diff --git a/daemons/execd/execd_commands.c b/daemons/execd/execd_commands.c index 1718dced41..68253a5c36 100644 --- a/daemons/execd/execd_commands.c +++ b/daemons/execd/execd_commands.c @@ -1,1684 +1,1695 @@ /* * Copyright 2012-2018 David Vossel * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include #include "pacemaker-execd.h" #ifdef HAVE_SYS_TIMEB_H # include #endif #define EXIT_REASON_MAX_LEN 128 GHashTable *rsc_list = NULL; typedef struct lrmd_cmd_s { int timeout; guint interval_ms; int start_delay; int timeout_orig; int call_id; int exec_rc; int lrmd_op_status; 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 *exit_reason; char *output; char *userdata_str; #ifdef HAVE_SYS_TIMEB_H /* 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 timeb t_first_run; /* Timestamp of when op first ran */ struct timeb t_run; /* Timestamp of when op most recently ran */ struct timeb t_first_queue; /* Timestamp of when op first was queued */ struct timeb t_queue; /* Timestamp of when op most recently was queued */ struct timeb t_rcchange; /* Timestamp of last rc change */ #endif int first_notify_sent; int last_notify_rc; int last_notify_op_status; int last_pid; GHashTable *params; } lrmd_cmd_t; static void cmd_finalize(lrmd_cmd_t * cmd, lrmd_rsc_t * rsc); static gboolean lrmd_rsc_dispatch(gpointer user_data); static void cancel_all_recurring(lrmd_rsc_t * rsc, const char *client_id); static void log_finished(lrmd_cmd_t * cmd, int exec_time, int queue_time) { char pid_str[32] = { 0, }; int log_level = LOG_INFO; if (cmd->last_pid) { snprintf(pid_str, 32, "%d", cmd->last_pid); } if (safe_str_eq(cmd->action, "monitor")) { log_level = LOG_DEBUG; } #ifdef HAVE_SYS_TIMEB_H do_crm_log(log_level, "finished - rsc:%s action:%s call_id:%d %s%s exit-code:%d exec-time:%dms queue-time:%dms", cmd->rsc_id, cmd->action, cmd->call_id, cmd->last_pid ? "pid:" : "", pid_str, cmd->exec_rc, exec_time, queue_time); #else do_crm_log(log_level, "finished - rsc:%s action:%s call_id:%d %s%s exit-code:%d", cmd->rsc_id, cmd->action, cmd->call_id, cmd->last_pid ? "pid:" : "", pid_str, cmd->exec_rc); #endif } static void log_execute(lrmd_cmd_t * cmd) { int log_level = LOG_INFO; if (safe_str_eq(cmd->action, "monitor")) { 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 (safe_str_eq(action, "monitor") && is_set(pcmk_get_ra_caps(rsc->class), pcmk_ra_cap_status)) { return "status"; } return action; } static lrmd_rsc_t * build_rsc_from_xml(xmlNode * msg) { xmlNode *rsc_xml = get_xpath_object("//" F_LRMD_RSC, msg, LOG_ERR); lrmd_rsc_t *rsc = NULL; rsc = calloc(1, sizeof(lrmd_rsc_t)); crm_element_value_int(msg, F_LRMD_CALLOPTS, &rsc->call_opts); rsc->rsc_id = crm_element_value_copy(rsc_xml, F_LRMD_RSC_ID); rsc->class = crm_element_value_copy(rsc_xml, F_LRMD_CLASS); rsc->provider = crm_element_value_copy(rsc_xml, F_LRMD_PROVIDER); rsc->type = crm_element_value_copy(rsc_xml, F_LRMD_TYPE); rsc->work = mainloop_add_trigger(G_PRIORITY_HIGH, lrmd_rsc_dispatch, rsc); return rsc; } static lrmd_cmd_t * create_lrmd_cmd(xmlNode * msg, crm_client_t * client) { int call_options = 0; xmlNode *rsc_xml = get_xpath_object("//" F_LRMD_RSC, msg, LOG_ERR); lrmd_cmd_t *cmd = NULL; cmd = calloc(1, sizeof(lrmd_cmd_t)); crm_element_value_int(msg, F_LRMD_CALLOPTS, &call_options); cmd->call_opts = call_options; cmd->client_id = strdup(client->id); crm_element_value_int(msg, F_LRMD_CALLID, &cmd->call_id); crm_element_value_ms(rsc_xml, F_LRMD_RSC_INTERVAL, &cmd->interval_ms); crm_element_value_int(rsc_xml, F_LRMD_TIMEOUT, &cmd->timeout); crm_element_value_int(rsc_xml, F_LRMD_RSC_START_DELAY, &cmd->start_delay); cmd->timeout_orig = cmd->timeout; cmd->origin = crm_element_value_copy(rsc_xml, F_LRMD_ORIGIN); cmd->action = crm_element_value_copy(rsc_xml, F_LRMD_RSC_ACTION); cmd->userdata_str = crm_element_value_copy(rsc_xml, F_LRMD_RSC_USERDATA_STR); cmd->rsc_id = crm_element_value_copy(rsc_xml, F_LRMD_RSC_ID); cmd->params = xml2list(rsc_xml); if (safe_str_eq(g_hash_table_lookup(cmd->params, "CRM_meta_on_fail"), "block")) { crm_debug("Setting flag to leave pid group on timeout and only kill action pid for " CRM_OP_FMT, cmd->rsc_id, cmd->action, cmd->interval_ms); cmd->service_flags |= SVC_ACTION_LEAVE_GROUP; } return cmd; } static void free_lrmd_cmd(lrmd_cmd_t * cmd) { if (cmd->stonith_recurring_id) { g_source_remove(cmd->stonith_recurring_id); } if (cmd->delay_id) { g_source_remove(cmd->delay_id); } if (cmd->params) { g_hash_table_destroy(cmd->params); } free(cmd->origin); free(cmd->action); free(cmd->real_action); free(cmd->userdata_str); free(cmd->rsc_id); free(cmd->output); free(cmd->exit_reason); free(cmd->client_id); free(cmd); } static gboolean stonith_recurring_op_helper(gpointer data) { lrmd_cmd_t *cmd = data; lrmd_rsc_t *rsc; cmd->stonith_recurring_id = 0; if (!cmd->rsc_id) { return FALSE; } rsc = g_hash_table_lookup(rsc_list, cmd->rsc_id); CRM_ASSERT(rsc != NULL); /* take it out of recurring_ops list, and put it in the pending ops * to be executed */ rsc->recurring_ops = g_list_remove(rsc->recurring_ops, cmd); rsc->pending_ops = g_list_append(rsc->pending_ops, cmd); #ifdef HAVE_SYS_TIMEB_H ftime(&cmd->t_queue); if (cmd->t_first_queue.time == 0) { cmd->t_first_queue = cmd->t_queue; } #endif mainloop_set_trigger(rsc->work); return FALSE; } 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; } static gboolean merge_recurring_duplicate(lrmd_rsc_t * rsc, lrmd_cmd_t * cmd) { GListPtr gIter = NULL; lrmd_cmd_t * dup = NULL; gboolean dup_pending = FALSE; if (cmd->interval_ms == 0) { return 0; } for (gIter = rsc->pending_ops; gIter != NULL; gIter = gIter->next) { dup = gIter->data; if (safe_str_eq(cmd->action, dup->action) && (cmd->interval_ms == dup->interval_ms)) { dup_pending = TRUE; goto merge_dup; } } /* if dup is in recurring_ops list, that means it has already executed * and is in the interval loop. we can't just remove it in this case. */ for (gIter = rsc->recurring_ops; gIter != NULL; gIter = gIter->next) { dup = gIter->data; if (safe_str_eq(cmd->action, dup->action) && (cmd->interval_ms == dup->interval_ms)) { goto merge_dup; } } return FALSE; merge_dup: /* 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 (" CRM_OP_FMT "), merging with previous op entry", rsc->rsc_id, normalize_action_name(rsc, dup->action), dup->interval_ms); /* merge */ dup->first_notify_sent = 0; free(dup->userdata_str); dup->userdata_str = cmd->userdata_str; cmd->userdata_str = NULL; dup->call_id = cmd->call_id; if (safe_str_eq(rsc->class, PCMK_RESOURCE_CLASS_STONITH)) { /* if we are waiting for the next interval, kick it off now */ if (dup_pending == TRUE) { g_source_remove(cmd->stonith_recurring_id); cmd->stonith_recurring_id = 0; stonith_recurring_op_helper(cmd); } } else if (dup_pending == FALSE) { /* if