diff --git a/daemons/controld/controld_cib.c b/daemons/controld/controld_cib.c index 7acff3064f..99a222f405 100644 --- a/daemons/controld/controld_cib.c +++ b/daemons/controld/controld_cib.c @@ -1,1066 +1,1067 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include /* sleep */ #include #include #include #include #include // Call ID of the most recent in-progress CIB resource update (or 0 if none) static int pending_rsc_update = 0; /*! * \internal * \brief Respond to a dropped CIB connection * * \param[in] user_data CIB connection that dropped */ static void handle_cib_disconnect(gpointer user_data) { CRM_LOG_ASSERT(user_data == controld_globals.cib_conn); controld_trigger_fsa(); controld_globals.cib_conn->state = cib_disconnected; if (pcmk_is_set(controld_globals.fsa_input_register, R_CIB_CONNECTED)) { // @TODO This should trigger a reconnect, not a shutdown crm_crit("Lost connection to the CIB manager, shutting down"); register_fsa_input(C_FSA_INTERNAL, I_ERROR, NULL); controld_clear_fsa_input_flags(R_CIB_CONNECTED); } else { // Expected crm_info("Disconnected from the CIB manager"); } } static void do_cib_updated(const char *event, xmlNode * msg) { const xmlNode *patchset = NULL; const char *client_name = NULL; crm_debug("Received CIB diff notification: DC=%s", pcmk__btoa(AM_I_DC)); if (cib__get_notify_patchset(msg, &patchset) != pcmk_rc_ok) { return; } if (cib__element_in_patchset(patchset, PCMK_XE_ALERTS) || cib__element_in_patchset(patchset, PCMK_XE_CRM_CONFIG)) { controld_trigger_config(); } if (!AM_I_DC) { // We're not in control of the join sequence return; } client_name = crm_element_value(msg, PCMK__XA_CIB_CLIENTNAME); if (!cib__client_triggers_refresh(client_name)) { // The CIB is still accurate return; } if (cib__element_in_patchset(patchset, PCMK_XE_NODES) || cib__element_in_patchset(patchset, PCMK_XE_STATUS)) { /* An unsafe client modified the PCMK_XE_NODES or PCMK_XE_STATUS * section. Ensure the node list is up-to-date, and start the join * process again so we get everyone's current resource history. */ if (client_name == NULL) { client_name = crm_element_value(msg, PCMK__XA_CIB_CLIENTID); } crm_notice("Populating nodes and starting an election after %s event " "triggered by %s", event, pcmk__s(client_name, "(unidentified client)")); populate_cib_nodes(node_update_quick|node_update_all, __func__); register_fsa_input(C_FSA_INTERNAL, I_ELECTION, NULL); } } void controld_disconnect_cib_manager(void) { cib_t *cib_conn = controld_globals.cib_conn; CRM_ASSERT(cib_conn != NULL); crm_debug("Disconnecting from the CIB manager"); controld_clear_fsa_input_flags(R_CIB_CONNECTED); cib_conn->cmds->del_notify_callback(cib_conn, PCMK__VALUE_CIB_DIFF_NOTIFY, do_cib_updated); cib_free_callbacks(cib_conn); if (cib_conn->state != cib_disconnected) { cib_conn->cmds->set_secondary(cib_conn, cib_scope_local|cib_discard_reply); cib_conn->cmds->signoff(cib_conn); } } /* A_CIB_STOP, A_CIB_START, O_CIB_RESTART */ void do_cib_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) { static int cib_retries = 0; cib_t *cib_conn = controld_globals.cib_conn; void (*dnotify_fn) (gpointer user_data) = handle_cib_disconnect; void (*update_cb) (const char *event, xmlNodePtr msg) = do_cib_updated; int rc = pcmk_ok; CRM_ASSERT(cib_conn != NULL); if (pcmk_is_set(action, A_CIB_STOP)) { if ((cib_conn->state != cib_disconnected) && (pending_rsc_update != 0)) { crm_info("Waiting for resource update %d to complete", pending_rsc_update); crmd_fsa_stall(FALSE); return; } controld_disconnect_cib_manager(); } if (!pcmk_is_set(action, A_CIB_START)) { return; } if (cur_state == S_STOPPING) { crm_err("Ignoring request to connect to the CIB manager after " "shutdown"); return; } rc = cib_conn->cmds->signon(cib_conn, CRM_SYSTEM_CRMD, cib_command_nonblocking); if (rc != pcmk_ok) { // A short wait that usually avoids stalling the FSA sleep(1); rc = cib_conn->cmds->signon(cib_conn, CRM_SYSTEM_CRMD, cib_command_nonblocking); } if (rc != pcmk_ok) { crm_info("Could not connect to the CIB manager: %s", pcmk_strerror(rc)); } else if (cib_conn->cmds->set_connection_dnotify(cib_conn, dnotify_fn) != pcmk_ok) { crm_err("Could not set dnotify callback"); } else if (cib_conn->cmds->add_notify_callback(cib_conn, PCMK__VALUE_CIB_DIFF_NOTIFY, update_cb) != pcmk_ok) { crm_err("Could not set CIB notification callback (update)"); } else { controld_set_fsa_input_flags(R_CIB_CONNECTED); cib_retries = 0; } if (!pcmk_is_set(controld_globals.fsa_input_register, R_CIB_CONNECTED)) { cib_retries++; if (cib_retries < 30) { crm_warn("Couldn't complete CIB registration %d times... " "pause and retry", cib_retries); controld_start_wait_timer(); crmd_fsa_stall(FALSE); } else { crm_err("Could not complete CIB registration %d times... " "hard error", cib_retries); register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); } } } #define MIN_CIB_OP_TIMEOUT (30) /*! * \internal * \brief Get the timeout (in seconds) that should be used with CIB operations * * \return The maximum of 30 seconds, the value of the PCMK_cib_timeout * environment variable, or 10 seconds times one more than the number of * nodes in the cluster. */ unsigned int cib_op_timeout(void) { // @COMPAT: Drop env_timeout at 3.0.0 static int env_timeout = -1; unsigned int calculated_timeout = 0; if (env_timeout == -1) { const char *env = pcmk__env_option(PCMK__ENV_CIB_TIMEOUT); pcmk__scan_min_int(env, &env_timeout, MIN_CIB_OP_TIMEOUT); crm_trace("Minimum CIB op timeout: %ds (environment: %s)", env_timeout, (env? env : "none")); } calculated_timeout = 10U * (1U + pcmk__cluster_num_active_nodes() + pcmk__cluster_num_remote_nodes()); calculated_timeout = QB_MAX(calculated_timeout, env_timeout); crm_trace("Calculated timeout: %us", calculated_timeout); if (controld_globals.cib_conn) { controld_globals.cib_conn->call_timeout = calculated_timeout; } return calculated_timeout; } /*! * \internal * \brief Get CIB call options to use local scope if primary is unavailable * * \return CIB call options */ int crmd_cib_smart_opt(void) { int call_opt = cib_none; if ((controld_globals.fsa_state == S_ELECTION) || (controld_globals.fsa_state == S_PENDING)) { crm_info("Sending update to local CIB in state: %s", fsa_state2string(controld_globals.fsa_state)); cib__set_call_options(call_opt, "update", cib_scope_local); } return call_opt; } static void cib_delete_callback(xmlNode *msg, int call_id, int rc, xmlNode *output, void *user_data) { char *desc = user_data; if (rc == 0) { crm_debug("Deletion of %s (via CIB call %d) succeeded", desc, call_id); } else { crm_warn("Deletion of %s (via CIB call %d) failed: %s " CRM_XS " rc=%d", desc, call_id, pcmk_strerror(rc), rc); } } // Searches for various portions of PCMK__XE_NODE_STATE to delete // Match a particular node's PCMK__XE_NODE_STATE (takes node name 1x) #define XPATH_NODE_STATE "//" PCMK__XE_NODE_STATE "[@" PCMK_XA_UNAME "='%s']" // Node's lrm section (name 1x) #define XPATH_NODE_LRM XPATH_NODE_STATE "/" PCMK__XE_LRM /* Node's PCMK__XE_LRM_RSC_OP entries and PCMK__XE_LRM_RESOURCE entries without * unexpired lock * (name 2x, (seconds_since_epoch - PCMK_OPT_SHUTDOWN_LOCK_LIMIT) 1x) */ #define XPATH_NODE_LRM_UNLOCKED XPATH_NODE_STATE "//" PCMK__XE_LRM_RSC_OP \ "|" XPATH_NODE_STATE \ "//" PCMK__XE_LRM_RESOURCE \ "[not(@" PCMK_OPT_SHUTDOWN_LOCK ") " \ "or " PCMK_OPT_SHUTDOWN_LOCK "<%lld]" // Node's PCMK__XE_TRANSIENT_ATTRIBUTES section (name 1x) #define XPATH_NODE_ATTRS XPATH_NODE_STATE "/" PCMK__XE_TRANSIENT_ATTRIBUTES // Everything under PCMK__XE_NODE_STATE (name 1x) #define XPATH_NODE_ALL XPATH_NODE_STATE "/*" /* Unlocked history + transient attributes * (name 2x, (seconds_since_epoch - PCMK_OPT_SHUTDOWN_LOCK_LIMIT) 1x, name 1x) */ #define XPATH_NODE_ALL_UNLOCKED XPATH_NODE_LRM_UNLOCKED "|" XPATH_NODE_ATTRS /*! * \internal * \brief Get the XPath and description of a node state section to be deleted * * \param[in] uname Desired node * \param[in] section Subsection of \c PCMK__XE_NODE_STATE to be deleted * \param[out] xpath Where to store XPath of \p section * \param[out] desc If not \c NULL, where to store description of \p section */ void controld_node_state_deletion_strings(const char *uname, enum controld_section_e section, char **xpath, char **desc) { const char *desc_pre = NULL; // Shutdown locks that started before this time are expired long long expire = (long long) time(NULL) - controld_globals.shutdown_lock_limit; switch (section) { case controld_section_lrm: *xpath = crm_strdup_printf(XPATH_NODE_LRM, uname); desc_pre = "resource history"; break; case controld_section_lrm_unlocked: *xpath = crm_strdup_printf(XPATH_NODE_LRM_UNLOCKED, uname, uname, expire); desc_pre = "resource history (other than shutdown locks)"; break; case controld_section_attrs: *xpath = crm_strdup_printf(XPATH_NODE_ATTRS, uname); desc_pre = "transient attributes"; break; case controld_section_all: *xpath = crm_strdup_printf(XPATH_NODE_ALL, uname); desc_pre = "all state"; break; case controld_section_all_unlocked: *xpath = crm_strdup_printf(XPATH_NODE_ALL_UNLOCKED, uname, uname, expire, uname); desc_pre = "all state (other than shutdown locks)"; break; default: // We called this function incorrectly CRM_ASSERT(false); break; } if (desc != NULL) { *desc = crm_strdup_printf("%s for node %s", desc_pre, uname); } } /*! * \internal * \brief Delete subsection of a node's CIB \c PCMK__XE_NODE_STATE * * \param[in] uname Desired node * \param[in] section Subsection of \c PCMK__XE_NODE_STATE to delete * \param[in] options CIB call options to use */ void controld_delete_node_state(const char *uname, enum controld_section_e section, int options) { cib_t *cib = controld_globals.cib_conn; char *xpath = NULL; char *desc = NULL; int cib_rc = pcmk_ok; CRM_ASSERT((uname != NULL) && (cib != NULL)); controld_node_state_deletion_strings(uname, section, &xpath, &desc); cib__set_call_options(options, "node state deletion", cib_xpath|cib_multiple); cib_rc = cib->cmds->remove(cib, xpath, NULL, options); fsa_register_cib_callback(cib_rc, desc, cib_delete_callback); crm_info("Deleting %s (via CIB call %d) " CRM_XS " xpath=%s", desc, cib_rc, xpath); // CIB library handles freeing desc free(xpath); } // Takes node name and resource ID #define XPATH_RESOURCE_HISTORY "//" PCMK__XE_NODE_STATE \ "[@" PCMK_XA_UNAME "='%s']/" \ PCMK__XE_LRM "/" PCMK__XE_LRM_RESOURCES \ "/" PCMK__XE_LRM_RESOURCE \ "[@" PCMK_XA_ID "='%s']" // @TODO could add "and @PCMK_OPT_SHUTDOWN_LOCK" to limit to locks /*! * \internal * \brief Clear resource history from CIB for a given resource and node * * \param[in] rsc_id ID of resource to be cleared * \param[in] node Node whose resource history should be cleared * \param[in] user_name ACL user name to use * \param[in] call_options CIB call options * * \return Standard Pacemaker return code */ int controld_delete_resource_history(const char *rsc_id, const char *node, const char *user_name, int call_options) { char *desc = NULL; char *xpath = NULL; int rc = pcmk_rc_ok; cib_t *cib = controld_globals.cib_conn; CRM_CHECK((rsc_id != NULL) && (node != NULL), return EINVAL); desc = crm_strdup_printf("resource history for %s on %s", rsc_id, node); if (cib == NULL) { crm_err("Unable to clear %s: no CIB connection", desc); free(desc); return ENOTCONN; } // Ask CIB to delete the entry xpath = crm_strdup_printf(XPATH_RESOURCE_HISTORY, node, rsc_id); cib->cmds->set_user(cib, user_name); rc = cib->cmds->remove(cib, xpath, NULL, call_options|cib_xpath); cib->cmds->set_user(cib, NULL); if (rc < 0) { rc = pcmk_legacy2rc(rc); crm_err("Could not delete resource status of %s on %s%s%s: %s " CRM_XS " rc=%d", rsc_id, node, (user_name? " for user " : ""), (user_name? user_name : ""), pcmk_rc_str(rc), rc); free(desc); free(xpath); return rc; } if (pcmk_is_set(call_options, cib_sync_call)) { if (pcmk_is_set(call_options, cib_dryrun)) { crm_debug("Deletion of %s would succeed", desc); } else { crm_debug("Deletion of %s succeeded", desc); } free(desc); } else { crm_info("Clearing %s (via CIB call %d) " CRM_XS " xpath=%s", desc, rc, xpath); fsa_register_cib_callback(rc, desc, cib_delete_callback); // CIB library handles freeing desc } free(xpath); return pcmk_rc_ok; } /*! * \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 * \p g_string_free() and the XML result with \p free_xml(). */ static GString * build_parameter_list(const lrmd_event_data_t *op, const struct ra_metadata_s *metadata, enum ra_param_flags_e param_type, xmlNode **result) { GString *list = NULL; *result = pcmk__xe_create(NULL, PCMK_XE_PARAMETERS); /* Consider all parameters only except private ones to be consistent with * what scheduler does with calculate_secure_digest(). */ if (param_type == ra_param_private && compare_version(controld_globals.dc_version, "3.16.0") >= 0) { g_hash_table_foreach(op->params, hash2field, *result); pcmk__filter_op_for_digest(*result); } 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, 256, " "); } pcmk__add_word(&list, 0, 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); } } else { crm_trace("Removing attr %s from the xml result", param->rap_name); pcmk__xe_remove_attr(*result, param->rap_name); } } if (list != NULL) { // We will later search for " WORD ", so end list with a space pcmk__add_word(&list, 0, " "); } return list; } static void append_restart_list(lrmd_event_data_t *op, struct ra_metadata_s *metadata, xmlNode *update, const char *version) { GString *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 PCMK__XA_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 * PCMK__XA_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 PCMK__XA_OP_FORCE_RESTART and PCMK__XA_OP_RESTART_DIGEST to indicate * the resource supports reload, no matter if it actually supports any * reloadable parameters */ crm_xml_add(update, PCMK__XA_OP_FORCE_RESTART, (list == NULL)? "" : (const char *) list->str); crm_xml_add(update, PCMK__XA_OP_RESTART_DIGEST, digest); if ((list != NULL) && (list->len > 0)) { crm_trace("%s: %s, %s", op->rsc_id, digest, (const char *) list->str); } else { crm_trace("%s: %s", op->rsc_id, digest); } if (list != NULL) { g_string_free(list, TRUE); } free_xml(restart); free(digest); } static void append_secure_list(lrmd_event_data_t *op, struct ra_metadata_s *metadata, xmlNode *update, const char *version) { GString *list = NULL; char *digest = NULL; xmlNode *secure = NULL; CRM_LOG_ASSERT(op->params != NULL); /* To keep PCMK__XA_OP_SECURE_PARAMS short, we want it to contain the secure * parameters but PCMK__XA_OP_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, PCMK__XA_OP_SECURE_PARAMS, (const char *) list->str); crm_xml_add(update, PCMK__XA_OP_SECURE_DIGEST, digest); crm_trace("%s: %s, %s", op->rsc_id, digest, (const char *) list->str); g_string_free(list, TRUE); } else { crm_trace("%s: no secure parameters", op->rsc_id); } free_xml(secure); free(digest); } /*! * \internal * \brief Create XML for a resource history entry * * \param[in] func Function name of caller * \param[in,out] parent XML to add entry to * \param[in] rsc Affected resource * \param[in,out] op Action to add an entry for (or NULL to do nothing) * \param[in] node_name Node where action occurred */ void controld_add_resource_history_xml_as(const char *func, xmlNode *parent, const lrmd_rsc_info_t *rsc, lrmd_event_data_t *op, const char *node_name) { 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; if (op == NULL) { return; } target_rc = rsc_op_expected_rc(op); caller_version = g_hash_table_lookup(op->params, PCMK_XA_CRM_FEATURE_SET); CRM_CHECK(caller_version != NULL, caller_version = CRM_FEATURE_SET); xml_op = pcmk__create_history_xml(parent, op, caller_version, target_rc, controld_globals.our_nodename, func); if (xml_op == NULL) { return; } 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; } 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; } /* Ideally the metadata is cached, and the agent is just a fallback. * * @TODO Go through all callers and ensure they get metadata asynchronously * first. */ metadata = controld_get_rsc_metadata(lrm_state, rsc, controld_metadata_from_agent |controld_metadata_from_cache); if (metadata == NULL) { return; } 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; } /*! * \internal * \brief Record an action as pending in the CIB, if appropriate * * \param[in] node_name Node where the action is pending * \param[in] rsc Resource that action is for * \param[in,out] op Pending action * * \return true if action was recorded in CIB, otherwise false */ bool controld_record_pending_op(const char *node_name, const lrmd_rsc_info_t *rsc, lrmd_event_data_t *op) { const char *record_pending = NULL; CRM_CHECK((node_name != NULL) && (rsc != NULL) && (op != NULL), return false); // Never record certain operation types as pending if ((op->op_type == NULL) || (op->params == NULL) || !controld_action_is_recordable(op->op_type)) { return false; } // Check action's PCMK_META_RECORD_PENDING meta-attribute (defaults to true) record_pending = crm_meta_value(op->params, PCMK_META_RECORD_PENDING); if ((record_pending != NULL) && !crm_is_true(record_pending)) { return false; } op->call_id = -1; + // coverity[store_truncates_time_t] op->t_run = time(NULL); op->t_rcchange = op->t_run; lrmd__set_result(op, PCMK_OCF_UNKNOWN, PCMK_EXEC_PENDING, NULL); crm_debug("Recording pending %s-interval %s for %s on %s in the CIB", pcmk__readable_interval(op->interval_ms), op->op_type, op->rsc_id, node_name); controld_update_resource_history(node_name, rsc, op, 0); return true; } 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 history update completed (call=%d rc=%d)", call_id, rc); break; default: if (call_id > 0) { crm_warn("Resource history update %d failed: %s " CRM_XS " rc=%d", call_id, pcmk_strerror(rc), rc); } else { crm_warn("Resource history update failed: %s " CRM_XS " rc=%d", pcmk_strerror(rc), rc); } } if (call_id == pending_rsc_update) { pending_rsc_update = 0; controld_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 (!pcmk_is_set(controld_globals.flags, controld_shutdown_lock_enabled)) { return false; } if (!strcmp(op->op_type, PCMK_ACTION_STOP) && (op->rc == PCMK_OCF_OK)) { return true; } if (!strcmp(op->op_type, PCMK_ACTION_MONITOR) && (op->rc == PCMK_OCF_NOT_RUNNING)) { return true; } return false; } /*! * \internal * \brief Request a CIB update * * \param[in] section Section of CIB to update * \param[in] data New XML of CIB section to update * \param[in] options CIB call options * \param[in] callback If not \c NULL, set this as the operation callback * * \return Standard Pacemaker return code * * \note If \p callback is \p cib_rsc_callback(), the CIB update's call ID is * stored in \p pending_rsc_update on success. */ int controld_update_cib(const char *section, xmlNode *data, int options, void (*callback)(xmlNode *, int, int, xmlNode *, void *)) { cib_t *cib = controld_globals.cib_conn; int cib_rc = -ENOTCONN; CRM_ASSERT(data != NULL); if (cib != NULL) { cib_rc = cib->cmds->modify(cib, section, data, options); if (cib_rc >= 0) { crm_debug("Submitted CIB update %d for %s section", cib_rc, section); } } if (callback == NULL) { if (cib_rc < 0) { crm_err("Failed to update CIB %s section: %s", section, pcmk_rc_str(pcmk_legacy2rc(cib_rc))); } } else { if ((cib_rc >= 0) && (callback == cib_rsc_callback)) { /* Checking for a particular callback is a little hacky, but it * didn't seem worth adding an output argument for cib_rc for just * one use case. */ pending_rsc_update = cib_rc; } fsa_register_cib_callback(cib_rc, NULL, callback); } return (cib_rc >= 0)? pcmk_rc_ok : pcmk_legacy2rc(cib_rc); } /*! * \internal * \brief Update resource history entry in CIB * * \param[in] node_name Node where action occurred * \param[in] rsc Resource that action is for * \param[in,out] op Action to record * \param[in] lock_time If nonzero, when resource was locked to node * * \note On success, the CIB update's call ID will be stored in * pending_rsc_update. */ void controld_update_resource_history(const char *node_name, const lrmd_rsc_info_t *rsc, lrmd_event_data_t *op, time_t lock_time) { xmlNode *update = NULL; xmlNode *xml = NULL; int call_opt = crmd_cib_smart_opt(); const char *node_id = NULL; const char *container = NULL; CRM_CHECK((node_name != NULL) && (op != NULL), return); if (rsc == NULL) { crm_warn("Resource %s no longer exists in the executor", op->rsc_id); controld_ack_event_directly(NULL, NULL, rsc, op, op->rsc_id); return; } // update = pcmk__xe_create(NULL, PCMK_XE_STATUS); // xml = pcmk__xe_create(update, PCMK__XE_NODE_STATE); if (pcmk__str_eq(node_name, controld_globals.our_nodename, pcmk__str_casei)) { node_id = controld_globals.our_uuid; } else { node_id = node_name; pcmk__xe_set_bool_attr(xml, PCMK_XA_REMOTE_NODE, true); } crm_xml_add(xml, PCMK_XA_ID, node_id); crm_xml_add(xml, PCMK_XA_UNAME, node_name); crm_xml_add(xml, PCMK_XA_CRM_DEBUG_ORIGIN, __func__); // xml = pcmk__xe_create(xml, PCMK__XE_LRM); crm_xml_add(xml, PCMK_XA_ID, node_id); // xml = pcmk__xe_create(xml, PCMK__XE_LRM_RESOURCES); // xml = pcmk__xe_create(xml, PCMK__XE_LRM_RESOURCE); crm_xml_add(xml, PCMK_XA_ID, op->rsc_id); crm_xml_add(xml, PCMK_XA_CLASS, rsc->standard); crm_xml_add(xml, PCMK_XA_PROVIDER, rsc->provider); crm_xml_add(xml, PCMK_XA_TYPE, rsc->type); 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(xml, PCMK_OPT_SHUTDOWN_LOCK, (long long) lock_time); } if (op->params != NULL) { container = g_hash_table_lookup(op->params, CRM_META "_" PCMK__META_CONTAINER); if (container != NULL) { crm_trace("Resource %s is a part of container resource %s", op->rsc_id, container); crm_xml_add(xml, PCMK__META_CONTAINER, container); } } // (possibly more than one) controld_add_resource_history_xml(xml, rsc, op, node_name); /* Update CIB asynchronously. Even if it fails, the resource state should be * discovered during the next election. Worst case, the node is wrongly * fenced for running a resource it isn't. */ crm_log_xml_trace(update, __func__); controld_update_cib(PCMK_XE_STATUS, update, call_opt, cib_rsc_callback); free_xml(update); } /*! * \internal * \brief Erase an LRM history entry from the CIB, given the operation data * * \param[in] op Operation whose history should be deleted */ void controld_delete_action_history(const lrmd_event_data_t *op) { xmlNode *xml_top = NULL; CRM_CHECK(op != NULL, return); xml_top = pcmk__xe_create(NULL, PCMK__XE_LRM_RSC_OP); crm_xml_add_int(xml_top, PCMK__XA_CALL_ID, op->call_id); crm_xml_add(xml_top, PCMK__XA_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, PCMK_XA_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); controld_globals.cib_conn->cmds->remove(controld_globals.cib_conn, PCMK_XE_STATUS, xml_top, cib_none); 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 \ "/" PCMK_XE_CIB "/" PCMK_XE_STATUS \ "/" PCMK__XE_NODE_STATE "[@" PCMK_XA_UNAME "='%s']" \ "/" PCMK__XE_LRM "/" PCMK__XE_LRM_RESOURCES \ "/" PCMK__XE_LRM_RESOURCE "[@" PCMK_XA_ID "='%s']" \ "/" PCMK__XE_LRM_RSC_OP /* ... and also by operation key */ #define XPATH_HISTORY_ID XPATH_HISTORY "[@" PCMK_XA_ID "='%s']" /* ... and also by operation key and operation call ID */ #define XPATH_HISTORY_CALL XPATH_HISTORY \ "[@" PCMK_XA_ID "='%s' and @" PCMK__XA_CALL_ID "='%d']" /* ... and also by operation key and original operation key */ #define XPATH_HISTORY_ORIG XPATH_HISTORY \ "[@" PCMK_XA_ID "='%s' and @" PCMK__XA_OPERATION_KEY "='%s']" /*! * \internal * \brief Delete a last_failure resource history entry from the CIB * * \param[in] rsc_id Name of resource to clear history for * \param[in] node Name of node to clear history for * \param[in] action If specified, delete only if this was failed action * \param[in] interval_ms If \p action is specified, it has this interval */ void controld_cib_delete_last_failure(const char *rsc_id, const char *node, const char *action, guint interval_ms) { char *xpath = NULL; char *last_failure_key = NULL; CRM_CHECK((rsc_id != NULL) && (node != NULL), return); // Generate XPath to match desired entry last_failure_key = pcmk__op_key(rsc_id, "last_failure", 0); if (action == NULL) { xpath = crm_strdup_printf(XPATH_HISTORY_ID, node, rsc_id, last_failure_key); } else { char *action_key = pcmk__op_key(rsc_id, action, interval_ms); xpath = crm_strdup_printf(XPATH_HISTORY_ORIG, node, rsc_id, last_failure_key, action_key); free(action_key); } free(last_failure_key); controld_globals.cib_conn->cmds->remove(controld_globals.cib_conn, xpath, NULL, cib_xpath); free(xpath); } /*! * \internal * \brief Delete resource history entry from the CIB, given operation key * * \param[in] rsc_id Name of resource to clear history for * \param[in] node Name of node to clear history for * \param[in] key Operation key of operation to clear history for * \param[in] call_id If specified, delete entry only if it has this call ID */ void controld_delete_action_history_by_key(const char *rsc_id, const char *node, const char *key, int call_id) { char *xpath = NULL; CRM_CHECK((rsc_id != NULL) && (node != NULL) && (key != NULL), return); if (call_id > 0) { xpath = crm_strdup_printf(XPATH_HISTORY_CALL, node, rsc_id, key, call_id); } else { xpath = crm_strdup_printf(XPATH_HISTORY_ID, node, rsc_id, key); } controld_globals.cib_conn->cmds->remove(controld_globals.cib_conn, xpath, NULL, cib_xpath); free(xpath); } diff --git a/daemons/controld/controld_execd.c b/daemons/controld/controld_execd.c index 917c8c0b4d..9979904515 100644 --- a/daemons/controld/controld_execd.c +++ b/daemons/controld/controld_execd.c @@ -1,2454 +1,2455 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include // lrmd_event_data_t, lrmd_rsc_info_t, etc. #include #include #include #include #include #include #define START_DELAY_THRESHOLD 5 * 60 * 1000 #define MAX_LRM_REG_FAILS 30 struct delete_event_s { int rc; const char *rsc; lrm_state_t *lrm_state; }; static gboolean is_rsc_active(lrm_state_t * lrm_state, const char *rsc_id); static gboolean build_active_RAs(lrm_state_t * lrm_state, xmlNode * rsc_list); static gboolean stop_recurring_actions(gpointer key, gpointer value, gpointer user_data); static lrmd_event_data_t *construct_op(const lrm_state_t *lrm_state, const xmlNode *rsc_op, const char *rsc_id, const char *operation); static void do_lrm_rsc_op(lrm_state_t *lrm_state, lrmd_rsc_info_t *rsc, xmlNode *msg, struct ra_metadata_s *md); static gboolean lrm_state_verify_stopped(lrm_state_t * lrm_state, enum crmd_fsa_state cur_state, int log_level); static void lrm_connection_destroy(void) { if (pcmk_is_set(controld_globals.fsa_input_register, R_LRM_CONNECTED)) { crm_crit("Lost connection to local executor"); register_fsa_input(C_FSA_INTERNAL, I_ERROR, NULL); controld_clear_fsa_input_flags(R_LRM_CONNECTED); } } static char * make_stop_id(const char *rsc, int call_id) { return crm_strdup_printf("%s:%d", rsc, call_id); } static void copy_instance_keys(gpointer key, gpointer value, gpointer user_data) { if (strstr(key, CRM_META "_") == NULL) { pcmk__insert_dup(user_data, (const char *) key, (const char *) value); } } static void copy_meta_keys(gpointer key, gpointer value, gpointer user_data) { if (strstr(key, CRM_META "_") != NULL) { pcmk__insert_dup(user_data, (const char *) key, (const char *) value); } } /*! * \internal * \brief Remove a recurring operation from a resource's history * * \param[in,out] history Resource history to modify * \param[in] op Operation to remove * * \return TRUE if the operation was found and removed, FALSE otherwise */ static gboolean history_remove_recurring_op(rsc_history_t *history, const lrmd_event_data_t *op) { GList *iter; for (iter = history->recurring_op_list; iter != NULL; iter = iter->next) { lrmd_event_data_t *existing = iter->data; if ((op->interval_ms == existing->interval_ms) && pcmk__str_eq(op->rsc_id, existing->rsc_id, pcmk__str_none) && pcmk__str_eq(op->op_type, existing->op_type, pcmk__str_casei)) { history->recurring_op_list = g_list_delete_link(history->recurring_op_list, iter); lrmd_free_event(existing); return TRUE; } } return FALSE; } /*! * \internal * \brief Free all recurring operations in resource history * * \param[in,out] history Resource history to modify */ static void history_free_recurring_ops(rsc_history_t *history) { GList *iter; for (iter = history->recurring_op_list; iter != NULL; iter = iter->next) { lrmd_free_event(iter->data); } g_list_free(history->recurring_op_list); history->recurring_op_list = NULL; } /*! * \internal * \brief Free resource history * * \param[in,out] history Resource history to free */ void history_free(gpointer data) { rsc_history_t *history = (rsc_history_t*)data; if (history->stop_params) { g_hash_table_destroy(history->stop_params); } /* Don't need to free history->rsc.id because it's set to history->id */ free(history->rsc.type); free(history->rsc.standard); free(history->rsc.provider); lrmd_free_event(history->failed); lrmd_free_event(history->last); free(history->id); history_free_recurring_ops(history); free(history); } static void update_history_cache(lrm_state_t * lrm_state, lrmd_rsc_info_t * rsc, lrmd_event_data_t * op) { int target_rc = 0; rsc_history_t *entry = NULL; if (op->rsc_deleted) { crm_debug("Purged history for '%s' after %s", op->rsc_id, op->op_type); controld_delete_resource_history(op->rsc_id, lrm_state->node_name, NULL, crmd_cib_smart_opt()); return; } if (pcmk__str_eq(op->op_type, PCMK_ACTION_NOTIFY, pcmk__str_casei)) { return; } crm_debug("Updating history for '%s' with %s op", op->rsc_id, op->op_type); entry = g_hash_table_lookup(lrm_state->resource_history, op->rsc_id); if (entry == NULL && rsc) { entry = pcmk__assert_alloc(1, sizeof(rsc_history_t)); entry->id = pcmk__str_copy(op->rsc_id); g_hash_table_insert(lrm_state->resource_history, entry->id, entry); entry->rsc.id = entry->id; entry->rsc.type = pcmk__str_copy(rsc->type); entry->rsc.standard = pcmk__str_copy(rsc->standard); entry->rsc.provider = pcmk__str_copy(rsc->provider); } else if (entry == NULL) { crm_info("Resource %s no longer exists, not updating cache", op->rsc_id); return; } entry->last_callid = op->call_id; target_rc = rsc_op_expected_rc(op); if (op->op_status == PCMK_EXEC_CANCELLED) { if (op->interval_ms > 0) { crm_trace("Removing cancelled recurring op: " PCMK__OP_FMT, op->rsc_id, op->op_type, op->interval_ms); history_remove_recurring_op(entry, op); return; } else { crm_trace("Skipping " PCMK__OP_FMT " rc=%d, status=%d", op->rsc_id, op->op_type, op->interval_ms, op->rc, op->op_status); } } else if (did_rsc_op_fail(op, target_rc)) { /* Store failed monitors here, otherwise the block below will cause them * to be forgotten when a stop happens. */ if (entry->failed) { lrmd_free_event(entry->failed); } entry->failed = lrmd_copy_event(op); } else if (op->interval_ms == 0) { if (entry->last) { lrmd_free_event(entry->last); } entry->last = lrmd_copy_event(op); if (op->params && pcmk__strcase_any_of(op->op_type, PCMK_ACTION_START, PCMK_ACTION_RELOAD, PCMK_ACTION_RELOAD_AGENT, PCMK_ACTION_MONITOR, NULL)) { if (entry->stop_params) { g_hash_table_destroy(entry->stop_params); } entry->stop_params = pcmk__strkey_table(free, free); g_hash_table_foreach(op->params, copy_instance_keys, entry->stop_params); } } if (op->interval_ms > 0) { /* Ensure there are no duplicates */ history_remove_recurring_op(entry, op); crm_trace("Adding recurring op: " PCMK__OP_FMT, op->rsc_id, op->op_type, op->interval_ms); entry->recurring_op_list = g_list_prepend(entry->recurring_op_list, lrmd_copy_event(op)); } else if ((entry->recurring_op_list != NULL) && !pcmk__str_eq(op->op_type, PCMK_ACTION_MONITOR, pcmk__str_casei)) { crm_trace("Dropping %d recurring ops because of: " PCMK__OP_FMT, g_list_length(entry->recurring_op_list), op->rsc_id, op->op_type, op->interval_ms); history_free_recurring_ops(entry); } } /*! * \internal * \brief Send a direct OK ack for a resource task * * \param[in] lrm_state LRM connection * \param[in] input Input message being ack'ed * \param[in] rsc_id ID of affected resource * \param[in] rsc Affected resource (if available) * \param[in] task Operation task being ack'ed * \param[in] ack_host Name of host to send ack to * \param[in] ack_sys IPC system name to ack */ static void send_task_ok_ack(const lrm_state_t *lrm_state, const ha_msg_input_t *input, const char *rsc_id, const lrmd_rsc_info_t *rsc, const char *task, const char *ack_host, const char *ack_sys) { lrmd_event_data_t *op = construct_op(lrm_state, input->xml, rsc_id, task); lrmd__set_result(op, PCMK_OCF_OK, PCMK_EXEC_DONE, NULL); controld_ack_event_directly(ack_host, ack_sys, rsc, op, rsc_id); lrmd_free_event(op); } static inline const char * op_node_name(lrmd_event_data_t *op) { return pcmk__s(op->remote_nodename, controld_globals.our_nodename); } void lrm_op_callback(lrmd_event_data_t * op) { CRM_CHECK(op != NULL, return); switch (op->type) { case lrmd_event_disconnect: if (op->remote_nodename == NULL) { /* If this is the local executor IPC connection, set the right * bits in the controller when the connection goes down. */ lrm_connection_destroy(); } break; case lrmd_event_exec_complete: { lrm_state_t *lrm_state = lrm_state_find(op_node_name(op)); CRM_ASSERT(lrm_state != NULL); process_lrm_event(lrm_state, op, NULL, NULL); } break; default: break; } } static void try_local_executor_connect(long long action, fsa_data_t *msg_data, lrm_state_t *lrm_state) { int rc = pcmk_rc_ok; crm_debug("Connecting to the local executor"); // If we can connect, great rc = controld_connect_local_executor(lrm_state); if (rc == pcmk_rc_ok) { controld_set_fsa_input_flags(R_LRM_CONNECTED); crm_info("Connection to the local executor established"); return; } // Otherwise, if we can try again, set a timer to do so if (lrm_state->num_lrm_register_fails < MAX_LRM_REG_FAILS) { crm_warn("Failed to connect to the local executor %d time%s " "(%d max): %s", lrm_state->num_lrm_register_fails, pcmk__plural_s(lrm_state->num_lrm_register_fails), MAX_LRM_REG_FAILS, pcmk_rc_str(rc)); controld_start_wait_timer(); crmd_fsa_stall(FALSE); return; } // Otherwise give up crm_err("Failed to connect to the executor the max allowed " "%d time%s: %s", lrm_state->num_lrm_register_fails, pcmk__plural_s(lrm_state->num_lrm_register_fails), pcmk_rc_str(rc)); register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); } /* A_LRM_CONNECT */ void do_lrm_control(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data) { /* This only pertains to local executor connections. Remote connections are * handled as resources within the scheduler. Connecting and disconnecting * from remote executor instances is handled differently. */ lrm_state_t *lrm_state = NULL; if (controld_globals.our_nodename == NULL) { return; /* Nothing to do */ } lrm_state = lrm_state_find_or_create(controld_globals.our_nodename); if (lrm_state == NULL) { register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); return; } if (action & A_LRM_DISCONNECT) { if (lrm_state_verify_stopped(lrm_state, cur_state, LOG_INFO) == FALSE) { if (action == A_LRM_DISCONNECT) { crmd_fsa_stall(FALSE); return; } } controld_clear_fsa_input_flags(R_LRM_CONNECTED); lrm_state_disconnect(lrm_state); lrm_state_reset_tables(lrm_state, FALSE); } if (action & A_LRM_CONNECT) { try_local_executor_connect(action, msg_data, lrm_state); } if (action & ~(A_LRM_CONNECT | A_LRM_DISCONNECT)) { crm_err("Unexpected action %s in %s", fsa_action2string(action), __func__); } } static gboolean lrm_state_verify_stopped(lrm_state_t * lrm_state, enum crmd_fsa_state cur_state, int log_level) { int counter = 0; gboolean rc = TRUE; const char *when = "lrm disconnect"; GHashTableIter gIter; const char *key = NULL; rsc_history_t *entry = NULL; active_op_t *pending = NULL; crm_debug("Checking for active resources before exit"); if (cur_state == S_TERMINATE) { log_level = LOG_ERR; when = "shutdown"; } else if (pcmk_is_set(controld_globals.fsa_input_register, R_SHUTDOWN)) { when = "shutdown... waiting"; } if ((lrm_state->active_ops != NULL) && lrm_state_is_connected(lrm_state)) { guint removed = g_hash_table_foreach_remove(lrm_state->active_ops, stop_recurring_actions, lrm_state); guint nremaining = g_hash_table_size(lrm_state->active_ops); if (removed || nremaining) { crm_notice("Stopped %u recurring operation%s at %s (%u remaining)", removed, pcmk__plural_s(removed), when, nremaining); } } if (lrm_state->active_ops != NULL) { g_hash_table_iter_init(&gIter, lrm_state->active_ops); while (g_hash_table_iter_next(&gIter, NULL, (void **)&pending)) { /* Ignore recurring actions in the shutdown calculations */ if (pending->interval_ms == 0) { counter++; } } } if (counter > 0) { do_crm_log(log_level, "%d pending executor operation%s at %s", counter, pcmk__plural_s(counter), when); if ((cur_state == S_TERMINATE) || !pcmk_is_set(controld_globals.fsa_input_register, R_SENT_RSC_STOP)) { g_hash_table_iter_init(&gIter, lrm_state->active_ops); while (g_hash_table_iter_next(&gIter, (gpointer*)&key, (gpointer*)&pending)) { do_crm_log(log_level, "Pending action: %s (%s)", key, pending->op_key); } } else { rc = FALSE; } return rc; } if (lrm_state->resource_history == NULL) { return rc; } if (pcmk_is_set(controld_globals.fsa_input_register, R_SHUTDOWN)) { /* At this point we're not waiting, we're just shutting down */ when = "shutdown"; } counter = 0; g_hash_table_iter_init(&gIter, lrm_state->resource_history); while (g_hash_table_iter_next(&gIter, NULL, (gpointer*)&entry)) { if (is_rsc_active(lrm_state, entry->id) == FALSE) { continue; } counter++; if (log_level == LOG_ERR) { crm_info("Found %s active at %s", entry->id, when); } else { crm_trace("Found %s active at %s", entry->id, when); } if (lrm_state->active_ops != NULL) { GHashTableIter hIter; g_hash_table_iter_init(&hIter, lrm_state->active_ops); while (g_hash_table_iter_next(&hIter, (gpointer*)&key, (gpointer*)&pending)) { if (pcmk__str_eq(entry->id, pending->rsc_id, pcmk__str_none)) { crm_notice("%sction %s (%s) incomplete at %s", pending->interval_ms == 0 ? "A" : "Recurring a", key, pending->op_key, when); } } } } if (counter) { crm_err("%d resource%s active at %s", counter, (counter == 1)? " was" : "s were", when); } return rc; } static gboolean is_rsc_active(lrm_state_t * lrm_state, const char *rsc_id) { rsc_history_t *entry = NULL; entry = g_hash_table_lookup(lrm_state->resource_history, rsc_id); if (entry == NULL || entry->last == NULL) { return FALSE; } crm_trace("Processing %s: %s.%d=%d", rsc_id, entry->last->op_type, entry->last->interval_ms, entry->last->rc); if ((entry->last->rc == PCMK_OCF_OK) && pcmk__str_eq(entry->last->op_type, PCMK_ACTION_STOP, pcmk__str_casei)) { return FALSE; } else if (entry->last->rc == PCMK_OCF_OK && pcmk__str_eq(entry->last->op_type, PCMK_ACTION_MIGRATE_TO, pcmk__str_casei)) { // A stricter check is too complex ... leave that to the scheduler return FALSE; } else if (entry->last->rc == PCMK_OCF_NOT_RUNNING) { return FALSE; } else if ((entry->last->interval_ms == 0) && (entry->last->rc == PCMK_OCF_NOT_CONFIGURED)) { /* Badly configured resources can't be reliably stopped */ return FALSE; } return TRUE; } static gboolean build_active_RAs(lrm_state_t * lrm_state, xmlNode * rsc_list) { GHashTableIter iter; rsc_history_t *entry = NULL; g_hash_table_iter_init(&iter, lrm_state->resource_history); while (g_hash_table_iter_next(&iter, NULL, (void **)&entry)) { GList *gIter = NULL; xmlNode *xml_rsc = pcmk__xe_create(rsc_list, PCMK__XE_LRM_RESOURCE); crm_xml_add(xml_rsc, PCMK_XA_ID, entry->id); crm_xml_add(xml_rsc, PCMK_XA_TYPE, entry->rsc.type); crm_xml_add(xml_rsc, PCMK_XA_CLASS, entry->rsc.standard); crm_xml_add(xml_rsc, PCMK_XA_PROVIDER, entry->rsc.provider); if (entry->last && entry->last->params) { static const char *name = CRM_META "_" PCMK__META_CONTAINER; const char *container = g_hash_table_lookup(entry->last->params, name); if (container) { crm_trace("Resource %s is a part of container resource %s", entry->id, container); crm_xml_add(xml_rsc, PCMK__META_CONTAINER, container); } } controld_add_resource_history_xml(xml_rsc, &(entry->rsc), entry->failed, lrm_state->node_name); controld_add_resource_history_xml(xml_rsc, &(entry->rsc), entry->last, lrm_state->node_name); for (gIter = entry->recurring_op_list; gIter != NULL; gIter = gIter->next) { controld_add_resource_history_xml(xml_rsc, &(entry->rsc), gIter->data, lrm_state->node_name); } } return FALSE; } xmlNode * controld_query_executor_state(void) { xmlNode *xml_state = NULL; xmlNode *xml_data = NULL; xmlNode *rsc_list = NULL; crm_node_t *peer = NULL; lrm_state_t *lrm_state = lrm_state_find(controld_globals.our_nodename); if (!lrm_state) { crm_err("Could not find executor state for node %s", controld_globals.our_nodename); return NULL; } peer = pcmk__get_node(0, lrm_state->node_name, NULL, pcmk__node_search_any); CRM_CHECK(peer != NULL, return NULL); xml_state = create_node_state_update(peer, node_update_cluster|node_update_peer, NULL, __func__); if (xml_state == NULL) { return NULL; } xml_data = pcmk__xe_create(xml_state, PCMK__XE_LRM); crm_xml_add(xml_data, PCMK_XA_ID, peer->uuid); rsc_list = pcmk__xe_create(xml_data, PCMK__XE_LRM_RESOURCES); /* Build a list of active (not always running) resources */ build_active_RAs(lrm_state, rsc_list); crm_log_xml_trace(xml_state, "Current executor state"); return xml_state; } /*! * \internal * \brief Map standard Pacemaker return code to operation status and OCF code * * \param[out] event Executor event whose status and return code should be set * \param[in] rc Standard Pacemaker return code */ void controld_rc2event(lrmd_event_data_t *event, int rc) { /* This is called for cleanup requests from controller peers/clients, not * for resource actions, so no exit reason is needed. */ switch (rc) { case pcmk_rc_ok: lrmd__set_result(event, PCMK_OCF_OK, PCMK_EXEC_DONE, NULL); break; case EACCES: lrmd__set_result(event, PCMK_OCF_INSUFFICIENT_PRIV, PCMK_EXEC_ERROR, NULL); break; default: lrmd__set_result(event, PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_ERROR, NULL); break; } } /*! * \internal * \brief Trigger a new transition after CIB status was deleted * * If a CIB status delete was not expected (as part of the transition graph), * trigger a new transition by updating the (arbitrary) "last-lrm-refresh" * cluster property. * * \param[in] from_sys IPC name that requested the delete * \param[in] rsc_id Resource whose status was deleted (for logging only) */ void controld_trigger_delete_refresh(const char *from_sys, const char *rsc_id) { if (!pcmk__str_eq(from_sys, CRM_SYSTEM_TENGINE, pcmk__str_casei)) { char *now_s = crm_strdup_printf("%lld", (long long) time(NULL)); crm_debug("Triggering a refresh after %s cleaned %s", from_sys, rsc_id); cib__update_node_attr(controld_globals.logger_out, controld_globals.cib_conn, cib_none, PCMK_XE_CRM_CONFIG, NULL, NULL, NULL, NULL, "last-lrm-refresh", now_s, NULL, NULL); free(now_s); } } static void notify_deleted(lrm_state_t * lrm_state, ha_msg_input_t * input, const char *rsc_id, int rc) { lrmd_event_data_t *op = NULL; const char *from_sys = crm_element_value(input->msg, PCMK__XA_CRM_SYS_FROM); const char *from_host = crm_element_value(input->msg, PCMK__XA_SRC); crm_info("Notifying %s on %s that %s was%s deleted", from_sys, (from_host? from_host : "localhost"), rsc_id, ((rc == pcmk_ok)? "" : " not")); op = construct_op(lrm_state, input->xml, rsc_id, PCMK_ACTION_DELETE); controld_rc2event(op, pcmk_legacy2rc(rc)); controld_ack_event_directly(from_host, from_sys, NULL, op, rsc_id); lrmd_free_event(op); controld_trigger_delete_refresh(from_sys, rsc_id); } static gboolean lrm_remove_deleted_rsc(gpointer key, gpointer value, gpointer user_data) { struct delete_event_s *event = user_data; struct pending_deletion_op_s *op = value; if (pcmk__str_eq(event->rsc, op->rsc, pcmk__str_none)) { notify_deleted(event->lrm_state, op->input, event->rsc, event->rc); return TRUE; } return FALSE; } static gboolean lrm_remove_deleted_op(gpointer key, gpointer value, gpointer user_data) { const char *rsc = user_data; active_op_t *pending = value; if (pcmk__str_eq(rsc, pending->rsc_id, pcmk__str_none)) { crm_info("Removing op %s:%d for deleted resource %s", pending->op_key, pending->call_id, rsc); return TRUE; } return FALSE; } static void delete_rsc_entry(lrm_state_t *lrm_state, ha_msg_input_t *input, const char *rsc_id, GHashTableIter *rsc_iter, int rc, const char *user_name, bool from_cib) { struct delete_event_s event; CRM_CHECK(rsc_id != NULL, return); if (rc == pcmk_ok) { char *rsc_id_copy = pcmk__str_copy(rsc_id); if (rsc_iter) { g_hash_table_iter_remove(rsc_iter); } else { g_hash_table_remove(lrm_state->resource_history, rsc_id_copy); } if (from_cib) { controld_delete_resource_history(rsc_id_copy, lrm_state->node_name, user_name, crmd_cib_smart_opt()); } g_hash_table_foreach_remove(lrm_state->active_ops, lrm_remove_deleted_op, rsc_id_copy); free(rsc_id_copy); } if (input) { notify_deleted(lrm_state, input, rsc_id, rc); } event.rc = rc; event.rsc = rsc_id; event.lrm_state = lrm_state; g_hash_table_foreach_remove(lrm_state->deletion_ops, lrm_remove_deleted_rsc, &event); } static inline gboolean last_failed_matches_op(rsc_history_t *entry, const char *op, guint interval_ms) { if (entry == NULL) { return FALSE; } if (op == NULL) { return TRUE; } return (pcmk__str_eq(op, entry->failed->op_type, pcmk__str_casei) && (interval_ms == entry->failed->interval_ms)); } /*! * \internal * \brief Clear a resource's last failure * * Erase a resource's last failure on a particular node from both the * LRM resource history in the CIB, and the resource history remembered * for the LRM state. * * \param[in] rsc_id Resource name * \param[in] node_name Node name * \param[in] operation If specified, only clear if matching this operation * \param[in] interval_ms If operation is specified, it has this interval */ void lrm_clear_last_failure(const char *rsc_id, const char *node_name, const char *operation, guint interval_ms) { lrm_state_t *lrm_state = lrm_state_find(node_name); if (lrm_state == NULL) { return; } if (lrm_state->resource_history != NULL) { rsc_history_t *entry = g_hash_table_lookup(lrm_state->resource_history, rsc_id); if (last_failed_matches_op(entry, operation, interval_ms)) { lrmd_free_event(entry->failed); entry->failed = NULL; } } } /* Returns: gboolean - cancellation is in progress */ static gboolean cancel_op(lrm_state_t * lrm_state, const char *rsc_id, const char *key, int op, gboolean remove) { int rc = pcmk_ok; char *local_key = NULL; active_op_t *pending = NULL; CRM_CHECK(op != 0, return FALSE); CRM_CHECK(rsc_id != NULL, return FALSE); if (key == NULL) { local_key = make_stop_id(rsc_id, op); key = local_key; } pending = g_hash_table_lookup(lrm_state->active_ops, key); if (pending) { if (remove && !pcmk_is_set(pending->flags, active_op_remove)) { controld_set_active_op_flags(pending, active_op_remove); crm_debug("Scheduling %s for removal", key); } if (pcmk_is_set(pending->flags, active_op_cancelled)) { crm_debug("Operation %s already cancelled", key); free(local_key); return FALSE; } controld_set_active_op_flags(pending, active_op_cancelled); } else { crm_info("No pending op found for %s", key); free(local_key); return FALSE; } crm_debug("Cancelling op %d for %s (%s)", op, rsc_id, key); rc = lrm_state_cancel(lrm_state, pending->rsc_id, pending->op_type, pending->interval_ms); if (rc == pcmk_ok) { crm_debug("Op %d for %s (%s): cancelled", op, rsc_id, key); free(local_key); return TRUE; } crm_debug("Op %d for %s (%s): Nothing to cancel", op, rsc_id, key); /* The caller needs to make sure the entry is * removed from the active operations list * * Usually by returning TRUE inside the worker function * supplied to g_hash_table_foreach_remove() * * Not removing the entry from active operations will block * the node from shutting down */ free(local_key); return FALSE; } struct cancel_data { gboolean done; gboolean remove; const char *key; lrmd_rsc_info_t *rsc; lrm_state_t *lrm_state; }; static gboolean cancel_action_by_key(gpointer key, gpointer value, gpointer user_data) { gboolean remove = FALSE; struct cancel_data *data = user_data; active_op_t *op = value; if (pcmk__str_eq(op->op_key, data->key, pcmk__str_none)) { data->done = TRUE; remove = !cancel_op(data->lrm_state, data->rsc->id, key, op->call_id, data->remove); } return remove; } static gboolean cancel_op_key(lrm_state_t * lrm_state, lrmd_rsc_info_t * rsc, const char *key, gboolean remove) { guint removed = 0; struct cancel_data data; CRM_CHECK(rsc != NULL, return FALSE); CRM_CHECK(key != NULL, return FALSE); data.key = key; data.rsc = rsc; data.done = FALSE; data.remove = remove; data.lrm_state = lrm_state; removed = g_hash_table_foreach_remove(lrm_state->active_ops, cancel_action_by_key, &data); crm_trace("Removed %u op cache entries, new size: %u", removed, g_hash_table_size(lrm_state->active_ops)); return data.done; } /*! * \internal * \brief Retrieve resource information from LRM * * \param[in,out] lrm_state Executor connection state to use * \param[in] rsc_xml XML containing resource configuration * \param[in] do_create If true, register resource if not already * \param[out] rsc_info Where to store information obtained from executor * * \retval pcmk_ok Success (and rsc_info holds newly allocated result) * \retval -EINVAL Required information is missing from arguments * \retval -ENOTCONN No active connection to LRM * \retval -ENODEV Resource not found * \retval -errno Error communicating with executor when registering resource * * \note Caller is responsible for freeing result on success. */ static int get_lrm_resource(lrm_state_t *lrm_state, const xmlNode *rsc_xml, gboolean do_create, lrmd_rsc_info_t **rsc_info) { const char *id = pcmk__xe_id(rsc_xml); CRM_CHECK(lrm_state && rsc_xml && rsc_info, return -EINVAL); CRM_CHECK(id, return -EINVAL); if (lrm_state_is_connected(lrm_state) == FALSE) { return -ENOTCONN; } crm_trace("Retrieving resource information for %s from the executor", id); *rsc_info = lrm_state_get_rsc_info(lrm_state, id, 0); // If resource isn't known by ID, try clone name, if provided if (!*rsc_info) { const char *long_id = crm_element_value(rsc_xml, PCMK__XA_LONG_ID); if (long_id) { *rsc_info = lrm_state_get_rsc_info(lrm_state, long_id, 0); } } if ((*rsc_info == NULL) && do_create) { const char *class = crm_element_value(rsc_xml, PCMK_XA_CLASS); const char *provider = crm_element_value(rsc_xml, PCMK_XA_PROVIDER); const char *type = crm_element_value(rsc_xml, PCMK_XA_TYPE); int rc; crm_trace("Registering resource %s with the executor", id); rc = lrm_state_register_rsc(lrm_state, id, class, provider, type, lrmd_opt_drop_recurring); if (rc != pcmk_ok) { fsa_data_t *msg_data = NULL; crm_err("Could not register resource %s with the executor on %s: %s " CRM_XS " rc=%d", id, lrm_state->node_name, pcmk_strerror(rc), rc); /* Register this as an internal error if this involves the local * executor. Otherwise, we're likely dealing with an unresponsive * remote node, which is not an FSA failure. */ if (lrm_state_is_local(lrm_state) == TRUE) { register_fsa_error(C_FSA_INTERNAL, I_FAIL, NULL); } return rc; } *rsc_info = lrm_state_get_rsc_info(lrm_state, id, 0); } return *rsc_info? pcmk_ok : -ENODEV; } static void delete_resource(lrm_state_t *lrm_state, const char *id, lrmd_rsc_info_t *rsc, GHashTableIter *iter, const char *sys, const char *user, ha_msg_input_t *request, bool unregister, bool from_cib) { int rc = pcmk_ok; crm_info("Removing resource %s from executor for %s%s%s", id, sys, (user? " as " : ""), (user? user : "")); if (rsc && unregister) { rc = lrm_state_unregister_rsc(lrm_state, id, 0); } if (rc == pcmk_ok) { crm_trace("Resource %s deleted from executor", id); } else if (rc == -EINPROGRESS) { crm_info("Deletion of resource '%s' from executor is pending", id); if (request) { struct pending_deletion_op_s *op = NULL; char *ref = crm_element_value_copy(request->msg, PCMK_XA_REFERENCE); op = pcmk__assert_alloc(1, sizeof(struct pending_deletion_op_s)); op->rsc = pcmk__str_copy(rsc->id); op->input = copy_ha_msg_input(request); g_hash_table_insert(lrm_state->deletion_ops, ref, op); } return; } else { crm_warn("Could not delete '%s' from executor for %s%s%s: %s " CRM_XS " rc=%d", id, sys, (user? " as " : ""), (user? user : ""), pcmk_strerror(rc), rc); } delete_rsc_entry(lrm_state, request, id, iter, rc, user, from_cib); } static int get_fake_call_id(lrm_state_t *lrm_state, const char *rsc_id) { int call_id = 999999999; rsc_history_t *entry = NULL; if(lrm_state) { entry = g_hash_table_lookup(lrm_state->resource_history, rsc_id); } /* Make sure the call id is greater than the last successful operation, * otherwise the failure will not result in a possible recovery of the resource * as it could appear the failure occurred before the successful start */ if (entry) { call_id = entry->last_callid + 1; } if (call_id < 0) { call_id = 1; } return call_id; } static void fake_op_status(lrm_state_t *lrm_state, lrmd_event_data_t *op, int op_status, enum ocf_exitcode op_exitcode, const char *exit_reason) { op->call_id = get_fake_call_id(lrm_state, op->rsc_id); + // coverity[store_truncates_time_t] op->t_run = time(NULL); op->t_rcchange = op->t_run; lrmd__set_result(op, op_exitcode, op_status, exit_reason); } static void force_reprobe(lrm_state_t *lrm_state, const char *from_sys, const char *from_host, const char *user_name, gboolean is_remote_node, bool reprobe_all_nodes) { GHashTableIter gIter; rsc_history_t *entry = NULL; crm_info("Clearing resource history on node %s", lrm_state->node_name); g_hash_table_iter_init(&gIter, lrm_state->resource_history); while (g_hash_table_iter_next(&gIter, NULL, (void **)&entry)) { /* only unregister the resource during a reprobe if it is not a remote connection * resource. otherwise unregistering the connection will terminate remote-node * membership */ bool unregister = true; if (is_remote_lrmd_ra(NULL, NULL, entry->id)) { unregister = false; if (reprobe_all_nodes) { lrm_state_t *remote_lrm_state = lrm_state_find(entry->id); if (remote_lrm_state != NULL) { /* If reprobing all nodes, be sure to reprobe the remote * node before clearing its connection resource */ force_reprobe(remote_lrm_state, from_sys, from_host, user_name, TRUE, reprobe_all_nodes); } } } /* Don't delete from the CIB, since we'll delete the whole node's LRM * state from the CIB soon */ delete_resource(lrm_state, entry->id, &entry->rsc, &gIter, from_sys, user_name, NULL, unregister, false); } /* Now delete the copy in the CIB */ controld_delete_node_state(lrm_state->node_name, controld_section_lrm, cib_scope_local); // @COMPAT DCs < 1.1.14 need this deleted (in case it was explicitly false) update_attrd(lrm_state->node_name, CRM_OP_PROBED, NULL, user_name, is_remote_node); } /*! * \internal * \brief Fail a requested action without actually executing it * * For an action that can't be executed, process it similarly to an actual * execution result, with specified error status (except for notify actions, * which will always be treated as successful). * * \param[in,out] lrm_state Executor connection that action is for * \param[in] action Action XML from request * \param[in] rc Desired return code to use * \param[in] op_status Desired operation status to use * \param[in] exit_reason Human-friendly detail, if error */ static void synthesize_lrmd_failure(lrm_state_t *lrm_state, const xmlNode *action, int op_status, enum ocf_exitcode rc, const char *exit_reason) { lrmd_event_data_t *op = NULL; const char *operation = crm_element_value(action, PCMK_XA_OPERATION); const char *target_node = crm_element_value(action, PCMK__META_ON_NODE); xmlNode *xml_rsc = pcmk__xe_first_child(action, PCMK_XE_PRIMITIVE, NULL, NULL); if ((xml_rsc == NULL) || (pcmk__xe_id(xml_rsc) == NULL)) { /* @TODO Should we do something else, like direct ack? */ crm_info("Can't fake %s failure (%d) on %s without resource configuration", crm_element_value(action, PCMK__XA_OPERATION_KEY), rc, target_node); return; } else if(operation == NULL) { /* This probably came from crm_resource -C, nothing to do */ crm_info("Can't fake %s failure (%d) on %s without operation", pcmk__xe_id(xml_rsc), rc, target_node); return; } op = construct_op(lrm_state, action, pcmk__xe_id(xml_rsc), operation); if (pcmk__str_eq(operation, PCMK_ACTION_NOTIFY, pcmk__str_casei)) { // Notifications can't fail fake_op_status(lrm_state, op, PCMK_EXEC_DONE, PCMK_OCF_OK, NULL); } else { fake_op_status(lrm_state, op, op_status, rc, exit_reason); } crm_info("Faking " PCMK__OP_FMT " result (%d) on %s", op->rsc_id, op->op_type, op->interval_ms, op->rc, target_node); // Process the result as if it came from the LRM process_lrm_event(lrm_state, op, NULL, action); lrmd_free_event(op); } /*! * \internal * \brief Get target of an LRM operation (replacing \p NULL with local node * name) * * \param[in] xml LRM operation data XML * * \return LRM operation target node name (local node or Pacemaker Remote node) */ static const char * lrm_op_target(const xmlNode *xml) { const char *target = NULL; if (xml) { target = crm_element_value(xml, PCMK__META_ON_NODE); } if (target == NULL) { target = controld_globals.our_nodename; } return target; } static void fail_lrm_resource(xmlNode *xml, lrm_state_t *lrm_state, const char *user_name, const char *from_host, const char *from_sys) { lrmd_event_data_t *op = NULL; lrmd_rsc_info_t *rsc = NULL; xmlNode *xml_rsc = pcmk__xe_first_child(xml, PCMK_XE_PRIMITIVE, NULL, NULL); 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, pcmk__xe_id(xml_rsc), "asyncmon"); free((char*) op->user_data); op->user_data = NULL; op->interval_ms = 0; if (user_name && !pcmk__is_privileged(user_name)) { crm_err("%s does not have permission to fail %s", user_name, pcmk__xe_id(xml_rsc)); fake_op_status(lrm_state, op, PCMK_EXEC_ERROR, PCMK_OCF_INSUFFICIENT_PRIV, "Unprivileged user cannot fail resources"); controld_ack_event_directly(from_host, from_sys, NULL, op, pcmk__xe_id(xml_rsc)); lrmd_free_event(op); return; } if (get_lrm_resource(lrm_state, xml_rsc, TRUE, &rsc) == pcmk_ok) { crm_info("Failing resource %s...", rsc->id); fake_op_status(lrm_state, op, PCMK_EXEC_DONE, PCMK_OCF_UNKNOWN_ERROR, "Simulated failure"); process_lrm_event(lrm_state, op, NULL, xml); op->rc = PCMK_OCF_OK; // The request to fail the resource succeeded lrmd_free_rsc_info(rsc); } else { crm_info("Cannot find/create resource in order to fail it..."); crm_log_xml_warn(xml, "bad input"); fake_op_status(lrm_state, op, PCMK_EXEC_ERROR, PCMK_OCF_UNKNOWN_ERROR, "Cannot fail unknown resource"); } controld_ack_event_directly(from_host, from_sys, NULL, op, pcmk__xe_id(xml_rsc)); lrmd_free_event(op); } static void handle_reprobe_op(lrm_state_t *lrm_state, const char *from_sys, const char *from_host, const char *user_name, gboolean is_remote_node, bool reprobe_all_nodes) { crm_notice("Forcing the status of all resources to be redetected"); force_reprobe(lrm_state, from_sys, from_host, user_name, is_remote_node, reprobe_all_nodes); if (!pcmk__strcase_any_of(from_sys, CRM_SYSTEM_PENGINE, CRM_SYSTEM_TENGINE, NULL)) { xmlNode *reply = create_request(CRM_OP_INVOKE_LRM, NULL, from_host, from_sys, CRM_SYSTEM_LRMD, controld_globals.our_uuid); crm_debug("ACK'ing re-probe from %s (%s)", from_sys, from_host); if (relay_message(reply, TRUE) == FALSE) { crm_log_xml_err(reply, "Unable to route reply"); } free_xml(reply); } } static bool do_lrm_cancel(ha_msg_input_t *input, lrm_state_t *lrm_state, lrmd_rsc_info_t *rsc, const char *from_host, const char *from_sys) { char *op_key = NULL; char *meta_key = NULL; int call = 0; const char *call_id = NULL; const char *op_task = NULL; guint interval_ms = 0; gboolean in_progress = FALSE; xmlNode *params = pcmk__xe_first_child(input->xml, PCMK__XE_ATTRIBUTES, NULL, NULL); CRM_CHECK(params != NULL, return FALSE); meta_key = crm_meta_name(PCMK_XA_OPERATION); op_task = crm_element_value(params, meta_key); free(meta_key); CRM_CHECK(op_task != NULL, return FALSE); meta_key = crm_meta_name(PCMK_META_INTERVAL); if (crm_element_value_ms(params, meta_key, &interval_ms) != pcmk_ok) { free(meta_key); return FALSE; } free(meta_key); op_key = pcmk__op_key(rsc->id, op_task, interval_ms); meta_key = crm_meta_name(PCMK__XA_CALL_ID); call_id = crm_element_value(params, meta_key); free(meta_key); crm_debug("Scheduler requested op %s (call=%s) be cancelled", op_key, (call_id? call_id : "NA")); pcmk__scan_min_int(call_id, &call, 0); if (call == 0) { // Normal case when the scheduler cancels a recurring op in_progress = cancel_op_key(lrm_state, rsc, op_key, TRUE); } else { // Normal case when the scheduler cancels an orphan op in_progress = cancel_op(lrm_state, rsc->id, NULL, call, TRUE); } // Acknowledge cancellation operation if for a remote connection resource if (!in_progress || is_remote_lrmd_ra(NULL, NULL, rsc->id)) { char *op_id = make_stop_id(rsc->id, call); if (is_remote_lrmd_ra(NULL, NULL, rsc->id) == FALSE) { crm_info("Nothing known about operation %d for %s", call, op_key); } controld_delete_action_history_by_key(rsc->id, lrm_state->node_name, op_key, call); send_task_ok_ack(lrm_state, input, rsc->id, rsc, op_task, from_host, from_sys); /* needed at least for cancellation of a remote operation */ if (lrm_state->active_ops != NULL) { g_hash_table_remove(lrm_state->active_ops, op_id); } free(op_id); } else { /* No ack is needed since abcdaa8, but peers with older versions * in a rolling upgrade need one. We didn't bump the feature set * at that commit, so we can only compare against the previous * CRM version (3.0.8). If any peers have feature set 3.0.9 but * not abcdaa8, they will time out waiting for the ack (no * released versions of Pacemaker are affected). */ const char *peer_version = crm_element_value(params, PCMK_XA_CRM_FEATURE_SET); if (compare_version(peer_version, "3.0.8") <= 0) { crm_info("Sending compatibility ack for %s cancellation to %s (CRM version %s)", op_key, from_host, peer_version); send_task_ok_ack(lrm_state, input, rsc->id, rsc, op_task, from_host, from_sys); } } free(op_key); return TRUE; } static void do_lrm_delete(ha_msg_input_t *input, lrm_state_t *lrm_state, lrmd_rsc_info_t *rsc, const char *from_sys, const char *from_host, bool crm_rsc_delete, const char *user_name) { bool unregister = true; int cib_rc = controld_delete_resource_history(rsc->id, lrm_state->node_name, user_name, cib_dryrun|cib_sync_call); if (cib_rc != pcmk_rc_ok) { lrmd_event_data_t *op = NULL; op = construct_op(lrm_state, input->xml, rsc->id, PCMK_ACTION_DELETE); /* These are resource clean-ups, not actions, so no exit reason is * needed. */ lrmd__set_result(op, pcmk_rc2ocf(cib_rc), PCMK_EXEC_ERROR, NULL); controld_ack_event_directly(from_host, from_sys, NULL, op, rsc->id); lrmd_free_event(op); return; } if (crm_rsc_delete && is_remote_lrmd_ra(NULL, NULL, rsc->id)) { unregister = false; } delete_resource(lrm_state, rsc->id, rsc, NULL, from_sys, user_name, input, unregister, true); } // User data for asynchronous metadata execution struct metadata_cb_data { lrmd_rsc_info_t *rsc; // Copy of resource information xmlNode *input_xml; // Copy of FSA input XML }; static struct metadata_cb_data * new_metadata_cb_data(lrmd_rsc_info_t *rsc, xmlNode *input_xml) { struct metadata_cb_data *data = NULL; data = pcmk__assert_alloc(1, sizeof(struct metadata_cb_data)); data->input_xml = pcmk__xml_copy(NULL, input_xml); data->rsc = lrmd_copy_rsc_info(rsc); return data; } static void free_metadata_cb_data(struct metadata_cb_data *data) { lrmd_free_rsc_info(data->rsc); free_xml(data->input_xml); free(data); } /*! * \internal * \brief Execute an action after metadata has been retrieved * * \param[in] pid Ignored * \param[in] result Result of metadata action * \param[in] user_data Metadata callback data */ static void metadata_complete(int pid, const pcmk__action_result_t *result, void *user_data) { struct metadata_cb_data *data = (struct metadata_cb_data *) user_data; struct ra_metadata_s *md = NULL; lrm_state_t *lrm_state = lrm_state_find(lrm_op_target(data->input_xml)); if ((lrm_state != NULL) && pcmk__result_ok(result)) { md = controld_cache_metadata(lrm_state->metadata_cache, data->rsc, result->action_stdout); } if (!pcmk_is_set(controld_globals.fsa_input_register, R_HA_DISCONNECTED)) { do_lrm_rsc_op(lrm_state, data->rsc, data->input_xml, md); } free_metadata_cb_data(data); } /* A_LRM_INVOKE */ void do_lrm_invoke(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data) { lrm_state_t *lrm_state = NULL; const char *crm_op = NULL; const char *from_sys = NULL; const char *from_host = NULL; const char *operation = NULL; ha_msg_input_t *input = fsa_typed_data(fsa_dt_ha_msg); const char *user_name = NULL; const char *target_node = lrm_op_target(input->xml); gboolean is_remote_node = FALSE; bool crm_rsc_delete = FALSE; // Message routed to the local node is targeting a specific, non-local node is_remote_node = !pcmk__str_eq(target_node, controld_globals.our_nodename, pcmk__str_casei); lrm_state = lrm_state_find(target_node); if ((lrm_state == NULL) && is_remote_node) { crm_err("Failing action because local node has never had connection to remote node %s", target_node); synthesize_lrmd_failure(NULL, input->xml, PCMK_EXEC_NOT_CONNECTED, PCMK_OCF_UNKNOWN_ERROR, "Local node has no connection to remote"); return; } CRM_ASSERT(lrm_state != NULL); user_name = pcmk__update_acl_user(input->msg, PCMK__XA_CRM_USER, NULL); crm_op = crm_element_value(input->msg, PCMK__XA_CRM_TASK); from_sys = crm_element_value(input->msg, PCMK__XA_CRM_SYS_FROM); if (!pcmk__str_eq(from_sys, CRM_SYSTEM_TENGINE, pcmk__str_none)) { from_host = crm_element_value(input->msg, PCMK__XA_SRC); } if (pcmk__str_eq(crm_op, PCMK_ACTION_LRM_DELETE, pcmk__str_none)) { if (!pcmk__str_eq(from_sys, CRM_SYSTEM_TENGINE, pcmk__str_none)) { crm_rsc_delete = TRUE; // from crm_resource } operation = PCMK_ACTION_DELETE; } else if (input->xml != NULL) { operation = crm_element_value(input->xml, PCMK_XA_OPERATION); } CRM_CHECK(!pcmk__str_empty(crm_op) || !pcmk__str_empty(operation), return); crm_trace("'%s' execution request from %s as %s user", pcmk__s(crm_op, operation), pcmk__s(from_sys, "unknown subsystem"), pcmk__s(user_name, "current")); if (pcmk__str_eq(crm_op, CRM_OP_LRM_FAIL, pcmk__str_none)) { fail_lrm_resource(input->xml, lrm_state, user_name, from_host, from_sys); } else if (pcmk__str_eq(crm_op, CRM_OP_LRM_REFRESH, pcmk__str_none)) { /* @COMPAT This can only be sent by crm_resource --refresh on a * Pacemaker Remote node running Pacemaker 1.1.9, which is extremely * unlikely. It previously would cause the controller to re-write its * resource history to the CIB. Just ignore it. */ crm_notice("Ignoring refresh request from Pacemaker Remote 1.1.9 node"); // @COMPAT DCs <1.1.14 in a rolling upgrade might schedule this op } else if (pcmk__str_eq(operation, CRM_OP_PROBED, pcmk__str_none)) { update_attrd(lrm_state->node_name, CRM_OP_PROBED, PCMK_VALUE_TRUE, user_name, is_remote_node); } else if (pcmk__str_eq(crm_op, CRM_OP_REPROBE, pcmk__str_none) || pcmk__str_eq(operation, CRM_OP_REPROBE, pcmk__str_none)) { const char *raw_target = NULL; if (input->xml != NULL) { // For CRM_OP_REPROBE, a NULL target means we're targeting all nodes raw_target = crm_element_value(input->xml, PCMK__META_ON_NODE); } handle_reprobe_op(lrm_state, from_sys, from_host, user_name, is_remote_node, (raw_target == NULL)); } else if (operation != NULL) { lrmd_rsc_info_t *rsc = NULL; xmlNode *xml_rsc = pcmk__xe_first_child(input->xml, PCMK_XE_PRIMITIVE, NULL, NULL); gboolean create_rsc = !pcmk__str_eq(operation, PCMK_ACTION_DELETE, pcmk__str_none); int rc; // We can't return anything meaningful without a resource ID CRM_CHECK((xml_rsc != NULL) && (pcmk__xe_id(xml_rsc) != NULL), return); rc = get_lrm_resource(lrm_state, xml_rsc, create_rsc, &rsc); if (rc == -ENOTCONN) { synthesize_lrmd_failure(lrm_state, input->xml, PCMK_EXEC_NOT_CONNECTED, PCMK_OCF_UNKNOWN_ERROR, "Not connected to remote executor"); return; } else if ((rc < 0) && !create_rsc) { /* Delete of malformed or nonexistent resource * (deleting something that does not exist is a success) */ crm_notice("Not registering resource '%s' for a %s event " CRM_XS " get-rc=%d (%s) transition-key=%s", pcmk__xe_id(xml_rsc), operation, rc, pcmk_strerror(rc), pcmk__xe_id(input->xml)); delete_rsc_entry(lrm_state, input, pcmk__xe_id(xml_rsc), NULL, pcmk_ok, user_name, true); return; } else if (rc == -EINVAL) { // Resource operation on malformed resource crm_err("Invalid resource definition for %s", pcmk__xe_id(xml_rsc)); crm_log_xml_warn(input->msg, "invalid resource"); synthesize_lrmd_failure(lrm_state, input->xml, PCMK_EXEC_ERROR, PCMK_OCF_NOT_CONFIGURED, // fatal error "Invalid resource definition"); return; } else if (rc < 0) { // Error communicating with the executor crm_err("Could not register resource '%s' with executor: %s " CRM_XS " rc=%d", pcmk__xe_id(xml_rsc), pcmk_strerror(rc), rc); crm_log_xml_warn(input->msg, "failed registration"); synthesize_lrmd_failure(lrm_state, input->xml, PCMK_EXEC_ERROR, PCMK_OCF_INVALID_PARAM, // hard error "Could not register resource with executor"); return; } if (pcmk__str_eq(operation, PCMK_ACTION_CANCEL, pcmk__str_none)) { if (!do_lrm_cancel(input, lrm_state, rsc, from_host, from_sys)) { crm_log_xml_warn(input->xml, "Bad command"); } } else if (pcmk__str_eq(operation, PCMK_ACTION_DELETE, pcmk__str_none)) { do_lrm_delete(input, lrm_state, rsc, from_sys, from_host, crm_rsc_delete, user_name); } else { struct ra_metadata_s *md = NULL; /* Getting metadata from cache is OK except for start actions -- * always refresh from the agent for those, in case the resource * agent was updated. * * @TODO Only refresh metadata for starts if the agent actually * changed (using something like inotify, or a hash or modification * time of the agent executable). */ if (strcmp(operation, PCMK_ACTION_START) != 0) { md = controld_get_rsc_metadata(lrm_state, rsc, controld_metadata_from_cache); } if ((md == NULL) && crm_op_needs_metadata(rsc->standard, operation)) { /* Most likely, we'll need the agent metadata to record the * pending operation and the operation result. Get it now rather * than wait until then, so the metadata action doesn't eat into * the real action's timeout. * * @TODO Metadata is retrieved via direct execution of the * agent, which has a couple of related issues: the executor * should execute agents, not the controller; and metadata for * Pacemaker Remote nodes should be collected on those nodes, * not locally. */ struct metadata_cb_data *data = NULL; data = new_metadata_cb_data(rsc, input->xml); crm_info("Retrieving metadata for %s (%s%s%s:%s) asynchronously", rsc->id, rsc->standard, ((rsc->provider == NULL)? "" : ":"), ((rsc->provider == NULL)? "" : rsc->provider), rsc->type); (void) lrmd__metadata_async(rsc, metadata_complete, (void *) data); } else { do_lrm_rsc_op(lrm_state, rsc, input->xml, md); } } lrmd_free_rsc_info(rsc); } else { crm_err("Invalid execution request: unknown command '%s' (bug?)", crm_op); register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); } } static lrmd_event_data_t * construct_op(const lrm_state_t *lrm_state, const xmlNode *rsc_op, const char *rsc_id, const char *operation) { lrmd_event_data_t *op = NULL; const char *op_delay = NULL; const char *op_timeout = NULL; GHashTable *params = NULL; xmlNode *primitive = NULL; const char *class = NULL; const char *transition = NULL; CRM_ASSERT(rsc_id && operation); op = lrmd_new_event(rsc_id, operation, 0); op->type = lrmd_event_exec_complete; op->timeout = 0; op->start_delay = 0; lrmd__set_result(op, PCMK_OCF_UNKNOWN, PCMK_EXEC_PENDING, NULL); if (rsc_op == NULL) { CRM_LOG_ASSERT(pcmk__str_eq(operation, PCMK_ACTION_STOP, pcmk__str_casei)); op->user_data = NULL; /* the stop_all_resources() case * by definition there is no DC (or they'd be shutting * us down). * So we should put our version here. */ op->params = pcmk__strkey_table(free, free); pcmk__insert_dup(op->params, PCMK_XA_CRM_FEATURE_SET, CRM_FEATURE_SET); crm_trace("Constructed %s op for %s", operation, rsc_id); return op; } params = xml2list(rsc_op); g_hash_table_remove(params, CRM_META "_" PCMK__META_OP_TARGET_RC); op_delay = crm_meta_value(params, PCMK_META_START_DELAY); pcmk__scan_min_int(op_delay, &op->start_delay, 0); op_timeout = crm_meta_value(params, PCMK_META_TIMEOUT); pcmk__scan_min_int(op_timeout, &op->timeout, 0); if (pcmk__guint_from_hash(params, CRM_META "_" PCMK_META_INTERVAL, 0, &(op->interval_ms)) != pcmk_rc_ok) { op->interval_ms = 0; } /* Use pcmk_monitor_timeout instead of meta timeout for stonith recurring monitor, if set */ primitive = pcmk__xe_first_child(rsc_op, PCMK_XE_PRIMITIVE, NULL, NULL); class = crm_element_value(primitive, PCMK_XA_CLASS); if (pcmk_is_set(pcmk_get_ra_caps(class), pcmk_ra_cap_fence_params) && pcmk__str_eq(operation, PCMK_ACTION_MONITOR, pcmk__str_casei) && (op->interval_ms > 0)) { op_timeout = g_hash_table_lookup(params, "pcmk_monitor_timeout"); if (op_timeout != NULL) { long long timeout_ms = crm_get_msec(op_timeout); op->timeout = (int) QB_MIN(timeout_ms, INT_MAX); } } if (!pcmk__str_eq(operation, PCMK_ACTION_STOP, pcmk__str_casei)) { op->params = params; } else { rsc_history_t *entry = NULL; if (lrm_state) { entry = g_hash_table_lookup(lrm_state->resource_history, rsc_id); } /* If we do not have stop parameters cached, use * whatever we are given */ if (!entry || !entry->stop_params) { op->params = params; } else { /* Copy the cached parameter list so that we stop the resource * with the old attributes, not the new ones */ op->params = pcmk__strkey_table(free, free); g_hash_table_foreach(params, copy_meta_keys, op->params); g_hash_table_foreach(entry->stop_params, copy_instance_keys, op->params); g_hash_table_destroy(params); params = NULL; } } /* sanity */ if (op->timeout <= 0) { op->timeout = op->interval_ms; } if (op->start_delay < 0) { op->start_delay = 0; } transition = crm_element_value(rsc_op, PCMK__XA_TRANSITION_KEY); CRM_CHECK(transition != NULL, return op); op->user_data = pcmk__str_copy(transition); if (op->interval_ms != 0) { if (pcmk__strcase_any_of(operation, PCMK_ACTION_START, PCMK_ACTION_STOP, NULL)) { crm_err("Start and Stop actions cannot have an interval: %u", op->interval_ms); op->interval_ms = 0; } } crm_trace("Constructed %s op for %s: interval=%u", operation, rsc_id, op->interval_ms); return op; } /*! * \internal * \brief Send a (synthesized) event result * * Reply with a synthesized event result directly, as opposed to going through * the executor. * * \param[in] to_host Host to send result to * \param[in] to_sys IPC name to send result (NULL for transition engine) * \param[in] rsc Type information about resource the result is for * \param[in,out] op Event with result to send * \param[in] rsc_id ID of resource the result is for */ void controld_ack_event_directly(const char *to_host, const char *to_sys, const lrmd_rsc_info_t *rsc, lrmd_event_data_t *op, const char *rsc_id) { xmlNode *reply = NULL; xmlNode *update, *iter; crm_node_t *peer = NULL; CRM_CHECK(op != NULL, return); if (op->rsc_id == NULL) { // op->rsc_id is a (const char *) but lrmd_free_event() frees it CRM_ASSERT(rsc_id != NULL); op->rsc_id = pcmk__str_copy(rsc_id); } if (to_sys == NULL) { to_sys = CRM_SYSTEM_TENGINE; } peer = pcmk__get_node(0, controld_globals.our_nodename, NULL, pcmk__node_search_cluster_member); update = create_node_state_update(peer, node_update_none, NULL, __func__); iter = pcmk__xe_create(update, PCMK__XE_LRM); crm_xml_add(iter, PCMK_XA_ID, controld_globals.our_uuid); iter = pcmk__xe_create(iter, PCMK__XE_LRM_RESOURCES); iter = pcmk__xe_create(iter, PCMK__XE_LRM_RESOURCE); crm_xml_add(iter, PCMK_XA_ID, op->rsc_id); controld_add_resource_history_xml(iter, rsc, op, controld_globals.our_nodename); reply = create_request(CRM_OP_INVOKE_LRM, update, to_host, to_sys, CRM_SYSTEM_LRMD, NULL); crm_log_xml_trace(update, "[direct ACK]"); crm_debug("ACK'ing resource op " PCMK__OP_FMT " from %s: %s", op->rsc_id, op->op_type, op->interval_ms, op->user_data, crm_element_value(reply, PCMK_XA_REFERENCE)); if (relay_message(reply, TRUE) == FALSE) { crm_log_xml_err(reply, "Unable to route reply"); } free_xml(update); free_xml(reply); } gboolean verify_stopped(enum crmd_fsa_state cur_state, int log_level) { gboolean res = TRUE; GList *lrm_state_list = lrm_state_get_list(); GList *state_entry; for (state_entry = lrm_state_list; state_entry != NULL; state_entry = state_entry->next) { lrm_state_t *lrm_state = state_entry->data; if (!lrm_state_verify_stopped(lrm_state, cur_state, log_level)) { /* keep iterating through all even when false is returned */ res = FALSE; } } controld_set_fsa_input_flags(R_SENT_RSC_STOP); g_list_free(lrm_state_list); lrm_state_list = NULL; return res; } struct stop_recurring_action_s { lrmd_rsc_info_t *rsc; lrm_state_t *lrm_state; }; static gboolean stop_recurring_action_by_rsc(gpointer key, gpointer value, gpointer user_data) { gboolean remove = FALSE; struct stop_recurring_action_s *event = user_data; active_op_t *op = value; if ((op->interval_ms != 0) && pcmk__str_eq(op->rsc_id, event->rsc->id, pcmk__str_none)) { crm_debug("Cancelling op %d for %s (%s)", op->call_id, op->rsc_id, (char*)key); remove = !cancel_op(event->lrm_state, event->rsc->id, key, op->call_id, FALSE); } return remove; } static gboolean stop_recurring_actions(gpointer key, gpointer value, gpointer user_data) { gboolean remove = FALSE; lrm_state_t *lrm_state = user_data; active_op_t *op = value; if (op->interval_ms != 0) { crm_info("Cancelling op %d for %s (%s)", op->call_id, op->rsc_id, (const char *) key); remove = !cancel_op(lrm_state, op->rsc_id, key, op->call_id, FALSE); } return remove; } /*! * \internal * \brief Check whether recurring actions should be cancelled before an action * * \param[in] rsc_id Resource that action is for * \param[in] action Action being performed * \param[in] interval_ms Operation interval of \p action (in milliseconds) * * \return true if recurring actions should be cancelled, otherwise false */ static bool should_cancel_recurring(const char *rsc_id, const char *action, guint interval_ms) { if (is_remote_lrmd_ra(NULL, NULL, rsc_id) && (interval_ms == 0) && (strcmp(action, PCMK_ACTION_MIGRATE_TO) == 0)) { /* Don't stop monitoring a migrating Pacemaker Remote connection * resource until the entire migration has completed. We must detect if * the connection is unexpectedly severed, even during a migration. */ return false; } // Cancel recurring actions before changing resource state return (interval_ms == 0) && !pcmk__str_any_of(action, PCMK_ACTION_MONITOR, PCMK_ACTION_NOTIFY, NULL); } /*! * \internal * \brief Check whether an action should not be performed at this time * * \param[in] operation Action to be performed * * \return Readable description of why action should not be performed, * or NULL if it should be performed */ static const char * should_nack_action(const char *action) { if (pcmk_is_set(controld_globals.fsa_input_register, R_SHUTDOWN) && pcmk__str_eq(action, PCMK_ACTION_START, pcmk__str_none)) { register_fsa_input(C_SHUTDOWN, I_SHUTDOWN, NULL); return "Not attempting start due to shutdown in progress"; } switch (controld_globals.fsa_state) { case S_NOT_DC: case S_POLICY_ENGINE: // Recalculating case S_TRANSITION_ENGINE: break; default: if (!pcmk__str_eq(action, PCMK_ACTION_STOP, pcmk__str_none)) { return "Controller cannot attempt actions at this time"; } break; } return NULL; } static void do_lrm_rsc_op(lrm_state_t *lrm_state, lrmd_rsc_info_t *rsc, xmlNode *msg, struct ra_metadata_s *md) { int rc; int call_id = 0; char *op_id = NULL; lrmd_event_data_t *op = NULL; fsa_data_t *msg_data = NULL; const char *transition = NULL; const char *operation = NULL; const char *nack_reason = NULL; CRM_CHECK((rsc != NULL) && (msg != NULL), return); operation = crm_element_value(msg, PCMK_XA_OPERATION); CRM_CHECK(!pcmk__str_empty(operation), return); transition = crm_element_value(msg, PCMK__XA_TRANSITION_KEY); if (pcmk__str_empty(transition)) { crm_log_xml_err(msg, "Missing transition number"); } if (lrm_state == NULL) { // This shouldn't be possible, but provide a failsafe just in case crm_err("Cannot execute %s of %s: No executor connection " CRM_XS " transition_key=%s", operation, rsc->id, pcmk__s(transition, "")); synthesize_lrmd_failure(NULL, msg, PCMK_EXEC_INVALID, PCMK_OCF_UNKNOWN_ERROR, "No executor connection"); return; } if (pcmk__str_any_of(operation, PCMK_ACTION_RELOAD, PCMK_ACTION_RELOAD_AGENT, NULL)) { /* Pre-2.1.0 DCs will schedule reload actions only, and 2.1.0+ DCs * will schedule reload-agent actions only. In either case, we need * to map that to whatever the resource agent actually supports. * Default to the OCF 1.1 name. */ if ((md != NULL) && pcmk_is_set(md->ra_flags, ra_supports_legacy_reload)) { operation = PCMK_ACTION_RELOAD; } else { operation = PCMK_ACTION_RELOAD_AGENT; } } op = construct_op(lrm_state, msg, rsc->id, operation); CRM_CHECK(op != NULL, return); if (should_cancel_recurring(rsc->id, operation, op->interval_ms)) { guint removed = 0; struct stop_recurring_action_s data; data.rsc = rsc; data.lrm_state = lrm_state; removed = g_hash_table_foreach_remove(lrm_state->active_ops, stop_recurring_action_by_rsc, &data); if (removed) { crm_debug("Stopped %u recurring operation%s in preparation for " PCMK__OP_FMT, removed, pcmk__plural_s(removed), rsc->id, operation, op->interval_ms); } } /* now do the op */ crm_notice("Requesting local execution of %s operation for %s on %s " CRM_XS " transition_key=%s op_key=" PCMK__OP_FMT, pcmk__readable_action(op->op_type, op->interval_ms), rsc->id, lrm_state->node_name, pcmk__s(transition, ""), rsc->id, operation, op->interval_ms); nack_reason = should_nack_action(operation); if (nack_reason != NULL) { crm_notice("Discarding attempt to perform action %s on %s in state %s " "(shutdown=%s)", operation, rsc->id, fsa_state2string(controld_globals.fsa_state), pcmk__flag_text(controld_globals.fsa_input_register, R_SHUTDOWN)); lrmd__set_result(op, PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_INVALID, nack_reason); controld_ack_event_directly(NULL, NULL, rsc, op, rsc->id); lrmd_free_event(op); free(op_id); return; } controld_record_pending_op(lrm_state->node_name, rsc, op); op_id = pcmk__op_key(rsc->id, op->op_type, op->interval_ms); if (op->interval_ms > 0) { /* cancel it so we can then restart it without conflict */ cancel_op_key(lrm_state, rsc, op_id, FALSE); } rc = controld_execute_resource_agent(lrm_state, rsc->id, op->op_type, op->user_data, op->interval_ms, op->timeout, op->start_delay, op->params, &call_id); if (rc == pcmk_rc_ok) { /* record all operations so we can wait * for them to complete during shutdown */ char *call_id_s = make_stop_id(rsc->id, call_id); active_op_t *pending = NULL; pending = pcmk__assert_alloc(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 = pcmk__str_copy(operation); pending->op_key = pcmk__str_copy(op_id); pending->rsc_id = pcmk__str_copy(rsc->id); pending->start_time = time(NULL); pending->user_data = pcmk__str_copy(op->user_data); if (crm_element_value_epoch(msg, PCMK_OPT_SHUTDOWN_LOCK, &(pending->lock_time)) != pcmk_ok) { pending->lock_time = 0; } g_hash_table_replace(lrm_state->active_ops, call_id_s, pending); if ((op->interval_ms > 0) && (op->start_delay > START_DELAY_THRESHOLD)) { int target_rc = PCMK_OCF_OK; crm_info("Faking confirmation of %s: execution postponed for over 5 minutes", op_id); decode_transition_key(op->user_data, NULL, NULL, NULL, &target_rc); lrmd__set_result(op, target_rc, PCMK_EXEC_DONE, NULL); controld_ack_event_directly(NULL, NULL, rsc, op, rsc->id); } pending->params = op->params; op->params = NULL; } else if (lrm_state_is_local(lrm_state)) { crm_err("Could not initiate %s action for resource %s locally: %s " CRM_XS " rc=%d", operation, rsc->id, pcmk_rc_str(rc), rc); fake_op_status(lrm_state, op, PCMK_EXEC_NOT_CONNECTED, PCMK_OCF_UNKNOWN_ERROR, pcmk_rc_str(rc)); process_lrm_event(lrm_state, op, NULL, NULL); register_fsa_error(C_FSA_INTERNAL, I_FAIL, NULL); } else { crm_err("Could not initiate %s action for resource %s remotely on %s: " "%s " CRM_XS " rc=%d", operation, rsc->id, lrm_state->node_name, pcmk_rc_str(rc), rc); fake_op_status(lrm_state, op, PCMK_EXEC_NOT_CONNECTED, PCMK_OCF_UNKNOWN_ERROR, pcmk_rc_str(rc)); process_lrm_event(lrm_state, op, NULL, NULL); } free(op_id); lrmd_free_event(op); } void do_lrm_event(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input cur_input, fsa_data_t * msg_data) { CRM_CHECK(FALSE, return); } static char * unescape_newlines(const char *string) { char *pch = NULL; char *ret = NULL; static const char *escaped_newline = "\\n"; if (!string) { return NULL; } ret = pcmk__str_copy(string); pch = strstr(ret, escaped_newline); while (pch != NULL) { /* Replace newline escape pattern with actual newline (and a space so we * don't have to shuffle the rest of the buffer) */ pch[0] = '\n'; pch[1] = ' '; pch = strstr(pch, escaped_newline); } return ret; } static bool did_lrm_rsc_op_fail(lrm_state_t *lrm_state, const char * rsc_id, const char * op_type, guint interval_ms) { rsc_history_t *entry = NULL; CRM_CHECK(lrm_state != NULL, return FALSE); CRM_CHECK(rsc_id != NULL, return FALSE); CRM_CHECK(op_type != NULL, return FALSE); entry = g_hash_table_lookup(lrm_state->resource_history, rsc_id); if (entry == NULL || entry->failed == NULL) { return FALSE; } if (pcmk__str_eq(entry->failed->rsc_id, rsc_id, pcmk__str_none) && pcmk__str_eq(entry->failed->op_type, op_type, pcmk__str_casei) && entry->failed->interval_ms == interval_ms) { return TRUE; } return FALSE; } /*! * \internal * \brief Log the result of an executor action (actual or synthesized) * * \param[in] op Executor action to log result for * \param[in] op_key Operation key for action * \param[in] node_name Name of node action was performed on, if known * \param[in] confirmed Whether to log that graph action was confirmed */ static void log_executor_event(const lrmd_event_data_t *op, const char *op_key, const char *node_name, gboolean confirmed) { int log_level = LOG_ERR; GString *str = g_string_sized_new(100); // reasonable starting size pcmk__g_strcat(str, "Result of ", pcmk__readable_action(op->op_type, op->interval_ms), " operation for ", op->rsc_id, NULL); if (node_name != NULL) { pcmk__g_strcat(str, " on ", node_name, NULL); } switch (op->op_status) { case PCMK_EXEC_DONE: log_level = LOG_NOTICE; pcmk__g_strcat(str, ": ", services_ocf_exitcode_str(op->rc), NULL); break; case PCMK_EXEC_TIMEOUT: pcmk__g_strcat(str, ": ", pcmk_exec_status_str(op->op_status), " after ", pcmk__readable_interval(op->timeout), NULL); break; case PCMK_EXEC_CANCELLED: log_level = LOG_INFO; /* order of __attribute__ and Fall through comment is IMPORTANT! * do not change it without proper testing with both clang and gcc * in multiple versions. * the clang check allows to build with all versions of clang. * the has_c_attribute check is to workaround a bug in clang version * in rhel7. has_attribute would happily return "YES SIR WE GOT IT" * and fail the build the next line. */ #ifdef __clang__ #ifdef __has_c_attribute #if __has_attribute(fallthrough) __attribute__((fallthrough)); #endif #endif #endif // Fall through default: pcmk__g_strcat(str, ": ", pcmk_exec_status_str(op->op_status), NULL); } if ((op->exit_reason != NULL) && ((op->op_status != PCMK_EXEC_DONE) || (op->rc != PCMK_OCF_OK))) { pcmk__g_strcat(str, " (", op->exit_reason, ")", NULL); } g_string_append(str, " " CRM_XS); g_string_append_printf(str, " graph action %sconfirmed; call=%d key=%s", (confirmed? "" : "un"), op->call_id, op_key); if (op->op_status == PCMK_EXEC_DONE) { g_string_append_printf(str, " rc=%d", op->rc); } do_crm_log(log_level, "%s", str->str); g_string_free(str, TRUE); /* The services library has already logged the output at info or debug * level, so just raise to notice if it looks like a failure. */ if ((op->output != NULL) && (op->rc != PCMK_OCF_OK)) { char *prefix = crm_strdup_printf(PCMK__OP_FMT "@%s output", op->rsc_id, op->op_type, op->interval_ms, node_name); crm_log_output(LOG_NOTICE, prefix, op->output); free(prefix); } } void process_lrm_event(lrm_state_t *lrm_state, lrmd_event_data_t *op, active_op_t *pending, const xmlNode *action_xml) { char *op_id = NULL; char *op_key = NULL; gboolean remove = FALSE; gboolean removed = FALSE; bool need_direct_ack = FALSE; lrmd_rsc_info_t *rsc = NULL; const char *node_name = NULL; CRM_CHECK(op != NULL, return); CRM_CHECK(op->rsc_id != NULL, return); // Remap new status codes for older DCs if (compare_version(controld_globals.dc_version, "3.2.0") < 0) { switch (op->op_status) { case PCMK_EXEC_NOT_CONNECTED: lrmd__set_result(op, PCMK_OCF_CONNECTION_DIED, PCMK_EXEC_ERROR, op->exit_reason); break; case PCMK_EXEC_INVALID: lrmd__set_result(op, CRM_DIRECT_NACK_RC, PCMK_EXEC_ERROR, op->exit_reason); break; default: break; } } op_id = make_stop_id(op->rsc_id, op->call_id); op_key = pcmk__op_key(op->rsc_id, op->op_type, op->interval_ms); // Get resource info if available (from executor state or action XML) if (lrm_state) { rsc = lrm_state_get_rsc_info(lrm_state, op->rsc_id, 0); } if ((rsc == NULL) && action_xml) { xmlNode *xml = pcmk__xe_first_child(action_xml, PCMK_XE_PRIMITIVE, NULL, NULL); const char *standard = crm_element_value(xml, PCMK_XA_CLASS); const char *provider = crm_element_value(xml, PCMK_XA_PROVIDER); const char *type = crm_element_value(xml, PCMK_XA_TYPE); if (standard && type) { crm_info("%s agent information not cached, using %s%s%s:%s from action XML", op->rsc_id, standard, (provider? ":" : ""), (provider? provider : ""), type); rsc = lrmd_new_rsc_info(op->rsc_id, standard, provider, type); } else { crm_err("Can't process %s result because %s agent information not cached or in XML", op_key, op->rsc_id); } } // Get node name if available (from executor state or action XML) if (lrm_state) { node_name = lrm_state->node_name; } else if (action_xml) { node_name = crm_element_value(action_xml, PCMK__META_ON_NODE); } if(pending == NULL) { remove = TRUE; if (lrm_state) { pending = g_hash_table_lookup(lrm_state->active_ops, op_id); } } if (op->op_status == PCMK_EXEC_ERROR) { switch(op->rc) { case PCMK_OCF_NOT_RUNNING: case PCMK_OCF_RUNNING_PROMOTED: case PCMK_OCF_DEGRADED: case PCMK_OCF_DEGRADED_PROMOTED: // Leave it to the TE/scheduler to decide if this is an error op->op_status = PCMK_EXEC_DONE; break; default: /* Nothing to do */ break; } } if (op->op_status != PCMK_EXEC_CANCELLED) { /* We might not record the result, so directly acknowledge it to the * originator instead, so it doesn't time out waiting for the result * (especially important if part of a transition). */ need_direct_ack = TRUE; if (controld_action_is_recordable(op->op_type)) { if (node_name && rsc) { // We should record the result, and happily, we can time_t lock_time = (pending == NULL)? 0 : pending->lock_time; controld_update_resource_history(node_name, rsc, op, lock_time); need_direct_ack = FALSE; } else if (op->rsc_deleted) { /* We shouldn't record the result (likely the resource was * refreshed, cleaned, or removed while this operation was * in flight). */ crm_notice("Not recording %s result in CIB because " "resource information was removed since it was initiated", op_key); } else { /* This shouldn't be possible; the executor didn't consider the * resource deleted, but we couldn't find resource or node * information. */ crm_err("Unable to record %s result in CIB: %s", op_key, (node_name? "No resource information" : "No node name")); } } } else if (op->interval_ms == 0) { /* A non-recurring operation was cancelled. Most likely, the * never-initiated action was removed from the executor's pending * operations list upon resource removal. */ need_direct_ack = TRUE; } else if (pending == NULL) { /* This recurring operation was cancelled, but was not pending. No * transition actions are waiting on it, nothing needs to be done. */ } else if (op->user_data == NULL) { /* This recurring operation was cancelled and pending, but we don't * have a transition key. This should never happen. */ crm_err("Recurring operation %s was cancelled without transition information", op_key); } else if (pcmk_is_set(pending->flags, active_op_remove)) { /* This recurring operation was cancelled (by us) and pending, and we * have been waiting for it to finish. */ if (lrm_state) { controld_delete_action_history(op); } /* Directly acknowledge failed recurring actions here. The above call to * controld_delete_action_history() will not erase any corresponding * last_failure entry, which means that the DC won't confirm the * cancellation via process_op_deletion(), and the transition would * otherwise wait for the action timer to pop. */ if (did_lrm_rsc_op_fail(lrm_state, pending->rsc_id, pending->op_type, pending->interval_ms)) { need_direct_ack = TRUE; } } else if (op->rsc_deleted) { /* This recurring operation was cancelled (but not by us, and the * executor does not have resource information, likely due to resource * cleanup, refresh, or removal) and pending. */ crm_debug("Recurring op %s was cancelled due to resource deletion", op_key); need_direct_ack = TRUE; } else { /* This recurring operation was cancelled (but not by us, likely by the * executor before stopping the resource) and pending. We don't need to * do anything special. */ } if (need_direct_ack) { controld_ack_event_directly(NULL, NULL, NULL, op, op->rsc_id); } if(remove == FALSE) { /* The caller will do this afterwards, but keep the logging consistent */ removed = TRUE; } else if (lrm_state && ((op->interval_ms == 0) || (op->op_status == PCMK_EXEC_CANCELLED))) { gboolean found = g_hash_table_remove(lrm_state->active_ops, op_id); if (op->interval_ms != 0) { removed = TRUE; } else if (found) { removed = TRUE; crm_trace("Op %s (call=%d, stop-id=%s, remaining=%u): Confirmed", op_key, op->call_id, op_id, g_hash_table_size(lrm_state->active_ops)); } } log_executor_event(op, op_key, node_name, removed); if (lrm_state) { if (!pcmk__str_eq(op->op_type, PCMK_ACTION_META_DATA, pcmk__str_casei)) { crmd_alert_resource_op(lrm_state->node_name, op); } else if (rsc && (op->rc == PCMK_OCF_OK)) { char *metadata = unescape_newlines(op->output); controld_cache_metadata(lrm_state->metadata_cache, rsc, metadata); free(metadata); } } if (op->rsc_deleted) { crm_info("Deletion of resource '%s' complete after %s", op->rsc_id, op_key); if (lrm_state) { delete_rsc_entry(lrm_state, NULL, op->rsc_id, NULL, pcmk_ok, NULL, true); } } /* If a shutdown was escalated while operations were pending, * then the FSA will be stalled right now... allow it to continue */ controld_trigger_fsa(); if (lrm_state && rsc) { update_history_cache(lrm_state, rsc, op); } lrmd_free_rsc_info(rsc); free(op_key); free(op_id); } diff --git a/daemons/controld/controld_execd_state.c b/daemons/controld/controld_execd_state.c index 19196848bd..5930fb123b 100644 --- a/daemons/controld/controld_execd_state.c +++ b/daemons/controld/controld_execd_state.c @@ -1,818 +1,820 @@ /* * Copyright 2012-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include #include static 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, "Action was pending when executor connection was dropped"); + // coverity[store_truncates_time_t] event.t_run = (unsigned int) op->start_time; + // coverity[store_truncates_time_t] 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); lrmd__reset_result(&event); return TRUE; } gboolean lrm_state_is_local(lrm_state_t *lrm_state) { return (lrm_state != NULL) && pcmk__str_eq(lrm_state->node_name, controld_globals.our_nodename, pcmk__str_casei); } /*! * \internal * \brief Create executor state entry for a node and add it to the state table * * \param[in] node_name Node to create entry for * * \return Newly allocated executor state object initialized for \p node_name */ static 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 = pcmk__assert_alloc(1, sizeof(lrm_state_t)); state->node_name = pcmk__str_copy(node_name); state->rsc_info_cache = pcmk__strkey_table(NULL, free_rsc_info); state->deletion_ops = pcmk__strkey_table(free, free_deletion_op); state->active_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; } 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 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; } static void internal_lrm_state_destroy(gpointer data) { lrm_state_t *lrm_state = data; if (!lrm_state) { return; } /* Rather than directly remove the recorded proxy entries from proxy_table, * make sure any connected proxies get disconnected. So that * remote_proxy_disconnected() will be called and as well remove the * entries from proxy_table. */ remote_proxy_disconnect_by_node(lrm_state->node_name); crm_trace("Destroying proxy table %s with %u members", lrm_state->node_name, g_hash_table_size(proxy_table)); // Just in case there's still any leftovers in 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 %u 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 %u 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 %u members", g_hash_table_size(lrm_state->deletion_ops)); g_hash_table_destroy(lrm_state->deletion_ops); } if (lrm_state->active_ops != NULL) { crm_trace("Destroying pending op cache with %u members", g_hash_table_size(lrm_state->active_ops)); g_hash_table_destroy(lrm_state->active_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("Resetting resource history cache with %u 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("Resetting deletion operations cache with %u members", g_hash_table_size(lrm_state->deletion_ops)); g_hash_table_remove_all(lrm_state->deletion_ops); } if (lrm_state->active_ops != NULL) { crm_trace("Resetting active operations cache with %u members", g_hash_table_size(lrm_state->active_ops)); g_hash_table_remove_all(lrm_state->active_ops); } if (lrm_state->rsc_info_cache) { crm_trace("Resetting resource information cache with %u 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 %u 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 %u 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 == NULL) || (lrm_state_table == NULL)) { 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; CRM_CHECK(lrm_state_table != NULL, return NULL); 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) { if (lrm_state_table == NULL) { return NULL; } return g_hash_table_get_values(lrm_state_table); } 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(controld_globals.fsa_input_register, R_SHUTDOWN)) { removed = g_hash_table_foreach_remove(lrm_state->active_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, PCMK__XA_CRM_SYS_FROM, session); if (controld_authorize_ipc_message(msg, NULL, session)) { route_message(C_IPC_MESSAGE, msg); } controld_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, PCMK_XE_CLUSTER_PROPERTY_SET, NULL, config_hash, PCMK_VALUE_CIB_BOOTSTRAP_OPTIONS, 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, PCMK__XA_LRMD_IPC_SESSION); remote_proxy_t *proxy = g_hash_table_lookup(proxy_table, session); const char *op = crm_element_value(msg, PCMK__XA_LRMD_IPC_OP); if (pcmk__str_eq(op, LRMD_IPC_OP_NEW, pcmk__str_casei)) { const char *channel = crm_element_value(msg, PCMK__XA_LRMD_IPC_SERVER); proxy = crmd_remote_proxy_new(lrmd, lrm_state->node_name, session, channel); if (!remote_ra_controlling_guest(lrm_state)) { if (proxy != NULL) { cib_t *cib_conn = controld_globals.cib_conn; /* Look up PCMK_OPT_STONITH_WATCHDOG_TIMEOUT and send to the * remote peer for validation */ int rc = cib_conn->cmds->query(cib_conn, PCMK_XE_CRM_CONFIG, NULL, cib_scope_local); cib_conn->cmds->register_callback_full(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, PCMK__NODE_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 *wrapper = pcmk__xe_first_child(msg, PCMK__XE_LRMD_IPC_MSG, NULL, NULL); xmlNode *request = pcmk__xe_first_child(wrapper, NULL, NULL, NULL); CRM_CHECK(request != NULL, return); CRM_CHECK(lrm_state->node_name, return); crm_xml_add(request, PCMK_XE_ACL_ROLE, "pacemaker-remote"); pcmk__update_acl_user(request, PCMK__XA_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, PCMK__XA_CRM_TASK), CRM_OP_NODE_INFO, pcmk__str_none)) { int node_id = 0; crm_element_value_int(request, PCMK_XA_ID, &node_id); if ((node_id <= 0) && (crm_element_value(request, PCMK_XA_UNAME) == NULL)) { crm_xml_add(request, PCMK_XA_UNAME, lrm_state->node_name); } } crmd_proxy_dispatch(session, request); crm_element_value_int(msg, PCMK__XA_LRMD_IPC_MSG_FLAGS, &flags); if (flags & crm_ipc_client_response) { int msg_id = 0; xmlNode *op_reply = pcmk__xe_create(NULL, PCMK__XE_ACK); crm_xml_add(op_reply, PCMK_XA_FUNCTION, __func__); crm_xml_add_int(op_reply, PCMK__XA_LINE, __LINE__); crm_element_value_int(msg, PCMK__XA_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 "_" PCMK__META_ON_NODE, 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,out] 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 ms) before initiating action * \param[in] parameters Hash table of resource parameters * \param[out] call_id Where to store call ID on success * * \return Standard Pacemaker return code */ 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, GHashTable *parameters, int *call_id) { int rc = pcmk_rc_ok; lrmd_key_value_t *params = NULL; if (lrm_state->conn == NULL) { return ENOTCONN; } // Convert parameters from hash table to list if (parameters != NULL) { const char *key = NULL; const char *value = NULL; GHashTableIter iter; g_hash_table_iter_init(&iter, parameters); while (g_hash_table_iter_next(&iter, (gpointer *) &key, (gpointer *) &value)) { params = lrmd_key_value_add(params, key, value); } } 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)) { g_hash_table_remove(lrm_state_table, 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); } diff --git a/daemons/controld/controld_remote_ra.c b/daemons/controld/controld_remote_ra.c index 4bbf80c41f..dbde467bd5 100644 --- a/daemons/controld/controld_remote_ra.c +++ b/daemons/controld/controld_remote_ra.c @@ -1,1471 +1,1475 @@ /* * Copyright 2013-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #define REMOTE_LRMD_RA "remote" /* The max start timeout before cmd retry */ #define MAX_START_TIMEOUT_MS 10000 #define cmd_set_flags(cmd, flags_to_set) do { \ (cmd)->status = pcmk__set_flags_as(__func__, __LINE__, LOG_TRACE, \ "Remote command", (cmd)->rsc_id, (cmd)->status, \ (flags_to_set), #flags_to_set); \ } while (0) #define cmd_clear_flags(cmd, flags_to_clear) do { \ (cmd)->status = pcmk__clear_flags_as(__func__, __LINE__, LOG_TRACE, \ "Remote command", (cmd)->rsc_id, (cmd)->status, \ (flags_to_clear), #flags_to_clear); \ } while (0) enum remote_cmd_status { cmd_reported_success = (1 << 0), cmd_cancel = (1 << 1), }; typedef struct remote_ra_cmd_s { /*! the local node the cmd is issued from */ char *owner; /*! the remote node the cmd is executed on */ char *rsc_id; /*! the action to execute */ char *action; /*! some string the client wants us to give it back */ char *userdata; /*! start delay in ms */ int start_delay; /*! timer id used for start delay. */ int delay_id; /*! timeout in ms for cmd */ int timeout; int remaining_timeout; /*! recurring interval in ms */ guint interval_ms; /*! interval timer id */ int interval_id; int monitor_timeout_id; int takeover_timeout_id; /*! action parameters */ lrmd_key_value_t *params; pcmk__action_result_t result; int call_id; time_t start_time; uint32_t status; } remote_ra_cmd_t; #define lrm_remote_set_flags(lrm_state, flags_to_set) do { \ lrm_state_t *lrm = (lrm_state); \ remote_ra_data_t *ra = lrm->remote_ra_data; \ ra->status = pcmk__set_flags_as(__func__, __LINE__, LOG_TRACE, "Remote", \ lrm->node_name, ra->status, \ (flags_to_set), #flags_to_set); \ } while (0) #define lrm_remote_clear_flags(lrm_state, flags_to_clear) do { \ lrm_state_t *lrm = (lrm_state); \ remote_ra_data_t *ra = lrm->remote_ra_data; \ ra->status = pcmk__clear_flags_as(__func__, __LINE__, LOG_TRACE, "Remote", \ lrm->node_name, ra->status, \ (flags_to_clear), #flags_to_clear); \ } while (0) enum remote_status { expect_takeover = (1 << 0), takeover_complete = (1 << 1), remote_active = (1 << 2), /* Maintenance mode is difficult to determine from the controller's context, * so we have it signalled back with the transition from the scheduler. */ remote_in_maint = (1 << 3), /* Similar for whether we are controlling a guest node or remote node. * Fortunately there is a meta-attribute in the transition already and * as the situation doesn't change over time we can use the * resource start for noting down the information for later use when * the attributes aren't at hand. */ controlling_guest = (1 << 4), }; typedef struct remote_ra_data_s { crm_trigger_t *work; remote_ra_cmd_t *cur_cmd; GList *cmds; GList *recurring_cmds; uint32_t status; } remote_ra_data_t; static int handle_remote_ra_start(lrm_state_t * lrm_state, remote_ra_cmd_t * cmd, int timeout_ms); static void handle_remote_ra_stop(lrm_state_t * lrm_state, remote_ra_cmd_t * cmd); static GList *fail_all_monitor_cmds(GList * list); static void free_cmd(gpointer user_data) { remote_ra_cmd_t *cmd = user_data; if (!cmd) { return; } if (cmd->delay_id) { g_source_remove(cmd->delay_id); } if (cmd->interval_id) { g_source_remove(cmd->interval_id); } if (cmd->monitor_timeout_id) { g_source_remove(cmd->monitor_timeout_id); } if (cmd->takeover_timeout_id) { g_source_remove(cmd->takeover_timeout_id); } free(cmd->owner); free(cmd->rsc_id); free(cmd->action); free(cmd->userdata); pcmk__reset_result(&(cmd->result)); lrmd_key_value_freeall(cmd->params); free(cmd); } static int generate_callid(void) { static int remote_ra_callid = 0; remote_ra_callid++; if (remote_ra_callid <= 0) { remote_ra_callid = 1; } return remote_ra_callid; } static gboolean recurring_helper(gpointer data) { remote_ra_cmd_t *cmd = data; lrm_state_t *connection_rsc = NULL; cmd->interval_id = 0; connection_rsc = lrm_state_find(cmd->rsc_id); if (connection_rsc && connection_rsc->remote_ra_data) { remote_ra_data_t *ra_data = connection_rsc->remote_ra_data; ra_data->recurring_cmds = g_list_remove(ra_data->recurring_cmds, cmd); ra_data->cmds = g_list_append(ra_data->cmds, cmd); mainloop_set_trigger(ra_data->work); } return FALSE; } static gboolean start_delay_helper(gpointer data) { remote_ra_cmd_t *cmd = data; lrm_state_t *connection_rsc = NULL; cmd->delay_id = 0; connection_rsc = lrm_state_find(cmd->rsc_id); if (connection_rsc && connection_rsc->remote_ra_data) { remote_ra_data_t *ra_data = connection_rsc->remote_ra_data; mainloop_set_trigger(ra_data->work); } return FALSE; } static bool should_purge_attributes(crm_node_t *node) { bool purge = true; crm_node_t *conn_node = NULL; lrm_state_t *connection_rsc = NULL; if (!node->conn_host) { return purge; } /* Get the node that was hosting the remote connection resource from the * peer cache. That's the one we really care about here. */ conn_node = pcmk__get_node(0, node->conn_host, NULL, pcmk__node_search_cluster_member); if (conn_node == NULL) { return purge; } /* Check the uptime of connection_rsc. If it hasn't been running long * enough, set purge=true. "Long enough" means it started running earlier * than the timestamp when we noticed it went away in the first place. */ connection_rsc = lrm_state_find(node->uname); if (connection_rsc != NULL) { lrmd_t *lrm = connection_rsc->conn; time_t uptime = lrmd__uptime(lrm); time_t now = time(NULL); /* Add 20s of fuzziness to give corosync a while to notice the remote * host is gone. On various error conditions (failure to get uptime, * peer_lost isn't set) we default to purging. */ if (uptime > 0 && conn_node->peer_lost > 0 && uptime + 20 >= now - conn_node->peer_lost) { purge = false; } } return purge; } static enum controld_section_e section_to_delete(bool purge) { if (pcmk_is_set(controld_globals.flags, controld_shutdown_lock_enabled)) { if (purge) { return controld_section_all_unlocked; } else { return controld_section_lrm_unlocked; } } else { if (purge) { return controld_section_all; } else { return controld_section_lrm; } } } static void purge_remote_node_attrs(int call_opt, crm_node_t *node) { bool purge = should_purge_attributes(node); enum controld_section_e section = section_to_delete(purge); /* Purge node from attrd's memory */ if (purge) { update_attrd_remote_node_removed(node->uname, NULL); } controld_delete_node_state(node->uname, section, call_opt); } /*! * \internal * \brief Handle cluster communication related to pacemaker_remote node joining * * \param[in] node_name Name of newly integrated pacemaker_remote node */ static void remote_node_up(const char *node_name) { int call_opt; xmlNode *update, *state; crm_node_t *node; lrm_state_t *connection_rsc = NULL; CRM_CHECK(node_name != NULL, return); crm_info("Announcing Pacemaker Remote node %s", node_name); call_opt = crmd_cib_smart_opt(); /* Delete node's probe_complete attribute. This serves two purposes: * * - @COMPAT DCs < 1.1.14 in a rolling upgrade might use it * - deleting it (or any attribute for that matter) here ensures the * attribute manager learns the node is remote */ update_attrd(node_name, CRM_OP_PROBED, NULL, NULL, TRUE); /* Ensure node is in the remote peer cache with member status */ node = pcmk__cluster_lookup_remote_node(node_name); CRM_CHECK(node != NULL, return); purge_remote_node_attrs(call_opt, node); pcmk__update_peer_state(__func__, node, CRM_NODE_MEMBER, 0); /* Apply any start state that we were given from the environment on the * remote node. */ connection_rsc = lrm_state_find(node->uname); if (connection_rsc != NULL) { lrmd_t *lrm = connection_rsc->conn; const char *start_state = lrmd__node_start_state(lrm); if (start_state) { set_join_state(start_state, node->uname, node->uuid, true); } } /* pacemaker_remote nodes don't participate in the membership layer, * so cluster nodes don't automatically get notified when they come and go. * We send a cluster message to the DC, and update the CIB node state entry, * so the DC will get it sooner (via message) or later (via CIB refresh), * and any other interested parties can query the CIB. */ broadcast_remote_state_message(node_name, true); update = pcmk__xe_create(NULL, PCMK_XE_STATUS); state = create_node_state_update(node, node_update_cluster, update, __func__); /* Clear the PCMK__XA_NODE_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, PCMK__XA_NODE_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 PCMK__XA_NODE_FENCED attribute was * previously set, because it won't have been cleared). This could prevent * actual fencing or allow recurring monitor failures to be cleared too * soon. Ideally, we wouldn't rely on the CIB for the fenced status. */ controld_update_cib(PCMK_XE_STATUS, update, call_opt, NULL); free_xml(update); } enum down_opts { DOWN_KEEP_LRM, DOWN_ERASE_LRM }; /*! * \internal * \brief Handle cluster communication related to pacemaker_remote node leaving * * \param[in] node_name Name of lost node * \param[in] opts Whether to keep or erase LRM history */ static void remote_node_down(const char *node_name, const enum down_opts opts) { xmlNode *update; int call_opt = crmd_cib_smart_opt(); crm_node_t *node; /* Purge node from attrd's memory */ update_attrd_remote_node_removed(node_name, NULL); /* Normally, only node attributes should be erased, and the resource history * should be kept until the node comes back up. However, after a successful * fence, we want to clear the history as well, so we don't think resources * are still running on the node. */ if (opts == DOWN_ERASE_LRM) { controld_delete_node_state(node_name, controld_section_all, call_opt); } else { controld_delete_node_state(node_name, controld_section_attrs, call_opt); } /* Ensure node is in the remote peer cache with lost state */ node = pcmk__cluster_lookup_remote_node(node_name); CRM_CHECK(node != NULL, return); pcmk__update_peer_state(__func__, node, CRM_NODE_LOST, 0); /* Notify DC */ broadcast_remote_state_message(node_name, false); /* Update CIB node state */ update = pcmk__xe_create(NULL, PCMK_XE_STATUS); create_node_state_update(node, node_update_cluster, update, __func__); controld_update_cib(PCMK_XE_STATUS, update, call_opt, NULL); free_xml(update); } /*! * \internal * \brief Handle effects of a remote RA command on node state * * \param[in] cmd Completed remote RA command */ static void check_remote_node_state(const remote_ra_cmd_t *cmd) { /* Only successful actions can change node state */ if (!pcmk__result_ok(&(cmd->result))) { return; } if (pcmk__str_eq(cmd->action, PCMK_ACTION_START, pcmk__str_casei)) { remote_node_up(cmd->rsc_id); } else if (pcmk__str_eq(cmd->action, PCMK_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 = pcmk__cluster_lookup_remote_node(cmd->rsc_id); CRM_CHECK(node != NULL, return); pcmk__update_peer_state(__func__, node, CRM_NODE_MEMBER, 0); } else if (pcmk__str_eq(cmd->action, PCMK_ACTION_STOP, pcmk__str_casei)) { lrm_state_t *lrm_state = lrm_state_find(cmd->rsc_id); remote_ra_data_t *ra_data = lrm_state? lrm_state->remote_ra_data : NULL; if (ra_data) { if (!pcmk_is_set(ra_data->status, takeover_complete)) { /* Stop means down if we didn't successfully migrate elsewhere */ remote_node_down(cmd->rsc_id, DOWN_KEEP_LRM); } else if (AM_I_DC == FALSE) { /* Only the connection host and DC track node state, * so if the connection migrated elsewhere and we aren't DC, * un-cache the node, so we don't have stale info */ pcmk__cluster_forget_remote_node(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; + // coverity[store_truncates_time_t] op.t_run = (unsigned int) cmd->start_time; + // coverity[store_truncates_time_t] op.t_rcchange = (unsigned int) cmd->start_time; lrmd__set_result(&op, cmd->result.exit_status, cmd->result.execution_status, cmd->result.exit_reason); if (pcmk_is_set(cmd->status, cmd_reported_success) && !pcmk__result_ok(&(cmd->result))) { + // coverity[store_truncates_time_t] 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) { pcmk__insert_dup(op.params, tmp->key, tmp->value); } } op.call_id = cmd->call_id; op.remote_nodename = cmd->owner; lrm_op_callback(&op); if (op.params) { g_hash_table_destroy(op.params); } lrmd__reset_result(&op); } static void update_remaining_timeout(remote_ra_cmd_t * cmd) { cmd->remaining_timeout = ((cmd->timeout / 1000) - (time(NULL) - cmd->start_time)) * 1000; } static gboolean retry_start_cmd_cb(gpointer data) { lrm_state_t *lrm_state = data; remote_ra_data_t *ra_data = lrm_state->remote_ra_data; remote_ra_cmd_t *cmd = NULL; int rc = ETIME; if (!ra_data || !ra_data->cur_cmd) { return FALSE; } cmd = ra_data->cur_cmd; if (!pcmk__strcase_any_of(cmd->action, PCMK_ACTION_START, PCMK_ACTION_MIGRATE_FROM, NULL)) { return FALSE; } update_remaining_timeout(cmd); if (cmd->remaining_timeout > 0) { rc = handle_remote_ra_start(lrm_state, cmd, cmd->remaining_timeout); } else { pcmk__set_result(&(cmd->result), PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_TIMEOUT, "Not enough time remains to retry remote connection"); } if (rc != pcmk_rc_ok) { report_remote_ra_result(cmd); if (ra_data->cmds) { mainloop_set_trigger(ra_data->work); } ra_data->cur_cmd = NULL; free_cmd(cmd); } else { /* wait for connection event */ } return FALSE; } static gboolean connection_takeover_timeout_cb(gpointer data) { lrm_state_t *lrm_state = NULL; remote_ra_cmd_t *cmd = data; crm_info("takeover event timed out for node %s", cmd->rsc_id); cmd->takeover_timeout_id = 0; lrm_state = lrm_state_find(cmd->rsc_id); handle_remote_ra_stop(lrm_state, cmd); free_cmd(cmd); return FALSE; } static gboolean monitor_timeout_cb(gpointer data) { lrm_state_t *lrm_state = NULL; remote_ra_cmd_t *cmd = data; lrm_state = lrm_state_find(cmd->rsc_id); crm_info("Timed out waiting for remote poke response from %s%s", cmd->rsc_id, (lrm_state? "" : " (no LRM state)")); cmd->monitor_timeout_id = 0; pcmk__set_result(&(cmd->result), PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_TIMEOUT, "Remote executor did not respond"); if (lrm_state && lrm_state->remote_ra_data) { remote_ra_data_t *ra_data = lrm_state->remote_ra_data; if (ra_data->cur_cmd == cmd) { ra_data->cur_cmd = NULL; } if (ra_data->cmds) { mainloop_set_trigger(ra_data->work); } } report_remote_ra_result(cmd); free_cmd(cmd); if(lrm_state) { lrm_state_disconnect(lrm_state); } return FALSE; } static void synthesize_lrmd_success(lrm_state_t *lrm_state, const char *rsc_id, const char *op_type) { lrmd_event_data_t op = { 0, }; if (lrm_state == NULL) { /* if lrm_state not given assume local */ lrm_state = lrm_state_find(controld_globals.our_nodename); } CRM_ASSERT(lrm_state != NULL); op.type = lrmd_event_exec_complete; op.rsc_id = rsc_id; op.op_type = op_type; + // coverity[store_truncates_time_t] op.t_run = (unsigned int) time(NULL); op.t_rcchange = op.t_run; op.call_id = generate_callid(); lrmd__set_result(&op, PCMK_OCF_OK, PCMK_EXEC_DONE, NULL); process_lrm_event(lrm_state, &op, NULL, NULL); } void remote_lrm_op_callback(lrmd_event_data_t * op) { gboolean cmd_handled = FALSE; lrm_state_t *lrm_state = NULL; remote_ra_data_t *ra_data = NULL; remote_ra_cmd_t *cmd = NULL; crm_debug("Processing '%s%s%s' event on remote connection to %s: %s " "(%d) status=%s (%d)", (op->op_type? op->op_type : ""), (op->op_type? " " : ""), lrmd_event_type2str(op->type), op->remote_nodename, services_ocf_exitcode_str(op->rc), op->rc, pcmk_exec_status_str(op->op_status), op->op_status); lrm_state = lrm_state_find(op->remote_nodename); if (!lrm_state || !lrm_state->remote_ra_data) { crm_debug("No state information found for remote connection event"); return; } ra_data = lrm_state->remote_ra_data; if (op->type == lrmd_event_new_client) { // Another client has connected to the remote daemon if (pcmk_is_set(ra_data->status, expect_takeover)) { // Great, we knew this was coming lrm_remote_clear_flags(lrm_state, expect_takeover); lrm_remote_set_flags(lrm_state, takeover_complete); } else { crm_err("Disconnecting from Pacemaker Remote node %s due to " "unexpected client takeover", op->remote_nodename); /* In this case, lrmd_tls_connection_destroy() will be called under the control of mainloop. */ /* Do not free lrm_state->conn yet. */ /* It'll be freed in the following stop action. */ lrm_state_disconnect_only(lrm_state); } return; } /* filter all EXEC events up */ if (op->type == lrmd_event_exec_complete) { if (pcmk_is_set(ra_data->status, takeover_complete)) { crm_debug("ignoring event, this connection is taken over by another node"); } else { lrm_op_callback(op); } return; } if ((op->type == lrmd_event_disconnect) && (ra_data->cur_cmd == NULL)) { if (!pcmk_is_set(ra_data->status, remote_active)) { crm_debug("Disconnection from Pacemaker Remote node %s complete", lrm_state->node_name); } else if (!remote_ra_is_in_maintenance(lrm_state)) { crm_err("Lost connection to Pacemaker Remote node %s", lrm_state->node_name); ra_data->recurring_cmds = fail_all_monitor_cmds(ra_data->recurring_cmds); ra_data->cmds = fail_all_monitor_cmds(ra_data->cmds); } else { crm_notice("Unmanaged Pacemaker Remote node %s disconnected", lrm_state->node_name); /* Do roughly what a 'stop' on the remote-resource would do */ handle_remote_ra_stop(lrm_state, NULL); remote_node_down(lrm_state->node_name, DOWN_KEEP_LRM); /* now fake the reply of a successful 'stop' */ synthesize_lrmd_success(NULL, lrm_state->node_name, PCMK_ACTION_STOP); } return; } if (!ra_data->cur_cmd) { crm_debug("no event to match"); return; } cmd = ra_data->cur_cmd; /* Start actions and migrate from actions complete after connection * comes back to us. */ if ((op->type == lrmd_event_connect) && pcmk__strcase_any_of(cmd->action, PCMK_ACTION_START, PCMK_ACTION_MIGRATE_FROM, NULL)) { if (op->connection_rc < 0) { update_remaining_timeout(cmd); if ((op->connection_rc == -ENOKEY) || (op->connection_rc == -EKEYREJECTED)) { // Hard error, don't retry pcmk__set_result(&(cmd->result), PCMK_OCF_INVALID_PARAM, PCMK_EXEC_ERROR, pcmk_strerror(op->connection_rc)); } else if (cmd->remaining_timeout > 3000) { crm_trace("rescheduling start, remaining timeout %d", cmd->remaining_timeout); g_timeout_add(1000, retry_start_cmd_cb, lrm_state); return; } else { crm_trace("can't reschedule start, remaining timeout too small %d", cmd->remaining_timeout); pcmk__format_result(&(cmd->result), PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_TIMEOUT, "%s without enough time to retry", pcmk_strerror(op->connection_rc)); } } else { lrm_state_reset_tables(lrm_state, TRUE); pcmk__set_result(&(cmd->result), PCMK_OCF_OK, PCMK_EXEC_DONE, NULL); lrm_remote_set_flags(lrm_state, remote_active); } crm_debug("Remote connection event matched %s action", cmd->action); report_remote_ra_result(cmd); cmd_handled = TRUE; } else if ((op->type == lrmd_event_poke) && pcmk__str_eq(cmd->action, PCMK_ACTION_MONITOR, pcmk__str_casei)) { if (cmd->monitor_timeout_id) { g_source_remove(cmd->monitor_timeout_id); cmd->monitor_timeout_id = 0; } /* Only report success the first time, after that only worry about failures. * For this function, if we get the poke pack, it is always a success. Pokes * only fail if the send fails, or the response times out. */ if (!pcmk_is_set(cmd->status, cmd_reported_success)) { pcmk__set_result(&(cmd->result), PCMK_OCF_OK, PCMK_EXEC_DONE, NULL); report_remote_ra_result(cmd); cmd_set_flags(cmd, cmd_reported_success); } crm_debug("Remote poke event matched %s action", cmd->action); /* success, keep rescheduling if interval is present. */ if (cmd->interval_ms && !pcmk_is_set(cmd->status, cmd_cancel)) { ra_data->recurring_cmds = g_list_append(ra_data->recurring_cmds, cmd); cmd->interval_id = g_timeout_add(cmd->interval_ms, recurring_helper, cmd); cmd = NULL; /* prevent free */ } cmd_handled = TRUE; } else if ((op->type == lrmd_event_disconnect) && pcmk__str_eq(cmd->action, PCMK_ACTION_MONITOR, pcmk__str_casei)) { if (pcmk_is_set(ra_data->status, remote_active) && !pcmk_is_set(cmd->status, cmd_cancel)) { pcmk__set_result(&(cmd->result), PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_ERROR, "Remote connection unexpectedly dropped " "during monitor"); report_remote_ra_result(cmd); crm_err("Remote connection to %s unexpectedly dropped during monitor", lrm_state->node_name); } cmd_handled = TRUE; } else if ((op->type == lrmd_event_new_client) && pcmk__str_eq(cmd->action, PCMK_ACTION_STOP, pcmk__str_casei)) { handle_remote_ra_stop(lrm_state, cmd); cmd_handled = TRUE; } else { crm_debug("Event did not match %s action", ra_data->cur_cmd->action); } if (cmd_handled) { ra_data->cur_cmd = NULL; if (ra_data->cmds) { mainloop_set_trigger(ra_data->work); } free_cmd(cmd); } } static void handle_remote_ra_stop(lrm_state_t * lrm_state, remote_ra_cmd_t * cmd) { remote_ra_data_t *ra_data = NULL; CRM_ASSERT(lrm_state); ra_data = lrm_state->remote_ra_data; if (!pcmk_is_set(ra_data->status, takeover_complete)) { /* delete pending ops when ever the remote connection is intentionally stopped */ g_hash_table_remove_all(lrm_state->active_ops); } else { /* we no longer hold the history if this connection has been migrated, * however, we keep metadata cache for future use */ lrm_state_reset_tables(lrm_state, FALSE); } lrm_remote_clear_flags(lrm_state, remote_active); lrm_state_disconnect(lrm_state); if (ra_data->cmds) { g_list_free_full(ra_data->cmds, free_cmd); } if (ra_data->recurring_cmds) { g_list_free_full(ra_data->recurring_cmds, free_cmd); } ra_data->cmds = NULL; ra_data->recurring_cmds = NULL; ra_data->cur_cmd = NULL; if (cmd) { pcmk__set_result(&(cmd->result), PCMK_OCF_OK, PCMK_EXEC_DONE, NULL); report_remote_ra_result(cmd); } } // \return Standard Pacemaker return code static int handle_remote_ra_start(lrm_state_t * lrm_state, remote_ra_cmd_t * cmd, int timeout_ms) { const char *server = NULL; lrmd_key_value_t *tmp = NULL; int port = 0; int timeout_used = timeout_ms > MAX_START_TIMEOUT_MS ? MAX_START_TIMEOUT_MS : timeout_ms; int rc = pcmk_rc_ok; for (tmp = cmd->params; tmp; tmp = tmp->next) { if (pcmk__strcase_any_of(tmp->key, PCMK_REMOTE_RA_ADDR, PCMK_REMOTE_RA_SERVER, NULL)) { server = tmp->value; } else if (pcmk__str_eq(tmp->key, PCMK_REMOTE_RA_PORT, pcmk__str_none)) { port = atoi(tmp->value); } else if (pcmk__str_eq(tmp->key, CRM_META "_" PCMK__META_CONTAINER, pcmk__str_none)) { lrm_remote_set_flags(lrm_state, controlling_guest); } } rc = controld_connect_remote_executor(lrm_state, server, port, timeout_used); if (rc != pcmk_rc_ok) { pcmk__format_result(&(cmd->result), PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_ERROR, "Could not connect to Pacemaker Remote node %s: %s", lrm_state->node_name, pcmk_rc_str(rc)); } return rc; } static gboolean handle_remote_ra_exec(gpointer user_data) { int rc = 0; lrm_state_t *lrm_state = user_data; remote_ra_data_t *ra_data = lrm_state->remote_ra_data; remote_ra_cmd_t *cmd; GList *first = NULL; if (ra_data->cur_cmd) { /* still waiting on previous cmd */ return TRUE; } while (ra_data->cmds) { first = ra_data->cmds; cmd = first->data; if (cmd->delay_id) { /* still waiting for start delay timer to trip */ return TRUE; } ra_data->cmds = g_list_remove_link(ra_data->cmds, first); g_list_free_1(first); if (pcmk__str_any_of(cmd->action, PCMK_ACTION_START, PCMK_ACTION_MIGRATE_FROM, NULL)) { lrm_remote_clear_flags(lrm_state, expect_takeover | takeover_complete); if (handle_remote_ra_start(lrm_state, cmd, cmd->timeout) == pcmk_rc_ok) { /* take care of this later when we get async connection result */ crm_debug("Initiated async remote connection, %s action will complete after connect event", cmd->action); ra_data->cur_cmd = cmd; return TRUE; } report_remote_ra_result(cmd); } else if (!strcmp(cmd->action, PCMK_ACTION_MONITOR)) { if (lrm_state_is_connected(lrm_state) == TRUE) { rc = lrm_state_poke_connection(lrm_state); if (rc < 0) { pcmk__set_result(&(cmd->result), PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_ERROR, pcmk_strerror(rc)); } } else { rc = -1; pcmk__set_result(&(cmd->result), PCMK_OCF_NOT_RUNNING, PCMK_EXEC_DONE, "Remote connection inactive"); } if (rc == 0) { crm_debug("Poked Pacemaker Remote at node %s, waiting for async response", cmd->rsc_id); ra_data->cur_cmd = cmd; cmd->monitor_timeout_id = g_timeout_add(cmd->timeout, monitor_timeout_cb, cmd); return TRUE; } report_remote_ra_result(cmd); } else if (!strcmp(cmd->action, PCMK_ACTION_STOP)) { if (pcmk_is_set(ra_data->status, expect_takeover)) { /* briefly wait on stop for the takeover event to occur. If the * takeover event does not occur during the wait period, that's fine. * It just means that the remote-node's lrm_status section is going to get * cleared which will require all the resources running in the remote-node * to be explicitly re-detected via probe actions. If the takeover does occur * successfully, then we can leave the status section intact. */ cmd->takeover_timeout_id = g_timeout_add((cmd->timeout/2), connection_takeover_timeout_cb, cmd); ra_data->cur_cmd = cmd; return TRUE; } handle_remote_ra_stop(lrm_state, cmd); } else if (strcmp(cmd->action, PCMK_ACTION_MIGRATE_TO) == 0) { lrm_remote_clear_flags(lrm_state, takeover_complete); lrm_remote_set_flags(lrm_state, expect_takeover); pcmk__set_result(&(cmd->result), PCMK_OCF_OK, PCMK_EXEC_DONE, NULL); report_remote_ra_result(cmd); } else if (pcmk__str_any_of(cmd->action, PCMK_ACTION_RELOAD, PCMK_ACTION_RELOAD_AGENT, NULL)) { /* Currently the only reloadable parameter is * PCMK_REMOTE_RA_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 = pcmk__assert_alloc(1, sizeof(remote_ra_data_t)); ra_data->work = mainloop_add_trigger(G_PRIORITY_HIGH, handle_remote_ra_exec, lrm_state); lrm_state->remote_ra_data = ra_data; } void remote_ra_cleanup(lrm_state_t * lrm_state) { remote_ra_data_t *ra_data = lrm_state->remote_ra_data; if (!ra_data) { return; } if (ra_data->cmds) { g_list_free_full(ra_data->cmds, free_cmd); } if (ra_data->recurring_cmds) { g_list_free_full(ra_data->recurring_cmds, free_cmd); } mainloop_destroy_trigger(ra_data->work); free(ra_data); lrm_state->remote_ra_data = NULL; } gboolean is_remote_lrmd_ra(const char *agent, const char *provider, const char *id) { if (agent && provider && !strcmp(agent, REMOTE_LRMD_RA) && !strcmp(provider, "pacemaker")) { return TRUE; } if ((id != NULL) && (lrm_state_find(id) != NULL) && !pcmk__str_eq(id, controld_globals.our_nodename, pcmk__str_casei)) { return TRUE; } return FALSE; } lrmd_rsc_info_t * remote_ra_get_rsc_info(lrm_state_t * lrm_state, const char *rsc_id) { lrmd_rsc_info_t *info = NULL; if ((lrm_state_find(rsc_id))) { info = pcmk__assert_alloc(1, sizeof(lrmd_rsc_info_t)); info->id = pcmk__str_copy(rsc_id); info->type = pcmk__str_copy(REMOTE_LRMD_RA); info->standard = pcmk__str_copy(PCMK_RESOURCE_CLASS_OCF); info->provider = pcmk__str_copy("pacemaker"); } return info; } static gboolean is_remote_ra_supported_action(const char *action) { return pcmk__str_any_of(action, PCMK_ACTION_START, PCMK_ACTION_STOP, PCMK_ACTION_MONITOR, PCMK_ACTION_MIGRATE_TO, PCMK_ACTION_MIGRATE_FROM, PCMK_ACTION_RELOAD_AGENT, PCMK_ACTION_RELOAD, NULL); } static GList * fail_all_monitor_cmds(GList * list) { GList *rm_list = NULL; remote_ra_cmd_t *cmd = NULL; GList *gIter = NULL; for (gIter = list; gIter != NULL; gIter = gIter->next) { cmd = gIter->data; if ((cmd->interval_ms > 0) && pcmk__str_eq(cmd->action, PCMK_ACTION_MONITOR, pcmk__str_casei)) { rm_list = g_list_append(rm_list, cmd); } } for (gIter = rm_list; gIter != NULL; gIter = gIter->next) { cmd = gIter->data; pcmk__set_result(&(cmd->result), PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_ERROR, "Lost connection to remote executor"); crm_trace("Pre-emptively failing %s %s (interval=%u, %s)", cmd->action, cmd->rsc_id, cmd->interval_ms, cmd->userdata); report_remote_ra_result(cmd); list = g_list_remove(list, cmd); free_cmd(cmd); } /* frees only the list data, not the cmds */ g_list_free(rm_list); return list; } static GList * remove_cmd(GList * list, const char *action, guint interval_ms) { remote_ra_cmd_t *cmd = NULL; GList *gIter = NULL; for (gIter = list; gIter != NULL; gIter = gIter->next) { cmd = gIter->data; if ((cmd->interval_ms == interval_ms) && pcmk__str_eq(cmd->action, action, pcmk__str_casei)) { break; } cmd = NULL; } if (cmd) { list = g_list_remove(list, cmd); free_cmd(cmd); } return list; } int remote_ra_cancel(lrm_state_t *lrm_state, const char *rsc_id, const char *action, guint interval_ms) { lrm_state_t *connection_rsc = NULL; remote_ra_data_t *ra_data = NULL; connection_rsc = lrm_state_find(rsc_id); if (!connection_rsc || !connection_rsc->remote_ra_data) { return -EINVAL; } ra_data = connection_rsc->remote_ra_data; ra_data->cmds = remove_cmd(ra_data->cmds, action, interval_ms); ra_data->recurring_cmds = remove_cmd(ra_data->recurring_cmds, action, interval_ms); if (ra_data->cur_cmd && (ra_data->cur_cmd->interval_ms == interval_ms) && (pcmk__str_eq(ra_data->cur_cmd->action, action, pcmk__str_casei))) { cmd_set_flags(ra_data->cur_cmd, cmd_cancel); } return 0; } static remote_ra_cmd_t * handle_dup_monitor(remote_ra_data_t *ra_data, guint interval_ms, const char *userdata) { GList *gIter = NULL; remote_ra_cmd_t *cmd = NULL; /* there are 3 places a potential duplicate monitor operation * could exist. * 1. recurring_cmds list. where the op is waiting for its next interval * 2. cmds list, where the op is queued to get executed immediately * 3. cur_cmd, which means the monitor op is in flight right now. */ if (interval_ms == 0) { return NULL; } if (ra_data->cur_cmd && !pcmk_is_set(ra_data->cur_cmd->status, cmd_cancel) && (ra_data->cur_cmd->interval_ms == interval_ms) && pcmk__str_eq(ra_data->cur_cmd->action, PCMK_ACTION_MONITOR, pcmk__str_casei)) { cmd = ra_data->cur_cmd; goto handle_dup; } for (gIter = ra_data->recurring_cmds; gIter != NULL; gIter = gIter->next) { cmd = gIter->data; if ((cmd->interval_ms == interval_ms) && pcmk__str_eq(cmd->action, PCMK_ACTION_MONITOR, pcmk__str_casei)) { goto handle_dup; } } for (gIter = ra_data->cmds; gIter != NULL; gIter = gIter->next) { cmd = gIter->data; if ((cmd->interval_ms == interval_ms) && pcmk__str_eq(cmd->action, PCMK_ACTION_MONITOR, pcmk__str_casei)) { goto handle_dup; } } return NULL; handle_dup: crm_trace("merging duplicate monitor cmd " PCMK__OP_FMT, cmd->rsc_id, PCMK_ACTION_MONITOR, interval_ms); /* update the userdata */ if (userdata) { free(cmd->userdata); cmd->userdata = pcmk__str_copy(userdata); } /* if we've already reported success, generate a new call id */ if (pcmk_is_set(cmd->status, cmd_reported_success)) { cmd->start_time = time(NULL); cmd->call_id = generate_callid(); cmd_clear_flags(cmd, cmd_reported_success); } /* if we have an interval_id set, that means we are in the process of * waiting for this cmd's next interval. instead of waiting, cancel * the timer and execute the action immediately */ if (cmd->interval_id) { g_source_remove(cmd->interval_id); cmd->interval_id = 0; recurring_helper(cmd); } return cmd; } /*! * \internal * \brief Execute an action using the (internal) ocf:pacemaker:remote agent * * \param[in] lrm_state Executor state object for remote connection * \param[in] rsc_id Connection resource ID * \param[in] action Action to execute * \param[in] userdata String to copy and pass to execution callback * \param[in] interval_ms Action interval (in milliseconds) * \param[in] timeout_ms Action timeout (in milliseconds) * \param[in] start_delay_ms Delay (in milliseconds) before executing action * \param[in,out] params Connection resource parameters * \param[out] call_id Where to store call ID on success * * \return Standard Pacemaker return code * \note This takes ownership of \p params, which should not be used or freed * after calling this function. */ int controld_execute_remote_agent(const lrm_state_t *lrm_state, const char *rsc_id, const char *action, const char *userdata, guint interval_ms, int timeout_ms, int start_delay_ms, lrmd_key_value_t *params, int *call_id) { lrm_state_t *connection_rsc = NULL; remote_ra_cmd_t *cmd = NULL; remote_ra_data_t *ra_data = NULL; *call_id = 0; CRM_CHECK((lrm_state != NULL) && (rsc_id != NULL) && (action != NULL) && (userdata != NULL) && (call_id != NULL), lrmd_key_value_freeall(params); return EINVAL); if (!is_remote_ra_supported_action(action)) { lrmd_key_value_freeall(params); return EOPNOTSUPP; } connection_rsc = lrm_state_find(rsc_id); if (connection_rsc == NULL) { lrmd_key_value_freeall(params); return ENOTCONN; } remote_ra_data_init(connection_rsc); ra_data = connection_rsc->remote_ra_data; cmd = handle_dup_monitor(ra_data, interval_ms, userdata); if (cmd) { *call_id = cmd->call_id; lrmd_key_value_freeall(params); return pcmk_rc_ok; } cmd = pcmk__assert_alloc(1, sizeof(remote_ra_cmd_t)); cmd->owner = pcmk__str_copy(lrm_state->node_name); cmd->rsc_id = pcmk__str_copy(rsc_id); cmd->action = pcmk__str_copy(action); cmd->userdata = pcmk__str_copy(userdata); 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 "/" PCMK__XE_PSEUDO_EVENT \ "[@" PCMK_XA_OPERATION "='stonith']/" PCMK__XE_DOWNED "/" PCMK_XE_NODE /*! * \internal * \brief Check a pseudo-action for Pacemaker Remote node side effects * * \param[in,out] xml XML of pseudo-action to check */ void remote_ra_process_pseudo(xmlNode *xml) { xmlXPathObjectPtr search = xpath_search(xml, XPATH_PSEUDO_FENCE); if (numXpathResults(search) == 1) { xmlNode *result = getXpathResult(search, 0); /* Normally, we handle the necessary side effects of a guest node stop * action when reporting the remote agent's result. However, if the stop * is implied due to fencing, it will be a fencing pseudo-event, and * there won't be a result to report. Handle that case here. * * This will result in a duplicate call to remote_node_down() if the * guest stop was real instead of implied, but that shouldn't hurt. * * There is still one corner case that isn't handled: if a guest node * isn't running any resources when its host is fenced, it will appear * to be cleanly stopped, so there will be no pseudo-fence, and our * peer cache state will be incorrect unless and until the guest is * recovered. */ if (result) { const char *remote = pcmk__xe_id(result); if (remote) { remote_node_down(remote, DOWN_ERASE_LRM); } } } freeXpathObject(search); } static void remote_ra_maintenance(lrm_state_t * lrm_state, gboolean maintenance) { xmlNode *update, *state; int call_opt; crm_node_t *node; call_opt = crmd_cib_smart_opt(); node = pcmk__cluster_lookup_remote_node(lrm_state->node_name); CRM_CHECK(node != NULL, return); update = pcmk__xe_create(NULL, PCMK_XE_STATUS); state = create_node_state_update(node, node_update_none, update, __func__); crm_xml_add(state, PCMK__XA_NODE_IN_MAINTENANCE, (maintenance? "1" : "0")); if (controld_update_cib(PCMK_XE_STATUS, update, call_opt, NULL) == pcmk_rc_ok) { /* TODO: still not 100% sure that async update will succeed ... */ if (maintenance) { lrm_remote_set_flags(lrm_state, remote_in_maint); } else { lrm_remote_clear_flags(lrm_state, remote_in_maint); } } free_xml(update); } #define XPATH_PSEUDO_MAINTENANCE "//" PCMK__XE_PSEUDO_EVENT \ "[@" PCMK_XA_OPERATION "='" PCMK_ACTION_MAINTENANCE_NODES "']/" \ PCMK__XE_MAINTENANCE /*! * \internal * \brief Check a pseudo-action holding updates for maintenance state * * \param[in,out] xml XML of pseudo-action to check */ void remote_ra_process_maintenance_nodes(xmlNode *xml) { xmlXPathObjectPtr search = xpath_search(xml, XPATH_PSEUDO_MAINTENANCE); if (numXpathResults(search) == 1) { xmlNode *node; int cnt = 0, cnt_remote = 0; for (node = pcmk__xe_first_child(getXpathResult(search, 0), PCMK_XE_NODE, NULL, NULL); node != NULL; node = pcmk__xe_next_same(node)) { lrm_state_t *lrm_state = lrm_state_find(pcmk__xe_id(node)); cnt++; if (lrm_state && lrm_state->remote_ra_data && pcmk_is_set(((remote_ra_data_t *) lrm_state->remote_ra_data)->status, remote_active)) { const char *in_maint_s = NULL; int in_maint; cnt_remote++; in_maint_s = crm_element_value(node, PCMK__XA_NODE_IN_MAINTENANCE); pcmk__scan_min_int(in_maint_s, &in_maint, 0); remote_ra_maintenance(lrm_state, in_maint); } } crm_trace("Action holds %d nodes (%d remotes found) adjusting " PCMK_OPT_MAINTENANCE_MODE, cnt, cnt_remote); } freeXpathObject(search); } gboolean remote_ra_is_in_maintenance(lrm_state_t * lrm_state) { remote_ra_data_t *ra_data = lrm_state->remote_ra_data; return pcmk_is_set(ra_data->status, remote_in_maint); } gboolean remote_ra_controlling_guest(lrm_state_t * lrm_state) { remote_ra_data_t *ra_data = lrm_state->remote_ra_data; return pcmk_is_set(ra_data->status, controlling_guest); } diff --git a/daemons/controld/controld_timers.c b/daemons/controld/controld_timers.c index 0d387b94e0..c4827631ca 100644 --- a/daemons/controld/controld_timers.c +++ b/daemons/controld/controld_timers.c @@ -1,482 +1,483 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include //! FSA mainloop timer type typedef struct fsa_timer_s { guint source_id; //!< Timer source ID guint period_ms; //!< Timer period enum crmd_fsa_input fsa_input; //!< Input to register if timer pops gboolean (*callback) (gpointer data); //!< What do if timer pops bool log_error; //!< Timer popping indicates error int counter; //!< For detecting loops } fsa_timer_t; //! Wait before retrying a failed cib or executor connection static fsa_timer_t *wait_timer = NULL; //! Periodically re-run scheduler (for date_spec evaluation and as a failsafe) static fsa_timer_t *recheck_timer = NULL; //! Wait at start-up, or after an election, for DC to make contact static fsa_timer_t *election_timer = NULL; //! Delay start of new transition with expectation something else might happen static fsa_timer_t *transition_timer = NULL; //! \c PCMK_OPT_JOIN_INTEGRATION_TIMEOUT static fsa_timer_t *integration_timer = NULL; //! \c PCMK_OPT_JOIN_FINALIZATION_TIMEOUT static fsa_timer_t *finalization_timer = NULL; // Wait for DC to stop all resources and give us the all-clear to shut down fsa_timer_t *shutdown_escalation_timer = NULL; //! Cluster recheck interval (from configuration) static guint recheck_interval_ms = 0; static const char * get_timer_desc(fsa_timer_t * timer) { if (timer == election_timer) { return "Election Trigger"; } else if (timer == shutdown_escalation_timer) { return "Shutdown Escalation"; } else if (timer == integration_timer) { return "Integration Timer"; } else if (timer == finalization_timer) { return "Finalization Timer"; } else if (timer == transition_timer) { return "New Transition Timer"; } else if (timer == wait_timer) { return "Wait Timer"; } else if (timer == recheck_timer) { return "Cluster Recheck Timer"; } return "Unknown Timer"; } /*! * \internal * \brief Stop an FSA timer * * \param[in,out] timer Timer to stop * * \return true if the timer was running, or false otherwise */ static bool controld_stop_timer(fsa_timer_t *timer) { CRM_CHECK(timer != NULL, return false); if (timer->source_id != 0) { crm_trace("Stopping %s (would inject %s if popped after %ums, src=%d)", get_timer_desc(timer), fsa_input2string(timer->fsa_input), timer->period_ms, timer->source_id); g_source_remove(timer->source_id); timer->source_id = 0; } else { crm_trace("%s already stopped (would inject %s if popped after %ums)", get_timer_desc(timer), fsa_input2string(timer->fsa_input), timer->period_ms); return false; } return true; } /*! * \internal * \brief Start an FSA timer * * \param[in,out] timer Timer to start */ static void controld_start_timer(fsa_timer_t *timer) { if (timer->source_id == 0 && timer->period_ms > 0) { timer->source_id = g_timeout_add(timer->period_ms, timer->callback, (void *)timer); CRM_ASSERT(timer->source_id != 0); crm_debug("Started %s (inject %s if pops after %ums, source=%d)", get_timer_desc(timer), fsa_input2string(timer->fsa_input), timer->period_ms, timer->source_id); } else { crm_debug("%s already running (inject %s if pops after %ums, source=%d)", get_timer_desc(timer), fsa_input2string(timer->fsa_input), timer->period_ms, timer->source_id); } } /* A_DC_TIMER_STOP, A_DC_TIMER_START, * A_FINALIZE_TIMER_STOP, A_FINALIZE_TIMER_START * A_INTEGRATE_TIMER_STOP, A_INTEGRATE_TIMER_START */ void do_timer_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) { gboolean timer_op_ok = TRUE; if (action & A_DC_TIMER_STOP) { timer_op_ok = controld_stop_timer(election_timer); } else if (action & A_FINALIZE_TIMER_STOP) { timer_op_ok = controld_stop_timer(finalization_timer); } else if (action & A_INTEGRATE_TIMER_STOP) { timer_op_ok = controld_stop_timer(integration_timer); } /* don't start a timer that wasn't already running */ if (action & A_DC_TIMER_START && timer_op_ok) { controld_start_timer(election_timer); if (AM_I_DC) { /* there can be only one */ register_fsa_input(cause, I_ELECTION, NULL); } } else if (action & A_FINALIZE_TIMER_START) { controld_start_timer(finalization_timer); } else if (action & A_INTEGRATE_TIMER_START) { controld_start_timer(integration_timer); } } static gboolean crm_timer_popped(gpointer data) { fsa_timer_t *timer = (fsa_timer_t *) data; if (timer->log_error) { crm_err("%s just popped in state %s! " CRM_XS " input=%s time=%ums", get_timer_desc(timer), fsa_state2string(controld_globals.fsa_state), fsa_input2string(timer->fsa_input), timer->period_ms); } else { crm_info("%s just popped " CRM_XS " input=%s time=%ums", get_timer_desc(timer), fsa_input2string(timer->fsa_input), timer->period_ms); timer->counter++; } if ((timer == election_timer) && (election_timer->counter > 5)) { crm_notice("We appear to be in an election loop, something may be wrong"); crm_write_blackbox(0, NULL); election_timer->counter = 0; } controld_stop_timer(timer); // Make timer _not_ go off again if (timer->fsa_input == I_INTEGRATED) { crm_info("Welcomed: %d, Integrated: %d", crmd_join_phase_count(crm_join_welcomed), crmd_join_phase_count(crm_join_integrated)); if (crmd_join_phase_count(crm_join_welcomed) == 0) { // If we don't even have ourselves, start again register_fsa_error_adv(C_FSA_INTERNAL, I_ELECTION, NULL, NULL, __func__); } else { register_fsa_input_before(C_TIMER_POPPED, timer->fsa_input, NULL); } } else if ((timer == recheck_timer) && (controld_globals.fsa_state != S_IDLE)) { crm_debug("Discarding %s event in state: %s", fsa_input2string(timer->fsa_input), fsa_state2string(controld_globals.fsa_state)); } else if ((timer == finalization_timer) && (controld_globals.fsa_state != S_FINALIZE_JOIN)) { crm_debug("Discarding %s event in state: %s", fsa_input2string(timer->fsa_input), fsa_state2string(controld_globals.fsa_state)); } else if (timer->fsa_input != I_NULL) { register_fsa_input(C_TIMER_POPPED, timer->fsa_input, NULL); } controld_trigger_fsa(); return TRUE; } bool controld_init_fsa_timers(void) { transition_timer = pcmk__assert_alloc(1, sizeof(fsa_timer_t)); integration_timer = pcmk__assert_alloc(1, sizeof(fsa_timer_t)); finalization_timer = pcmk__assert_alloc(1, sizeof(fsa_timer_t)); election_timer = pcmk__assert_alloc(1, sizeof(fsa_timer_t)); shutdown_escalation_timer = pcmk__assert_alloc(1, sizeof(fsa_timer_t)); wait_timer = pcmk__assert_alloc(1, sizeof(fsa_timer_t)); recheck_timer = pcmk__assert_alloc(1, sizeof(fsa_timer_t)); election_timer->source_id = 0; election_timer->period_ms = 0; election_timer->fsa_input = I_DC_TIMEOUT; election_timer->callback = crm_timer_popped; election_timer->log_error = FALSE; transition_timer->source_id = 0; transition_timer->period_ms = 0; transition_timer->fsa_input = I_PE_CALC; transition_timer->callback = crm_timer_popped; transition_timer->log_error = FALSE; integration_timer->source_id = 0; integration_timer->period_ms = 0; integration_timer->fsa_input = I_INTEGRATED; integration_timer->callback = crm_timer_popped; integration_timer->log_error = TRUE; finalization_timer->source_id = 0; finalization_timer->period_ms = 0; finalization_timer->fsa_input = I_FINALIZED; finalization_timer->callback = crm_timer_popped; finalization_timer->log_error = FALSE; /* We can't use I_FINALIZED here, because that creates a bug in the join * process where a joining node can be stuck in S_PENDING while we think it * is in S_NOT_DC. This created an infinite transition loop in which we * continually send probes which the node NACKs because it's pending. * * If we have nodes where the cluster layer is active but the controller is * not, we can avoid this causing an election/join loop, in the integration * phase. */ finalization_timer->fsa_input = I_ELECTION; shutdown_escalation_timer->source_id = 0; shutdown_escalation_timer->period_ms = 0; shutdown_escalation_timer->fsa_input = I_STOP; shutdown_escalation_timer->callback = crm_timer_popped; shutdown_escalation_timer->log_error = TRUE; wait_timer->source_id = 0; wait_timer->period_ms = 2000; wait_timer->fsa_input = I_NULL; wait_timer->callback = crm_timer_popped; wait_timer->log_error = FALSE; recheck_timer->source_id = 0; recheck_timer->period_ms = 0; recheck_timer->fsa_input = I_PE_CALC; recheck_timer->callback = crm_timer_popped; recheck_timer->log_error = FALSE; return TRUE; } /*! * \internal * \brief Configure timers based on the CIB * * \param[in,out] options Name/value pairs for configured options */ void controld_configure_fsa_timers(GHashTable *options) { const char *value = NULL; // Election timer value = g_hash_table_lookup(options, PCMK_OPT_DC_DEADTIME); pcmk_parse_interval_spec(value, &(election_timer->period_ms)); // Integration timer value = g_hash_table_lookup(options, PCMK_OPT_JOIN_INTEGRATION_TIMEOUT); pcmk_parse_interval_spec(value, &(integration_timer->period_ms)); // Finalization timer value = g_hash_table_lookup(options, PCMK_OPT_JOIN_FINALIZATION_TIMEOUT); pcmk_parse_interval_spec(value, &(finalization_timer->period_ms)); // Shutdown escalation timer value = g_hash_table_lookup(options, PCMK_OPT_SHUTDOWN_ESCALATION); pcmk_parse_interval_spec(value, &(shutdown_escalation_timer->period_ms)); crm_debug("Shutdown escalation occurs if DC has not responded to request " "in %ums", shutdown_escalation_timer->period_ms); // Transition timer value = g_hash_table_lookup(options, PCMK_OPT_TRANSITION_DELAY); pcmk_parse_interval_spec(value, &(transition_timer->period_ms)); // Recheck interval value = g_hash_table_lookup(options, PCMK_OPT_CLUSTER_RECHECK_INTERVAL); pcmk_parse_interval_spec(value, &recheck_interval_ms); crm_debug("Re-run scheduler after %dms of inactivity", recheck_interval_ms); } void controld_free_fsa_timers(void) { controld_stop_timer(transition_timer); controld_stop_timer(integration_timer); controld_stop_timer(finalization_timer); controld_stop_timer(election_timer); controld_stop_timer(shutdown_escalation_timer); controld_stop_timer(wait_timer); controld_stop_timer(recheck_timer); free(transition_timer); transition_timer = NULL; free(integration_timer); integration_timer = NULL; free(finalization_timer); finalization_timer = NULL; free(election_timer); election_timer = NULL; free(shutdown_escalation_timer); shutdown_escalation_timer = NULL; free(wait_timer); wait_timer = NULL; free(recheck_timer); recheck_timer = NULL; } /*! * \internal * \brief Check whether the transition timer is started * \return true if the transition timer is started, or false otherwise */ bool controld_is_started_transition_timer(void) { return (transition_timer->period_ms > 0) && (transition_timer->source_id != 0); } /*! * \internal * \brief Start the recheck timer */ void controld_start_recheck_timer(void) { // Default to recheck interval configured in CIB (if any) guint period_ms = recheck_interval_ms; // If scheduler supplied a "recheck by" time, check whether that's sooner if (controld_globals.transition_graph->recheck_by > 0) { time_t diff_seconds = controld_globals.transition_graph->recheck_by - time(NULL); if (diff_seconds < 1) { // We're already past the desired time period_ms = 500; } else { + // coverity[store_truncates_time_t] period_ms = (guint) diff_seconds * 1000; } // Use "recheck by" only if it's sooner than interval from CIB if (period_ms > recheck_interval_ms) { period_ms = recheck_interval_ms; } } if (period_ms > 0) { recheck_timer->period_ms = period_ms; controld_start_timer(recheck_timer); } } /*! * \internal * \brief Start the wait timer */ void controld_start_wait_timer(void) { controld_start_timer(wait_timer); } /*! * \internal * \brief Stop the recheck timer * * \return true if the recheck timer was running, or false otherwise */ bool controld_stop_recheck_timer(void) { return controld_stop_timer(recheck_timer); } /*! * \brief Get the transition timer's configured period * \return The transition_timer's period */ guint controld_get_period_transition_timer(void) { return transition_timer->period_ms; } /*! * \internal * \brief Reset the election timer's counter to 0 */ void controld_reset_counter_election_timer(void) { election_timer->counter = 0; } /*! * \internal * \brief Stop the transition timer * * \return true if the transition timer was running, or false otherwise */ bool controld_stop_transition_timer(void) { return controld_stop_timer(transition_timer); } /*! * \internal * \brief Start the transition timer */ void controld_start_transition_timer(void) { controld_start_timer(transition_timer); } /*! * \internal * \brief Start the countdown sequence for a shutdown * * \param[in] default_period_ms Period to use if the shutdown escalation * timer's period is 0 */ void controld_shutdown_start_countdown(guint default_period_ms) { if (shutdown_escalation_timer->period_ms == 0) { shutdown_escalation_timer->period_ms = default_period_ms; } crm_notice("Initiating controller shutdown sequence " CRM_XS " limit=%ums", shutdown_escalation_timer->period_ms); controld_start_timer(shutdown_escalation_timer); } diff --git a/lib/lrmd/lrmd_client.c b/lib/lrmd/lrmd_client.c index dac316fbb7..406205c809 100644 --- a/lib/lrmd/lrmd_client.c +++ b/lib/lrmd/lrmd_client.c @@ -1,2712 +1,2714 @@ /* * Copyright 2012-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include // uint32_t, uint64_t #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include // stonith__* #ifdef HAVE_GNUTLS_GNUTLS_H # include #endif #include #include #include #include #include #define MAX_TLS_RECV_WAIT 10000 CRM_TRACE_INIT_DATA(lrmd); static int lrmd_api_disconnect(lrmd_t * lrmd); static int lrmd_api_is_connected(lrmd_t * lrmd); /* IPC proxy functions */ int lrmd_internal_proxy_send(lrmd_t * lrmd, xmlNode *msg); static void lrmd_internal_proxy_dispatch(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)); #ifdef HAVE_GNUTLS_GNUTLS_H // GnuTLS client handshake timeout in seconds #define TLS_HANDSHAKE_TIMEOUT 5 gnutls_psk_client_credentials_t psk_cred_s; static void lrmd_tls_disconnect(lrmd_t * lrmd); static int global_remote_msg_id = 0; static void lrmd_tls_connection_destroy(gpointer userdata); static int add_tls_to_mainloop(lrmd_t *lrmd, bool do_handshake); #endif static void report_async_connection_result(lrmd_t * lrmd, int rc); typedef struct lrmd_private_s { uint64_t type; char *token; mainloop_io_t *source; /* IPC parameters */ crm_ipc_t *ipc; pcmk__remote_t *remote; /* Extra TLS parameters */ char *remote_nodename; #ifdef HAVE_GNUTLS_GNUTLS_H char *server; int port; gnutls_psk_client_credentials_t psk_cred_c; /* while the async connection is occurring, this is the id * of the connection timeout timer. */ int async_timer; int sock; /* since tls requires a round trip across the network for a * request/reply, there are times where we just want to be able * to send a request from the client and not wait around (or even care * about) what the reply is. */ int expected_late_replies; GList *pending_notify; crm_trigger_t *process_notify; crm_trigger_t *handshake_trigger; #endif lrmd_event_callback callback; /* Internal IPC proxy msg passing for remote guests */ void (*proxy_callback)(lrmd_t *lrmd, void *userdata, xmlNode *msg); void *proxy_callback_userdata; char *peer_version; } lrmd_private_t; static lrmd_list_t * lrmd_list_add(lrmd_list_t * head, const char *value) { lrmd_list_t *p, *end; p = pcmk__assert_alloc(1, sizeof(lrmd_list_t)); p->val = strdup(value); end = head; while (end && end->next) { end = end->next; } if (end) { end->next = p; } else { head = p; } return head; } void lrmd_list_freeall(lrmd_list_t * head) { lrmd_list_t *p; while (head) { char *val = (char *)head->val; p = head->next; free(val); free(head); head = p; } } lrmd_key_value_t * lrmd_key_value_add(lrmd_key_value_t * head, const char *key, const char *value) { lrmd_key_value_t *p, *end; p = pcmk__assert_alloc(1, sizeof(lrmd_key_value_t)); p->key = strdup(key); p->value = strdup(value); end = head; while (end && end->next) { end = end->next; } if (end) { end->next = p; } else { head = p; } return head; } void lrmd_key_value_freeall(lrmd_key_value_t * head) { lrmd_key_value_t *p; while (head) { p = head->next; free(head->key); free(head->value); free(head); head = p; } } /*! * \brief Create a new lrmd_event_data_t object * * \param[in] rsc_id ID of resource involved in event * \param[in] task Action name * \param[in] interval_ms Action interval * * \return Newly allocated and initialized lrmd_event_data_t * \note This functions asserts on memory errors, so the return value is * guaranteed to be non-NULL. The caller is responsible for freeing the * result with lrmd_free_event(). */ lrmd_event_data_t * lrmd_new_event(const char *rsc_id, const char *task, guint interval_ms) { lrmd_event_data_t *event = pcmk__assert_alloc(1, sizeof(lrmd_event_data_t)); // lrmd_event_data_t has (const char *) members that lrmd_free_event() frees event->rsc_id = pcmk__str_copy(rsc_id); event->op_type = pcmk__str_copy(task); event->interval_ms = interval_ms; return event; } lrmd_event_data_t * lrmd_copy_event(lrmd_event_data_t * event) { lrmd_event_data_t *copy = NULL; copy = pcmk__assert_alloc(1, sizeof(lrmd_event_data_t)); copy->type = event->type; // lrmd_event_data_t has (const char *) members that lrmd_free_event() frees copy->rsc_id = pcmk__str_copy(event->rsc_id); copy->op_type = pcmk__str_copy(event->op_type); copy->user_data = pcmk__str_copy(event->user_data); copy->output = pcmk__str_copy(event->output); copy->remote_nodename = pcmk__str_copy(event->remote_nodename); copy->exit_reason = pcmk__str_copy(event->exit_reason); copy->call_id = event->call_id; copy->timeout = event->timeout; copy->interval_ms = event->interval_ms; copy->start_delay = event->start_delay; copy->rsc_deleted = event->rsc_deleted; copy->rc = event->rc; copy->op_status = event->op_status; copy->t_run = event->t_run; copy->t_rcchange = event->t_rcchange; copy->exec_time = event->exec_time; copy->queue_time = event->queue_time; copy->connection_rc = event->connection_rc; copy->params = pcmk__str_table_dup(event->params); return copy; } /*! * \brief Free an executor event * * \param[in,out] Executor event object to free */ void lrmd_free_event(lrmd_event_data_t *event) { if (event == NULL) { return; } // @TODO Why are these const char *? free((void *) event->rsc_id); free((void *) event->op_type); free((void *) event->user_data); free((void *) event->remote_nodename); lrmd__reset_result(event); if (event->params != NULL) { g_hash_table_destroy(event->params); } free(event); } static void lrmd_dispatch_internal(lrmd_t * lrmd, xmlNode * msg) { const char *type; const char *proxy_session = crm_element_value(msg, PCMK__XA_LRMD_IPC_SESSION); lrmd_private_t *native = lrmd->lrmd_private; lrmd_event_data_t event = { 0, }; if (proxy_session != NULL) { /* this is proxy business */ lrmd_internal_proxy_dispatch(lrmd, msg); return; } else if (!native->callback) { /* no callback set */ crm_trace("notify event received but client has not set callback"); return; } event.remote_nodename = native->remote_nodename; type = crm_element_value(msg, PCMK__XA_LRMD_OP); crm_element_value_int(msg, PCMK__XA_LRMD_CALLID, &event.call_id); event.rsc_id = crm_element_value(msg, PCMK__XA_LRMD_RSC_ID); if (pcmk__str_eq(type, LRMD_OP_RSC_REG, pcmk__str_none)) { event.type = lrmd_event_register; } else if (pcmk__str_eq(type, LRMD_OP_RSC_UNREG, pcmk__str_none)) { event.type = lrmd_event_unregister; } else if (pcmk__str_eq(type, LRMD_OP_RSC_EXEC, pcmk__str_none)) { int rc = 0; int exec_time = 0; int queue_time = 0; time_t epoch = 0; crm_element_value_int(msg, PCMK__XA_LRMD_TIMEOUT, &event.timeout); crm_element_value_ms(msg, PCMK__XA_LRMD_RSC_INTERVAL, &event.interval_ms); crm_element_value_int(msg, PCMK__XA_LRMD_RSC_START_DELAY, &event.start_delay); crm_element_value_int(msg, PCMK__XA_LRMD_EXEC_RC, &rc); event.rc = (enum ocf_exitcode) rc; crm_element_value_int(msg, PCMK__XA_LRMD_EXEC_OP_STATUS, &event.op_status); crm_element_value_int(msg, PCMK__XA_LRMD_RSC_DELETED, &event.rsc_deleted); crm_element_value_epoch(msg, PCMK__XA_LRMD_RUN_TIME, &epoch); + // coverity[store_truncates_time_t] event.t_run = (unsigned int) epoch; crm_element_value_epoch(msg, PCMK__XA_LRMD_RCCHANGE_TIME, &epoch); + // coverity[store_truncates_time_t] event.t_rcchange = (unsigned int) epoch; crm_element_value_int(msg, PCMK__XA_LRMD_EXEC_TIME, &exec_time); CRM_LOG_ASSERT(exec_time >= 0); event.exec_time = QB_MAX(0, exec_time); crm_element_value_int(msg, PCMK__XA_LRMD_QUEUE_TIME, &queue_time); CRM_LOG_ASSERT(queue_time >= 0); event.queue_time = QB_MAX(0, queue_time); event.op_type = crm_element_value(msg, PCMK__XA_LRMD_RSC_ACTION); event.user_data = crm_element_value(msg, PCMK__XA_LRMD_RSC_USERDATA_STR); event.type = lrmd_event_exec_complete; /* output and exit_reason may be freed by a callback */ event.output = crm_element_value_copy(msg, PCMK__XA_LRMD_RSC_OUTPUT); lrmd__set_result(&event, event.rc, event.op_status, crm_element_value(msg, PCMK__XA_LRMD_RSC_EXIT_REASON)); event.params = xml2list(msg); } else if (pcmk__str_eq(type, LRMD_OP_NEW_CLIENT, pcmk__str_none)) { event.type = lrmd_event_new_client; } else if (pcmk__str_eq(type, LRMD_OP_POKE, pcmk__str_none)) { event.type = lrmd_event_poke; } else { return; } crm_trace("op %s notify event received", type); native->callback(&event); if (event.params) { g_hash_table_destroy(event.params); } lrmd__reset_result(&event); } // \return Always 0, to indicate that IPC mainloop source should be kept static int lrmd_ipc_dispatch(const char *buffer, ssize_t length, gpointer userdata) { lrmd_t *lrmd = userdata; lrmd_private_t *native = lrmd->lrmd_private; if (native->callback != NULL) { xmlNode *msg = pcmk__xml_parse(buffer); lrmd_dispatch_internal(lrmd, msg); free_xml(msg); } return 0; } #ifdef HAVE_GNUTLS_GNUTLS_H static void lrmd_free_xml(gpointer userdata) { free_xml((xmlNode *) userdata); } static bool remote_executor_connected(lrmd_t * lrmd) { lrmd_private_t *native = lrmd->lrmd_private; return (native->remote->tls_session != NULL); } /*! * \internal * \brief TLS dispatch function (for both trigger and file descriptor sources) * * \param[in,out] userdata API connection * * \return Always return a nonnegative value, which as a file descriptor * dispatch function means keep the mainloop source, and as a * trigger dispatch function, 0 means remove the trigger from the * mainloop while 1 means keep it (and job completed) */ static int lrmd_tls_dispatch(gpointer userdata) { lrmd_t *lrmd = userdata; lrmd_private_t *native = lrmd->lrmd_private; xmlNode *xml = NULL; int rc = pcmk_rc_ok; if (!remote_executor_connected(lrmd)) { crm_trace("TLS dispatch triggered after disconnect"); return 0; } crm_trace("TLS dispatch triggered"); /* First check if there are any pending notifies to process that came * while we were waiting for replies earlier. */ if (native->pending_notify) { GList *iter = NULL; crm_trace("Processing pending notifies"); for (iter = native->pending_notify; iter; iter = iter->next) { lrmd_dispatch_internal(lrmd, iter->data); } g_list_free_full(native->pending_notify, lrmd_free_xml); native->pending_notify = NULL; } /* Next read the current buffer and see if there are any messages to handle. */ switch (pcmk__remote_ready(native->remote, 0)) { case pcmk_rc_ok: rc = pcmk__read_remote_message(native->remote, -1); xml = pcmk__remote_message_xml(native->remote); break; case ETIME: // Nothing to read, check if a full message is already in buffer xml = pcmk__remote_message_xml(native->remote); break; default: rc = ENOTCONN; break; } while (xml) { const char *msg_type = crm_element_value(xml, PCMK__XA_LRMD_REMOTE_MSG_TYPE); if (pcmk__str_eq(msg_type, "notify", pcmk__str_casei)) { lrmd_dispatch_internal(lrmd, xml); } else if (pcmk__str_eq(msg_type, "reply", pcmk__str_casei)) { if (native->expected_late_replies > 0) { native->expected_late_replies--; } else { int reply_id = 0; crm_element_value_int(xml, PCMK__XA_LRMD_CALLID, &reply_id); /* if this happens, we want to know about it */ crm_err("Got outdated Pacemaker Remote reply %d", reply_id); } } free_xml(xml); xml = pcmk__remote_message_xml(native->remote); } if (rc == ENOTCONN) { crm_info("Lost %s executor connection while reading data", (native->remote_nodename? native->remote_nodename : "local")); lrmd_tls_disconnect(lrmd); return 0; } return 1; } #endif /* Not used with mainloop */ int lrmd_poll(lrmd_t * lrmd, int timeout) { lrmd_private_t *native = lrmd->lrmd_private; switch (native->type) { case pcmk__client_ipc: return crm_ipc_ready(native->ipc); #ifdef HAVE_GNUTLS_GNUTLS_H case pcmk__client_tls: if (native->pending_notify) { return 1; } else { int rc = pcmk__remote_ready(native->remote, 0); switch (rc) { case pcmk_rc_ok: return 1; case ETIME: return 0; default: return pcmk_rc2legacy(rc); } } #endif default: crm_err("Unsupported executor connection type (bug?): %d", native->type); return -EPROTONOSUPPORT; } } /* Not used with mainloop */ bool lrmd_dispatch(lrmd_t * lrmd) { lrmd_private_t *private = NULL; CRM_ASSERT(lrmd != NULL); private = lrmd->lrmd_private; switch (private->type) { case pcmk__client_ipc: while (crm_ipc_ready(private->ipc)) { if (crm_ipc_read(private->ipc) > 0) { const char *msg = crm_ipc_buffer(private->ipc); lrmd_ipc_dispatch(msg, strlen(msg), lrmd); } } break; #ifdef HAVE_GNUTLS_GNUTLS_H case pcmk__client_tls: lrmd_tls_dispatch(lrmd); break; #endif default: crm_err("Unsupported executor connection type (bug?): %d", private->type); } if (lrmd_api_is_connected(lrmd) == FALSE) { crm_err("Connection closed"); return FALSE; } return TRUE; } static xmlNode * lrmd_create_op(const char *token, const char *op, xmlNode *data, int timeout, enum lrmd_call_options options) { xmlNode *op_msg = NULL; CRM_CHECK(token != NULL, return NULL); op_msg = pcmk__xe_create(NULL, PCMK__XE_LRMD_COMMAND); crm_xml_add(op_msg, PCMK__XA_T, PCMK__VALUE_LRMD); crm_xml_add(op_msg, PCMK__XA_LRMD_OP, op); crm_xml_add_int(op_msg, PCMK__XA_LRMD_TIMEOUT, timeout); crm_xml_add_int(op_msg, PCMK__XA_LRMD_CALLOPT, options); if (data != NULL) { xmlNode *wrapper = pcmk__xe_create(op_msg, PCMK__XE_LRMD_CALLDATA); pcmk__xml_copy(wrapper, data); } crm_trace("Created executor %s command with call options %.8lx (%d)", op, (long)options, options); return op_msg; } static void lrmd_ipc_connection_destroy(gpointer userdata) { lrmd_t *lrmd = userdata; lrmd_private_t *native = lrmd->lrmd_private; switch (native->type) { case pcmk__client_ipc: crm_info("Disconnected from local executor"); break; #ifdef HAVE_GNUTLS_GNUTLS_H case pcmk__client_tls: crm_info("Disconnected from remote executor on %s", native->remote_nodename); break; #endif default: crm_err("Unsupported executor connection type %d (bug?)", native->type); } /* Prevent these from being cleaned up in lrmd_api_disconnect() */ native->ipc = NULL; native->source = NULL; if (native->callback) { lrmd_event_data_t event = { 0, }; event.type = lrmd_event_disconnect; event.remote_nodename = native->remote_nodename; native->callback(&event); } } #ifdef HAVE_GNUTLS_GNUTLS_H static void lrmd_tls_connection_destroy(gpointer userdata) { lrmd_t *lrmd = userdata; lrmd_private_t *native = lrmd->lrmd_private; crm_info("TLS connection destroyed"); if (native->remote->tls_session) { gnutls_bye(*native->remote->tls_session, GNUTLS_SHUT_RDWR); gnutls_deinit(*native->remote->tls_session); gnutls_free(native->remote->tls_session); native->remote->tls_session = NULL; } if (native->psk_cred_c) { gnutls_psk_free_client_credentials(native->psk_cred_c); } if (native->sock) { close(native->sock); } if (native->process_notify) { mainloop_destroy_trigger(native->process_notify); native->process_notify = NULL; } if (native->pending_notify) { g_list_free_full(native->pending_notify, lrmd_free_xml); native->pending_notify = NULL; } if (native->handshake_trigger != NULL) { mainloop_destroy_trigger(native->handshake_trigger); native->handshake_trigger = NULL; } free(native->remote->buffer); free(native->remote->start_state); native->remote->buffer = NULL; native->remote->start_state = NULL; native->source = 0; native->sock = 0; native->psk_cred_c = NULL; native->sock = 0; if (native->callback) { lrmd_event_data_t event = { 0, }; event.remote_nodename = native->remote_nodename; event.type = lrmd_event_disconnect; native->callback(&event); } return; } // \return Standard Pacemaker return code int lrmd__remote_send_xml(pcmk__remote_t *session, xmlNode *msg, uint32_t id, const char *msg_type) { crm_xml_add_int(msg, PCMK__XA_LRMD_REMOTE_MSG_ID, id); crm_xml_add(msg, PCMK__XA_LRMD_REMOTE_MSG_TYPE, msg_type); return pcmk__remote_send_xml(session, msg); } // \return Standard Pacemaker return code static int read_remote_reply(lrmd_t *lrmd, int total_timeout, int expected_reply_id, xmlNode **reply) { lrmd_private_t *native = lrmd->lrmd_private; time_t start = time(NULL); const char *msg_type = NULL; int reply_id = 0; int remaining_timeout = 0; int rc = pcmk_rc_ok; /* A timeout of 0 here makes no sense. We have to wait a period of time * for the response to come back. If -1 or 0, default to 10 seconds. */ if (total_timeout <= 0 || total_timeout > MAX_TLS_RECV_WAIT) { total_timeout = MAX_TLS_RECV_WAIT; } for (*reply = NULL; *reply == NULL; ) { *reply = pcmk__remote_message_xml(native->remote); if (*reply == NULL) { /* read some more off the tls buffer if we still have time left. */ if (remaining_timeout) { remaining_timeout = total_timeout - ((time(NULL) - start) * 1000); } else { remaining_timeout = total_timeout; } if (remaining_timeout <= 0) { return ETIME; } rc = pcmk__read_remote_message(native->remote, remaining_timeout); if (rc != pcmk_rc_ok) { return rc; } *reply = pcmk__remote_message_xml(native->remote); if (*reply == NULL) { return ENOMSG; } } crm_element_value_int(*reply, PCMK__XA_LRMD_REMOTE_MSG_ID, &reply_id); msg_type = crm_element_value(*reply, PCMK__XA_LRMD_REMOTE_MSG_TYPE); if (!msg_type) { crm_err("Empty msg type received while waiting for reply"); free_xml(*reply); *reply = NULL; } else if (pcmk__str_eq(msg_type, "notify", pcmk__str_casei)) { /* got a notify while waiting for reply, trigger the notify to be processed later */ crm_info("queueing notify"); native->pending_notify = g_list_append(native->pending_notify, *reply); if (native->process_notify) { crm_info("notify trigger set."); mainloop_set_trigger(native->process_notify); } *reply = NULL; } else if (!pcmk__str_eq(msg_type, "reply", pcmk__str_casei)) { /* msg isn't a reply, make some noise */ crm_err("Expected a reply, got %s", msg_type); free_xml(*reply); *reply = NULL; } else if (reply_id != expected_reply_id) { if (native->expected_late_replies > 0) { native->expected_late_replies--; } else { crm_err("Got outdated reply, expected id %d got id %d", expected_reply_id, reply_id); } free_xml(*reply); *reply = NULL; } } if (native->remote->buffer && native->process_notify) { mainloop_set_trigger(native->process_notify); } return rc; } // \return Standard Pacemaker return code static int send_remote_message(lrmd_t *lrmd, xmlNode *msg) { int rc = pcmk_rc_ok; lrmd_private_t *native = lrmd->lrmd_private; global_remote_msg_id++; if (global_remote_msg_id <= 0) { global_remote_msg_id = 1; } rc = lrmd__remote_send_xml(native->remote, msg, global_remote_msg_id, "request"); if (rc != pcmk_rc_ok) { crm_err("Disconnecting because TLS message could not be sent to " "Pacemaker Remote: %s", pcmk_rc_str(rc)); lrmd_tls_disconnect(lrmd); } return rc; } static int lrmd_tls_send_recv(lrmd_t * lrmd, xmlNode * msg, int timeout, xmlNode ** reply) { int rc = 0; xmlNode *xml = NULL; if (!remote_executor_connected(lrmd)) { return -ENOTCONN; } rc = send_remote_message(lrmd, msg); if (rc != pcmk_rc_ok) { return pcmk_rc2legacy(rc); } rc = read_remote_reply(lrmd, timeout, global_remote_msg_id, &xml); if (rc != pcmk_rc_ok) { crm_err("Disconnecting remote after request %d reply not received: %s " CRM_XS " rc=%d timeout=%dms", global_remote_msg_id, pcmk_rc_str(rc), rc, timeout); lrmd_tls_disconnect(lrmd); } if (reply) { *reply = xml; } else { free_xml(xml); } return pcmk_rc2legacy(rc); } #endif static int lrmd_send_xml(lrmd_t * lrmd, xmlNode * msg, int timeout, xmlNode ** reply) { int rc = pcmk_ok; lrmd_private_t *native = lrmd->lrmd_private; switch (native->type) { case pcmk__client_ipc: rc = crm_ipc_send(native->ipc, msg, crm_ipc_client_response, timeout, reply); break; #ifdef HAVE_GNUTLS_GNUTLS_H case pcmk__client_tls: rc = lrmd_tls_send_recv(lrmd, msg, timeout, reply); break; #endif default: crm_err("Unsupported executor connection type (bug?): %d", native->type); rc = -EPROTONOSUPPORT; } return rc; } static int lrmd_send_xml_no_reply(lrmd_t * lrmd, xmlNode * msg) { int rc = pcmk_ok; lrmd_private_t *native = lrmd->lrmd_private; switch (native->type) { case pcmk__client_ipc: rc = crm_ipc_send(native->ipc, msg, crm_ipc_flags_none, 0, NULL); break; #ifdef HAVE_GNUTLS_GNUTLS_H case pcmk__client_tls: rc = send_remote_message(lrmd, msg); if (rc == pcmk_rc_ok) { /* we don't want to wait around for the reply, but * since the request/reply protocol needs to behave the same * as libqb, a reply will eventually come later anyway. */ native->expected_late_replies++; } rc = pcmk_rc2legacy(rc); break; #endif default: crm_err("Unsupported executor connection type (bug?): %d", native->type); rc = -EPROTONOSUPPORT; } return rc; } static int lrmd_api_is_connected(lrmd_t * lrmd) { lrmd_private_t *native = lrmd->lrmd_private; switch (native->type) { case pcmk__client_ipc: return crm_ipc_connected(native->ipc); #ifdef HAVE_GNUTLS_GNUTLS_H case pcmk__client_tls: return remote_executor_connected(lrmd); #endif default: crm_err("Unsupported executor connection type (bug?): %d", native->type); return 0; } } /*! * \internal * \brief Send a prepared API command to the executor * * \param[in,out] lrmd Existing connection to the executor * \param[in] op Name of API command to send * \param[in] data Command data XML to add to the sent command * \param[out] output_data If expecting a reply, it will be stored here * \param[in] timeout Timeout in milliseconds (if 0, defaults to * a sensible value per the type of connection, * standard vs. pacemaker remote); * also propagated to the command XML * \param[in] call_options Call options to pass to server when sending * \param[in] expect_reply If TRUE, wait for a reply from the server; * must be TRUE for IPC (as opposed to TLS) clients * * \return pcmk_ok on success, -errno on error */ static int lrmd_send_command(lrmd_t *lrmd, const char *op, xmlNode *data, xmlNode **output_data, int timeout, enum lrmd_call_options options, gboolean expect_reply) { int rc = pcmk_ok; lrmd_private_t *native = lrmd->lrmd_private; xmlNode *op_msg = NULL; xmlNode *op_reply = NULL; if (!lrmd_api_is_connected(lrmd)) { return -ENOTCONN; } if (op == NULL) { crm_err("No operation specified"); return -EINVAL; } CRM_CHECK(native->token != NULL,; ); crm_trace("Sending %s op to executor", op); op_msg = lrmd_create_op(native->token, op, data, timeout, options); if (op_msg == NULL) { return -EINVAL; } if (expect_reply) { rc = lrmd_send_xml(lrmd, op_msg, timeout, &op_reply); } else { rc = lrmd_send_xml_no_reply(lrmd, op_msg); goto done; } if (rc < 0) { crm_perror(LOG_ERR, "Couldn't perform %s operation (timeout=%d): %d", op, timeout, rc); goto done; } else if(op_reply == NULL) { rc = -ENOMSG; goto done; } rc = pcmk_ok; crm_trace("%s op reply received", op); if (crm_element_value_int(op_reply, PCMK__XA_LRMD_RC, &rc) != 0) { rc = -ENOMSG; goto done; } crm_log_xml_trace(op_reply, "Reply"); if (output_data) { *output_data = op_reply; op_reply = NULL; /* Prevent subsequent free */ } done: if (lrmd_api_is_connected(lrmd) == FALSE) { crm_err("Executor disconnected"); } free_xml(op_msg); free_xml(op_reply); return rc; } static int lrmd_api_poke_connection(lrmd_t * lrmd) { int rc; lrmd_private_t *native = lrmd->lrmd_private; xmlNode *data = pcmk__xe_create(NULL, PCMK__XE_LRMD_RSC); crm_xml_add(data, PCMK__XA_LRMD_ORIGIN, __func__); rc = lrmd_send_command(lrmd, LRMD_OP_POKE, data, NULL, 0, 0, (native->type == pcmk__client_ipc)); free_xml(data); return rc < 0 ? rc : pcmk_ok; } // \return Standard Pacemaker return code int lrmd__validate_remote_settings(lrmd_t *lrmd, GHashTable *hash) { int rc = pcmk_rc_ok; const char *value; lrmd_private_t *native = lrmd->lrmd_private; xmlNode *data = pcmk__xe_create(NULL, PCMK__XA_LRMD_OP); crm_xml_add(data, PCMK__XA_LRMD_ORIGIN, __func__); value = g_hash_table_lookup(hash, PCMK_OPT_STONITH_WATCHDOG_TIMEOUT); if ((value) && (stonith__watchdog_fencing_enabled_for_node(native->remote_nodename))) { crm_xml_add(data, PCMK__XA_LRMD_WATCHDOG, value); } rc = lrmd_send_command(lrmd, LRMD_OP_CHECK, data, NULL, 0, 0, (native->type == pcmk__client_ipc)); free_xml(data); return (rc < 0)? pcmk_legacy2rc(rc) : pcmk_rc_ok; } static int lrmd_handshake(lrmd_t * lrmd, const char *name) { int rc = pcmk_ok; lrmd_private_t *native = lrmd->lrmd_private; xmlNode *reply = NULL; xmlNode *hello = pcmk__xe_create(NULL, PCMK__XE_LRMD_COMMAND); crm_xml_add(hello, PCMK__XA_T, PCMK__VALUE_LRMD); crm_xml_add(hello, PCMK__XA_LRMD_OP, CRM_OP_REGISTER); crm_xml_add(hello, PCMK__XA_LRMD_CLIENTNAME, name); crm_xml_add(hello, PCMK__XA_LRMD_PROTOCOL_VERSION, LRMD_PROTOCOL_VERSION); /* advertise that we are a proxy provider */ if (native->proxy_callback) { pcmk__xe_set_bool_attr(hello, PCMK__XA_LRMD_IS_IPC_PROVIDER, true); } rc = lrmd_send_xml(lrmd, hello, -1, &reply); if (rc < 0) { crm_perror(LOG_DEBUG, "Couldn't complete registration with the executor API: %d", rc); rc = -ECOMM; } else if (reply == NULL) { crm_err("Did not receive registration reply"); rc = -EPROTO; } else { const char *version = crm_element_value(reply, PCMK__XA_LRMD_PROTOCOL_VERSION); const char *msg_type = crm_element_value(reply, PCMK__XA_LRMD_OP); const char *tmp_ticket = crm_element_value(reply, PCMK__XA_LRMD_CLIENTID); const char *start_state = crm_element_value(reply, PCMK__XA_NODE_START_STATE); long long uptime = -1; crm_element_value_int(reply, PCMK__XA_LRMD_RC, &rc); /* The remote executor may add its uptime to the XML reply, which is * useful in handling transient attributes when the connection to the * remote node unexpectedly drops. If no parameter is given, just * default to -1. */ crm_element_value_ll(reply, PCMK__XA_UPTIME, &uptime); native->remote->uptime = uptime; if (start_state) { native->remote->start_state = strdup(start_state); } if (rc == -EPROTO) { crm_err("Executor protocol version mismatch between client (%s) and server (%s)", LRMD_PROTOCOL_VERSION, version); crm_log_xml_err(reply, "Protocol Error"); } else if (!pcmk__str_eq(msg_type, CRM_OP_REGISTER, pcmk__str_casei)) { crm_err("Invalid registration message: %s", msg_type); crm_log_xml_err(reply, "Bad reply"); rc = -EPROTO; } else if (tmp_ticket == NULL) { crm_err("No registration token provided"); crm_log_xml_err(reply, "Bad reply"); rc = -EPROTO; } else { crm_trace("Obtained registration token: %s", tmp_ticket); native->token = strdup(tmp_ticket); native->peer_version = strdup(version?version:"1.0"); /* Included since 1.1 */ rc = pcmk_ok; } } free_xml(reply); free_xml(hello); if (rc != pcmk_ok) { lrmd_api_disconnect(lrmd); } return rc; } static int lrmd_ipc_connect(lrmd_t * lrmd, int *fd) { int rc = pcmk_ok; lrmd_private_t *native = lrmd->lrmd_private; struct ipc_client_callbacks lrmd_callbacks = { .dispatch = lrmd_ipc_dispatch, .destroy = lrmd_ipc_connection_destroy }; crm_info("Connecting to executor"); if (fd) { /* No mainloop */ native->ipc = crm_ipc_new(CRM_SYSTEM_LRMD, 0); if (native->ipc != NULL) { rc = pcmk__connect_generic_ipc(native->ipc); if (rc == pcmk_rc_ok) { rc = pcmk__ipc_fd(native->ipc, fd); } if (rc != pcmk_rc_ok) { crm_err("Connection to executor failed: %s", pcmk_rc_str(rc)); rc = -ENOTCONN; } } } else { native->source = mainloop_add_ipc_client(CRM_SYSTEM_LRMD, G_PRIORITY_HIGH, 0, lrmd, &lrmd_callbacks); native->ipc = mainloop_get_ipc_client(native->source); } if (native->ipc == NULL) { crm_debug("Could not connect to the executor API"); rc = -ENOTCONN; } return rc; } #ifdef HAVE_GNUTLS_GNUTLS_H static void copy_gnutls_datum(gnutls_datum_t *dest, gnutls_datum_t *source) { CRM_ASSERT((dest != NULL) && (source != NULL) && (source->data != NULL)); dest->data = gnutls_malloc(source->size); pcmk__mem_assert(dest->data); memcpy(dest->data, source->data, source->size); dest->size = source->size; } static void clear_gnutls_datum(gnutls_datum_t *datum) { gnutls_free(datum->data); datum->data = NULL; datum->size = 0; } #define KEY_READ_LEN 256 // Chunk size for reading key from file // \return Standard Pacemaker return code static int read_gnutls_key(const char *location, gnutls_datum_t *key) { FILE *stream = NULL; size_t buf_len = KEY_READ_LEN; if ((location == NULL) || (key == NULL)) { return EINVAL; } stream = fopen(location, "r"); if (stream == NULL) { return errno; } key->data = gnutls_malloc(buf_len); key->size = 0; while (!feof(stream)) { int next = fgetc(stream); if (next == EOF) { if (!feof(stream)) { crm_warn("Pacemaker Remote key read was partially successful " "(copy in memory may be corrupted)"); } break; } if (key->size == buf_len) { buf_len = key->size + KEY_READ_LEN; key->data = gnutls_realloc(key->data, buf_len); CRM_ASSERT(key->data); } key->data[key->size++] = (unsigned char) next; } fclose(stream); if (key->size == 0) { clear_gnutls_datum(key); return ENOKEY; } return pcmk_rc_ok; } // Cache the most recently used Pacemaker Remote authentication key struct key_cache_s { time_t updated; // When cached key was read (valid for 1 minute) const char *location; // Where cached key was read from gnutls_datum_t key; // Cached key }; static bool key_is_cached(struct key_cache_s *key_cache) { return key_cache->updated != 0; } static bool key_cache_expired(struct key_cache_s *key_cache) { return (time(NULL) - key_cache->updated) >= 60; } static void clear_key_cache(struct key_cache_s *key_cache) { clear_gnutls_datum(&(key_cache->key)); if ((key_cache->updated != 0) || (key_cache->location != NULL)) { key_cache->updated = 0; key_cache->location = NULL; crm_debug("Cleared Pacemaker Remote key cache"); } } static void get_cached_key(struct key_cache_s *key_cache, gnutls_datum_t *key) { copy_gnutls_datum(key, &(key_cache->key)); crm_debug("Using cached Pacemaker Remote key from %s", pcmk__s(key_cache->location, "unknown location")); } static void cache_key(struct key_cache_s *key_cache, gnutls_datum_t *key, const char *location) { key_cache->updated = time(NULL); key_cache->location = location; copy_gnutls_datum(&(key_cache->key), key); crm_debug("Using (and cacheing) Pacemaker Remote key from %s", pcmk__s(location, "unknown location")); } /*! * \internal * \brief Get Pacemaker Remote authentication key from file or cache * * \param[in] location Path to key file to try (this memory must * persist across all calls of this function) * \param[out] key Key from location or cache * * \return Standard Pacemaker return code */ static int get_remote_key(const char *location, gnutls_datum_t *key) { static struct key_cache_s key_cache = { 0, }; int rc = pcmk_rc_ok; if ((location == NULL) || (key == NULL)) { return EINVAL; } if (key_is_cached(&key_cache)) { if (key_cache_expired(&key_cache)) { clear_key_cache(&key_cache); } else { get_cached_key(&key_cache, key); return pcmk_rc_ok; } } rc = read_gnutls_key(location, key); if (rc != pcmk_rc_ok) { return rc; } cache_key(&key_cache, key, location); return pcmk_rc_ok; } /*! * \internal * \brief Initialize the Pacemaker Remote authentication key * * Try loading the Pacemaker Remote authentication key from cache if available, * otherwise from these locations, in order of preference: the value of the * PCMK_authkey_location environment variable, if set; the Pacemaker default key * file location; or (for historical reasons) /etc/corosync/authkey. * * \param[out] key Where to store key * * \return Standard Pacemaker return code */ int lrmd__init_remote_key(gnutls_datum_t *key) { static const char *env_location = NULL; static bool need_env = true; int env_rc = pcmk_rc_ok; int default_rc = pcmk_rc_ok; int alt_rc = pcmk_rc_ok; bool env_is_default = false; bool env_is_fallback = false; if (need_env) { env_location = pcmk__env_option(PCMK__ENV_AUTHKEY_LOCATION); need_env = false; } // Try location in environment variable, if set if (env_location != NULL) { env_rc = get_remote_key(env_location, key); if (env_rc == pcmk_rc_ok) { return pcmk_rc_ok; } env_is_default = !strcmp(env_location, DEFAULT_REMOTE_KEY_LOCATION); env_is_fallback = !strcmp(env_location, ALT_REMOTE_KEY_LOCATION); /* @TODO It would be more secure to fail, rather than fall back to the * default, if an explicitly set key location is not readable, and it * would be better to never use the Corosync location as a fallback. * However, that would break any deployments currently working with the * fallbacks. * * @COMPAT Change at 3.0.0 */ } // Try default location, if environment wasn't explicitly set to it if (env_is_default) { default_rc = env_rc; } else { default_rc = get_remote_key(DEFAULT_REMOTE_KEY_LOCATION, key); } // Try fallback location, if environment wasn't set to it and default failed // @COMPAT Drop at 3.0.0 if (env_is_fallback) { alt_rc = env_rc; } else if (default_rc != pcmk_rc_ok) { alt_rc = get_remote_key(ALT_REMOTE_KEY_LOCATION, key); } // We have all results, so log and return if ((env_rc != pcmk_rc_ok) && (default_rc != pcmk_rc_ok) && (alt_rc != pcmk_rc_ok)) { // Environment set, everything failed crm_warn("Could not read Pacemaker Remote key from %s (%s%s%s%s%s): %s", env_location, env_is_default? "" : "or default location ", env_is_default? "" : DEFAULT_REMOTE_KEY_LOCATION, !env_is_default && !env_is_fallback? " " : "", env_is_fallback? "" : "or fallback location ", env_is_fallback? "" : ALT_REMOTE_KEY_LOCATION, pcmk_rc_str(env_rc)); return ENOKEY; } if (env_rc != pcmk_rc_ok) { // Environment set but failed, using a default crm_warn("Could not read Pacemaker Remote key from %s " "(using %s location %s instead): %s", env_location, (default_rc == pcmk_rc_ok)? "default" : "fallback", (default_rc == pcmk_rc_ok)? DEFAULT_REMOTE_KEY_LOCATION : ALT_REMOTE_KEY_LOCATION, pcmk_rc_str(env_rc)); crm_warn("This undocumented behavior is deprecated and unsafe and will " "be removed in a future release"); return pcmk_rc_ok; } if (default_rc != pcmk_rc_ok) { if (alt_rc == pcmk_rc_ok) { // Environment variable unset, used alternate location // This gets caught by the default return below, but we additionally // warn on this behavior here. crm_warn("Read Pacemaker Remote key from alternate location %s", ALT_REMOTE_KEY_LOCATION); crm_warn("This undocumented behavior is deprecated and unsafe and will " "be removed in a future release"); } else { // Environment unset, defaults failed crm_warn("Could not read Pacemaker Remote key from default location %s" " (or fallback location %s): %s", DEFAULT_REMOTE_KEY_LOCATION, ALT_REMOTE_KEY_LOCATION, pcmk_rc_str(default_rc)); return ENOKEY; } } return pcmk_rc_ok; // Environment variable unset, a default worked } static void lrmd_gnutls_global_init(void) { static int gnutls_init = 0; if (!gnutls_init) { crm_gnutls_global_init(); } gnutls_init = 1; } static void tls_handshake_failed(lrmd_t *lrmd, int tls_rc, int rc) { lrmd_private_t *native = lrmd->lrmd_private; crm_warn("Disconnecting after TLS handshake with " "Pacemaker Remote server %s:%d failed: %s", native->server, native->port, (rc == EPROTO)? gnutls_strerror(tls_rc) : pcmk_rc_str(rc)); report_async_connection_result(lrmd, pcmk_rc2legacy(rc)); gnutls_deinit(*native->remote->tls_session); gnutls_free(native->remote->tls_session); native->remote->tls_session = NULL; lrmd_tls_connection_destroy(lrmd); } static void tls_handshake_succeeded(lrmd_t *lrmd) { lrmd_private_t *native = lrmd->lrmd_private; crm_info("TLS connection to Pacemaker Remote server %s:%d succeeded", native->server, native->port); add_tls_to_mainloop(lrmd, true); report_async_connection_result(lrmd, pcmk_rc2legacy(pcmk_rc_ok)); } #endif static void report_async_connection_result(lrmd_t * lrmd, int rc) { lrmd_private_t *native = lrmd->lrmd_private; if (native->callback) { lrmd_event_data_t event = { 0, }; event.type = lrmd_event_connect; event.remote_nodename = native->remote_nodename; event.connection_rc = rc; native->callback(&event); } } #ifdef HAVE_GNUTLS_GNUTLS_H /*! * \internal * \brief Perform a TLS client handshake with a Pacemaker Remote server * * \param[in] lrmd Newly established Pacemaker Remote executor connection * * \return Standard Pacemaker return code */ static int tls_client_handshake(lrmd_t *lrmd) { lrmd_private_t *native = lrmd->lrmd_private; int tls_rc = GNUTLS_E_SUCCESS; int rc = pcmk__tls_client_handshake(native->remote, TLS_HANDSHAKE_TIMEOUT, &tls_rc); if (rc != pcmk_rc_ok) { tls_handshake_failed(lrmd, tls_rc, rc); } return rc; } /*! * \internal * \brief Add trigger and file descriptor mainloop sources for TLS * * \param[in,out] lrmd API connection with established TLS session * \param[in] do_handshake Whether to perform executor handshake * * \return Standard Pacemaker return code */ static int add_tls_to_mainloop(lrmd_t *lrmd, bool do_handshake) { lrmd_private_t *native = lrmd->lrmd_private; int rc = pcmk_rc_ok; char *name = crm_strdup_printf("pacemaker-remote-%s:%d", native->server, native->port); struct mainloop_fd_callbacks tls_fd_callbacks = { .dispatch = lrmd_tls_dispatch, .destroy = lrmd_tls_connection_destroy, }; native->process_notify = mainloop_add_trigger(G_PRIORITY_HIGH, lrmd_tls_dispatch, lrmd); native->source = mainloop_add_fd(name, G_PRIORITY_HIGH, native->sock, lrmd, &tls_fd_callbacks); /* Async connections lose the client name provided by the API caller, so we * have to use our generated name here to perform the executor handshake. * * @TODO Keep track of the caller-provided name. Perhaps we should be using * that name in this function instead of generating one anyway. */ if (do_handshake) { rc = lrmd_handshake(lrmd, name); rc = pcmk_legacy2rc(rc); } free(name); return rc; } struct handshake_data_s { lrmd_t *lrmd; time_t start_time; int timeout_sec; }; static gboolean try_handshake_cb(gpointer user_data) { struct handshake_data_s *hs = user_data; lrmd_t *lrmd = hs->lrmd; lrmd_private_t *native = lrmd->lrmd_private; pcmk__remote_t *remote = native->remote; int rc = pcmk_rc_ok; int tls_rc = GNUTLS_E_SUCCESS; if (time(NULL) >= hs->start_time + hs->timeout_sec) { rc = ETIME; tls_handshake_failed(lrmd, GNUTLS_E_TIMEDOUT, rc); free(hs); return 0; } rc = pcmk__tls_client_try_handshake(remote, &tls_rc); if (rc == pcmk_rc_ok) { tls_handshake_succeeded(lrmd); free(hs); return 0; } else if (rc == EAGAIN) { mainloop_set_trigger(native->handshake_trigger); return 1; } else { rc = EKEYREJECTED; tls_handshake_failed(lrmd, tls_rc, rc); free(hs); return 0; } } static void lrmd_tcp_connect_cb(void *userdata, int rc, int sock) { lrmd_t *lrmd = userdata; lrmd_private_t *native = lrmd->lrmd_private; gnutls_datum_t psk_key = { NULL, 0 }; int tls_rc = GNUTLS_E_SUCCESS; native->async_timer = 0; if (rc != pcmk_rc_ok) { lrmd_tls_connection_destroy(lrmd); crm_info("Could not connect to Pacemaker Remote at %s:%d: %s " CRM_XS " rc=%d", native->server, native->port, pcmk_rc_str(rc), rc); report_async_connection_result(lrmd, pcmk_rc2legacy(rc)); return; } /* The TCP connection was successful, so establish the TLS connection. */ native->sock = sock; rc = lrmd__init_remote_key(&psk_key); if (rc != pcmk_rc_ok) { crm_info("Could not connect to Pacemaker Remote at %s:%d: %s " CRM_XS " rc=%d", native->server, native->port, pcmk_rc_str(rc), rc); lrmd_tls_connection_destroy(lrmd); report_async_connection_result(lrmd, pcmk_rc2legacy(rc)); return; } gnutls_psk_allocate_client_credentials(&native->psk_cred_c); gnutls_psk_set_client_credentials(native->psk_cred_c, DEFAULT_REMOTE_USERNAME, &psk_key, GNUTLS_PSK_KEY_RAW); gnutls_free(psk_key.data); native->remote->tls_session = pcmk__new_tls_session(sock, GNUTLS_CLIENT, GNUTLS_CRD_PSK, native->psk_cred_c); if (native->remote->tls_session == NULL) { lrmd_tls_connection_destroy(lrmd); report_async_connection_result(lrmd, -EPROTO); return; } /* If the TLS handshake immediately succeeds or fails, we can handle that * now without having to deal with mainloops and retries. Otherwise, add a * trigger to keep trying until we get a result (or it times out). */ rc = pcmk__tls_client_try_handshake(native->remote, &tls_rc); if (rc == EAGAIN) { struct handshake_data_s *hs = NULL; if (native->handshake_trigger != NULL) { return; } hs = pcmk__assert_alloc(1, sizeof(struct handshake_data_s)); hs->lrmd = lrmd; hs->start_time = time(NULL); hs->timeout_sec = TLS_HANDSHAKE_TIMEOUT; native->handshake_trigger = mainloop_add_trigger(G_PRIORITY_LOW, try_handshake_cb, hs); mainloop_set_trigger(native->handshake_trigger); } else if (rc == pcmk_rc_ok) { tls_handshake_succeeded(lrmd); } else { tls_handshake_failed(lrmd, tls_rc, rc); } } static int lrmd_tls_connect_async(lrmd_t * lrmd, int timeout /*ms */ ) { int rc; int timer_id = 0; lrmd_private_t *native = lrmd->lrmd_private; lrmd_gnutls_global_init(); native->sock = -1; rc = pcmk__connect_remote(native->server, native->port, timeout, &timer_id, &(native->sock), lrmd, lrmd_tcp_connect_cb); if (rc != pcmk_rc_ok) { crm_warn("Pacemaker Remote connection to %s:%d failed: %s " CRM_XS " rc=%d", native->server, native->port, pcmk_rc_str(rc), rc); return pcmk_rc2legacy(rc); } native->async_timer = timer_id; return pcmk_ok; } static int lrmd_tls_connect(lrmd_t * lrmd, int *fd) { int rc; lrmd_private_t *native = lrmd->lrmd_private; gnutls_datum_t psk_key = { NULL, 0 }; lrmd_gnutls_global_init(); native->sock = -1; rc = pcmk__connect_remote(native->server, native->port, 0, NULL, &(native->sock), NULL, NULL); if (rc != pcmk_rc_ok) { crm_warn("Pacemaker Remote connection to %s:%d failed: %s " CRM_XS " rc=%d", native->server, native->port, pcmk_rc_str(rc), rc); lrmd_tls_connection_destroy(lrmd); return -ENOTCONN; } rc = lrmd__init_remote_key(&psk_key); if (rc != pcmk_rc_ok) { lrmd_tls_connection_destroy(lrmd); return pcmk_rc2legacy(rc); } gnutls_psk_allocate_client_credentials(&native->psk_cred_c); gnutls_psk_set_client_credentials(native->psk_cred_c, DEFAULT_REMOTE_USERNAME, &psk_key, GNUTLS_PSK_KEY_RAW); gnutls_free(psk_key.data); native->remote->tls_session = pcmk__new_tls_session(native->sock, GNUTLS_CLIENT, GNUTLS_CRD_PSK, native->psk_cred_c); if (native->remote->tls_session == NULL) { lrmd_tls_connection_destroy(lrmd); return -EPROTO; } if (tls_client_handshake(lrmd) != pcmk_rc_ok) { return -EKEYREJECTED; } crm_info("Client TLS connection established with Pacemaker Remote server %s:%d", native->server, native->port); if (fd) { *fd = native->sock; } else { add_tls_to_mainloop(lrmd, false); } return pcmk_ok; } #endif static int lrmd_api_connect(lrmd_t * lrmd, const char *name, int *fd) { int rc = -ENOTCONN; lrmd_private_t *native = lrmd->lrmd_private; switch (native->type) { case pcmk__client_ipc: rc = lrmd_ipc_connect(lrmd, fd); break; #ifdef HAVE_GNUTLS_GNUTLS_H case pcmk__client_tls: rc = lrmd_tls_connect(lrmd, fd); break; #endif default: crm_err("Unsupported executor connection type (bug?): %d", native->type); rc = -EPROTONOSUPPORT; } if (rc == pcmk_ok) { rc = lrmd_handshake(lrmd, name); } return rc; } static int lrmd_api_connect_async(lrmd_t * lrmd, const char *name, int timeout) { int rc = pcmk_ok; lrmd_private_t *native = lrmd->lrmd_private; CRM_CHECK(native && native->callback, return -EINVAL); switch (native->type) { case pcmk__client_ipc: /* fake async connection with ipc. it should be fast * enough that we gain very little from async */ rc = lrmd_api_connect(lrmd, name, NULL); if (!rc) { report_async_connection_result(lrmd, rc); } break; #ifdef HAVE_GNUTLS_GNUTLS_H case pcmk__client_tls: rc = lrmd_tls_connect_async(lrmd, timeout); break; #endif default: crm_err("Unsupported executor connection type (bug?): %d", native->type); rc = -EPROTONOSUPPORT; } return rc; } static void lrmd_ipc_disconnect(lrmd_t * lrmd) { lrmd_private_t *native = lrmd->lrmd_private; if (native->source != NULL) { /* Attached to mainloop */ mainloop_del_ipc_client(native->source); native->source = NULL; native->ipc = NULL; } else if (native->ipc) { /* Not attached to mainloop */ crm_ipc_t *ipc = native->ipc; native->ipc = NULL; crm_ipc_close(ipc); crm_ipc_destroy(ipc); } } #ifdef HAVE_GNUTLS_GNUTLS_H static void lrmd_tls_disconnect(lrmd_t * lrmd) { lrmd_private_t *native = lrmd->lrmd_private; if (native->remote->tls_session) { gnutls_bye(*native->remote->tls_session, GNUTLS_SHUT_RDWR); gnutls_deinit(*native->remote->tls_session); gnutls_free(native->remote->tls_session); native->remote->tls_session = NULL; } if (native->async_timer) { g_source_remove(native->async_timer); native->async_timer = 0; } if (native->source != NULL) { /* Attached to mainloop */ mainloop_del_ipc_client(native->source); native->source = NULL; } else if (native->sock) { close(native->sock); native->sock = 0; } if (native->pending_notify) { g_list_free_full(native->pending_notify, lrmd_free_xml); native->pending_notify = NULL; } } #endif static int lrmd_api_disconnect(lrmd_t * lrmd) { lrmd_private_t *native = lrmd->lrmd_private; int rc = pcmk_ok; switch (native->type) { case pcmk__client_ipc: crm_debug("Disconnecting from local executor"); lrmd_ipc_disconnect(lrmd); break; #ifdef HAVE_GNUTLS_GNUTLS_H case pcmk__client_tls: crm_debug("Disconnecting from remote executor on %s", native->remote_nodename); lrmd_tls_disconnect(lrmd); break; #endif default: crm_err("Unsupported executor connection type (bug?): %d", native->type); rc = -EPROTONOSUPPORT; } free(native->token); native->token = NULL; free(native->peer_version); native->peer_version = NULL; return rc; } static int lrmd_api_register_rsc(lrmd_t * lrmd, const char *rsc_id, const char *class, const char *provider, const char *type, enum lrmd_call_options options) { int rc = pcmk_ok; xmlNode *data = NULL; if (!class || !type || !rsc_id) { return -EINVAL; } if (pcmk_is_set(pcmk_get_ra_caps(class), pcmk_ra_cap_provider) && (provider == NULL)) { return -EINVAL; } data = pcmk__xe_create(NULL, PCMK__XE_LRMD_RSC); crm_xml_add(data, PCMK__XA_LRMD_ORIGIN, __func__); crm_xml_add(data, PCMK__XA_LRMD_RSC_ID, rsc_id); crm_xml_add(data, PCMK__XA_LRMD_CLASS, class); crm_xml_add(data, PCMK__XA_LRMD_PROVIDER, provider); crm_xml_add(data, PCMK__XA_LRMD_TYPE, type); rc = lrmd_send_command(lrmd, LRMD_OP_RSC_REG, data, NULL, 0, options, TRUE); free_xml(data); return rc; } static int lrmd_api_unregister_rsc(lrmd_t * lrmd, const char *rsc_id, enum lrmd_call_options options) { int rc = pcmk_ok; xmlNode *data = pcmk__xe_create(NULL, PCMK__XE_LRMD_RSC); crm_xml_add(data, PCMK__XA_LRMD_ORIGIN, __func__); crm_xml_add(data, PCMK__XA_LRMD_RSC_ID, rsc_id); rc = lrmd_send_command(lrmd, LRMD_OP_RSC_UNREG, data, NULL, 0, options, TRUE); free_xml(data); return rc; } lrmd_rsc_info_t * lrmd_new_rsc_info(const char *rsc_id, const char *standard, const char *provider, const char *type) { lrmd_rsc_info_t *rsc_info = pcmk__assert_alloc(1, sizeof(lrmd_rsc_info_t)); rsc_info->id = pcmk__str_copy(rsc_id); rsc_info->standard = pcmk__str_copy(standard); rsc_info->provider = pcmk__str_copy(provider); rsc_info->type = pcmk__str_copy(type); return rsc_info; } lrmd_rsc_info_t * lrmd_copy_rsc_info(lrmd_rsc_info_t * rsc_info) { return lrmd_new_rsc_info(rsc_info->id, rsc_info->standard, rsc_info->provider, rsc_info->type); } void lrmd_free_rsc_info(lrmd_rsc_info_t * rsc_info) { if (!rsc_info) { return; } free(rsc_info->id); free(rsc_info->type); free(rsc_info->standard); free(rsc_info->provider); free(rsc_info); } static lrmd_rsc_info_t * lrmd_api_get_rsc_info(lrmd_t * lrmd, const char *rsc_id, enum lrmd_call_options options) { lrmd_rsc_info_t *rsc_info = NULL; xmlNode *data = pcmk__xe_create(NULL, PCMK__XE_LRMD_RSC); xmlNode *output = NULL; const char *class = NULL; const char *provider = NULL; const char *type = NULL; crm_xml_add(data, PCMK__XA_LRMD_ORIGIN, __func__); crm_xml_add(data, PCMK__XA_LRMD_RSC_ID, rsc_id); lrmd_send_command(lrmd, LRMD_OP_RSC_INFO, data, &output, 0, options, TRUE); free_xml(data); if (!output) { return NULL; } class = crm_element_value(output, PCMK__XA_LRMD_CLASS); provider = crm_element_value(output, PCMK__XA_LRMD_PROVIDER); type = crm_element_value(output, PCMK__XA_LRMD_TYPE); if (!class || !type) { free_xml(output); return NULL; } else if (pcmk_is_set(pcmk_get_ra_caps(class), pcmk_ra_cap_provider) && !provider) { free_xml(output); return NULL; } rsc_info = lrmd_new_rsc_info(rsc_id, class, provider, type); free_xml(output); return rsc_info; } void lrmd_free_op_info(lrmd_op_info_t *op_info) { if (op_info) { free(op_info->rsc_id); free(op_info->action); free(op_info->interval_ms_s); free(op_info->timeout_ms_s); free(op_info); } } static int lrmd_api_get_recurring_ops(lrmd_t *lrmd, const char *rsc_id, int timeout_ms, enum lrmd_call_options options, GList **output) { xmlNode *data = NULL; xmlNode *output_xml = NULL; int rc = pcmk_ok; if (output == NULL) { return -EINVAL; } *output = NULL; // Send request if (rsc_id) { data = pcmk__xe_create(NULL, PCMK__XE_LRMD_RSC); crm_xml_add(data, PCMK__XA_LRMD_ORIGIN, __func__); crm_xml_add(data, PCMK__XA_LRMD_RSC_ID, rsc_id); } rc = lrmd_send_command(lrmd, LRMD_OP_GET_RECURRING, data, &output_xml, timeout_ms, options, TRUE); if (data) { free_xml(data); } // Process reply if ((rc != pcmk_ok) || (output_xml == NULL)) { return rc; } for (const xmlNode *rsc_xml = pcmk__xe_first_child(output_xml, PCMK__XE_LRMD_RSC, NULL, NULL); (rsc_xml != NULL) && (rc == pcmk_ok); rsc_xml = pcmk__xe_next_same(rsc_xml)) { rsc_id = crm_element_value(rsc_xml, PCMK__XA_LRMD_RSC_ID); if (rsc_id == NULL) { crm_err("Could not parse recurring operation information from executor"); continue; } for (const xmlNode *op_xml = pcmk__xe_first_child(rsc_xml, PCMK__XE_LRMD_RSC_OP, NULL, NULL); op_xml != NULL; op_xml = pcmk__xe_next_same(op_xml)) { lrmd_op_info_t *op_info = calloc(1, sizeof(lrmd_op_info_t)); if (op_info == NULL) { rc = -ENOMEM; break; } op_info->rsc_id = strdup(rsc_id); op_info->action = crm_element_value_copy(op_xml, PCMK__XA_LRMD_RSC_ACTION); op_info->interval_ms_s = crm_element_value_copy(op_xml, PCMK__XA_LRMD_RSC_INTERVAL); op_info->timeout_ms_s = crm_element_value_copy(op_xml, PCMK__XA_LRMD_TIMEOUT); *output = g_list_prepend(*output, op_info); } } free_xml(output_xml); return rc; } static void lrmd_api_set_callback(lrmd_t * lrmd, lrmd_event_callback callback) { lrmd_private_t *native = lrmd->lrmd_private; native->callback = callback; } void lrmd_internal_set_proxy_callback(lrmd_t * lrmd, void *userdata, void (*callback)(lrmd_t *lrmd, void *userdata, xmlNode *msg)) { lrmd_private_t *native = lrmd->lrmd_private; native->proxy_callback = callback; native->proxy_callback_userdata = userdata; } void lrmd_internal_proxy_dispatch(lrmd_t *lrmd, xmlNode *msg) { lrmd_private_t *native = lrmd->lrmd_private; if (native->proxy_callback) { crm_log_xml_trace(msg, "PROXY_INBOUND"); native->proxy_callback(lrmd, native->proxy_callback_userdata, msg); } } int lrmd_internal_proxy_send(lrmd_t * lrmd, xmlNode *msg) { if (lrmd == NULL) { return -ENOTCONN; } crm_xml_add(msg, PCMK__XA_LRMD_OP, CRM_OP_IPC_FWD); crm_log_xml_trace(msg, "PROXY_OUTBOUND"); return lrmd_send_xml_no_reply(lrmd, msg); } static int stonith_get_metadata(const char *provider, const char *type, char **output) { int rc = pcmk_ok; stonith_t *stonith_api = stonith_api_new(); if (stonith_api == NULL) { crm_err("Could not get fence agent meta-data: API memory allocation failed"); return -ENOMEM; } rc = stonith_api->cmds->metadata(stonith_api, st_opt_sync_call, type, provider, output, 0); if ((rc == pcmk_ok) && (*output == NULL)) { rc = -EIO; } stonith_api->cmds->free(stonith_api); return rc; } static int lrmd_api_get_metadata(lrmd_t *lrmd, const char *standard, const char *provider, const char *type, char **output, enum lrmd_call_options options) { return lrmd->cmds->get_metadata_params(lrmd, standard, provider, type, output, options, NULL); } static int lrmd_api_get_metadata_params(lrmd_t *lrmd, const char *standard, const char *provider, const char *type, char **output, enum lrmd_call_options options, lrmd_key_value_t *params) { svc_action_t *action = NULL; GHashTable *params_table = NULL; if (!standard || !type) { lrmd_key_value_freeall(params); return -EINVAL; } if (pcmk__str_eq(standard, PCMK_RESOURCE_CLASS_STONITH, pcmk__str_casei)) { lrmd_key_value_freeall(params); return stonith_get_metadata(provider, type, output); } params_table = pcmk__strkey_table(free, free); for (const lrmd_key_value_t *param = params; param; param = param->next) { pcmk__insert_dup(params_table, param->key, param->value); } action = services__create_resource_action(type, standard, provider, type, PCMK_ACTION_META_DATA, 0, PCMK_DEFAULT_METADATA_TIMEOUT_MS, params_table, 0); lrmd_key_value_freeall(params); if (action == NULL) { return -ENOMEM; } if (action->rc != PCMK_OCF_UNKNOWN) { services_action_free(action); return -EINVAL; } if (!services_action_sync(action)) { crm_err("Failed to retrieve meta-data for %s:%s:%s", standard, provider, type); services_action_free(action); return -EIO; } if (!action->stdout_data) { crm_err("Failed to receive meta-data for %s:%s:%s", standard, provider, type); services_action_free(action); return -EIO; } *output = strdup(action->stdout_data); services_action_free(action); return pcmk_ok; } static int lrmd_api_exec(lrmd_t *lrmd, const char *rsc_id, const char *action, const char *userdata, guint interval_ms, int timeout, /* ms */ int start_delay, /* ms */ enum lrmd_call_options options, lrmd_key_value_t * params) { int rc = pcmk_ok; xmlNode *data = pcmk__xe_create(NULL, PCMK__XE_LRMD_RSC); xmlNode *args = pcmk__xe_create(data, PCMK__XE_ATTRIBUTES); lrmd_key_value_t *tmp = NULL; crm_xml_add(data, PCMK__XA_LRMD_ORIGIN, __func__); crm_xml_add(data, PCMK__XA_LRMD_RSC_ID, rsc_id); crm_xml_add(data, PCMK__XA_LRMD_RSC_ACTION, action); crm_xml_add(data, PCMK__XA_LRMD_RSC_USERDATA_STR, userdata); crm_xml_add_ms(data, PCMK__XA_LRMD_RSC_INTERVAL, interval_ms); crm_xml_add_int(data, PCMK__XA_LRMD_TIMEOUT, timeout); crm_xml_add_int(data, PCMK__XA_LRMD_RSC_START_DELAY, start_delay); for (tmp = params; tmp; tmp = tmp->next) { hash2smartfield((gpointer) tmp->key, (gpointer) tmp->value, args); } rc = lrmd_send_command(lrmd, LRMD_OP_RSC_EXEC, data, NULL, timeout, options, TRUE); free_xml(data); lrmd_key_value_freeall(params); return rc; } /* timeout is in ms */ static int lrmd_api_exec_alert(lrmd_t *lrmd, const char *alert_id, const char *alert_path, int timeout, lrmd_key_value_t *params) { int rc = pcmk_ok; xmlNode *data = pcmk__xe_create(NULL, PCMK__XE_LRMD_ALERT); xmlNode *args = pcmk__xe_create(data, PCMK__XE_ATTRIBUTES); lrmd_key_value_t *tmp = NULL; crm_xml_add(data, PCMK__XA_LRMD_ORIGIN, __func__); crm_xml_add(data, PCMK__XA_LRMD_ALERT_ID, alert_id); crm_xml_add(data, PCMK__XA_LRMD_ALERT_PATH, alert_path); crm_xml_add_int(data, PCMK__XA_LRMD_TIMEOUT, timeout); for (tmp = params; tmp; tmp = tmp->next) { hash2smartfield((gpointer) tmp->key, (gpointer) tmp->value, args); } rc = lrmd_send_command(lrmd, LRMD_OP_ALERT_EXEC, data, NULL, timeout, lrmd_opt_notify_orig_only, TRUE); free_xml(data); lrmd_key_value_freeall(params); return rc; } static int lrmd_api_cancel(lrmd_t *lrmd, const char *rsc_id, const char *action, guint interval_ms) { int rc = pcmk_ok; xmlNode *data = pcmk__xe_create(NULL, PCMK__XE_LRMD_RSC); crm_xml_add(data, PCMK__XA_LRMD_ORIGIN, __func__); crm_xml_add(data, PCMK__XA_LRMD_RSC_ACTION, action); crm_xml_add(data, PCMK__XA_LRMD_RSC_ID, rsc_id); crm_xml_add_ms(data, PCMK__XA_LRMD_RSC_INTERVAL, interval_ms); rc = lrmd_send_command(lrmd, LRMD_OP_RSC_CANCEL, data, NULL, 0, 0, TRUE); free_xml(data); return rc; } static int list_stonith_agents(lrmd_list_t ** resources) { int rc = 0; stonith_t *stonith_api = stonith_api_new(); stonith_key_value_t *stonith_resources = NULL; stonith_key_value_t *dIter = NULL; if (stonith_api == NULL) { crm_err("Could not list fence agents: API memory allocation failed"); return -ENOMEM; } stonith_api->cmds->list_agents(stonith_api, st_opt_sync_call, NULL, &stonith_resources, 0); stonith_api->cmds->free(stonith_api); for (dIter = stonith_resources; dIter; dIter = dIter->next) { rc++; if (resources) { *resources = lrmd_list_add(*resources, dIter->value); } } stonith_key_value_freeall(stonith_resources, 1, 0); return rc; } static int lrmd_api_list_agents(lrmd_t * lrmd, lrmd_list_t ** resources, const char *class, const char *provider) { int rc = 0; int stonith_count = 0; // Initially, whether to include stonith devices if (pcmk__str_eq(class, PCMK_RESOURCE_CLASS_STONITH, pcmk__str_casei)) { stonith_count = 1; } else { GList *gIter = NULL; GList *agents = resources_list_agents(class, provider); for (gIter = agents; gIter != NULL; gIter = gIter->next) { *resources = lrmd_list_add(*resources, (const char *)gIter->data); rc++; } g_list_free_full(agents, free); if (!class) { stonith_count = 1; } } if (stonith_count) { // Now, if stonith devices are included, how many there are stonith_count = list_stonith_agents(resources); if (stonith_count > 0) { rc += stonith_count; } } if (rc == 0) { crm_notice("No agents found for class %s", class); rc = -EPROTONOSUPPORT; } return rc; } static bool does_provider_have_agent(const char *agent, const char *provider, const char *class) { bool found = false; GList *agents = NULL; GList *gIter2 = NULL; agents = resources_list_agents(class, provider); for (gIter2 = agents; gIter2 != NULL; gIter2 = gIter2->next) { if (pcmk__str_eq(agent, gIter2->data, pcmk__str_casei)) { found = true; } } g_list_free_full(agents, free); return found; } static int lrmd_api_list_ocf_providers(lrmd_t * lrmd, const char *agent, lrmd_list_t ** providers) { int rc = pcmk_ok; char *provider = NULL; GList *ocf_providers = NULL; GList *gIter = NULL; ocf_providers = resources_list_providers(PCMK_RESOURCE_CLASS_OCF); for (gIter = ocf_providers; gIter != NULL; gIter = gIter->next) { provider = gIter->data; if (!agent || does_provider_have_agent(agent, provider, PCMK_RESOURCE_CLASS_OCF)) { *providers = lrmd_list_add(*providers, (const char *)gIter->data); rc++; } } g_list_free_full(ocf_providers, free); return rc; } static int lrmd_api_list_standards(lrmd_t * lrmd, lrmd_list_t ** supported) { int rc = 0; GList *standards = NULL; GList *gIter = NULL; standards = resources_list_standards(); for (gIter = standards; gIter != NULL; gIter = gIter->next) { *supported = lrmd_list_add(*supported, (const char *)gIter->data); rc++; } if (list_stonith_agents(NULL) > 0) { *supported = lrmd_list_add(*supported, PCMK_RESOURCE_CLASS_STONITH); rc++; } g_list_free_full(standards, free); return rc; } /*! * \internal * \brief Create an executor API object * * \param[out] api Will be set to newly created API object (it is the * caller's responsibility to free this value with * lrmd_api_delete() if this function succeeds) * \param[in] nodename If the object will be used for a remote connection, * the node name to use in cluster for remote executor * \param[in] server If the object will be used for a remote connection, * the resolvable host name to connect to * \param[in] port If the object will be used for a remote connection, * port number on \p server to connect to * * \return Standard Pacemaker return code * \note If the caller leaves one of \p nodename or \p server NULL, the other's * value will be used for both. If the caller leaves both NULL, an API * object will be created for a local executor connection. */ int lrmd__new(lrmd_t **api, const char *nodename, const char *server, int port) { lrmd_private_t *pvt = NULL; if (api == NULL) { return EINVAL; } *api = NULL; // Allocate all memory needed *api = calloc(1, sizeof(lrmd_t)); if (*api == NULL) { return ENOMEM; } pvt = calloc(1, sizeof(lrmd_private_t)); if (pvt == NULL) { lrmd_api_delete(*api); *api = NULL; return ENOMEM; } (*api)->lrmd_private = pvt; // @TODO Do we need to do this for local connections? pvt->remote = calloc(1, sizeof(pcmk__remote_t)); (*api)->cmds = calloc(1, sizeof(lrmd_api_operations_t)); if ((pvt->remote == NULL) || ((*api)->cmds == NULL)) { lrmd_api_delete(*api); *api = NULL; return ENOMEM; } // Set methods (*api)->cmds->connect = lrmd_api_connect; (*api)->cmds->connect_async = lrmd_api_connect_async; (*api)->cmds->is_connected = lrmd_api_is_connected; (*api)->cmds->poke_connection = lrmd_api_poke_connection; (*api)->cmds->disconnect = lrmd_api_disconnect; (*api)->cmds->register_rsc = lrmd_api_register_rsc; (*api)->cmds->unregister_rsc = lrmd_api_unregister_rsc; (*api)->cmds->get_rsc_info = lrmd_api_get_rsc_info; (*api)->cmds->get_recurring_ops = lrmd_api_get_recurring_ops; (*api)->cmds->set_callback = lrmd_api_set_callback; (*api)->cmds->get_metadata = lrmd_api_get_metadata; (*api)->cmds->exec = lrmd_api_exec; (*api)->cmds->cancel = lrmd_api_cancel; (*api)->cmds->list_agents = lrmd_api_list_agents; (*api)->cmds->list_ocf_providers = lrmd_api_list_ocf_providers; (*api)->cmds->list_standards = lrmd_api_list_standards; (*api)->cmds->exec_alert = lrmd_api_exec_alert; (*api)->cmds->get_metadata_params = lrmd_api_get_metadata_params; if ((nodename == NULL) && (server == NULL)) { pvt->type = pcmk__client_ipc; } else { #ifdef HAVE_GNUTLS_GNUTLS_H if (nodename == NULL) { nodename = server; } else if (server == NULL) { server = nodename; } pvt->type = pcmk__client_tls; pvt->remote_nodename = strdup(nodename); pvt->server = strdup(server); if ((pvt->remote_nodename == NULL) || (pvt->server == NULL)) { lrmd_api_delete(*api); *api = NULL; return ENOMEM; } pvt->port = port; if (pvt->port == 0) { pvt->port = crm_default_remote_port(); } #else crm_err("Cannot communicate with Pacemaker Remote " "because GnuTLS is not enabled for this build"); lrmd_api_delete(*api); *api = NULL; return EOPNOTSUPP; #endif } return pcmk_rc_ok; } lrmd_t * lrmd_api_new(void) { lrmd_t *api = NULL; CRM_ASSERT(lrmd__new(&api, NULL, NULL, 0) == pcmk_rc_ok); return api; } lrmd_t * lrmd_remote_api_new(const char *nodename, const char *server, int port) { lrmd_t *api = NULL; CRM_ASSERT(lrmd__new(&api, nodename, server, port) == pcmk_rc_ok); return api; } void lrmd_api_delete(lrmd_t * lrmd) { if (lrmd == NULL) { return; } if (lrmd->cmds != NULL) { // Never NULL, but make static analysis happy if (lrmd->cmds->disconnect != NULL) { // Also never really NULL lrmd->cmds->disconnect(lrmd); // No-op if already disconnected } free(lrmd->cmds); } if (lrmd->lrmd_private != NULL) { lrmd_private_t *native = lrmd->lrmd_private; #ifdef HAVE_GNUTLS_GNUTLS_H free(native->server); #endif free(native->remote_nodename); free(native->remote); free(native->token); free(native->peer_version); free(lrmd->lrmd_private); } free(lrmd); } struct metadata_cb { void (*callback)(int pid, const pcmk__action_result_t *result, void *user_data); void *user_data; }; /*! * \internal * \brief Process asynchronous metadata completion * * \param[in,out] action Metadata action that completed */ static void metadata_complete(svc_action_t *action) { struct metadata_cb *metadata_cb = (struct metadata_cb *) action->cb_data; pcmk__action_result_t result = PCMK__UNKNOWN_RESULT; pcmk__set_result(&result, action->rc, action->status, services__exit_reason(action)); pcmk__set_result_output(&result, action->stdout_data, action->stderr_data); metadata_cb->callback(0, &result, metadata_cb->user_data); result.action_stdout = NULL; // Prevent free, because action owns it result.action_stderr = NULL; // Prevent free, because action owns it pcmk__reset_result(&result); free(metadata_cb); } /*! * \internal * \brief Retrieve agent metadata asynchronously * * \param[in] rsc Resource agent specification * \param[in] callback Function to call with result (this will always be * called, whether by this function directly or later * via the main loop, and on success the metadata will * be in its result argument's action_stdout) * \param[in,out] user_data User data to pass to callback * * \return Standard Pacemaker return code * \note This function is not a lrmd_api_operations_t method because it does not * need an lrmd_t object and does not go through the executor, but * executes the agent directly. */ int lrmd__metadata_async(const lrmd_rsc_info_t *rsc, void (*callback)(int pid, const pcmk__action_result_t *result, void *user_data), void *user_data) { svc_action_t *action = NULL; struct metadata_cb *metadata_cb = NULL; pcmk__action_result_t result = PCMK__UNKNOWN_RESULT; CRM_CHECK(callback != NULL, return EINVAL); if ((rsc == NULL) || (rsc->standard == NULL) || (rsc->type == NULL)) { pcmk__set_result(&result, PCMK_OCF_NOT_CONFIGURED, PCMK_EXEC_ERROR_FATAL, "Invalid resource specification"); callback(0, &result, user_data); pcmk__reset_result(&result); return EINVAL; } if (strcmp(rsc->standard, PCMK_RESOURCE_CLASS_STONITH) == 0) { return stonith__metadata_async(rsc->type, PCMK_DEFAULT_METADATA_TIMEOUT_MS / 1000, callback, user_data); } action = services__create_resource_action(pcmk__s(rsc->id, rsc->type), rsc->standard, rsc->provider, rsc->type, PCMK_ACTION_META_DATA, 0, PCMK_DEFAULT_METADATA_TIMEOUT_MS, NULL, 0); if (action == NULL) { pcmk__set_result(&result, PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_ERROR, "Out of memory"); callback(0, &result, user_data); pcmk__reset_result(&result); return ENOMEM; } if (action->rc != PCMK_OCF_UNKNOWN) { pcmk__set_result(&result, action->rc, action->status, services__exit_reason(action)); callback(0, &result, user_data); pcmk__reset_result(&result); services_action_free(action); return EINVAL; } action->cb_data = calloc(1, sizeof(struct metadata_cb)); if (action->cb_data == NULL) { services_action_free(action); pcmk__set_result(&result, PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_ERROR, "Out of memory"); callback(0, &result, user_data); pcmk__reset_result(&result); return ENOMEM; } metadata_cb = (struct metadata_cb *) action->cb_data; metadata_cb->callback = callback; metadata_cb->user_data = user_data; if (!services_action_async(action, metadata_complete)) { services_action_free(action); return pcmk_rc_error; // @TODO Derive from action->rc and ->status } // The services library has taken responsibility for action return pcmk_rc_ok; } /*! * \internal * \brief Set the result of an executor event * * \param[in,out] event Executor event to set * \param[in] rc OCF exit status of event * \param[in] op_status Executor status of event * \param[in] exit_reason Human-friendly description of event */ void lrmd__set_result(lrmd_event_data_t *event, enum ocf_exitcode rc, int op_status, const char *exit_reason) { if (event == NULL) { return; } event->rc = rc; event->op_status = op_status; // lrmd_event_data_t has (const char *) members that lrmd_free_event() frees pcmk__str_update((char **) &event->exit_reason, exit_reason); } /*! * \internal * \brief Clear an executor event's exit reason, output, and error output * * \param[in,out] event Executor event to reset */ void lrmd__reset_result(lrmd_event_data_t *event) { if (event == NULL) { return; } free((void *) event->exit_reason); event->exit_reason = NULL; free((void *) event->output); event->output = NULL; } /*! * \internal * \brief Get the uptime of a remote resource connection * * When the cluster connects to a remote resource, part of that resource's * handshake includes the uptime of the remote resource's connection. This * uptime is stored in the lrmd_t object. * * \return The connection's uptime, or -1 if unknown */ time_t lrmd__uptime(lrmd_t *lrmd) { lrmd_private_t *native = lrmd->lrmd_private; if (native->remote == NULL) { return -1; } else { return native->remote->uptime; } } const char * lrmd__node_start_state(lrmd_t *lrmd) { lrmd_private_t *native = lrmd->lrmd_private; if (native->remote == NULL) { return NULL; } else { return native->remote->start_state; } } diff --git a/lib/pacemaker/pcmk_graph_consumer.c b/lib/pacemaker/pcmk_graph_consumer.c index e1449bac33..aa3f5460b3 100644 --- a/lib/pacemaker/pcmk_graph_consumer.c +++ b/lib/pacemaker/pcmk_graph_consumer.c @@ -1,886 +1,887 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include /* * Functions for freeing transition graph objects */ /*! * \internal * \brief Free a transition graph action object * * \param[in,out] user_data Action to free */ static void free_graph_action(gpointer user_data) { pcmk__graph_action_t *action = user_data; if (action->timer != 0) { crm_warn("Cancelling timer for graph action %d", action->id); g_source_remove(action->timer); } if (action->params != NULL) { g_hash_table_destroy(action->params); } free_xml(action->xml); free(action); } /*! * \internal * \brief Free a transition graph synapse object * * \param[in,out] user_data Synapse to free */ static void free_graph_synapse(gpointer user_data) { pcmk__graph_synapse_t *synapse = user_data; g_list_free_full(synapse->actions, free_graph_action); g_list_free_full(synapse->inputs, free_graph_action); free(synapse); } /*! * \internal * \brief Free a transition graph object * * \param[in,out] graph Transition graph to free */ void pcmk__free_graph(pcmk__graph_t *graph) { if (graph != NULL) { g_list_free_full(graph->synapses, free_graph_synapse); free(graph->source); free(graph->failed_stop_offset); free(graph->failed_start_offset); free(graph); } } /* * Functions for updating graph */ /*! * \internal * \brief Update synapse after completed prerequisite * * A synapse is ready to be executed once all its prerequisite actions (inputs) * complete. Given a completed action, check whether it is an input for a given * synapse, and if so, mark the input as confirmed, and mark the synapse as * ready if appropriate. * * \param[in,out] synapse Transition graph synapse to update * \param[in] action_id ID of an action that completed * * \note The only substantial effect here is confirming synapse inputs. * should_fire_synapse() will recalculate pcmk__synapse_ready, so the only * thing that uses the pcmk__synapse_ready from here is * synapse_state_str(). */ static void update_synapse_ready(pcmk__graph_synapse_t *synapse, int action_id) { if (pcmk_is_set(synapse->flags, pcmk__synapse_ready)) { return; // All inputs have already been confirmed } // Presume ready until proven otherwise pcmk__set_synapse_flags(synapse, pcmk__synapse_ready); for (GList *lpc = synapse->inputs; lpc != NULL; lpc = lpc->next) { pcmk__graph_action_t *prereq = (pcmk__graph_action_t *) lpc->data; if (prereq->id == action_id) { crm_trace("Confirming input %d of synapse %d", action_id, synapse->id); pcmk__set_graph_action_flags(prereq, pcmk__graph_action_confirmed); } else if (!pcmk_is_set(prereq->flags, pcmk__graph_action_confirmed)) { pcmk__clear_synapse_flags(synapse, pcmk__synapse_ready); crm_trace("Synapse %d still not ready after action %d", synapse->id, action_id); } } if (pcmk_is_set(synapse->flags, pcmk__synapse_ready)) { crm_trace("Synapse %d is now ready to execute", synapse->id); } } /*! * \internal * \brief Update action and synapse confirmation after action completion * * \param[in,out] synapse Transition graph synapse that action belongs to * \param[in] action_id ID of action that completed */ static void update_synapse_confirmed(pcmk__graph_synapse_t *synapse, int action_id) { bool all_confirmed = true; for (GList *lpc = synapse->actions; lpc != NULL; lpc = lpc->next) { pcmk__graph_action_t *action = (pcmk__graph_action_t *) lpc->data; if (action->id == action_id) { crm_trace("Confirmed action %d of synapse %d", action_id, synapse->id); pcmk__set_graph_action_flags(action, pcmk__graph_action_confirmed); } else if (all_confirmed && !pcmk_is_set(action->flags, pcmk__graph_action_confirmed)) { all_confirmed = false; crm_trace("Synapse %d still not confirmed after action %d", synapse->id, action_id); } } if (all_confirmed && !pcmk_is_set(synapse->flags, pcmk__synapse_confirmed)) { crm_trace("Confirmed synapse %d", synapse->id); pcmk__set_synapse_flags(synapse, pcmk__synapse_confirmed); } } /*! * \internal * \brief Update the transition graph with a completed action result * * \param[in,out] graph Transition graph to update * \param[in] action Action that completed */ void pcmk__update_graph(pcmk__graph_t *graph, const pcmk__graph_action_t *action) { for (GList *lpc = graph->synapses; lpc != NULL; lpc = lpc->next) { pcmk__graph_synapse_t *synapse = (pcmk__graph_synapse_t *) lpc->data; if (pcmk_any_flags_set(synapse->flags, pcmk__synapse_confirmed|pcmk__synapse_failed)) { continue; // This synapse already completed } else if (pcmk_is_set(synapse->flags, pcmk__synapse_executed)) { update_synapse_confirmed(synapse, action->id); } else if (!pcmk_is_set(action->flags, pcmk__graph_action_failed) || (synapse->priority == PCMK_SCORE_INFINITY)) { update_synapse_ready(synapse, action->id); } } } /* * Functions for executing graph */ /* A transition graph consists of various types of actions. The library caller * registers execution functions for each action type, which will be stored * here. */ static pcmk__graph_functions_t *graph_fns = NULL; /*! * \internal * \brief Set transition graph execution functions * * \param[in] Execution functions to use */ void pcmk__set_graph_functions(pcmk__graph_functions_t *fns) { crm_debug("Setting custom functions for executing transition graphs"); graph_fns = fns; CRM_ASSERT(graph_fns != NULL); CRM_ASSERT(graph_fns->rsc != NULL); CRM_ASSERT(graph_fns->cluster != NULL); CRM_ASSERT(graph_fns->pseudo != NULL); CRM_ASSERT(graph_fns->fence != NULL); } /*! * \internal * \brief Check whether a graph synapse is ready to be executed * * \param[in,out] graph Transition graph that synapse is part of * \param[in,out] synapse Synapse to check * * \return true if synapse is ready, false otherwise */ static bool should_fire_synapse(pcmk__graph_t *graph, pcmk__graph_synapse_t *synapse) { GList *lpc = NULL; pcmk__set_synapse_flags(synapse, pcmk__synapse_ready); for (lpc = synapse->inputs; lpc != NULL; lpc = lpc->next) { pcmk__graph_action_t *prereq = (pcmk__graph_action_t *) lpc->data; if (!(pcmk_is_set(prereq->flags, pcmk__graph_action_confirmed))) { crm_trace("Input %d for synapse %d not yet confirmed", prereq->id, synapse->id); pcmk__clear_synapse_flags(synapse, pcmk__synapse_ready); break; } else if (pcmk_is_set(prereq->flags, pcmk__graph_action_failed) && !pcmk_is_set(prereq->flags, pcmk__graph_action_can_fail)) { crm_trace("Input %d for synapse %d confirmed but failed", prereq->id, synapse->id); pcmk__clear_synapse_flags(synapse, pcmk__synapse_ready); break; } } if (pcmk_is_set(synapse->flags, pcmk__synapse_ready)) { crm_trace("Synapse %d is ready to execute", synapse->id); } else { return false; } for (lpc = synapse->actions; lpc != NULL; lpc = lpc->next) { pcmk__graph_action_t *a = (pcmk__graph_action_t *) lpc->data; if (a->type == pcmk__pseudo_graph_action) { /* None of the below applies to pseudo ops */ } else if (synapse->priority < graph->abort_priority) { crm_trace("Skipping synapse %d: priority %d is less than " "abort priority %d", synapse->id, synapse->priority, graph->abort_priority); graph->skipped++; return false; } else if (graph_fns->allowed && !(graph_fns->allowed(graph, a))) { crm_trace("Deferring synapse %d: not allowed", synapse->id); return false; } } return true; } /*! * \internal * \brief Initiate an action from a transition graph * * \param[in,out] graph Transition graph containing action * \param[in,out] action Action to execute * * \return Standard Pacemaker return code */ static int initiate_action(pcmk__graph_t *graph, pcmk__graph_action_t *action) { const char *id = pcmk__xe_id(action->xml); CRM_CHECK(id != NULL, return EINVAL); CRM_CHECK(!pcmk_is_set(action->flags, pcmk__graph_action_executed), return pcmk_rc_already); pcmk__set_graph_action_flags(action, pcmk__graph_action_executed); switch (action->type) { case pcmk__pseudo_graph_action: crm_trace("Executing pseudo-action %d (%s)", action->id, id); return graph_fns->pseudo(graph, action); case pcmk__rsc_graph_action: crm_trace("Executing resource action %d (%s)", action->id, id); return graph_fns->rsc(graph, action); case pcmk__cluster_graph_action: if (pcmk__str_eq(crm_element_value(action->xml, PCMK_XA_OPERATION), PCMK_ACTION_STONITH, pcmk__str_none)) { crm_trace("Executing fencing action %d (%s)", action->id, id); return graph_fns->fence(graph, action); } crm_trace("Executing cluster action %d (%s)", action->id, id); return graph_fns->cluster(graph, action); default: crm_err("Unsupported graph action type <%s " PCMK_XA_ID "='%s'> " "(bug?)", action->xml->name, id); return EINVAL; } } /*! * \internal * \brief Execute a graph synapse * * \param[in,out] graph Transition graph with synapse to execute * \param[in,out] synapse Synapse to execute * * \return Standard Pacemaker return value */ static int fire_synapse(pcmk__graph_t *graph, pcmk__graph_synapse_t *synapse) { pcmk__set_synapse_flags(synapse, pcmk__synapse_executed); for (GList *lpc = synapse->actions; lpc != NULL; lpc = lpc->next) { pcmk__graph_action_t *action = (pcmk__graph_action_t *) lpc->data; int rc = initiate_action(graph, action); if (rc != pcmk_rc_ok) { crm_err("Failed initiating <%s " PCMK_XA_ID "=%d> in synapse %d: " "%s", action->xml->name, action->id, synapse->id, pcmk_rc_str(rc)); pcmk__set_synapse_flags(synapse, pcmk__synapse_confirmed); pcmk__set_graph_action_flags(action, pcmk__graph_action_confirmed |pcmk__graph_action_failed); return pcmk_rc_error; } } return pcmk_rc_ok; } /*! * \internal * \brief Dummy graph method that can be used with simulations * * \param[in,out] graph Transition graph containing action * \param[in,out] action Graph action to be initiated * * \return Standard Pacemaker return code * \note If the PE_fail environment variable is set to the action ID, * then the graph action will be marked as failed. */ static int pseudo_action_dummy(pcmk__graph_t *graph, pcmk__graph_action_t *action) { static int fail = -1; if (fail < 0) { long long fail_ll; if ((pcmk__scan_ll(getenv("PE_fail"), &fail_ll, 0LL) == pcmk_rc_ok) && (fail_ll > 0LL) && (fail_ll <= INT_MAX)) { fail = (int) fail_ll; } else { fail = 0; } } if (action->id == fail) { crm_err("Dummy event handler: pretending action %d failed", action->id); pcmk__set_graph_action_flags(action, pcmk__graph_action_failed); graph->abort_priority = PCMK_SCORE_INFINITY; } else { crm_trace("Dummy event handler: action %d initiated", action->id); } pcmk__set_graph_action_flags(action, pcmk__graph_action_confirmed); pcmk__update_graph(graph, action); return pcmk_rc_ok; } static pcmk__graph_functions_t default_fns = { pseudo_action_dummy, pseudo_action_dummy, pseudo_action_dummy, pseudo_action_dummy }; /*! * \internal * \brief Execute all actions in a transition graph * * \param[in,out] graph Transition graph to execute * * \return Status of transition after execution */ enum pcmk__graph_status pcmk__execute_graph(pcmk__graph_t *graph) { GList *lpc = NULL; int log_level = LOG_DEBUG; enum pcmk__graph_status pass_result = pcmk__graph_active; const char *status = "In progress"; if (graph_fns == NULL) { graph_fns = &default_fns; } if (graph == NULL) { return pcmk__graph_complete; } graph->fired = 0; graph->pending = 0; graph->skipped = 0; graph->completed = 0; graph->incomplete = 0; // Count completed and in-flight synapses for (lpc = graph->synapses; lpc != NULL; lpc = lpc->next) { pcmk__graph_synapse_t *synapse = (pcmk__graph_synapse_t *) lpc->data; if (pcmk_is_set(synapse->flags, pcmk__synapse_confirmed)) { graph->completed++; } else if (!pcmk_is_set(synapse->flags, pcmk__synapse_failed) && pcmk_is_set(synapse->flags, pcmk__synapse_executed)) { graph->pending++; } } crm_trace("Executing graph %d (%d synapses already completed, %d pending)", graph->id, graph->completed, graph->pending); // Execute any synapses that are ready for (lpc = graph->synapses; lpc != NULL; lpc = lpc->next) { pcmk__graph_synapse_t *synapse = (pcmk__graph_synapse_t *) lpc->data; if ((graph->batch_limit > 0) && (graph->pending >= graph->batch_limit)) { crm_debug("Throttling graph execution: batch limit (%d) reached", graph->batch_limit); break; } else if (pcmk_is_set(synapse->flags, pcmk__synapse_failed)) { graph->skipped++; continue; } else if (pcmk_any_flags_set(synapse->flags, pcmk__synapse_confirmed |pcmk__synapse_executed)) { continue; // Already handled } else if (should_fire_synapse(graph, synapse)) { graph->fired++; if (fire_synapse(graph, synapse) != pcmk_rc_ok) { crm_err("Synapse %d failed to fire", synapse->id); log_level = LOG_ERR; graph->abort_priority = PCMK_SCORE_INFINITY; graph->incomplete++; graph->fired--; } if (!