diff --git a/daemons/execd/execd_commands.c b/daemons/execd/execd_commands.c index cc9d2afbdc..3c738289aa 100644 --- a/daemons/execd/execd_commands.c +++ b/daemons/execd/execd_commands.c @@ -1,2000 +1,2001 @@ /* * Copyright 2012-2025 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #include #include #include #include // xmlNode // 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); } switch (cmd->result.execution_status) { case PCMK_EXEC_DONE: g_string_append_printf(str, ") exited with status %d", cmd->result.exit_status); break; case PCMK_EXEC_CANCELLED: g_string_append_printf(str, ") cancelled"); break; default: pcmk__g_strcat(str, ") could not be executed: ", pcmk_exec_status_str(cmd->result.execution_status), NULL); break; } 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 = pcmk__xpath_find_one(msg->doc, "//" PCMK__XE_LRMD_RSC, LOG_ERR); lrmd_rsc_t *rsc = NULL; rsc = pcmk__assert_alloc(1, sizeof(lrmd_rsc_t)); pcmk__xe_get_int(msg, PCMK__XA_LRMD_CALLOPT, &rsc->call_opts); rsc->rsc_id = pcmk__xe_get_copy(rsc_xml, PCMK__XA_LRMD_RSC_ID); rsc->class = pcmk__xe_get_copy(rsc_xml, PCMK__XA_LRMD_CLASS); rsc->provider = pcmk__xe_get_copy(rsc_xml, PCMK__XA_LRMD_PROVIDER); rsc->type = pcmk__xe_get_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 = pcmk__xpath_find_one(msg->doc, "//" PCMK__XE_LRMD_RSC, LOG_ERR); lrmd_cmd_t *cmd = NULL; cmd = pcmk__assert_alloc(1, sizeof(lrmd_cmd_t)); pcmk__xe_get_int(msg, PCMK__XA_LRMD_CALLOPT, &call_options); cmd->call_opts = call_options; cmd->client_id = pcmk__str_copy(client->id); pcmk__xe_get_int(msg, PCMK__XA_LRMD_CALLID, &cmd->call_id); pcmk__xe_get_guint(rsc_xml, PCMK__XA_LRMD_RSC_INTERVAL, &cmd->interval_ms); pcmk__xe_get_int(rsc_xml, PCMK__XA_LRMD_TIMEOUT, &cmd->timeout); pcmk__xe_get_int(rsc_xml, PCMK__XA_LRMD_RSC_START_DELAY, &cmd->start_delay); cmd->timeout_orig = cmd->timeout; cmd->origin = pcmk__xe_get_copy(rsc_xml, PCMK__XA_LRMD_ORIGIN); cmd->action = pcmk__xe_get_copy(rsc_xml, PCMK__XA_LRMD_RSC_ACTION); cmd->userdata_str = pcmk__xe_get_copy(rsc_xml, PCMK__XA_LRMD_RSC_USERDATA_STR); cmd->rsc_id = pcmk__xe_get_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)) { return; } cmd->stonith_recurring_id = pcmk__create_timer(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 = pcmk__create_timer(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); + pcmk__xe_set_guint(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); pcmk__xe_set_time(notify, PCMK__XA_LRMD_RUN_TIME, cmd->epoch_last_run); pcmk__xe_set_time(notify, PCMK__XA_LRMD_RCCHANGE_TIME, 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 = pcmk__xpath_find_one(request->doc, "//" PCMK__XE_LRMD_RSC, LOG_ERR); const char *rsc_id = pcmk__xe_get(rsc_xml, PCMK__XA_LRMD_RSC_ID); const char *op = pcmk__xe_get(request, PCMK__XA_LRMD_OP); pcmk__xe_get_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; int time_sum = 0; int timeout_left = 0; int delay = 0; #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)) { goto finalize; } if (pcmk__result_ok(&(cmd->result)) && pcmk__strcase_any_of(cmd->action, PCMK_ACTION_START, PCMK_ACTION_STOP, NULL)) { /* Getting results for when a start or stop action completes is now * handled by watching for JobRemoved() signals from systemd and * reacting to them. So, we can bypass the rest of the code in this * function for those actions, and simply finalize cmd. * * @TODO When monitors are handled in the same way, this function * can either be drastically simplified or done away with entirely. */ services__copy_result(action, &(cmd->result)); goto finalize; } else if (cmd->result.execution_status == PCMK_EXEC_PENDING && pcmk__str_any_of(cmd->action, PCMK_ACTION_MONITOR, PCMK_ACTION_STATUS, NULL) && cmd->interval_ms == 0 && cmd->real_action == NULL) { /* If the state is Pending at the time of probe, execute follow-up monitor. */ 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/probe(monitor) 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__str_any_of(cmd->action, PCMK_ACTION_MONITOR, PCMK_ACTION_STATUS, NULL) && (cmd->interval_ms > 0)) { /* For monitors, excluding follow-up monitors, */ /* if the pending state persists from the first notification until its timeout, */ /* it will be treated as a timeout. */ 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); } } } if (!goagain) { goto finalize; } time_sum = time_diff_ms(NULL, &(cmd->t_first_run)); timeout_left = cmd->timeout_orig - time_sum; 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 finalize: 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, pcmk__timeout_ms2s(cmd->timeout)); 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) { services__copy_result(action, &(cmd->result)); 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. */ services__copy_result(action, &(cmd->result)); 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 = pcmk__xe_get(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 = pcmk__xe_get(request, PCMK__XA_LRMD_OP); // This is a remote connection from a cluster node's controller ipc_proxy_add_provider(client); /* @TODO Allowing multiple proxies makes no sense given that clients * have no way to choose between them. Maybe always use the most * recent one and switch any existing IPC connections to use it, * by iterating over ipc_clients here, and if client->id doesn't * match the client's userdata, replace the userdata with the new * ID. After the iteration, call lrmd_remote_client_destroy() on any * of the replaced values in ipc_providers. */ /* 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); pcmk__xe_set_time(*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 = pcmk__xpath_find_one(request->doc, "//" PCMK__XE_LRMD_RSC, LOG_ERR); const char *rsc_id = pcmk__xe_get(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 = pcmk__xpath_find_one(request->doc, "//" PCMK__XE_LRMD_RSC, LOG_ERR); const char *rsc_id = pcmk__xe_get(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 = pcmk__xpath_find_one(request->doc, "//" PCMK__XE_LRMD_RSC, LOG_ERR); const char *rsc_id = pcmk__xe_get(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 = pcmk__xpath_find_one(request->doc, "//" PCMK__XE_LRMD_RSC, LOG_ERR); const char *rsc_id = pcmk__xe_get(rsc_xml, PCMK__XA_LRMD_RSC_ID); const char *action = pcmk__xe_get(rsc_xml, PCMK__XA_LRMD_RSC_ACTION); guint interval_ms = 0; pcmk__xe_get_guint(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); + pcmk__xe_set_guint(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 = pcmk__xe_get(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 = pcmk__xe_get(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); pcmk__xe_get_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 = pcmk__xe_get(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_element_internal.h b/include/crm/common/xml_element_internal.h index 2bfa2f64ff..dfe66625e2 100644 --- a/include/crm/common/xml_element_internal.h +++ b/include/crm/common/xml_element_internal.h @@ -1,200 +1,203 @@ /* * Copyright 2017-2025 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #ifndef PCMK__CRM_COMMON_XML_ELEMENT_INTERNAL__H #define PCMK__CRM_COMMON_XML_ELEMENT_INTERNAL__H /* * Internal-only wrappers for and extensions to libxml2 for processing XML * elements */ #include // bool #include // uint32_t #include // NULL #include // strcmp() #include // guint #include // xmlNode, etc. #include // crm_time_t #include // pcmk__str_copy() #include // PCMK_XA_ID #ifdef __cplusplus extern "C" { #endif const char *pcmk__xe_add_last_written(xmlNode *xe); xmlNode *pcmk__xe_first_child(const xmlNode *parent, const char *node_name, const char *attr_n, const char *attr_v); void pcmk__xe_remove_attr(xmlNode *element, const char *name); bool pcmk__xe_remove_attr_cb(xmlNode *xml, void *user_data); void pcmk__xe_remove_matching_attrs(xmlNode *element, bool force, bool (*match)(xmlAttrPtr, void *), void *user_data); int pcmk__xe_delete_match(xmlNode *xml, xmlNode *search); int pcmk__xe_replace_match(xmlNode *xml, xmlNode *replace); int pcmk__xe_update_match(xmlNode *xml, xmlNode *update, uint32_t flags); /*! * \internal * \brief Check whether an XML element is of a particular type * * \param[in] xml XML element to compare * \param[in] name XML element name to compare * * \return \c true if \p xml is of type \p name, otherwise \c false */ static inline bool pcmk__xe_is(const xmlNode *xml, const char *name) { return (xml != NULL) && (xml->name != NULL) && (name != NULL) && (strcmp((const char *) xml->name, name) == 0); } xmlNode *pcmk__xe_create(xmlNode *parent, const char *name); xmlNode *pcmk__xe_next(const xmlNode *node, const char *element_name); void pcmk__xe_set_content(xmlNode *node, const char *format, ...) G_GNUC_PRINTF(2, 3); int pcmk__xe_get_score(const xmlNode *xml, const char *name, int *score, int default_score); int pcmk__xe_copy_attrs(xmlNode *target, const xmlNode *src, uint32_t flags); void pcmk__xe_sort_attrs(xmlNode *xml); void pcmk__xe_set_id(xmlNode *xml, const char *format, ...) G_GNUC_PRINTF(2, 3); /*! * \internal * \brief Like pcmk__xe_set_props, but takes a va_list instead of * arguments directly. * * \param[in,out] node XML to add attributes to * \param[in] pairs NULL-terminated list of name/value pairs to add */ void pcmk__xe_set_propv(xmlNodePtr node, va_list pairs); /*! * \internal * \brief Add a NULL-terminated list of name/value pairs to the given * XML node as properties. * * \param[in,out] node XML node to add properties to * \param[in] ... NULL-terminated list of name/value pairs * * \note A NULL name terminates the arguments; a NULL value will be skipped. */ void pcmk__xe_set_props(xmlNodePtr node, ...) G_GNUC_NULL_TERMINATED; /*! * \internal * \brief Get first attribute of an XML element * * \param[in] xe XML element to check * * \return First attribute of \p xe (or NULL if \p xe is NULL or has none) */ static inline xmlAttr * pcmk__xe_first_attr(const xmlNode *xe) { return (xe == NULL)? NULL : xe->properties; } /*! * \internal * \brief Iterate over child elements of \p xml * * This function iterates over the children of \p xml, performing the * callback function \p handler on each node. If the callback returns * a value other than pcmk_rc_ok, the iteration stops and the value is * returned. It is therefore possible that not all children will be * visited. * * \param[in,out] xml The starting XML node. Can be NULL. * \param[in] child_element_name The name that the node must match in order * for \p handler to be run. If NULL, all * child elements will match. * \param[in] handler The callback function. * \param[in,out] userdata User data to pass to the callback function. * Can be NULL. * * \return Standard Pacemaker return code */ int pcmk__xe_foreach_child(xmlNode *xml, const char *child_element_name, int (*handler)(xmlNode *xml, void *userdata), void *userdata); const char *pcmk__xe_get(const xmlNode *xml, const char *attr_name); int pcmk__xe_get_datetime(const xmlNode *xml, const char *attr, crm_time_t **t); int pcmk__xe_get_flags(const xmlNode *xml, const char *name, uint32_t *dest, uint32_t default_value); + int pcmk__xe_get_guint(const xmlNode *xml, const char *attr, guint *dest); -int pcmk__xe_get_int(const xmlNode *xml, const char *attr, int *dest); -int pcmk__xe_get_ll(const xmlNode *xml, const char *attr, long long *dest); +void pcmk__xe_set_guint(xmlNode *xml, const char *attr, guint value); + +int pcmk__xe_get_int(const xmlNode *xml, const char *name, int *dest); +int pcmk__xe_get_ll(const xmlNode *xml, const char *name, long long *dest); int pcmk__xe_get_time(const xmlNode *xml, const char *attr, time_t *dest); void pcmk__xe_set_time(xmlNode *xml, const char *attr, time_t value); int pcmk__xe_get_timeval(const xmlNode *xml, const char *sec_attr, const char *usec_attr, struct timeval *dest); void pcmk__xe_set_timeval(xmlNode *xml, const char *sec_attr, const char *usec_attr, const struct timeval *value); void pcmk__xe_set_bool_attr(xmlNodePtr node, const char *name, bool value); int pcmk__xe_get_bool_attr(const xmlNode *node, const char *name, bool *value); bool pcmk__xe_attr_is_true(const xmlNode *node, const char *name); /*! * \internal * \brief Retrieve a copy of the value of an XML attribute * * This is like \c pcmk__xe_get() but allocates new memory for the result. * * \param[in] xml XML element whose attribute to get * \param[in] attr Attribute name * * \return Value of specified attribute (or \c NULL if not set) * * \note The caller is responsible for freeing the result using \c free(). */ static inline char * pcmk__xe_get_copy(const xmlNode *xml, const char *attr) { return pcmk__str_copy(pcmk__xe_get(xml, attr)); } /*! * \internal * \brief Retrieve the value of the \c PCMK_XA_ID XML attribute * * \param[in] xml XML element to check * * \return Value of the \c PCMK_XA_ID attribute (may be \c NULL) */ static inline const char * pcmk__xe_id(const xmlNode *xml) { return pcmk__xe_get(xml, PCMK_XA_ID); } #ifdef __cplusplus } #endif #endif // PCMK__CRM_COMMON_XML_ELEMENT_INTERNAL__H diff --git a/lib/common/xml_element.c b/lib/common/xml_element.c index 843953bb97..9f7c439991 100644 --- a/lib/common/xml_element.c +++ b/lib/common/xml_element.c @@ -1,1784 +1,1806 @@ /* * Copyright 2004-2025 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #include #include // va_start(), etc. #include // uint32_t #include // NULL, etc. #include // free(), etc. #include // strchr(), etc. #include // time_t, etc. #include // xmlNode, etc. #include // xmlValidateNameValue() #include // xmlChar #include #include // crm_xml_add(), etc. #include // pcmk_rc_ok, etc. #include #include "crmcommon_private.h" /*! * \internal * \brief Find first XML child element matching given criteria * * \param[in] parent XML element to search (can be \c NULL) * \param[in] node_name If not \c NULL, only match children of this type * \param[in] attr_n If not \c NULL, only match children with an attribute * of this name. * \param[in] attr_v If \p attr_n and this are not NULL, only match children * with an attribute named \p attr_n and this value * * \return Matching XML child element, or \c NULL if none found */ xmlNode * pcmk__xe_first_child(const xmlNode *parent, const char *node_name, const char *attr_n, const char *attr_v) { xmlNode *child = NULL; CRM_CHECK((attr_v == NULL) || (attr_n != NULL), return NULL); if (parent == NULL) { return NULL; } child = parent->children; while ((child != NULL) && (child->type != XML_ELEMENT_NODE)) { child = child->next; } for (; child != NULL; child = pcmk__xe_next(child, NULL)) { const char *value = NULL; if ((node_name != NULL) && !pcmk__xe_is(child, node_name)) { // Node name mismatch continue; } if (attr_n == NULL) { // No attribute match needed return child; } value = pcmk__xe_get(child, attr_n); if ((attr_v == NULL) && (value != NULL)) { // attr_v == NULL: Attribute attr_n must be set (to any value) return child; } if ((attr_v != NULL) && (pcmk__str_eq(value, attr_v, pcmk__str_none))) { // attr_v != NULL: Attribute attr_n must be set to value attr_v return child; } } if (attr_n == NULL) { crm_trace("%s XML has no child element of %s type", (const char *) parent->name, pcmk__s(node_name, "any")); } else { crm_trace("%s XML has no child element of %s type with %s='%s'", (const char *) parent->name, pcmk__s(node_name, "any"), attr_n, attr_v); } return NULL; } /*! * \internal * \brief Return next sibling element of an XML element * * \param[in] xml XML element to check * \param[in] element_name If not NULL, get next sibling with this element name * * \return Next desired sibling of \p xml (or NULL if none) */ xmlNode * pcmk__xe_next(const xmlNode *xml, const char *element_name) { for (xmlNode *next = (xml == NULL)? NULL : xml->next; next != NULL; next = next->next) { if ((next->type == XML_ELEMENT_NODE) && ((element_name == NULL) || pcmk__xe_is(next, element_name))) { return next; } } return NULL; } /*! * \internal * \brief Parse an integer score from an XML attribute * * \param[in] xml XML element with attribute to parse * \param[in] name Name of attribute to parse * \param[out] score Where to store parsed score (can be NULL to * just validate) * \param[in] default_score What to return if the attribute value is not * present or invalid * * \return Standard Pacemaker return code */ int pcmk__xe_get_score(const xmlNode *xml, const char *name, int *score, int default_score) { const char *value = NULL; CRM_CHECK((xml != NULL) && (name != NULL), return EINVAL); value = pcmk__xe_get(xml, name); return pcmk_parse_score(value, score, default_score); } /*! * \internal * \brief Set an XML attribute, expanding \c ++ and \c += where appropriate * * If \p target already has an attribute named \p name set to an integer value * and \p value is an addition assignment expression on \p name, then expand * \p value to an integer and set attribute \p name to the expanded value in * \p target. * * Otherwise, set attribute \p name on \p target using the literal \p value. * * The original attribute value in \p target and the number in an assignment * expression in \p value are parsed and added as scores (that is, their values * are capped at \c INFINITY and \c -INFINITY). For more details, refer to * \c pcmk_parse_score(). * * For example, suppose \p target has an attribute named \c "X" with value * \c "5", and that \p name is \c "X". * * If \p value is \c "X++", the new value of \c "X" in \p target is \c "6". * * If \p value is \c "X+=3", the new value of \c "X" in \p target is \c "8". * * If \p value is \c "val", the new value of \c "X" in \p target is \c "val". * * If \p value is \c "Y++", the new value of \c "X" in \p target is \c "Y++". * * \param[in,out] target XML node whose attribute to set * \param[in] name Name of the attribute to set * \param[in] value New value of attribute to set (if NULL, initial value * will be left unchanged) * * \return Standard Pacemaker return code (specifically, \c EINVAL on invalid * argument, or \c pcmk_rc_ok otherwise) */ int pcmk__xe_set_score(xmlNode *target, const char *name, const char *value) { const char *old_value = NULL; CRM_CHECK((target != NULL) && (name != NULL), return EINVAL); if (value == NULL) { // @TODO Maybe instead delete the attribute or set it to 0 return pcmk_rc_ok; } old_value = pcmk__xe_get(target, name); // If no previous value, skip to default case and set the value unexpanded. if (old_value != NULL) { const char *n = name; const char *v = value; // Stop at first character that differs between name and value for (; (*n == *v) && (*n != '\0'); n++, v++); // If value begins with name followed by a "++" or "+=" if ((*n == '\0') && (*v++ == '+') && ((*v == '+') || (*v == '='))) { int add = 1; int old_value_i = 0; int rc = pcmk_rc_ok; // If we're expanding ourselves, no previous value was set; use 0 if (old_value != value) { rc = pcmk_parse_score(old_value, &old_value_i, 0); if (rc != pcmk_rc_ok) { // @TODO This is inconsistent with old_value==NULL crm_trace("Using 0 before incrementing %s because '%s' " "is not a score", name, old_value); } } /* value="X++": new value of X is old_value + 1 * value="X+=Y": new value of X is old_value + Y (for some number Y) */ if (*v != '+') { rc = pcmk_parse_score(++v, &add, 0); if (rc != pcmk_rc_ok) { // @TODO We should probably skip expansion instead crm_trace("Not incrementing %s because '%s' does not have " "a valid increment", name, value); } } crm_xml_add_int(target, name, pcmk__add_scores(old_value_i, add)); return pcmk_rc_ok; } } // Default case: set the attribute unexpanded (with value treated literally) if (old_value != value) { crm_xml_add(target, name, value); } return pcmk_rc_ok; } /*! * \internal * \brief Copy XML attributes from a source element to a target element * * This is similar to \c xmlCopyPropList() except that attributes are marked * as dirty for change tracking purposes. * * \param[in,out] target XML element to receive copied attributes from \p src * \param[in] src XML element whose attributes to copy to \p target * \param[in] flags Group of enum pcmk__xa_flags * * \return Standard Pacemaker return code */ int pcmk__xe_copy_attrs(xmlNode *target, const xmlNode *src, uint32_t flags) { CRM_CHECK((src != NULL) && (target != NULL), return EINVAL); for (xmlAttr *attr = pcmk__xe_first_attr(src); attr != NULL; attr = attr->next) { const char *name = (const char *) attr->name; const char *value = pcmk__xml_attr_value(attr); if (pcmk_is_set(flags, pcmk__xaf_no_overwrite) && (pcmk__xe_get(target, name) != NULL)) { continue; } if (pcmk_is_set(flags, pcmk__xaf_score_update)) { pcmk__xe_set_score(target, name, value); } else { crm_xml_add(target, name, value); } } return pcmk_rc_ok; } /*! * \internal * \brief Compare two XML attributes by name * * \param[in] a First XML attribute to compare * \param[in] b Second XML attribute to compare * * \retval negative \c a->name is \c NULL or comes before \c b->name * lexicographically * \retval 0 \c a->name and \c b->name are equal * \retval positive \c b->name is \c NULL or comes before \c a->name * lexicographically */ static gint compare_xml_attr(gconstpointer a, gconstpointer b) { const xmlAttr *attr_a = a; const xmlAttr *attr_b = b; return pcmk__strcmp((const char *) attr_a->name, (const char *) attr_b->name, pcmk__str_none); } /*! * \internal * \brief Sort an XML element's attributes by name * * This does not consider ACLs and does not mark the attributes as deleted or * dirty. Upon return, all attributes still exist and are set to the same values * as before the call. The only thing that may change is the order of the * attribute list. * * \param[in,out] xml XML element whose attributes to sort */ void pcmk__xe_sort_attrs(xmlNode *xml) { GSList *attr_list = NULL; for (xmlAttr *iter = pcmk__xe_first_attr(xml); iter != NULL; iter = iter->next) { attr_list = g_slist_prepend(attr_list, iter); } attr_list = g_slist_sort(attr_list, compare_xml_attr); for (GSList *iter = attr_list; iter != NULL; iter = iter->next) { xmlNode *attr = iter->data; xmlUnlinkNode(attr); xmlAddChild(xml, attr); } g_slist_free(attr_list); } /*! * \internal * \brief Remove a named attribute from an XML element * * \param[in,out] element XML element to remove an attribute from * \param[in] name Name of attribute to remove */ void pcmk__xe_remove_attr(xmlNode *element, const char *name) { if (name != NULL) { pcmk__xa_remove(xmlHasProp(element, (const xmlChar *) name), false); } } /*! * \internal * \brief Remove a named attribute from an XML element * * This is a wrapper for \c pcmk__xe_remove_attr() for use with * \c pcmk__xml_tree_foreach(). * * \param[in,out] xml XML element to remove an attribute from * \param[in] user_data Name of attribute to remove * * \return \c true (to continue traversing the tree) * * \note This is compatible with \c pcmk__xml_tree_foreach(). */ bool pcmk__xe_remove_attr_cb(xmlNode *xml, void *user_data) { const char *name = user_data; pcmk__xe_remove_attr(xml, name); return true; } /*! * \internal * \brief Remove an XML element's attributes that match some criteria * * \param[in,out] element XML element to modify * \param[in] force