we've already handed this to the service lib, kick off an early execution */ services_action_kick(rsc->rsc_id, normalize_action_name(rsc, dup->action), dup->interval_ms); } free_lrmd_cmd(cmd); return TRUE; } static void schedule_lrmd_cmd(lrmd_rsc_t * rsc, lrmd_cmd_t * cmd) { gboolean dup_processed = FALSE; CRM_CHECK(cmd != NULL, return); CRM_CHECK(rsc != NULL, return); crm_trace("Scheduling %s on %s", cmd->action, rsc->rsc_id); dup_processed = merge_recurring_duplicate(rsc, cmd); if (dup_processed) { /* duplicate recurring cmd found, cmds merged */ return; } /* The controller expects the executor to automatically cancel * recurring operations before a resource stops. */ if (safe_str_eq(cmd->action, "stop")) { cancel_all_recurring(rsc, NULL); } rsc->pending_ops = g_list_append(rsc->pending_ops, cmd); #ifdef HAVE_SYS_TIMEB_H ftime(&cmd->t_queue); if (cmd->t_first_queue.time == 0) { cmd->t_first_queue = cmd->t_queue; } #endif mainloop_set_trigger(rsc->work); if (cmd->start_delay) { cmd->delay_id = g_timeout_add(cmd->start_delay, start_delay_helper, cmd); } } static xmlNode * create_lrmd_reply(const char *origin, int rc, int call_id) { xmlNode *reply = create_xml_node(NULL, T_LRMD_REPLY); crm_xml_add(reply, F_LRMD_ORIGIN, origin); crm_xml_add_int(reply, F_LRMD_RC, rc); crm_xml_add_int(reply, F_LRMD_CALLID, call_id); return reply; } static void send_client_notify(gpointer key, gpointer value, gpointer user_data) { xmlNode *update_msg = user_data; crm_client_t *client = value; int rc; if (client == NULL) { crm_err("Asked to send event to NULL client"); return; } else if (client->name == NULL) { crm_trace("Asked to send event to client with no name"); return; } rc = lrmd_server_send_notify(client, update_msg); if ((rc <= 0) && (rc != -ENOTCONN)) { crm_warn("Could not notify client %s/%s: %s " CRM_XS " rc=%d", client->name, client->id, (rc? pcmk_strerror(rc) : "no data sent"), rc); } } #ifdef HAVE_SYS_TIMEB_H /*! * \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 has time zero) */ static int time_diff_ms(struct timeb *now, struct timeb *old) { struct timeb local_now = { 0, }; if (now == NULL) { ftime(&local_now); now = &local_now; } if ((old == NULL) || (old->time == 0)) { return 0; } return difftime(now->time, old->time) * 1000 + now->millitm - old->millitm; } /*! * \internal * \brief Reset a command's operation times to their original values. * * Reset a command's run and queued timestamps to the timestamps of the original * command, so we report the entire time since then and not just the time since * the most recent command (for recurring and systemd operations). * * /param[in] cmd Executor command object to reset * * /note It's not obvious what the queued time should be for a systemd * start/stop operation, which might go like this: * initial command queued 5ms, runs 3s * monitor command queued 10ms, runs 10s * monitor command queued 10ms, runs 10s * Is the queued time for that operation 5ms, 10ms or 25ms? The current * implementation will report 5ms. If it's 25ms, then we need to * subtract 20ms from the total exec time so as not to count it twice. * We can implement that later if it matters to anyone ... */ static void cmd_original_times(lrmd_cmd_t * cmd) { cmd->t_run = cmd->t_first_run; cmd->t_queue = cmd->t_first_queue; } #endif static void send_cmd_complete_notify(lrmd_cmd_t * cmd) { int exec_time = 0; int queue_time = 0; xmlNode *notify = NULL; #ifdef HAVE_SYS_TIMEB_H 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 first notify result for a cmd has already been sent earlier, and the * the option to only send notifies on result changes is set. Check to see * if the last result is the same as the new one. If so, suppress this update */ if (cmd->first_notify_sent && (cmd->call_opts & lrmd_opt_notify_changes_only)) { if (cmd->last_notify_rc == cmd->exec_rc && cmd->last_notify_op_status == cmd->lrmd_op_status) { /* only send changes */ return; } } cmd->first_notify_sent = 1; cmd->last_notify_rc = cmd->exec_rc; cmd->last_notify_op_status = cmd->lrmd_op_status; notify = create_xml_node(NULL, T_LRMD_NOTIFY); crm_xml_add(notify, F_LRMD_ORIGIN, __FUNCTION__); crm_xml_add_int(notify, F_LRMD_TIMEOUT, cmd->timeout); crm_xml_add_ms(notify, F_LRMD_RSC_INTERVAL, cmd->interval_ms); crm_xml_add_int(notify, F_LRMD_RSC_START_DELAY, cmd->start_delay); crm_xml_add_int(notify, F_LRMD_EXEC_RC, cmd->exec_rc); crm_xml_add_int(notify, F_LRMD_OP_STATUS, cmd->lrmd_op_status); crm_xml_add_int(notify, F_LRMD_CALLID, cmd->call_id); crm_xml_add_int(notify, F_LRMD_RSC_DELETED, cmd->rsc_deleted); #ifdef HAVE_SYS_TIMEB_H crm_xml_add_int(notify, F_LRMD_RSC_RUN_TIME, cmd->t_run.time); crm_xml_add_int(notify, F_LRMD_RSC_RCCHANGE_TIME, cmd->t_rcchange.time); crm_xml_add_int(notify, F_LRMD_RSC_EXEC_TIME, exec_time); crm_xml_add_int(notify, F_LRMD_RSC_QUEUE_TIME, queue_time); #endif crm_xml_add(notify, F_LRMD_OPERATION, LRMD_OP_RSC_EXEC); crm_xml_add(notify, F_LRMD_RSC_ID, cmd->rsc_id); if(cmd->real_action) { crm_xml_add(notify, F_LRMD_RSC_ACTION, cmd->real_action); } else { crm_xml_add(notify, F_LRMD_RSC_ACTION, cmd->action); } crm_xml_add(notify, F_LRMD_RSC_USERDATA_STR, cmd->userdata_str); crm_xml_add(notify, F_LRMD_RSC_OUTPUT, cmd->output); crm_xml_add(notify, F_LRMD_RSC_EXIT_REASON, cmd->exit_reason); if (cmd->params) { char *key = NULL; char *value = NULL; GHashTableIter iter; xmlNode *args = create_xml_node(notify, XML_TAG_ATTRS); g_hash_table_iter_init(&iter, cmd->params); while (g_hash_table_iter_next(&iter, (gpointer *) & key, (gpointer *) & value)) { hash2smartfield((gpointer) key, (gpointer) value, args); } } if (cmd->client_id && (cmd->call_opts & lrmd_opt_notify_orig_only)) { crm_client_t *client = crm_client_get_by_id(cmd->client_id); if (client) { send_client_notify(client->id, client, notify); } } else if (client_connections != NULL) { g_hash_table_foreach(client_connections, send_client_notify, notify); } free_xml(notify); } static void send_generic_notify(int rc, xmlNode * request) { if (client_connections != NULL) { int call_id = 0; xmlNode *notify = NULL; xmlNode *rsc_xml = get_xpath_object("//" F_LRMD_RSC, request, LOG_ERR); const char *rsc_id = crm_element_value(rsc_xml, F_LRMD_RSC_ID); const char *op = crm_element_value(request, F_LRMD_OPERATION); crm_element_value_int(request, F_LRMD_CALLID, &call_id); notify = create_xml_node(NULL, T_LRMD_NOTIFY); crm_xml_add(notify, F_LRMD_ORIGIN, __FUNCTION__); crm_xml_add_int(notify, F_LRMD_RC, rc); crm_xml_add_int(notify, F_LRMD_CALLID, call_id); crm_xml_add(notify, F_LRMD_OPERATION, op); crm_xml_add(notify, F_LRMD_RSC_ID, rsc_id); g_hash_table_foreach(client_connections, send_client_notify, notify); free_xml(notify); } } static void cmd_reset(lrmd_cmd_t * cmd) { cmd->lrmd_op_status = 