diff --git a/daemons/controld/controld_execd.c b/daemons/controld/controld_execd.c index b4ba16a534..4fc2b8861a 100644 --- a/daemons/controld/controld_execd.c +++ b/daemons/controld/controld_execd.c @@ -1,2893 +1,2900 @@ /* * Copyright 2004-2021 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(lrm_state_t * lrm_state, 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, const char *operation, xmlNode *msg); static gboolean lrm_state_verify_stopped(lrm_state_t * lrm_state, enum crmd_fsa_state cur_state, int log_level); static int do_update_resource(const char *node_name, lrmd_rsc_info_t *rsc, lrmd_event_data_t *op, time_t lock_time); static void lrm_connection_destroy(void) { if (pcmk_is_set(fsa_input_register, R_LRM_CONNECTED)) { crm_crit("Connection to executor failed"); register_fsa_input(C_FSA_INTERNAL, I_ERROR, NULL); controld_clear_fsa_input_flags(R_LRM_CONNECTED); } else { crm_info("Disconnected from executor"); } } static char * make_stop_id(const char *rsc, int call_id) { return crm_strdup_printf("%s:%d", rsc, call_id); } static void copy_instance_keys(gpointer key, gpointer value, gpointer user_data) { if (strstr(key, CRM_META "_") == NULL) { g_hash_table_replace(user_data, strdup((const char *)key), strdup((const char *)value)); } } static void copy_meta_keys(gpointer key, gpointer value, gpointer user_data) { if (strstr(key, CRM_META "_") != NULL) { g_hash_table_replace(user_data, strdup((const char *)key), strdup((const char *)value)); } } /*! * \internal * \brief Remove a recurring operation from a resource's history * * \param[in,out] history Resource history to modify * \param[in] op Operation to remove * * \return TRUE if the operation was found and removed, FALSE otherwise */ static gboolean history_remove_recurring_op(rsc_history_t *history, const lrmd_event_data_t *op) { GList *iter; for (iter = history->recurring_op_list; iter != NULL; iter = iter->next) { lrmd_event_data_t *existing = iter->data; if ((op->interval_ms == existing->interval_ms) && pcmk__str_eq(op->rsc_id, existing->rsc_id, pcmk__str_none) && pcmk__str_eq(op->op_type, existing->op_type, pcmk__str_casei)) { history->recurring_op_list = g_list_delete_link(history->recurring_op_list, iter); lrmd_free_event(existing); return TRUE; } } return FALSE; } /*! * \internal * \brief Free all recurring operations in resource history * * \param[in,out] history Resource history to modify */ static void history_free_recurring_ops(rsc_history_t *history) { GList *iter; for (iter = history->recurring_op_list; iter != NULL; iter = iter->next) { lrmd_free_event(iter->data); } g_list_free(history->recurring_op_list); history->recurring_op_list = NULL; } /*! * \internal * \brief Free resource history * * \param[in,out] history Resource history to free */ void history_free(gpointer data) { rsc_history_t *history = (rsc_history_t*)data; if (history->stop_params) { g_hash_table_destroy(history->stop_params); } /* Don't need to free history->rsc.id because it's set to history->id */ free(history->rsc.type); free(history->rsc.standard); free(history->rsc.provider); lrmd_free_event(history->failed); lrmd_free_event(history->last); free(history->id); history_free_recurring_ops(history); free(history); } static void update_history_cache(lrm_state_t * lrm_state, lrmd_rsc_info_t * rsc, lrmd_event_data_t * op) { int target_rc = 0; rsc_history_t *entry = NULL; if (op->rsc_deleted) { crm_debug("Purged history for '%s' after %s", op->rsc_id, op->op_type); controld_delete_resource_history(op->rsc_id, lrm_state->node_name, NULL, crmd_cib_smart_opt()); return; } if (pcmk__str_eq(op->op_type, RSC_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); if (rsc->provider) { entry->rsc.provider = strdup(rsc->provider); } else { entry->rsc.provider = NULL; } } 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, CRMD_ACTION_START, CRMD_ACTION_RELOAD, CRMD_ACTION_RELOAD_AGENT, CRMD_ACTION_STATUS, 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 && !pcmk__str_eq(op->op_type, RSC_STATUS, 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(lrm_state_t *lrm_state, ha_msg_input_t *input, const char *rsc_id, 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 op->remote_nodename? op->remote_nodename : fsa_our_uname; } 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_timer(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(fsa_our_uname == NULL) { return; /* Nothing to do */ } lrm_state = lrm_state_find_or_create(fsa_our_uname); if (lrm_state == NULL) { register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); return; } if (action & A_LRM_DISCONNECT) { if (lrm_state_verify_stopped(lrm_state, cur_state, LOG_INFO) == FALSE) { if (action == A_LRM_DISCONNECT) { crmd_fsa_stall(FALSE); return; } } controld_clear_fsa_input_flags(R_LRM_CONNECTED); crm_info("Disconnecting from the executor"); lrm_state_disconnect(lrm_state); lrm_state_reset_tables(lrm_state, FALSE); crm_notice("Disconnected from the executor"); } if (action & A_LRM_CONNECT) { try_local_executor_connect(action, msg_data, lrm_state); } if (action & ~(A_LRM_CONNECT | A_LRM_DISCONNECT)) { crm_err("Unexpected action %s in %s", fsa_action2string(action), __func__); } } static gboolean lrm_state_verify_stopped(lrm_state_t * lrm_state, enum crmd_fsa_state cur_state, int log_level) { int counter = 0; gboolean rc = TRUE; const char *when = "lrm disconnect"; GHashTableIter gIter; const char *key = NULL; rsc_history_t *entry = NULL; active_op_t *pending = NULL; crm_debug("Checking for active resources before exit"); if (cur_state == S_TERMINATE) { log_level = LOG_ERR; when = "shutdown"; } else if (pcmk_is_set(fsa_input_register, R_SHUTDOWN)) { when = "shutdown... waiting"; } if (lrm_state->pending_ops && lrm_state_is_connected(lrm_state) == TRUE) { guint removed = g_hash_table_foreach_remove( lrm_state->pending_ops, stop_recurring_actions, lrm_state); guint nremaining = g_hash_table_size(lrm_state->pending_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->pending_ops) { g_hash_table_iter_init(&gIter, lrm_state->pending_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(fsa_input_register, R_SENT_RSC_STOP)) { g_hash_table_iter_init(&gIter, lrm_state->pending_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(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->pending_ops) { GHashTableIter hIter; g_hash_table_iter_init(&hIter, lrm_state->pending_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; } /*! * \internal * \brief Build XML and string of parameters meeting some criteria, for digest * * \param[in] op Executor event with parameter table to use * \param[in] metadata Parsed meta-data for executed resource agent * \param[in] param_type Flag used for selection criteria * \param[out] result Will be set to newly created XML with selected * parameters as attributes * * \return Newly allocated space-separated string of parameter names * \note Selection criteria varies by param_type: for the restart digest, we * want parameters that are *not* marked reloadable (OCF 1.1) or that * *are* marked unique (pre-1.1), for both string and XML results; for the * secure digest, we want parameters that *are* marked private for the * string, but parameters that are *not* marked private for the XML. * \note It is the caller's responsibility to free the string return value with * free() and the XML result with free_xml(). */ static char * build_parameter_list(const lrmd_event_data_t *op, const struct ra_metadata_s *metadata, enum ra_param_flags_e param_type, xmlNode **result) { char *list = NULL; size_t len = 0; *result = create_xml_node(NULL, XML_TAG_PARAMS); for (GList *iter = metadata->ra_params; iter != NULL; iter = iter->next) { struct ra_param_s *param = (struct ra_param_s *) iter->data; bool accept_for_list = false; bool accept_for_xml = false; switch (param_type) { case ra_param_reloadable: accept_for_list = !pcmk_is_set(param->rap_flags, param_type); accept_for_xml = accept_for_list; break; case ra_param_unique: accept_for_list = pcmk_is_set(param->rap_flags, param_type); accept_for_xml = accept_for_list; break; case ra_param_private: accept_for_list = pcmk_is_set(param->rap_flags, param_type); accept_for_xml = !accept_for_list; break; } if (accept_for_list) { crm_trace("Attr %s is %s", param->rap_name, ra_param_flag2text(param_type)); if (list == NULL) { // We will later search for " WORD ", so start list with a space pcmk__add_word(&list, &len, " "); } pcmk__add_word(&list, &len, param->rap_name); } else { crm_trace("Rejecting %s for %s", param->rap_name, ra_param_flag2text(param_type)); } if (accept_for_xml) { const char *v = g_hash_table_lookup(op->params, param->rap_name); if (v != NULL) { crm_trace("Adding attr %s=%s to the xml result", param->rap_name, v); crm_xml_add(*result, param->rap_name, v); } } } if (list != NULL) { // We will later search for " WORD ", so end list with a space pcmk__add_word(&list, &len, " "); } return list; } static void append_restart_list(lrmd_event_data_t *op, struct ra_metadata_s *metadata, xmlNode *update, const char *version) { char *list = NULL; char *digest = NULL; xmlNode *restart = NULL; CRM_LOG_ASSERT(op->params != NULL); if (op->interval_ms > 0) { /* monitors are not reloadable */ return; } if (pcmk_is_set(metadata->ra_flags, ra_supports_reload_agent)) { // Add parameters not marked reloadable to the "op-force-restart" list list = build_parameter_list(op, metadata, ra_param_reloadable, &restart); } else if (pcmk_is_set(metadata->ra_flags, ra_supports_legacy_reload)) { /* @COMPAT pre-OCF-1.1 resource agents * * Before OCF 1.1, Pacemaker abused "unique=0" to indicate * reloadability. Add any parameters with unique="1" to the * "op-force-restart" list. */ list = build_parameter_list(op, metadata, ra_param_unique, &restart); } else { // Resource does not support agent reloads return; } digest = calculate_operation_digest(restart, version); /* Add "op-force-restart" and "op-restart-digest" to indicate the resource supports reload, * no matter if it actually supports any parameters with unique="1"). */ crm_xml_add(update, XML_LRM_ATTR_OP_RESTART, list? list: ""); crm_xml_add(update, XML_LRM_ATTR_RESTART_DIGEST, digest); crm_trace("%s: %s, %s", op->rsc_id, digest, list); crm_log_xml_trace(restart, "restart digest source"); free_xml(restart); free(digest); free(list); } static void append_secure_list(lrmd_event_data_t *op, struct ra_metadata_s *metadata, xmlNode *update, const char *version) { char *list = NULL; char *digest = NULL; xmlNode *secure = NULL; CRM_LOG_ASSERT(op->params != NULL); /* * To keep XML_LRM_ATTR_OP_SECURE short, we want it to contain the * secure parameters but XML_LRM_ATTR_SECURE_DIGEST to be based on * the insecure ones */ list = build_parameter_list(op, metadata, ra_param_private, &secure); if (list != NULL) { digest = calculate_operation_digest(secure, version); crm_xml_add(update, XML_LRM_ATTR_OP_SECURE, list); crm_xml_add(update, XML_LRM_ATTR_SECURE_DIGEST, digest); crm_trace("%s: %s, %s", op->rsc_id, digest, list); crm_log_xml_trace(secure, "secure digest source"); } else { crm_trace("%s: no secure parameters", op->rsc_id); } free_xml(secure); free(digest); free(list); } static gboolean build_operation_update(xmlNode * parent, lrmd_rsc_info_t * rsc, lrmd_event_data_t * op, const char *node_name, const char *src) { int target_rc = 0; xmlNode *xml_op = NULL; struct ra_metadata_s *metadata = NULL; const char *caller_version = NULL; lrm_state_t *lrm_state = NULL; uint32_t metadata_source = controld_metadata_from_agent; if (op == NULL) { return FALSE; } target_rc = rsc_op_expected_rc(op); /* there is a small risk in formerly mixed clusters that it will * be sub-optimal. * * however with our upgrade policy, the update we send should * still be completely supported anyway */ caller_version = g_hash_table_lookup(op->params, XML_ATTR_CRM_VERSION); CRM_LOG_ASSERT(caller_version != NULL); if(caller_version == NULL) { caller_version = CRM_FEATURE_SET; } crm_trace("Building %s operation update with originator version: %s", op->rsc_id, caller_version); xml_op = pcmk__create_history_xml(parent, op, caller_version, target_rc, fsa_our_uname, src, LOG_DEBUG); if (xml_op == NULL) { return TRUE; } if ((rsc == NULL) || (op->params == NULL) || !crm_op_needs_metadata(rsc->standard, op->op_type)) { crm_trace("No digests needed for %s action on %s (params=%p rsc=%p)", op->op_type, op->rsc_id, op->params, rsc); return TRUE; } lrm_state = lrm_state_find(node_name); if (lrm_state == NULL) { crm_warn("Cannot calculate digests for operation " PCMK__OP_FMT " because we have no connection to executor for %s", op->rsc_id, op->op_type, op->interval_ms, node_name); return TRUE; } /* Getting meta-data from cache is OK unless this is a successful start * action -- always refresh from the agent for those, in case the * resource agent was updated. * * @TODO Only refresh the meta-data after starts if the agent actually * changed (using something like inotify, or a hash or modification time of * the agent executable). */ if ((op->op_status != PCMK_EXEC_DONE) || (op->rc != target_rc) || !pcmk__str_eq(op->op_type, CRMD_ACTION_START, pcmk__str_none)) { metadata_source |= controld_metadata_from_cache; } metadata = controld_get_rsc_metadata(lrm_state, rsc, metadata_source); if (metadata == NULL) { return TRUE; } #if ENABLE_VERSIONED_ATTRS crm_xml_add(xml_op, XML_ATTR_RA_VERSION, metadata->ra_version); #endif crm_trace("Including additional digests for %s:%s:%s", rsc->standard, rsc->provider, rsc->type); append_restart_list(op, metadata, xml_op, caller_version); append_secure_list(op, metadata, xml_op, caller_version); return TRUE; } 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, CRMD_ACTION_STOP, pcmk__str_casei)) { return FALSE; } else if (entry->last->rc == PCMK_OCF_OK && pcmk__str_eq(entry->last->op_type, CRMD_ACTION_MIGRATE, pcmk__str_casei)) { // A stricter check is too complex ... leave that to the scheduler return FALSE; } else if (entry->last->rc == PCMK_OCF_NOT_RUNNING) { return FALSE; } else if ((entry->last->interval_ms == 0) && (entry->last->rc == PCMK_OCF_NOT_CONFIGURED)) { /* Badly configured resources can't be reliably stopped */ return FALSE; } return TRUE; } static gboolean build_active_RAs(lrm_state_t * lrm_state, xmlNode * rsc_list) { GHashTableIter iter; rsc_history_t *entry = NULL; g_hash_table_iter_init(&iter, lrm_state->resource_history); while (g_hash_table_iter_next(&iter, NULL, (void **)&entry)) { GList *gIter = NULL; xmlNode *xml_rsc = create_xml_node(rsc_list, XML_LRM_TAG_RESOURCE); crm_xml_add(xml_rsc, XML_ATTR_ID, entry->id); crm_xml_add(xml_rsc, XML_ATTR_TYPE, entry->rsc.type); crm_xml_add(xml_rsc, XML_AGENT_ATTR_CLASS, entry->rsc.standard); crm_xml_add(xml_rsc, XML_AGENT_ATTR_PROVIDER, entry->rsc.provider); if (entry->last && entry->last->params) { const char *container = g_hash_table_lookup(entry->last->params, CRM_META"_"XML_RSC_ATTR_CONTAINER); if (container) { crm_trace("Resource %s is a part of container resource %s", entry->id, container); crm_xml_add(xml_rsc, XML_RSC_ATTR_CONTAINER, container); } } build_operation_update(xml_rsc, &(entry->rsc), entry->failed, lrm_state->node_name, __func__); build_operation_update(xml_rsc, &(entry->rsc), entry->last, lrm_state->node_name, __func__); for (gIter = entry->recurring_op_list; gIter != NULL; gIter = gIter->next) { build_operation_update(xml_rsc, &(entry->rsc), gIter->data, lrm_state->node_name, __func__); } } return FALSE; } static xmlNode * do_lrm_query_internal(lrm_state_t *lrm_state, int update_flags) { xmlNode *xml_state = NULL; xmlNode *xml_data = NULL; xmlNode *rsc_list = NULL; crm_node_t *peer = NULL; peer = crm_get_peer_full(0, lrm_state->node_name, CRM_GET_PEER_ANY); CRM_CHECK(peer != NULL, return NULL); xml_state = create_node_state_update(peer, update_flags, NULL, __func__); if (xml_state == NULL) { return NULL; } xml_data = create_xml_node(xml_state, XML_CIB_TAG_LRM); crm_xml_add(xml_data, XML_ATTR_ID, peer->uuid); rsc_list = create_xml_node(xml_data, XML_LRM_TAG_RESOURCES); /* Build a list of active (not always running) resources */ build_active_RAs(lrm_state, rsc_list); crm_log_xml_trace(xml_state, "Current executor state"); return xml_state; } xmlNode * controld_query_executor_state(const char *node_name) { lrm_state_t *lrm_state = lrm_state_find(node_name); if (!lrm_state) { crm_err("Could not find executor state for node %s", node_name); return NULL; } return do_lrm_query_internal(lrm_state, node_update_cluster|node_update_peer); } /*! * \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); update_attr_delegate(fsa_cib_conn, cib_none, XML_CIB_TAG_CRMCONFIG, NULL, NULL, NULL, NULL, "last-lrm-refresh", now_s, FALSE, NULL, NULL); free(now_s); } } static void notify_deleted(lrm_state_t * lrm_state, ha_msg_input_t * input, const char *rsc_id, int rc) { lrmd_event_data_t *op = NULL; const char *from_sys = crm_element_value(input->msg, F_CRM_SYS_FROM); const char *from_host = crm_element_value(input->msg, F_CRM_HOST_FROM); crm_info("Notifying %s on %s that %s was%s deleted", from_sys, (from_host? from_host : "localhost"), rsc_id, ((rc == pcmk_ok)? "" : " not")); op = construct_op(lrm_state, input->xml, rsc_id, CRMD_ACTION_DELETE); controld_rc2event(op, pcmk_legacy2rc(rc)); controld_ack_event_directly(from_host, from_sys, NULL, op, rsc_id); lrmd_free_event(op); controld_trigger_delete_refresh(from_sys, rsc_id); } static gboolean lrm_remove_deleted_rsc(gpointer key, gpointer value, gpointer user_data) { struct delete_event_s *event = user_data; struct pending_deletion_op_s *op = value; if (pcmk__str_eq(event->rsc, op->rsc, pcmk__str_none)) { notify_deleted(event->lrm_state, op->input, event->rsc, event->rc); return TRUE; } return FALSE; } static gboolean lrm_remove_deleted_op(gpointer key, gpointer value, gpointer user_data) { const char *rsc = user_data; active_op_t *pending = value; if (pcmk__str_eq(rsc, pending->rsc_id, pcmk__str_none)) { crm_info("Removing op %s:%d for deleted resource %s", pending->op_key, pending->call_id, rsc); return TRUE; } return FALSE; } static void delete_rsc_entry(lrm_state_t * lrm_state, ha_msg_input_t * input, const char *rsc_id, GHashTableIter * rsc_gIter, int rc, const char *user_name) { struct delete_event_s event; CRM_CHECK(rsc_id != NULL, return); if (rc == pcmk_ok) { char *rsc_id_copy = strdup(rsc_id); if (rsc_gIter) { g_hash_table_iter_remove(rsc_gIter); } else { g_hash_table_remove(lrm_state->resource_history, rsc_id_copy); } controld_delete_resource_history(rsc_id_copy, lrm_state->node_name, user_name, crmd_cib_smart_opt()); g_hash_table_foreach_remove(lrm_state->pending_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); } /*! * \internal * \brief Erase an LRM history entry from the CIB, given the operation data * * \param[in] lrm_state LRM state of the desired node * \param[in] op Operation whose history should be deleted */ static void erase_lrm_history_by_op(lrm_state_t *lrm_state, lrmd_event_data_t *op) { xmlNode *xml_top = NULL; CRM_CHECK(op != NULL, return); xml_top = create_xml_node(NULL, XML_LRM_TAG_RSC_OP); crm_xml_add_int(xml_top, XML_LRM_ATTR_CALLID, op->call_id); crm_xml_add(xml_top, XML_ATTR_TRANSITION_KEY, op->user_data); if (op->interval_ms > 0) { char *op_id = pcmk__op_key(op->rsc_id, op->op_type, op->interval_ms); /* Avoid deleting last_failure too (if it was a result of this recurring op failing) */ crm_xml_add(xml_top, XML_ATTR_ID, op_id); free(op_id); } crm_debug("Erasing resource operation history for " PCMK__OP_FMT " (call=%d)", op->rsc_id, op->op_type, op->interval_ms, op->call_id); fsa_cib_conn->cmds->remove(fsa_cib_conn, XML_CIB_TAG_STATUS, xml_top, cib_quorum_override); crm_log_xml_trace(xml_top, "op:cancel"); free_xml(xml_top); } /* Define xpath to find LRM resource history entry by node and resource */ #define XPATH_HISTORY \ "/" XML_TAG_CIB "/" XML_CIB_TAG_STATUS \ "/" XML_CIB_TAG_STATE "[@" XML_ATTR_UNAME "='%s']" \ "/" XML_CIB_TAG_LRM "/" XML_LRM_TAG_RESOURCES \ "/" XML_LRM_TAG_RESOURCE "[@" XML_ATTR_ID "='%s']" \ "/" XML_LRM_TAG_RSC_OP /* ... and also by operation key */ #define XPATH_HISTORY_ID XPATH_HISTORY \ "[@" XML_ATTR_ID "='%s']" /* ... and also by operation key and operation call ID */ #define XPATH_HISTORY_CALL XPATH_HISTORY \ "[@" XML_ATTR_ID "='%s' and @" XML_LRM_ATTR_CALLID "='%d']" /* ... and also by operation key and original operation key */ #define XPATH_HISTORY_ORIG XPATH_HISTORY \ "[@" XML_ATTR_ID "='%s' and @" XML_LRM_ATTR_TASK_KEY "='%s']" /*! * \internal * \brief Erase an LRM history entry from the CIB, given operation identifiers * * \param[in] lrm_state LRM state of the node to clear history for * \param[in] rsc_id Name of resource to clear history for * \param[in] key Operation key of operation to clear history for * \param[in] orig_op If specified, delete only if it has this original op * \param[in] call_id If specified, delete entry only if it has this call ID */ static void erase_lrm_history_by_id(lrm_state_t *lrm_state, const char *rsc_id, const char *key, const char *orig_op, int call_id) { char *op_xpath = NULL; CRM_CHECK((rsc_id != NULL) && (key != NULL), return); if (call_id > 0) { op_xpath = crm_strdup_printf(XPATH_HISTORY_CALL, lrm_state->node_name, rsc_id, key, call_id); } else if (orig_op) { op_xpath = crm_strdup_printf(XPATH_HISTORY_ORIG, lrm_state->node_name, rsc_id, key, orig_op); } else { op_xpath = crm_strdup_printf(XPATH_HISTORY_ID, lrm_state->node_name, rsc_id, key); } crm_debug("Erasing resource operation history for %s on %s (call=%d)", key, rsc_id, call_id); fsa_cib_conn->cmds->remove(fsa_cib_conn, op_xpath, NULL, cib_quorum_override | cib_xpath); free(op_xpath); } 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) { char *op_key = NULL; char *orig_op_key = NULL; lrm_state_t *lrm_state = NULL; lrm_state = lrm_state_find(node_name); if (lrm_state == NULL) { return; } /* Erase from CIB */ op_key = pcmk__op_key(rsc_id, "last_failure", 0); if (operation) { orig_op_key = pcmk__op_key(rsc_id, operation, interval_ms); } erase_lrm_history_by_id(lrm_state, rsc_id, op_key, orig_op_key, 0); free(op_key); free(orig_op_key); /* Remove from memory */ if (lrm_state->resource_history) { 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->pending_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 pending_ops list * * Usually by returning TRUE inside the worker function * supplied to g_hash_table_foreach_remove() * * Not removing the entry from pending_ops 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->pending_ops, cancel_action_by_key, &data); crm_trace("Removed %u op cache entries, new size: %u", removed, g_hash_table_size(lrm_state->pending_ops)); return data.done; } /*! * \internal * \brief Retrieve resource information from LRM * * \param[in] lrm_state LRM connection to use * \param[in] rsc_xml XML containing resource configuration * \param[in] do_create If true, register resource with LRM if not already * \param[out] rsc_info Where to store resource information obtained from LRM * * \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, xmlNode *rsc_xml, gboolean do_create, lrmd_rsc_info_t **rsc_info) { const char *id = ID(rsc_xml); CRM_CHECK(lrm_state && rsc_xml && rsc_info, return -EINVAL); CRM_CHECK(id, return -EINVAL); if (lrm_state_is_connected(lrm_state) == FALSE) { return -ENOTCONN; } crm_trace("Retrieving resource information for %s from the executor", id); *rsc_info = lrm_state_get_rsc_info(lrm_state, id, 0); // If resource isn't known by ID, try clone name, if provided if (!