If \c true, remove matching attributes immediately, * ignoring ACLs and change tracking * \param[in] match If not NULL, only remove attributes for which * this function returns true * \param[in,out] user_data Data to pass to \p match */ void pcmk__xe_remove_matching_attrs(xmlNode *element, bool force, bool (*match)(xmlAttrPtr, void *), void *user_data) { xmlAttrPtr next = NULL; for (xmlAttrPtr a = pcmk__xe_first_attr(element); a != NULL; a = next) { next = a->next; // Grab now because attribute might get removed if ((match == NULL) || match(a, user_data)) { if (pcmk__xa_remove(a, force) != pcmk_rc_ok) { return; } } } } /*! * \internal * \brief Create a new XML element under a given parent * * \param[in,out] parent XML element that will be the new element's parent * (\c NULL to create a new XML document with the new * node as root) * \param[in] name Name of new element * * \return Newly created XML element (guaranteed not to be \c NULL) */ xmlNode * pcmk__xe_create(xmlNode *parent, const char *name) { xmlNode *node = NULL; pcmk__assert(!pcmk__str_empty(name)); if (parent == NULL) { xmlDoc *doc = pcmk__xml_new_doc(); node = xmlNewDocRawNode(doc, NULL, (const xmlChar *) name, NULL); pcmk__mem_assert(node); xmlDocSetRootElement(doc, node); } else { node = xmlNewChild(parent, NULL, (const xmlChar *) name, NULL); pcmk__mem_assert(node); } pcmk__xml_new_private_data(node); return node; } /*! * \internal * \brief Set a formatted string as an XML node's content * * \param[in,out] node Node whose content to set * \param[in] format printf(3)-style format string * \param[in] ... Arguments for \p format * * \note This function escapes special characters. \c xmlNodeSetContent() does * not. */ G_GNUC_PRINTF(2, 3) void pcmk__xe_set_content(xmlNode *node, const char *format, ...) { if (node != NULL) { const char *content = NULL; char *buf = NULL; /* xmlNodeSetContent() frees node->children and replaces it with new * text. If this function is called for a node that already has a non- * text child, it's a bug. */ CRM_CHECK((node->children == NULL) || (node->children->type == XML_TEXT_NODE), return); if (strchr(format, '%') == NULL) { // Nothing to format content = format; } else { va_list ap; va_start(ap, format); if (pcmk__str_eq(format, "%s", pcmk__str_none)) { // No need to make a copy content = va_arg(ap, const char *); } else { pcmk__assert(vasprintf(&buf, format, ap) >= 0); content = buf; } va_end(ap); } xmlNodeSetContent(node, (const xmlChar *) content); free(buf); } } /*! * \internal * \brief Set a formatted string as an XML element's ID * * If the formatted string would not be a valid ID, it's first sanitized by * \c pcmk__xml_sanitize_id(). * * \param[in,out] node Node whose ID to set * \param[in] format printf(3)-style format string * \param[in] ... Arguments for \p format */ G_GNUC_PRINTF(2, 3) void pcmk__xe_set_id(xmlNode *node, const char *format, ...) { char *id = NULL; va_list ap; pcmk__assert(!pcmk__str_empty(format)); if (node == NULL) { return; } va_start(ap, format); pcmk__assert(vasprintf(&id, format, ap) >= 0); va_end(ap); if (!xmlValidateNameValue((const xmlChar *) id)) { pcmk__xml_sanitize_id(id); } crm_xml_add(node, PCMK_XA_ID, id); free(id); } /*! * \internal * \brief Add a "last written" attribute to an XML element, set to current time * * \param[in,out] xe XML element to add attribute to * * \return Value that was set, or NULL on error */ const char * pcmk__xe_add_last_written(xmlNode *xe) { char *now_s = pcmk__epoch2str(NULL, 0); const char *result = NULL; result = crm_xml_add(xe, PCMK_XA_CIB_LAST_WRITTEN, pcmk__s(now_s, "Could not determine current time")); free(now_s); return result; } /*! * \internal * \brief Merge one XML tree into another * * Here, "merge" means: * 1. Copy attribute values from \p update to the target, overwriting in case of * conflict. * 2. Descend through \p update and the target in parallel. At each level, for * each child of \p update, look for a matching child of the target. * a. For each child, if a match is found, go to step 1, recursively merging * the child of \p update into the child of the target. * b. Otherwise, copy the child of \p update as a child of the target. * * A match is defined as the first child of the same type within the target, * with: * * the \c PCMK_XA_ID attribute matching, if set in \p update; otherwise, * * the \c PCMK_XA_ID_REF attribute matching, if set in \p update * * This function does not delete any elements or attributes from the target. It * may add elements or overwrite attributes, as described above. * * \param[in,out] parent If \p target is NULL and this is not, add or update * child of this XML node that matches \p update * \param[in,out] target If not NULL, update this XML * \param[in] update Make the desired XML match this (must not be \c NULL) * \param[in] flags Group of enum pcmk__xa_flags * * \note At least one of \p parent and \p target must be non-NULL. * \note This function is recursive. For the top-level call, \p parent is * \c NULL and \p target is not \c NULL. For recursive calls, \p target is * \c NULL and \p parent is not \c NULL. */ static void update_xe(xmlNode *parent, xmlNode *target, xmlNode *update, uint32_t flags) { // @TODO Try to refactor further, possibly using pcmk__xml_tree_foreach() const char *update_name = NULL; const char *update_id_attr = NULL; const char *update_id_val = NULL; char *trace_s = NULL; crm_log_xml_trace(update, "update"); crm_log_xml_trace(target, "target"); CRM_CHECK(update != NULL, goto done); if (update->type == XML_COMMENT_NODE) { pcmk__xc_update(parent, target, update); goto done; } update_name = (const char *) update->name; CRM_CHECK(update_name != NULL, goto done); CRM_CHECK((target != NULL) || (parent != NULL), goto done); update_id_val = pcmk__xe_id(update); if (update_id_val != NULL) { update_id_attr = PCMK_XA_ID; } else { update_id_val = pcmk__xe_get(update, PCMK_XA_ID_REF); if (update_id_val != NULL) { update_id_attr = PCMK_XA_ID_REF; } } pcmk__if_tracing( { if (update_id_attr != NULL) { trace_s = crm_strdup_printf("<%s %s=%s/>", update_name, update_id_attr, update_id_val); } else { trace_s = crm_strdup_printf("<%s/>", update_name); } }, {} ); if (target == NULL) { // Recursive call target = pcmk__xe_first_child(parent, update_name, update_id_attr, update_id_val); } if (target == NULL) { // Recursive call with no existing matching child target = pcmk__xe_create(parent, update_name); crm_trace("Added %s", pcmk__s(trace_s, update_name)); } else { // Either recursive call with match, or top-level call crm_trace("Found node %s to update", pcmk__s(trace_s, update_name)); } CRM_CHECK(pcmk__xe_is(target, (const char *) update->name), return); pcmk__xe_copy_attrs(target, update, flags); for (xmlNode *child = pcmk__xml_first_child(update); child != NULL; child = pcmk__xml_next(child)) { crm_trace("Updating child of %s", pcmk__s(trace_s, update_name)); update_xe(target, NULL, child, flags); } crm_trace("Finished with %s", pcmk__s(trace_s, update_name)); done: free(trace_s); } /*! * \internal * \brief Delete an XML subtree if it matches a search element * * A match is defined as follows: * * \p xml and \p user_data are both element nodes of the same type. * * If \p user_data has attributes set, \p xml has those attributes set to the * same values. (\p xml may have additional attributes set to arbitrary * values.) * * \param[in,out] xml XML subtree to delete upon match * \param[in] user_data Search element * * \return \c true to continue traversing the tree, or \c false to stop (because * \p xml was deleted) * * \note This is compatible with \c pcmk__xml_tree_foreach(). */ static bool delete_xe_if_matching(xmlNode *xml, void *user_data) { xmlNode *search = user_data; if (!pcmk__xe_is(search, (const char *) xml->name)) { // No match: either not both elements, or different element types return true; } for (const xmlAttr *attr = pcmk__xe_first_attr(search); attr != NULL; attr = attr->next) { const char *search_val = pcmk__xml_attr_value(attr); const char *xml_val = pcmk__xe_get(xml, (const char *) attr->name); if (!pcmk__str_eq(search_val, xml_val, pcmk__str_casei)) { // No match: an attr in xml doesn't match the attr in search return true; } } crm_log_xml_trace(xml, "delete-match"); crm_log_xml_trace(search, "delete-search"); pcmk__xml_free(xml); // Found a match and deleted it; stop traversing tree return false; } /*! * \internal * \brief Search an XML tree depth-first and delete the first matching element * * This function does not attempt to match the tree root (\p xml). * * A match with a node \c node is defined as follows: * * \c node and \p search are both element nodes of the same type. * * If \p search has attributes set, \c node has those attributes set to the * same values. (\c node may have additional attributes set to arbitrary * values.) * * \param[in,out] xml XML subtree to search * \param[in] search Element to match against * * \return Standard Pacemaker return code (specifically, \c pcmk_rc_ok on * successful deletion and an error code otherwise) */ int pcmk__xe_delete_match(xmlNode *xml, xmlNode *search) { // See @COMPAT comment in pcmk__xe_replace_match() CRM_CHECK((xml != NULL) && (search != NULL), return EINVAL); for (xml = pcmk__xe_first_child(xml, NULL, NULL, NULL); xml != NULL; xml = pcmk__xe_next(xml, NULL)) { if (!pcmk__xml_tree_foreach(xml, delete_xe_if_matching, search)) { // Found and deleted an element return pcmk_rc_ok; } } // No match found in this subtree return ENXIO; } /*! * \internal * \brief Replace one XML node with a copy of another XML node * * This function handles change tracking and applies ACLs. * * \param[in,out] old XML node to replace * \param[in] new XML node to copy as replacement for \p old * * \note This frees \p old. */ static void replace_node(xmlNode *old, xmlNode *new) { // Pass old for its doc; it won't remain the parent of new new = pcmk__xml_copy(old, new); old = xmlReplaceNode(old, new); // old == NULL means memory allocation error pcmk__assert(old != NULL); // May be unnecessary but avoids slight changes to some test outputs pcmk__xml_tree_foreach(new, pcmk__xml_reset_node_flags, NULL); if (pcmk__xml_doc_all_flags_set(new->doc, pcmk__xf_tracking)) { // Replaced sections may have included relevant ACLs pcmk__apply_acl(new); } pcmk__xml_mark_changes(old, new); pcmk__xml_free_node(old); } /*! * \internal * \brief Replace one XML subtree with a copy of another if the two match * * A match is defined as follows: * * \p xml and \p user_data are both element nodes of the same type. * * If \p user_data has the \c PCMK_XA_ID attribute set, then \p xml has * \c PCMK_XA_ID set to the same value. * * \param[in,out] xml XML subtree to replace with \p user_data upon match * \param[in] user_data XML to replace \p xml with a copy of upon match * * \return \c true to continue traversing the tree, or \c false to stop (because * \p xml was replaced by \p user_data) * * \note This is compatible with \c pcmk__xml_tree_foreach(). */ static bool replace_xe_if_matching(xmlNode *xml, void *user_data) { xmlNode *replace = user_data; const char *xml_id = NULL; const char *replace_id = NULL; xml_id = pcmk__xe_id(xml); replace_id = pcmk__xe_id(replace); if (!pcmk__xe_is(replace, (const char *) xml->name)) { // No match: either not both elements, or different element types return true; } if ((replace_id != NULL) && !pcmk__str_eq(replace_id, xml_id, pcmk__str_none)) { // No match: ID was provided in replace and doesn't match xml's ID return true; } crm_log_xml_trace(xml, "replace-match"); crm_log_xml_trace(replace, "replace-with"); replace_node(xml, replace); // Found a match and replaced it; stop traversing tree return false; } /*! * \internal * \brief Search an XML tree depth-first and replace the first matching element * * This function does not attempt to match the tree root (\p xml). * * A match with a node \c node is defined as follows: * * \c node and \p replace are both element nodes of the same type. * * If \p replace has the \c PCMK_XA_ID attribute set, then \c node has * \c PCMK_XA_ID set to the same value. * * \param[in,out] xml XML tree to search * \param[in] replace XML to replace a matching element with a copy of * * \return Standard Pacemaker return code (specifically, \c pcmk_rc_ok on * successful replacement and an error code otherwise) */ int pcmk__xe_replace_match(xmlNode *xml, xmlNode *replace) { /* @COMPAT Some of this behavior (like not matching the tree root, which is * allowed by pcmk__xe_update_match()) is questionable for general use but * required for backward compatibility by cib_process_replace() and * cib_process_delete(). Behavior can change at a major version release if * desired. */ CRM_CHECK((xml != NULL) && (replace != NULL), return EINVAL); for (xml = pcmk__xe_first_child(xml, NULL, NULL, NULL); xml != NULL; xml = pcmk__xe_next(xml, NULL)) { if (!pcmk__xml_tree_foreach(xml, replace_xe_if_matching, replace)) { // Found and replaced an element return pcmk_rc_ok; } } // No match found in this subtree return ENXIO; } //! User data for \c update_xe_if_matching() struct update_data { xmlNode *update; //!< Update source uint32_t flags; //!< Group of enum pcmk__xa_flags }; /*! * \internal * \brief Update one XML subtree with another if the two match * * "Update" means to merge a source subtree into a target subtree (see * \c update_xe()). * * A match is defined as follows: * * \p xml and \p user_data->update are both element nodes of the same type. * * \p xml and \p user_data->update have the same \c PCMK_XA_ID attribute * value, or \c PCMK_XA_ID is unset in both * * \param[in,out] xml XML subtree to update with \p user_data->update * upon match * \param[in] user_data struct update_data object * * \return \c true to continue traversing the tree, or \c false to stop (because * \p xml was updated by \p user_data->update) * * \note This is compatible with \c pcmk__xml_tree_foreach(). */ static bool update_xe_if_matching(xmlNode *xml, void *user_data) { struct update_data *data = user_data; xmlNode *update = data->update; if (!pcmk__xe_is(update, (const char *) xml->name)) { // No match: either not both elements, or different element types return true; } if (!pcmk__str_eq(pcmk__xe_id(xml), pcmk__xe_id(update), pcmk__str_none)) { // No match: ID mismatch return true; } crm_log_xml_trace(xml, "update-match"); crm_log_xml_trace(update, "update-with"); update_xe(NULL, xml, update, data->flags); // Found a match and replaced it; stop traversing tree return false; } /*! * \internal * \brief Search an XML tree depth-first and update the first matching element * * "Update" means to merge a source subtree into a target subtree (see * \c update_xe()). * * A match with a node \c node is defined as follows: * * \c node and \p update are both element nodes of the same type. * * \c node and \p update have the same \c PCMK_XA_ID attribute value, or * \c PCMK_XA_ID is unset in both * * \param[in,out] xml XML tree to search * \param[in] update XML to update a matching element with * \param[in] flags Group of enum pcmk__xa_flags * * \return Standard Pacemaker return code (specifically, \c pcmk_rc_ok on * successful update and an error code otherwise) */ int pcmk__xe_update_match(xmlNode *xml, xmlNode *update, uint32_t flags) { /* @COMPAT In pcmk__xe_delete_match() and pcmk__xe_replace_match(), we * compare IDs only if the equivalent of the update argument has an ID. * Here, we're stricter: we consider it a mismatch if only one element has * an ID attribute, or if both elements have IDs but they don't match. * * Perhaps we should align the behavior at a major version release. */ struct update_data data = { .update = update, .flags = flags, }; CRM_CHECK((xml != NULL) && (update != NULL), return EINVAL); if (!pcmk__xml_tree_foreach(xml, update_xe_if_matching, &data)) { // Found and updated an element return pcmk_rc_ok; } // No match found in this subtree return ENXIO; } void pcmk__xe_set_propv(xmlNodePtr node, va_list pairs) { while (true) { const char *name, *value; name = va_arg(pairs, const char *); if (name == NULL) { return; } value = va_arg(pairs, const char *); if (value != NULL) { crm_xml_add(node, name, value); } } } void pcmk__xe_set_props(xmlNodePtr node, ...) { va_list pairs; va_start(pairs, node); pcmk__xe_set_propv(node, pairs); va_end(pairs); } int pcmk__xe_foreach_child(xmlNode *xml, const char *child_element_name, int (*handler)(xmlNode *xml, void *userdata), void *userdata) { xmlNode *children = (xml? xml->children : NULL); pcmk__assert(handler != NULL); for (xmlNode *node = children; node != NULL; node = node->next) { if ((node->type == XML_ELEMENT_NODE) && ((child_element_name == NULL) || pcmk__xe_is(node, child_element_name))) { int rc = handler(node, userdata); if (rc != pcmk_rc_ok) { return rc; } } } return pcmk_rc_ok; } // XML attribute handling /*! * \brief Create an XML attribute with specified name and value * * \param[in,out] node XML node to modify * \param[in] name Attribute name to set * \param[in] value Attribute value to set * * \return New value on success, \c NULL otherwise * \note This does nothing if node, name, or value are \c NULL or empty. */ const char * crm_xml_add(xmlNode *node, const char *name, const char *value) { // @TODO Replace with internal function that returns the new attribute bool dirty = FALSE; xmlAttr *attr = NULL; CRM_CHECK(node != NULL, return NULL); CRM_CHECK(name != NULL, return NULL); if (value == NULL) { return NULL; } if (pcmk__xml_doc_all_flags_set(node->doc, pcmk__xf_tracking)) { const char *old = pcmk__xe_get(node, name); if (old == NULL || value == NULL || strcmp(old, value) != 0) { dirty = TRUE; } } if (dirty && (pcmk__check_acl(node, name, pcmk__xf_acl_create) == FALSE)) { crm_trace("Cannot add %s=%s to %s", name, value, node->name); return NULL; } attr = xmlSetProp(node, (const xmlChar *) name, (const xmlChar *) value); /* If the attribute already exists, this does nothing. Attribute values * don't get private data. */ pcmk__xml_new_private_data((xmlNode *) attr); if (dirty) { pcmk__mark_xml_attr_dirty(attr); } CRM_CHECK(attr && attr->children && attr->children->content, return NULL); return (char *)attr->children->content; } /*! * \brief Create an XML attribute with specified name and integer value * * This is like \c crm_xml_add() but taking an integer value. * * \param[in,out] node XML node to modify * \param[in] name Attribute name to set * \param[in] value Attribute value to set * * \return New value as string on success, \c NULL otherwise * \note This does nothing if node or name are \c NULL or empty. */ const char * crm_xml_add_int(xmlNode *node, const char *name, int value) { char *number = pcmk__itoa(value); const char *added = crm_xml_add(node, name, number); free(number); return added; } /*! * \brief Create an XML attribute with specified name and unsigned value * * This is like \c crm_xml_add() but taking a guint value. * * \param[in,out] node XML node to modify * \param[in] name Attribute name to set * \param[in] ms Attribute value to set * * \return New value as string on success, \c NULL otherwise * \note This does nothing if node or name are \c NULL or empty. */ const char * crm_xml_add_ms(xmlNode *node, const char *name, guint ms) { char *number = crm_strdup_printf("%u", ms); const char *added = crm_xml_add(node, name, number); free(number); return added; } // Maximum size of null-terminated string representation of 64-bit integer // -9223372036854775808 #define LLSTRSIZE 21 /*! * \brief Create an XML attribute with specified name and long long int value * * This is like \c crm_xml_add() but taking a long long int value. It is a * useful equivalent for defined types like time_t, etc. * * \param[in,out] xml XML node to modify * \param[in] name Attribute name to set * \param[in] value Attribute value to set * * \return New value as string on success, \c NULL otherwise * \note This does nothing if xml or name are \c NULL or empty. * This does not support greater than 64-bit values. */ const char * crm_xml_add_ll(xmlNode *xml, const char *name, long long value) { char s[LLSTRSIZE] = { '\0', }; if (snprintf(s, LLSTRSIZE, "%lld", (long long) value) == LLSTRSIZE) { return NULL; } return crm_xml_add(xml, name, s); } /*! * \internal * \brief Retrieve the value of an XML attribute * * \param[in] xml XML element whose attribute to get * \param[in] attr_name Attribute name * * \return Value of specified attribute (may be \c NULL) */ const char * pcmk__xe_get(const xmlNode *xml, const char *attr_name) { xmlAttr *attr = NULL; CRM_CHECK((xml != NULL) && (attr_name != NULL), return NULL); attr = xmlHasProp(xml, (const xmlChar *) attr_name); if ((attr == NULL) || (attr->children == NULL)) { return NULL; } return (const char *) attr->children->content; } /*! * \internal * \brief Retrieve a flag group from an XML attribute value * * This is like \c pcmk__xe_get() but returns the value as a \c uint32_t. * * \param[in] xml XML node to check * \param[in] name Attribute name to check (must not be NULL) * \param[out] dest Where to store flags (may be NULL to just * validate type) * \param[in] default_value What to use for missing or invalid value * * \return Standard Pacemaker return code */ int pcmk__xe_get_flags(const xmlNode *xml, const char *name, uint32_t *dest, uint32_t default_value) { const char *value = NULL; long long value_ll = 0LL; int rc = pcmk_rc_ok; if (dest != NULL) { *dest = default_value; } if (name == NULL) { return EINVAL; } if (xml == NULL) { return pcmk_rc_ok; } value = pcmk__xe_get(xml, name); if (value == NULL) { return pcmk_rc_ok; } rc = pcmk__scan_ll(value, &value_ll, default_value); if ((value_ll < 0) || (value_ll > UINT32_MAX)) { value_ll = default_value; if (rc == pcmk_rc_ok) { rc = pcmk_rc_bad_input; } } if (dest != NULL) { *dest = (uint32_t) value_ll; } return rc; } /*! * \internal * \brief Retrieve a \c guint value from an XML attribute * * This is like \c pcmk__xe_get() but returns the value as a \c guint. * * \param[in] xml XML element whose attribute to get * \param[in] attr Attribute name * \param[out] dest Where to store attribute value (unchanged on error) * * \return Standard Pacemaker return code */ int pcmk__xe_get_guint(const xmlNode *xml, const char *attr, guint *dest) { long long value_ll = 0; int rc = pcmk_rc_ok; CRM_CHECK((xml != NULL) && (attr != NULL) && (dest != NULL), return EINVAL); rc = pcmk__xe_get_ll(xml, attr, &value_ll); if (rc != pcmk_rc_ok) { return rc; } if ((value_ll < 0) || (value_ll > G_MAXUINT)) { return ERANGE; } *dest = (guint) value_ll; return pcmk_rc_ok; } +/*! + * \internal + * \brief Set an XML attribute using a \c guint value + * + * This is like \c crm_xml_add() but takes a \c guint. + * + * \param[in,out] xml XML node to modify + * \param[in] attr Attribute name + * \param[in] value Attribute value to set + */ +void +pcmk__xe_set_guint(xmlNode *xml, const char *attr, guint value) +{ + char *value_s = NULL; + + CRM_CHECK((xml != NULL) && (attr != NULL), return); + + value_s = crm_strdup_printf("%u", value); + crm_xml_add(xml, attr, value_s); + free(value_s); +} + /*! * \internal * \brief Retrieve an \c int value from an XML attribute * * This is like \c pcmk__xe_get() but returns the value as an \c int. * * \param[in] xml XML element whose attribute to get * \param[in] attr Attribute name * \param[out] dest Where to store element value (unchanged on error) * * \return Standard Pacemaker return code */ int pcmk__xe_get_int(const xmlNode *xml, const char *attr, int *dest) { long long value_ll = 0; int rc = pcmk_rc_ok; CRM_CHECK((xml != NULL) && (attr != NULL) && (dest != NULL), return EINVAL); rc = pcmk__xe_get_ll(xml, attr, &value_ll); if (rc != pcmk_rc_ok) { return rc; } if ((value_ll < INT_MIN) || (value_ll > INT_MAX)) { return ERANGE; } *dest = (int) value_ll; return pcmk_rc_ok; } /*! * \internal * \brief Retrieve a long long value from an XML attribute * * This is like \c pcmk__xe_get() but returns the value as a long long. * * \param[in] xml XML element whose attribute to get * \param[in] attr Attribute name * \param[out] dest Where to store element value (unchanged on error) * * \return Standard Pacemaker return code */ int pcmk__xe_get_ll(const xmlNode *xml, const char *attr, long long *dest) { const char *value = NULL; long long value_ll = 0; int rc = pcmk_rc_ok; CRM_CHECK((xml != NULL) && (attr != NULL) && (dest != NULL), return EINVAL); value = pcmk__xe_get(xml, attr); if (value == NULL) { return ENXIO; } rc = pcmk__scan_ll(value, &value_ll, PCMK__PARSE_INT_DEFAULT); if (rc != pcmk_rc_ok) { return rc; } *dest = value_ll; return pcmk_rc_ok; } /*! * \internal * \brief Retrieve a \c time_t value from an XML attribute * * This is like \c pcmk__xe_get() but returns the value as a \c time_t. * * \param[in] xml XML element whose attribute to get * \param[in] attr Attribute name * \param[out] dest Where to store attribute value (unchanged on error) * * \return Standard Pacemaker return code */ int pcmk__xe_get_time(const xmlNode *xml, const char *attr, time_t *dest) { long long value_ll = 0; int rc = pcmk_rc_ok; CRM_CHECK((xml != NULL) && (attr != NULL) && (dest != NULL), return EINVAL); rc = pcmk__xe_get_ll(xml, attr, &value_ll); if (rc != pcmk_rc_ok) { return rc; } /* We don't do any bounds checking, since there are no constants provided * for the bounds of time_t, and calculating them isn't worth the effort. If * there are XML values beyond the native sizes, there will likely be worse * problems anyway. */ *dest = (time_t) value_ll; return pcmk_rc_ok; } /*! * \internal * \brief Set an XML attribute using a \c time_t value * * This is like \c crm_xml_add() but takes a \c time_t. * * \param[in,out] xml XML element whose attribute to set * \param[in] attr Attribute name * \param[in] value Attribute value to set (in seconds) */ void pcmk__xe_set_time(xmlNode *xml, const char *attr, time_t value) { // Could be inline, but keep it underneath pcmk__xe_get_time() CRM_CHECK((xml != NULL) && (attr != NULL), return); crm_xml_add_ll(xml, attr, (long long) value); } /*! * \internal * \brief Retrieve the values of XML second/microsecond attributes as time * * This is like \c pcmk__xe_get() but returns the value as a * struct timeval. * * \param[in] xml XML element whose attributes to get * \param[in] sec_attr Name of XML attribute for seconds * \param[in] usec_attr Name of XML attribute for microseconds * \param[out] dest Where to store result (unchanged on error) * * \return Standard Pacemaker return code */ int pcmk__xe_get_timeval(const xmlNode *xml, const char *sec_attr, const char *usec_attr, struct timeval *dest) { long long value_ll = 0; struct timeval result = { 0, 0 }; int rc = pcmk_rc_ok; // Could allow one of sec_attr and usec_attr to be NULL in the future CRM_CHECK((xml != NULL) && (sec_attr != NULL) && (usec_attr != NULL) && (dest != NULL), return EINVAL); // No bounds checking; see comment in pcmk__xe_get_time() // Parse seconds rc = pcmk__xe_get_time(xml, sec_attr, &(result.tv_sec)); if (rc != pcmk_rc_ok) { return rc; } // Parse microseconds rc = pcmk__xe_get_ll(xml, usec_attr, &value_ll); if (rc != pcmk_rc_ok) { return rc; } result.tv_usec = (suseconds_t) value_ll; *dest = result; return pcmk_rc_ok; } /*! * \internal * \brief Set XML attribute values for seconds and microseconds * * This is like \c crm_xml_add() but takes a struct timeval *. * * \param[in,out] xml XML element whose attributes to set * \param[in] sec_attr Name of XML attribute for seconds * \param[in] usec_attr Name of XML attribute for microseconds * \param[in] value Attribute values to set * * \note This does nothing if \p value is \c NULL. */ void pcmk__xe_set_timeval(xmlNode *xml, const char *sec_attr, const char *usec_attr, const struct timeval *value) { CRM_CHECK((xml != NULL) && (sec_attr != NULL) && (usec_attr != NULL), return); if (value == NULL) { return; } if (crm_xml_add_ll(xml, sec_attr, (long long) value->tv_sec) == NULL) { return; } /* Seconds were added successfully. Ignore any errors adding microseconds. * * It would be nice to make this atomic: revert the seconds attribute if * adding the microseconds attribute fails. That's somewhat complicated due * to change tracking: the chain of parents is already marked dirty, etc. In * practice, microseconds should succeed if seconds succeeded, unless it's * due to memory allocation failure. Nothing checks the return values of * these setter functions at time of writing, anyway. */ crm_xml_add_ll(xml, usec_attr, (long long) value->tv_usec); } /*! * \internal * \brief Get a date/time object from an XML attribute value * * \param[in] xml XML with attribute to parse (from CIB) * \param[in] attr Name of attribute to parse * \param[out] t Where to create date/time object * (\p *t must be NULL initially) * * \return Standard Pacemaker return code * \note The caller is responsible for freeing \p *t using crm_time_free(). */ int pcmk__xe_get_datetime(const xmlNode *xml, const char *attr, crm_time_t **t) { const char *value = NULL; if ((t == NULL) || (*t != NULL) || (xml == NULL) || (attr == NULL)) { return EINVAL; } value = pcmk__xe_get(xml, attr); if (value != NULL) { *t = crm_time_new(value); if (*t == NULL) { return pcmk_rc_unpack_error; } } return pcmk_rc_ok; } /*! * \internal * \brief Add a boolean attribute to an XML node. * * \param[in,out] node XML node to add attributes to * \param[in] name XML attribute to create * \param[in] value Value to give to the attribute */ void pcmk__xe_set_bool_attr(xmlNodePtr node, const char *name, bool value) { crm_xml_add(node, name, pcmk__btoa(value)); } /*! * \internal * \brief Extract a boolean attribute's value from an XML element, with * error checking * * \param[in] node XML node to get attribute from * \param[in] name XML attribute to get * \param[out] value Destination for the value of the attribute * * \return EINVAL if \p name or \p value are NULL, ENODATA if \p node is * NULL or the attribute does not exist, pcmk_rc_unknown_format * if the attribute is not a boolean, and pcmk_rc_ok otherwise. * * \note \p value only has any meaning if the return value is pcmk_rc_ok. */ int pcmk__xe_get_bool_attr(const xmlNode *node, const char *name, bool *value) { const char *xml_value = NULL; int ret, rc; if (node == NULL) { return ENODATA; } else if (name == NULL || value == NULL) { return EINVAL; } xml_value = pcmk__xe_get(node, name); if (xml_value == NULL) { return ENODATA; } rc = crm_str_to_boolean(xml_value, &ret); if (rc == 1) { *value = ret; return pcmk_rc_ok; } else { return pcmk_rc_bad_input; } } /*! * \internal * \brief Extract a boolean attribute's value from an XML element * * \param[in] node XML node to get attribute from * \param[in] name XML attribute to get * * \return True if the given \p name is an attribute on \p node and has * the value \c PCMK_VALUE_TRUE, False in all other cases */ bool pcmk__xe_attr_is_true(const xmlNode *node, const char *name) { bool value = false; int rc; rc = pcmk__xe_get_bool_attr(node, name, &value); return rc == pcmk_rc_ok && value == true; } // Deprecated functions kept only for backward API compatibility // LCOV_EXCL_START #include // gboolean, GSList #include // pcmk_xml_attrs2nvpairs(), etc. #include // crm_xml_sanitize_id() #include xmlNode * expand_idref(xmlNode *input, xmlNode *top) { return pcmk__xe_resolve_idref(input, top); } void crm_xml_set_id(xmlNode *xml, const char *format, ...) { va_list ap; int len = 0; char *id = NULL; /* equivalent to crm_strdup_printf() */ va_start(ap, format); len = vasprintf(&id, format, ap); va_end(ap); pcmk__assert(len > 0); crm_xml_sanitize_id(id); crm_xml_add(xml, PCMK_XA_ID, id); free(id); } xmlNode * sorted_xml(xmlNode *input, xmlNode *parent, gboolean recursive) { xmlNode *child = NULL; GSList *nvpairs = NULL; xmlNode *result = NULL; CRM_CHECK(input != NULL, return NULL); result = pcmk__xe_create(parent, (const char *) input->name); nvpairs = pcmk_xml_attrs2nvpairs(input); nvpairs = pcmk_sort_nvpairs(nvpairs); pcmk_nvpairs2xml_attrs(nvpairs, result); pcmk_free_nvpairs(nvpairs); for (child = pcmk__xe_first_child(input, NULL, NULL, NULL); child != NULL; child = pcmk__xe_next(child, NULL)) { if (recursive) { sorted_xml(child, result, recursive); } else { pcmk__xml_copy(result, child); } } return result; } const char * crm_element_value(const xmlNode *data, const char *name) { xmlAttr *attr = NULL; if (data == NULL) { crm_err("Couldn't find %s in NULL", name ? name : ""); CRM_LOG_ASSERT(data != NULL); return NULL; } else if (name == NULL) { crm_err("Couldn't find NULL in %s", data->name); return NULL; } attr = xmlHasProp(data, (const xmlChar *) name); if (!attr || !attr->children) { return NULL; } return (const char *) attr->children->content; } const char * crm_copy_xml_element(const xmlNode *obj1, xmlNode *obj2, const char *element) { const char *value = crm_element_value(obj1, element); crm_xml_add(obj2, element, value); return value; } int crm_element_value_ll(const xmlNode *data, const char *name, long long *dest) { const char *value = NULL; CRM_CHECK(dest != NULL, return -1); value = crm_element_value(data, name); if (value != NULL) { int rc = pcmk__scan_ll(value, dest, PCMK__PARSE_INT_DEFAULT); if (rc == pcmk_rc_ok) { return 0; } crm_warn("Using default for %s " "because '%s' is not a valid integer: %s", name, value, pcmk_rc_str(rc)); } return -1; } int crm_element_value_timeval(const xmlNode *xml, const char *name_sec, const char *name_usec, struct timeval *dest) { long long value_i = 0; CRM_CHECK(dest != NULL, return -EINVAL); dest->tv_sec = 0; dest->tv_usec = 0; if (xml == NULL) { return pcmk_ok; } // No bounds checking; see comment in pcmk__xe_get_time() // Parse seconds errno = 0; if (crm_element_value_ll(xml, name_sec, &value_i) < 0) { return -errno; } dest->tv_sec = (time_t) value_i; // Parse microseconds if (crm_element_value_ll(xml, name_usec, &value_i) < 0) { return -errno; } dest->tv_usec = (suseconds_t) value_i; return pcmk_ok; } int crm_element_value_epoch(const xmlNode *xml, const char *name, time_t *dest) { long long value_ll = 0; if (crm_element_value_ll(xml, name, &value_ll) < 0) { return -1; } // No bounds checking; see comment in pcmk__xe_get_time() *dest = (time_t) value_ll; return pcmk_ok; } int crm_element_value_ms(const xmlNode *data, const char *name, guint *dest) { const char *value = NULL; long long value_ll; int rc = pcmk_rc_ok; CRM_CHECK(dest != NULL, return -1); *dest = 0; value = crm_element_value(data, name); rc = pcmk__scan_ll(value, &value_ll, 0LL); if (rc != pcmk_rc_ok) { crm_warn("Using default for %s " "because '%s' is not valid milliseconds: %s", name, value, pcmk_rc_str(rc)); return -1; } if ((value_ll < 0) || (value_ll > G_MAXUINT)) { crm_warn("Using default for %s because '%s' is out of range", name, value); return -1; } *dest = (guint) value_ll; return pcmk_ok; } int crm_element_value_int(const xmlNode *data, const char *name, int *dest) { const char *value = NULL; CRM_CHECK(dest != NULL, return -1); value = crm_element_value(data, name); if (value) { long long value_ll; int rc = pcmk__scan_ll(value, &value_ll, 0LL); *dest = PCMK__PARSE_INT_DEFAULT; if (rc != pcmk_rc_ok) { crm_warn("Using default for %s " "because '%s' is not a valid integer: %s", name, value, pcmk_rc_str(rc)); } else if ((value_ll < INT_MIN) || (value_ll > INT_MAX)) { crm_warn("Using default for %s because '%s' is out of range", name, value); } else { *dest = (int) value_ll; return 0; } } return -1; } char * crm_element_value_copy(const xmlNode *data, const char *name) { CRM_CHECK((data != NULL) && (name != NULL), return NULL); return pcmk__str_copy(pcmk__xe_get(data, name)); } const char * crm_xml_add_timeval(xmlNode *xml, const char *name_sec, const char *name_usec, const struct timeval *value) { const char *added = NULL; if (xml && name_sec && value) { added = crm_xml_add_ll(xml, name_sec, (long long) value->tv_sec); if (added && name_usec) { // Any error is ignored (we successfully added seconds) crm_xml_add_ll(xml, name_usec, (long long) value->tv_usec); } } return added; } // LCOV_EXCL_STOP // End deprecated API diff --git a/lib/lrmd/lrmd_client.c b/lib/lrmd/lrmd_client.c index 71b9fb0e02..b47e0068ef 100644 --- a/lib/lrmd/lrmd_client.c +++ b/lib/lrmd/lrmd_client.c @@ -1,2672 +1,2672 @@ /* * Copyright 2012-2025 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include // uint32_t, uint64_t #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include // stonith__* #include #include #include #include #include #include #define MAX_TLS_RECV_WAIT 10000 CRM_TRACE_INIT_DATA(lrmd); static int lrmd_api_disconnect(lrmd_t * lrmd); static int lrmd_api_is_connected(lrmd_t * lrmd); /* IPC proxy functions */ int lrmd_internal_proxy_send(lrmd_t * lrmd, xmlNode *msg); static void lrmd_internal_proxy_dispatch(lrmd_t *lrmd, xmlNode *msg); void lrmd_internal_set_proxy_callback(lrmd_t * lrmd, void *userdata, void (*callback)(lrmd_t *lrmd, void *userdata, xmlNode *msg)); // GnuTLS client handshake timeout in seconds #define TLS_HANDSHAKE_TIMEOUT 5 static void lrmd_tls_disconnect(lrmd_t * lrmd); static int global_remote_msg_id = 0; static void lrmd_tls_connection_destroy(gpointer userdata); static int add_tls_to_mainloop(lrmd_t *lrmd, bool do_api_handshake); typedef struct lrmd_private_s { uint64_t type; char *token; mainloop_io_t *source; /* IPC parameters */ crm_ipc_t *ipc; pcmk__remote_t *remote; /* Extra TLS parameters */ char *remote_nodename; char *server; int port; pcmk__tls_t *tls; /* while the async connection is occurring, this is the id * of the connection timeout timer. */ int async_timer; int sock; /* since tls requires a round trip across the network for a * request/reply, there are times where we just want to be able * to send a request from the client and not wait around (or even care * about) what the reply is. */ int expected_late_replies; GList *pending_notify; crm_trigger_t *process_notify; crm_trigger_t *handshake_trigger; lrmd_event_callback callback; /* Internal IPC proxy msg passing for remote guests */ void (*proxy_callback)(lrmd_t *lrmd, void *userdata, xmlNode *msg); void *proxy_callback_userdata; char *peer_version; } lrmd_private_t; static int process_lrmd_handshake_reply(xmlNode *reply, lrmd_private_t *native); static void report_async_connection_result(lrmd_t * lrmd, int rc); static lrmd_list_t * lrmd_list_add(lrmd_list_t * head, const char *value) { lrmd_list_t *p, *end; p = pcmk__assert_alloc(1, sizeof(lrmd_list_t)); p->val = strdup(value); end = head; while (end && end->next) { end = end->next; } if (end) { end->next = p; } else { head = p; } return head; } void lrmd_list_freeall(lrmd_list_t * head) { lrmd_list_t *p; while (head) { char *val = (char *)head->val; p = head->next; free(val); free(head); head = p; } } lrmd_key_value_t * lrmd_key_value_add(lrmd_key_value_t * head, const char *key, const char *value) { lrmd_key_value_t *p, *end; p = pcmk__assert_alloc(1, sizeof(lrmd_key_value_t)); p->key = strdup(key); p->value = strdup(value); end = head; while (end && end->next) { end = end->next; } if (end) { end->next = p; } else { head = p; } return head; } void lrmd_key_value_freeall(lrmd_key_value_t * head) { lrmd_key_value_t *p; while (head) { p = head->next; free(head->key); free(head->value); free(head); head = p; } } /*! * \brief Create a new lrmd_event_data_t object * * \param[in] rsc_id ID of resource involved in event * \param[in] task Action name * \param[in] interval_ms Action interval * * \return Newly allocated and initialized lrmd_event_data_t * \note This functions asserts on memory errors, so the return value is * guaranteed to be non-NULL. The caller is responsible for freeing the * result with lrmd_free_event(). */ lrmd_event_data_t * lrmd_new_event(const char *rsc_id, const char *task, guint interval_ms) { lrmd_event_data_t *event = pcmk__assert_alloc(1, sizeof(lrmd_event_data_t)); // lrmd_event_data_t has (const char *) members that lrmd_free_event() frees event->rsc_id = pcmk__str_copy(rsc_id); event->op_type = pcmk__str_copy(task); event->interval_ms = interval_ms; return event; } lrmd_event_data_t * lrmd_copy_event(lrmd_event_data_t * event) { lrmd_event_data_t *copy = NULL; copy = pcmk__assert_alloc(1, sizeof(lrmd_event_data_t)); copy->type = event->type; // lrmd_event_data_t has (const char *) members that lrmd_free_event() frees copy->rsc_id = pcmk__str_copy(event->rsc_id); copy->op_type = pcmk__str_copy(event->op_type); copy->user_data = pcmk__str_copy(event->user_data); copy->output = pcmk__str_copy(event->output); copy->remote_nodename = pcmk__str_copy(event->remote_nodename); copy->exit_reason = pcmk__str_copy(event->exit_reason); copy->call_id = event->call_id; copy->timeout = event->timeout; copy->interval_ms = event->interval_ms; copy->start_delay = event->start_delay; copy->rsc_deleted = event->rsc_deleted; copy->rc = event->rc; copy->op_status = event->op_status; copy->t_run = event->t_run; copy->t_rcchange = event->t_rcchange; copy->exec_time = event->exec_time; copy->queue_time = event->queue_time; copy->connection_rc = event->connection_rc; copy->params = pcmk__str_table_dup(event->params); return copy; } /*! * \brief Free an executor event * * \param[in,out] Executor event object to free */ void lrmd_free_event(lrmd_event_data_t *event) { if (event == NULL) { return; } // @TODO Why are these const char *? free((void *) event->rsc_id); free((void *) event->op_type); free((void *) event->user_data); free((void *) event->remote_nodename); lrmd__reset_result(event); if (event->params != NULL) { g_hash_table_destroy(event->params); } free(event); } static void lrmd_dispatch_internal(gpointer data, gpointer user_data) { xmlNode *msg = data; lrmd_t *lrmd = user_data; const char *type; const char *proxy_session = pcmk__xe_get(msg, PCMK__XA_LRMD_IPC_SESSION); lrmd_private_t *native = lrmd->lrmd_private; lrmd_event_data_t event = { 0, }; if (proxy_session != NULL) { /* this is proxy business */ lrmd_internal_proxy_dispatch(lrmd, msg); return; } else if (!native->callback) { /* no callback set */ crm_trace("notify event received but client has not set callback"); return; } event.remote_nodename = native->remote_nodename; type = pcmk__xe_get(msg, PCMK__XA_LRMD_OP); pcmk__xe_get_int(msg, PCMK__XA_LRMD_CALLID, &event.call_id); event.rsc_id = pcmk__xe_get(msg, PCMK__XA_LRMD_RSC_ID); if (pcmk__str_eq(type, LRMD_OP_RSC_REG, pcmk__str_none)) { event.type = lrmd_event_register; } else if (pcmk__str_eq(type, LRMD_OP_RSC_UNREG, pcmk__str_none)) { event.type = lrmd_event_unregister; } else if (pcmk__str_eq(type, LRMD_OP_RSC_EXEC, pcmk__str_none)) { int rc = 0; int exec_time = 0; int queue_time = 0; pcmk__xe_get_int(msg, PCMK__XA_LRMD_TIMEOUT, &event.timeout); pcmk__xe_get_guint(msg, PCMK__XA_LRMD_RSC_INTERVAL, &event.interval_ms); pcmk__xe_get_int(msg, PCMK__XA_LRMD_RSC_START_DELAY, &event.start_delay); pcmk__xe_get_int(msg, PCMK__XA_LRMD_EXEC_RC, &rc); event.rc = (enum ocf_exitcode) rc; pcmk__xe_get_int(msg, PCMK__XA_LRMD_EXEC_OP_STATUS, &event.op_status); pcmk__xe_get_int(msg, PCMK__XA_LRMD_RSC_DELETED, &event.rsc_deleted); pcmk__xe_get_time(msg, PCMK__XA_LRMD_RUN_TIME, &event.t_run); pcmk__xe_get_time(msg, PCMK__XA_LRMD_RCCHANGE_TIME, &event.t_rcchange); pcmk__xe_get_int(msg, PCMK__XA_LRMD_EXEC_TIME, &exec_time); CRM_LOG_ASSERT(exec_time >= 0); event.exec_time = QB_MAX(0, exec_time); pcmk__xe_get_int(msg, PCMK__XA_LRMD_QUEUE_TIME, &queue_time); CRM_LOG_ASSERT(queue_time >= 0); event.queue_time = QB_MAX(0, queue_time); event.op_type = pcmk__xe_get(msg, PCMK__XA_LRMD_RSC_ACTION); event.user_data = pcmk__xe_get(msg, PCMK__XA_LRMD_RSC_USERDATA_STR); event.type = lrmd_event_exec_complete; /* output and exit_reason may be freed by a callback */ event.output = pcmk__xe_get_copy(msg, PCMK__XA_LRMD_RSC_OUTPUT); lrmd__set_result(&event, event.rc, event.op_status, pcmk__xe_get(msg, PCMK__XA_LRMD_RSC_EXIT_REASON)); event.params = xml2list(msg); } else if (pcmk__str_eq(type, LRMD_OP_NEW_CLIENT, pcmk__str_none)) { event.type = lrmd_event_new_client; } else if (pcmk__str_eq(type, LRMD_OP_POKE, pcmk__str_none)) { event.type = lrmd_event_poke; } else { return; } crm_trace("op %s notify event received", type); native->callback(&event); if (event.params) { g_hash_table_destroy(event.params); } lrmd__reset_result(&event); } // \return Always 0, to indicate that IPC mainloop source should be kept static int lrmd_ipc_dispatch(const char *buffer, ssize_t length, gpointer userdata) { lrmd_t *lrmd = userdata; lrmd_private_t *native = lrmd->lrmd_private; if (native->callback != NULL) { xmlNode *msg = pcmk__xml_parse(buffer); lrmd_dispatch_internal(msg, lrmd); pcmk__xml_free(msg); } return 0; } static void lrmd_free_xml(gpointer userdata) { pcmk__xml_free((xmlNode *) userdata); } static bool remote_executor_connected(lrmd_t * lrmd) { lrmd_private_t *native = lrmd->lrmd_private; return (native->remote->tls_session != NULL); } static void handle_remote_msg(xmlNode *xml, lrmd_t *lrmd) { lrmd_private_t *native = lrmd->lrmd_private; const char *msg_type = NULL; msg_type = pcmk__xe_get(xml, PCMK__XA_LRMD_REMOTE_MSG_TYPE); if (pcmk__str_eq(msg_type, "notify", pcmk__str_casei)) { lrmd_dispatch_internal(xml, lrmd); } else if (pcmk__str_eq(msg_type, "reply", pcmk__str_casei)) { const char *op = pcmk__xe_get(xml, PCMK__XA_LRMD_OP); if (native->expected_late_replies > 0) { native->expected_late_replies--; /* The register op message we get as a response to lrmd_handshake_async * is a reply, so we have to handle that here. */ if (pcmk__str_eq(op, "register", pcmk__str_casei)) { int rc = process_lrmd_handshake_reply(xml, native); report_async_connection_result(lrmd, pcmk_rc2legacy(rc)); } } else { int reply_id = 0; pcmk__xe_get_int(xml, PCMK__XA_LRMD_CALLID, &reply_id); /* if this happens, we want to know about it */ crm_err("Got outdated Pacemaker Remote reply %d", reply_id); } } } /*! * \internal * \brief Notify trigger handler * * \param[in,out] userdata API connection * * \return Always return G_SOURCE_CONTINUE to leave this trigger handler in the * mainloop */ static int process_pending_notifies(gpointer userdata) { lrmd_t *lrmd = userdata; lrmd_private_t *native = lrmd->lrmd_private; if (native->pending_notify == NULL) { return G_SOURCE_CONTINUE; } crm_trace("Processing pending notifies"); g_list_foreach(native->pending_notify, lrmd_dispatch_internal, lrmd); g_list_free_full(native->pending_notify, lrmd_free_xml); native->pending_notify = NULL; return G_SOURCE_CONTINUE; } /*! * \internal * \brief TLS dispatch function for file descriptor sources * * \param[in,out] userdata API connection * * \return -1 on error to remove the source from the mainloop, or 0 otherwise * to leave it in the mainloop */ static int lrmd_tls_dispatch(gpointer userdata) { lrmd_t *lrmd = userdata; lrmd_private_t *native = lrmd->lrmd_private; xmlNode *xml = NULL; int rc = pcmk_rc_ok; if (!remote_executor_connected(lrmd)) { crm_trace("TLS dispatch triggered after disconnect"); return -1; } crm_trace("TLS dispatch triggered"); rc = pcmk__remote_ready(native->remote, 0); if (rc == pcmk_rc_ok) { rc = pcmk__read_remote_message(native->remote, -1); } if (rc != pcmk_rc_ok && rc != ETIME) { crm_info("Lost %s executor connection while reading data", (native->remote_nodename? native->remote_nodename : "local")); lrmd_tls_disconnect(lrmd); return -1; } /* If rc is ETIME, there was nothing to read but we may already have a * full message in the buffer */ xml = pcmk__remote_message_xml(native->remote); if (xml == NULL) { return 0; } handle_remote_msg(xml, lrmd); pcmk__xml_free(xml); return 0; } /* Not used with mainloop */ int lrmd_poll(lrmd_t * lrmd, int timeout) { lrmd_private_t *native = lrmd->lrmd_private; switch (native->type) { case pcmk__client_ipc: return crm_ipc_ready(native->ipc); case pcmk__client_tls: if (native->pending_notify) { return 1; } else { int rc = pcmk__remote_ready(native->remote, 0); switch (rc) { case pcmk_rc_ok: return 1; case ETIME: return 0; default: return pcmk_rc2legacy(rc); } } default: crm_err("Unsupported executor connection type (bug?): %d", native->type); return -EPROTONOSUPPORT; } } /* Not used with mainloop */ bool lrmd_dispatch(lrmd_t * lrmd) { lrmd_private_t *private = NULL; pcmk__assert(lrmd != NULL); private = lrmd->lrmd_private; switch (private->type) { case pcmk__client_ipc: while (crm_ipc_ready(private->ipc)) { if (crm_ipc_read(private->ipc) > 0) { const char *msg = crm_ipc_buffer(private->ipc); lrmd_ipc_dispatch(msg, strlen(msg), lrmd); } } break; case pcmk__client_tls: lrmd_tls_dispatch(lrmd); break; default: crm_err("Unsupported executor connection type (bug?): %d", private->type); } if (lrmd_api_is_connected(lrmd) == FALSE) { crm_err("Connection closed"); return FALSE; } return TRUE; } static xmlNode * lrmd_create_op(const char *token, const char *op, xmlNode *data, int timeout, enum lrmd_call_options options) { xmlNode *op_msg = NULL; CRM_CHECK(token != NULL, return NULL); op_msg = pcmk__xe_create(NULL, PCMK__XE_LRMD_COMMAND); crm_xml_add(op_msg, PCMK__XA_T, PCMK__VALUE_LRMD); crm_xml_add(op_msg, PCMK__XA_LRMD_OP, op); crm_xml_add_int(op_msg, PCMK__XA_LRMD_TIMEOUT, timeout); crm_xml_add_int(op_msg, PCMK__XA_LRMD_CALLOPT, options); if (data != NULL) { xmlNode *wrapper = pcmk__xe_create(op_msg, PCMK__XE_LRMD_CALLDATA); pcmk__xml_copy(wrapper, data); } crm_trace("Created executor %s command with call options %.8lx (%d)", op, (long)options, options); return op_msg; } static void lrmd_ipc_connection_destroy(gpointer userdata) { lrmd_t *lrmd = userdata; lrmd_private_t *native = lrmd->lrmd_private; switch (native->type) { case pcmk__client_ipc: crm_info("Disconnected from local executor"); break; case pcmk__client_tls: crm_info("Disconnected from remote executor on %s", native->remote_nodename); break; default: crm_err("Unsupported executor connection type %d (bug?)", native->type); } /* Prevent these from being cleaned up in lrmd_api_disconnect() */ native->ipc = NULL; native->source = NULL; if (native->callback) { lrmd_event_data_t event = { 0, }; event.type = lrmd_event_disconnect; event.remote_nodename = native->remote_nodename; native->callback(&event); } } static void lrmd_tls_connection_destroy(gpointer userdata) { lrmd_t *lrmd = userdata; lrmd_private_t *native = lrmd->lrmd_private; crm_info("TLS connection destroyed"); if (native->remote->tls_session) { gnutls_bye(native->remote->tls_session, GNUTLS_SHUT_RDWR); gnutls_deinit(native->remote->tls_session); native->remote->tls_session = NULL; } if (native->tls) { pcmk__free_tls(native->tls); native->tls = NULL; } if (native->sock >= 0) { close(native->sock); } if (native->process_notify) { mainloop_destroy_trigger(native->process_notify); native->process_notify = NULL; } if (native->pending_notify) { g_list_free_full(native->pending_notify, lrmd_free_xml); native->pending_notify = NULL; } if (native->handshake_trigger != NULL) { mainloop_destroy_trigger(native->handshake_trigger); native->handshake_trigger = NULL; } free(native->remote->buffer); free(native->remote->start_state); native->remote->buffer = NULL; native->remote->start_state = NULL; native->source = 0; native->sock = -1; if (native->callback) { lrmd_event_data_t event = { 0, }; event.remote_nodename = native->remote_nodename; event.type = lrmd_event_disconnect; native->callback(&event); } return; } // \return Standard Pacemaker return code int lrmd__remote_send_xml(pcmk__remote_t *session, xmlNode *msg, uint32_t id, const char *msg_type) { crm_xml_add_int(msg, PCMK__XA_LRMD_REMOTE_MSG_ID, id); crm_xml_add(msg, PCMK__XA_LRMD_REMOTE_MSG_TYPE, msg_type); return pcmk__remote_send_xml(session, msg); } // \return Standard Pacemaker return code static int read_remote_reply(lrmd_t *lrmd, int total_timeout, int expected_reply_id, xmlNode **reply) { lrmd_private_t *native = lrmd->lrmd_private; time_t start = time(NULL); const char *msg_type = NULL; int reply_id = 0; int remaining_timeout = 0; int rc = pcmk_rc_ok; /* A timeout of 0 here makes no sense. We have to wait a period of time * for the response to come back. If -1 or 0, default to 10 seconds. */ if (total_timeout <= 0 || total_timeout > MAX_TLS_RECV_WAIT) { total_timeout = MAX_TLS_RECV_WAIT; } for (*reply = NULL; *reply == NULL; ) { *reply = pcmk__remote_message_xml(native->remote); if (*reply == NULL) { /* read some more off the tls buffer if we still have time left. */ if (remaining_timeout) { remaining_timeout = total_timeout - ((time(NULL) - start) * 1000); } else { remaining_timeout = total_timeout; } if (remaining_timeout <= 0) { return ETIME; } rc = pcmk__read_remote_message(native->remote, remaining_timeout); if (rc != pcmk_rc_ok) { return rc; } *reply = pcmk__remote_message_xml(native->remote); if (*reply == NULL) { return ENOMSG; } } pcmk__xe_get_int(*reply, PCMK__XA_LRMD_REMOTE_MSG_ID, &reply_id); msg_type = pcmk__xe_get(*reply, PCMK__XA_LRMD_REMOTE_MSG_TYPE); if (!msg_type) { crm_err("Empty msg type received while waiting for reply"); pcmk__xml_free(*reply); *reply = NULL; } else if (pcmk__str_eq(msg_type, "notify", pcmk__str_casei)) { /* got a notify while waiting for reply, trigger the notify to be processed later */ crm_info("queueing notify"); native->pending_notify = g_list_append(native->pending_notify, *reply); if (native->process_notify) { crm_info("notify trigger set."); mainloop_set_trigger(native->process_notify); } *reply = NULL; } else if (!pcmk__str_eq(msg_type, "reply", pcmk__str_casei)) { /* msg isn't a reply, make some noise */ crm_err("Expected a reply, got %s", msg_type); pcmk__xml_free(*reply); *reply = NULL; } else if (reply_id != expected_reply_id) { if (native->expected_late_replies > 0) { native->expected_late_replies--; } else { crm_err("Got outdated reply, expected id %d got id %d", expected_reply_id, reply_id); } pcmk__xml_free(*reply); *reply = NULL; } } if (native->remote->buffer && native->process_notify) { mainloop_set_trigger(native->process_notify); } return rc; } // \return Standard Pacemaker return code static int send_remote_message(lrmd_t *lrmd, xmlNode *msg) { int rc = pcmk_rc_ok; lrmd_private_t *native = lrmd->lrmd_private; global_remote_msg_id++; if (global_remote_msg_id <= 0) { global_remote_msg_id = 1; } rc = lrmd__remote_send_xml(native->remote, msg, global_remote_msg_id, "request"); if (rc != pcmk_rc_ok) { crm_err("Disconnecting because TLS message could not be sent to " "Pacemaker Remote: %s", pcmk_rc_str(rc)); lrmd_tls_disconnect(lrmd); } return rc; } static int lrmd_tls_send_recv(lrmd_t * lrmd, xmlNode * msg, int timeout, xmlNode ** reply) { int rc = 0; xmlNode *xml = NULL; if (!remote_executor_connected(lrmd)) { return -ENOTCONN; } rc = send_remote_message(lrmd, msg); if (rc != pcmk_rc_ok) { return pcmk_rc2legacy(rc); } rc = read_remote_reply(lrmd, timeout, global_remote_msg_id, &xml); if (rc != pcmk_rc_ok) { crm_err("Disconnecting remote after request %d reply not received: %s " QB_XS " rc=%d timeout=%dms", global_remote_msg_id, pcmk_rc_str(rc), rc, timeout); lrmd_tls_disconnect(lrmd); } if (reply) { *reply = xml; } else { pcmk__xml_free(xml); } return pcmk_rc2legacy(rc); } static int lrmd_send_xml(lrmd_t * lrmd, xmlNode * msg, int timeout, xmlNode ** reply) { int rc = pcmk_ok; lrmd_private_t *native = lrmd->lrmd_private; switch (native->type) { case pcmk__client_ipc: rc = crm_ipc_send(native->ipc, msg, crm_ipc_client_response, timeout, reply); break; case pcmk__client_tls: rc = lrmd_tls_send_recv(lrmd, msg, timeout, reply); break; default: crm_err("Unsupported executor connection type (bug?): %d", native->type); rc = -EPROTONOSUPPORT; } return rc; } static int lrmd_send_xml_no_reply(lrmd_t * lrmd, xmlNode * msg) { int rc = pcmk_ok; lrmd_private_t *native = lrmd->lrmd_private; switch (native->type) { case pcmk__client_ipc: rc = crm_ipc_send(native->ipc, msg, crm_ipc_flags_none, 0, NULL); break; case pcmk__client_tls: rc = send_remote_message(lrmd, msg); if (rc == pcmk_rc_ok) { /* we don't want to wait around for the reply, but * since the request/reply protocol needs to behave the same * as libqb, a reply will eventually come later anyway. */ native->expected_late_replies++; } rc = pcmk_rc2legacy(rc); break; default: crm_err("Unsupported executor connection type (bug?): %d", native->type); rc = -EPROTONOSUPPORT; } return rc; } static int lrmd_api_is_connected(lrmd_t * lrmd) { lrmd_private_t *native = lrmd->lrmd_private; switch (native->type) { case pcmk__client_ipc: return crm_ipc_connected(native->ipc); case pcmk__client_tls: return remote_executor_connected(lrmd); default: crm_err("Unsupported executor connection type (bug?): %d", native->type); return 0; } } /*! * \internal * \brief Send a prepared API command to the executor * * \param[in,out] lrmd Existing connection to the executor * \param[in] op Name of API command to send * \param[in] data Command data XML to add to the sent command * \param[out] output_data If expecting a reply, it will be stored here * \param[in] timeout Timeout in milliseconds (if 0, defaults to * a sensible value per the type of connection, * standard vs. pacemaker remote); * also propagated to the command XML * \param[in] call_options Call options to pass to server when sending * \param[in] expect_reply If true, wait for a reply from the server; * must be true for IPC (as opposed to TLS) clients * * \return pcmk_ok on success, -errno on error */ static int lrmd_send_command(lrmd_t *lrmd, const char *op, xmlNode *data, xmlNode **output_data, int timeout, enum lrmd_call_options options, bool expect_reply) { int rc = pcmk_ok; lrmd_private_t *native = lrmd->lrmd_private; xmlNode *op_msg = NULL; xmlNode *op_reply = NULL; if (!lrmd_api_is_connected(lrmd)) { return -ENOTCONN; } if (op == NULL) { crm_err("No operation specified"); return -EINVAL; } CRM_LOG_ASSERT(native->token != NULL); crm_trace("Sending %s op to executor", op); op_msg = lrmd_create_op(native->token, op, data, timeout, options); if (op_msg == NULL) { return -EINVAL; } if (expect_reply) { rc = lrmd_send_xml(lrmd, op_msg, timeout, &op_reply); } else { rc = lrmd_send_xml_no_reply(lrmd, op_msg); goto done; } if (rc < 0) { crm_perror(LOG_ERR, "Couldn't perform %s operation (timeout=%d): %d", op, timeout, rc); goto done; } else if (op_reply == NULL) { rc = -ENOMSG; goto done; } rc = pcmk_ok; crm_trace("%s op reply received", op); if (pcmk__xe_get_int(op_reply, PCMK__XA_LRMD_RC, &rc) != pcmk_rc_ok) { rc = -ENOMSG; goto done; } crm_log_xml_trace(op_reply, "Reply"); if (output_data) { *output_data = op_reply; op_reply = NULL; /* Prevent subsequent free */ } done: if (lrmd_api_is_connected(lrmd) == FALSE) { crm_err("Executor disconnected"); } pcmk__xml_free(op_msg); pcmk__xml_free(op_reply); return rc; } static int lrmd_api_poke_connection(lrmd_t * lrmd) { int rc; lrmd_private_t *native = lrmd->lrmd_private; xmlNode *data = pcmk__xe_create(NULL, PCMK__XE_LRMD_RSC); crm_xml_add(data, PCMK__XA_LRMD_ORIGIN, __func__); rc = lrmd_send_command(lrmd, LRMD_OP_POKE, data, NULL, 0, 0, (native->type == pcmk__client_ipc)); pcmk__xml_free(data); return rc < 0 ? rc : pcmk_ok; } // \return Standard Pacemaker return code int lrmd__validate_remote_settings(lrmd_t *lrmd, GHashTable *hash) { int rc = pcmk_rc_ok; const char *value; lrmd_private_t *native = lrmd->lrmd_private; xmlNode *data = pcmk__xe_create(NULL, PCMK__XA_LRMD_OP); crm_xml_add(data, PCMK__XA_LRMD_ORIGIN, __func__); value = g_hash_table_lookup(hash, PCMK_OPT_STONITH_WATCHDOG_TIMEOUT); if ((value) && (stonith__watchdog_fencing_enabled_for_node(native->remote_nodename))) { crm_xml_add(data, PCMK__XA_LRMD_WATCHDOG, value); } rc = lrmd_send_command(lrmd, LRMD_OP_CHECK, data, NULL, 0, 0, (native->type == pcmk__client_ipc)); pcmk__xml_free(data); return (rc < 0)? pcmk_legacy2rc(rc) : pcmk_rc_ok; } static xmlNode * lrmd_handshake_hello_msg(const char *name, bool is_proxy) { xmlNode *hello = pcmk__xe_create(NULL, PCMK__XE_LRMD_COMMAND); crm_xml_add(hello, PCMK__XA_T, PCMK__VALUE_LRMD); crm_xml_add(hello, PCMK__XA_LRMD_OP, CRM_OP_REGISTER); crm_xml_add(hello, PCMK__XA_LRMD_CLIENTNAME, name); crm_xml_add(hello, PCMK__XA_LRMD_PROTOCOL_VERSION, LRMD_PROTOCOL_VERSION); /* advertise that we are a proxy provider */ if (is_proxy) { pcmk__xe_set_bool_attr(hello, PCMK__XA_LRMD_IS_IPC_PROVIDER, true); } return hello; } static int process_lrmd_handshake_reply(xmlNode *reply, lrmd_private_t *native) { int rc = pcmk_rc_ok; const char *version = pcmk__xe_get(reply, PCMK__XA_LRMD_PROTOCOL_VERSION); const char *msg_type = pcmk__xe_get(reply, PCMK__XA_LRMD_OP); const char *tmp_ticket = pcmk__xe_get(reply, PCMK__XA_LRMD_CLIENTID); const char *start_state = pcmk__xe_get(reply, PCMK__XA_NODE_START_STATE); pcmk__xe_get_int(reply, PCMK__XA_LRMD_RC, &rc); rc = pcmk_legacy2rc(rc); /* The remote executor may add its uptime to the XML reply, which is useful * in handling transient attributes when the connection to the remote node * unexpectedly drops. If no parameter is given, just default to -1. */ native->remote->uptime = -1; pcmk__xe_get_time(reply, PCMK__XA_UPTIME, &native->remote->uptime); if (start_state) { native->remote->start_state = strdup(start_state); } if (rc == EPROTO) { crm_err("Executor protocol version mismatch between client (%s) and server (%s)", LRMD_PROTOCOL_VERSION, version); crm_log_xml_err(reply, "Protocol Error"); } else if (!pcmk__str_eq(msg_type, CRM_OP_REGISTER, pcmk__str_casei)) { crm_err("Invalid registration message: %s", msg_type); crm_log_xml_err(reply, "Bad reply"); rc = EPROTO; } else if (tmp_ticket == NULL) { crm_err("No registration token provided"); crm_log_xml_err(reply, "Bad reply"); rc = EPROTO; } else { crm_trace("Obtained registration token: %s", tmp_ticket); native->token = strdup(tmp_ticket); native->peer_version = strdup(version?version:"1.0"); /* Included since 1.1 */ rc = pcmk_rc_ok; } return rc; } static int lrmd_handshake(lrmd_t * lrmd, const char *name) { int rc = pcmk_rc_ok; lrmd_private_t *native = lrmd->lrmd_private; xmlNode *reply = NULL; xmlNode *hello = lrmd_handshake_hello_msg(name, native->proxy_callback != NULL); rc = lrmd_send_xml(lrmd, hello, -1, &reply); if (rc < 0) { crm_perror(LOG_DEBUG, "Couldn't complete registration with the executor API: %d", rc); rc = ECOMM; } else if (reply == NULL) { crm_err("Did not receive registration reply"); rc = EPROTO; } else { rc = process_lrmd_handshake_reply(reply, native); } pcmk__xml_free(reply); pcmk__xml_free(hello); if (rc != pcmk_rc_ok) { lrmd_api_disconnect(lrmd); } return rc; } static int lrmd_handshake_async(lrmd_t * lrmd, const char *name) { int rc = pcmk_rc_ok; lrmd_private_t *native = lrmd->lrmd_private; xmlNode *hello = lrmd_handshake_hello_msg(name, native->proxy_callback != NULL); rc = send_remote_message(lrmd, hello); if (rc == pcmk_rc_ok) { native->expected_late_replies++; } else { lrmd_api_disconnect(lrmd); } pcmk__xml_free(hello); return rc; } static int lrmd_ipc_connect(lrmd_t * lrmd, int *fd) { int rc = pcmk_ok; lrmd_private_t *native = lrmd->lrmd_private; struct ipc_client_callbacks lrmd_callbacks = { .dispatch = lrmd_ipc_dispatch, .destroy = lrmd_ipc_connection_destroy }; crm_info("Connecting to executor"); if (fd) { /* No mainloop */ native->ipc = crm_ipc_new(CRM_SYSTEM_LRMD, 0); if (native->ipc != NULL) { rc = pcmk__connect_generic_ipc(native->ipc); if (rc == pcmk_rc_ok) { rc = pcmk__ipc_fd(native->ipc, fd); } if (rc != pcmk_rc_ok) { crm_err("Connection to executor failed: %s", pcmk_rc_str(rc)); rc = -ENOTCONN; } } } else { native->source = mainloop_add_ipc_client(CRM_SYSTEM_LRMD, G_PRIORITY_HIGH, 0, lrmd, &lrmd_callbacks); native->ipc = mainloop_get_ipc_client(native->source); } if (native->ipc == NULL) { crm_debug("Could not connect to the executor API"); rc = -ENOTCONN; } return rc; } static void copy_gnutls_datum(gnutls_datum_t *dest, gnutls_datum_t *source) { pcmk__assert((dest != NULL) && (source != NULL) && (source->data != NULL)); dest->data = gnutls_malloc(source->size); pcmk__mem_assert(dest->data); memcpy(dest->data, source->data, source->size); dest->size = source->size; } static void clear_gnutls_datum(gnutls_datum_t *datum) { gnutls_free(datum->data); datum->data = NULL; datum->size = 0; } #define KEY_READ_LEN 256 // Chunk size for reading key from file // \return Standard Pacemaker return code static int read_gnutls_key(const char *location, gnutls_datum_t *key) { FILE *stream = NULL; size_t buf_len = KEY_READ_LEN; if ((location == NULL) || (key == NULL)) { return EINVAL; } stream = fopen(location, "r"); if (stream == NULL) { return errno; } key->data = gnutls_malloc(buf_len); key->size = 0; while (!feof(stream)) { int next = fgetc(stream); if (next == EOF) { if (!feof(stream)) { crm_warn("Pacemaker Remote key read was partially successful " "(copy in memory may be corrupted)"); } break; } if (key->size == buf_len) { buf_len = key->size + KEY_READ_LEN; key->data = gnutls_realloc(key->data, buf_len); pcmk__assert(key->data); } key->data[key->size++] = (unsigned char) next; } fclose(stream); if (key->size == 0) { clear_gnutls_datum(key); return ENOKEY; } return pcmk_rc_ok; } // Cache the most recently used Pacemaker Remote authentication key struct key_cache_s { time_t updated; // When cached key was read (valid for 1 minute) const char *location; // Where cached key was read from gnutls_datum_t key; // Cached key }; static bool key_is_cached(struct key_cache_s *key_cache) { return key_cache->updated != 0; } static bool key_cache_expired(struct key_cache_s *key_cache) { return (time(NULL) - key_cache->updated) >= 60; } static void clear_key_cache(struct key_cache_s *key_cache) { clear_gnutls_datum(&(key_cache->key)); if ((key_cache->updated != 0) || (key_cache->location != NULL)) { key_cache->updated = 0; key_cache->location = NULL; crm_debug("Cleared Pacemaker Remote key cache"); } } static void get_cached_key(struct key_cache_s *key_cache, gnutls_datum_t *key) { copy_gnutls_datum(key, &(key_cache->key)); crm_debug("Using cached Pacemaker Remote key from %s", pcmk__s(key_cache->location, "unknown location")); } static void cache_key(struct key_cache_s *key_cache, gnutls_datum_t *key, const char *location) { key_cache->updated = time(NULL); key_cache->location = location; copy_gnutls_datum(&(key_cache->key), key); crm_debug("Using (and cacheing) Pacemaker Remote key from %s", pcmk__s(location, "unknown location")); } /*! * \internal * \brief Get Pacemaker Remote authentication key from file or cache * * \param[in] location Path to key file to try (this memory must * persist across all calls of this function) * \param[out] key Key from location or cache * * \return Standard Pacemaker return code */ static int get_remote_key(const char *location, gnutls_datum_t *key) { static struct key_cache_s key_cache = { 0, }; int rc = pcmk_rc_ok; if ((location == NULL) || (key == NULL)) { return EINVAL; } if (key_is_cached(&key_cache)) { if (key_cache_expired(&key_cache)) { clear_key_cache(&key_cache); } else { get_cached_key(&key_cache, key); return pcmk_rc_ok; } } rc = read_gnutls_key(location, key); if (rc != pcmk_rc_ok) { return rc; } cache_key(&key_cache, key, location); return pcmk_rc_ok; } /*! * \internal * \brief Initialize the Pacemaker Remote authentication key * * Try loading the Pacemaker Remote authentication key from cache if available, * otherwise from these locations, in order of preference: * * - The value of the PCMK_authkey_location environment variable, if set * - The Pacemaker default key file location * * \param[out] key Where to store key * * \return Standard Pacemaker return code */ int lrmd__init_remote_key(gnutls_datum_t *key) { static const char *env_location = NULL; static bool need_env = true; int rc = pcmk_rc_ok; if (need_env) { env_location = pcmk__env_option(PCMK__ENV_AUTHKEY_LOCATION); need_env = false; } // Try location in environment variable, if set if (env_location != NULL) { rc = get_remote_key(env_location, key); if (rc == pcmk_rc_ok) { return pcmk_rc_ok; } crm_warn("Could not read Pacemaker Remote key from %s: %s", env_location, pcmk_rc_str(rc)); return ENOKEY; } // Try default location, if environment wasn't explicitly set to it rc = get_remote_key(DEFAULT_REMOTE_KEY_LOCATION, key); if (rc == pcmk_rc_ok) { return pcmk_rc_ok; } crm_warn("Could not read Pacemaker Remote key from default location %s: %s", DEFAULT_REMOTE_KEY_LOCATION, pcmk_rc_str(rc)); return ENOKEY; } static void report_async_connection_result(lrmd_t * lrmd, int rc) { lrmd_private_t *native = lrmd->lrmd_private; if (native->callback) { lrmd_event_data_t event = { 0, }; event.type = lrmd_event_connect; event.remote_nodename = native->remote_nodename; event.connection_rc = rc; native->callback(&event); } } static void tls_handshake_failed(lrmd_t *lrmd, int tls_rc, int rc) { lrmd_private_t *native = lrmd->lrmd_private; crm_warn("Disconnecting after TLS handshake with " "Pacemaker Remote server %s:%d failed: %s", native->server, native->port, (rc == EPROTO)? gnutls_strerror(tls_rc) : pcmk_rc_str(rc)); report_async_connection_result(lrmd, pcmk_rc2legacy(rc)); gnutls_deinit(native->remote->tls_session); native->remote->tls_session = NULL; lrmd_tls_connection_destroy(lrmd); } static void tls_handshake_succeeded(lrmd_t *lrmd) { int rc = pcmk_rc_ok; lrmd_private_t *native = lrmd->lrmd_private; /* Now that the handshake is done, see if any client TLS certificate is * close to its expiration date and