diff --git a/daemons/controld/controld_remote_ra.c b/daemons/controld/controld_remote_ra.c index 6260a79759..faf3a32de7 100644 --- a/daemons/controld/controld_remote_ra.c +++ b/daemons/controld/controld_remote_ra.c @@ -1,1375 +1,1375 @@ /* * Copyright 2013-2023 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #define REMOTE_LRMD_RA "remote" /* The max start timeout before cmd retry */ #define MAX_START_TIMEOUT_MS 10000 #define cmd_set_flags(cmd, flags_to_set) do { \ (cmd)->status = pcmk__set_flags_as(__func__, __LINE__, LOG_TRACE, \ "Remote command", (cmd)->rsc_id, (cmd)->status, \ (flags_to_set), #flags_to_set); \ } while (0) #define cmd_clear_flags(cmd, flags_to_clear) do { \ (cmd)->status = pcmk__clear_flags_as(__func__, __LINE__, LOG_TRACE, \ "Remote command", (cmd)->rsc_id, (cmd)->status, \ (flags_to_clear), #flags_to_clear); \ } while (0) enum remote_cmd_status { cmd_reported_success = (1 << 0), cmd_cancel = (1 << 1), }; typedef struct remote_ra_cmd_s { /*! the local node the cmd is issued from */ char *owner; /*! the remote node the cmd is executed on */ char *rsc_id; /*! the action to execute */ char *action; /*! some string the client wants us to give it back */ char *userdata; /*! start delay in ms */ int start_delay; /*! timer id used for start delay. */ int delay_id; /*! timeout in ms for cmd */ int timeout; int remaining_timeout; /*! recurring interval in ms */ guint interval_ms; /*! interval timer id */ int interval_id; int monitor_timeout_id; int takeover_timeout_id; /*! action parameters */ lrmd_key_value_t *params; pcmk__action_result_t result; int call_id; time_t start_time; uint32_t status; } remote_ra_cmd_t; #define lrm_remote_set_flags(lrm_state, flags_to_set) do { \ lrm_state_t *lrm = (lrm_state); \ remote_ra_data_t *ra = lrm->remote_ra_data; \ ra->status = pcmk__set_flags_as(__func__, __LINE__, LOG_TRACE, "Remote", \ lrm->node_name, ra->status, \ (flags_to_set), #flags_to_set); \ } while (0) #define lrm_remote_clear_flags(lrm_state, flags_to_clear) do { \ lrm_state_t *lrm = (lrm_state); \ remote_ra_data_t *ra = lrm->remote_ra_data; \ ra->status = pcmk__clear_flags_as(__func__, __LINE__, LOG_TRACE, "Remote", \ lrm->node_name, ra->status, \ (flags_to_clear), #flags_to_clear); \ } while (0) enum remote_status { expect_takeover = (1 << 0), takeover_complete = (1 << 1), remote_active = (1 << 2), /* Maintenance mode is difficult to determine from the controller's context, * so we have it signalled back with the transition from the scheduler. */ remote_in_maint = (1 << 3), /* Similar for whether we are controlling a guest node or remote node. * Fortunately there is a meta-attribute in the transition already and * as the situation doesn't change over time we can use the * resource start for noting down the information for later use when * the attributes aren't at hand. */ controlling_guest = (1 << 4), }; typedef struct remote_ra_data_s { crm_trigger_t *work; remote_ra_cmd_t *cur_cmd; GList *cmds; GList *recurring_cmds; uint32_t status; } remote_ra_data_t; static int handle_remote_ra_start(lrm_state_t * lrm_state, remote_ra_cmd_t * cmd, int timeout_ms); static void handle_remote_ra_stop(lrm_state_t * lrm_state, remote_ra_cmd_t * cmd); static GList *fail_all_monitor_cmds(GList * list); static void free_cmd(gpointer user_data) { remote_ra_cmd_t *cmd = user_data; if (!cmd) { return; } if (cmd->delay_id) { g_source_remove(cmd->delay_id); } if (cmd->interval_id) { g_source_remove(cmd->interval_id); } if (cmd->monitor_timeout_id) { g_source_remove(cmd->monitor_timeout_id); } if (cmd->takeover_timeout_id) { g_source_remove(cmd->takeover_timeout_id); } free(cmd->owner); free(cmd->rsc_id); free(cmd->action); free(cmd->userdata); pcmk__reset_result(&(cmd->result)); lrmd_key_value_freeall(cmd->params); free(cmd); } static int generate_callid(void) { static int remote_ra_callid = 0; remote_ra_callid++; if (remote_ra_callid <= 0) { remote_ra_callid = 1; } return remote_ra_callid; } static gboolean recurring_helper(gpointer data) { remote_ra_cmd_t *cmd = data; lrm_state_t *connection_rsc = NULL; cmd->interval_id = 0; connection_rsc = lrm_state_find(cmd->rsc_id); if (connection_rsc && connection_rsc->remote_ra_data) { remote_ra_data_t *ra_data = connection_rsc->remote_ra_data; ra_data->recurring_cmds = g_list_remove(ra_data->recurring_cmds, cmd); ra_data->cmds = g_list_append(ra_data->cmds, cmd); mainloop_set_trigger(ra_data->work); } return FALSE; } static gboolean start_delay_helper(gpointer data) { remote_ra_cmd_t *cmd = data; lrm_state_t *connection_rsc = NULL; cmd->delay_id = 0; connection_rsc = lrm_state_find(cmd->rsc_id); if (connection_rsc && connection_rsc->remote_ra_data) { remote_ra_data_t *ra_data = connection_rsc->remote_ra_data; mainloop_set_trigger(ra_data->work); } return FALSE; } /*! * \internal * \brief Handle cluster communication related to pacemaker_remote node joining * * \param[in] node_name Name of newly integrated pacemaker_remote node */ static void remote_node_up(const char *node_name) { int call_opt; xmlNode *update, *state; crm_node_t *node; enum controld_section_e section = controld_section_all; CRM_CHECK(node_name != NULL, return); crm_info("Announcing Pacemaker Remote node %s", node_name); /* Clear node's entire state (resource history and transient attributes) * other than shutdown locks. The transient attributes should and normally * will be cleared when the node leaves, but since remote node state has a * number of corner cases, clear them here as well, to be sure. */ call_opt = crmd_cib_smart_opt(); if (pcmk_is_set(controld_globals.flags, controld_shutdown_lock_enabled)) { section = controld_section_all_unlocked; } /* Purge node from attrd's memory */ update_attrd_remote_node_removed(node_name, NULL); controld_delete_node_state(node_name, section, call_opt); /* Delete node's probe_complete attribute. This serves two purposes: * * - @COMPAT DCs < 1.1.14 in a rolling upgrade might use it * - deleting it (or any attribute for that matter) here ensures the * attribute manager learns the node is remote */ update_attrd(node_name, CRM_OP_PROBED, NULL, NULL, TRUE); /* Ensure node is in the remote peer cache with member status */ node = crm_remote_peer_get(node_name); CRM_CHECK(node != NULL, return); pcmk__update_peer_state(__func__, node, CRM_NODE_MEMBER, 0); /* pacemaker_remote nodes don't participate in the membership layer, * so cluster nodes don't automatically get notified when they come and go. * We send a cluster message to the DC, and update the CIB node state entry, * so the DC will get it sooner (via message) or later (via CIB refresh), * and any other interested parties can query the CIB. */ send_remote_state_message(node_name, TRUE); update = create_xml_node(NULL, XML_CIB_TAG_STATUS); state = create_node_state_update(node, node_update_cluster, update, __func__); /* Clear the XML_NODE_IS_FENCED flag in the node state. If the node ever * needs to be fenced, this flag will allow various actions to determine * whether the fencing has happened yet. */ crm_xml_add(state, XML_NODE_IS_FENCED, "0"); /* TODO: If the remote connection drops, and this (async) CIB update either * failed or has not yet completed, later actions could mistakenly think the * node has already been fenced (if the XML_NODE_IS_FENCED attribute was * previously set, because it won't have been cleared). This could prevent * actual fencing or allow recurring monitor failures to be cleared too * soon. Ideally, we wouldn't rely on the CIB for the fenced status. */ controld_update_cib(XML_CIB_TAG_STATUS, update, call_opt, NULL); free_xml(update); } enum down_opts { DOWN_KEEP_LRM, DOWN_ERASE_LRM }; /*! * \internal * \brief Handle cluster communication related to pacemaker_remote node leaving * * \param[in] node_name Name of lost node * \param[in] opts Whether to keep or erase LRM history */ static void remote_node_down(const char *node_name, const enum down_opts opts) { xmlNode *update; int call_opt = crmd_cib_smart_opt(); crm_node_t *node; /* Purge node from attrd's memory */ update_attrd_remote_node_removed(node_name, NULL); /* Normally, only node attributes should be erased, and the resource history * should be kept until the node comes back up. However, after a successful * fence, we want to clear the history as well, so we don't think resources * are still running on the