*rsc_info) { const char *long_id = crm_element_value(rsc_xml, XML_ATTR_ID_LONG); if (long_id) { *rsc_info = lrm_state_get_rsc_info(lrm_state, long_id, 0); } } if ((*rsc_info == NULL) && do_create) { const char *class = crm_element_value(rsc_xml, XML_AGENT_ATTR_CLASS); const char *provider = crm_element_value(rsc_xml, XML_AGENT_ATTR_PROVIDER); const char *type = crm_element_value(rsc_xml, XML_ATTR_TYPE); int rc; crm_trace("Registering resource %s with the executor", id); rc = lrm_state_register_rsc(lrm_state, id, class, provider, type, lrmd_opt_drop_recurring); if (rc != pcmk_ok) { fsa_data_t *msg_data = NULL; crm_err("Could not register resource %s with the executor on %s: %s " CRM_XS " rc=%d", id, lrm_state->node_name, pcmk_strerror(rc), rc); /* Register this as an internal error if this involves the local * executor. Otherwise, we're likely dealing with an unresponsive * remote node, which is not an FSA failure. */ if (lrm_state_is_local(lrm_state) == TRUE) { register_fsa_error(C_FSA_INTERNAL, I_FAIL, NULL); } return rc; } *rsc_info = lrm_state_get_rsc_info(lrm_state, id, 0); } return *rsc_info? pcmk_ok : -ENODEV; } static void delete_resource(lrm_state_t * lrm_state, const char *id, lrmd_rsc_info_t * rsc, GHashTableIter * gIter, const char *sys, const char *user, ha_msg_input_t * request, gboolean unregister) { int rc = pcmk_ok; crm_info("Removing resource %s from executor for %s%s%s", id, sys, (user? " as " : ""), (user? user : "")); if (rsc && unregister) { rc = lrm_state_unregister_rsc(lrm_state, id, 0); } if (rc == pcmk_ok) { crm_trace("Resource %s deleted from executor", id); } else if (rc == -EINPROGRESS) { crm_info("Deletion of resource '%s' from executor is pending", id); if (request) { struct pending_deletion_op_s *op = NULL; char *ref = crm_element_value_copy(request->msg, XML_ATTR_REFERENCE); op = calloc(1, sizeof(struct pending_deletion_op_s)); op->rsc = strdup(rsc->id); op->input = copy_ha_msg_input(request); g_hash_table_insert(lrm_state->deletion_ops, ref, op); } return; } else { crm_warn("Could not delete '%s' from executor for %s%s%s: %s " CRM_XS " rc=%d", id, sys, (user? " as " : ""), (user? user : ""), pcmk_strerror(rc), rc); } delete_rsc_entry(lrm_state, request, id, gIter, rc, user); } 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) { 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, NULL); } 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) { 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 */ gboolean unregister = TRUE; if (is_remote_lrmd_ra(NULL, NULL, entry->id)) { lrm_state_t *remote_lrm_state = lrm_state_find(entry->id); if (remote_lrm_state) { /* when forcing a reprobe, make sure to clear remote node before * clearing the remote node's connection resource */ force_reprobe(remote_lrm_state, from_sys, from_host, user_name, TRUE); } unregister = FALSE; } delete_resource(lrm_state, entry->id, &entry->rsc, &gIter, from_sys, user_name, NULL, unregister); } /* Now delete the copy in the CIB */ controld_delete_node_state(lrm_state->node_name, controld_section_lrm, cib_scope_local); /* Finally, _delete_ the value in pacemaker-attrd -- setting it to FALSE * would result in the scheduler sending us back here again */ 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] 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 */ static void synthesize_lrmd_failure(lrm_state_t *lrm_state, xmlNode *action, int op_status, enum ocf_exitcode rc) { lrmd_event_data_t *op = NULL; const char *operation = crm_element_value(action, XML_LRM_ATTR_TASK); const char *target_node = crm_element_value(action, XML_LRM_ATTR_TARGET); xmlNode *xml_rsc = find_xml_node(action, XML_CIB_TAG_RESOURCE, TRUE); if ((xml_rsc == NULL) || (ID(xml_rsc) == NULL)) { /* @TODO Should we do something else, like direct ack? */ crm_info("Can't fake %s failure (%d) on %s without resource configuration", crm_element_value(action, XML_LRM_ATTR_TASK_KEY), rc, target_node); return; } else if(operation == NULL) { /* This probably came from crm_resource -C, nothing to do */ crm_info("Can't fake %s failure (%d) on %s without operation", ID(xml_rsc), rc, target_node); return; } op = construct_op(lrm_state, action, ID(xml_rsc), operation); if (pcmk__str_eq(operation, RSC_NOTIFY, pcmk__str_casei)) { // Notifications can't fail fake_op_status(lrm_state, op, PCMK_EXEC_DONE, PCMK_OCF_OK); } else { fake_op_status(lrm_state, op, op_status, rc); } 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 * * \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(xmlNode *xml) { const char *target = NULL; if (xml) { target = crm_element_value(xml, XML_LRM_ATTR_TARGET); } if (target == NULL) { target = fsa_our_uname; } return target; } static void fail_lrm_resource(xmlNode *xml, lrm_state_t *lrm_state, const char *user_name, const char *from_host, const char *from_sys) { lrmd_event_data_t *op = NULL; lrmd_rsc_info_t *rsc = NULL; xmlNode *xml_rsc = find_xml_node(xml, XML_CIB_TAG_RESOURCE, TRUE); CRM_CHECK(xml_rsc != NULL, return); /* The executor simply executes operations and reports the results, without * any concept of success or failure, so to fail a resource, we must fake * what a failure looks like. * * To do this, we create a fake executor operation event for the resource, * and pass that event to the executor client callback so it will be * processed as if it came from the executor. */ op = construct_op(lrm_state, xml, ID(xml_rsc), "asyncmon"); fake_op_status(lrm_state, op, PCMK_EXEC_DONE, PCMK_OCF_UNKNOWN_ERROR); 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)); 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); lrmd__set_result(op, PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_DONE, "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"); } controld_ack_event_directly(from_host, from_sys, NULL, op, ID(xml_rsc)); lrmd_free_event(op); } static void handle_refresh_op(lrm_state_t *lrm_state, const char *user_name, const char *from_host, const char *from_sys) { int rc = pcmk_ok; xmlNode *fragment = do_lrm_query_internal(lrm_state, node_update_all); fsa_cib_update(XML_CIB_TAG_STATUS, fragment, cib_quorum_override, rc, user_name); crm_info("Forced a local resource history refresh: call=%d", rc); if (!pcmk__str_eq(CRM_SYSTEM_CRMD, from_sys, pcmk__str_casei)) { xmlNode *reply = create_request(CRM_OP_INVOKE_LRM, fragment, from_host, from_sys, CRM_SYSTEM_LRMD, fsa_our_uuid); crm_debug("ACK'ing refresh 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); } free_xml(fragment); } static void handle_query_op(xmlNode *msg, lrm_state_t *lrm_state) { xmlNode *data = do_lrm_query_internal(lrm_state, node_update_all); xmlNode *reply = create_reply(msg, data); if (relay_message(reply, TRUE) == FALSE) { crm_err("Unable to route reply"); crm_log_xml_err(reply, "reply"); } free_xml(reply); free_xml(data); } 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) { crm_notice("Forcing the status of all resources to be redetected"); force_reprobe(lrm_state, from_sys, from_host, user_name, is_remote_node); 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, fsa_our_uuid); crm_debug("ACK'ing re-probe from %s (%s)", from_sys, from_host); if (relay_message(reply, TRUE) == FALSE) { crm_log_xml_err(reply, "Unable to route reply"); } free_xml(reply); } } static bool do_lrm_cancel(ha_msg_input_t *input, lrm_state_t *lrm_state, lrmd_rsc_info_t *rsc, const char *from_host, const char *from_sys) { char *op_key = NULL; char *meta_key = NULL; int call = 0; const char *call_id = NULL; const char *op_task = NULL; guint interval_ms = 0; gboolean in_progress = FALSE; xmlNode *params = find_xml_node(input->xml, XML_TAG_ATTRS, TRUE); CRM_CHECK(params != NULL, return FALSE); meta_key = crm_meta_name(XML_LRM_ATTR_TASK); op_task = crm_element_value(params, meta_key); free(meta_key); CRM_CHECK(op_task != NULL, return FALSE); meta_key = crm_meta_name(XML_LRM_ATTR_INTERVAL_MS); if (crm_element_value_ms(params, meta_key, &interval_ms) != pcmk_ok) { free(meta_key); return FALSE; } free(meta_key); op_key = pcmk__op_key(rsc->id, op_task, interval_ms); meta_key = crm_meta_name(XML_LRM_ATTR_CALLID); call_id = crm_element_value(params, meta_key); free(meta_key); crm_debug("Scheduler requested op %s (call=%s) be cancelled", op_key, (call_id? call_id : "NA")); pcmk__scan_min_int(call_id, &call, 0); if (call == 0) { // Normal case when the scheduler cancels a recurring op in_progress = cancel_op_key(lrm_state, rsc, op_key, TRUE); } else { // Normal case when the scheduler cancels an orphan op in_progress = cancel_op(lrm_state, rsc->id, NULL, call, TRUE); } // Acknowledge cancellation operation if for a remote connection resource if (!in_progress || is_remote_lrmd_ra(NULL, NULL, rsc->id)) { char *op_id = make_stop_id(rsc->id, call); if (is_remote_lrmd_ra(NULL, NULL, rsc->id) == FALSE) { crm_info("Nothing known about operation %d for %s", call, op_key); } erase_lrm_history_by_id(lrm_state, rsc->id, op_key, NULL, 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 */ g_hash_table_remove(lrm_state->pending_ops, op_id); free(op_id); } else { /* No ack is needed since abcdaa8, but peers with older versions * in a rolling upgrade need one. We didn't bump the feature set * at that commit, so we can only compare against the previous * CRM version (3.0.8). If any peers have feature set 3.0.9 but * not abcdaa8, they will time out waiting for the ack (no * released versions of Pacemaker are affected). */ const char *peer_version = crm_element_value(params, XML_ATTR_CRM_VERSION); if (compare_version(peer_version, "3.0.8") <= 0) { crm_info("Sending compatibility ack for %s cancellation to %s (CRM version %s)", op_key, from_host, peer_version); send_task_ok_ack(lrm_state, input, rsc->id, rsc, op_task, from_host, from_sys); } } free(op_key); return TRUE; } static void do_lrm_delete(ha_msg_input_t *input, lrm_state_t *lrm_state, lrmd_rsc_info_t *rsc, const char *from_sys, const char *from_host, bool crm_rsc_delete, const char *user_name) { gboolean unregister = TRUE; int cib_rc = controld_delete_resource_history(rsc->id, lrm_state->node_name, user_name, cib_dryrun|cib_sync_call); if (cib_rc != pcmk_rc_ok) { lrmd_event_data_t *op = NULL; op = construct_op(lrm_state, input->xml, rsc->id, CRMD_ACTION_DELETE); + /* These are resource clean-ups, not actions, so no exit reason is + * needed. + */ if (cib_rc == EACCES) { lrmd__set_result(op, PCMK_OCF_INSUFFICIENT_PRIV, PCMK_EXEC_ERROR, NULL); } else { lrmd__set_result(op, PCMK_OCF_UNKNOWN_ERROR, 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); } /* 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 = NULL; gboolean is_remote_node = FALSE; bool crm_rsc_delete = FALSE; target_node = lrm_op_target(input->xml); is_remote_node = !pcmk__str_eq(target_node, fsa_our_uname, 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); return; } CRM_ASSERT(lrm_state != NULL); user_name = pcmk__update_acl_user(input->msg, F_CRM_USER, NULL); crm_op = crm_element_value(input->msg, F_CRM_TASK); from_sys = crm_element_value(input->msg, F_CRM_SYS_FROM); if (!pcmk__str_eq(from_sys, CRM_SYSTEM_TENGINE, pcmk__str_casei)) { from_host = crm_element_value(input->msg, F_CRM_HOST_FROM); } crm_trace("Executor %s command from %s as user %s", crm_op, from_sys, user_name); if (pcmk__str_eq(crm_op, CRM_OP_LRM_DELETE, pcmk__str_casei)) { if (!pcmk__str_eq(from_sys, CRM_SYSTEM_TENGINE, pcmk__str_casei)) { crm_rsc_delete = TRUE; // from crm_resource } operation = CRMD_ACTION_DELETE; } else if (pcmk__str_eq(crm_op, CRM_OP_LRM_FAIL, pcmk__str_casei)) { fail_lrm_resource(input->xml, lrm_state, user_name, from_host, from_sys); return; } else if (input->xml != NULL) { operation = crm_element_value(input->xml, XML_LRM_ATTR_TASK); } if (pcmk__str_eq(crm_op, CRM_OP_LRM_REFRESH, pcmk__str_casei)) { handle_refresh_op(lrm_state, user_name, from_host, from_sys); } else if (pcmk__str_eq(crm_op, CRM_OP_LRM_QUERY, pcmk__str_casei)) { handle_query_op(input->msg, lrm_state); } else if (pcmk__str_eq(operation, CRM_OP_PROBED, pcmk__str_casei)) { update_attrd(lrm_state->node_name, CRM_OP_PROBED, XML_BOOLEAN_TRUE, user_name, is_remote_node); } else if (pcmk__str_eq(crm_op, CRM_OP_REPROBE, pcmk__str_casei) || pcmk__str_eq(operation, CRM_OP_REPROBE, pcmk__str_casei)) { handle_reprobe_op(lrm_state, from_sys, from_host, user_name, is_remote_node); } else if (operation != NULL) { lrmd_rsc_info_t *rsc = NULL; xmlNode *xml_rsc = find_xml_node(input->xml, XML_CIB_TAG_RESOURCE, TRUE); gboolean create_rsc = !pcmk__str_eq(operation, CRMD_ACTION_DELETE, pcmk__str_casei); 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); 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); 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 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 return; } if (pcmk__str_eq(operation, CRMD_ACTION_CANCEL, pcmk__str_casei)) { if (!do_lrm_cancel(input, lrm_state, rsc, from_host, from_sys)) { crm_log_xml_warn(input->xml, "Bad command"); } } else if (pcmk__str_eq(operation, CRMD_ACTION_DELETE, pcmk__str_casei)) { do_lrm_delete(input, lrm_state, rsc, from_sys, from_host, crm_rsc_delete, user_name); } else if (pcmk__str_any_of(operation, CRMD_ACTION_RELOAD, CRMD_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. */ struct ra_metadata_s *md = NULL; const char *reload_name = CRMD_ACTION_RELOAD_AGENT; md = controld_get_rsc_metadata(lrm_state, rsc, controld_metadata_from_cache); if ((md != NULL) && pcmk_is_set(md->ra_flags, ra_supports_legacy_reload)) { reload_name = CRMD_ACTION_RELOAD; } do_lrm_rsc_op(lrm_state, rsc, reload_name, input->xml); } else { do_lrm_rsc_op(lrm_state, rsc, operation, input->xml); } lrmd_free_rsc_info(rsc); } else { crm_err("Cannot perform operation %s of unknown type", crm_str(crm_op)); register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); } } #if ENABLE_VERSIONED_ATTRS static void resolve_versioned_parameters(lrm_state_t *lrm_state, const char *rsc_id, const xmlNode *rsc_op, GHashTable *params) { /* Resource info *should* already be cached, so we don't get * executor call */ lrmd_rsc_info_t *rsc = lrm_state_get_rsc_info(lrm_state, rsc_id, 0); struct ra_metadata_s *metadata; metadata = controld_get_rsc_metadata(lrm_state, rsc, controld_metadata_from_cache); if (metadata) { xmlNode *versioned_attrs = NULL; GHashTable *hash = NULL; char *key = NULL; char *value = NULL; GHashTableIter iter; versioned_attrs = first_named_child(rsc_op, XML_TAG_OP_VER_ATTRS); hash = pe_unpack_versioned_parameters(versioned_attrs, metadata->ra_version); g_hash_table_iter_init(&iter, hash); while (g_hash_table_iter_next(&iter, (gpointer *) &key, (gpointer *) &value)) { g_hash_table_iter_steal(&iter); g_hash_table_replace(params, key, value); } g_hash_table_destroy(hash); versioned_attrs = first_named_child(rsc_op, XML_TAG_OP_VER_META); hash = pe_unpack_versioned_parameters(versioned_attrs, metadata->ra_version); g_hash_table_iter_init(&iter, hash); while (g_hash_table_iter_next(&iter, (gpointer *) &key, (gpointer *) &value)) { g_hash_table_replace(params, crm_meta_name(key), strdup(value)); if (pcmk__str_eq(key, XML_ATTR_TIMEOUT, pcmk__str_casei)) { pcmk__scan_min_int(value, &op->timeout, 0); } else if (pcmk__str_eq(key, XML_OP_ATTR_START_DELAY, pcmk__str_casei)) { pcmk__scan_min_int(value, &op->start_delay, 0); } } g_hash_table_destroy(hash); versioned_attrs = first_named_child(rsc_op, XML_TAG_RSC_VER_ATTRS); hash = pe_unpack_versioned_parameters(versioned_attrs, metadata->ra_version); g_hash_table_iter_init(&iter, hash); while (g_hash_table_iter_next(&iter, (gpointer *) &key, (gpointer *) &value)) { g_hash_table_iter_steal(&iter); g_hash_table_replace(params, key, value); } g_hash_table_destroy(hash); } lrmd_free_rsc_info(rsc); } #endif static lrmd_event_data_t * construct_op(lrm_state_t *lrm_state, 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(CRMD_ACTION_STOP, operation, pcmk__str_casei)); op->user_data = NULL; /* the stop_all_resources() case * by definition there is no DC (or they'd be shutting * us down). * So we should put our version here. */ op->params = pcmk__strkey_table(free, free); g_hash_table_insert(op->params, strdup(XML_ATTR_CRM_VERSION), strdup(CRM_FEATURE_SET)); crm_trace("Constructed %s op for %s", operation, rsc_id); return op; } params = xml2list(rsc_op); g_hash_table_remove(params, CRM_META "_op_target_rc"); op_delay = crm_meta_value(params, XML_OP_ATTR_START_DELAY); pcmk__scan_min_int(op_delay, &op->start_delay, 0); op_timeout = crm_meta_value(params, XML_ATTR_TIMEOUT); pcmk__scan_min_int(op_timeout, &op->timeout, 0); if (pcmk__guint_from_hash(params, CRM_META "_" XML_LRM_ATTR_INTERVAL_MS, 0, &(op->interval_ms)) != pcmk_rc_ok) { op->interval_ms = 0; } /* Use pcmk_monitor_timeout instead of meta timeout for stonith recurring monitor, if set */ primitive = find_xml_node(rsc_op, XML_CIB_TAG_RESOURCE, FALSE); class = crm_element_value(primitive, XML_AGENT_ATTR_CLASS); if (pcmk_is_set(pcmk_get_ra_caps(class), pcmk_ra_cap_fence_params) && pcmk__str_eq(operation, CRMD_ACTION_STATUS, pcmk__str_casei) && (op->interval_ms > 0)) { op_timeout = g_hash_table_lookup(params, "pcmk_monitor_timeout"); if (op_timeout != NULL) { op->timeout = crm_get_msec(op_timeout); } } #if ENABLE_VERSIONED_ATTRS if (lrm_state && !is_remote_lrmd_ra(NULL, NULL, rsc_id) && !pcmk__strcase_any_of(op_type, CRMD_ACTION_METADATA, CRMD_ACTION_DELETE, NULL)) { resolve_versioned_parameters(lrm_state, rsc_id, rsc_op, params); } #endif if (!pcmk__str_eq(operation, RSC_STOP, pcmk__str_casei)) { op->params = params; } else { rsc_history_t *entry = NULL; if (lrm_state) { entry = g_hash_table_lookup(lrm_state->resource_history, rsc_id); } /* If we do not have stop parameters cached, use * whatever we are given */ if (!entry || !entry->stop_params) { op->params = params; } else { /* Copy the cached parameter list so that we stop the resource * with the old attributes, not the new ones */ op->params = pcmk__strkey_table(free, free); g_hash_table_foreach(params, copy_meta_keys, op->params); g_hash_table_foreach(entry->stop_params, copy_instance_keys, op->params); g_hash_table_destroy(params); params = NULL; } } /* sanity */ if (op->timeout <= 0) { op->timeout = op->interval_ms; } if (op->start_delay < 0) { op->start_delay = 0; } transition = crm_element_value(rsc_op, XML_ATTR_TRANSITION_KEY); CRM_CHECK(transition != NULL, return op); op->user_data = strdup(transition); if (op->interval_ms != 0) { if (pcmk__strcase_any_of(operation, CRMD_ACTION_START, CRMD_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 to (NULL for transition engine) * \param[in] rsc Type information about resource the result is for * \param[in] 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, lrmd_rsc_info_t *rsc, lrmd_event_data_t *op, const char *rsc_id) { xmlNode *reply = NULL; xmlNode *update, *iter; crm_node_t *peer = NULL; CRM_CHECK(op != NULL, return); if (op->rsc_id == NULL) { CRM_ASSERT(rsc_id != NULL); op->rsc_id = strdup(rsc_id); } if (to_sys == NULL) { to_sys = CRM_SYSTEM_TENGINE; } peer = crm_get_peer(0, fsa_our_uname); update = create_node_state_update(peer, node_update_none, NULL, __func__); iter = create_xml_node(update, XML_CIB_TAG_LRM); crm_xml_add(iter, XML_ATTR_ID, fsa_our_uuid); iter = create_xml_node(iter, XML_LRM_TAG_RESOURCES); iter = create_xml_node(iter, XML_LRM_TAG_RESOURCE); crm_xml_add(iter, XML_ATTR_ID, op->rsc_id); build_operation_update(iter, rsc, op, fsa_our_uname, __func__); reply = create_request(CRM_OP_INVOKE_LRM, update, to_host, to_sys, CRM_SYSTEM_LRMD, NULL); crm_log_xml_trace(update, "[direct ACK]"); crm_debug("ACK'ing resource op " PCMK__OP_FMT " from %s: %s", op->rsc_id, op->op_type, op->interval_ms, op->user_data, crm_element_value(reply, XML_ATTR_REFERENCE)); if (relay_message(reply, TRUE) == FALSE) { crm_log_xml_err(reply, "Unable to route reply"); } free_xml(update); free_xml(reply); } gboolean verify_stopped(enum crmd_fsa_state cur_state, int log_level) { gboolean res = TRUE; GList *lrm_state_list = lrm_state_get_list(); GList *state_entry; for (state_entry = lrm_state_list; state_entry != NULL; state_entry = state_entry->next) { lrm_state_t *lrm_state = state_entry->data; if (!lrm_state_verify_stopped(lrm_state, cur_state, log_level)) { /* keep iterating through all even when false is returned */ res = FALSE; } } controld_set_fsa_input_flags(R_SENT_RSC_STOP); g_list_free(lrm_state_list); lrm_state_list = NULL; return res; } struct stop_recurring_action_s { lrmd_rsc_info_t *rsc; lrm_state_t *lrm_state; }; static gboolean stop_recurring_action_by_rsc(gpointer key, gpointer value, gpointer user_data) { gboolean remove = FALSE; struct stop_recurring_action_s *event = user_data; active_op_t *op = value; if ((op->interval_ms != 0) && pcmk__str_eq(op->rsc_id, event->rsc->id, pcmk__str_none)) { crm_debug("Cancelling op %d for %s (%s)", op->call_id, op->rsc_id, (char*)key); remove = !cancel_op(event->lrm_state, event->rsc->id, key, op->call_id, FALSE); } return remove; } static gboolean stop_recurring_actions(gpointer key, gpointer value, gpointer user_data) { gboolean remove = FALSE; lrm_state_t *lrm_state = user_data; active_op_t *op = value; if (op->interval_ms != 0) { crm_info("Cancelling op %d for %s (%s)", op->call_id, op->rsc_id, (const char *) key); remove = !cancel_op(lrm_state, op->rsc_id, key, op->call_id, FALSE); } return remove; } static void record_pending_op(const char *node_name, lrmd_rsc_info_t *rsc, lrmd_event_data_t *op) { const char *record_pending = NULL; CRM_CHECK(node_name != NULL, return); CRM_CHECK(rsc != NULL, return); CRM_CHECK(op != NULL, return); // Never record certain operation types as pending if ((op->op_type == NULL) || (op->params == NULL) || !controld_action_is_recordable(op->op_type)) { return; } // defaults to true record_pending = crm_meta_value(op->params, XML_OP_ATTR_PENDING); if (record_pending && !crm_is_true(record_pending)) { return; } op->call_id = -1; lrmd__set_result(op, PCMK_OCF_UNKNOWN, PCMK_EXEC_PENDING, NULL); op->t_run = time(NULL); op->t_rcchange = op->t_run; /* write a "pending" entry to the CIB, inhibit notification */ crm_debug("Recording pending op " PCMK__OP_FMT " on %s in the CIB", op->rsc_id, op->op_type, op->interval_ms, node_name); do_update_resource(node_name, rsc, op, 0); } static void do_lrm_rsc_op(lrm_state_t *lrm_state, lrmd_rsc_info_t *rsc, const char *operation, xmlNode *msg) { int rc; int call_id = 0; char *op_id = NULL; lrmd_event_data_t *op = NULL; lrmd_key_value_t *params = NULL; fsa_data_t *msg_data = NULL; const char *transition = NULL; gboolean stop_recurring = FALSE; - bool send_nack = FALSE; + const char *nack_reason = NULL; CRM_CHECK(rsc != NULL, return); CRM_CHECK(operation != NULL, return); if (msg != NULL) { transition = crm_element_value(msg, XML_ATTR_TRANSITION_KEY); if (transition == NULL) { crm_log_xml_err(msg, "Missing transition number"); } } op = construct_op(lrm_state, msg, rsc->id, operation); CRM_CHECK(op != NULL, return); if (is_remote_lrmd_ra(NULL, NULL, rsc->id) && (op->interval_ms == 0) && strcmp(operation, CRMD_ACTION_MIGRATE) == 0) { /* pcmk remote connections are a special use case. * We never ever want to stop monitoring a connection resource until * the entire migration has completed. If the connection is unexpectedly * severed, even during a migration, this is an event we must detect.