diff --git a/daemons/controld/controld_control.c b/daemons/controld/controld_control.c index 2fdb0189f3..c966e35022 100644 --- a/daemons/controld/controld_control.c +++ b/daemons/controld/controld_control.c @@ -1,715 +1,715 @@ /* * Copyright 2004-2025 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 qb_ipcs_service_t *ipcs = NULL; static crm_trigger_t *config_read_trigger = NULL; #if SUPPORT_COROSYNC extern gboolean crm_connect_corosync(pcmk_cluster_t *cluster); #endif static void crm_shutdown(int nsig); static gboolean crm_read_options(gpointer user_data); /* A_HA_CONNECT */ void do_ha_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 registered = FALSE; if (controld_globals.cluster == NULL) { controld_globals.cluster = pcmk_cluster_new(); } if (action & A_HA_DISCONNECT) { pcmk_cluster_disconnect(controld_globals.cluster); crm_info("Disconnected from the cluster"); controld_set_fsa_input_flags(R_HA_DISCONNECTED); } if (action & A_HA_CONNECT) { pcmk__cluster_set_status_callback(&peer_update_callback); pcmk__cluster_set_autoreap(false); #if SUPPORT_COROSYNC if (pcmk_get_cluster_layer() == pcmk_cluster_layer_corosync) { registered = crm_connect_corosync(controld_globals.cluster); } #endif // SUPPORT_COROSYNC if (registered) { pcmk__node_status_t *node = controld_get_local_node_status(); controld_election_init(); free(controld_globals.our_uuid); controld_globals.our_uuid = pcmk__str_copy(pcmk__cluster_get_xml_id(node)); if (controld_globals.our_uuid == NULL) { crm_err("Could not obtain local uuid"); registered = FALSE; } } if (!registered) { controld_set_fsa_input_flags(R_HA_DISCONNECTED); register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); return; } populate_cib_nodes(controld_node_update_none, __func__); controld_clear_fsa_input_flags(R_HA_DISCONNECTED); crm_info("Connected to the cluster"); } if (action & ~(A_HA_CONNECT | A_HA_DISCONNECT)) { crm_err("Unexpected action %s in %s", fsa_action2string(action), __func__); } } /* A_SHUTDOWN */ void do_shutdown(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) { /* just in case */ controld_set_fsa_input_flags(R_SHUTDOWN); controld_disconnect_fencer(FALSE); } /* A_SHUTDOWN_REQ */ void do_shutdown_req(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) { xmlNode *msg = NULL; controld_set_fsa_input_flags(R_SHUTDOWN); //controld_set_fsa_input_flags(R_STAYDOWN); crm_info("Sending shutdown request to all peers (DC is %s)", pcmk__s(controld_globals.dc_name, "not set")); msg = pcmk__new_request(pcmk_ipc_controld, CRM_SYSTEM_CRMD, NULL, CRM_SYSTEM_CRMD, CRM_OP_SHUTDOWN_REQ, NULL); if (!pcmk__cluster_send_message(NULL, pcmk_ipc_controld, msg)) { register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); } pcmk__xml_free(msg); } void crmd_fast_exit(crm_exit_t exit_code) { if (pcmk_is_set(controld_globals.fsa_input_register, R_STAYDOWN)) { crm_warn("Inhibiting respawn " QB_XS " remapping exit code %d to %d", exit_code, CRM_EX_FATAL); exit_code = CRM_EX_FATAL; } else if ((exit_code == CRM_EX_OK) && pcmk_is_set(controld_globals.fsa_input_register, R_IN_RECOVERY)) { crm_err("Could not recover from internal error"); exit_code = CRM_EX_ERROR; } if (controld_globals.logger_out != NULL) { controld_globals.logger_out->finish(controld_globals.logger_out, exit_code, true, NULL); pcmk__output_free(controld_globals.logger_out); controld_globals.logger_out = NULL; } crm_exit(exit_code); } crm_exit_t crmd_exit(crm_exit_t exit_code) { GMainLoop *mloop = controld_globals.mainloop; static bool in_progress = FALSE; if (in_progress && (exit_code == CRM_EX_OK)) { crm_debug("Exit is already in progress"); return exit_code; } else if(in_progress) { crm_notice("Error during shutdown process, exiting now with status %d (%s)", exit_code, crm_exit_str(exit_code)); crm_write_blackbox(SIGTRAP, NULL); crmd_fast_exit(exit_code); } in_progress = TRUE; crm_trace("Preparing to exit with status %d (%s)", exit_code, crm_exit_str(exit_code)); /* Suppress secondary errors resulting from us disconnecting everything */ controld_set_fsa_input_flags(R_HA_DISCONNECTED); /* Close all IPC servers and clients to ensure any and all shared memory files are cleaned up */ if(ipcs) { crm_trace("Closing IPC server"); mainloop_del_ipc_server(ipcs); ipcs = NULL; } controld_close_attrd_ipc(); controld_shutdown_schedulerd_ipc(); controld_disconnect_fencer(TRUE); if ((exit_code == CRM_EX_OK) && (controld_globals.mainloop == NULL)) { crm_debug("No mainloop detected"); exit_code = CRM_EX_ERROR; } /* On an error, just get out. * * Otherwise, make the effort to have mainloop exit gracefully so * that it (mostly) cleans up after itself and valgrind has less * to report on - allowing real errors stand out */ if (exit_code != CRM_EX_OK) { crm_notice("Forcing immediate exit with status %d (%s)", exit_code, crm_exit_str(exit_code)); crm_write_blackbox(SIGTRAP, NULL); crmd_fast_exit(exit_code); } /* Clean up as much memory as possible for valgrind */ controld_clear_fsa_input_flags(R_MEMBERSHIP); g_queue_free_full(controld_globals.fsa_message_queue, (GDestroyNotify) delete_fsa_input); controld_globals.fsa_message_queue = NULL; controld_free_node_pending_timers(); election_reset(controld_globals.cluster); // Stop any election timer /* Tear down the CIB manager connection, but don't free it yet -- it could * be used when we drain the mainloop later. */ controld_disconnect_cib_manager(); verify_stopped(controld_globals.fsa_state, LOG_WARNING); controld_clear_fsa_input_flags(R_LRM_CONNECTED); lrm_state_destroy_all(); mainloop_destroy_trigger(config_read_trigger); config_read_trigger = NULL; controld_destroy_fsa_trigger(); controld_destroy_transition_trigger(); pcmk__client_cleanup(); pcmk__cluster_destroy_node_caches(); controld_free_fsa_timers(); - te_cleanup_stonith_history_sync(NULL, TRUE); + controld_cleanup_fencing_history_sync(NULL, true); controld_free_sched_timer(); free(controld_globals.our_uuid); controld_globals.our_uuid = NULL; free(controld_globals.dc_name); controld_globals.dc_name = NULL; free(controld_globals.dc_version); controld_globals.dc_version = NULL; free(controld_globals.cluster_name); controld_globals.cluster_name = NULL; free(controld_globals.te_uuid); controld_globals.te_uuid = NULL; free_max_generation(); controld_destroy_failed_sync_table(); controld_destroy_outside_events_table(); mainloop_destroy_signal(SIGPIPE); mainloop_destroy_signal(SIGUSR1); mainloop_destroy_signal(SIGTERM); mainloop_destroy_signal(SIGTRAP); /* leave SIGCHLD engaged as we might still want to drain some service-actions */ if (mloop) { GMainContext *ctx = g_main_loop_get_context(controld_globals.mainloop); /* Don't re-enter this block */ controld_globals.mainloop = NULL; /* no signals on final draining anymore */ mainloop_destroy_signal(SIGCHLD); crm_trace("Draining mainloop %d %d", g_main_loop_is_running(mloop), g_main_context_pending(ctx)); { int lpc = 0; while((g_main_context_pending(ctx) && lpc < 10)) { lpc++; crm_trace("Iteration %d", lpc); g_main_context_dispatch(ctx); } } crm_trace("Closing mainloop %d %d", g_main_loop_is_running(mloop), g_main_context_pending(ctx)); g_main_loop_quit(mloop); /* Won't do anything yet, since we're inside it now */ g_main_loop_unref(mloop); } else { mainloop_destroy_signal(SIGCHLD); } cib_delete(controld_globals.cib_conn); controld_globals.cib_conn = NULL; throttle_fini(); pcmk_cluster_free(controld_globals.cluster); controld_globals.cluster = NULL; /* Graceful */ crm_trace("Done preparing for exit with status %d (%s)", exit_code, crm_exit_str(exit_code)); return exit_code; } /* A_EXIT_0, A_EXIT_1 */ void do_exit(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) { crm_exit_t exit_code = CRM_EX_OK; if (pcmk_is_set(action, A_EXIT_1)) { exit_code = CRM_EX_ERROR; crm_err("Exiting now due to errors"); } verify_stopped(cur_state, LOG_ERR); crmd_exit(exit_code); } static void sigpipe_ignore(int nsig) { return; } /* A_STARTUP */ void do_startup(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) { crm_debug("Registering Signal Handlers"); mainloop_add_signal(SIGTERM, crm_shutdown); mainloop_add_signal(SIGPIPE, sigpipe_ignore); config_read_trigger = mainloop_add_trigger(G_PRIORITY_HIGH, crm_read_options, NULL); controld_init_fsa_trigger(); controld_init_transition_trigger(); crm_debug("Creating CIB manager and executor objects"); controld_globals.cib_conn = cib_new(); lrm_state_init_local(); if (controld_init_fsa_timers() == FALSE) { register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); } } // \return libqb error code (0 on success, -errno on error) static int32_t accept_controller_client(qb_ipcs_connection_t *c, uid_t uid, gid_t gid) { crm_trace("Accepting new IPC client connection"); if (pcmk__new_client(c, uid, gid) == NULL) { return -ENOMEM; } return 0; } // \return libqb error code (0 on success, -errno on error) static int32_t dispatch_controller_ipc(qb_ipcs_connection_t * c, void *data, size_t size) { int rc = pcmk_rc_ok; uint32_t id = 0; uint32_t flags = 0; pcmk__client_t *client = pcmk__find_client(c); xmlNode *msg = NULL; rc = pcmk__ipc_msg_append(&client->buffer, data); if (rc == pcmk_rc_ipc_more) { /* We haven't read the complete message yet, so just return. */ return 0; } else if (rc == pcmk_rc_ok) { /* We've read the complete message and there's already a header on * the front. Pass it off for processing. */ msg = pcmk__client_data2xml(client, &id, &flags); g_byte_array_free(client->buffer, TRUE); client->buffer = NULL; } else { /* Some sort of error occurred reassembling the message. All we can * do is clean up, log an error and return. */ crm_err("Error when reading IPC message: %s", pcmk_rc_str(rc)); if (client->buffer != NULL) { g_byte_array_free(client->buffer, TRUE); client->buffer = NULL; } return 0; } if (msg == NULL) { pcmk__ipc_send_ack(client, id, flags, PCMK__XE_ACK, NULL, CRM_EX_PROTOCOL); return 0; } pcmk__ipc_send_ack(client, id, flags, PCMK__XE_ACK, NULL, CRM_EX_INDETERMINATE); pcmk__assert(client->user != NULL); pcmk__update_acl_user(msg, PCMK__XA_CRM_USER, client->user); pcmk__xe_set(msg, PCMK__XA_CRM_SYS_FROM, client->id); if (controld_authorize_ipc_message(msg, client, NULL)) { crm_trace("Processing IPC message from client %s", pcmk__client_name(client)); route_message(C_IPC_MESSAGE, msg); } controld_trigger_fsa(); pcmk__xml_free(msg); return 0; } static int32_t ipc_client_disconnected(qb_ipcs_connection_t *c) { pcmk__client_t *client = pcmk__find_client(c); if (client) { crm_trace("Disconnecting %sregistered client %s (%p/%p)", (client->userdata? "" : "un"), pcmk__client_name(client), c, client); free(client->userdata); pcmk__free_client(client); controld_trigger_fsa(); } return 0; } static void ipc_connection_destroyed(qb_ipcs_connection_t *c) { crm_trace("Connection %p", c); ipc_client_disconnected(c); } /* A_STOP */ void do_stop(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) { crm_trace("Closing IPC server"); mainloop_del_ipc_server(ipcs); ipcs = NULL; register_fsa_input(C_FSA_INTERNAL, I_TERMINATE, NULL); } /* A_STARTED */ void do_started(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 struct qb_ipcs_service_handlers crmd_callbacks = { .connection_accept = accept_controller_client, .connection_created = NULL, .msg_process = dispatch_controller_ipc, .connection_closed = ipc_client_disconnected, .connection_destroyed = ipc_connection_destroyed }; if (cur_state != S_STARTING) { crm_err("Start cancelled... %s", fsa_state2string(cur_state)); return; } else if (!pcmk_is_set(controld_globals.fsa_input_register, R_MEMBERSHIP)) { crm_info("Delaying start, no membership data (%.16llx)", R_MEMBERSHIP); crmd_fsa_stall(TRUE); return; } else if (!pcmk_is_set(controld_globals.fsa_input_register, R_LRM_CONNECTED)) { crm_info("Delaying start, not connected to executor (%.16llx)", R_LRM_CONNECTED); crmd_fsa_stall(TRUE); return; } else if (!pcmk_is_set(controld_globals.fsa_input_register, R_CIB_CONNECTED)) { crm_info("Delaying start, CIB not connected (%.16llx)", R_CIB_CONNECTED); crmd_fsa_stall(TRUE); return; } else if (!pcmk_is_set(controld_globals.fsa_input_register, R_READ_CONFIG)) { crm_info("Delaying start, Config not read (%.16llx)", R_READ_CONFIG); crmd_fsa_stall(TRUE); return; } else if (!pcmk_is_set(controld_globals.fsa_input_register, R_PEER_DATA)) { crm_info("Delaying start, No peer data (%.16llx)", R_PEER_DATA); crmd_fsa_stall(TRUE); return; } crm_debug("Init server comms"); ipcs = pcmk__serve_controld_ipc(&crmd_callbacks); if (ipcs == NULL) { crm_err("Failed to create IPC server: shutting down and inhibiting respawn"); register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); } else { crm_notice("Pacemaker controller successfully started and accepting connections"); } controld_set_fsa_input_flags(R_ST_REQUIRED); controld_timer_fencer_connect(GINT_TO_POINTER(TRUE)); controld_clear_fsa_input_flags(R_STARTING); register_fsa_input(msg_data->fsa_cause, I_PENDING, NULL); } /* A_RECOVER */ void do_recover(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) { controld_set_fsa_input_flags(R_IN_RECOVERY); crm_warn("Fast-tracking shutdown in response to errors"); register_fsa_input(C_FSA_INTERNAL, I_TERMINATE, NULL); } static void config_query_callback(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data) { const char *value = NULL; GHashTable *config_hash = NULL; crm_time_t *now = crm_time_new(NULL); xmlNode *crmconfig = NULL; xmlNode *alerts = NULL; pcmk_rule_input_t rule_input = { .now = now, }; if (rc != pcmk_ok) { fsa_data_t *msg_data = NULL; crm_err("Local CIB query resulted in an error: %s", pcmk_strerror(rc)); register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); if (rc == -EACCES || rc == -pcmk_err_schema_validation) { crm_err("The cluster is mis-configured - shutting down and staying down"); controld_set_fsa_input_flags(R_STAYDOWN); } goto bail; } crmconfig = output; if ((crmconfig != NULL) && !pcmk__xe_is(crmconfig, PCMK_XE_CRM_CONFIG)) { crmconfig = pcmk__xe_first_child(crmconfig, PCMK_XE_CRM_CONFIG, NULL, NULL); } if (!crmconfig) { fsa_data_t *msg_data = NULL; crm_err("Local CIB query for " PCMK_XE_CRM_CONFIG " section failed"); register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); goto bail; } crm_debug("Call %d : Parsing CIB options", call_id); config_hash = pcmk__strkey_table(free, free); pcmk_unpack_nvpair_blocks(crmconfig, PCMK_XE_CLUSTER_PROPERTY_SET, PCMK_VALUE_CIB_BOOTSTRAP_OPTIONS, &rule_input, config_hash, NULL); // Validate all options, and use defaults if not already present in hash pcmk__validate_cluster_options(config_hash); /* Validate the watchdog timeout in the context of the local node * environment. If invalid, the controller will exit with a fatal error. * * We do this via a wrapper in the controller, so that we call * pcmk__valid_fencing_watchdog_timeout() only if watchdog fencing is * enabled for the local node. Otherwise, we may exit unnecessarily. * * A validator function in libcrmcommon can't act as such a wrapper, because * it doesn't have a stonith API connection or the local node name. */ value = g_hash_table_lookup(config_hash, PCMK_OPT_FENCING_WATCHDOG_TIMEOUT); controld_valid_fencing_watchdog_timeout(value); value = g_hash_table_lookup(config_hash, PCMK_OPT_NO_QUORUM_POLICY); if (pcmk__strcase_any_of(value, PCMK_VALUE_FENCE, PCMK_VALUE_FENCE_LEGACY, NULL) && (pcmk__locate_sbd() != 0)) { controld_set_global_flags(controld_no_quorum_panic); } value = g_hash_table_lookup(config_hash, PCMK_OPT_SHUTDOWN_LOCK); if (pcmk__is_true(value)) { controld_set_global_flags(controld_shutdown_lock_enabled); } else { controld_clear_global_flags(controld_shutdown_lock_enabled); } value = g_hash_table_lookup(config_hash, PCMK_OPT_SHUTDOWN_LOCK_LIMIT); pcmk_parse_interval_spec(value, &controld_globals.shutdown_lock_limit); controld_globals.shutdown_lock_limit /= 1000; value = g_hash_table_lookup(config_hash, PCMK_OPT_NODE_PENDING_TIMEOUT); pcmk_parse_interval_spec(value, &controld_globals.node_pending_timeout); controld_globals.node_pending_timeout /= 1000; value = g_hash_table_lookup(config_hash, PCMK_OPT_CLUSTER_NAME); pcmk__str_update(&(controld_globals.cluster_name), value); // Let subcomponents initialize their own static variables controld_configure_election(config_hash); controld_configure_fencing(config_hash); controld_configure_fsa_timers(config_hash); controld_configure_throttle(config_hash); alerts = pcmk__xe_first_child(output, PCMK_XE_ALERTS, NULL, NULL); crmd_unpack_alerts(alerts); controld_set_fsa_input_flags(R_READ_CONFIG); controld_trigger_fsa(); g_hash_table_destroy(config_hash); bail: crm_time_free(now); } /*! * \internal * \brief Trigger read and processing of the configuration * * \param[in] fn Calling function name * \param[in] line Line number where call occurred */ void controld_trigger_config_as(const char *fn, int line) { if (config_read_trigger != NULL) { crm_trace("%s:%d - Triggered config processing", fn, line); mainloop_set_trigger(config_read_trigger); } } gboolean crm_read_options(gpointer user_data) { cib_t *cib_conn = controld_globals.cib_conn; int call_id = cib_conn->cmds->query(cib_conn, "//" PCMK_XE_CRM_CONFIG " | //" PCMK_XE_ALERTS, NULL, cib_xpath); fsa_register_cib_callback(call_id, NULL, config_query_callback); crm_trace("Querying the CIB... call %d", call_id); return TRUE; } /* A_READCONFIG */ void do_read_config(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) { throttle_init(); controld_trigger_config(); } static void crm_shutdown(int nsig) { const char *value = NULL; guint default_period_ms = 0; if ((controld_globals.mainloop == NULL) || !g_main_loop_is_running(controld_globals.mainloop)) { crmd_exit(CRM_EX_OK); return; } if (pcmk_is_set(controld_globals.fsa_input_register, R_SHUTDOWN)) { crm_err("Escalating shutdown"); register_fsa_input_before(C_SHUTDOWN, I_ERROR, NULL); return; } controld_set_fsa_input_flags(R_SHUTDOWN); register_fsa_input(C_SHUTDOWN, I_SHUTDOWN, NULL); /* If shutdown timer doesn't have a period set, use the default * * @TODO: Evaluate whether this is still necessary. As long as * config_query_callback() has been run at least once, it doesn't look like * anything could have changed the timer period since then. */ value = pcmk__cluster_option(NULL, PCMK_OPT_SHUTDOWN_ESCALATION); pcmk_parse_interval_spec(value, &default_period_ms); controld_shutdown_start_countdown(default_period_ms); } diff --git a/daemons/controld/controld_fencing.c b/daemons/controld/controld_fencing.c index 50de6a800a..ebea24a135 100644 --- a/daemons/controld/controld_fencing.c +++ b/daemons/controld/controld_fencing.c @@ -1,1110 +1,1110 @@ /* * Copyright 2004-2025 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 