node. */ if (opts == DOWN_ERASE_LRM) { controld_delete_node_state(node_name, controld_section_all, call_opt); } else { controld_delete_node_state(node_name, controld_section_attrs, call_opt); } /* Ensure node is in the remote peer cache with lost state */ node = crm_remote_peer_get(node_name); CRM_CHECK(node != NULL, return); pcmk__update_peer_state(__func__, node, CRM_NODE_LOST, 0); /* Notify DC */ send_remote_state_message(node_name, FALSE); /* Update CIB node state */ update = create_xml_node(NULL, XML_CIB_TAG_STATUS); create_node_state_update(node, node_update_cluster, update, __func__); controld_update_cib(XML_CIB_TAG_STATUS, update, call_opt, NULL); free_xml(update); } /*! * \internal * \brief Handle effects of a remote RA command on node state * * \param[in] cmd Completed remote RA command */ static void check_remote_node_state(const remote_ra_cmd_t *cmd) { /* Only successful actions can change node state */ if (!pcmk__result_ok(&(cmd->result))) { return; } if (pcmk__str_eq(cmd->action, "start", pcmk__str_casei)) { remote_node_up(cmd->rsc_id); } else if (pcmk__str_eq(cmd->action, "migrate_from", pcmk__str_casei)) { /* After a successful migration, we don't need to do remote_node_up() * because the DC already knows the node is up, and we don't want to * clear LRM history etc. We do need to add the remote node to this * host's remote peer cache, because (unless it happens to be DC) * it hasn't been tracking the remote node, and other code relies on * the cache to distinguish remote nodes from unseen cluster nodes. */ crm_node_t *node = crm_remote_peer_get(cmd->rsc_id); CRM_CHECK(node != NULL, return); pcmk__update_peer_state(__func__, node, CRM_NODE_MEMBER, 0); } else if (pcmk__str_eq(cmd->action, "stop", pcmk__str_casei)) { lrm_state_t *lrm_state = lrm_state_find(cmd->rsc_id); remote_ra_data_t *ra_data = lrm_state? lrm_state->remote_ra_data : NULL; if (ra_data) { if (!pcmk_is_set(ra_data->status, takeover_complete)) { /* Stop means down if we didn't successfully migrate elsewhere */ remote_node_down(cmd->rsc_id, DOWN_KEEP_LRM); } else if (AM_I_DC == FALSE) { /* Only the connection host and DC track node state, * so if the connection migrated elsewhere and we aren't DC, * un-cache the node, so we don't have stale info */ crm_remote_peer_cache_remove(cmd->rsc_id); } } } /* We don't do anything for successful monitors, which is correct for * routine recurring monitors, and for monitors on nodes where the * connection isn't supposed to be (the cluster will stop the connection in * that case). However, if the initial probe finds the connection already * active on the node where we want it, we probably should do * remote_node_up(). Unfortunately, we can't distinguish that case here. * Given that connections have to be initiated by the cluster, the chance of * that should be close to zero. */ } static void report_remote_ra_result(remote_ra_cmd_t * cmd) { lrmd_event_data_t op = { 0, }; check_remote_node_state(cmd); op.type = lrmd_event_exec_complete; op.rsc_id = cmd->rsc_id; op.op_type = cmd->action; op.user_data = cmd->userdata; op.timeout = cmd->timeout; op.interval_ms = cmd->interval_ms; op.t_run = (unsigned int) cmd->start_time; op.t_rcchange = (unsigned int) cmd->start_time; lrmd__set_result(&op, cmd->result.exit_status, cmd->result.execution_status, cmd->result.exit_reason); if (pcmk_is_set(cmd->status, cmd_reported_success) && !pcmk__result_ok(&(cmd->result))) { op.t_rcchange = (unsigned int) time(NULL); /* This edge case will likely never ever occur, but if it does the * result is that a failure will not be processed correctly. This is only * remotely possible because we are able to detect a connection resource's tcp * connection has failed at any moment after start has completed. The actual * recurring operation is just a connectivity ping. * * basically, we are not guaranteed that the first successful monitor op and * a subsequent failed monitor op will not occur in the same timestamp. We have to * make it look like the operations occurred at separate times though. */ if (op.t_rcchange == op.t_run) { op.t_rcchange++; } } if (cmd->params) { lrmd_key_value_t *tmp; op.params = pcmk__strkey_table(free, free); for (tmp = cmd->params; tmp; tmp = tmp->next) { g_hash_table_insert(op.params, strdup(tmp->key), strdup(tmp->value)); } } op.call_id = cmd->call_id; op.remote_nodename = cmd->owner; lrm_op_callback(&op); if (op.params) { g_hash_table_destroy(op.params); } lrmd__reset_result(&op); } static void update_remaining_timeout(remote_ra_cmd_t * cmd) { cmd->remaining_timeout = ((cmd->timeout / 1000) - (time(NULL) - cmd->start_time)) * 1000; } static gboolean retry_start_cmd_cb(gpointer data) { lrm_state_t *lrm_state = data; remote_ra_data_t *ra_data = lrm_state->remote_ra_data; remote_ra_cmd_t *cmd = NULL; int rc = ETIME; if (!ra_data || !ra_data->cur_cmd) { return FALSE; } cmd = ra_data->cur_cmd; if (!pcmk__strcase_any_of(cmd->action, "start", "migrate_from", NULL)) { return FALSE; } update_remaining_timeout(cmd); if (cmd->remaining_timeout > 0) { rc = handle_remote_ra_start(lrm_state, cmd, cmd->remaining_timeout); } else { pcmk__set_result(&(cmd->result), PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_TIMEOUT, "Not enough time remains to retry remote connection"); } if (rc != pcmk_rc_ok) { report_remote_ra_result(cmd); if (ra_data->cmds) { mainloop_set_trigger(ra_data->work); } ra_data->cur_cmd = NULL; free_cmd(cmd); } else { /* wait for connection event */ } return FALSE; } static gboolean connection_takeover_timeout_cb(gpointer data) { lrm_state_t *lrm_state = NULL; remote_ra_cmd_t *cmd = data; crm_info("takeover event timed out for node %s", cmd->rsc_id); cmd->takeover_timeout_id = 0; lrm_state = lrm_state_find(cmd->rsc_id); handle_remote_ra_stop(lrm_state, cmd); free_cmd(cmd); return FALSE; } static gboolean monitor_timeout_cb(gpointer data) { lrm_state_t *lrm_state = NULL; remote_ra_cmd_t *cmd = data; lrm_state = lrm_state_find(cmd->rsc_id); crm_info("Timed out waiting for remote poke response from %s%s", cmd->rsc_id, (lrm_state? "" : " (no LRM state)")); cmd->monitor_timeout_id = 0; pcmk__set_result(&(cmd->result), PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_TIMEOUT, "Remote executor did not respond"); if (lrm_state && lrm_state->remote_ra_data) { remote_ra_data_t *ra_data = lrm_state->remote_ra_data; if (ra_data->cur_cmd == cmd) { ra_data->cur_cmd = NULL; } if (ra_data->cmds) { mainloop_set_trigger(ra_data->work); } } report_remote_ra_result(cmd); free_cmd(cmd); if(lrm_state) { lrm_state_disconnect(lrm_state); } return FALSE; } static void synthesize_lrmd_success(lrm_state_t *lrm_state, const char *rsc_id, const char *op_type) { lrmd_event_data_t op = { 0, }; if (lrm_state == NULL) { /* if lrm_state not given assume local */ lrm_state = lrm_state_find(controld_globals.our_nodename); } CRM_ASSERT(lrm_state != NULL); op.type = lrmd_event_exec_complete; op.rsc_id = rsc_id; op.op_type = op_type; op.t_run = (unsigned int) time(NULL); op.t_rcchange = op.t_run; op.call_id = generate_callid(); lrmd__set_result(&op, PCMK_OCF_OK, PCMK_EXEC_DONE, NULL); process_lrm_event(lrm_state, &op, NULL, NULL); } void remote_lrm_op_callback(lrmd_event_data_t * op) { gboolean cmd_handled = FALSE; lrm_state_t *lrm_state = NULL; remote_ra_data_t *ra_data = NULL; remote_ra_cmd_t *cmd = NULL; crm_debug("Processing '%s%s%s' event on remote connection to %s: %s " "(%d) status=%s (%d)", (op->op_type? op->op_type : ""), (op->op_type? " " : ""), lrmd_event_type2str(op->type), op->remote_nodename, services_ocf_exitcode_str(op->rc), op->rc, pcmk_exec_status_str(op->op_status), op->op_status); lrm_state = lrm_state_find(op->remote_nodename); if (!lrm_state || !lrm_state->remote_ra_data) { crm_debug("No state information found for remote connection event"); return; } ra_data = lrm_state->remote_ra_data; if (op->type == lrmd_event_new_client) { // Another client has connected to the remote daemon if (pcmk_is_set(ra_data->status, expect_takeover)) { // Great, we knew this was coming lrm_remote_clear_flags(lrm_state, expect_takeover); lrm_remote_set_flags(lrm_state, takeover_complete); } else { crm_err("Disconnecting from Pacemaker Remote node %s due to " "unexpected client takeover", op->remote_nodename); /* In this case, lrmd_tls_connection_destroy() will be called under the control of