diff --git a/daemons/attrd/attrd_messages.c b/daemons/attrd/attrd_messages.c index 2e104dc26e..ce69a138c3 100644 --- a/daemons/attrd/attrd_messages.c +++ b/daemons/attrd/attrd_messages.c @@ -1,345 +1,345 @@ /* * Copyright 2022-2024 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 // pcmk__get_node() #include #include "pacemaker-attrd.h" int minimum_protocol_version = -1; static GHashTable *attrd_handlers = NULL; static bool is_sync_point_attr(xmlAttrPtr attr, void *data) { return pcmk__str_eq((const char *) attr->name, PCMK__XA_ATTR_SYNC_POINT, pcmk__str_none); } static int remove_sync_point_attribute(xmlNode *xml, void *data) { pcmk__xe_remove_matching_attrs(xml, is_sync_point_attr, NULL); pcmk__xe_foreach_child(xml, PCMK_XE_OP, remove_sync_point_attribute, NULL); return pcmk_rc_ok; } /* Sync points on a multi-update IPC message to an attrd too old to support * multi-update messages won't work. Strip the sync point attribute off here * so we don't pretend to support this situation and instead ACK the client * immediately. */ static void remove_unsupported_sync_points(pcmk__request_t *request) { if (request->xml->children != NULL && !ATTRD_SUPPORTS_MULTI_MESSAGE(minimum_protocol_version) && attrd_request_has_sync_point(request->xml)) { crm_warn("Ignoring sync point in request from %s because not all nodes support it", pcmk__request_origin(request)); remove_sync_point_attribute(request->xml, NULL); } } static xmlNode * handle_unknown_request(pcmk__request_t *request) { crm_err("Unknown IPC request %s from %s %s", request->op, pcmk__request_origin_type(request), pcmk__request_origin(request)); pcmk__format_result(&request->result, CRM_EX_PROTOCOL, PCMK_EXEC_INVALID, "Unknown request type '%s' (bug?)", request->op); return NULL; } static xmlNode * handle_clear_failure_request(pcmk__request_t *request) { if (request->peer != NULL) { /* It is not currently possible to receive this as a peer command, * but will be, if we one day enable propagating this operation. */ attrd_peer_clear_failure(request); pcmk__set_result(&request->result, CRM_EX_OK, PCMK_EXEC_DONE, NULL); return NULL; } else { remove_unsupported_sync_points(request); if (attrd_request_has_sync_point(request->xml)) { /* If this client supplied a sync point it wants to wait for, add it to * the wait list. Clients on this list will not receive an ACK until * their sync point is hit which will result in the client stalled there * until it receives a response. * * All other clients will receive the expected response as normal. */ attrd_add_client_to_waitlist(request); } else { /* If the client doesn't want to wait for a sync point, go ahead and send * the ACK immediately. Otherwise, we'll send the ACK when the appropriate * sync point is reached. */ attrd_send_ack(request->ipc_client, request->ipc_id, request->ipc_flags); } return attrd_client_clear_failure(request); } } static xmlNode * handle_confirm_request(pcmk__request_t *request) { if (request->peer != NULL) { int callid; crm_debug("Received confirmation from %s", request->peer); if (crm_element_value_int(request->xml, PCMK__XA_CALL_ID, &callid) == -1) { pcmk__set_result(&request->result, CRM_EX_PROTOCOL, PCMK_EXEC_INVALID, "Could not get callid from XML"); } else { attrd_handle_confirmation(callid, request->peer); } pcmk__set_result(&request->result, CRM_EX_OK, PCMK_EXEC_DONE, NULL); return NULL; } else { return handle_unknown_request(request); } } static xmlNode * handle_query_request(pcmk__request_t *request) { if (request->peer != NULL) { return handle_unknown_request(request); } else { return attrd_client_query(request); } } static xmlNode * handle_remove_request(pcmk__request_t *request) { if (request->peer != NULL) { const char *host = crm_element_value(request->xml, PCMK__XA_ATTR_HOST); bool reap = false; if (pcmk__xe_get_bool_attr(request->xml, PCMK__XA_REAP, &reap) != pcmk_rc_ok) { reap = true; // Default to true for backward compatibility } attrd_peer_remove(host, reap, request->peer); pcmk__set_result(&request->result, CRM_EX_OK, PCMK_EXEC_DONE, NULL); return NULL; } else { return attrd_client_peer_remove(request); } } static xmlNode * handle_refresh_request(pcmk__request_t *request) { if (request->peer != NULL) { return handle_unknown_request(request); } else { return attrd_client_refresh(request); } } static xmlNode * handle_sync_response_request(pcmk__request_t *request) { if (request->ipc_client != NULL) { return handle_unknown_request(request); } else { if (request->peer != NULL) { pcmk__node_status_t *peer = pcmk__get_node(0, request->peer, NULL, pcmk__node_search_cluster_member); bool peer_won = attrd_check_for_new_writer(peer, request->xml); if (!pcmk__str_eq(peer->name, attrd_cluster->uname, pcmk__str_casei)) { attrd_peer_sync_response(peer, peer_won, request->xml); } } pcmk__set_result(&request->result, CRM_EX_OK, PCMK_EXEC_DONE, NULL); return NULL; } } static xmlNode * handle_update_request(pcmk__request_t *request) { if (request->peer != NULL) { const char *host = crm_element_value(request->xml, PCMK__XA_ATTR_HOST); pcmk__node_status_t *peer = pcmk__get_node(0, request->peer, NULL, pcmk__node_search_cluster_member); attrd_peer_update(peer, request->xml, host, false); pcmk__set_result(&request->result, CRM_EX_OK, PCMK_EXEC_DONE, NULL); return NULL; } else { remove_unsupported_sync_points(request); if (attrd_request_has_sync_point(request->xml)) { /* If this client supplied a sync point it wants to wait for, add it to * the wait list. Clients on this list will not receive an ACK until * their sync point is hit which will result in the client stalled there * until it receives a response. * * All other clients will receive the expected response as normal. */ attrd_add_client_to_waitlist(request); } else { /* If the client doesn't want to wait for a sync point, go ahead and send * the ACK immediately. Otherwise, we'll send the ACK when the appropriate * sync point is reached. * * In the normal case, attrd_client_update can be called recursively which * makes where to send the ACK tricky. Doing it here ensures the client * only ever receives one. */ attrd_send_ack(request->ipc_client, request->ipc_id, request->flags|crm_ipc_client_response); } return attrd_client_update(request); } } static void attrd_register_handlers(void) { pcmk__server_command_t handlers[] = { { PCMK__ATTRD_CMD_CLEAR_FAILURE, handle_clear_failure_request }, { PCMK__ATTRD_CMD_CONFIRM, handle_confirm_request }, { PCMK__ATTRD_CMD_PEER_REMOVE, handle_remove_request }, { PCMK__ATTRD_CMD_QUERY, handle_query_request }, { PCMK__ATTRD_CMD_REFRESH, handle_refresh_request }, { PCMK__ATTRD_CMD_SYNC_RESPONSE, handle_sync_response_request }, { PCMK__ATTRD_CMD_UPDATE, handle_update_request }, { PCMK__ATTRD_CMD_UPDATE_DELAY, handle_update_request }, { PCMK__ATTRD_CMD_UPDATE_BOTH, handle_update_request }, { NULL, handle_unknown_request }, }; attrd_handlers = pcmk__register_handlers(handlers); } void attrd_unregister_handlers(void) { if (attrd_handlers != NULL) { g_hash_table_destroy(attrd_handlers); attrd_handlers = NULL; } } void attrd_handle_request(pcmk__request_t *request) { xmlNode *reply = NULL; char *log_msg = NULL; const char *reason = NULL; if (attrd_handlers == NULL) { attrd_register_handlers(); } reply = pcmk__process_request(request, attrd_handlers); if (reply != NULL) { crm_log_xml_trace(reply, "Reply"); if (request->ipc_client != NULL) { pcmk__ipc_send_xml(request->ipc_client, request->ipc_id, reply, request->ipc_flags); } else { crm_err("Not sending CPG reply to client"); } pcmk__xml_free(reply); } reason = request->result.exit_reason; log_msg = crm_strdup_printf("Processed %s request from %s %s: %s%s%s%s", request->op, pcmk__request_origin_type(request), pcmk__request_origin(request), pcmk_exec_status_str(request->result.execution_status), (reason == NULL)? "" : " (", pcmk__s(reason, ""), (reason == NULL)? "" : ")"); if (!pcmk__result_ok(&request->result)) { crm_warn("%s", log_msg); } else { crm_debug("%s", log_msg); } free(log_msg); pcmk__reset_request(request); } /*! \internal \brief Broadcast private attribute for local node with protocol version */ void attrd_broadcast_protocol(void) { xmlNode *attrd_op = pcmk__xe_create(NULL, __func__); crm_xml_add(attrd_op, PCMK__XA_T, PCMK__VALUE_ATTRD); crm_xml_add(attrd_op, PCMK__XA_SRC, crm_system_name); crm_xml_add(attrd_op, PCMK_XA_TASK, PCMK__ATTRD_CMD_UPDATE); crm_xml_add(attrd_op, PCMK__XA_ATTR_NAME, CRM_ATTR_PROTOCOL); crm_xml_add(attrd_op, PCMK__XA_ATTR_VALUE, ATTRD_PROTOCOL_VERSION); crm_xml_add_int(attrd_op, PCMK__XA_ATTR_IS_PRIVATE, 1); pcmk__xe_add_node(attrd_op, attrd_cluster->uname, attrd_cluster->nodeid); crm_debug("Broadcasting attrd protocol version %s for node %s", ATTRD_PROTOCOL_VERSION, attrd_cluster->uname); attrd_send_message(NULL, attrd_op, false); /* ends up at attrd_peer_message() */ pcmk__xml_free(attrd_op); } gboolean attrd_send_message(pcmk__node_status_t *node, xmlNode *data, bool confirm) { const char *op = crm_element_value(data, PCMK_XA_TASK); crm_xml_add(data, PCMK__XA_T, PCMK__VALUE_ATTRD); crm_xml_add(data, PCMK__XA_ATTR_VERSION, ATTRD_PROTOCOL_VERSION); /* Request a confirmation from the destination peer node (which could * be all if node is NULL) that the message has been received and * acted upon. */ if (!pcmk__str_eq(op, PCMK__ATTRD_CMD_CONFIRM, pcmk__str_none)) { pcmk__xe_set_bool_attr(data, PCMK__XA_CONFIRM, confirm); } attrd_xml_add_writer(data); - return pcmk__cluster_send_message(node, crm_msg_attrd, data); + return pcmk__cluster_send_message(node, pcmk__cluster_msg_attrd, data); } diff --git a/daemons/based/based_callbacks.c b/daemons/based/based_callbacks.c index 20a37f1a3b..1bf2202dc6 100644 --- a/daemons/based/based_callbacks.c +++ b/daemons/based/based_callbacks.c @@ -1,1377 +1,1377 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include // uint32_t, uint64_t, UINT64_C() #include #include #include // PRIu64 #include #include #include #include #include #include #include #include #define EXIT_ESCALATION_MS 10000 qb_ipcs_service_t *ipcs_ro = NULL; qb_ipcs_service_t *ipcs_rw = NULL; qb_ipcs_service_t *ipcs_shm = NULL; static int cib_process_command(xmlNode *request, const cib__operation_t *operation, cib__op_fn_t op_function, xmlNode **reply, xmlNode **cib_diff, bool privileged); static gboolean cib_common_callback(qb_ipcs_connection_t *c, void *data, size_t size, gboolean privileged); static int32_t cib_ipc_accept(qb_ipcs_connection_t * c, uid_t uid, gid_t gid) { if (cib_shutdown_flag) { crm_info("Ignoring new IPC client [%d] during shutdown", pcmk__client_pid(c)); return -ECONNREFUSED; } if (pcmk__new_client(c, uid, gid) == NULL) { return -ENOMEM; } return 0; } static int32_t cib_ipc_dispatch_rw(qb_ipcs_connection_t * c, void *data, size_t size) { pcmk__client_t *client = pcmk__find_client(c); crm_trace("%p message from %s", c, client->id); return cib_common_callback(c, data, size, TRUE); } static int32_t cib_ipc_dispatch_ro(qb_ipcs_connection_t * c, void *data, size_t size) { pcmk__client_t *client = pcmk__find_client(c); crm_trace("%p message from %s", c, client->id); return cib_common_callback(c, data, size, FALSE); } /* Error code means? */ static int32_t cib_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); pcmk__free_client(client); return 0; } static void cib_ipc_destroy(qb_ipcs_connection_t * c) { crm_trace("Connection %p", c); cib_ipc_closed(c); if (cib_shutdown_flag) { cib_shutdown(0); } } struct qb_ipcs_service_handlers ipc_ro_callbacks = { .connection_accept = cib_ipc_accept, .connection_created = NULL, .msg_process = cib_ipc_dispatch_ro, .connection_closed = cib_ipc_closed, .connection_destroyed = cib_ipc_destroy }; struct qb_ipcs_service_handlers ipc_rw_callbacks = { .connection_accept = cib_ipc_accept, .connection_created = NULL, .msg_process = cib_ipc_dispatch_rw, .connection_closed = cib_ipc_closed, .connection_destroyed = cib_ipc_destroy }; /*! * \internal * \brief Create reply XML for a CIB request * * \param[in] op CIB operation type * \param[in] call_id CIB call ID * \param[in] client_id CIB client ID * \param[in] call_options Group of enum cib_call_options flags * \param[in] rc Request return code * \param[in] call_data Request output data * * \return Reply XML (guaranteed not to be \c NULL) * * \note The caller is responsible for freeing the return value using * \p pcmk__xml_free(). */ static xmlNode * create_cib_reply(const char *op, const char *call_id, const char *client_id, int call_options, int rc, xmlNode *call_data) { xmlNode *reply = pcmk__xe_create(NULL, PCMK__XE_CIB_REPLY); crm_xml_add(reply, PCMK__XA_T, PCMK__VALUE_CIB); crm_xml_add(reply, PCMK__XA_CIB_OP, op); crm_xml_add(reply, PCMK__XA_CIB_CALLID, call_id); crm_xml_add(reply, PCMK__XA_CIB_CLIENTID, client_id); crm_xml_add_int(reply, PCMK__XA_CIB_CALLOPT, call_options); crm_xml_add_int(reply, PCMK__XA_CIB_RC, rc); if (call_data != NULL) { xmlNode *wrapper = pcmk__xe_create(reply, PCMK__XE_CIB_CALLDATA); crm_trace("Attaching reply output"); pcmk__xml_copy(wrapper, call_data); } crm_log_xml_explicit(reply, "cib:reply"); return reply; } static void do_local_notify(const xmlNode *notify_src, const char *client_id, bool sync_reply, bool from_peer) { int msg_id = 0; int rc = pcmk_rc_ok; pcmk__client_t *client_obj = NULL; uint32_t flags = crm_ipc_server_event; CRM_CHECK((notify_src != NULL) && (client_id != NULL), return); crm_element_value_int(notify_src, PCMK__XA_CIB_CALLID, &msg_id); client_obj = pcmk__find_client_by_id(client_id); if (client_obj == NULL) { crm_debug("Could not notify client %s%s %s of call %d result: " "client no longer exists", client_id, (from_peer? " (originator of delegated request)" : ""), (sync_reply? "synchronously" : "asynchronously"), msg_id); return; } if (sync_reply) { flags = crm_ipc_flags_none; if (client_obj->ipcs != NULL) { msg_id = client_obj->request_id; client_obj->request_id = 0; } } switch (PCMK__CLIENT_TYPE(client_obj)) { case pcmk__client_ipc: rc = pcmk__ipc_send_xml(client_obj, msg_id, notify_src, flags); break; case pcmk__client_tls: case pcmk__client_tcp: rc = pcmk__remote_send_xml(client_obj->remote, notify_src); break; default: rc = EPROTONOSUPPORT; break; } if (rc == pcmk_rc_ok) { crm_trace("Notified %s client %s%s %s of call %d result", pcmk__client_type_str(PCMK__CLIENT_TYPE(client_obj)), pcmk__client_name(client_obj), (from_peer? " (originator of delegated request)" : ""), (sync_reply? "synchronously" : "asynchronously"), msg_id); } else { crm_warn("Could not notify %s client %s%s %s of call %d result: %s", pcmk__client_type_str(PCMK__CLIENT_TYPE(client_obj)), pcmk__client_name(client_obj), (from_peer? " (originator of delegated request)" : ""), (sync_reply? "synchronously" : "asynchronously"), msg_id, pcmk_rc_str(rc)); } } void cib_common_callback_worker(uint32_t id, uint32_t flags, xmlNode * op_request, pcmk__client_t *cib_client, gboolean privileged) { const char *op = crm_element_value(op_request, PCMK__XA_CIB_OP); int call_options = cib_none; crm_element_value_int(op_request, PCMK__XA_CIB_CALLOPT, &call_options); /* Requests with cib_transaction set should not be sent to based directly * (outside of a commit-transaction request) */ if (pcmk_is_set(call_options, cib_transaction)) { return; } if (pcmk__str_eq(op, CRM_OP_REGISTER, pcmk__str_none)) { if (flags & crm_ipc_client_response) { xmlNode *ack = pcmk__xe_create(NULL, __func__); crm_xml_add(ack, PCMK__XA_CIB_OP, CRM_OP_REGISTER); crm_xml_add(ack, PCMK__XA_CIB_CLIENTID, cib_client->id); pcmk__ipc_send_xml(cib_client, id, ack, flags); cib_client->request_id = 0; pcmk__xml_free(ack); } return; } else if (pcmk__str_eq(op, PCMK__VALUE_CIB_NOTIFY, pcmk__str_none)) { /* Update the notify filters for this client */ int on_off = 0; crm_exit_t status = CRM_EX_OK; uint64_t bit = UINT64_C(0); const char *type = crm_element_value(op_request, PCMK__XA_CIB_NOTIFY_TYPE); crm_element_value_int(op_request, PCMK__XA_CIB_NOTIFY_ACTIVATE, &on_off); crm_debug("Setting %s callbacks %s for client %s", type, (on_off? "on" : "off"), pcmk__client_name(cib_client)); if (pcmk__str_eq(type, PCMK__VALUE_CIB_POST_NOTIFY, pcmk__str_none)) { bit = cib_notify_post; } else if (pcmk__str_eq(type, PCMK__VALUE_CIB_PRE_NOTIFY, pcmk__str_none)) { bit = cib_notify_pre; } else if (pcmk__str_eq(type, PCMK__VALUE_CIB_UPDATE_CONFIRMATION, pcmk__str_none)) { bit = cib_notify_confirm; } else if (pcmk__str_eq(type, PCMK__VALUE_CIB_DIFF_NOTIFY, pcmk__str_none)) { bit = cib_notify_diff; } else { status = CRM_EX_INVALID_PARAM; } if (bit != 0) { if (on_off) { pcmk__set_client_flags(cib_client, bit); } else { pcmk__clear_client_flags(cib_client, bit); } } pcmk__ipc_send_ack(cib_client, id, flags, PCMK__XE_ACK, NULL, status); return; } cib_process_request(op_request, privileged, cib_client); } int32_t cib_common_callback(qb_ipcs_connection_t * c, void *data, size_t size, gboolean privileged) { uint32_t id = 0; uint32_t flags = 0; int call_options = 0; pcmk__client_t *cib_client = pcmk__find_client(c); xmlNode *op_request = pcmk__client_data2xml(cib_client, data, &id, &flags); if (op_request) { crm_element_value_int(op_request, PCMK__XA_CIB_CALLOPT, &call_options); } if (op_request == NULL) { crm_trace("Invalid message from %p", c); pcmk__ipc_send_ack(cib_client, id, flags, PCMK__XE_NACK, NULL, CRM_EX_PROTOCOL); return 0; } else if(cib_client == NULL) { crm_trace("Invalid client %p", c); return 0; } if (pcmk_is_set(call_options, cib_sync_call)) { CRM_LOG_ASSERT(flags & crm_ipc_client_response); CRM_LOG_ASSERT(cib_client->request_id == 0); /* This means the client has two synchronous events in-flight */ cib_client->request_id = id; /* Reply only to the last one */ } if (cib_client->name == NULL) { const char *value = crm_element_value(op_request, PCMK__XA_CIB_CLIENTNAME); if (value == NULL) { cib_client->name = pcmk__itoa(cib_client->pid); } else { cib_client->name = pcmk__str_copy(value); if (crm_is_daemon_name(value)) { pcmk__set_client_flags(cib_client, cib_is_daemon); } } } /* Allow cluster daemons more leeway before being evicted */ if (pcmk_is_set(cib_client->flags, cib_is_daemon)) { const char *qmax = cib_config_lookup(PCMK_OPT_CLUSTER_IPC_LIMIT); if (pcmk__set_client_queue_max(cib_client, qmax)) { crm_trace("IPC threshold for client %s[%u] is now %u", pcmk__client_name(cib_client), cib_client->pid, cib_client->queue_max); } } crm_xml_add(op_request, PCMK__XA_CIB_CLIENTID, cib_client->id); crm_xml_add(op_request, PCMK__XA_CIB_CLIENTNAME, cib_client->name); CRM_LOG_ASSERT(cib_client->user != NULL); pcmk__update_acl_user(op_request, PCMK__XA_CIB_USER, cib_client->user); cib_common_callback_worker(id, flags, op_request, cib_client, privileged); pcmk__xml_free(op_request); return 0; } static uint64_t ping_seq = 0; static char *ping_digest = NULL; static bool ping_modified_since = FALSE; static gboolean cib_digester_cb(gpointer data) { if (based_is_primary) { char buffer[32]; xmlNode *ping = pcmk__xe_create(NULL, PCMK__XE_PING); ping_seq++; free(ping_digest); ping_digest = NULL; ping_modified_since = FALSE; snprintf(buffer, 32, "%" PRIu64, ping_seq); crm_trace("Requesting peer digests (%s)", buffer); crm_xml_add(ping, PCMK__XA_T, PCMK__VALUE_CIB); crm_xml_add(ping, PCMK__XA_CIB_OP, CRM_OP_PING); crm_xml_add(ping, PCMK__XA_CIB_PING_ID, buffer); crm_xml_add(ping, PCMK_XA_CRM_FEATURE_SET, CRM_FEATURE_SET); - pcmk__cluster_send_message(NULL, crm_msg_cib, ping); + pcmk__cluster_send_message(NULL, pcmk__cluster_msg_based, ping); pcmk__xml_free(ping); } return FALSE; } static void process_ping_reply(xmlNode *reply) { uint64_t seq = 0; const char *host = crm_element_value(reply, PCMK__XA_SRC); xmlNode *wrapper = pcmk__xe_first_child(reply, PCMK__XE_CIB_CALLDATA, NULL, NULL); xmlNode *pong = pcmk__xe_first_child(wrapper, NULL, NULL, NULL); const char *seq_s = crm_element_value(pong, PCMK__XA_CIB_PING_ID); const char *digest = crm_element_value(pong, PCMK__XA_DIGEST); if (seq_s == NULL) { crm_debug("Ignoring ping reply with no " PCMK__XA_CIB_PING_ID); return; } else { long long seq_ll; if (pcmk__scan_ll(seq_s, &seq_ll, 0LL) != pcmk_rc_ok) { return; } seq = (uint64_t) seq_ll; } if(digest == NULL) { crm_trace("Ignoring ping reply %s from %s with no digest", seq_s, host); } else if(seq != ping_seq) { crm_trace("Ignoring out of sequence ping reply %s from %s", seq_s, host); } else if(ping_modified_since) { crm_trace("Ignoring ping reply %s from %s: cib updated since", seq_s, host); } else { if(ping_digest == NULL) { crm_trace("Calculating new digest"); ping_digest = pcmk__digest_xml(the_cib, true); } crm_trace("Processing ping reply %s from %s (%s)", seq_s, host, digest); if (!pcmk__str_eq(ping_digest, digest, pcmk__str_casei)) { xmlNode *wrapper = pcmk__xe_first_child(pong, PCMK__XE_CIB_CALLDATA, NULL, NULL); xmlNode *remote_cib = pcmk__xe_first_child(wrapper, NULL, NULL, NULL); const char *admin_epoch_s = NULL; const char *epoch_s = NULL; const char *num_updates_s = NULL; if (remote_cib != NULL) { admin_epoch_s = crm_element_value(remote_cib, PCMK_XA_ADMIN_EPOCH); epoch_s = crm_element_value(remote_cib, PCMK_XA_EPOCH); num_updates_s = crm_element_value(remote_cib, PCMK_XA_NUM_UPDATES); } crm_notice("Local CIB %s.%s.%s.%s differs from %s: %s.%s.%s.