = gIter->next) { pcmk_action_t *action = (pcmk_action_t *) gIter->data; if ((action->node != NULL) && pcmk__str_eq(key, action->uuid, pcmk__str_casei) && pcmk__str_eq(on_node->details->id, action->node->details->id, pcmk__str_casei)) { crm_trace("Action %s on %s matches", key, pcmk__node_name(on_node)); result = g_list_prepend(result, action); } } return result; } /*! * \brief Find all actions of given type for a resource * * \param[in] rsc Resource to search * \param[in] node Find only actions scheduled on this node * \param[in] task Action name to search for * \param[in] require_node If TRUE, NULL node or action node will not match * * \return List of actions found (or NULL if none) * \note If node is not NULL and require_node is FALSE, matching actions * without a node will be assigned to node. */ GList * pe__resource_actions(const pcmk_resource_t *rsc, const pcmk_node_t *node, const char *task, bool require_node) { GList *result = NULL; char *key = pcmk__op_key(rsc->id, task, 0); if (require_node) { result = find_actions_exact(rsc->actions, key, node); } else { result = find_actions(rsc->actions, key, node); } free(key); return result; } /*! * \internal * \brief Create an action reason string based on the action itself * * \param[in] action Action to create reason string for * \param[in] flag Action flag that was cleared * * \return Newly allocated string suitable for use as action reason * \note It is the caller's responsibility to free() the result. */ char * pe__action2reason(const pcmk_action_t *action, enum pe_action_flags flag) { const char *change = NULL; switch (flag) { case pcmk_action_runnable: change = "unrunnable"; break; case pcmk_action_migratable: change = "unmigrateable"; break; case pcmk_action_optional: change = "required"; break; default: // Bug: caller passed unsupported flag CRM_CHECK(change != NULL, change = ""); break; } return crm_strdup_printf("%s%s%s %s", change, (action->rsc == NULL)? "" : " ", (action->rsc == NULL)? "" : action->rsc->id, action->task); } void pe_action_set_reason(pcmk_action_t *action, const char *reason, bool overwrite) { if (action->reason != NULL && overwrite) { pcmk__rsc_trace(action->rsc, "Changing %s reason from '%s' to '%s'", action->uuid, action->reason, pcmk__s(reason, "(none)")); } else if (action->reason == NULL) { pcmk__rsc_trace(action->rsc, "Set %s reason to '%s'", action->uuid, pcmk__s(reason, "(none)")); } else { // crm_assert(action->reason != NULL && !overwrite); return; } pcmk__str_update(&action->reason, reason); } /*! * \internal * \brief Create an action to clear a resource's history from CIB * * \param[in,out] rsc Resource to clear * \param[in] node Node to clear history on */ void pe__clear_resource_history(pcmk_resource_t *rsc, const pcmk_node_t *node) { CRM_ASSERT((rsc != NULL) && (node != NULL)); custom_action(rsc, pcmk__op_key(rsc->id, PCMK_ACTION_LRM_DELETE, 0), PCMK_ACTION_LRM_DELETE, node, FALSE, rsc->cluster); } #define sort_return(an_int, why) do { \ free(a_uuid); \ free(b_uuid); \ crm_trace("%s (%d) %c %s (%d) : %s", \ a_xml_id, a_call_id, an_int>0?'>':an_int<0?'<':'=', \ b_xml_id, b_call_id, why); \ return an_int; \ } while(0) int pe__is_newer_op(const xmlNode *xml_a, const xmlNode *xml_b, bool same_node_default) { int a_call_id = -1; int b_call_id = -1; char *a_uuid = NULL; char *b_uuid = NULL; const char *a_xml_id = crm_element_value(xml_a, PCMK_XA_ID); const char *b_xml_id = crm_element_value(xml_b, PCMK_XA_ID); const char *a_node = crm_element_value(xml_a, PCMK__META_ON_NODE); const char *b_node = crm_element_value(xml_b, PCMK__META_ON_NODE); bool same_node = true; /* @COMPAT The on_node attribute was added to last_failure as of 1.1.13 (via * 8b3ca1c) and the other entries as of 1.1.12 (via 0b07b5c). * * In case that any of the lrm_rsc_op entries doesn't have on_node * attribute, we need to explicitly tell whether the two operations are on * the same node. */ if (a_node == NULL || b_node == NULL) { same_node = same_node_default; } else { same_node = pcmk__str_eq(a_node, b_node, pcmk__str_casei); } if (same_node && pcmk__str_eq(a_xml_id, b_xml_id, pcmk__str_none)) { /* We have duplicate lrm_rsc_op entries in the status * section which is unlikely to be a good thing * - we can handle it easily enough, but we need to get * to the bottom of why it's happening. */ pcmk__config_err("Duplicate lrm_rsc_op entries named %s", a_xml_id); sort_return(0, "duplicate"); } crm_element_value_int(xml_a, PCMK__XA_CALL_ID, &a_call_id); crm_element_value_int(xml_b, PCMK__XA_CALL_ID, &b_call_id); if (a_call_id == -1 && b_call_id == -1) { /* both are pending ops so it doesn't matter since * stops are never pending */ sort_return(0, "pending"); } else if (same_node && a_call_id >= 0 && a_call_id < b_call_id) { sort_return(-1, "call id"); } else if (same_node && b_call_id >= 0 && a_call_id > b_call_id) { sort_return(1, "call id"); } else if (a_call_id >= 0 && b_call_id >= 0 && (!same_node || a_call_id == b_call_id)) { /* The op and last_failed_op are the same. Order on * PCMK_XA_LAST_RC_CHANGE. */ time_t last_a = -1; time_t last_b = -1; crm_element_value_epoch(xml_a, PCMK_XA_LAST_RC_CHANGE, &last_a); crm_element_value_epoch(xml_b, PCMK_XA_LAST_RC_CHANGE, &last_b); crm_trace("rc-change: %lld vs %lld", (long long) last_a, (long long) last_b); if (last_a >= 0 && last_a < last_b) { sort_return(-1, "rc-change"); } else if (last_b >= 0 && last_a > last_b) { sort_return(1, "rc-change"); } sort_return(0, "rc-change"); } else { /* One of the inputs is a pending operation. * Attempt to use PCMK__XA_TRANSITION_MAGIC to determine its age relative * to the other. */ int a_id = -1; int b_id = -1; const char *a_magic = crm_element_value(xml_a, PCMK__XA_TRANSITION_MAGIC); const char *b_magic = crm_element_value(xml_b, PCMK__XA_TRANSITION_MAGIC); CRM_CHECK(a_magic != NULL && b_magic != NULL, sort_return(0, "No magic")); if (!decode_transition_magic(a_magic, &a_uuid, &a_id, NULL, NULL, NULL, NULL)) { sort_return(0, "bad magic a"); } if (!decode_transition_magic(b_magic, &b_uuid, &b_id, NULL, NULL, NULL, NULL)) { sort_return(0, "bad magic b"); } /* try to determine the relative age of the operation... * some pending operations (e.g. a start) may have been superseded * by a subsequent stop * * [a|b]_id == -1 means it's a shutdown operation and _always_ comes last */ if (!pcmk__str_eq(a_uuid, b_uuid, pcmk__str_casei) || a_id == b_id) { /* * some of the logic in here may be redundant... * * if the UUID from the TE doesn't match then one better * be a pending operation. * pending operations don't survive between elections and joins * because we query the LRM directly */ if (b_call_id == -1) { sort_return(-1, "transition + call"); } else if (a_call_id == -1) { sort_return(1, "transition + call"); } } else if ((a_id >= 0 && a_id < b_id) || b_id == -1) { sort_return(-1, "transition"); } else if ((b_id >= 0 && a_id > b_id) || a_id == -1) { sort_return(1, "transition"); } } /* we should never end up here */ CRM_CHECK(FALSE, sort_return(0, "default")); } gint sort_op_by_callid(gconstpointer a, gconstpointer b) { const xmlNode *xml_a = a; const xmlNode *xml_b = b; return pe__is_newer_op(xml_a, xml_b, true); } /*! * \internal * \brief Create a new pseudo-action for a resource * * \param[in,out] rsc Resource to create action for * \param[in] task Action name * \param[in] optional Whether action should be considered optional * \param[in] runnable Whethe action should be considered runnable * * \return New action object corresponding to arguments */ pcmk_action_t * pe__new_rsc_pseudo_action(pcmk_resource_t *rsc, const char *task, bool optional, bool runnable) { pcmk_action_t *action = NULL; CRM_ASSERT((rsc != NULL) && (task != NULL)); action = custom_action(rsc, pcmk__op_key(rsc->id, task, 0), task, NULL, optional, rsc->cluster); pcmk__set_action_flags(action, pcmk_action_pseudo); if (runnable) { pcmk__set_action_flags(action, pcmk_action_runnable); } return action; } /*! * \internal * \brief Add the expected result to an action * * \param[in,out] action Action to add expected result to * \param[in] expected_result Expected result to add * * \note This is more efficient than calling add_hash_param(). */ void pe__add_action_expected_result(pcmk_action_t *action, int expected_result) { char *name = NULL; CRM_ASSERT((action != NULL) && (action->meta != NULL)); name = strdup(PCMK__META_OP_TARGET_RC); CRM_ASSERT (name != NULL); g_hash_table_insert(action->meta, name, pcmk__itoa(expected_result)); } diff --git a/lib/pengine/rules_alerts.c b/lib/pengine/rules_alerts.c index 537d92a05d..a5e97de84e 100644 --- a/lib/pengine/rules_alerts.c +++ b/lib/pengine/rules_alerts.c @@ -1,301 +1,303 @@ /* * Copyright 2015-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include /*! * \internal * \brief Unpack an alert's or alert recipient's meta attributes * * \param[in,out] basenode Alert or recipient XML * \param[in,out] entry Where to store unpacked values * \param[in,out] max_timeout Max timeout of all alerts and recipients thus far * * \return Standard Pacemaker return code */ static int get_meta_attrs_from_cib(xmlNode *basenode, pcmk__alert_t *entry, guint *max_timeout) { GHashTable *config_hash = pcmk__strkey_table(free, free); crm_time_t *now = crm_time_new(NULL); const char *value = NULL; int rc = pcmk_rc_ok; pe_unpack_nvpairs(basenode, basenode, PCMK_XE_META_ATTRIBUTES, NULL, config_hash, NULL, FALSE, now, NULL); crm_time_free(now); value = g_hash_table_lookup(config_hash, PCMK_META_ENABLED); if ((value != NULL) && !crm_is_true(value)) { // No need to continue unpacking rc = pcmk_rc_disabled; goto done; } value = g_hash_table_lookup(config_hash, PCMK_META_TIMEOUT); if (value) { - entry->timeout = crm_get_msec(value); + long long timeout_ms = crm_get_msec(value); + + entry->timeout = (int) QB_MIN(timeout_ms, INT_MAX); if (entry->timeout <= 0) { if (entry->timeout == 0) { crm_trace("Alert %s uses default timeout of %dmsec", entry->id, PCMK__ALERT_DEFAULT_TIMEOUT_MS); } else { pcmk__config_warn("Alert %s has invalid timeout value '%s', " "using default (%d ms)", entry->id, value, PCMK__ALERT_DEFAULT_TIMEOUT_MS); } entry->timeout = PCMK__ALERT_DEFAULT_TIMEOUT_MS; } else { crm_trace("Alert %s uses timeout of %dmsec", entry->id, entry->timeout); } if (entry->timeout > *max_timeout) { *max_timeout = entry->timeout; } } value = g_hash_table_lookup(config_hash, PCMK_META_TIMESTAMP_FORMAT); if (value) { /* hard to do any checks here as merely anything can * can be a valid time-format-string */ entry->tstamp_format = strdup(value); crm_trace("Alert %s uses timestamp format '%s'", entry->id, entry->tstamp_format); } done: g_hash_table_destroy(config_hash); return rc; } static void get_envvars_from_cib(xmlNode *basenode, pcmk__alert_t *entry) { xmlNode *child; if ((basenode == NULL) || (entry == NULL)) { return; } child = first_named_child(basenode, PCMK_XE_INSTANCE_ATTRIBUTES); if (child == NULL) { return; } if (entry->envvars == NULL) { entry->envvars = pcmk__strkey_table(free, free); } for (child = first_named_child(child, PCMK_XE_NVPAIR); child != NULL; child = crm_next_same_xml(child)) { const char *name = crm_element_value(child, PCMK_XA_NAME); const char *value = crm_element_value(child, PCMK_XA_VALUE); if (value == NULL) { value = ""; } g_hash_table_insert(entry->envvars, strdup(name), strdup(value)); crm_trace("Alert %s: added environment variable %s='%s'", entry->id, name, value); } } static void unpack_alert_filter(xmlNode *basenode, pcmk__alert_t *entry) { xmlNode *select = first_named_child(basenode, PCMK_XE_SELECT); xmlNode *event_type = NULL; uint32_t flags = pcmk__alert_none; for (event_type = pcmk__xe_first_child(select); event_type != NULL; event_type = pcmk__xe_next(event_type)) { if (pcmk__xe_is(event_type, PCMK_XE_SELECT_FENCING)) { flags |= pcmk__alert_fencing; } else if (pcmk__xe_is(event_type, PCMK_XE_SELECT_NODES)) { flags |= pcmk__alert_node; } else