log if so. If a TLS certificate is not * in use, this function will just return so we don't need to check for the * session type here. */ pcmk__tls_check_cert_expiration(native->remote->tls_session); crm_info("TLS connection to Pacemaker Remote server %s:%d succeeded", native->server, native->port); rc = add_tls_to_mainloop(lrmd, true); /* If add_tls_to_mainloop failed, report that right now. Otherwise, we have * to wait until we read the async reply to report anything. */ if (rc != pcmk_rc_ok) { report_async_connection_result(lrmd, pcmk_rc2legacy(rc)); } } /*! * \internal * \brief Perform a TLS client handshake with a Pacemaker Remote server * * \param[in] lrmd Newly established Pacemaker Remote executor connection * * \return Standard Pacemaker return code */ static int tls_client_handshake(lrmd_t *lrmd) { lrmd_private_t *native = lrmd->lrmd_private; int tls_rc = GNUTLS_E_SUCCESS; int rc = pcmk__tls_client_handshake(native->remote, TLS_HANDSHAKE_TIMEOUT, &tls_rc); if (rc != pcmk_rc_ok) { tls_handshake_failed(lrmd, tls_rc, rc); } return rc; } /*! * \internal * \brief Add trigger and file descriptor mainloop sources for TLS * * \param[in,out] lrmd API connection with established TLS session * \param[in] do_api_handshake Whether to perform executor handshake * * \return Standard Pacemaker return code */ static int add_tls_to_mainloop(lrmd_t *lrmd, bool do_api_handshake) { lrmd_private_t *native = lrmd->lrmd_private; int rc = pcmk_rc_ok; char *name = crm_strdup_printf("pacemaker-remote-%s:%d", native->server, native->port); struct mainloop_fd_callbacks tls_fd_callbacks = { .dispatch = lrmd_tls_dispatch, .destroy = lrmd_tls_connection_destroy, }; native->process_notify = mainloop_add_trigger(G_PRIORITY_HIGH, process_pending_notifies, lrmd); native->source = mainloop_add_fd(name, G_PRIORITY_HIGH, native->sock, lrmd, &tls_fd_callbacks); /* Async connections lose the client name provided by the API caller, so we * have to use our generated name here to perform the executor handshake. * * @TODO Keep track of the caller-provided name. Perhaps we should be using * that name in this function instead of generating one anyway. */ if (do_api_handshake) { rc = lrmd_handshake_async(lrmd, name); } free(name); return rc; } struct handshake_data_s { lrmd_t *lrmd; time_t start_time; int timeout_sec; }; static gboolean try_handshake_cb(gpointer user_data) { struct handshake_data_s *hs = user_data; lrmd_t *lrmd = hs->lrmd; lrmd_private_t *native = lrmd->lrmd_private; pcmk__remote_t *remote = native->remote; int rc = pcmk_rc_ok; int tls_rc = GNUTLS_E_SUCCESS; if (time(NULL) >= hs->start_time + hs->timeout_sec) { rc = ETIME; tls_handshake_failed(lrmd, GNUTLS_E_TIMEDOUT, rc); free(hs); return 0; } rc = pcmk__tls_client_try_handshake(remote, &tls_rc); if (rc == pcmk_rc_ok) { tls_handshake_succeeded(lrmd); free(hs); return 0; } else if (rc == EAGAIN) { mainloop_set_trigger(native->handshake_trigger); return 1; } else { rc = EKEYREJECTED; tls_handshake_failed(lrmd, tls_rc, rc); free(hs); return 0; } } static void lrmd_tcp_connect_cb(void *userdata, int rc, int sock) { lrmd_t *lrmd = userdata; lrmd_private_t *native = lrmd->lrmd_private; int tls_rc = GNUTLS_E_SUCCESS; bool use_cert = pcmk__x509_enabled(); native->async_timer = 0; if (rc != pcmk_rc_ok) { lrmd_tls_connection_destroy(lrmd); crm_info("Could not connect to Pacemaker Remote at %s:%d: %s " QB_XS " rc=%d", native->server, native->port, pcmk_rc_str(rc), rc); report_async_connection_result(lrmd, pcmk_rc2legacy(rc)); return; } /* The TCP connection was successful, so establish the TLS connection. */ native->sock = sock; if (native->tls == NULL) { rc = pcmk__init_tls(&native->tls, false, use_cert ? GNUTLS_CRD_CERTIFICATE : GNUTLS_CRD_PSK); if (rc != pcmk_rc_ok) { lrmd_tls_connection_destroy(lrmd); report_async_connection_result(lrmd, pcmk_rc2legacy(rc)); return; } } if (!use_cert) { gnutls_datum_t psk_key = { NULL, 0 }; rc = lrmd__init_remote_key(&psk_key); if (rc != pcmk_rc_ok) { crm_info("Could not connect to Pacemaker Remote at %s:%d: %s " QB_XS " rc=%d", native->server, native->port, pcmk_rc_str(rc), rc); lrmd_tls_connection_destroy(lrmd); report_async_connection_result(lrmd, pcmk_rc2legacy(rc)); return; } pcmk__tls_add_psk_key(native->tls, &psk_key); gnutls_free(psk_key.data); } native->remote->tls_session = pcmk__new_tls_session(native->tls, sock); if (native->remote->tls_session == NULL) { lrmd_tls_connection_destroy(lrmd); report_async_connection_result(lrmd, -EPROTO); return; } /* If the TLS handshake immediately succeeds or fails, we can handle that * now without having to deal with mainloops and retries. Otherwise, add a * trigger to keep trying until we get a result (or it times out). */ rc = pcmk__tls_client_try_handshake(native->remote, &tls_rc); if (rc == EAGAIN) { struct handshake_data_s *hs = NULL; if (native->handshake_trigger != NULL) { return; } hs = pcmk__assert_alloc(1, sizeof(struct handshake_data_s)); hs->lrmd = lrmd; hs->start_time = time(NULL); hs->timeout_sec = TLS_HANDSHAKE_TIMEOUT; native->handshake_trigger = mainloop_add_trigger(G_PRIORITY_LOW, try_handshake_cb, hs); mainloop_set_trigger(native->handshake_trigger); } else if (rc == pcmk_rc_ok) { tls_handshake_succeeded(lrmd); } else { tls_handshake_failed(lrmd, tls_rc, rc); } } static int lrmd_tls_connect_async(lrmd_t * lrmd, int timeout /*ms */ ) { int rc = pcmk_rc_ok; int timer_id = 0; lrmd_private_t *native = lrmd->lrmd_private; native->sock = -1; rc = pcmk__connect_remote(native->server, native->port, timeout, &timer_id, &(native->sock), lrmd, lrmd_tcp_connect_cb); if (rc != pcmk_rc_ok) { crm_warn("Pacemaker Remote connection to %s:%d failed: %s " QB_XS " rc=%d", native->server, native->port, pcmk_rc_str(rc), rc); return rc; } native->async_timer = timer_id; return rc; } static int lrmd_tls_connect(lrmd_t * lrmd, int *fd) { int rc = pcmk_rc_ok; bool use_cert = pcmk__x509_enabled(); lrmd_private_t *native = lrmd->lrmd_private; native->sock = -1; rc = pcmk__connect_remote(native->server, native->port, 0, NULL, &(native->sock), NULL, NULL); if (rc != pcmk_rc_ok) { crm_warn("Pacemaker Remote connection to %s:%d failed: %s " QB_XS " rc=%d", native->server, native->port, pcmk_rc_str(rc), rc); lrmd_tls_connection_destroy(lrmd); return ENOTCONN; } if (native->tls == NULL) { rc = pcmk__init_tls(&native->tls, false, use_cert ? GNUTLS_CRD_CERTIFICATE : GNUTLS_CRD_PSK); if (rc != pcmk_rc_ok) { lrmd_tls_connection_destroy(lrmd); return rc; } } if (!use_cert) { gnutls_datum_t psk_key = { NULL, 0 }; rc = lrmd__init_remote_key(&psk_key); if (rc != pcmk_rc_ok) { lrmd_tls_connection_destroy(lrmd); return rc; } pcmk__tls_add_psk_key(native->tls, &psk_key); gnutls_free(psk_key.data); } native->remote->tls_session = pcmk__new_tls_session(native->tls, native->sock); if (native->remote->tls_session == NULL) { lrmd_tls_connection_destroy(lrmd); return EPROTO; } if (tls_client_handshake(lrmd) != pcmk_rc_ok) { return EKEYREJECTED; } crm_info("Client TLS connection established with Pacemaker Remote server %s:%d", native->server, native->port); if (fd) { *fd = native->sock; } else { rc = add_tls_to_mainloop(lrmd, false); } return rc; } static int lrmd_api_connect(lrmd_t * lrmd, const char *name, int *fd) { int rc = -ENOTCONN; lrmd_private_t *native = lrmd->lrmd_private; switch (native->type) { case pcmk__client_ipc: rc = lrmd_ipc_connect(lrmd, fd); break; case pcmk__client_tls: rc = lrmd_tls_connect(lrmd, fd); rc = pcmk_rc2legacy(rc); break; default: crm_err("Unsupported executor connection type (bug?): %d", native->type); rc = -EPROTONOSUPPORT; } if (rc == pcmk_ok) { rc = lrmd_handshake(lrmd, name); rc = pcmk_rc2legacy(rc); } return rc; } static int lrmd_api_connect_async(lrmd_t * lrmd, const char *name, int timeout) { int rc = pcmk_ok; lrmd_private_t *native = lrmd->lrmd_private; CRM_CHECK(native && native->callback, return -EINVAL); switch (native->type) { case pcmk__client_ipc: /* fake async connection with ipc. it should be fast * enough that we gain very little from async */ rc = lrmd_api_connect(lrmd, name, NULL); if (!rc) { report_async_connection_result(lrmd, rc); } break; case pcmk__client_tls: rc = lrmd_tls_connect_async(lrmd, timeout); rc = pcmk_rc2legacy(rc); break; default: crm_err("Unsupported executor connection type (bug?): %d", native->type); rc = -EPROTONOSUPPORT; } return rc; } static void lrmd_ipc_disconnect(lrmd_t * lrmd) { lrmd_private_t *native = lrmd->lrmd_private; if (native->source != NULL) { /* Attached to mainloop */ mainloop_del_ipc_client(native->source); native->source = NULL; native->ipc = NULL; } else if (native->ipc) { /* Not attached to mainloop */ crm_ipc_t *ipc = native->ipc; native->ipc = NULL; crm_ipc_close(ipc); crm_ipc_destroy(ipc); } } static void lrmd_tls_disconnect(lrmd_t * lrmd) { lrmd_private_t *native = lrmd->lrmd_private; if (native->remote->tls_session) { gnutls_bye(native->remote->tls_session, GNUTLS_SHUT_RDWR); gnutls_deinit(native->remote->tls_session); native->remote->tls_session = NULL; } if (native->async_timer) { g_source_remove(native->async_timer); native->async_timer = 0; } if (native->source != NULL) { /* Attached to mainloop */ mainloop_del_ipc_client(native->source); native->source = NULL; } else if (native->sock >= 0) { close(native->sock); native->sock = -1; } if (native->pending_notify) { g_list_free_full(native->pending_notify, lrmd_free_xml); native->pending_notify = NULL; } } static int lrmd_api_disconnect(lrmd_t * lrmd) { lrmd_private_t *native = lrmd->lrmd_private; int rc = pcmk_ok; switch (native->type) { case pcmk__client_ipc: crm_debug("Disconnecting from local executor"); lrmd_ipc_disconnect(lrmd); break; case pcmk__client_tls: crm_debug("Disconnecting from remote executor on %s", native->remote_nodename); lrmd_tls_disconnect(lrmd); break; default: crm_err("Unsupported executor connection type (bug?): %d", native->type); rc = -EPROTONOSUPPORT; } free(native->token); native->token = NULL; free(native->peer_version); native->peer_version = NULL; return rc; } static int lrmd_api_register_rsc(lrmd_t * lrmd, const char *rsc_id, const char *class, const char *provider, const char *type, enum lrmd_call_options options) { int rc = pcmk_ok; xmlNode *data = NULL; if (!class || !type || !rsc_id) { return -EINVAL; } if (pcmk_is_set(pcmk_get_ra_caps(class), pcmk_ra_cap_provider) && (provider == NULL)) { return -EINVAL; } data = pcmk__xe_create(NULL, PCMK__XE_LRMD_RSC); crm_xml_add(data, PCMK__XA_LRMD_ORIGIN, __func__); crm_xml_add(data, PCMK__XA_LRMD_RSC_ID, rsc_id); crm_xml_add(data, PCMK__XA_LRMD_CLASS, class); crm_xml_add(data, PCMK__XA_LRMD_PROVIDER, provider); crm_xml_add(data, PCMK__XA_LRMD_TYPE, type); rc = lrmd_send_command(lrmd, LRMD_OP_RSC_REG, data, NULL, 0, options, true); pcmk__xml_free(data); return rc; } static int lrmd_api_unregister_rsc(lrmd_t * lrmd, const char *rsc_id, enum lrmd_call_options options) { int rc = pcmk_ok; xmlNode *data = pcmk__xe_create(NULL, PCMK__XE_LRMD_RSC); crm_xml_add(data, PCMK__XA_LRMD_ORIGIN, __func__); crm_xml_add(data, PCMK__XA_LRMD_RSC_ID, rsc_id); rc = lrmd_send_command(lrmd, LRMD_OP_RSC_UNREG, data, NULL, 0, options, true); pcmk__xml_free(data); return rc; } lrmd_rsc_info_t * lrmd_new_rsc_info(const char *rsc_id, const char *standard, const char *provider, const char *type) { lrmd_rsc_info_t *rsc_info = pcmk__assert_alloc(1, sizeof(lrmd_rsc_info_t)); rsc_info->id = pcmk__str_copy(rsc_id); rsc_info->standard = pcmk__str_copy(standard); rsc_info->provider = pcmk__str_copy(provider); rsc_info->type = pcmk__str_copy(type); return rsc_info; } lrmd_rsc_info_t * lrmd_copy_rsc_info(lrmd_rsc_info_t * rsc_info) { return lrmd_new_rsc_info(rsc_info->id, rsc_info->standard, rsc_info->provider, rsc_info->type); } void lrmd_free_rsc_info(lrmd_rsc_info_t * rsc_info) { if (!rsc_info) { return; } free(rsc_info->id); free(rsc_info->type); free(rsc_info->standard); free(rsc_info->provider); free(rsc_info); } static lrmd_rsc_info_t * lrmd_api_get_rsc_info(lrmd_t * lrmd, const char *rsc_id, enum lrmd_call_options options) { lrmd_rsc_info_t *rsc_info = NULL; xmlNode *data = pcmk__xe_create(NULL, PCMK__XE_LRMD_RSC); xmlNode *output = NULL; const char *class = NULL; const char *provider = NULL; const char *type = NULL; crm_xml_add(data, PCMK__XA_LRMD_ORIGIN, __func__); crm_xml_add(data, PCMK__XA_LRMD_RSC_ID, rsc_id); lrmd_send_command(lrmd, LRMD_OP_RSC_INFO, data, &output, 0, options, true); pcmk__xml_free(data); if (!output) { return NULL; } class = pcmk__xe_get(output, PCMK__XA_LRMD_CLASS); provider = pcmk__xe_get(output, PCMK__XA_LRMD_PROVIDER); type = pcmk__xe_get(output, PCMK__XA_LRMD_TYPE); if (!class || !type) { pcmk__xml_free(output); return NULL; } else if (pcmk_is_set(pcmk_get_ra_caps(class), pcmk_ra_cap_provider) && !provider) { pcmk__xml_free(output); return NULL; } rsc_info = lrmd_new_rsc_info(rsc_id, class, provider, type); pcmk__xml_free(output); return rsc_info; } void lrmd_free_op_info(lrmd_op_info_t *op_info) { if (op_info) { free(op_info->rsc_id); free(op_info->action); free(op_info->interval_ms_s); free(op_info->timeout_ms_s); free(op_info); } } static int lrmd_api_get_recurring_ops(lrmd_t *lrmd, const char *rsc_id, int timeout_ms, enum lrmd_call_options options, GList **output) { xmlNode *data = NULL; xmlNode *output_xml = NULL; int rc = pcmk_ok; if (output == NULL) { return -EINVAL; } *output = NULL; // Send request if (rsc_id) { data = pcmk__xe_create(NULL, PCMK__XE_LRMD_RSC); crm_xml_add(data, PCMK__XA_LRMD_ORIGIN, __func__); crm_xml_add(data, PCMK__XA_LRMD_RSC_ID, rsc_id); } rc = lrmd_send_command(lrmd, LRMD_OP_GET_RECURRING, data, &output_xml, timeout_ms, options, true); if (data) { pcmk__xml_free(data); } // Process reply if ((rc != pcmk_ok) || (output_xml == NULL)) { return rc; } for (const xmlNode *rsc_xml = pcmk__xe_first_child(output_xml, PCMK__XE_LRMD_RSC, NULL, NULL); (rsc_xml != NULL) && (rc == pcmk_ok); rsc_xml = pcmk__xe_next(rsc_xml, PCMK__XE_LRMD_RSC)) { rsc_id = pcmk__xe_get(rsc_xml, PCMK__XA_LRMD_RSC_ID); if (rsc_id == NULL) { crm_err("Could not parse recurring operation information from executor"); continue; } for (const xmlNode *op_xml = pcmk__xe_first_child(rsc_xml, PCMK__XE_LRMD_RSC_OP, NULL, NULL); op_xml != NULL; op_xml = pcmk__xe_next(op_xml, PCMK__XE_LRMD_RSC_OP)) { lrmd_op_info_t *op_info = calloc(1, sizeof(lrmd_op_info_t)); if (op_info == NULL) { rc = -ENOMEM; break; } op_info->rsc_id = strdup(rsc_id); op_info->action = pcmk__xe_get_copy(op_xml, PCMK__XA_LRMD_RSC_ACTION); op_info->interval_ms_s = pcmk__xe_get_copy(op_xml, PCMK__XA_LRMD_RSC_INTERVAL); op_info->timeout_ms_s = pcmk__xe_get_copy(op_xml, PCMK__XA_LRMD_TIMEOUT); *output = g_list_prepend(*output, op_info); } } pcmk__xml_free(output_xml); return rc; } static void lrmd_api_set_callback(lrmd_t * lrmd, lrmd_event_callback callback) { lrmd_private_t *native = lrmd->lrmd_private; native->callback = callback; } void lrmd_internal_set_proxy_callback(lrmd_t * lrmd, void *userdata, void (*callback)(lrmd_t *lrmd, void *userdata, xmlNode *msg)) { lrmd_private_t *native = lrmd->lrmd_private; native->proxy_callback = callback; native->proxy_callback_userdata = userdata; } void lrmd_internal_proxy_dispatch(lrmd_t *lrmd, xmlNode *msg) { lrmd_private_t *native = lrmd->lrmd_private; if (native->proxy_callback) { crm_log_xml_trace(msg, "PROXY_INBOUND"); native->proxy_callback(lrmd, native->proxy_callback_userdata, msg); } } int lrmd_internal_proxy_send(lrmd_t * lrmd, xmlNode *msg) { if (lrmd == NULL) { return -ENOTCONN; } crm_xml_add(msg, PCMK__XA_LRMD_OP, CRM_OP_IPC_FWD); crm_log_xml_trace(msg, "PROXY_OUTBOUND"); return lrmd_send_xml_no_reply(lrmd, msg); } static int stonith_get_metadata(const char *provider, const char *type, char **output) { int rc = pcmk_ok; stonith_t *stonith_api = stonith_api_new(); if (stonith_api == NULL) { crm_err("Could not get fence agent meta-data: API memory allocation failed"); return -ENOMEM; } rc = stonith_api->cmds->metadata(stonith_api, st_opt_sync_call, type, provider, output, 0); if ((rc == pcmk_ok) && (*output == NULL)) { rc = -EIO; } stonith_api->cmds->free(stonith_api); return rc; } static int lrmd_api_get_metadata(lrmd_t *lrmd, const char *standard, const char *provider, const char *type, char **output, enum lrmd_call_options options) { return lrmd->cmds->get_metadata_params(lrmd, standard, provider, type, output, options, NULL); } static int lrmd_api_get_metadata_params(lrmd_t *lrmd, const char *standard, const char *provider, const char *type, char **output, enum lrmd_call_options options, lrmd_key_value_t *params) { svc_action_t *action = NULL; GHashTable *params_table = NULL; if (!standard || !type) { lrmd_key_value_freeall(params); return -EINVAL; } if (pcmk__str_eq(standard, PCMK_RESOURCE_CLASS_STONITH, pcmk__str_casei)) { lrmd_key_value_freeall(params); return stonith_get_metadata(provider, type, output); } params_table = pcmk__strkey_table(free, free); for (const lrmd_key_value_t *param = params; param; param = param->next) { pcmk__insert_dup(params_table, param->key, param->value); } action = services__create_resource_action(type, standard, provider, type, PCMK_ACTION_META_DATA, 0, PCMK_DEFAULT_ACTION_TIMEOUT_MS, params_table, 0); lrmd_key_value_freeall(params); if (action == NULL) { return -ENOMEM; } if (action->rc != PCMK_OCF_UNKNOWN) { services_action_free(action); return -EINVAL; } if (!services_action_sync(action)) { crm_err("Failed to retrieve meta-data for %s:%s:%s", standard, provider, type); services_action_free(action); return -EIO; } if (!action->stdout_data) { crm_err("Failed to receive meta-data for %s:%s:%s", standard, provider, type); services_action_free(action); return -EIO; } *output = strdup(action->stdout_data); services_action_free(action); return pcmk_ok; } static int lrmd_api_exec(lrmd_t *lrmd, const char *rsc_id, const char *action, const char *userdata, guint interval_ms, int timeout, /* ms */ int start_delay, /* ms */ enum lrmd_call_options options, lrmd_key_value_t * params) { int rc = pcmk_ok; xmlNode *data = pcmk__xe_create(NULL, PCMK__XE_LRMD_RSC); xmlNode *args = pcmk__xe_create(data, PCMK__XE_ATTRIBUTES); lrmd_key_value_t *tmp = NULL; crm_xml_add(data, PCMK__XA_LRMD_ORIGIN, __func__); crm_xml_add(data, PCMK__XA_LRMD_RSC_ID, rsc_id); crm_xml_add(data, PCMK__XA_LRMD_RSC_ACTION, action); crm_xml_add(data, PCMK__XA_LRMD_RSC_USERDATA_STR, userdata); - crm_xml_add_ms(data, PCMK__XA_LRMD_RSC_INTERVAL, interval_ms); + pcmk__xe_set_guint(data, PCMK__XA_LRMD_RSC_INTERVAL, interval_ms); crm_xml_add_int(data, PCMK__XA_LRMD_TIMEOUT, timeout); crm_xml_add_int(data, PCMK__XA_LRMD_RSC_START_DELAY, start_delay); for (tmp = params; tmp; tmp = tmp->next) { hash2smartfield((gpointer) tmp->key, (gpointer) tmp->value, args); } rc = lrmd_send_command(lrmd, LRMD_OP_RSC_EXEC, data, NULL, timeout, options, true); pcmk__xml_free(data); lrmd_key_value_freeall(params); return rc; } /* timeout is in ms */ static int lrmd_api_exec_alert(lrmd_t *lrmd, const char *alert_id, const char *alert_path, int timeout, lrmd_key_value_t *params) { int rc = pcmk_ok; xmlNode *data = pcmk__xe_create(NULL, PCMK__XE_LRMD_ALERT); xmlNode *args = pcmk__xe_create(data, PCMK__XE_ATTRIBUTES); lrmd_key_value_t *tmp = NULL; crm_xml_add(data, PCMK__XA_LRMD_ORIGIN, __func__); crm_xml_add(data, PCMK__XA_LRMD_ALERT_ID, alert_id); crm_xml_add(data, PCMK__XA_LRMD_ALERT_PATH, alert_path); crm_xml_add_int(data, PCMK__XA_LRMD_TIMEOUT, timeout); for (tmp = params; tmp; tmp = tmp->next) { hash2smartfield((gpointer) tmp->key, (gpointer) tmp->value, args); } rc = lrmd_send_command(lrmd, LRMD_OP_ALERT_EXEC, data, NULL, timeout, lrmd_opt_notify_orig_only, true); pcmk__xml_free(data); lrmd_key_value_freeall(params); return rc; } static int lrmd_api_cancel(lrmd_t *lrmd, const char *rsc_id, const char *action, guint interval_ms) { int rc = pcmk_ok; xmlNode *data = pcmk__xe_create(NULL, PCMK__XE_LRMD_RSC); crm_xml_add(data, PCMK__XA_LRMD_ORIGIN, __func__); crm_xml_add(data, PCMK__XA_LRMD_RSC_ACTION, action); crm_xml_add(data, PCMK__XA_LRMD_RSC_ID, rsc_id); - crm_xml_add_ms(data, PCMK__XA_LRMD_RSC_INTERVAL, interval_ms); + pcmk__xe_set_guint(data, PCMK__XA_LRMD_RSC_INTERVAL, interval_ms); rc = lrmd_send_command(lrmd, LRMD_OP_RSC_CANCEL, data, NULL, 0, 0, true); pcmk__xml_free(data); return rc; } static int list_stonith_agents(lrmd_list_t ** resources) { int rc = 0; stonith_t *stonith_api = stonith_api_new(); stonith_key_value_t *stonith_resources = NULL; stonith_key_value_t *dIter = NULL; if (stonith_api == NULL) { crm_err("Could not list fence agents: API memory allocation failed"); return -ENOMEM; } stonith_api->cmds->list_agents(stonith_api, st_opt_sync_call, NULL, &stonith_resources, 0); stonith_api->cmds->free(stonith_api); for (dIter = stonith_resources; dIter; dIter = dIter->next) { rc++; if (resources) { *resources = lrmd_list_add(*resources, dIter->value); } } stonith_key_value_freeall(stonith_resources, 1, 0); return rc; } static int lrmd_api_list_agents(lrmd_t * lrmd, lrmd_list_t ** resources, const char *class, const char *provider) { int rc = 0; int stonith_count = 0; // Initially, whether to include stonith devices if (pcmk__str_eq(class, PCMK_RESOURCE_CLASS_STONITH, pcmk__str_casei)) { stonith_count = 1; } else { GList *gIter = NULL; GList *agents = resources_list_agents(class, provider); for (gIter = agents; gIter != NULL; gIter = gIter->next) { *resources = lrmd_list_add(*resources, (const char *)gIter->data); rc++; } g_list_free_full(agents, free); if (!class) { stonith_count = 1; } } if (stonith_count) { // Now, if stonith devices are included, how many there are stonith_count = list_stonith_agents(resources); if (stonith_count > 0) { rc += stonith_count; } } if (rc == 0) { crm_notice("No agents found for class %s", class); rc = -EPROTONOSUPPORT; } return rc; } static bool does_provider_have_agent(const char *agent, const char *provider, const char *class) { bool found = false; GList *agents = NULL; GList *gIter2 = NULL; agents = resources_list_agents(class, provider); for (gIter2 = agents; gIter2 != NULL; gIter2 = gIter2->next) { if (pcmk__str_eq(agent, gIter2->data, pcmk__str_casei)) { found = true; } } g_list_free_full(agents, free); return found; } static int lrmd_api_list_ocf_providers(lrmd_t * lrmd, const char *agent, lrmd_list_t ** providers) { int rc = pcmk_ok; char *provider = NULL; GList *ocf_providers = NULL; GList *gIter = NULL; ocf_providers = resources_list_providers(PCMK_RESOURCE_CLASS_OCF); for (gIter = ocf_providers; gIter != NULL; gIter = gIter->next) { provider = gIter->data; if (!agent || does_provider_have_agent(agent, provider, PCMK_RESOURCE_CLASS_OCF)) { *providers = lrmd_list_add(*providers, (const char *)gIter->data); rc++; } } g_list_free_full(ocf_providers, free); return rc; } static int lrmd_api_list_standards(lrmd_t * lrmd, lrmd_list_t ** supported) { int rc = 0; GList *standards = NULL; GList *gIter = NULL; standards = resources_list_standards(); for (gIter = standards; gIter != NULL; gIter = gIter->next) { *supported = lrmd_list_add(*supported, (const char *)gIter->data); rc++; } if (list_stonith_agents(NULL) > 