mainloop. */ /* Do not free lrm_state->conn yet. */ /* It'll be freed in the following stop action. */ lrm_state_disconnect_only(lrm_state); } return; } /* filter all EXEC events up */ if (op->type == lrmd_event_exec_complete) { if (pcmk_is_set(ra_data->status, takeover_complete)) { crm_debug("ignoring event, this connection is taken over by another node"); } else { lrm_op_callback(op); } return; } if ((op->type == lrmd_event_disconnect) && (ra_data->cur_cmd == NULL)) { if (!pcmk_is_set(ra_data->status, remote_active)) { crm_debug("Disconnection from Pacemaker Remote node %s complete", lrm_state->node_name); } else if (!remote_ra_is_in_maintenance(lrm_state)) { crm_err("Lost connection to Pacemaker Remote node %s", lrm_state->node_name); ra_data->recurring_cmds = fail_all_monitor_cmds(ra_data->recurring_cmds); ra_data->cmds = fail_all_monitor_cmds(ra_data->cmds); } else { crm_notice("Unmanaged Pacemaker Remote node %s disconnected", lrm_state->node_name); /* Do roughly what a 'stop' on the remote-resource would do */ handle_remote_ra_stop(lrm_state, NULL); remote_node_down(lrm_state->node_name, DOWN_KEEP_LRM); /* now fake the reply of a successful 'stop' */ synthesize_lrmd_success(NULL, lrm_state->node_name, "stop"); } return; } if (!ra_data->cur_cmd) { crm_debug("no event to match"); return; } cmd = ra_data->cur_cmd; /* Start actions and migrate from actions complete after connection * comes back to us. */ if (op->type == lrmd_event_connect && pcmk__strcase_any_of(cmd->action, "start", "migrate_from", NULL)) { if (op->connection_rc < 0) { update_remaining_timeout(cmd); if ((op->connection_rc == -ENOKEY) || (op->connection_rc == -EKEYREJECTED)) { // Hard error, don't retry pcmk__set_result(&(cmd->result), PCMK_OCF_INVALID_PARAM, PCMK_EXEC_ERROR, pcmk_strerror(op->connection_rc)); } else if (cmd->remaining_timeout > 3000) { crm_trace("rescheduling start, remaining timeout %d", cmd->remaining_timeout); g_timeout_add(1000, retry_start_cmd_cb, lrm_state); return; } else { crm_trace("can't reschedule start, remaining timeout too small %d", cmd->remaining_timeout); pcmk__format_result(&(cmd->result), PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_TIMEOUT, "%s without enough time to retry", pcmk_strerror(op->connection_rc)); } } else { lrm_state_reset_tables(lrm_state, TRUE); pcmk__set_result(&(cmd->result), PCMK_OCF_OK, PCMK_EXEC_DONE, NULL); lrm_remote_set_flags(lrm_state, remote_active); } crm_debug("Remote connection event matched %s action", cmd->action); report_remote_ra_result(cmd); cmd_handled = TRUE; } else if (op->type == lrmd_event_poke && pcmk__str_eq(cmd->action, "monitor", pcmk__str_casei)) { if (cmd->monitor_timeout_id) { g_source_remove(cmd->monitor_timeout_id); cmd->monitor_timeout_id = 0; } /* Only report success the first time, after that only worry about failures. * For this function, if we get the poke pack, it is always a success. Pokes * only fail if the send fails, or the response times out. */ if (!pcmk_is_set(cmd->status, cmd_reported_success)) { pcmk__set_result(&(cmd->result), PCMK_OCF_OK, PCMK_EXEC_DONE, NULL); report_remote_ra_result(cmd); cmd_set_flags(cmd, cmd_reported_success); } crm_debug("Remote poke event matched %s action", cmd->action); /* success, keep rescheduling if interval is present. */ if (cmd->interval_ms && !pcmk_is_set(cmd->status, cmd_cancel)) { ra_data->recurring_cmds = g_list_append(ra_data->recurring_cmds, cmd); cmd->interval_id = g_timeout_add(cmd->interval_ms, recurring_helper, cmd); cmd = NULL; /* prevent free */ } cmd_handled = TRUE; } else if (op->type == lrmd_event_disconnect && pcmk__str_eq(cmd->action, "monitor", pcmk__str_casei)) { if (pcmk_is_set(ra_data->status, remote_active) && !pcmk_is_set(cmd->status, cmd_cancel)) { pcmk__set_result(&(cmd->result), PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_ERROR, "Remote connection unexpectedly dropped " "during monitor"); report_remote_ra_result(cmd); crm_err("Remote connection to %s unexpectedly dropped during monitor", lrm_state->node_name); } cmd_handled = TRUE; } else if (op->type == lrmd_event_new_client && pcmk__str_eq(cmd->action, "stop", pcmk__str_casei)) { handle_remote_ra_stop(lrm_state, cmd); cmd_handled = TRUE; } else { crm_debug("Event did not match %s action", ra_data->cur_cmd->action); } if (cmd_handled) { ra_data->cur_cmd = NULL; if (ra_data->cmds) { mainloop_set_trigger(ra_data->work); } free_cmd(cmd); } } static void handle_remote_ra_stop(lrm_state_t * lrm_state, remote_ra_cmd_t * cmd) { remote_ra_data_t *ra_data = NULL; CRM_ASSERT(lrm_state); ra_data = lrm_state->remote_ra_data; if (!pcmk_is_set(ra_data->status, takeover_complete)) { /* delete pending ops when ever the remote connection is intentionally stopped */ g_hash_table_remove_all(lrm_state->active_ops); } else { /* we no longer hold the history if this connection has been migrated, * however, we keep metadata cache for future use */ lrm_state_reset_tables(lrm_state, FALSE); } lrm_remote_clear_flags(lrm_state, remote_active); lrm_state_disconnect(lrm_state); if (ra_data->cmds) { g_list_free_full(ra_data->cmds, free_cmd); } if (ra_data->recurring_cmds) { g_list_free_full(ra_data->recurring_cmds, free_cmd); } ra_data->cmds = NULL; ra_data->recurring_cmds = NULL; ra_data->cur_cmd = NULL; if (cmd) { pcmk__set_result(&(cmd->result), PCMK_OCF_OK, PCMK_EXEC_DONE, NULL); report_remote_ra_result(cmd); } } // \return Standard Pacemaker return code static int handle_remote_ra_start(lrm_state_t * lrm_state, remote_ra_cmd_t * cmd, int timeout_ms) { const char *server = NULL; lrmd_key_value_t *tmp = NULL; int port = 0; int timeout_used = timeout_ms > MAX_START_TIMEOUT_MS ? MAX_START_TIMEOUT_MS : timeout_ms; int rc = pcmk_rc_ok; for (tmp = cmd->params; tmp; tmp = tmp->next) { if (pcmk__strcase_any_of(tmp->key, XML_RSC_ATTR_REMOTE_RA_ADDR, XML_RSC_ATTR_REMOTE_RA_SERVER, NULL)) { server = tmp->value; } else if (pcmk__str_eq(tmp->key, XML_RSC_ATTR_REMOTE_RA_PORT, pcmk__str_casei)) { port = atoi(tmp->value); } else if (pcmk__str_eq(tmp->key, CRM_META "_" XML_RSC_ATTR_CONTAINER, pcmk__str_casei)) { lrm_remote_set_flags(lrm_state, controlling_guest); } } rc = controld_connect_remote_executor(lrm_state, server, port, timeout_used); if (rc != pcmk_rc_ok) { pcmk__format_result(&(cmd->result), PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_ERROR, "Could not connect to Pacemaker Remote node %s: %s", lrm_state->node_name, pcmk_rc_str(rc)); } return rc; } static gboolean handle_remote_ra_exec(gpointer user_data) { int rc = 0; lrm_state_t *lrm_state = user_data; remote_ra_data_t *ra_data = lrm_state->remote_ra_data; remote_ra_cmd_t *cmd; GList *first = NULL; if (ra_data->cur_cmd) { /* still waiting on previous cmd */ return TRUE; } while (ra_data->cmds) { first = ra_data->cmds; cmd = first->data; if (cmd->delay_id) { /* still waiting for start delay timer to trip */ return TRUE; } ra_data->cmds = g_list_remove_link(ra_data->cmds, first); g_list_free_1(first); if (!strcmp(cmd->action, "start") || !strcmp(cmd->action, "migrate_from")) { lrm_remote_clear_flags(lrm_state, expect_takeover | takeover_complete); if (handle_remote_ra_start(lrm_state, cmd, cmd->timeout) == pcmk_rc_ok) { /* take care of this later when we get async connection result */ crm_debug("Initiated async remote connection, %s action will complete after connect event", cmd->action); ra_data->cur_cmd = cmd; return TRUE; } report_remote_ra_result(cmd); } else if (!strcmp(cmd->action, "monitor")) { if (lrm_state_is_connected(lrm_state) == TRUE) { rc = lrm_state_poke_connection(lrm_state); if (rc < 0) { pcmk__set_result(&(cmd->result), PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_ERROR, pcmk_strerror(rc)); } } else { rc = -1; pcmk__set_result(&(cmd->result), PCMK_OCF_NOT_RUNNING, PCMK_EXEC_DONE, "Remote connection inactive"); } if (rc == 0) { crm_debug("Poked Pacemaker Remote at node %s, waiting for async response", cmd->rsc_id); ra_data->cur_cmd = cmd; cmd->monitor_timeout_id = g_timeout_add(cmd->timeout, monitor_timeout_cb, cmd); return TRUE; } report_remote_ra_result(cmd); } else if (!strcmp(cmd->action, "stop")) { if (pcmk_is_set(ra_data->status, expect_takeover)) { /* briefly wait on stop for the takeover event to occur. If the * takeover event does not occur during the wait period, that's fine. * It just means that the remote-node's