if (pcmk__xe_is(event_type, PCMK_XE_SELECT_RESOURCES)) { flags |= pcmk__alert_resource; } else if (pcmk__xe_is(event_type, PCMK_XE_SELECT_ATTRIBUTES)) { xmlNode *attr; const char *attr_name; int nattrs = 0; flags |= pcmk__alert_attribute; for (attr = first_named_child(event_type, PCMK_XE_ATTRIBUTE); attr != NULL; attr = crm_next_same_xml(attr)) { attr_name = crm_element_value(attr, PCMK_XA_NAME); if (attr_name) { if (nattrs == 0) { g_strfreev(entry->select_attribute_name); entry->select_attribute_name = NULL; } ++nattrs; entry->select_attribute_name = pcmk__realloc(entry->select_attribute_name, (nattrs + 1) * sizeof(char*)); entry->select_attribute_name[nattrs - 1] = strdup(attr_name); entry->select_attribute_name[nattrs] = NULL; } } } } if (flags != pcmk__alert_none) { entry->flags = flags; crm_debug("Alert %s receives events: attributes:%s%s%s%s", entry->id, (pcmk_is_set(flags, pcmk__alert_attribute)? (entry->select_attribute_name? "some" : "all") : "none"), (pcmk_is_set(flags, pcmk__alert_fencing)? " fencing" : ""), (pcmk_is_set(flags, pcmk__alert_node)? " nodes" : ""), (pcmk_is_set(flags, pcmk__alert_resource)? " resources" : "")); } } /*! * \internal * \brief Unpack an alert or an alert recipient * * \param[in,out] alert Alert or recipient XML * \param[in,out] entry Where to store unpacked values * \param[in,out] max_timeout Max timeout of all alerts and recipients thus far * * \return Standard Pacemaker return code */ static int unpack_alert(xmlNode *alert, pcmk__alert_t *entry, guint *max_timeout) { int rc = pcmk_rc_ok; get_envvars_from_cib(alert, entry); rc = get_meta_attrs_from_cib(alert, entry, max_timeout); if (rc == pcmk_rc_ok) { unpack_alert_filter(alert, entry); } return rc; } /*! * \internal * \brief Unpack a CIB alerts section * * \param[in] alerts XML of alerts section * * \return List of unpacked alert entries * * \note Unlike most unpack functions, this is not used by the scheduler itself, * but is supplied for use by daemons that need to send alerts. */ GList * pe_unpack_alerts(const xmlNode *alerts) { xmlNode *alert; pcmk__alert_t *entry; guint max_timeout = 0; GList *alert_list = NULL; if (alerts == NULL) { return alert_list; } for (alert = first_named_child(alerts, PCMK_XE_ALERT); alert != NULL; alert = crm_next_same_xml(alert)) { xmlNode *recipient; int recipients = 0; const char *alert_id = ID(alert); const char *alert_path = crm_element_value(alert, PCMK_XA_PATH); /* The schema should enforce this, but to be safe ... */ if (alert_id == NULL) { pcmk__config_warn("Ignoring invalid alert without " PCMK_XA_ID); crm_log_xml_info(alert, "missing-id"); continue; } if (alert_path == NULL) { pcmk__config_warn("Ignoring alert %s: No " PCMK_XA_PATH, alert_id); continue; } entry = pcmk__alert_new(alert_id, alert_path); if (unpack_alert(alert, entry, &max_timeout) != pcmk_rc_ok) { // Don't allow recipients to override if entire alert is disabled crm_debug("Alert %s is disabled", entry->id); pcmk__free_alert(entry); continue; } if (entry->tstamp_format == NULL) { entry->tstamp_format = strdup(PCMK__ALERT_DEFAULT_TSTAMP_FORMAT); } crm_debug("Alert %s: path=%s timeout=%dms tstamp-format='%s' %u vars", entry->id, entry->path, entry->timeout, entry->tstamp_format, (entry->envvars? g_hash_table_size(entry->envvars) : 0)); for (recipient = first_named_child(alert, PCMK_XE_RECIPIENT); recipient != NULL; recipient = crm_next_same_xml(recipient)) { pcmk__alert_t *recipient_entry = pcmk__dup_alert(entry); recipients++; recipient_entry->recipient = crm_element_value_copy(recipient, PCMK_XA_VALUE); if (unpack_alert(recipient, recipient_entry, &max_timeout) != pcmk_rc_ok) { crm_debug("Alert %s: recipient %s is disabled", entry->id, recipient_entry->id); pcmk__free_alert(recipient_entry); continue; } alert_list = g_list_prepend(alert_list, recipient_entry); crm_debug("Alert %s has recipient %s with value %s and %d envvars", entry->id, ID(recipient), recipient_entry->recipient, (recipient_entry->envvars? g_hash_table_size(recipient_entry->envvars) : 0)); } if (recipients == 0) { alert_list = g_list_prepend(alert_list, entry); } else { pcmk__free_alert(entry); } } return alert_list; } /*! * \internal * \brief Free an alert list generated by pe_unpack_alerts() * * \param[in,out] alert_list Alert list to free */ void pe_free_alert_list(GList *alert_list) { if (alert_list) { g_list_free_full(alert_list, (GDestroyNotify) pcmk__free_alert); } } diff --git a/tools/crm_resource.c b/tools/crm_resource.c index 7198ec248c..fb92fa3eff 100644 --- a/tools/crm_resource.c +++ b/tools/crm_resource.c @@ -1,2088 +1,2095 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #define SUMMARY "crm_resource - perform tasks related to Pacemaker cluster resources" enum rsc_command { cmd_none = 0, // No command option given (yet) cmd_ban, cmd_cleanup, cmd_clear, cmd_colocations, cmd_cts, cmd_delete, cmd_delete_param, cmd_digests, cmd_execute_agent, cmd_fail, cmd_get_param, cmd_get_property, cmd_list_active_ops, cmd_list_agents, cmd_list_all_ops, cmd_list_alternatives, cmd_list_instances, cmd_list_providers, cmd_list_resources, cmd_list_standards, cmd_locate, cmd_metadata, cmd_move, cmd_query_raw_xml, cmd_query_xml, cmd_refresh, cmd_restart, cmd_set_param, cmd_set_property, cmd_wait, cmd_why, }; struct { enum rsc_command rsc_cmd; // crm_resource command to perform // Infrastructure that given command needs to work gboolean require_cib; // Whether command requires CIB IPC int cib_options; // Options to use with CIB IPC calls gboolean require_crmd; // Whether command requires controller IPC gboolean require_scheduler; // Whether command requires scheduler data gboolean require_resource; // Whether command requires resource specified gboolean require_node; // Whether command requires node specified int find_flags; // Flags to use when searching for resource // Command-line option values gchar *rsc_id; // Value of --resource gchar *rsc_type; // Value of --resource-type gboolean force; // --force was given gboolean clear_expired; // --expired was given gboolean recursive; // --recursive was given gboolean promoted_role_only; // --promoted was given gchar *host_uname; // Value of --node gchar *interval_spec; // Value of --interval gchar *move_lifetime; // Value of --lifetime gchar *operation; // Value of --operation const char *attr_set_type; // Instance, meta, utilization, or element attribute gchar *prop_id; // --nvpair (attribute XML ID) char *prop_name; // Attribute name gchar *prop_set; // --set-name (attribute block XML ID) gchar *prop_value; // --parameter-value (attribute value) - int timeout_ms; // Parsed from --timeout value + long long timeout_ms; // Parsed from --timeout value char *agent_spec; // Standard and/or provider and/or agent gchar *xml_file; // Value of (deprecated) --xml-file int check_level; // Optional value of --validate or --force-check // Resource configuration specified via command-line arguments gboolean cmdline_config; // Resource configuration was via arguments char *v_agent; // Value of --agent char *v_class; // Value of --class char *v_provider; // Value of --provider GHashTable *cmdline_params; // Resource parameters specified // Positional command-line arguments gchar **remainder; // Positional arguments as given GHashTable *override_params; // Resource parameter values that override config } options = { .attr_set_type = PCMK_XE_INSTANCE_ATTRIBUTES, .check_level = -1, .cib_options = cib_sync_call, .require_cib = TRUE, .require_scheduler = TRUE, .require_resource = TRUE, }; #if 0 // @COMPAT @TODO enable this at next backward compatibility break #define SET_COMMAND(cmd) do { \ if (options.rsc_cmd != cmd_none) { \ g_set_error(error, PCMK__EXITC_ERROR, CRM_EX_USAGE, \ "Only one command option may be specified"); \ return FALSE; \ } \ options.rsc_cmd = (cmd); \ } while (0) #else #define SET_COMMAND(cmd) do { \ if (options.rsc_cmd != cmd_none) { \ reset_options(); \ } \ options.rsc_cmd = (cmd); \ } while (0) #endif gboolean agent_provider_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error); gboolean attr_set_type_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error); gboolean class_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error); gboolean cleanup_refresh_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error); gboolean delete_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error); gboolean expired_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error); gboolean list_agents_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error); gboolean list_providers_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error); gboolean list_standards_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error); gboolean list_alternatives_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error); gboolean metadata_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error); gboolean option_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error); gboolean fail_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error); gboolean flag_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error); gboolean get_param_prop_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error); gboolean list_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error); gboolean set_delete_param_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error); gboolean set_prop_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error); gboolean timeout_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error); gboolean validate_or_force_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error); gboolean restart_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error); gboolean digests_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error); gboolean wait_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error); gboolean why_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error); static crm_exit_t exit_code = CRM_EX_OK; static pcmk__output_t *out = NULL; static pcmk__common_args_t *args = NULL; // Things that should be cleaned up on exit static GError *error = NULL; static GMainLoop *mainloop = NULL; static cib_t *cib_conn = NULL; static pcmk_ipc_api_t *controld_api = NULL; static pcmk_scheduler_t *scheduler = NULL; #define MESSAGE_TIMEOUT_S 60 #define INDENT " " static pcmk__supported_format_t formats[] = { PCMK__SUPPORTED_FORMAT_NONE, PCMK__SUPPORTED_FORMAT_TEXT, PCMK__SUPPORTED_FORMAT_XML, { NULL, NULL, NULL } }; // Clean up and exit static crm_exit_t bye(crm_exit_t ec) { pcmk__output_and_clear_error(&error, out); if (out != NULL) { out->finish(out, ec, true, NULL); pcmk__output_free(out); } pcmk__unregister_formats(); if (cib_conn != NULL) { cib_t *save_cib_conn = cib_conn; cib_conn = NULL; // Ensure we can't free this twice cib__clean_up_connection(&save_cib_conn); } if (controld_api != NULL) { pcmk_ipc_api_t *save_controld_api = controld_api; controld_api = NULL; // Ensure we can't free this twice pcmk_free_ipc_api(save_controld_api); } if (mainloop != NULL) { g_main_loop_unref(mainloop); mainloop = NULL; } pe_free_working_set(scheduler); scheduler = NULL; crm_exit(ec); return ec; } static void quit_main_loop(crm_exit_t ec) { exit_code = ec; if (mainloop != NULL) { GMainLoop *mloop = mainloop; mainloop = NULL; // Don't re-enter this block pcmk_quit_main_loop(mloop, 10); g_main_loop_unref(mloop); } } static gboolean resource_ipc_timeout(gpointer data) { // Start with newline because "Waiting for ..." message doesn't have one if (error != NULL) { g_clear_error(&error); } g_set_error(&error, PCMK__EXITC_ERROR, CRM_EX_TIMEOUT, _("Aborting because no messages received in %d seconds"), MESSAGE_TIMEOUT_S); quit_main_loop(CRM_EX_TIMEOUT); return FALSE; } static void controller_event_callback(pcmk_ipc_api_t *api, enum pcmk_ipc_event event_type, crm_exit_t status, void *event_data, void *user_data) { switch (event_type) { case pcmk_ipc_event_disconnect: if (exit_code == CRM_EX_DISCONNECT) { // Unexpected crm_info("Connection to controller was terminated"); } quit_main_loop(exit_code); break; case pcmk_ipc_event_reply: if (status != CRM_EX_OK) { out->err(out, "Error: bad reply from controller: %s", crm_exit_str(status)); pcmk_disconnect_ipc(api); quit_main_loop(status); } else { if ((pcmk_controld_api_replies_expected(api) == 0) && mainloop && g_main_loop_is_running(mainloop)) { out->info(out, "... got reply (done)"); crm_debug("Got all the replies we expected"); pcmk_disconnect_ipc(api); quit_main_loop(CRM_EX_OK); } else { out->info(out, "... got reply"); } } break; default: break; } } static void start_mainloop(pcmk_ipc_api_t *capi) { unsigned int count = pcmk_controld_api_replies_expected(capi); if (count > 0) { out->info(out, "Waiting for %u %s from the controller", count, pcmk__plural_alt(count, "reply", "replies")); exit_code = CRM_EX_DISCONNECT; // For unexpected disconnects mainloop = g_main_loop_new(NULL, FALSE); g_timeout_add(MESSAGE_TIMEOUT_S * 1000, resource_ipc_timeout, NULL); g_main_loop_run(mainloop); } } static int compare_id(gconstpointer a, gconstpointer b) { return strcmp((const char *)a, (const char *)b); } static GList * build_constraint_list(xmlNode *root) { GList *retval = NULL; xmlNode *cib_constraints = NULL; xmlXPathObjectPtr xpathObj = NULL; int ndx = 0; cib_constraints = pcmk_find_cib_element(root, PCMK_XE_CONSTRAINTS); xpathObj = xpath_search(cib_constraints, "//" XML_CONS_TAG_RSC_LOCATION); for (ndx = 0; ndx < numXpathResults(xpathObj); ndx++) { xmlNode *match = getXpathResult(xpathObj, ndx); retval = g_list_insert_sorted(retval, (gpointer) ID(match), compare_id); } freeXpathObject(xpathObj); return retval; } /* short option letters still available: eEJkKXyYZ */ static GOptionEntry query_entries[] = { { "list", 'L', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, list_cb, "List all cluster resources with status", NULL }, { "list-raw", 'l', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, list_cb, "List IDs of all instantiated resources (individual members\n" INDENT "rather than groups etc.)", NULL }, { "list-cts", 'c', G_OPTION_FLAG_HIDDEN|G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, list_cb, NULL, NULL }, { "list-operations", 'O', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, list_cb, "List active resource operations, optionally filtered by\n" INDENT "--resource and/or --node", NULL }, { "list-all-operations", 'o', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, list_cb, "List all resource operations, optionally filtered by\n" INDENT "--resource and/or --node", NULL }, { "list-standards", 0, G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, list_standards_cb, "List supported standards", NULL }, { "list-ocf-providers", 0, G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, list_providers_cb, "List all available OCF providers", NULL }, { "list-agents", 0, G_OPTION_FLAG_NONE, G_OPTION_ARG_CALLBACK, list_agents_cb, "List all agents available for the named standard and/or provider", "STD:PROV" }, { "list-ocf-alternatives", 0, G_OPTION_FLAG_NONE, G_OPTION_ARG_CALLBACK, list_alternatives_cb, "List all available providers for the named OCF agent", "AGENT" }, { "show-metadata", 0, G_OPTION_FLAG_NONE, G_OPTION_ARG_CALLBACK, metadata_cb, "Show the metadata for the named class:provider:agent", "SPEC" }, { "query-xml", 'q', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, flag_cb, "Show XML configuration of resource (after any template expansion)", NULL }, { "query-xml-raw", 'w', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, flag_cb, "Show XML configuration of resource (before any template expansion)", NULL }, { "get-parameter", 'g', G_OPTION_FLAG_NONE, G_OPTION_ARG_CALLBACK, get_param_prop_cb, "Display named parameter for resource (use instance attribute\n" INDENT "unless --element, --meta, or --utilization is specified)", "PARAM" }, { "get-property", 'G', G_OPTION_FLAG_HIDDEN, G_OPTION_ARG_CALLBACK, get_param_prop_cb, "Display named property of resource ('class', 'type', or 'provider') " "(requires --resource)", "PROPERTY" }, { "locate", 'W', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, flag_cb, "Show node(s) currently running resource", NULL }, { "constraints", 'a', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, flag_cb, "Display the location and colocation constraints that apply to a\n" INDENT "resource, and if --recursive is specified, to the resources\n" INDENT "directly or indirectly involved in those colocations.\n" INDENT "If the named resource is part of a group, or a clone or\n" INDENT "bundle instance, constraints for the collective resource\n" INDENT "will be shown unless --force is given.", NULL }, { "stack", 'A', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, flag_cb, "Equivalent to --constraints --recursive", NULL }, { "why", 'Y', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, why_cb, "Show why resources are not running, optionally filtered by\n" INDENT "--resource and/or --node", NULL }, { NULL } }; static GOptionEntry command_entries[] = { { "validate", 0, G_OPTION_FLAG_OPTIONAL_ARG, G_OPTION_ARG_CALLBACK, validate_or_force_cb, "Validate resource configuration by calling agent's validate-all\n" INDENT "action. The configuration may be specified either by giving an\n" INDENT "existing resource name with -r, or by specifying --class,\n" INDENT "--agent, and --provider arguments, along with any number of\n" INDENT "--option arguments. An optional LEVEL argument can be given\n" INDENT "to control the level of checking performed.", "LEVEL" }, { "cleanup", 'C', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, cleanup_refresh_cb, "If resource has any past failures, clear its history and fail\n" INDENT "count. Optionally filtered by --resource, --node, --operation\n" INDENT "and --interval (otherwise all). --operation and --interval\n" INDENT "apply to fail counts, but entire history is always clear, to\n" INDENT "allow current state to be rechecked. If the named resource is\n" INDENT "part of a group, or one numbered instance of a clone or bundled\n" INDENT "resource, the clean-up applies to the whole collective resource\n" INDENT "unless --force is given.", NULL }, { "refresh", 'R', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, cleanup_refresh_cb, "Delete resource's history (including failures) so its current state\n" INDENT "is rechecked. Optionally filtered by --resource and --node\n" INDENT "(otherwise all). If the named resource is part of a group, or one\n" INDENT "numbered instance of a clone or bundled resource, the refresh\n" INDENT "applies to the whole collective resource unless --force is given.", NULL }, { "set-parameter", 'p', G_OPTION_FLAG_NONE, G_OPTION_ARG_CALLBACK, set_delete_param_cb, "Set named parameter for resource (requires -v). Use instance\n" INDENT "attribute unless --element, --meta, or --utilization is " "specified.", "PARAM" }, { "delete-parameter", 'd', G_OPTION_FLAG_NONE, G_OPTION_ARG_CALLBACK, set_delete_param_cb, "Delete named parameter for resource. Use instance attribute\n" INDENT "unless --element, --meta or, --utilization is specified.", "PARAM" }, { "set-property", 'S', G_OPTION_FLAG_HIDDEN, G_OPTION_ARG_CALLBACK, set_prop_cb, "Set named property of resource ('class', 'type', or 'provider') " "(requires -r, -t, -v)", "PROPERTY" }, { NULL } }; static GOptionEntry location_entries[] = { { "move", 'M', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, flag_cb, "Create a constraint to move resource. If --node is specified,\n" INDENT "the constraint will be to move to that node, otherwise it\n" INDENT "will be to ban the current node. Unless --force is specified\n" INDENT "this will return an error if the resource is already running\n" INDENT "on the specified node. If --force is specified, this will\n" INDENT "always ban the current node.\n" INDENT "Optional: --lifetime, --promoted. NOTE: This may prevent the\n" INDENT "resource from running on its previous location until the\n" INDENT "implicit constraint expires or is removed with --clear.", NULL }, { "ban", 'B', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, flag_cb, "Create a constraint to keep resource off a node.\n" INDENT "Optional: --node, --lifetime, --promoted.\n" INDENT "NOTE: This will prevent the resource from running on the\n" INDENT "affected node until the implicit constraint expires or is\n" INDENT "removed with --clear. If --node is not specified, it defaults\n" INDENT "to the node currently running the resource for primitives\n" INDENT "and groups, or the promoted instance of promotable clones with\n" INDENT PCMK_META_PROMOTED_MAX "=1 (all other situations result in an\n" INDENT "error as there is no sane default).", NULL }, { "clear", 'U', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, flag_cb, "Remove all constraints created by the --ban and/or --move\n" INDENT "commands. Requires: --resource. Optional: --node, --promoted,\n" INDENT "--expired. If --node is not specified, all constraints created\n" INDENT "by --ban and --move will be removed for the named resource. If\n" INDENT "--node and --force are specified, any constraint created by\n" INDENT "--move will be cleared, even if it is not for the specified\n" INDENT "node. If --expired is specified, only those constraints whose\n" INDENT "lifetimes have expired will be removed.", NULL }, { "expired", 'e', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, expired_cb, "Modifies the --clear argument to remove constraints with\n" INDENT "expired lifetimes.", NULL }, { "lifetime", 'u', G_OPTION_FLAG_NONE, G_OPTION_ARG_STRING, &options.move_lifetime, "Lifespan (as ISO 8601 duration) of created constraints (with\n" INDENT "-B, -M) see https://en.wikipedia.org/wiki/ISO_8601#Durations)", "TIMESPEC" }, { "promoted", 0, G_OPTION_FLAG_NONE, G_OPTION_ARG_NONE, &options.promoted_role_only, "Limit scope of command to promoted role (with -B, -M, -U). For\n" INDENT "-B and -M, previously promoted instances may remain\n" INDENT "active in the unpromoted role.", NULL }, // Deprecated since 2.1.0 { "master", 0, G_OPTION_FLAG_NONE, G_OPTION_ARG_NONE, &options.promoted_role_only, "Deprecated: Use --promoted instead", NULL }, { NULL } }; static GOptionEntry advanced_entries[] = { { "delete", 'D', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, delete_cb, "(Advanced) Delete a resource from the CIB. Required: -t", NULL }, { "fail", 'F', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, fail_cb, "(Advanced) Tell the cluster this resource has failed", NULL }, { "restart", 0, G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, restart_cb, "(Advanced) Tell the cluster to restart this resource and\n" INDENT "anything that depends on it", NULL }, { "wait", 0, G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, wait_cb, "(Advanced) Wait until the cluster settles into a stable state", NULL }, { "digests", 0, G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, digests_cb, "(Advanced) Show parameter hashes that Pacemaker uses to detect\n" INDENT "configuration changes (only accurate if there is resource\n" INDENT "history on the specified node). Required: --resource, --node.