(pcmk_is_set(synapse->flags, pcmk__synapse_confirmed))) { graph->pending++; } } else { crm_trace("Synapse %d cannot fire", synapse->id); graph->incomplete++; } } if ((graph->pending == 0) && (graph->fired == 0)) { graph->complete = true; if ((graph->incomplete != 0) && (graph->abort_priority <= 0)) { log_level = LOG_WARNING; pass_result = pcmk__graph_terminated; status = "Terminated"; } else if (graph->skipped != 0) { log_level = LOG_NOTICE; pass_result = pcmk__graph_complete; status = "Stopped"; } else { log_level = LOG_NOTICE; pass_result = pcmk__graph_complete; status = "Complete"; } } else if (graph->fired == 0) { pass_result = pcmk__graph_pending; } do_crm_log(log_level, "Transition %d (Complete=%d, Pending=%d," " Fired=%d, Skipped=%d, Incomplete=%d, Source=%s): %s", graph->id, graph->completed, graph->pending, graph->fired, graph->skipped, graph->incomplete, graph->source, status); return pass_result; } /* * Functions for unpacking transition graph XML into structs */ /*! * \internal * \brief Unpack a transition graph action from XML * * \param[in] parent Synapse that action is part of * \param[in] xml_action Action XML to unparse * * \return Newly allocated action on success, or NULL otherwise */ static pcmk__graph_action_t * unpack_action(pcmk__graph_synapse_t *parent, xmlNode *xml_action) { enum pcmk__graph_action_type action_type; pcmk__graph_action_t *action = NULL; const char *value = pcmk__xe_id(xml_action); if (value == NULL) { crm_err("Ignoring transition graph action without " PCMK_XA_ID " (bug?)"); crm_log_xml_trace(xml_action, "invalid"); return NULL; } if (pcmk__xe_is(xml_action, PCMK__XE_RSC_OP)) { action_type = pcmk__rsc_graph_action; } else if (pcmk__xe_is(xml_action, PCMK__XE_PSEUDO_EVENT)) { action_type = pcmk__pseudo_graph_action; } else if (pcmk__xe_is(xml_action, PCMK__XE_CRM_EVENT)) { action_type = pcmk__cluster_graph_action; } else { crm_err("Ignoring transition graph action of unknown type '%s' (bug?)", xml_action->name); crm_log_xml_trace(xml_action, "invalid"); return NULL; } action = calloc(1, sizeof(pcmk__graph_action_t)); if (action == NULL) { crm_perror(LOG_CRIT, "Cannot unpack transition graph action"); crm_log_xml_trace(xml_action, "lost"); return NULL; } pcmk__scan_min_int(value, &(action->id), -1); action->type = pcmk__rsc_graph_action; action->xml = pcmk__xml_copy(NULL, xml_action); action->synapse = parent; action->type = action_type; action->params = xml2list(action->xml); value = crm_meta_value(action->params, PCMK_META_TIMEOUT); pcmk__scan_min_int(value, &(action->timeout), 0); /* Take PCMK_META_START_DELAY into account for the timeout of the action * timer */ value = crm_meta_value(action->params, PCMK_META_START_DELAY); { int start_delay; pcmk__scan_min_int(value, &start_delay, 0); action->timeout += start_delay; } if (pcmk__guint_from_hash(action->params, CRM_META "_" PCMK_META_INTERVAL, 0, &(action->interval_ms)) != pcmk_rc_ok) { action->interval_ms = 0; } value = crm_meta_value(action->params, PCMK__META_CAN_FAIL); if (value != NULL) { int can_fail = 0; if ((crm_str_to_boolean(value, &can_fail) > 0) && (can_fail > 0)) { pcmk__set_graph_action_flags(action, pcmk__graph_action_can_fail); } else { pcmk__clear_graph_action_flags(action, pcmk__graph_action_can_fail); } if (pcmk_is_set(action->flags, pcmk__graph_action_can_fail)) { crm_warn("Support for the " PCMK__META_CAN_FAIL " meta-attribute " "is deprecated and will be removed in a future release"); } } crm_trace("Action %d has timer set to %dms", action->id, action->timeout); return action; } /*! * \internal * \brief Unpack transition graph synapse from XML * * \param[in,out] new_graph Transition graph that synapse is part of * \param[in] xml_synapse Synapse XML * * \return Newly allocated synapse on success, or NULL otherwise */ static pcmk__graph_synapse_t * unpack_synapse(pcmk__graph_t *new_graph, const xmlNode *xml_synapse) { const char *value = NULL; xmlNode *action_set = NULL; pcmk__graph_synapse_t *new_synapse = NULL; crm_trace("Unpacking synapse %s", pcmk__xe_id(xml_synapse)); new_synapse = calloc(1, sizeof(pcmk__graph_synapse_t)); if (new_synapse == NULL) { return NULL; } pcmk__scan_min_int(pcmk__xe_id(xml_synapse), &(new_synapse->id), 0); value = crm_element_value(xml_synapse, PCMK__XA_PRIORITY); pcmk__scan_min_int(value, &(new_synapse->priority), 0); CRM_CHECK(new_synapse->id >= 0, free_graph_synapse((gpointer) new_synapse); return NULL); new_graph->num_synapses++; crm_trace("Unpacking synapse %s action sets", crm_element_value(xml_synapse, PCMK_XA_ID)); for (action_set = pcmk__xe_first_child(xml_synapse, "action_set", NULL, NULL); action_set != NULL; action_set = pcmk__xe_next_same(action_set)) { for (xmlNode *action = pcmk__xe_first_child(action_set, NULL, NULL, NULL); action != NULL; action = pcmk__xe_next(action)) { pcmk__graph_action_t *new_action = unpack_action(new_synapse, action); if (new_action == NULL) { continue; } crm_trace("Adding action %d to synapse %d", new_action->id, new_synapse->id); new_graph->num_actions++; new_synapse->actions = g_list_append(new_synapse->actions, new_action); } } crm_trace("Unpacking synapse %s inputs", pcmk__xe_id(xml_synapse)); for (xmlNode *inputs = pcmk__xe_first_child(xml_synapse, "inputs", NULL, NULL); inputs != NULL; inputs = pcmk__xe_next_same(inputs)) { for (xmlNode *trigger = pcmk__xe_first_child(inputs, "trigger", NULL, NULL); trigger != NULL; trigger = pcmk__xe_next_same(trigger)) { for (xmlNode *input = pcmk__xe_first_child(trigger, NULL, NULL, NULL); input != NULL; input = pcmk__xe_next(input)) { pcmk__graph_action_t *new_input = unpack_action(new_synapse, input); if (new_input == NULL) { continue; } crm_trace("Adding input %d to synapse %d", new_input->id, new_synapse->id); new_synapse->inputs = g_list_append(new_synapse->inputs, new_input); } } } return new_synapse; } /*! * \internal * \brief Unpack transition graph XML * * \param[in] xml_graph Transition graph XML to unpack * \param[in] reference Where the XML came from (for logging) * * \return Newly allocated transition graph on success, NULL otherwise * \note The caller is responsible for freeing the return value using * pcmk__free_graph(). * \note The XML is expected to be structured like: ... ... */ pcmk__graph_t * pcmk__unpack_graph(const xmlNode *xml_graph, const char *reference) { pcmk__graph_t *new_graph = NULL; new_graph = calloc(1, sizeof(pcmk__graph_t)); if (new_graph == NULL) { return NULL; } new_graph->source = strdup(pcmk__s(reference, "unknown")); if (new_graph->source == NULL) { pcmk__free_graph(new_graph); return NULL; } new_graph->completion_action = pcmk__graph_done; // Parse top-level attributes from PCMK__XE_TRANSITION_GRAPH if (xml_graph != NULL) { const char *buf = crm_element_value(xml_graph, "transition_id"); CRM_CHECK(buf != NULL, pcmk__free_graph(new_graph); return NULL); pcmk__scan_min_int(buf, &(new_graph->id), 1); buf = crm_element_value(xml_graph, PCMK_OPT_CLUSTER_DELAY); CRM_CHECK(buf != NULL, pcmk__free_graph(new_graph); return NULL); pcmk_parse_interval_spec(buf, &(new_graph->network_delay)); buf = crm_element_value(xml_graph, PCMK_OPT_STONITH_TIMEOUT); if (buf == NULL) { new_graph->stonith_timeout = new_graph->network_delay; } else { pcmk_parse_interval_spec(buf, &(new_graph->stonith_timeout)); } // Use 0 (dynamic limit) as default/invalid, -1 (no limit) as minimum buf = crm_element_value(xml_graph, PCMK_OPT_BATCH_LIMIT); if ((buf == NULL) || (pcmk__scan_min_int(buf, &(new_graph->batch_limit), -1) != pcmk_rc_ok)) { new_graph->batch_limit = 0; } buf = crm_element_value(xml_graph, PCMK_OPT_MIGRATION_LIMIT); pcmk__scan_min_int(buf, &(new_graph->migration_limit), -1); new_graph->failed_stop_offset = crm_element_value_copy(xml_graph, "failed-stop-offset"); new_graph->failed_start_offset = crm_element_value_copy(xml_graph, "failed-start-offset"); if (crm_element_value_epoch(xml_graph, "recheck-by", &(new_graph->recheck_by)) != pcmk_ok) { new_graph->recheck_by = 0; } } // Unpack each child element for (const xmlNode *synapse_xml = pcmk__xe_first_child(xml_graph, "synapse", NULL, NULL); synapse_xml != NULL; synapse_xml = pcmk__xe_next_same(synapse_xml)) { pcmk__graph_synapse_t *new_synapse = unpack_synapse(new_graph, synapse_xml); if (new_synapse != NULL) { new_graph->synapses = g_list_append(new_graph->synapses, new_synapse); } } crm_debug("Unpacked transition %d from %s: %d actions in %d synapses", new_graph->id, new_graph->source, new_graph->num_actions, new_graph->num_synapses); return new_graph; } /* * Other transition graph utilities */ /*! * \internal * \brief Synthesize an executor event from a graph action * * \param[in] resource If not NULL, use greater call ID than in this XML * \param[in] action Graph action * \param[in] status What to use as event execution status * \param[in] rc What to use as event exit status * \param[in] exit_reason What to use as event exit reason * * \return Newly allocated executor event on success, or NULL otherwise */ lrmd_event_data_t * pcmk__event_from_graph_action(const xmlNode *resource, const pcmk__graph_action_t *action, int status, int rc, const char *exit_reason) { lrmd_event_data_t *op = NULL; GHashTableIter iter; const char *name = NULL; const char *value = NULL; xmlNode *action_resource = NULL; CRM_CHECK(action != NULL, return NULL); CRM_CHECK(action->type == pcmk__rsc_graph_action, return NULL); action_resource = pcmk__xe_first_child(action->xml, PCMK_XE_PRIMITIVE, NULL, NULL); CRM_CHECK(action_resource != NULL, crm_log_xml_warn(action->xml, "invalid"); return NULL); op = lrmd_new_event(pcmk__xe_id(action_resource), crm_element_value(action->xml, PCMK_XA_OPERATION), action->interval_ms); lrmd__set_result(op, rc, status, exit_reason); + // coverity[store_truncates_time_t] op->t_run = time(NULL); op->t_rcchange = op->t_run; op->params = pcmk__strkey_table(free, free); g_hash_table_iter_init(&iter, action->params); while (g_hash_table_iter_next(&iter, (void **)&name, (void **)&value)) { pcmk__insert_dup(op->params, name, value); } for (xmlNode *xop = pcmk__xe_first_child(resource, NULL, NULL, NULL); xop != NULL; xop = pcmk__xe_next(xop)) { int tmp = 0; crm_element_value_int(xop, PCMK__XA_CALL_ID, &tmp); crm_debug("Got call_id=%d for %s", tmp, pcmk__xe_id(resource)); if (tmp > op->call_id) { op->call_id = tmp; } } op->call_id++; return op; } diff --git a/lib/pacemaker/pcmk_injections.c b/lib/pacemaker/pcmk_injections.c index 41bdea649f..61f254761b 100644 --- a/lib/pacemaker/pcmk_injections.c +++ b/lib/pacemaker/pcmk_injections.c @@ -1,786 +1,787 @@ /* * Copyright 2009-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include // lrmd_event_data_t, etc. #include #include #include #include "libpacemaker_private.h" bool pcmk__simulate_node_config = false; #define XPATH_NODE_CONFIG "//" PCMK_XE_NODE "[@" PCMK_XA_UNAME "='%s']" #define XPATH_NODE_STATE "//" PCMK__XE_NODE_STATE "[@" PCMK_XA_UNAME "='%s']" #define XPATH_NODE_STATE_BY_ID "//" PCMK__XE_NODE_STATE "[@" PCMK_XA_ID "='%s']" #define XPATH_RSC_HISTORY XPATH_NODE_STATE \ "//" PCMK__XE_LRM_RESOURCE "[@" PCMK_XA_ID "='%s']" /*! * \internal * \brief Inject a fictitious transient node attribute into scheduler input * * \param[in,out] out Output object for displaying error messages * \param[in,out] cib_node \c PCMK__XE_NODE_STATE XML to inject attribute into * \param[in] name Transient node attribute name to inject * \param[in] value Transient node attribute value to inject */ static void inject_transient_attr(pcmk__output_t *out, xmlNode *cib_node, const char *name, const char *value) { xmlNode *attrs = NULL; xmlNode *instance_attrs = NULL; const char *node_uuid = pcmk__xe_id(cib_node); out->message(out, "inject-attr", name, value, cib_node); attrs = pcmk__xe_first_child(cib_node, PCMK__XE_TRANSIENT_ATTRIBUTES, NULL, NULL); if (attrs == NULL) { attrs = pcmk__xe_create(cib_node, PCMK__XE_TRANSIENT_ATTRIBUTES); crm_xml_add(attrs, PCMK_XA_ID, node_uuid); } instance_attrs = pcmk__xe_first_child(attrs, PCMK_XE_INSTANCE_ATTRIBUTES, NULL, NULL); if (instance_attrs == NULL) { instance_attrs = pcmk__xe_create(attrs, PCMK_XE_INSTANCE_ATTRIBUTES); crm_xml_add(instance_attrs, PCMK_XA_ID, node_uuid); } crm_create_nvpair_xml(instance_attrs, NULL, name, value); } /*! * \internal * \brief Inject a fictitious fail count into a scheduler input * * \param[in,out] out Output object for displaying error messages * \param[in,out] cib_conn CIB connection * \param[in,out] cib_node Node state XML to inject into * \param[in] resource ID of resource for fail count to inject * \param[in] task Action name for fail count to inject * \param[in] interval_ms Action interval (in milliseconds) for fail count * \param[in] exit_status Action result for fail count to inject (if * \c PCMK_OCF_OK, or \c PCMK_OCF_NOT_RUNNING when * \p interval_ms is 0, inject nothing) */ void pcmk__inject_failcount(pcmk__output_t *out, cib_t *cib_conn, xmlNode *cib_node, const char *resource, const char *task, guint interval_ms, int exit_status) { char *name = NULL; char *value = NULL; int failcount = 0; xmlNode *output = NULL; CRM_CHECK((out != NULL) && (cib_conn != NULL) && (cib_node != NULL) && (resource != NULL) && (task != NULL), return); if ((exit_status == PCMK_OCF_OK) || ((exit_status == PCMK_OCF_NOT_RUNNING) && (interval_ms == 0))) { return; } // Get current failcount and increment it name = pcmk__failcount_name(resource, task, interval_ms); if (cib__get_node_attrs(out, cib_conn, PCMK_XE_STATUS, pcmk__xe_id(cib_node), NULL, NULL, NULL, name, NULL, &output) == pcmk_rc_ok) { if (crm_element_value_int(output, PCMK_XA_VALUE, &failcount) != 0) { failcount = 0; } } value = pcmk__itoa(failcount + 1); inject_transient_attr(out, cib_node, name, value); free(name); free(value); free_xml(output); name = pcmk__lastfailure_name(resource, task, interval_ms); value = pcmk__ttoa(time(NULL)); inject_transient_attr(out, cib_node, name, value); free(name); free(value); } /*! * \internal * \brief Create a CIB configuration entry for a fictitious node * * \param[in,out] cib_conn CIB object to use * \param[in] node Node name to use */ static void create_node_entry(cib_t *cib_conn, const char *node) { int rc = pcmk_ok; char *xpath = crm_strdup_printf(XPATH_NODE_CONFIG, node); rc = cib_conn->cmds->query(cib_conn, xpath, NULL, cib_xpath|cib_sync_call|cib_scope_local); if (rc == -ENXIO) { // Only add if not already existing xmlNode *cib_object = pcmk__xe_create(NULL, PCMK_XE_NODE); crm_xml_add(cib_object, PCMK_XA_ID, node); // Use node name as ID crm_xml_add(cib_object, PCMK_XA_UNAME, node); cib_conn->cmds->create(cib_conn, PCMK_XE_NODES, cib_object, cib_sync_call|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); } /*! * \internal * \brief Synthesize a fake executor event for an action * * \param[in] cib_resource XML for any existing resource action history * \param[in] task Name of action to synthesize * \param[in] interval_ms Interval of action to synthesize * \param[in] outcome Result of action to synthesize * * \return Newly allocated executor event * \note It is the caller's responsibility to free the result with * lrmd_free_event(). */ static lrmd_event_data_t * create_op(const xmlNode *cib_resource, const char *task, guint interval_ms, int outcome) { lrmd_event_data_t *op = NULL; xmlNode *xop = NULL; op = lrmd_new_event(pcmk__xe_id(cib_resource), task, interval_ms); lrmd__set_result(op, outcome, PCMK_EXEC_DONE, "Simulated action result"); op->params = NULL; // Not needed for simulation purposes + // coverity[store_truncates_time_t] op->t_run = (unsigned int) time(NULL); op->t_rcchange = op->t_run; // Use a call ID higher than any existing history entries op->call_id = 0; for (xop = pcmk__xe_first_child(cib_resource, NULL, NULL, NULL); xop != NULL; xop = pcmk__xe_next(xop)) { int tmp = 0; crm_element_value_int(xop, PCMK__XA_CALL_ID, &tmp); if (tmp > op->call_id) { op->call_id = tmp; } } op->call_id++; return op; } /*! * \internal * \brief Inject a fictitious resource history entry into a scheduler input * * \param[in,out] cib_resource Resource history XML to inject entry into * \param[in,out] op Action result to inject * \param[in] target_rc Expected result for action to inject * * \return XML of injected resource history entry */ xmlNode * pcmk__inject_action_result(xmlNode *cib_resource, lrmd_event_data_t *op, int target_rc) { return pcmk__create_history_xml(cib_resource, op, CRM_FEATURE_SET, target_rc, NULL, crm_system_name); } /*! * \internal * \brief Inject a fictitious node into a scheduler input * * \param[in,out] cib_conn Scheduler input CIB to inject node into * \param[in] node Name of node to inject * \param[in] uuid UUID of node to inject * * \return XML of \c PCMK__XE_NODE_STATE entry for new node * \note If the global pcmk__simulate_node_config has been set to true, a * node entry in the configuration section will be added, as well as a * node state entry in the status section. */ xmlNode * pcmk__inject_node(cib_t *cib_conn, const char *node, const char *uuid) { int rc = pcmk_ok; xmlNode *cib_object = NULL; char *xpath = crm_strdup_printf(XPATH_NODE_STATE, node); bool duplicate = false; char *found_uuid = NULL; if (pcmk__simulate_node_config) { create_node_entry(cib_conn, node); } rc = cib_conn->cmds->query(cib_conn, xpath, &cib_object, cib_xpath|cib_sync_call|cib_scope_local); if ((cib_object != NULL) && (pcmk__xe_id(cib_object) == NULL)) { crm_err("Detected multiple " PCMK__XE_NODE_STATE " entries for " "xpath=%s, bailing", xpath); duplicate = true; goto done; } if (rc == -ENXIO) { if (uuid == NULL) { query_node_uuid(cib_conn, node, &found_uuid, NULL); } else { found_uuid = strdup(uuid); } if (found_uuid) { char *xpath_by_uuid = crm_strdup_printf(XPATH_NODE_STATE_BY_ID, found_uuid); /* It's possible that a PCMK__XE_NODE_STATE entry doesn't have a * PCMK_XA_UNAME yet */ rc = cib_conn->cmds->query(cib_conn, xpath_by_uuid, &cib_object, cib_xpath|cib_sync_call|cib_scope_local); if ((cib_object != NULL) && (pcmk__xe_id(cib_object) == NULL)) { crm_err("Can't inject node state for %s because multiple " "state entries found for ID %s", node, found_uuid); duplicate = true; free(xpath_by_uuid); goto done; } else if (cib_object != NULL) { crm_xml_add(cib_object, PCMK_XA_UNAME, node); rc = cib_conn->cmds->modify(cib_conn, PCMK_XE_STATUS, cib_object, cib_sync_call|cib_scope_local); } free(xpath_by_uuid); } } if (rc == -ENXIO) { cib_object = pcmk__xe_create(NULL, PCMK__XE_NODE_STATE); crm_xml_add(cib_object, PCMK_XA_ID, found_uuid); crm_xml_add(cib_object, PCMK_XA_UNAME, node); cib_conn->cmds->create(cib_conn, PCMK_XE_STATUS, cib_object, cib_sync_call|cib_scope_local); free_xml(cib_object); 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=%d)", node, rc); } done: free(found_uuid); free(xpath); if (duplicate) { crm_log_xml_warn(cib_object, "Duplicates"); crm_exit(CRM_EX_SOFTWARE); return NULL; // not reached, but makes static analysis happy } CRM_ASSERT(rc == pcmk_ok); return cib_object; } /*! * \internal * \brief Inject a fictitious node state change into a scheduler input * * \param[in,out] cib_conn Scheduler input CIB to inject into * \param[in] node Name of node to inject change for * \param[in] up If true, change state to online, otherwise offline * * \return XML of changed (or added) node state entry */ xmlNode * pcmk__inject_node_state_change(cib_t *cib_conn, const char *node, bool up) { xmlNode *cib_node = pcmk__inject_node(cib_conn, node, NULL); if (up) { pcmk__xe_set_props(cib_node, PCMK__XA_IN_CCM, PCMK_VALUE_TRUE, PCMK_XA_CRMD, PCMK_VALUE_ONLINE, PCMK__XA_JOIN, CRMD_JOINSTATE_MEMBER, PCMK_XA_EXPECTED, CRMD_JOINSTATE_MEMBER, NULL); } else { pcmk__xe_set_props(cib_node, PCMK__XA_IN_CCM, PCMK_VALUE_FALSE, PCMK_XA_CRMD, PCMK_VALUE_OFFLINE, PCMK__XA_JOIN, CRMD_JOINSTATE_DOWN, PCMK_XA_EXPECTED, CRMD_JOINSTATE_DOWN, NULL); } crm_xml_add(cib_node, PCMK_XA_CRM_DEBUG_ORIGIN, crm_system_name); return cib_node; } /*! * \internal * \brief Check whether a node has history for a given resource * * \param[in,out] cib_node Node state XML to check * \param[in] resource Resource name to check for * * \return Resource's \c PCMK__XE_LRM_RESOURCE XML entry beneath \p cib_node if * found, otherwise \c NULL */ static xmlNode * find_resource_xml(xmlNode *cib_node, const char *resource) { const char *node = crm_element_value(cib_node, PCMK_XA_UNAME); char *xpath = crm_strdup_printf(XPATH_RSC_HISTORY, node, resource); xmlNode *match = get_xpath_object(xpath, cib_node, LOG_TRACE); free(xpath); return match; } /*! * \internal * \brief Inject a resource history element into a scheduler input * * \param[in,out] out Output object for displaying error messages * \param[in,out] cib_node Node state XML to inject resource history entry into * \param[in] resource ID (in configuration) of resource to inject * \param[in] lrm_name ID as used in history (could be clone instance) * \param[in] rclass Resource agent class of resource to inject * \param[in] rtype Resource agent type of resource to inject * \param[in] rprovider Resource agent provider of resource to inject * * \return XML of injected resource history element * \note If a history element already exists under either \p resource or * \p lrm_name, this will return it rather than injecting a new one. */ xmlNode * pcmk__inject_resource_history(pcmk__output_t *out, xmlNode *cib_node, const char *resource, const char *lrm_name, const char *rclass, const char *rtype, const char *rprovider) { xmlNode *lrm = NULL; xmlNode *container = NULL; xmlNode *cib_resource = NULL; cib_resource = find_resource_xml(cib_node, resource); if (cib_resource != NULL) { /* If an existing LRM history entry uses the resource name, * continue using it, even if lrm_name is different. */ return cib_resource; } // Check for history entry under preferred name if (strcmp(resource, lrm_name) != 0) { cib_resource = find_resource_xml(cib_node, lrm_name); if (cib_resource != NULL) { return cib_resource; } } if ((rclass == NULL) || (rtype == NULL)) { // @TODO query configuration for class, provider, type out->err(out, "Resource %s not found in the status section of %s " "(supply class and type to continue)", resource, pcmk__xe_id(cib_node)); return NULL; } else if (!pcmk__strcase_any_of(rclass, PCMK_RESOURCE_CLASS_OCF, PCMK_RESOURCE_CLASS_STONITH, PCMK_RESOURCE_CLASS_SERVICE, PCMK_RESOURCE_CLASS_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)) { // @TODO query configuration for provider out->err(out, "Please specify the provider for resource %s", resource); return NULL; } crm_info("Injecting new resource %s into node state '%s'", lrm_name, pcmk__xe_id(cib_node)); lrm = pcmk__xe_first_child(cib_node, PCMK__XE_LRM, NULL, NULL); if (lrm == NULL) { const char *node_uuid = pcmk__xe_id(cib_node); lrm = pcmk__xe_create(cib_node, PCMK__XE_LRM); crm_xml_add(lrm, PCMK_XA_ID, node_uuid); } container = pcmk__xe_first_child(lrm, PCMK__XE_LRM_RESOURCES, NULL, NULL); if (container == NULL) { container = pcmk__xe_create(lrm, PCMK__XE_LRM_RESOURCES); } cib_resource = pcmk__xe_create(container, PCMK__XE_LRM_RESOURCE); // If we're creating a new entry, use the preferred name crm_xml_add(cib_resource, PCMK_XA_ID, lrm_name); crm_xml_add(cib_resource, PCMK_XA_CLASS, rclass); crm_xml_add(cib_resource, PCMK_XA_PROVIDER, rprovider); crm_xml_add(cib_resource, PCMK_XA_TYPE, rtype); return cib_resource; } /*! * \internal * \brief Inject a ticket attribute into ticket state * * \param[in,out] out Output object for displaying error messages * \param[in] ticket_id Ticket whose state should be changed * \param[in] attr_name Ticket attribute name to inject * \param[in] attr_value Boolean value of ticket attribute to inject * \param[in,out] cib CIB object to use * * \return Standard Pacemaker return code */ static int set_ticket_state_attr(pcmk__output_t *out, const char *ticket_id, const char *attr_name, bool attr_value, cib_t *cib) { int rc = pcmk_rc_ok; xmlNode *xml_top = NULL; xmlNode *ticket_state_xml = NULL; // Check for an existing ticket state entry rc = pcmk__get_ticket_state(cib, ticket_id, &ticket_state_xml); if (rc == pcmk_rc_duplicate_id) { out->err(out, "Multiple " PCMK__XE_TICKET_STATE "s match ticket_id=%s", ticket_id); rc = pcmk_rc_ok; } if (rc == pcmk_rc_ok) { // Ticket state found, use it crm_debug("Injecting attribute into existing ticket state %s", ticket_id); xml_top = ticket_state_xml; } else if (rc == ENXIO) { // No ticket state, create it xmlNode *xml_obj = NULL; xml_top = pcmk__xe_create(NULL, PCMK_XE_STATUS); xml_obj = pcmk__xe_create(xml_top, PCMK_XE_TICKETS); ticket_state_xml = pcmk__xe_create(xml_obj, PCMK__XE_TICKET_STATE); crm_xml_add(ticket_state_xml, PCMK_XA_ID, ticket_id); } else { // Error return rc; } // Add the attribute to the ticket state pcmk__xe_set_bool_attr(ticket_state_xml, attr_name, attr_value); crm_log_xml_debug(xml_top, "Update"); // Commit the change to the CIB rc = cib->cmds->modify(cib, PCMK_XE_STATUS, xml_top, cib_sync_call|cib_scope_local); rc = pcmk_legacy2rc(rc); free_xml(xml_top); return rc; } /*! * \internal * \brief Inject a fictitious action into the cluster * * \param[in,out] out Output object for displaying error messages * \param[in] spec Action specification to inject * \param[in,out] cib CIB object for scheduler input * \param[in] scheduler Scheduler data */ static void inject_action(pcmk__output_t *out, const char *spec, cib_t *cib, const pcmk_scheduler_t *scheduler) { int rc; int outcome = PCMK_OCF_OK; guint interval_ms = 0; char *key = NULL; char *node = NULL; char *task = NULL; char *resource = NULL; const char *rtype = NULL; const char *rclass = NULL; const char *rprovider = NULL; xmlNode *cib_op = NULL; xmlNode *cib_node = NULL; xmlNode *cib_resource = NULL; const pcmk_resource_t *rsc = NULL; lrmd_event_data_t *op = NULL; out->message(out, "inject-spec", spec); key = pcmk__assert_alloc(1, strlen(spec) + 1); node = pcmk__assert_alloc(1, strlen(spec) + 1); rc = sscanf(spec, "%[^@]@%[^=]=%d", key, node, &outcome); if (rc != 3) { out->err(out, "Invalid operation spec: %s. Only found %d fields", spec, rc); goto done; } parse_op_key(key, &resource, &task, &interval_ms); rsc = pe_find_resource(scheduler->resources, resource); if (rsc == NULL) { out->err(out, "Invalid resource name: %s", resource); goto done; } rclass = crm_element_value(rsc->xml, PCMK_XA_CLASS); rtype = crm_element_value(rsc->xml, PCMK_XA_TYPE); rprovider = crm_element_value(rsc->xml, PCMK_XA_PROVIDER); cib_node = pcmk__inject_node(cib, node, NULL); CRM_ASSERT(cib_node != NULL); pcmk__inject_failcount(out, cib, cib_node, resource, task, interval_ms, outcome); cib_resource = pcmk__inject_resource_history(out, 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 = pcmk__inject_action_result(cib_resource, op, 0); CRM_ASSERT(cib_op != NULL); lrmd_free_event(op); rc = cib->cmds->modify(cib, PCMK_XE_STATUS, cib_node, cib_sync_call|cib_scope_local); CRM_ASSERT(rc == pcmk_ok); done: free(task); free(node); free(key); } /*! * \internal * \brief Inject fictitious scheduler inputs * * \param[in,out] scheduler Scheduler data * \param[in,out] cib CIB object for scheduler input to modify * \param[in] injections Injections to apply */ void pcmk__inject_scheduler_input(pcmk_scheduler_t *scheduler, cib_t *cib, const pcmk_injections_t *injections) { int rc = pcmk_ok; const GList *iter = NULL; xmlNode *cib_node = NULL; pcmk__output_t *out = scheduler->priv; out->message(out, "inject-modify-config", injections->quorum, injections->watchdog); if (injections->quorum != NULL) { xmlNode *top = pcmk__xe_create(NULL, PCMK_XE_CIB); /* crm_xml_add(top, PCMK_XA_DC_UUID, dc_uuid); */ crm_xml_add(top, PCMK_XA_HAVE_QUORUM, injections->quorum); rc = cib->cmds->modify(cib, NULL, top, cib_sync_call|cib_scope_local); CRM_ASSERT(rc == pcmk_ok); } if (injections->watchdog != NULL) { rc = cib__update_node_attr(out, cib, cib_sync_call|cib_scope_local, PCMK_XE_CRM_CONFIG, NULL, NULL, NULL, NULL, PCMK_OPT_HAVE_WATCHDOG, injections->watchdog, NULL, NULL); CRM_ASSERT(rc == pcmk_rc_ok); } for (iter = injections->node_up; iter != NULL; iter = iter->next) { const char *node = (const char *) iter->data; out->message(out, "inject-modify-node", "Online", node); cib_node = pcmk__inject_node_state_change(cib, node, true); CRM_ASSERT(cib_node != NULL); rc = cib->cmds->modify(cib, PCMK_XE_STATUS, cib_node, cib_sync_call|cib_scope_local); CRM_ASSERT(rc == pcmk_ok); free_xml(cib_node); } for (iter = injections->node_down; iter != NULL; iter = iter->next) { const char *node = (const char *) iter->data; char *xpath = NULL; out->message(out, "inject-modify-node", "Offline", node); cib_node = pcmk__inject_node_state_change(cib, node, false); CRM_ASSERT(cib_node != NULL); rc = cib->cmds->modify(cib, PCMK_XE_STATUS, cib_node, cib_sync_call|cib_scope_local); CRM_ASSERT(rc == pcmk_ok); free_xml(cib_node); xpath = crm_strdup_printf("//" PCMK__XE_NODE_STATE "[@" PCMK_XA_UNAME "='%s']" "/" PCMK__XE_LRM, node); cib->cmds->remove(cib, xpath, NULL, cib_xpath|cib_sync_call|cib_scope_local); free(xpath); xpath = crm_strdup_printf("//" PCMK__XE_NODE_STATE "[@" PCMK_XA_UNAME "='%s']" "/" PCMK__XE_TRANSIENT_ATTRIBUTES, node); cib->cmds->remove(cib, xpath, NULL, cib_xpath|cib_sync_call|cib_scope_local); free(xpath); } for (iter = injections->node_fail; iter != NULL; iter = iter->next) { const char *node = (const char *) iter->data; out->message(out, "inject-modify-node", "Failing", node); cib_node = pcmk__inject_node_state_change(cib, node, true); crm_xml_add(cib_node, PCMK__XA_IN_CCM, PCMK_VALUE_FALSE); CRM_ASSERT(cib_node != NULL); rc = cib->cmds->modify(cib, PCMK_XE_STATUS, cib_node, cib_sync_call|cib_scope_local); CRM_ASSERT(rc == pcmk_ok); free_xml(cib_node); } for (iter = injections->ticket_grant; iter != NULL; iter = iter->next) { const char *ticket_id = (const char *) iter->data; out->message(out, "inject-modify-ticket", "Granting", ticket_id); rc = set_ticket_state_attr(out, ticket_id, PCMK__XA_GRANTED, true, cib); CRM_ASSERT(rc == pcmk_rc_ok); } for (iter = injections->ticket_revoke; iter != NULL; iter = iter->next) { const char *ticket_id = (const char *) iter->data; out->message(out, "inject-modify-ticket", "Revoking", ticket_id); rc = set_ticket_state_attr(out, ticket_id, PCMK__XA_GRANTED, false, cib); CRM_ASSERT(rc == pcmk_rc_ok); } for (iter = injections->ticket_standby; iter != NULL; iter = iter->next) { const char *ticket_id = (const char *) iter->data; out->message(out, "inject-modify-ticket", "Standby", ticket_id); rc = set_ticket_state_attr(out, ticket_id, PCMK_XA_STANDBY, true, cib); CRM_ASSERT(rc == pcmk_rc_ok); } for (iter = injections->ticket_activate; iter != NULL; iter = iter->next) { const char *ticket_id = (const char *) iter->data; out->message(out, "inject-modify-ticket", "Activating", ticket_id); rc = set_ticket_state_attr(out, ticket_id, PCMK_XA_STANDBY, false, cib); CRM_ASSERT(rc == pcmk_rc_ok); } for (iter = injections->op_inject; iter != NULL; iter = iter->next) { inject_action(out, (const char *) iter->data, cib, scheduler); } if (!out->is_quiet(out)) { out->end_list(out); } } void pcmk_free_injections(pcmk_injections_t *injections) { if (injections == NULL) { return; } g_list_free_full(injections->node_up, g_free); g_list_free_full(injections->node_down, g_free); g_list_free_full(injections->node_fail, g_free); g_list_free_full(injections->op_fail, g_free); g_list_free_full(injections->op_inject, g_free); g_list_free_full(injections->ticket_grant, g_free); g_list_free_full(injections->ticket_revoke, g_free); g_list_free_full(injections->ticket_standby, g_free); g_list_free_full(injections->ticket_activate, g_free); free(injections->quorum); free(injections->watchdog); free(injections); }