static void tengine_stonith_history_synced(stonith_t *st, stonith_event_t *st_event); /* * stonith failure counting * * We don't want to get stuck in a permanent fencing loop. Keep track of the * number of fencing failures for each target node, and the most we'll restart a * transition for. */ struct st_fail_rec { int count; }; #define DEFAULT_FENCING_MAX_ATTEMPTS 10 static bool fence_reaction_panic = false; static unsigned long int fencing_max_attempts = DEFAULT_FENCING_MAX_ATTEMPTS; static GHashTable *stonith_failures = NULL; /*! * \internal * \brief Update max fencing attempts before giving up * * \param[in] value New max fencing attempts */ static void update_fencing_max_attempts(const char *value) { int score = 0; int rc = pcmk_parse_score(value, &score, DEFAULT_FENCING_MAX_ATTEMPTS); // The option validator ensures invalid values shouldn't be possible CRM_CHECK((rc == pcmk_rc_ok) && (score > 0), return); if (fencing_max_attempts != score) { crm_debug("Maximum fencing attempts per transition is now %d (was %lu)", score, fencing_max_attempts); } fencing_max_attempts = score; } /*! * \internal * \brief Configure reaction to notification of local node being fenced * * \param[in] reaction_s Reaction type */ static void set_fence_reaction(const char *reaction_s) { if (pcmk__str_eq(reaction_s, "panic", pcmk__str_casei)) { fence_reaction_panic = true; } else { if (!pcmk__str_eq(reaction_s, PCMK_VALUE_STOP, pcmk__str_casei)) { crm_warn("Invalid value '%s' for %s, using 'stop'", reaction_s, PCMK_OPT_FENCING_REACTION); } fence_reaction_panic = false; } } /*! * \internal * \brief Configure fencing options based on the CIB * * \param[in,out] options Name/value pairs for configured options */ void controld_configure_fencing(GHashTable *options) { const char *value = NULL; value = g_hash_table_lookup(options, PCMK_OPT_FENCING_REACTION); set_fence_reaction(value); value = g_hash_table_lookup(options, PCMK_OPT_FENCING_MAX_ATTEMPTS); update_fencing_max_attempts(value); } static gboolean too_many_st_failures(const char *target) { GHashTableIter iter; const char *key = NULL; struct st_fail_rec *value = NULL; if (stonith_failures == NULL) { return FALSE; } if (target == NULL) { g_hash_table_iter_init(&iter, stonith_failures); while (g_hash_table_iter_next(&iter, (gpointer *) &key, (gpointer *) &value)) { if (value->count >= fencing_max_attempts) { target = (const char*)key; goto too_many; } } } else { value = g_hash_table_lookup(stonith_failures, target); if ((value != NULL) && (value->count >= fencing_max_attempts)) { goto too_many; } } return FALSE; too_many: crm_warn("Too many failures (%d) to fence %s, giving up", value->count, target); return TRUE; } /*! * \internal * \brief Reset a stonith fail count * * \param[in] target Name of node to reset, or NULL for all */ void st_fail_count_reset(const char *target) { if (stonith_failures == NULL) { return; } if (target) { struct st_fail_rec *rec = NULL; rec = g_hash_table_lookup(stonith_failures, target); if (rec) { rec->count = 0; } } else { GHashTableIter iter; const char *key = NULL; struct st_fail_rec *rec = NULL; g_hash_table_iter_init(&iter, stonith_failures); while (g_hash_table_iter_next(&iter, (gpointer *) &key, (gpointer *) &rec)) { rec->count = 0; } } } static void st_fail_count_increment(const char *target) { struct st_fail_rec *rec = NULL; if (stonith_failures == NULL) { stonith_failures = pcmk__strkey_table(free, free); } rec = g_hash_table_lookup(stonith_failures, target); if (rec) { rec->count++; } else { rec = malloc(sizeof(struct st_fail_rec)); if(rec == NULL) { return; } rec->count = 1; g_hash_table_insert(stonith_failures, pcmk__str_copy(target), rec); } } /* end stonith fail count functions */ static void cib_fencing_updated(xmlNode *msg, int call_id, int rc, xmlNode *output, void *user_data) { if (rc < pcmk_ok) { crm_err("Fencing update %d for %s: failed - %s (%d)", call_id, (char *)user_data, pcmk_strerror(rc), rc); crm_log_xml_warn(msg, "Failed update"); abort_transition(PCMK_SCORE_INFINITY, pcmk__graph_shutdown, "CIB update failed", NULL); } else { crm_info("Fencing update %d for %s: complete", call_id, (char *)user_data); } } /*! * \internal * \brief Update a fencing target's node state * * \param[in] target Node that was successfully fenced * \param[in] target_xml_id CIB XML ID of target */ static void update_node_state_after_fencing(const char *target, const char *target_xml_id) { int rc = pcmk_ok; pcmk__node_status_t *peer = NULL; xmlNode *node_state = NULL; /* We (usually) rely on the membership layer to do * controld_node_update_cluster, and the peer status callback to do * controld_node_update_peer, because the node might have already rejoined * before we get the stonith result here. */ uint32_t flags = controld_node_update_join|controld_node_update_expected; CRM_CHECK((target != NULL) && (target_xml_id != NULL), return); // Ensure target is cached peer = pcmk__get_node(0, target, target_xml_id, pcmk__node_search_any); CRM_CHECK(peer != NULL, return); if (peer->state == NULL) { /* Usually, we rely on the membership layer to update the cluster state * in the CIB. However, if the node has never been seen, do it here, so * the node is not considered unclean. */ flags |= controld_node_update_cluster; } if (peer->xml_id == NULL) { crm_info("Recording XML ID '%s' for node '%s'", target_xml_id, target); peer->xml_id = pcmk__str_copy(target_xml_id); } crmd_peer_down(peer, TRUE); node_state = create_node_state_update(peer, flags, NULL, __func__); pcmk__xe_set(node_state, PCMK_XA_ID, target_xml_id); if (pcmk_is_set(peer->flags, pcmk__node_status_remote)) { char *now_s = pcmk__ttoa(time(NULL)); pcmk__xe_set(node_state, PCMK__XA_NODE_FENCED, now_s); free(now_s); } rc = controld_globals.cib_conn->cmds->modify(controld_globals.cib_conn, PCMK_XE_STATUS, node_state, cib_can_create); pcmk__xml_free(node_state); crm_debug("Updating node state for %s after fencing (call %d)", target, rc); fsa_register_cib_callback(rc, pcmk__str_copy(target), cib_fencing_updated); controld_delete_node_state(peer->name, controld_section_all, cib_none); } /*! * \internal * \brief Abort transition due to stonith failure * * \param[in] abort_action Whether to restart or