\n" INDENT "Optional: any NAME=VALUE parameters will be used to override\n" INDENT "the configuration (to see what the hash would be with those\n" INDENT "changes).", NULL }, { "force-demote", 0, G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, validate_or_force_cb, "(Advanced) Bypass the cluster and demote a resource on the local\n" INDENT "node. Unless --force is specified, this will refuse to do so if\n" INDENT "the cluster believes the resource is a clone instance already\n" INDENT "running on the local node.", NULL }, { "force-stop", 0, G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, validate_or_force_cb, "(Advanced) Bypass the cluster and stop a resource on the local node", NULL }, { "force-start", 0, G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, validate_or_force_cb, "(Advanced) Bypass the cluster and start a resource on the local\n" INDENT "node. Unless --force is specified, this will refuse to do so if\n" INDENT "the cluster believes the resource is a clone instance already\n" INDENT "running on the local node.", NULL }, { "force-promote", 0, G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, validate_or_force_cb, "(Advanced) Bypass the cluster and promote a resource on the local\n" INDENT "node. Unless --force is specified, this will refuse to do so if\n" INDENT "the cluster believes the resource is a clone instance already\n" INDENT "running on the local node.", NULL }, { "force-check", 0, G_OPTION_FLAG_OPTIONAL_ARG, G_OPTION_ARG_CALLBACK, validate_or_force_cb, "(Advanced) Bypass the cluster and check the state of a resource on\n" INDENT "the local node. An optional LEVEL argument can be given\n" INDENT "to control the level of checking performed.", "LEVEL" }, { NULL } }; static GOptionEntry addl_entries[] = { { "node", 'N', G_OPTION_FLAG_NONE, G_OPTION_ARG_STRING, &options.host_uname, "Node name", "NAME" }, { "recursive", 0, G_OPTION_FLAG_NONE, G_OPTION_ARG_NONE, &options.recursive, "Follow colocation chains when using --set-parameter or --constraints", NULL }, { "resource-type", 't', G_OPTION_FLAG_NONE, G_OPTION_ARG_STRING, &options.rsc_type, "Resource XML element (primitive, group, etc.) (with -D)", "ELEMENT" }, { "parameter-value", 'v', G_OPTION_FLAG_NONE, G_OPTION_ARG_STRING, &options.prop_value, "Value to use with -p", "PARAM" }, { "meta", 'm', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, attr_set_type_cb, "Use resource meta-attribute instead of instance attribute\n" INDENT "(with -p, -g, -d)", NULL }, { "utilization", 'z', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, attr_set_type_cb, "Use resource utilization attribute instead of instance attribute\n" INDENT "(with -p, -g, -d)", NULL }, { "element", 0, G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, attr_set_type_cb, "Use resource element attribute instead of instance attribute\n" INDENT "(with -p, -g, -d)", NULL }, { "operation", 'n', G_OPTION_FLAG_NONE, G_OPTION_ARG_STRING, &options.operation, "Operation to clear instead of all (with -C -r)", "OPERATION" }, { "interval", 'I', G_OPTION_FLAG_NONE, G_OPTION_ARG_STRING, &options.interval_spec, "Interval of operation to clear (default 0) (with -C -r -n)", "N" }, { "class", 0, G_OPTION_FLAG_NONE, G_OPTION_ARG_CALLBACK, class_cb, "The standard the resource agent conforms to (for example, ocf).\n" INDENT "Use with --agent, --provider, --option, and --validate.", "CLASS" }, { "agent", 0, G_OPTION_FLAG_NONE, G_OPTION_ARG_CALLBACK, agent_provider_cb, "The agent to use (for example, IPaddr). Use with --class,\n" INDENT "--provider, --option, and --validate.", "AGENT" }, { "provider", 0, G_OPTION_FLAG_NONE, G_OPTION_ARG_CALLBACK, agent_provider_cb, "The vendor that supplies the resource agent (for example,\n" INDENT "heartbeat). Use with --class, --agent, --option, and --validate.", "PROVIDER" }, { "option", 0, G_OPTION_FLAG_NONE, G_OPTION_ARG_CALLBACK, option_cb, "Specify a device configuration parameter as NAME=VALUE (may be\n" INDENT "specified multiple times). Use with --validate and without the\n" INDENT "-r option.", "PARAM" }, { "set-name", 's', G_OPTION_FLAG_NONE, G_OPTION_ARG_STRING, &options.prop_set, "(Advanced) XML ID of attributes element to use (with -p, -d)", "ID" }, { "nvpair", 'i', G_OPTION_FLAG_NONE, G_OPTION_ARG_STRING, &options.prop_id, "(Advanced) XML ID of nvpair element to use (with -p, -d)", "ID" }, { "timeout", 'T', G_OPTION_FLAG_NONE, G_OPTION_ARG_CALLBACK, timeout_cb, "(Advanced) Abort if command does not finish in this time (with\n" INDENT "--restart, --wait, --force-*)", "N" }, { "force", 'f', G_OPTION_FLAG_NONE, G_OPTION_ARG_NONE, &options.force, "Force the action to be performed. See help for individual commands for\n" INDENT "additional behavior.", NULL }, { "xml-file", 'x', G_OPTION_FLAG_HIDDEN, G_OPTION_ARG_FILENAME, &options.xml_file, NULL, "FILE" }, { "host-uname", 'H', G_OPTION_FLAG_HIDDEN, G_OPTION_ARG_STRING, &options.host_uname, NULL, "HOST" }, { NULL } }; static void reset_options(void) { options.require_crmd = FALSE; options.require_node = FALSE; options.require_cib = TRUE; options.require_scheduler = TRUE; options.require_resource = TRUE; options.find_flags = 0; } gboolean agent_provider_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error) { options.cmdline_config = TRUE; options.require_resource = FALSE; if (pcmk__str_eq(option_name, "--provider", pcmk__str_casei)) { pcmk__str_update(&options.v_provider, optarg); } else { pcmk__str_update(&options.v_agent, optarg); } return TRUE; } gboolean attr_set_type_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error) { if (pcmk__str_any_of(option_name, "-m", "--meta", NULL)) { options.attr_set_type = PCMK_XE_META_ATTRIBUTES; } else if (pcmk__str_any_of(option_name, "-z", "--utilization", NULL)) { options.attr_set_type = PCMK_XE_UTILIZATION; } else if (pcmk__str_eq(option_name, "--element", pcmk__str_casei)) { options.attr_set_type = ATTR_SET_ELEMENT; } return TRUE; } gboolean class_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error) { pcmk__str_update(&options.v_class, optarg); options.cmdline_config = TRUE; options.require_resource = FALSE; return TRUE; } gboolean cleanup_refresh_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error) { if (pcmk__str_any_of(option_name, "-C", "--cleanup", NULL)) { SET_COMMAND(cmd_cleanup); } else { SET_COMMAND(cmd_refresh); } options.require_resource = FALSE; if (getenv("CIB_file") == NULL) { options.require_crmd = TRUE; } options.find_flags = pcmk_rsc_match_history|pcmk_rsc_match_anon_basename; return TRUE; } gboolean delete_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error) { SET_COMMAND(cmd_delete); options.require_scheduler = FALSE; options.find_flags = pcmk_rsc_match_history|pcmk_rsc_match_basename; return TRUE; } gboolean expired_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error) { options.clear_expired = TRUE; options.require_resource = FALSE; return TRUE; } static void get_agent_spec(const gchar *optarg) { options.require_cib = FALSE; options.require_scheduler = FALSE; options.require_resource = FALSE; pcmk__str_update(&options.agent_spec, optarg); } gboolean list_agents_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error) { SET_COMMAND(cmd_list_agents); get_agent_spec(optarg); return TRUE; } gboolean list_providers_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error) { SET_COMMAND(cmd_list_providers); get_agent_spec(optarg); return TRUE; } gboolean list_standards_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error) { SET_COMMAND(cmd_list_standards); options.require_cib = FALSE; options.require_scheduler = FALSE; options.require_resource = FALSE; return TRUE; } gboolean list_alternatives_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error) { SET_COMMAND(cmd_list_alternatives); get_agent_spec(optarg); return TRUE; } gboolean metadata_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error) { SET_COMMAND(cmd_metadata); get_agent_spec(optarg); return TRUE; } gboolean option_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error) { char *name = NULL; char *value = NULL; if (pcmk__scan_nvpair(optarg, &name, &value) != 2) { return FALSE; } if (options.cmdline_params == NULL) { options.cmdline_params = pcmk__strkey_table(free, free); } g_hash_table_replace(options.cmdline_params, name, value); return TRUE; } gboolean fail_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error) { SET_COMMAND(cmd_fail); options.require_crmd = TRUE; options.require_node = TRUE; return TRUE; } gboolean flag_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error) { if (pcmk__str_any_of(option_name, "-U", "--clear", NULL)) { SET_COMMAND(cmd_clear); options.find_flags = pcmk_rsc_match_history |pcmk_rsc_match_anon_basename; } else if (pcmk__str_any_of(option_name, "-B", "--ban", NULL)) { SET_COMMAND(cmd_ban); options.find_flags = pcmk_rsc_match_history |pcmk_rsc_match_anon_basename; } else if (pcmk__str_any_of(option_name, "-M", "--move", NULL)) { SET_COMMAND(cmd_move); options.find_flags = pcmk_rsc_match_history |pcmk_rsc_match_anon_basename; } else if (pcmk__str_any_of(option_name, "-q", "--query-xml", NULL)) { SET_COMMAND(cmd_query_xml); options.find_flags = pcmk_rsc_match_history|pcmk_rsc_match_basename; } else if (pcmk__str_any_of(option_name, "-w", "--query-xml-raw", NULL)) { SET_COMMAND(cmd_query_raw_xml); options.find_flags = pcmk_rsc_match_history|pcmk_rsc_match_basename; } else if (pcmk__str_any_of(option_name, "-W", "--locate", NULL)) { SET_COMMAND(cmd_locate); options.find_flags = pcmk_rsc_match_history |pcmk_rsc_match_anon_basename; } else if (pcmk__str_any_of(option_name, "-a", "--constraints", NULL)) { SET_COMMAND(cmd_colocations); options.find_flags = pcmk_rsc_match_history |pcmk_rsc_match_anon_basename; } else if (pcmk__str_any_of(option_name, "-A", "--stack", NULL)) { SET_COMMAND(cmd_colocations); options.find_flags = pcmk_rsc_match_history |pcmk_rsc_match_anon_basename; options.recursive = TRUE; } return TRUE; } gboolean get_param_prop_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error) { if (pcmk__str_any_of(option_name, "-g", "--get-parameter", NULL)) { SET_COMMAND(cmd_get_param); } else { SET_COMMAND(cmd_get_property); } pcmk__str_update(&options.prop_name, optarg); options.find_flags = pcmk_rsc_match_history|pcmk_rsc_match_basename; return TRUE; } gboolean list_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error) { if (pcmk__str_any_of(option_name, "-c", "--list-cts", NULL)) { SET_COMMAND(cmd_cts); } else if (pcmk__str_any_of(option_name, "-L", "--list", NULL)) { SET_COMMAND(cmd_list_resources); } else if (pcmk__str_any_of(option_name, "-l", "--list-raw", NULL)) { SET_COMMAND(cmd_list_instances); } else if (pcmk__str_any_of(option_name, "-O", "--list-operations", NULL)) { SET_COMMAND(cmd_list_active_ops); } else { SET_COMMAND(cmd_list_all_ops); } options.require_resource = FALSE; return TRUE; } gboolean set_delete_param_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error) { if (pcmk__str_any_of(option_name, "-p", "--set-parameter", NULL)) { SET_COMMAND(cmd_set_param); } else { SET_COMMAND(cmd_delete_param); } pcmk__str_update(&options.prop_name, optarg); options.find_flags = pcmk_rsc_match_history|pcmk_rsc_match_basename; return TRUE; } gboolean set_prop_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error) { SET_COMMAND(cmd_set_property); options.require_scheduler = FALSE; pcmk__str_update(&options.prop_name, optarg); options.find_flags = pcmk_rsc_match_history|pcmk_rsc_match_basename; return TRUE; } gboolean timeout_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error) { options.timeout_ms = crm_get_msec(optarg); + + if (options.timeout_ms < 0) { + crm_warn("Ignoring invalid timeout '%s'", optarg); + options.timeout_ms = 0; + } else { + options.timeout_ms = QB_MIN(options.timeout_ms, INT_MAX); + } return TRUE; } gboolean validate_or_force_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error) { SET_COMMAND(cmd_execute_agent); if (options.operation) { g_free(options.operation); } options.operation = g_strdup(option_name + 2); // skip "--" options.find_flags = pcmk_rsc_match_history|pcmk_rsc_match_anon_basename; if (options.override_params == NULL) { options.override_params = pcmk__strkey_table(free, free); } if (optarg != NULL) { if (pcmk__scan_min_int(optarg, &options.check_level, 0) != pcmk_rc_ok) { g_set_error(error, G_OPTION_ERROR, CRM_EX_INVALID_PARAM, _("Invalid check level setting: %s"), optarg); return FALSE; } } return TRUE; } gboolean restart_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error) { SET_COMMAND(cmd_restart); options.find_flags = pcmk_rsc_match_history|pcmk_rsc_match_anon_basename; return TRUE; } gboolean digests_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error) { SET_COMMAND(cmd_digests); options.find_flags = pcmk_rsc_match_history|pcmk_rsc_match_anon_basename; if (options.override_params == NULL) { options.override_params = pcmk__strkey_table(free, free); } options.require_node = TRUE; options.require_scheduler = TRUE; return TRUE; } gboolean wait_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error) { SET_COMMAND(cmd_wait); options.require_resource = FALSE; options.require_scheduler = FALSE; return TRUE; } gboolean why_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error) { SET_COMMAND(cmd_why); options.require_resource = FALSE; options.find_flags = pcmk_rsc_match_history|pcmk_rsc_match_anon_basename; return TRUE; } static int ban_or_move(pcmk__output_t *out, pcmk_resource_t *rsc, const char *move_lifetime) { int rc = pcmk_rc_ok; pcmk_node_t *current = NULL; unsigned int nactive = 0; CRM_CHECK(rsc != NULL, return EINVAL); current = pe__find_active_requires(rsc, &nactive); if (nactive == 1) { rc = cli_resource_ban(out, options.rsc_id, current->details->uname, move_lifetime, cib_conn, options.cib_options, options.promoted_role_only, PCMK__ROLE_PROMOTED); } else if (pcmk_is_set(rsc->flags, pcmk_rsc_promotable)) { int count = 0; GList *iter = NULL; current = NULL; for(iter = rsc->children; iter; iter = iter->next) { pcmk_resource_t *child = (pcmk_resource_t *)iter->data; enum rsc_role_e child_role = child->fns->state(child, TRUE); if (child_role == pcmk_role_promoted) { count++; current = pcmk__current_node(child); } } if(count == 1 && current) { rc = cli_resource_ban(out, options.rsc_id, current->details->uname, move_lifetime, cib_conn, options.cib_options, options.promoted_role_only, PCMK__ROLE_PROMOTED); } else { rc = EINVAL; g_set_error(&error, PCMK__EXITC_ERROR, CRM_EX_USAGE, _("Resource '%s' not moved: active in %d locations (promoted in %d).\n" "To prevent '%s' from running on a specific location, " "specify a node." "To prevent '%s' from being promoted at a specific " "location, specify a node and the --promoted option."), options.rsc_id, nactive, count, options.rsc_id, options.rsc_id); } } else { rc = EINVAL; g_set_error(&error, PCMK__EXITC_ERROR, CRM_EX_USAGE, _("Resource '%s' not moved: active in %d locations.\n" "To prevent '%s' from running on a specific location, " "specify a node."), options.rsc_id, nactive, options.rsc_id); } return rc; } static void cleanup(pcmk__output_t *out, pcmk_resource_t *rsc, pcmk_node_t *node) { int rc = pcmk_rc_ok; if (options.force == FALSE) { rsc = uber_parent(rsc); } crm_debug("Erasing failures of %s (%s requested) on %s", rsc->id, options.rsc_id, (options.host_uname? options.host_uname: "all nodes")); rc = cli_resource_delete(controld_api, options.host_uname, rsc, options.operation, options.interval_spec, TRUE, scheduler, options.force); if ((rc == pcmk_rc_ok) && !out->is_quiet(out)) { // Show any reasons why resource might stay stopped cli_resource_check(out, rsc, node); } if (rc == pcmk_rc_ok) { start_mainloop(controld_api); } } static int clear_constraints(pcmk__output_t *out, xmlNodePtr *cib_xml_copy) { GList *before = NULL; GList *after = NULL; GList *remaining = NULL; GList *ele = NULL; pcmk_node_t *dest = NULL; int rc = pcmk_rc_ok; if (!out->is_quiet(out)) { before = build_constraint_list(scheduler->input); } if (options.clear_expired) { rc = cli_resource_clear_all_expired(scheduler->input, cib_conn, options.cib_options, options.rsc_id, options.host_uname, options.promoted_role_only); } else if (options.host_uname) { dest = pe_find_node(scheduler->nodes, options.host_uname); if (dest == NULL) { rc = pcmk_rc_node_unknown; if (!out->is_quiet(out)) { g_list_free(before); } return rc; } rc = cli_resource_clear(options.rsc_id, dest->details->uname, NULL, cib_conn, options.cib_options, TRUE, options.force); } else { rc = cli_resource_clear(options.rsc_id, NULL, scheduler->nodes, cib_conn, options.cib_options, TRUE, options.force); } if (!out->is_quiet(out)) { rc = cib_conn->cmds->query(cib_conn, NULL, cib_xml_copy, cib_scope_local | cib_sync_call); rc = pcmk_legacy2rc(rc); if (rc != pcmk_rc_ok) { g_set_error(&error, PCMK__RC_ERROR, rc, _("Could not get modified CIB: %s\n"), pcmk_rc_str(rc)); g_list_free(before); free_xml(*cib_xml_copy); *cib_xml_copy = NULL; return rc; } scheduler->input = *cib_xml_copy; cluster_status(scheduler); after = build_constraint_list(scheduler->input); remaining = pcmk__subtract_lists(before, after, (GCompareFunc) strcmp); for (ele = remaining; ele != NULL; ele = ele->next) { out->info(out, "Removing constraint: %s", (char *) ele->data); } g_list_free(before); g_list_free(after); g_list_free(remaining); } return rc; } static int initialize_scheduler_data(xmlNodePtr *cib_xml_copy) { int rc = pcmk_rc_ok; if (options.xml_file != NULL) { *cib_xml_copy = filename2xml(options.xml_file); if (*cib_xml_copy == NULL) { rc = pcmk_rc_cib_corrupt; } } else { rc = cib_conn->cmds->query(cib_conn, NULL, cib_xml_copy, cib_scope_local | cib_sync_call); rc = pcmk_legacy2rc(rc); } if (rc == pcmk_rc_ok) { scheduler = pe_new_working_set(); if (scheduler == NULL) { rc = ENOMEM; } else { pcmk__set_scheduler_flags(scheduler, pcmk_sched_no_counts |pcmk_sched_no_compat); scheduler->priv = out; rc = update_scheduler_input(scheduler, cib_xml_copy); } } if (rc != pcmk_rc_ok) { free_xml(*cib_xml_copy); *cib_xml_copy = NULL; return rc; } cluster_status(scheduler); return pcmk_rc_ok; } static int refresh(pcmk__output_t *out) { int rc = pcmk_rc_ok; const char *router_node = options.host_uname; int attr_options = pcmk__node_attr_none; if (options.host_uname) { pcmk_node_t *node = pe_find_node(scheduler->nodes, options.host_uname); if (pe__is_guest_or_remote_node(node)) { node = pcmk__current_node(node->details->remote_rsc); if (node == NULL) { rc = ENXIO; g_set_error(&error, PCMK__RC_ERROR, rc, _("No cluster connection to Pacemaker Remote node %s detected"), options.host_uname); return rc; } router_node = node->details->uname; attr_options |= pcmk__node_attr_remote; } } if (controld_api == NULL) { out->info(out, "Dry run: skipping clean-up of %s due to CIB_file", options.host_uname? options.host_uname : "all nodes"); rc = pcmk_rc_ok; return rc; } crm_debug("Re-checking the state of all resources on %s", options.host_uname?options.host_uname:"all nodes"); rc = pcmk__attrd_api_clear_failures(NULL, options.host_uname, NULL, NULL, NULL, NULL, attr_options); if (pcmk_controld_api_reprobe(controld_api, options.host_uname, router_node) == pcmk_rc_ok) { start_mainloop(controld_api); } return rc; } static void refresh_resource(pcmk__output_t *out, pcmk_resource_t *rsc, pcmk_node_t *node) { int rc = pcmk_rc_ok; if (options.force == FALSE) { rsc = uber_parent(rsc); } crm_debug("Re-checking the state of %s (%s requested) on %s", rsc->id, options.rsc_id, (options.host_uname? options.host_uname: "all nodes")); rc = cli_resource_delete(controld_api, options.host_uname, rsc, NULL, 0, FALSE, scheduler, options.force); if ((rc == pcmk_rc_ok) && !out->is_quiet(out)) { // Show any reasons why resource might stay stopped cli_resource_check(out, rsc, node); } if (rc == pcmk_rc_ok) { start_mainloop(controld_api); } } static int set_property(void) { int rc = pcmk_rc_ok; xmlNode *msg_data = NULL; if (pcmk__str_empty(options.rsc_type)) { g_set_error(&error, PCMK__EXITC_ERROR, CRM_EX_USAGE, _("Must specify -t with resource type")); rc = ENXIO; return rc; } else if (pcmk__str_empty(options.prop_value)) { g_set_error(&error, PCMK__EXITC_ERROR, CRM_EX_USAGE, _("Must supply -v with new value")); rc = ENXIO; return rc; } CRM_LOG_ASSERT(options.prop_name != NULL); msg_data = create_xml_node(NULL, options.rsc_type); crm_xml_add(msg_data, PCMK_XA_ID, options.rsc_id); crm_xml_add(msg_data, options.prop_name, options.prop_value); rc = cib_conn->cmds->modify(cib_conn, PCMK_XE_RESOURCES, msg_data, options.cib_options); rc = pcmk_legacy2rc(rc); free_xml(msg_data); return rc; } static int show_metadata(pcmk__output_t *out, const char *agent_spec) { int rc = pcmk_rc_ok; char *standard = NULL; char *provider = NULL; char *type = NULL; char *metadata = NULL; lrmd_t *lrmd_conn = NULL; rc = lrmd__new(&lrmd_conn, NULL, NULL, 0); if (rc != pcmk_rc_ok) { g_set_error(&error, PCMK__RC_ERROR, rc, _("Could not create executor connection")); lrmd_api_delete(lrmd_conn); return rc; } rc = crm_parse_agent_spec(agent_spec, &standard, &provider, &type); rc = pcmk_legacy2rc(rc); if (rc == pcmk_rc_ok) { rc = lrmd_conn->cmds->get_metadata(lrmd_conn, standard, provider, type, &metadata, 0); rc = pcmk_legacy2rc(rc); if (metadata) { out->output_xml(out, "metadata", metadata); free(metadata); } else { /* We were given a validly formatted spec, but it doesn't necessarily * match up with anything that exists. Use ENXIO as the return code * here because that maps to an exit code of CRM_EX_NOSUCH, which * probably is the most common reason to get here. */ rc = ENXIO; g_set_error(&error, PCMK__RC_ERROR, rc, _("Metadata query for %s failed: %s"), agent_spec, pcmk_rc_str(rc)); } } else { rc = ENXIO; g_set_error(&error, PCMK__RC_ERROR, rc, _("'%s' is not a valid agent specification"), agent_spec); } lrmd_api_delete(lrmd_conn); return rc; } static void validate_cmdline_config(void) { // Cannot use both --resource and command-line resource configuration if (options.rsc_id != NULL) { g_set_error(&error, PCMK__EXITC_ERROR, CRM_EX_USAGE, _("--resource cannot be used with --class, --agent, and --provider")); // Not all commands support command-line resource configuration } else if (options.rsc_cmd != cmd_execute_agent) { g_set_error(&error, PCMK__EXITC_ERROR, CRM_EX_USAGE, _("--class, --agent, and --provider can only be used with " "--validate and --force-*")); // Not all of --class, --agent, and --provider need to be given. Not all // classes support the concept of a provider. Check that what we were given // is valid. } else if (pcmk__str_eq(options.v_class, "stonith", pcmk__str_none)) { if (options.v_provider != NULL) { g_set_error(&error, PCMK__EXITC_ERROR, CRM_EX_USAGE, _("stonith does not support providers")); } else if (stonith_agent_exists(options.v_agent, 0) == FALSE) { g_set_error(&error, PCMK__EXITC_ERROR, CRM_EX_USAGE, _("%s is not a known stonith agent"), options.v_agent ? options.v_agent : ""); } } else if (resources_agent_exists(options.v_class, options.v_provider, options.v_agent) == FALSE) { g_set_error(&error, PCMK__EXITC_ERROR, CRM_EX_USAGE, _("%s:%s:%s is not a known resource"), options.v_class ? options.v_class : "", options.v_provider ? options.v_provider : "", options.v_agent ? options.v_agent : ""); } if (error != NULL) { return; } if (options.cmdline_params == NULL) { options.cmdline_params = pcmk__strkey_table(free, free); } options.require_resource = FALSE; options.require_scheduler = FALSE; options.require_cib = FALSE; } static GOptionContext * build_arg_context(pcmk__common_args_t *args, GOptionGroup **group) { GOptionContext *context = NULL; GOptionEntry extra_prog_entries[] = { { "quiet", 'Q', G_OPTION_FLAG_NONE, G_OPTION_ARG_NONE, &(args->quiet), "Be less descriptive in output.", NULL }, { "resource", 'r', G_OPTION_FLAG_NONE, G_OPTION_ARG_STRING, &options.rsc_id, "Resource ID", "ID" }, { G_OPTION_REMAINING, 0, G_OPTION_FLAG_NONE, G_OPTION_ARG_STRING_ARRAY, &options.remainder, NULL, NULL }, { NULL } }; const char *description = "Examples:\n\n" "List the available OCF agents:\n\n" "\t# crm_resource --list-agents ocf\n\n" "List the available OCF agents from the linux-ha project:\n\n" "\t# crm_resource --list-agents ocf:heartbeat\n\n" "Move 'myResource' to a specific node:\n\n" "\t# crm_resource --resource myResource --move --node altNode\n\n" "Allow (but not force) 'myResource' to move back to its original " "location:\n\n" "\t# crm_resource --resource myResource --clear\n\n" "Stop 'myResource' (and anything that depends on it):\n\n" "\t# crm_resource --resource myResource --set-parameter " PCMK_META_TARGET_ROLE "--meta --parameter-value Stopped\n\n" "Tell the cluster not to manage 'myResource' (the cluster will not " "attempt to start or stop the\n" "resource under any circumstances; useful when performing maintenance " "tasks on a resource):\n\n" "\t# crm_resource --resource myResource --set-parameter " PCMK_META_IS_MANAGED "--meta --parameter-value false\n\n" "Erase the operation history of 'myResource' on 'aNode' (the cluster " "will 'forget' the existing\n" "resource state, including any errors, and attempt to recover the" "resource; useful when a resource\n" "had failed permanently and has been repaired by an administrator):\n\n" "\t# crm_resource --resource myResource --cleanup --node aNode\n\n"; context = pcmk__build_arg_context(args, "text (default), xml", group, NULL); g_option_context_set_description(context, description); /* Add the -Q option, which cannot be part of the globally supported options * because some tools use that flag for something else. */ pcmk__add_main_args(context, extra_prog_entries); pcmk__add_arg_group(context, "queries", "Queries:", "Show query help", query_entries); pcmk__add_arg_group(context, "commands", "Commands:", "Show command help", command_entries); pcmk__add_arg_group(context, "locations", "Locations:", "Show location help", location_entries); pcmk__add_arg_group(context, "advanced", "Advanced:", "Show advanced option help", advanced_entries); pcmk__add_arg_group(context, "additional", "Additional Options:", "Show additional options", addl_entries); return context; } int main(int argc, char **argv) { xmlNode *cib_xml_copy = NULL; pcmk_resource_t *rsc = NULL; pcmk_node_t *node = NULL; int rc = pcmk_rc_ok; GOptionGroup *output_group = NULL; gchar **processed_args = NULL; GOptionContext *context = NULL; /* * Parse command line arguments */ args = pcmk__new_common_args(SUMMARY); processed_args = pcmk__cmdline_preproc(argv, "GHINSTdginpstuvx"); context = build_arg_context(args, &output_group); pcmk__register_formats(output_group, formats); if (!g_option_context_parse_strv(context, &processed_args, &error)) { exit_code = CRM_EX_USAGE; goto done; } pcmk__cli_init_logging("crm_resource", args->verbosity); rc = pcmk__output_new(&out, args->output_ty, args->output_dest, argv); if (rc != pcmk_rc_ok) { exit_code = CRM_EX_ERROR; g_set_error(&error, PCMK__EXITC_ERROR, exit_code, _("Error creating output format %s: %s"), args->output_ty, pcmk_rc_str(rc)); goto done; } pe__register_messages(out); crm_resource_register_messages(out); lrmd__register_messages(out); pcmk__register_lib_messages(out); out->quiet = args->quiet; crm_log_args(argc, argv); /* * Validate option combinations */ // If the user didn't explicitly specify a command, list resources if (options.rsc_cmd == cmd_none) { options.rsc_cmd = cmd_list_resources; options.require_resource = FALSE; } // --expired without --clear/-U doesn't make sense if (options.clear_expired && (options.rsc_cmd != cmd_clear)) { exit_code = CRM_EX_USAGE; g_set_error(&error, PCMK__EXITC_ERROR, exit_code, _("--expired requires --clear or -U")); goto done; } if ((options.remainder != NULL) && (options.override_params != NULL)) { // Commands that use positional arguments will create override_params for (gchar **s = options.remainder; *s; s++) { char *name = calloc(1, strlen(*s)); char *value = calloc(1, strlen(*s)); int rc = sscanf(*s, "%[^=]=%s", name, value); if (rc == 2) { g_hash_table_replace(options.override_params, name, value); } else { exit_code = CRM_EX_USAGE; g_set_error(&error, PCMK__EXITC_ERROR, exit_code, _("Error parsing '%s' as a name=value pair"), argv[optind]); free(value); free(name); goto done; } } } else if (options.remainder != NULL) { gchar **strv = NULL; gchar *msg = NULL; int i = 1; int len = 0; for (gchar **s = options.remainder; *s; s++) { len++; } CRM_ASSERT(len > 0); /* Add 1 for the strv[0] string below, and add another 1 for the NULL * at the end of the array so g_strjoinv knows when to stop. */ strv = calloc(len+2, sizeof(char *)); strv[0] = strdup("non-option ARGV-elements:\n"); for (gchar **s = options.remainder; *s; s++) { strv[i] = crm_strdup_printf("[%d of %d] %s\n", i, len, *s); i++; } strv[i] = NULL; exit_code = CRM_EX_USAGE; msg = g_strjoinv("", strv); g_set_error(&error, PCMK__EXITC_ERROR, exit_code, "%s", msg); g_free(msg); /* Don't try to free the last element, which is just NULL. */ for(i = 0; i < len+1; i++) { free(strv[i]); } free(strv); goto done; } if (pcmk__str_eq(args->output_ty, "xml", pcmk__str_none)) { /* Kind of a hack to display XML lists using a real tag instead of . This just * saves from having to write custom messages to build the lists around all these things */ switch (options.rsc_cmd) { case cmd_execute_agent: case cmd_list_resources: case cmd_query_xml: case cmd_query_raw_xml: case cmd_list_active_ops: case cmd_list_all_ops: case cmd_colocations: pcmk__force_args(context, &error, "%s --xml-simple-list --xml-substitute", g_get_prgname()); break; default: pcmk__force_args(context, &error, "%s --xml-substitute", g_get_prgname()); break; } } else if (pcmk__str_eq(args->output_ty, "text", pcmk__str_null_matches)) { if ((options.rsc_cmd == cmd_colocations) || options.rsc_cmd == cmd_list_resources) { pcmk__force_args(context, &error, "%s --text-fancy", g_get_prgname()); } } if (args->version) { out->version(out, false); goto done; } if (options.cmdline_config) { /* A resource configuration was given on the command line. Sanity-check * the values and set error if they don't make sense. */ validate_cmdline_config(); if (error != NULL) { exit_code = CRM_EX_USAGE; goto done; } } else if (options.cmdline_params != NULL) { // @COMPAT @TODO error out here when we can break backward compatibility g_hash_table_destroy(options.cmdline_params); options.cmdline_params = NULL; } if (options.require_resource && (options.rsc_id == NULL)) { exit_code = CRM_EX_USAGE; g_set_error(&error, PCMK__EXITC_ERROR, exit_code, _("Must supply a resource id with -r")); goto done; } if (options.require_node && (options.host_uname == NULL)) { exit_code = CRM_EX_USAGE; g_set_error(&error, PCMK__EXITC_ERROR, exit_code, _("Must supply a node name with -N")); goto done; } /* * Set up necessary connections */ if (options.find_flags && options.rsc_id) { options.require_scheduler = TRUE; } // Establish a connection to the CIB if needed if (options.require_cib) { cib_conn = cib_new(); if ((cib_conn == NULL) || (cib_conn->cmds == NULL)) { exit_code = CRM_EX_DISCONNECT; g_set_error(&error, PCMK__EXITC_ERROR, exit_code, _("Could not create CIB connection")); goto done; } rc = cib_conn->cmds->signon(cib_conn, crm_system_name, cib_command); rc = pcmk_legacy2rc(rc); if (rc != pcmk_rc_ok) { exit_code = pcmk_rc2exitc(rc); g_set_error(&error, PCMK__EXITC_ERROR, exit_code, _("Could not connect to the CIB: %s"), pcmk_rc_str(rc)); goto done; } } // Populate scheduler data from XML file if specified or CIB query otherwise if (options.require_scheduler) { rc = initialize_scheduler_data(&cib_xml_copy); if (rc != pcmk_rc_ok) { exit_code = pcmk_rc2exitc(rc); goto done; } } // If command requires that resource exist if specified, find it if (options.find_flags && options.rsc_id) { rsc = pe_find_resource_with_flags(scheduler->resources, options.rsc_id, options.find_flags); if (rsc == NULL) { exit_code = CRM_EX_NOSUCH; g_set_error(&error, PCMK__EXITC_ERROR, exit_code, _("Resource '%s' not found"), options.rsc_id); goto done; } /* The --ban, --clear, --move, and --restart commands do not work with * instances of clone resourcs. */ if (strchr(options.rsc_id, ':') != NULL && pe_rsc_is_clone(rsc->parent) && (options.rsc_cmd == cmd_ban || options.rsc_cmd == cmd_clear || options.rsc_cmd == cmd_move || options.rsc_cmd == cmd_restart)) { exit_code = CRM_EX_INVALID_PARAM; g_set_error(&error, PCMK__EXITC_ERROR, exit_code, _("Cannot operate on clone resource instance '%s'"), options.rsc_id); goto done; } } // If user supplied a node name, check whether it exists if ((options.host_uname != NULL) && (scheduler != NULL)) { node = pe_find_node(scheduler->nodes, options.host_uname); if (node == NULL) { exit_code = CRM_EX_NOSUCH; g_set_error(&error, PCMK__EXITC_ERROR, exit_code, _("Node '%s' not found"), options.host_uname); goto done; } } // Establish a connection to the controller if needed if (options.require_crmd) { rc = pcmk_new_ipc_api(&controld_api, pcmk_ipc_controld); if (rc != pcmk_rc_ok) { exit_code = pcmk_rc2exitc(rc); g_set_error(&error, PCMK__EXITC_ERROR, exit_code, _("Error connecting to the controller: %s"), pcmk_rc_str(rc)); goto done; } pcmk_register_ipc_callback(controld_api, controller_event_callback, NULL); rc = pcmk__connect_ipc(controld_api, pcmk_ipc_dispatch_main, 5); if (rc != pcmk_rc_ok) { exit_code = pcmk_rc2exitc(rc); g_set_error(&error, PCMK__EXITC_ERROR, exit_code, _("Error connecting to %s: %s"), pcmk_ipc_name(controld_api, true), pcmk_rc_str(rc)); goto done; } } /* * Handle requested command */ switch (options.rsc_cmd) { case cmd_list_resources: { GList *all = NULL; all = g_list_prepend(all, (gpointer) "*"); rc = out->message(out, "resource-list", scheduler, pcmk_show_inactive_rscs | pcmk_show_rsc_only | pcmk_show_pending, true, all, all, false); g_list_free(all); if (rc == pcmk_rc_no_output) { rc = ENXIO; } break; } case cmd_list_instances: rc = out->message(out, "resource-names-list", scheduler->resources); if (rc != pcmk_rc_ok) { rc = ENXIO; } break; case cmd_list_alternatives: rc = pcmk__list_alternatives(out, options.agent_spec); break; case cmd_list_agents: rc = pcmk__list_agents(out, options.agent_spec); break; case cmd_list_standards: rc = pcmk__list_standards(out); break; case cmd_list_providers: rc = pcmk__list_providers(out, options.agent_spec); break; case cmd_metadata: rc = show_metadata(out, options.agent_spec); break; case cmd_restart: /* We don't pass scheduler because rsc needs to stay valid for the * entire lifetime of cli_resource_restart(), but it will reset and * update the scheduler data multiple times, so it needs to use its * own copy. */ rc = cli_resource_restart(out, rsc, node, options.move_lifetime, options.timeout_ms, cib_conn, options.cib_options, options.promoted_role_only, options.force); break; case cmd_wait: rc = wait_till_stable(out, options.timeout_ms, cib_conn); break; case cmd_execute_agent: if (options.cmdline_config) { exit_code = cli_resource_execute_from_params(out, NULL, options.v_class, options.v_provider, options.v_agent, options.operation, options.cmdline_params, options.override_params, options.timeout_ms, args->verbosity, options.force, options.check_level); } else { exit_code = cli_resource_execute(rsc, options.rsc_id, options.operation, options.override_params, options.timeout_ms, cib_conn, scheduler, args->verbosity, options.force, options.check_level); } goto done; case cmd_digests: node = pe_find_node(scheduler->nodes, options.host_uname); if (node == NULL) { rc = pcmk_rc_node_unknown; } else { rc = pcmk__resource_digests(out, rsc, node, options.override_params); } break; case cmd_colocations: rc = out->message(out, "locations-and-colocations", rsc, options.recursive, (bool) options.force); break; case cmd_cts: rc = pcmk_rc_ok; g_list_foreach(scheduler->resources, (GFunc) cli_resource_print_cts, out); cli_resource_print_cts_constraints(scheduler); break; case cmd_fail: rc = cli_resource_fail(controld_api, options.host_uname, options.rsc_id, scheduler); if (rc == pcmk_rc_ok) { start_mainloop(controld_api); } break; case cmd_list_active_ops: rc = cli_resource_print_operations(options.rsc_id, options.host_uname, TRUE, scheduler); break; case cmd_list_all_ops: rc = cli_resource_print_operations(options.rsc_id, options.host_uname, FALSE, scheduler); break; case cmd_locate: { GList *nodes = cli_resource_search(rsc, options.rsc_id, scheduler); rc = out->message(out, "resource-search-list", nodes, options.rsc_id); g_list_free_full(nodes, free); break; } case cmd_query_xml: rc = cli_resource_print(rsc, scheduler, true); break; case cmd_query_raw_xml: rc = cli_resource_print(rsc, scheduler, false); break; case cmd_why: if ((options.host_uname != NULL) && (node == NULL)) { rc = pcmk_rc_node_unknown; } else { rc = out->message(out, "resource-reasons-list", scheduler->resources, rsc, node); } break; case cmd_clear: rc = clear_constraints(out, &cib_xml_copy); break; case cmd_move: if (options.host_uname == NULL) { rc = ban_or_move(out, rsc, options.move_lifetime); } else { rc = cli_resource_move(rsc, options.rsc_id, options.host_uname, options.move_lifetime, cib_conn, options.cib_options, scheduler, options.promoted_role_only, options.force); } if (rc == EINVAL) { exit_code = CRM_EX_USAGE; goto done; } break; case cmd_ban: if (options.host_uname == NULL) { rc = ban_or_move(out, rsc, options.move_lifetime); } else if (node == NULL) { rc = pcmk_rc_node_unknown; } else { rc = cli_resource_ban(out, options.rsc_id, node->details->uname, options.move_lifetime, cib_conn, options.cib_options, options.promoted_role_only, PCMK__ROLE_PROMOTED); } if (rc == EINVAL) { exit_code = CRM_EX_USAGE; goto done; } break; case cmd_get_property: rc = out->message(out, "property-list", rsc, options.prop_name); if (rc == pcmk_rc_no_output) { rc = ENXIO; } break; case cmd_set_property: rc = set_property(); break; case cmd_get_param: { unsigned int count = 0; GHashTable *params = NULL; pcmk_node_t *current = rsc->fns->active_node(rsc, &count, NULL); bool free_params = true; const char* value = NULL; if (count > 1) { out->err(out, "%s is active on more than one node," " returning the default value for %s", rsc->id, pcmk__s(options.prop_name, "unspecified property")); current = NULL; } crm_debug("Looking up %s in %s", options.prop_name, rsc->id); if (pcmk__str_eq(options.attr_set_type, PCMK_XE_INSTANCE_ATTRIBUTES, pcmk__str_none)) { params = pe_rsc_params(rsc, current, scheduler); free_params = false; value = g_hash_table_lookup(params, options.prop_name); } else if (pcmk__str_eq(options.attr_set_type, PCMK_XE_META_ATTRIBUTES, pcmk__str_none)) { params = pcmk__strkey_table(free, free); get_meta_attributes(params, rsc, current, scheduler); value = g_hash_table_lookup(params, options.prop_name); } else if (pcmk__str_eq(options.attr_set_type, ATTR_SET_ELEMENT, pcmk__str_none)) { value = crm_element_value(rsc->xml, options.prop_name); free_params = false; } else { params = pcmk__strkey_table(free, free); pe__unpack_dataset_nvpairs(rsc->xml, PCMK_XE_UTILIZATION, NULL, params, NULL, FALSE, scheduler); value = g_hash_table_lookup(params, options.prop_name); } rc = out->message(out, "attribute-list", rsc, options.prop_name, value); if (free_params) { g_hash_table_destroy(params); } break; } case cmd_set_param: if (pcmk__str_empty(options.prop_value)) { exit_code = CRM_EX_USAGE; g_set_error(&error, PCMK__EXITC_ERROR, exit_code, _("You need to supply a value with the -v option")); goto done; } /* coverity[var_deref_model] False positive */ rc = cli_resource_update_attribute(rsc, options.rsc_id, options.prop_set, options.attr_set_type, options.prop_id, options.prop_name, options.prop_value, options.recursive, cib_conn, options.cib_options, options.force); break; case cmd_delete_param: /* coverity[var_deref_model] False positive */ rc = cli_resource_delete_attribute(rsc, options.rsc_id, options.prop_set, options.attr_set_type, options.prop_id, options.prop_name, cib_conn, options.cib_options, options.force); break; case cmd_cleanup: if (rsc == NULL) { rc = cli_cleanup_all(controld_api, options.host_uname, options.operation, options.interval_spec, scheduler); if (rc == pcmk_rc_ok) { start_mainloop(controld_api); } } else { cleanup(out, rsc, node); } break; case cmd_refresh: if (rsc == NULL) { rc = refresh(out); } else { refresh_resource(out, rsc, node); } break; case cmd_delete: /* rsc_id was already checked for NULL much earlier when validating * command line arguments. */ if (options.rsc_type == NULL) { // @COMPAT @TODO change this to exit_code = CRM_EX_USAGE rc = ENXIO; g_set_error(&error, PCMK__RC_ERROR, rc, _("You need to specify a resource type with -t")); } else { rc = pcmk__resource_delete(cib_conn, options.cib_options, options.rsc_id, options.rsc_type); if (rc != pcmk_rc_ok) { g_set_error(&error, PCMK__RC_ERROR, rc, _("Could not delete resource %s: %s"), options.rsc_id, pcmk_rc_str(rc)); } } break; default: exit_code = CRM_EX_USAGE; g_set_error(&error, PCMK__EXITC_ERROR, exit_code, _("Unimplemented command: %d"), (int) options.rsc_cmd); goto done; } /* Convert rc into an exit code. */ if (rc != pcmk_rc_ok && rc != pcmk_rc_no_output) { exit_code = pcmk_rc2exitc(rc); } /* * Clean up and exit */ done: /* When we get here, exit_code has been set one of two ways - either at one of * the spots where there's a "goto done" (which itself could have happened either * directly or by calling pcmk_rc2exitc), or just up above after any of the break * statements. * * Thus, we can use just exit_code here to decide what to do. */ if (exit_code != CRM_EX_OK && exit_code != CRM_EX_USAGE) { if (error != NULL) { char *msg = crm_strdup_printf("%s\nError performing operation: %s", error->message, crm_exit_str(exit_code)); g_clear_error(&error); g_set_error(&error, PCMK__EXITC_ERROR, exit_code, "%s", msg); free(msg); } else { g_set_error(&error, PCMK__EXITC_ERROR, exit_code, _("Error performing operation: %s"), crm_exit_str(exit_code)); } } g_free(options.host_uname); g_free(options.interval_spec); g_free(options.move_lifetime); g_free(options.operation); g_free(options.prop_id); free(options.prop_name); g_free(options.prop_set); g_free(options.prop_value); g_free(options.rsc_id); g_free(options.rsc_type); free(options.agent_spec); free(options.v_agent); free(options.v_class); free(options.v_provider); g_free(options.xml_file); g_strfreev(options.remainder); if (options.override_params != NULL) { g_hash_table_destroy(options.override_params); } /* options.cmdline_params does not need to be destroyed here. See the * comments in cli_resource_execute_from_params. */ g_strfreev(processed_args); g_option_context_free(context); return bye(exit_code); } diff --git a/tools/stonith_admin.c b/tools/stonith_admin.c index 01f72d59d7..ca2178aa72 100644 --- a/tools/stonith_admin.c +++ b/tools/stonith_admin.c @@ -1,678 +1,686 @@ /* * Copyright 2009-2023 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 #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #define SUMMARY "stonith_admin - Access the Pacemaker fencing API" char action = 0; struct { gboolean as_nodeid; gboolean broadcast; gboolean cleanup; gboolean installed; gboolean metadata; gboolean registered; gboolean validate_cfg; stonith_key_value_t *devices; stonith_key_value_t *params; int fence_level; int timeout ; - int tolerance; + long long tolerance_ms; int delay; char *agent; char *confirm_host; char *fence_host; char *history; char *last_fenced; char *query; char *reboot_host; char *register_dev; char *register_level; char *targets; char *terminate; char *unfence_host; char *unregister_dev; char *unregister_level; } options = { .timeout = 120, .delay = 0 }; gboolean add_env_params(const gchar *option_name, const gchar *optarg, gpointer data, GError **error); gboolean add_stonith_device(const gchar *option_name, const gchar *optarg, gpointer data, GError **error); gboolean add_stonith_params(const gchar *option_name, const gchar *optarg, gpointer data, GError **error); gboolean add_tolerance(const gchar *option_name, const gchar *optarg, gpointer data, GError **error); gboolean set_tag(const gchar *option_name, const gchar *optarg, gpointer data, GError **error); #define INDENT " " /* *INDENT-OFF* */ static GOptionEntry defn_entries[] = { { "register", 'R', 0, G_OPTION_ARG_STRING, &options.register_dev, "Register the named stonith device. Requires: --agent.\n" INDENT "Optional: --option, --env-option.", "DEVICE" }, { "deregister", 'D', 0, G_OPTION_ARG_STRING, &options.unregister_dev, "De-register the named stonith device.", "DEVICE" }, { "register-level", 'r', 0, G_OPTION_ARG_STRING, &options.register_level, "Register a stonith level for the named target,\n" INDENT "specified as one of NAME, @PATTERN, or ATTR=VALUE.\n" INDENT "Requires: --index and one or more --device entries.", "TARGET" }, { "deregister-level", 'd', 0, G_OPTION_ARG_STRING, &options.unregister_level, "Unregister a stonith level for the named target,\n" INDENT "specified as for --register-level. Requires: --index", "TARGET" }, { NULL } }; static GOptionEntry query_entries[] = { { "list", 'l', 0, G_OPTION_ARG_STRING, &options.terminate, "List devices that can terminate the specified host.\n" INDENT "Optional: --timeout", "HOST" }, { "list-registered", 'L', 0, G_OPTION_ARG_NONE, &options.registered, "List all registered devices. Optional: --timeout.", NULL }, { "list-installed", 'I', 0, G_OPTION_ARG_NONE, &options.installed, "List all installed devices. Optional: --timeout.", NULL }, { "list-targets", 's', 0, G_OPTION_ARG_STRING, &options.targets, "List the targets that can be fenced by the\n" INDENT "named device. Optional: --timeout.", "DEVICE" }, { "metadata", 'M', 0, G_OPTION_ARG_NONE, &options.metadata, "Show agent metadata. Requires: --agent.\n" INDENT "Optional: --timeout.", NULL }, { "query", 'Q', 0, G_OPTION_ARG_STRING, &options.query, "Check the named device's status. Optional: --timeout.", "DEVICE" }, { "history", 'H', 0, G_OPTION_ARG_STRING, &options.history, "Show last successful fencing operation for named node\n" INDENT "(or '*' for all nodes). Optional: --timeout, --cleanup,\n" INDENT "--quiet (show only the operation's epoch timestamp),\n" INDENT "--verbose (show all recorded and pending operations),\n" INDENT "--broadcast (update history from all nodes available).", "NODE" }, { "last", 'h', 0, G_OPTION_ARG_STRING, &options.last_fenced, "Indicate when the named node was last fenced.\n" INDENT "Optional: --as-node-id.", "NODE" }, { "validate", 'K', 0, G_OPTION_ARG_NONE, &options.validate_cfg, "Validate a fence device configuration.