*/ stop_recurring = FALSE; } else if ((op->interval_ms == 0) && strcmp(operation, CRMD_ACTION_STATUS) != 0 && strcmp(operation, CRMD_ACTION_NOTIFY) != 0) { /* stop any previous monitor operations before changing the resource state */ stop_recurring = TRUE; } if (stop_recurring == TRUE) { 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->pending_ops, stop_recurring_action_by_rsc, &data); if (removed) { crm_debug("Stopped %u recurring operation%s in preparation for " PCMK__OP_FMT, removed, pcmk__plural_s(removed), rsc->id, operation, op->interval_ms); } } /* now do the op */ crm_notice("Requesting local execution of %s operation for %s on %s " CRM_XS " transition_key=%s op_key=" PCMK__OP_FMT, crm_action_str(op->op_type, op->interval_ms), rsc->id, lrm_state->node_name, transition, rsc->id, operation, op->interval_ms); if (pcmk_is_set(fsa_input_register, R_SHUTDOWN) && pcmk__str_eq(operation, RSC_START, pcmk__str_casei)) { register_fsa_input(C_SHUTDOWN, I_SHUTDOWN, NULL); - send_nack = TRUE; + nack_reason = "Not attempting start due to shutdown in progress"; } else if (fsa_state != S_NOT_DC && fsa_state != S_POLICY_ENGINE /* Recalculating */ && fsa_state != S_TRANSITION_ENGINE && !pcmk__str_eq(operation, CRMD_ACTION_STOP, pcmk__str_casei)) { - send_nack = TRUE; + nack_reason = "Controller cannot attempt actions at this time"; } - if(send_nack) { + if (nack_reason != NULL) { crm_notice("Discarding attempt to perform action %s on %s in state %s (shutdown=%s)", operation, rsc->id, fsa_state2string(fsa_state), pcmk__btoa(pcmk_is_set(fsa_input_register, R_SHUTDOWN))); - lrmd__set_result(op, PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_INVALID, NULL); + 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; } 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); } if (op->params) { char *key = NULL; char *value = NULL; GHashTableIter iter; g_hash_table_iter_init(&iter, op->params); while (g_hash_table_iter_next(&iter, (gpointer *) & key, (gpointer *) & value)) { params = lrmd_key_value_add(params, key, value); } } rc = controld_execute_resource_agent(lrm_state, rsc->id, op->op_type, op->user_data, op->interval_ms, op->timeout, op->start_delay, 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); pending->user_data = op->user_data? strdup(op->user_data) : NULL; if (crm_element_value_epoch(msg, XML_CONFIG_ATTR_SHUTDOWN_LOCK, &(pending->lock_time)) != pcmk_ok) { pending->lock_time = 0; } g_hash_table_replace(lrm_state->pending_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); 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); process_lrm_event(lrm_state, op, NULL, NULL); } free(op_id); lrmd_free_event(op); } int last_resource_update = 0; static void cib_rsc_callback(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data) { switch (rc) { case pcmk_ok: case -pcmk_err_diff_failed: case -pcmk_err_diff_resync: crm_trace("Resource update %d complete: rc=%d", call_id, rc); break; default: crm_warn("Resource update %d failed: (rc=%d) %s", call_id, rc, pcmk_strerror(rc)); } if (call_id == last_resource_update) { last_resource_update = 0; trigger_fsa(); } } /* Only successful stops, and probes that found the resource inactive, get locks * recorded in the history. This ensures the resource stays locked to the node * until it is active there again after the node comes back up. */ static bool should_preserve_lock(lrmd_event_data_t *op) { if (!controld_shutdown_lock_enabled) { return false; } if (!strcmp(op->op_type, RSC_STOP) && (op->rc == PCMK_OCF_OK)) { return true; } if (!strcmp(op->op_type, RSC_STATUS) && (op->rc == PCMK_OCF_NOT_RUNNING)) { return true; } return false; } static int do_update_resource(const char *node_name, lrmd_rsc_info_t *rsc, lrmd_event_data_t *op, time_t lock_time) { /* */ int rc = pcmk_ok; xmlNode *update, *iter = NULL; int call_opt = crmd_cib_smart_opt(); const char *uuid = NULL; CRM_CHECK(op != NULL, return 0); iter = create_xml_node(iter, XML_CIB_TAG_STATUS); update = iter; iter = create_xml_node(iter, XML_CIB_TAG_STATE); if (pcmk__str_eq(node_name, fsa_our_uname, pcmk__str_casei)) { uuid = fsa_our_uuid; } else { /* remote nodes uuid and uname are equal */ uuid = node_name; crm_xml_add(iter, XML_NODE_IS_REMOTE, "true"); } CRM_LOG_ASSERT(uuid != NULL); if(uuid == NULL) { rc = -EINVAL; goto done; } crm_xml_add(iter, XML_ATTR_UUID, uuid); crm_xml_add(iter, XML_ATTR_UNAME, node_name); crm_xml_add(iter, XML_ATTR_ORIGIN, __func__); iter = create_xml_node(iter, XML_CIB_TAG_LRM); crm_xml_add(iter, XML_ATTR_ID, uuid); iter = create_xml_node(iter, XML_LRM_TAG_RESOURCES); iter = create_xml_node(iter, XML_LRM_TAG_RESOURCE); crm_xml_add(iter, XML_ATTR_ID, op->rsc_id); build_operation_update(iter, rsc, op, node_name, __func__); if (rsc) { const char *container = NULL; crm_xml_add(iter, XML_ATTR_TYPE, rsc->type); crm_xml_add(iter, XML_AGENT_ATTR_CLASS, rsc->standard); crm_xml_add(iter, XML_AGENT_ATTR_PROVIDER, rsc->provider); if (lock_time != 0) { /* Actions on a locked resource should either preserve the lock by * recording it with the action result, or clear it. */ if (!should_preserve_lock(op)) { lock_time = 0; } crm_xml_add_ll(iter, XML_CONFIG_ATTR_SHUTDOWN_LOCK, (long long) lock_time); } if (op->params) { container = g_hash_table_lookup(op->params, CRM_META"_"XML_RSC_ATTR_CONTAINER); } if (container) { crm_trace("Resource %s is a part of container resource %s", op->rsc_id, container); crm_xml_add(iter, XML_RSC_ATTR_CONTAINER, container); } } else { crm_warn("Resource %s no longer exists in the executor", op->rsc_id); controld_ack_event_directly(NULL, NULL, rsc, op, op->rsc_id); goto cleanup; } crm_log_xml_trace(update, __func__); /* make it an asynchronous call and be done with it * * Best case: * the resource state will be discovered during * the next signup or election. * * Bad case: * we are shutting down and there is no DC at the time, * but then why were we shutting down then anyway? * (probably because of an internal error) * * Worst case: * we get shot for having resources "running" that really weren't * * the alternative however means blocking here for too long, which * isn't acceptable */ fsa_cib_update(XML_CIB_TAG_STATUS, update, call_opt, rc, NULL); if (rc > 0) { last_resource_update = rc; } done: /* the return code is a call number, not an error code */ crm_trace("Sent resource state update message: %d for %s=%u on %s", rc, op->op_type, op->interval_ms, op->rsc_id); fsa_register_cib_callback(rc, FALSE, NULL, cib_rsc_callback); cleanup: free_xml(update); return rc; } 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; } void process_lrm_event(lrm_state_t *lrm_state, lrmd_event_data_t *op, active_op_t *pending, xmlNode *action_xml) { char *op_id = NULL; char *op_key = NULL; int update_id = 0; 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(fsa_our_dc_version, "3.2.0") < 0) { switch (op->op_status) { case PCMK_EXEC_NOT_CONNECTED: lrmd__set_result(op, PCMK_OCF_CONNECTION_DIED, PCMK_EXEC_ERROR, op->exit_reason); break; case PCMK_EXEC_INVALID: lrmd__set_result(op, CRM_DIRECT_NACK_RC, PCMK_EXEC_ERROR, op->exit_reason); break; default: break; } } op_id = make_stop_id(op->rsc_id, op->call_id); op_key = pcmk__op_key(op->rsc_id, op->op_type, op->interval_ms); // Get resource info if available (from executor state or action XML) if (lrm_state) { rsc = lrm_state_get_rsc_info(lrm_state, op->rsc_id, 0); } if ((rsc == NULL) && action_xml) { xmlNode *xml = find_xml_node(action_xml, XML_CIB_TAG_RESOURCE, TRUE); const char *standard = crm_element_value(xml, XML_AGENT_ATTR_CLASS); const char *provider = crm_element_value(xml, XML_AGENT_ATTR_PROVIDER); const char *type = crm_element_value(xml, XML_ATTR_TYPE); if (standard && type) { crm_info("%s agent information not cached, using %s%s%s:%s from action XML", op->rsc_id, standard, (provider? ":" : ""), (provider? provider : ""), type); rsc = lrmd_new_rsc_info(op->rsc_id, standard, provider, type); } else { crm_err("Can't process %s result because %s agent information not cached or in XML", op_key, op->rsc_id); } } // Get node name if available (from executor state or action XML) if (lrm_state) { node_name = lrm_state->node_name; } else if (action_xml) { node_name = crm_element_value(action_xml, XML_LRM_ATTR_TARGET); } if(pending == NULL) { remove = TRUE; if (lrm_state) { pending = g_hash_table_lookup(lrm_state->pending_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 update_id = do_update_resource(node_name, rsc, op, pending? pending->lock_time : 0); 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) { erase_lrm_history_by_op(lrm_state, op); } /* If the recurring operation had failed, the lrm_rsc_op is recorded as * "last_failure" which won't get erased from the cib given the logic on * purpose in erase_lrm_history_by_op(). So that the cancel action won't * have a chance to get confirmed by DC with process_op_deletion(). * Cluster transition would get stuck waiting for the remaining action * timer to time out. * * Directly acknowledge the cancel operation in this case. */ 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->pending_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->pending_ops)); } } if (node_name == NULL) { node_name = "unknown node"; // for logging } switch (op->op_status) { case PCMK_EXEC_CANCELLED: crm_info("Result of %s operation for %s on %s: %s " CRM_XS " call=%d key=%s confirmed=%s", crm_action_str(op->op_type, op->interval_ms), op->rsc_id, node_name, pcmk_exec_status_str(op->op_status), op->call_id, op_key, pcmk__btoa(removed)); break; case PCMK_EXEC_DONE: crm_notice("Result of %s operation for %s on %s: %s " CRM_XS " rc=%d call=%d key=%s confirmed=%s cib-update=%d", crm_action_str(op->op_type, op->interval_ms), op->rsc_id, node_name, services_ocf_exitcode_str(op->rc), op->rc, op->call_id, op_key, pcmk__btoa(removed), update_id); break; case PCMK_EXEC_TIMEOUT: crm_err("Result of %s operation for %s on %s: %s " CRM_XS " call=%d key=%s timeout=%dms", crm_action_str(op->op_type, op->interval_ms), op->rsc_id, node_name, pcmk_exec_status_str(op->op_status), op->call_id, op_key, op->timeout); break; default: crm_err("Result of %s operation for %s on %s: %s " CRM_XS " call=%d key=%s confirmed=%s status=%d cib-update=%d", crm_action_str(op->op_type, op->interval_ms), op->rsc_id, node_name, pcmk_exec_status_str(op->op_status), op->call_id, op_key, pcmk__btoa(removed), op->op_status, update_id); } if (op->output) { char *prefix = crm_strdup_printf("%s-" PCMK__OP_FMT ":%d", node_name, op->rsc_id, op->op_type, op->interval_ms, op->call_id); if (op->rc) { crm_log_output(LOG_NOTICE, prefix, op->output); } else { crm_log_output(LOG_DEBUG, prefix, op->output); } free(prefix); } if (lrm_state) { if (!pcmk__str_eq(op->op_type, RSC_METADATA, 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); metadata_cache_update(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); } } /* If a shutdown was escalated while operations were pending, * then the FSA will be stalled right now... allow it to continue */ mainloop_set_trigger(fsa_source); 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_execd_state.c b/daemons/controld/controld_execd_state.c index 5435402717..a0129bd514 100644 --- a/daemons/controld/controld_execd_state.c +++ b/daemons/controld/controld_execd_state.c @@ -1,853 +1,853 @@ /* * Copyright 2012-2021 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 GHashTable *lrm_state_table = NULL; extern GHashTable *proxy_table; int lrmd_internal_proxy_send(lrmd_t * lrmd, xmlNode *msg); void lrmd_internal_set_proxy_callback(lrmd_t * lrmd, void *userdata, void (*callback)(lrmd_t *lrmd, void *userdata, xmlNode *msg)); static void free_rsc_info(gpointer value) { lrmd_rsc_info_t *rsc_info = value; lrmd_free_rsc_info(rsc_info); } static void free_deletion_op(gpointer value) { struct pending_deletion_op_s *op = value; free(op->rsc); delete_ha_msg_input(op->input); free(op); } static void free_recurring_op(gpointer value) { active_op_t *op = value; free(op->user_data); free(op->rsc_id); free(op->op_type); free(op->op_key); if (op->params) { g_hash_table_destroy(op->params); } free(op); } static gboolean fail_pending_op(gpointer key, gpointer value, gpointer user_data) { lrmd_event_data_t event = { 0, }; lrm_state_t *lrm_state = user_data; active_op_t *op = value; crm_trace("Pre-emptively failing " PCMK__OP_FMT " on %s (call=%s, %s)", op->rsc_id, op->op_type, op->interval_ms, lrm_state->node_name, (char*)key, op->user_data); event.type = lrmd_event_exec_complete; event.rsc_id = op->rsc_id; event.op_type = op->op_type; event.user_data = op->user_data; event.timeout = 0; event.interval_ms = op->interval_ms; lrmd__set_result(&event, PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_NOT_CONNECTED, - NULL); + "Action was pending when executor connection was dropped"); event.t_run = (unsigned int) op->start_time; event.t_rcchange = (unsigned int) op->start_time; event.call_id = op->call_id; event.remote_nodename = lrm_state->node_name; event.params = op->params; process_lrm_event(lrm_state, &event, op, NULL); return TRUE; } gboolean lrm_state_is_local(lrm_state_t *lrm_state) { if (lrm_state == NULL || fsa_our_uname == NULL) { return FALSE; } if (strcmp(lrm_state->node_name, fsa_our_uname) != 0) { return FALSE; } return TRUE; } lrm_state_t * lrm_state_create(const char *node_name) { lrm_state_t *state = NULL; if (!node_name) { crm_err("No node name given for lrm state object"); return NULL; } state = calloc(1, sizeof(lrm_state_t)); if (!state) { return NULL; } state->node_name = strdup(node_name); state->rsc_info_cache = pcmk__strkey_table(NULL, free_rsc_info); state->deletion_ops = pcmk__strkey_table(free, free_deletion_op); state->pending_ops = pcmk__strkey_table(free, free_recurring_op); state->resource_history = pcmk__strkey_table(NULL, history_free); state->metadata_cache = metadata_cache_new(); g_hash_table_insert(lrm_state_table, (char *)state->node_name, state); return state; } void lrm_state_destroy(const char *node_name) { g_hash_table_remove(lrm_state_table, node_name); } static gboolean remote_proxy_remove_by_node(gpointer key, gpointer value, gpointer user_data) { remote_proxy_t *proxy = value; const char *node_name = user_data; if (pcmk__str_eq(node_name, proxy->node_name, pcmk__str_casei)) { return TRUE; } return FALSE; } static void internal_lrm_state_destroy(gpointer data) { lrm_state_t *lrm_state = data; if (!lrm_state) { return; } crm_trace("Destroying proxy table %s with %d members", lrm_state->node_name, g_hash_table_size(proxy_table)); g_hash_table_foreach_remove(proxy_table, remote_proxy_remove_by_node, (char *) lrm_state->node_name); remote_ra_cleanup(lrm_state); lrmd_api_delete(lrm_state->conn); if (lrm_state->rsc_info_cache) { crm_trace("Destroying rsc info cache with %d members", g_hash_table_size(lrm_state->rsc_info_cache)); g_hash_table_destroy(lrm_state->rsc_info_cache); } if (lrm_state->resource_history) { crm_trace("Destroying history op cache with %d members", g_hash_table_size(lrm_state->resource_history)); g_hash_table_destroy(lrm_state->resource_history); } if (lrm_state->deletion_ops) { crm_trace("Destroying deletion op cache with %d members", g_hash_table_size(lrm_state->deletion_ops)); g_hash_table_destroy(lrm_state->deletion_ops); } if (lrm_state->pending_ops) { crm_trace("Destroying pending op cache with %d members", g_hash_table_size(lrm_state->pending_ops)); g_hash_table_destroy(lrm_state->pending_ops); } metadata_cache_free(lrm_state->metadata_cache); free((char *)lrm_state->node_name); free(lrm_state); } void lrm_state_reset_tables(lrm_state_t * lrm_state, gboolean reset_metadata) { if (lrm_state->resource_history) { crm_trace("Re-setting history op cache with %d members", g_hash_table_size(lrm_state->resource_history)); g_hash_table_remove_all(lrm_state->resource_history); } if (lrm_state->deletion_ops) { crm_trace("Re-setting deletion op cache with %d members", g_hash_table_size(lrm_state->deletion_ops)); g_hash_table_remove_all(lrm_state->deletion_ops); } if (lrm_state->pending_ops) { crm_trace("Re-setting pending op cache with %d members", g_hash_table_size(lrm_state->pending_ops)); g_hash_table_remove_all(lrm_state->pending_ops); } if (lrm_state->rsc_info_cache) { crm_trace("Re-setting rsc info cache with %d members", g_hash_table_size(lrm_state->rsc_info_cache)); g_hash_table_remove_all(lrm_state->rsc_info_cache); } if (reset_metadata) { metadata_cache_reset(lrm_state->metadata_cache); } } gboolean lrm_state_init_local(void) { if (lrm_state_table) { return TRUE; } lrm_state_table = pcmk__strikey_table(NULL, internal_lrm_state_destroy); if (!lrm_state_table) { return FALSE; } proxy_table = pcmk__strikey_table(NULL, remote_proxy_free); if (!proxy_table) { g_hash_table_destroy(lrm_state_table); lrm_state_table = NULL; return FALSE; } return TRUE; } void lrm_state_destroy_all(void) { if (lrm_state_table) { crm_trace("Destroying state table with %d members", g_hash_table_size(lrm_state_table)); g_hash_table_destroy(lrm_state_table); lrm_state_table = NULL; } if(proxy_table) { crm_trace("Destroying proxy table with %d members", g_hash_table_size(proxy_table)); g_hash_table_destroy(proxy_table); proxy_table = NULL; } } lrm_state_t * lrm_state_find(const char *node_name) { if (!node_name) { return NULL; } return g_hash_table_lookup(lrm_state_table, node_name); } lrm_state_t * lrm_state_find_or_create(const char *node_name) { lrm_state_t *lrm_state; lrm_state = g_hash_table_lookup(lrm_state_table, node_name); if (!lrm_state) { lrm_state = lrm_state_create(node_name); } return lrm_state; } GList * lrm_state_get_list(void) { return g_hash_table_get_values(lrm_state_table); } static remote_proxy_t * find_connected_proxy_by_node(const char * node_name) { GHashTableIter gIter; remote_proxy_t *proxy = NULL; CRM_CHECK(proxy_table != NULL, return NULL); g_hash_table_iter_init(&gIter, proxy_table); while (g_hash_table_iter_next(&gIter, NULL, (gpointer *) &proxy)) { if (proxy->source && pcmk__str_eq(node_name, proxy->node_name, pcmk__str_casei)) { return proxy; } } return NULL; } static void remote_proxy_disconnect_by_node(const char * node_name) { remote_proxy_t *proxy = NULL; CRM_CHECK(proxy_table != NULL, return); while ((proxy = find_connected_proxy_by_node(node_name)) != NULL) { /* mainloop_del_ipc_client() eventually calls remote_proxy_disconnected() * , which removes the entry from proxy_table. * Do not do this in a g_hash_table_iter_next() loop. */ if (proxy->source) { mainloop_del_ipc_client(proxy->source); } } return; } void lrm_state_disconnect_only(lrm_state_t * lrm_state) { int removed = 0; if (!lrm_state->conn) { return; } crm_trace("Disconnecting %s", lrm_state->node_name); remote_proxy_disconnect_by_node(lrm_state->node_name); ((lrmd_t *) lrm_state->conn)->cmds->disconnect(lrm_state->conn); if (!pcmk_is_set(fsa_input_register, R_SHUTDOWN)) { removed = g_hash_table_foreach_remove(lrm_state->pending_ops, fail_pending_op, lrm_state); crm_trace("Synthesized %d operation failures for %s", removed, lrm_state->node_name); } } void lrm_state_disconnect(lrm_state_t * lrm_state) { if (!lrm_state->conn) { return; } lrm_state_disconnect_only(lrm_state); lrmd_api_delete(lrm_state->conn); lrm_state->conn = NULL; } int lrm_state_is_connected(lrm_state_t * lrm_state) { if (!lrm_state->conn) { return FALSE; } return ((lrmd_t *) lrm_state->conn)->cmds->is_connected(lrm_state->conn); } int lrm_state_poke_connection(lrm_state_t * lrm_state) { if (!lrm_state->conn) { return -ENOTCONN; } return ((lrmd_t *) lrm_state->conn)->cmds->poke_connection(lrm_state->conn); } // \return Standard Pacemaker return code int controld_connect_local_executor(lrm_state_t *lrm_state) { int rc = pcmk_rc_ok; if (lrm_state->conn == NULL) { lrmd_t *api = NULL; rc = lrmd__new(&api, NULL, NULL, 0); if (rc != pcmk_rc_ok) { return rc; } api->cmds->set_callback(api, lrm_op_callback); lrm_state->conn = api; } rc = ((lrmd_t *) lrm_state->conn)->cmds->connect(lrm_state->conn, CRM_SYSTEM_CRMD, NULL); rc = pcmk_legacy2rc(rc); if (rc == pcmk_rc_ok) { lrm_state->num_lrm_register_fails = 0; } else { lrm_state->num_lrm_register_fails++; } return rc; } static remote_proxy_t * crmd_remote_proxy_new(lrmd_t *lrmd, const char *node_name, const char *session_id, const char *channel) { struct ipc_client_callbacks proxy_callbacks = { .dispatch = remote_proxy_dispatch, .destroy = remote_proxy_disconnected }; remote_proxy_t *proxy = remote_proxy_new(lrmd, &proxy_callbacks, node_name, session_id, channel); return proxy; } gboolean crmd_is_proxy_session(const char *session) { return g_hash_table_lookup(proxy_table, session) ? TRUE : FALSE; } void crmd_proxy_send(const char *session, xmlNode *msg) { remote_proxy_t *proxy = g_hash_table_lookup(proxy_table, session); lrm_state_t *lrm_state = NULL; if (!proxy) { return; } crm_log_xml_trace(msg, "to-proxy"); lrm_state = lrm_state_find(proxy->node_name); if (lrm_state) { crm_trace("Sending event to %.8s on %s", proxy->session_id, proxy->node_name); remote_proxy_relay_event(proxy, msg); } } static void crmd_proxy_dispatch(const char *session, xmlNode *msg) { crm_trace("Processing proxied IPC message from session %s", session); crm_log_xml_trace(msg, "controller[inbound]"); crm_xml_add(msg, F_CRM_SYS_FROM, session); if (controld_authorize_ipc_message(msg, NULL, session)) { route_message(C_IPC_MESSAGE, msg); } trigger_fsa(); } static void remote_config_check(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data) { if (rc != pcmk_ok) { crm_err("Query resulted in an error: %s", pcmk_strerror(rc)); if (rc == -EACCES || rc == -pcmk_err_schema_validation) { crm_err("The cluster is mis-configured - shutting down and staying down"); } } else { lrmd_t * lrmd = (lrmd_t *)user_data; crm_time_t *now = crm_time_new(NULL); GHashTable *config_hash = pcmk__strkey_table(free, free); crm_debug("Call %d : Parsing CIB options", call_id); pe_unpack_nvpairs(output, output, XML_CIB_TAG_PROPSET, NULL, config_hash, CIB_OPTIONS_FIRST, FALSE, now, NULL); /* Now send it to the remote peer */ lrmd__validate_remote_settings(lrmd, config_hash); g_hash_table_destroy(config_hash); crm_time_free(now); } } static void crmd_remote_proxy_cb(lrmd_t *lrmd, void *userdata, xmlNode *msg) { lrm_state_t *lrm_state = userdata; const char *session = crm_element_value(msg, F_LRMD_IPC_SESSION); remote_proxy_t *proxy = g_hash_table_lookup(proxy_table, session); const char *op = crm_element_value(msg, F_LRMD_IPC_OP); if (pcmk__str_eq(op, LRMD_IPC_OP_NEW, pcmk__str_casei)) { const char *channel = crm_element_value(msg, F_LRMD_IPC_IPC_SERVER); proxy = crmd_remote_proxy_new(lrmd, lrm_state->node_name, session, channel); if (!remote_ra_controlling_guest(lrm_state)) { if (proxy != NULL) { /* Look up stonith-watchdog-timeout and send to the remote peer for validation */ int rc = fsa_cib_conn->cmds->query(fsa_cib_conn, XML_CIB_TAG_CRMCONFIG, NULL, cib_scope_local); fsa_cib_conn->cmds->register_callback_full(fsa_cib_conn, rc, 10, FALSE, lrmd, "remote_config_check", remote_config_check, NULL); } } else { crm_debug("Skipping remote_config_check for guest-nodes"); } } else if (pcmk__str_eq(op, LRMD_IPC_OP_SHUTDOWN_REQ, pcmk__str_casei)) { char *now_s = NULL; crm_notice("%s requested shutdown of its remote connection", lrm_state->node_name); if (!remote_ra_is_in_maintenance(lrm_state)) { now_s = pcmk__ttoa(time(NULL)); update_attrd(lrm_state->node_name, XML_CIB_ATTR_SHUTDOWN, now_s, NULL, TRUE); free(now_s); remote_proxy_ack_shutdown(lrmd); crm_warn("Reconnection attempts to %s may result in failures that must be cleared", lrm_state->node_name); } else { remote_proxy_nack_shutdown(lrmd); crm_notice("Remote resource for %s is not managed so no ordered shutdown happening", lrm_state->node_name); } return; } else if (pcmk__str_eq(op, LRMD_IPC_OP_REQUEST, pcmk__str_casei) && proxy && proxy->is_local) { /* This is for the controller, which we are, so don't try * to send to ourselves over IPC -- do it directly. */ int flags = 0; xmlNode *request = get_message_xml(msg, F_LRMD_IPC_MSG); CRM_CHECK(request != NULL, return); CRM_CHECK(lrm_state->node_name, return); crm_xml_add(request, XML_ACL_TAG_ROLE, "pacemaker-remote"); pcmk__update_acl_user(request, F_LRMD_IPC_USER, lrm_state->node_name); /* Pacemaker Remote nodes don't know their own names (as known to the * cluster). When getting a node info request with no name or ID, add * the name, so we don't return info for ourselves instead of the * Pacemaker Remote node. */ if (pcmk__str_eq(crm_element_value(request, F_CRM_TASK), CRM_OP_NODE_INFO, pcmk__str_casei)) { int node_id = 0; crm_element_value_int(request, XML_ATTR_ID, &node_id); if ((node_id <= 0) && (crm_element_value(request, XML_ATTR_UNAME) == NULL)) { crm_xml_add(request, XML_ATTR_UNAME, lrm_state->node_name); } } crmd_proxy_dispatch(session, request); crm_element_value_int(msg, F_LRMD_IPC_MSG_FLAGS, &flags); if (flags & crm_ipc_client_response) { int msg_id = 0; xmlNode *op_reply = create_xml_node(NULL, "ack"); crm_xml_add(op_reply, "function", __func__); crm_xml_add_int(op_reply, "line", __LINE__); crm_element_value_int(msg, F_LRMD_IPC_MSG_ID, &msg_id); remote_proxy_relay_response(proxy, op_reply, msg_id); free_xml(op_reply); } } else { remote_proxy_cb(lrmd, lrm_state->node_name, msg); } } // \return Standard Pacemaker return code int controld_connect_remote_executor(lrm_state_t *lrm_state, const char *server, int port, int timeout_ms) { int rc = pcmk_rc_ok; if (lrm_state->conn == NULL) { lrmd_t *api = NULL; rc = lrmd__new(&api, lrm_state->node_name, server, port); if (rc != pcmk_rc_ok) { crm_warn("Pacemaker Remote connection to %s:%s failed: %s " CRM_XS " rc=%d", server, port, pcmk_rc_str(rc), rc); return rc; } lrm_state->conn = api; api->cmds->set_callback(api, remote_lrm_op_callback); lrmd_internal_set_proxy_callback(api, lrm_state, crmd_remote_proxy_cb); } crm_trace("Initiating remote connection to %s:%d with timeout %dms", server, port, timeout_ms); rc = ((lrmd_t *) lrm_state->conn)->cmds->connect_async(lrm_state->conn, lrm_state->node_name, timeout_ms); if (rc == pcmk_ok) { lrm_state->num_lrm_register_fails = 0; } else { lrm_state->num_lrm_register_fails++; // Ignored for remote connections } return pcmk_legacy2rc(rc); } int lrm_state_get_metadata(lrm_state_t * lrm_state, const char *class, const char *provider, const char *agent, char **output, enum lrmd_call_options options) { lrmd_key_value_t *params = NULL; if (!lrm_state->conn) { return -ENOTCONN; } /* Add the node name to the environment, as is done with normal resource * action calls. Meta-data calls shouldn't need it, but some agents are * written with an ocf_local_nodename call at the beginning regardless of * action. Without the environment variable, the agent would try to contact * the controller to get the node name -- but the controller would be * blocking on the synchronous meta-data call. * * At this point, we have to assume that agents are unlikely to make other * calls that require the controller, such as crm_node --quorum or * --cluster-id. * * @TODO Make meta-data calls asynchronous. (This will be part of a larger * project to make meta-data calls via the executor rather than directly.) */ params = lrmd_key_value_add(params, CRM_META "_" XML_LRM_ATTR_TARGET, lrm_state->node_name); return ((lrmd_t *) lrm_state->conn)->cmds->get_metadata_params(lrm_state->conn, class, provider, agent, output, options, params); } int lrm_state_cancel(lrm_state_t *lrm_state, const char *rsc_id, const char *action, guint interval_ms) { if (!lrm_state->conn) { return -ENOTCONN; } /* Figure out a way to make this async? * NOTICE: Currently it's synced and directly acknowledged in do_lrm_invoke(). */ if (is_remote_lrmd_ra(NULL, NULL, rsc_id)) { return remote_ra_cancel(lrm_state, rsc_id, action, interval_ms); } return ((lrmd_t *) lrm_state->conn)->cmds->cancel(lrm_state->conn, rsc_id, action, interval_ms); } lrmd_rsc_info_t * lrm_state_get_rsc_info(lrm_state_t * lrm_state, const char *rsc_id, enum lrmd_call_options options) { lrmd_rsc_info_t *rsc = NULL; if (!lrm_state->conn) { return NULL; } if (is_remote_lrmd_ra(NULL, NULL, rsc_id)) { return remote_ra_get_rsc_info(lrm_state, rsc_id); } rsc = g_hash_table_lookup(lrm_state->rsc_info_cache, rsc_id); if (rsc == NULL) { /* only contact the lrmd if we don't already have a cached rsc info */ rsc = ((lrmd_t *) lrm_state->conn)->cmds->get_rsc_info(lrm_state->conn, rsc_id, options); if (rsc == NULL) { return NULL; } /* cache the result */ g_hash_table_insert(lrm_state->rsc_info_cache, rsc->id, rsc); } return lrmd_copy_rsc_info(rsc); } /*! * \internal * \brief Initiate a resource agent action * * \param[in] lrm_state Executor state object * \param[in] rsc_id ID of resource for action * \param[in] action Action to execute * \param[in] userdata String to copy and pass to execution callback * \param[in] interval_ms Action interval (in milliseconds) * \param[in] timeout_ms Action timeout (in milliseconds) * \param[in] start_delay_ms Delay (in milliseconds) before initiating action * \param[in] params Resource parameters * \param[out] call_id Where to store call ID on success * * \return Standard Pacemaker return code * \note This takes ownership of \p params, which should not be used or freed * after calling this function. */ int controld_execute_resource_agent(lrm_state_t *lrm_state, const char *rsc_id, const char *action, const char *userdata, guint interval_ms, int timeout_ms, int start_delay_ms, lrmd_key_value_t *params, int *call_id) { int rc = pcmk_rc_ok; if (lrm_state->conn == NULL) { lrmd_key_value_freeall(params); rc = ENOTCONN; } else if (is_remote_lrmd_ra(NULL, NULL, rsc_id)) { rc = controld_execute_remote_agent(lrm_state, rsc_id, action, userdata, interval_ms, timeout_ms, start_delay_ms, params, call_id); } else { rc = ((lrmd_t *) lrm_state->conn)->cmds->exec(lrm_state->conn, rsc_id, action, userdata, interval_ms, timeout_ms, start_delay_ms, lrmd_opt_notify_changes_only, params); if (rc < 0) { rc = pcmk_legacy2rc(rc); } else { *call_id = rc; rc = pcmk_rc_ok; } } return rc; } int lrm_state_register_rsc(lrm_state_t * lrm_state, const char *rsc_id, const char *class, const char *provider, const char *agent, enum lrmd_call_options options) { lrmd_t *conn = (lrmd_t *) lrm_state->conn; if (conn == NULL) { return -ENOTCONN; } if (is_remote_lrmd_ra(agent, provider, NULL)) { return lrm_state_find_or_create(rsc_id)? pcmk_ok : -EINVAL; } /* @TODO Implement an asynchronous version of this (currently a blocking * call to the lrmd). */ return conn->cmds->register_rsc(lrm_state->conn, rsc_id, class, provider, agent, options); } int lrm_state_unregister_rsc(lrm_state_t * lrm_state, const char *rsc_id, enum lrmd_call_options options) { if (!lrm_state->conn) { return -ENOTCONN; } if (is_remote_lrmd_ra(NULL, NULL, rsc_id)) { lrm_state_destroy(rsc_id); return pcmk_ok; } g_hash_table_remove(lrm_state->rsc_info_cache, rsc_id); /* @TODO Optimize this ... this function is a blocking round trip from * client to daemon. The controld_execd_state.c code path that uses this * function should always treat it as an async operation. The executor API * should make an async version available. */ return ((lrmd_t *) lrm_state->conn)->cmds->unregister_rsc(lrm_state->conn, rsc_id, options); } /* * Functions for sending alerts via local executor connection */ static GList *crmd_alert_list = NULL; void crmd_unpack_alerts(xmlNode *alerts) { pe_free_alert_list(crmd_alert_list); crmd_alert_list = pe_unpack_alerts(alerts); } void crmd_alert_node_event(crm_node_t *node) { lrm_state_t *lrm_state; if (crmd_alert_list == NULL) { return; } lrm_state = lrm_state_find(fsa_our_uname); if (lrm_state == NULL) { return; } lrmd_send_node_alert((lrmd_t *) lrm_state->conn, crmd_alert_list, node->uname, node->id, node->state); } void crmd_alert_fencing_op(stonith_event_t * e) { char *desc; lrm_state_t *lrm_state; if (crmd_alert_list == NULL) { return; } lrm_state = lrm_state_find(fsa_our_uname); if (lrm_state == NULL) { return; } desc = crm_strdup_printf("Operation %s of %s by %s for %s@%s: %s (ref=%s)", e->action, e->target, (e->executioner? e->executioner : ""), e->client_origin, e->origin, pcmk_strerror(e->result), e->id); lrmd_send_fencing_alert((lrmd_t *) lrm_state->conn, crmd_alert_list, e->target, e->operation, desc, e->result); free(desc); } void crmd_alert_resource_op(const char *node, lrmd_event_data_t * op) { lrm_state_t *lrm_state; if (crmd_alert_list == NULL) { return; } lrm_state = lrm_state_find(fsa_our_uname); if (lrm_state == NULL) { return; } lrmd_send_resource_alert((lrmd_t *) lrm_state->conn, crmd_alert_list, node, op); } diff --git a/daemons/controld/controld_remote_ra.c b/daemons/controld/controld_remote_ra.c index e36df711c4..ffddd2f3cf 100644 --- a/daemons/controld/controld_remote_ra.c +++ b/daemons/controld/controld_remote_ra.c @@ -1,1337 +1,1340 @@ /* * Copyright 2013-2021 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 #define REMOTE_LRMD_RA "remote" /* The max start timeout before cmd retry */ #define MAX_START_TIMEOUT_MS 10000 typedef struct remote_ra_cmd_s { /*! the local node the cmd is issued from */ char *owner; /*! the remote node the cmd is executed on */ char *rsc_id; /*! the action to execute */ char *action; /*! some string the client wants us to give it back */ char *userdata; /*! start delay in ms */ int start_delay; /*! timer id used for start delay. */ int delay_id; /*! timeout in ms for cmd */ int timeout; int remaining_timeout; /*! recurring interval in ms */ guint interval_ms; /*! interval timer id */ int interval_id; int reported_success; int monitor_timeout_id; int takeover_timeout_id; /*! action parameters */ lrmd_key_value_t *params; pcmk__action_result_t result; int call_id; time_t start_time; gboolean cancel; } remote_ra_cmd_t; enum remote_migration_status { expect_takeover = 1, takeover_complete, }; typedef struct remote_ra_data_s { crm_trigger_t *work; remote_ra_cmd_t *cur_cmd; GList *cmds; GList *recurring_cmds; enum remote_migration_status migrate_status; gboolean active; /* Maintenance mode is difficult to determine from the controller's context, * so we have it signalled back with the transition from the scheduler. */ gboolean is_maintenance; /* Similar for whether we are controlling a guest node or remote node. * Fortunately there is a meta-attribute in the transition already and * as the situation doesn't change over time we can use the * resource start for noting down the information for later use when * the attributes aren't at hand. */ gboolean controlling_guest; } remote_ra_data_t; static int handle_remote_ra_start(lrm_state_t * lrm_state, remote_ra_cmd_t * cmd, int timeout_ms); static void handle_remote_ra_stop(lrm_state_t * lrm_state, remote_ra_cmd_t * cmd); static GList *fail_all_monitor_cmds(GList * list); static void free_cmd(gpointer user_data) { remote_ra_cmd_t *cmd = user_data; if (!cmd) { return; } if (cmd->delay_id) { g_source_remove(cmd->delay_id); } if (cmd->interval_id) { g_source_remove(cmd->interval_id); } if (cmd->monitor_timeout_id) { g_source_remove(cmd->monitor_timeout_id); } if (cmd->takeover_timeout_id) { g_source_remove(cmd->takeover_timeout_id); } free(cmd->owner); free(cmd->rsc_id); free(cmd->action); free(cmd->userdata); pcmk__reset_result(&(cmd->result)); lrmd_key_value_freeall(cmd->params); free(cmd); } static int generate_callid(void) { static int remote_ra_callid = 0; remote_ra_callid++; if (remote_ra_callid <= 0) { remote_ra_callid = 1; } return remote_ra_callid; } static gboolean recurring_helper(gpointer data) { remote_ra_cmd_t *cmd = data; lrm_state_t *connection_rsc = NULL; cmd->interval_id = 0; connection_rsc = lrm_state_find(cmd->rsc_id); if (connection_rsc && connection_rsc->remote_ra_data) { remote_ra_data_t *ra_data = connection_rsc->remote_ra_data; ra_data->recurring_cmds = g_list_remove(ra_data->recurring_cmds, cmd); ra_data->cmds = g_list_append(ra_data->cmds, cmd); mainloop_set_trigger(ra_data->work); } return FALSE; } static gboolean start_delay_helper(gpointer data) { remote_ra_cmd_t *cmd = data; lrm_state_t *connection_rsc = NULL; cmd->delay_id = 0; connection_rsc = lrm_state_find(cmd->rsc_id); if (connection_rsc && connection_rsc->remote_ra_data) { remote_ra_data_t *ra_data = connection_rsc->remote_ra_data; mainloop_set_trigger(ra_data->work); } return FALSE; } /*! * \internal * \brief Handle cluster communication related to pacemaker_remote node joining * * \param[in] node_name Name of newly integrated pacemaker_remote node */ static void remote_node_up(const char *node_name) { int call_opt, call_id = 0; xmlNode *update, *state; crm_node_t *node; enum controld_section_e section = controld_section_all; CRM_CHECK(node_name != NULL, return); crm_info("Announcing Pacemaker Remote node %s", node_name); /* Clear node's entire state (resource history and transient attributes) * other than shutdown locks. The transient attributes should and normally * will be cleared when the node leaves, but since remote node state has a * number of corner cases, clear them here as well, to be sure. */ call_opt = crmd_cib_smart_opt(); if (controld_shutdown_lock_enabled) { section = controld_section_all_unlocked; } /* Purge node from attrd's memory */ update_attrd_remote_node_removed(node_name, NULL); controld_delete_node_state(node_name, section, call_opt); /* Clear node's probed attribute */ update_attrd(node_name, CRM_OP_PROBED, NULL, NULL, TRUE); /* Ensure node is in the remote peer cache with member status */ node = crm_remote_peer_get(node_name); CRM_CHECK(node != NULL, return); pcmk__update_peer_state(__func__, node, CRM_NODE_MEMBER, 0); /* pacemaker_remote nodes don't participate in the membership layer, * so cluster nodes don't automatically get notified when they come and go. * We send a cluster message to the DC, and update the CIB node state entry, * so the DC will get it sooner (via message) or later (via CIB refresh), * and any other interested parties can query the CIB. */ send_remote_state_message(node_name, TRUE); update = create_xml_node(NULL, XML_CIB_TAG_STATUS); state = create_node_state_update(node, node_update_cluster, update, __func__); /* Clear the XML_NODE_IS_FENCED flag in the node state. If the node ever * needs to be fenced, this flag will allow various actions to determine * whether the fencing has happened yet. */ crm_xml_add(state, XML_NODE_IS_FENCED, "0"); /* TODO: If the remote connection drops, and this (async) CIB update either * failed or has not yet completed, later actions could mistakenly think the * node has already been fenced (if the XML_NODE_IS_FENCED attribute was * previously set, because it won't have been cleared). This could prevent * actual fencing or allow recurring monitor failures to be cleared too * soon. Ideally, we wouldn't rely on the CIB for the fenced status. */ fsa_cib_update(XML_CIB_TAG_STATUS, update, call_opt, call_id, NULL); if (call_id < 0) { crm_perror(LOG_WARNING, "%s CIB node state setup", node_name); } free_xml(update); } enum down_opts { DOWN_KEEP_LRM, DOWN_ERASE_LRM }; /*! * \internal * \brief Handle cluster communication related to pacemaker_remote node leaving * * \param[in] node_name Name of lost node * \param[in] opts Whether to keep or erase LRM history */ static void remote_node_down(const char *node_name, const enum down_opts opts) { xmlNode *update; int call_id = 0; int call_opt = crmd_cib_smart_opt(); crm_node_t *node; /* Purge node from attrd's memory */ update_attrd_remote_node_removed(node_name, NULL); /* Normally, only node attributes should be erased, and the resource history * should be kept until the node comes back up. However, after a successful * fence, we want to clear the history as well, so we don't think resources * are still running on the node. */ if (opts == DOWN_ERASE_LRM) { controld_delete_node_state(node_name, controld_section_all, call_opt); } else { controld_delete_node_state(node_name, controld_section_attrs, call_opt); } /* Ensure node is in the remote peer cache with lost state */ node = crm_remote_peer_get(node_name); CRM_CHECK(node != NULL, return); pcmk__update_peer_state(__func__, node, CRM_NODE_LOST, 0); /* Notify DC */ send_remote_state_message(node_name, FALSE); /* Update CIB node state */ update = create_xml_node(NULL, XML_CIB_TAG_STATUS); create_node_state_update(node, node_update_cluster, update, __func__); fsa_cib_update(XML_CIB_TAG_STATUS, update, call_opt, call_id, NULL); if (call_id < 0) { crm_perror(LOG_ERR, "%s CIB node state update", node_name); } free_xml(update); } /*! * \internal * \brief Handle effects of a remote RA command on node state * * \param[in] cmd Completed remote RA command */ static void check_remote_node_state(remote_ra_cmd_t *cmd) { /* Only successful actions can change node state */ if (cmd->result.exit_status != PCMK_OCF_OK) { return; } if (pcmk__str_eq(cmd->action, "start", pcmk__str_casei)) { remote_node_up(cmd->rsc_id); } else if (pcmk__str_eq(cmd->action, "migrate_from", pcmk__str_casei)) { /* After a successful migration, we don't need to do remote_node_up() * because the DC already knows the node is up, and we don't want to * clear LRM history etc. We do need to add the remote node to this * host's remote peer cache, because (unless it happens to be DC) * it hasn't been tracking the remote node, and other code relies on * the cache to distinguish remote nodes from unseen cluster nodes. */ crm_node_t *node = crm_remote_peer_get(cmd->rsc_id); CRM_CHECK(node != NULL, return); pcmk__update_peer_state(__func__, node, CRM_NODE_MEMBER, 0); } else if (pcmk__str_eq(cmd->action, "stop", pcmk__str_casei)) { lrm_state_t *lrm_state = lrm_state_find(cmd->rsc_id); remote_ra_data_t *ra_data = lrm_state? lrm_state->remote_ra_data : NULL; if (ra_data) { if (ra_data->migrate_status != takeover_complete) { /* Stop means down if we didn't successfully migrate elsewhere */ remote_node_down(cmd->rsc_id, DOWN_KEEP_LRM); } else if (AM_I_DC == FALSE) { /* Only the connection host and DC track node state, * so if the connection migrated elsewhere and we aren't DC, * un-cache the node, so we don't have stale info */ crm_remote_peer_cache_remove(cmd->rsc_id); } } } /* We don't do anything for successful monitors, which is correct for * routine recurring monitors, and for monitors on nodes where the * connection isn't supposed to be (the cluster will stop the connection in * that case). However, if the initial probe finds the connection already * active on the node where we want it, we probably should do * remote_node_up(). Unfortunately, we can't distinguish that case here. * Given that connections have to be initiated by the cluster, the chance of * that should be close to zero. */ } static void report_remote_ra_result(remote_ra_cmd_t * cmd) { lrmd_event_data_t op = { 0, }; check_remote_node_state(cmd); op.type = lrmd_event_exec_complete; op.rsc_id = cmd->rsc_id; op.op_type = cmd->action; op.user_data = cmd->userdata; op.timeout = cmd->timeout; op.interval_ms = cmd->interval_ms; op.t_run = (unsigned int) cmd->start_time; op.t_rcchange = (unsigned int) cmd->start_time; lrmd__set_result(&op, cmd->result.exit_status, cmd->result.exec_status, cmd->result.exit_reason); if (cmd->reported_success && (cmd->result.exit_status != PCMK_OCF_OK)) { op.t_rcchange = (unsigned int) time(NULL); /* This edge case will likely never ever occur, but if it does the * result is that a failure will not be processed correctly. This is only * remotely possible because we are able to detect a connection resource's tcp * connection has failed at any moment after start has completed. The actual * recurring operation is just a connectivity ping. * * basically, we are not guaranteed that the first successful monitor op and * a subsequent failed monitor op will not occur in the same timestamp. We have to * make it look like the operations occurred at separate times though. */ if (op.t_rcchange == op.t_run) { op.t_rcchange++; } } if (cmd->params) { lrmd_key_value_t *tmp; op.params = pcmk__strkey_table(free, free); for (tmp = cmd->params; tmp; tmp = tmp->next) { g_hash_table_insert(op.params, strdup(tmp->key), strdup(tmp->value)); } } op.call_id = cmd->call_id; op.remote_nodename = cmd->owner; lrm_op_callback(&op); if (op.params) { g_hash_table_destroy(op.params); } lrmd__reset_result(&op); } static void update_remaining_timeout(remote_ra_cmd_t * cmd) { cmd->remaining_timeout = ((cmd->timeout / 1000) - (time(NULL) - cmd->start_time)) * 1000; } static gboolean retry_start_cmd_cb(gpointer data) { lrm_state_t *lrm_state = data; remote_ra_data_t *ra_data = lrm_state->remote_ra_data; remote_ra_cmd_t *cmd = NULL; int rc = pcmk_rc_error; if (!ra_data || !ra_data->cur_cmd) { return FALSE; } cmd = ra_data->cur_cmd; if (!pcmk__strcase_any_of(cmd->action, "start", "migrate_from", NULL)) { return FALSE; } update_remaining_timeout(cmd); if (cmd->remaining_timeout > 0) { rc = handle_remote_ra_start(lrm_state, cmd, cmd->remaining_timeout); } if (rc != pcmk_rc_ok) { report_remote_ra_result(cmd); if (ra_data->cmds) { mainloop_set_trigger(ra_data->work); } ra_data->cur_cmd = NULL; free_cmd(cmd); } else { /* wait for connection event */ } return FALSE; } static gboolean connection_takeover_timeout_cb(gpointer data) { lrm_state_t *lrm_state = NULL; remote_ra_cmd_t *cmd = data; crm_info("takeover event timed out for node %s", cmd->rsc_id); cmd->takeover_timeout_id = 0; lrm_state = lrm_state_find(cmd->rsc_id); handle_remote_ra_stop(lrm_state, cmd); free_cmd(cmd); return FALSE; } static gboolean monitor_timeout_cb(gpointer data) { lrm_state_t *lrm_state = NULL; remote_ra_cmd_t *cmd = data; lrm_state = lrm_state_find(cmd->rsc_id); crm_info("Timed out waiting for remote poke response from %s%s", cmd->rsc_id, (lrm_state? "" : " (no LRM state)")); cmd->monitor_timeout_id = 0; pcmk__set_result(&(cmd->result), PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_TIMEOUT, - NULL); + "Remote executor did not respond"); if (lrm_state && lrm_state->remote_ra_data) { remote_ra_data_t *ra_data = lrm_state->remote_ra_data; if (ra_data->cur_cmd == cmd) { ra_data->cur_cmd = NULL; } if (ra_data->cmds) { mainloop_set_trigger(ra_data->work); } } report_remote_ra_result(cmd); free_cmd(cmd); if(lrm_state) { lrm_state_disconnect(lrm_state); } return FALSE; } static void synthesize_lrmd_success(lrm_state_t *lrm_state, const char *rsc_id, const char *op_type) { lrmd_event_data_t op = { 0, }; if (lrm_state == NULL) { /* if lrm_state not given assume local */ lrm_state = lrm_state_find(fsa_our_uname); } CRM_ASSERT(lrm_state != NULL); op.type = lrmd_event_exec_complete; op.rsc_id = rsc_id; op.op_type = op_type; op.t_run = (unsigned int) time(NULL); op.t_rcchange = op.t_run; op.call_id = generate_callid(); lrmd__set_result(&op, PCMK_OCF_OK, PCMK_EXEC_DONE, NULL); process_lrm_event(lrm_state, &op, NULL, NULL); } void remote_lrm_op_callback(lrmd_event_data_t * op) { gboolean cmd_handled = FALSE; lrm_state_t *lrm_state = NULL; remote_ra_data_t *ra_data = NULL; remote_ra_cmd_t *cmd = NULL; crm_debug("Processing '%s%s%s' event on remote connection to %s: %s " "(%d) status=%s (%d)", (op->op_type? op->op_type : ""), (op->op_type? " " : ""), lrmd_event_type2str(op->type), op->remote_nodename, services_ocf_exitcode_str(op->rc), op->rc, pcmk_exec_status_str(op->op_status), op->op_status); lrm_state = lrm_state_find(op->remote_nodename); if (!lrm_state || !lrm_state->remote_ra_data) { crm_debug("No state information found for remote connection event"); return; } ra_data = lrm_state->remote_ra_data; if (op->type == lrmd_event_new_client) { // Another client has connected to the remote daemon if (ra_data->migrate_status == expect_takeover) { // Great, we knew this was coming ra_data->migrate_status = takeover_complete; } else { crm_err("Disconnecting from Pacemaker Remote node %s due to " "unexpected client takeover", op->remote_nodename); /* In this case, lrmd_tls_connection_destroy() will be called under the control of mainloop. */ /* Do not free lrm_state->conn yet. */ /* It'll be freed in the following stop action. */ lrm_state_disconnect_only(lrm_state); } return; } /* filter all EXEC events up */ if (op->type == lrmd_event_exec_complete) { if (ra_data->migrate_status == takeover_complete) { crm_debug("ignoring event, this connection is taken over by another node"); } else { lrm_op_callback(op); } return; } if ((op->type == lrmd_event_disconnect) && (ra_data->cur_cmd == NULL)) { if (ra_data->active == FALSE) { crm_debug("Disconnection from Pacemaker Remote node %s complete", lrm_state->node_name); } else if (!remote_ra_is_in_maintenance(lrm_state)) { crm_err("Lost connection to Pacemaker Remote node %s", lrm_state->node_name); ra_data->recurring_cmds = fail_all_monitor_cmds(ra_data->recurring_cmds); ra_data->cmds = fail_all_monitor_cmds(ra_data->cmds); } else { crm_notice("Unmanaged Pacemaker Remote node %s disconnected", lrm_state->node_name); /* Do roughly what a 'stop' on the remote-resource would do */ handle_remote_ra_stop(lrm_state, NULL); remote_node_down(lrm_state->node_name, DOWN_KEEP_LRM); /* now fake the reply of a successful 'stop' */ synthesize_lrmd_success(NULL, lrm_state->node_name, "stop"); } return; } if (!ra_data->cur_cmd) { crm_debug("no event to match"); return; } cmd = ra_data->cur_cmd; /* Start actions and migrate from actions complete after connection * comes back to us. */ if (op->type == lrmd_event_connect && pcmk__strcase_any_of(cmd->action, "start", "migrate_from", NULL)) { if (op->connection_rc < 0) { update_remaining_timeout(cmd); if (op->connection_rc == -ENOKEY) { // Hard error, don't retry pcmk__set_result(&(cmd->result), PCMK_OCF_INVALID_PARAM, PCMK_EXEC_ERROR, "Authentication key not readable"); } else if (cmd->remaining_timeout > 3000) { crm_trace("rescheduling start, remaining timeout %d", cmd->remaining_timeout); g_timeout_add(1000, retry_start_cmd_cb, lrm_state); return; } else { crm_trace("can't reschedule start, remaining timeout too small %d", cmd->remaining_timeout); pcmk__set_result(&(cmd->result), PCMK_OCF_UNKNOWN_ERROR, - PCMK_EXEC_TIMEOUT, NULL); + PCMK_EXEC_TIMEOUT, + pcmk_strerror(op->connection_rc)); } } else { lrm_state_reset_tables(lrm_state, TRUE); pcmk__set_result(&(cmd->result), PCMK_OCF_OK, PCMK_EXEC_DONE, NULL); ra_data->active = TRUE; } crm_debug("Remote connection event matched %s action", cmd->action); report_remote_ra_result(cmd); cmd_handled = TRUE; } else if (op->type == lrmd_event_poke && pcmk__str_eq(cmd->action, "monitor", pcmk__str_casei)) { if (cmd->monitor_timeout_id) { g_source_remove(cmd->monitor_timeout_id); cmd->monitor_timeout_id = 0; } /* Only report success the first time, after that only worry about failures. * For this function, if we get the poke pack, it is always a success. Pokes * only fail if the send fails, or the response times out. */ if (!cmd->reported_success) { pcmk__set_result(&(cmd->result), PCMK_OCF_OK, PCMK_EXEC_DONE, NULL); report_remote_ra_result(cmd); cmd->reported_success = 1; } crm_debug("Remote poke event matched %s action", cmd->action); /* success, keep rescheduling if interval is present. */ if (cmd->interval_ms && (cmd->cancel == FALSE)) { ra_data->recurring_cmds = g_list_append(ra_data->recurring_cmds, cmd); cmd->interval_id = g_timeout_add(cmd->interval_ms, recurring_helper, cmd); cmd = NULL; /* prevent free */ } cmd_handled = TRUE; } else if (op->type == lrmd_event_disconnect && pcmk__str_eq(cmd->action, "monitor", pcmk__str_casei)) { if (ra_data->active == TRUE && (cmd->cancel == FALSE)) { pcmk__set_result(&(cmd->result), PCMK_OCF_UNKNOWN_ERROR, - PCMK_EXEC_ERROR, NULL); + PCMK_EXEC_ERROR, + "Remote connection unexpectedly dropped " + "during monitor"); report_remote_ra_result(cmd); crm_err("Remote connection to %s unexpectedly dropped during monitor", lrm_state->node_name); } cmd_handled = TRUE; } else if (op->type == lrmd_event_new_client && pcmk__str_eq(cmd->action, "stop", pcmk__str_casei)) { handle_remote_ra_stop(lrm_state, cmd); cmd_handled = TRUE; } else { crm_debug("Event did not match %s action", ra_data->cur_cmd->action); } if (cmd_handled) { ra_data->cur_cmd = NULL; if (ra_data->cmds) { mainloop_set_trigger(ra_data->work); } free_cmd(cmd); } } static void handle_remote_ra_stop(lrm_state_t * lrm_state, remote_ra_cmd_t * cmd) { remote_ra_data_t *ra_data = NULL; CRM_ASSERT(lrm_state); ra_data = lrm_state->remote_ra_data; if (ra_data->migrate_status != takeover_complete) { /* delete pending ops when ever the remote connection is intentionally stopped */ g_hash_table_remove_all(lrm_state->pending_ops); } else { /* we no longer hold the history if this connection has been migrated, * however, we keep metadata cache for future use */ lrm_state_reset_tables(lrm_state, FALSE); } ra_data->active = FALSE; lrm_state_disconnect(lrm_state); if (ra_data->cmds) { g_list_free_full(ra_data->cmds, free_cmd); } if (ra_data->recurring_cmds) { g_list_free_full(ra_data->recurring_cmds, free_cmd); } ra_data->cmds = NULL; ra_data->recurring_cmds = NULL; ra_data->cur_cmd = NULL; if (cmd) { pcmk__set_result(&(cmd->result), PCMK_OCF_OK, PCMK_EXEC_DONE, NULL); report_remote_ra_result(cmd); } } // \return Standard Pacemaker return code static int handle_remote_ra_start(lrm_state_t * lrm_state, remote_ra_cmd_t * cmd, int timeout_ms) { const char *server = NULL; lrmd_key_value_t *tmp = NULL; int port = 0; remote_ra_data_t *ra_data = lrm_state->remote_ra_data; int timeout_used = timeout_ms > MAX_START_TIMEOUT_MS ? MAX_START_TIMEOUT_MS : timeout_ms; int rc = pcmk_rc_ok; for (tmp = cmd->params; tmp; tmp = tmp->next) { if (pcmk__strcase_any_of(tmp->key, XML_RSC_ATTR_REMOTE_RA_ADDR, XML_RSC_ATTR_REMOTE_RA_SERVER, NULL)) { server = tmp->value; } else if (pcmk__str_eq(tmp->key, XML_RSC_ATTR_REMOTE_RA_PORT, pcmk__str_casei)) { port = atoi(tmp->value); } else if (pcmk__str_eq(tmp->key, CRM_META "_" XML_RSC_ATTR_CONTAINER, pcmk__str_casei)) { ra_data->controlling_guest = TRUE; } } rc = controld_connect_remote_executor(lrm_state, server, port, timeout_used); if (rc != pcmk_rc_ok) { pcmk__set_result(&(cmd->result), PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_ERROR, pcmk_rc_str(rc)); } return rc; } static gboolean handle_remote_ra_exec(gpointer user_data) { int rc = 0; lrm_state_t *lrm_state = user_data; remote_ra_data_t *ra_data = lrm_state->remote_ra_data; remote_ra_cmd_t *cmd; GList *first = NULL; if (ra_data->cur_cmd) { /* still waiting on previous cmd */ return TRUE; } while (ra_data->cmds) { first = ra_data->cmds; cmd = first->data; if (cmd->delay_id) { /* still waiting for start delay timer to trip */ return TRUE; } ra_data->cmds = g_list_remove_link(ra_data->cmds, first); g_list_free_1(first); if (!strcmp(cmd->action, "start") || !strcmp(cmd->action, "migrate_from")) { ra_data->migrate_status = 0; if (handle_remote_ra_start(lrm_state, cmd, cmd->timeout) == pcmk_rc_ok) { /* take care of this later when we get async connection result */ crm_debug("Initiated async remote connection, %s action will complete after connect event", cmd->action); ra_data->cur_cmd = cmd; return TRUE; } report_remote_ra_result(cmd); } else if (!strcmp(cmd->action, "monitor")) { if (lrm_state_is_connected(lrm_state) == TRUE) { rc = lrm_state_poke_connection(lrm_state); if (rc < 0) { pcmk__set_result(&(cmd->result), PCMK_OCF_UNKNOWN_ERROR, - PCMK_EXEC_ERROR, NULL); + PCMK_EXEC_ERROR, pcmk_strerror(rc)); } } else { rc = -1; pcmk__set_result(&(cmd->result), PCMK_OCF_NOT_RUNNING, - PCMK_EXEC_DONE, NULL); + PCMK_EXEC_DONE, "Remote connection inactive"); } if (rc == 0) { crm_debug("Poked Pacemaker Remote at node %s, waiting for async response", cmd->rsc_id); ra_data->cur_cmd = cmd; cmd->monitor_timeout_id = g_timeout_add(cmd->timeout, monitor_timeout_cb, cmd); return TRUE; } report_remote_ra_result(cmd); } else if (!strcmp(cmd->action, "stop")) { if (ra_data->migrate_status == expect_takeover) { /* briefly wait on stop for the takeover event to occur. If the * takeover event does not occur during the wait period, that's fine. * It just means that the remote-node's lrm_status section is going to get * cleared which will require all the resources running in the remote-node * to be explicitly re-detected via probe actions. If the takeover does occur * successfully, then we can leave the status section intact. */ cmd->takeover_timeout_id = g_timeout_add((cmd->timeout/2), connection_takeover_timeout_cb, cmd); ra_data->cur_cmd = cmd; return TRUE; } handle_remote_ra_stop(lrm_state, cmd); } else if (!strcmp(cmd->action, "migrate_to")) { ra_data->migrate_status = expect_takeover; pcmk__set_result(&(cmd->result), PCMK_OCF_OK, PCMK_EXEC_DONE, NULL); report_remote_ra_result(cmd); } else if (pcmk__str_any_of(cmd->action, CRMD_ACTION_RELOAD, CRMD_ACTION_RELOAD_AGENT, NULL)) { /* Currently the only reloadable parameter is reconnect_interval, * which is only used by the scheduler via the CIB, so reloads are a * no-op. * * @COMPAT DC <2.1.0: We only need to check for "reload" in case * we're in a rolling upgrade with a DC scheduling "reload" instead * of "reload-agent". An OCF 1.1 "reload" would be a no-op anyway, * so this would work for that purpose as well. */ pcmk__set_result(&(cmd->result), PCMK_OCF_OK, PCMK_EXEC_DONE, NULL); report_remote_ra_result(cmd); } free_cmd(cmd); } return TRUE; } static void remote_ra_data_init(lrm_state_t * lrm_state) { remote_ra_data_t *ra_data = NULL; if (lrm_state->remote_ra_data) { return; } ra_data = calloc(1, sizeof(remote_ra_data_t)); ra_data->work = mainloop_add_trigger(G_PRIORITY_HIGH, handle_remote_ra_exec, lrm_state); lrm_state->remote_ra_data = ra_data; } void remote_ra_cleanup(lrm_state_t * lrm_state) { remote_ra_data_t *ra_data = lrm_state->remote_ra_data; if (!ra_data) { return; } if (ra_data->cmds) { g_list_free_full(ra_data->cmds, free_cmd); } if (ra_data->recurring_cmds) { g_list_free_full(ra_data->recurring_cmds, free_cmd); } mainloop_destroy_trigger(ra_data->work); free(ra_data); lrm_state->remote_ra_data = NULL; } gboolean is_remote_lrmd_ra(const char *agent, const char *provider, const char *id) { if (agent && provider && !strcmp(agent, REMOTE_LRMD_RA) && !strcmp(provider, "pacemaker")) { return TRUE; } if (id && lrm_state_find(id) && !pcmk__str_eq(id, fsa_our_uname, pcmk__str_casei)) { return TRUE; } return FALSE; } lrmd_rsc_info_t * remote_ra_get_rsc_info(lrm_state_t * lrm_state, const char *rsc_id) { lrmd_rsc_info_t *info = NULL; if ((lrm_state_find(rsc_id))) { info = calloc(1, sizeof(lrmd_rsc_info_t)); info->id = strdup(rsc_id); info->type = strdup(REMOTE_LRMD_RA); info->standard = strdup(PCMK_RESOURCE_CLASS_OCF); info->provider = strdup("pacemaker"); } return info; } static gboolean is_remote_ra_supported_action(const char *action) { return pcmk__str_any_of(action, CRMD_ACTION_START, CRMD_ACTION_STOP, CRMD_ACTION_STATUS, CRMD_ACTION_MIGRATE, CRMD_ACTION_MIGRATED, CRMD_ACTION_RELOAD_AGENT, CRMD_ACTION_RELOAD, NULL); } static GList * fail_all_monitor_cmds(GList * list) { GList *rm_list = NULL; remote_ra_cmd_t *cmd = NULL; GList *gIter = NULL; for (gIter = list; gIter != NULL; gIter = gIter->next) { cmd = gIter->data; if ((cmd->interval_ms > 0) && pcmk__str_eq(cmd->action, "monitor", pcmk__str_casei)) { rm_list = g_list_append(rm_list, cmd); } } for (gIter = rm_list; gIter != NULL; gIter = gIter->next) { cmd = gIter->data; pcmk__set_result(&(cmd->result), PCMK_OCF_UNKNOWN_ERROR, - PCMK_EXEC_ERROR, NULL); + PCMK_EXEC_ERROR, "Lost connection to remote executor"); crm_trace("Pre-emptively failing %s %s (interval=%u, %s)", cmd->action, cmd->rsc_id, cmd->interval_ms, cmd->userdata); report_remote_ra_result(cmd); list = g_list_remove(list, cmd); free_cmd(cmd); } /* frees only the list data, not the cmds */ g_list_free(rm_list); return list; } static GList * remove_cmd(GList * list, const char *action, guint interval_ms) { remote_ra_cmd_t *cmd = NULL; GList *gIter = NULL; for (gIter = list; gIter != NULL; gIter = gIter->next) { cmd = gIter->data; if ((cmd->interval_ms == interval_ms) && pcmk__str_eq(cmd->action, action, pcmk__str_casei)) { break; } cmd = NULL; } if (cmd) { list = g_list_remove(list, cmd); free_cmd(cmd); } return list; } int remote_ra_cancel(lrm_state_t *lrm_state, const char *rsc_id, const char *action, guint interval_ms) { lrm_state_t *connection_rsc = NULL; remote_ra_data_t *ra_data = NULL; connection_rsc = lrm_state_find(rsc_id); if (!connection_rsc || !connection_rsc->remote_ra_data) { return -EINVAL; } ra_data = connection_rsc->remote_ra_data; ra_data->cmds = remove_cmd(ra_data->cmds, action, interval_ms); ra_data->recurring_cmds = remove_cmd(ra_data->recurring_cmds, action, interval_ms); if (ra_data->cur_cmd && (ra_data->cur_cmd->interval_ms == interval_ms) && (pcmk__str_eq(ra_data->cur_cmd->action, action, pcmk__str_casei))) { ra_data->cur_cmd->cancel = TRUE; } return 0; } static remote_ra_cmd_t * handle_dup_monitor(remote_ra_data_t *ra_data, guint interval_ms, const char *userdata) { GList *gIter = NULL; remote_ra_cmd_t *cmd = NULL; /* there are 3 places a potential duplicate monitor operation * could exist. * 1. recurring_cmds list. where the op is waiting for its next interval * 2. cmds list, where the op is queued to get executed immediately * 3. cur_cmd, which means the monitor op is in flight right now. */ if (interval_ms == 0) { return NULL; } if (ra_data->cur_cmd && ra_data->cur_cmd->cancel == FALSE && (ra_data->cur_cmd->interval_ms == interval_ms) && pcmk__str_eq(ra_data->cur_cmd->action, "monitor", pcmk__str_casei)) { cmd = ra_data->cur_cmd; goto handle_dup; } for (gIter = ra_data->recurring_cmds; gIter != NULL; gIter = gIter->next) { cmd = gIter->data; if ((cmd->interval_ms == interval_ms) && pcmk__str_eq(cmd->action, "monitor", pcmk__str_casei)) { goto handle_dup; } } for (gIter = ra_data->cmds; gIter != NULL; gIter = gIter->next) { cmd = gIter->data; if ((cmd->interval_ms == interval_ms) && pcmk__str_eq(cmd->action, "monitor", pcmk__str_casei)) { goto handle_dup; } } return NULL; handle_dup: crm_trace("merging duplicate monitor cmd " PCMK__OP_FMT, cmd->rsc_id, "monitor", interval_ms); /* update the userdata */ if (userdata) { free(cmd->userdata); cmd->userdata = strdup(userdata); } /* if we've already reported success, generate a new call id */ if (cmd->reported_success) { cmd->start_time = time(NULL); cmd->call_id = generate_callid(); cmd->reported_success = 0; } /* if we have an interval_id set, that means we are in the process of * waiting for this cmd's next interval. instead of waiting, cancel * the timer and execute the action immediately */ if (cmd->interval_id) { g_source_remove(cmd->interval_id); cmd->interval_id = 0; recurring_helper(cmd); } return cmd; } /*! * \internal * \brief Execute an action using the (internal) ocf:pacemaker:remote agent * * \param[in] lrm_state Executor state object for remote connection * \param[in] rsc_id Connection resource ID * \param[in] action Action to execute * \param[in] userdata String to copy and pass to execution callback * \param[in] interval_ms Action interval (in milliseconds) * \param[in] timeout_ms Action timeout (in milliseconds) * \param[in] start_delay_ms Delay (in milliseconds) before initiating action * \param[in] params Connection resource parameters * \param[out] call_id Where to store call ID on success * * \return Standard Pacemaker return code * \note This takes ownership of \p params, which should not be used or freed * after calling this function. */ int controld_execute_remote_agent(lrm_state_t *lrm_state, const char *rsc_id, const char *action, const char *userdata, guint interval_ms, int timeout_ms, int start_delay_ms, lrmd_key_value_t *params, int *call_id) { lrm_state_t *connection_rsc = NULL; remote_ra_cmd_t *cmd = NULL; remote_ra_data_t *ra_data = NULL; *call_id = 0; CRM_CHECK((lrm_state != NULL) && (rsc_id != NULL) && (action != NULL) && (userdata != NULL) && (call_id != NULL), lrmd_key_value_freeall(params); return EINVAL); if (!is_remote_ra_supported_action(action)) { lrmd_key_value_freeall(params); return EOPNOTSUPP; } connection_rsc = lrm_state_find(rsc_id); if (connection_rsc == NULL) { lrmd_key_value_freeall(params); return ENOTCONN; } remote_ra_data_init(connection_rsc); ra_data = connection_rsc->remote_ra_data; cmd = handle_dup_monitor(ra_data, interval_ms, userdata); if (cmd) { *call_id = cmd->call_id; lrmd_key_value_freeall(params); return pcmk_rc_ok; } cmd = calloc(1, sizeof(remote_ra_cmd_t)); if (cmd == NULL) { lrmd_key_value_freeall(params); return ENOMEM; } cmd->owner = strdup(lrm_state->node_name); cmd->rsc_id = strdup(rsc_id); cmd->action = strdup(action); cmd->userdata = strdup(userdata); if ((cmd->owner == NULL) || (cmd->rsc_id == NULL) || (cmd->action == NULL) || (cmd->userdata == NULL)) { free_cmd(cmd); lrmd_key_value_freeall(params); return ENOMEM; } cmd->interval_ms = interval_ms; cmd->timeout = timeout_ms; cmd->start_delay = start_delay_ms; cmd->params = params; cmd->start_time = time(NULL); cmd->call_id = generate_callid(); if (cmd->start_delay) { cmd->delay_id = g_timeout_add(cmd->start_delay, start_delay_helper, cmd); } ra_data->cmds = g_list_append(ra_data->cmds, cmd); mainloop_set_trigger(ra_data->work); *call_id = cmd->call_id; return pcmk_rc_ok; } /*! * \internal * \brief Immediately fail all monitors of a remote node, if proxied here * * \param[in] node_name Name of pacemaker_remote node */ void remote_ra_fail(const char *node_name) { lrm_state_t *lrm_state = lrm_state_find(node_name); if (lrm_state && lrm_state_is_connected(lrm_state)) { remote_ra_data_t *ra_data = lrm_state->remote_ra_data; crm_info("Failing monitors on Pacemaker Remote node %s", node_name); ra_data->recurring_cmds = fail_all_monitor_cmds(ra_data->recurring_cmds); ra_data->cmds = fail_all_monitor_cmds(ra_data->cmds); } } /* A guest node fencing implied by host fencing looks like: * * * * * * * */ #define XPATH_PSEUDO_FENCE "//" XML_GRAPH_TAG_PSEUDO_EVENT \ "[@" XML_LRM_ATTR_TASK "='stonith']/" XML_GRAPH_TAG_DOWNED \ "/" XML_CIB_TAG_NODE /*! * \internal * \brief Check a pseudo-action for Pacemaker Remote node side effects * * \param[in] xml XML of pseudo-action to check */ void remote_ra_process_pseudo(xmlNode *xml) { xmlXPathObjectPtr search = xpath_search(xml, XPATH_PSEUDO_FENCE); if (numXpathResults(search) == 1) { xmlNode *result = getXpathResult(search, 0); /* Normally, we handle the necessary side effects of a guest node stop * action when reporting the remote agent's result. However, if the stop * is implied due to fencing, it will be a fencing pseudo-event, and * there won't be a result to report. Handle that case here. * * This will result in a duplicate call to remote_node_down() if the * guest stop was real instead of implied, but that shouldn't hurt. * * There is still one corner case that isn't handled: if a guest node * isn't running any resources when its host is fenced, it will appear * to be cleanly stopped, so there will be no pseudo-fence, and our * peer cache state will be incorrect unless and until the guest is * recovered. */ if (result) { const char *remote = ID(result); if (remote) { remote_node_down(remote, DOWN_ERASE_LRM); } } } freeXpathObject(search); } static void remote_ra_maintenance(lrm_state_t * lrm_state, gboolean maintenance) { remote_ra_data_t *ra_data = lrm_state->remote_ra_data; xmlNode *update, *state; int call_opt, call_id = 0; crm_node_t *node; call_opt = crmd_cib_smart_opt(); node = crm_remote_peer_get(lrm_state->node_name); CRM_CHECK(node != NULL, return); update = create_xml_node(NULL, XML_CIB_TAG_STATUS); state = create_node_state_update(node, node_update_none, update, __func__); crm_xml_add(state, XML_NODE_IS_MAINTENANCE, maintenance?"1":"0"); fsa_cib_update(XML_CIB_TAG_STATUS, update, call_opt, call_id, NULL); if (call_id < 0) { crm_perror(LOG_WARNING, "%s CIB node state update failed", lrm_state->node_name); } else { /* TODO: still not 100% sure that async update will succeed ... */ ra_data->is_maintenance = maintenance; } free_xml(update); } #define XPATH_PSEUDO_MAINTENANCE "//" XML_GRAPH_TAG_PSEUDO_EVENT \ "[@" XML_LRM_ATTR_TASK "='" CRM_OP_MAINTENANCE_NODES "']/" \ XML_GRAPH_TAG_MAINTENANCE /*! * \internal * \brief Check a pseudo-action holding updates for maintenance state * * \param[in] xml XML of pseudo-action to check */ void remote_ra_process_maintenance_nodes(xmlNode *xml) { xmlXPathObjectPtr search = xpath_search(xml, XPATH_PSEUDO_MAINTENANCE); if (numXpathResults(search) == 1) { xmlNode *node; int cnt = 0, cnt_remote = 0; for (node = first_named_child(getXpathResult(search, 0), XML_CIB_TAG_NODE); node != NULL; node = pcmk__xml_next(node)) { lrm_state_t *lrm_state = lrm_state_find(ID(node)); cnt++; if (lrm_state && lrm_state->remote_ra_data && ((remote_ra_data_t *) lrm_state->remote_ra_data)->active) { int is_maint; cnt_remote++; pcmk__scan_min_int(crm_element_value(node, XML_NODE_IS_MAINTENANCE), &is_maint, 0); remote_ra_maintenance(lrm_state, is_maint); } } crm_trace("Action holds %d nodes (%d remotes found) " "adjusting maintenance-mode", cnt, cnt_remote); } freeXpathObject(search); } gboolean remote_ra_is_in_maintenance(lrm_state_t * lrm_state) { remote_ra_data_t *ra_data = lrm_state->remote_ra_data; return ra_data->is_maintenance; } gboolean remote_ra_controlling_guest(lrm_state_t * lrm_state) { remote_ra_data_t *ra_data = lrm_state->remote_ra_data; return ra_data->controlling_guest; } diff --git a/daemons/execd/execd_commands.c b/daemons/execd/execd_commands.c index 5ca4735704..3390518759 100644 --- a/daemons/execd/execd_commands.c +++ b/daemons/execd/execd_commands.c @@ -1,2000 +1,2003 @@ /* * Copyright 2012-2021 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 // Check whether we have a high-resolution monotonic clock #undef PCMK__TIME_USE_CGT #if HAVE_DECL_CLOCK_MONOTONIC && defined(CLOCK_MONOTONIC) # define PCMK__TIME_USE_CGT # include /* clock_gettime */ #endif #include #include #include #include #include #include #include #include #include "pacemaker-execd.h" #define EXIT_REASON_MAX_LEN 128 GHashTable *rsc_list = NULL; typedef struct lrmd_cmd_s { int timeout; guint interval_ms; int start_delay; int timeout_orig; int call_id; int call_opts; /* Timer ids, must be removed on cmd destruction. */ int delay_id; int stonith_recurring_id; int rsc_deleted; int service_flags; char *client_id; char *origin; char *rsc_id; char *action; char *real_action; char *userdata_str; pcmk__action_result_t result; /* We can track operation queue time and run time, to be saved with the CIB * resource history (and displayed in cluster status). We need * high-resolution monotonic time for this purpose, so we use * clock_gettime(CLOCK_MONOTONIC, ...) (if available, otherwise this feature * is disabled). * * However, we also need epoch timestamps for recording the time the command * last ran and the time its return value last changed, for use in time * displays (as opposed to interval calculations). We keep time_t values for * this purpose. * * The last run time is used for both purposes, so we keep redundant * monotonic and epoch values for this. Technically the two could represent * different times, but since time_t has only second resolution and the * values are used for distinct purposes, that is not significant. */ #ifdef PCMK__TIME_USE_CGT /* Recurring and systemd operations may involve more than one executor * command per operation, so they need info about the original and the most * recent. */ struct timespec t_first_run; // When op first ran struct timespec t_run; // When op most recently ran struct timespec t_first_queue; // When op was first queued struct timespec t_queue; // When op was most recently queued #endif time_t epoch_last_run; // Epoch timestamp of when op last ran time_t epoch_rcchange; // Epoch timestamp of when rc last changed bool first_notify_sent; int last_notify_rc; int last_notify_op_status; int last_pid; GHashTable *params; } lrmd_cmd_t; static void cmd_finalize(lrmd_cmd_t * cmd, lrmd_rsc_t * rsc); static gboolean lrmd_rsc_dispatch(gpointer user_data); static void cancel_all_recurring(lrmd_rsc_t * rsc, const char *client_id); #ifdef PCMK__TIME_USE_CGT /*! * \internal * \brief Check whether a struct timespec has been set * * \param[in] timespec Time to check * * \return true if timespec has been set (i.e. is nonzero), false otherwise */ static inline bool time_is_set(struct timespec *timespec) { return (timespec != NULL) && ((timespec->tv_sec != 0) || (timespec->tv_nsec != 0)); } /* * \internal * \brief Set a timespec (and its original if unset) to the current time * * \param[out] t_current Where to store current time * \param[out] t_orig Where to copy t_current if unset */ static void