0; cmd->last_pid = 0; #ifdef HAVE_SYS_TIMEB_H memset(&cmd->t_run, 0, sizeof(cmd->t_run)); memset(&cmd->t_queue, 0, sizeof(cmd->t_queue)); #endif free(cmd->exit_reason); cmd->exit_reason = NULL; free(cmd->output); cmd->output = NULL; } 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 && (cmd->lrmd_op_status == PCMK_LRM_OP_CANCELLED)) { if (rsc) { rsc->recurring_ops = g_list_remove(rsc->recurring_ops, cmd); rsc->pending_ops = g_list_remove(rsc->pending_ops, cmd); } free_lrmd_cmd(cmd); } else if (cmd->interval_ms == 0) { if (rsc) { rsc->pending_ops = g_list_remove(rsc->pending_ops, cmd); } free_lrmd_cmd(cmd); } else { /* Clear all the values pertaining just to the last iteration of a recurring op. */ cmd_reset(cmd); } } static int ocf2uniform_rc(int rc) { if (rc < 0 || rc > PCMK_OCF_FAILED_MASTER) { return PCMK_OCF_UNKNOWN_ERROR; } return rc; } static int stonith2uniform_rc(const char *action, int rc) { if (rc == -ENODEV) { if (safe_str_eq(action, "stop")) { rc = PCMK_OCF_OK; } else if (safe_str_eq(action, "start")) { rc = PCMK_OCF_NOT_INSTALLED; } else { rc = PCMK_OCF_NOT_RUNNING; } } else if (rc != 0) { rc = PCMK_OCF_UNKNOWN_ERROR; } return rc; } #if SUPPORT_NAGIOS static int nagios2uniform_rc(const char *action, int rc) { if (rc < 0) { return PCMK_OCF_UNKNOWN_ERROR; } switch (rc) { case NAGIOS_STATE_OK: return PCMK_OCF_OK; case NAGIOS_INSUFFICIENT_PRIV: return PCMK_OCF_INSUFFICIENT_PRIV; case NAGIOS_NOT_INSTALLED: return PCMK_OCF_NOT_INSTALLED; case NAGIOS_STATE_WARNING: case NAGIOS_STATE_CRITICAL: case NAGIOS_STATE_UNKNOWN: case NAGIOS_STATE_DEPENDENT: default: return PCMK_OCF_UNKNOWN_ERROR; } return PCMK_OCF_UNKNOWN_ERROR; } #endif static int get_uniform_rc(const char *standard, const char *action, int rc) { if (safe_str_eq(standard, PCMK_RESOURCE_CLASS_OCF)) { return ocf2uniform_rc(rc); } else if (safe_str_eq(standard, PCMK_RESOURCE_CLASS_STONITH)) { return stonith2uniform_rc(action, rc); } else if (safe_str_eq(standard, PCMK_RESOURCE_CLASS_SYSTEMD)) { return rc; } else if (safe_str_eq(standard, PCMK_RESOURCE_CLASS_UPSTART)) { return rc; #if SUPPORT_NAGIOS } else if (safe_str_eq(standard, PCMK_RESOURCE_CLASS_NAGIOS)) { return nagios2uniform_rc(action, rc); #endif } else { return services_get_ocf_exitcode(action, rc); } } static int action_get_uniform_rc(svc_action_t * action) { lrmd_cmd_t *cmd = action->cb_data; return get_uniform_rc(action->standard, cmd->action, action->rc); } void notify_of_new_client(crm_client_t *new_client) { crm_client_t *client = NULL; GHashTableIter iter; xmlNode *notify = NULL; char *key = NULL; notify = create_xml_node(NULL, T_LRMD_NOTIFY); crm_xml_add(notify, F_LRMD_ORIGIN, __FUNCTION__); crm_xml_add(notify, F_LRMD_OPERATION, LRMD_OP_NEW_CLIENT); g_hash_table_iter_init(&iter, client_connections); while (g_hash_table_iter_next(&iter, (gpointer *) & key, (gpointer *) & client)) { if (safe_str_eq(client->id, new_client->id)) { continue; } send_client_notify((gpointer) key, (gpointer) client, (gpointer) notify); } free_xml(notify); } static char * parse_exit_reason(const char *output) { const char *cur = NULL; const char *last = NULL; static int cookie_len = 0; char *eol = NULL; size_t reason_len = EXIT_REASON_MAX_LEN; if (output == NULL) { return NULL; } if (!cookie_len) { cookie_len = strlen(PCMK_OCF_REASON_PREFIX); } cur = strstr(output, PCMK_OCF_REASON_PREFIX); for (; cur != NULL; cur = strstr(cur, PCMK_OCF_REASON_PREFIX)) { /* skip over the cookie delimiter string */ cur += cookie_len; last = cur; } if (last == NULL) { return NULL; } // Truncate everything after a new line, and limit reason string size eol = strchr(last, '\n'); if (eol) { reason_len = QB_MIN(reason_len, eol - last); } return strndup(last, reason_len); } void client_disconnect_cleanup(const char *client_id) { GHashTableIter iter; lrmd_rsc_t *rsc = NULL; char *key = NULL; g_hash_table_iter_init(&iter, rsc_list); while (g_hash_table_iter_next(&iter, (gpointer *) & key, (gpointer *) & rsc)) { if (rsc->call_opts & lrmd_opt_drop_recurring) { /* This client is disconnecting, drop any recurring operations * it may have initiated on the resource */ cancel_all_recurring(rsc, client_id); } } } static void action_complete(svc_action_t * action) { lrmd_rsc_t *rsc; lrmd_cmd_t *cmd = action->cb_data; const char *rclass = NULL; bool goagain = false; if (!cmd) { crm_err("Completed executor action (%s) does not match any known operations", action->id); return; } #ifdef HAVE_SYS_TIMEB_H if (cmd->exec_rc != action->rc) { ftime(&cmd->t_rcchange); } #endif cmd->last_pid = action->pid; cmd->exec_rc = action_get_uniform_rc(action); cmd->lrmd_op_status = action->status; rsc = cmd->rsc_id ? g_hash_table_lookup(rsc_list, cmd->rsc_id) : NULL; if (rsc && safe_str_eq(rsc->class, PCMK_RESOURCE_CLASS_SERVICE)) { rclass = resources_find_service_class(rsc->type); } else if(rsc) { rclass = rsc->class; } if (safe_str_eq(rclass, PCMK_RESOURCE_CLASS_SYSTEMD)) { if(cmd->exec_rc == PCMK_OCF_OK && safe_str_eq(cmd->action, "start")) { /* systemd I curse thee! * * systemd returns from start actions after the start _begins_ * not after it completes. * * So we have to jump through a few hoops so that we don't * report 'complete' to the rest of pacemaker until, you know, * it's actually done. */ goagain = true; cmd->real_action = cmd->action; cmd->action = strdup("monitor"); } else if(cmd->exec_rc == PCMK_OCF_OK && safe_str_eq(cmd->action, "stop")) { goagain = true; cmd->real_action = cmd->action; cmd->action = strdup("monitor"); } else if(cmd->real_action) { /* Ok, so this is the follow up monitor action to check if start actually completed */ if(cmd->lrmd_op_status == PCMK_LRM_OP_DONE && cmd->exec_rc == PCMK_OCF_PENDING) { goagain = true; } else if(cmd->exec_rc == PCMK_OCF_OK && safe_str_eq(cmd->real_action, "stop")) { goagain = true; } else { #ifdef HAVE_SYS_TIMEB_H int time_sum = time_diff_ms(NULL, &cmd->t_first_run); int timeout_left = cmd->timeout_orig - time_sum; crm_debug("%s %s is now complete (elapsed=%dms, remaining=%dms): %s (%d)", cmd->rsc_id, cmd->real_action, time_sum, timeout_left, services_ocf_exitcode_str(cmd->exec_rc), cmd->exec_rc); cmd_original_times(cmd); #endif if(cmd->lrmd_op_status == PCMK_LRM_OP_DONE && cmd->exec_rc == PCMK_OCF_NOT_RUNNING && safe_str_eq(cmd->real_action, "stop")) { cmd->exec_rc = PCMK_OCF_OK; } } } } #if SUPPORT_NAGIOS if (rsc && safe_str_eq(rsc->class, PCMK_RESOURCE_CLASS_NAGIOS)) { if (safe_str_eq(cmd->action, "monitor") && (cmd->interval_ms == 0) && cmd->exec_rc == PCMK_OCF_OK) { /* Successfully executed --version for the nagios plugin */ cmd->exec_rc = PCMK_OCF_NOT_RUNNING; } else if (safe_str_eq(cmd->action, "start") && cmd->exec_rc != PCMK_OCF_OK) { goagain = true; } } #endif /* Wrapping this section in ifdef implies that