stop transition * \param[in] target Don't restart if this (NULL for any) has too many failures * \param[in] reason Log this stonith action XML as abort reason (or NULL) */ static void abort_for_stonith_failure(enum pcmk__graph_next abort_action, const char *target, const xmlNode *reason) { /* If stonith repeatedly fails, we eventually give up on starting a new * transition for that reason. */ if ((abort_action != pcmk__graph_wait) && too_many_st_failures(target)) { abort_action = pcmk__graph_wait; } abort_transition(PCMK_SCORE_INFINITY, abort_action, "Stonith failed", reason); } /* * Fencing cleanup list * * If the DC is fenced, proper notifications might not go out. The fencing * cleanup list allows the cluster to (re-)send notifications once a new DC is * elected. */ static GList *fencing_cleanup_list = NULL; /*! * \internal * \brief Add a node to the fencing cleanup list * * \param[in] target Name of node to add */ static void add_fencing_cleanup(const char *target) { fencing_cleanup_list = g_list_append(fencing_cleanup_list, pcmk__str_copy(target)); } /*! * \internal * \brief Remove a node from the fencing cleanup list * * \param[in] Name of node to remove */ void controld_remove_fencing_cleanup(const char *target) { GList *iter = fencing_cleanup_list; while (iter != NULL) { GList *tmp = iter; char *iter_name = tmp->data; iter = iter->next; if (pcmk__str_eq(target, iter_name, pcmk__str_casei)) { crm_trace("Removing %s from the cleanup list", iter_name); fencing_cleanup_list = g_list_delete_link(fencing_cleanup_list, tmp); free(iter_name); } } } /*! * \internal * \brief Purge all entries from the fencing cleanup list */ void controld_purge_fencing_cleanup(void) { for (GList *iter = fencing_cleanup_list; iter != NULL; iter = iter->next) { char *target = iter->data; crm_info("Purging %s from fencing cleanup list", target); free(target); } g_list_free(fencing_cleanup_list); fencing_cleanup_list = NULL; } /*! * \internal * \brief Send fencing updates for all entries in cleanup list, then purge it */ void controld_execute_fencing_cleanup(void) { for (GList *iter = fencing_cleanup_list; iter != NULL; iter = iter->next) { char *target = iter->data; pcmk__node_status_t *target_node = pcmk__get_node(0, target, NULL, pcmk__node_search_cluster_member); const char *uuid = pcmk__cluster_get_xml_id(target_node); crm_notice("Marking %s, target of a previous fencing action, as clean", target); update_node_state_after_fencing(target, uuid); free(target); } g_list_free(fencing_cleanup_list); fencing_cleanup_list = NULL; } /* end fencing cleanup list functions */ /* stonith API client * * Functions that need to interact directly with the fencer via its API */ static stonith_t *stonith_api = NULL; static mainloop_timer_t *controld_fencer_connect_timer = NULL; static char *te_client_id = NULL; static gboolean fail_incompletable_stonith(pcmk__graph_t *graph) { GList *lpc = NULL; const char *task = NULL; xmlNode *last_action = NULL; if (graph == NULL) { return FALSE; } for (lpc = graph->synapses; lpc != NULL; lpc = lpc->next) { GList *lpc2 = NULL; pcmk__graph_synapse_t *synapse = (pcmk__graph_synapse_t *) lpc->data; if (pcmk_is_set(synapse->flags, pcmk__synapse_confirmed)) { continue; } for (lpc2 = synapse->actions; lpc2 != NULL; lpc2 = lpc2->next) { pcmk__graph_action_t *action = (pcmk__graph_action_t *) lpc2->data; if ((action->type != pcmk__cluster_graph_action) || pcmk_is_set(action->flags, pcmk__graph_action_confirmed)) { continue; } task = pcmk__xe_get(action->xml, PCMK_XA_OPERATION); if (pcmk__str_eq(task, PCMK_ACTION_STONITH, pcmk__str_casei)) { pcmk__set_graph_action_flags(action, pcmk__graph_action_failed); last_action = action->xml; pcmk__update_graph(graph, action); crm_notice("Failing action %d (%s): fencer terminated", action->id, pcmk__xe_id(action->xml)); } } } if (last_action != NULL) { crm_warn("Fencer failure resulted in unrunnable actions"); abort_for_stonith_failure(pcmk__graph_restart, NULL, last_action); return TRUE; } return FALSE; } static void tengine_stonith_connection_destroy(stonith_t *st, stonith_event_t *e) { - te_cleanup_stonith_history_sync(st, FALSE); + controld_cleanup_fencing_history_sync(st, false); if (pcmk_is_set(controld_globals.fsa_input_register, R_ST_REQUIRED)) { crm_err("Lost fencer connection (will attempt to reconnect)"); if (!mainloop_timer_running(controld_fencer_connect_timer)) { mainloop_timer_start(controld_fencer_connect_timer); } } else { crm_info("Disconnected from fencer"); } if (stonith_api) { /* the client API won't properly reconnect notifications * if they are still in the table - so remove them */ if (stonith_api->state != stonith_disconnected) { stonith_api->cmds->disconnect(st); } stonith_api->cmds->remove_notification(stonith_api, NULL); } if (AM_I_DC) { fail_incompletable_stonith(controld_globals.transition_graph); trigger_graph(); } } /*! * \internal * \brief Handle an event notification from the fencing API * * \param[in] st Fencing API connection (ignored) * \param[in] event Fencing API event notification */ static void handle_fence_notification(stonith_t *st, stonith_event_t *event) { bool succeeded = true; const char *executioner = "the cluster"; const char *client = "a client"; const char *reason = NULL; int exec_status; if (te_client_id == NULL) { te_client_id = crm_strdup_printf("%s.