diff --git a/daemons/execd/execd_commands.c b/daemons/execd/execd_commands.c index 482707455a..9da1a2a276 100644 --- a/daemons/execd/execd_commands.c +++ b/daemons/execd/execd_commands.c @@ -1,1944 +1,1966 @@ /* * 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 // Check whether we have a high-resolution monotonic clock #undef PCMK__TIME_USE_CGT #if HAVE_DECL_CLOCK_MONOTONIC && defined(CLOCK_MONOTONIC) # define PCMK__TIME_USE_CGT # include /* clock_gettime */ #endif #include #include #include #include #include #include #include #include #include #include "pacemaker-execd.h" GHashTable *rsc_list = NULL; typedef struct lrmd_cmd_s { int timeout; guint interval_ms; int start_delay; int timeout_orig; int call_id; int call_opts; /* Timer ids, must be removed on cmd destruction. */ int delay_id; int stonith_recurring_id; int rsc_deleted; int service_flags; char *client_id; char *origin; char *rsc_id; char *action; char *real_action; char *userdata_str; pcmk__action_result_t result; /* We can track operation queue time and run time, to be saved with the CIB * resource history (and displayed in cluster status). We need * high-resolution monotonic time for this purpose, so we use * clock_gettime(CLOCK_MONOTONIC, ...) (if available, otherwise this feature * is disabled). * * However, we also need epoch timestamps for recording the time the command * last ran and the time its return value last changed, for use in time * displays (as opposed to interval calculations). We keep time_t values for * this purpose. * * The last run time is used for both purposes, so we keep redundant * monotonic and epoch values for this. Technically the two could represent * different times, but since time_t has only second resolution and the * values are used for distinct purposes, that is not significant. */ #ifdef PCMK__TIME_USE_CGT /* Recurring and systemd operations may involve more than one executor * command per operation, so they need info about the original and the most * recent. */ struct timespec t_first_run; // When op first ran struct timespec t_run; // When op most recently ran struct timespec t_first_queue; // When op was first queued struct timespec t_queue; // When op was most recently queued #endif time_t epoch_last_run; // Epoch timestamp of when op last ran time_t epoch_rcchange; // Epoch timestamp of when rc last changed bool first_notify_sent; int last_notify_rc; int last_notify_op_status; int last_pid; GHashTable *params; } lrmd_cmd_t; static void cmd_finalize(lrmd_cmd_t * cmd, lrmd_rsc_t * rsc); static gboolean execute_resource_action(gpointer user_data); static void cancel_all_recurring(lrmd_rsc_t * rsc, const char *client_id); #ifdef PCMK__TIME_USE_CGT /*! * \internal * \brief Check whether a struct timespec has been set * * \param[in] timespec Time to check * * \return true if timespec has been set (i.e. is nonzero), false otherwise */ static inline bool time_is_set(const struct timespec *timespec) { return (timespec != NULL) && ((timespec->tv_sec != 0) || (timespec->tv_nsec != 0)); } /* * \internal * \brief Set a timespec (and its original if unset) to the current time * * \param[out] t_current Where to store current time * \param[out] t_orig Where to copy t_current if unset */ static void get_current_time(struct timespec *t_current, struct timespec *t_orig) { clock_gettime(CLOCK_MONOTONIC, t_current); if ((t_orig != NULL) && !time_is_set(t_orig)) { *t_orig = *t_current; } } /*! * \internal * \brief Return difference between two times in milliseconds * * \param[in] now More recent time (or NULL to use current time) * \param[in] old Earlier time * * \return milliseconds difference (or 0 if old is NULL or unset) * * \note Can overflow on 32bit machines when the differences is around * 24 days or more. */ static int time_diff_ms(const struct timespec *now, const struct timespec *old) { int diff_ms = 0; if (time_is_set(old)) { struct timespec local_now = { 0, }; if (now == NULL) { clock_gettime(CLOCK_MONOTONIC, &local_now); now = &local_now; } diff_ms = (now->tv_sec - old->tv_sec) * 1000 + (now->tv_nsec - old->tv_nsec) / 1000000; } return diff_ms; } /*! * \internal * \brief Reset a command's operation times to their original values. * * Reset a command's run and queued timestamps to the timestamps of the original * command, so we report the entire time since then and not just the time since * the most recent command (for recurring and systemd operations). * * \param[in,out] cmd Executor command object to reset * * \note It's not obvious what the queued time should be for a systemd * start/stop operation, which might go like this: * initial command queued 5ms, runs 3s * monitor command queued 10ms, runs 10s * monitor command queued 10ms, runs 10s * Is the queued time for that operation 5ms, 10ms or 25ms? The current * implementation will report 5ms. If it's 25ms, then we need to * subtract 20ms from the total exec time so as not to count it twice. * We can implement that later if it matters to anyone ... */ static void cmd_original_times(lrmd_cmd_t * cmd) { cmd->t_run = cmd->t_first_run; cmd->t_queue = cmd->t_first_queue; } #endif static inline bool action_matches(const lrmd_cmd_t *cmd, const char *action, guint interval_ms) { return (cmd->interval_ms == interval_ms) && pcmk__str_eq(cmd->action, action, pcmk__str_casei); } /*! * \internal * \brief Log the result of an asynchronous command * * \param[in] cmd Command to log result for * \param[in] exec_time_ms Execution time in milliseconds, if known * \param[in] queue_time_ms Queue time in milliseconds, if known */ static void log_finished(const lrmd_cmd_t *cmd, int exec_time_ms, int queue_time_ms) { int log_level = LOG_INFO; GString *str = g_string_sized_new(100); // reasonable starting size if (pcmk__str_eq(cmd->action, PCMK_ACTION_MONITOR, pcmk__str_casei)) { log_level = LOG_DEBUG; } g_string_append_printf(str, "%s %s (call %d", cmd->rsc_id, cmd->action, cmd->call_id); if (cmd->last_pid != 0) { g_string_append_printf(str, ", PID %d", cmd->last_pid); } if (cmd->result.execution_status == PCMK_EXEC_DONE) { g_string_append_printf(str, ") exited with status %d", cmd->result.exit_status); } else { pcmk__g_strcat(str, ") could not be executed: ", pcmk_exec_status_str(cmd->result.execution_status), NULL); } if (cmd->result.exit_reason != NULL) { pcmk__g_strcat(str, " (", cmd->result.exit_reason, ")", NULL); } #ifdef PCMK__TIME_USE_CGT pcmk__g_strcat(str, " (execution time ", pcmk__readable_interval(exec_time_ms), NULL); if (queue_time_ms > 0) { pcmk__g_strcat(str, " after being queued ", pcmk__readable_interval(queue_time_ms), NULL); } g_string_append_c(str, ')'); #endif do_crm_log(log_level, "%s", str->str); g_string_free(str, TRUE); } static void log_execute(lrmd_cmd_t * cmd) { int log_level = LOG_INFO; if (pcmk__str_eq(cmd->action, PCMK_ACTION_MONITOR, pcmk__str_casei)) { log_level = LOG_DEBUG; } do_crm_log(log_level, "executing - rsc:%s action:%s call_id:%d", cmd->rsc_id, cmd->action, cmd->call_id); } static const char * normalize_action_name(lrmd_rsc_t * rsc, const char *action) { if (pcmk__str_eq(action, PCMK_ACTION_MONITOR, pcmk__str_casei) && pcmk_is_set(pcmk_get_ra_caps(rsc->class), pcmk_ra_cap_status)) { return PCMK_ACTION_STATUS; } return action; } static lrmd_rsc_t * build_rsc_from_xml(xmlNode * msg) { xmlNode *rsc_xml = get_xpath_object("//" PCMK__XE_LRMD_RSC, msg, LOG_ERR); lrmd_rsc_t *rsc = NULL; rsc = pcmk__assert_alloc(1, sizeof(lrmd_rsc_t)); crm_element_value_int(msg, PCMK__XA_LRMD_CALLOPT, &rsc->call_opts); rsc->rsc_id = crm_element_value_copy(rsc_xml, PCMK__XA_LRMD_RSC_ID); rsc->class = crm_element_value_copy(rsc_xml, PCMK__XA_LRMD_CLASS); rsc->provider = crm_element_value_copy(rsc_xml, PCMK__XA_LRMD_PROVIDER); rsc->type = crm_element_value_copy(rsc_xml, PCMK__XA_LRMD_TYPE); rsc->work = mainloop_add_trigger(G_PRIORITY_HIGH, execute_resource_action, rsc); // Initialize fence device probes (to return "not running") pcmk__set_result(&rsc->fence_probe_result, CRM_EX_ERROR, PCMK_EXEC_NO_FENCE_DEVICE, NULL); return rsc; } static lrmd_cmd_t * create_lrmd_cmd(xmlNode *msg, pcmk__client_t *client) { int call_options = 0; xmlNode *rsc_xml = get_xpath_object("//" PCMK__XE_LRMD_RSC, msg, LOG_ERR); lrmd_cmd_t *cmd = NULL; cmd = pcmk__assert_alloc(1, sizeof(lrmd_cmd_t)); crm_element_value_int(msg, PCMK__XA_LRMD_CALLOPT, &call_options); cmd->call_opts = call_options; cmd->client_id = pcmk__str_copy(client->id); crm_element_value_int(msg, PCMK__XA_LRMD_CALLID, &cmd->call_id); crm_element_value_ms(rsc_xml, PCMK__XA_LRMD_RSC_INTERVAL, &cmd->interval_ms); crm_element_value_int(rsc_xml, PCMK__XA_LRMD_TIMEOUT, &cmd->timeout); crm_element_value_int(rsc_xml, PCMK__XA_LRMD_RSC_START_DELAY, &cmd->start_delay); cmd->timeout_orig = cmd->timeout; cmd->origin = crm_element_value_copy(rsc_xml, PCMK__XA_LRMD_ORIGIN); cmd->action = crm_element_value_copy(rsc_xml, PCMK__XA_LRMD_RSC_ACTION); cmd->userdata_str = crm_element_value_copy(rsc_xml, PCMK__XA_LRMD_RSC_USERDATA_STR); cmd->rsc_id = crm_element_value_copy(rsc_xml, PCMK__XA_LRMD_RSC_ID); cmd->params = xml2list(rsc_xml); if (pcmk__str_eq(g_hash_table_lookup(cmd->params, "CRM_meta_on_fail"), PCMK_VALUE_BLOCK, pcmk__str_casei)) { crm_debug("Setting flag to leave pid group on timeout and " "only kill action pid for " PCMK__OP_FMT, cmd->rsc_id, cmd->action, cmd->interval_ms); cmd->service_flags = pcmk__set_flags_as(__func__, __LINE__, LOG_TRACE, "Action", cmd->action, 0, SVC_ACTION_LEAVE_GROUP, "SVC_ACTION_LEAVE_GROUP"); } return cmd; } static void stop_recurring_timer(lrmd_cmd_t *cmd) { if (cmd) { if (cmd->stonith_recurring_id) { g_source_remove(cmd->stonith_recurring_id); } cmd->stonith_recurring_id = 0; } } static void free_lrmd_cmd(lrmd_cmd_t * cmd) { stop_recurring_timer(cmd); if (cmd->delay_id) { g_source_remove(cmd->delay_id); } if (cmd->params) { g_hash_table_destroy(cmd->params); } pcmk__reset_result(&(cmd->result)); free(cmd->origin); free(cmd->action); free(cmd->real_action); free(cmd->userdata_str); free(cmd->rsc_id); free(cmd->client_id); free(cmd); } static gboolean stonith_recurring_op_helper(gpointer data) { lrmd_cmd_t *cmd = data; lrmd_rsc_t *rsc; cmd->stonith_recurring_id = 0; if (!cmd->rsc_id) { return FALSE; } rsc = g_hash_table_lookup(rsc_list, cmd->rsc_id); pcmk__assert(rsc != NULL); /* take it out of recurring_ops list, and put it in the pending ops * to be executed */ rsc->recurring_ops = g_list_remove(rsc->recurring_ops, cmd); rsc->pending_ops = g_list_append(rsc->pending_ops, cmd); #ifdef PCMK__TIME_USE_CGT get_current_time(&(cmd->t_queue), &(cmd->t_first_queue)); #endif mainloop_set_trigger(rsc->work); return FALSE; } static inline void start_recurring_timer(lrmd_cmd_t *cmd) { if (cmd && (cmd->interval_ms > 0)) { cmd->stonith_recurring_id = g_timeout_add(cmd->interval_ms, stonith_recurring_op_helper, cmd); } } static gboolean start_delay_helper(gpointer data) { lrmd_cmd_t *cmd = data; lrmd_rsc_t *rsc = NULL; cmd->delay_id = 0; rsc = cmd->rsc_id ? g_hash_table_lookup(rsc_list, cmd->rsc_id) : NULL; if (rsc) { mainloop_set_trigger(rsc->work); } return FALSE; } /*! * \internal * \brief Check whether a list already contains the equivalent of a given action * * \param[in] action_list List to search * \param[in] cmd Action to search for */ static lrmd_cmd_t * find_duplicate_action(const GList *action_list, const lrmd_cmd_t *cmd) { for (const GList *item = action_list; item != NULL; item = item->next) { lrmd_cmd_t *dup = item->data; if (action_matches(cmd, dup->action, dup->interval_ms)) { return dup; } } return NULL; } static bool merge_recurring_duplicate(lrmd_rsc_t * rsc, lrmd_cmd_t * cmd) { lrmd_cmd_t * dup = NULL; bool dup_pending = true; if (cmd->interval_ms == 0) { return false; } // Search for a duplicate of this action (in-flight or not) dup = find_duplicate_action(rsc->pending_ops, cmd); if (dup == NULL) { dup_pending = false; dup = find_duplicate_action(rsc->recurring_ops, cmd); if (dup == NULL) { return false; } } /* Do not merge fencing monitors marked for cancellation, so we can reply to * the cancellation separately. */ if (pcmk__str_eq(rsc->class, PCMK_RESOURCE_CLASS_STONITH, pcmk__str_casei) && (dup->result.execution_status == PCMK_EXEC_CANCELLED)) { return false; } /* This should not occur. If it does, we need to investigate how something * like this is possible in the controller. */ crm_warn("Duplicate recurring op entry detected (" PCMK__OP_FMT "), merging with previous op entry", rsc->rsc_id, normalize_action_name(rsc, dup->action), dup->interval_ms); // Merge new action's call ID and user data into existing action dup->first_notify_sent = false; free(dup->userdata_str); dup->userdata_str = cmd->userdata_str; cmd->userdata_str = NULL; dup->call_id = cmd->call_id; free_lrmd_cmd(cmd); cmd = NULL; /* If dup is not pending, that means it has already executed at least once * and is waiting in the interval. In that case, stop waiting and initiate * a new instance now. */ if (!dup_pending) { if (pcmk__str_eq(rsc->class, PCMK_RESOURCE_CLASS_STONITH, pcmk__str_casei)) { stop_recurring_timer(dup); stonith_recurring_op_helper(dup); } else { services_action_kick(rsc->rsc_id, normalize_action_name(rsc, dup->action), dup->interval_ms); } } return true; } static void schedule_lrmd_cmd(lrmd_rsc_t * rsc, lrmd_cmd_t * cmd) { CRM_CHECK(cmd != NULL, return); CRM_CHECK(rsc != NULL, return); crm_trace("Scheduling %s on %s", cmd->action, rsc->rsc_id); if (merge_recurring_duplicate(rsc, cmd)) { // Equivalent of cmd has already been scheduled return; } /* The controller expects the executor to automatically cancel * recurring operations before a resource stops. */ if (pcmk__str_eq(cmd->action, PCMK_ACTION_STOP, pcmk__str_casei)) { cancel_all_recurring(rsc, NULL); } rsc->pending_ops = g_list_append(rsc->pending_ops, cmd); #ifdef PCMK__TIME_USE_CGT get_current_time(&(cmd->t_queue), &(cmd->t_first_queue)); #endif mainloop_set_trigger(rsc->work); if (cmd->start_delay) { cmd->delay_id = g_timeout_add(cmd->start_delay, start_delay_helper, cmd); } } static xmlNode * create_lrmd_reply(const char *origin, int rc, int call_id) { xmlNode *reply = pcmk__xe_create(NULL, PCMK__XE_LRMD_REPLY); crm_xml_add(reply, PCMK__XA_LRMD_ORIGIN, origin); crm_xml_add_int(reply, PCMK__XA_LRMD_RC, rc); crm_xml_add_int(reply, PCMK__XA_LRMD_CALLID, call_id); return reply; } static void send_client_notify(gpointer key, gpointer value, gpointer user_data) { xmlNode *update_msg = user_data; pcmk__client_t *client = value; int rc; int log_level = LOG_WARNING; const char *msg = NULL; CRM_CHECK(client != NULL, return); if (client->name == NULL) { crm_trace("Skipping notification to client without name"); return; } if (pcmk_is_set(client->flags, pcmk__client_to_proxy)) { /* We only want to notify clients of the executor IPC API. If we are * running as Pacemaker Remote, we may have clients proxied to other * IPC services in the cluster, so skip those. */ crm_trace("Skipping executor API notification to client %s", pcmk__client_name(client)); return; } rc = lrmd_server_send_notify(client, update_msg); if (rc == pcmk_rc_ok) { return; } switch (rc) { case ENOTCONN: case EPIPE: // Client exited without waiting for notification log_level = LOG_INFO; msg = "Disconnected"; break; default: msg = pcmk_rc_str(rc); break; } do_crm_log(log_level, "Could not notify client %s: %s " QB_XS " rc=%d", pcmk__client_name(client), msg, rc); } static void send_cmd_complete_notify(lrmd_cmd_t * cmd) { xmlNode *notify = NULL; int exec_time = 0; int queue_time = 0; #ifdef PCMK__TIME_USE_CGT exec_time = time_diff_ms(NULL, &(cmd->t_run)); queue_time = time_diff_ms(&cmd->t_run, &(cmd->t_queue)); #endif log_finished(cmd, exec_time, queue_time); /* If the originator requested to be notified only for changes in recurring * operation results, skip the notification if the result hasn't changed. */ if (cmd->first_notify_sent && pcmk_is_set(cmd->call_opts, lrmd_opt_notify_changes_only) && (cmd->last_notify_rc == cmd->result.exit_status) && (cmd->last_notify_op_status == cmd->result.execution_status)) { return; } cmd->first_notify_sent = true; cmd->last_notify_rc = cmd->result.exit_status; cmd->last_notify_op_status = cmd->result.execution_status; notify = pcmk__xe_create(NULL, PCMK__XE_LRMD_NOTIFY); crm_xml_add(notify, PCMK__XA_LRMD_ORIGIN, __func__); crm_xml_add_int(notify, PCMK__XA_LRMD_TIMEOUT, cmd->timeout); crm_xml_add_ms(notify, PCMK__XA_LRMD_RSC_INTERVAL, cmd->interval_ms); crm_xml_add_int(notify, PCMK__XA_LRMD_RSC_START_DELAY, cmd->start_delay); crm_xml_add_int(notify, PCMK__XA_LRMD_EXEC_RC, cmd->result.exit_status); crm_xml_add_int(notify, PCMK__XA_LRMD_EXEC_OP_STATUS, cmd->result.execution_status); crm_xml_add_int(notify, PCMK__XA_LRMD_CALLID, cmd->call_id); crm_xml_add_int(notify, PCMK__XA_LRMD_RSC_DELETED, cmd->rsc_deleted); crm_xml_add_ll(notify, PCMK__XA_LRMD_RUN_TIME, (long long) cmd->epoch_last_run); crm_xml_add_ll(notify, PCMK__XA_LRMD_RCCHANGE_TIME, (long long) cmd->epoch_rcchange); #ifdef PCMK__TIME_USE_CGT crm_xml_add_int(notify, PCMK__XA_LRMD_EXEC_TIME, exec_time); crm_xml_add_int(notify, PCMK__XA_LRMD_QUEUE_TIME, queue_time); #endif crm_xml_add(notify, PCMK__XA_LRMD_OP, LRMD_OP_RSC_EXEC); crm_xml_add(notify, PCMK__XA_LRMD_RSC_ID, cmd->rsc_id); if(cmd->real_action) { crm_xml_add(notify, PCMK__XA_LRMD_RSC_ACTION, cmd->real_action); } else { crm_xml_add(notify, PCMK__XA_LRMD_RSC_ACTION, cmd->action); } crm_xml_add(notify, PCMK__XA_LRMD_RSC_USERDATA_STR, cmd->userdata_str); crm_xml_add(notify, PCMK__XA_LRMD_RSC_EXIT_REASON, cmd->result.exit_reason); if (cmd->result.action_stderr != NULL) { crm_xml_add(notify, PCMK__XA_LRMD_RSC_OUTPUT, cmd->result.action_stderr); } else if (cmd->result.action_stdout != NULL) { crm_xml_add(notify, PCMK__XA_LRMD_RSC_OUTPUT, cmd->result.action_stdout); } if (cmd->params) { char *key = NULL; char *value = NULL; GHashTableIter iter; xmlNode *args = pcmk__xe_create(notify, PCMK__XE_ATTRIBUTES); g_hash_table_iter_init(&iter, cmd->params); while (g_hash_table_iter_next(&iter, (gpointer *) & key, (gpointer *) & value)) { hash2smartfield((gpointer) key, (gpointer) value, args); } } if ((cmd->client_id != NULL) && pcmk_is_set(cmd->call_opts, lrmd_opt_notify_orig_only)) { pcmk__client_t *client = pcmk__find_client_by_id(cmd->client_id); if (client != NULL) { send_client_notify(client->id, client, notify); } } else { pcmk__foreach_ipc_client(send_client_notify, notify); } pcmk__xml_free(notify); } static void send_generic_notify(int rc, xmlNode * request) { if (pcmk__ipc_client_count() != 0) { int call_id = 0; xmlNode *notify = NULL; xmlNode *rsc_xml = get_xpath_object("//" PCMK__XE_LRMD_RSC, request, LOG_ERR); const char *rsc_id = crm_element_value(rsc_xml, PCMK__XA_LRMD_RSC_ID); const char *op = crm_element_value(request, PCMK__XA_LRMD_OP); crm_element_value_int(request, PCMK__XA_LRMD_CALLID, &call_id); notify = pcmk__xe_create(NULL, PCMK__XE_LRMD_NOTIFY); crm_xml_add(notify, PCMK__XA_LRMD_ORIGIN, __func__); crm_xml_add_int(notify, PCMK__XA_LRMD_RC, rc); crm_xml_add_int(notify, PCMK__XA_LRMD_CALLID, call_id); crm_xml_add(notify, PCMK__XA_LRMD_OP, op); crm_xml_add(notify, PCMK__XA_LRMD_RSC_ID, rsc_id); pcmk__foreach_ipc_client(send_client_notify, notify); pcmk__xml_free(notify); } } static void cmd_reset(lrmd_cmd_t * cmd) { cmd->last_pid = 0; #ifdef PCMK__TIME_USE_CGT memset(&cmd->t_run, 0, sizeof(cmd->t_run)); memset(&cmd->t_queue, 0, sizeof(cmd->t_queue)); #endif cmd->epoch_last_run = 0; pcmk__reset_result(&(cmd->result)); cmd->result.execution_status = PCMK_EXEC_DONE; } static void cmd_finalize(lrmd_cmd_t * cmd, lrmd_rsc_t * rsc) { crm_trace("Resource operation rsc:%s action:%s completed (%p %p)", cmd->rsc_id, cmd->action, rsc ? rsc->active : NULL, cmd); if (rsc && (rsc->active == cmd)) { rsc->active = NULL; mainloop_set_trigger(rsc->work); } if (!rsc) { cmd->rsc_deleted = 1; } /* reset original timeout so client notification has correct information */ cmd->timeout = cmd->timeout_orig; send_cmd_complete_notify(cmd); if ((cmd->interval_ms != 0) && (cmd->result.execution_status == PCMK_EXEC_CANCELLED)) { if (rsc) { rsc->recurring_ops = g_list_remove(rsc->recurring_ops, cmd); rsc->pending_ops = g_list_remove(rsc->pending_ops, cmd); } free_lrmd_cmd(cmd); } else if (cmd->interval_ms == 0) { if (rsc) { rsc->pending_ops = g_list_remove(rsc->pending_ops, cmd); } free_lrmd_cmd(cmd); } else { /* Clear all the values pertaining just to the last iteration of a recurring op. */ cmd_reset(cmd); } } struct notify_new_client_data { xmlNode *notify; pcmk__client_t *new_client; }; static void notify_one_client(gpointer key, gpointer value, gpointer user_data) { pcmk__client_t *client = value; struct notify_new_client_data *data = user_data; if (!pcmk__str_eq(client->id, data->new_client->id, pcmk__str_casei)) { send_client_notify(key, (gpointer) client, (gpointer) data->notify); } } void notify_of_new_client(pcmk__client_t *new_client) { struct notify_new_client_data data; data.new_client = new_client; data.notify = pcmk__xe_create(NULL, PCMK__XE_LRMD_NOTIFY); crm_xml_add(data.notify, PCMK__XA_LRMD_ORIGIN, __func__); crm_xml_add(data.notify, PCMK__XA_LRMD_OP, LRMD_OP_NEW_CLIENT); pcmk__foreach_ipc_client(notify_one_client, &data); pcmk__xml_free(data.notify); } void client_disconnect_cleanup(const char *client_id) { GHashTableIter iter; lrmd_rsc_t *rsc = NULL; char *key = NULL; g_hash_table_iter_init(&iter, rsc_list); while (g_hash_table_iter_next(&iter, (gpointer *) & key, (gpointer *) & rsc)) { if (pcmk_all_flags_set(rsc->call_opts, lrmd_opt_drop_recurring)) { /* This client is disconnecting, drop any recurring operations * it may have initiated on the resource */ cancel_all_recurring(rsc, client_id); } } } static void action_complete(svc_action_t * action) { lrmd_rsc_t *rsc; lrmd_cmd_t *cmd = action->cb_data; enum ocf_exitcode code; #ifdef PCMK__TIME_USE_CGT const char *rclass = NULL; bool goagain = false; #endif if (!cmd) { crm_err("Completed executor action (%s) does not match any known operations", action->id); return; } #ifdef PCMK__TIME_USE_CGT if (cmd->result.exit_status != action->rc) { cmd->epoch_rcchange = time(NULL); } #endif cmd->last_pid = action->pid; // Cast variable instead of function return to keep compilers happy code = services_result2ocf(action->standard, cmd->action, action->rc); pcmk__set_result(&(cmd->result), (int) code, action->status, services__exit_reason(action)); rsc = cmd->rsc_id ? g_hash_table_lookup(rsc_list, cmd->rsc_id) : NULL; #ifdef PCMK__TIME_USE_CGT if (rsc != NULL) { rclass = rsc->class; #if PCMK__ENABLE_SERVICE if (pcmk__str_eq(rclass, PCMK_RESOURCE_CLASS_SERVICE, pcmk__str_casei)) { rclass = resources_find_service_class(rsc->type); } #endif } if (pcmk__str_eq(rclass, PCMK_RESOURCE_CLASS_SYSTEMD, pcmk__str_casei)) { if (pcmk__result_ok(&(cmd->result)) && pcmk__strcase_any_of(cmd->action, PCMK_ACTION_START, PCMK_ACTION_STOP, NULL)) { /* systemd returns from start and stop actions after the action * begins, not after it completes. We have to jump through a few * hoops so that we don't report 'complete' to the rest of pacemaker * until it's actually done. */ goagain = true; cmd->real_action = cmd->action; cmd->action = pcmk__str_copy(PCMK_ACTION_MONITOR); } else if (cmd->real_action != NULL) { // This is follow-up monitor to check whether start/stop completed if (cmd->result.execution_status == PCMK_EXEC_PENDING) { goagain = true; } else if (pcmk__result_ok(&(cmd->result)) && pcmk__str_eq(cmd->real_action, PCMK_ACTION_STOP, pcmk__str_casei)) { goagain = true; } else { int time_sum = time_diff_ms(NULL, &(cmd->t_first_run)); int timeout_left = cmd->timeout_orig - time_sum; crm_debug("%s systemd %s is now complete (elapsed=%dms, " "remaining=%dms): %s (%d)", cmd->rsc_id, cmd->real_action, time_sum, timeout_left, crm_exit_str(cmd->result.exit_status), cmd->result.exit_status); cmd_original_times(cmd); // Monitors may return "not running", but start/stop shouldn't if ((cmd->result.execution_status == PCMK_EXEC_DONE) && (cmd->result.exit_status == PCMK_OCF_NOT_RUNNING)) { if (pcmk__str_eq(cmd->real_action, PCMK_ACTION_START, pcmk__str_casei)) { cmd->result.exit_status = PCMK_OCF_UNKNOWN_ERROR; } else if (pcmk__str_eq(cmd->real_action, PCMK_ACTION_STOP, pcmk__str_casei)) { cmd->result.exit_status = PCMK_OCF_OK; } } } + } else if (pcmk__strcase_any_of(cmd->action, PCMK_ACTION_MONITOR, PCMK_ACTION_STATUS, NULL) && + (cmd->interval_ms > 0)) { + /* For monitors other than follow up monitors, if "use-monitor-pending-timeout" */ + /* is enabled and the pending state continues from the time of the first notification */ + /* until the timeout, it will be treated as a timeout. */ + if (pcmk__str_eq(g_hash_table_lookup(cmd->params, PCMK_XA_USE_MONITOR_PENDING_TIMEOUT), "true", pcmk__str_casei)) { + if ((cmd->result.execution_status == PCMK_EXEC_PENDING) && + (cmd->last_notify_op_status == PCMK_EXEC_PENDING)) { + int time_left = time(NULL) - (cmd->epoch_rcchange + (cmd->timeout_orig/1000)); + + if (time_left >= 0) { + crm_notice("Giving up on %s %s (rc=%d): monitor pending timeout (first pending notification=%s timeout=%ds)", + cmd->rsc_id, cmd->action, + cmd->result.exit_status, pcmk__trim(ctime(&cmd->epoch_rcchange)), cmd->timeout_orig); + pcmk__set_result(&(cmd->result), PCMK_OCF_UNKNOWN_ERROR, + PCMK_EXEC_TIMEOUT, + "Investigate reason for timeout, and adjust " + "configured operation timeout if necessary"); + cmd_original_times(cmd); + } + } + } } } #endif #ifdef PCMK__TIME_USE_CGT if (goagain) { int time_sum = time_diff_ms(NULL, &(cmd->t_first_run)); int timeout_left = cmd->timeout_orig - time_sum; int delay = cmd->timeout_orig / 10; if(delay >= timeout_left && timeout_left > 20) { delay = timeout_left/2; } delay = QB_MIN(2000, delay); if (delay < timeout_left) { cmd->start_delay = delay; cmd->timeout = timeout_left; if (pcmk__result_ok(&(cmd->result))) { crm_debug("%s %s may still be in progress: re-scheduling (elapsed=%dms, remaining=%dms, start_delay=%dms)", cmd->rsc_id, cmd->real_action, time_sum, timeout_left, delay); } else if (cmd->result.execution_status == PCMK_EXEC_PENDING) { crm_info("%s %s is still in progress: re-scheduling (elapsed=%dms, remaining=%dms, start_delay=%dms)", cmd->rsc_id, cmd->action, time_sum, timeout_left, delay); } else { crm_notice("%s %s failed: %s: Re-scheduling (remaining " "timeout %s) " QB_XS " exitstatus=%d elapsed=%dms start_delay=%dms)", cmd->rsc_id, cmd->action, crm_exit_str(cmd->result.exit_status), pcmk__readable_interval(timeout_left), cmd->result.exit_status, time_sum, delay); } cmd_reset(cmd); if(rsc) { rsc->active = NULL; } schedule_lrmd_cmd(rsc, cmd); /* Don't finalize cmd, we're not done with it yet */ return; } else { crm_notice("Giving up on %s %s (rc=%d): timeout (elapsed=%dms, remaining=%dms)", cmd->rsc_id, (cmd->real_action? cmd->real_action : cmd->action), cmd->result.exit_status, time_sum, timeout_left); pcmk__set_result(&(cmd->result), PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_TIMEOUT, "Investigate reason for timeout, and adjust " "configured operation timeout if necessary"); cmd_original_times(cmd); } } #endif pcmk__set_result_output(&(cmd->result), services__grab_stdout(action), services__grab_stderr(action)); cmd_finalize(cmd, rsc); } /*! * \internal * \brief Process the result of a fence device action (start, stop, or monitor) * * \param[in,out] cmd Fence device action that completed * \param[in] exit_status Fencer API exit status for action * \param[in] execution_status Fencer API execution status for action * \param[in] exit_reason Human-friendly detail, if action failed */ static void stonith_action_complete(lrmd_cmd_t *cmd, int exit_status, enum pcmk_exec_status execution_status, const char *exit_reason) { // This can be NULL if resource was removed before command completed lrmd_rsc_t *rsc = g_hash_table_lookup(rsc_list, cmd->rsc_id); // Simplify fencer exit status to uniform exit status if (exit_status != CRM_EX_OK) { exit_status = PCMK_OCF_UNKNOWN_ERROR; } if (cmd->result.execution_status == PCMK_EXEC_CANCELLED) { /* An in-flight fence action was cancelled. The execution status is * already correct, so don't overwrite it. */ execution_status = PCMK_EXEC_CANCELLED; } else { /* Some execution status codes have specific meanings for the fencer * that executor clients may not expect, so map them to a simple error * status. */ switch (execution_status) { case PCMK_EXEC_NOT_CONNECTED: case PCMK_EXEC_INVALID: execution_status = PCMK_EXEC_ERROR; break; case PCMK_EXEC_NO_FENCE_DEVICE: /* This should be possible only for probes in practice, but * interpret for all actions to be safe. */ if (pcmk__str_eq(cmd->action, PCMK_ACTION_MONITOR, pcmk__str_none)) { exit_status = PCMK_OCF_NOT_RUNNING; } else if (pcmk__str_eq(cmd->action, PCMK_ACTION_STOP, pcmk__str_none)) { exit_status = PCMK_OCF_OK; } else { exit_status = PCMK_OCF_NOT_INSTALLED; } execution_status = PCMK_EXEC_ERROR; break; case PCMK_EXEC_NOT_SUPPORTED: exit_status = PCMK_OCF_UNIMPLEMENT_FEATURE; break; default: break; } } pcmk__set_result(&cmd->result, exit_status, execution_status, exit_reason); // Certain successful actions change the known state of the resource if ((rsc != NULL) && pcmk__result_ok(&(cmd->result))) { if (pcmk__str_eq(cmd->action, PCMK_ACTION_START, pcmk__str_casei)) { pcmk__set_result(&rsc->fence_probe_result, CRM_EX_OK, PCMK_EXEC_DONE, NULL); // "running" } else if (pcmk__str_eq(cmd->action, PCMK_ACTION_STOP, pcmk__str_casei)) { pcmk__set_result(&rsc->fence_probe_result, CRM_EX_ERROR, PCMK_EXEC_NO_FENCE_DEVICE, NULL); // "not running" } } /* The recurring timer should not be running at this point in any case, but * as a failsafe, stop it if it is. */ stop_recurring_timer(cmd); /* Reschedule this command if appropriate. If a recurring command is *not* * rescheduled, its status must be PCMK_EXEC_CANCELLED, otherwise it will * not be removed from recurring_ops by cmd_finalize(). */ if (rsc && (cmd->interval_ms > 0) && (cmd->result.execution_status != PCMK_EXEC_CANCELLED)) { start_recurring_timer(cmd); } cmd_finalize(cmd, rsc); } static void lrmd_stonith_callback(stonith_t * stonith, stonith_callback_data_t * data) { if ((data == NULL) || (data->userdata == NULL)) { crm_err("Ignoring fence action result: " "Invalid callback arguments (bug?)"); } else { stonith_action_complete((lrmd_cmd_t *) data->userdata, stonith__exit_status(data), stonith__execution_status(data), stonith__exit_reason(data)); } } void stonith_connection_failed(void) { GHashTableIter iter; lrmd_rsc_t *rsc = NULL; crm_warn("Connection to fencer lost (any pending operations for " "fence devices will be considered failed)"); g_hash_table_iter_init(&iter, rsc_list); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &rsc)) { if (!pcmk__str_eq(rsc->class, PCMK_RESOURCE_CLASS_STONITH, pcmk__str_none)) { continue; } /* If we registered this fence device, we don't know whether the * fencer still has the registration or not. Cause future probes to * return an error until the resource is stopped or started * successfully. This is especially important if the controller also * went away (possibly due to a cluster layer restart) and won't * receive our client notification of any monitors finalized below. */ if (rsc->fence_probe_result.execution_status == PCMK_EXEC_DONE) { pcmk__set_result(&rsc->fence_probe_result, CRM_EX_ERROR, PCMK_EXEC_NOT_CONNECTED, "Lost connection to fencer"); } // Consider any active, pending, or recurring operations as failed for (GList *op = rsc->recurring_ops; op != NULL; op = op->next) { lrmd_cmd_t *cmd = op->data; /* This won't free a recurring op but instead restart its timer. * If cmd is rsc->active, this will set rsc->active to NULL, so we * don't have to worry about finalizing it a second time below. */ stonith_action_complete(cmd, CRM_EX_ERROR, PCMK_EXEC_NOT_CONNECTED, "Lost connection to fencer"); } if (rsc->active != NULL) { rsc->pending_ops = g_list_prepend(rsc->pending_ops, rsc->active); } while (rsc->pending_ops != NULL) { // This will free the op and remove it from rsc->pending_ops stonith_action_complete((lrmd_cmd_t *) rsc->pending_ops->data, CRM_EX_ERROR, PCMK_EXEC_NOT_CONNECTED, "Lost connection to fencer"); } } } /*! * \internal * \brief Execute a stonith resource "start" action * * Start a stonith resource by registering it with the fencer. * (Stonith agents don't have a start command.) * * \param[in,out] stonith_api Connection to fencer * \param[in] rsc Stonith resource to start * \param[in] cmd Start command to execute * * \return pcmk_ok on success, -errno otherwise */ static int execd_stonith_start(stonith_t *stonith_api, const lrmd_rsc_t *rsc, const lrmd_cmd_t *cmd) { char *key = NULL; char *value = NULL; stonith_key_value_t *device_params = NULL; int rc = pcmk_ok; // Convert command parameters to stonith API key/values if (cmd->params) { GHashTableIter iter; g_hash_table_iter_init(&iter, cmd->params); while (g_hash_table_iter_next(&iter, (gpointer *) & key, (gpointer *) & value)) { device_params = stonith_key_value_add(device_params, key, value); } } /* The fencer will automatically register devices via CIB notifications * when the CIB changes, but to avoid a possible race condition between * the fencer receiving the notification and the executor requesting that * resource, the executor registers the device as well. The fencer knows how * to handle duplicate registrations. */ rc = stonith_api->cmds->register_device(stonith_api, st_opt_sync_call, cmd->rsc_id, rsc->provider, rsc->type, device_params); stonith_key_value_freeall(device_params, 1, 1); return rc; } /*! * \internal * \brief Execute a stonith resource "stop" action * * Stop a stonith resource by unregistering it with the fencer. * (Stonith agents don't have a stop command.) * * \param[in,out] stonith_api Connection to fencer * \param[in] rsc Stonith resource to stop * * \return pcmk_ok on success, -errno otherwise */ static inline int execd_stonith_stop(stonith_t *stonith_api, const lrmd_rsc_t *rsc) { /* @TODO Failure would indicate a problem communicating with fencer; * perhaps we should try reconnecting and retrying a few times? */ return stonith_api->cmds->remove_device(stonith_api, st_opt_sync_call, rsc->rsc_id); } /*! * \internal * \brief Initiate a stonith resource agent recurring "monitor" action * * \param[in,out] stonith_api Connection to fencer * \param[in,out] rsc Stonith resource to monitor * \param[in] cmd Monitor command being executed * * \return pcmk_ok if monitor was successfully initiated, -errno otherwise */ static inline int execd_stonith_monitor(stonith_t *stonith_api, lrmd_rsc_t *rsc, lrmd_cmd_t *cmd) { int rc = stonith_api->cmds->monitor(stonith_api, 0, cmd->rsc_id, cmd->timeout / 1000); rc = stonith_api->cmds->register_callback(stonith_api, rc, 0, 0, cmd, "lrmd_stonith_callback", lrmd_stonith_callback); if (rc == TRUE) { rsc->active = cmd; rc = pcmk_ok; } else { rc = -pcmk_err_generic; } return rc; } static void execute_stonith_action(lrmd_rsc_t *rsc, lrmd_cmd_t *cmd) { int rc = 0; bool do_monitor = FALSE; stonith_t *stonith_api = get_stonith_connection(); if (pcmk__str_eq(cmd->action, PCMK_ACTION_MONITOR, pcmk__str_casei) && (cmd->interval_ms == 0)) { // Probes don't require a fencer connection stonith_action_complete(cmd, rsc->fence_probe_result.exit_status, rsc->fence_probe_result.execution_status, rsc->fence_probe_result.exit_reason); return; } else if (stonith_api == NULL) { stonith_action_complete(cmd, PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_NOT_CONNECTED, "No connection to fencer"); return; } else if (pcmk__str_eq(cmd->action, PCMK_ACTION_START, pcmk__str_casei)) { rc = execd_stonith_start(stonith_api, rsc, cmd); if (rc == pcmk_ok) { do_monitor = TRUE; } } else if (pcmk__str_eq(cmd->action, PCMK_ACTION_STOP, pcmk__str_casei)) { rc = execd_stonith_stop(stonith_api, rsc); } else if (pcmk__str_eq(cmd->action, PCMK_ACTION_MONITOR, pcmk__str_casei)) { do_monitor = TRUE; } else { stonith_action_complete(cmd, PCMK_OCF_UNIMPLEMENT_FEATURE, PCMK_EXEC_ERROR, "Invalid fence device action (bug?)"); return; } if (do_monitor) { rc = execd_stonith_monitor(stonith_api, rsc, cmd); if (rc == pcmk_ok) { // Don't clean up yet, we will find out result of the monitor later return; } } stonith_action_complete(cmd, ((rc == pcmk_ok)? CRM_EX_OK : CRM_EX_ERROR), stonith__legacy2status(rc), ((rc == -pcmk_err_generic)? NULL : pcmk_strerror(rc))); } static void execute_nonstonith_action(lrmd_rsc_t *rsc, lrmd_cmd_t *cmd) { svc_action_t *action = NULL; GHashTable *params_copy = NULL; pcmk__assert((rsc != NULL) && (cmd != NULL)); crm_trace("Creating action, resource:%s action:%s class:%s provider:%s agent:%s", rsc->rsc_id, cmd->action, rsc->class, rsc->provider, rsc->type); params_copy = pcmk__str_table_dup(cmd->params); action = services__create_resource_action(rsc->rsc_id, rsc->class, rsc->provider, rsc->type, normalize_action_name(rsc, cmd->action), cmd->interval_ms, cmd->timeout, params_copy, cmd->service_flags); if (action == NULL) { pcmk__set_result(&(cmd->result), PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_ERROR, strerror(ENOMEM)); cmd_finalize(cmd, rsc); return; } if (action->rc != PCMK_OCF_UNKNOWN) { pcmk__set_result(&(cmd->result), action->rc, action->status, services__exit_reason(action)); services_action_free(action); cmd_finalize(cmd, rsc); return; } action->cb_data = cmd; if (services_action_async(action, action_complete)) { /* The services library has taken responsibility for the action. It * could be pending, blocked, or merged into a duplicate recurring * action, in which case the action callback (action_complete()) * will be called when the action completes, otherwise the callback has * already been called. * * action_complete() calls cmd_finalize() which can free cmd, so cmd * cannot be used here. */ } else { /* This is a recurring action that is not being cancelled and could not * be initiated. It has been rescheduled, and the action callback * (action_complete()) has been called, which in this case has already * called cmd_finalize(), which in this case should only reset (not * free) cmd. */ pcmk__set_result(&(cmd->result), action->rc, action->status, services__exit_reason(action)); services_action_free(action); } } static gboolean execute_resource_action(gpointer user_data) { lrmd_rsc_t *rsc = (lrmd_rsc_t *) user_data; lrmd_cmd_t *cmd = NULL; CRM_CHECK(rsc != NULL, return FALSE); if (rsc->active) { crm_trace("%s is still active", rsc->rsc_id); return TRUE; } if (rsc->pending_ops) { GList *first = rsc->pending_ops; cmd = first->data; if (cmd->delay_id) { crm_trace ("Command %s %s was asked to run too early, waiting for start_delay timeout of %dms", cmd->rsc_id, cmd->action, cmd->start_delay); return TRUE; } rsc->pending_ops = g_list_remove_link(rsc->pending_ops, first); g_list_free_1(first); #ifdef PCMK__TIME_USE_CGT get_current_time(&(cmd->t_run), &(cmd->t_first_run)); #endif cmd->epoch_last_run = time(NULL); } if (!cmd) { crm_trace("Nothing further to do for %s", rsc->rsc_id); return TRUE; } rsc->active = cmd; /* only one op at a time for a rsc */ if (cmd->interval_ms) { rsc->recurring_ops = g_list_append(rsc->recurring_ops, cmd); } log_execute(cmd); if (pcmk__str_eq(rsc->class, PCMK_RESOURCE_CLASS_STONITH, pcmk__str_casei)) { execute_stonith_action(rsc, cmd); } else { execute_nonstonith_action(rsc, cmd); } return TRUE; } void free_rsc(gpointer data) { GList *gIter = NULL; lrmd_rsc_t *rsc = data; int is_stonith = pcmk__str_eq(rsc->class, PCMK_RESOURCE_CLASS_STONITH, pcmk__str_casei); gIter = rsc->pending_ops; while (gIter != NULL) { GList *next = gIter->next; lrmd_cmd_t *cmd = gIter->data; /* command was never executed */ cmd->result.execution_status = PCMK_EXEC_CANCELLED; cmd_finalize(cmd, NULL); gIter = next; } /* frees list, but not list elements. */ g_list_free(rsc->pending_ops); gIter = rsc->recurring_ops; while (gIter != NULL) { GList *next = gIter->next; lrmd_cmd_t *cmd = gIter->data; if (is_stonith) { cmd->result.execution_status = PCMK_EXEC_CANCELLED; /* If a stonith command is in-flight, just mark it as cancelled; * it is not safe to finalize/free the cmd until the stonith api * says it has either completed or timed out. */ if (rsc->active != cmd) { cmd_finalize(cmd, NULL); } } else { /* This command is already handed off to service library, * let service library cancel it and tell us via the callback * when it is cancelled. The rsc can be safely destroyed * even if we are waiting for the cancel result */ services_action_cancel(rsc->rsc_id, normalize_action_name(rsc, cmd->action), cmd->interval_ms); } gIter = next; } /* frees list, but not list elements. */ g_list_free(rsc->recurring_ops); free(rsc->rsc_id); free(rsc->class); free(rsc->provider); free(rsc->type); mainloop_destroy_trigger(rsc->work); free(rsc); } static int process_lrmd_signon(pcmk__client_t *client, xmlNode *request, int call_id, xmlNode **reply) { int rc = pcmk_ok; time_t now = time(NULL); const char *protocol_version = crm_element_value(request, PCMK__XA_LRMD_PROTOCOL_VERSION); const char *start_state = pcmk__env_option(PCMK__ENV_NODE_START_STATE); if (compare_version(protocol_version, LRMD_COMPATIBLE_PROTOCOL) < 0) { crm_err("Cluster API version must be greater than or equal to %s, not %s", LRMD_COMPATIBLE_PROTOCOL, protocol_version); rc = -EPROTO; } if (pcmk__xe_attr_is_true(request, PCMK__XA_LRMD_IS_IPC_PROVIDER)) { #ifdef PCMK__COMPILE_REMOTE if ((client->remote != NULL) && pcmk_is_set(client->flags, pcmk__client_tls_handshake_complete)) { const char *op = crm_element_value(request, PCMK__XA_LRMD_OP); // This is a remote connection from a cluster node's controller ipc_proxy_add_provider(client); /* If this was a register operation, also ask for new schema files but * only if it's supported by the protocol version. */ if (pcmk__str_eq(op, CRM_OP_REGISTER, pcmk__str_none) && LRMD_SUPPORTS_SCHEMA_XFER(protocol_version)) { remoted_request_cib_schema_files(); } } else { rc = -EACCES; } #else rc = -EPROTONOSUPPORT; #endif } *reply = create_lrmd_reply(__func__, rc, call_id); crm_xml_add(*reply, PCMK__XA_LRMD_OP, CRM_OP_REGISTER); crm_xml_add(*reply, PCMK__XA_LRMD_CLIENTID, client->id); crm_xml_add(*reply, PCMK__XA_LRMD_PROTOCOL_VERSION, LRMD_PROTOCOL_VERSION); crm_xml_add_ll(*reply, PCMK__XA_UPTIME, now - start_time); if (start_state) { crm_xml_add(*reply, PCMK__XA_NODE_START_STATE, start_state); } return rc; } static int process_lrmd_rsc_register(pcmk__client_t *client, uint32_t id, xmlNode *request) { int rc = pcmk_ok; lrmd_rsc_t *rsc = build_rsc_from_xml(request); lrmd_rsc_t *dup = g_hash_table_lookup(rsc_list, rsc->rsc_id); if (dup && pcmk__str_eq(rsc->class, dup->class, pcmk__str_casei) && pcmk__str_eq(rsc->provider, dup->provider, pcmk__str_casei) && pcmk__str_eq(rsc->type, dup->type, pcmk__str_casei)) { crm_notice("Ignoring duplicate registration of '%s'", rsc->rsc_id); free_rsc(rsc); return rc; } g_hash_table_replace(rsc_list, rsc->rsc_id, rsc); crm_info("Cached agent information for '%s'", rsc->rsc_id); return rc; } static xmlNode * process_lrmd_get_rsc_info(xmlNode *request, int call_id) { int rc = pcmk_ok; xmlNode *rsc_xml = get_xpath_object("//" PCMK__XE_LRMD_RSC, request, LOG_ERR); const char *rsc_id = crm_element_value(rsc_xml, PCMK__XA_LRMD_RSC_ID); xmlNode *reply = NULL; lrmd_rsc_t *rsc = NULL; if (rsc_id == NULL) { rc = -ENODEV; } else { rsc = g_hash_table_lookup(rsc_list, rsc_id); if (rsc == NULL) { crm_info("Agent information for '%s' not in cache", rsc_id); rc = -ENODEV; } } reply = create_lrmd_reply(__func__, rc, call_id); if (rsc) { crm_xml_add(reply, PCMK__XA_LRMD_RSC_ID, rsc->rsc_id); crm_xml_add(reply, PCMK__XA_LRMD_CLASS, rsc->class); crm_xml_add(reply, PCMK__XA_LRMD_PROVIDER, rsc->provider); crm_xml_add(reply, PCMK__XA_LRMD_TYPE, rsc->type); } return reply; } static int process_lrmd_rsc_unregister(pcmk__client_t *client, uint32_t id, xmlNode *request) { int rc = pcmk_ok; lrmd_rsc_t *rsc = NULL; xmlNode *rsc_xml = get_xpath_object("//" PCMK__XE_LRMD_RSC, request, LOG_ERR); const char *rsc_id = crm_element_value(rsc_xml, PCMK__XA_LRMD_RSC_ID); if (!rsc_id) { return -ENODEV; } rsc = g_hash_table_lookup(rsc_list, rsc_id); if (rsc == NULL) { crm_info("Ignoring unregistration of resource '%s', which is not registered", rsc_id); return pcmk_ok; } if (rsc->active) { /* let the caller know there are still active ops on this rsc to watch for */ crm_trace("Operation (%p) still in progress for unregistered resource %s", rsc->active, rsc_id); rc = -EINPROGRESS; } g_hash_table_remove(rsc_list, rsc_id); return rc; } static int process_lrmd_rsc_exec(pcmk__client_t *client, uint32_t id, xmlNode *request) { lrmd_rsc_t *rsc = NULL; lrmd_cmd_t *cmd = NULL; xmlNode *rsc_xml = get_xpath_object("//" PCMK__XE_LRMD_RSC, request, LOG_ERR); const char *rsc_id = crm_element_value(rsc_xml, PCMK__XA_LRMD_RSC_ID); int call_id; if (!rsc_id) { return -EINVAL; } if (!(rsc = g_hash_table_lookup(rsc_list, rsc_id))) { crm_info("Resource '%s' not found (%d active resources)", rsc_id, g_hash_table_size(rsc_list)); return -ENODEV; } cmd = create_lrmd_cmd(request, client); call_id = cmd->call_id; /* Don't reference cmd after handing it off to be scheduled. * The cmd could get merged and freed. */ schedule_lrmd_cmd(rsc, cmd); return call_id; } static int cancel_op(const char *rsc_id, const char *action, guint interval_ms) { GList *gIter = NULL; lrmd_rsc_t *rsc = g_hash_table_lookup(rsc_list, rsc_id); /* How to cancel an action. * 1. Check pending ops list, if it hasn't been handed off * to the service library or stonith recurring list remove * it there and that will stop it. * 2. If it isn't in the pending ops list, then it's either a * recurring op in the stonith recurring list, or the service * library's recurring list. Stop it there * 3. If not found in any lists, then this operation has either * been executed already and is not a recurring operation, or * never existed. */ if (!rsc) { return -ENODEV; } for (gIter = rsc->pending_ops; gIter != NULL; gIter = gIter->next) { lrmd_cmd_t *cmd = gIter->data; if (action_matches(cmd, action, interval_ms)) { cmd->result.execution_status = PCMK_EXEC_CANCELLED; cmd_finalize(cmd, rsc); return pcmk_ok; } } if (pcmk__str_eq(rsc->class, PCMK_RESOURCE_CLASS_STONITH, pcmk__str_casei)) { /* The service library does not handle stonith operations. * We have to handle recurring stonith operations ourselves. */ for (gIter = rsc->recurring_ops; gIter != NULL; gIter = gIter->next) { lrmd_cmd_t *cmd = gIter->data; if (action_matches(cmd, action, interval_ms)) { cmd->result.execution_status = PCMK_EXEC_CANCELLED; if (rsc->active != cmd) { cmd_finalize(cmd, rsc); } return pcmk_ok; } } } else if (services_action_cancel(rsc_id, normalize_action_name(rsc, action), interval_ms) == TRUE) { /* The service library will tell the action_complete callback function * this action was cancelled, which will destroy the cmd and remove * it from the recurring_op list. Do not do that in this function * if the service library says it cancelled it. */ return pcmk_ok; } return -EOPNOTSUPP; } static void cancel_all_recurring(lrmd_rsc_t * rsc, const char *client_id) { GList *cmd_list = NULL; GList *cmd_iter = NULL; /* Notice a copy of each list is created when concat is called. * This prevents odd behavior from occurring when the cmd_list * is iterated through later on. It is possible the cancel_op * function may end up modifying the recurring_ops and pending_ops * lists. If we did not copy those lists, our cmd_list iteration * could get messed up.