systemd resources are not * fully supported on platforms without sys/timeb.h. Since timeb is * obsolete, we should eventually prefer a clock_gettime() implementation * (wrapped in its own ifdef) with timeb as a fallback. */ if(goagain) { #ifdef HAVE_SYS_TIMEB_H int time_sum = time_diff_ms(NULL, &cmd->t_first_run); int timeout_left = cmd->timeout_orig - time_sum; int delay = cmd->timeout_orig / 10; if(delay >= timeout_left && timeout_left > 20) { delay = timeout_left/2; } delay = QB_MIN(2000, delay); if (delay < timeout_left) { cmd->start_delay = delay; cmd->timeout = timeout_left; if(cmd->exec_rc == PCMK_OCF_OK) { crm_debug("%s %s may still be in progress: re-scheduling (elapsed=%dms, remaining=%dms, start_delay=%dms)", cmd->rsc_id, cmd->real_action, time_sum, timeout_left, delay); } else if(cmd->exec_rc == PCMK_OCF_PENDING) { crm_info("%s %s is still in progress: re-scheduling (elapsed=%dms, remaining=%dms, start_delay=%dms)", cmd->rsc_id, cmd->action, time_sum, timeout_left, delay); } else { crm_notice("%s %s failed '%s' (%d): re-scheduling (elapsed=%dms, remaining=%dms, start_delay=%dms)", cmd->rsc_id, cmd->action, services_ocf_exitcode_str(cmd->exec_rc), cmd->exec_rc, time_sum, timeout_left, delay); } cmd_reset(cmd); if(rsc) { rsc->active = NULL; } schedule_lrmd_cmd(rsc, cmd); /* Don't finalize cmd, we're not done with it yet */ return; } else { crm_notice("Giving up on %s %s (rc=%d): timeout (elapsed=%dms, remaining=%dms)", cmd->rsc_id, cmd->real_action?cmd->real_action:cmd->action, cmd->exec_rc, time_sum, timeout_left); cmd->lrmd_op_status = PCMK_LRM_OP_TIMEOUT; cmd->exec_rc = PCMK_OCF_TIMEOUT; cmd_original_times(cmd); } #endif } if (action->stderr_data) { cmd->output = strdup(action->stderr_data); cmd->exit_reason = parse_exit_reason(action->stderr_data); } else if (action->stdout_data) { cmd->output = strdup(action->stdout_data); } cmd_finalize(cmd, rsc); } static void stonith_action_complete(lrmd_cmd_t * cmd, int rc) { bool recurring = (cmd->interval_ms > 0); lrmd_rsc_t *rsc = NULL; cmd->exec_rc = get_uniform_rc(PCMK_RESOURCE_CLASS_STONITH, cmd->action, rc); rsc = g_hash_table_lookup(rsc_list, cmd->rsc_id); if (cmd->lrmd_op_status == PCMK_LRM_OP_CANCELLED) { recurring = FALSE; /* do nothing */ } else if (rc == -ENODEV && safe_str_eq(cmd->action, "monitor")) { // The device is not registered with the fencer if (recurring) { /* If we get here, the fencer somehow lost the registration of a * previously active device (possibly due to crash and respawn). In * that case, we need to indicate that the recurring monitor needs * to be cancelled. */ cmd->lrmd_op_status = PCMK_LRM_OP_CANCELLED; recurring = FALSE; } else { cmd->lrmd_op_status = PCMK_LRM_OP_DONE; } cmd->exec_rc = PCMK_OCF_NOT_RUNNING; } else if (rc) { /* Attempt to map return codes to op status if possible */ switch (rc) { case -EPROTONOSUPPORT: cmd->lrmd_op_status = PCMK_LRM_OP_NOTSUPPORTED; break; case -ETIME: cmd->lrmd_op_status = PCMK_LRM_OP_TIMEOUT; break; default: /* TODO: This looks wrong. Status should be _DONE and exec_rc set to an error */ cmd->lrmd_op_status = PCMK_LRM_OP_ERROR; } } else { /* command successful */ cmd->lrmd_op_status = PCMK_LRM_OP_DONE; if (safe_str_eq(cmd->action, "start") && rsc) { rsc->stonith_started = 1; } } if (recurring && rsc) { if (cmd->stonith_recurring_id) { g_source_remove(cmd->stonith_recurring_id); } cmd->stonith_recurring_id = g_timeout_add(cmd->interval_ms, stonith_recurring_op_helper, cmd); } cmd_finalize(cmd, rsc); } static void lrmd_stonith_callback(stonith_t * stonith, stonith_callback_data_t * data) { stonith_action_complete(data->userdata, data->rc); } void stonith_connection_failed(void) { GHashTableIter iter; GList *cmd_list = NULL; GList *cmd_iter = NULL; lrmd_rsc_t *rsc = NULL; char *key = NULL; g_hash_table_iter_init(&iter, rsc_list); while (g_hash_table_iter_next(&iter, (gpointer *) & key, (gpointer *) & rsc)) { if (safe_str_eq(rsc->class, PCMK_RESOURCE_CLASS_STONITH)) { if (rsc->active) { cmd_list = g_list_append(cmd_list, rsc->active); } if (rsc->recurring_ops) { cmd_list = g_list_concat(cmd_list, rsc->recurring_ops); } if (rsc->pending_ops) { cmd_list = g_list_concat(cmd_list, rsc->pending_ops); } rsc->pending_ops = rsc->recurring_ops = NULL; } } if (!cmd_list) { return; } crm_err("Connection to fencer failed, finalizing %d pending operations", g_list_length(cmd_list)); for (cmd_iter = cmd_list; cmd_iter; cmd_iter = cmd_iter->next) { stonith_action_complete(cmd_iter->data, -ENOTCONN); } g_list_free(cmd_list); } static int lrmd_rsc_execute_stonith(lrmd_rsc_t * rsc, lrmd_cmd_t * cmd) { int rc = 0; int do_monitor = 0; stonith_t *stonith_api = get_stonith_connection(); if (!stonith_api) { cmd->exec_rc = get_uniform_rc(PCMK_RESOURCE_CLASS_STONITH, cmd->action, -ENOTCONN); cmd->lrmd_op_status = PCMK_LRM_OP_ERROR; cmd_finalize(cmd, rsc); return -EUNATCH; } if (safe_str_eq(cmd->action, "start")) { char *key = NULL; char *value = NULL; stonith_key_value_t *device_params = NULL; 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); } } /* Stonith automatically registers devices from the IPC when changes * occur, but to avoid a possible race condition between stonith * receiving the IPC update and the executor requesting that resource, * the executor still registers the device as well. Stonith knows how to * handle duplicate device registrations correctly. */ 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); if (rc == 0) { do_monitor = 1; } } else if (safe_str_eq(cmd->action, "stop")) { rc = stonith_api->cmds->remove_device(stonith_api, st_opt_sync_call, cmd->rsc_id); rsc->stonith_started = 0; } else if (safe_str_eq(cmd->action, "monitor")) { if (cmd->interval_ms > 0) { do_monitor = 1; } else { rc = rsc->stonith_started ? 