%s %p", crm_element_value(the_cib, PCMK_XA_ADMIN_EPOCH), crm_element_value(the_cib, PCMK_XA_EPOCH), crm_element_value(the_cib, PCMK_XA_NUM_UPDATES), ping_digest, host, pcmk__s(admin_epoch_s, "_"), pcmk__s(epoch_s, "_"), pcmk__s(num_updates_s, "_"), digest, remote_cib); if(remote_cib && remote_cib->children) { // Additional debug xml_calculate_changes(the_cib, remote_cib); pcmk__log_xml_changes(LOG_INFO, remote_cib); crm_trace("End of differences"); } pcmk__xml_free(remote_cib); sync_our_cib(reply, FALSE); } } } static void parse_local_options(const pcmk__client_t *cib_client, const cib__operation_t *operation, int call_options, const char *host, const char *op, gboolean *local_notify, gboolean *needs_reply, gboolean *process, gboolean *needs_forward) { // Process locally and notify local client *process = TRUE; *needs_reply = FALSE; *local_notify = TRUE; *needs_forward = FALSE; if (pcmk_is_set(operation->flags, cib__op_attr_local)) { /* Always process locally if cib__op_attr_local is set. * * @COMPAT: Currently host is ignored. At a compatibility break, throw * an error (from cib_process_request() or earlier) if host is not NULL or * OUR_NODENAME. */ crm_trace("Processing always-local %s op from client %s", op, pcmk__client_name(cib_client)); if (!pcmk__str_eq(host, OUR_NODENAME, pcmk__str_casei|pcmk__str_null_matches)) { crm_warn("Operation '%s' is always local but its target host is " "set to '%s'", op, host); } return; } if (pcmk_is_set(operation->flags, cib__op_attr_modifies) || !pcmk__str_eq(host, OUR_NODENAME, pcmk__str_casei|pcmk__str_null_matches)) { // Forward modifying and non-local requests via cluster *process = FALSE; *needs_reply = FALSE; *local_notify = FALSE; *needs_forward = TRUE; crm_trace("%s op from %s needs to be forwarded to %s", op, pcmk__client_name(cib_client), pcmk__s(host, "all nodes")); return; } if (stand_alone) { crm_trace("Processing %s op from client %s (stand-alone)", op, pcmk__client_name(cib_client)); } else { crm_trace("Processing %saddressed %s op from client %s", ((host != NULL)? "locally " : "un"), op, pcmk__client_name(cib_client)); } } static gboolean parse_peer_options(const cib__operation_t *operation, xmlNode *request, gboolean *local_notify, gboolean *needs_reply, gboolean *process) { /* TODO: What happens when an update comes in after node A * requests the CIB from node B, but before it gets the reply (and * sends out the replace operation)? * * (This may no longer be relevant since legacy mode was dropped; need to * trace code more closely to check.) */ const char *host = NULL; const char *delegated = crm_element_value(request, PCMK__XA_CIB_DELEGATED_FROM); const char *op = crm_element_value(request, PCMK__XA_CIB_OP); const char *originator = crm_element_value(request, PCMK__XA_SRC); const char *reply_to = crm_element_value(request, PCMK__XA_CIB_ISREPLYTO); gboolean is_reply = pcmk__str_eq(reply_to, OUR_NODENAME, pcmk__str_casei); if (originator == NULL) { // Shouldn't be possible originator = "peer"; } if (pcmk__str_eq(op, PCMK__CIB_REQUEST_REPLACE, pcmk__str_none)) { // sync_our_cib() sets PCMK__XA_CIB_ISREPLYTO if (reply_to) { delegated = reply_to; } goto skip_is_reply; } else if (pcmk__str_eq(op, PCMK__CIB_REQUEST_SYNC_TO_ALL, pcmk__str_none)) { // Nothing to do } else if (is_reply && pcmk__str_eq(op, CRM_OP_PING, pcmk__str_casei)) { process_ping_reply(request); return FALSE; } else if (pcmk__str_eq(op, PCMK__CIB_REQUEST_UPGRADE, pcmk__str_none)) { /* Only the DC (node with the oldest software) should process * this operation if PCMK__XA_CIB_SCHEMA_MAX is unset. * * If the DC is happy it will then send out another * PCMK__CIB_REQUEST_UPGRADE which will tell all nodes to do the actual * upgrade. * * Except this time PCMK__XA_CIB_SCHEMA_MAX will be set which puts a * limit on how far newer nodes will go */ const char *max = crm_element_value(request, PCMK__XA_CIB_SCHEMA_MAX); const char *upgrade_rc = crm_element_value(request, PCMK__XA_CIB_UPGRADE_RC); crm_trace("Parsing upgrade %s for %s with max=%s and upgrade_rc=%s", (is_reply? "reply" : "request"), (based_is_primary? "primary" : "secondary"), pcmk__s(max, "none"), pcmk__s(upgrade_rc, "none")); if (upgrade_rc != NULL) { // Our upgrade request was rejected by DC, notify clients of result crm_xml_add(request, PCMK__XA_CIB_RC, upgrade_rc); } else if ((max == NULL) && based_is_primary) { /* We are the DC, check if this upgrade is allowed */ goto skip_is_reply; } else if(max) { /* Ok, go ahead and upgrade to 'max' */ goto skip_is_reply; } else { // Ignore broadcast client requests when we're not primary return FALSE; } } else if (pcmk__xe_attr_is_true(request, PCMK__XA_CIB_UPDATE)) { crm_info("Detected legacy %s global update from %s", op, originator); send_sync_request(NULL); return FALSE; } else if (is_reply && pcmk_is_set(operation->flags, cib__op_attr_modifies)) { crm_trace("Ignoring legacy %s reply sent from %s to local clients", op, originator); return FALSE; } else if (pcmk__str_eq(op, PCMK__CIB_REQUEST_SHUTDOWN, pcmk__str_none)) { *local_notify = FALSE; if (reply_to == NULL) { *process = TRUE; } else { // Not possible? crm_debug("Ignoring shutdown request from %s because reply_to=%s", originator, reply_to); } return *process; } if (is_reply) { crm_trace("Will notify local clients for %s reply from %s", op, originator); *process = FALSE; *needs_reply = FALSE; *local_notify = TRUE; return TRUE; } skip_is_reply: *process = TRUE; *needs_reply = FALSE; *local_notify = pcmk__str_eq(delegated, OUR_NODENAME, pcmk__str_casei); host = crm_element_value(request, PCMK__XA_CIB_HOST); if (pcmk__str_eq(host, OUR_NODENAME, pcmk__str_casei)) { crm_trace("Processing %s request sent to us from %s", op, originator); *needs_reply = TRUE; return TRUE; } else if (host != NULL) { crm_trace("Ignoring %s request intended for CIB manager on %s", op, host); return FALSE; } else if(is_reply == FALSE && pcmk__str_eq(op, CRM_OP_PING, pcmk__str_casei)) { *needs_reply = TRUE; } crm_trace("Processing %s request broadcast by %s call %s on %s " "(local clients will%s be notified)", op, pcmk__s(crm_element_value(request, PCMK__XA_CIB_CLIENTNAME), "client"), pcmk__s(crm_element_value(request, PCMK__XA_CIB_CALLID), "without ID"), originator, (*local_notify? "" : "not")); return TRUE; } /*! * \internal * \brief Forward a CIB request to the appropriate target host(s) * * \param[in] request CIB request to forward */ static void forward_request(xmlNode *request) { const char *op = crm_element_value(request, PCMK__XA_CIB_OP); const char *section = crm_element_value(request, PCMK__XA_CIB_SECTION); const char *host = crm_element_value(request, PCMK__XA_CIB_HOST); const char *originator = crm_element_value(request, PCMK__XA_SRC); const char *client_name = crm_element_value(request, PCMK__XA_CIB_CLIENTNAME); const char *call_id = crm_element_value(request, PCMK__XA_CIB_CALLID); pcmk__node_status_t *peer = NULL; int log_level = LOG_INFO; if (pcmk__str_eq(op, PCMK__CIB_REQUEST_NOOP, pcmk__str_none)) { log_level = LOG_DEBUG; } do_crm_log(log_level, "Forwarding %s operation for section %s to %s (origin=%s/%s/%s)", pcmk__s(op, "invalid"), pcmk__s(section, "all"), pcmk__s(host, "all"), pcmk__s(originator, "local"), pcmk__s(client_name, "unspecified"), pcmk__s(call_id, "unspecified")); crm_xml_add(request, PCMK__XA_CIB_DELEGATED_FROM, OUR_NODENAME); if (host != NULL) { peer = pcmk__get_node(0, host, NULL, pcmk__node_search_cluster_member); } - pcmk__cluster_send_message(peer, crm_msg_cib, request); + pcmk__cluster_send_message(peer, pcmk__cluster_msg_based, request); // Return the request to its original state pcmk__xe_remove_attr(request, PCMK__XA_CIB_DELEGATED_FROM); } static void send_peer_reply(xmlNode *msg, const char *originator) { const pcmk__node_status_t *node = NULL; if ((msg == NULL) || (originator == NULL)) { return; } // Send reply via cluster to originating node node = pcmk__get_node(0, originator, NULL, pcmk__node_search_cluster_member); crm_trace("Sending request result to %s only", originator); crm_xml_add(msg, PCMK__XA_CIB_ISREPLYTO, originator); - pcmk__cluster_send_message(node, crm_msg_cib, msg); + pcmk__cluster_send_message(node, pcmk__cluster_msg_based, msg); } /*! * \internal * \brief Handle an IPC or CPG message containing a request * * \param[in,out] request Request XML * \param[in] privileged Whether privileged commands may be run * (see cib_server_ops[] definition) * \param[in] cib_client IPC client that sent request (or NULL if CPG) * * \return Legacy Pacemaker return code */ int cib_process_request(xmlNode *request, gboolean privileged, const pcmk__client_t *cib_client) { // @TODO: Break into multiple smaller functions int call_options = 0; gboolean process = TRUE; // Whether to process request locally now gboolean is_update = TRUE; // Whether request would modify CIB gboolean needs_reply = TRUE; // Whether to build a reply gboolean local_notify = FALSE; // Whether to notify (local) requester gboolean needs_forward = FALSE; // Whether to forward request somewhere else xmlNode *op_reply = NULL; xmlNode *result_diff = NULL; int rc = pcmk_ok; const char *op = crm_element_value(request, PCMK__XA_CIB_OP); const char *originator = crm_element_value(request, PCMK__XA_SRC); const char *host = crm_element_value(request, PCMK__XA_CIB_HOST); const char *call_id = crm_element_value(request, PCMK__XA_CIB_CALLID); const char *client_id = crm_element_value(request, PCMK__XA_CIB_CLIENTID); const char *client_name = crm_element_value(request, PCMK__XA_CIB_CLIENTNAME); const char *reply_to = crm_element_value(request, PCMK__XA_CIB_ISREPLYTO); const cib__operation_t *operation = NULL; cib__op_fn_t op_function = NULL; crm_element_value_int(request, PCMK__XA_CIB_CALLOPT, &call_options); if ((host != NULL) && (*host == '\0')) { host = NULL; } if (cib_client == NULL) { crm_trace("Processing peer %s operation from %s/%s on %s intended for %s (reply=%s)", op, client_name, call_id, originator, pcmk__s(host, "all"), reply_to); } else { crm_xml_add(request, PCMK__XA_SRC, OUR_NODENAME); crm_trace("Processing local %s operation from %s/%s intended for %s", op, client_name, call_id, pcmk__s(host, "all")); } rc = cib__get_operation(op, &operation); rc = pcmk_rc2legacy(rc); if (rc != pcmk_ok) { /* TODO: construct error reply? */ crm_err("Pre-processing of command failed: %s", pcmk_strerror(rc)); return rc; } op_function = based_get_op_function(operation); if (op_function == NULL) { crm_err("Operation %s not supported by CIB manager", op); return -EOPNOTSUPP; } if (cib_client != NULL) { parse_local_options(cib_client, operation, call_options, host, op, &local_notify, &needs_reply, &process, &needs_forward); } else if (!parse_peer_options(operation, request, &local_notify, &needs_reply, &process)) { return rc; } if (pcmk_is_set(call_options, cib_transaction)) { /* All requests in a transaction are processed locally against a working * CIB copy, and we don't notify for individual requests because the * entire transaction is atomic. * * We still call the option parser functions above, for the sake of log * messages and checking whether we're the target for peer requests. */ process = TRUE; needs_reply = FALSE; local_notify = FALSE; needs_forward = FALSE; } is_update = pcmk_is_set(operation->flags, cib__op_attr_modifies); if (pcmk_is_set(call_options, cib_discard_reply)) { /* If the request will modify the CIB, and we are in legacy mode, we * need to build a reply so we can broadcast a diff, even if the * requester doesn't want one. */ needs_reply = FALSE; local_notify = FALSE; crm_trace("Client is not interested in the reply"); } if (needs_forward) { forward_request(request); return rc; } if (cib_status != pcmk_ok) { rc = cib_status; crm_err("Operation ignored, cluster configuration is invalid." " Please repair and restart: %s", pcmk_strerror(cib_status)); op_reply = create_cib_reply(op, call_id, client_id, call_options, rc, the_cib); } else if (process) { time_t finished = 0; time_t now = time(NULL); int level = LOG_INFO; const char *section = crm_element_value(request, PCMK__XA_CIB_SECTION); const char *admin_epoch_s = NULL; const char *epoch_s = NULL; const char *num_updates_s = NULL; rc = cib_process_command(request, operation, op_function, &op_reply, &result_diff, privileged); if (!is_update) { level = LOG_TRACE; } else if (pcmk__xe_attr_is_true(request, PCMK__XA_CIB_UPDATE)) { switch (rc) { case pcmk_ok: level = LOG_INFO; break; case -pcmk_err_old_data: case -pcmk_err_diff_resync: case -pcmk_err_diff_failed: level = LOG_TRACE; break; default: level = LOG_ERR; } } else if (rc != pcmk_ok) { level = LOG_WARNING; } if (the_cib != NULL) { admin_epoch_s = crm_element_value(the_cib, PCMK_XA_ADMIN_EPOCH); epoch_s = crm_element_value(the_cib, PCMK_XA_EPOCH); num_updates_s = crm_element_value(the_cib, PCMK_XA_NUM_UPDATES); } do_crm_log(level, "Completed %s operation for section %s: %s (rc=%d, origin=%s/%s/%s, version=%s.%s.%s)", op, section ? section : "'all'", pcmk_strerror(rc), rc, originator ? originator : "local", client_name, call_id, pcmk__s(admin_epoch_s, "0"), pcmk__s(epoch_s, "0"), pcmk__s(num_updates_s, "0")); finished = time(NULL); if ((finished - now) > 3) { crm_trace("%s operation took %lds to complete", op, (long)(finished - now)); crm_write_blackbox(0, NULL); } if (op_reply == NULL && (needs_reply || local_notify)) { crm_err("Unexpected NULL reply to message"); crm_log_xml_err(request, "null reply"); needs_reply = FALSE; local_notify = FALSE; } } if (is_update) { crm_trace("Completed pre-sync update from %s/%s/%s%s", originator ? originator : "local", client_name, call_id, local_notify?" with local notification":""); } else if (!needs_reply || stand_alone) { // This was a non-originating secondary update crm_trace("Completed update as secondary"); } else if ((cib_client == NULL) && !pcmk_is_set(call_options, cib_discard_reply)) { if (is_update == FALSE || result_diff == NULL) { crm_trace("Request not broadcast: R/O call"); } else if (rc != pcmk_ok) { crm_trace("Request not broadcast: call failed: %s", pcmk_strerror(rc)); } else { crm_trace("Directing reply to %s", originator); } send_peer_reply(op_reply, originator); } if (local_notify && client_id) { crm_trace("Performing local %ssync notification for %s", (pcmk_is_set(call_options, cib_sync_call)? "" : "a"), client_id); if (process == FALSE) { do_local_notify(request, client_id, pcmk_is_set(call_options, cib_sync_call), (cib_client == NULL)); } else { do_local_notify(op_reply, client_id, pcmk_is_set(call_options, cib_sync_call), (cib_client == NULL)); } } pcmk__xml_free(op_reply); pcmk__xml_free(result_diff); return rc; } /*! * \internal * \brief Get a CIB operation's input from the request XML * * \param[in] request CIB request XML * \param[in] type CIB operation type * \param[out] section Where to store CIB section name * * \return Input XML for CIB operation * * \note If not \c NULL, the return value is a non-const pointer to part of * \p request. The caller should not free it directly. */ static xmlNode * prepare_input(const xmlNode *request, enum cib__op_type type, const char **section) { xmlNode *wrapper = pcmk__xe_first_child(request, PCMK__XE_CIB_CALLDATA, NULL, NULL); xmlNode *input = pcmk__xe_first_child(wrapper, NULL, NULL, NULL); if (type == cib__op_apply_patch) { *section = NULL; } else { *section = crm_element_value(request, PCMK__XA_CIB_SECTION); } // Grab the specified section if ((*section != NULL) && pcmk__xe_is(input, PCMK_XE_CIB)) { input = pcmk_find_cib_element(input, *section); } return input; } #define XPATH_CONFIG_CHANGE \ "//" PCMK_XE_CHANGE \ "[contains(@" PCMK_XA_PATH ",'/" PCMK_XE_CRM_CONFIG "/')]" static bool contains_config_change(xmlNode *diff) { bool changed = false; if (diff) { xmlXPathObject *xpathObj = xpath_search(diff, XPATH_CONFIG_CHANGE); if (numXpathResults(xpathObj) > 0) { changed = true; } freeXpathObject(xpathObj); } return changed; } static int cib_process_command(xmlNode *request, const cib__operation_t *operation, cib__op_fn_t op_function, xmlNode **reply, xmlNode **cib_diff, bool privileged) { xmlNode *input = NULL; xmlNode *output = NULL; xmlNode *result_cib = NULL; int call_options = 0; const char *op = NULL; const char *section = NULL; const char *call_id = crm_element_value(request, PCMK__XA_CIB_CALLID); const char *client_id = crm_element_value(request, PCMK__XA_CIB_CLIENTID); const char *client_name = crm_element_value(request, PCMK__XA_CIB_CLIENTNAME); const char *originator = crm_element_value(request, PCMK__XA_SRC); int rc = pcmk_ok; bool config_changed = false; bool manage_counters = true; static mainloop_timer_t *digest_timer = NULL; CRM_ASSERT(cib_status == pcmk_ok); if(digest_timer == NULL) { digest_timer = mainloop_timer_add("digester", 5000, FALSE, cib_digester_cb, NULL); } *reply = NULL; *cib_diff = NULL; /* Start processing the request... */ op = crm_element_value(request, PCMK__XA_CIB_OP); crm_element_value_int(request, PCMK__XA_CIB_CALLOPT, &call_options); if (!privileged && pcmk_is_set(operation->flags, cib__op_attr_privileged)) { rc = -EACCES; crm_trace("Failed due to lack of privileges: %s", pcmk_strerror(rc)); goto done; } input = prepare_input(request, operation->type, §ion); if (!pcmk_is_set(operation->flags, cib__op_attr_modifies)) { rc = cib_perform_op(NULL, op, call_options, op_function, true, section, request, input, false, &config_changed, &the_cib, &result_cib, NULL, &output); CRM_CHECK(result_cib == NULL, pcmk__xml_free(result_cib)); goto done; } /* @COMPAT: Handle a valid write action (legacy) * * @TODO: Re-evaluate whether this is all truly legacy. The cib_force_diff * portion is. However, PCMK__XA_CIB_UPDATE may be set by a sync operation * even in non-legacy mode, and manage_counters tells xml_create_patchset() * whether to update version/epoch info. */ if (pcmk__xe_attr_is_true(request, PCMK__XA_CIB_UPDATE)) { manage_counters = false; cib__set_call_options(call_options, "call", cib_force_diff); crm_trace("Global update detected"); CRM_LOG_ASSERT(pcmk__str_any_of(op, PCMK__CIB_REQUEST_APPLY_PATCH, PCMK__CIB_REQUEST_REPLACE, NULL)); } ping_modified_since = TRUE; // result_cib must not be modified after cib_perform_op() returns rc = cib_perform_op(NULL, op, call_options, op_function, false, section, request, input, manage_counters, &config_changed, &the_cib, &result_cib, cib_diff, &output); /* Always write to disk for successful ops with the flag set. This also * negates the need to detect ordering changes. */ if ((rc == pcmk_ok) && pcmk_is_set(operation->flags, cib__op_attr_writes_through)) { config_changed = true; } if ((rc == pcmk_ok) && !pcmk_any_flags_set(call_options, cib_dryrun|cib_transaction)) { if (result_cib != the_cib) { if (pcmk_is_set(operation->flags, cib__op_attr_writes_through)) { config_changed = true; } crm_trace("Activating %s->%s%s", crm_element_value(the_cib, PCMK_XA_NUM_UPDATES), crm_element_value(result_cib, PCMK_XA_NUM_UPDATES), (config_changed? " changed" : "")); rc = activateCibXml(result_cib, config_changed, op); if (rc != pcmk_ok) { crm_err("Failed to activate new CIB: %s", pcmk_strerror(rc)); } } if ((rc == pcmk_ok) && contains_config_change(*cib_diff)) { cib_read_config(config_hash, result_cib); } /* @COMPAT Nodes older than feature set 3.19.0 don't support * transactions. In a mixed-version cluster with nodes <3.19.0, we must * sync the updated CIB, so that the older nodes receive the changes. * Any node that has already applied the transaction will ignore the * synced CIB. * * To ensure the updated CIB is synced from only one node, we sync it * from the originator. */ if ((operation->type == cib__op_commit_transact) && pcmk__str_eq(originator, OUR_NODENAME, pcmk__str_casei) && compare_version(crm_element_value(the_cib, PCMK_XA_CRM_FEATURE_SET), "3.19.0") < 0) { sync_our_cib(request, TRUE); } mainloop_timer_stop(digest_timer); mainloop_timer_start(digest_timer); } else if (rc == -pcmk_err_schema_validation) { CRM_ASSERT(result_cib != the_cib); if (output != NULL) { crm_log_xml_info(output, "cib:output"); pcmk__xml_free(output); } output = result_cib; } else { crm_trace("Not activating %d %d %s", rc, pcmk_is_set(call_options, cib_dryrun), crm_element_value(result_cib, PCMK_XA_NUM_UPDATES)); if (result_cib != the_cib) { pcmk__xml_free(result_cib); } } if (!pcmk_any_flags_set(call_options, cib_dryrun|cib_inhibit_notify|cib_transaction)) { crm_trace("Sending notifications %d", pcmk_is_set(call_options, cib_dryrun)); cib_diff_notify(op, rc, call_id, client_id, client_name, originator, input, *cib_diff); } pcmk__log_xml_patchset(LOG_TRACE, *cib_diff); done: if (!pcmk_is_set(call_options, cib_discard_reply)) { *reply = create_cib_reply(op, call_id, client_id, call_options, rc, output); } if (output != the_cib) { pcmk__xml_free(output); } crm_trace("done"); return rc; } void cib_peer_callback(xmlNode * msg, void *private_data) { const char *reason = NULL; const char *originator = crm_element_value(msg, PCMK__XA_SRC); if (crm_peer_cache == NULL) { reason = "membership not established"; goto bail; } if (crm_element_value(msg, PCMK__XA_CIB_CLIENTNAME) == NULL) { crm_xml_add(msg, PCMK__XA_CIB_CLIENTNAME, originator); } /* crm_log_xml_trace(msg, "Peer[inbound]"); */ cib_process_request(msg, TRUE, NULL); return; bail: if (reason) { const char *op = crm_element_value(msg, PCMK__XA_CIB_OP); crm_warn("Discarding %s message from %s: %s", op, originator, reason); } } static gboolean cib_force_exit(gpointer data) { crm_notice("Forcing exit!"); terminate_cib(__func__, CRM_EX_ERROR); return FALSE; } static void disconnect_remote_client(gpointer key, gpointer value, gpointer user_data) { pcmk__client_t *a_client = value; crm_err("Can't disconnect client %s: Not implemented", pcmk__client_name(a_client)); } static void initiate_exit(void) { int active = 0; xmlNode *leaving = NULL; active = pcmk__cluster_num_active_nodes(); if (active < 2) { // This is the last active node terminate_cib(__func__, 0); return; } crm_info("Sending shutdown request to %d peers", active); leaving = pcmk__xe_create(NULL, PCMK__XE_EXIT_NOTIFICATION); crm_xml_add(leaving, PCMK__XA_T, PCMK__VALUE_CIB); crm_xml_add(leaving, PCMK__XA_CIB_OP, PCMK__CIB_REQUEST_SHUTDOWN); - pcmk__cluster_send_message(NULL, crm_msg_cib, leaving); + pcmk__cluster_send_message(NULL, pcmk__cluster_msg_based, leaving); pcmk__xml_free(leaving); g_timeout_add(EXIT_ESCALATION_MS, cib_force_exit, NULL); } void cib_shutdown(int nsig) { struct qb_ipcs_stats srv_stats; if (cib_shutdown_flag == FALSE) { int disconnects = 0; qb_ipcs_connection_t *c = NULL; cib_shutdown_flag = TRUE; c = qb_ipcs_connection_first_get(ipcs_rw); while (c != NULL) { qb_ipcs_connection_t *last = c; c = qb_ipcs_connection_next_get(ipcs_rw, last); crm_debug("Disconnecting r/w client %p...", last); qb_ipcs_disconnect(last); qb_ipcs_connection_unref(last); disconnects++; } c = qb_ipcs_connection_first_get(ipcs_ro); while (c != NULL) { qb_ipcs_connection_t *last = c; c = qb_ipcs_connection_next_get(ipcs_ro, last); crm_debug("Disconnecting r/o client %p...", last); qb_ipcs_disconnect(last); qb_ipcs_connection_unref(last); disconnects++; } c = qb_ipcs_connection_first_get(ipcs_shm); while (c != NULL) { qb_ipcs_connection_t *last = c; c = qb_ipcs_connection_next_get(ipcs_shm, last); crm_debug("Disconnecting non-blocking r/w client %p...", last); qb_ipcs_disconnect(last); qb_ipcs_connection_unref(last); disconnects++; } disconnects += pcmk__ipc_client_count(); crm_debug("Disconnecting %d remote clients", pcmk__ipc_client_count()); pcmk__foreach_ipc_client(disconnect_remote_client, NULL); crm_info("Disconnected %d clients", disconnects); } qb_ipcs_stats_get(ipcs_rw, &srv_stats, QB_FALSE); if (pcmk__ipc_client_count() == 0) { crm_info("All clients disconnected (%d)", srv_stats.active_connections); initiate_exit(); } else { crm_info("Waiting on %d clients to disconnect (%d)", pcmk__ipc_client_count(), srv_stats.active_connections); } } extern int remote_fd; extern int remote_tls_fd; /*! * \internal * \brief Close remote sockets, free the global CIB and quit * * \param[in] caller Name of calling function (for log message) * \param[in] fast If -1, skip disconnect; if positive, exit that */ void terminate_cib(const char *caller, int fast) { crm_info("%s: Exiting%s...", caller, (fast > 0)? " fast" : mainloop ? " from mainloop" : ""); if (remote_fd > 0) { close(remote_fd); remote_fd = 0; } if (remote_tls_fd > 0) { close(remote_tls_fd); remote_tls_fd = 0; } uninitializeCib(); if (fast > 0) { /* Quit fast on error */ pcmk__stop_based_ipc(ipcs_ro, ipcs_rw, ipcs_shm); crm_exit(fast); } else if ((mainloop != NULL) && g_main_loop_is_running(mainloop)) { /* Quit via returning from the main loop. If fast == -1, we skip the * disconnect here, and it will be done when the main loop returns * (this allows the peer status callback to avoid messing with the * peer caches). */ if (fast == 0) { pcmk_cluster_disconnect(crm_cluster); } g_main_loop_quit(mainloop); } else { /* Quit via clean exit. Even the peer status callback can disconnect * here, because we're not returning control to the caller. */ pcmk_cluster_disconnect(crm_cluster); pcmk__stop_based_ipc(ipcs_ro, ipcs_rw, ipcs_shm); crm_exit(CRM_EX_OK); } } diff --git a/daemons/based/based_messages.c b/daemons/based/based_messages.c index 86400292de..4af5264acd 100644 --- a/daemons/based/based_messages.c +++ b/daemons/based/based_messages.c @@ -1,521 +1,523 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* Maximum number of diffs to ignore while waiting for a resync */ #define MAX_DIFF_RETRY 5 bool based_is_primary = false; xmlNode *the_cib = NULL; int cib_process_shutdown_req(const char *op, int options, const char *section, xmlNode * req, xmlNode * input, xmlNode * existing_cib, xmlNode ** result_cib, xmlNode ** answer) { const char *host = crm_element_value(req, PCMK__XA_SRC); *answer = NULL; if (crm_element_value(req, PCMK__XA_CIB_ISREPLYTO) == NULL) { crm_info("Peer %s is requesting to shut down", host); return pcmk_ok; } if (cib_shutdown_flag == FALSE) { crm_err("Peer %s mistakenly thinks we wanted to shut down", host); return -EINVAL; } crm_info("Peer %s has acknowledged our shutdown request", host); terminate_cib(__func__, 0); return pcmk_ok; } // @COMPAT: Remove when PCMK__CIB_REQUEST_NOOP is removed int cib_process_noop(const char *op, int options, const char *section, xmlNode *req, xmlNode *input, xmlNode *existing_cib, xmlNode **result_cib, xmlNode **answer) { crm_trace("Processing \"%s\" event", op); *answer = NULL; return pcmk_ok; } int cib_process_readwrite(const char *op, int options, const char *section, xmlNode * req, xmlNode * input, xmlNode * existing_cib, xmlNode ** result_cib, xmlNode ** answer) { int result = pcmk_ok; crm_trace("Processing \"%s\" event", op); if (pcmk__str_eq(op, PCMK__CIB_REQUEST_IS_PRIMARY, pcmk__str_none)) { if (based_is_primary) { result = pcmk_ok; } else { result = -EPERM; } return result; } if (pcmk__str_eq(op, PCMK__CIB_REQUEST_PRIMARY, pcmk__str_none)) { if (!based_is_primary) { crm_info("We are now in R/W mode"); based_is_primary = true; } else { crm_debug("We are still in R/W mode"); } } else if (based_is_primary) { crm_info("We are now in R/O mode"); based_is_primary = false; } return result; } /* Set to 1 when a sync is requested, incremented when a diff is ignored, * reset to 0 when a sync is received */ static int sync_in_progress = 0; void send_sync_request(const char *host) { xmlNode *sync_me = pcmk__xe_create(NULL, "sync-me"); pcmk__node_status_t *peer = NULL; crm_info("Requesting re-sync from %s", (host? host : "all peers")); sync_in_progress = 1; crm_xml_add(sync_me, PCMK__XA_T, PCMK__VALUE_CIB); crm_xml_add(sync_me, PCMK__XA_CIB_OP, PCMK__CIB_REQUEST_SYNC_TO_ONE); crm_xml_add(sync_me, PCMK__XA_CIB_DELEGATED_FROM, stand_alone? "localhost" : crm_cluster->uname); if (host != NULL) { peer = pcmk__get_node(0, host, NULL, pcmk__node_search_cluster_member); } - pcmk__cluster_send_message(peer, crm_msg_cib, sync_me); + pcmk__cluster_send_message(peer, pcmk__cluster_msg_based, sync_me); pcmk__xml_free(sync_me); } int cib_process_ping(const char *op, int options, const char *section, xmlNode * req, xmlNode * input, xmlNode * existing_cib, xmlNode ** result_cib, xmlNode ** answer) { const char *host = crm_element_value(req, PCMK__XA_SRC); const char *seq = crm_element_value(req, PCMK__XA_CIB_PING_ID); char *digest = pcmk__digest_xml(the_cib, true); xmlNode *wrapper = NULL; crm_trace("Processing \"%s\" event %s from %s", op, seq, host); *answer = pcmk__xe_create(NULL, PCMK__XE_PING_RESPONSE); crm_xml_add(*answer, PCMK_XA_CRM_FEATURE_SET, CRM_FEATURE_SET); crm_xml_add(*answer, PCMK__XA_DIGEST, digest); crm_xml_add(*answer, PCMK__XA_CIB_PING_ID, seq); wrapper = pcmk__xe_create(*answer, PCMK__XE_CIB_CALLDATA); if (the_cib != NULL) { pcmk__if_tracing( { /* Append additional detail so the receiver can log the * differences */ pcmk__xml_copy(wrapper, the_cib); }, { // Always include at least the version details const char *name = (const char *) the_cib->name; xmlNode *shallow = pcmk__xe_create(wrapper, name); pcmk__xe_copy_attrs(shallow, the_cib, pcmk__xaf_none); } ); } crm_info("Reporting our current digest to %s: %s for %s.%s.%s", host, digest, crm_element_value(existing_cib, PCMK_XA_ADMIN_EPOCH), crm_element_value(existing_cib, PCMK_XA_EPOCH), crm_element_value(existing_cib, PCMK_XA_NUM_UPDATES)); free(digest); return pcmk_ok; } int cib_process_sync(const char *op, int options, const char *section, xmlNode * req, xmlNode * input, xmlNode * existing_cib, xmlNode ** result_cib, xmlNode ** answer) { return sync_our_cib(req, TRUE); } int cib_process_upgrade_server(const char *op, int options, const char *section, xmlNode * req, xmlNode * input, xmlNode * existing_cib, xmlNode ** result_cib, xmlNode ** answer) { int rc = pcmk_ok; *answer = NULL; if (crm_element_value(req, PCMK__XA_CIB_SCHEMA_MAX) != NULL) { /* The originator of an upgrade request sends it to the DC, without * PCMK__XA_CIB_SCHEMA_MAX. If an upgrade is needed, the DC * re-broadcasts the request with PCMK__XA_CIB_SCHEMA_MAX, and each node * performs the upgrade (and notifies its local clients) here. */ return cib_process_upgrade( op, options, section, req, input, existing_cib, result_cib, answer); } else { xmlNode *scratch = pcmk__xml_copy(NULL, existing_cib); const char *host = crm_element_value(req, PCMK__XA_SRC); const char *original_schema = NULL; const char *new_schema = NULL; const char *client_id = crm_element_value(req, PCMK__XA_CIB_CLIENTID); const char *call_opts = crm_element_value(req, PCMK__XA_CIB_CALLOPT); const char *call_id = crm_element_value(req, PCMK__XA_CIB_CALLID); crm_trace("Processing \"%s\" event", op); original_schema = crm_element_value(existing_cib, PCMK_XA_VALIDATE_WITH); rc = pcmk__update_schema(&scratch, NULL, true, true); rc = pcmk_rc2legacy(rc); new_schema = crm_element_value(scratch, PCMK_XA_VALIDATE_WITH); if (pcmk__cmp_schemas_by_name(new_schema, original_schema) > 0) { xmlNode *up = pcmk__xe_create(NULL, __func__); rc = pcmk_ok; crm_notice("Upgrade request from %s verified", host); crm_xml_add(up, PCMK__XA_T, PCMK__VALUE_CIB); crm_xml_add(up, PCMK__XA_CIB_OP, PCMK__CIB_REQUEST_UPGRADE); crm_xml_add(up, PCMK__XA_CIB_SCHEMA_MAX, new_schema); crm_xml_add(up, PCMK__XA_CIB_DELEGATED_FROM, host); crm_xml_add(up, PCMK__XA_CIB_CLIENTID, client_id); crm_xml_add(up, PCMK__XA_CIB_CALLOPT, call_opts); crm_xml_add(up, PCMK__XA_CIB_CALLID, call_id); - pcmk__cluster_send_message(NULL, crm_msg_cib, up); + pcmk__cluster_send_message(NULL, pcmk__cluster_msg_based, up); pcmk__xml_free(up); } else if(rc == pcmk_ok) { rc = -pcmk_err_schema_unchanged; } if (rc != pcmk_ok) { // Notify originating peer so it can notify its local clients pcmk__node_status_t *origin = NULL; origin = pcmk__search_node_caches(0, host, pcmk__node_search_cluster_member); crm_info("Rejecting upgrade request from %s: %s " QB_XS " rc=%d peer=%s", host, pcmk_strerror(rc), rc, (origin? origin->name : "lost")); if (origin) { xmlNode *up = pcmk__xe_create(NULL, __func__); crm_xml_add(up, PCMK__XA_T, PCMK__VALUE_CIB); crm_xml_add(up, PCMK__XA_CIB_OP, PCMK__CIB_REQUEST_UPGRADE); crm_xml_add(up, PCMK__XA_CIB_DELEGATED_FROM, host); crm_xml_add(up, PCMK__XA_CIB_ISREPLYTO, host); crm_xml_add(up, PCMK__XA_CIB_CLIENTID, client_id); crm_xml_add(up, PCMK__XA_CIB_CALLOPT, call_opts); crm_xml_add(up, PCMK__XA_CIB_CALLID, call_id); crm_xml_add_int(up, PCMK__XA_CIB_UPGRADE_RC, rc); - if (!pcmk__cluster_send_message(origin, crm_msg_cib, up)) { + if (!pcmk__cluster_send_message(origin, pcmk__cluster_msg_based, + up)) { crm_warn("Could not send CIB upgrade result to %s", host); } pcmk__xml_free(up); } } pcmk__xml_free(scratch); } return rc; } int cib_process_sync_one(const char *op, int options, const char *section, xmlNode * req, xmlNode * input, xmlNode * existing_cib, xmlNode ** result_cib, xmlNode ** answer) { return sync_our_cib(req, FALSE); } int cib_server_process_diff(const char *op, int options, const char *section, xmlNode * req, xmlNode * input, xmlNode * existing_cib, xmlNode ** result_cib, xmlNode ** answer) { int rc = pcmk_ok; if (sync_in_progress > MAX_DIFF_RETRY) { /* Don't ignore diffs forever; the last request may have been lost. * If the diff fails, we'll ask for another full resync. */ sync_in_progress = 0; } // The primary instance should never ignore a diff if (sync_in_progress && !based_is_primary) { int diff_add_updates = 0; int diff_add_epoch = 0; int diff_add_admin_epoch = 0; int diff_del_updates = 0; int diff_del_epoch = 0; int diff_del_admin_epoch = 0; cib_diff_version_details(input, &diff_add_admin_epoch, &diff_add_epoch, &diff_add_updates, &diff_del_admin_epoch, &diff_del_epoch, &diff_del_updates); sync_in_progress++; crm_notice("Not applying diff %d.%d.%d -> %d.%d.%d (sync in progress)", diff_del_admin_epoch, diff_del_epoch, diff_del_updates, diff_add_admin_epoch, diff_add_epoch, diff_add_updates); return -pcmk_err_diff_resync; } rc = cib_process_diff(op, options, section, req, input, existing_cib, result_cib, answer); crm_trace("result: %s (%d), %s", pcmk_strerror(rc), rc, (based_is_primary? "primary": "secondary")); if ((rc == -pcmk_err_diff_resync) && !based_is_primary) { pcmk__xml_free(*result_cib); *result_cib = NULL; send_sync_request(NULL); } else if (rc == -pcmk_err_diff_resync) { rc = -pcmk_err_diff_failed; if (options & cib_force_diff) { crm_warn("Not requesting full refresh in R/W mode"); } } return rc; } int cib_process_replace_svr(const char *op, int options, const char *section, xmlNode * req, xmlNode * input, xmlNode * existing_cib, xmlNode ** result_cib, xmlNode ** answer) { int rc = cib_process_replace(op, options, section, req, input, existing_cib, result_cib, answer); if ((rc == pcmk_ok) && pcmk__xe_is(input, PCMK_XE_CIB)) { sync_in_progress = 0; } return rc; } // @COMPAT: Remove when PCMK__CIB_REQUEST_ABS_DELETE is removed int cib_process_delete_absolute(const char *op, int options, const char *section, xmlNode * req, xmlNode * input, xmlNode * existing_cib, xmlNode ** result_cib, xmlNode ** answer) { return -EINVAL; } static xmlNode * cib_msg_copy(xmlNode *msg) { static const char *field_list[] = { PCMK__XA_T, PCMK__XA_CIB_CLIENTID, PCMK__XA_CIB_CALLOPT, PCMK__XA_CIB_CALLID, PCMK__XA_CIB_OP, PCMK__XA_CIB_ISREPLYTO, PCMK__XA_CIB_SECTION, PCMK__XA_CIB_HOST, PCMK__XA_CIB_RC, PCMK__XA_CIB_DELEGATED_FROM, PCMK__XA_CIB_UPDATE, PCMK__XA_CIB_CLIENTNAME, PCMK__XA_CIB_USER, PCMK__XA_CIB_NOTIFY_TYPE, PCMK__XA_CIB_NOTIFY_ACTIVATE, }; xmlNode *copy = pcmk__xe_create(NULL, PCMK__XE_COPY); for (int lpc = 0; lpc < PCMK__NELEM(field_list); lpc++) { const char *field = field_list[lpc]; const char *value = crm_element_value(msg, field); if (value != NULL) { crm_xml_add(copy, field, value); } } return copy; } int sync_our_cib(xmlNode * request, gboolean all) { int result = pcmk_ok; char *digest = NULL; const char *host = crm_element_value(request, PCMK__XA_SRC); const char *op = crm_element_value(request, PCMK__XA_CIB_OP); pcmk__node_status_t *peer = NULL; xmlNode *replace_request = NULL; xmlNode *wrapper = NULL; CRM_CHECK(the_cib != NULL, return -EINVAL); CRM_CHECK(all || (host != NULL), return -EINVAL); crm_debug("Syncing CIB to %s", all ? "all peers" : host); replace_request = cib_msg_copy(request); if (host != NULL) { crm_xml_add(replace_request, PCMK__XA_CIB_ISREPLYTO, host); } if (all) { pcmk__xe_remove_attr(replace_request, PCMK__XA_CIB_HOST); } crm_xml_add(replace_request, PCMK__XA_CIB_OP, PCMK__CIB_REQUEST_REPLACE); // @TODO Keep for tracing, or drop? crm_xml_add(replace_request, PCMK__XA_ORIGINAL_CIB_OP, op); pcmk__xe_set_bool_attr(replace_request, PCMK__XA_CIB_UPDATE, true); crm_xml_add(replace_request, PCMK_XA_CRM_FEATURE_SET, CRM_FEATURE_SET); digest = pcmk__digest_xml(the_cib, true); crm_xml_add(replace_request, PCMK__XA_DIGEST, digest); wrapper = pcmk__xe_create(replace_request, PCMK__XE_CIB_CALLDATA); pcmk__xml_copy(wrapper, the_cib); if (!all) { peer = pcmk__get_node(0, host, NULL, pcmk__node_search_cluster_member); } - if (!pcmk__cluster_send_message(peer, crm_msg_cib, replace_request)) { + if (!pcmk__cluster_send_message(peer, pcmk__cluster_msg_based, + replace_request)) { result = -ENOTCONN; } pcmk__xml_free(replace_request); free(digest); return result; } int cib_process_commit_transaction(const char *op, int options, const char *section, xmlNode *req, xmlNode *input, xmlNode *existing_cib, xmlNode **result_cib, xmlNode **answer) { /* On success, our caller will activate *result_cib locally, trigger a * replace notification if appropriate, and sync *result_cib to all nodes. * On failure, our caller will free *result_cib. */ int rc = pcmk_rc_ok; const char *client_id = crm_element_value(req, PCMK__XA_CIB_CLIENTID); const char *origin = crm_element_value(req, PCMK__XA_SRC); pcmk__client_t *client = pcmk__find_client_by_id(client_id); rc = based_commit_transaction(input, client, origin, result_cib); if (rc != pcmk_rc_ok) { char *source = based_transaction_source_str(client, origin); crm_err("Could not commit transaction for %s: %s", source, pcmk_rc_str(rc)); free(source); } return pcmk_rc2legacy(rc); } int cib_process_schemas(const char *op, int options, const char *section, xmlNode *req, xmlNode *input, xmlNode *existing_cib, xmlNode **result_cib, xmlNode **answer) { xmlNode *wrapper = NULL; xmlNode *data = NULL; const char *after_ver = NULL; GList *schemas = NULL; GList *already_included = NULL; *answer = pcmk__xe_create(NULL, PCMK__XA_SCHEMAS); wrapper = pcmk__xe_first_child(req, PCMK__XE_CIB_CALLDATA, NULL, NULL); data = pcmk__xe_first_child(wrapper, NULL, NULL, NULL); if (data == NULL) { crm_warn("No data specified in request"); return -EPROTO; } after_ver = crm_element_value(data, PCMK_XA_VERSION); if (after_ver == NULL) { crm_warn("No version specified in request"); return -EPROTO; } /* The client requested all schemas after the latest one we know about, which * means the client is fully up-to-date. Return a properly formatted reply * with no schemas. */ if (pcmk__str_eq(after_ver, pcmk__highest_schema_name(), pcmk__str_none)) { return pcmk_ok; } schemas = pcmk__schema_files_later_than(after_ver); for (GList *iter = schemas; iter != NULL; iter = iter->next) { pcmk__build_schema_xml_node(*answer, iter->data, &already_included); } g_list_free_full(schemas, free); g_list_free_full(already_included, free); return pcmk_ok; } diff --git a/daemons/controld/controld_callbacks.c b/daemons/controld/controld_callbacks.c index ef0490827e..ed4da4ec00 100644 --- a/daemons/controld/controld_callbacks.c +++ b/daemons/controld/controld_callbacks.c @@ -1,392 +1,392 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include // PRIu32 #include // bool #include // NULL #include #include #include #include #include #include #include /* From join_dc... */ extern gboolean check_join_state(enum crmd_fsa_state cur_state, const char *source); void crmd_ha_msg_filter(xmlNode * msg) { if (AM_I_DC) { const char *sys_from = crm_element_value(msg, PCMK__XA_CRM_SYS_FROM); if (pcmk__str_eq(sys_from, CRM_SYSTEM_DC, pcmk__str_casei)) { const char *from = crm_element_value(msg, PCMK__XA_SRC); if (!pcmk__str_eq(from, controld_globals.our_nodename, pcmk__str_casei)) { int level = LOG_INFO; const char *op = crm_element_value(msg, PCMK__XA_CRM_TASK); /* make sure the election happens NOW */ if (controld_globals.fsa_state != S_ELECTION) { ha_msg_input_t new_input; level = LOG_WARNING; new_input.msg = msg; register_fsa_error_adv(C_FSA_INTERNAL, I_ELECTION, NULL, &new_input, __func__); } do_crm_log(level, "Another DC detected: %s (op=%s)", from, op); goto done; } } } else { const char *sys_to = crm_element_value(msg, PCMK__XA_CRM_SYS_TO); if (pcmk__str_eq(sys_to, CRM_SYSTEM_DC, pcmk__str_casei)) { return; } } /* crm_log_xml_trace(msg, "HA[inbound]"); */ route_message(C_HA_MESSAGE, msg); done: controld_trigger_fsa(); } /*! * \internal * \brief Check whether a node is online * * \param[in] node Node to check * * \retval -1 if completely dead * \retval 0 if partially alive * \retval 1 if completely alive */ static int node_alive(const pcmk__node_status_t *node) { if (pcmk_is_set(node->flags, pcmk__node_status_remote)) { // Pacemaker Remote nodes can't be partially alive return pcmk__str_eq(node->state, CRM_NODE_MEMBER, pcmk__str_casei) ? 1: -1; } else if (pcmk__cluster_is_node_active(node)) { // Completely up cluster node: both cluster member and peer return 1; } else if (!pcmk_is_set(node->processes, crm_get_cluster_proc()) && !pcmk__str_eq(node->state, CRM_NODE_MEMBER, pcmk__str_casei)) { // Completely down cluster node: neither cluster member nor peer return -1; } // Partially up cluster node: only cluster member or only peer return 0; } #define state_text(state) ((state)? (const char *)(state) : "in unknown state") void peer_update_callback(enum pcmk__node_update type, pcmk__node_status_t *node, const void *data) { uint32_t old = 0; bool appeared = FALSE; bool is_remote = pcmk_is_set(node->flags, pcmk__node_status_remote); controld_node_pending_timer(node); /* The controller waits to receive some information from the membership * layer before declaring itself operational. If this is being called for a * cluster node, indicate that we have it. */ if (!is_remote) { controld_set_fsa_input_flags(R_PEER_DATA); } if ((type == pcmk__node_update_processes) && pcmk_is_set(node->processes, crm_get_cluster_proc()) && !AM_I_DC && !is_remote) { /* * This is a hack until we can send to a nodeid and/or we fix node name lookups * These messages are ignored in crmd_ha_msg_filter() */ xmlNode *query = create_request(CRM_OP_HELLO, NULL, NULL, CRM_SYSTEM_CRMD, CRM_SYSTEM_CRMD, NULL); crm_debug("Sending hello to node %" PRIu32 " so that it learns our " "node name", node->cluster_layer_id); - pcmk__cluster_send_message(node, crm_msg_crmd, query); + pcmk__cluster_send_message(node, pcmk__cluster_msg_controld, query); pcmk__xml_free(query); } if (node->name == NULL) { return; } switch (type) { case pcmk__node_update_name: /* If we've never seen the node, then it also won't be in the status section */ crm_info("%s node %s is now %s", (is_remote? "Remote" : "Cluster"), node->name, state_text(node->state)); return; case pcmk__node_update_state: /* This callback should not be called unless the state actually * changed, but here's a failsafe just in case. */ CRM_CHECK(!pcmk__str_eq(data, node->state, pcmk__str_casei), return); crm_info("%s node %s is now %s (was %s)", (is_remote? "Remote" : "Cluster"), node->name, state_text(node->state), state_text(data)); if (pcmk__str_eq(CRM_NODE_MEMBER, node->state, pcmk__str_casei)) { appeared = TRUE; if (!is_remote) { remove_stonith_cleanup(node->name); } } else { controld_remove_failed_sync_node(node->name); controld_remove_voter(node->name); } crmd_alert_node_event(node); break; case pcmk__node_update_processes: CRM_CHECK(data != NULL, return); old = *(const uint32_t *)data; appeared = pcmk_is_set(node->processes, crm_get_cluster_proc()); { const char *dc_s = controld_globals.dc_name; if ((dc_s == NULL) && AM_I_DC) { dc_s = PCMK_VALUE_TRUE; } crm_info("Node %s is %s a peer " QB_XS " DC=%s old=%#07x new=%#07x", node->name, (appeared? "now" : "no longer"), pcmk__s(dc_s, ""), old, node->processes); } if (!pcmk_is_set((node->processes ^ old), crm_get_cluster_proc())) { /* Peer status did not change. This should not be possible, * since we don't track process flags other than peer status. */ crm_trace("Process flag %#7x did not change from %#7x to %#7x", crm_get_cluster_proc(), old, node->processes); return; } if (!appeared) { node->peer_lost = time(NULL); controld_remove_failed_sync_node(node->name); controld_remove_voter(node->name); } if (!pcmk_is_set(controld_globals.fsa_input_register, R_CIB_CONNECTED)) { crm_trace("Ignoring peer status change because not connected to CIB"); return; } else if (controld_globals.fsa_state == S_STOPPING) { crm_trace("Ignoring peer status change because stopping"); return; } if (!appeared && pcmk__str_eq(node->name, controld_globals.our_nodename, pcmk__str_casei)) { /* Did we get evicted? */ crm_notice("Our peer connection failed"); register_fsa_input(C_CRMD_STATUS_CALLBACK, I_ERROR, NULL); } else if (pcmk__str_eq(node->name, controld_globals.dc_name, pcmk__str_casei) && !pcmk__cluster_is_node_active(node)) { /* The DC has left, so delete its transient attributes and * trigger a new election. * * A DC sends its shutdown request to all peers, who update the * DC's expected state to down. This avoids fencing upon * deletion of its transient attributes. */ crm_notice("Our peer on the DC (%s) is dead", controld_globals.dc_name); register_fsa_input(C_CRMD_STATUS_CALLBACK, I_ELECTION, NULL); controld_delete_node_state(node->name, controld_section_attrs, cib_none); } else