*/ if (rsc->recurring_ops) { cmd_list = g_list_concat(cmd_list, g_list_copy(rsc->recurring_ops)); } if (rsc->pending_ops) { cmd_list = g_list_concat(cmd_list, g_list_copy(rsc->pending_ops)); } if (!cmd_list) { return; } for (cmd_iter = cmd_list; cmd_iter; cmd_iter = cmd_iter->next) { lrmd_cmd_t *cmd = cmd_iter->data; if (cmd->interval_ms == 0) { continue; } if (client_id && !pcmk__str_eq(cmd->client_id, client_id, pcmk__str_casei)) { continue; } cancel_op(rsc->rsc_id, cmd->action, cmd->interval_ms); } /* frees only the copied list data, not the cmds */ g_list_free(cmd_list); } static int process_lrmd_rsc_cancel(pcmk__client_t *client, uint32_t id, xmlNode *request) { xmlNode *rsc_xml = get_xpath_object("//" PCMK__XE_LRMD_RSC, request, LOG_ERR); const char *rsc_id = crm_element_value(rsc_xml, PCMK__XA_LRMD_RSC_ID); const char *action = crm_element_value(rsc_xml, PCMK__XA_LRMD_RSC_ACTION); guint interval_ms = 0; crm_element_value_ms(rsc_xml, PCMK__XA_LRMD_RSC_INTERVAL, &interval_ms); if (!rsc_id || !action) { return -EINVAL; } return cancel_op(rsc_id, action, interval_ms); } static void add_recurring_op_xml(xmlNode *reply, lrmd_rsc_t *rsc) { xmlNode *rsc_xml = pcmk__xe_create(reply, PCMK__XE_LRMD_RSC); crm_xml_add(rsc_xml, PCMK__XA_LRMD_RSC_ID, rsc->rsc_id); for (GList *item = rsc->recurring_ops; item != NULL; item = item->next) { lrmd_cmd_t *cmd = item->data; xmlNode *op_xml = pcmk__xe_create(rsc_xml, PCMK__XE_LRMD_RSC_OP); crm_xml_add(op_xml, PCMK__XA_LRMD_RSC_ACTION, pcmk__s(cmd->real_action, cmd->action)); crm_xml_add_ms(op_xml, PCMK__XA_LRMD_RSC_INTERVAL, cmd->interval_ms); crm_xml_add_int(op_xml, PCMK__XA_LRMD_TIMEOUT, cmd->timeout_orig); } } static xmlNode * process_lrmd_get_recurring(xmlNode *request, int call_id) { int rc = pcmk_ok; const char *rsc_id = NULL; lrmd_rsc_t *rsc = NULL; xmlNode *reply = NULL; xmlNode *rsc_xml = NULL; // Resource ID is optional rsc_xml = pcmk__xe_first_child(request, PCMK__XE_LRMD_CALLDATA, NULL, NULL); if (rsc_xml) { rsc_xml = pcmk__xe_first_child(rsc_xml, PCMK__XE_LRMD_RSC, NULL, NULL); } if (rsc_xml) { rsc_id = crm_element_value(rsc_xml, PCMK__XA_LRMD_RSC_ID); } // If resource ID is specified, resource must exist if (rsc_id != NULL) { rsc = g_hash_table_lookup(rsc_list, rsc_id); if (rsc == NULL) { crm_info("Resource '%s' not found (%d active resources)", rsc_id, g_hash_table_size(rsc_list)); rc = -ENODEV; } } reply = create_lrmd_reply(__func__, rc, call_id); // If resource ID is not specified, check all resources if (rsc_id == NULL) { GHashTableIter iter; char *key = NULL; g_hash_table_iter_init(&iter, rsc_list); while (g_hash_table_iter_next(&iter, (gpointer *) &key, (gpointer *) &rsc)) { add_recurring_op_xml(reply, rsc); } } else if (rsc) { add_recurring_op_xml(reply, rsc); } return reply; } void process_lrmd_message(pcmk__client_t *client, uint32_t id, xmlNode *request) { int rc = pcmk_ok; int call_id = 0; const char *op = crm_element_value(request, PCMK__XA_LRMD_OP); int do_reply = 0; int do_notify = 0; xmlNode *reply = NULL; /* Certain IPC commands may be done only by privileged users (i.e. root or * hacluster), because they would otherwise provide a means of bypassing * ACLs. */ bool allowed = pcmk_is_set(client->flags, pcmk__client_privileged); crm_trace("Processing %s operation from %s", op, client->id); crm_element_value_int(request, PCMK__XA_LRMD_CALLID, &call_id); if (pcmk__str_eq(op, CRM_OP_IPC_FWD, pcmk__str_none)) { #ifdef PCMK__COMPILE_REMOTE if (allowed) { ipc_proxy_forward_client(client, request); } else { rc = -EACCES; } #else rc = -EPROTONOSUPPORT; #endif do_reply = 1; } else if (pcmk__str_eq(op, CRM_OP_REGISTER, pcmk__str_none)) { rc = process_lrmd_signon(client, request, call_id, &reply); do_reply = 1; } else if (pcmk__str_eq(op, LRMD_OP_RSC_REG, pcmk__str_none)) { if (allowed) { rc = process_lrmd_rsc_register(client, id, request); do_notify = 1; } else { rc = -EACCES; } do_reply = 1; } else if (pcmk__str_eq(op, LRMD_OP_RSC_INFO, pcmk__str_none)) { if (allowed) { reply = process_lrmd_get_rsc_info(request, call_id); } else { rc = -EACCES; } do_reply = 1; } else if (pcmk__str_eq(op, LRMD_OP_RSC_UNREG, pcmk__str_none)) { if (allowed) { rc = process_lrmd_rsc_unregister(client, id, request); /* don't notify anyone about failed un-registers */ if (rc == pcmk_ok || rc == -EINPROGRESS) { do_notify = 1; } } else { rc = -EACCES; } do_reply = 1; } else if (pcmk__str_eq(op, LRMD_OP_RSC_EXEC, pcmk__str_none)) { if (allowed) { rc = process_lrmd_rsc_exec(client, id, request); } else { rc = -EACCES; } do_reply = 1; } else if (pcmk__str_eq(op, LRMD_OP_RSC_CANCEL, pcmk__str_none)) { if (allowed) { rc = process_lrmd_rsc_cancel(client, id, request); } else { rc = -EACCES; } do_reply = 1; } else if (pcmk__str_eq(op, LRMD_OP_POKE, pcmk__str_none)) { do_notify = 1; do_reply = 1; } else if (pcmk__str_eq(op, LRMD_OP_CHECK, pcmk__str_none)) { if (allowed) { xmlNode *wrapper = pcmk__xe_first_child(request, PCMK__XE_LRMD_CALLDATA, NULL, NULL); xmlNode *data = pcmk__xe_first_child(wrapper, NULL, NULL, NULL); const char *timeout = NULL; CRM_LOG_ASSERT(data != NULL); timeout = crm_element_value(data, PCMK__XA_LRMD_WATCHDOG); pcmk__valid_stonith_watchdog_timeout(timeout); } else { rc = -EACCES; } } else if (pcmk__str_eq(op, LRMD_OP_ALERT_EXEC, pcmk__str_none)) { if (allowed) { rc = process_lrmd_alert_exec(client, id, request); } else { rc = -EACCES; } do_reply = 1; } else if (pcmk__str_eq(op, LRMD_OP_GET_RECURRING, pcmk__str_none)) { if (allowed) { reply = process_lrmd_get_recurring(request, call_id); } else { rc = -EACCES; } do_reply = 1; } else { rc = -EOPNOTSUPP; do_reply = 1; crm_err("Unknown IPC request '%s' from client %s", op, pcmk__client_name(client)); } if (rc == -EACCES) { crm_warn("Rejecting IPC request '%s' from unprivileged client %s", op, pcmk__client_name(client)); } crm_debug("Processed %s operation from %s: rc=%d, reply=%d, notify=%d", op, client->id, rc, do_reply, do_notify); if (do_reply) { int send_rc = pcmk_rc_ok; if (reply == NULL) { reply = create_lrmd_reply(__func__, rc, call_id); } send_rc = lrmd_server_send_reply(client, id, reply); pcmk__xml_free(reply); if (send_rc != pcmk_rc_ok) { crm_warn("Reply to client %s failed: %s " QB_XS " rc=%d", pcmk__client_name(client), pcmk_rc_str(send_rc), send_rc); } } if (do_notify) { send_generic_notify(rc, request); } } diff --git a/include/crm/common/xml_names.h b/include/crm/common/xml_names.h index 6dfd23af8b..ba067b7be2 100644 --- a/include/crm/common/xml_names.h +++ b/include/crm/common/xml_names.h @@ -1,463 +1,464 @@ /* * 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. */ #ifndef PCMK__CRM_COMMON_XML_NAMES__H #define PCMK__CRM_COMMON_XML_NAMES__H #ifdef __cplusplus extern "C" { #endif /*! * \file * \brief Defined string constants for XML element and attribute names * \ingroup core */ /* For consistency, new constants should start with "PCMK_", followed by: * - "XE" for XML element names * - "XA" for XML attribute names * - "OPT" for cluster option (property) names * - "META" for meta-attribute names * - "VALUE" for enumerated values (such as for options or for XML attributes) * - "NODE_ATTR" for node attribute names * * Old names that don't follow this policy should eventually be deprecated and * replaced with names that do. * * Symbols should be public if the user may specify them somewhere (especially * the CIB) or if they're part of a well-defined structure that a user may need * to parse. They should be internal if they're used only internally to * Pacemaker (such as daemon IPC/CPG message XML). * * Constants belong in the following locations: * * "XE" and "XA": xml_names.h and xml_names_internal.h * * "OPT", "META", and "VALUE": options.h and options_internal.h * * "NODE_ATTR": nodes.h and nodes_internal.h * * For meta-attributes that can be specified as either XML attributes or nvpair * names, use "META" unless using both "XA" and "META" constants adds clarity. * An example is operation attributes, which can be specified either as * attributes of the PCMK_XE_OP element or as nvpairs in a meta-attribute set * beneath the PCMK_XE_OP element. */ /* * XML elements */ #define PCMK_XE_ACL_GROUP "acl_group" #define PCMK_XE_ACL_PERMISSION "acl_permission" #define PCMK_XE_ACL_ROLE "acl_role" #define PCMK_XE_ACL_TARGET "acl_target" #define PCMK_XE_ACLS "acls" #define PCMK_XE_ACTION "action" #define PCMK_XE_ACTIONS "actions" #define PCMK_XE_AGENT "agent" #define PCMK_XE_AGENT_STATUS "agent-status" #define PCMK_XE_AGENTS "agents" #define PCMK_XE_ALERT "alert" #define PCMK_XE_ALERTS "alerts" #define PCMK_XE_ALLOCATIONS "allocations" #define PCMK_XE_ALLOCATIONS_UTILIZATIONS "allocations_utilizations" #define PCMK_XE_ATTRIBUTE "attribute" #define PCMK_XE_BAN "ban" #define PCMK_XE_BANS "bans" #define PCMK_XE_BUNDLE "bundle" #define PCMK_XE_CAPACITY "capacity" #define PCMK_XE_CHANGE "change" #define PCMK_XE_CHANGE_ATTR "change-attr" #define PCMK_XE_CHANGE_LIST "change-list" #define PCMK_XE_CHANGE_RESULT "change-result" #define PCMK_XE_CHECK "check" #define PCMK_XE_CIB "cib" #define PCMK_XE_CLONE "clone" #define PCMK_XE_CLUSTER_ACTION "cluster_action" #define PCMK_XE_CLUSTER_INFO "cluster-info" #define PCMK_XE_CLUSTER_OPTIONS "cluster_options" #define PCMK_XE_CLUSTER_PROPERTY_SET "cluster_property_set" #define PCMK_XE_CLUSTER_STATUS "cluster_status" #define PCMK_XE_COMMAND "command" #define PCMK_XE_CONFIGURATION "configuration" #define PCMK_XE_CONSTRAINT "constraint" #define PCMK_XE_CONSTRAINTS "constraints" #define PCMK_XE_CONTENT "content" #define PCMK_XE_CRM_CONFIG "crm_config" #define PCMK_XE_CRM_MON "crm_mon" #define PCMK_XE_CRM_MON_DISCONNECTED "crm-mon-disconnected" #define PCMK_XE_CURRENT_DC "current_dc" #define PCMK_XE_DATE "date" #define PCMK_XE_DATE_EXPRESSION "date_expression" #define PCMK_XE_DATE_SPEC "date_spec" #define PCMK_XE_DC "dc" #define PCMK_XE_DEPRECATED "deprecated" #define PCMK_XE_DIFF "diff" #define PCMK_XE_DIGEST "digest" #define PCMK_XE_DIGESTS "digests" #define PCMK_XE_DOCKER "docker" #define PCMK_XE_DURATION "duration" #define PCMK_XE_DURATION_ENDS "duration_ends" #define PCMK_XE_END "end" #define PCMK_XE_ERROR "error" #define PCMK_XE_ERRORS "errors" #define PCMK_XE_EXPRESSION "expression" #define PCMK_XE_FAILURE "failure" #define PCMK_XE_FAILURES "failures" #define PCMK_XE_FEATURE "feature" #define PCMK_XE_FEATURES "features" #define PCMK_XE_FENCE_EVENT "fence_event" #define PCMK_XE_FENCE_HISTORY "fence_history" #define PCMK_XE_FENCING_ACTION "fencing_action" #define PCMK_XE_FENCING_LEVEL "fencing-level" #define PCMK_XE_FENCING_TOPOLOGY "fencing-topology" #define PCMK_XE_GROUP "group" #define PCMK_XE_INJECT_ATTR "inject_attr" #define PCMK_XE_INJECT_SPEC "inject_spec" #define PCMK_XE_INSTANCE_ATTRIBUTES "instance_attributes" #define PCMK_XE_INSTRUCTION "instruction" #define PCMK_XE_ITEM "item" #define PCMK_XE_LAST_CHANGE "last_change" #define PCMK_XE_LAST_FENCED "last-fenced" #define PCMK_XE_LAST_UPDATE "last_update" #define PCMK_XE_LIST "list" #define PCMK_XE_LONGDESC "longdesc" #define PCMK_XE_META_ATTRIBUTES "meta_attributes" #define PCMK_XE_METADATA "metadata" #define PCMK_XE_MODIFICATIONS "modifications" #define PCMK_XE_MODIFY_NODE "modify_node" #define PCMK_XE_MODIFY_TICKET "modify_ticket" #define PCMK_XE_NETWORK "network" #define PCMK_XE_NODE "node" #define PCMK_XE_NODE_ACTION "node_action" #define PCMK_XE_NODE_ATTRIBUTES "node_attributes" #define PCMK_XE_NODE_HISTORY "node_history" #define PCMK_XE_NODE_INFO "node-info" #define PCMK_XE_NODE_WEIGHT "node_weight" #define PCMK_XE_NODES "nodes" #define PCMK_XE_NODES_CONFIGURED "nodes_configured" #define PCMK_XE_NVPAIR "nvpair" #define PCMK_XE_OBJ_REF "obj_ref" #define PCMK_XE_OP "op" #define PCMK_XE_OP_DEFAULTS "op_defaults" #define PCMK_XE_OP_EXPRESSION "op_expression" #define PCMK_XE_OPERATION "operation" #define PCMK_XE_OPERATION_HISTORY "operation_history" #define PCMK_XE_OPERATIONS "operations" #define PCMK_XE_OPTION "option" #define PCMK_XE_OUTPUT "output" #define PCMK_XE_OVERRIDE "override" #define PCMK_XE_OVERRIDES "overrides" #define PCMK_XE_PACEMAKER_RESULT "pacemaker-result" #define PCMK_XE_PACEMAKERD "pacemakerd" #define PCMK_XE_PARAMETER "parameter" #define PCMK_XE_PARAMETERS "parameters" #define PCMK_XE_PERIOD "period" #define PCMK_XE_PODMAN "podman" #define PCMK_XE_PORT_MAPPING "port-mapping" #define PCMK_XE_POSITION "position" #define PCMK_XE_PRIMITIVE "primitive" #define PCMK_XE_PROMOTION_SCORE "promotion_score" #define PCMK_XE_PROVIDER "provider" #define PCMK_XE_PROVIDERS "providers" #define PCMK_XE_PSEUDO_ACTION "pseudo_action" #define PCMK_XE_REASON "reason" #define PCMK_XE_RECIPIENT "recipient" #define PCMK_XE_REPLICA "replica" #define PCMK_XE_RESOURCE "resource" #define PCMK_XE_RESOURCE_AGENT "resource-agent" #define PCMK_XE_RESOURCE_AGENT_ACTION "resource-agent-action" #define PCMK_XE_RESOURCE_CONFIG "resource_config" #define PCMK_XE_RESOURCE_HISTORY "resource_history" #define PCMK_XE_RESOURCE_REF "resource_ref" #define PCMK_XE_RESOURCE_SET "resource_set" #define PCMK_XE_RESOURCES "resources" #define PCMK_XE_RESOURCES_CONFIGURED "resources_configured" #define PCMK_XE_RESULT_CODE "result-code" #define PCMK_XE_REVISED_CLUSTER_STATUS "revised_cluster_status" #define PCMK_XE_ROLE "role" #define PCMK_XE_RSC_ACTION "rsc_action" #define PCMK_XE_RSC_COLOCATION "rsc_colocation" #define PCMK_XE_RSC_DEFAULTS "rsc_defaults" #define PCMK_XE_RSC_EXPRESSION "rsc_expression" #define PCMK_XE_RSC_LOCATION "rsc_location" #define PCMK_XE_RSC_ORDER "rsc_order" #define PCMK_XE_RSC_TICKET "rsc_ticket" #define PCMK_XE_RULE "rule" #define PCMK_XE_RULE_CHECK "rule-check" #define PCMK_XE_SELECT "select" #define PCMK_XE_SELECT_ATTRIBUTES "select_attributes" #define PCMK_XE_SELECT_FENCING "select_fencing" #define PCMK_XE_SELECT_NODES "select_nodes" #define PCMK_XE_SELECT_RESOURCES "select_resources" #define PCMK_XE_SHADOW "shadow" #define PCMK_XE_SHORTDESC "shortdesc" #define PCMK_XE_SOURCE "source" #define PCMK_XE_SPECIAL "special" #define PCMK_XE_STACK "stack" #define PCMK_XE_START "start" #define PCMK_XE_STATUS "status" #define PCMK_XE_STORAGE "storage" #define PCMK_XE_STORAGE_MAPPING "storage-mapping" #define PCMK_XE_SUMMARY "summary" #define PCMK_XE_TAG "tag" #define PCMK_XE_TAGS "tags" #define PCMK_XE_TARGET "target" #define PCMK_XE_TEMPLATE "template" #define PCMK_XE_TICKET "ticket" #define PCMK_XE_TICKETS "tickets" #define PCMK_XE_TIMING "timing" #define PCMK_XE_TIMINGS "timings" #define PCMK_XE_TRANSITION "transition" #define PCMK_XE_UTILIZATION "utilization" #define PCMK_XE_UTILIZATIONS "utilizations" #define PCMK_XE_VALIDATE "validate" #define PCMK_XE_VERSION "version" #define PCMK_XE_XML "xml" #define PCMK_XE_XML_PATCHSET "xml-patchset" /* * XML attributes */ #define PCMK_XA_ACTION "action" #define PCMK_XA_ACTIVE "active" #define PCMK_XA_ADD_HOST "add-host" #define PCMK_XA_ADMIN_EPOCH "admin_epoch" #define PCMK_XA_ADVANCED "advanced" #define PCMK_XA_AGENT "agent" #define PCMK_XA_API_VERSION "api-version" #define PCMK_XA_ATTRIBUTE "attribute" #define PCMK_XA_AUTHOR "author" #define PCMK_XA_AUTOMATIC "automatic" #define PCMK_XA_BLOCKED "blocked" #define PCMK_XA_BOOLEAN_OP "boolean-op" #define PCMK_XA_BUILD "build" #define PCMK_XA_CACHED "cached" #define PCMK_XA_CALL "call" #define PCMK_XA_CIB_LAST_WRITTEN "cib-last-written" #define PCMK_XA_CIB_NODE "cib_node" #define PCMK_XA_CLASS "class" #define PCMK_XA_CLIENT "client" #define PCMK_XA_CODE "code" #define PCMK_XA_COMMENT "comment" #define PCMK_XA_COMPLETED "completed" #define PCMK_XA_CONTROL_PORT "control-port" #define PCMK_XA_COUNT "count" #define PCMK_XA_CRM_DEBUG_ORIGIN "crm-debug-origin" #define PCMK_XA_CRM_FEATURE_SET "crm_feature_set" #define PCMK_XA_CRM_TIMESTAMP "crm-timestamp" #define PCMK_XA_CRMD "crmd" #define PCMK_XA_DAYS "days" #define PCMK_XA_DC_UUID "dc-uuid" #define PCMK_XA_DEFAULT "default" #define PCMK_XA_DELEGATE "delegate" #define PCMK_XA_DESCRIPTION "description" #define PCMK_XA_DEST "dest" #define PCMK_XA_DEVICE "device" #define PCMK_XA_DEVICES "devices" #define PCMK_XA_DISABLED "disabled" #define PCMK_XA_DURATION "duration" #define PCMK_XA_END "end" #define PCMK_XA_EPOCH "epoch" #define PCMK_XA_EXEC "exec" #define PCMK_XA_EXEC_TIME "exec-time" #define PCMK_XA_EXECUTION_CODE "execution_code" #define PCMK_XA_EXECUTION_DATE "execution-date" #define PCMK_XA_EXECUTION_MESSAGE "execution_message" #define PCMK_XA_EXIT_REASON "exit-reason" #define PCMK_XA_EXITCODE "exitcode" #define PCMK_XA_EXITREASON "exitreason" #define PCMK_XA_EXITSTATUS "exitstatus" #define PCMK_XA_EXPECTED "expected" #define PCMK_XA_EXPECTED_UP "expected_up" #define PCMK_XA_EXPIRES "expires" #define PCMK_XA_EXTENDED_STATUS "extended-status" #define PCMK_XA_FAIL_COUNT "fail-count" #define PCMK_XA_FAILED "failed" #define PCMK_XA_FAILURE_IGNORED "failure_ignored" #define PCMK_XA_FEATURE_SET "feature_set" #define PCMK_XA_FEATURES "features" #define PCMK_XA_FILE "file" #define PCMK_XA_FIRST "first" #define PCMK_XA_FIRST_ACTION "first-action" #define PCMK_XA_FOR "for" #define PCMK_XA_FORMAT "format" #define PCMK_XA_FUNCTION "function" #define PCMK_XA_GENERATED "generated" #define PCMK_XA_HASH "hash" #define PCMK_XA_HAVE_QUORUM "have-quorum" #define PCMK_XA_HEALTH "health" #define PCMK_XA_HOST "host" #define PCMK_XA_HOST_INTERFACE "host-interface" #define PCMK_XA_HOST_NETMASK "host-netmask" #define PCMK_XA_HOURS "hours" #define PCMK_XA_ID "id" #define PCMK_XA_ID_AS_RESOURCE "id_as_resource" #define PCMK_XA_ID_REF "id-ref" #define PCMK_XA_IMAGE "image" #define PCMK_XA_INDEX "index" #define PCMK_XA_INFLUENCE "influence" #define PCMK_XA_INSTANCE "instance" #define PCMK_XA_INTERNAL_PORT "internal-port" #define PCMK_XA_INTERVAL "interval" #define PCMK_XA_IP_RANGE_START "ip-range-start" #define PCMK_XA_IS_DC "is_dc" #define PCMK_XA_KIND "kind" #define PCMK_XA_LANG "lang" #define PCMK_XA_LAST_FAILURE "last-failure" #define PCMK_XA_LAST_GRANTED "last-granted" #define PCMK_XA_LAST_RC_CHANGE "last-rc-change" #define PCMK_XA_LAST_UPDATED "last_updated" #define PCMK_XA_LOCKED_TO "locked_to" #define PCMK_XA_LOCKED_TO_HYPHEN "locked-to" #define PCMK_XA_LOSS_POLICY "loss-policy" #define PCMK_XA_MAINTENANCE "maintenance" #define PCMK_XA_MAINTENANCE_MODE "maintenance-mode" #define PCMK_XA_MANAGED "managed" #define PCMK_XA_MESSAGE "message" #define PCMK_XA_MINUTES "minutes" #define PCMK_XA_MIXED_VERSION "mixed_version" #define PCMK_XA_MONTHDAYS "monthdays" #define PCMK_XA_MONTHS "months" #define PCMK_XA_MULTI_STATE "multi_state" #define