0 : -ENODEV; } } if (!do_monitor) { goto cleanup_stonith_exec; } rc = stonith_api->cmds->monitor(stonith_api, 0, cmd->rsc_id, cmd->timeout / 1000); rc = stonith_api->cmds->register_callback(stonith_api, rc, 0, 0, cmd, "lrmd_stonith_callback", lrmd_stonith_callback); /* don't cleanup yet, we will find out the result of the monitor later */ if (rc > 0) { rsc->active = cmd; return rc; } else if (rc == 0) { rc = -1; } cleanup_stonith_exec: stonith_action_complete(cmd, rc); return rc; } static int lrmd_rsc_execute_service_lib(lrmd_rsc_t * rsc, lrmd_cmd_t * cmd) { svc_action_t *action = NULL; GHashTable *params_copy = NULL; CRM_ASSERT(rsc); CRM_ASSERT(cmd); crm_trace("Creating action, resource:%s action:%s class:%s provider:%s agent:%s", rsc->rsc_id, cmd->action, rsc->class, rsc->provider, rsc->type); #if SUPPORT_NAGIOS /* Recurring operations are cancelled anyway for a stop operation */ if (safe_str_eq(rsc->class, PCMK_RESOURCE_CLASS_NAGIOS) && safe_str_eq(cmd->action, "stop")) { cmd->exec_rc = PCMK_OCF_OK; goto exec_done; } #endif params_copy = crm_str_table_dup(cmd->params); action = resources_action_create(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) { crm_err("Failed to create action, action:%s on resource %s", cmd->action, rsc->rsc_id); cmd->lrmd_op_status = PCMK_LRM_OP_ERROR; goto exec_done; } action->cb_data = cmd; /* 'cmd' may not be valid after this point if * services_action_async() returned TRUE * * Upstart and systemd both synchronously determine monitor/status * results and call action_complete (which may free 'cmd') if necessary. */ if (services_action_async(action, action_complete)) { return TRUE; } cmd->exec_rc = action->rc; if(action->status != PCMK_LRM_OP_DONE) { cmd->lrmd_op_status = action->status; } else { cmd->lrmd_op_status = PCMK_LRM_OP_ERROR; } services_action_free(action); action = NULL; exec_done: cmd_finalize(cmd, rsc); return TRUE; } static gboolean lrmd_rsc_execute(lrmd_rsc_t * rsc) { lrmd_cmd_t *cmd = NULL; CRM_CHECK(rsc != NULL, return FALSE); if (rsc->active) { crm_trace("%s is still active", rsc->rsc_id); return TRUE; } if (rsc->pending_ops) { GList *first = rsc->pending_ops; cmd = first->data; if (cmd->delay_id) { crm_trace ("Command %s %s was asked to run too early, waiting for start_delay timeout of %dms", cmd->rsc_id, cmd->action, cmd->start_delay); return TRUE; } rsc->pending_ops = g_list_remove_link(rsc->pending_ops, first); g_list_free_1(first); #ifdef HAVE_SYS_TIMEB_H if (cmd->t_first_run.time == 0) { ftime(&cmd->t_first_run); } ftime(&cmd->t_run); #endif } 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 (safe_str_eq(rsc->class, PCMK_RESOURCE_CLASS_STONITH)) { lrmd_rsc_execute_stonith(rsc, cmd); } else { lrmd_rsc_execute_service_lib(rsc, cmd); } return TRUE; } static gboolean lrmd_rsc_dispatch(gpointer user_data) { return lrmd_rsc_execute(user_data); } void free_rsc(gpointer data) { GListPtr gIter = NULL; lrmd_rsc_t *rsc = data; int is_stonith = safe_str_eq(rsc->class, PCMK_RESOURCE_CLASS_STONITH); gIter = rsc->pending_ops; while (gIter != NULL) { GListPtr next = gIter->next; lrmd_cmd_t *cmd = gIter->data; /* command was never executed */ cmd->lrmd_op_status = PCMK_LRM_OP_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) { GListPtr next = gIter->next; lrmd_cmd_t *cmd = gIter->data; if (is_stonith) { cmd->lrmd_op_status = PCMK_LRM_OP_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 xmlNode * process_lrmd_signon(crm_client_t *client, xmlNode *request, int call_id) { xmlNode *reply = NULL; int rc = pcmk_ok; const char *is_ipc_provider = crm_element_value(request, F_LRMD_IS_IPC_PROVIDER); const char *protocol_version = crm_element_value(request, F_LRMD_PROTOCOL_VERSION); if (compare_version(protocol_version, LRMD_MIN_PROTOCOL_VERSION) < 0) { crm_err("Cluster API version must be greater than or equal to %s, not %s", LRMD_MIN_PROTOCOL_VERSION, protocol_version); rc = -EPROTO; } reply = create_lrmd_reply(__FUNCTION__, rc, call_id); crm_xml_add(reply, F_LRMD_OPERATION, CRM_OP_REGISTER); crm_xml_add(reply, F_LRMD_CLIENTID, client->id); crm_xml_add(reply, F_LRMD_PROTOCOL_VERSION, LRMD_PROTOCOL_VERSION); if (crm_is_true(is_ipc_provider)) { // This is a remote connection from a cluster node's controller #ifdef SUPPORT_REMOTE ipc_proxy_add_provider(client); #endif } return reply; } static int process_lrmd_rsc_register(crm_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 && safe_str_eq(rsc->class, dup->class) && safe_str_eq(rsc->provider, dup->provider) && safe_str_eq(rsc->type, dup->type)) { crm_warn("Can't add, RSC '%s' already present in the rsc list (%d active resources)", rsc->rsc_id, g_hash_table_size(rsc_list)); free_rsc(rsc); return rc; } g_hash_table_replace(rsc_list, rsc->rsc_id, rsc); crm_info("Added '%s' to the rsc list (%d active resources)", rsc->rsc_id, g_hash_table_size(rsc_list)); return rc; } static xmlNode * process_lrmd_get_rsc_info(xmlNode *request, int call_id) { int rc = pcmk_ok; xmlNode *rsc_xml = get_xpath_object("//" F_LRMD_RSC, request, LOG_ERR); const char *rsc_id = crm_element_value(rsc_xml, F_LRMD_RSC_ID); xmlNode *reply = NULL; lrmd_rsc_t *rsc = NULL; if (rsc_id == NULL) { rc = -ENODEV; } else { rsc = g_hash_table_lookup(rsc_list, rsc_id); if (rsc == NULL) { crm_info("Resource '%s' not found (%d active resources)", rsc_id, g_hash_table_size(rsc_list)); rc = -ENODEV; } } reply = create_lrmd_reply(__FUNCTION__, rc, call_id); if (rsc) { crm_xml_add(reply, F_LRMD_RSC_ID, rsc->rsc_id); crm_xml_add(reply, F_LRMD_CLASS, rsc->class); crm_xml_add(reply, F_LRMD_PROVIDER, rsc->provider); crm_xml_add(reply, F_LRMD_TYPE, rsc->type); } return reply; } static int process_lrmd_rsc_unregister(crm_client_t * client, uint32_t id, xmlNode * request) { int rc = pcmk_ok; lrmd_rsc_t *rsc = NULL; xmlNode *rsc_xml = get_xpath_object("//" F_LRMD_RSC, request, LOG_ERR); const char *rsc_id = crm_element_value(rsc_xml, F_LRMD_RSC_ID); if (!rsc_id) { return -ENODEV; } 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 pcmk_ok; } if (rsc->active) { /* let the caller know there are still active ops on this rsc to watch for */ crm_trace("Operation still in progress: %p", rsc->active); rc = -EINPROGRESS; } g_hash_table_remove(rsc_list, rsc_id); return rc; } static int process_lrmd_rsc_exec(crm_client_t * client, uint32_t id, xmlNode * request) { lrmd_rsc_t *rsc = NULL; lrmd_cmd_t *cmd = NULL; xmlNode *rsc_xml = get_xpath_object("//" F_LRMD_RSC, request, LOG_ERR); const char *rsc_id = crm_element_value(rsc_xml, F_LRMD_RSC_ID); int call_id; if (!rsc_id) { return -EINVAL; } if (!(rsc = g_hash_table_lookup(rsc_list, rsc_id))) { crm_info("Resource '%s' not found (%d active resources)", rsc_id, g_hash_table_size(rsc_list)); return -ENODEV; } cmd = create_lrmd_cmd(request, client); call_id = cmd->call_id; /* Don't reference cmd after handing it off to be scheduled. * The cmd could get merged and freed. */ schedule_lrmd_cmd(rsc, cmd); return call_id; } static int cancel_op(const char *rsc_id, const char *action, guint interval_ms) { GListPtr 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 (safe_str_eq(cmd->action, action) && (cmd->interval_ms == interval_ms)) { cmd->lrmd_op_status = PCMK_LRM_OP_CANCELLED; cmd_finalize(cmd, rsc); return pcmk_ok; } } if (safe_str_eq(rsc->class, PCMK_RESOURCE_CLASS_STONITH)) { /* 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 (safe_str_eq(cmd->action, action) && (cmd->interval_ms == interval_ms)) { cmd->lrmd_op_status = PCMK_LRM_OP_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 && safe_str_neq(cmd->client_id, client_id)) { 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(crm_client_t * client, uint32_t id, xmlNode * request) { xmlNode *rsc_xml = get_xpath_object("//" F_LRMD_RSC, request, LOG_ERR); const char *rsc_id = crm_element_value(rsc_xml, F_LRMD_RSC_ID); const char *action = crm_element_value(rsc_xml, F_LRMD_RSC_ACTION); guint interval_ms = 0; crm_element_value_ms(rsc_xml, F_LRMD_RSC_INTERVAL, &interval_ms); if (!rsc_id || !action) { return -EINVAL; } return cancel_op(rsc_id, action, interval_ms); } static void add_recurring_op_xml(xmlNode *reply, lrmd_rsc_t *rsc) { xmlNode *rsc_xml = create_xml_node(reply, F_LRMD_RSC); crm_xml_add(rsc_xml, F_LRMD_RSC_ID, rsc->rsc_id); for (GList *item = rsc->recurring_ops; item != NULL; item = item->next) { lrmd_cmd_t *cmd = item->data; xmlNode *op_xml = create_xml_node(rsc_xml, T_LRMD_RSC_OP); crm_xml_add(op_xml, F_LRMD_RSC_ACTION, (cmd->real_action? cmd->real_action : cmd->action)); crm_xml_add_ms(op_xml, F_LRMD_RSC_INTERVAL, cmd->interval_ms); crm_xml_add_int(op_xml, F_LRMD_TIMEOUT, cmd->timeout_orig); } } static xmlNode * process_lrmd_get_recurring(xmlNode *request, int call_id) { int rc = pcmk_ok; const char *rsc_id = NULL; lrmd_rsc_t *rsc = NULL; xmlNode *reply = NULL; xmlNode *rsc_xml = NULL; // Resource ID is optional rsc_xml = first_named_child(request, F_LRMD_CALLDATA); if (rsc_xml) { rsc_xml = first_named_child(rsc_xml, F_LRMD_RSC); } if (rsc_xml) { rsc_id = crm_element_value(rsc_xml, F_LRMD_RSC_ID); } // If resource ID is specified, resource must exist if (rsc_id != NULL) { rsc = g_hash_table_lookup(rsc_list, rsc_id); if (rsc == NULL) { crm_info("Resource '%s' not found (%d active resources)", rsc_id, g_hash_table_size(rsc_list)); rc = -ENODEV; } } reply = create_lrmd_reply(__FUNCTION__, 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; } +static bool sbd_check_logged = FALSE; + void process_lrmd_message(crm_client_t * client, uint32_t id, xmlNode * request) { int rc = pcmk_ok; int call_id = 0; const char *op = crm_element_value(request, F_LRMD_OPERATION); int do_reply = 0; int do_notify = 0; xmlNode *reply = NULL; crm_trace("Processing %s operation from %s", op, client->id); crm_element_value_int(request, F_LRMD_CALLID, &call_id); if (crm_str_eq(op, CRM_OP_IPC_FWD, TRUE)) { #ifdef SUPPORT_REMOTE ipc_proxy_forward_client(client, request); #endif do_reply = 1; } else if (crm_str_eq(op, CRM_OP_REGISTER, TRUE)) { reply = process_lrmd_signon(client, request, call_id); do_reply = 1; } else if (crm_str_eq(op, LRMD_OP_RSC_REG, TRUE)) { rc = process_lrmd_rsc_register(client, id, request); do_notify = 1; do_reply = 1; } else if (crm_str_eq(op, LRMD_OP_RSC_INFO, TRUE)) { reply = process_lrmd_get_rsc_info(request, call_id); do_reply = 1; } else if (crm_str_eq(op, LRMD_OP_RSC_UNREG, TRUE)) { 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; } do_reply = 1; } else if (crm_str_eq(op, LRMD_OP_RSC_EXEC, TRUE)) { rc = process_lrmd_rsc_exec(client, id, request); do_reply = 1; } else if (crm_str_eq(op, LRMD_OP_RSC_CANCEL, TRUE)) { rc = process_lrmd_rsc_cancel(client, id, request); do_reply = 1; } else if (crm_str_eq(op, LRMD_OP_POKE, TRUE)) { do_notify = 1; do_reply = 1; } else if (crm_str_eq(op, LRMD_OP_CHECK, TRUE)) { - xmlNode *data = get_message_xml(request, F_LRMD_CALLDATA); - const char *timeout = crm_element_value(data, F_LRMD_WATCHDOG); - CRM_LOG_ASSERT(data != NULL); - check_sbd_timeout(timeout); + if(disable_sbd_check) { + if(sbd_check_logged == FALSE) { + crm_warn("Explicitly ignoring SBD check, this is only safe inside bundles"); + sbd_check_logged = TRUE; + } + + } else { + xmlNode *data = get_message_xml(request, F_LRMD_CALLDATA); + const char *timeout = crm_element_value(data, F_LRMD_WATCHDOG); + CRM_LOG_ASSERT(data != NULL); + check_sbd_timeout(timeout); + } + } else if (crm_str_eq(op, LRMD_OP_ALERT_EXEC, TRUE)) { rc = process_lrmd_alert_exec(client, id, request); do_reply = 1; } else if (crm_str_eq(op, LRMD_OP_GET_RECURRING, TRUE)) { reply = process_lrmd_get_recurring(request, call_id); do_reply = 1; } else { rc = -EOPNOTSUPP; do_reply = 1; crm_err("Unknown %s from %s", op, client->name); crm_log_xml_warn(request, "UnknownOp"); } 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_ok; if (reply == NULL) { reply = create_lrmd_reply(__FUNCTION__, rc, call_id); } send_rc = lrmd_server_send_reply(client, id, reply); free_xml(reply); if (send_rc < 0) { crm_warn("Reply to client %s failed: %s " CRM_XS " %d", client->name, pcmk_strerror(send_rc), send_rc); } } if (do_notify) { send_generic_notify(rc, request); } } diff --git a/daemons/execd/pacemaker-execd.c b/daemons/execd/pacemaker-execd.c index 198e0249b1..e9be5a9c66 100644 --- a/daemons/execd/pacemaker-execd.c +++ b/daemons/execd/pacemaker-execd.c @@ -1,620 +1,625 @@ /* * Copyright 2012-2018 David Vossel * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include "pacemaker-execd.h" #if defined(HAVE_GNUTLS_GNUTLS_H) && defined(SUPPORT_REMOTE) # define ENABLE_PCMK_REMOTE // Hidden in liblrmd extern int lrmd_tls_send_msg(crm_remote_t *session, xmlNode *msg, uint32_t id, const char *msg_type); #endif static GMainLoop *mainloop = NULL; static qb_ipcs_service_t *ipcs = NULL; static stonith_t *stonith_api = NULL; int lrmd_call_id = 0; +int disable_sbd_check = 0; #ifdef ENABLE_PCMK_REMOTE /* whether shutdown request has been sent */ static volatile sig_atomic_t shutting_down = FALSE; /* timer for waiting for acknowledgment of shutdown request */ static volatile guint shutdown_ack_timer = 0; static gboolean lrmd_exit(gpointer data); #endif static void stonith_connection_destroy_cb(stonith_t * st, stonith_event_t * e) { stonith_api->state = stonith_disconnected; crm_err("Connection to fencer lost"); stonith_connection_failed(); } stonith_t * get_stonith_connection(void) { if (stonith_api && stonith_api->state == stonith_disconnected) { stonith_api_delete(stonith_api); stonith_api = NULL; } if (!stonith_api) { int rc = 0; int tries = 10; stonith_api = stonith_api_new(); do { rc = stonith_api->cmds->connect(stonith_api, "pacemaker-execd", NULL); if (rc == pcmk_ok) { stonith_api->cmds->register_notification(stonith_api, T_STONITH_NOTIFY_DISCONNECT, stonith_connection_destroy_cb); break; } sleep(1); tries--; } while (tries); if (rc) { crm_err("Unable to connect to stonith daemon to execute command. error: %s", pcmk_strerror(rc)); stonith_api_delete(stonith_api); stonith_api = NULL; } } return stonith_api; } static int32_t lrmd_ipc_accept(qb_ipcs_connection_t * c, uid_t uid, gid_t gid) { crm_trace("Connection %p", c); if (crm_client_new(c, uid, gid) == NULL) { return -EIO; } return 0; } static void lrmd_ipc_created(qb_ipcs_connection_t * c) { crm_client_t *new_client = crm_client_get(c); crm_trace("Connection %p", c); CRM_ASSERT(new_client != NULL); /* Now that the connection is offically established, alert * the other clients a new connection exists. */ notify_of_new_client(new_client); } static int32_t lrmd_ipc_dispatch(qb_ipcs_connection_t * c, void *data, size_t size) { uint32_t id = 0; uint32_t flags = 0; crm_client_t *client = crm_client_get(c); xmlNode *request = crm_ipcs_recv(client, data, size, &id, &flags); CRM_CHECK(client != NULL, crm_err("Invalid client"); return FALSE); CRM_CHECK(client->id != NULL, crm_err("Invalid client: %p", client); return FALSE); CRM_CHECK(flags & crm_ipc_client_response, crm_err("Invalid client request: %p", client); return FALSE); if (!request) { return 0; } if (!client->name) { const char *value = crm_element_value(request, F_LRMD_CLIENTNAME); if (value == NULL) { client->name = crm_itoa(crm_ipcs_client_pid(c)); } else { client->name = strdup(value); } } lrmd_call_id++; if (lrmd_call_id < 1) { lrmd_call_id = 1; } crm_xml_add(request, F_LRMD_CLIENTID, client->id); crm_xml_add(request, F_LRMD_CLIENTNAME, client->name); crm_xml_add_int(request, F_LRMD_CALLID, lrmd_call_id); process_lrmd_message(client, id, request); free_xml(request); return 0; } /*! * \internal * \brief Free a client connection, and exit if appropriate * * \param[in] client Client connection to free */ void lrmd_client_destroy(crm_client_t *client) { crm_client_destroy(client); #ifdef ENABLE_PCMK_REMOTE /* If we were waiting to shut down, we can now safely do so * if there are no more proxied IPC providers */ if (shutting_down && (ipc_proxy_get_provider() == NULL)) { lrmd_exit(NULL); } #endif } static int32_t lrmd_ipc_closed(qb_ipcs_connection_t * c) { crm_client_t *client = crm_client_get(c); if (client == NULL) { return 0; } crm_trace("Connection %p", c); client_disconnect_cleanup(client->id); #ifdef ENABLE_PCMK_REMOTE ipc_proxy_remove_provider(client); #endif lrmd_client_destroy(client); return 0; } static void lrmd_ipc_destroy(qb_ipcs_connection_t * c) { lrmd_ipc_closed(c); crm_trace("Connection %p", c); } static struct qb_ipcs_service_handlers lrmd_ipc_callbacks = { .connection_accept = lrmd_ipc_accept, .connection_created = lrmd_ipc_created, .msg_process = lrmd_ipc_dispatch, .connection_closed = lrmd_ipc_closed, .connection_destroyed = lrmd_ipc_destroy }; int lrmd_server_send_reply(crm_client_t * client, uint32_t id, xmlNode * reply) { crm_trace("Sending reply (%d) to client (%s)", id, client->id); switch (client->kind) { case CRM_CLIENT_IPC: return crm_ipcs_send(client, id, reply, FALSE); #ifdef ENABLE_PCMK_REMOTE case CRM_CLIENT_TLS: return lrmd_tls_send_msg(client->remote, reply, id, "reply"); #endif default: crm_err("Could not send reply: unknown client type %d", client->kind); } return -ENOTCONN; } int lrmd_server_send_notify(crm_client_t * client, xmlNode * msg) { crm_trace("Sending notification to client (%s)", client->id); switch (client->kind) { case CRM_CLIENT_IPC: if (client->ipcs == NULL) { crm_trace("Could not notify local client: disconnected"); return -ENOTCONN; } return crm_ipcs_send(client, 0, msg, crm_ipc_server_event); #ifdef ENABLE_PCMK_REMOTE case CRM_CLIENT_TLS: if (client->remote == NULL) { crm_trace("Could not notify remote client: disconnected"); return -ENOTCONN; } return lrmd_tls_send_msg(client->remote, msg, 0, "notify"); #endif default: crm_err("Could not notify client: unknown type %d", client->kind); } return -ENOTCONN; } /*! * \internal * \brief Clean up and exit immediately * * \param[in] data Ignored * * \return Doesn't return * \note This can be used as a timer callback. */ static gboolean lrmd_exit(gpointer data) { crm_info("Terminating with %d clients", crm_hash_table_size(client_connections)); if (stonith_api) { stonith_api->cmds->remove_notification(stonith_api, T_STONITH_NOTIFY_DISCONNECT); stonith_api->cmds->disconnect(stonith_api); stonith_api_delete(stonith_api); } if (ipcs) { mainloop_del_ipc_server(ipcs); } #ifdef ENABLE_PCMK_REMOTE lrmd_tls_server_destroy(); ipc_proxy_cleanup(); #endif crm_client_cleanup(); g_hash_table_destroy(rsc_list); if (mainloop) { lrmd_drain_alerts(g_main_loop_get_context(mainloop)); } crm_exit(CRM_EX_OK); return FALSE; } /*! * \internal * \brief Request cluster shutdown if appropriate, otherwise exit immediately * * \param[in] nsig Signal that caused invocation (ignored) */ static void lrmd_shutdown(int nsig) { #ifdef ENABLE_PCMK_REMOTE crm_client_t *ipc_proxy = ipc_proxy_get_provider(); /* If there are active proxied IPC providers, then we may be running * resources, so notify the cluster that we wish to shut down. */ if (ipc_proxy) { if (shutting_down) { crm_notice("Waiting for cluster to stop resources before exiting"); return; } crm_info("Sending shutdown request to cluster"); if (ipc_proxy_shutdown_req(ipc_proxy) < 0) { crm_crit("Shutdown request failed, exiting immediately"); } else { /* We requested a shutdown. Now, we need to wait for an * acknowledgement from the proxy host (which ensures the proxy host * supports shutdown requests), then wait for all proxy hosts to * disconnect (which ensures that all resources have been stopped). */ shutting_down = TRUE; /* Stop accepting new proxy connections */ lrmd_tls_server_destroy(); /* Older controller versions will never acknowledge our request, so * set a fairly short timeout to exit quickly in that case. If we * get the ack, we'll defuse this timer. */ shutdown_ack_timer = g_timeout_add_seconds(20, lrmd_exit, NULL); /* Currently, we let the OS kill us if the clients don't disconnect * in a reasonable time. We could instead set a long timer here * (shorter than what the OS is likely to use) and exit immediately * if it pops. */ return; } } #endif lrmd_exit(NULL); } /*! * \internal * \brief Defuse short exit timer if shutting down */ void handle_shutdown_ack() { #ifdef ENABLE_PCMK_REMOTE if (shutting_down) { crm_info("Received shutdown ack"); if (shutdown_ack_timer > 0) { g_source_remove(shutdown_ack_timer); shutdown_ack_timer = 0; } return; } #endif crm_debug("Ignoring unexpected shutdown ack"); } /*! * \internal * \brief Make short exit timer fire immediately */ void handle_shutdown_nack() { #ifdef ENABLE_PCMK_REMOTE if (shutting_down) { crm_info("Received shutdown nack"); if (shutdown_ack_timer > 0) { g_source_remove(shutdown_ack_timer); shutdown_ack_timer = g_timeout_add(0, lrmd_exit, NULL); } return; } #endif crm_debug("Ignoring unexpected shutdown nack"); } static pid_t main_pid = 0; static void sigdone(void) { exit(CRM_EX_OK); } static void sigreap(void) { pid_t pid = 0; int status; do { /* * Opinions seem to differ as to what to put here: * -1, any child process * 0, any child process whose process group ID is equal to that of the calling process */ pid = waitpid(-1, &status, WNOHANG); if(pid == main_pid) { /* Exit when pacemaker-remote exits and use the same return code */ if (WIFEXITED(status)) { exit(WEXITSTATUS(status)); } exit(CRM_EX_ERROR); } } while (pid > 0); } static struct { int sig; void (*handler)(void); } sigmap[] = { { SIGCHLD, sigreap }, { SIGINT, sigdone }, }; static void spawn_pidone(int argc, char **argv, char **envp) { sigset_t set; if (getpid() != 1) { return; } sigfillset(&set); sigprocmask(SIG_BLOCK, &set, 0); main_pid = fork(); switch (main_pid) { case 0: sigprocmask(SIG_UNBLOCK, &set, NULL); setsid(); setpgid(0, 0); /* Child remains as pacemaker-remoted */ return; case -1: perror("fork"); } /* Parent becomes the reaper of zombie processes */ /* Safe to initialize logging now if needed */ #ifdef HAVE___PROGNAME /* Differentiate ourselves in the 'ps' output */ { char *p; int i, maxlen; char *LastArgv = NULL; const char *name = "pcmk-init"; for(i = 0; i < argc; i++) { if(!i || (LastArgv + 1 == argv[i])) LastArgv = argv[i] + strlen(argv[i]); } for(i = 0; envp[i] != NULL; i++) { if((LastArgv + 1) == envp[i]) { LastArgv = envp[i] + strlen(envp[i]); } } maxlen = (LastArgv - argv[0]) - 2; i = strlen(name); /* We can overwrite individual argv[] arguments */ snprintf(argv[0], maxlen, "%s", name); /* Now zero out everything else */ p = &argv[0][i]; while(p < LastArgv) *p++ = '\0'; argv[1] = NULL; } #endif /* HAVE___PROGNAME */ while (1) { int sig; size_t i; sigwait(&set, &sig); for (i = 0; i < DIMOF(sigmap); i++) { if (sigmap[i].sig == sig) { sigmap[i].handler(); break; } } } } /* *INDENT-OFF* */ static struct crm_option long_options[] = { /* Top-level Options */ {"help", 0, 0, '?', "\tThis text"}, {"version", 0, 0, '$', "\tVersion information" }, {"verbose", 0, 0, 'V', "\tIncrease debug output"}, + {"disable-sbd-check", 0, 0, 'D', "\tDisable the check for sbd - only safe for bundles"}, {"logfile", 1, 0, 'l', "\tSend logs to the additional named logfile"}, #ifdef ENABLE_PCMK_REMOTE {"port", 1, 0, 'p', "\tPort to listen on"}, #endif {0, 0, 0, 0} }; /* *INDENT-ON* */ int main(int argc, char **argv, char **envp) { int flag = 0; int index = 0; int bump_log_num = 0; const char *option = NULL; /* If necessary, create PID1 now before any FDs are opened */ spawn_pidone(argc, argv, envp); #ifndef ENABLE_PCMK_REMOTE crm_log_preinit("pacemaker-execd", argc, argv); crm_set_options(NULL, "[options]", long_options, "Resource agent executor daemon for cluster nodes"); #else crm_log_preinit("pacemaker-remoted", argc, argv); crm_set_options(NULL, "[options]", long_options, "Resource agent executor daemon for Pacemaker Remote nodes"); #endif while (1) { flag = crm_get_option(argc, argv, &index); if (flag == -1) { break; } switch (flag) { case 'l': crm_add_logfile(optarg); break; case 'p': setenv("PCMK_remote_port", optarg, 1); break; case 'V': bump_log_num++; break; + case 'D': + disable_sbd_check = 1; + break; case '?': case '$': crm_help(flag, CRM_EX_OK); break; default: crm_help('?', CRM_EX_USAGE); break; } } crm_log_init(NULL, LOG_INFO, TRUE, FALSE, argc, argv, FALSE); while (bump_log_num > 0) { crm_bump_log_level(argc, argv); bump_log_num--; } option = daemon_option("logfacility"); if(option && safe_str_neq(option, "none")) { setenv("HA_LOGFACILITY", option, 1); /* Used by the ocf_log/ha_log OCF macro */ } option = daemon_option("logfile"); if(option && safe_str_neq(option, "none")) { setenv("HA_LOGFILE", option, 1); /* Used by the ocf_log/ha_log OCF macro */ if (daemon_option_enabled(crm_system_name, "debug")) { setenv("HA_DEBUGLOG", option, 1); /* Used by the ocf_log/ha_debug OCF macro */ } } /* The presence of this variable allegedly controls whether child * processes like httpd will try and use Systemd's sd_notify * API */ unsetenv("NOTIFY_SOCKET"); /* Used by RAs - Leave owned by root */ crm_build_path(CRM_RSCTMP_DIR, 0755); rsc_list = g_hash_table_new_full(crm_str_hash, g_str_equal, NULL, free_rsc); ipcs = mainloop_add_ipc_server(CRM_SYSTEM_LRMD, QB_IPC_SHM, &lrmd_ipc_callbacks); if (ipcs == NULL) { crm_err("Failed to create IPC server: shutting down and inhibiting respawn"); crm_exit(CRM_EX_FATAL); } #ifdef ENABLE_PCMK_REMOTE if (lrmd_init_remote_tls_server() < 0) { crm_err("Failed to create TLS listener: shutting down and staying down"); crm_exit(CRM_EX_FATAL); } ipc_proxy_init(); #endif mainloop_add_signal(SIGTERM, lrmd_shutdown); mainloop = g_main_loop_new(NULL, FALSE); crm_info("Starting"); g_main_loop_run(mainloop); /* should never get here */ lrmd_exit(NULL); return CRM_EX_OK; } diff --git a/daemons/execd/pacemaker-execd.h b/daemons/execd/pacemaker-execd.h index 47b96601c5..30f451e855 100644 --- a/daemons/execd/pacemaker-execd.h +++ b/daemons/execd/pacemaker-execd.h @@ -1,98 +1,99 @@ /* * Copyright 2012-2018 David Vossel * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #ifndef PACEMAKER_EXECD__H # define PACEMAKER_EXECD__H # include # include # include # include # ifdef HAVE_GNUTLS_GNUTLS_H # undef KEYFILE # include # endif GHashTable *rsc_list; +extern int disable_sbd_check; typedef struct lrmd_rsc_s { char *rsc_id; char *class; char *provider; char *type; int call_opts; /* NEVER dereference this pointer, * It simply exists as a switch to let us know * when the currently active operation has completed */ void *active; /* Operations in this list * have not been executed yet. */ GList *pending_ops; /* Operations in this list are recurring operations * that have been handed off from the pending ops list. */ GList *recurring_ops; int stonith_started; crm_trigger_t *work; } lrmd_rsc_t; # ifdef HAVE_GNUTLS_GNUTLS_H // in remoted_tls.c int lrmd_init_remote_tls_server(void); void lrmd_tls_server_destroy(void); # endif int lrmd_server_send_reply(crm_client_t * client, uint32_t id, xmlNode * reply); int lrmd_server_send_notify(crm_client_t * client, xmlNode * msg); void notify_of_new_client(crm_client_t *new_client); void process_lrmd_message(crm_client_t * client, uint32_t id, xmlNode * request); void free_rsc(gpointer data); void handle_shutdown_ack(void); void handle_shutdown_nack(void); void lrmd_client_destroy(crm_client_t *client); void client_disconnect_cleanup(const char *client_id); /*! * \brief Don't worry about freeing this connection. It is * taken care of after mainloop exits by the main() function. */ stonith_t *get_stonith_connection(void); /*! * \brief This is a callback that tells the lrmd * the current stonith connection has gone away. This allows * us to timeout any pending stonith commands */ void stonith_connection_failed(void); #ifdef SUPPORT_REMOTE void ipc_proxy_init(void); void ipc_proxy_cleanup(void); void ipc_proxy_add_provider(crm_client_t *client); void ipc_proxy_remove_provider(crm_client_t *client); void ipc_proxy_forward_client(crm_client_t *client, xmlNode *xml); crm_client_t *ipc_proxy_get_provider(void); int ipc_proxy_shutdown_req(crm_client_t *ipc_proxy); #endif int process_lrmd_alert_exec(crm_client_t *client, uint32_t id, xmlNode *request); void lrmd_drain_alerts(GMainContext *ctx); #endif // PACEMAKER_EXECD__H