lrm_status section is going to get * cleared which will require all the resources running in the remote-node * to be explicitly re-detected via probe actions. If the takeover does occur * successfully, then we can leave the status section intact. */ cmd->takeover_timeout_id = g_timeout_add((cmd->timeout/2), connection_takeover_timeout_cb, cmd); ra_data->cur_cmd = cmd; return TRUE; } handle_remote_ra_stop(lrm_state, cmd); } else if (!strcmp(cmd->action, "migrate_to")) { lrm_remote_clear_flags(lrm_state, takeover_complete); lrm_remote_set_flags(lrm_state, expect_takeover); pcmk__set_result(&(cmd->result), PCMK_OCF_OK, PCMK_EXEC_DONE, NULL); report_remote_ra_result(cmd); } else if (pcmk__str_any_of(cmd->action, CRMD_ACTION_RELOAD, CRMD_ACTION_RELOAD_AGENT, NULL)) { /* Currently the only reloadable parameter is reconnect_interval, * which is only used by the scheduler via the CIB, so reloads are a * no-op. * * @COMPAT DC <2.1.0: We only need to check for "reload" in case * we're in a rolling upgrade with a DC scheduling "reload" instead * of "reload-agent". An OCF 1.1 "reload" would be a no-op anyway, * so this would work for that purpose as well. */ pcmk__set_result(&(cmd->result), PCMK_OCF_OK, PCMK_EXEC_DONE, NULL); report_remote_ra_result(cmd); } free_cmd(cmd); } return TRUE; } static void remote_ra_data_init(lrm_state_t * lrm_state) { remote_ra_data_t *ra_data = NULL; if (lrm_state->remote_ra_data) { return; } ra_data = calloc(1, sizeof(remote_ra_data_t)); ra_data->work = mainloop_add_trigger(G_PRIORITY_HIGH, handle_remote_ra_exec, lrm_state); lrm_state->remote_ra_data = ra_data; } void remote_ra_cleanup(lrm_state_t * lrm_state) { remote_ra_data_t *ra_data = lrm_state->remote_ra_data; if (!ra_data) { return; } if (ra_data->cmds) { g_list_free_full(ra_data->cmds, free_cmd); } if (ra_data->recurring_cmds) { g_list_free_full(ra_data->recurring_cmds, free_cmd); } mainloop_destroy_trigger(ra_data->work); free(ra_data); lrm_state->remote_ra_data = NULL; } gboolean is_remote_lrmd_ra(const char *agent, const char *provider, const char *id) { if (agent && provider && !strcmp(agent, REMOTE_LRMD_RA) && !strcmp(provider, "pacemaker")) { return TRUE; } if ((id != NULL) && (lrm_state_find(id) != NULL) && !pcmk__str_eq(id, controld_globals.our_nodename, pcmk__str_casei)) { return TRUE; } return FALSE; } lrmd_rsc_info_t * remote_ra_get_rsc_info(lrm_state_t * lrm_state, const char *rsc_id) { lrmd_rsc_info_t *info = NULL; if ((lrm_state_find(rsc_id))) { info = calloc(1, sizeof(lrmd_rsc_info_t)); info->id = strdup(rsc_id); info->type = strdup(REMOTE_LRMD_RA); info->standard = strdup(PCMK_RESOURCE_CLASS_OCF); info->provider = strdup("pacemaker"); } return info; } static gboolean is_remote_ra_supported_action(const char *action) { return pcmk__str_any_of(action, CRMD_ACTION_START, CRMD_ACTION_STOP, CRMD_ACTION_STATUS, CRMD_ACTION_MIGRATE, CRMD_ACTION_MIGRATED, CRMD_ACTION_RELOAD_AGENT, CRMD_ACTION_RELOAD, NULL); } static GList * fail_all_monitor_cmds(GList * list) { GList *rm_list = NULL; remote_ra_cmd_t *cmd = NULL; GList *gIter = NULL; for (gIter = list; gIter != NULL; gIter = gIter->next) { cmd = gIter->data; if ((cmd->interval_ms > 0) && pcmk__str_eq(cmd->action, "monitor", pcmk__str_casei)) { rm_list = g_list_append(rm_list, cmd); } } for (gIter = rm_list; gIter != NULL; gIter = gIter->next) { cmd = gIter->data; pcmk__set_result(&(cmd->result), PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_ERROR, "Lost connection to remote executor"); crm_trace("Pre-emptively failing %s %s (interval=%u, %s)", cmd->action, cmd->rsc_id, cmd->interval_ms, cmd->userdata); report_remote_ra_result(cmd); list = g_list_remove(list, cmd); free_cmd(cmd); } /* frees only the list data, not the cmds */ g_list_free(rm_list); return list; } static GList * remove_cmd(GList * list, const char *action, guint interval_ms) { remote_ra_cmd_t *cmd = NULL; GList *gIter = NULL; for (gIter = list; gIter != NULL; gIter = gIter->next) { cmd = gIter->data; if ((cmd->interval_ms == interval_ms) && pcmk__str_eq(cmd->action, action, pcmk__str_casei)) { break; } cmd = NULL; } if (cmd) { list = g_list_remove(list, cmd); free_cmd(cmd); } return list; } int remote_ra_cancel(lrm_state_t *lrm_state, const char *rsc_id, const char *action, guint interval_ms) { lrm_state_t *connection_rsc = NULL; remote_ra_data_t *ra_data = NULL; connection_rsc = lrm_state_find(rsc_id); if (!connection_rsc || !connection_rsc->remote_ra_data) { return -EINVAL; } ra_data = connection_rsc->remote_ra_data; ra_data->cmds = remove_cmd(ra_data->cmds, action, interval_ms); ra_data->recurring_cmds = remove_cmd(ra_data->recurring_cmds, action, interval_ms); if (ra_data->cur_cmd && (ra_data->cur_cmd->interval_ms == interval_ms) && (pcmk__str_eq(ra_data->cur_cmd->action, action, pcmk__str_casei))) { cmd_set_flags(ra_data->cur_cmd, cmd_cancel); } return 0; } static remote_ra_cmd_t * handle_dup_monitor(remote_ra_data_t *ra_data, guint interval_ms, const char *userdata) { GList *gIter = NULL; remote_ra_cmd_t *cmd = NULL; /* there are 3 places a potential duplicate monitor operation * could exist. * 1. recurring_cmds list. where the op is waiting for its next interval * 2. cmds list, where the op is queued to get executed immediately * 3. cur_cmd, which means the monitor op is in flight right now. */ if (interval_ms == 0) { return NULL; } if (ra_data->cur_cmd && !pcmk_is_set(ra_data->cur_cmd->status, cmd_cancel) && (ra_data->cur_cmd->interval_ms == interval_ms) && pcmk__str_eq(ra_data->cur_cmd->action, "monitor", pcmk__str_casei)) { cmd = ra_data->cur_cmd; goto handle_dup; } for (gIter = ra_data->recurring_cmds; gIter != NULL; gIter = gIter->next) { cmd = gIter->data; if ((cmd->interval_ms == interval_ms) && pcmk__str_eq(cmd->action, "monitor", pcmk__str_casei)) { goto handle_dup; } } for (gIter = ra_data->cmds; gIter != NULL; gIter = gIter->next) { cmd = gIter->data; if ((cmd->interval_ms == interval_ms) && pcmk__str_eq(cmd->action, "monitor", pcmk__str_casei)) { goto handle_dup; } } return NULL; handle_dup: crm_trace("merging duplicate monitor cmd " PCMK__OP_FMT, cmd->rsc_id, "monitor", interval_ms); /* update the userdata */ if (userdata) { free(cmd->userdata); cmd->userdata = strdup(userdata); } /* if we've already reported success, generate a new call id */ if (pcmk_is_set(cmd->status, cmd_reported_success)) { cmd->start_time = time(NULL); cmd->call_id = generate_callid(); cmd_clear_flags(cmd, cmd_reported_success); } /* if we have an interval_id set, that means we are in the process of * waiting for this cmd's next interval. instead of waiting, cancel * the timer and execute the action immediately */ if (cmd->interval_id) { g_source_remove(cmd->interval_id); cmd->interval_id = 0; recurring_helper(cmd); } return cmd; } /*! * \internal * \brief Execute an action using the (internal) ocf:pacemaker:remote agent * * \param[in] lrm_state Executor state object for remote connection * \param[in] rsc_id Connection resource ID * \param[in] action Action to execute * \param[in] userdata String to copy and pass to execution callback * \param[in] interval_ms Action interval (in milliseconds) * \param[in] timeout_ms Action timeout (in milliseconds) * \param[in] start_delay_ms Delay (in milliseconds) before executing action * \param[in,out] params Connection resource parameters * \param[out] call_id Where to store call ID on success * * \return Standard Pacemaker return code * \note This takes ownership of \p params, which should not be used or freed * after calling this function. */ int controld_execute_remote_agent(const lrm_state_t *lrm_state, const char *rsc_id, const char *action, const char *userdata, guint interval_ms, int timeout_ms, int start_delay_ms, lrmd_key_value_t *params, int *call_id) { lrm_state_t *connection_rsc = NULL; remote_ra_cmd_t *cmd = NULL; remote_ra_data_t *ra_data = NULL; *call_id = 0; CRM_CHECK((lrm_state != NULL) && (rsc_id != NULL) && (action != NULL) && (userdata != NULL) && (call_id != NULL), lrmd_key_value_freeall(params); return EINVAL); if (!is_remote_ra_supported_action(action)) { lrmd_key_value_freeall(params); return EOPNOTSUPP; } connection_rsc = lrm_state_find(rsc_id); if (connection_rsc == NULL) { lrmd_key_value_freeall(params); return ENOTCONN; } remote_ra_data_init(connection_rsc); ra_data = connection_rsc->remote_ra_data; cmd = handle_dup_monitor(ra_data, interval_ms, userdata); if (cmd) { *call_id = cmd->call_id; lrmd_key_value_freeall(params); return pcmk_rc_ok; } cmd = calloc(1, sizeof(remote_ra_cmd_t)); if (cmd == NULL) { lrmd_key_value_freeall(params); return ENOMEM; } cmd->owner = strdup(lrm_state->node_name); cmd->rsc_id = strdup(rsc_id); cmd->action = strdup(action); cmd->userdata = strdup(userdata); if ((cmd->owner == NULL) || (cmd->rsc_id == NULL) || (cmd->action == NULL) || (cmd->userdata == NULL)) { free_cmd(cmd); lrmd_key_value_freeall(params); return ENOMEM; } cmd->interval_ms = interval_ms; cmd->timeout = timeout_ms; cmd->start_delay = start_delay_ms; cmd->params = params; cmd->start_time = time(NULL); cmd->call_id = generate_callid(); if (cmd->start_delay) { cmd->delay_id = g_timeout_add(cmd->start_delay, start_delay_helper, cmd); } ra_data->cmds = g_list_append(ra_data->cmds, cmd); mainloop_set_trigger(ra_data->work); *call_id = cmd->call_id; return pcmk_rc_ok; } /*! * \internal * \brief Immediately fail all monitors of a remote node, if proxied here * * \param[in] node_name Name of pacemaker_remote node */ void remote_ra_fail(const char *node_name) { lrm_state_t *lrm_state = lrm_state_find(node_name); if (lrm_state && lrm_state_is_connected(lrm_state)) { remote_ra_data_t *ra_data = lrm_state->remote_ra_data; crm_info("Failing monitors on Pacemaker Remote node %s", node_name); ra_data->recurring_cmds = fail_all_monitor_cmds(ra_data->recurring_cmds); ra_data->cmds = fail_all_monitor_cmds(ra_data->cmds); } } /* A guest node fencing implied by host fencing looks like: * * * * * * * */ -#define XPATH_PSEUDO_FENCE "//" XML_GRAPH_TAG_PSEUDO_EVENT \ +#define XPATH_PSEUDO_FENCE "/" XML_GRAPH_TAG_PSEUDO_EVENT \ "[@" XML_LRM_ATTR_TASK "='stonith']/" XML_GRAPH_TAG_DOWNED \ "/" XML_CIB_TAG_NODE /*! * \internal * \brief Check a pseudo-action for Pacemaker Remote node side effects * * \param[in,out] xml XML of pseudo-action to check */ void remote_ra_process_pseudo(xmlNode *xml) { xmlXPathObjectPtr search = xpath_search(xml, XPATH_PSEUDO_FENCE); if (numXpathResults(search) == 1) { xmlNode *result = getXpathResult(search, 0); /* Normally, we handle the necessary side effects of a guest node stop * action when reporting the remote agent's result. However, if the stop * is implied due to fencing, it will be a fencing pseudo-event, and * there won't be a result to report. Handle that case here. * * This will result in a duplicate call to remote_node_down() if the * guest stop was real instead of implied, but that shouldn't hurt. * * There is still one corner case that isn't handled: if a guest node * isn't running any resources when its host is fenced, it will appear * to be cleanly stopped, so there will be no pseudo-fence, and our * peer cache state will be incorrect unless and until the guest is * recovered. */ if (result) { const char *remote = ID(result); if (remote) { remote_node_down(remote, DOWN_ERASE_LRM); } } } freeXpathObject(search); } static void remote_ra_maintenance(lrm_state_t * lrm_state, gboolean maintenance) { xmlNode *update, *state; int call_opt; crm_node_t *node; call_opt = crmd_cib_smart_opt(); node = crm_remote_peer_get(lrm_state->node_name); CRM_CHECK(node != NULL, return); update = create_xml_node(NULL, XML_CIB_TAG_STATUS); state = create_node_state_update(node, node_update_none, update, __func__); crm_xml_add(state, XML_NODE_IS_MAINTENANCE, maintenance?"1":"0"); if (controld_update_cib(XML_CIB_TAG_STATUS, update, call_opt, NULL) == pcmk_rc_ok) { /* TODO: still not 100% sure that async update will succeed ... */ if (maintenance) { lrm_remote_set_flags(lrm_state, remote_in_maint); } else { lrm_remote_clear_flags(lrm_state, remote_in_maint); } } free_xml(update); } #define XPATH_PSEUDO_MAINTENANCE "//" XML_GRAPH_TAG_PSEUDO_EVENT \ "[@" XML_LRM_ATTR_TASK "='" CRM_OP_MAINTENANCE_NODES "']/" \ XML_GRAPH_TAG_MAINTENANCE /*! * \internal * \brief Check a pseudo-action holding updates for maintenance state * * \param[in,out] xml XML of pseudo-action to check */ void remote_ra_process_maintenance_nodes(xmlNode *xml) { xmlXPathObjectPtr search = xpath_search(xml, XPATH_PSEUDO_MAINTENANCE); if (numXpathResults(search) == 1) { xmlNode *node; int cnt = 0, cnt_remote = 0; for (node = first_named_child(getXpathResult(search, 0), XML_CIB_TAG_NODE); node != NULL; node = pcmk__xml_next(node)) { lrm_state_t *lrm_state = lrm_state_find(ID(node)); cnt++; if (lrm_state && lrm_state->remote_ra_data && pcmk_is_set(((remote_ra_data_t *) lrm_state->remote_ra_data)->status, remote_active)) { int is_maint; cnt_remote++; pcmk__scan_min_int(crm_element_value(node, XML_NODE_IS_MAINTENANCE), &is_maint, 0); remote_ra_maintenance(lrm_state, is_maint); } } crm_trace("Action holds %d nodes (%d remotes found) " "adjusting maintenance-mode", cnt, cnt_remote); } freeXpathObject(search); } gboolean remote_ra_is_in_maintenance(lrm_state_t * lrm_state) { remote_ra_data_t *ra_data = lrm_state->remote_ra_data; return pcmk_is_set(ra_data->status, remote_in_maint); } gboolean remote_ra_controlling_guest(lrm_state_t * lrm_state) { remote_ra_data_t *ra_data = lrm_state->remote_ra_data; return pcmk_is_set(ra_data->status, controlling_guest); } diff --git a/daemons/execd/pacemaker-execd.c b/daemons/execd/pacemaker-execd.c index ea93601023..c75663cf45 100644 --- a/daemons/execd/pacemaker-execd.c +++ b/daemons/execd/pacemaker-execd.c @@ -1,578 +1,579 @@ /* * Copyright 2012-2023 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include "pacemaker-execd.h" #ifdef PCMK__COMPILE_REMOTE # define EXECD_TYPE "remote" # define EXECD_NAME "pacemaker-remoted" # define SUMMARY "resource agent executor daemon for Pacemaker Remote nodes" #else # define EXECD_TYPE "local" # define EXECD_NAME "pacemaker-execd" # define SUMMARY "resource agent executor daemon for Pacemaker cluster nodes" #endif static GMainLoop *mainloop = NULL; static qb_ipcs_service_t *ipcs = NULL; static stonith_t *stonith_api = NULL; int lrmd_call_id = 0; static struct { gchar **log_files; #ifdef PCMK__COMPILE_REMOTE gchar *port; #endif // PCMK__COMPILE_REMOTE } options; #ifdef PCMK__COMPILE_REMOTE /* whether shutdown request has been sent */ static gboolean shutting_down = FALSE; /* timer for waiting for acknowledgment of shutdown request */ static 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; 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 == NULL) { int rc = pcmk_ok; stonith_api = stonith_api_new(); if (stonith_api == NULL) { crm_err("Could not connect to fencer: API memory allocation failed"); return NULL; } rc = stonith_api_connect_retry(stonith_api, crm_system_name, 10); if (rc != pcmk_ok) { crm_err("Could not connect to fencer in 10 attempts: %s " CRM_XS " rc=%d", pcmk_strerror(rc), rc); stonith_api_delete(stonith_api); stonith_api = NULL; } else { stonith_api->cmds->register_notification(stonith_api, T_STONITH_NOTIFY_DISCONNECT, stonith_connection_destroy_cb); } } 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 (pcmk__new_client(c, uid, gid) == NULL) { return -EIO; } return 0; } static void lrmd_ipc_created(qb_ipcs_connection_t * c) { pcmk__client_t *new_client = pcmk__find_client(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; pcmk__client_t *client = pcmk__find_client(c); xmlNode *request = pcmk__client_data2xml(client, data, &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 = pcmk__itoa(pcmk__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,out] client Client connection to free */ void lrmd_client_destroy(pcmk__client_t *client) { pcmk__free_client(client); #ifdef PCMK__COMPILE_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) { pcmk__client_t *client = pcmk__find_client(c); if (client == NULL) { return 0; } crm_trace("Connection %p", c); client_disconnect_cleanup(client->id); #ifdef PCMK__COMPILE_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 }; // \return Standard Pacemaker return code int lrmd_server_send_reply(pcmk__client_t *client, uint32_t id, xmlNode *reply) { crm_trace("Sending reply (%d) to client (%s)", id, client->id); switch (PCMK__CLIENT_TYPE(client)) { case pcmk__client_ipc: return pcmk__ipc_send_xml(client, id, reply, FALSE); #ifdef PCMK__COMPILE_REMOTE case pcmk__client_tls: return lrmd__remote_send_xml(client->remote, reply, id, "reply"); #endif default: crm_err("Could not send reply: unknown type