\n" INDENT "Requires: --agent. Optional: --option, --env-option,\n" INDENT "--quiet (print no output, only return status).", NULL }, { NULL } }; static GOptionEntry fence_entries[] = { { "fence", 'F', 0, G_OPTION_ARG_STRING, &options.fence_host, "Fence named host. Optional: --timeout, --tolerance, --delay.", "HOST" }, { "unfence", 'U', 0, G_OPTION_ARG_STRING, &options.unfence_host, "Unfence named host. Optional: --timeout, --tolerance, --delay.", "HOST" }, { "reboot", 'B', 0, G_OPTION_ARG_STRING, &options.reboot_host, "Reboot named host. Optional: --timeout, --tolerance, --delay.", "HOST" }, { "confirm", 'C', 0, G_OPTION_ARG_STRING, &options.confirm_host, "Tell cluster that named host is now safely down.", "HOST", }, { NULL } }; static GOptionEntry addl_entries[] = { { "cleanup", 'c', 0, G_OPTION_ARG_NONE, &options.cleanup, "Cleanup wherever appropriate. Requires --history.", NULL }, { "broadcast", 'b', 0, G_OPTION_ARG_NONE, &options.broadcast, "Broadcast wherever appropriate.", NULL }, { "agent", 'a', 0, G_OPTION_ARG_STRING, &options.agent, "The agent to use (for example, fence_xvm;\n" INDENT "with --register, --metadata, --validate).", "AGENT" }, { "option", 'o', 0, G_OPTION_ARG_CALLBACK, add_stonith_params, "Specify a device configuration parameter as NAME=VALUE\n" INDENT "(may be specified multiple times; with --register,\n" INDENT "--validate).", "PARAM" }, { "env-option", 'e', 0, G_OPTION_ARG_CALLBACK, add_env_params, "Specify a device configuration parameter with the\n" INDENT "specified name, using the value of the\n" INDENT "environment variable of the same name prefixed with\n" INDENT "OCF_RESKEY_ (may be specified multiple times;\n" INDENT "with --register, --validate).", "PARAM" }, { "tag", 'T', 0, G_OPTION_ARG_CALLBACK, set_tag, "Identify fencing operations in logs with the specified\n" INDENT "tag; useful when multiple entities might invoke\n" INDENT "stonith_admin (used with most commands).", "TAG" }, { "device", 'v', 0, G_OPTION_ARG_CALLBACK, add_stonith_device, "Device ID (with --register-level, device to associate with\n" INDENT "a given host and level; may be specified multiple times)" #if SUPPORT_CIBSECRETS "\n" INDENT "(with --validate, name to use to load CIB secrets)" #endif ".", "DEVICE" }, { "index", 'i', 0, G_OPTION_ARG_INT, &options.fence_level, "The stonith level (1-9) (with --register-level,\n" INDENT "--deregister-level).", "LEVEL" }, { "timeout", 't', 0, G_OPTION_ARG_INT, &options.timeout, "Operation timeout in seconds (default 120;\n" INDENT "used with most commands).", "SECONDS" }, { "delay", 'y', 0, G_OPTION_ARG_INT, &options.delay, "Apply a fencing delay in seconds. Any static/random delays from\n" INDENT "pcmk_delay_base/max will be added, otherwise all\n" INDENT "disabled with the value -1\n" INDENT "(default 0; with --fence, --reboot, --unfence).", "SECONDS" }, { "as-node-id", 'n', 0, G_OPTION_ARG_NONE, &options.as_nodeid, "(Advanced) The supplied node is the corosync node ID\n" INDENT "(with --last).", NULL }, { "tolerance", 0, 0, G_OPTION_ARG_CALLBACK, add_tolerance, "(Advanced) Do nothing if an equivalent --fence request\n" INDENT "succeeded less than this many seconds earlier\n" INDENT "(with --fence, --unfence, --reboot).", "SECONDS" }, { NULL } }; /* *INDENT-ON* */ static pcmk__supported_format_t formats[] = { PCMK__SUPPORTED_FORMAT_HTML, PCMK__SUPPORTED_FORMAT_NONE, PCMK__SUPPORTED_FORMAT_TEXT, PCMK__SUPPORTED_FORMAT_XML, { NULL, NULL, NULL } }; static const int st_opts = st_opt_sync_call | st_opt_allow_suicide; static char *name = NULL; gboolean add_env_params(const gchar *option_name, const gchar *optarg, gpointer data, GError **error) { char *key = crm_strdup_printf("OCF_RESKEY_%s", optarg); const char *env = getenv(key); gboolean retval = TRUE; if (env == NULL) { g_set_error(error, PCMK__EXITC_ERROR, CRM_EX_INVALID_PARAM, "Invalid option: -e %s", optarg); retval = FALSE; } else { crm_info("Got: '%s'='%s'", optarg, env); options.params = stonith_key_value_add(options.params, optarg, env); } free(key); return retval; } gboolean add_stonith_device(const gchar *option_name, const gchar *optarg, gpointer data, GError **error) { options.devices = stonith_key_value_add(options.devices, NULL, optarg); return TRUE; } gboolean add_tolerance(const gchar *option_name, const gchar *optarg, gpointer data, GError **error) { - options.tolerance = crm_get_msec(optarg) / 1000; + // pcmk__request_fencing() expects an unsigned int + options.tolerance_ms = crm_get_msec(optarg); + + if (options.tolerance_ms < 0) { + crm_warn("Ignoring invalid tolerance '%s'", optarg); + options.tolerance_ms = 0; + } else { + options.tolerance_ms = QB_MIN(options.tolerance_ms, UINT_MAX); + } return TRUE; } gboolean add_stonith_params(const gchar *option_name, const gchar *optarg, gpointer data, GError **error) { char *name = NULL; char *value = NULL; int rc = 0; gboolean retval = TRUE; crm_info("Scanning: -o %s", optarg); rc = pcmk__scan_nvpair(optarg, &name, &value); if (rc != 2) { rc = pcmk_legacy2rc(rc); g_set_error(error, PCMK__RC_ERROR, rc, "Invalid option: -o %s: %s", optarg, pcmk_rc_str(rc)); retval = FALSE; } else { crm_info("Got: '%s'='%s'", name, value); options.params = stonith_key_value_add(options.params, name, value); } free(name); free(value); return retval; } gboolean set_tag(const gchar *option_name, const gchar *optarg, gpointer data, GError **error) { free(name); name = crm_strdup_printf("%s.%s", crm_system_name, optarg); return TRUE; } static GOptionContext * build_arg_context(pcmk__common_args_t *args, GOptionGroup **group) { GOptionContext *context = NULL; GOptionEntry extra_prog_entries[] = { { "quiet", 'q', 0, G_OPTION_ARG_NONE, &(args->quiet), "Be less descriptive in output.", NULL }, { NULL } }; context = pcmk__build_arg_context(args, "text (default), html, xml", group, NULL); /* Add the -q option, which cannot be part of the globally supported options * because some tools use that flag for something else. */ pcmk__add_main_args(context, extra_prog_entries); pcmk__add_arg_group(context, "definition", "Device Definition Commands:", "Show device definition help", defn_entries); pcmk__add_arg_group(context, "queries", "Queries:", "Show query help", query_entries); pcmk__add_arg_group(context, "fence", "Fencing Commands:", "Show fence help", fence_entries); pcmk__add_arg_group(context, "additional", "Additional Options:", "Show additional options", addl_entries); return context; } // \return Standard Pacemaker return code static int request_fencing(stonith_t *st, const char *target, const char *command, GError **error) { char *reason = NULL; int rc = pcmk__request_fencing(st, target, command, name, options.timeout * 1000, - options.tolerance * 1000, - options.delay, &reason); + options.tolerance_ms, options.delay, + &reason); if (rc != pcmk_rc_ok) { const char *rc_str = pcmk_rc_str(rc); const char *what = "fence"; if (strcmp(command, PCMK_ACTION_ON) == 0) { what = "unfence"; } // If reason is identical to return code string, don't display it twice if (pcmk__str_eq(rc_str, reason, pcmk__str_none)) { free(reason); reason = NULL; } g_set_error(error, PCMK__RC_ERROR, rc, "Couldn't %s %s: %s%s%s%s", what, target, rc_str, ((reason == NULL)? "" : " ("), ((reason == NULL)? "" : reason), ((reason == NULL)? "" : ")")); } free(reason); return rc; } int main(int argc, char **argv) { int rc = 0; crm_exit_t exit_code = CRM_EX_OK; bool no_connect = false; bool required_agent = false; char *target = NULL; const char *device = NULL; stonith_t *st = NULL; GError *error = NULL; pcmk__output_t *out = NULL; GOptionGroup *output_group = NULL; pcmk__common_args_t *args = pcmk__new_common_args(SUMMARY); gchar **processed_args = pcmk__cmdline_preproc(argv, "adehilorstvyBCDFHQRTU"); GOptionContext *context = build_arg_context(args, &output_group); pcmk__register_formats(output_group, formats); if (!g_option_context_parse_strv(context, &processed_args, &error)) { exit_code = CRM_EX_USAGE; goto done; } pcmk__cli_init_logging("stonith_admin", args->verbosity); if (name == NULL) { name = strdup(crm_system_name); } rc = pcmk__output_new(&out, args->output_ty, args->output_dest, argv); if (rc != pcmk_rc_ok) { exit_code = CRM_EX_ERROR; g_set_error(&error, PCMK__EXITC_ERROR, exit_code, "Error creating output format %s: %s", args->output_ty, pcmk_rc_str(rc)); goto done; } stonith__register_messages(out); if (args->version) { out->version(out, false); goto done; } if (options.validate_cfg) { required_agent = true; no_connect = true; action = 'K'; } if (options.installed) { no_connect = true; action = 'I'; } if (options.registered) { action = 'L'; } if (options.register_dev != NULL) { required_agent = true; action = 'R'; device = options.register_dev; } if (options.query != NULL) { action = 'Q'; device = options.query; } if (options.unregister_dev != NULL) { action = 'D'; device = options.unregister_dev; } if (options.targets != NULL) { action = 's'; device = options.targets; } if (options.terminate != NULL) { action = 'L'; target = options.terminate; } if (options.metadata) { no_connect = true; required_agent = true; action = 'M'; } if (options.reboot_host != NULL) { no_connect = true; action = 'B'; target = options.reboot_host; crm_log_args(argc, argv); } if (options.fence_host != NULL) { no_connect = true; action = 'F'; target = options.fence_host; crm_log_args(argc, argv); } if (options.unfence_host != NULL) { no_connect = true; action = 'U'; target = options.unfence_host; crm_log_args(argc, argv); } if (options.confirm_host != NULL) { action = 'C'; target = options.confirm_host; crm_log_args(argc, argv); } if (options.last_fenced != NULL) { action = 'h'; target = options.last_fenced; } if (options.history != NULL) { action = 'H'; target = options.history; } if (options.register_level != NULL) { action = 'r'; target = options.register_level; } if (options.unregister_level != NULL) { action = 'd'; target = options.unregister_level; } if (action == 0) { char *help = g_option_context_get_help(context, TRUE, NULL); out->err(out, "%s", help); g_free(help); exit_code = CRM_EX_USAGE; goto done; } if (required_agent && options.agent == NULL) { char *help = g_option_context_get_help(context, TRUE, NULL); out->err(out, "Please specify an agent to query using -a,--agent [value]"); out->err(out, "%s", help); g_free(help); exit_code = CRM_EX_USAGE; goto done; } out->quiet = args->quiet; st = stonith_api_new(); if (st == NULL) { rc = -ENOMEM; } else if (!no_connect) { rc = st->cmds->connect(st, name, NULL); } if (rc < 0) { out->err(out, "Could not connect to fencer: %s", pcmk_strerror(rc)); exit_code = CRM_EX_DISCONNECT; goto done; } switch (action) { case 'I': rc = pcmk__fence_installed(out, st, options.timeout*1000); if (rc != pcmk_rc_ok) { out->err(out, "Failed to list installed devices: %s", pcmk_rc_str(rc)); } break; case 'L': rc = pcmk__fence_registered(out, st, target, options.timeout*1000); if (rc != pcmk_rc_ok) { out->err(out, "Failed to list registered devices: %s", pcmk_rc_str(rc)); } break; case 'Q': rc = st->cmds->monitor(st, st_opts, device, options.timeout); if (rc != pcmk_rc_ok) { rc = st->cmds->list(st, st_opts, device, NULL, options.timeout); } rc = pcmk_legacy2rc(rc); break; case 's': rc = pcmk__fence_list_targets(out, st, device, options.timeout*1000); if (rc != pcmk_rc_ok) { out->err(out, "Couldn't list targets: %s", pcmk_rc_str(rc)); } break; case 'R': rc = st->cmds->register_device(st, st_opts, device, NULL, options.agent, options.params); rc = pcmk_legacy2rc(rc); if (rc != pcmk_rc_ok) { out->err(out, "Can't register device %s using agent %s: %s", device, options.agent, pcmk_rc_str(rc)); } break; case 'D': rc = st->cmds->remove_device(st, st_opts, device); rc = pcmk_legacy2rc(rc); if (rc != pcmk_rc_ok) { out->err(out, "Can't unregister device %s: %s", device, pcmk_rc_str(rc)); } break; case 'd': rc = pcmk__fence_unregister_level(st, target, options.fence_level); if (rc != pcmk_rc_ok) { out->err(out, "Can't unregister topology level %d for %s: %s", options.fence_level, target, pcmk_rc_str(rc)); } break; case 'r': rc = pcmk__fence_register_level(st, target, options.fence_level, options.devices); if (rc != pcmk_rc_ok) { out->err(out, "Can't register topology level %d for %s: %s", options.fence_level, target, pcmk_rc_str(rc)); } break; case 'M': rc = pcmk__fence_metadata(out, st, options.agent, options.timeout*1000); if (rc != pcmk_rc_ok) { out->err(out, "Can't get fence agent meta-data: %s", pcmk_rc_str(rc)); } break; case 'C': rc = st->cmds->confirm(st, st_opts, target); rc = pcmk_legacy2rc(rc); break; case 'B': rc = request_fencing(st, target, PCMK_ACTION_REBOOT, &error); break; case 'F': rc = request_fencing(st, target, PCMK_ACTION_OFF, &error); break; case 'U': rc = request_fencing(st, target, PCMK_ACTION_ON, &error); break; case 'h': rc = pcmk__fence_last(out, target, options.as_nodeid); break; case 'H': rc = pcmk__fence_history(out, st, target, options.timeout*1000, args->verbosity, options.broadcast, options.cleanup); break; case 'K': device = options.devices ? options.devices->key : NULL; rc = pcmk__fence_validate(out, st, options.agent, device, options.params, options.timeout*1000); break; } crm_info("Command returned: %s (%d)", pcmk_rc_str(rc), rc); exit_code = pcmk_rc2exitc(rc); done: g_strfreev(processed_args); pcmk__free_arg_context(context); pcmk__output_and_clear_error(&error, out); if (out != NULL) { out->finish(out, exit_code, true, NULL); pcmk__output_free(out); } pcmk__unregister_formats(); free(name); stonith_key_value_freeall(options.params, 1, 1); if (st != NULL) { st->cmds->disconnect(st); stonith_api_delete(st); } return exit_code; }