if (AM_I_DC || pcmk_is_set(controld_globals.flags, controld_dc_left) || (controld_globals.dc_name == NULL)) { /* This only needs to be done once, so normally the DC should do * it. However if there is no DC, every node must do it, since * there is no other way to ensure some one node does it. */ if (appeared) { te_trigger_stonith_history_sync(FALSE); } else { controld_delete_node_state(node->name, controld_section_attrs, cib_none); } } break; } if (AM_I_DC) { xmlNode *update = NULL; int flags = node_update_peer; int alive = node_alive(node); pcmk__graph_action_t *down = match_down_event(node->xml_id); crm_trace("Alive=%d, appeared=%d, down=%d", alive, appeared, (down? down->id : -1)); if (appeared && (alive > 0) && !is_remote) { register_fsa_input_before(C_FSA_INTERNAL, I_NODE_JOIN, NULL); } if (down) { const char *task = crm_element_value(down->xml, PCMK_XA_OPERATION); if (pcmk__str_eq(task, PCMK_ACTION_STONITH, pcmk__str_casei)) { const bool confirmed = pcmk_is_set(down->flags, pcmk__graph_action_confirmed); /* tengine_stonith_callback() confirms fence actions */ crm_trace("Updating CIB %s fencer reported fencing of %s complete", (confirmed? "after" : "before"), node->name); } else if (!appeared && pcmk__str_eq(task, PCMK_ACTION_DO_SHUTDOWN, pcmk__str_casei)) { // Shutdown actions are immediately confirmed (i.e. no_wait) if (!is_remote) { flags |= node_update_join | node_update_expected; crmd_peer_down(node, FALSE); check_join_state(controld_globals.fsa_state, __func__); } if (alive >= 0) { crm_info("%s of peer %s is in progress " QB_XS " action=%d", task, node->name, down->id); } else { crm_notice("%s of peer %s is complete " QB_XS " action=%d", task, node->name, down->id); pcmk__update_graph(controld_globals.transition_graph, down); trigger_graph(); } } else { const char *liveness = "alive"; if (alive == 0) { liveness = "partially alive"; } else if (alive < 0) { liveness = "dead"; } crm_trace("Node %s is %s, was expected to %s (op %d)", node->name, liveness, task, down->id); } } else if (appeared == FALSE) { if ((controld_globals.transition_graph == NULL) || (controld_globals.transition_graph->id == -1)) { crm_info("Stonith/shutdown of node %s is unknown to the " "current DC", node->name); } else { crm_warn("Stonith/shutdown of node %s was not expected", node->name); } if (!is_remote) { crm_update_peer_join(__func__, node, crm_join_none); check_join_state(controld_globals.fsa_state, __func__); } abort_transition(PCMK_SCORE_INFINITY, pcmk__graph_restart, "Node failure", NULL); fail_incompletable_actions(controld_globals.transition_graph, node->xml_id); } else { crm_trace("Node %s came up, was not expected to be down", node->name); } if (is_remote) { /* A pacemaker_remote node won't have its cluster status updated * in the CIB by membership-layer callbacks, so do it here. */ flags |= node_update_cluster; /* Trigger resource placement on newly integrated nodes */ if (appeared) { abort_transition(PCMK_SCORE_INFINITY, pcmk__graph_restart, "Pacemaker Remote node integrated", NULL); } } if (!appeared && (type == pcmk__node_update_processes) && (node->when_member > 1)) { /* The node left CPG but is still a cluster member. Set its * membership time to 1 to record it in the cluster state as a * boolean, so we don't fence it due to * PCMK_OPT_NODE_PENDING_TIMEOUT. */ node->when_member = 1; flags |= node_update_cluster; controld_node_pending_timer(node); } /* Update the CIB node state */ update = create_node_state_update(node, flags, NULL, __func__); if (update == NULL) { crm_debug("Node state update not yet possible for %s", node->name); } else { fsa_cib_anon_update(PCMK_XE_STATUS, update); } pcmk__xml_free(update); } controld_trigger_fsa(); } gboolean crm_fsa_trigger(gpointer user_data) { crm_trace("Invoked (queue len: %d)", g_list_length(controld_globals.fsa_message_queue)); s_crmd_fsa(C_FSA_INTERNAL); crm_trace("Exited (queue len: %d)", g_list_length(controld_globals.fsa_message_queue)); return TRUE; } diff --git a/daemons/controld/controld_control.c b/daemons/controld/controld_control.c index 2673b829c2..0e09d3b526 100644 --- a/daemons/controld/controld_control.c +++ b/daemons/controld/controld_control.c @@ -1,689 +1,689 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include #include #include static qb_ipcs_service_t *ipcs = NULL; static crm_trigger_t *config_read_trigger = NULL; #if SUPPORT_COROSYNC extern gboolean crm_connect_corosync(pcmk_cluster_t *cluster); #endif static void crm_shutdown(int nsig); static gboolean crm_read_options(gpointer user_data); /* A_HA_CONNECT */ void do_ha_control(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data) { gboolean registered = FALSE; static pcmk_cluster_t *cluster = NULL; if (cluster == NULL) { cluster = pcmk_cluster_new(); } if (action & A_HA_DISCONNECT) { pcmk_cluster_disconnect(cluster); crm_info("Disconnected from the cluster"); controld_set_fsa_input_flags(R_HA_DISCONNECTED); } if (action & A_HA_CONNECT) { pcmk__cluster_set_status_callback(&peer_update_callback); pcmk__cluster_set_autoreap(false); #if SUPPORT_COROSYNC if (pcmk_get_cluster_layer() == pcmk_cluster_layer_corosync) { registered = crm_connect_corosync(cluster); } #endif // SUPPORT_COROSYNC if (registered) { controld_election_init(cluster->uname); controld_globals.our_nodename = cluster->uname; controld_globals.our_uuid = cluster->uuid; if(cluster->uuid == NULL) { crm_err("Could not obtain local uuid"); registered = FALSE; } } if (!registered) { controld_set_fsa_input_flags(R_HA_DISCONNECTED); register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); return; } populate_cib_nodes(node_update_none, __func__); controld_clear_fsa_input_flags(R_HA_DISCONNECTED); crm_info("Connected to the cluster"); } if (action & ~(A_HA_CONNECT | A_HA_DISCONNECT)) { crm_err("Unexpected action %s in %s", fsa_action2string(action), __func__); } } /* A_SHUTDOWN */ void do_shutdown(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data) { /* just in case */ controld_set_fsa_input_flags(R_SHUTDOWN); controld_disconnect_fencer(FALSE); } /* A_SHUTDOWN_REQ */ void do_shutdown_req(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data) { xmlNode *msg = NULL; controld_set_fsa_input_flags(R_SHUTDOWN); //controld_set_fsa_input_flags(R_STAYDOWN); crm_info("Sending shutdown request to all peers (DC is %s)", pcmk__s(controld_globals.dc_name, "not set")); msg = create_request(CRM_OP_SHUTDOWN_REQ, NULL, NULL, CRM_SYSTEM_CRMD, CRM_SYSTEM_CRMD, NULL); - if (!pcmk__cluster_send_message(NULL, crm_msg_crmd, msg)) { + if (!pcmk__cluster_send_message(NULL, pcmk__cluster_msg_controld, msg)) { register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); } pcmk__xml_free(msg); } void crmd_fast_exit(crm_exit_t exit_code) { if (pcmk_is_set(controld_globals.fsa_input_register, R_STAYDOWN)) { crm_warn("Inhibiting respawn " QB_XS " remapping exit code %d to %d", exit_code, CRM_EX_FATAL); exit_code = CRM_EX_FATAL; } else if ((exit_code == CRM_EX_OK) && pcmk_is_set(controld_globals.fsa_input_register, R_IN_RECOVERY)) { crm_err("Could not recover from internal error"); exit_code = CRM_EX_ERROR; } if (controld_globals.logger_out != NULL) { controld_globals.logger_out->finish(controld_globals.logger_out, exit_code, true, NULL); pcmk__output_free(controld_globals.logger_out); controld_globals.logger_out = NULL; } crm_exit(exit_code); } crm_exit_t crmd_exit(crm_exit_t exit_code) { GMainLoop *mloop = controld_globals.mainloop; static bool in_progress = FALSE; if (in_progress && (exit_code == CRM_EX_OK)) { crm_debug("Exit is already in progress"); return exit_code; } else if(in_progress) { crm_notice("Error during shutdown process, exiting now with status %d (%s)", exit_code, crm_exit_str(exit_code)); crm_write_blackbox(SIGTRAP, NULL); crmd_fast_exit(exit_code); } in_progress = TRUE; crm_trace("Preparing to exit with status %d (%s)", exit_code, crm_exit_str(exit_code)); /* Suppress secondary errors resulting from us disconnecting everything */ controld_set_fsa_input_flags(R_HA_DISCONNECTED); /* Close all IPC servers and clients to ensure any and all shared memory files are cleaned up */ if(ipcs) { crm_trace("Closing IPC server"); mainloop_del_ipc_server(ipcs); ipcs = NULL; } controld_close_attrd_ipc(); controld_shutdown_schedulerd_ipc(); controld_disconnect_fencer(TRUE); if ((exit_code == CRM_EX_OK) && (controld_globals.mainloop == NULL)) { crm_debug("No mainloop detected"); exit_code = CRM_EX_ERROR; } /* On an error, just get out. * * Otherwise, make the effort to have mainloop exit gracefully so * that it (mostly) cleans up after itself and valgrind has less * to report on - allowing real errors stand out */ if (exit_code != CRM_EX_OK) { crm_notice("Forcing immediate exit with status %d (%s)", exit_code, crm_exit_str(exit_code)); crm_write_blackbox(SIGTRAP, NULL); crmd_fast_exit(exit_code); } /* Clean up as much memory as possible for valgrind */ for (GList *iter = controld_globals.fsa_message_queue; iter != NULL; iter = iter->next) { fsa_data_t *fsa_data = (fsa_data_t *) iter->data; crm_info("Dropping %s: [ state=%s cause=%s origin=%s ]", fsa_input2string(fsa_data->fsa_input), fsa_state2string(controld_globals.fsa_state), fsa_cause2string(fsa_data->fsa_cause), fsa_data->origin); delete_fsa_input(fsa_data); } controld_clear_fsa_input_flags(R_MEMBERSHIP); g_list_free(controld_globals.fsa_message_queue); controld_globals.fsa_message_queue = NULL; controld_free_node_pending_timers(); controld_election_fini(); /* Tear down the CIB manager connection, but don't free it yet -- it could * be used when we drain the mainloop later. */ controld_disconnect_cib_manager(); verify_stopped(controld_globals.fsa_state, LOG_WARNING); controld_clear_fsa_input_flags(R_LRM_CONNECTED); lrm_state_destroy_all(); mainloop_destroy_trigger(config_read_trigger); config_read_trigger = NULL; controld_destroy_fsa_trigger(); controld_destroy_transition_trigger(); pcmk__client_cleanup(); pcmk__cluster_destroy_node_caches(); controld_free_fsa_timers(); te_cleanup_stonith_history_sync(NULL, TRUE); controld_free_sched_timer(); free(controld_globals.our_nodename); controld_globals.our_nodename = NULL; free(controld_globals.our_uuid); controld_globals.our_uuid = NULL; free(controld_globals.dc_name); controld_globals.dc_name = NULL; free(controld_globals.dc_version); controld_globals.dc_version = NULL; free(controld_globals.cluster_name); controld_globals.cluster_name = NULL; free(controld_globals.te_uuid); controld_globals.te_uuid = NULL; free_max_generation(); controld_destroy_failed_sync_table(); controld_destroy_outside_events_table(); mainloop_destroy_signal(SIGPIPE); mainloop_destroy_signal(SIGUSR1); mainloop_destroy_signal(SIGTERM); mainloop_destroy_signal(SIGTRAP); /* leave SIGCHLD engaged as we might still want to drain some service-actions */ if (mloop) { GMainContext *ctx = g_main_loop_get_context(controld_globals.mainloop); /* Don't re-enter this block */ controld_globals.mainloop = NULL; /* no signals on final draining anymore */ mainloop_destroy_signal(SIGCHLD); crm_trace("Draining mainloop %d %d", g_main_loop_is_running(mloop), g_main_context_pending(ctx)); { int lpc = 0; while((g_main_context_pending(ctx) && lpc < 10)) { lpc++; crm_trace("Iteration %d", lpc); g_main_context_dispatch(ctx); } } crm_trace("Closing mainloop %d %d", g_main_loop_is_running(mloop), g_main_context_pending(ctx)); g_main_loop_quit(mloop); /* Won't do anything yet, since we're inside it now */ g_main_loop_unref(mloop); } else { mainloop_destroy_signal(SIGCHLD); } cib_delete(controld_globals.cib_conn); controld_globals.cib_conn = NULL; throttle_fini(); /* Graceful */ crm_trace("Done preparing for exit with status %d (%s)", exit_code, crm_exit_str(exit_code)); return exit_code; } /* A_EXIT_0, A_EXIT_1 */ void do_exit(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data) { crm_exit_t exit_code = CRM_EX_OK; if (pcmk_is_set(action, A_EXIT_1)) { exit_code = CRM_EX_ERROR; crm_err("Exiting now due to errors"); } verify_stopped(cur_state, LOG_ERR); crmd_exit(exit_code); } static void sigpipe_ignore(int nsig) { return; } /* A_STARTUP */ void do_startup(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data) { crm_debug("Registering Signal Handlers"); mainloop_add_signal(SIGTERM, crm_shutdown); mainloop_add_signal(SIGPIPE, sigpipe_ignore); config_read_trigger = mainloop_add_trigger(G_PRIORITY_HIGH, crm_read_options, NULL); controld_init_fsa_trigger(); controld_init_transition_trigger(); crm_debug("Creating CIB manager and executor objects"); controld_globals.cib_conn = cib_new(); lrm_state_init_local(); if (controld_init_fsa_timers() == FALSE) { register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); } } // \return libqb error code (0 on success, -errno on error) static int32_t accept_controller_client(qb_ipcs_connection_t *c, uid_t uid, gid_t gid) { crm_trace("Accepting new IPC client connection"); if (pcmk__new_client(c, uid, gid) == NULL) { return -ENOMEM; } return 0; } // \return libqb error code (0 on success, -errno on error) static int32_t dispatch_controller_ipc(qb_ipcs_connection_t * c, void *data, size_t size) { uint32_t id = 0; uint32_t flags = 0; pcmk__client_t *client = pcmk__find_client(c); xmlNode *msg = pcmk__client_data2xml(client, data, &id, &flags); if (msg == NULL) { pcmk__ipc_send_ack(client, id, flags, PCMK__XE_ACK, NULL, CRM_EX_PROTOCOL); return 0; } pcmk__ipc_send_ack(client, id, flags, PCMK__XE_ACK, NULL, CRM_EX_INDETERMINATE); CRM_ASSERT(client->user != NULL); pcmk__update_acl_user(msg, PCMK__XA_CRM_USER, client->user); crm_xml_add(msg, PCMK__XA_CRM_SYS_FROM, client->id); if (controld_authorize_ipc_message(msg, client, NULL)) { crm_trace("Processing IPC message from client %s", pcmk__client_name(client)); route_message(C_IPC_MESSAGE, msg); } controld_trigger_fsa(); pcmk__xml_free(msg); return 0; } static int32_t ipc_client_disconnected(qb_ipcs_connection_t *c) { pcmk__client_t *client = pcmk__find_client(c); if (client) { crm_trace("Disconnecting %sregistered client %s (%p/%p)", (client->userdata? "" : "un"), pcmk__client_name(client), c, client); free(client->userdata); pcmk__free_client(client); controld_trigger_fsa(); } return 0; } static void ipc_connection_destroyed(qb_ipcs_connection_t *c) { crm_trace("Connection %p", c); ipc_client_disconnected(c); } /* A_STOP */ void do_stop(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data) { crm_trace("Closing IPC server"); mainloop_del_ipc_server(ipcs); ipcs = NULL; register_fsa_input(C_FSA_INTERNAL, I_TERMINATE, NULL); } /* A_STARTED */ void do_started(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data) { static struct qb_ipcs_service_handlers crmd_callbacks = { .connection_accept = accept_controller_client, .connection_created = NULL, .msg_process = dispatch_controller_ipc, .connection_closed = ipc_client_disconnected, .connection_destroyed = ipc_connection_destroyed }; if (cur_state != S_STARTING) { crm_err("Start cancelled... %s", fsa_state2string(cur_state)); return; } else if (!pcmk_is_set(controld_globals.fsa_input_register, R_MEMBERSHIP)) { crm_info("Delaying start, no membership data (%.16llx)", R_MEMBERSHIP); crmd_fsa_stall(TRUE); return; } else if (!pcmk_is_set(controld_globals.fsa_input_register, R_LRM_CONNECTED)) { crm_info("Delaying start, not connected to executor (%.16llx)", R_LRM_CONNECTED); crmd_fsa_stall(TRUE); return; } else if (!pcmk_is_set(controld_globals.fsa_input_register, R_CIB_CONNECTED)) { crm_info("Delaying start, CIB not connected (%.16llx)", R_CIB_CONNECTED); crmd_fsa_stall(TRUE); return; } else if (!pcmk_is_set(controld_globals.fsa_input_register, R_READ_CONFIG)) { crm_info("Delaying start, Config not read (%.16llx)", R_READ_CONFIG); crmd_fsa_stall(TRUE); return; } else if (!pcmk_is_set(controld_globals.fsa_input_register, R_PEER_DATA)) { crm_info("Delaying start, No peer data (%.16llx)", R_PEER_DATA); crmd_fsa_stall(TRUE); return; } crm_debug("Init server comms"); ipcs = pcmk__serve_controld_ipc(&crmd_callbacks); if (ipcs == NULL) { crm_err("Failed to create IPC server: shutting down and inhibiting respawn"); register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); } else { crm_notice("Pacemaker controller successfully started and accepting connections"); } controld_set_fsa_input_flags(R_ST_REQUIRED); controld_timer_fencer_connect(GINT_TO_POINTER(TRUE)); controld_clear_fsa_input_flags(R_STARTING); register_fsa_input(msg_data->fsa_cause, I_PENDING, NULL); } /* A_RECOVER */ void do_recover(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data) { controld_set_fsa_input_flags(R_IN_RECOVERY); crm_warn("Fast-tracking shutdown in response to errors"); register_fsa_input(C_FSA_INTERNAL, I_TERMINATE, NULL); } static void config_query_callback(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data) { const char *value = NULL; GHashTable *config_hash = NULL; crm_time_t *now = crm_time_new(NULL); xmlNode *crmconfig = NULL; xmlNode *alerts = NULL; if (rc != pcmk_ok) { fsa_data_t *msg_data = NULL; crm_err("Local CIB query resulted in an error: %s", pcmk_strerror(rc)); register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); if (rc == -EACCES || rc == -pcmk_err_schema_validation) { crm_err("The cluster is mis-configured - shutting down and staying down"); controld_set_fsa_input_flags(R_STAYDOWN); } goto bail; } crmconfig = output; if ((crmconfig != NULL) && !pcmk__xe_is(crmconfig, PCMK_XE_CRM_CONFIG)) { crmconfig = pcmk__xe_first_child(crmconfig, PCMK_XE_CRM_CONFIG, NULL, NULL); } if (!crmconfig) { fsa_data_t *msg_data = NULL; crm_err("Local CIB query for " PCMK_XE_CRM_CONFIG " section failed"); register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); goto bail; } crm_debug("Call %d : Parsing CIB options", call_id); config_hash = pcmk__strkey_table(free, free); pe_unpack_nvpairs(crmconfig, crmconfig, PCMK_XE_CLUSTER_PROPERTY_SET, NULL, config_hash, PCMK_VALUE_CIB_BOOTSTRAP_OPTIONS, FALSE, now, NULL); // Validate all options, and use defaults if not already present in hash pcmk__validate_cluster_options(config_hash); /* Validate the watchdog timeout in the context of the local node * environment. If invalid, the controller will exit with a fatal error. * * We do this via a wrapper in the controller, so that we call * pcmk__valid_stonith_watchdog_timeout() only if watchdog fencing is * enabled for the local node. Otherwise, we may exit unnecessarily. * * A validator function in libcrmcommon can't act as such a wrapper, because * it doesn't have a stonith API connection or the local node name. */ value = g_hash_table_lookup(config_hash, PCMK_OPT_STONITH_WATCHDOG_TIMEOUT); controld_verify_stonith_watchdog_timeout(value); value = g_hash_table_lookup(config_hash, PCMK_OPT_NO_QUORUM_POLICY); if (pcmk__str_eq(value, PCMK_VALUE_FENCE_LEGACY, pcmk__str_casei) && (pcmk__locate_sbd() != 0)) { controld_set_global_flags(controld_no_quorum_suicide); } value = g_hash_table_lookup(config_hash, PCMK_OPT_SHUTDOWN_LOCK); if (crm_is_true(value)) { controld_set_global_flags(controld_shutdown_lock_enabled); } else { controld_clear_global_flags(controld_shutdown_lock_enabled); } value = g_hash_table_lookup(config_hash, PCMK_OPT_SHUTDOWN_LOCK_LIMIT); pcmk_parse_interval_spec(value, &controld_globals.shutdown_lock_limit); controld_globals.shutdown_lock_limit /= 1000; value = g_hash_table_lookup(config_hash, PCMK_OPT_NODE_PENDING_TIMEOUT); pcmk_parse_interval_spec(value, &controld_globals.node_pending_timeout); controld_globals.node_pending_timeout /= 1000; value = g_hash_table_lookup(config_hash, PCMK_OPT_CLUSTER_NAME); pcmk__str_update(&(controld_globals.cluster_name), value); // Let subcomponents initialize their own static variables controld_configure_election(config_hash); controld_configure_fencing(config_hash); controld_configure_fsa_timers(config_hash); controld_configure_throttle(config_hash); alerts = pcmk__xe_first_child(output, PCMK_XE_ALERTS, NULL, NULL); crmd_unpack_alerts(alerts); controld_set_fsa_input_flags(R_READ_CONFIG); controld_trigger_fsa(); g_hash_table_destroy(config_hash); bail: crm_time_free(now); } /*! * \internal * \brief Trigger read and processing of the configuration * * \param[in] fn Calling function name * \param[in] line Line number where call occurred */ void controld_trigger_config_as(const char *fn, int line) { if (config_read_trigger != NULL) { crm_trace("%s:%d - Triggered config processing", fn, line); mainloop_set_trigger(config_read_trigger); } } gboolean crm_read_options(gpointer user_data) { cib_t *cib_conn = controld_globals.cib_conn; int call_id = cib_conn->cmds->query(cib_conn, "//" PCMK_XE_CRM_CONFIG " | //" PCMK_XE_ALERTS, NULL, cib_xpath); fsa_register_cib_callback(call_id, NULL, config_query_callback); crm_trace("Querying the CIB... call %d", call_id); return TRUE; } /* A_READCONFIG */ void do_read_config(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data) { throttle_init(); controld_trigger_config(); } static void crm_shutdown(int nsig) { const char *value = NULL; guint default_period_ms = 0; if ((controld_globals.mainloop == NULL) || !g_main_loop_is_running(controld_globals.mainloop)) { crmd_exit(CRM_EX_OK); return; } if (pcmk_is_set(controld_globals.fsa_input_register, R_SHUTDOWN)) { crm_err("Escalating shutdown"); register_fsa_input_before(C_SHUTDOWN, I_ERROR, NULL); return; } controld_set_fsa_input_flags(R_SHUTDOWN); register_fsa_input(C_SHUTDOWN, I_SHUTDOWN, NULL); /* If shutdown timer doesn't have a period set, use the default * * @TODO: Evaluate whether this is still necessary. As long as * config_query_callback() has been run at least once, it doesn't look like * anything could have changed the timer period since then. */ value = pcmk__cluster_option(NULL, PCMK_OPT_SHUTDOWN_ESCALATION); pcmk_parse_interval_spec(value, &default_period_ms); controld_shutdown_start_countdown(default_period_ms); } diff --git a/daemons/controld/controld_join_client.c b/daemons/controld/controld_join_client.c index b6d038df44..5df40d37ce 100644 --- a/daemons/controld/controld_join_client.c +++ b/daemons/controld/controld_join_client.c @@ -1,370 +1,370 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include void join_query_callback(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data); extern ha_msg_input_t *copy_ha_msg_input(ha_msg_input_t * orig); /*! * \internal * \brief Remember if DC is shutting down as we join * * If we're joining while the current DC is shutting down, update its expected * state, so we don't fence it if we become the new DC. (We weren't a peer * when it broadcast its shutdown request.) * * \param[in] msg A join message from the DC */ static void update_dc_expected(const xmlNode *msg) { if ((controld_globals.dc_name != NULL) && pcmk__xe_attr_is_true(msg, PCMK__XA_DC_LEAVING)) { pcmk__node_status_t *dc_node = pcmk__get_node(0, controld_globals.dc_name, NULL, pcmk__node_search_cluster_member); pcmk__update_peer_expected(__func__, dc_node, CRMD_JOINSTATE_DOWN); } } /* A_CL_JOIN_QUERY */ /* is there a DC out there? */ void do_cl_join_query(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data) { xmlNode *req = create_request(CRM_OP_JOIN_ANNOUNCE, NULL, NULL, CRM_SYSTEM_DC, CRM_SYSTEM_CRMD, NULL); sleep(1); // Give the cluster layer time to propagate to the DC update_dc(NULL); /* Unset any existing value so that the result is not discarded */ crm_debug("Querying for a DC"); - pcmk__cluster_send_message(NULL, crm_msg_crmd, req); + pcmk__cluster_send_message(NULL, pcmk__cluster_msg_controld, req); pcmk__xml_free(req); } /* A_CL_JOIN_ANNOUNCE */ /* this is kind of a workaround for the fact that we may not be around or * are otherwise unable to reply when the DC sends out A_DC_JOIN_OFFER_ALL */ void do_cl_join_announce(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data) { /* don't announce if we're in one of these states */ if (cur_state != S_PENDING) { crm_warn("Not announcing cluster join because in state %s", fsa_state2string(cur_state)); return; } if (!pcmk_is_set(controld_globals.fsa_input_register, R_STARTING)) { /* send as a broadcast */ xmlNode *req = create_request(CRM_OP_JOIN_ANNOUNCE, NULL, NULL, CRM_SYSTEM_DC, CRM_SYSTEM_CRMD, NULL); crm_debug("Announcing availability"); update_dc(NULL); - pcmk__cluster_send_message(NULL, crm_msg_crmd, req); + pcmk__cluster_send_message(NULL, pcmk__cluster_msg_controld, req); pcmk__xml_free(req); } else { /* Delay announce until we have finished local startup */ crm_warn("Delaying announce of cluster join until local startup is complete"); return; } } static int query_call_id = 0; /* A_CL_JOIN_REQUEST */ /* aka. accept the welcome offer */ void do_cl_join_offer_respond(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data) { cib_t *cib_conn = controld_globals.cib_conn; ha_msg_input_t *input = fsa_typed_data(fsa_dt_ha_msg); const char *welcome_from; const char *join_id; CRM_CHECK(input != NULL, return); welcome_from = crm_element_value(input->msg, PCMK__XA_SRC); join_id = crm_element_value(input->msg, PCMK__XA_JOIN_ID); crm_trace("Accepting cluster join offer from node %s " QB_XS " join-%s", welcome_from, crm_element_value(input->msg, PCMK__XA_JOIN_ID)); /* we only ever want the last one */ if (query_call_id > 0) { crm_trace("Cancelling previous join query: %d", query_call_id); remove_cib_op_callback(query_call_id, FALSE); query_call_id = 0; } if (update_dc(input->msg) == FALSE) { crm_warn("Discarding cluster join offer from node %s (expected %s)", welcome_from, controld_globals.dc_name); return; } update_dc_expected(input->msg); query_call_id = cib_conn->cmds->query(cib_conn, NULL, NULL, cib_no_children); fsa_register_cib_callback(query_call_id, pcmk__str_copy(join_id), join_query_callback); crm_trace("Registered join query callback: %d", query_call_id); controld_set_fsa_action_flags(A_DC_TIMER_STOP); controld_trigger_fsa(); } void join_query_callback(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data) { char *join_id = user_data; xmlNode *generation = pcmk__xe_create(NULL, PCMK__XE_GENERATION_TUPLE); CRM_LOG_ASSERT(join_id != NULL); if (query_call_id != call_id) { crm_trace("Query %d superseded", call_id); goto done; } query_call_id = 0; if(rc != pcmk_ok || output == NULL) { crm_err("Could not retrieve version details for join-%s: %s (%d)", join_id, pcmk_strerror(rc), rc); register_fsa_error_adv(C_FSA_INTERNAL, I_ERROR, NULL, NULL, __func__); } else if (controld_globals.dc_name == NULL) { crm_debug("Membership is in flux, not continuing join-%s", join_id); } else { xmlNode *reply = NULL; const pcmk__node_status_t *dc_node = pcmk__get_node(0, controld_globals.dc_name, NULL, pcmk__node_search_cluster_member); crm_debug("Respond to join offer join-%s from %s", join_id, controld_globals.dc_name); pcmk__xe_copy_attrs(generation, output, pcmk__xaf_none); reply = create_request(CRM_OP_JOIN_REQUEST, generation, controld_globals.dc_name, CRM_SYSTEM_DC, CRM_SYSTEM_CRMD, NULL); crm_xml_add(reply, PCMK__XA_JOIN_ID, join_id); crm_xml_add(reply, PCMK_XA_CRM_FEATURE_SET, CRM_FEATURE_SET); - pcmk__cluster_send_message(dc_node, crm_msg_crmd, reply); + pcmk__cluster_send_message(dc_node, pcmk__cluster_msg_controld, reply); pcmk__xml_free(reply); } done: pcmk__xml_free(generation); } void set_join_state(const char *start_state, const char *node_name, const char *node_uuid, bool remote) { if (pcmk__str_eq(start_state, PCMK_VALUE_STANDBY, pcmk__str_casei)) { crm_notice("Forcing node %s to join in %s state per configured " "environment", node_name, start_state); cib__update_node_attr(controld_globals.logger_out, controld_globals.cib_conn, cib_sync_call, PCMK_XE_NODES, node_uuid, NULL, NULL, NULL, PCMK_NODE_ATTR_STANDBY, PCMK_VALUE_TRUE, NULL, (remote? PCMK_VALUE_REMOTE : NULL)); } else if (pcmk__str_eq(start_state, PCMK_VALUE_ONLINE, pcmk__str_casei)) { crm_notice("Forcing node %s to join in %s state per configured " "environment", node_name, start_state); cib__update_node_attr(controld_globals.logger_out, controld_globals.cib_conn, cib_sync_call, PCMK_XE_NODES, node_uuid, NULL, NULL, NULL, PCMK_NODE_ATTR_STANDBY, PCMK_VALUE_FALSE, NULL, (remote? PCMK_VALUE_REMOTE : NULL)); } else if (pcmk__str_eq(start_state, PCMK_VALUE_DEFAULT, pcmk__str_casei)) { crm_debug("Not forcing a starting state on node %s", node_name); } else { crm_warn("Unrecognized start state '%s', using " "'" PCMK_VALUE_DEFAULT "' (%s)", start_state, node_name); } } static int update_conn_host_cache(xmlNode *node, void *userdata) { const char *remote = crm_element_value(node, PCMK_XA_ID); const char *conn_host = crm_element_value(node, PCMK__XA_CONNECTION_HOST); const char *state = crm_element_value(node, PCMK__XA_NODE_STATE); pcmk__node_status_t *remote_peer = pcmk__cluster_lookup_remote_node(remote); if (remote_peer == NULL) { return pcmk_rc_ok; } if (conn_host != NULL) { pcmk__str_update(&remote_peer->conn_host, conn_host); } if (state != NULL) { pcmk__update_peer_state(__func__, remote_peer, state, 0); } return pcmk_rc_ok; } /* A_CL_JOIN_RESULT */ /* aka. this is notification that we have (or have not) been accepted */ void do_cl_join_finalize_respond(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data) { xmlNode *tmp1 = NULL; gboolean was_nack = TRUE; static gboolean first_join = TRUE; ha_msg_input_t *input = fsa_typed_data(fsa_dt_ha_msg); const char *start_state = pcmk__env_option(PCMK__ENV_NODE_START_STATE); int join_id = -1; const char *op = crm_element_value(input->msg, PCMK__XA_CRM_TASK); const char *welcome_from = crm_element_value(input->msg, PCMK__XA_SRC); if (!pcmk__str_eq(op, CRM_OP_JOIN_ACKNAK, pcmk__str_casei)) { crm_trace("Ignoring op=%s message", op); return; } /* calculate if it was an ack or a nack */ if (pcmk__xe_attr_is_true(input->msg, CRM_OP_JOIN_ACKNAK)) { was_nack = FALSE; } crm_element_value_int(input->msg, PCMK__XA_JOIN_ID, &join_id); if (was_nack) { crm_err("Shutting down because cluster join with leader %s failed " QB_XS " join-%d NACK'd", welcome_from, join_id); register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); controld_set_fsa_input_flags(R_STAYDOWN); return; } if (!AM_I_DC && pcmk__str_eq(welcome_from, controld_globals.our_nodename, pcmk__str_casei)) { crm_warn("Discarding our own welcome - we're no longer the DC"); return; } if (update_dc(input->msg) == FALSE) { crm_warn("Discarding %s from node %s (expected from %s)", op, welcome_from, controld_globals.dc_name); return; } update_dc_expected(input->msg); /* record the node's feature set as a transient attribute */ update_attrd(controld_globals.our_nodename, CRM_ATTR_FEATURE_SET, CRM_FEATURE_SET, NULL, FALSE); /* send our status section to the DC */ tmp1 = controld_query_executor_state(); if (tmp1 != NULL) { xmlNode *remotes = NULL; xmlNode *reply = create_request(CRM_OP_JOIN_CONFIRM, tmp1, controld_globals.dc_name, CRM_SYSTEM_DC, CRM_SYSTEM_CRMD, NULL); const pcmk__node_status_t *dc_node = pcmk__get_node(0, controld_globals.dc_name, NULL, pcmk__node_search_cluster_member); crm_xml_add_int(reply, PCMK__XA_JOIN_ID, join_id); crm_debug("Confirming join-%d: sending local operation history to %s", join_id, controld_globals.dc_name); /* * If this is the node's first join since the controller started on it, * set its initial state (standby or member) according to the user's * preference. * * We do not clear the LRM history here. Even if the DC failed to do it * when we last left, removing them here creates a race condition if the * controller is being recovered. Instead of a list of active resources * from the executor, we may end up with a blank status section. If we * are _NOT_ lucky, we will probe for the "wrong" instance of anonymous * clones and end up with multiple active instances on the machine. */ if (first_join && !pcmk_is_set(controld_globals.fsa_input_register, R_SHUTDOWN)) { first_join = FALSE; if (start_state) { set_join_state(start_state, controld_globals.our_nodename, controld_globals.our_uuid, false); } } - pcmk__cluster_send_message(dc_node, crm_msg_crmd, reply); + pcmk__cluster_send_message(dc_node, pcmk__cluster_msg_controld, reply); pcmk__xml_free(reply); if (AM_I_DC == FALSE) { register_fsa_input_adv(cause, I_NOT_DC, NULL, A_NOTHING, TRUE, __func__); } pcmk__xml_free(tmp1); /* Update the remote node cache with information about which node * is hosting the connection. */ remotes = pcmk__xe_first_child(input->msg, PCMK_XE_NODES, NULL, NULL); if (remotes != NULL) { pcmk__xe_foreach_child(remotes, PCMK_XE_NODE, update_conn_host_cache, NULL); } } else { crm_err("Could not confirm join-%d with %s: Local operation history " "failed", join_id, controld_globals.dc_name); register_fsa_error(C_FSA_INTERNAL, I_FAIL, NULL); } } diff --git a/daemons/controld/controld_join_dc.c b/daemons/controld/controld_join_dc.c index f7740f7082..c2d8ef3cb1 100644 --- a/daemons/controld/controld_join_dc.c +++ b/daemons/controld/controld_join_dc.c @@ -1,1057 +1,1057 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include // PRIu32 #include // bool, true, false #include // NULL #include // free(), etc. #include // gboolean, etc. #include // xmlNode #include #include #include #include static char *max_generation_from = NULL; static xmlNodePtr max_generation_xml = NULL; /*! * \internal * \brief Nodes from which a CIB sync has failed since the peer joined * * This table is of the form (node_name -> join_id). \p node_name is * the name of a client node from which a CIB \p sync_from() call has failed in * \p do_dc_join_finalize() since the client joined the cluster as a peer. * \p join_id is the ID of the join round in which the \p sync_from() failed, * and is intended for use in nack log messages. */ static GHashTable *failed_sync_nodes = NULL; void finalize_join_for(gpointer key, gpointer value, gpointer user_data); void finalize_sync_callback(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data); gboolean check_join_state(enum crmd_fsa_state cur_state, const char *source); /* Numeric counter used to identify join rounds (an unsigned int would be * appropriate, except we get and set it in XML as int) */ static int current_join_id = 0; /*! * \internal * \brief Destroy the hash table containing failed sync nodes */ void controld_destroy_failed_sync_table(void) { if (failed_sync_nodes != NULL) { g_hash_table_destroy(failed_sync_nodes); failed_sync_nodes = NULL; } } /*! * \internal * \brief Remove a node from the failed sync nodes table if present * * \param[in] node_name Node name to remove */ void controld_remove_failed_sync_node(const char *node_name) { if (failed_sync_nodes != NULL) { g_hash_table_remove(failed_sync_nodes, (gchar *) node_name); } } /*! * \internal * \brief Add to a hash table a node whose CIB failed to sync * * \param[in] node_name Name of node whose CIB failed to sync * \param[in] join_id Join round when the failure occurred */ static void record_failed_sync_node(const char *node_name, gint join_id) { if (failed_sync_nodes == NULL) { failed_sync_nodes = pcmk__strikey_table(g_free, NULL); } /* If the node is already in the table then we failed to nack it during the * filter offer step */ CRM_LOG_ASSERT(g_hash_table_insert(failed_sync_nodes, g_strdup(node_name), GINT_TO_POINTER(join_id))); } /*! * \internal * \brief Look up a node name in the failed sync table * * \param[in] node_name Name of node to look up * \param[out] join_id Where to store the join ID of when the sync failed * * \return Standard Pacemaker return code. Specifically, \p pcmk_rc_ok if the * node name was found, or \p pcmk_rc_node_unknown otherwise. * \note \p *join_id is set to -1 if the node is not found. */ static int lookup_failed_sync_node(const char *node_name, gint *join_id) { *join_id = -1; if (failed_sync_nodes != NULL) { gpointer result = g_hash_table_lookup(failed_sync_nodes, (gchar *) node_name); if (result != NULL) { *join_id = GPOINTER_TO_INT(result); return pcmk_rc_ok; } } return pcmk_rc_node_unknown; } void crm_update_peer_join(const char *source, pcmk__node_status_t *node, enum crm_join_phase phase) { enum crm_join_phase last = 0; CRM_CHECK(node != NULL, return); /* Remote nodes do not participate in joins */ if (pcmk_is_set(node->flags, pcmk__node_status_remote)) { return; } last = node->join; if(phase == last) { crm_trace("Node %s join-%d phase is still %s " QB_XS " nodeid=%" PRIu32 " source=%s", node->name, current_join_id, controld_join_phase_text(last), node->cluster_layer_id, source); } else if ((phase <= crm_join_none) || (phase == (last + 1))) { node->join = phase; crm_trace("Node %s join-%d phase is now %s (was %s) " QB_XS " nodeid=%" PRIu32 " source=%s", node->name, current_join_id, controld_join_phase_text(phase), controld_join_phase_text(last), node->cluster_layer_id, source); } else { crm_warn("Rejecting join-%d phase update for node %s because " "can't go from %s to %s " QB_XS " nodeid=%" PRIu32 " source=%s", current_join_id, node->name, controld_join_phase_text(last), controld_join_phase_text(phase), node->cluster_layer_id, source); } } static void start_join_round(void) { GHashTableIter iter; pcmk__node_status_t *peer = NULL; crm_debug("Starting new join round join-%d", current_join_id); g_hash_table_iter_init(&iter, crm_peer_cache); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &peer)) { crm_update_peer_join(__func__, peer, crm_join_none); } if (max_generation_from != NULL) { free(max_generation_from); max_generation_from = NULL; } if (max_generation_xml != NULL) { pcmk__xml_free(max_generation_xml); max_generation_xml = NULL; } controld_clear_fsa_input_flags(R_HAVE_CIB); } /*! * \internal * \brief Create a join message from the DC * * \param[in] join_op Join operation name * \param[in] host_to Recipient of message */ static xmlNode * create_dc_message(const char *join_op, const char *host_to) { xmlNode *msg = create_request(join_op, NULL, host_to, CRM_SYSTEM_CRMD, CRM_SYSTEM_DC, NULL); /* Identify which election this is a part of */ crm_xml_add_int(msg, PCMK__XA_JOIN_ID, current_join_id); /* Add a field specifying whether the DC is shutting down. This keeps the * joining node from fencing the old DC if it becomes the new DC. */ pcmk__xe_set_bool_attr(msg, PCMK__XA_DC_LEAVING, pcmk_is_set(controld_globals.fsa_input_register, R_SHUTDOWN)); return msg; } static void join_make_offer(gpointer key, gpointer value, gpointer user_data) { xmlNode *offer = NULL; pcmk__node_status_t *member = (pcmk__node_status_t *) value; CRM_ASSERT(member != NULL); if (!pcmk__cluster_is_node_active(member)) { crm_info("Not making join-%d offer to inactive node %s", current_join_id, pcmk__s(member->name, "with unknown name")); if(member->expected == NULL && pcmk__str_eq(member->state, CRM_NODE_LOST, pcmk__str_casei)) { /* You would think this unsafe, but in fact this plus an * active resource is what causes it to be fenced. * * Yes, this does mean that any node that dies at the same * time as the old DC and is not running resource (still) * won't be fenced. * * I'm not happy about this either. */ pcmk__update_peer_expected(__func__, member, CRMD_JOINSTATE_DOWN); } return; } if (member->name == NULL) { crm_info("Not making join-%d offer to node uuid %s with unknown name", current_join_id, member->xml_id); return; } if (controld_globals.membership_id != crm_peer_seq) { controld_globals.membership_id = crm_peer_seq; crm_info("Making join-%d offers based on membership event %llu", current_join_id, crm_peer_seq); } if(user_data && member->join > crm_join_none) { crm_info("Not making join-%d offer to already known node %s (%s)", current_join_id, member->name, controld_join_phase_text(member->join)); return; } crm_update_peer_join(__func__, (pcmk__node_status_t*) member, crm_join_none); offer = create_dc_message(CRM_OP_JOIN_OFFER, member->name); // Advertise our feature set so the joining node can bail if not compatible crm_xml_add(offer, PCMK_XA_CRM_FEATURE_SET, CRM_FEATURE_SET); crm_info("Sending join-%d offer to %s", current_join_id, member->name); - pcmk__cluster_send_message(member, crm_msg_crmd, offer); + pcmk__cluster_send_message(member, pcmk__cluster_msg_controld, offer); pcmk__xml_free(offer); crm_update_peer_join(__func__, member, crm_join_welcomed); } /* A_DC_JOIN_OFFER_ALL */ void do_dc_join_offer_all(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data) { int count; /* Reset everyone's status back to down or in_ccm in the CIB. * Any nodes that are active in the CIB but not in the cluster membership * will be seen as offline by the scheduler anyway. */ current_join_id++; start_join_round(); update_dc(NULL); if (cause == C_HA_MESSAGE && current_input == I_NODE_JOIN) { crm_info("A new node joined the cluster"); } g_hash_table_foreach(crm_peer_cache, join_make_offer, NULL); count = crmd_join_phase_count(crm_join_welcomed); crm_info("Waiting on join-%d requests from %d outstanding node%s", current_join_id, count, pcmk__plural_s(count)); // Don't waste time by invoking the scheduler yet } /* A_DC_JOIN_OFFER_ONE */ void do_dc_join_offer_one(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data) { pcmk__node_status_t *member = NULL; ha_msg_input_t *welcome = NULL; int count; const char *join_to = NULL; if (msg_data->data == NULL) { crm_info("Making join-%d offers to any unconfirmed nodes " "because an unknown node joined", current_join_id); g_hash_table_foreach(crm_peer_cache, join_make_offer, &member); check_join_state(cur_state, __func__); return; } welcome = fsa_typed_data(fsa_dt_ha_msg); if (welcome == NULL) { // fsa_typed_data() already logged an error return; } join_to = crm_element_value(welcome->msg, PCMK__XA_SRC); if (join_to == NULL) { crm_err("Can't make join-%d offer to unknown node", current_join_id); return; } member = pcmk__get_node(0, join_to, NULL, pcmk__node_search_cluster_member); /* It is possible that a node will have been sick or starting up when the * original offer was made. However, it will either re-announce itself in * due course, or we can re-store the original offer on the client. */ crm_update_peer_join(__func__, member, crm_join_none); join_make_offer(NULL, member, NULL); /* If the offer isn't to the local node, make an offer to the local node as * well, to ensure the correct value for max_generation_from. */ if (strcasecmp(join_to, controld_globals.our_nodename) != 0) { member = pcmk__get_node(0, controld_globals.our_nodename, NULL, pcmk__node_search_cluster_member); join_make_offer(NULL, member, NULL); } /* This was a genuine join request; cancel any existing transition and * invoke the scheduler. */ abort_transition(PCMK_SCORE_INFINITY, pcmk__graph_restart, "Node join", NULL); count = crmd_join_phase_count(crm_join_welcomed); crm_info("Waiting on join-%d requests from %d outstanding node%s", current_join_id, count, pcmk__plural_s(count)); // Don't waste time by invoking the scheduler yet } static int compare_int_fields(xmlNode * left, xmlNode * right, const char *field) { const char *elem_l = crm_element_value(left, field); const char *elem_r = crm_element_value(right, field); long long int_elem_l; long long int_elem_r; pcmk__scan_ll(elem_l, &int_elem_l, -1LL); pcmk__scan_ll(elem_r, &int_elem_r, -1LL); if (int_elem_l < int_elem_r) { return -1; } else if (int_elem_l > int_elem_r) { return 1; } return 0; } /* A_DC_JOIN_PROCESS_REQ */ void do_dc_join_filter_offer(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data) { xmlNode *generation = NULL; int cmp = 0; int join_id = -1; int count = 0; gint value = 0; gboolean ack_nack_bool = TRUE; ha_msg_input_t *join_ack = fsa_typed_data(fsa_dt_ha_msg); const char *join_from = crm_element_value(join_ack->msg, PCMK__XA_SRC); const char *ref = crm_element_value(join_ack->msg, PCMK_XA_REFERENCE); const char *join_version = crm_element_value(join_ack->msg, PCMK_XA_CRM_FEATURE_SET); pcmk__node_status_t *join_node = NULL; if (join_from == NULL) { crm_err("Ignoring invalid join request without node name"); return; } join_node = pcmk__get_node(0, join_from, NULL, pcmk__node_search_cluster_member); crm_element_value_int(join_ack->msg, PCMK__XA_JOIN_ID, &join_id); if (join_id != current_join_id) { crm_debug("Ignoring join-%d request from %s because we are on join-%d", join_id, join_from, current_join_id); check_join_state(cur_state, __func__); return; } generation = join_ack->xml; if (max_generation_xml != NULL && generation != NULL) { int lpc = 0; const char *attributes[] = { PCMK_XA_ADMIN_EPOCH, PCMK_XA_EPOCH, PCMK_XA_NUM_UPDATES, }; /* It's not obvious that join_ack->xml is the PCMK__XE_GENERATION_TUPLE * element from the join client. The "if" guard is for clarity. */ if (pcmk__xe_is(generation, PCMK__XE_GENERATION_TUPLE)) { for (lpc = 0; cmp == 0 && lpc < PCMK__NELEM(attributes); lpc++) { cmp = compare_int_fields(max_generation_xml, generation, attributes[lpc]); } } else { // Should always be PCMK__XE_GENERATION_TUPLE CRM_LOG_ASSERT(false); } } if (ref == NULL) { ref = "none"; // for logging only } if (lookup_failed_sync_node(join_from, &value) == pcmk_rc_ok) { crm_err("Rejecting join-%d request from node %s because we failed to " "sync its CIB in join-%d " QB_XS " ref=%s", join_id, join_from, value, ref); ack_nack_bool = FALSE; } else if (!pcmk__cluster_is_node_active(join_node)) { if (match_down_event(join_from) != NULL) { /* The join request was received after the node was fenced or * otherwise shutdown in a way that we're aware of. No need to log * an error in this rare occurrence; we know the client was recently * shut down, and receiving a lingering in-flight request is not * cause for alarm. */ crm_debug("Rejecting join-%d request from inactive node %s " QB_XS " ref=%s", join_id, join_from, ref); } else { crm_err("Rejecting join-%d request from inactive node %s " QB_XS " ref=%s", join_id, join_from, ref); } ack_nack_bool = FALSE; } else if (generation == NULL) { crm_err("Rejecting invalid join-%d request from node %s " "missing CIB generation " QB_XS " ref=%s", join_id, join_from, ref); ack_nack_bool = FALSE; } else if ((join_version == NULL) || !feature_set_compatible(CRM_FEATURE_SET, join_version)) { crm_err("Rejecting join-%d request from node %s because feature set %s" " is incompatible with ours (%s) " QB_XS " ref=%s", join_id, join_from, (join_version? join_version : "pre-3.1.0"), CRM_FEATURE_SET, ref); ack_nack_bool = FALSE; } else if (max_generation_xml == NULL) { const char *validation = crm_element_value(generation, PCMK_XA_VALIDATE_WITH); if (pcmk__get_schema(validation) == NULL) { crm_err("Rejecting join-%d request from %s (with first CIB " "generation) due to unknown schema version %s " QB_XS " ref=%s", join_id, join_from, pcmk__s(validation, "(missing)"), ref); ack_nack_bool = FALSE; } else { crm_debug("Accepting join-%d request from %s (with first CIB " "generation) " QB_XS " ref=%s", join_id, join_from, ref); max_generation_xml = pcmk__xml_copy(NULL, generation); pcmk__str_update(&max_generation_from, join_from); } } else if ((cmp < 0) || ((cmp == 0) && pcmk__str_eq(join_from, controld_globals.our_nodename, pcmk__str_casei))) { const char *validation = crm_element_value(generation, PCMK_XA_VALIDATE_WITH); if (pcmk__get_schema(validation) == NULL) { crm_err("Rejecting join-%d request from %s (with better CIB " "generation than current best from %s) due to unknown " "schema version %s " QB_XS " ref=%s", join_id, join_from, max_generation_from, pcmk__s(validation, "(missing)"), ref); ack_nack_bool = FALSE; } else { crm_debug("Accepting join-%d request from %s (with better CIB " "generation than current best from %s) " QB_XS " ref=%s", join_id, join_from, max_generation_from, ref); crm_log_xml_debug(max_generation_xml, "Old max generation"); crm_log_xml_debug(generation, "New max generation"); pcmk__xml_free(max_generation_xml); max_generation_xml = pcmk__xml_copy(NULL, join_ack->xml); pcmk__str_update(&max_generation_from, join_from); } } else { crm_debug("Accepting join-%d request from %s " QB_XS " ref=%s", join_id, join_from, ref); } if (!ack_nack_bool) { if (compare_version(join_version, "3.17.0") < 0) { /* Clients with CRM_FEATURE_SET < 3.17.0 may respawn infinitely * after a nack message, don't send one */ crm_update_peer_join(__func__, join_node, crm_join_nack_quiet); } else { crm_update_peer_join(__func__, join_node, crm_join_nack); } pcmk__update_peer_expected(__func__, join_node, CRMD_JOINSTATE_NACK); } else { crm_update_peer_join(__func__, join_node, crm_join_integrated); pcmk__update_peer_expected(__func__, join_node, CRMD_JOINSTATE_MEMBER); } count = crmd_join_phase_count(crm_join_integrated); crm_debug("%d node%s currently integrated in join-%d", count, pcmk__plural_s(count), join_id); if (check_join_state(cur_state, __func__) == FALSE) { // Don't waste time by invoking the scheduler yet count = crmd_join_phase_count(crm_join_welcomed); crm_debug("Waiting on join-%d requests from %d outstanding node%s", join_id, count, pcmk__plural_s(count)); } } /* A_DC_JOIN_FINALIZE */ void do_dc_join_finalize(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data) { char *sync_from = NULL; int rc = pcmk_ok; int count_welcomed = crmd_join_phase_count(crm_join_welcomed); int count_finalizable = crmd_join_phase_count(crm_join_integrated) + crmd_join_phase_count(crm_join_nack) + crmd_join_phase_count(crm_join_nack_quiet); /* This we can do straight away and avoid clients timing us out * while we compute the latest CIB */ if (count_welcomed != 0) { crm_debug("Waiting on join-%d requests from %d outstanding node%s " "before finalizing join", current_join_id, count_welcomed, pcmk__plural_s(count_welcomed)); crmd_join_phase_log(LOG_DEBUG); /* crmd_fsa_stall(FALSE); Needed? */ return; } else if (count_finalizable == 0) { crm_debug("Finalization not needed for join-%d at the current time", current_join_id); crmd_join_phase_log(LOG_DEBUG); check_join_state(controld_globals.fsa_state, __func__); return; } controld_clear_fsa_input_flags(R_HAVE_CIB); if (pcmk__str_eq(max_generation_from, controld_globals.our_nodename, pcmk__str_null_matches|pcmk__str_casei)) { controld_set_fsa_input_flags(R_HAVE_CIB); } if (!controld_globals.transition_graph->complete) { crm_warn("Delaying join-%d finalization while transition in progress", current_join_id); crmd_join_phase_log(LOG_DEBUG); crmd_fsa_stall(FALSE); return; } if (pcmk_is_set(controld_globals.fsa_input_register, R_HAVE_CIB)) { // Send our CIB out to everyone sync_from = pcmk__str_copy(controld_globals.our_nodename); crm_debug("Finalizing join-%d for %d node%s (sync'ing from local CIB)", current_join_id, count_finalizable, pcmk__plural_s(count_finalizable)); crm_log_xml_debug(max_generation_xml, "Requested CIB version"); } else { // Ask for the agreed best CIB sync_from = pcmk__str_copy(max_generation_from); crm_notice("Finalizing join-%d for %d node%s (sync'ing CIB from %s)", current_join_id, count_finalizable, pcmk__plural_s(count_finalizable), sync_from); crm_log_xml_notice(max_generation_xml, "Requested CIB version"); } crmd_join_phase_log(LOG_DEBUG); rc = controld_globals.cib_conn->cmds->sync_from(controld_globals.cib_conn, sync_from, NULL, cib_none); fsa_register_cib_callback(rc, sync_from, finalize_sync_callback); } void free_max_generation(void) { free(max_generation_from); max_generation_from = NULL; pcmk__xml_free(max_generation_xml); max_generation_xml = NULL; } void finalize_sync_callback(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data) { CRM_LOG_ASSERT(-EPERM != rc); if (rc != pcmk_ok) { const char *sync_from = (const char *) user_data; do_crm_log(((rc == -pcmk_err_old_data)? LOG_WARNING : LOG_ERR), "Could not sync CIB from %s in join-%d: %s", sync_from, current_join_id, pcmk_strerror(rc)); if (rc != -pcmk_err_old_data) { record_failed_sync_node(sync_from, current_join_id); } /* restart the whole join process */ register_fsa_error_adv(C_FSA_INTERNAL, I_ELECTION_DC, NULL, NULL, __func__); } else if (!AM_I_DC) { crm_debug("Sync'ed CIB for join-%d but no longer DC", current_join_id); } else if (controld_globals.fsa_state != S_FINALIZE_JOIN) { crm_debug("Sync'ed CIB for join-%d but no longer in S_FINALIZE_JOIN " "(%s)", current_join_id, fsa_state2string(controld_globals.fsa_state)); } else { controld_set_fsa_input_flags(R_HAVE_CIB); /* make sure dc_uuid is re-set to us */ if (!check_join_state(controld_globals.fsa_state, __func__)) { int count_finalizable = 0; count_finalizable = crmd_join_phase_count(crm_join_integrated) + crmd_join_phase_count(crm_join_nack) + crmd_join_phase_count(crm_join_nack_quiet); crm_debug("Notifying %d node%s of join-%d results", count_finalizable, pcmk__plural_s(count_finalizable), current_join_id); g_hash_table_foreach(crm_peer_cache, finalize_join_for, NULL); } } } static void join_node_state_commit_callback(xmlNode *msg, int call_id, int rc, xmlNode *output, void *user_data) { const char *node = user_data; if (rc != pcmk_ok) { fsa_data_t *msg_data = NULL; // for register_fsa_error() macro crm_crit("join-%d node history update (via CIB call %d) for node %s " "failed: %s", current_join_id, call_id, node, pcmk_strerror(rc)); crm_log_xml_debug(msg, "failed"); register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); } crm_debug("join-%d node history update (via CIB call %d) for node %s " "complete", current_join_id, call_id, node); check_join_state(controld_globals.fsa_state, __func__); } /* A_DC_JOIN_PROCESS_ACK */ void do_dc_join_ack(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data) { int join_id = -1; ha_msg_input_t *join_ack = fsa_typed_data(fsa_dt_ha_msg); const char *op = crm_element_value(join_ack->msg, PCMK__XA_CRM_TASK); char *join_from = crm_element_value_copy(join_ack->msg, PCMK__XA_SRC); pcmk__node_status_t *peer = NULL; enum controld_section_e section = controld_section_lrm; char *xpath = NULL; xmlNode *state = join_ack->xml; xmlNode *execd_state = NULL; cib_t *cib = controld_globals.cib_conn; int rc = pcmk_ok; // Sanity checks if (join_from == NULL) { crm_warn("Ignoring message received without node identification"); goto done; } if (op == NULL) { crm_warn("Ignoring message received from %s without task", join_from); goto done; } if (strcmp(op, CRM_OP_JOIN_CONFIRM)) { crm_debug("Ignoring '%s' message from %s while waiting for '%s'", op, join_from, CRM_OP_JOIN_CONFIRM); goto done; } if (crm_element_value_int(join_ack->msg, PCMK__XA_JOIN_ID, &join_id) != 0) { crm_warn("Ignoring join confirmation from %s without valid join ID", join_from); goto done; } peer = pcmk__get_node(0, join_from, NULL, pcmk__node_search_cluster_member); if (peer->join != crm_join_finalized) { crm_info("Ignoring out-of-sequence join-%d confirmation from %s " "(currently %s not %s)", join_id, join_from, controld_join_phase_text(peer->join), controld_join_phase_text(crm_join_finalized)); goto done; } if (join_id != current_join_id) { crm_err("Rejecting join-%d confirmation from %s " "because currently on join-%d", join_id, join_from, current_join_id); crm_update_peer_join(__func__, peer, crm_join_nack); goto done; } crm_update_peer_join(__func__, peer, crm_join_confirmed); /* Update CIB with node's current executor state. A new transition will be * triggered later, when the CIB manager notifies us of the change. * * The delete and modify requests are part of an atomic transaction. */ rc = cib->cmds->init_transaction(cib); if (rc != pcmk_ok) { goto done; } // Delete relevant parts of node's current executor state from CIB if (pcmk_is_set(controld_globals.flags, controld_shutdown_lock_enabled)) { section = controld_section_lrm_unlocked; } controld_node_state_deletion_strings(join_from, section, &xpath, NULL); rc = cib->cmds->remove(cib, xpath, NULL, cib_xpath|cib_multiple|cib_transaction); if (rc != pcmk_ok) { goto done; } // Update CIB with node's latest known executor state if (pcmk__str_eq(join_from, controld_globals.our_nodename, pcmk__str_casei)) { // Use the latest possible state if processing our own join ack execd_state = controld_query_executor_state(); if (execd_state != NULL) { crm_debug("Updating local node history for join-%d from query " "result", current_join_id); state = execd_state; } else { crm_warn("Updating local node history from join-%d confirmation " "because query failed", current_join_id); } } else { crm_debug("Updating node history for %s from join-%d confirmation", join_from, current_join_id); } rc = cib->cmds->modify(cib, PCMK_XE_STATUS, state, cib_can_create|cib_transaction); pcmk__xml_free(execd_state); if (rc != pcmk_ok) { goto done; } // Commit the transaction rc = cib->cmds->end_transaction(cib, true, cib_none); fsa_register_cib_callback(rc, join_from, join_node_state_commit_callback); if (rc > 0) { // join_from will be freed after callback join_from = NULL; rc = pcmk_ok; } done: if (rc != pcmk_ok) { crm_crit("join-%d node history update for node %s failed: %s", current_join_id, join_from, pcmk_strerror(rc)); register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); } free(join_from); free(xpath); } void finalize_join_for(gpointer key, gpointer value, gpointer user_data) { xmlNode *acknak = NULL; xmlNode *tmp1 = NULL; pcmk__node_status_t *join_node = value; const char *join_to = join_node->name; bool integrated = false; switch (join_node->join) { case crm_join_integrated: integrated = true; break; case crm_join_nack: case crm_join_nack_quiet: break; default: crm_trace("Not updating non-integrated and non-nacked node %s (%s) " "for join-%d", join_to, controld_join_phase_text(join_node->join), current_join_id); return; } /* Update the element with the node's name and UUID, in case they * weren't known before */ crm_trace("Updating node name and UUID in CIB for %s", join_to); tmp1 = pcmk__xe_create(NULL, PCMK_XE_NODE); crm_xml_add(tmp1, PCMK_XA_ID, pcmk__cluster_node_uuid(join_node)); crm_xml_add(tmp1, PCMK_XA_UNAME, join_to); fsa_cib_anon_update(PCMK_XE_NODES, tmp1); pcmk__xml_free(tmp1); if (join_node->join == crm_join_nack_quiet) { crm_trace("Not sending nack message to node %s with feature set older " "than 3.17.0", join_to); return; } join_node = pcmk__get_node(0, join_to, NULL, pcmk__node_search_cluster_member); if (!pcmk__cluster_is_node_active(join_node)) { /* * NACK'ing nodes that the membership layer doesn't know about yet * simply creates more churn * * Better to leave them waiting and let the join restart when * the new membership event comes in * * All other NACKs (due to versions etc) should still be processed */ pcmk__update_peer_expected(__func__, join_node, CRMD_JOINSTATE_PENDING); return; } // Acknowledge or nack node's join request crm_debug("%sing join-%d request from %s", integrated? "Acknowledg" : "Nack", current_join_id, join_to); acknak = create_dc_message(CRM_OP_JOIN_ACKNAK, join_to); pcmk__xe_set_bool_attr(acknak, CRM_OP_JOIN_ACKNAK, integrated); if (integrated) { // No change needed for a nacked node crm_update_peer_join(__func__, join_node, crm_join_finalized); pcmk__update_peer_expected(__func__, join_node, CRMD_JOINSTATE_MEMBER); /* Iterate through the remote peer cache and add information on which * node hosts each to the ACK message. This keeps new controllers in * sync with what has already happened. */ if (pcmk__cluster_num_remote_nodes() > 0) { GHashTableIter iter; pcmk__node_status_t *node = NULL; xmlNode *remotes = pcmk__xe_create(acknak, PCMK_XE_NODES); g_hash_table_iter_init(&iter, crm_remote_peer_cache); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) { xmlNode *remote = NULL; if (!node->conn_host) { continue; } remote = pcmk__xe_create(remotes, PCMK_XE_NODE); pcmk__xe_set_props(remote, PCMK_XA_ID, node->name, PCMK__XA_NODE_STATE, node->state, PCMK__XA_CONNECTION_HOST, node->conn_host, NULL); } } } - pcmk__cluster_send_message(join_node, crm_msg_crmd, acknak); + pcmk__cluster_send_message(join_node, pcmk__cluster_msg_controld, acknak); pcmk__xml_free(acknak); return; } gboolean check_join_state(enum crmd_fsa_state cur_state, const char *source) { static unsigned long long highest_seq = 0; if (controld_globals.membership_id != crm_peer_seq) { crm_debug("join-%d: Membership changed from %llu to %llu " QB_XS " highest=%llu state=%s for=%s", current_join_id, controld_globals.membership_id, crm_peer_seq, highest_seq, fsa_state2string(cur_state), source); if(highest_seq < crm_peer_seq) { /* Don't spam the FSA with duplicates */ highest_seq = crm_peer_seq; register_fsa_input_before(C_FSA_INTERNAL, I_NODE_JOIN, NULL); } } else if (cur_state == S_INTEGRATION) { if (crmd_join_phase_count(crm_join_welcomed) == 0) { int count = crmd_join_phase_count(crm_join_integrated); crm_debug("join-%d: Integration of %d peer%s complete " QB_XS " state=%s for=%s", current_join_id, count, pcmk__plural_s(count), fsa_state2string(cur_state), source); register_fsa_input_before(C_FSA_INTERNAL, I_INTEGRATED, NULL); return TRUE; } } else if (cur_state == S_FINALIZE_JOIN) { if (!pcmk_is_set(controld_globals.fsa_input_register, R_HAVE_CIB)) { crm_debug("join-%d: Delaying finalization until we have CIB " QB_XS " state=%s for=%s", current_join_id, fsa_state2string(cur_state), source); return TRUE; } else if (crmd_join_phase_count(crm_join_welcomed) != 0) { int count = crmd_join_phase_count(crm_join_welcomed); crm_debug("join-%d: Still waiting on %d welcomed node%s " QB_XS " state=%s for=%s", current_join_id, count, pcmk__plural_s(count), fsa_state2string(cur_state), source); crmd_join_phase_log(LOG_DEBUG); } else if (crmd_join_phase_count(crm_join_integrated) != 0) { int count = crmd_join_phase_count(crm_join_integrated); crm_debug("join-%d: Still waiting on %d integrated node%s " QB_XS " state=%s for=%s", current_join_id, count, pcmk__plural_s(count), fsa_state2string(cur_state), source); crmd_join_phase_log(LOG_DEBUG); } else if (crmd_join_phase_count(crm_join_finalized) != 0) { int count = crmd_join_phase_count(crm_join_finalized); crm_debug("join-%d: Still waiting on %d finalized node%s " QB_XS " state=%s for=%s", current_join_id, count, pcmk__plural_s(count), fsa_state2string(cur_state), source); crmd_join_phase_log(LOG_DEBUG); } else { crm_debug("join-%d: Complete " QB_XS " state=%s for=%s", current_join_id, fsa_state2string(cur_state), source); register_fsa_input_later(C_FSA_INTERNAL, I_FINALIZED, NULL); return TRUE; } } return FALSE; } void do_dc_join_final(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data) { crm_debug("Ensuring DC, quorum and node attributes are up-to-date"); crm_update_quorum(pcmk__cluster_has_quorum(), TRUE); } int crmd_join_phase_count(enum crm_join_phase phase) { int count = 0; pcmk__node_status_t *peer; GHashTableIter iter; g_hash_table_iter_init(&iter, crm_peer_cache); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &peer)) { if(peer->join == phase) { count++; } } return count; } void crmd_join_phase_log(int level) { pcmk__node_status_t *peer; GHashTableIter iter; g_hash_table_iter_init(&iter, crm_peer_cache); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &peer)) { do_crm_log(level, "join-%d: %s=%s", current_join_id, peer->name, controld_join_phase_text(peer->join)); } } diff --git a/daemons/controld/controld_membership.c b/daemons/controld/controld_membership.c index 29a5fdcddd..359c5c6510 100644 --- a/daemons/controld/controld_membership.c +++ b/daemons/controld/controld_membership.c @@ -1,498 +1,498 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ /* put these first so that uuid_t is defined without conflicts */ #include #include #include #include #include #include #include void post_cache_update(int instance); extern gboolean check_join_state(enum crmd_fsa_state cur_state, const char *source); /*! * \internal * \brief Get log-friendly string equivalent of a controller group join phase * * \param[in] phase Join phase * * \return Log-friendly string equivalent of \p phase */ const char * controld_join_phase_text(enum crm_join_phase phase) { switch (phase) { case crm_join_nack_quiet: return "nack_quiet"; case crm_join_nack: return "nack"; case crm_join_none: return "none"; case crm_join_welcomed: return "welcomed"; case crm_join_integrated: return "integrated"; case crm_join_finalized: return "finalized"; case crm_join_confirmed: return "confirmed"; default: return "invalid"; } } static void reap_dead_nodes(gpointer key, gpointer value, gpointer user_data) { pcmk__node_status_t *node = value; if (pcmk__cluster_is_node_active(node)) { return; } crm_update_peer_join(__func__, node, crm_join_none); if ((node != NULL) && (node->name != NULL)) { if (pcmk__str_eq(controld_globals.our_nodename, node->name, pcmk__str_casei)) { crm_err("We're not part of the cluster anymore"); register_fsa_input(C_FSA_INTERNAL, I_ERROR, NULL); } else if (!AM_I_DC && pcmk__str_eq(node->name, controld_globals.dc_name, pcmk__str_casei)) { crm_warn("Our DC node (%s) left the cluster", node->name); register_fsa_input(C_FSA_INTERNAL, I_ELECTION, NULL); } } if ((controld_globals.fsa_state == S_INTEGRATION) || (controld_globals.fsa_state == S_FINALIZE_JOIN)) { check_join_state(controld_globals.fsa_state, __func__); } if ((node != NULL) && (node->xml_id != NULL)) { fail_incompletable_actions(controld_globals.transition_graph, node->xml_id); } } void post_cache_update(int instance) { xmlNode *no_op = NULL; crm_peer_seq = instance; crm_debug("Updated cache after membership event %d.", instance); g_hash_table_foreach(crm_peer_cache, reap_dead_nodes, NULL); controld_set_fsa_input_flags(R_MEMBERSHIP); if (AM_I_DC) { populate_cib_nodes(node_update_quick | node_update_cluster | node_update_peer | node_update_expected, __func__); } /* * If we lost nodes, we should re-check the election status * Safe to call outside of an election */ controld_set_fsa_action_flags(A_ELECTION_CHECK); controld_trigger_fsa(); /* Membership changed, remind everyone we're here. * This will aid detection of duplicate DCs */ no_op = create_request(CRM_OP_NOOP, NULL, NULL, CRM_SYSTEM_CRMD, AM_I_DC ? CRM_SYSTEM_DC : CRM_SYSTEM_CRMD, NULL); - pcmk__cluster_send_message(NULL, crm_msg_crmd, no_op); + pcmk__cluster_send_message(NULL, pcmk__cluster_msg_controld, no_op); pcmk__xml_free(no_op); } static void crmd_node_update_complete(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data) { fsa_data_t *msg_data = NULL; if (rc == pcmk_ok) { crm_trace("Node update %d complete", call_id); } else if(call_id < pcmk_ok) { crm_err("Node update failed: %s (%d)", pcmk_strerror(call_id), call_id); crm_log_xml_debug(msg, "failed"); register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); } else { crm_err("Node update %d failed: %s (%d)", call_id, pcmk_strerror(rc), rc); crm_log_xml_debug(msg, "failed"); register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); } } /*! * \internal * \brief Create an XML node state tag with updates * * \param[in,out] node Node whose state will be used for update * \param[in] flags Bitmask of node_update_flags indicating what to update * \param[in,out] parent XML node to contain update (or NULL) * \param[in] source Who requested the update (only used for logging) * * \return Pointer to created node state tag */ xmlNode * create_node_state_update(pcmk__node_status_t *node, int flags, xmlNode *parent, const char *source) { const char *value = NULL; xmlNode *node_state; if (!node->state) { crm_info("Node update for %s cancelled: no state, not seen yet", node->name); return NULL; } node_state = pcmk__xe_create(parent, PCMK__XE_NODE_STATE); if (pcmk_is_set(node->flags, pcmk__node_status_remote)) { pcmk__xe_set_bool_attr(node_state, PCMK_XA_REMOTE_NODE, true); } if (crm_xml_add(node_state, PCMK_XA_ID, pcmk__cluster_node_uuid(node)) == NULL) { crm_info("Node update for %s cancelled: no ID", node->name); pcmk__xml_free(node_state); return NULL; } crm_xml_add(node_state, PCMK_XA_UNAME, node->name); if ((flags & node_update_cluster) && node->state) { if (compare_version(controld_globals.dc_version, "3.18.0") >= 0) { // A value 0 means the node is not a cluster member. crm_xml_add_ll(node_state, PCMK__XA_IN_CCM, node->when_member); } else { pcmk__xe_set_bool_attr(node_state, PCMK__XA_IN_CCM, pcmk__str_eq(node->state, CRM_NODE_MEMBER, pcmk__str_casei)); } } if (!pcmk_is_set(node->flags, pcmk__node_status_remote)) { if (flags & node_update_peer) { if (compare_version(controld_globals.dc_version, "3.18.0") >= 0) { // A value 0 means the peer is offline in CPG. crm_xml_add_ll(node_state, PCMK_XA_CRMD, node->when_online); } else { // @COMPAT DCs < 2.1.7 use online/offline rather than timestamp value = PCMK_VALUE_OFFLINE; if (pcmk_is_set(node->processes, crm_get_cluster_proc())) { value = PCMK_VALUE_ONLINE; } crm_xml_add(node_state, PCMK_XA_CRMD, value); } } if (flags & node_update_join) { if (node->join <= crm_join_none) { value = CRMD_JOINSTATE_DOWN; } else { value = CRMD_JOINSTATE_MEMBER; } crm_xml_add(node_state, PCMK__XA_JOIN, value); } if (flags & node_update_expected) { crm_xml_add(node_state, PCMK_XA_EXPECTED, node->expected); } } crm_xml_add(node_state, PCMK_XA_CRM_DEBUG_ORIGIN, source); return node_state; } static void remove_conflicting_node_callback(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data) { char *node_uuid = user_data; do_crm_log_unlikely(rc == 0 ? LOG_DEBUG : LOG_NOTICE, "Deletion of the unknown conflicting node \"%s\": %s (rc=%d)", node_uuid, pcmk_strerror(rc), rc); } static void search_conflicting_node_callback(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data) { char *new_node_uuid = user_data; xmlNode *node_xml = NULL; if (rc != pcmk_ok) { if (rc != -ENXIO) { crm_notice("Searching conflicting nodes for %s failed: %s (%d)", new_node_uuid, pcmk_strerror(rc), rc); } return; } else if (output == NULL) { return; } if (pcmk__xe_is(output, PCMK_XE_NODE)) { node_xml = output; } else { node_xml = pcmk__xe_first_child(output, PCMK_XE_NODE, NULL, NULL); } for (; node_xml != NULL; node_xml = pcmk__xe_next_same(node_xml)) { const char *node_uuid = NULL; const char *node_uname = NULL; GHashTableIter iter; pcmk__node_status_t *node = NULL; gboolean known = FALSE; node_uuid = crm_element_value(node_xml, PCMK_XA_ID); node_uname = crm_element_value(node_xml, PCMK_XA_UNAME); if (node_uuid == NULL || node_uname == NULL) { continue; } g_hash_table_iter_init(&iter, crm_peer_cache); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) { if ((node != NULL) && pcmk__str_eq(node->xml_id, node_uuid, pcmk__str_casei) && pcmk__str_eq(node->name, node_uname, pcmk__str_casei)) { known = TRUE; break; } } if (known == FALSE) { cib_t *cib_conn = controld_globals.cib_conn; int delete_call_id = 0; xmlNode *node_state_xml = NULL; crm_notice("Deleting unknown node %s/%s which has conflicting uname with %s", node_uuid, node_uname, new_node_uuid); delete_call_id = cib_conn->cmds->remove(cib_conn, PCMK_XE_NODES, node_xml, cib_none); fsa_register_cib_callback(delete_call_id, pcmk__str_copy(node_uuid), remove_conflicting_node_callback); node_state_xml = pcmk__xe_create(NULL, PCMK__XE_NODE_STATE); crm_xml_add(node_state_xml, PCMK_XA_ID, node_uuid); crm_xml_add(node_state_xml, PCMK_XA_UNAME, node_uname); delete_call_id = cib_conn->cmds->remove(cib_conn, PCMK_XE_STATUS, node_state_xml, cib_none); fsa_register_cib_callback(delete_call_id, pcmk__str_copy(node_uuid), remove_conflicting_node_callback); pcmk__xml_free(node_state_xml); } } } static void node_list_update_callback(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data) { fsa_data_t *msg_data = NULL; if(call_id < pcmk_ok) { crm_err("Node list update failed: %s (%d)", pcmk_strerror(call_id), call_id); crm_log_xml_debug(msg, "update:failed"); register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); } else if(rc < pcmk_ok) { crm_err("Node update %d failed: %s (%d)", call_id, pcmk_strerror(rc), rc); crm_log_xml_debug(msg, "update:failed"); register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); } } void populate_cib_nodes(enum node_update_flags flags, const char *source) { cib_t *cib_conn = controld_globals.cib_conn; int call_id = 0; gboolean from_hashtable = TRUE; xmlNode *node_list = pcmk__xe_create(NULL, PCMK_XE_NODES); #if SUPPORT_COROSYNC if (!pcmk_is_set(flags, node_update_quick) && (pcmk_get_cluster_layer() == pcmk_cluster_layer_corosync)) { from_hashtable = pcmk__corosync_add_nodes(node_list); } #endif if (from_hashtable) { GHashTableIter iter; pcmk__node_status_t *node = NULL; GString *xpath = NULL; g_hash_table_iter_init(&iter, crm_peer_cache); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) { xmlNode *new_node = NULL; if ((node->xml_id != NULL) && (node->name != NULL)) { crm_trace("Creating node entry for %s/%s", node->name, node->xml_id); if (xpath == NULL) { xpath = g_string_sized_new(512); } else { g_string_truncate(xpath, 0); } /* We need both to be valid */ new_node = pcmk__xe_create(node_list, PCMK_XE_NODE); crm_xml_add(new_node, PCMK_XA_ID, node->xml_id); crm_xml_add(new_node, PCMK_XA_UNAME, node->name); /* Search and remove unknown nodes with the conflicting uname from CIB */ pcmk__g_strcat(xpath, "/" PCMK_XE_CIB "/" PCMK_XE_CONFIGURATION "/" PCMK_XE_NODES "/" PCMK_XE_NODE "[@" PCMK_XA_UNAME "='", node->name, "']" "[@" PCMK_XA_ID "!='", node->xml_id, "']", NULL); call_id = cib_conn->cmds->query(cib_conn, (const char *) xpath->str, NULL, cib_xpath); fsa_register_cib_callback(call_id, pcmk__str_copy(node->xml_id), search_conflicting_node_callback); } } if (xpath != NULL) { g_string_free(xpath, TRUE); } } crm_trace("Populating section from %s", from_hashtable ? "hashtable" : "cluster"); if ((controld_update_cib(PCMK_XE_NODES, node_list, cib_none, node_list_update_callback) == pcmk_rc_ok) && (crm_peer_cache != NULL) && AM_I_DC) { /* * There is no need to update the local CIB with our values if * we've not seen valid membership data */ GHashTableIter iter; pcmk__node_status_t *node = NULL; pcmk__xml_free(node_list); node_list = pcmk__xe_create(NULL, PCMK_XE_STATUS); g_hash_table_iter_init(&iter, crm_peer_cache); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) { create_node_state_update(node, flags, node_list, source); } if (crm_remote_peer_cache) { g_hash_table_iter_init(&iter, crm_remote_peer_cache); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) { create_node_state_update(node, flags, node_list, source); } } controld_update_cib(PCMK_XE_STATUS, node_list, cib_none, crmd_node_update_complete); } pcmk__xml_free(node_list); } static void cib_quorum_update_complete(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data) { fsa_data_t *msg_data = NULL; if (rc == pcmk_ok) { crm_trace("Quorum update %d complete", call_id); } else { crm_err("Quorum update %d failed: %s (%d)", call_id, pcmk_strerror(rc), rc); crm_log_xml_debug(msg, "failed"); register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); } } void crm_update_quorum(gboolean quorum, gboolean force_update) { bool has_quorum = pcmk_is_set(controld_globals.flags, controld_has_quorum); if (quorum) { controld_set_global_flags(controld_ever_had_quorum); } else if (pcmk_all_flags_set(controld_globals.flags, controld_ever_had_quorum |controld_no_quorum_suicide)) { pcmk__panic(__func__); } if (AM_I_DC && ((has_quorum && !quorum) || (!has_quorum && quorum) || force_update)) { xmlNode *update = NULL; update = pcmk__xe_create(NULL, PCMK_XE_CIB); crm_xml_add_int(update, PCMK_XA_HAVE_QUORUM, quorum); crm_xml_add(update, PCMK_XA_DC_UUID, controld_globals.our_uuid); crm_debug("Updating quorum status to %s", pcmk__btoa(quorum)); controld_update_cib(PCMK_XE_CIB, update, cib_none, cib_quorum_update_complete); pcmk__xml_free(update); /* Quorum changes usually cause a new transition via other activity: * quorum gained via a node joining will abort via the node join, * and quorum lost via a node leaving will usually abort via resource * activity and/or fencing. * * However, it is possible that nothing else causes a transition (e.g. * someone forces quorum via corosync-cmaptcl, or quorum is lost due to * a node in standby shutting down cleanly), so here ensure a new * transition is triggered. */ if (quorum) { /* If quorum was gained, abort after a short delay, in case multiple * nodes are joining around the same time, so the one that brings us * to quorum doesn't cause all the remaining ones to be fenced. */ abort_after_delay(PCMK_SCORE_INFINITY, pcmk__graph_restart, "Quorum gained", 5000); } else { abort_transition(PCMK_SCORE_INFINITY, pcmk__graph_restart, "Quorum lost", NULL); } } if (quorum) { controld_set_global_flags(controld_has_quorum); } else { controld_clear_global_flags(controld_has_quorum); } } diff --git a/daemons/controld/controld_messages.c b/daemons/controld/controld_messages.c index ae1892ef58..ab6da9befc 100644 --- a/daemons/controld/controld_messages.c +++ b/daemons/controld/controld_messages.c @@ -1,1360 +1,1361 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include #include static enum crmd_fsa_input handle_message(xmlNode *msg, enum crmd_fsa_cause cause); static void handle_response(xmlNode *stored_msg); static enum crmd_fsa_input handle_request(xmlNode *stored_msg, enum crmd_fsa_cause cause); static enum crmd_fsa_input handle_shutdown_request(xmlNode *stored_msg); static void send_msg_via_ipc(xmlNode * msg, const char *sys); /* debug only, can wrap all it likes */ static int last_data_id = 0; void register_fsa_error_adv(enum crmd_fsa_cause cause, enum crmd_fsa_input input, fsa_data_t * cur_data, void *new_data, const char *raised_from) { /* save the current actions if any */ if (controld_globals.fsa_actions != A_NOTHING) { register_fsa_input_adv(cur_data ? cur_data->fsa_cause : C_FSA_INTERNAL, I_NULL, cur_data ? cur_data->data : NULL, controld_globals.fsa_actions, TRUE, __func__); } /* reset the action list */ crm_info("Resetting the current action list"); fsa_dump_actions(controld_globals.fsa_actions, "Drop"); controld_globals.fsa_actions = A_NOTHING; /* register the error */ register_fsa_input_adv(cause, input, new_data, A_NOTHING, TRUE, raised_from); } void register_fsa_input_adv(enum crmd_fsa_cause cause, enum crmd_fsa_input input, void *data, uint64_t with_actions, gboolean prepend, const char *raised_from) { unsigned old_len = g_list_length(controld_globals.fsa_message_queue); fsa_data_t *fsa_data = NULL; if (raised_from == NULL) { raised_from = ""; } if (input == I_NULL && with_actions == A_NOTHING /* && data == NULL */ ) { /* no point doing anything */ crm_err("Cannot add entry to queue: no input and no action"); return; } if (input == I_WAIT_FOR_EVENT) { controld_set_global_flags(controld_fsa_is_stalled); crm_debug("Stalling the FSA pending further input: source=%s cause=%s data=%p queue=%d", raised_from, fsa_cause2string(cause), data, old_len); if (old_len > 0) { fsa_dump_queue(LOG_TRACE); prepend = FALSE; } if (data == NULL) { controld_set_fsa_action_flags(with_actions); fsa_dump_actions(with_actions, "Restored"); return; } /* Store everything in the new event and reset * controld_globals.fsa_actions */ with_actions |= controld_globals.fsa_actions; controld_globals.fsa_actions = A_NOTHING; } last_data_id++; crm_trace("%s %s FSA input %d (%s) due to %s, %s data", raised_from, (prepend? "prepended" : "appended"), last_data_id, fsa_input2string(input), fsa_cause2string(cause), (data? "with" : "without")); fsa_data = pcmk__assert_alloc(1, sizeof(fsa_data_t)); fsa_data->id = last_data_id; fsa_data->fsa_input = input; fsa_data->fsa_cause = cause; fsa_data->origin = raised_from; fsa_data->data = NULL; fsa_data->data_type = fsa_dt_none; fsa_data->actions = with_actions; if (with_actions != A_NOTHING) { crm_trace("Adding actions %.16llx to input", (unsigned long long) with_actions); } if (data != NULL) { switch (cause) { case C_FSA_INTERNAL: case C_CRMD_STATUS_CALLBACK: case C_IPC_MESSAGE: case C_HA_MESSAGE: CRM_CHECK(((ha_msg_input_t *) data)->msg != NULL, crm_err("Bogus data from %s", raised_from)); crm_trace("Copying %s data from %s as cluster message data", fsa_cause2string(cause), raised_from); fsa_data->data = copy_ha_msg_input(data); fsa_data->data_type = fsa_dt_ha_msg; break; case C_LRM_OP_CALLBACK: crm_trace("Copying %s data from %s as lrmd_event_data_t", fsa_cause2string(cause), raised_from); fsa_data->data = lrmd_copy_event((lrmd_event_data_t *) data); fsa_data->data_type = fsa_dt_lrm; break; case C_TIMER_POPPED: case C_SHUTDOWN: case C_UNKNOWN: case C_STARTUP: crm_crit("Copying %s data (from %s) is not yet implemented", fsa_cause2string(cause), raised_from); crmd_exit(CRM_EX_SOFTWARE); break; } } /* make sure to free it properly later */ if (prepend) { controld_globals.fsa_message_queue = g_list_prepend(controld_globals.fsa_message_queue, fsa_data); } else { controld_globals.fsa_message_queue = g_list_append(controld_globals.fsa_message_queue, fsa_data); } crm_trace("FSA message queue length is %d", g_list_length(controld_globals.fsa_message_queue)); /* fsa_dump_queue(LOG_TRACE); */ if (old_len == g_list_length(controld_globals.fsa_message_queue)) { crm_err("Couldn't add message to the queue"); } if (input != I_WAIT_FOR_EVENT) { controld_trigger_fsa(); } } void fsa_dump_queue(int log_level) { int offset = 0; for (GList *iter = controld_globals.fsa_message_queue; iter != NULL; iter = iter->next) { fsa_data_t *data = (fsa_data_t *) iter->data; do_crm_log_unlikely(log_level, "queue[%d.%d]: input %s raised by %s(%p.%d)\t(cause=%s)", offset++, data->id, fsa_input2string(data->fsa_input), data->origin, data->data, data->data_type, fsa_cause2string(data->fsa_cause)); } } ha_msg_input_t * copy_ha_msg_input(ha_msg_input_t * orig) { xmlNode *wrapper = NULL; ha_msg_input_t *copy = pcmk__assert_alloc(1, sizeof(ha_msg_input_t)); copy->msg = (orig != NULL)? pcmk__xml_copy(NULL, orig->msg) : NULL; wrapper = pcmk__xe_first_child(copy->msg, PCMK__XE_CRM_XML, NULL, NULL); copy->xml = pcmk__xe_first_child(wrapper, NULL, NULL, NULL); return copy; } void delete_fsa_input(fsa_data_t * fsa_data) { lrmd_event_data_t *op = NULL; xmlNode *foo = NULL; if (fsa_data == NULL) { return; } crm_trace("About to free %s data", fsa_cause2string(fsa_data->fsa_cause)); if (fsa_data->data != NULL) { switch (fsa_data->data_type) { case fsa_dt_ha_msg: delete_ha_msg_input(fsa_data->data); break; case fsa_dt_xml: foo = fsa_data->data; pcmk__xml_free(foo); break; case fsa_dt_lrm: op = (lrmd_event_data_t *) fsa_data->data; lrmd_free_event(op); break; case fsa_dt_none: if (fsa_data->data != NULL) { crm_err("Don't know how to free %s data from %s", fsa_cause2string(fsa_data->fsa_cause), fsa_data->origin); crmd_exit(CRM_EX_SOFTWARE); } break; } crm_trace("%s data freed", fsa_cause2string(fsa_data->fsa_cause)); } free(fsa_data); } /* returns the next message */ fsa_data_t * get_message(void) { fsa_data_t *message = (fsa_data_t *) controld_globals.fsa_message_queue->data; controld_globals.fsa_message_queue = g_list_remove(controld_globals.fsa_message_queue, message); crm_trace("Processing input %d", message->id); return message; } void * fsa_typed_data_adv(fsa_data_t * fsa_data, enum fsa_data_type a_type, const char *caller) { void *ret_val = NULL; if (fsa_data == NULL) { crm_err("%s: No FSA data available", caller); } else if (fsa_data->data == NULL) { crm_err("%s: No message data available. Origin: %s", caller, fsa_data->origin); } else if (fsa_data->data_type != a_type) { crm_crit("%s: Message data was the wrong type! %d vs. requested=%d. Origin: %s", caller, fsa_data->data_type, a_type, fsa_data->origin); CRM_ASSERT(fsa_data->data_type == a_type); } else { ret_val = fsa_data->data; } return ret_val; } /* A_MSG_ROUTE */ void do_msg_route(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data) { ha_msg_input_t *input = fsa_typed_data(fsa_dt_ha_msg); route_message(msg_data->fsa_cause, input->msg); } void route_message(enum crmd_fsa_cause cause, xmlNode * input) { ha_msg_input_t fsa_input; enum crmd_fsa_input result = I_NULL; fsa_input.msg = input; CRM_CHECK(cause == C_IPC_MESSAGE || cause == C_HA_MESSAGE, return); /* try passing the buck first */ if (relay_message(input, cause == C_IPC_MESSAGE)) { return; } /* handle locally */ result = handle_message(input, cause); /* done or process later? */ switch (result) { case I_NULL: case I_CIB_OP: case I_ROUTER: case I_NODE_JOIN: case I_JOIN_REQUEST: case I_JOIN_RESULT: break; default: /* Defering local processing of message */ register_fsa_input_later(cause, result, &fsa_input); return; } if (result != I_NULL) { /* add to the front of the queue */ register_fsa_input(cause, result, &fsa_input); } } gboolean relay_message(xmlNode * msg, gboolean originated_locally) { - enum crm_ais_msg_types dest = crm_msg_none; + enum pcmk__cluster_msg dest = pcmk__cluster_msg_unknown; bool is_for_dc = false; bool is_for_dcib = false; bool is_for_te = false; bool is_for_crm = false; bool is_for_cib = false; bool is_local = false; bool broadcast = false; const char *host_to = NULL; const char *sys_to = NULL; const char *sys_from = NULL; const char *type = NULL; const char *task = NULL; const char *ref = NULL; pcmk__node_status_t *node_to = NULL; CRM_CHECK(msg != NULL, return TRUE); host_to = crm_element_value(msg, PCMK__XA_CRM_HOST_TO); sys_to = crm_element_value(msg, PCMK__XA_CRM_SYS_TO); sys_from = crm_element_value(msg, PCMK__XA_CRM_SYS_FROM); type = crm_element_value(msg, PCMK__XA_T); task = crm_element_value(msg, PCMK__XA_CRM_TASK); ref = crm_element_value(msg, PCMK_XA_REFERENCE); broadcast = pcmk__str_empty(host_to); if (ref == NULL) { ref = "without reference ID"; } if (pcmk__str_eq(task, CRM_OP_HELLO, pcmk__str_casei)) { crm_trace("Received hello %s from %s (no processing needed)", ref, pcmk__s(sys_from, "unidentified source")); crm_log_xml_trace(msg, "hello"); return TRUE; } // Require message type (set by create_request()) if (!pcmk__str_eq(type, PCMK__VALUE_CRMD, pcmk__str_none)) { crm_warn("Ignoring invalid message %s with type '%s' " "(not '" PCMK__VALUE_CRMD "')", ref, pcmk__s(type, "")); crm_log_xml_trace(msg, "ignored"); return TRUE; } // Require a destination subsystem (also set by create_request()) if (sys_to == NULL) { crm_warn("Ignoring invalid message %s with no " PCMK__XA_CRM_SYS_TO, ref); crm_log_xml_trace(msg, "ignored"); return TRUE; } // Get the message type appropriate to the destination subsystem if (pcmk_get_cluster_layer() == pcmk_cluster_layer_corosync) { dest = pcmk__cluster_parse_msg_type(sys_to); - if (dest == crm_msg_none) { + if (dest == pcmk__cluster_msg_unknown) { /* Unrecognized value, use a sane default * * @TODO Maybe we should bail instead */ - dest = crm_msg_crmd; + dest = pcmk__cluster_msg_controld; } } is_for_dc = (strcasecmp(CRM_SYSTEM_DC, sys_to) == 0); is_for_dcib = (strcasecmp(CRM_SYSTEM_DCIB, sys_to) == 0); is_for_te = (strcasecmp(CRM_SYSTEM_TENGINE, sys_to) == 0); is_for_cib = (strcasecmp(CRM_SYSTEM_CIB, sys_to) == 0); is_for_crm = (strcasecmp(CRM_SYSTEM_CRMD, sys_to) == 0); // Check whether message should be processed locally is_local = false; if (broadcast) { if (is_for_dc || is_for_te) { is_local = false; } else if (is_for_crm) { if (pcmk__strcase_any_of(task, CRM_OP_NODE_INFO, PCMK__CONTROLD_CMD_NODES, NULL)) { /* Node info requests do not specify a host, which is normally * treated as "all hosts", because the whole point is that the * client may not know the local node name. Always handle these * requests locally. */ is_local = true; } else { is_local = !originated_locally; } } else { is_local = true; } } else if (pcmk__str_eq(controld_globals.our_nodename, host_to, pcmk__str_casei)) { is_local = true; } else if (is_for_crm && pcmk__str_eq(task, CRM_OP_LRM_DELETE, pcmk__str_casei)) { xmlNode *wrapper = pcmk__xe_first_child(msg, PCMK__XE_CRM_XML, NULL, NULL); xmlNode *msg_data = pcmk__xe_first_child(wrapper, NULL, NULL, NULL); const char *mode = crm_element_value(msg_data, PCMK__XA_MODE); if (pcmk__str_eq(mode, PCMK__VALUE_CIB, pcmk__str_none)) { // Local delete of an offline node's resource history is_local = true; } } // Check whether message should be relayed if (is_for_dc || is_for_dcib || is_for_te) { if (AM_I_DC) { if (is_for_te) { crm_trace("Route message %s locally as transition request", ref); crm_log_xml_trace(msg, sys_to); send_msg_via_ipc(msg, sys_to); return TRUE; // No further processing of message is needed } crm_trace("Route message %s locally as DC request", ref); return FALSE; // More to be done by caller } if (originated_locally && !pcmk__strcase_any_of(sys_from, CRM_SYSTEM_PENGINE, CRM_SYSTEM_TENGINE, NULL)) { crm_trace("Relay message %s to DC (via %s)", ref, pcmk__s(host_to, "broadcast")); crm_log_xml_trace(msg, "relayed"); if (!broadcast) { node_to = pcmk__get_node(0, host_to, NULL, pcmk__node_search_cluster_member); } pcmk__cluster_send_message(node_to, dest, msg); return TRUE; } /* Transition engine and scheduler messages are sent only to the DC on * the same node. If we are no longer the DC, discard this message. */ crm_trace("Ignoring message %s because we are no longer DC", ref); crm_log_xml_trace(msg, "ignored"); return TRUE; // No further processing of message is needed } if (is_local) { if (is_for_crm || is_for_cib) { crm_trace("Route message %s locally as controller request", ref); return FALSE; // More to be done by caller } crm_trace("Relay message %s locally to %s", ref, sys_to); crm_log_xml_trace(msg, "IPC-relay"); send_msg_via_ipc(msg, sys_to); return TRUE; } if (!broadcast) { node_to = pcmk__search_node_caches(0, host_to, pcmk__node_search_cluster_member); if (node_to == NULL) { crm_warn("Ignoring message %s because node %s is unknown", ref, host_to); crm_log_xml_trace(msg, "ignored"); return TRUE; } } crm_trace("Relay message %s to %s", ref, pcmk__s(host_to, "all