%lu", crm_system_name, (unsigned long) getpid()); } if (event == NULL) { crm_err("Notify data not found"); return; } if (event->executioner != NULL) { executioner = event->executioner; } if (event->client_origin != NULL) { client = event->client_origin; } exec_status = stonith__event_execution_status(event); if ((stonith__event_exit_status(event) != CRM_EX_OK) || (exec_status != PCMK_EXEC_DONE)) { succeeded = false; if (exec_status == PCMK_EXEC_DONE) { exec_status = PCMK_EXEC_ERROR; } } reason = stonith__event_exit_reason(event); crmd_alert_fencing_op(event); if (pcmk__str_eq(PCMK_ACTION_ON, event->action, pcmk__str_none)) { // Unfencing doesn't need special handling, just a log message if (succeeded) { crm_notice("%s was unfenced by %s at the request of %s@%s", event->target, executioner, client, event->origin); } else { crm_err("Unfencing of %s by %s failed (%s%s%s) with exit status %d", event->target, executioner, pcmk_exec_status_str(exec_status), ((reason == NULL)? "" : ": "), ((reason == NULL)? "" : reason), stonith__event_exit_status(event)); } return; } if (succeeded && controld_is_local_node(event->target)) { /* We were notified of our own fencing. Most likely, either fencing was * misconfigured, or fabric fencing that doesn't cut cluster * communication is in use. * * Either way, shutting down the local host is a good idea, to require * administrator intervention. Also, other nodes would otherwise likely * set our status to lost because of the fencing callback and discard * our subsequent election votes as "not part of our cluster". */ crm_crit("We were allegedly just fenced by %s for %s!", executioner, event->origin); // Dumps blackbox if enabled if (fence_reaction_panic) { pcmk__panic("Notified of own fencing"); } else { crm_exit(CRM_EX_FATAL); } return; // Should never get here } /* Update the count of fencing failures for this target, in case we become * DC later. The current DC has already updated its fail count in * tengine_stonith_callback(). */ if (!AM_I_DC) { if (succeeded) { st_fail_count_reset(event->target); } else { st_fail_count_increment(event->target); } } crm_notice("Peer %s was%s terminated (%s) by %s on behalf of %s@%s: " "%s%s%s%s " QB_XS " event=%s", event->target, (succeeded? "" : " not"), event->action, executioner, client, event->origin, (succeeded? "OK" : pcmk_exec_status_str(exec_status)), ((reason == NULL)? "" : " ("), ((reason == NULL)? "" : reason), ((reason == NULL)? "" : ")"), event->id); if (succeeded) { const uint32_t flags = pcmk__node_search_any |pcmk__node_search_cluster_cib; pcmk__node_status_t *peer = pcmk__search_node_caches(0, event->target, NULL, flags); const char *uuid = NULL; if (peer == NULL) { return; } uuid = pcmk__cluster_get_xml_id(peer); if (AM_I_DC) { /* The DC always sends updates */ update_node_state_after_fencing(event->target, uuid); /* @TODO Ideally, at this point, we'd check whether the fenced node * hosted any guest nodes, and call remote_node_down() for them. * Unfortunately, the controller doesn't have a simple, reliable way * to map hosts to guests. It might be possible to track this in the * peer cache via refresh_remote_nodes(). For now, we rely on the * scheduler creating fence pseudo-events for the guests. */ if (!pcmk__str_eq(client, te_client_id, pcmk__str_casei)) { /* Abort the current transition if it wasn't the cluster that * initiated fencing. */ crm_info("External fencing operation from %s fenced %s", client, event->target); abort_transition(PCMK_SCORE_INFINITY, pcmk__graph_restart, "External Fencing Operation", NULL); } } else if (pcmk__str_eq(controld_globals.dc_name, event->target, pcmk__str_null_matches|pcmk__str_casei) && !pcmk_is_set(peer->flags, pcmk__node_status_remote)) { // Assume the target was our DC if we don't currently have one if (controld_globals.dc_name != NULL) { crm_notice("Fencing target %s was our DC", event->target); } else { crm_notice("Fencing target %s may have been our DC", event->target); } /* Given the CIB resyncing that occurs around elections, * have one node update the CIB now and, if the new DC is different, * have them do so too after the election */ if (controld_is_local_node(event->executioner)) { update_node_state_after_fencing(event->target, uuid); } add_fencing_cleanup(event->target); } /* If the target is a remote node, and we host its connection, * immediately fail all monitors so it can be recovered quickly. * The connection won't necessarily drop when a remote node is fenced, * so the failure might not otherwise be detected until the next poke. */ if (pcmk_is_set(peer->flags, pcmk__node_status_remote)) { remote_ra_fail(event->target); } crmd_peer_down(peer, TRUE); } } /*! * \brief Connect to fencer * * \param[in] user_data If NULL, retry failures now, otherwise retry in mainloop timer * * \return G_SOURCE_REMOVE on success, G_SOURCE_CONTINUE to retry * \note If user_data is NULL, this will wait 2s between attempts, for up to * 30 attempts, meaning the controller could be blocked as long as 58s. */ gboolean controld_timer_fencer_connect(gpointer user_data) { int rc = pcmk_ok; if (stonith_api == NULL) { stonith_api = stonith__api_new(); if (stonith_api == NULL) { crm_err("Could not connect to fencer: API memory allocation failed"); return G_SOURCE_REMOVE; } } if (stonith_api->state != stonith_disconnected) { crm_trace("Already connected to fencer, no need to retry"); return G_SOURCE_REMOVE; } if (user_data == NULL) { // Blocking (retry failures now until successful) rc = stonith__api_connect_retry(stonith_api, crm_system_name, 30); if (rc != pcmk_rc_ok) { crm_err("Could not connect to fencer in 30 attempts: %s " QB_XS " rc=%d", pcmk_rc_str(rc), rc); } } else { // Non-blocking (retry failures later in main loop) rc = stonith_api->cmds->connect(stonith_api, crm_system_name, NULL); if (controld_fencer_connect_timer == NULL) { controld_fencer_connect_timer = mainloop_timer_add("controld_fencer_connect", 1000, TRUE, controld_timer_fencer_connect, GINT_TO_POINTER(TRUE)); } if (rc != pcmk_ok) { if (pcmk_is_set(controld_globals.fsa_input_register, R_ST_REQUIRED)) { crm_notice("Fencer connection failed (will retry): %s " QB_XS " rc=%d", pcmk_strerror(rc), rc); if (!mainloop_timer_running(controld_fencer_connect_timer)) { mainloop_timer_start(controld_fencer_connect_timer); } return G_SOURCE_CONTINUE; } else { crm_info("Fencer connection failed (ignoring because no longer required): %s " QB_XS " rc=%d", pcmk_strerror(rc), rc); } return G_SOURCE_REMOVE; } } if (rc == pcmk_ok) { stonith_api_operations_t *cmds = stonith_api->cmds; cmds->register_notification(stonith_api, PCMK__VALUE_ST_NOTIFY_DISCONNECT, tengine_stonith_connection_destroy); cmds->register_notification(stonith_api, PCMK__VALUE_ST_NOTIFY_FENCE, handle_fence_notification); cmds->register_notification(stonith_api, PCMK__VALUE_ST_NOTIFY_HISTORY_SYNCED, tengine_stonith_history_synced); controld_trigger_fencing_history_sync(true); crm_notice("Fencer successfully connected"); } return G_SOURCE_REMOVE; } void controld_disconnect_fencer(bool destroy) { if (stonith_api) { // Prevent fencer connection