get_current_time(struct timespec *t_current, struct timespec *t_orig) { clock_gettime(CLOCK_MONOTONIC, t_current); if ((t_orig != NULL) && !time_is_set(t_orig)) { *t_orig = *t_current; } } /*! * \internal * \brief Return difference between two times in milliseconds * * \param[in] now More recent time (or NULL to use current time) * \param[in] old Earlier time * * \return milliseconds difference (or 0 if old is NULL or unset) * * \note Can overflow on 32bit machines when the differences is around * 24 days or more. */ static int time_diff_ms(struct timespec *now, struct timespec *old) { int diff_ms = 0; if (time_is_set(old)) { struct timespec local_now = { 0, }; if (now == NULL) { clock_gettime(CLOCK_MONOTONIC, &local_now); now = &local_now; } diff_ms = (now->tv_sec - old->tv_sec) * 1000 + (now->tv_nsec - old->tv_nsec) / 1000000; } return diff_ms; } /*! * \internal * \brief Reset a command's operation times to their original values. * * Reset a command's run and queued timestamps to the timestamps of the original * command, so we report the entire time since then and not just the time since * the most recent command (for recurring and systemd operations). * * \param[in] cmd Executor command object to reset * * \note It's not obvious what the queued time should be for a systemd * start/stop operation, which might go like this: * initial command queued 5ms, runs 3s * monitor command queued 10ms, runs 10s * monitor command queued 10ms, runs 10s * Is the queued time for that operation 5ms, 10ms or 25ms? The current * implementation will report 5ms. If it's 25ms, then we need to * subtract 20ms from the total exec time so as not to count it twice. * We can implement that later if it matters to anyone ... */ static void cmd_original_times(lrmd_cmd_t * cmd) { cmd->t_run = cmd->t_first_run; cmd->t_queue = cmd->t_first_queue; } #endif static inline bool action_matches(lrmd_cmd_t *cmd, const char *action, guint interval_ms) { return (cmd->interval_ms == interval_ms) && pcmk__str_eq(cmd->action, action, pcmk__str_casei); } static void log_finished(lrmd_cmd_t * cmd, int exec_time, int queue_time) { char pid_str[32] = { 0, }; int log_level = LOG_INFO; if (cmd->last_pid) { snprintf(pid_str, 32, "%d", cmd->last_pid); } if (pcmk__str_eq(cmd->action, "monitor", pcmk__str_casei)) { log_level = LOG_DEBUG; } #ifdef PCMK__TIME_USE_CGT do_crm_log(log_level, "%s %s (call %d%s%s) exited with status %d" " (execution time %dms, queue time %dms)", cmd->rsc_id, cmd->action, cmd->call_id, (cmd->last_pid? ", PID " : ""), pid_str, cmd->result.exit_status, exec_time, queue_time); #else do_crm_log(log_level, "%s %s (call %d%s%s) exited with status %d", cmd->rsc_id, cmd->action, cmd->call_id, (cmd->last_pid? ", PID " : ""), pid_str, cmd->result.exit_status); #endif } static void log_execute(lrmd_cmd_t * cmd) { int log_level = LOG_INFO; if (pcmk__str_eq(cmd->action, "monitor", pcmk__str_casei)) { log_level = LOG_DEBUG; } do_crm_log(log_level, "executing - rsc:%s action:%s call_id:%d", cmd->rsc_id, cmd->action, cmd->call_id); } static const char * normalize_action_name(lrmd_rsc_t * rsc, const char *action) { if (pcmk__str_eq(action, "monitor", pcmk__str_casei) && pcmk_is_set(pcmk_get_ra_caps(rsc->class), pcmk_ra_cap_status)) { return "status"; } return action; } static lrmd_rsc_t * build_rsc_from_xml(xmlNode * msg) { xmlNode *rsc_xml = get_xpath_object("//" F_LRMD_RSC, msg, LOG_ERR); lrmd_rsc_t *rsc = NULL; rsc = calloc(1, sizeof(lrmd_rsc_t)); crm_element_value_int(msg, F_LRMD_CALLOPTS, &rsc->call_opts); rsc->rsc_id = crm_element_value_copy(rsc_xml, F_LRMD_RSC_ID); rsc->class = crm_element_value_copy(rsc_xml, F_LRMD_CLASS); rsc->provider = crm_element_value_copy(rsc_xml, F_LRMD_PROVIDER); rsc->type = crm_element_value_copy(rsc_xml, F_LRMD_TYPE); rsc->work = mainloop_add_trigger(G_PRIORITY_HIGH, lrmd_rsc_dispatch, rsc); rsc->st_probe_rc = -ENODEV; // if stonith, initialize to "not running" return rsc; } static lrmd_cmd_t * create_lrmd_cmd(xmlNode *msg, pcmk__client_t *client) { int call_options = 0; xmlNode *rsc_xml = get_xpath_object("//" F_LRMD_RSC, msg, LOG_ERR); lrmd_cmd_t *cmd = NULL; cmd = calloc(1, sizeof(lrmd_cmd_t)); crm_element_value_int(msg, F_LRMD_CALLOPTS, &call_options); cmd->call_opts = call_options; cmd->client_id = strdup(client->id); crm_element_value_int(msg, F_LRMD_CALLID, &cmd->call_id); crm_element_value_ms(rsc_xml, F_LRMD_RSC_INTERVAL, &cmd->interval_ms); crm_element_value_int(rsc_xml, F_LRMD_TIMEOUT, &cmd->timeout); crm_element_value_int(rsc_xml, F_LRMD_RSC_START_DELAY, &cmd->start_delay); cmd->timeout_orig = cmd->timeout; cmd->origin = crm_element_value_copy(rsc_xml, F_LRMD_ORIGIN); cmd->action = crm_element_value_copy(rsc_xml, F_LRMD_RSC_ACTION); cmd->userdata_str = crm_element_value_copy(rsc_xml, F_LRMD_RSC_USERDATA_STR); cmd->rsc_id = crm_element_value_copy(rsc_xml, F_LRMD_RSC_ID); cmd->params = xml2list(rsc_xml); if (pcmk__str_eq(g_hash_table_lookup(cmd->params, "CRM_meta_on_fail"), "block", pcmk__str_casei)) { crm_debug("Setting flag to leave pid group on timeout and " "only kill action pid for " PCMK__OP_FMT, cmd->rsc_id, cmd->action, cmd->interval_ms); cmd->service_flags = pcmk__set_flags_as(__func__, __LINE__, LOG_TRACE, "Action", cmd->action, 0, SVC_ACTION_LEAVE_GROUP, "SVC_ACTION_LEAVE_GROUP"); } return cmd; } static void stop_recurring_timer(lrmd_cmd_t *cmd) { if (cmd) { if (cmd->stonith_recurring_id) { g_source_remove(cmd->stonith_recurring_id); } cmd->stonith_recurring_id = 0; } } static void free_lrmd_cmd(lrmd_cmd_t * cmd) { stop_recurring_timer(cmd); if (cmd->delay_id) { g_source_remove(cmd->delay_id); } if (cmd->params) { g_hash_table_destroy(cmd->params); } pcmk__reset_result(&(cmd->result)); free(cmd->origin); free(cmd->action); free(cmd->real_action); free(cmd->userdata_str); free(cmd->rsc_id); free(cmd->client_id); free(cmd); } static gboolean stonith_recurring_op_helper(gpointer data) { lrmd_cmd_t *cmd = data; lrmd_rsc_t *rsc; cmd->stonith_recurring_id = 0; if (!cmd->rsc_id) { return FALSE; } rsc = g_hash_table_lookup(rsc_list, cmd->rsc_id); CRM_ASSERT(rsc != NULL); /* take it out of recurring_ops list, and put it in the pending ops * to be executed */ rsc->recurring_ops = g_list_remove(rsc->recurring_ops, cmd); rsc->pending_ops = g_list_append(rsc->pending_ops, cmd); #ifdef PCMK__TIME_USE_CGT get_current_time(&(cmd->t_queue), &(cmd->t_first_queue)); #endif mainloop_set_trigger(rsc->work); return FALSE; } static inline void start_recurring_timer(lrmd_cmd_t *cmd) { if (cmd && (cmd->interval_ms > 0)) { cmd->stonith_recurring_id = g_timeout_add(cmd->interval_ms, stonith_recurring_op_helper, cmd); } } static gboolean start_delay_helper(gpointer data) { lrmd_cmd_t *cmd = data; lrmd_rsc_t *rsc = NULL; cmd->delay_id = 0; rsc = cmd->rsc_id ? g_hash_table_lookup(rsc_list, cmd->rsc_id) : NULL; if (rsc) { mainloop_set_trigger(rsc->work); } return FALSE; } /*! * \internal * \brief Check whether a list already contains the equivalent of a given action */ static lrmd_cmd_t * find_duplicate_action(GList *action_list, lrmd_cmd_t *cmd) { for (GList *item = action_list; item != NULL; item = item->next) { lrmd_cmd_t *dup = item->data; if (action_matches(cmd, dup->action, dup->interval_ms)) { return dup; } } return NULL; } static bool merge_recurring_duplicate(lrmd_rsc_t * rsc, lrmd_cmd_t * cmd) { lrmd_cmd_t * dup = NULL; bool dup_pending = true; if (cmd->interval_ms == 0) { return false; } // Search for a duplicate of this action (in-flight or not) dup = find_duplicate_action(rsc->pending_ops, cmd); if (dup == NULL) { dup_pending = false; dup = find_duplicate_action(rsc->recurring_ops, cmd); if (dup == NULL) { return false; } } /* Do not merge fencing monitors marked for cancellation, so we can reply to * the cancellation separately. */ if (pcmk__str_eq(rsc->class, PCMK_RESOURCE_CLASS_STONITH, pcmk__str_casei) && (dup->result.exec_status == PCMK_EXEC_CANCELLED)) { return false; } /* This should not occur. If it does, we need to investigate how something * like this is possible in the controller. */ crm_warn("Duplicate recurring op entry detected (" PCMK__OP_FMT "), merging with previous op entry", rsc->rsc_id, normalize_action_name(rsc, dup->action), dup->interval_ms); // Merge new action's call ID and user data into existing action dup->first_notify_sent = false; free(dup->userdata_str); dup->userdata_str = cmd->userdata_str; cmd->userdata_str = NULL; dup->call_id = cmd->call_id; free_lrmd_cmd(cmd); cmd = NULL; /* If dup is not pending, that means it has already executed at least once * and is waiting in the interval. In that case, stop waiting and initiate * a new instance now. */ if (!dup_pending) { if (pcmk__str_eq(rsc->class, PCMK_RESOURCE_CLASS_STONITH, pcmk__str_casei)) { stop_recurring_timer(dup); stonith_recurring_op_helper(dup); } else { services_action_kick(rsc->rsc_id, normalize_action_name(rsc, dup->action), dup->interval_ms); } } return true; } static void schedule_lrmd_cmd(lrmd_rsc_t * rsc, lrmd_cmd_t * cmd) { CRM_CHECK(cmd != NULL, return); CRM_CHECK(rsc != NULL, return); crm_trace("Scheduling %s on %s", cmd->action, rsc->rsc_id); if (merge_recurring_duplicate(rsc, cmd)) { // Equivalent of cmd has already been scheduled return; } /* The controller expects the executor to automatically cancel * recurring operations before a resource stops. */ if (pcmk__str_eq(cmd->action, "stop", pcmk__str_casei)) { cancel_all_recurring(rsc, NULL); } rsc->pending_ops = g_list_append(rsc->pending_ops, cmd); #ifdef PCMK__TIME_USE_CGT get_current_time(&(cmd->t_queue), &(cmd->t_first_queue)); #endif mainloop_set_trigger(rsc->work); if (cmd->start_delay) { cmd->delay_id = g_timeout_add(cmd->start_delay, start_delay_helper, cmd); } } static xmlNode * create_lrmd_reply(const char *origin, int rc, int call_id) { xmlNode *reply = create_xml_node(NULL, T_LRMD_REPLY); crm_xml_add(reply, F_LRMD_ORIGIN, origin); crm_xml_add_int(reply, F_LRMD_RC, rc); crm_xml_add_int(reply, F_LRMD_CALLID, call_id); return reply; } static void send_client_notify(gpointer key, gpointer value, gpointer user_data) { xmlNode *update_msg = user_data; pcmk__client_t *client = value; int rc; int log_level = LOG_WARNING; const char *msg = NULL; CRM_CHECK(client != NULL, return); if (client->name == NULL) { crm_trace("Skipping notification to client without name"); return; } if (pcmk_is_set(client->flags, pcmk__client_to_proxy)) { /* We only want to notify clients of the executor IPC API. If we are * running as Pacemaker Remote, we may have clients proxied to other * IPC services in the cluster, so skip those. */ crm_trace("Skipping executor API notification to client %s", pcmk__client_name(client)); return; } rc = lrmd_server_send_notify(client, update_msg); if (rc == pcmk_rc_ok) { return; } switch (rc) { case ENOTCONN: case EPIPE: // Client exited without waiting for notification log_level = LOG_INFO; msg = "Disconnected"; break; default: msg = pcmk_rc_str(rc); break; } do_crm_log(log_level, "Could not notify client %s: %s " CRM_XS " rc=%d", pcmk__client_name(client), msg, rc); } static void send_cmd_complete_notify(lrmd_cmd_t * cmd) { xmlNode *notify = NULL; #ifdef PCMK__TIME_USE_CGT int exec_time = time_diff_ms(NULL, &(cmd->t_run)); int queue_time = time_diff_ms(&cmd->t_run, &(cmd->t_queue)); log_finished(cmd, exec_time, queue_time); #else log_finished(cmd, 0, 0); #endif /* if the first notify result for a cmd has already been sent earlier, and the * the option to only send notifies on result changes is set. Check to see * if the last result is the same as the new one. If so, suppress this update */ if (cmd->first_notify_sent && (cmd->call_opts & lrmd_opt_notify_changes_only)) { if ((cmd->last_notify_rc == cmd->result.exit_status) && (cmd->last_notify_op_status == cmd->result.exec_status)) { /* only send changes */ return; } } cmd->first_notify_sent = true; cmd->last_notify_rc = cmd->result.exit_status; cmd->last_notify_op_status = cmd->result.exec_status; notify = create_xml_node(NULL, T_LRMD_NOTIFY); crm_xml_add(notify, F_LRMD_ORIGIN, __func__); crm_xml_add_int(notify, F_LRMD_TIMEOUT, cmd->timeout); crm_xml_add_ms(notify, F_LRMD_RSC_INTERVAL, cmd->interval_ms); crm_xml_add_int(notify, F_LRMD_RSC_START_DELAY, cmd->start_delay); crm_xml_add_int(notify, F_LRMD_EXEC_RC, cmd->result.exit_status); crm_xml_add_int(notify, F_LRMD_OP_STATUS, cmd->result.exec_status); crm_xml_add_int(notify, F_LRMD_CALLID, cmd->call_id); crm_xml_add_int(notify, F_LRMD_RSC_DELETED, cmd->rsc_deleted); crm_xml_add_ll(notify, F_LRMD_RSC_RUN_TIME, (long long) cmd->epoch_last_run); crm_xml_add_ll(notify, F_LRMD_RSC_RCCHANGE_TIME, (long long) cmd->epoch_rcchange); #ifdef PCMK__TIME_USE_CGT crm_xml_add_int(notify, F_LRMD_RSC_EXEC_TIME, exec_time); crm_xml_add_int(notify, F_LRMD_RSC_QUEUE_TIME, queue_time); #endif crm_xml_add(notify, F_LRMD_OPERATION, LRMD_OP_RSC_EXEC); crm_xml_add(notify, F_LRMD_RSC_ID, cmd->rsc_id); if(cmd->real_action) { crm_xml_add(notify, F_LRMD_RSC_ACTION, cmd->real_action); } else { crm_xml_add(notify, F_LRMD_RSC_ACTION, cmd->action); } crm_xml_add(notify, F_LRMD_RSC_USERDATA_STR, cmd->userdata_str); crm_xml_add(notify, F_LRMD_RSC_EXIT_REASON, cmd->result.exit_reason); if (cmd->result.action_stderr != NULL) { crm_xml_add(notify, F_LRMD_RSC_OUTPUT, cmd->result.action_stderr); } else if (cmd->result.action_stdout != NULL) { crm_xml_add(notify, F_LRMD_RSC_OUTPUT, cmd->result.action_stdout); } if (cmd->params) { char *key = NULL; char *value = NULL; GHashTableIter iter; xmlNode *args = create_xml_node(notify, XML_TAG_ATTRS); g_hash_table_iter_init(&iter, cmd->params); while (g_hash_table_iter_next(&iter, (gpointer *) & key, (gpointer *) & value)) { hash2smartfield((gpointer) key, (gpointer) value, args); } } if (cmd->client_id && (cmd->call_opts & lrmd_opt_notify_orig_only)) { pcmk__client_t *client = pcmk__find_client_by_id(cmd->client_id); if (client) { send_client_notify(client->id, client, notify); } } else { pcmk__foreach_ipc_client(send_client_notify, notify); } free_xml(notify); } static void send_generic_notify(int rc, xmlNode * request) { if (pcmk__ipc_client_count() != 0) { int call_id = 0; xmlNode *notify = NULL; xmlNode *rsc_xml = get_xpath_object("//" F_LRMD_RSC, request, LOG_ERR); const char *rsc_id = crm_element_value(rsc_xml, F_LRMD_RSC_ID); const char *op = crm_element_value(request, F_LRMD_OPERATION); crm_element_value_int(request, F_LRMD_CALLID, &call_id); notify = create_xml_node(NULL, T_LRMD_NOTIFY); crm_xml_add(notify, F_LRMD_ORIGIN, __func__); crm_xml_add_int(notify, F_LRMD_RC, rc); crm_xml_add_int(notify, F_LRMD_CALLID, call_id); crm_xml_add(notify, F_LRMD_OPERATION, op); crm_xml_add(notify, F_LRMD_RSC_ID, rsc_id); pcmk__foreach_ipc_client(send_client_notify, notify); free_xml(notify); } } static void cmd_reset(lrmd_cmd_t * cmd) { cmd->last_pid = 0; #ifdef PCMK__TIME_USE_CGT memset(&cmd->t_run, 0, sizeof(cmd->t_run)); memset(&cmd->t_queue, 0, sizeof(cmd->t_queue)); #endif cmd->epoch_last_run = 0; pcmk__reset_result(&(cmd->result)); cmd->result.exec_status = PCMK_EXEC_DONE; } static void cmd_finalize(lrmd_cmd_t * cmd, lrmd_rsc_t * rsc) { crm_trace("Resource operation rsc:%s action:%s completed (%p %p)", cmd->rsc_id, cmd->action, rsc ? rsc->active : NULL, cmd); if (rsc && (rsc->active == cmd)) { rsc->active = NULL; mainloop_set_trigger(rsc->work); } if (!rsc) { cmd->rsc_deleted = 1; } /* reset original timeout so client notification has correct information */ cmd->timeout = cmd->timeout_orig; send_cmd_complete_notify(cmd); if ((cmd->interval_ms != 0) && (cmd->result.exec_status == PCMK_EXEC_CANCELLED)) { if (rsc) { rsc->recurring_ops = g_list_remove(rsc->recurring_ops, cmd); rsc->pending_ops = g_list_remove(rsc->pending_ops, cmd); } free_lrmd_cmd(cmd); } else if (cmd->interval_ms == 0) { if (rsc) { rsc->pending_ops = g_list_remove(rsc->pending_ops, cmd); } free_lrmd_cmd(cmd); } else { /* Clear all the values pertaining just to the last iteration of a recurring op. */ cmd_reset(cmd); } } static int ocf2uniform_rc(int rc) { switch (rc) { case PCMK_OCF_DEGRADED: case PCMK_OCF_DEGRADED_PROMOTED: break; default: if (rc < 0 || rc > PCMK_OCF_FAILED_PROMOTED) { return PCMK_OCF_UNKNOWN_ERROR; } } return rc; } static int stonith2uniform_rc(const char *action, int rc) { switch (rc) { case pcmk_ok: rc = PCMK_OCF_OK; break; case -ENODEV: /* This should be possible only for probes in practice, but * interpret for all actions to be safe. */ if (pcmk__str_eq(action, "monitor", pcmk__str_casei)) { rc = PCMK_OCF_NOT_RUNNING; } else if (pcmk__str_eq(action, "stop", pcmk__str_casei)) { rc = PCMK_OCF_OK; } else { rc = PCMK_OCF_NOT_INSTALLED; } break; case -EOPNOTSUPP: rc = PCMK_OCF_UNIMPLEMENT_FEATURE; break; case -ETIME: case -ETIMEDOUT: rc = PCMK_OCF_TIMEOUT; break; default: rc = PCMK_OCF_UNKNOWN_ERROR; break; } return rc; } #if SUPPORT_NAGIOS static int nagios2uniform_rc(const char *action, int rc) { if (rc < 0) { return PCMK_OCF_UNKNOWN_ERROR; } switch (rc) { case NAGIOS_STATE_OK: return PCMK_OCF_OK; case NAGIOS_INSUFFICIENT_PRIV: return PCMK_OCF_INSUFFICIENT_PRIV; case NAGIOS_NOT_INSTALLED: return PCMK_OCF_NOT_INSTALLED; case NAGIOS_STATE_WARNING: case NAGIOS_STATE_CRITICAL: case NAGIOS_STATE_UNKNOWN: case NAGIOS_STATE_DEPENDENT: default: return PCMK_OCF_UNKNOWN_ERROR; } return PCMK_OCF_UNKNOWN_ERROR; } #endif static int get_uniform_rc(const char *standard, const char *action, int rc) { if (pcmk__str_eq(standard, PCMK_RESOURCE_CLASS_OCF, pcmk__str_casei)) { return ocf2uniform_rc(rc); } else if (pcmk__str_eq(standard, PCMK_RESOURCE_CLASS_STONITH, pcmk__str_casei)) { return stonith2uniform_rc(action, rc); } else if (pcmk__str_eq(standard, PCMK_RESOURCE_CLASS_SYSTEMD, pcmk__str_casei)) { return rc; } else if (pcmk__str_eq(standard, PCMK_RESOURCE_CLASS_UPSTART, pcmk__str_casei)) { return rc; #if SUPPORT_NAGIOS } else if (pcmk__str_eq(standard, PCMK_RESOURCE_CLASS_NAGIOS, pcmk__str_casei)) { return nagios2uniform_rc(action, rc); #endif } else { return services_get_ocf_exitcode(action, rc); } } static int action_get_uniform_rc(svc_action_t * action) { lrmd_cmd_t *cmd = action->cb_data; return get_uniform_rc(action->standard, cmd->action, action->rc); } struct notify_new_client_data { xmlNode *notify; pcmk__client_t *new_client; }; static void notify_one_client(gpointer key, gpointer value, gpointer user_data) { pcmk__client_t *client = value; struct notify_new_client_data *data = user_data; if (!pcmk__str_eq(client->id, data->new_client->id, pcmk__str_casei)) { send_client_notify(key, (gpointer) client, (gpointer) data->notify); } } void notify_of_new_client(pcmk__client_t *new_client) { struct notify_new_client_data data; data.new_client = new_client; data.notify = create_xml_node(NULL, T_LRMD_NOTIFY); crm_xml_add(data.notify, F_LRMD_ORIGIN, __func__); crm_xml_add(data.notify, F_LRMD_OPERATION, LRMD_OP_NEW_CLIENT); pcmk__foreach_ipc_client(notify_one_client, &data); free_xml(data.notify); } static char * parse_exit_reason(const char *output) { const char *cur = NULL; const char *last = NULL; static int cookie_len = 0; char *eol = NULL; size_t reason_len = EXIT_REASON_MAX_LEN; if (output == NULL) { return NULL; } if (!cookie_len) { cookie_len = strlen(PCMK_OCF_REASON_PREFIX); } cur = strstr(output, PCMK_OCF_REASON_PREFIX); for (; cur != NULL; cur = strstr(cur, PCMK_OCF_REASON_PREFIX)) { /* skip over the cookie delimiter string */ cur += cookie_len; last = cur; } if (last == NULL) { return NULL; } // Truncate everything after a new line, and limit reason string size eol = strchr(last, '\n'); if (eol) { reason_len = QB_MIN(reason_len, eol - last); } return strndup(last, reason_len); } void client_disconnect_cleanup(const char *client_id) { GHashTableIter iter; lrmd_rsc_t *rsc = NULL; char *key = NULL; g_hash_table_iter_init(&iter, rsc_list); while (g_hash_table_iter_next(&iter, (gpointer *) & key, (gpointer *) & rsc)) { if (rsc->call_opts & lrmd_opt_drop_recurring) { /* This client is disconnecting, drop any recurring operations * it may have initiated on the resource */ cancel_all_recurring(rsc, client_id); } } } static void action_complete(svc_action_t * action) { lrmd_rsc_t *rsc; lrmd_cmd_t *cmd = action->cb_data; #ifdef PCMK__TIME_USE_CGT const char *rclass = NULL; bool goagain = false; #endif if (!cmd) { crm_err("Completed executor action (%s) does not match any known operations", action->id); return; } #ifdef PCMK__TIME_USE_CGT if (cmd->result.exit_status != action->rc) { cmd->epoch_rcchange = time(NULL); } #endif cmd->last_pid = action->pid; pcmk__set_result(&(cmd->result), action_get_uniform_rc(action), action->status, NULL); rsc = cmd->rsc_id ? g_hash_table_lookup(rsc_list, cmd->rsc_id) : NULL; #ifdef PCMK__TIME_USE_CGT if (rsc && pcmk__str_eq(rsc->class, PCMK_RESOURCE_CLASS_SERVICE, pcmk__str_casei)) { rclass = resources_find_service_class(rsc->type); } else if(rsc) { rclass = rsc->class; } if (pcmk__str_eq(rclass, PCMK_RESOURCE_CLASS_SYSTEMD, pcmk__str_casei)) { if ((cmd->result.exit_status == PCMK_OCF_OK) && pcmk__strcase_any_of(cmd->action, "start", "stop", NULL)) { /* systemd returns from start and stop actions after the action * begins, not after it completes. We have to jump through a few * hoops so that we don't report 'complete' to the rest of pacemaker * until it's actually done. */ goagain = true; cmd->real_action = cmd->action; cmd->action = strdup("monitor"); } else if (cmd->real_action != NULL) { // This is follow-up monitor to check whether start/stop completed if ((cmd->result.exec_status == PCMK_EXEC_DONE) && (cmd->result.exit_status == PCMK_OCF_PENDING)) { goagain = true; } else if ((cmd->result.exit_status == PCMK_OCF_OK) && pcmk__str_eq(cmd->real_action, "stop", pcmk__str_casei)) { goagain = true; } else { int time_sum = time_diff_ms(NULL, &(cmd->t_first_run)); int timeout_left = cmd->timeout_orig - time_sum; crm_debug("%s systemd %s is now complete (elapsed=%dms, " "remaining=%dms): %s (%d)", cmd->rsc_id, cmd->real_action, time_sum, timeout_left, services_ocf_exitcode_str(cmd->result.exit_status), cmd->result.exit_status); cmd_original_times(cmd); // Monitors may return "not running", but start/stop shouldn't if ((cmd->result.exec_status == PCMK_EXEC_DONE) && (cmd->result.exit_status == PCMK_OCF_NOT_RUNNING)) { if (pcmk__str_eq(cmd->real_action, "start", pcmk__str_casei)) { cmd->result.exit_status = PCMK_OCF_UNKNOWN_ERROR; } else if (pcmk__str_eq(cmd->real_action, "stop", pcmk__str_casei)) { cmd->result.exit_status = PCMK_OCF_OK; } } } } } #endif #if SUPPORT_NAGIOS if (rsc && pcmk__str_eq(rsc->class, PCMK_RESOURCE_CLASS_NAGIOS, pcmk__str_casei)) { if (action_matches(cmd, "monitor", 0) && (cmd->result.exit_status == PCMK_OCF_OK)) { /* Successfully executed --version for the nagios plugin */ cmd->result.exit_status = PCMK_OCF_NOT_RUNNING; } else if (pcmk__str_eq(cmd->action, "start", pcmk__str_casei) && (cmd->result.exit_status != PCMK_OCF_OK)) { #ifdef PCMK__TIME_USE_CGT goagain = true; #endif } } #endif #ifdef PCMK__TIME_USE_CGT if (goagain) { int time_sum = time_diff_ms(NULL, &(cmd->t_first_run)); int timeout_left = cmd->timeout_orig - time_sum; int delay = cmd->timeout_orig / 10; if(delay >= timeout_left && timeout_left > 20) { delay = timeout_left/2; } delay = QB_MIN(2000, delay); if (delay < timeout_left) { cmd->start_delay = delay; cmd->timeout = timeout_left; if (cmd->result.exit_status == PCMK_OCF_OK) { crm_debug("%s %s may still be in progress: re-scheduling (elapsed=%dms, remaining=%dms, start_delay=%dms)", cmd->rsc_id, cmd->real_action, time_sum, timeout_left, delay); } else if (cmd->result.exit_status == PCMK_OCF_PENDING) { crm_info("%s %s is still in progress: re-scheduling (elapsed=%dms, remaining=%dms, start_delay=%dms)", cmd->rsc_id, cmd->action, time_sum, timeout_left, delay); } else { crm_notice("%s %s failed '%s' (%d): re-scheduling (elapsed=%dms, remaining=%dms, start_delay=%dms)", cmd->rsc_id, cmd->action, services_ocf_exitcode_str(cmd->result.exit_status), cmd->result.exit_status, time_sum, timeout_left, delay); } cmd_reset(cmd); if(rsc) { rsc->active = NULL; } schedule_lrmd_cmd(rsc, cmd); /* Don't finalize cmd, we're not done with it yet */ return; } else { crm_notice("Giving up on %s %s (rc=%d): timeout (elapsed=%dms, remaining=%dms)", cmd->rsc_id, (cmd->real_action? cmd->real_action : cmd->action), cmd->result.exit_status, time_sum, timeout_left); pcmk__set_result(&(cmd->result), PCMK_OCF_TIMEOUT, - PCMK_EXEC_TIMEOUT, NULL); + PCMK_EXEC_TIMEOUT, + "Investigate reason for timeout, and adjust " + "configured operation timeout if necessary"); cmd_original_times(cmd); } } #endif pcmk__set_result_output(&(cmd->result), action->stdout_data, action->stderr_data); if (action->stderr_data) { cmd->result.exit_reason = parse_exit_reason(action->stderr_data); } cmd_finalize(cmd, rsc); } /*! * \internal * \brief Determine operation status of a stonith operation * * Non-stonith resource operations get their operation status directly from the * service library, but the fencer does not have an equivalent, so we must infer * an operation status from the fencer API's return code. * * \param[in] action Name of action performed on stonith resource * \param[in] interval_ms Action interval * \param[in] rc Action result from fencer * * \return Operation status corresponding to fencer API return code */ static int stonith_rc2status(const char *action, guint interval_ms, int rc) { int status = PCMK_EXEC_DONE; switch (rc) { case pcmk_ok: break; case -EOPNOTSUPP: case -EPROTONOSUPPORT: status = PCMK_EXEC_NOT_SUPPORTED; break; case -ETIME: case -ETIMEDOUT: status = PCMK_EXEC_TIMEOUT; break; case -ENOTCONN: case -ECOMM: // Couldn't talk to fencer status = PCMK_EXEC_ERROR; break; case -ENODEV: // The device is not registered with the fencer status = PCMK_EXEC_ERROR; break; default: break; } return status; } static void stonith_action_complete(lrmd_cmd_t * cmd, int rc) { // This can be NULL if resource was removed before command completed lrmd_rsc_t *rsc = g_hash_table_lookup(rsc_list, cmd->rsc_id); cmd->result.exit_status = stonith2uniform_rc(cmd->action, rc); /* This function may be called with status already set to cancelled, if a * pending action was aborted. Otherwise, we need to determine status from * the fencer return code. */ if (cmd->result.exec_status != PCMK_EXEC_CANCELLED) { cmd->result.exec_status = stonith_rc2status(cmd->action, cmd->interval_ms, rc); // Certain successful actions change the known state of the resource if ((rsc != NULL) && (cmd->result.exit_status == PCMK_OCF_OK)) { if (pcmk__str_eq(cmd->action, "start", pcmk__str_casei)) { rsc->st_probe_rc = pcmk_ok; // maps to PCMK_OCF_OK } else if (pcmk__str_eq(cmd->action, "stop", pcmk__str_casei)) { rsc->st_probe_rc = -ENODEV; // maps to PCMK_OCF_NOT_RUNNING } } } /* The recurring timer should not be running at this point in any case, but * as a failsafe, stop it if it is. */ stop_recurring_timer(cmd); /* Reschedule this command if appropriate. If a recurring command is *not* * rescheduled, its status must be PCMK_EXEC_CANCELLED, otherwise it will * not be removed from recurring_ops by cmd_finalize(). */ if (rsc && (cmd->interval_ms > 0) && (cmd->result.exec_status != PCMK_EXEC_CANCELLED)) { start_recurring_timer(cmd); } cmd_finalize(cmd, rsc); } static void lrmd_stonith_callback(stonith_t * stonith, stonith_callback_data_t * data) { stonith_action_complete(data->userdata, data->rc); } void stonith_connection_failed(void) { GHashTableIter iter; GList *cmd_list = NULL; GList *cmd_iter = NULL; lrmd_rsc_t *rsc = NULL; char *key = NULL; g_hash_table_iter_init(&iter, rsc_list); while (g_hash_table_iter_next(&iter, (gpointer *) & key, (gpointer *) & rsc)) { if (pcmk__str_eq(rsc->class, PCMK_RESOURCE_CLASS_STONITH, pcmk__str_casei)) { /* If we registered this fence device, we don't know whether the * fencer still has the registration or not. Cause future probes to * return PCMK_OCF_UNKNOWN_ERROR until the resource is stopped or * started successfully. This is especially important if the * controller also went away (possibly due to a cluster layer * restart) and won't receive our client notification of any * monitors finalized below. */ if (rsc->st_probe_rc == pcmk_ok) { rsc->st_probe_rc = pcmk_err_generic; } if (rsc->active) { cmd_list = g_list_append(cmd_list, rsc->active); } if (rsc->recurring_ops) { cmd_list = g_list_concat(cmd_list, rsc->recurring_ops); } if (rsc->pending_ops) { cmd_list = g_list_concat(cmd_list, rsc->pending_ops); } rsc->pending_ops = rsc->recurring_ops = NULL; } } if (!cmd_list) { return; } crm_err("Connection to fencer failed, finalizing %d pending operations", g_list_length(cmd_list)); for (cmd_iter = cmd_list; cmd_iter; cmd_iter = cmd_iter->next) { stonith_action_complete(cmd_iter->data, -ENOTCONN); } g_list_free(cmd_list); } /*! * \internal * \brief Execute a stonith resource "start" action * * Start a stonith resource by registering it with the fencer. * (Stonith agents don't have a start command.) * * \param[in] stonith_api Connection to fencer * \param[in] rsc Stonith resource to start * \param[in] cmd Start command to execute * * \return pcmk_ok on success, -errno otherwise */ static int execd_stonith_start(stonith_t *stonith_api, lrmd_rsc_t *rsc, lrmd_cmd_t *cmd) { char *key = NULL; char *value = NULL; stonith_key_value_t *device_params = NULL; int rc = pcmk_ok; // Convert command parameters to stonith API key/values if (cmd->params) { GHashTableIter iter; g_hash_table_iter_init(&iter, cmd->params); while (g_hash_table_iter_next(&iter, (gpointer *) & key, (gpointer *) & value)) { device_params = stonith_key_value_add(device_params, key, value); } } /* The fencer will automatically register devices via CIB notifications * when the CIB changes, but to avoid a possible race condition between * the fencer receiving the notification and the executor requesting that * resource, the executor registers the device as well. The fencer knows how * to handle duplicate registrations. */ rc = stonith_api->cmds->register_device(stonith_api, st_opt_sync_call, cmd->rsc_id, rsc->provider, rsc->type, device_params); stonith_key_value_freeall(device_params, 1, 1); return rc; } /*! * \internal * \brief Execute a stonith resource "stop" action * * Stop a stonith resource by unregistering it with the fencer. * (Stonith agents don't have a stop command.) * * \param[in] stonith_api Connection to fencer * \param[in] rsc Stonith resource to stop * * \return pcmk_ok on success, -errno otherwise */ static inline int execd_stonith_stop(stonith_t *stonith_api, const lrmd_rsc_t *rsc) { /* @TODO Failure would indicate a problem communicating with fencer; * perhaps we should try reconnecting and retrying a few times? */ return stonith_api->cmds->remove_device(stonith_api, st_opt_sync_call, rsc->rsc_id); } /*! * \internal * \brief Initiate a stonith resource agent recurring "monitor" action * * \param[in] stonith_api Connection to fencer * \param[in] rsc Stonith resource to monitor * \param[in] cmd Monitor command being executed * * \return pcmk_ok if monitor was successfully initiated, -errno otherwise */ static inline int execd_stonith_monitor(stonith_t *stonith_api, lrmd_rsc_t *rsc, lrmd_cmd_t *cmd) { int rc = stonith_api->cmds->monitor(stonith_api, 0, cmd->rsc_id, cmd->timeout / 1000); rc = stonith_api->cmds->register_callback(stonith_api, rc, 0, 0, cmd, "lrmd_stonith_callback", lrmd_stonith_callback); if (rc == TRUE) { rsc->active = cmd; rc = pcmk_ok; } else { rc = -pcmk_err_generic; } return rc; } static void lrmd_rsc_execute_stonith(lrmd_rsc_t * rsc, lrmd_cmd_t * cmd) { int rc = 0; bool do_monitor = FALSE; stonith_t *stonith_api = get_stonith_connection(); if (!stonith_api) { rc = -ENOTCONN; } else if (pcmk__str_eq(cmd->action, "start", pcmk__str_casei)) { rc = execd_stonith_start(stonith_api, rsc, cmd); if (rc == 0) { do_monitor = TRUE; } } else if (pcmk__str_eq(cmd->action, "stop", pcmk__str_casei)) { rc = execd_stonith_stop(stonith_api, rsc); } else if (pcmk__str_eq(cmd->action, "monitor", pcmk__str_casei)) { if (cmd->interval_ms > 0) { do_monitor = TRUE; } else { rc = rsc->st_probe_rc; } } if (do_monitor) { rc = execd_stonith_monitor(stonith_api, rsc, cmd); if (rc == pcmk_ok) { // Don't clean up yet, we will find out result of the monitor later return; } } stonith_action_complete(cmd, rc); } static int lrmd_rsc_execute_service_lib(lrmd_rsc_t * rsc, lrmd_cmd_t * cmd) { svc_action_t *action = NULL; GHashTable *params_copy = NULL; CRM_ASSERT(rsc); CRM_ASSERT(cmd); crm_trace("Creating action, resource:%s action:%s class:%s provider:%s agent:%s", rsc->rsc_id, cmd->action, rsc->class, rsc->provider, rsc->type); #if SUPPORT_NAGIOS /* Recurring operations are cancelled anyway for a stop operation */ if (pcmk__str_eq(rsc->class, PCMK_RESOURCE_CLASS_NAGIOS, pcmk__str_casei) && pcmk__str_eq(cmd->action, "stop", pcmk__str_casei)) { cmd->result.exit_status = PCMK_OCF_OK; goto exec_done; } #endif params_copy = pcmk__str_table_dup(cmd->params); action = services__create_resource_action(rsc->rsc_id, rsc->class, rsc->provider, rsc->type, normalize_action_name(rsc, cmd->action), cmd->interval_ms, cmd->timeout, params_copy, cmd->service_flags); if (!action) { + // Invalid arguments (which would be a bug) or out-of-memory crm_err("Failed to create action, action:%s on resource %s", cmd->action, rsc->rsc_id); pcmk__set_result(&(cmd->result), PCMK_OCF_UNKNOWN_ERROR, - PCMK_EXEC_ERROR, NULL); + PCMK_EXEC_ERROR, "Internal Pacemaker error"); goto exec_done; } if (action->rc != PCMK_OCF_OK) { pcmk__set_result(&(cmd->result), action->rc, action->status, NULL); services_action_free(action); goto exec_done; } action->cb_data = cmd; /* 'cmd' may not be valid after this point if * services_action_async() returned TRUE * * Upstart and systemd both synchronously determine monitor/status * results and call action_complete (which may free 'cmd') if necessary. */ if (services_action_async(action, action_complete)) { return TRUE; } if (action->status == PCMK_EXEC_DONE) { action->status = PCMK_EXEC_ERROR; } pcmk__set_result(&(cmd->result), action->rc, action->status, NULL); services_action_free(action); action = NULL; exec_done: cmd_finalize(cmd, rsc); return TRUE; } static gboolean lrmd_rsc_execute(lrmd_rsc_t * rsc) { lrmd_cmd_t *cmd = NULL; CRM_CHECK(rsc != NULL, return FALSE); if (rsc->active) { crm_trace("%s is still active", rsc->rsc_id); return TRUE; } if (rsc->pending_ops) { GList *first = rsc->pending_ops; cmd = first->data; if (cmd->delay_id) { crm_trace ("Command %s %s was asked to run too early, waiting for start_delay timeout of %dms", cmd->rsc_id, cmd->action, cmd->start_delay); return TRUE; } rsc->pending_ops = g_list_remove_link(rsc->pending_ops, first); g_list_free_1(first); #ifdef PCMK__TIME_USE_CGT get_current_time(&(cmd->t_run), &(cmd->t_first_run)); #endif cmd->epoch_last_run = time(NULL); } if (!cmd) { crm_trace("Nothing further to do for %s", rsc->rsc_id); return TRUE; } rsc->active = cmd; /* only one op at a time for a rsc */ if (cmd->interval_ms) { rsc->recurring_ops = g_list_append(rsc->recurring_ops, cmd); } log_execute(cmd); if (pcmk__str_eq(rsc->class, PCMK_RESOURCE_CLASS_STONITH, pcmk__str_casei)) { lrmd_rsc_execute_stonith(rsc, cmd); } else { lrmd_rsc_execute_service_lib(rsc, cmd); } return TRUE; } static gboolean lrmd_rsc_dispatch(gpointer user_data) { return lrmd_rsc_execute(user_data); } void free_rsc(gpointer data) { GList *gIter = NULL; lrmd_rsc_t *rsc = data; int is_stonith = pcmk__str_eq(rsc->class, PCMK_RESOURCE_CLASS_STONITH, pcmk__str_casei); gIter = rsc->pending_ops; while (gIter != NULL) { GList *next = gIter->next; lrmd_cmd_t *cmd = gIter->data; /* command was never executed */ cmd->result.exec_status = PCMK_EXEC_CANCELLED; cmd_finalize(cmd, NULL); gIter = next; } /* frees list, but not list elements. */ g_list_free(rsc->pending_ops); gIter = rsc->recurring_ops; while (gIter != NULL) { GList *next = gIter->next; lrmd_cmd_t *cmd = gIter->data; if (is_stonith) { cmd->result.exec_status = PCMK_EXEC_CANCELLED; /* If a stonith command is in-flight, just mark it as cancelled; * it is not safe to finalize/free the cmd until the stonith api * says it has either completed or timed out. */ if (rsc->active != cmd) { cmd_finalize(cmd, NULL); } } else { /* This command is already handed off to service library, * let service library cancel it and tell us via the callback * when it is cancelled. The rsc can be safely destroyed * even if we are waiting for the cancel result */ services_action_cancel(rsc->rsc_id, normalize_action_name(rsc, cmd->action), cmd->interval_ms); } gIter = next; } /* frees list, but not list elements. */ g_list_free(rsc->recurring_ops); free(rsc->rsc_id); free(rsc->class); free(rsc->provider); free(rsc->type); mainloop_destroy_trigger(rsc->work); free(rsc); } static int process_lrmd_signon(pcmk__client_t *client, xmlNode *request, int call_id, xmlNode **reply) { int rc = pcmk_ok; const char *is_ipc_provider = crm_element_value(request, F_LRMD_IS_IPC_PROVIDER); const char *protocol_version = crm_element_value(request, F_LRMD_PROTOCOL_VERSION); if (compare_version(protocol_version, LRMD_MIN_PROTOCOL_VERSION) < 0) { crm_err("Cluster API version must be greater than or equal to %s, not %s", LRMD_MIN_PROTOCOL_VERSION, protocol_version); rc = -EPROTO; } if (crm_is_true(is_ipc_provider)) { #ifdef PCMK__COMPILE_REMOTE if ((client->remote != NULL) && client->remote->tls_handshake_complete) { // This is a remote connection from a cluster node's controller ipc_proxy_add_provider(client); } else { rc = -EACCES; } #else rc = -EPROTONOSUPPORT; #endif } *reply = create_lrmd_reply(__func__, rc, call_id); crm_xml_add(*reply, F_LRMD_OPERATION, CRM_OP_REGISTER); crm_xml_add(*reply, F_LRMD_CLIENTID, client->id); crm_xml_add(*reply, F_LRMD_PROTOCOL_VERSION, LRMD_PROTOCOL_VERSION); return rc; } static int process_lrmd_rsc_register(pcmk__client_t *client, uint32_t id, xmlNode *request) { int rc = pcmk_ok; lrmd_rsc_t *rsc = build_rsc_from_xml(request); lrmd_rsc_t *dup = g_hash_table_lookup(rsc_list, rsc->rsc_id); if (dup && pcmk__str_eq(rsc->class, dup->class, pcmk__str_casei) && pcmk__str_eq(rsc->provider, dup->provider, pcmk__str_casei) && pcmk__str_eq(rsc->type, dup->type, pcmk__str_casei)) { crm_notice("Ignoring duplicate registration of '%s'", rsc->rsc_id); free_rsc(rsc); return rc; } g_hash_table_replace(rsc_list, rsc->rsc_id, rsc); crm_info("Cached agent information for '%s'", rsc->rsc_id); return rc; } static xmlNode * process_lrmd_get_rsc_info(xmlNode *request, int call_id) { int rc = pcmk_ok; xmlNode *rsc_xml = get_xpath_object("//" F_LRMD_RSC, request, LOG_ERR); const char *rsc_id = crm_element_value(rsc_xml, F_LRMD_RSC_ID); xmlNode *reply = NULL; lrmd_rsc_t *rsc = NULL; if (rsc_id == NULL) { rc = -ENODEV; } else { rsc = g_hash_table_lookup(rsc_list, rsc_id); if (rsc == NULL) { crm_info("Agent information for '%s' not in cache", rsc_id); rc = -ENODEV; } } reply = create_lrmd_reply(__func__, rc, call_id); if (rsc) { crm_xml_add(reply, F_LRMD_RSC_ID, rsc->rsc_id); crm_xml_add(reply, F_LRMD_CLASS, rsc->class); crm_xml_add(reply, F_LRMD_PROVIDER, rsc->provider); crm_xml_add(reply, F_LRMD_TYPE, rsc->type); } return reply; } static int process_lrmd_rsc_unregister(pcmk__client_t *client, uint32_t id, xmlNode *request) { int rc = pcmk_ok; lrmd_rsc_t *rsc = NULL; xmlNode *rsc_xml = get_xpath_object("//" F_LRMD_RSC, request, LOG_ERR); const char *rsc_id = crm_element_value(rsc_xml, F_LRMD_RSC_ID); if (!rsc_id) { return -ENODEV; } rsc = g_hash_table_lookup(rsc_list, rsc_id); if (rsc == NULL) { crm_info("Ignoring unregistration of resource '%s', which is not registered", rsc_id); return pcmk_ok; } if (rsc->active) { /* let the caller know there are still active ops on this rsc to watch for */ crm_trace("Operation (0x%p) still in progress for unregistered resource %s", rsc->active, rsc_id); rc = -EINPROGRESS; } g_hash_table_remove(rsc_list, rsc_id); return rc; } static int process_lrmd_rsc_exec(pcmk__client_t *client, uint32_t id, xmlNode *request) { lrmd_rsc_t *rsc = NULL; lrmd_cmd_t *cmd = NULL; xmlNode *rsc_xml = get_xpath_object("//" F_LRMD_RSC, request, LOG_ERR); const char *rsc_id = crm_element_value(rsc_xml, F_LRMD_RSC_ID); int call_id; if (!rsc_id) { return -EINVAL; } if (!(rsc = g_hash_table_lookup(rsc_list, rsc_id))) { crm_info("Resource '%s' not found (%d active resources)", rsc_id, g_hash_table_size(rsc_list)); return -ENODEV; } cmd = create_lrmd_cmd(request, client); call_id = cmd->call_id; /* Don't reference cmd after handing it off to be scheduled. * The cmd could get merged and freed. */ schedule_lrmd_cmd(rsc, cmd); return call_id; } static int cancel_op(const char *rsc_id, const char *action, guint interval_ms) { GList *gIter = NULL; lrmd_rsc_t *rsc = g_hash_table_lookup(rsc_list, rsc_id); /* How to cancel an action. * 1. Check pending ops list, if it hasn't been handed off * to the service library or stonith recurring list remove * it there and that will stop it. * 2. If it isn't in the pending ops list, then it's either a * recurring op in the stonith recurring list, or the service * library's recurring list. Stop it there * 3. If not found in any lists, then this operation has either * been executed already and is not a recurring operation, or * never existed. */ if (!rsc) { return -ENODEV; } for (gIter = rsc->pending_ops; gIter != NULL; gIter = gIter->next) { lrmd_cmd_t *cmd = gIter->data; if (action_matches(cmd, action, interval_ms)) { cmd->result.exec_status = PCMK_EXEC_CANCELLED; cmd_finalize(cmd, rsc); return pcmk_ok; } } if (pcmk__str_eq(rsc->class, PCMK_RESOURCE_CLASS_STONITH, pcmk__str_casei)) { /* The service library does not handle stonith operations. * We have to handle recurring stonith operations ourselves. */ for (gIter = rsc->recurring_ops; gIter != NULL; gIter = gIter->next) { lrmd_cmd_t *cmd = gIter->data; if (action_matches(cmd, action, interval_ms)) { cmd->result.exec_status = PCMK_EXEC_CANCELLED; if (rsc->active != cmd) { cmd_finalize(cmd, rsc); } return pcmk_ok; } } } else if (services_action_cancel(rsc_id, normalize_action_name(rsc, action), interval_ms) == TRUE) { /* The service library will tell the action_complete callback function * this action was cancelled, which will destroy the cmd and remove * it from the recurring_op list. Do not do that in this function * if the service library says it cancelled it. */ return pcmk_ok; } return -EOPNOTSUPP; } static void cancel_all_recurring(lrmd_rsc_t * rsc, const char *client_id) { GList *cmd_list = NULL; GList *cmd_iter = NULL; /* Notice a copy of each list is created when concat is called. * This prevents odd behavior from occurring when the cmd_list * is iterated through later on. It is possible the cancel_op * function may end up modifying the recurring_ops and pending_ops * lists. If we did not copy those lists, our cmd_list iteration * could get messed up.*/ if (rsc->recurring_ops) { cmd_list = g_list_concat(cmd_list, g_list_copy(rsc->recurring_ops)); } if (rsc->pending_ops) { cmd_list = g_list_concat(cmd_list, g_list_copy(rsc->pending_ops)); } if (!cmd_list) { return; } for (cmd_iter = cmd_list; cmd_iter; cmd_iter = cmd_iter->next) { lrmd_cmd_t *cmd = cmd_iter->data; if (cmd->interval_ms == 0) { continue; } if (client_id && !pcmk__str_eq(cmd->client_id, client_id, pcmk__str_casei)) { continue; } cancel_op(rsc->rsc_id, cmd->action, cmd->interval_ms); } /* frees only the copied list data, not the cmds */ g_list_free(cmd_list); } static int process_lrmd_rsc_cancel(pcmk__client_t *client, uint32_t id, xmlNode *request) { xmlNode *rsc_xml = get_xpath_object("//" F_LRMD_RSC, request, LOG_ERR); const char *rsc_id = crm_element_value(rsc_xml, F_LRMD_RSC_ID); const char *action = crm_element_value(rsc_xml, F_LRMD_RSC_ACTION); guint interval_ms = 0; crm_element_value_ms(rsc_xml, F_LRMD_RSC_INTERVAL, &interval_ms); if (!rsc_id || !action) { return -EINVAL; } return cancel_op(rsc_id, action, interval_ms); } static void add_recurring_op_xml(xmlNode *reply, lrmd_rsc_t *rsc) { xmlNode *rsc_xml = create_xml_node(reply, F_LRMD_RSC); crm_xml_add(rsc_xml, F_LRMD_RSC_ID, rsc->rsc_id); for (GList *item = rsc->recurring_ops; item != NULL; item = item->next) { lrmd_cmd_t *cmd = item->data; xmlNode *op_xml = create_xml_node(rsc_xml, T_LRMD_RSC_OP); crm_xml_add(op_xml, F_LRMD_RSC_ACTION, (cmd->real_action? cmd->real_action : cmd->action)); crm_xml_add_ms(op_xml, F_LRMD_RSC_INTERVAL, cmd->interval_ms); crm_xml_add_int(op_xml, F_LRMD_TIMEOUT, cmd->timeout_orig); } } static xmlNode * process_lrmd_get_recurring(xmlNode *request, int call_id) { int rc = pcmk_ok; const char *rsc_id = NULL; lrmd_rsc_t *rsc = NULL; xmlNode *reply = NULL; xmlNode *rsc_xml = NULL; // Resource ID is optional rsc_xml = first_named_child(request, F_LRMD_CALLDATA); if (rsc_xml) { rsc_xml = first_named_child(rsc_xml, F_LRMD_RSC); } if (rsc_xml) { rsc_id = crm_element_value(rsc_xml, F_LRMD_RSC_ID); } // If resource ID is specified, resource must exist if (rsc_id != NULL) { rsc = g_hash_table_lookup(rsc_list, rsc_id); if (rsc == NULL) { crm_info("Resource '%s' not found (%d active resources)", rsc_id, g_hash_table_size(rsc_list)); rc = -ENODEV; } } reply = create_lrmd_reply(__func__, rc, call_id); // If resource ID is not specified, check all resources if (rsc_id == NULL) { GHashTableIter iter; char *key = NULL; g_hash_table_iter_init(&iter, rsc_list); while (g_hash_table_iter_next(&iter, (gpointer *) &key, (gpointer *) &rsc)) { add_recurring_op_xml(reply, rsc); } } else if (rsc) { add_recurring_op_xml(reply, rsc); } return reply; } void process_lrmd_message(pcmk__client_t *client, uint32_t id, xmlNode *request) { int rc = pcmk_ok; int call_id = 0; const char *op = crm_element_value(request, F_LRMD_OPERATION); int do_reply = 0; int do_notify = 0; xmlNode *reply = NULL; /* Certain IPC commands may be done only by privileged users (i.e. root or * hacluster), because they would otherwise provide a means of bypassing * ACLs. */ bool allowed = pcmk_is_set(client->flags, pcmk__client_privileged); crm_trace("Processing %s operation from %s", op, client->id); crm_element_value_int(request, F_LRMD_CALLID, &call_id); if (pcmk__str_eq(op, CRM_OP_IPC_FWD, pcmk__str_none)) { #ifdef PCMK__COMPILE_REMOTE if (allowed) { ipc_proxy_forward_client(client, request); } else { rc = -EACCES; } #else rc = -EPROTONOSUPPORT; #endif do_reply = 1; } else if (pcmk__str_eq(op, CRM_OP_REGISTER, pcmk__str_none)) { rc = process_lrmd_signon(client, request, call_id, &reply); do_reply = 1; } else if (pcmk__str_eq(op, LRMD_OP_RSC_REG, pcmk__str_none)) { if (allowed) { rc = process_lrmd_rsc_register(client, id, request); do_notify = 1; } else { rc = -EACCES; } do_reply = 1; } else if (pcmk__str_eq(op, LRMD_OP_RSC_INFO, pcmk__str_none)) { if (allowed) { reply = process_lrmd_get_rsc_info(request, call_id); } else { rc = -EACCES; } do_reply = 1; } else if (pcmk__str_eq(op, LRMD_OP_RSC_UNREG, pcmk__str_none)) { if (allowed) { rc = process_lrmd_rsc_unregister(client, id, request); /* don't notify anyone about failed un-registers */ if (rc == pcmk_ok || rc == -EINPROGRESS) { do_notify = 1; } } else { rc = -EACCES; } do_reply = 1; } else if (pcmk__str_eq(op, LRMD_OP_RSC_EXEC, pcmk__str_none)) { if (allowed) { rc = process_lrmd_rsc_exec(client, id, request); } else { rc = -EACCES; } do_reply = 1; } else if (pcmk__str_eq(op, LRMD_OP_RSC_CANCEL, pcmk__str_none)) { if (allowed) { rc = process_lrmd_rsc_cancel(client, id, request); } else { rc = -EACCES; } do_reply = 1; } else if (pcmk__str_eq(op, LRMD_OP_POKE, pcmk__str_none)) { do_notify = 1; do_reply = 1; } else if (pcmk__str_eq(op, LRMD_OP_CHECK, pcmk__str_none)) { if (allowed) { xmlNode *data = get_message_xml(request, F_LRMD_CALLDATA); CRM_LOG_ASSERT(data != NULL); pcmk__valid_sbd_timeout(crm_element_value(data, F_LRMD_WATCHDOG)); } else { rc = -EACCES; } } else if (pcmk__str_eq(op, LRMD_OP_ALERT_EXEC, pcmk__str_none)) { if (allowed) { rc = process_lrmd_alert_exec(client, id, request); } else { rc = -EACCES; } do_reply = 1; } else if (pcmk__str_eq(op, LRMD_OP_GET_RECURRING, pcmk__str_none)) { if (allowed) { reply = process_lrmd_get_recurring(request, call_id); } else { rc = -EACCES; } do_reply = 1; } else { rc = -EOPNOTSUPP; do_reply = 1; crm_err("Unknown IPC request '%s' from client %s", op, pcmk__client_name(client)); } if (rc == -EACCES) { crm_warn("Rejecting IPC request '%s' from unprivileged client %s", op, pcmk__client_name(client)); } crm_debug("Processed %s operation from %s: rc=%d, reply=%d, notify=%d", op, client->id, rc, do_reply, do_notify); if (do_reply) { int send_rc = pcmk_rc_ok; if (reply == NULL) { reply = create_lrmd_reply(__func__, rc, call_id); } send_rc = lrmd_server_send_reply(client, id, reply); free_xml(reply); if (send_rc != pcmk_rc_ok) { crm_warn("Reply to client %s failed: %s " CRM_XS " rc=%d", pcmk__client_name(client), pcmk_rc_str(send_rc), send_rc); } } if (do_notify) { send_generic_notify(rc, request); } } diff --git a/lib/pacemaker/pcmk_sched_transition.c b/lib/pacemaker/pcmk_sched_transition.c index 5b1882e3dd..1862c9cd3c 100644 --- a/lib/pacemaker/pcmk_sched_transition.c +++ b/lib/pacemaker/pcmk_sched_transition.c @@ -1,850 +1,850 @@ /* * Copyright 2009-2021 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 // lrmd_event_data_t, lrmd_free_event() #include #include #include #include #include #include #include static pcmk__output_t *out = NULL; static cib_t *fake_cib = NULL; static GList *fake_resource_list = NULL; static GList *fake_op_fail_list = NULL; gboolean bringing_nodes_online = FALSE; #define STATUS_PATH_MAX 512 #define NEW_NODE_TEMPLATE "//"XML_CIB_TAG_NODE"[@uname='%s']" #define NODE_TEMPLATE "//"XML_CIB_TAG_STATE"[@uname='%s']" #define RSC_TEMPLATE "//"XML_CIB_TAG_STATE"[@uname='%s']//"XML_LRM_TAG_RESOURCE"[@id='%s']" static void inject_transient_attr(xmlNode * cib_node, const char *name, const char *value) { xmlNode *attrs = NULL; xmlNode *instance_attrs = NULL; const char *node_uuid = ID(cib_node); out->message(out, "inject-attr", name, value, cib_node); attrs = first_named_child(cib_node, XML_TAG_TRANSIENT_NODEATTRS); if (attrs == NULL) { attrs = create_xml_node(cib_node, XML_TAG_TRANSIENT_NODEATTRS); crm_xml_add(attrs, XML_ATTR_ID, node_uuid); } instance_attrs = first_named_child(attrs, XML_TAG_ATTR_SETS); if (instance_attrs == NULL) { instance_attrs = create_xml_node(attrs, XML_TAG_ATTR_SETS); crm_xml_add(instance_attrs, XML_ATTR_ID, node_uuid); } crm_create_nvpair_xml(instance_attrs, NULL, name, value); } static void update_failcounts(xmlNode * cib_node, const char *resource, const char *task, guint interval_ms, int rc) { if (rc == 0) { return; } else if ((rc == 7) && (interval_ms == 0)) { return; } else { char *name = NULL; char *now = pcmk__ttoa(time(NULL)); name = pcmk__failcount_name(resource, task, interval_ms); inject_transient_attr(cib_node, name, "value++"); free(name); name = pcmk__lastfailure_name(resource, task, interval_ms); inject_transient_attr(cib_node, name, now); free(name); free(now); } } static void create_node_entry(cib_t * cib_conn, const char *node) { int rc = pcmk_ok; char *xpath = crm_strdup_printf(NEW_NODE_TEMPLATE, node); rc = cib_conn->cmds->query(cib_conn, xpath, NULL, cib_xpath | cib_sync_call | cib_scope_local); if (rc == -ENXIO) { xmlNode *cib_object = create_xml_node(NULL, XML_CIB_TAG_NODE); crm_xml_add(cib_object, XML_ATTR_ID, node); // Use node name as ID crm_xml_add(cib_object, XML_ATTR_UNAME, node); cib_conn->cmds->create(cib_conn, XML_CIB_TAG_NODES, cib_object, cib_sync_call | cib_scope_local); /* Not bothering with subsequent query to see if it exists, we'll bomb out later in the call to query_node_uuid()... */ free_xml(cib_object); } free(xpath); } static lrmd_event_data_t * create_op(xmlNode *cib_resource, const char *task, guint interval_ms, int outcome) { lrmd_event_data_t *op = NULL; xmlNode *xop = NULL; op = lrmd_new_event(ID(cib_resource), task, interval_ms); - lrmd__set_result(op, outcome, PCMK_EXEC_DONE, NULL); + lrmd__set_result(op, outcome, PCMK_EXEC_DONE, "Simulated action result"); op->params = NULL; /* TODO: Fill me in */ op->t_run = (unsigned int) time(NULL); op->t_rcchange = op->t_run; op->call_id = 0; for (xop = pcmk__xe_first_child(cib_resource); xop != NULL; xop = pcmk__xe_next(xop)) { int tmp = 0; crm_element_value_int(xop, XML_LRM_ATTR_CALLID, &tmp); if (tmp > op->call_id) { op->call_id = tmp; } } op->call_id++; return op; } static xmlNode * inject_op(xmlNode * cib_resource, lrmd_event_data_t * op, int target_rc) { return pcmk__create_history_xml(cib_resource, op, CRM_FEATURE_SET, target_rc, NULL, crm_system_name, LOG_TRACE); } static xmlNode * inject_node_state(cib_t * cib_conn, const char *node, const char *uuid) { int rc = pcmk_ok; xmlNode *cib_object = NULL; char *xpath = crm_strdup_printf(NODE_TEMPLATE, node); if (bringing_nodes_online) { create_node_entry(cib_conn, node); } rc = cib_conn->cmds->query(cib_conn, xpath, &cib_object, cib_xpath | cib_sync_call | cib_scope_local); if (cib_object && ID(cib_object) == NULL) { crm_err("Detected multiple node_state entries for xpath=%s, bailing", xpath); crm_log_xml_warn(cib_object, "Duplicates"); free(xpath); crm_exit(CRM_EX_SOFTWARE); return NULL; // not reached, but makes static analysis happy } if (rc == -ENXIO) { char *found_uuid = NULL; if (uuid == NULL) { query_node_uuid(cib_conn, node, &found_uuid, NULL); } else { found_uuid = strdup(uuid); } cib_object = create_xml_node(NULL, XML_CIB_TAG_STATE); crm_xml_add(cib_object, XML_ATTR_UUID, found_uuid); crm_xml_add(cib_object, XML_ATTR_UNAME, node); cib_conn->cmds->create(cib_conn, XML_CIB_TAG_STATUS, cib_object, cib_sync_call | cib_scope_local); free_xml(cib_object); free(found_uuid); rc = cib_conn->cmds->query(cib_conn, xpath, &cib_object, cib_xpath | cib_sync_call | cib_scope_local); crm_trace("injecting node state for %s. rc is %d", node, rc); } free(xpath); CRM_ASSERT(rc == pcmk_ok); return cib_object; } static xmlNode * modify_node(cib_t * cib_conn, char *node, gboolean up) { xmlNode *cib_node = inject_node_state(cib_conn, node, NULL); if (up) { crm_xml_add(cib_node, XML_NODE_IN_CLUSTER, XML_BOOLEAN_YES); crm_xml_add(cib_node, XML_NODE_IS_PEER, ONLINESTATUS); crm_xml_add(cib_node, XML_NODE_JOIN_STATE, CRMD_JOINSTATE_MEMBER); crm_xml_add(cib_node, XML_NODE_EXPECTED, CRMD_JOINSTATE_MEMBER); } else { crm_xml_add(cib_node, XML_NODE_IN_CLUSTER, XML_BOOLEAN_NO); crm_xml_add(cib_node, XML_NODE_IS_PEER, OFFLINESTATUS); crm_xml_add(cib_node, XML_NODE_JOIN_STATE, CRMD_JOINSTATE_DOWN); crm_xml_add(cib_node, XML_NODE_EXPECTED, CRMD_JOINSTATE_DOWN); } crm_xml_add(cib_node, XML_ATTR_ORIGIN, crm_system_name); return cib_node; } static xmlNode * find_resource_xml(xmlNode * cib_node, const char *resource) { xmlNode *match = NULL; const char *node = crm_element_value(cib_node, XML_ATTR_UNAME); char *xpath = crm_strdup_printf(RSC_TEMPLATE, node, resource); match = get_xpath_object(xpath, cib_node, LOG_TRACE); free(xpath); return match; } static xmlNode * inject_resource(xmlNode * cib_node, const char *resource, const char *lrm_name, const char *rclass, const char *rtype, const char *rprovider) { xmlNode *lrm = NULL; xmlNode *container = NULL; xmlNode *cib_resource = NULL; char *xpath = NULL; cib_resource = find_resource_xml(cib_node, resource); if (cib_resource != NULL) { /* If an existing LRM history entry uses the resource name, * continue using it, even if lrm_name is different. */ return cib_resource; } // Check for history entry under preferred name if (strcmp(resource, lrm_name)) { cib_resource = find_resource_xml(cib_node, lrm_name); if (cib_resource != NULL) { return cib_resource; } } /* One day, add query for class, provider, type */ if (rclass == NULL || rtype == NULL) { out->err(out, "Resource %s not found in the status section of %s." " Please supply the class and type to continue", resource, ID(cib_node)); return NULL; } else if (!pcmk__strcase_any_of(rclass, PCMK_RESOURCE_CLASS_OCF, PCMK_RESOURCE_CLASS_STONITH, PCMK_RESOURCE_CLASS_SERVICE, PCMK_RESOURCE_CLASS_UPSTART, PCMK_RESOURCE_CLASS_SYSTEMD, PCMK_RESOURCE_CLASS_LSB, NULL)) { out->err(out, "Invalid class for %s: %s", resource, rclass); return NULL; } else if (pcmk_is_set(pcmk_get_ra_caps(rclass), pcmk_ra_cap_provider) && (rprovider == NULL)) { out->err(out, "Please specify the provider for resource %s", resource); return NULL; } xpath = (char *)xmlGetNodePath(cib_node); crm_info("Injecting new resource %s into %s '%s'", lrm_name, xpath, ID(cib_node)); free(xpath); lrm = first_named_child(cib_node, XML_CIB_TAG_LRM); if (lrm == NULL) { const char *node_uuid = ID(cib_node); lrm = create_xml_node(cib_node, XML_CIB_TAG_LRM); crm_xml_add(lrm, XML_ATTR_ID, node_uuid); } container = first_named_child(lrm, XML_LRM_TAG_RESOURCES); if (container == NULL) { container = create_xml_node(lrm, XML_LRM_TAG_RESOURCES); } cib_resource = create_xml_node(container, XML_LRM_TAG_RESOURCE); // If we're creating a new entry, use the preferred name crm_xml_add(cib_resource, XML_ATTR_ID, lrm_name); crm_xml_add(cib_resource, XML_AGENT_ATTR_CLASS, rclass); crm_xml_add(cib_resource, XML_AGENT_ATTR_PROVIDER, rprovider); crm_xml_add(cib_resource, XML_ATTR_TYPE, rtype); return cib_resource; } #define XPATH_MAX 1024 static int find_ticket_state(cib_t * the_cib, const char *ticket_id, xmlNode ** ticket_state_xml) { int offset = 0; int rc = pcmk_ok; xmlNode *xml_search = NULL; char *xpath_string = NULL; CRM_ASSERT(ticket_state_xml != NULL); *ticket_state_xml = NULL; xpath_string = calloc(1, XPATH_MAX); offset += snprintf(xpath_string + offset, XPATH_MAX - offset, "%s", "/cib/status/tickets"); if (ticket_id) { offset += snprintf(xpath_string + offset, XPATH_MAX - offset, "/%s[@id=\"%s\"]", XML_CIB_TAG_TICKET_STATE, ticket_id); } CRM_LOG_ASSERT(offset > 0); rc = the_cib->cmds->query(the_cib, xpath_string, &xml_search, cib_sync_call | cib_scope_local | cib_xpath); if (rc != pcmk_ok) { goto bail; } crm_log_xml_debug(xml_search, "Match"); if (xml_has_children(xml_search)) { if (ticket_id) { out->err(out, "Multiple ticket_states match ticket_id=%s", ticket_id); } *ticket_state_xml = xml_search; } else { *ticket_state_xml = xml_search; } bail: free(xpath_string); return rc; } static int set_ticket_state_attr(const char *ticket_id, const char *attr_name, const char *attr_value, cib_t * cib, int cib_options) { int rc = pcmk_ok; xmlNode *xml_top = NULL; xmlNode *ticket_state_xml = NULL; rc = find_ticket_state(cib, ticket_id, &ticket_state_xml); if (rc == pcmk_ok) { crm_debug("Found a match state for ticket: id=%s", ticket_id); xml_top = ticket_state_xml; } else if (rc != -ENXIO) { return rc; } else { xmlNode *xml_obj = NULL; xml_top = create_xml_node(NULL, XML_CIB_TAG_STATUS); xml_obj = create_xml_node(xml_top, XML_CIB_TAG_TICKETS); ticket_state_xml = create_xml_node(xml_obj, XML_CIB_TAG_TICKET_STATE); crm_xml_add(ticket_state_xml, XML_ATTR_ID, ticket_id); } crm_xml_add(ticket_state_xml, attr_name, attr_value); crm_log_xml_debug(xml_top, "Update"); rc = cib->cmds->modify(cib, XML_CIB_TAG_STATUS, xml_top, cib_options); free_xml(xml_top); return rc; } void modify_configuration(pe_working_set_t * data_set, cib_t *cib, pcmk_injections_t *injections) { int rc = pcmk_ok; GList *gIter = NULL; xmlNode *cib_op = NULL; xmlNode *cib_node = NULL; xmlNode *cib_resource = NULL; lrmd_event_data_t *op = NULL; out = data_set->priv; out->message(out, "inject-modify-config", injections->quorum, injections->watchdog); if (injections->quorum) { xmlNode *top = create_xml_node(NULL, XML_TAG_CIB); /* crm_xml_add(top, XML_ATTR_DC_UUID, dc_uuid); */ crm_xml_add(top, XML_ATTR_HAVE_QUORUM, injections->quorum); rc = cib->cmds->modify(cib, NULL, top, cib_sync_call | cib_scope_local); CRM_ASSERT(rc == pcmk_ok); } if (injections->watchdog) { rc = update_attr_delegate(cib, cib_sync_call | cib_scope_local, XML_CIB_TAG_CRMCONFIG, NULL, NULL, NULL, NULL, XML_ATTR_HAVE_WATCHDOG, injections->watchdog, FALSE, NULL, NULL); CRM_ASSERT(rc == pcmk_ok); } for (gIter = injections->node_up; gIter != NULL; gIter = gIter->next) { char *node = (char *)gIter->data; out->message(out, "inject-modify-node", "Online", node); cib_node = modify_node(cib, node, TRUE); CRM_ASSERT(cib_node != NULL); rc = cib->cmds->modify(cib, XML_CIB_TAG_STATUS, cib_node, cib_sync_call | cib_scope_local); CRM_ASSERT(rc == pcmk_ok); free_xml(cib_node); } for (gIter = injections->node_down; gIter != NULL; gIter = gIter->next) { char xpath[STATUS_PATH_MAX]; char *node = (char *)gIter->data; out->message(out, "inject-modify-node", "Offline", node); cib_node = modify_node(cib, node, FALSE); CRM_ASSERT(cib_node != NULL); rc = cib->cmds->modify(cib, XML_CIB_TAG_STATUS, cib_node, cib_sync_call | cib_scope_local); CRM_ASSERT(rc == pcmk_ok); free_xml(cib_node); snprintf(xpath, STATUS_PATH_MAX, "//node_state[@uname='%s']/%s", node, XML_CIB_TAG_LRM); cib->cmds->remove(cib, xpath, NULL, cib_xpath | cib_sync_call | cib_scope_local); snprintf(xpath, STATUS_PATH_MAX, "//node_state[@uname='%s']/%s", node, XML_TAG_TRANSIENT_NODEATTRS); cib->cmds->remove(cib, xpath, NULL, cib_xpath | cib_sync_call | cib_scope_local); } for (gIter = injections->node_fail; gIter != NULL; gIter = gIter->next) { char *node = (char *)gIter->data; out->message(out, "inject-modify-node", "Failing", node); cib_node = modify_node(cib, node, TRUE); crm_xml_add(cib_node, XML_NODE_IN_CLUSTER, XML_BOOLEAN_NO); CRM_ASSERT(cib_node != NULL); rc = cib->cmds->modify(cib, XML_CIB_TAG_STATUS, cib_node, cib_sync_call | cib_scope_local); CRM_ASSERT(rc == pcmk_ok); free_xml(cib_node); } for (gIter = injections->ticket_grant; gIter != NULL; gIter = gIter->next) { char *ticket_id = (char *)gIter->data; out->message(out, "inject-modify-ticket", "Granting", ticket_id); rc = set_ticket_state_attr(ticket_id, "granted", "true", cib, cib_sync_call | cib_scope_local); CRM_ASSERT(rc == pcmk_ok); } for (gIter = injections->ticket_revoke; gIter != NULL; gIter = gIter->next) { char *ticket_id = (char *)gIter->data; out->message(out, "inject-modify-ticket", "Revoking", ticket_id); rc = set_ticket_state_attr(ticket_id, "granted", "false", cib, cib_sync_call | cib_scope_local); CRM_ASSERT(rc == pcmk_ok); } for (gIter = injections->ticket_standby; gIter != NULL; gIter = gIter->next) { char *ticket_id = (char *)gIter->data; out->message(out, "inject-modify-ticket", "Standby", ticket_id); rc = set_ticket_state_attr(ticket_id, "standby", "true", cib, cib_sync_call | cib_scope_local); CRM_ASSERT(rc == pcmk_ok); } for (gIter = injections->ticket_activate; gIter != NULL; gIter = gIter->next) { char *ticket_id = (char *)gIter->data; out->message(out, "inject-modify-ticket", "Activating", ticket_id); rc = set_ticket_state_attr(ticket_id, "standby", "false", cib, cib_sync_call | cib_scope_local); CRM_ASSERT(rc == pcmk_ok); } for (gIter = injections->op_inject; gIter != NULL; gIter = gIter->next) { char *spec = (char *)gIter->data; int rc = 0; int outcome = PCMK_OCF_OK; guint interval_ms = 0; char *key = NULL; char *node = NULL; char *task = NULL; char *resource = NULL; const char *rtype = NULL; const char *rclass = NULL; const char *rprovider = NULL; pe_resource_t *rsc = NULL; out->message(out, "inject-spec", spec); key = calloc(1, strlen(spec) + 1); node = calloc(1, strlen(spec) + 1); rc = sscanf(spec, "%[^@]@%[^=]=%d", key, node, &outcome); if (rc != 3) { out->err(out, "Invalid operation spec: %s. Only found %d fields", spec, rc); free(key); free(node); continue; } parse_op_key(key, &resource, &task, &interval_ms); rsc = pe_find_resource(data_set->resources, resource); if (rsc == NULL) { out->err(out, "Invalid resource name: %s", resource); } else { rclass = crm_element_value(rsc->xml, XML_AGENT_ATTR_CLASS); rtype = crm_element_value(rsc->xml, XML_ATTR_TYPE); rprovider = crm_element_value(rsc->xml, XML_AGENT_ATTR_PROVIDER); cib_node = inject_node_state(cib, node, NULL); CRM_ASSERT(cib_node != NULL); update_failcounts(cib_node, resource, task, interval_ms, outcome); cib_resource = inject_resource(cib_node, resource, resource, rclass, rtype, rprovider); CRM_ASSERT(cib_resource != NULL); op = create_op(cib_resource, task, interval_ms, outcome); CRM_ASSERT(op != NULL); cib_op = inject_op(cib_resource, op, 0); CRM_ASSERT(cib_op != NULL); lrmd_free_event(op); rc = cib->cmds->modify(cib, XML_CIB_TAG_STATUS, cib_node, cib_sync_call | cib_scope_local); CRM_ASSERT(rc == pcmk_ok); } free(task); free(node); free(key); } if (!out->is_quiet(out)) { out->end_list(out); } } static gboolean exec_pseudo_action(crm_graph_t * graph, crm_action_t * action) { const char *node = crm_element_value(action->xml, XML_LRM_ATTR_TARGET); const char *task = crm_element_value(action->xml, XML_LRM_ATTR_TASK_KEY); action->confirmed = TRUE; out->message(out, "inject-pseudo-action", node, task); pcmk__update_graph(graph, action); return TRUE; } static gboolean exec_rsc_action(crm_graph_t * graph, crm_action_t * action) { int rc = 0; GList *gIter = NULL; lrmd_event_data_t *op = NULL; int target_outcome = PCMK_OCF_OK; const char *rtype = NULL; const char *rclass = NULL; const char *resource = NULL; const char *rprovider = NULL; const char *lrm_name = NULL; const char *operation = crm_element_value(action->xml, "operation"); const char *target_rc_s = crm_meta_value(action->params, XML_ATTR_TE_TARGET_RC); xmlNode *cib_node = NULL; xmlNode *cib_resource = NULL; xmlNode *action_rsc = first_named_child(action->xml, XML_CIB_TAG_RESOURCE); char *node = crm_element_value_copy(action->xml, XML_LRM_ATTR_TARGET); char *uuid = crm_element_value_copy(action->xml, XML_LRM_ATTR_TARGET_UUID); const char *router_node = crm_element_value(action->xml, XML_LRM_ATTR_ROUTER_NODE); if (pcmk__strcase_any_of(operation, CRM_OP_PROBED, CRM_OP_REPROBE, NULL)) { crm_info("Skipping %s op for %s", operation, node); goto done; } if (action_rsc == NULL) { crm_log_xml_err(action->xml, "Bad"); free(node); free(uuid); return FALSE; } /* Look for the preferred name * If not found, try the expected 'local' name * If not found use the preferred name anyway */ resource = crm_element_value(action_rsc, XML_ATTR_ID); CRM_ASSERT(resource != NULL); // makes static analysis happy lrm_name = resource; // Preferred name when writing history if (pe_find_resource(fake_resource_list, resource) == NULL) { const char *longname = crm_element_value(action_rsc, XML_ATTR_ID_LONG); if (longname && pe_find_resource(fake_resource_list, longname)) { resource = longname; } } if (pcmk__strcase_any_of(operation, "delete", RSC_METADATA, NULL)) { out->message(out, "inject-rsc-action", resource, operation, node, (guint) 0); goto done; } rclass = crm_element_value(action_rsc, XML_AGENT_ATTR_CLASS); rtype = crm_element_value(action_rsc, XML_ATTR_TYPE); rprovider = crm_element_value(action_rsc, XML_AGENT_ATTR_PROVIDER); pcmk__scan_min_int(target_rc_s, &target_outcome, 0); CRM_ASSERT(fake_cib->cmds->query(fake_cib, NULL, NULL, cib_sync_call | cib_scope_local) == pcmk_ok); cib_node = inject_node_state(fake_cib, node, (router_node? node : uuid)); CRM_ASSERT(cib_node != NULL); cib_resource = inject_resource(cib_node, resource, lrm_name, rclass, rtype, rprovider); if (cib_resource == NULL) { crm_err("invalid resource in transition"); free(node); free(uuid); free_xml(cib_node); return FALSE; } op = pcmk__event_from_graph_action(cib_resource, action, PCMK_EXEC_DONE, target_outcome); out->message(out, "inject-rsc-action", resource, op->op_type, node, op->interval_ms); for (gIter = fake_op_fail_list; gIter != NULL; gIter = gIter->next) { char *spec = (char *)gIter->data; char *key = NULL; const char *match_name = NULL; // Allow user to specify anonymous clone with or without instance number key = crm_strdup_printf(PCMK__OP_FMT "@%s=", resource, op->op_type, op->interval_ms, node); if (strncasecmp(key, spec, strlen(key)) == 0) { match_name = resource; } free(key); if ((match_name == NULL) && strcmp(resource, lrm_name)) { key = crm_strdup_printf(PCMK__OP_FMT "@%s=", lrm_name, op->op_type, op->interval_ms, node); if (strncasecmp(key, spec, strlen(key)) == 0) { match_name = lrm_name; } free(key); } if (match_name != NULL) { rc = sscanf(spec, "%*[^=]=%d", (int *) &op->rc); // ${match_name}_${task}_${interval_in_ms}@${node}=${rc} if (rc != 1) { out->err(out, "Invalid failed operation spec: %s. Result code must be integer", spec); continue; } action->failed = TRUE; graph->abort_priority = INFINITY; out->info(out, "Pretending action %d failed with rc=%d", action->id, op->rc); update_failcounts(cib_node, match_name, op->op_type, op->interval_ms, op->rc); break; } } inject_op(cib_resource, op, target_outcome); lrmd_free_event(op); rc = fake_cib->cmds->modify(fake_cib, XML_CIB_TAG_STATUS, cib_node, cib_sync_call | cib_scope_local); CRM_ASSERT(rc == pcmk_ok); done: free(node); free(uuid); free_xml(cib_node); action->confirmed = TRUE; pcmk__update_graph(graph, action); return TRUE; } static gboolean exec_crmd_action(crm_graph_t * graph, crm_action_t * action) { const char *node = crm_element_value(action->xml, XML_LRM_ATTR_TARGET); const char *task = crm_element_value(action->xml, XML_LRM_ATTR_TASK); xmlNode *rsc = first_named_child(action->xml, XML_CIB_TAG_RESOURCE); action->confirmed = TRUE; out->message(out, "inject-cluster-action", node, task, rsc); pcmk__update_graph(graph, action); return TRUE; } static gboolean exec_stonith_action(crm_graph_t * graph, crm_action_t * action) { const char *op = crm_meta_value(action->params, "stonith_action"); char *target = crm_element_value_copy(action->xml, XML_LRM_ATTR_TARGET); out->message(out, "inject-fencing-action", target, op); if(!pcmk__str_eq(op, "on", pcmk__str_casei)) { int rc = 0; char xpath[STATUS_PATH_MAX]; xmlNode *cib_node = modify_node(fake_cib, target, FALSE); crm_xml_add(cib_node, XML_ATTR_ORIGIN, __func__); CRM_ASSERT(cib_node != NULL); rc = fake_cib->cmds->replace(fake_cib, XML_CIB_TAG_STATUS, cib_node, cib_sync_call | cib_scope_local); CRM_ASSERT(rc == pcmk_ok); snprintf(xpath, STATUS_PATH_MAX, "//node_state[@uname='%s']/%s", target, XML_CIB_TAG_LRM); fake_cib->cmds->remove(fake_cib, xpath, NULL, cib_xpath | cib_sync_call | cib_scope_local); snprintf(xpath, STATUS_PATH_MAX, "//node_state[@uname='%s']/%s", target, XML_TAG_TRANSIENT_NODEATTRS); fake_cib->cmds->remove(fake_cib, xpath, NULL, cib_xpath | cib_sync_call | cib_scope_local); free_xml(cib_node); } action->confirmed = TRUE; pcmk__update_graph(graph, action); free(target); return TRUE; } enum transition_status run_simulation(pe_working_set_t * data_set, cib_t *cib, GList *op_fail_list) { crm_graph_t *transition = NULL; enum transition_status graph_rc; crm_graph_functions_t exec_fns = { exec_pseudo_action, exec_rsc_action, exec_crmd_action, exec_stonith_action, }; out = data_set->priv; fake_cib = cib; fake_op_fail_list = op_fail_list; if (!out->is_quiet(out)) { out->begin_list(out, NULL, NULL, "Executing Cluster Transition"); } pcmk__set_graph_functions(&exec_fns); transition = pcmk__unpack_graph(data_set->graph, crm_system_name); pcmk__log_graph(LOG_DEBUG, transition); fake_resource_list = data_set->resources; do { graph_rc = pcmk__execute_graph(transition); } while (graph_rc == transition_active); fake_resource_list = NULL; if (graph_rc != transition_complete) { out->err(out, "Transition failed: %s", pcmk__graph_status2text(graph_rc)); pcmk__log_graph(LOG_ERR, transition); } pcmk__free_graph(transition); if (graph_rc != transition_complete) { out->err(out, "An invalid transition was produced"); } if (!out->is_quiet(out)) { xmlNode *cib_object = NULL; int rc = fake_cib->cmds->query(fake_cib, NULL, &cib_object, cib_sync_call | cib_scope_local); CRM_ASSERT(rc == pcmk_ok); pe_reset_working_set(data_set); data_set->input = cib_object; out->end_list(out); } return graph_rc; }