for client %s " CRM_XS " flags=%#llx", pcmk__client_name(client), client->flags); } return ENOTCONN; } // \return Standard Pacemaker return code int lrmd_server_send_notify(pcmk__client_t *client, xmlNode *msg) { crm_trace("Sending notification to client (%s)", client->id); switch (PCMK__CLIENT_TYPE(client)) { case pcmk__client_ipc: if (client->ipcs == NULL) { crm_trace("Could not notify local client: disconnected"); return ENOTCONN; } return pcmk__ipc_send_xml(client, 0, msg, crm_ipc_server_event); #ifdef PCMK__COMPILE_REMOTE case pcmk__client_tls: if (client->remote == NULL) { crm_trace("Could not notify remote client: disconnected"); return ENOTCONN; } else { return lrmd__remote_send_xml(client->remote, msg, 0, "notify"); } #endif default: crm_err("Could not notify client %s with unknown transport " CRM_XS " flags=%#llx", pcmk__client_name(client), client->flags); } 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", pcmk__ipc_client_count()); 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 PCMK__COMPILE_REMOTE execd_stop_tls_server(); ipc_proxy_cleanup(); #endif pcmk__client_cleanup(); g_hash_table_destroy(rsc_list); if (mainloop) { lrmd_drain_alerts(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 PCMK__COMPILE_REMOTE pcmk__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 */ execd_stop_tls_server(); /* 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(void) { #ifdef PCMK__COMPILE_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(void) { #ifdef PCMK__COMPILE_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 GOptionEntry entries[] = { { "logfile", 'l', G_OPTION_FLAG_NONE, G_OPTION_ARG_FILENAME_ARRAY, &options.log_files, "Send logs to the additional named logfile", NULL }, #ifdef PCMK__COMPILE_REMOTE + { "port", 'p', G_OPTION_FLAG_NONE, G_OPTION_ARG_STRING, &options.port, - "Port to listen on", NULL }, + "Port to listen on (defaults to " G_STRINGIFY(DEFAULT_REMOTE_PORT) ")", NULL }, #endif // PCMK__COMPILE_REMOTE { NULL } }; static pcmk__supported_format_t formats[] = { PCMK__SUPPORTED_FORMAT_NONE, PCMK__SUPPORTED_FORMAT_TEXT, PCMK__SUPPORTED_FORMAT_XML, { NULL, NULL, NULL } }; static GOptionContext * build_arg_context(pcmk__common_args_t *args, GOptionGroup **group) { GOptionContext *context = NULL; context = pcmk__build_arg_context(args, "text (default), xml", group, NULL); pcmk__add_main_args(context, entries); return context; } int main(int argc, char **argv, char **envp) { int rc = pcmk_rc_ok; crm_exit_t exit_code = CRM_EX_OK; const char *option = NULL; pcmk__output_t *out = NULL; GError *error = NULL; GOptionGroup *output_group = NULL; pcmk__common_args_t *args = pcmk__new_common_args(SUMMARY); #ifdef PCMK__COMPILE_REMOTE gchar **processed_args = pcmk__cmdline_preproc(argv, "lp"); #else gchar **processed_args = pcmk__cmdline_preproc(argv, "l"); #endif // PCMK__COMPILE_REMOTE GOptionContext *context = build_arg_context(args, &output_group); #ifdef PCMK__COMPILE_REMOTE // If necessary, create PID 1 now before any file descriptors are opened remoted_spawn_pidone(argc, argv, envp); #endif crm_log_preinit(EXECD_NAME, argc, argv); pcmk__register_formats(output_group, formats); if (!g_option_context_parse_strv(context, &processed_args, &error)) { exit_code = CRM_EX_USAGE; goto done; } rc = pcmk__output_new(&out, args->output_ty, args->output_dest, argv); if (rc != pcmk_rc_ok) { exit_code = CRM_EX_ERROR; g_set_error(&error, PCMK__EXITC_ERROR, exit_code, "Error creating output format %s: %s", args->output_ty, pcmk_rc_str(rc)); goto done; } if (args->version) { out->version(out, false); goto done; } // Open additional log files if (options.log_files != NULL) { for (gchar **fname = options.log_files; *fname != NULL; fname++) { rc = pcmk__add_logfile(*fname); if (rc != pcmk_rc_ok) { out->err(out, "Logging to %s is disabled: %s", *fname, pcmk_rc_str(rc)); } } } pcmk__cli_init_logging(EXECD_NAME, args->verbosity); crm_log_init(NULL, LOG_INFO, TRUE, FALSE, argc, argv, FALSE); option = pcmk__env_option(PCMK__ENV_LOGFACILITY); if (!pcmk__str_eq(option, PCMK__VALUE_NONE, pcmk__str_casei|pcmk__str_null_matches) && !pcmk__str_eq(option, "/dev/null", pcmk__str_none)) { setenv("HA_LOGFACILITY", option, 1); /* Used by the ocf_log/ha_log OCF macro */ } option = pcmk__env_option(PCMK__ENV_LOGFILE); if (!pcmk__str_eq(option, PCMK__VALUE_NONE, pcmk__str_casei|pcmk__str_null_matches)) { setenv("HA_LOGFILE", option, 1); /* Used by the ocf_log/ha_log OCF macro */ if (pcmk__env_option_enabled(crm_system_name, PCMK__ENV_DEBUG)) { setenv("HA_DEBUGLOG", option, 1); /* Used by the ocf_log/ha_debug OCF macro */ } } #ifdef PCMK__COMPILE_REMOTE if (options.port != NULL) { setenv("PCMK_remote_port", options.port, 1); } #endif // PCMK__COMPILE_REMOTE crm_notice("Starting Pacemaker " EXECD_TYPE " executor"); /* The presence of this variable allegedly controls whether child * processes like httpd will try and use Systemd's sd_notify * API */ unsetenv("NOTIFY_SOCKET"); { // Temporary directory for resource agent use (leave owned by root) int rc = pcmk__build_path(CRM_RSCTMP_DIR, 0755); if (rc != pcmk_rc_ok) { crm_warn("Could not create resource agent temporary directory " CRM_RSCTMP_DIR ": %s", pcmk_rc_str(rc)); } } rsc_list = pcmk__strkey_table(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"); exit_code = CRM_EX_FATAL; goto done; } #ifdef PCMK__COMPILE_REMOTE if (lrmd_init_remote_tls_server() < 0) { crm_err("Failed to create TLS listener: shutting down and staying down"); exit_code = CRM_EX_FATAL; goto done; } ipc_proxy_init(); #endif mainloop_add_signal(SIGTERM, lrmd_shutdown); mainloop = g_main_loop_new(NULL, FALSE); crm_notice("Pacemaker " EXECD_TYPE " executor successfully started and accepting connections"); crm_notice("OCF resource agent search path is %s", OCF_RA_PATH); g_main_loop_run(mainloop); /* should never get here */ lrmd_exit(NULL); done: g_strfreev(options.log_files); #ifdef PCMK__COMPILE_REMOTE g_free(options.port); #endif // PCMK__COMPILE_REMOTE g_strfreev(processed_args); pcmk__free_arg_context(context); pcmk__output_and_clear_error(&error, out); if (out != NULL) { out->finish(out, exit_code, true, NULL); pcmk__output_free(out); } pcmk__unregister_formats(); crm_exit(exit_code); } diff --git a/lib/common/ipc_controld.c b/lib/common/ipc_controld.c index b6672fc643..9303afd895 100644 --- a/lib/common/ipc_controld.c +++ b/lib/common/ipc_controld.c @@ -1,671 +1,671 @@ /* * Copyright 2020-2022 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include #include #include #include "crmcommon_private.h" struct controld_api_private_s { char *client_uuid; unsigned int replies_expected; }; /*! * \internal * \brief Get a string representation of a controller API reply type * * \param[in] reply Controller API reply type * * \return String representation of a controller API reply type */ const char * pcmk__controld_api_reply2str(enum pcmk_controld_api_reply reply) { switch (reply) { case pcmk_controld_reply_reprobe: return "reprobe"; case pcmk_controld_reply_info: return "info"; case pcmk_controld_reply_resource: return "resource"; case pcmk_controld_reply_ping: return "ping"; case pcmk_controld_reply_nodes: return "nodes"; default: return "unknown"; } } // \return Standard Pacemaker return code static int new_data(pcmk_ipc_api_t *api) { struct controld_api_private_s *private = NULL; api->api_data = calloc(1, sizeof(struct controld_api_private_s)); if (api->api_data == NULL) { return errno; } private = api->api_data; /* This is set to the PID because that's how it was always done, but PIDs * are not unique because clients can be remote. The value appears to be * unused other than as part of F_CRM_SYS_FROM in IPC requests, which is * only compared against the internal system names (CRM_SYSTEM_TENGINE, * etc.), so it shouldn't be a problem. */ private->client_uuid = pcmk__getpid_s(); /* @TODO Implement a call ID model similar to the CIB, executor, and fencer * IPC APIs, so that requests and replies can be matched, and * duplicate replies can be discarded. */ return pcmk_rc_ok; } static void free_data(void *data) { free(((struct controld_api_private_s *) data)->client_uuid); free(data); } // \return Standard Pacemaker return code static int post_connect(pcmk_ipc_api_t *api) { /* The controller currently requires clients to register via a hello * request, but does not