PCMK_XA_NAME "name" #define PCMK_XA_NETWORK "network" #define PCMK_XA_NEXT_ROLE "next-role" #define PCMK_XA_NO_QUORUM_PANIC "no-quorum-panic" #define PCMK_XA_NO_QUORUM_POLICY "no-quorum-policy" #define PCMK_XA_NODE "node" #define PCMK_XA_NODE_ATTRIBUTE "node-attribute" #define PCMK_XA_NODE_NAME "node_name" #define PCMK_XA_NODE_PATH "node_path" #define PCMK_XA_NODEID "nodeid" #define PCMK_XA_NODES_RUNNING_ON "nodes_running_on" #define PCMK_XA_NUM_UPDATES "num_updates" #define PCMK_XA_NUMBER "number" #define PCMK_XA_NUMBER_RESOURCES "number_resources" #define PCMK_XA_OBJECT_TYPE "object-type" #define PCMK_XA_ON_TARGET "on_target" #define PCMK_XA_ONLINE "online" #define PCMK_XA_OP "op" #define PCMK_XA_OP_KEY "op_key" #define PCMK_XA_OPERATION "operation" #define PCMK_XA_OPTIONS "options" #define PCMK_XA_ORIGIN "origin" #define PCMK_XA_ORPHAN "orphan" #define PCMK_XA_ORPHANED "orphaned" #define PCMK_XA_PACEMAKERD_STATE "pacemakerd-state" #define PCMK_XA_PATH "path" #define PCMK_XA_PENDING "pending" #define PCMK_XA_PORT "port" #define PCMK_XA_PRESENT "present" #define PCMK_XA_PRIORITY_FENCING_DELAY_MS "priority-fencing-delay-ms" #define PCMK_XA_PROGRAM "program" #define PCMK_XA_PROMOTABLE "promotable" #define PCMK_XA_PROMOTED_MAX "promoted-max" #define PCMK_XA_PROMOTED_ONLY "promoted-only" #define PCMK_XA_PROVIDER "provider" #define PCMK_XA_QUEUE_TIME "queue-time" #define PCMK_XA_QUEUED "queued" #define PCMK_XA_QUORUM "quorum" #define PCMK_XA_RANGE "range" #define PCMK_XA_RC "rc" #define PCMK_XA_RC_TEXT "rc_text" #define PCMK_XA_REASON "reason" #define PCMK_XA_REFERENCE "reference" #define PCMK_XA_RELOADABLE "reloadable" #define PCMK_XA_REMAIN_STOPPED "remain_stopped" #define PCMK_XA_REMOTE_CLEAR_PORT "remote-clear-port" #define PCMK_XA_REMOTE_NODE "remote_node" #define PCMK_XA_REMOTE_TLS_PORT "remote-tls-port" #define PCMK_XA_REPLICAS "replicas" #define PCMK_XA_REPLICAS_PER_HOST "replicas-per-host" #define PCMK_XA_REQUEST "request" #define PCMK_XA_REQUIRE_ALL "require-all" #define PCMK_XA_RESOURCE "resource" #define PCMK_XA_RESOURCE_AGENT "resource_agent" #define PCMK_XA_RESOURCE_DISCOVERY "resource-discovery" #define PCMK_XA_RESOURCES_RUNNING "resources_running" #define PCMK_XA_RESULT "result" #define PCMK_XA_ROLE "role" #define PCMK_XA_RSC "rsc" #define PCMK_XA_RSC_PATTERN "rsc-pattern" #define PCMK_XA_RSC_ROLE "rsc-role" #define PCMK_XA_RULE_ID "rule-id" #define PCMK_XA_RUN_COMMAND "run-command" #define PCMK_XA_RUNNING "running" #define PCMK_XA_RUNNING_ON "running_on" #define PCMK_XA_SCOPE "scope" #define PCMK_XA_SCORE "score" #define PCMK_XA_SCORE_ATTRIBUTE "score-attribute" #define PCMK_XA_SEQUENTIAL "sequential" #define PCMK_XA_SECONDS "seconds" #define PCMK_XA_SHUTDOWN "shutdown" #define PCMK_XA_SOURCE "source" #define PCMK_XA_SOURCE_DIR "source-dir" #define PCMK_XA_SOURCE_DIR_ROOT "source-dir-root" #define PCMK_XA_SPEC "spec" #define PCMK_XA_STANDARD "standard" #define PCMK_XA_STANDBY "standby" #define PCMK_XA_STANDBY_ONFAIL "standby_onfail" #define PCMK_XA_START "start" #define PCMK_XA_STATE "state" #define PCMK_XA_STATUS "status" #define PCMK_XA_STONITH_ENABLED "stonith-enabled" #define PCMK_XA_STONITH_TIMEOUT_MS "stonith-timeout-ms" #define PCMK_XA_STOP_ALL_RESOURCES "stop-all-resources" #define PCMK_XA_SYMMETRIC_CLUSTER "symmetric-cluster" #define PCMK_XA_SYMMETRICAL "symmetrical" #define PCMK_XA_SYS_FROM "sys_from" #define PCMK_XA_TAG "tag" #define PCMK_XA_TARGET "target" #define PCMK_XA_TARGET_ATTRIBUTE "target-attribute" #define PCMK_XA_TARGET_DIR "target-dir" #define PCMK_XA_TARGET_PATTERN "target-pattern" #define PCMK_XA_TARGET_ROLE "target_role" #define PCMK_XA_TARGET_VALUE "target-value" #define PCMK_XA_TASK "task" #define PCMK_XA_TEMPLATE "template" #define PCMK_XA_TICKET "ticket" #define PCMK_XA_TIME "time" #define PCMK_XA_THEN "then" #define PCMK_XA_THEN_ACTION "then-action" #define PCMK_XA_TYPE "type" #define PCMK_XA_UNAME "uname" #define PCMK_XA_UNCLEAN "unclean" #define PCMK_XA_UNHEALTHY "unhealthy" #define PCMK_XA_UNIQUE "unique" #define PCMK_XA_UNMANAGED "unmanaged" #define PCMK_XA_UPDATE_CLIENT "update-client" #define PCMK_XA_UPDATE_ORIGIN "update-origin" #define PCMK_XA_UPDATE_USER "update-user" #define PCMK_XA_USER "user" #define PCMK_XA_VALID "valid" #define PCMK_XA_VALIDATE_WITH "validate-with" #define PCMK_XA_VALUE "value" #define PCMK_XA_VALUE_SOURCE "value-source" #define PCMK_XA_VERSION "version" #define PCMK_XA_WATCHDOG "watchdog" #define PCMK_XA_WEEKDAYS "weekdays" #define PCMK_XA_WEEKS "weeks" #define PCMK_XA_WEEKYEARS "weekyears" #define PCMK_XA_WEIGHT "weight" #define PCMK_XA_WHEN "when" #define PCMK_XA_WITH_QUORUM "with_quorum" #define PCMK_XA_WITH_RSC "with-rsc" #define PCMK_XA_WITH_RSC_ROLE "with-rsc-role" #define PCMK_XA_XPATH "xpath" #define PCMK_XA_YEARDAYS "yeardays" #define PCMK_XA_YEARS "years" +#define PCMK_XA_USE_MONITOR_PENDING_TIMEOUT "use-monitor-pending-timeout" #ifdef __cplusplus } #endif #endif // PCMK__CRM_COMMON_XML_NAMES__H diff --git a/lib/services/services.c b/lib/services/services.c index c65eb9e73f..d6080346fd 100644 --- a/lib/services/services.c +++ b/lib/services/services.c @@ -1,1358 +1,1368 @@ /* * Copyright 2010-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include "services_private.h" #include "services_ocf.h" #if PCMK__ENABLE_LSB #include "services_lsb.h" #endif #if SUPPORT_SYSTEMD # include #endif /* TODO: Develop a rollover strategy */ static int operations = 0; static GHashTable *recurring_actions = NULL; /* ops waiting to run async because of conflicting active * pending ops */ static GList *blocked_ops = NULL; /* ops currently active (in-flight) */ static GList *inflight_ops = NULL; static void handle_blocked_ops(void); /*! * \brief Find first service class that can provide a specified agent * * \param[in] agent Name of agent to search for * * \return Service class if found, NULL otherwise * * \note The priority is LSB then systemd. It would be preferable to put systemd * first, but LSB merely requires a file existence check, while systemd * requires contacting DBus. */ const char * resources_find_service_class(const char *agent) { #if PCMK__ENABLE_LSB if (services__lsb_agent_exists(agent)) { return PCMK_RESOURCE_CLASS_LSB; } #endif #if SUPPORT_SYSTEMD if (systemd_unit_exists(agent)) { return PCMK_RESOURCE_CLASS_SYSTEMD; } #endif return NULL; } static inline void init_recurring_actions(void) { if (recurring_actions == NULL) { recurring_actions = pcmk__strkey_table(NULL, NULL); } } /*! * \internal * \brief Check whether op is in-flight systemd op * * \param[in] op Operation to check * * \return TRUE if op is in-flight systemd op */ static inline gboolean inflight_systemd(const svc_action_t *op) { return pcmk__str_eq(op->standard, PCMK_RESOURCE_CLASS_SYSTEMD, pcmk__str_casei) && (g_list_find(inflight_ops, op) != NULL); } /*! * \internal * \brief Expand "service" alias to an actual resource class * * \param[in] rsc Resource name (for logging only) * \param[in] standard Resource class as configured * \param[in] agent Agent name to look for * * \return Newly allocated string with actual resource class * * \note The caller is responsible for calling free() on the result. */ static char * expand_resource_class(const char *rsc, const char *standard, const char *agent) { char *expanded_class = NULL; #if PCMK__ENABLE_SERVICE if (strcasecmp(standard, PCMK_RESOURCE_CLASS_SERVICE) == 0) { const char *found_class = resources_find_service_class(agent); if (found_class != NULL) { crm_debug("Found %s agent %s for %s", found_class, agent, rsc); expanded_class = pcmk__str_copy(found_class); } else { const char *default_standard = NULL; #if PCMK__ENABLE_LSB default_standard = PCMK_RESOURCE_CLASS_LSB; #elif SUPPORT_SYSTEMD default_standard = PCMK_RESOURCE_CLASS_SYSTEMD; #else #error No standards supported for service alias (configure script bug) #endif crm_info("Assuming resource class %s for agent %s for %s", default_standard, agent, rsc); expanded_class = pcmk__str_copy(default_standard); } } #endif if (expanded_class == NULL) { expanded_class = pcmk__str_copy(standard); } return expanded_class; } /*! * \internal * \brief Create a simple svc_action_t instance * * \return Newly allocated instance (or NULL if not enough memory) */ static svc_action_t * new_action(void) { svc_action_t *op = calloc(1, sizeof(svc_action_t)); if (op == NULL) { return NULL; } op->opaque = calloc(1, sizeof(svc_action_private_t)); if (op->opaque == NULL) { free(op); return NULL; } // Initialize result services__set_result(op, PCMK_OCF_UNKNOWN, PCMK_EXEC_UNKNOWN, NULL); return op; } static bool required_argument_missing(uint32_t ra_caps, const char *name, const char *standard, const char *provider, const char *agent, const char *action) { if (pcmk__str_empty(name)) { crm_info("Cannot create operation without resource name (bug?)"); return true; } if (pcmk__str_empty(standard)) { crm_info("Cannot create operation for %s without resource class (bug?)", name); return true; } if (pcmk_is_set(ra_caps, pcmk_ra_cap_provider) && pcmk__str_empty(provider)) { crm_info("Cannot create operation for %s resource %s " "without provider (bug?)", standard, name); return true; } if (pcmk__str_empty(agent)) { crm_info("Cannot create operation for %s without agent name (bug?)", name); return true; } if (pcmk__str_empty(action)) { crm_info("Cannot create operation for %s without action name (bug?)", name); return true; } return false; } // \return Standard Pacemaker return code (pcmk_rc_ok or ENOMEM) static int copy_action_arguments(svc_action_t *op, uint32_t ra_caps, const char *name, const char *standard, const char *provider, const char *agent, const char *action) { op->rsc = strdup(name); if (op->rsc == NULL) { return ENOMEM; } op->agent = strdup(agent); if (op->agent == NULL) { return ENOMEM; } op->standard = expand_resource_class(name, standard, agent); if (op->standard == NULL) { return ENOMEM; } if (pcmk_is_set(ra_caps, pcmk_ra_cap_status) && pcmk__str_eq(action, PCMK_ACTION_MONITOR, pcmk__str_casei)) { action = PCMK_ACTION_STATUS; } op->action = strdup(action); if (op->action == NULL) { return ENOMEM; } if (pcmk_is_set(ra_caps, pcmk_ra_cap_provider)) { op->provider = strdup(provider); if (op->provider == NULL) { return ENOMEM; } } return pcmk_rc_ok; } svc_action_t * services__create_resource_action(const char *name, const char *standard, const char *provider, const char *agent, const char *action, guint interval_ms, int timeout, GHashTable *params, enum svc_action_flags flags) { svc_action_t *op = NULL; uint32_t ra_caps = pcmk_get_ra_caps(standard); int rc = pcmk_rc_ok; op = new_action(); if (op == NULL) { crm_crit("Cannot prepare action: %s", strerror(ENOMEM)); if (params != NULL) { g_hash_table_destroy(params); } return NULL; } op->interval_ms = interval_ms; op->timeout = timeout; op->flags = flags; op->sequence = ++operations; // Take ownership of params - if (pcmk_is_set(ra_caps, pcmk_ra_cap_params)) { + if (pcmk_is_set(ra_caps, pcmk_ra_cap_params) || pcmk_is_set(ra_caps, pcmk_ra_cap_status)) { op->params = params; } else if (params != NULL) { g_hash_table_destroy(params); params = NULL; } if (required_argument_missing(ra_caps, name, standard, provider, agent, action)) { services__set_result(op, services__generic_error(op), PCMK_EXEC_ERROR_FATAL, "Required agent or action information missing"); return op; } op->id = pcmk__op_key(name, action, interval_ms); if (copy_action_arguments(op, ra_caps, name, standard, provider, agent, action) != pcmk_rc_ok) { crm_crit("Cannot prepare %s action for %s: %s", action, name, strerror(ENOMEM)); services__handle_exec_error(op, ENOMEM); return op; } if (strcasecmp(op->standard, PCMK_RESOURCE_CLASS_OCF) == 0) { rc = services__ocf_prepare(op); #if PCMK__ENABLE_LSB } else if (strcasecmp(op->standard, PCMK_RESOURCE_CLASS_LSB) == 0) { rc = services__lsb_prepare(op); #endif #if SUPPORT_SYSTEMD } else if (strcasecmp(op->standard, PCMK_RESOURCE_CLASS_SYSTEMD) == 0) { + if (pcmk__strcase_any_of(action, PCMK_ACTION_MONITOR, PCMK_ACTION_STATUS, NULL) && (interval_ms > 0)) { + if (pcmk__str_eq(g_hash_table_lookup(op->params, PCMK_XA_USE_MONITOR_PENDING_TIMEOUT), "true", pcmk__str_casei)) { + if (interval_ms > timeout) { + crm_warn("If monitor-pending-timeout is set to true and the interval is longer than the monitor timeout, " + "two consecutive monitor deactivations will be treated as an error. " + "We recommend setting a shorter interval to ensure that " + "multiple consecutive deactivations before the timeout are treated as an error."); + } + } + } rc = services__systemd_prepare(op); #endif } else { crm_info("Unknown resource standard: %s", op->standard); rc = ENOENT; } if (rc != pcmk_rc_ok) { crm_info("Cannot prepare %s operation for %s: %s", action, name, strerror(rc)); services__handle_exec_error(op, rc); } return op; } svc_action_t * resources_action_create(const char *name, const char *standard, const char *provider, const char *agent, const char *action, guint interval_ms, int timeout, GHashTable *params, enum svc_action_flags flags) { svc_action_t *op = services__create_resource_action(name, standard, provider, agent, action, interval_ms, timeout, params, flags); if (op == NULL || op->rc != 0) { services_action_free(op); return NULL; } else { // Preserve public API backward compatibility op->rc = PCMK_OCF_OK; op->status = PCMK_EXEC_DONE; return op; } } svc_action_t * services_action_create_generic(const char *exec, const char *args[]) { svc_action_t *op = new_action(); pcmk__mem_assert(op); op->opaque->exec = strdup(exec); op->opaque->args[0] = strdup(exec); if ((op->opaque->exec == NULL) || (op->opaque->args[0] == NULL)) { crm_crit("Cannot prepare action for '%s': %s", exec, strerror(ENOMEM)); services__set_result(op, PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_ERROR, strerror(ENOMEM)); return op; } if (args == NULL) { return op; } for (int cur_arg = 1; args[cur_arg - 1] != NULL; cur_arg++) { if (cur_arg == PCMK__NELEM(op->opaque->args)) { crm_info("Cannot prepare action for '%s': Too many arguments", exec); services__set_result(op, PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_ERROR_HARD, "Too many arguments"); break; } op->opaque->args[cur_arg] = strdup(args[cur_arg - 1]); if (op->opaque->args[cur_arg] == NULL) { crm_crit("Cannot prepare action for '%s': %s", exec, strerror(ENOMEM)); services__set_result(op, PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_ERROR, strerror(ENOMEM)); break; } } return op; } /*! * \brief Create an alert agent action * * \param[in] id Alert ID * \param[in] exec Path to alert agent executable * \param[in] timeout Action timeout * \param[in] params Parameters to use with action * \param[in] sequence Action sequence number * \param[in] cb_data Data to pass to callback function * * \return New action on success, NULL on error * \note It is the caller's responsibility to free cb_data. * The caller should not free params explicitly. */ svc_action_t * services_alert_create(const char *id, const char *exec, int timeout, GHashTable *params, int sequence, void *cb_data) { svc_action_t *action = services_action_create_generic(exec, NULL); action->id = pcmk__str_copy(id); action->standard = pcmk__str_copy(PCMK_RESOURCE_CLASS_ALERT); action->timeout = timeout; action->params = params; action->sequence = sequence; action->cb_data = cb_data; return action; } /*! * \brief Set the user and group that an action will execute as * * \param[in,out] op Action to modify * \param[in] user Name of user to execute action as * \param[in] group Name of group to execute action as * * \return pcmk_ok on success, -errno otherwise * * \note This will have no effect unless the process executing the action runs * as root and the action is not a systemd action. We could implement this * for systemd by adding User= and Group= to [Service] in the override * file, but that seems more likely to cause problems than be useful. */ int services_action_user(svc_action_t *op, const char *user) { CRM_CHECK((op != NULL) && (user != NULL), return -EINVAL); return crm_user_lookup(user, &(op->opaque->uid), &(op->opaque->gid)); } /*! * \brief Execute an alert agent action * * \param[in,out] action Action to execute * \param[in] cb Function to call when action completes * * \return TRUE if the library will free action, FALSE otherwise * * \note If this function returns FALSE, it is the caller's responsibility to * free the action with services_action_free(). However, unless someone * intentionally