from coming up again controld_clear_fsa_input_flags(R_ST_REQUIRED); if (stonith_api->state != stonith_disconnected) { stonith_api->cmds->disconnect(stonith_api); } stonith_api->cmds->remove_notification(stonith_api, NULL); } if (destroy) { if (stonith_api) { stonith_api->cmds->free(stonith_api); stonith_api = NULL; } if (controld_fencer_connect_timer) { mainloop_timer_del(controld_fencer_connect_timer); controld_fencer_connect_timer = NULL; } if (te_client_id) { free(te_client_id); te_client_id = NULL; } } } static gboolean do_stonith_history_sync(gpointer user_data) { if (stonith_api && (stonith_api->state != stonith_disconnected)) { stonith_history_t *history = NULL; - te_cleanup_stonith_history_sync(stonith_api, FALSE); + controld_cleanup_fencing_history_sync(stonith_api, false); stonith_api->cmds->history(stonith_api, st_opt_sync_call | st_opt_broadcast, NULL, &history, 5); stonith__history_free(history); return TRUE; } else { crm_info("Skip triggering stonith history-sync as stonith is disconnected"); return FALSE; } } static void tengine_stonith_callback(stonith_t *stonith, stonith_callback_data_t *data) { char *uuid = NULL; int stonith_id = -1; int transition_id = -1; pcmk__graph_action_t *action = NULL; const char *target = NULL; if ((data == NULL) || (data->userdata == NULL)) { crm_err("Ignoring fence operation %d result: " "No transition key given (bug?)", ((data == NULL)? -1 : data->call_id)); return; } if (!AM_I_DC) { const char *reason = stonith__exit_reason(data); if (reason == NULL) { reason = pcmk_exec_status_str(stonith__execution_status(data)); } crm_notice("Result of fence operation %d: %d (%s) " QB_XS " key=%s", data->call_id, stonith__exit_status(data), reason, (const char *) data->userdata); return; } CRM_CHECK(decode_transition_key(data->userdata, &uuid, &transition_id, &stonith_id, NULL), goto bail); if (controld_globals.transition_graph->complete || (stonith_id < 0) || !pcmk__str_eq(uuid, controld_globals.te_uuid, pcmk__str_none) || (controld_globals.transition_graph->id != transition_id)) { crm_info("Ignoring fence operation %d result: " "Not from current transition " QB_XS " complete=%s action=%d uuid=%s (vs %s) transition=%d (vs %d)", data->call_id, pcmk__btoa(controld_globals.transition_graph->complete), stonith_id, uuid, controld_globals.te_uuid, transition_id, controld_globals.transition_graph->id); goto bail; } action = controld_get_action(stonith_id); if (action == NULL) { crm_err("Ignoring fence operation %d result: " "Action %d not found in transition graph (bug?) " QB_XS " uuid=%s transition=%d", data->call_id, stonith_id, uuid, transition_id); goto bail; } target = pcmk__xe_get(action->xml, PCMK__META_ON_NODE); if (target == NULL) { crm_err("Ignoring fence operation %d result: No target given (bug?)", data->call_id); goto bail; } stop_te_timer(action); if (stonith__exit_status(data) == CRM_EX_OK) { const char *uuid = pcmk__xe_get(action->xml, PCMK__META_ON_NODE_UUID); const char *op = crm_meta_value(action->params, PCMK__META_STONITH_ACTION); crm_info("Fence operation %d for %s succeeded", data->call_id, target); if (!(pcmk_is_set(action->flags, pcmk__graph_action_confirmed))) { te_action_confirmed(action, NULL); if (pcmk__str_eq(PCMK_ACTION_ON, op, pcmk__str_casei)) { const char *value = NULL; char *now = pcmk__ttoa(time(NULL)); gboolean is_remote_node = FALSE; /* This check is not 100% reliable, since this node is not * guaranteed to have the remote node cached. However, it * doesn't have to be reliable, since the attribute manager can * learn a node's "remoteness" by other means sooner or later. * This allows it to learn more quickly if this node does have * the information. */ if (g_hash_table_lookup(pcmk__remote_peer_cache, uuid) != NULL) { is_remote_node = TRUE; } update_attrd(target, CRM_ATTR_UNFENCED, now, NULL, is_remote_node); free(now); value = crm_meta_value(action->params, PCMK__META_DIGESTS_ALL); update_attrd(target, CRM_ATTR_DIGESTS_ALL, value, NULL, is_remote_node); value = crm_meta_value(action->params, PCMK__META_DIGESTS_SECURE); update_attrd(target, CRM_ATTR_DIGESTS_SECURE, value, NULL, is_remote_node); } else if (!(pcmk_is_set(action->flags, pcmk__graph_action_sent_update))) { update_node_state_after_fencing(target, uuid); pcmk__set_graph_action_flags(action, pcmk__graph_action_sent_update); } } st_fail_count_reset(target); } else { enum pcmk__graph_next abort_action = pcmk__graph_restart; int status = stonith__execution_status(data); const char *reason = stonith__exit_reason(data); if (reason == NULL) { if (status == PCMK_EXEC_DONE) { reason = "Agent returned error"; } else { reason = pcmk_exec_status_str(status); } } pcmk__set_graph_action_flags(action, pcmk__graph_action_failed); /* If no fence devices were available, there's no use in immediately * checking again, so don't start a new transition in that case. */ if (status == PCMK_EXEC_NO_FENCE_DEVICE) { crm_warn("Fence operation %d for %s failed: %s " "(aborting transition and giving up for now)", data->call_id, target, reason); abort_action = pcmk__graph_wait; } else { crm_notice("Fence operation %d for %s failed: %s " "(aborting transition)", data->call_id, target, reason); } /* Increment the fail count now, so abort_for_stonith_failure() can * check it. Non-DC nodes will increment it in * handle_fence_notification(). */ st_fail_count_increment(target); abort_for_stonith_failure(abort_action, target, NULL); } pcmk__update_graph(controld_globals.transition_graph, action); trigger_graph(); bail: free(data->userdata); free(uuid); return; } static int fence_with_delay(const char *target, const char *type, int delay) { uint32_t options = st_opt_none; // Group of enum stonith_call_options int timeout_sec = pcmk__timeout_ms2s(controld_globals.transition_graph->fencing_timeout); if (crmd_join_phase_count(controld_join_confirmed) == 1) { stonith__set_call_options(options, target, st_opt_allow_self_fencing); } return stonith_api->cmds->fence_with_delay(stonith_api, options, target, type, timeout_sec, 0, delay); } /*! * \internal * \brief Execute a fencing action from a transition graph * * \param[in] graph Transition graph being executed (ignored) * \param[in] action Fencing action to execute * * \return Standard Pacemaker return code */ int controld_execute_fence_action(pcmk__graph_t *graph, pcmk__graph_action_t *action) { int rc = 0; const char *id = pcmk__xe_id(action->xml); const char *uuid = pcmk__xe_get(action->xml, PCMK__META_ON_NODE_UUID); const char *target = pcmk__xe_get(action->xml, PCMK__META_ON_NODE); const char *type = crm_meta_value(action->params, PCMK__META_STONITH_ACTION); char *transition_key = NULL; const char *priority_delay = NULL; int delay_i = 0; gboolean invalid_action = FALSE; int timeout_sec = pcmk__timeout_ms2s(controld_globals.transition_graph->fencing_timeout); CRM_CHECK(id != NULL, invalid_action = TRUE); CRM_CHECK(uuid != NULL, invalid_action = TRUE); CRM_CHECK(type != NULL, invalid_action = TRUE); CRM_CHECK(target != NULL, invalid_action = TRUE); if (invalid_action) { crm_log_xml_warn(action->xml, "BadAction"); return EPROTO; } priority_delay = crm_meta_value(action->params, PCMK_OPT_PRIORITY_FENCING_DELAY); crm_notice("Requesting fencing (%s) targeting node %s " QB_XS " action=%s timeout=%i%s%s", type, target, id, timeout_sec, priority_delay ? " priority_delay=" : "", priority_delay ? priority_delay : ""); /* Passing NULL means block until we can connect... */ controld_timer_fencer_connect(NULL); pcmk__scan_min_int(priority_delay, &delay_i, 0); rc = fence_with_delay(target, type, delay_i); transition_key = pcmk__transition_key(controld_globals.transition_graph->id, action->id, 0, controld_globals.te_uuid), stonith_api->cmds->register_callback(stonith_api, rc, (timeout_sec + (delay_i > 0 ? delay_i : 0)), st_opt_timeout_updates, transition_key, "tengine_stonith_callback", tengine_stonith_callback); return pcmk_rc_ok; } bool controld_valid_fencing_watchdog_timeout(const char *value) { const char *our_nodename = controld_globals.cluster->priv->node_name; if ((stonith_api == NULL) || (stonith_api->state == stonith_disconnected) || !stonith__watchdog_fencing_enabled_for_node_api(stonith_api, our_nodename)) { // Anything is valid since it won't be used return true; } return pcmk__valid_fencing_watchdog_timeout(value); } /* end stonith API client functions */ /* * stonith history synchronization * * Each node's fencer keeps track of a cluster-wide fencing history. When a node * joins or leaves, we need to synchronize the history across all nodes. */ static crm_trigger_t *stonith_history_sync_trigger = NULL; static mainloop_timer_t *stonith_history_sync_timer_short = NULL; static mainloop_timer_t *stonith_history_sync_timer_long = NULL; void -te_cleanup_stonith_history_sync(stonith_t *st, bool free_timers) +controld_cleanup_fencing_history_sync(stonith_t *st, bool free_timers) { if (free_timers) { mainloop_timer_del(stonith_history_sync_timer_short); stonith_history_sync_timer_short = NULL; mainloop_timer_del(stonith_history_sync_timer_long); stonith_history_sync_timer_long = NULL; } else { mainloop_timer_stop(stonith_history_sync_timer_short); mainloop_timer_stop(stonith_history_sync_timer_long); } if (st) { st->cmds->remove_notification(st, PCMK__VALUE_ST_NOTIFY_HISTORY_SYNCED); } } static void tengine_stonith_history_synced(stonith_t *st, stonith_event_t *st_event) { - te_cleanup_stonith_history_sync(st, FALSE); + controld_cleanup_fencing_history_sync(st, false); crm_debug("Fence-history synced - cancel all timers"); } static gboolean stonith_history_sync_set_trigger(gpointer user_data) { mainloop_set_trigger(stonith_history_sync_trigger); return FALSE; } void controld_trigger_fencing_history_sync(bool long_timeout) { /* trigger a sync in 5s to give more nodes the * chance to show up so that we don't create * unnecessary stonith-history-sync traffic * * the long timeout of 30s is there as a fallback * so that after a successful connection to fenced * we will wait for 30s for the DC to trigger a * history-sync * if this doesn't happen we trigger a sync locally * (e.g. fenced segfaults and is restarted by pacemakerd) */ /* as we are finally checking the stonith-connection * in do_stonith_history_sync we should be fine * leaving stonith_history_sync_time & stonith_history_sync_trigger * around */ if (stonith_history_sync_trigger == NULL) { stonith_history_sync_trigger = mainloop_add_trigger(G_PRIORITY_LOW, do_stonith_history_sync, NULL); } if (long_timeout) { if(stonith_history_sync_timer_long == NULL) { stonith_history_sync_timer_long = mainloop_timer_add("history_sync_long", 30000, FALSE, stonith_history_sync_set_trigger, NULL); } crm_info("Fence history will be synchronized cluster-wide within 30 seconds"); mainloop_timer_start(stonith_history_sync_timer_long); } else { if(stonith_history_sync_timer_short == NULL) { stonith_history_sync_timer_short = mainloop_timer_add("history_sync_short", 5000, FALSE, stonith_history_sync_set_trigger, NULL); } crm_info("Fence history will be synchronized cluster-wide within 5 seconds"); mainloop_timer_start(stonith_history_sync_timer_short); } } /* end stonith history synchronization functions */ diff --git a/daemons/controld/controld_fencing.h b/daemons/controld/controld_fencing.h index 6d1b0e86ca..bad90fe5bf 100644 --- a/daemons/controld/controld_fencing.h +++ b/daemons/controld/controld_fencing.h @@ -1,37 +1,37 @@ /* * Copyright 2004-2025 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #ifndef CONTROLD_FENCING__H # define CONTROLD_FENCING__H #include // bool #include // pcmk__graph_t, pcmk__graph_action_t void controld_configure_fencing(GHashTable *options); // stonith fail counts void st_fail_count_reset(const char * target); // stonith API client gboolean controld_timer_fencer_connect(gpointer user_data); void controld_disconnect_fencer(bool destroy); int controld_execute_fence_action(pcmk__graph_t *graph, pcmk__graph_action_t *action); bool controld_valid_fencing_watchdog_timeout(const char *value); // Fencing cleanup list void controld_remove_fencing_cleanup(const char *target); void controld_purge_fencing_cleanup(void); void controld_execute_fencing_cleanup(void); // Fencing history synchronization void controld_trigger_fencing_history_sync(bool long_timeout); -void te_cleanup_stonith_history_sync(stonith_t *st, bool free_timers); +void controld_cleanup_fencing_history_sync(stonith_t *st, bool free_timers); #endif