reply back. */ struct controld_api_private_s *private = api->api_data; const char *client_name = crm_system_name? crm_system_name : "client"; xmlNode *hello; int rc; hello = create_hello_message(private->client_uuid, client_name, PCMK__CONTROLD_API_MAJOR, PCMK__CONTROLD_API_MINOR); rc = pcmk__send_ipc_request(api, hello); free_xml(hello); if (rc != pcmk_rc_ok) { crm_info("Could not send IPC hello to %s: %s " CRM_XS " rc=%s", pcmk_ipc_name(api, true), pcmk_rc_str(rc), rc); } else { crm_debug("Sent IPC hello to %s", pcmk_ipc_name(api, true)); } return rc; } static void set_node_info_data(pcmk_controld_api_reply_t *data, xmlNode *msg_data) { data->reply_type = pcmk_controld_reply_info; if (msg_data == NULL) { return; } data->data.node_info.have_quorum = pcmk__xe_attr_is_true(msg_data, XML_ATTR_HAVE_QUORUM); data->data.node_info.is_remote = pcmk__xe_attr_is_true(msg_data, XML_NODE_IS_REMOTE); /* Integer node_info.id is currently valid only for Corosync nodes. * - * @TODO: Improve handling after crm_peer_t is refactored to handle layer- + * @TODO: Improve handling after crm_node_t is refactored to handle layer- * specific data better. */ crm_element_value_int(msg_data, XML_ATTR_ID, &(data->data.node_info.id)); data->data.node_info.uuid = crm_element_value(msg_data, XML_ATTR_ID); data->data.node_info.uname = crm_element_value(msg_data, XML_ATTR_UNAME); data->data.node_info.state = crm_element_value(msg_data, XML_NODE_IS_PEER); } static void set_ping_data(pcmk_controld_api_reply_t *data, xmlNode *msg_data) { data->reply_type = pcmk_controld_reply_ping; if (msg_data == NULL) { return; } data->data.ping.sys_from = crm_element_value(msg_data, XML_PING_ATTR_SYSFROM); data->data.ping.fsa_state = crm_element_value(msg_data, XML_PING_ATTR_CRMDSTATE); data->data.ping.result = crm_element_value(msg_data, XML_PING_ATTR_STATUS); } static void set_nodes_data(pcmk_controld_api_reply_t *data, xmlNode *msg_data) { pcmk_controld_api_node_t *node_info; data->reply_type = pcmk_controld_reply_nodes; for (xmlNode *node = first_named_child(msg_data, XML_CIB_TAG_NODE); node != NULL; node = crm_next_same_xml(node)) { long long id_ll = 0; node_info = calloc(1, sizeof(pcmk_controld_api_node_t)); crm_element_value_ll(node, XML_ATTR_ID, &id_ll); if (id_ll > 0) { node_info->id = id_ll; } node_info->uname = crm_element_value(node, XML_ATTR_UNAME); node_info->state = crm_element_value(node, XML_NODE_IN_CLUSTER); data->data.nodes = g_list_prepend(data->data.nodes, node_info); } } static bool reply_expected(pcmk_ipc_api_t *api, xmlNode *request) { const char *command = crm_element_value(request, F_CRM_TASK); if (command == NULL) { return false; } // We only need to handle commands that functions in this file can send return !strcmp(command, CRM_OP_REPROBE) || !strcmp(command, CRM_OP_NODE_INFO) || !strcmp(command, CRM_OP_PING) || !strcmp(command, CRM_OP_LRM_FAIL) || !strcmp(command, CRM_OP_LRM_DELETE); } static bool dispatch(pcmk_ipc_api_t *api, xmlNode *reply) { struct controld_api_private_s *private = api->api_data; crm_exit_t status = CRM_EX_OK; xmlNode *msg_data = NULL; const char *value = NULL; pcmk_controld_api_reply_t reply_data = { pcmk_controld_reply_unknown, NULL, NULL, }; /* If we got an ACK, return true so the caller knows to expect more responses * from the IPC server. We do this before decrementing replies_expected because * ACKs are not going to be included in that value. * * Note that we cannot do the same kind of status checking here that we do in * ipc_pacemakerd.c. The ACK message we receive does not necessarily contain * a status attribute. That is, we may receive this: * * * * Instead of this: * * */ if (pcmk__str_eq(crm_element_name(reply), "ack", pcmk__str_none)) { return true; // More replies needed } if (private->replies_expected > 0) { private->replies_expected--; } // Do some basic validation of the reply /* @TODO We should be able to verify that value is always a response, but * currently the controller doesn't always properly set the type. Even * if we fix the controller, we'll still need to handle replies from * old versions (feature set could be used to differentiate). */ value = crm_element_value(reply, F_CRM_MSG_TYPE); if (pcmk__str_empty(value) || !pcmk__str_any_of(value, XML_ATTR_REQUEST, XML_ATTR_RESPONSE, NULL)) { crm_info("Unrecognizable message from controller: " "invalid message type '%s'", pcmk__s(value, "")); status = CRM_EX_PROTOCOL; goto done; } if (pcmk__str_empty(crm_element_value(reply, XML_ATTR_REFERENCE))) { crm_info("Unrecognizable message from controller: no reference"); status = CRM_EX_PROTOCOL; goto done; } value = crm_element_value(reply, F_CRM_TASK); if (pcmk__str_empty(value)) { crm_info("Unrecognizable message from controller: no command name"); status = CRM_EX_PROTOCOL; goto done; } // Parse useful info from reply reply_data.feature_set = crm_element_value(reply, XML_ATTR_VERSION); reply_data.host_from = crm_element_value(reply, F_CRM_HOST_FROM); msg_data = get_message_xml(reply, F_CRM_DATA); if (!strcmp(value, CRM_OP_REPROBE)) { reply_data.reply_type = pcmk_controld_reply_reprobe; } else if (!strcmp(value, CRM_OP_NODE_INFO)) { set_node_info_data(&reply_data, msg_data); } else if (!strcmp(value, CRM_OP_INVOKE_LRM)) { reply_data.reply_type = pcmk_controld_reply_resource; reply_data.data.resource.node_state = msg_data; } else if (!strcmp(value, CRM_OP_PING)) { set_ping_data(&reply_data, msg_data); } else if (!strcmp(value, PCMK__CONTROLD_CMD_NODES)) { set_nodes_data(&reply_data, msg_data); } else { crm_info("Unrecognizable message from controller: unknown command '%s'", value); status = CRM_EX_PROTOCOL; } done: pcmk__call_ipc_callback(api, pcmk_ipc_event_reply, status, &reply_data); // Free any reply data that was allocated if (pcmk__str_eq(value, PCMK__CONTROLD_CMD_NODES, pcmk__str_casei)) { g_list_free_full(reply_data.data.nodes, free); } return false; // No further replies needed } pcmk__ipc_methods_t * pcmk__controld_api_methods(void) { pcmk__ipc_methods_t *cmds = calloc(1, sizeof(pcmk__ipc_methods_t)); if (cmds != NULL) { cmds->new_data = new_data; cmds->free_data = free_data; cmds->post_connect = post_connect; cmds->reply_expected = reply_expected; cmds->dispatch = dispatch; } return cmds; } /*! * \internal * \brief Create XML for a controller IPC request * * \param[in] api Controller connection * \param[in] op Controller IPC command name * \param[in] node Node name to set as destination host * \param[in] msg_data XML to attach to request as message data * * \return Newly allocated XML for request */ static xmlNode * create_controller_request(const pcmk_ipc_api_t *api, const char *op, const char *node, xmlNode *msg_data) { struct controld_api_private_s *private = NULL; const char *sys_to = NULL; if (api == NULL) { return NULL; } private = api->api_data; if ((node == NULL) && !strcmp(op, CRM_OP_PING)) { sys_to = CRM_SYSTEM_DC; } else { sys_to = CRM_SYSTEM_CRMD; } return create_request(op, msg_data, node, sys_to, (crm_system_name? crm_system_name : "client"), private->client_uuid); } // \return Standard Pacemaker return code static int send_controller_request(pcmk_ipc_api_t *api, xmlNode *request, bool reply_is_expected) { int rc; if (crm_element_value(request, XML_ATTR_REFERENCE) == NULL) { return EINVAL; } rc = pcmk__send_ipc_request(api, request); if ((rc == pcmk_rc_ok) && reply_is_expected) { struct controld_api_private_s *private = api->api_data; private->replies_expected++; } return rc; } static xmlNode * create_reprobe_message_data(const char *target_node, const char *router_node) { xmlNode *msg_data; msg_data = create_xml_node(NULL, "data_for_" CRM_OP_REPROBE); crm_xml_add(msg_data, XML_LRM_ATTR_TARGET, target_node); if ((router_node != NULL) && !pcmk__str_eq(router_node, target_node, pcmk__str_casei)) { crm_xml_add(msg_data, XML_LRM_ATTR_ROUTER_NODE, router_node); } return msg_data; } /*! * \brief Send a reprobe controller operation * * \param[in,out] api Controller connection * \param[in] target_node Name of node to reprobe * \param[in] router_node Router node for host * * \return Standard Pacemaker return code * \note Event callback will get a reply of type pcmk_controld_reply_reprobe. */ int pcmk_controld_api_reprobe(pcmk_ipc_api_t *api, const char *target_node, const char *router_node) { xmlNode *request; xmlNode *msg_data; int rc = pcmk_rc_ok; if (api == NULL) { return EINVAL; } if (router_node == NULL) { router_node = target_node; } crm_debug("Sending %s IPC request to reprobe %s via %s", pcmk_ipc_name(api, true), pcmk__s(target_node, "local node"), pcmk__s(router_node, "local node")); msg_data = create_reprobe_message_data(target_node, router_node); request = create_controller_request(api, CRM_OP_REPROBE, router_node, msg_data); rc = send_controller_request(api, request, true); free_xml(msg_data); free_xml(request); return rc; } /*! * \brief Send a "node info" controller operation * * \param[in,out] api Controller connection * \param[in] nodeid ID of node to get info for (or 0 for local node) * * \return Standard Pacemaker return code * \note Event callback will get a reply of type pcmk_controld_reply_info. */ int pcmk_controld_api_node_info(pcmk_ipc_api_t *api, uint32_t nodeid) { xmlNode *request; int rc = pcmk_rc_ok; request = create_controller_request(api, CRM_OP_NODE_INFO, NULL, NULL); if (request == NULL) { return EINVAL; } if (nodeid > 0) { crm_xml_set_id(request, "%lu", (unsigned long) nodeid); } rc = send_controller_request(api, request, true); free_xml(request); return rc; } /*! * \brief Ask the controller for status * * \param[in,out] api Controller connection * \param[in] node_name Name of node whose status is desired (NULL for DC) * * \return Standard Pacemaker return code * \note Event callback will get a reply of type pcmk_controld_reply_ping. */ int pcmk_controld_api_ping(pcmk_ipc_api_t *api, const char *node_name) { xmlNode *request; int rc = pcmk_rc_ok; request = create_controller_request(api, CRM_OP_PING, node_name, NULL); if (request == NULL) { return EINVAL; } rc = send_controller_request(api, request, true); free_xml(request); return rc; } /*! * \brief Ask the controller for cluster information * * \param[in,out] api Controller connection * * \return Standard Pacemaker return code * \note Event callback will get a reply of type pcmk_controld_reply_nodes. */ int pcmk_controld_api_list_nodes(pcmk_ipc_api_t *api) { xmlNode *request; int rc = EINVAL; request = create_controller_request(api, PCMK__CONTROLD_CMD_NODES, NULL, NULL); if (request != NULL) { rc = send_controller_request(api, request, true); free_xml(request); } return rc; } // \return Standard Pacemaker return code static int controller_resource_op(pcmk_ipc_api_t *api, const char *op, const char *target_node, const char *router_node, bool cib_only, const char *rsc_id, const char *rsc_long_id, const char *standard, const char *provider, const char *type) { int rc = pcmk_rc_ok; char *key; xmlNode *request, *msg_data, *xml_rsc, *params; if (api == NULL) { return EINVAL; } if (router_node == NULL) { router_node = target_node; } msg_data = create_xml_node(NULL, XML_GRAPH_TAG_RSC_OP); /* The controller logs the transition key from resource op requests, so we * need to have *something* for it. * @TODO don't use "crm-resource" */ key = pcmk__transition_key(0, getpid(), 0, "xxxxxxxx-xrsc-opxx-xcrm-resourcexxxx"); crm_xml_add(msg_data, XML_ATTR_TRANSITION_KEY, key); free(key); crm_xml_add(msg_data, XML_LRM_ATTR_TARGET, target_node); if (!pcmk__str_eq(router_node, target_node, pcmk__str_casei)) { crm_xml_add(msg_data, XML_LRM_ATTR_ROUTER_NODE, router_node); } if (cib_only) { // Indicate that only the CIB needs to be cleaned crm_xml_add(msg_data, PCMK__XA_MODE, XML_TAG_CIB); } xml_rsc = create_xml_node(msg_data, XML_CIB_TAG_RESOURCE); crm_xml_add(xml_rsc, XML_ATTR_ID, rsc_id); crm_xml_add(xml_rsc, XML_ATTR_ID_LONG, rsc_long_id); crm_xml_add(xml_rsc, XML_AGENT_ATTR_CLASS, standard); crm_xml_add(xml_rsc, XML_AGENT_ATTR_PROVIDER, provider); crm_xml_add(xml_rsc, XML_ATTR_TYPE, type); params = create_xml_node(msg_data, XML_TAG_ATTRS); crm_xml_add(params, XML_ATTR_CRM_VERSION, CRM_FEATURE_SET); // The controller parses the timeout from the request key = crm_meta_name(XML_ATTR_TIMEOUT); crm_xml_add(params, key, "60000"); /* 1 minute */ //@TODO pass as arg free(key); request = create_controller_request(api, op, router_node, msg_data); rc = send_controller_request(api, request, true); free_xml(msg_data); free_xml(request); return rc; } /*! * \brief Ask the controller to fail a resource * * \param[in,out] api Controller connection * \param[in] target_node Name of node resource is on * \param[in] router_node Router node for target * \param[in] rsc_id ID of resource to fail * \param[in] rsc_long_id Long ID of resource (if any) * \param[in] standard Standard of resource * \param[in] provider Provider of resource (if any) * \param[in] type Type of resource to fail * * \return Standard Pacemaker return code * \note Event callback will get a reply of type pcmk_controld_reply_resource. */ int pcmk_controld_api_fail(pcmk_ipc_api_t *api, const char *target_node, const char *router_node, const char *rsc_id, const char *rsc_long_id, const char *standard, const char *provider, const char *type) { crm_debug("Sending %s IPC request to fail %s (a.k.a. %s) on %s via %s", pcmk_ipc_name(api, true), pcmk__s(rsc_id, "unknown resource"), pcmk__s(rsc_long_id, "no other names"), pcmk__s(target_node, "unspecified node"), pcmk__s(router_node, "unspecified node")); return controller_resource_op(api, CRM_OP_LRM_FAIL, target_node, router_node, false, rsc_id, rsc_long_id, standard, provider, type); } /*! * \brief Ask the controller to refresh a resource * * \param[in,out] api Controller connection * \param[in] target_node Name of node resource is on * \param[in] router_node Router node for target * \param[in] rsc_id ID of resource to refresh * \param[in] rsc_long_id Long ID of resource (if any) * \param[in] standard Standard of resource * \param[in] provider Provider of resource (if any) * \param[in] type Type of resource * \param[in] cib_only If true, clean resource from CIB only * * \return Standard Pacemaker return code * \note Event callback will get a reply of type pcmk_controld_reply_resource. */ int pcmk_controld_api_refresh(pcmk_ipc_api_t *api, const char *target_node, const char *router_node, const char *rsc_id, const char *rsc_long_id, const char *standard, const char *provider, const char *type, bool cib_only) { crm_debug("Sending %s IPC request to refresh %s (a.k.a. %s) on %s via %s", pcmk_ipc_name(api, true), pcmk__s(rsc_id, "unknown resource"), pcmk__s(rsc_long_id, "no other names"), pcmk__s(target_node, "unspecified node"), pcmk__s(router_node, "unspecified node")); return controller_resource_op(api, CRM_OP_LRM_DELETE, target_node, router_node, cib_only, rsc_id, rsc_long_id, standard, provider, type); } /*! * \brief Get the number of IPC replies currently expected from the controller * * \param[in] api Controller IPC API connection * * \return Number of replies expected */ unsigned int pcmk_controld_api_replies_expected(const pcmk_ipc_api_t *api) { struct controld_api_private_s *private = api->api_data; return private->replies_expected; } /*! * \brief Create XML for a controller IPC "hello" message * * \deprecated This function is deprecated as part of the public C API. */ // \todo make this static to this file when breaking API backward compatibility xmlNode * create_hello_message(const char *uuid, const char *client_name, const char *major_version, const char *minor_version) { xmlNode *hello_node = NULL; xmlNode *hello = NULL; if (pcmk__str_empty(uuid) || pcmk__str_empty(client_name) || pcmk__str_empty(major_version) || pcmk__str_empty(minor_version)) { crm_err("Could not create IPC hello message from %s (UUID %s): " "missing information", client_name? client_name : "unknown client", uuid? uuid : "unknown"); return NULL; } hello_node = create_xml_node(NULL, XML_TAG_OPTIONS); if (hello_node == NULL) { crm_err("Could not create IPC hello message from %s (UUID %s): " "Message data creation failed", client_name, uuid); return NULL; } crm_xml_add(hello_node, "major_version", major_version); crm_xml_add(hello_node, "minor_version", minor_version); crm_xml_add(hello_node, "client_name", client_name); crm_xml_add(hello_node, "client_uuid", uuid); hello = create_request(CRM_OP_HELLO, hello_node, NULL, NULL, client_name, uuid); if (hello == NULL) { crm_err("Could not create IPC hello message from %s (UUID %s): " "Request creation failed", client_name, uuid); return NULL; } free_xml(hello_node); crm_trace("Created hello message from %s (UUID %s)", client_name, uuid); return hello; }