creates a recurring alert action, this will never return * FALSE. */ gboolean services_alert_async(svc_action_t *action, void (*cb)(svc_action_t *op)) { action->synchronous = false; action->opaque->callback = cb; return services__execute_file(action) == pcmk_rc_ok; } #if HAVE_DBUS /*! * \internal * \brief Update operation's pending DBus call, unreferencing old one if needed * * \param[in,out] op Operation to modify * \param[in] pending Pending call to set */ void services_set_op_pending(svc_action_t *op, DBusPendingCall *pending) { if (op->opaque->pending && (op->opaque->pending != pending)) { if (pending) { crm_info("Lost pending %s DBus call (%p)", op->id, op->opaque->pending); } else { crm_trace("Done with pending %s DBus call (%p)", op->id, op->opaque->pending); } dbus_pending_call_unref(op->opaque->pending); } op->opaque->pending = pending; if (pending) { crm_trace("Updated pending %s DBus call (%p)", op->id, pending); } else { crm_trace("Cleared pending %s DBus call", op->id); } } #endif void services_action_cleanup(svc_action_t * op) { if ((op == NULL) || (op->opaque == NULL)) { return; } #if HAVE_DBUS if(op->opaque->timerid != 0) { crm_trace("Removing timer for call %s to %s", op->action, op->rsc); g_source_remove(op->opaque->timerid); op->opaque->timerid = 0; } if(op->opaque->pending) { if (dbus_pending_call_get_completed(op->opaque->pending)) { // This should never be the case crm_warn("Result of %s op %s was unhandled", op->standard, op->id); } else { crm_debug("Will ignore any result of canceled %s op %s", op->standard, op->id); } dbus_pending_call_cancel(op->opaque->pending); services_set_op_pending(op, NULL); } #endif if (op->opaque->stderr_gsource) { mainloop_del_fd(op->opaque->stderr_gsource); op->opaque->stderr_gsource = NULL; } if (op->opaque->stdout_gsource) { mainloop_del_fd(op->opaque->stdout_gsource); op->opaque->stdout_gsource = NULL; } } /*! * \internal * \brief Map an actual resource action result to a standard OCF result * * \param[in] standard Agent standard (must not be "service") * \param[in] action Action that result is for * \param[in] exit_status Actual agent exit status * * \return Standard OCF result */ enum ocf_exitcode services_result2ocf(const char *standard, const char *action, int exit_status) { if (pcmk__str_eq(standard, PCMK_RESOURCE_CLASS_OCF, pcmk__str_casei)) { return services__ocf2ocf(exit_status); #if SUPPORT_SYSTEMD } else if (pcmk__str_eq(standard, PCMK_RESOURCE_CLASS_SYSTEMD, pcmk__str_casei)) { return services__systemd2ocf(exit_status); #endif #if PCMK__ENABLE_LSB } else if (pcmk__str_eq(standard, PCMK_RESOURCE_CLASS_LSB, pcmk__str_casei)) { return services__lsb2ocf(action, exit_status); #endif } else { crm_warn("Treating result from unknown standard '%s' as OCF", ((standard == NULL)? "unspecified" : standard)); return services__ocf2ocf(exit_status); } } void services_action_free(svc_action_t * op) { unsigned int i; if (op == NULL) { return; } /* The operation should be removed from all tracking lists by this point. * If it's not, we have a bug somewhere, so bail. That may lead to a * memory leak, but it's better than a use-after-free segmentation fault. */ CRM_CHECK(g_list_find(inflight_ops, op) == NULL, return); CRM_CHECK(g_list_find(blocked_ops, op) == NULL, return); CRM_CHECK((recurring_actions == NULL) || (g_hash_table_lookup(recurring_actions, op->id) == NULL), return); services_action_cleanup(op); if (op->opaque->repeat_timer) { g_source_remove(op->opaque->repeat_timer); op->opaque->repeat_timer = 0; } free(op->id); free(op->opaque->exec); for (i = 0; i < PCMK__NELEM(op->opaque->args); i++) { free(op->opaque->args[i]); } free(op->opaque->exit_reason); free(op->opaque); free(op->rsc); free(op->action); free(op->standard); free(op->agent); free(op->provider); free(op->stdout_data); free(op->stderr_data); if (op->params) { g_hash_table_destroy(op->params); op->params = NULL; } free(op); } gboolean cancel_recurring_action(svc_action_t * op) { crm_info("Cancelling %s operation %s", op->standard, op->id); if (recurring_actions) { g_hash_table_remove(recurring_actions, op->id); } if (op->opaque->repeat_timer) { g_source_remove(op->opaque->repeat_timer); op->opaque->repeat_timer = 0; } return TRUE; } /*! * \brief Cancel a recurring action * * \param[in] name Name of resource that operation is for * \param[in] action Name of operation to cancel * \param[in] interval_ms Interval of operation to cancel * * \return TRUE if action was successfully cancelled, FALSE otherwise */ gboolean services_action_cancel(const char *name, const char *action, guint interval_ms) { gboolean cancelled = FALSE; char *id = pcmk__op_key(name, action, interval_ms); svc_action_t *op = NULL; /* We can only cancel a recurring action */ init_recurring_actions(); op = g_hash_table_lookup(recurring_actions, id); if (op == NULL) { goto done; } // Tell services__finalize_async_op() not to reschedule the operation op->cancel = TRUE; /* Stop tracking it as a recurring operation, and stop its repeat timer */ cancel_recurring_action(op); /* If the op has a PID, it's an in-flight child process, so kill it. * * Whether the kill succeeds or fails, the main loop will send the op to * async_action_complete() (and thus services__finalize_async_op()) when the * process goes away. */ if (op->pid != 0) { crm_info("Terminating in-flight op %s[%d] early because it was cancelled", id, op->pid); cancelled = mainloop_child_kill(op->pid); if (cancelled == FALSE) { crm_err("Termination of %s[%d] failed", id, op->pid); } goto done; } #if HAVE_DBUS // In-flight systemd ops don't have a pid if (inflight_systemd(op)) { inflight_ops = g_list_remove(inflight_ops, op); /* This will cause any result that comes in later to be discarded, so we * don't call the callback and free the operation twice. */ services_action_cleanup(op); } #endif /* The rest of this is essentially equivalent to * services__finalize_async_op(), minus the handle_blocked_ops() call. */ // Report operation as cancelled services__set_cancelled(op); if (op->opaque->callback) { op->opaque->callback(op); } blocked_ops = g_list_remove(blocked_ops, op); services_action_free(op); cancelled = TRUE; // @TODO Initiate handle_blocked_ops() asynchronously done: free(id); return cancelled; } gboolean services_action_kick(const char *name, const char *action, guint interval_ms) { svc_action_t * op = NULL; char *id = pcmk__op_key(name, action, interval_ms); init_recurring_actions(); op = g_hash_table_lookup(recurring_actions, id); free(id); if (op == NULL) { return FALSE; } if (op->pid || inflight_systemd(op)) { return TRUE; } else { if (op->opaque->repeat_timer) { g_source_remove(op->opaque->repeat_timer); op->opaque->repeat_timer = 0; } recurring_action_timer(op); return TRUE; } } /*! * \internal * \brief Add a new recurring operation, checking for duplicates * * \param[in,out] op Operation to add * * \return TRUE if duplicate found (and reschedule), FALSE otherwise */ static gboolean handle_duplicate_recurring(svc_action_t *op) { svc_action_t * dup = NULL; /* check for duplicates */ dup = g_hash_table_lookup(recurring_actions, op->id); if (dup && (dup != op)) { /* update user data */ if (op->opaque->callback) { dup->opaque->callback = op->opaque->callback; dup->cb_data = op->cb_data; op->cb_data = NULL; } /* immediately execute the next interval */ if (dup->pid != 0) { if (op->opaque->repeat_timer) { g_source_remove(op->opaque->repeat_timer); op->opaque->repeat_timer = 0; } recurring_action_timer(dup); } /* free the duplicate */ services_action_free(op); return TRUE; } return FALSE; } /*! * \internal * \brief Execute an action appropriately according to its standard * * \param[in,out] op Action to execute * * \return Standard Pacemaker return code * \retval EBUSY Recurring operation could not be initiated * \retval pcmk_rc_error Synchronous action failed * \retval pcmk_rc_ok Synchronous action succeeded, or asynchronous action * should not be freed (because it's pending or because * it failed to execute and was already freed) * * \note If the return value for an asynchronous action is not pcmk_rc_ok, the * caller is responsible for freeing the action. */ static int execute_action(svc_action_t *op) { #if SUPPORT_SYSTEMD if (pcmk__str_eq(op->standard, PCMK_RESOURCE_CLASS_SYSTEMD, pcmk__str_casei)) { return services__execute_systemd(op); } #endif return services__execute_file(op); } void services_add_inflight_op(svc_action_t * op) { if (op == NULL) { return; } pcmk__assert(op->synchronous == FALSE); /* keep track of ops that are in-flight to avoid collisions in the same namespace */ if (op->rsc) { inflight_ops = g_list_append(inflight_ops, op); } } /*! * \internal * \brief Stop tracking an operation that completed * * \param[in] op Operation to stop tracking */ void services_untrack_op(const svc_action_t *op) { /* Op is no longer in-flight or blocked */ inflight_ops = g_list_remove(inflight_ops, op); blocked_ops = g_list_remove(blocked_ops, op); /* Op is no longer blocking other ops, so check if any need to run */ handle_blocked_ops(); } gboolean services_action_async_fork_notify(svc_action_t * op, void (*action_callback) (svc_action_t *), void (*action_fork_callback) (svc_action_t *)) { CRM_CHECK(op != NULL, return TRUE); op->synchronous = false; if (action_callback != NULL) { op->opaque->callback = action_callback; } if (action_fork_callback != NULL) { op->opaque->fork_callback = action_fork_callback; } if (op->interval_ms > 0) { init_recurring_actions(); if (handle_duplicate_recurring(op)) { /* entry rescheduled, dup freed */ /* exit early */ return TRUE; } g_hash_table_replace(recurring_actions, op->id, op); } if (!pcmk_is_set(op->flags, SVC_ACTION_NON_BLOCKED) && op->rsc && is_op_blocked(op->rsc)) { blocked_ops = g_list_append(blocked_ops, op); return TRUE; } return execute_action(op) == pcmk_rc_ok; } gboolean services_action_async(svc_action_t * op, void (*action_callback) (svc_action_t *)) { return services_action_async_fork_notify(op, action_callback, NULL); } static gboolean processing_blocked_ops = FALSE; gboolean is_op_blocked(const char *rsc) { GList *gIter = NULL; svc_action_t *op = NULL; for (gIter = inflight_ops; gIter != NULL; gIter = gIter->next) { op = gIter->data; if (pcmk__str_eq(op->rsc, rsc, pcmk__str_casei)) { return TRUE; } } return FALSE; } static void handle_blocked_ops(void) { GList *executed_ops = NULL; GList *gIter = NULL; svc_action_t *op = NULL; if (processing_blocked_ops) { /* avoid nested calling of this function */ return; } processing_blocked_ops = TRUE; /* n^2 operation here, but blocked ops are incredibly rare. this list * will be empty 99% of the time. */ for (gIter = blocked_ops; gIter != NULL; gIter = gIter->next) { op = gIter->data; if (is_op_blocked(op->rsc)) { continue; } executed_ops = g_list_append(executed_ops, op); if (execute_action(op) != pcmk_rc_ok) { /* this can cause this function to be called recursively * which is why we have processing_blocked_ops static variable */ services__finalize_async_op(op); } } for (gIter = executed_ops; gIter != NULL; gIter = gIter->next) { op = gIter->data; blocked_ops = g_list_remove(blocked_ops, op); } g_list_free(executed_ops); processing_blocked_ops = FALSE; } /*! * \internal * \brief Execute a meta-data action appropriately to standard * * \param[in,out] op Meta-data action to execute * * \return Standard Pacemaker return code */ static int execute_metadata_action(svc_action_t *op) { const char *class = op->standard; if (op->agent == NULL) { crm_info("Meta-data requested without specifying agent"); services__set_result(op, services__generic_error(op), PCMK_EXEC_ERROR_FATAL, "Agent not specified"); return EINVAL; } if (class == NULL) { crm_info("Meta-data requested for agent %s without specifying class", op->agent); services__set_result(op, services__generic_error(op), PCMK_EXEC_ERROR_FATAL, "Agent standard not specified"); return EINVAL; } #if PCMK__ENABLE_SERVICE if (!strcmp(class, PCMK_RESOURCE_CLASS_SERVICE)) { class = resources_find_service_class(op->agent); } if (class == NULL) { crm_info("Meta-data requested for %s, but could not determine class", op->agent); services__set_result(op, services__generic_error(op), PCMK_EXEC_ERROR_HARD, "Agent standard could not be determined"); return EINVAL; } #endif #if PCMK__ENABLE_LSB if (pcmk__str_eq(class, PCMK_RESOURCE_CLASS_LSB, pcmk__str_casei)) { return pcmk_legacy2rc(services__get_lsb_metadata(op->agent, &op->stdout_data)); } #endif return execute_action(op); } gboolean services_action_sync(svc_action_t * op) { gboolean rc = TRUE; if (op == NULL) { crm_trace("No operation to execute"); return FALSE; } op->synchronous = true; if (pcmk__str_eq(op->action, PCMK_ACTION_META_DATA, pcmk__str_casei)) { /* Synchronous meta-data operations are handled specially. Since most * resource classes don't provide any meta-data, it has to be * synthesized from available information about the agent. * * services_action_async() doesn't treat meta-data actions specially, so * it will result in an error for classes that don't support the action. */ rc = (execute_metadata_action(op) == pcmk_rc_ok); } else { rc = (execute_action(op) == pcmk_rc_ok); } crm_trace(" > " PCMK__OP_FMT ": %s = %d", op->rsc, op->action, op->interval_ms, op->opaque->exec, op->rc); if (op->stdout_data) { crm_trace(" > stdout: %s", op->stdout_data); } if (op->stderr_data) { crm_trace(" > stderr: %s", op->stderr_data); } return rc; } GList * get_directory_list(const char *root, gboolean files, gboolean executable) { return services_os_get_directory_list(root, files, executable); } GList * resources_list_standards(void) { GList *standards = NULL; standards = g_list_append(standards, strdup(PCMK_RESOURCE_CLASS_OCF)); #if PCMK__ENABLE_SERVICE standards = g_list_append(standards, strdup(PCMK_RESOURCE_CLASS_SERVICE)); #endif #if PCMK__ENABLE_LSB standards = g_list_append(standards, strdup(PCMK_RESOURCE_CLASS_LSB)); #endif #if SUPPORT_SYSTEMD { GList *agents = systemd_unit_listall(); if (agents != NULL) { standards = g_list_append(standards, strdup(PCMK_RESOURCE_CLASS_SYSTEMD)); g_list_free_full(agents, free); } } #endif return standards; } GList * resources_list_providers(const char *standard) { if (pcmk_is_set(pcmk_get_ra_caps(standard), pcmk_ra_cap_provider)) { return resources_os_list_ocf_providers(); } return NULL; } GList * resources_list_agents(const char *standard, const char *provider) { if ((standard == NULL) #if PCMK__ENABLE_SERVICE || (strcasecmp(standard, PCMK_RESOURCE_CLASS_SERVICE) == 0) #endif ) { GList *tmp1; GList *tmp2; GList *result = NULL; if (standard == NULL) { tmp1 = result; tmp2 = resources_os_list_ocf_agents(NULL); if (tmp2) { result = g_list_concat(tmp1, tmp2); } } #if PCMK__ENABLE_LSB result = g_list_concat(result, services__list_lsb_agents()); #endif #if SUPPORT_SYSTEMD tmp1 = result; tmp2 = systemd_unit_listall(); if (tmp2) { result = g_list_concat(tmp1, tmp2); } #endif return result; } else if (strcasecmp(standard, PCMK_RESOURCE_CLASS_OCF) == 0) { return resources_os_list_ocf_agents(provider); #if PCMK__ENABLE_LSB } else if (strcasecmp(standard, PCMK_RESOURCE_CLASS_LSB) == 0) { return services__list_lsb_agents(); #endif #if SUPPORT_SYSTEMD } else if (strcasecmp(standard, PCMK_RESOURCE_CLASS_SYSTEMD) == 0) { return systemd_unit_listall(); #endif } return NULL; } gboolean resources_agent_exists(const char *standard, const char *provider, const char *agent) { GList *standards = NULL; GList *providers = NULL; GList *iter = NULL; gboolean rc = FALSE; gboolean has_providers = FALSE; standards = resources_list_standards(); for (iter = standards; iter != NULL; iter = iter->next) { if (pcmk__str_eq(iter->data, standard, pcmk__str_none)) { rc = TRUE; break; } } if (rc == FALSE) { goto done; } rc = FALSE; has_providers = pcmk_is_set(pcmk_get_ra_caps(standard), pcmk_ra_cap_provider); if (has_providers == TRUE && provider != NULL) { providers = resources_list_providers(standard); for (iter = providers; iter != NULL; iter = iter->next) { if (pcmk__str_eq(iter->data, provider, pcmk__str_none)) { rc = TRUE; break; } } } else if (has_providers == FALSE && provider == NULL) { rc = TRUE; } if (rc == FALSE) { goto done; } #if PCMK__ENABLE_SERVICE if (pcmk__str_eq(standard, PCMK_RESOURCE_CLASS_SERVICE, pcmk__str_casei)) { #if PCMK__ENABLE_LSB if (services__lsb_agent_exists(agent)) { rc = TRUE; goto done; } #endif #if SUPPORT_SYSTEMD if (systemd_unit_exists(agent)) { rc = TRUE; goto done; } #endif rc = FALSE; goto done; } #endif if (pcmk__str_eq(standard, PCMK_RESOURCE_CLASS_OCF, pcmk__str_casei)) { rc = services__ocf_agent_exists(provider, agent); #if PCMK__ENABLE_LSB } else if (pcmk__str_eq(standard, PCMK_RESOURCE_CLASS_LSB, pcmk__str_casei)) { rc = services__lsb_agent_exists(agent); #endif #if SUPPORT_SYSTEMD } else if (pcmk__str_eq(standard, PCMK_RESOURCE_CLASS_SYSTEMD, pcmk__str_casei)) { rc = systemd_unit_exists(agent); #endif } else { rc = FALSE; } done: g_list_free(standards); g_list_free(providers); return rc; } /*! * \internal * \brief Set the result of an action * * \param[out] action Where to set action result * \param[in] agent_status Exit status to set * \param[in] exec_status Execution status to set * \param[in] reason Human-friendly description of event to set */ void services__set_result(svc_action_t *action, int agent_status, enum pcmk_exec_status exec_status, const char *reason) { if (action == NULL) { return; } action->rc = agent_status; action->status = exec_status; if (!pcmk__str_eq(action->opaque->exit_reason, reason, pcmk__str_none)) { free(action->opaque->exit_reason); action->opaque->exit_reason = (reason == NULL)? NULL : strdup(reason); } } /*! * \internal * \brief Set the result of an action, with a formatted exit reason * * \param[out] action Where to set action result * \param[in] agent_status Exit status to set * \param[in] exec_status Execution status to set * \param[in] format printf-style format for a human-friendly * description of reason for result * \param[in] ... arguments for \p format */ void services__format_result(svc_action_t *action, int agent_status, enum pcmk_exec_status exec_status, const char *format, ...) { va_list ap; int len = 0; char *reason = NULL; if (action == NULL) { return; } action->rc = agent_status; action->status = exec_status; if (format != NULL) { va_start(ap, format); len = vasprintf(&reason, format, ap); pcmk__assert(len > 0); va_end(ap); } free(action->opaque->exit_reason); action->opaque->exit_reason = reason; } /*! * \internal * \brief Set the result of an action to cancelled * * \param[out] action Where to set action result * * \note This sets execution status but leaves the exit status unchanged */ void services__set_cancelled(svc_action_t *action) { if (action != NULL) { action->status = PCMK_EXEC_CANCELLED; free(action->opaque->exit_reason); action->opaque->exit_reason = NULL; } } /*! * \internal * \brief Get a readable description of what an action is for * * \param[in] action Action to check * * \return Readable name for the kind of \p action */ const char * services__action_kind(const svc_action_t *action) { if ((action == NULL) || (action->standard == NULL)) { return "Process"; } else if (pcmk__str_eq(action->standard, PCMK_RESOURCE_CLASS_STONITH, pcmk__str_none)) { return "Fence agent"; } else if (pcmk__str_eq(action->standard, PCMK_RESOURCE_CLASS_ALERT, pcmk__str_none)) { return "Alert agent"; } else { return "Resource agent"; } } /*! * \internal * \brief Get the exit reason of an action * * \param[in] action Action to check * * \return Action's exit reason (or NULL if none) */ const char * services__exit_reason(const svc_action_t *action) { return action->opaque->exit_reason; } /*! * \internal * \brief Steal stdout from an action * * \param[in,out] action Action whose stdout is desired * * \return Action's stdout (which may be NULL) * \note Upon return, \p action will no longer track the output, so it is the * caller's responsibility to free the return value. */ char * services__grab_stdout(svc_action_t *action) { char *output = action->stdout_data; action->stdout_data = NULL; return output; } /*! * \internal * \brief Steal stderr from an action * * \param[in,out] action Action whose stderr is desired * * \return Action's stderr (which may be NULL) * \note Upon return, \p action will no longer track the output, so it is the * caller's responsibility to free the return value. */ char * services__grab_stderr(svc_action_t *action) { char *output = action->stderr_data; action->stderr_data = NULL; return output; } diff --git a/lib/services/systemd.c b/lib/services/systemd.c index a2cdb2fd48..74bf1a9228 100644 --- a/lib/services/systemd.c +++ b/lib/services/systemd.c @@ -1,1125 +1,1138 @@ /* * 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 #include #include #include #include #include #include #include static void invoke_unit_by_path(svc_action_t *op, const char *unit); #define BUS_NAME "org.freedesktop.systemd1" #define BUS_NAME_MANAGER BUS_NAME ".Manager" #define BUS_NAME_UNIT BUS_NAME ".Unit" #define BUS_PATH "/org/freedesktop/systemd1" /*! * \internal * \brief Prepare a systemd action * * \param[in,out] op Action to prepare * * \return Standard Pacemaker return code */ int services__systemd_prepare(svc_action_t *op) { op->opaque->exec = strdup("systemd-dbus"); if (op->opaque->exec == NULL) { return ENOMEM; } return pcmk_rc_ok; } /*! * \internal * \brief Map a systemd result to a standard OCF result * * \param[in] exit_status Systemd result * * \return Standard OCF result */ enum ocf_exitcode services__systemd2ocf(int exit_status) { // This library uses OCF codes for systemd actions return (enum ocf_exitcode) exit_status; } static inline DBusMessage * systemd_new_method(const char *method) { crm_trace("Calling: %s on " BUS_NAME_MANAGER, method); return dbus_message_new_method_call(BUS_NAME, BUS_PATH, BUS_NAME_MANAGER, method); } /* * Functions to manage a static DBus connection */ static DBusConnection* systemd_proxy = NULL; static inline DBusPendingCall * systemd_send(DBusMessage *msg, void(*done)(DBusPendingCall *pending, void *user_data), void *user_data, int timeout) { return pcmk_dbus_send(msg, systemd_proxy, done, user_data, timeout); } static inline DBusMessage * systemd_send_recv(DBusMessage *msg, DBusError *error, int timeout) { return pcmk_dbus_send_recv(msg, systemd_proxy, error, timeout); } /*! * \internal * \brief Send a method to systemd without arguments, and wait for reply * * \param[in] method Method to send * * \return Systemd reply on success, NULL (and error will be logged) otherwise * * \note The caller must call dbus_message_unref() on the reply after * handling it. */ static DBusMessage * systemd_call_simple_method(const char *method) { DBusMessage *msg = systemd_new_method(method); DBusMessage *reply = NULL; DBusError error; /* Don't call systemd_init() here, because that calls this */ CRM_CHECK(systemd_proxy, return NULL); if (msg == NULL) { crm_err("Could not create message to send %s to systemd", method); return NULL; } dbus_error_init(&error); reply = systemd_send_recv(msg, &error, DBUS_TIMEOUT_USE_DEFAULT); dbus_message_unref(msg); if (dbus_error_is_set(&error)) { crm_err("Could not send %s to systemd: %s (%s)", method, error.message, error.name); dbus_error_free(&error); return NULL; } else if (reply == NULL) { crm_err("Could not send %s to systemd: no reply received", method); return NULL; } return reply; } static gboolean systemd_init(void) { static int need_init = 1; // https://dbus.freedesktop.org/doc/api/html/group__DBusConnection.html if (systemd_proxy && dbus_connection_get_is_connected(systemd_proxy) == FALSE) { crm_warn("Connection to System DBus is closed. Reconnecting..."); pcmk_dbus_disconnect(systemd_proxy); systemd_proxy = NULL; need_init = 1; } if (need_init) { need_init = 0; systemd_proxy = pcmk_dbus_connect(); } if (systemd_proxy == NULL) { return FALSE; } return TRUE; } static inline char * systemd_get_property(const char *unit, const char *name, void (*callback)(const char *name, const char *value, void *userdata), void *userdata, DBusPendingCall **pending, int timeout) { return systemd_proxy? pcmk_dbus_get_property(systemd_proxy, BUS_NAME, unit, BUS_NAME_UNIT, name, callback, userdata, pending, timeout) : NULL; } void systemd_cleanup(void) { if (systemd_proxy) { pcmk_dbus_disconnect(systemd_proxy); systemd_proxy = NULL; } } /* * end of systemd_proxy functions */ /*! * \internal * \brief Check whether a file name represents a manageable systemd unit * * \param[in] name File name to check * * \return Pointer to "dot" before filename extension if so, NULL otherwise */ static const char * systemd_unit_extension(const char *name) { if (name) { const char *dot = strrchr(name, '.'); if (dot && (!strcmp(dot, ".service") || !strcmp(dot, ".socket") || !strcmp(dot, ".mount") || !strcmp(dot, ".timer") || !strcmp(dot, ".path"))) { return dot; } } return NULL; } static char * systemd_service_name(const char *name, bool add_instance_name) { const char *dot = NULL; if (pcmk__str_empty(name)) { return NULL; } /* Services that end with an @ sign are systemd templates. They expect an * instance name to follow the service name. If no instance name was * provided, just add "pacemaker" to the string as the instance name. It * doesn't seem to matter for purposes of looking up whether a service * exists or not. * * A template can be specified either with or without the unit extension, * so this block handles both cases. */ dot = systemd_unit_extension(name); if (dot) { if (dot != name && *(dot-1) == '@') { char *s = NULL; if (asprintf(&s, "%.*spacemaker%s", (int) (dot-name), name, dot) == -1) { /* If asprintf fails, just return name. */ return strdup(name); } return s; } else { return strdup(name); } } else if (add_instance_name && *(name+strlen(name)-1) == '@') { return crm_strdup_printf("%spacemaker.service", name); } else { return crm_strdup_printf("%s.service", name); } } static void systemd_daemon_reload_complete(DBusPendingCall *pending, void *user_data) { DBusError error; DBusMessage *reply = NULL; unsigned int reload_count = GPOINTER_TO_UINT(user_data); dbus_error_init(&error); if(pending) { reply = dbus_pending_call_steal_reply(pending); } if (pcmk_dbus_find_error(pending, reply, &error)) { crm_warn("Could not issue systemd reload %d: %s", reload_count, error.message); dbus_error_free(&error); } else { crm_trace("Reload %d complete", reload_count); } if(pending) { dbus_pending_call_unref(pending); } if(reply) { dbus_message_unref(reply); } } static bool systemd_daemon_reload(int timeout) { static unsigned int reload_count = 0; DBusMessage *msg = systemd_new_method("Reload"); reload_count++; pcmk__assert(msg != NULL); systemd_send(msg, systemd_daemon_reload_complete, GUINT_TO_POINTER(reload_count), timeout); dbus_message_unref(msg); return TRUE; } /*! * \internal * \brief Set an action result based on a method error * * \param[in,out] op Action to set result for * \param[in] error Method error */ static void set_result_from_method_error(svc_action_t *op, const DBusError *error) { services__set_result(op, PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_ERROR, "Unable to invoke systemd DBus method"); if (strstr(error->name, "org.freedesktop.systemd1.InvalidName") || strstr(error->name, "org.freedesktop.systemd1.LoadFailed") || strstr(error->name, "org.freedesktop.systemd1.NoSuchUnit")) { if (pcmk__str_eq(op->action, PCMK_ACTION_STOP, pcmk__str_casei)) { crm_trace("Masking systemd stop failure (%s) for %s " "because unknown service can be considered stopped", error->name, pcmk__s(op->rsc, "unknown resource")); services__set_result(op, PCMK_OCF_OK, PCMK_EXEC_DONE, NULL); return; } services__format_result(op, PCMK_OCF_NOT_INSTALLED, PCMK_EXEC_NOT_INSTALLED, "systemd unit %s not found", op->agent); } crm_info("DBus request for %s of systemd unit %s%s%s failed: %s", op->action, op->agent, ((op->rsc == NULL)? "" : " for resource "), pcmk__s(op->rsc, ""), error->message); } /*! * \internal * \brief Extract unit path from LoadUnit reply, and execute action * * \param[in] reply LoadUnit reply * \param[in,out] op Action to execute (or NULL to just return path) * * \return DBus object path for specified unit if successful (only valid for * lifetime of \p reply), otherwise NULL */ static const char * execute_after_loadunit(DBusMessage *reply, svc_action_t *op) { const char *path = NULL; DBusError error; /* path here is not used other than as a non-NULL flag to indicate that a * request was indeed sent */ if (pcmk_dbus_find_error((void *) &path, reply, &error)) { if (op != NULL) { set_result_from_method_error(op, &error); } dbus_error_free(&error); } else if (!pcmk_dbus_type_check(reply, NULL, DBUS_TYPE_OBJECT_PATH, __func__, __LINE__)) { if (op != NULL) { services__set_result(op, PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_ERROR, "systemd DBus method had unexpected reply"); crm_info("Could not load systemd unit %s for %s: " "DBus reply has unexpected type", op->agent, op->id); } else { crm_info("Could not load systemd unit: " "DBus reply has unexpected type"); } } else { dbus_message_get_args (reply, NULL, DBUS_TYPE_OBJECT_PATH, &path, DBUS_TYPE_INVALID); } if (op != NULL) { if (path != NULL) { invoke_unit_by_path(op, path); } else if (!(op->synchronous)) { services__format_result(op, PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_ERROR, "No DBus object found for systemd unit %s", op->agent); services__finalize_async_op(op); } } return path; } /*! * \internal * \brief Execute a systemd action after its LoadUnit completes * * \param[in,out] pending If not NULL, DBus call associated with LoadUnit * \param[in,out] user_data Action to execute */ static void loadunit_completed(DBusPendingCall *pending, void *user_data) { DBusMessage *reply = NULL; svc_action_t *op = user_data; crm_trace("LoadUnit result for %s arrived", op->id); // Grab the reply if (pending != NULL) { reply = dbus_pending_call_steal_reply(pending); } // The call is no longer pending CRM_LOG_ASSERT(pending == op->opaque->pending); services_set_op_pending(op, NULL); // Execute the desired action based on the reply execute_after_loadunit(reply, user_data); if (reply != NULL) { dbus_message_unref(reply); } } /*! * \internal * \brief Execute a systemd action, given the unit name * * \param[in] arg_name Unit name (possibly without ".service" extension) * \param[in,out] op Action to execute (if NULL, just get object path) * \param[out] path If non-NULL and \p op is NULL or synchronous, where * to store DBus object path for specified unit * * \return Standard Pacemaker return code (for NULL \p op, pcmk_rc_ok means unit * was found; for synchronous actions, pcmk_rc_ok means unit was * executed, with the actual result stored in \p op; for asynchronous * actions, pcmk_rc_ok means action was initiated) * \note It is the caller's responsibility to free the path. */ static int invoke_unit_by_name(const char *arg_name, svc_action_t *op, char **path) { DBusMessage *msg; DBusMessage *reply = NULL; DBusPendingCall *pending = NULL; char *name = NULL; if (!systemd_init()) { if (op != NULL) { services__set_result(op, PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_ERROR, "No DBus connection"); } return ENOTCONN; } /* Create a LoadUnit DBus method (equivalent to GetUnit if already loaded), * which makes the unit usable via further DBus methods. * * * * * */ msg = systemd_new_method("LoadUnit"); pcmk__assert(msg != NULL); // Add the (expanded) unit name as the argument name = systemd_service_name(arg_name, (op == NULL) || pcmk__str_eq(op->action, PCMK_ACTION_META_DATA, pcmk__str_none)); CRM_LOG_ASSERT(dbus_message_append_args(msg, DBUS_TYPE_STRING, &name, DBUS_TYPE_INVALID)); free(name); if ((op == NULL) || op->synchronous) { // For synchronous ops, wait for a reply and extract the result const char *unit = NULL; int rc = pcmk_rc_ok; reply = systemd_send_recv(msg, NULL, (op? op->timeout : DBUS_TIMEOUT_USE_DEFAULT)); dbus_message_unref(msg); unit = execute_after_loadunit(reply, op); if (unit == NULL) { rc = ENOENT; if (path != NULL) { *path = NULL; } } else if (path != NULL) { *path = strdup(unit); if (*path == NULL) { rc = ENOMEM; } } if (reply != NULL) { dbus_message_unref(reply); } return rc; } // For asynchronous ops, initiate the LoadUnit call and return pending = systemd_send(msg, loadunit_completed, op, op->timeout); if (pending == NULL) { services__set_result(op, PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_ERROR, "Unable to send DBus message"); dbus_message_unref(msg); return ECOMM; } // LoadUnit was successfully initiated services__set_result(op, PCMK_OCF_UNKNOWN, PCMK_EXEC_PENDING, NULL); services_set_op_pending(op, pending); dbus_message_unref(msg); return pcmk_rc_ok; } /*! * \internal * \brief Compare two strings alphabetically (case-insensitive) * * \param[in] a First string to compare * \param[in] b Second string to compare * * \return 0 if strings are equal, -1 if a < b, 1 if a > b * * \note Usable