0) { *supported = lrmd_list_add(*supported, PCMK_RESOURCE_CLASS_STONITH); rc++; } g_list_free_full(standards, free); return rc; } /*! * \internal * \brief Create an executor API object * * \param[out] api Will be set to newly created API object (it is the * caller's responsibility to free this value with * lrmd_api_delete() if this function succeeds) * \param[in] nodename If the object will be used for a remote connection, * the node name to use in cluster for remote executor * \param[in] server If the object will be used for a remote connection, * the resolvable host name to connect to * \param[in] port If the object will be used for a remote connection, * port number on \p server to connect to * * \return Standard Pacemaker return code * \note If the caller leaves one of \p nodename or \p server NULL, the other's * value will be used for both. If the caller leaves both NULL, an API * object will be created for a local executor connection. */ int lrmd__new(lrmd_t **api, const char *nodename, const char *server, int port) { lrmd_private_t *pvt = NULL; if (api == NULL) { return EINVAL; } *api = NULL; // Allocate all memory needed *api = calloc(1, sizeof(lrmd_t)); if (*api == NULL) { return ENOMEM; } pvt = calloc(1, sizeof(lrmd_private_t)); if (pvt == NULL) { lrmd_api_delete(*api); *api = NULL; return ENOMEM; } (*api)->lrmd_private = pvt; // @TODO Do we need to do this for local connections? pvt->remote = calloc(1, sizeof(pcmk__remote_t)); (*api)->cmds = calloc(1, sizeof(lrmd_api_operations_t)); if ((pvt->remote == NULL) || ((*api)->cmds == NULL)) { lrmd_api_delete(*api); *api = NULL; return ENOMEM; } // Set methods (*api)->cmds->connect = lrmd_api_connect; (*api)->cmds->connect_async = lrmd_api_connect_async; (*api)->cmds->is_connected = lrmd_api_is_connected; (*api)->cmds->poke_connection = lrmd_api_poke_connection; (*api)->cmds->disconnect = lrmd_api_disconnect; (*api)->cmds->register_rsc = lrmd_api_register_rsc; (*api)->cmds->unregister_rsc = lrmd_api_unregister_rsc; (*api)->cmds->get_rsc_info = lrmd_api_get_rsc_info; (*api)->cmds->get_recurring_ops = lrmd_api_get_recurring_ops; (*api)->cmds->set_callback = lrmd_api_set_callback; (*api)->cmds->get_metadata = lrmd_api_get_metadata; (*api)->cmds->exec = lrmd_api_exec; (*api)->cmds->cancel = lrmd_api_cancel; (*api)->cmds->list_agents = lrmd_api_list_agents; (*api)->cmds->list_ocf_providers = lrmd_api_list_ocf_providers; (*api)->cmds->list_standards = lrmd_api_list_standards; (*api)->cmds->exec_alert = lrmd_api_exec_alert; (*api)->cmds->get_metadata_params = lrmd_api_get_metadata_params; if ((nodename == NULL) && (server == NULL)) { pvt->type = pcmk__client_ipc; } else { if (nodename == NULL) { nodename = server; } else if (server == NULL) { server = nodename; } pvt->type = pcmk__client_tls; pvt->remote_nodename = strdup(nodename); pvt->server = strdup(server); if ((pvt->remote_nodename == NULL) || (pvt->server == NULL)) { lrmd_api_delete(*api); *api = NULL; return ENOMEM; } pvt->port = port; if (pvt->port == 0) { pvt->port = crm_default_remote_port(); } } return pcmk_rc_ok; } lrmd_t * lrmd_api_new(void) { lrmd_t *api = NULL; pcmk__assert(lrmd__new(&api, NULL, NULL, 0) == pcmk_rc_ok); return api; } lrmd_t * lrmd_remote_api_new(const char *nodename, const char *server, int port) { lrmd_t *api = NULL; pcmk__assert(lrmd__new(&api, nodename, server, port) == pcmk_rc_ok); return api; } void lrmd_api_delete(lrmd_t * lrmd) { if (lrmd == NULL) { return; } if (lrmd->cmds != NULL) { // Never NULL, but make static analysis happy if (lrmd->cmds->disconnect != NULL) { // Also never really NULL lrmd->cmds->disconnect(lrmd); // No-op if already disconnected } free(lrmd->cmds); } if (lrmd->lrmd_private != NULL) { lrmd_private_t *native = lrmd->lrmd_private; free(native->server); free(native->remote_nodename); free(native->remote); free(native->token); free(native->peer_version); free(lrmd->lrmd_private); } free(lrmd); } struct metadata_cb { void (*callback)(int pid, const pcmk__action_result_t *result, void *user_data); void *user_data; }; /*! * \internal * \brief Process asynchronous metadata completion * * \param[in,out] action Metadata action that completed */ static void metadata_complete(svc_action_t *action) { struct metadata_cb *metadata_cb = (struct metadata_cb *) action->cb_data; pcmk__action_result_t result = PCMK__UNKNOWN_RESULT; services__copy_result(action, &result); pcmk__set_result_output(&result, action->stdout_data, action->stderr_data); metadata_cb->callback(0, &result, metadata_cb->user_data); result.action_stdout = NULL; // Prevent free, because action owns it result.action_stderr = NULL; // Prevent free, because action owns it pcmk__reset_result(&result); free(metadata_cb); } /*! * \internal * \brief Retrieve agent metadata asynchronously * * \param[in] rsc Resource agent specification * \param[in] callback Function to call with result (this will always be * called, whether by this function directly or later * via the main loop, and on success the metadata will * be in its result argument's action_stdout) * \param[in,out] user_data User data to pass to callback * * \return Standard Pacemaker return code * \note This function is not a lrmd_api_operations_t method because it does not * need an lrmd_t object and does not go through the executor, but * executes the agent directly. */ int lrmd__metadata_async(const lrmd_rsc_info_t *rsc, void (*callback)(int pid, const pcmk__action_result_t *result, void *user_data), void *user_data) { svc_action_t *action = NULL; struct metadata_cb *metadata_cb = NULL; pcmk__action_result_t result = PCMK__UNKNOWN_RESULT; CRM_CHECK(callback != NULL, return EINVAL); if ((rsc == NULL) || (rsc->standard == NULL) || (rsc->type == NULL)) { pcmk__set_result(&result, PCMK_OCF_NOT_CONFIGURED, PCMK_EXEC_ERROR_FATAL, "Invalid resource specification"); callback(0, &result, user_data); pcmk__reset_result(&result); return EINVAL; } if (strcmp(rsc->standard, PCMK_RESOURCE_CLASS_STONITH) == 0) { return stonith__metadata_async(rsc->type, pcmk__timeout_ms2s(PCMK_DEFAULT_ACTION_TIMEOUT_MS), callback, user_data); } action = services__create_resource_action(pcmk__s(rsc->id, rsc->type), rsc->standard, rsc->provider, rsc->type, PCMK_ACTION_META_DATA, 0, PCMK_DEFAULT_ACTION_TIMEOUT_MS, NULL, 0); if (action == NULL) { pcmk__set_result(&result, PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_ERROR, "Out of memory"); callback(0, &result, user_data); pcmk__reset_result(&result); return ENOMEM; } if (action->rc != PCMK_OCF_UNKNOWN) { services__copy_result(action, &result); callback(0, &result, user_data); pcmk__reset_result(&result); services_action_free(action); return EINVAL; } action->cb_data = calloc(1, sizeof(struct metadata_cb)); if (action->cb_data == NULL) { services_action_free(action); pcmk__set_result(&result, PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_ERROR, "Out of memory"); callback(0, &result, user_data); pcmk__reset_result(&result); return ENOMEM; } metadata_cb = (struct metadata_cb *) action->cb_data; metadata_cb->callback = callback; metadata_cb->user_data = user_data; if (!services_action_async(action, metadata_complete)) { services_action_free(action); return pcmk_rc_error; // @TODO Derive from action->rc and ->status } // The services library has taken responsibility for action return pcmk_rc_ok; } /*! * \internal * \brief Set the result of an executor event * * \param[in,out] event Executor event to set * \param[in] rc OCF exit status of event * \param[in] op_status Executor status of event * \param[in] exit_reason Human-friendly description of event */ void lrmd__set_result(lrmd_event_data_t *event, enum ocf_exitcode rc, int op_status, const char *exit_reason) { if (event == NULL) { return; } event->rc = rc; event->op_status = op_status; // lrmd_event_data_t has (const char *) members that lrmd_free_event() frees pcmk__str_update((char **) &event->exit_reason, exit_reason); } /*! * \internal * \brief Clear an executor event's exit reason, output, and error output * * \param[in,out] event Executor event to reset */ void lrmd__reset_result(lrmd_event_data_t *event) { if (event == NULL) { return; } free((void *) event->exit_reason); event->exit_reason = NULL; free((void *) event->output); event->output = NULL; } /*! * \internal * \brief Get the uptime of a remote resource connection * * When the cluster connects to a remote resource, part of that resource's * handshake includes the uptime of the remote resource's connection. This * uptime is stored in the lrmd_t object. * * \return The connection's uptime, or -1 if unknown */ time_t lrmd__uptime(lrmd_t *lrmd) { lrmd_private_t *native = lrmd->lrmd_private; if (native->remote == NULL) { return -1; } else { return native->remote->uptime; } } const char * lrmd__node_start_state(lrmd_t *lrmd) { lrmd_private_t *native = lrmd->lrmd_private; if (native->remote == NULL) { return NULL; } else { return native->remote->start_state; } } diff --git a/lib/pacemaker/pcmk_sched_actions.c b/lib/pacemaker/pcmk_sched_actions.c index 7216b71067..01896567e7 100644 --- a/lib/pacemaker/pcmk_sched_actions.c +++ b/lib/pacemaker/pcmk_sched_actions.c @@ -1,1955 +1,1955 @@ /* * Copyright 2004-2025 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include // bool, true, false #include #include #include #include // xmlNode #include #include #include #include "libpacemaker_private.h" /*! * \internal * \brief Get the action flags relevant to ordering constraints * * \param[in,out] action Action to check * \param[in] node Node that *other* action in the ordering is on * (used only for clone resource actions) * * \return Action flags that should be used for orderings */ static uint32_t action_flags_for_ordering(pcmk_action_t *action, const pcmk_node_t *node) { bool runnable = false; uint32_t flags; // For non-resource actions, return the action flags if (action->rsc == NULL) { return action->flags; } /* For non-clone resources, or a clone action not assigned to a node, * return the flags as determined by the resource method without a node * specified. */ flags = action->rsc->priv->cmds->action_flags(action, NULL); if ((node == NULL) || !pcmk__is_clone(action->rsc)) { return flags; } /* Otherwise (i.e., for clone resource actions on a specific node), first * remember whether the non-node-specific action is runnable. */ runnable = pcmk_is_set(flags, pcmk__action_runnable); // Then recheck the resource method with the node flags = action->rsc->priv->cmds->action_flags(action, node); /* For clones in ordering constraints, the node-specific "runnable" doesn't * matter, just the non-node-specific setting (i.e., is the action runnable * anywhere). * * This applies only to runnable, and only for ordering constraints. This * function shouldn't be used for other types of constraints without * changes. Not very satisfying, but it's logical and appears to work well. */ if (runnable && !pcmk_is_set(flags, pcmk__action_runnable)) { pcmk__set_raw_action_flags(flags, action->rsc->id, pcmk__action_runnable); } return flags; } /*! * \internal * \brief Get action UUID that should be used with a resource ordering * * When an action is ordered relative to an action for a collective resource * (clone, group, or bundle), it actually needs to be ordered after all * instances of the collective have completed the relevant action (for example, * given "start CLONE then start RSC", RSC must wait until all instances of * CLONE have started). Given the UUID and resource of the first action in an * ordering, this returns the UUID of the action that should actually be used * for ordering (for example, "CLONE_started_0" instead of "CLONE_start_0"). * * \param[in] first_uuid UUID of first action in ordering * \param[in] first_rsc Resource of first action in ordering * * \return Newly allocated copy of UUID to use with ordering * \note It is the caller's responsibility to free the return value. */ static char * action_uuid_for_ordering(const char *first_uuid, const pcmk_resource_t *first_rsc) { guint interval_ms = 0; char *uuid = NULL; char *rid = NULL; char *first_task_str = NULL; enum pcmk__action_type first_task = pcmk__action_unspecified; enum pcmk__action_type remapped_task = pcmk__action_unspecified; // Only non-notify actions for collective resources need remapping if ((strstr(first_uuid, PCMK_ACTION_NOTIFY) != NULL) || (first_rsc->priv->variant < pcmk__rsc_variant_group)) { goto done; } // Only non-recurring actions need remapping pcmk__assert(parse_op_key(first_uuid, &rid, &first_task_str, &interval_ms)); if (interval_ms > 0) { goto done; } first_task = pcmk__parse_action(first_task_str); switch (first_task) { case pcmk__action_stop: case pcmk__action_start: case pcmk__action_notify: case pcmk__action_promote: case pcmk__action_demote: remapped_task = first_task + 1; break; case pcmk__action_stopped: case pcmk__action_started: case pcmk__action_notified: case pcmk__action_promoted: case pcmk__action_demoted: remapped_task = first_task; break; case pcmk__action_monitor: case pcmk__action_shutdown: case pcmk__action_fence: break; default: crm_err("Unknown action '%s' in ordering", first_task_str); break; } if (remapped_task != pcmk__action_unspecified) { /* If a clone or bundle has notifications enabled, the ordering will be * relative to when notifications have been sent for the remapped task. */ if (pcmk_is_set(first_rsc->flags, pcmk__rsc_notify) && (pcmk__is_clone(first_rsc) || pcmk__is_bundled(first_rsc))) { uuid = pcmk__notify_key(rid, "confirmed-post", pcmk__action_text(remapped_task)); } else { uuid = pcmk__op_key(rid, pcmk__action_text(remapped_task), 0); } pcmk__rsc_trace(first_rsc, "Remapped action UUID %s to %s for ordering purposes", first_uuid, uuid); } done: free(first_task_str); free(rid); return (uuid != NULL)? uuid : pcmk__str_copy(first_uuid); } /*! * \internal * \brief Get actual action that should be used with an ordering * * When an action is ordered relative to an action for a collective resource * (clone, group, or bundle), it actually needs to be ordered after all * instances of the collective have completed the relevant action (for example, * given "start CLONE then start RSC", RSC must wait until all instances of * CLONE have started). Given the first action in an ordering, this returns the * the action that should actually be used for ordering (for example, the * started action instead of the start action). * * \param[in] action First action in an ordering * * \return Actual action that should be used for the ordering */ static pcmk_action_t * action_for_ordering(pcmk_action_t *action) { pcmk_action_t *result = action; pcmk_resource_t *rsc = action->rsc; if (rsc == NULL) { return result; } if ((rsc->priv->variant >= pcmk__rsc_variant_group) && (action->uuid != NULL)) { char *uuid = action_uuid_for_ordering(action->uuid, rsc); result = find_first_action(rsc->priv->actions, uuid, NULL, NULL); if (result == NULL) { crm_warn("Not remapping %s to %s because %s does not have " "remapped action", action->uuid, uuid, rsc->id); result = action; } free(uuid); } return result; } /*! * \internal * \brief Wrapper for update_ordered_actions() method for readability * * \param[in,out] rsc Resource to call method for * \param[in,out] first 'First' action in an ordering * \param[in,out] then 'Then' action in an ordering * \param[in] node If not NULL, limit scope of ordering to this * node (only used when interleaving instances) * \param[in] flags Action flags for \p first for ordering purposes * \param[in] filter Action flags to limit scope of certain updates * (may include pcmk__action_optional to affect only * mandatory actions, and pe_action_runnable to * affect only runnable actions) * \param[in] type Group of enum pcmk__action_relation_flags to apply * \param[in,out] scheduler Scheduler data * * \return Group of enum pcmk__updated flags indicating what was updated */ static inline uint32_t update(pcmk_resource_t *rsc, pcmk_action_t *first, pcmk_action_t *then, const pcmk_node_t *node, uint32_t flags, uint32_t filter, uint32_t type, pcmk_scheduler_t *scheduler) { return rsc->priv->cmds->update_ordered_actions(first, then, node, flags, filter, type, scheduler); } /*! * \internal * \brief Update flags for ordering's actions appropriately for ordering's flags * * \param[in,out] first First action in an ordering * \param[in,out] then Then action in an ordering * \param[in] first_flags Action flags for \p first for ordering purposes * \param[in] then_flags Action flags for \p then for ordering purposes * \param[in,out] order Action wrapper for \p first in ordering * \param[in,out] scheduler Scheduler data * * \return Group of enum pcmk__updated flags */ static uint32_t update_action_for_ordering_flags(pcmk_action_t *first, pcmk_action_t *then, uint32_t first_flags, uint32_t then_flags, pcmk__related_action_t *order, pcmk_scheduler_t *scheduler) { uint32_t changed = pcmk__updated_none; /* The node will only be used for clones. If interleaved, node will be NULL, * otherwise the ordering scope will be limited to the node. Normally, the * whole 'then' clone should restart if 'first' is restarted, so then->node * is needed. */ pcmk_node_t *node = then->node; if (pcmk_is_set(order->flags, pcmk__ar_first_implies_same_node_then)) { /* For unfencing, only instances of 'then' on the same node as 'first' * (the unfencing operation) should restart, so reset node to * first->node, at which point this case is handled like a normal * pcmk__ar_first_implies_then. */ pcmk__clear_relation_flags(order->flags, pcmk__ar_first_implies_same_node_then); pcmk__set_relation_flags(order->flags, pcmk__ar_first_implies_then); node = first->node; pcmk__rsc_trace(then->rsc, "%s then %s: mapped " "pcmk__ar_first_implies_same_node_then to " "pcmk__ar_first_implies_then on %s", first->uuid, then->uuid, pcmk__node_name(node)); } if (pcmk_is_set(order->flags, pcmk__ar_first_implies_then)) { if (then->rsc != NULL) { changed |= update(then->rsc, first, then, node, first_flags & pcmk__action_optional, pcmk__action_optional, pcmk__ar_first_implies_then, scheduler); } else if (!pcmk_is_set(first_flags, pcmk__action_optional) && pcmk_is_set(then->flags, pcmk__action_optional)) { pcmk__clear_action_flags(then, pcmk__action_optional); pcmk__set_updated_flags(changed, first, pcmk__updated_then); } pcmk__rsc_trace(then->rsc, "%s then %s: %s after pcmk__ar_first_implies_then", first->uuid, then->uuid, (changed? "changed" : "unchanged")); } if (pcmk_is_set(order->flags, pcmk__ar_intermediate_stop) && (then->rsc != NULL)) { enum pcmk__action_flags restart = pcmk__action_optional |pcmk__action_runnable; changed |= update(then->rsc, first, then, node, first_flags, restart, pcmk__ar_intermediate_stop, scheduler); pcmk__rsc_trace(then->rsc, "%s then %s: %s after pcmk__ar_intermediate_stop", first->uuid, then->uuid, (changed? "changed" : "unchanged")); } if (pcmk_is_set(order->flags, pcmk__ar_then_implies_first)) { if (first->rsc != NULL) { changed |= update(first->rsc, first, then, node, first_flags, pcmk__action_optional, pcmk__ar_then_implies_first, scheduler); } else if (!pcmk_is_set(first_flags, pcmk__action_optional) && pcmk_is_set(first->flags, pcmk__action_runnable)) { pcmk__clear_action_flags(first, pcmk__action_runnable); pcmk__set_updated_flags(changed, first, pcmk__updated_first); } pcmk__rsc_trace(then->rsc, "%s then %s: %s after pcmk__ar_then_implies_first", first->uuid, then->uuid, (changed? "changed" : "unchanged")); } if (pcmk_is_set(order->flags, pcmk__ar_promoted_then_implies_first)) { if (then->rsc != NULL) { changed |= update(then->rsc, first, then, node, first_flags & pcmk__action_optional, pcmk__action_optional, pcmk__ar_promoted_then_implies_first, scheduler); } pcmk__rsc_trace(then->rsc, "%s then %s: %s after " "pcmk__ar_promoted_then_implies_first", first->uuid, then->uuid, (changed? "changed" : "unchanged")); } if (pcmk_is_set(order->flags, pcmk__ar_min_runnable)) { if (then->rsc != NULL) { changed |= update(then->rsc, first, then, node, first_flags, pcmk__action_runnable, pcmk__ar_min_runnable, scheduler); } else if (pcmk_is_set(first_flags, pcmk__action_runnable)) { // We have another runnable instance of "first" then->runnable_before++; /* Mark "then" as runnable if it requires a certain number of * "before" instances to be runnable, and they now are. */ if ((then->runnable_before >= then->required_runnable_before) && !pcmk_is_set(then->flags, pcmk__action_runnable)) { pcmk__set_action_flags(then, pcmk__action_runnable); pcmk__set_updated_flags(changed, first, pcmk__updated_then); } } pcmk__rsc_trace(then->rsc, "%s then %s: %s after pcmk__ar_min_runnable", first->uuid, then->uuid, (changed? "changed" : "unchanged")); } if (pcmk_is_set(order->flags, pcmk__ar_nested_remote_probe) && (then->rsc != NULL)) { if (!pcmk_is_set(first_flags, pcmk__action_runnable) && (first->rsc != NULL) && (first->rsc->priv->active_nodes != NULL)) { pcmk__rsc_trace(then->rsc, "%s then %s: ignoring because first is stopping", first->uuid, then->uuid); order->flags = pcmk__ar_none; } else { changed |= update(then->rsc, first, then, node, first_flags, pcmk__action_runnable, pcmk__ar_unrunnable_first_blocks, scheduler); } pcmk__rsc_trace(then->rsc, "%s then %s: %s after pcmk__ar_nested_remote_probe", first->uuid, then->uuid, (changed? "changed" : "unchanged")); } if (pcmk_is_set(order->flags, pcmk__ar_unrunnable_first_blocks)) { if (then->rsc != NULL) { changed |= update(then->rsc, first, then, node, first_flags, pcmk__action_runnable, pcmk__ar_unrunnable_first_blocks, scheduler); } else if (!pcmk_is_set(first_flags, pcmk__action_runnable) && pcmk_is_set(then->flags, pcmk__action_runnable)) { pcmk__clear_action_flags(then, pcmk__action_runnable); pcmk__set_updated_flags(changed, first, pcmk__updated_then); } pcmk__rsc_trace(then->rsc, "%s then %s: %s after pcmk__ar_unrunnable_first_blocks", first->uuid, then->uuid, (changed? "changed" : "unchanged")); } if (pcmk_is_set(order->flags, pcmk__ar_unmigratable_then_blocks)) { if (then->rsc != NULL) { changed |= update(then->rsc, first, then, node, first_flags, pcmk__action_optional, pcmk__ar_unmigratable_then_blocks, scheduler); } pcmk__rsc_trace(then->rsc, "%s then %s: %s after " "pcmk__ar_unmigratable_then_blocks", first->uuid, then->uuid, (changed? "changed" : "unchanged")); } if (pcmk_is_set(order->flags, pcmk__ar_first_else_then)) { if (then->rsc != NULL) { changed |= update(then->rsc, first, then, node, first_flags, pcmk__action_optional, pcmk__ar_first_else_then, scheduler); } pcmk__rsc_trace(then->rsc, "%s then %s: %s after pcmk__ar_first_else_then", first->uuid, then->uuid, (changed? "changed" : "unchanged")); } if (pcmk_is_set(order->flags, pcmk__ar_ordered)) { if (then->rsc != NULL) { changed |= update(then->rsc, first, then, node, first_flags, pcmk__action_runnable, pcmk__ar_ordered, scheduler); } pcmk__rsc_trace(then->rsc, "%s then %s: %s after pcmk__ar_ordered", first->uuid, then->uuid, (changed? "changed" : "unchanged")); } if (pcmk_is_set(order->flags, pcmk__ar_asymmetric)) { if (then->rsc != NULL) { changed |= update(then->rsc, first, then, node, first_flags, pcmk__action_runnable, pcmk__ar_asymmetric, scheduler); } pcmk__rsc_trace(then->rsc, "%s then %s: %s after pcmk__ar_asymmetric", first->uuid, then->uuid, (changed? "changed" : "unchanged")); } if (pcmk_is_set(first->flags, pcmk__action_runnable) && pcmk_is_set(order->flags, pcmk__ar_first_implies_then_graphed) && !pcmk_is_set(first_flags, pcmk__action_optional)) { pcmk__rsc_trace(then->rsc, "%s will be in graph because %s is required", then->uuid, first->uuid); pcmk__set_action_flags(then, pcmk__action_always_in_graph); // Don't bother marking 'then' as changed just for this } if (pcmk_is_set(order->flags, pcmk__ar_then_implies_first_graphed) && !pcmk_is_set(then_flags, pcmk__action_optional)) { pcmk__rsc_trace(then->rsc, "%s will be in graph because %s is required", first->uuid, then->uuid); pcmk__set_action_flags(first, pcmk__action_always_in_graph); // Don't bother marking 'first' as changed just for this } if (pcmk_any_flags_set(order->flags, pcmk__ar_first_implies_then |pcmk__ar_then_implies_first |pcmk__ar_intermediate_stop) && (first->rsc != NULL) && !pcmk_is_set(first->rsc->flags, pcmk__rsc_managed) && pcmk_is_set(first->rsc->flags, pcmk__rsc_blocked) && !pcmk_is_set(first->flags, pcmk__action_runnable) && pcmk__str_eq(first->task, PCMK_ACTION_STOP, pcmk__str_none)) { /* @TODO This seems odd; why wouldn't an unrunnable "first" already * block "then" before this? Note that the unmanaged-stop-{1,2} * scheduler regression tests and the test CIB for T209 have tests for * "stop then stop" relations that would be good for checking any * changes. */ if (pcmk_is_set(then->flags, pcmk__action_runnable)) { pcmk__clear_action_flags(then, pcmk__action_runnable); pcmk__set_updated_flags(changed, first, pcmk__updated_then); } pcmk__rsc_trace(then->rsc, "%s then %s: %s after checking whether first " "is blocked, unmanaged, unrunnable stop", first->uuid, then->uuid, (changed? "changed" : "unchanged")); } return changed; } // Convenience macros for logging action properties #define action_type_str(flags) \ (pcmk_is_set((flags), pcmk__action_pseudo)? "pseudo-action" : "action") #define action_optional_str(flags) \ (pcmk_is_set((flags), pcmk__action_optional)? "optional" : "required") #define action_runnable_str(flags) \ (pcmk_is_set((flags), pcmk__action_runnable)? "runnable" : "unrunnable") #define action_node_str(a) \ (((a)->node == NULL)? "no node" : (a)->node->priv->name) /*! * \internal * \brief Update an action's flags for all orderings where it is "then" * * \param[in,out] then Action to update * \param[in,out] scheduler Scheduler data */ void pcmk__update_action_for_orderings(pcmk_action_t *then, pcmk_scheduler_t *scheduler) { GList *lpc = NULL; uint32_t changed = pcmk__updated_none; int last_flags = then->flags; pcmk__rsc_trace(then->rsc, "Updating %s %s (%s %s) on %s", action_type_str(then->flags), then->uuid, action_optional_str(then->flags), action_runnable_str(then->flags), action_node_str(then)); if (then->required_runnable_before > 0) { /* Initialize current known "runnable before" actions. As * update_action_for_ordering_flags() is called for each of then's * before actions, this number will increment as runnable 'first' * actions are encountered. */ then->runnable_before = 0; /* The pcmk__ar_min_runnable clause of * update_action_for_ordering_flags() (called below) * will reset runnable if appropriate. */ pcmk__clear_action_flags(then, pcmk__action_runnable); } for (lpc = then->actions_before; lpc != NULL; lpc = lpc->next) { pcmk__related_action_t *other = lpc->data; pcmk_action_t *first = other->action; pcmk_node_t *then_node = then->node; pcmk_node_t *first_node = first->node; const uint32_t target = pcmk__rsc_node_assigned; if ((first->rsc != NULL) && pcmk__is_group(first->rsc) && pcmk__str_eq(first->task, PCMK_ACTION_START, pcmk__str_none)) { first_node = first->rsc->priv->fns->location(first->rsc, NULL, target); if (first_node != NULL) { pcmk__rsc_trace(first->rsc, "Found %s for 'first' %s", pcmk__node_name(first_node), first->uuid); } } if (pcmk__is_group(then->rsc) && pcmk__str_eq(then->task, PCMK_ACTION_START, pcmk__str_none)) { then_node = then->rsc->priv->fns->location(then->rsc, NULL, target); if (then_node != NULL) { pcmk__rsc_trace(then->rsc, "Found %s for 'then' %s", pcmk__node_name(then_node), then->uuid); } } // Disable constraint if it only applies when on same node, but isn't if (pcmk_is_set(other->flags, pcmk__ar_if_on_same_node) && (first_node != NULL) && (then_node != NULL) && !pcmk__same_node(first_node, then_node)) { pcmk__rsc_trace(then->rsc, "Disabled ordering %s on %s then %s on %s: " "not same node", other->action->uuid, pcmk__node_name(first_node), then->uuid, pcmk__node_name(then_node)); other->flags = pcmk__ar_none; continue; } pcmk__clear_updated_flags(changed, then, pcmk__updated_first); if ((first->rsc != NULL) && pcmk_is_set(other->flags, pcmk__ar_then_cancels_first) && !pcmk_is_set(then->flags, pcmk__action_optional)) { /* 'then' is required, so we must abandon 'first' * (e.g. a required stop cancels any agent reload). */ pcmk__set_action_flags(other->action, pcmk__action_optional); if (!strcmp(first->task, PCMK_ACTION_RELOAD_AGENT)) { pcmk__clear_rsc_flags(first->rsc, pcmk__rsc_reload); } } if ((first->rsc != NULL) && (then->rsc != NULL) && (first->rsc != then->rsc) && !is_parent(then->rsc, first->rsc)) { first = action_for_ordering(first); } if (first != other->action) { pcmk__rsc_trace(then->rsc, "Ordering %s after %s instead of %s", then->uuid, first->uuid, other->action->uuid); } pcmk__rsc_trace(then->rsc, "%s (%#.6x) then %s (%#.6x): type=%#.6x node=%s", first->uuid, first->flags, then->uuid, then->flags, other->flags, action_node_str(first)); if (first == other->action) { /* 'first' was not remapped (e.g. from 'start' to 'running'), which * could mean it is a non-resource action, a primitive resource * action, or already expanded. */ uint32_t first_flags, then_flags; first_flags = action_flags_for_ordering(first, then_node); then_flags = action_flags_for_ordering(then, first_node); changed |= update_action_for_ordering_flags(first, then, first_flags, then_flags, other, scheduler); /* 'first' was for a complex resource (clone, group, etc), * create a new dependency if necessary */ } else if (order_actions(first, then, other->flags)) { /* This was the first time 'first' and 'then' were associated, * start again to get the new actions_before list */ pcmk__set_updated_flags(changed, then, pcmk__updated_then); pcmk__rsc_trace(then->rsc, "Disabled ordering %s then %s in favor of %s " "then %s", other->action->uuid, then->uuid, first->uuid, then->uuid); other->flags = pcmk__ar_none; } if (pcmk_is_set(changed, pcmk__updated_first)) { crm_trace("Re-processing %s and its 'after' actions " "because it changed", first->uuid); for (GList *lpc2 = first->actions_after; lpc2 != NULL; lpc2 = lpc2->next) { pcmk__related_action_t *other = lpc2->data; pcmk__update_action_for_orderings(other->action, scheduler); } pcmk__update_action_for_orderings(first, scheduler); } } if (then->required_runnable_before > 0) { if (last_flags == then->flags) { pcmk__clear_updated_flags(changed, then, pcmk__updated_then); } else { pcmk__set_updated_flags(changed, then, pcmk__updated_then); } } if (pcmk_is_set(changed, pcmk__updated_then)) { crm_trace("Re-processing %s and its 'after' actions because it changed", then->uuid); if (pcmk_is_set(last_flags, pcmk__action_runnable) && !pcmk_is_set(then->flags, pcmk__action_runnable)) { pcmk__block_colocation_dependents(then); } pcmk__update_action_for_orderings(then, scheduler); for (lpc = then->actions_after; lpc != NULL; lpc = lpc->next) { pcmk__related_action_t *other = lpc->data; pcmk__update_action_for_orderings(other->action, scheduler); } } } static inline bool is_primitive_action(const pcmk_action_t *action) { return (action != NULL) && pcmk__is_primitive(action->rsc); } /*! * \internal * \brief Clear a single action flag and set reason text * * \param[in,out] action Action whose flag should be cleared * \param[in] flag Action flag that should be cleared * \param[in] reason Action that is the reason why flag is being cleared */ #define clear_action_flag_because(action, flag, reason) do { \ if (pcmk_is_set((action)->flags, (flag))) { \ pcmk__clear_action_flags(action, flag); \ if ((action)->rsc != (reason)->rsc) { \ char *reason_text = pe__action2reason((reason), (flag)); \ pe_action_set_reason((action), reason_text, false); \ free(reason_text); \ } \ } \ } while (0) /*! * \internal * \brief Update actions in an asymmetric ordering * * If the "first" action in an asymmetric ordering is unrunnable, make the * "second" action unrunnable as well, if appropriate. * * \param[in] first 'First' action in an asymmetric ordering * \param[in,out] then 'Then' action in an asymmetric ordering */ static void handle_asymmetric_ordering(const pcmk_action_t *first, pcmk_action_t *then) { /* Only resource actions after an unrunnable 'first' action need updates for * asymmetric ordering. */ if ((then->rsc == NULL) || pcmk_is_set(first->flags, pcmk__action_runnable)) { return; } // Certain optional 'then' actions are unaffected by unrunnable 'first' if (pcmk_is_set(then->flags, pcmk__action_optional)) { enum rsc_role_e then_rsc_role; then_rsc_role = then->rsc->priv->fns->state(then->rsc, true); if ((then_rsc_role == pcmk_role_stopped) && pcmk__str_eq(then->task, PCMK_ACTION_STOP, pcmk__str_none)) { /* If 'then' should stop after 'first' but is already stopped, the * ordering is irrelevant. */ return; } else if ((then_rsc_role >= pcmk_role_started) && pcmk__str_eq(then->task, PCMK_ACTION_START, pcmk__str_none) && pe__rsc_running_on_only(then->rsc, then->node)) { /* Similarly if 'then' should start after 'first' but is already * started on a single node. */ return; } } // 'First' can't run, so 'then' can't either clear_action_flag_because(then, pcmk__action_optional, first); clear_action_flag_because(then, pcmk__action_runnable, first); } /*! * \internal * \brief Set action bits appropriately when pcmk__ar_intermediate_stop is used * * \param[in,out] first 'First' action in ordering * \param[in,out] then 'Then' action in ordering * \param[in] filter What action flags to care about * * \note pcmk__ar_intermediate_stop is set for "stop resource before starting * it" and "stop later group member before stopping earlier group member" */ static void handle_restart_ordering(pcmk_action_t *first, pcmk_action_t *then, uint32_t filter) { const char *reason = NULL; pcmk__assert(is_primitive_action(first) && is_primitive_action(then)); // We need to update the action in two cases: // ... if 'then' is required if (pcmk_is_set(filter, pcmk__action_optional) && !pcmk_is_set(then->flags, pcmk__action_optional)) { reason = "restart"; } /* ... if 'then' is unrunnable action on same resource (if a resource * should restart but can't start, we still want to stop) */ if (pcmk_is_set(filter, pcmk__action_runnable) && !pcmk_is_set(then->flags, pcmk__action_runnable) && pcmk_is_set(then->rsc->flags, pcmk__rsc_managed) && (first->rsc == then->rsc)) { reason = "stop"; } if (reason == NULL) { return; } pcmk__rsc_trace(first->rsc, "Handling %s -> %s for %s", first->uuid, then->uuid, reason); // Make 'first' required if it is runnable if (pcmk_is_set(first->flags, pcmk__action_runnable)) { clear_action_flag_because(first, pcmk__action_optional, then); } // Make 'first' required if 'then' is required if (!pcmk_is_set(then->flags, pcmk__action_optional)) { clear_action_flag_because(first, pcmk__action_optional, then); } // Make 'first' unmigratable if 'then' is unmigratable if (!pcmk_is_set(then->flags, pcmk__action_migratable)) { clear_action_flag_because(first, pcmk__action_migratable, then); } // Make 'then' unrunnable if 'first' is required but unrunnable if (!pcmk_is_set(first->flags, pcmk__action_optional) && !pcmk_is_set(first->flags, pcmk__action_runnable)) { clear_action_flag_because(then, pcmk__action_runnable, first); } } /*! * \internal * \brief Update two actions according to an ordering between them * * Given information about an ordering of two actions, update the actions' flags * (and runnable_before members if appropriate) as appropriate for the ordering. * Effects may cascade to other orderings involving the actions as well. * * \param[in,out] first 'First' action in an ordering * \param[in,out] then 'Then' action in an ordering * \param[in] node If not NULL, limit scope of ordering to this node * (ignored) * \param[in] flags Action flags for \p first for ordering purposes * \param[in] filter Action flags to limit scope of certain updates (may * include pcmk__action_optional to affect only * mandatory actions, and pcmk__action_runnable to * affect only runnable actions) * \param[in] type Group of enum pcmk__action_relation_flags to apply * \param[in,out] scheduler Scheduler data * * \return Group of enum pcmk__updated flags indicating what was updated */ uint32_t pcmk__update_ordered_actions(pcmk_action_t *first, pcmk_action_t *then, const pcmk_node_t *node, uint32_t flags, uint32_t filter, uint32_t type, pcmk_scheduler_t *scheduler) { uint32_t changed = pcmk__updated_none; uint32_t then_flags = 0U; uint32_t first_flags = 0U; pcmk__assert((first != NULL) && (then != NULL) && (scheduler != NULL)); then_flags = then->flags; first_flags = first->flags; if (pcmk_is_set(type, pcmk__ar_asymmetric)) { handle_asymmetric_ordering(first, then); } if (pcmk_is_set(type, pcmk__ar_then_implies_first) && !pcmk_is_set(then_flags, pcmk__action_optional)) { // Then is required, and implies first should be, too if (pcmk_is_set(filter, pcmk__action_optional) && !pcmk_is_set(flags, pcmk__action_optional) && pcmk_is_set(first_flags, pcmk__action_optional)) { clear_action_flag_because(first, pcmk__action_optional, then); } if (pcmk_is_set(flags, pcmk__action_migratable) && !pcmk_is_set(then->flags, pcmk__action_migratable)) { clear_action_flag_because(first, pcmk__action_migratable, then); } } if (pcmk_is_set(type, pcmk__ar_promoted_then_implies_first) && (then->rsc != NULL) && (then->rsc->priv->orig_role == pcmk_role_promoted) && pcmk_is_set(filter, pcmk__action_optional) && !pcmk_is_set(then->flags, pcmk__action_optional)) { clear_action_flag_because(first, pcmk__action_optional, then); if (pcmk_is_set(first->flags, pcmk__action_migratable) && !pcmk_is_set(then->flags, pcmk__action_migratable)) { clear_action_flag_because(first, pcmk__action_migratable, then); } } if (pcmk_is_set(type, pcmk__ar_unmigratable_then_blocks) && pcmk_is_set(filter, pcmk__action_optional)) { if (!pcmk_all_flags_set(then->flags, pcmk__action_migratable |pcmk__action_runnable)) { clear_action_flag_because(first, pcmk__action_runnable, then); } if (!pcmk_is_set(then->flags, pcmk__action_optional)) { clear_action_flag_because(first, pcmk__action_optional, then); } } if (pcmk_is_set(type, pcmk__ar_first_else_then) && pcmk_is_set(filter, pcmk__action_optional) && !pcmk_is_set(first->flags, pcmk__action_runnable)) { clear_action_flag_because(then, pcmk__action_migratable, first); pcmk__clear_action_flags(then, pcmk__action_pseudo); } if (pcmk_is_set(type, pcmk__ar_unrunnable_first_blocks) && pcmk_is_set(filter, pcmk__action_runnable) && pcmk_is_set(then->flags, pcmk__action_runnable) && !pcmk_is_set(flags, pcmk__action_runnable)) { clear_action_flag_because(then, pcmk__action_runnable, first); clear_action_flag_because(then, pcmk__action_migratable, first); } if (pcmk_is_set(type, pcmk__ar_first_implies_then) && pcmk_is_set(filter, pcmk__action_optional) && pcmk_is_set(then->flags, pcmk__action_optional) && !pcmk_is_set(flags, pcmk__action_optional) && !pcmk_is_set(first->flags, pcmk__action_migratable)) { clear_action_flag_because(then, pcmk__action_optional, first); } if (pcmk_is_set(type, pcmk__ar_intermediate_stop)) { handle_restart_ordering(first, then, filter); } if (then_flags != then->flags) { pcmk__set_updated_flags(changed, first, pcmk__updated_then); pcmk__rsc_trace(then->rsc, "%s on %s: flags are now %#.6x (was %#.6x) " "because of 'first' %s (%#.6x)", then->uuid, pcmk__node_name(then->node), then->flags, then_flags, first->uuid, first->flags); if ((then->rsc != NULL) && (then->rsc->priv->parent != NULL)) { // Required to handle "X_stop then X_start" for cloned groups pcmk__update_action_for_orderings(then, scheduler); } } if (first_flags != first->flags) { pcmk__set_updated_flags(changed, first, pcmk__updated_first); pcmk__rsc_trace(first->rsc, "%s on %s: flags are now %#.6x (was %#.6x) " "because of 'then' %s (%#.6x)", first->uuid, pcmk__node_name(first->node), first->flags, first_flags, then->uuid, then->flags); } return changed; } /*! * \internal * \brief Trace-log an action (optionally with its dependent actions) * * \param[in] pre_text If not NULL, prefix the log with this plus ": " * \param[in] action Action to log * \param[in] details If true, recursively log dependent actions */ void pcmk__log_action(const char *pre_text, const pcmk_action_t *action, bool details) { const char *node_uname = NULL; const char *node_uuid = NULL; const char *desc = NULL; CRM_CHECK(action != NULL, return); if (!pcmk_is_set(action->flags, pcmk__action_pseudo)) { if (action->node != NULL) { node_uname = action->node->priv->name; node_uuid = action->node->priv->id; } else { node_uname = ""; } } switch (pcmk__parse_action(action->task)) { case pcmk__action_fence: case pcmk__action_shutdown: if (pcmk_is_set(action->flags, pcmk__action_pseudo)) { desc = "Pseudo "; } else if (pcmk_is_set(action->flags, pcmk__action_optional)) { desc = "Optional "; } else if (!pcmk_is_set(action->flags, pcmk__action_runnable)) { desc = "!!Non-Startable!! "; } else { desc = "(Provisional) "; } crm_trace("%s%s%sAction %d: %s%s%s%s%s%s", ((pre_text == NULL)? "" : pre_text), ((pre_text == NULL)? "" : ": "), desc, action->id, action->uuid, (node_uname? "\ton " : ""), (node_uname? node_uname : ""), (node_uuid? "\t\t(" : ""), (node_uuid? node_uuid : ""), (node_uuid? ")" : "")); break; default: if (pcmk_is_set(action->flags, pcmk__action_optional)) { desc = "Optional "; } else if (pcmk_is_set(action->flags, pcmk__action_pseudo)) { desc = "Pseudo "; } else if (!pcmk_is_set(action->flags, pcmk__action_runnable)) { desc = "!!Non-Startable!! "; } else { desc = "(Provisional) "; } crm_trace("%s%s%sAction %d: %s %s%s%s%s%s%s", ((pre_text == NULL)? "" : pre_text), ((pre_text == NULL)? "" : ": "), desc, action->id, action->uuid, (action->rsc? action->rsc->id : ""), (node_uname? "\ton " : ""), (node_uname? node_uname : ""), (node_uuid? "\t\t(" : ""), (node_uuid? node_uuid : ""), (node_uuid? ")" : "")); break; } if (details) { const GList *iter = NULL; const pcmk__related_action_t *other = NULL; crm_trace("\t\t====== Preceding Actions"); for (iter = action->actions_before; iter != NULL; iter = iter->next) { other = (const pcmk__related_action_t *) iter->data; pcmk__log_action("\t\t", other->action, false); } crm_trace("\t\t====== Subsequent Actions"); for (iter = action->actions_after; iter != NULL; iter = iter->next) { other = (const pcmk__related_action_t *) iter->data; pcmk__log_action("\t\t", other->action, false); } crm_trace("\t\t====== End"); } else { crm_trace("\t\t(before=%d, after=%d)", g_list_length(action->actions_before), g_list_length(action->actions_after)); } } /*! * \internal * \brief Create a new shutdown action for a node * * \param[in,out] node Node being shut down * * \return Newly created shutdown action for \p node */ pcmk_action_t * pcmk__new_shutdown_action(pcmk_node_t *node) { char *shutdown_id = NULL; pcmk_action_t *shutdown_op = NULL; pcmk__assert(node != NULL); shutdown_id = crm_strdup_printf("%s-%s", PCMK_ACTION_DO_SHUTDOWN, node->priv->name); shutdown_op = custom_action(NULL, shutdown_id, PCMK_ACTION_DO_SHUTDOWN, node, FALSE, node->priv->scheduler); pcmk__order_stops_before_shutdown(node, shutdown_op); pcmk__insert_meta(shutdown_op, PCMK__META_OP_NO_WAIT, PCMK_VALUE_TRUE); return shutdown_op; } /*! * \internal * \brief Calculate and add an operation digest to XML * * Calculate an operation digest, which enables us to later determine when a * restart is needed due to the resource's parameters being changed, and add it * to given XML. * * \param[in] op Operation result from executor * \param[in,out] update XML to add digest to */ static void add_op_digest_to_xml(const lrmd_event_data_t *op, xmlNode *update) { char *digest = NULL; xmlNode *args_xml = NULL; if (op->params == NULL) { return; } args_xml = pcmk__xe_create(NULL, PCMK_XE_PARAMETERS); g_hash_table_foreach(op->params, hash2field, args_xml); pcmk__filter_op_for_digest(args_xml); digest = pcmk__digest_operation(args_xml); crm_xml_add(update, PCMK__XA_OP_DIGEST, digest); pcmk__xml_free(args_xml); free(digest); } #define FAKE_TE_ID "xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx" /*! * \internal * \brief Create XML for resource operation history update * * \param[in,out] parent Parent XML node to add to * \param[in,out] op Operation event data * \param[in] caller_version DC feature set * \param[in] target_rc Expected result of operation * \param[in] node Name of node on which operation was performed * \param[in] origin Arbitrary description of update source * * \return Newly created XML node for history update */ xmlNode * pcmk__create_history_xml(xmlNode *parent, lrmd_event_data_t *op, const char *caller_version, int target_rc, const char *node, const char *origin) { char *key = NULL; char *magic = NULL; char *op_id = NULL; char *op_id_additional = NULL; char *local_user_data = NULL; const char *exit_reason = NULL; xmlNode *xml_op = NULL; const char *task = NULL; CRM_CHECK(op != NULL, return NULL); crm_trace("Creating history XML for %s-interval %s action for %s on %s " "(DC version: %s, origin: %s)", pcmk__readable_interval(op->interval_ms), op->op_type, op->rsc_id, ((node == NULL)? "no node" : node), caller_version, origin); task = op->op_type; /* Record a successful agent reload as a start, and a failed one as a * monitor, to make life easier for the scheduler when determining the * current state. * * @COMPAT We should check "reload" here only if the operation was for a * pre-OCF-1.1 resource agent, but we don't know that here, and we should * only ever get results for actions scheduled by us, so we can reasonably * assume any "reload" is actually a pre-1.1 agent reload. * * @TODO This remapping can make log messages with task confusing for users * (for example, an "Initiating reload ..." followed by "... start ... * confirmed"). Either do this remapping in the scheduler if possible, or * store the original task in a new XML attribute for later logging. */ if (pcmk__str_any_of(task, PCMK_ACTION_RELOAD, PCMK_ACTION_RELOAD_AGENT, NULL)) { if (op->op_status == PCMK_EXEC_DONE) { task = PCMK_ACTION_START; } else { task = PCMK_ACTION_MONITOR; } } key = pcmk__op_key(op->rsc_id, task, op->interval_ms); if (pcmk__str_eq(task, PCMK_ACTION_NOTIFY, pcmk__str_none)) { const char *n_type = crm_meta_value(op->params, "notify_type"); const char *n_task = crm_meta_value(op->params, "notify_operation"); CRM_LOG_ASSERT(n_type != NULL); CRM_LOG_ASSERT(n_task != NULL); op_id = pcmk__notify_key(op->rsc_id, n_type, n_task); if (op->op_status != PCMK_EXEC_PENDING) { /* Ignore notify errors. * * @TODO It might be better to keep the correct result here, and * ignore it in process_graph_event(). */ lrmd__set_result(op, PCMK_OCF_OK, PCMK_EXEC_DONE, NULL); } /* Migration history is preserved separately, which usually matters for * multiple nodes and is important for future cluster transitions. */ } else if (pcmk__str_any_of(op->op_type, PCMK_ACTION_MIGRATE_TO, PCMK_ACTION_MIGRATE_FROM, NULL)) { op_id = strdup(key); } else if (did_rsc_op_fail(op, target_rc)) { op_id = pcmk__op_key(op->rsc_id, "last_failure", 0); if (op->interval_ms == 0) { /* Ensure 'last' gets updated, in case PCMK_META_RECORD_PENDING is * true */ op_id_additional = pcmk__op_key(op->rsc_id, "last", 0); } exit_reason = op->exit_reason; } else if (op->interval_ms > 0) { op_id = strdup(key); } else { op_id = pcmk__op_key(op->rsc_id, "last", 0); } again: xml_op = pcmk__xe_first_child(parent, PCMK__XE_LRM_RSC_OP, PCMK_XA_ID, op_id); if (xml_op == NULL) { xml_op = pcmk__xe_create(parent, PCMK__XE_LRM_RSC_OP); } if (op->user_data == NULL) { crm_debug("Generating fake transition key for: " PCMK__OP_FMT " %d from %s", op->rsc_id, op->op_type, op->interval_ms, op->call_id, origin); local_user_data = pcmk__transition_key(-1, op->call_id, target_rc, FAKE_TE_ID); op->user_data = local_user_data; } if (magic == NULL) { magic = crm_strdup_printf("%d:%d;%s", op->op_status, op->rc, (const char *) op->user_data); } crm_xml_add(xml_op, PCMK_XA_ID, op_id); crm_xml_add(xml_op, PCMK__XA_OPERATION_KEY, key); crm_xml_add(xml_op, PCMK_XA_OPERATION, task); crm_xml_add(xml_op, PCMK_XA_CRM_DEBUG_ORIGIN, origin); crm_xml_add(xml_op, PCMK_XA_CRM_FEATURE_SET, caller_version); crm_xml_add(xml_op, PCMK__XA_TRANSITION_KEY, op->user_data); crm_xml_add(xml_op, PCMK__XA_TRANSITION_MAGIC, magic); crm_xml_add(xml_op, PCMK_XA_EXIT_REASON, pcmk__s(exit_reason, "")); crm_xml_add(xml_op, PCMK__META_ON_NODE, node); // For context during triage crm_xml_add_int(xml_op, PCMK__XA_CALL_ID, op->call_id); crm_xml_add_int(xml_op, PCMK__XA_RC_CODE, op->rc); crm_xml_add_int(xml_op, PCMK__XA_OP_STATUS, op->op_status); - crm_xml_add_ms(xml_op, PCMK_META_INTERVAL, op->interval_ms); + pcmk__xe_set_guint(xml_op, PCMK_META_INTERVAL, op->interval_ms); if ((op->t_run > 0) || (op->t_rcchange > 0) || (op->exec_time > 0) || (op->queue_time > 0)) { crm_trace("Timing data (" PCMK__OP_FMT "): " "last=%lld change=%lld exec=%u queue=%u", op->rsc_id, op->op_type, op->interval_ms, (long long) op->t_run, (long long) op->t_rcchange, op->exec_time, op->queue_time); if ((op->interval_ms > 0) && (op->t_rcchange > 0)) { // Recurring ops may have changed rc after initial run pcmk__xe_set_time(xml_op, PCMK_XA_LAST_RC_CHANGE, op->t_rcchange); } else { pcmk__xe_set_time(xml_op, PCMK_XA_LAST_RC_CHANGE, op->t_run); } crm_xml_add_int(xml_op, PCMK_XA_EXEC_TIME, op->exec_time); crm_xml_add_int(xml_op, PCMK_XA_QUEUE_TIME, op->queue_time); } if (pcmk__str_any_of(op->op_type, PCMK_ACTION_MIGRATE_TO, PCMK_ACTION_MIGRATE_FROM, NULL)) { /* Record PCMK__META_MIGRATE_SOURCE and PCMK__META_MIGRATE_TARGET always * for migrate ops. */ const char *name = PCMK__META_MIGRATE_SOURCE; crm_xml_add(xml_op, name, crm_meta_value(op->params, name)); name = PCMK__META_MIGRATE_TARGET; crm_xml_add(xml_op, name, crm_meta_value(op->params, name)); } add_op_digest_to_xml(op, xml_op); if (op_id_additional) { free(op_id); op_id = op_id_additional; op_id_additional = NULL; goto again; } if (local_user_data) { free(local_user_data); op->user_data = NULL; } free(magic); free(op_id); free(key); return xml_op; } /*! * \internal * \brief Check whether an action shutdown-locks a resource to a node * * If the PCMK_OPT_SHUTDOWN_LOCK cluster property is set, resources will not be * recovered on a different node if cleanly stopped, and may start only on that * same node. This function checks whether that applies to a given action, so * that the transition graph can be marked appropriately. * * \param[in] action Action to check * * \return true if \p action locks its resource to the action's node, * otherwise false */ bool pcmk__action_locks_rsc_to_node(const pcmk_action_t *action) { // Only resource actions taking place on resource's lock node are locked if ((action == NULL) || (action->rsc == NULL) || !pcmk__same_node(action->node, action->rsc->priv->lock_node)) { return false; } /* During shutdown, only stops are locked (otherwise, another action such as * a demote would cause the controller to clear the lock) */ if (action->node->details->shutdown && (action->task != NULL) && (strcmp(action->task, PCMK_ACTION_STOP) != 0)) { return false; } return true; } /* lowest to highest */ static gint sort_action_id(gconstpointer a, gconstpointer b) { const pcmk__related_action_t *action_wrapper2 = a; const pcmk__related_action_t *action_wrapper1 = b; if (a == NULL) { return 1; } if (b == NULL) { return -1; } if (action_wrapper1->action->id < action_wrapper2->action->id) { return 1; } if (action_wrapper1->action->id > action_wrapper2->action->id) { return -1; } return 0; } /*! * \internal * \brief Remove any duplicate action inputs, merging action flags * * \param[in,out] action Action whose inputs should be checked */ void pcmk__deduplicate_action_inputs(pcmk_action_t *action) { GList *item = NULL; GList *next = NULL; pcmk__related_action_t *last_input = NULL; action->actions_before = g_list_sort(action->actions_before, sort_action_id); for (item = action->actions_before; item != NULL; item = next) { pcmk__related_action_t *input = item->data; next = item->next; if ((last_input != NULL) && (input->action->id == last_input->action->id)) { crm_trace("Input %s (%d) duplicate skipped for action %s (%d)", input->action->uuid, input->action->id, action->uuid, action->id); /* For the purposes of scheduling, the ordering flags no longer * matter, but crm_simulate looks at certain ones when creating a * dot graph. Combining the flags is sufficient for that purpose. */ pcmk__set_relation_flags(last_input->flags, input->flags); if (input->graphed) { last_input->graphed = true; } free(item->data); action->actions_before = g_list_delete_link(action->actions_before, item); } else { last_input = input; input->graphed = false; } } } /*! * \internal * \brief Output all scheduled actions * * \param[in,out] scheduler Scheduler data */ void pcmk__output_actions(pcmk_scheduler_t *scheduler) { pcmk__output_t *out = scheduler->priv->out; // Output node (non-resource) actions for (GList *iter = scheduler->priv->actions; iter != NULL; iter = iter->next) { char *node_name = NULL; char *task = NULL; pcmk_action_t *action = (pcmk_action_t *) iter->data; if (action->rsc != NULL) { continue; // Resource actions will be output later } else if (pcmk_is_set(action->flags, pcmk__action_optional)) { continue; // This action was not scheduled } if (pcmk__str_eq(action->task, PCMK_ACTION_DO_SHUTDOWN, pcmk__str_none)) { task = strdup("Shutdown"); } else if (pcmk__str_eq(action->task, PCMK_ACTION_STONITH, pcmk__str_none)) { const char *op = g_hash_table_lookup(action->meta, PCMK__META_STONITH_ACTION); task = crm_strdup_printf("Fence (%s)", op); } else { continue; // Don't display other node action types } if (pcmk__is_guest_or_bundle_node(action->node)) { const pcmk_resource_t *remote = action->node->priv->remote; node_name = crm_strdup_printf("%s (resource: %s)", pcmk__node_name(action->node), remote->priv->launcher->id); } else if (action->node != NULL) { node_name = crm_strdup_printf("%s", pcmk__node_name(action->node)); } out->message(out, "node-action", task, node_name, action->reason); free(node_name); free(task); } // Output resource actions for (GList *iter = scheduler->priv->resources; iter != NULL; iter = iter->next) { pcmk_resource_t *rsc = (pcmk_resource_t *) iter->data; rsc->priv->cmds->output_actions(rsc); } } /*! * \internal * \brief Get action name needed to compare digest for configuration changes * * \param[in] task Action name from history * \param[in] interval_ms Action interval (in milliseconds) * * \return Action name whose digest should be compared */ static const char * task_for_digest(const char *task, guint interval_ms) { /* Certain actions need to be compared against the parameters used to start * the resource. */ if ((interval_ms == 0) && pcmk__str_any_of(task, PCMK_ACTION_MONITOR, PCMK_ACTION_MIGRATE_FROM, PCMK_ACTION_PROMOTE, NULL)) { task = PCMK_ACTION_START; } return task; } /*! * \internal * \brief Check whether only sanitized parameters to an action changed * * When collecting CIB files for troubleshooting, crm_report will mask * sensitive resource parameters. If simulations were run using that, affected * resources would appear to need a restart, which would complicate * troubleshooting. To avoid that, we save a "secure digest" of non-sensitive * parameters. This function used that digest to check whether only masked * parameters are different. * * \param[in] xml_op Resource history entry with secure digest * \param[in] digest_data Operation digest information being compared * \param[in] scheduler Scheduler data * * \return true if only sanitized parameters changed, otherwise false */ static bool only_sanitized_changed(const xmlNode *xml_op, const pcmk__op_digest_t *digest_data, const pcmk_scheduler_t *scheduler) { const char *digest_secure = NULL; if (!pcmk_is_set(scheduler->flags, pcmk__sched_sanitized)) { // The scheduler is not being run as a simulation return false; } digest_secure = pcmk__xe_get(xml_op, PCMK__XA_OP_SECURE_DIGEST); return (digest_data->rc != pcmk__digest_match) && (digest_secure != NULL) && (digest_data->digest_secure_calc != NULL) && (strcmp(digest_data->digest_secure_calc, digest_secure) == 0); } /*! * \internal * \brief Force a restart due to a configuration change * * \param[in,out] rsc Resource that action is for * \param[in] task Name of action whose configuration changed * \param[in] interval_ms Action interval (in milliseconds) * \param[in,out] node Node where resource should be restarted */ static void force_restart(pcmk_resource_t *rsc, const char *task, guint interval_ms, pcmk_node_t *node) { char *key = pcmk__op_key(rsc->id, task, interval_ms); pcmk_action_t *required = custom_action(rsc, key, task, NULL, FALSE, rsc->priv->scheduler); pe_action_set_reason(required, "resource definition change", true); trigger_unfencing(rsc, node, "Device parameters changed", NULL, rsc->priv->scheduler); } /*! * \internal * \brief Schedule a reload of a resource on a node * * \param[in,out] data Resource to reload * \param[in] user_data Where resource should be reloaded */ static void schedule_reload(gpointer data, gpointer user_data) { pcmk_resource_t *rsc = data; const pcmk_node_t *node = user_data; pcmk_action_t *reload = NULL; // For collective resources, just call recursively for children if (rsc->priv->variant > pcmk__rsc_variant_primitive) { g_list_foreach(rsc->priv->children, schedule_reload, user_data); return; } // Skip the reload in certain situations if ((node == NULL) || !pcmk_is_set(rsc->flags, pcmk__rsc_managed) || pcmk_is_set(rsc->flags, pcmk__rsc_failed)) { pcmk__rsc_trace(rsc, "Skip reload of %s:%s%s %s", rsc->id, pcmk_is_set(rsc->flags, pcmk__rsc_managed)? "" : " unmanaged", pcmk_is_set(rsc->flags, pcmk__rsc_failed)? " failed" : "", (node == NULL)? "inactive" : node->priv->name); return; } /* If a resource's configuration changed while a start was pending, * force a full restart instead of a reload. */ if (pcmk_is_set(rsc->flags, pcmk__rsc_start_pending)) { pcmk__rsc_trace(rsc, "%s: preventing agent reload because start pending", rsc->id); custom_action(rsc, stop_key(rsc), PCMK_ACTION_STOP, node, FALSE, rsc->priv->scheduler); return; } // Schedule the reload pcmk__set_rsc_flags(rsc, pcmk__rsc_reload); reload = custom_action(rsc, reload_key(rsc), PCMK_ACTION_RELOAD_AGENT, node, FALSE, rsc->priv->scheduler); pe_action_set_reason(reload, "resource definition change", FALSE); // Set orderings so that a required stop or demote cancels the reload pcmk__new_ordering(NULL, NULL, reload, rsc, stop_key(rsc), NULL, pcmk__ar_ordered|pcmk__ar_then_cancels_first, rsc->priv->scheduler); pcmk__new_ordering(NULL, NULL, reload, rsc, demote_key(rsc), NULL, pcmk__ar_ordered|pcmk__ar_then_cancels_first, rsc->priv->scheduler); } /*! * \internal * \brief Handle any configuration change for an action * * Given an action from resource history, if the resource's configuration * changed since the action was done, schedule any actions needed (restart, * reload, unfencing, rescheduling recurring actions, etc.). * * \param[in,out] rsc Resource that action is for * \param[in,out] node Node that action was on * \param[in] xml_op Action XML from resource history * * \return true if action configuration changed, otherwise false */ bool pcmk__check_action_config(pcmk_resource_t *rsc, pcmk_node_t *node, const xmlNode *xml_op) { guint interval_ms = 0; const char *task = NULL; const pcmk__op_digest_t *digest_data = NULL; CRM_CHECK((rsc != NULL) && (node != NULL) && (xml_op != NULL), return false); task = pcmk__xe_get(xml_op, PCMK_XA_OPERATION); CRM_CHECK(task != NULL, return false); pcmk__xe_get_guint(xml_op, PCMK_META_INTERVAL, &interval_ms); // If this is a recurring action, check whether it has been orphaned if (interval_ms > 0) { if (pcmk__find_action_config(rsc, task, interval_ms, false) != NULL) { pcmk__rsc_trace(rsc, "%s-interval %s for %s on %s is in configuration", pcmk__readable_interval(interval_ms), task, rsc->id, pcmk__node_name(node)); } else if (pcmk_is_set(rsc->priv->scheduler->flags, pcmk__sched_cancel_removed_actions)) { pcmk__schedule_cancel(rsc, pcmk__xe_get(xml_op, PCMK__XA_CALL_ID), task, interval_ms, node, "orphan"); return true; } else { pcmk__rsc_debug(rsc, "%s-interval %s for %s on %s is orphaned", pcmk__readable_interval(interval_ms), task, rsc->id, pcmk__node_name(node)); return true; } } crm_trace("Checking %s-interval %s for %s on %s for configuration changes", pcmk__readable_interval(interval_ms), task, rsc->id, pcmk__node_name(node)); task = task_for_digest(task, interval_ms); digest_data = rsc_action_digest_cmp(rsc, xml_op, node, rsc->priv->scheduler); if (only_sanitized_changed(xml_op, digest_data, rsc->priv->scheduler)) { if (!pcmk__is_daemon && (rsc->priv->scheduler->priv->out != NULL)) { pcmk__output_t *out = rsc->priv->scheduler->priv->out; out->info(out, "Only 'private' parameters to %s-interval %s for %s " "on %s changed: %s", pcmk__readable_interval(interval_ms), task, rsc->id, pcmk__node_name(node), pcmk__xe_get(xml_op, PCMK__XA_TRANSITION_MAGIC)); } return false; } switch (digest_data->rc) { case pcmk__digest_restart: crm_log_xml_debug(digest_data->params_restart, "params:restart"); force_restart(rsc, task, interval_ms, node); return true; case pcmk__digest_unknown: case pcmk__digest_mismatch: // Changes that can potentially be handled by an agent reload if (interval_ms > 0) { /* Recurring actions aren't reloaded per se, they are just * re-scheduled so the next run uses the new parameters. * The old instance will be cancelled automatically. */ crm_log_xml_debug(digest_data->params_all, "params:reschedule"); pcmk__reschedule_recurring(rsc, task, interval_ms, node); } else if (pcmk__xe_get(xml_op, PCMK__XA_OP_RESTART_DIGEST) != NULL) { // Agent supports reload, so use it trigger_unfencing(rsc, node, "Device parameters changed (reload)", NULL, rsc->priv->scheduler); crm_log_xml_debug(digest_data->params_all, "params:reload"); schedule_reload((gpointer) rsc, (gpointer) node); } else { pcmk__rsc_trace(rsc, "Restarting %s " "because agent doesn't support reload", rsc->id); crm_log_xml_debug(digest_data->params_restart, "params:restart"); force_restart(rsc, task, interval_ms, node); } return true; default: break; } return false; } /*! * \internal * \brief Create a list of resource's action history entries, sorted by call ID * * \param[in] rsc_entry Resource's \c PCMK__XE_LRM_RSC_OP status XML * \param[out] start_index Where to store index of start-like action, if any * \param[out] stop_index Where to store index of stop action, if any */ static GList * rsc_history_as_list(const xmlNode *rsc_entry, int *start_index, int *stop_index) { GList *ops = NULL; for (xmlNode *rsc_op = pcmk__xe_first_child(rsc_entry, PCMK__XE_LRM_RSC_OP, NULL, NULL); rsc_op != NULL; rsc_op = pcmk__xe_next(rsc_op, PCMK__XE_LRM_RSC_OP)) { ops = g_list_prepend(ops, rsc_op); } ops = g_list_sort(ops, sort_op_by_callid); calculate_active_ops(ops, start_index, stop_index); return ops; } /*! * \internal * \brief Process a resource's action history from the CIB status * * Given a resource's action history, if the resource's configuration * changed since the actions were done, schedule any actions needed (restart, * reload, unfencing, rescheduling recurring actions, clean-up, etc.). * (This also cancels recurring actions for maintenance mode, which is not * entirely related but convenient to do here.) * * \param[in] rsc_entry Resource's \c PCMK__XE_LRM_RSC_OP status XML * \param[in,out] rsc Resource whose history is being processed * \param[in,out] node Node whose history is being processed */ static void process_rsc_history(const xmlNode *rsc_entry, pcmk_resource_t *rsc, pcmk_node_t *node) { int offset = -1; int stop_index = 0; int start_index = 0; GList *sorted_op_list = NULL; if (pcmk_is_set(rsc->flags, pcmk__rsc_removed)) { if (pcmk__is_anonymous_clone(pe__const_top_resource(rsc, false))) { /* @TODO Should this be done for bundled primitives as well? Added * by 2ac43ae31 */ pcmk__rsc_trace(rsc, "Skipping configuration check " "for orphaned clone instance %s", rsc->id); } else { pcmk__rsc_trace(rsc, "Skipping configuration check and scheduling " "clean-up for orphaned resource %s", rsc->id); pcmk__schedule_cleanup(rsc, node, false); } return; } if (pe_find_node_id(rsc->priv->active_nodes, node->priv->id) == NULL) { if (pcmk__rsc_agent_changed(rsc, node, rsc_entry, false)) { pcmk__schedule_cleanup(rsc, node, false); } pcmk__rsc_trace(rsc, "Skipping configuration check for %s " "because no longer active on %s", rsc->id, pcmk__node_name(node)); return; } pcmk__rsc_trace(rsc, "Checking for configuration changes for %s on %s", rsc->id, pcmk__node_name(node)); if (pcmk__rsc_agent_changed(rsc, node, rsc_entry, true)) { pcmk__schedule_cleanup(rsc, node, false); } sorted_op_list = rsc_history_as_list(rsc_entry, &start_index, &stop_index); if (start_index < stop_index) { return; // Resource is stopped } for (GList *iter = sorted_op_list; iter != NULL; iter = iter->next) { xmlNode *rsc_op = (xmlNode *) iter->data; const char *task = NULL; guint interval_ms = 0; if (++offset < start_index) { // Skip actions that happened before a start continue; } task = pcmk__xe_get(rsc_op, PCMK_XA_OPERATION); pcmk__xe_get_guint(rsc_op, PCMK_META_INTERVAL, &interval_ms); if ((interval_ms > 0) && (pcmk_is_set(rsc->flags, pcmk__rsc_maintenance) || node->details->maintenance)) { // Maintenance mode cancels recurring operations pcmk__schedule_cancel(rsc, pcmk__xe_get(rsc_op, PCMK__XA_CALL_ID), task, interval_ms, node, "maintenance mode"); } else if ((interval_ms > 0) || pcmk__strcase_any_of(task, PCMK_ACTION_MONITOR, PCMK_ACTION_START, PCMK_ACTION_PROMOTE, PCMK_ACTION_MIGRATE_FROM, NULL)) { /* If a resource operation failed, and the operation's definition * has changed, clear any fail count so they can be retried fresh. */ if (pe__bundle_needs_remote_name(rsc)) { /* We haven't assigned resources to nodes yet, so if the * REMOTE_CONTAINER_HACK is used, we may calculate the digest * based on the literal "#uname" value rather than the properly * substituted value. That would mistakenly make the action * definition appear to have been changed. Defer the check until * later in this case. */ pcmk__add_param_check(rsc_op, rsc, node, pcmk__check_active); } else if (pcmk__check_action_config(rsc, node, rsc_op) && (pe_get_failcount(node, rsc, NULL, pcmk__fc_effective, NULL) != 0)) { pe__clear_failcount(rsc, node, "action definition changed", rsc->priv->scheduler); } } } g_list_free(sorted_op_list); } /*! * \internal * \brief Process a node's action history from the CIB status * * Given a node's resource history, if the resource's configuration changed * since the actions were done, schedule any actions needed (restart, * reload, unfencing, rescheduling recurring actions, clean-up, etc.). * (This also cancels recurring actions for maintenance mode, which is not * entirely related but convenient to do here.) * * \param[in,out] node Node whose history is being processed * \param[in] lrm_rscs Node's \c PCMK__XE_LRM_RESOURCES from CIB status XML */ static void process_node_history(pcmk_node_t *node, const xmlNode *lrm_rscs) { crm_trace("Processing node history for %s", pcmk__node_name(node)); for (const xmlNode *rsc_entry = pcmk__xe_first_child(lrm_rscs, PCMK__XE_LRM_RESOURCE, NULL, NULL); rsc_entry != NULL; rsc_entry = pcmk__xe_next(rsc_entry, PCMK__XE_LRM_RESOURCE)) { if (rsc_entry->children != NULL) { GList *result = pcmk__rscs_matching_id(pcmk__xe_id(rsc_entry), node->priv->scheduler); for (GList *iter = result; iter != NULL; iter = iter->next) { pcmk_resource_t *rsc = (pcmk_resource_t *) iter->data; if (pcmk__is_primitive(rsc)) { process_rsc_history(rsc_entry, rsc, node); } } g_list_free(result); } } } // XPath to find a node's resource history #define XPATH_NODE_HISTORY "/" PCMK_XE_CIB "/" PCMK_XE_STATUS \ "/" PCMK__XE_NODE_STATE \ "[@" PCMK_XA_UNAME "='%s']" \ "/" PCMK__XE_LRM "/" PCMK__XE_LRM_RESOURCES /*! * \internal * \brief Process any resource configuration changes in the CIB status * * Go through all nodes' resource history, and if a resource's configuration * changed since its actions were done, schedule any actions needed (restart, * reload, unfencing, rescheduling recurring actions, clean-up, etc.). * (This also cancels recurring actions for maintenance mode, which is not * entirely related but convenient to do here.) * * \param[in,out] scheduler Scheduler data */ void pcmk__handle_rsc_config_changes(pcmk_scheduler_t *scheduler) { crm_trace("Check resource and action configuration for changes"); /* Rather than iterate through the status section, iterate through the nodes * and search for the appropriate status subsection for each. This skips * orphaned nodes and lets us eliminate some cases before searching the XML. */ for (GList *iter = scheduler->nodes; iter != NULL; iter = iter->next) { pcmk_node_t *node = (pcmk_node_t *) iter->data; /* Don't bother checking actions for a node that can't run actions ... * unless it's in maintenance mode, in which case we still need to * cancel any existing recurring monitors. */ if (node->details->maintenance || pcmk__node_available(node, false, false)) { char *xpath = NULL; xmlNode *history = NULL; xpath = crm_strdup_printf(XPATH_NODE_HISTORY, node->priv->name); history = pcmk__xpath_find_one(scheduler->input->doc, xpath, LOG_NEVER); free(xpath); process_node_history(node, history); } } }