as a GCompareFunc with g_list_sort(). * NULL is considered less than non-NULL. */ static gint sort_str(gconstpointer a, gconstpointer b) { if (!a && !b) { return 0; } else if (!a) { return -1; } else if (!b) { return 1; } return strcasecmp(a, b); } GList * systemd_unit_listall(void) { int nfiles = 0; GList *units = NULL; DBusMessageIter args; DBusMessageIter unit; DBusMessageIter elem; DBusMessage *reply = NULL; if (systemd_init() == FALSE) { return NULL; } /* " \n" \ " \n" \ " \n" \ */ reply = systemd_call_simple_method("ListUnitFiles"); if (reply == NULL) { return NULL; } if (!dbus_message_iter_init(reply, &args)) { crm_err("Could not list systemd unit files: systemd reply has no arguments"); dbus_message_unref(reply); return NULL; } if (!pcmk_dbus_type_check(reply, &args, DBUS_TYPE_ARRAY, __func__, __LINE__)) { crm_err("Could not list systemd unit files: systemd reply has invalid arguments"); dbus_message_unref(reply); return NULL; } dbus_message_iter_recurse(&args, &unit); for (; dbus_message_iter_get_arg_type(&unit) != DBUS_TYPE_INVALID; dbus_message_iter_next(&unit)) { DBusBasicValue value; const char *match = NULL; char *unit_name = NULL; char *basename = NULL; if(!pcmk_dbus_type_check(reply, &unit, DBUS_TYPE_STRUCT, __func__, __LINE__)) { crm_warn("Skipping systemd reply argument with unexpected type"); continue; } dbus_message_iter_recurse(&unit, &elem); if(!pcmk_dbus_type_check(reply, &elem, DBUS_TYPE_STRING, __func__, __LINE__)) { crm_warn("Skipping systemd reply argument with no string"); continue; } dbus_message_iter_get_basic(&elem, &value); if (value.str == NULL) { crm_debug("ListUnitFiles reply did not provide a string"); continue; } crm_trace("DBus ListUnitFiles listed: %s", value.str); match = systemd_unit_extension(value.str); if (match == NULL) { // This is not a unit file type we know how to manage crm_debug("ListUnitFiles entry '%s' is not supported as resource", value.str); continue; } // ListUnitFiles returns full path names, we just want base name basename = strrchr(value.str, '/'); if (basename) { basename = basename + 1; } else { basename = value.str; } if (!strcmp(match, ".service")) { // Service is the "default" unit type, so strip it unit_name = strndup(basename, match - basename); } else { unit_name = strdup(basename); } nfiles++; units = g_list_prepend(units, unit_name); } dbus_message_unref(reply); crm_trace("Found %d manageable systemd unit files", nfiles); units = g_list_sort(units, sort_str); return units; } gboolean systemd_unit_exists(const char *name) { char *path = NULL; char *state = NULL; /* Note: Makes a blocking dbus calls * Used by resources_find_service_class() when resource class=service */ if ((invoke_unit_by_name(name, NULL, &path) != pcmk_rc_ok) || (path == NULL)) { return FALSE; } /* A successful LoadUnit is not sufficient to determine the unit's * existence; it merely means the LoadUnit request received a reply. * We must make another blocking call to check the LoadState property. */ state = systemd_get_property(path, "LoadState", NULL, NULL, NULL, DBUS_TIMEOUT_USE_DEFAULT); free(path); if (pcmk__str_any_of(state, "loaded", "masked", NULL)) { free(state); return TRUE; } free(state); return FALSE; } // @TODO Use XML string constants and maybe a real XML object #define METADATA_FORMAT \ "\n" \ "<" PCMK_XE_RESOURCE_AGENT " " \ PCMK_XA_NAME "=\"%s\" " \ PCMK_XA_VERSION "=\"" PCMK_DEFAULT_AGENT_VERSION "\">\n" \ " <" PCMK_XE_VERSION ">1.1\n" \ " <" PCMK_XE_LONGDESC " " PCMK_XA_LANG "=\"" PCMK__VALUE_EN "\">\n" \ " %s\n" \ " \n" \ " <" PCMK_XE_SHORTDESC " " PCMK_XA_LANG "=\"" PCMK__VALUE_EN "\">" \ "systemd unit file for %s" \ "\n" \ - " <" PCMK_XE_PARAMETERS "/>\n" \ + " <" PCMK_XE_PARAMETERS ">\n" \ + " <" PCMK_XE_PARAMETER " " PCMK_XA_NAME "=\"" \ + PCMK_XA_USE_MONITOR_PENDING_TIMEOUT "\"" \ + " " PCMK_XA_UNIQUE "=\"0\">\n" \ + " <" PCMK_XE_LONGDESC " " PCMK_XA_LANG "=\"" PCMK__VALUE_EN "\">" \ + "It specifies whether a timeout will be triggered when the monitor" \ + " remains in a pending state.\n" \ + " \n" \ + " <" PCMK_XE_SHORTDESC " " PCMK_XA_LANG "=\"" PCMK__VALUE_EN "\">" \ + "Enables timeout when monitor is pending.\n" \ + " <" PCMK_XE_CONTENT " type=\"boolean\"" \ + " " PCMK_XA_DEFAULT "=\"false\" />\n" \ + " \n" \ + " \n" \ " <" PCMK_XE_ACTIONS ">\n" \ " <" PCMK_XE_ACTION " " PCMK_XA_NAME "=\"" PCMK_ACTION_START "\"" \ " " PCMK_META_TIMEOUT "=\"100s\" />\n" \ " <" PCMK_XE_ACTION " " PCMK_XA_NAME "=\"" PCMK_ACTION_STOP "\"" \ " " PCMK_META_TIMEOUT "=\"100s\" />\n" \ " <" PCMK_XE_ACTION " " PCMK_XA_NAME "=\"" PCMK_ACTION_STATUS "\"" \ " " PCMK_META_TIMEOUT "=\"100s\" />\n" \ " <" PCMK_XE_ACTION " " PCMK_XA_NAME "=\"" PCMK_ACTION_MONITOR "\"" \ " " PCMK_META_TIMEOUT "=\"100s\"" \ " " PCMK_META_INTERVAL "=\"60s\" />\n" \ " <" PCMK_XE_ACTION " " PCMK_XA_NAME "=\"" PCMK_ACTION_META_DATA "\"" \ " " PCMK_META_TIMEOUT "=\"5s\" />\n" \ " \n" \ " <" PCMK_XE_SPECIAL " " PCMK_XA_TAG "=\"systemd\"/>\n" \ "\n" static char * systemd_unit_metadata(const char *name, int timeout) { char *meta = NULL; char *desc = NULL; char *path = NULL; if (invoke_unit_by_name(name, NULL, &path) == pcmk_rc_ok) { /* TODO: Worth a making blocking call for? Probably not. Possibly if cached. */ desc = systemd_get_property(path, "Description", NULL, NULL, NULL, timeout); } else { desc = crm_strdup_printf("Systemd unit file for %s", name); } if (pcmk__xml_needs_escape(desc, pcmk__xml_escape_text)) { gchar *escaped = pcmk__xml_escape(desc, pcmk__xml_escape_text); meta = crm_strdup_printf(METADATA_FORMAT, name, escaped, name); g_free(escaped); } else { meta = crm_strdup_printf(METADATA_FORMAT, name, desc, name); } free(desc); free(path); return meta; } /*! * \internal * \brief Determine result of method from reply * * \param[in] reply Reply to start, stop, or restart request * \param[in,out] op Action that was executed */ static void process_unit_method_reply(DBusMessage *reply, svc_action_t *op) { DBusError error; dbus_error_init(&error); /* The first use of error here is not used other than as a non-NULL flag to * indicate that a request was indeed sent */ if (pcmk_dbus_find_error((void *) &error, reply, &error)) { set_result_from_method_error(op, &error); dbus_error_free(&error); } else if (!pcmk_dbus_type_check(reply, NULL, DBUS_TYPE_OBJECT_PATH, __func__, __LINE__)) { crm_info("DBus request for %s of %s succeeded but " "return type was unexpected", op->action, pcmk__s(op->rsc, "unknown resource")); services__set_result(op, PCMK_OCF_OK, PCMK_EXEC_DONE, "systemd DBus method had unexpected reply"); } else { const char *path = NULL; dbus_message_get_args(reply, NULL, DBUS_TYPE_OBJECT_PATH, &path, DBUS_TYPE_INVALID); crm_debug("DBus request for %s of %s using %s succeeded", op->action, pcmk__s(op->rsc, "unknown resource"), path); services__set_result(op, PCMK_OCF_OK, PCMK_EXEC_DONE, NULL); } } /*! * \internal * \brief Process the completion of an asynchronous unit start, stop, or restart * * \param[in,out] pending If not NULL, DBus call associated with request * \param[in,out] user_data Action that was executed */ static void unit_method_complete(DBusPendingCall *pending, void *user_data) { DBusMessage *reply = NULL; svc_action_t *op = user_data; crm_trace("Result for %s arrived", op->id); // Grab the reply if (pending != NULL) { reply = dbus_pending_call_steal_reply(pending); } // The call is no longer pending CRM_LOG_ASSERT(pending == op->opaque->pending); services_set_op_pending(op, NULL); // Determine result and finalize action process_unit_method_reply(reply, op); services__finalize_async_op(op); if (reply != NULL) { dbus_message_unref(reply); } } #define SYSTEMD_OVERRIDE_ROOT "/run/systemd/system/" /* When the cluster manages a systemd resource, we create a unit file override * to order the service "before" pacemaker. The "before" relationship won't * actually be used, since systemd won't ever start the resource -- we're * interested in the reverse shutdown ordering it creates, to ensure that * systemd doesn't stop the resource at shutdown while pacemaker is still * running. * * @TODO Add start timeout */ #define SYSTEMD_OVERRIDE_TEMPLATE \ "[Unit]\n" \ "Description=Cluster Controlled %s\n" \ "Before=pacemaker.service pacemaker_remote.service\n" \ "\n" \ "[Service]\n" \ "Restart=no\n" // Temporarily use rwxr-xr-x umask when opening a file for writing static FILE * create_world_readable(const char *filename) { mode_t orig_umask = umask(S_IWGRP | S_IWOTH); FILE *fp = fopen(filename, "w"); umask(orig_umask); return fp; } static void create_override_dir(const char *agent) { char *override_dir = crm_strdup_printf(SYSTEMD_OVERRIDE_ROOT "/%s.service.d", agent); int rc = pcmk__build_path(override_dir, 0755); if (rc != pcmk_rc_ok) { crm_warn("Could not create systemd override directory %s: %s", override_dir, pcmk_rc_str(rc)); } free(override_dir); } static char * get_override_filename(const char *agent) { return crm_strdup_printf(SYSTEMD_OVERRIDE_ROOT "/%s.service.d/50-pacemaker.conf", agent); } static void systemd_create_override(const char *agent, int timeout) { FILE *file_strm = NULL; char *override_file = get_override_filename(agent); create_override_dir(agent); /* Ensure the override file is world-readable. This is not strictly * necessary, but it avoids a systemd warning in the logs. */ file_strm = create_world_readable(override_file); if (file_strm == NULL) { crm_err("Cannot open systemd override file %s for writing", override_file); } else { char *override = crm_strdup_printf(SYSTEMD_OVERRIDE_TEMPLATE, agent); int rc = fprintf(file_strm, "%s\n", override); free(override); if (rc < 0) { crm_perror(LOG_WARNING, "Cannot write to systemd override file %s", override_file); } fflush(file_strm); fclose(file_strm); systemd_daemon_reload(timeout); } free(override_file); } static void systemd_remove_override(const char *agent, int timeout) { char *override_file = get_override_filename(agent); int rc = unlink(override_file); if (rc < 0) { // Stop may be called when already stopped, which is fine crm_perror(LOG_DEBUG, "Cannot remove systemd override file %s", override_file); } else { systemd_daemon_reload(timeout); } free(override_file); } /*! * \internal * \brief Parse result of systemd status check * * Set a status action's exit status and execution status based on a DBus * property check result, and finalize the action if asynchronous. * * \param[in] name DBus interface name for property that was checked * \param[in] state Property value * \param[in,out] userdata Status action that check was done for */ static void parse_status_result(const char *name, const char *state, void *userdata) { svc_action_t *op = userdata; crm_trace("Resource %s has %s='%s'", pcmk__s(op->rsc, "(unspecified)"), name, pcmk__s(state, "")); if (pcmk__str_eq(state, "active", pcmk__str_none)) { services__set_result(op, PCMK_OCF_OK, PCMK_EXEC_DONE, NULL); } else if (pcmk__str_eq(state, "reloading", pcmk__str_none)) { services__set_result(op, PCMK_OCF_OK, PCMK_EXEC_DONE, NULL); } else if (pcmk__str_eq(state, "activating", pcmk__str_none)) { services__set_result(op, PCMK_OCF_UNKNOWN, PCMK_EXEC_PENDING, NULL); } else if (pcmk__str_eq(state, "deactivating", pcmk__str_none)) { services__set_result(op, PCMK_OCF_UNKNOWN, PCMK_EXEC_PENDING, NULL); } else { services__set_result(op, PCMK_OCF_NOT_RUNNING, PCMK_EXEC_DONE, state); } if (!(op->synchronous)) { services_set_op_pending(op, NULL); services__finalize_async_op(op); } } /*! * \internal * \brief Invoke a systemd unit, given its DBus object path * * \param[in,out] op Action to execute * \param[in] unit DBus object path of systemd unit to invoke */ static void invoke_unit_by_path(svc_action_t *op, const char *unit) { const char *method = NULL; DBusMessage *msg = NULL; DBusMessage *reply = NULL; if (pcmk__str_any_of(op->action, PCMK_ACTION_MONITOR, PCMK_ACTION_STATUS, NULL)) { DBusPendingCall *pending = NULL; char *state; state = systemd_get_property(unit, "ActiveState", (op->synchronous? NULL : parse_status_result), op, (op->synchronous? NULL : &pending), op->timeout); if (op->synchronous) { parse_status_result("ActiveState", state, op); free(state); } else if (pending == NULL) { // Could not get ActiveState property services__format_result(op, PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_ERROR, "Could not get state for unit %s from DBus", op->agent); services__finalize_async_op(op); } else { services_set_op_pending(op, pending); } return; } else if (pcmk__str_eq(op->action, PCMK_ACTION_START, pcmk__str_none)) { method = "StartUnit"; systemd_create_override(op->agent, op->timeout); } else if (pcmk__str_eq(op->action, PCMK_ACTION_STOP, pcmk__str_none)) { method = "StopUnit"; systemd_remove_override(op->agent, op->timeout); } else if (pcmk__str_eq(op->action, "restart", pcmk__str_none)) { method = "RestartUnit"; } else { services__format_result(op, PCMK_OCF_UNIMPLEMENT_FEATURE, PCMK_EXEC_ERROR, "Action %s not implemented " "for systemd resources", pcmk__s(op->action, "(unspecified)")); if (!(op->synchronous)) { services__finalize_async_op(op); } return; } crm_trace("Calling %s for unit path %s%s%s", method, unit, ((op->rsc == NULL)? "" : " for resource "), pcmk__s(op->rsc, "")); msg = systemd_new_method(method); pcmk__assert(msg != NULL); /* (ss) */ { const char *replace_s = "replace"; char *name = systemd_service_name(op->agent, pcmk__str_eq(op->action, PCMK_ACTION_META_DATA, pcmk__str_none)); CRM_LOG_ASSERT(dbus_message_append_args(msg, DBUS_TYPE_STRING, &name, DBUS_TYPE_INVALID)); CRM_LOG_ASSERT(dbus_message_append_args(msg, DBUS_TYPE_STRING, &replace_s, DBUS_TYPE_INVALID)); free(name); } if (op->synchronous) { reply = systemd_send_recv(msg, NULL, op->timeout); dbus_message_unref(msg); process_unit_method_reply(reply, op); if (reply != NULL) { dbus_message_unref(reply); } } else { DBusPendingCall *pending = systemd_send(msg, unit_method_complete, op, op->timeout); dbus_message_unref(msg); if (pending == NULL) { services__set_result(op, PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_ERROR, "Unable to send DBus message"); services__finalize_async_op(op); } else { services_set_op_pending(op, pending); } } } static gboolean systemd_timeout_callback(gpointer p) { svc_action_t * op = p; op->opaque->timerid = 0; crm_info("%s action for systemd unit %s named '%s' timed out", op->action, op->agent, op->rsc); services__format_result(op, PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_TIMEOUT, "%s action for systemd unit %s " "did not complete in time", op->action, op->agent); services__finalize_async_op(op); return FALSE; } /*! * \internal * \brief Execute a systemd action * * \param[in,out] op Action to execute * * \return Standard Pacemaker return code * \retval EBUSY Recurring operation could not be initiated * \retval pcmk_rc_error Synchronous action failed * \retval pcmk_rc_ok Synchronous action succeeded, or asynchronous action * should not be freed (because it's pending or because * it failed to execute and was already freed) * * \note If the return value for an asynchronous action is not pcmk_rc_ok, the * caller is responsible for freeing the action. */ int services__execute_systemd(svc_action_t *op) { pcmk__assert(op != NULL); if ((op->action == NULL) || (op->agent == NULL)) { services__set_result(op, PCMK_OCF_NOT_CONFIGURED, PCMK_EXEC_ERROR_FATAL, "Bug in action caller"); goto done; } if (!systemd_init()) { services__set_result(op, PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_ERROR, "No DBus connection"); goto done; } crm_debug("Performing %ssynchronous %s op on systemd unit %s%s%s", (op->synchronous? "" : "a"), op->action, op->agent, ((op->rsc == NULL)? "" : " for resource "), pcmk__s(op->rsc, "")); if (pcmk__str_eq(op->action, PCMK_ACTION_META_DATA, pcmk__str_casei)) { op->stdout_data = systemd_unit_metadata(op->agent, op->timeout); services__set_result(op, PCMK_OCF_OK, PCMK_EXEC_DONE, NULL); goto done; } /* invoke_unit_by_name() should always override these values, which are here * just as a fail-safe in case there are any code paths that neglect to */ services__set_result(op, PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_ERROR, "Bug in service library"); if (invoke_unit_by_name(op->agent, op, NULL) == pcmk_rc_ok) { op->opaque->timerid = g_timeout_add(op->timeout + 5000, systemd_timeout_callback, op); services_add_inflight_op(op); return pcmk_rc_ok; } done: if (op->synchronous) { return (op->rc == PCMK_OCF_OK)? pcmk_rc_ok : pcmk_rc_error; } else { return services__finalize_async_op(op); } }