peers")); crm_log_xml_trace(msg, "relayed"); pcmk__cluster_send_message(node_to, dest, msg); return TRUE; } // Return true if field contains a positive integer static bool authorize_version(xmlNode *message_data, const char *field, const char *client_name, const char *ref, const char *uuid) { const char *version = crm_element_value(message_data, field); long long version_num; if ((pcmk__scan_ll(version, &version_num, -1LL) != pcmk_rc_ok) || (version_num < 0LL)) { crm_warn("Rejected IPC hello from %s: '%s' is not a valid protocol %s " QB_XS " ref=%s uuid=%s", client_name, ((version == NULL)? "" : version), field, (ref? ref : "none"), uuid); return false; } return true; } /*! * \internal * \brief Check whether a client IPC message is acceptable * * If a given client IPC message is a hello, "authorize" it by ensuring it has * valid information such as a protocol version, and return false indicating * that nothing further needs to be done with the message. If the message is not * a hello, just return true to indicate it needs further processing. * * \param[in] client_msg XML of IPC message * \param[in,out] curr_client If IPC is not proxied, client that sent message * \param[in] proxy_session If IPC is proxied, the session ID * * \return true if message needs further processing, false if it doesn't */ bool controld_authorize_ipc_message(const xmlNode *client_msg, pcmk__client_t *curr_client, const char *proxy_session) { xmlNode *wrapper = NULL; xmlNode *message_data = NULL; const char *client_name = NULL; const char *op = crm_element_value(client_msg, PCMK__XA_CRM_TASK); const char *ref = crm_element_value(client_msg, PCMK_XA_REFERENCE); const char *uuid = (curr_client? curr_client->id : proxy_session); if (uuid == NULL) { crm_warn("IPC message from client rejected: No client identifier " QB_XS " ref=%s", (ref? ref : "none")); goto rejected; } if (!pcmk__str_eq(CRM_OP_HELLO, op, pcmk__str_casei)) { // Only hello messages need to be authorized return true; } wrapper = pcmk__xe_first_child(client_msg, PCMK__XE_CRM_XML, NULL, NULL); message_data = pcmk__xe_first_child(wrapper, NULL, NULL, NULL); client_name = crm_element_value(message_data, PCMK__XA_CLIENT_NAME); if (pcmk__str_empty(client_name)) { crm_warn("IPC hello from client rejected: No client name", QB_XS " ref=%s uuid=%s", (ref? ref : "none"), uuid); goto rejected; } if (!authorize_version(message_data, PCMK__XA_MAJOR_VERSION, client_name, ref, uuid)) { goto rejected; } if (!authorize_version(message_data, PCMK__XA_MINOR_VERSION, client_name, ref, uuid)) { goto rejected; } crm_trace("Validated IPC hello from client %s", client_name); crm_log_xml_trace(client_msg, "hello"); if (curr_client) { curr_client->userdata = pcmk__str_copy(client_name); } controld_trigger_fsa(); return false; rejected: crm_log_xml_trace(client_msg, "rejected"); if (curr_client) { qb_ipcs_disconnect(curr_client->ipcs); } return false; } static enum crmd_fsa_input handle_message(xmlNode *msg, enum crmd_fsa_cause cause) { const char *type = NULL; CRM_CHECK(msg != NULL, return I_NULL); type = crm_element_value(msg, PCMK__XA_SUBT); if (pcmk__str_eq(type, PCMK__VALUE_REQUEST, pcmk__str_none)) { return handle_request(msg, cause); } if (pcmk__str_eq(type, PCMK__VALUE_RESPONSE, pcmk__str_none)) { handle_response(msg); return I_NULL; } crm_warn("Ignoring message with unknown " PCMK__XA_SUBT" '%s'", pcmk__s(type, "")); crm_log_xml_trace(msg, "bad"); return I_NULL; } static enum crmd_fsa_input handle_failcount_op(xmlNode * stored_msg) { const char *rsc = NULL; const char *uname = NULL; const char *op = NULL; char *interval_spec = NULL; guint interval_ms = 0; gboolean is_remote_node = FALSE; xmlNode *wrapper = pcmk__xe_first_child(stored_msg, PCMK__XE_CRM_XML, NULL, NULL); xmlNode *xml_op = pcmk__xe_first_child(wrapper, NULL, NULL, NULL); if (xml_op) { xmlNode *xml_rsc = pcmk__xe_first_child(xml_op, PCMK_XE_PRIMITIVE, NULL, NULL); xmlNode *xml_attrs = pcmk__xe_first_child(xml_op, PCMK__XE_ATTRIBUTES, NULL, NULL); if (xml_rsc) { rsc = pcmk__xe_id(xml_rsc); } if (xml_attrs) { op = crm_element_value(xml_attrs, CRM_META "_" PCMK__META_CLEAR_FAILURE_OP); crm_element_value_ms(xml_attrs, CRM_META "_" PCMK__META_CLEAR_FAILURE_INTERVAL, &interval_ms); } } uname = crm_element_value(xml_op, PCMK__META_ON_NODE); if ((rsc == NULL) || (uname == NULL)) { crm_log_xml_warn(stored_msg, "invalid failcount op"); return I_NULL; } if (crm_element_value(xml_op, PCMK__XA_ROUTER_NODE)) { is_remote_node = TRUE; } crm_debug("Clearing failures for %s-interval %s on %s " "from attribute manager, CIB, and executor state", pcmk__readable_interval(interval_ms), rsc, uname); if (interval_ms) { interval_spec = crm_strdup_printf("%ums", interval_ms); } update_attrd_clear_failures(uname, rsc, op, interval_spec, is_remote_node); free(interval_spec); controld_cib_delete_last_failure(rsc, uname, op, interval_ms); lrm_clear_last_failure(rsc, uname, op, interval_ms); return I_NULL; } static enum crmd_fsa_input handle_lrm_delete(xmlNode *stored_msg) { const char *mode = NULL; xmlNode *wrapper = pcmk__xe_first_child(stored_msg, PCMK__XE_CRM_XML, NULL, NULL); xmlNode *msg_data = pcmk__xe_first_child(wrapper, NULL, NULL, NULL); CRM_CHECK(msg_data != NULL, return I_NULL); /* CRM_OP_LRM_DELETE has two distinct modes. The default behavior is to * relay the operation to the affected node, which will unregister the * resource from the local executor, clear the resource's history from the * CIB, and do some bookkeeping in the controller. * * However, if the affected node is offline, the client will specify * mode=PCMK__VALUE_CIB which means the controller receiving the operation * should clear the resource's history from the CIB and nothing else. This * is used to clear shutdown locks. */ mode = crm_element_value(msg_data, PCMK__XA_MODE); if (!pcmk__str_eq(mode, PCMK__VALUE_CIB, pcmk__str_none)) { // Relay to affected node crm_xml_add(stored_msg, PCMK__XA_CRM_SYS_TO, CRM_SYSTEM_LRMD); return I_ROUTER; } else { // Delete CIB history locally (compare with do_lrm_delete()) const char *from_sys = NULL; const char *user_name = NULL; const char *rsc_id = NULL; const char *node = NULL; xmlNode *rsc_xml = NULL; int rc = pcmk_rc_ok; rsc_xml = pcmk__xe_first_child(msg_data, PCMK_XE_PRIMITIVE, NULL, NULL); CRM_CHECK(rsc_xml != NULL, return I_NULL); rsc_id = pcmk__xe_id(rsc_xml); from_sys = crm_element_value(stored_msg, PCMK__XA_CRM_SYS_FROM); node = crm_element_value(msg_data, PCMK__META_ON_NODE); user_name = pcmk__update_acl_user(stored_msg, PCMK__XA_CRM_USER, NULL); crm_debug("Handling " CRM_OP_LRM_DELETE " for %s on %s locally%s%s " "(clearing CIB resource history only)", rsc_id, node, (user_name? " for user " : ""), (user_name? user_name : "")); rc = controld_delete_resource_history(rsc_id, node, user_name, cib_dryrun|cib_sync_call); if (rc == pcmk_rc_ok) { rc = controld_delete_resource_history(rsc_id, node, user_name, crmd_cib_smart_opt()); } /* Notify client. Also notify tengine if mode=PCMK__VALUE_CIB and * op=CRM_OP_LRM_DELETE. */ if (from_sys) { lrmd_event_data_t *op = NULL; const char *from_host = crm_element_value(stored_msg, PCMK__XA_SRC); const char *transition; if (strcmp(from_sys, CRM_SYSTEM_TENGINE)) { transition = crm_element_value(msg_data, PCMK__XA_TRANSITION_KEY); } else { transition = crm_element_value(stored_msg, PCMK__XA_TRANSITION_KEY); } crm_info("Notifying %s on %s that %s was%s deleted", from_sys, (from_host? from_host : "local node"), rsc_id, ((rc == pcmk_rc_ok)? "" : " not")); op = lrmd_new_event(rsc_id, PCMK_ACTION_DELETE, 0); op->type = lrmd_event_exec_complete; op->user_data = pcmk__str_copy(pcmk__s(transition, FAKE_TE_ID)); op->params = pcmk__strkey_table(free, free); pcmk__insert_dup(op->params, PCMK_XA_CRM_FEATURE_SET, CRM_FEATURE_SET); controld_rc2event(op, rc); controld_ack_event_directly(from_host, from_sys, NULL, op, rsc_id); lrmd_free_event(op); controld_trigger_delete_refresh(from_sys, rsc_id); } return I_NULL; } } /*! * \brief Handle a CRM_OP_REMOTE_STATE message by updating remote peer cache * * \param[in] msg Message XML * * \return Next FSA input */ static enum crmd_fsa_input handle_remote_state(const xmlNode *msg) { const char *conn_host = NULL; const char *remote_uname = pcmk__xe_id(msg); pcmk__node_status_t *remote_peer; bool remote_is_up = false; int rc = pcmk_rc_ok; rc = pcmk__xe_get_bool_attr(msg, PCMK__XA_IN_CCM, &remote_is_up); CRM_CHECK(remote_uname && rc == pcmk_rc_ok, return I_NULL); remote_peer = pcmk__cluster_lookup_remote_node(remote_uname); CRM_CHECK(remote_peer, return I_NULL); pcmk__update_peer_state(__func__, remote_peer, remote_is_up ? CRM_NODE_MEMBER : CRM_NODE_LOST, 0); conn_host = crm_element_value(msg, PCMK__XA_CONNECTION_HOST); if (conn_host) { pcmk__str_update(&remote_peer->conn_host, conn_host); } else if (remote_peer->conn_host) { free(remote_peer->conn_host); remote_peer->conn_host = NULL; } return I_NULL; } /*! * \brief Handle a CRM_OP_PING message * * \param[in] msg Message XML * * \return Next FSA input */ static enum crmd_fsa_input handle_ping(const xmlNode *msg) { const char *value = NULL; xmlNode *ping = NULL; xmlNode *reply = NULL; // Build reply ping = pcmk__xe_create(NULL, PCMK__XE_PING_RESPONSE); value = crm_element_value(msg, PCMK__XA_CRM_SYS_TO); crm_xml_add(ping, PCMK__XA_CRM_SUBSYSTEM, value); // Add controller state value = fsa_state2string(controld_globals.fsa_state); crm_xml_add(ping, PCMK__XA_CRMD_STATE, value); crm_notice("Current ping state: %s", value); // CTS needs this // Add controller health // @TODO maybe do some checks to determine meaningful status crm_xml_add(ping, PCMK_XA_RESULT, "ok"); // Send reply reply = create_reply(msg, ping); pcmk__xml_free(ping); if (reply != NULL) { (void) relay_message(reply, TRUE); pcmk__xml_free(reply); } // Nothing further to do return I_NULL; } /*! * \brief Handle a PCMK__CONTROLD_CMD_NODES message * * \param[in] request Message XML * * \return Next FSA input */ static enum crmd_fsa_input handle_node_list(const xmlNode *request) { GHashTableIter iter; pcmk__node_status_t *node = NULL; xmlNode *reply = NULL; xmlNode *reply_data = NULL; // Create message data for reply reply_data = pcmk__xe_create(NULL, PCMK_XE_NODES); g_hash_table_iter_init(&iter, crm_peer_cache); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) & node)) { xmlNode *xml = pcmk__xe_create(reply_data, PCMK_XE_NODE); crm_xml_add_ll(xml, PCMK_XA_ID, (long long) node->cluster_layer_id); // uint32_t crm_xml_add(xml, PCMK_XA_UNAME, node->name); crm_xml_add(xml, PCMK__XA_IN_CCM, node->state); } // Create and send reply reply = create_reply(request, reply_data); pcmk__xml_free(reply_data); if (reply) { (void) relay_message(reply, TRUE); pcmk__xml_free(reply); } // Nothing further to do return I_NULL; } /*! * \brief Handle a CRM_OP_NODE_INFO request * * \param[in] msg Message XML * * \return Next FSA input */ static enum crmd_fsa_input handle_node_info_request(const xmlNode *msg) { const char *value = NULL; pcmk__node_status_t *node = NULL; int node_id = 0; xmlNode *reply = NULL; xmlNode *reply_data = NULL; // Build reply reply_data = pcmk__xe_create(NULL, PCMK_XE_NODE); crm_xml_add(reply_data, PCMK__XA_CRM_SUBSYSTEM, CRM_SYSTEM_CRMD); // Add whether current partition has quorum pcmk__xe_set_bool_attr(reply_data, PCMK_XA_HAVE_QUORUM, pcmk_is_set(controld_globals.flags, controld_has_quorum)); // Check whether client requested node info by ID and/or name crm_element_value_int(msg, PCMK_XA_ID, &node_id); if (node_id < 0) { node_id = 0; } value = crm_element_value(msg, PCMK_XA_UNAME); // Default to local node if none given if ((node_id == 0) && (value == NULL)) { value = controld_globals.our_nodename; } node = pcmk__search_node_caches(node_id, value, pcmk__node_search_any); if (node) { crm_xml_add(reply_data, PCMK_XA_ID, node->xml_id); crm_xml_add(reply_data, PCMK_XA_UNAME, node->name); crm_xml_add(reply_data, PCMK_XA_CRMD, node->state); pcmk__xe_set_bool_attr(reply_data, PCMK_XA_REMOTE_NODE, pcmk_is_set(node->flags, pcmk__node_status_remote)); } // Send reply reply = create_reply(msg, reply_data); pcmk__xml_free(reply_data); if (reply != NULL) { (void) relay_message(reply, TRUE); pcmk__xml_free(reply); } // Nothing further to do return I_NULL; } static void verify_feature_set(xmlNode *msg) { const char *dc_version = crm_element_value(msg, PCMK_XA_CRM_FEATURE_SET); if (dc_version == NULL) { /* All we really know is that the DC feature set is older than 3.1.0, * but that's also all that really matters. */ dc_version = "3.0.14"; } if (feature_set_compatible(dc_version, CRM_FEATURE_SET)) { crm_trace("Local feature set (%s) is compatible with DC's (%s)", CRM_FEATURE_SET, dc_version); } else { crm_err("Local feature set (%s) is incompatible with DC's (%s)", CRM_FEATURE_SET, dc_version); // Nothing is likely to improve without administrator involvement controld_set_fsa_input_flags(R_STAYDOWN); crmd_exit(CRM_EX_FATAL); } } // DC gets own shutdown all-clear static enum crmd_fsa_input handle_shutdown_self_ack(xmlNode *stored_msg) { const char *host_from = crm_element_value(stored_msg, PCMK__XA_SRC); if (pcmk_is_set(controld_globals.fsa_input_register, R_SHUTDOWN)) { // The expected case -- we initiated own shutdown sequence crm_info("Shutting down controller"); return I_STOP; } if (pcmk__str_eq(host_from, controld_globals.dc_name, pcmk__str_casei)) { // Must be logic error -- DC confirming its own unrequested shutdown crm_err("Shutting down controller immediately due to " "unexpected shutdown confirmation"); return I_TERMINATE; } if (controld_globals.fsa_state != S_STOPPING) { // Shouldn't happen -- non-DC confirming unrequested shutdown crm_err("Starting new DC election because %s is " "confirming shutdown we did not request", (host_from? host_from : "another node")); return I_ELECTION; } // Shouldn't happen, but we are already stopping anyway crm_debug("Ignoring unexpected shutdown confirmation from %s", (host_from? host_from : "another node")); return I_NULL; } // Non-DC gets shutdown all-clear from DC static enum crmd_fsa_input handle_shutdown_ack(xmlNode *stored_msg) { const char *host_from = crm_element_value(stored_msg, PCMK__XA_SRC); if (host_from == NULL) { crm_warn("Ignoring shutdown request without origin specified"); return I_NULL; } if (pcmk__str_eq(host_from, controld_globals.dc_name, pcmk__str_null_matches|pcmk__str_casei)) { if (pcmk_is_set(controld_globals.fsa_input_register, R_SHUTDOWN)) { crm_info("Shutting down controller after confirmation from %s", host_from); } else { crm_err("Shutting down controller after unexpected " "shutdown request from %s", host_from); controld_set_fsa_input_flags(R_STAYDOWN); } return I_STOP; } crm_warn("Ignoring shutdown request from %s because DC is %s", host_from, controld_globals.dc_name); return I_NULL; } static enum crmd_fsa_input handle_request(xmlNode *stored_msg, enum crmd_fsa_cause cause) { xmlNode *msg = NULL; const char *op = crm_element_value(stored_msg, PCMK__XA_CRM_TASK); /* Optimize this for the DC - it has the most to do */ crm_log_xml_trace(stored_msg, "request"); if (op == NULL) { crm_warn("Ignoring request without " PCMK__XA_CRM_TASK); return I_NULL; } if (strcmp(op, CRM_OP_SHUTDOWN_REQ) == 0) { const char *from = crm_element_value(stored_msg, PCMK__XA_SRC); pcmk__node_status_t *node = pcmk__search_node_caches(0, from, pcmk__node_search_cluster_member); pcmk__update_peer_expected(__func__, node, CRMD_JOINSTATE_DOWN); if(AM_I_DC == FALSE) { return I_NULL; /* Done */ } } /*========== DC-Only Actions ==========*/ if (AM_I_DC) { if (strcmp(op, CRM_OP_JOIN_ANNOUNCE) == 0) { return I_NODE_JOIN; } else if (strcmp(op, CRM_OP_JOIN_REQUEST) == 0) { return I_JOIN_REQUEST; } else if (strcmp(op, CRM_OP_JOIN_CONFIRM) == 0) { return I_JOIN_RESULT; } else if (strcmp(op, CRM_OP_SHUTDOWN) == 0) { return handle_shutdown_self_ack(stored_msg); } else if (strcmp(op, CRM_OP_SHUTDOWN_REQ) == 0) { // Another controller wants to shut down its node return handle_shutdown_request(stored_msg); } } /*========== common actions ==========*/ if (strcmp(op, CRM_OP_NOVOTE) == 0) { ha_msg_input_t fsa_input; fsa_input.msg = stored_msg; register_fsa_input_adv(C_HA_MESSAGE, I_NULL, &fsa_input, A_ELECTION_COUNT | A_ELECTION_CHECK, FALSE, __func__); } else if (strcmp(op, CRM_OP_REMOTE_STATE) == 0) { /* a remote connection host is letting us know the node state */ return handle_remote_state(stored_msg); } else if (strcmp(op, CRM_OP_THROTTLE) == 0) { throttle_update(stored_msg); if (AM_I_DC && (controld_globals.transition_graph != NULL) && !controld_globals.transition_graph->complete) { crm_debug("The throttle changed. Trigger a graph."); trigger_graph(); } return I_NULL; } else if (strcmp(op, CRM_OP_CLEAR_FAILCOUNT) == 0) { return handle_failcount_op(stored_msg); } else if (strcmp(op, CRM_OP_VOTE) == 0) { /* count the vote and decide what to do after that */ ha_msg_input_t fsa_input; fsa_input.msg = stored_msg; register_fsa_input_adv(C_HA_MESSAGE, I_NULL, &fsa_input, A_ELECTION_COUNT | A_ELECTION_CHECK, FALSE, __func__); /* Sometimes we _must_ go into S_ELECTION */ if (controld_globals.fsa_state == S_HALT) { crm_debug("Forcing an election from S_HALT"); return I_ELECTION; } } else if (strcmp(op, CRM_OP_JOIN_OFFER) == 0) { verify_feature_set(stored_msg); crm_debug("Raising I_JOIN_OFFER: join-%s", crm_element_value(stored_msg, PCMK__XA_JOIN_ID)); return I_JOIN_OFFER; } else if (strcmp(op, CRM_OP_JOIN_ACKNAK) == 0) { crm_debug("Raising I_JOIN_RESULT: join-%s", crm_element_value(stored_msg, PCMK__XA_JOIN_ID)); return I_JOIN_RESULT; } else if (strcmp(op, CRM_OP_LRM_DELETE) == 0) { return handle_lrm_delete(stored_msg); } else if ((strcmp(op, CRM_OP_LRM_FAIL) == 0) || (strcmp(op, CRM_OP_LRM_REFRESH) == 0) // @COMPAT || (strcmp(op, CRM_OP_REPROBE) == 0)) { crm_xml_add(stored_msg, PCMK__XA_CRM_SYS_TO, CRM_SYSTEM_LRMD); return I_ROUTER; } else if (strcmp(op, CRM_OP_NOOP) == 0) { return I_NULL; } else if (strcmp(op, CRM_OP_PING) == 0) { return handle_ping(stored_msg); } else if (strcmp(op, CRM_OP_NODE_INFO) == 0) { return handle_node_info_request(stored_msg); } else if (strcmp(op, CRM_OP_RM_NODE_CACHE) == 0) { int id = 0; const char *name = NULL; crm_element_value_int(stored_msg, PCMK_XA_ID, &id); name = crm_element_value(stored_msg, PCMK_XA_UNAME); if(cause == C_IPC_MESSAGE) { msg = create_request(CRM_OP_RM_NODE_CACHE, NULL, NULL, CRM_SYSTEM_CRMD, CRM_SYSTEM_CRMD, NULL); - if (!pcmk__cluster_send_message(NULL, crm_msg_crmd, msg)) { + if (!pcmk__cluster_send_message(NULL, pcmk__cluster_msg_controld, + msg)) { crm_err("Could not instruct peers to remove references to node %s/%u", name, id); } else { crm_notice("Instructing peers to remove references to node %s/%u", name, id); } pcmk__xml_free(msg); } else { pcmk__cluster_forget_cluster_node(id, name); /* If we're forgetting this node, also forget any failures to fence * it, so we don't carry that over to any node added later with the * same name. */ st_fail_count_reset(name); } } else if (strcmp(op, CRM_OP_MAINTENANCE_NODES) == 0) { xmlNode *wrapper = pcmk__xe_first_child(stored_msg, PCMK__XE_CRM_XML, NULL, NULL); xmlNode *xml = pcmk__xe_first_child(wrapper, NULL, NULL, NULL); remote_ra_process_maintenance_nodes(xml); } else if (strcmp(op, PCMK__CONTROLD_CMD_NODES) == 0) { return handle_node_list(stored_msg); /*========== (NOT_DC)-Only Actions ==========*/ } else if (!AM_I_DC) { if (strcmp(op, CRM_OP_SHUTDOWN) == 0) { return handle_shutdown_ack(stored_msg); } } else { crm_err("Unexpected request (%s) sent to %s", op, AM_I_DC ? "the DC" : "non-DC node"); crm_log_xml_err(stored_msg, "Unexpected"); } return I_NULL; } static void handle_response(xmlNode *stored_msg) { const char *op = crm_element_value(stored_msg, PCMK__XA_CRM_TASK); crm_log_xml_trace(stored_msg, "reply"); if (op == NULL) { crm_warn("Ignoring reply without " PCMK__XA_CRM_TASK); } else if (AM_I_DC && strcmp(op, CRM_OP_PECALC) == 0) { // Check whether scheduler answer been superseded by subsequent request const char *msg_ref = crm_element_value(stored_msg, PCMK_XA_REFERENCE); if (msg_ref == NULL) { crm_err("%s - Ignoring calculation with no reference", op); } else if (pcmk__str_eq(msg_ref, controld_globals.fsa_pe_ref, pcmk__str_none)) { ha_msg_input_t fsa_input; controld_stop_sched_timer(); fsa_input.msg = stored_msg; register_fsa_input_later(C_IPC_MESSAGE, I_PE_SUCCESS, &fsa_input); } else { crm_info("%s calculation %s is obsolete", op, msg_ref); } } else if (strcmp(op, CRM_OP_VOTE) == 0 || strcmp(op, CRM_OP_SHUTDOWN_REQ) == 0 || strcmp(op, CRM_OP_SHUTDOWN) == 0) { } else { const char *host_from = crm_element_value(stored_msg, PCMK__XA_SRC); crm_err("Unexpected response (op=%s, src=%s) sent to the %s", op, host_from, AM_I_DC ? "DC" : "controller"); } } static enum crmd_fsa_input handle_shutdown_request(xmlNode * stored_msg) { /* handle here to avoid potential version issues * where the shutdown message/procedure may have * been changed in later versions. * * This way the DC is always in control of the shutdown */ char *now_s = NULL; const char *host_from = crm_element_value(stored_msg, PCMK__XA_SRC); if (host_from == NULL) { /* we're shutting down and the DC */ host_from = controld_globals.our_nodename; } crm_info("Creating shutdown request for %s (state=%s)", host_from, fsa_state2string(controld_globals.fsa_state)); crm_log_xml_trace(stored_msg, "message"); now_s = pcmk__ttoa(time(NULL)); update_attrd(host_from, PCMK__NODE_ATTR_SHUTDOWN, now_s, NULL, FALSE); free(now_s); /* will be picked up by the TE as long as its running */ return I_NULL; } static void send_msg_via_ipc(xmlNode * msg, const char *sys) { pcmk__client_t *client_channel = NULL; CRM_CHECK(sys != NULL, return); client_channel = pcmk__find_client_by_id(sys); if (crm_element_value(msg, PCMK__XA_SRC) == NULL) { crm_xml_add(msg, PCMK__XA_SRC, controld_globals.our_nodename); } if (client_channel != NULL) { /* Transient clients such as crmadmin */ pcmk__ipc_send_xml(client_channel, 0, msg, crm_ipc_server_event); } else if (pcmk__str_eq(sys, CRM_SYSTEM_TENGINE, pcmk__str_none)) { xmlNode *wrapper = pcmk__xe_first_child(msg, PCMK__XE_CRM_XML, NULL, NULL); xmlNode *data = pcmk__xe_first_child(wrapper, NULL, NULL, NULL); process_te_message(msg, data); } else if (pcmk__str_eq(sys, CRM_SYSTEM_LRMD, pcmk__str_none)) { fsa_data_t fsa_data; ha_msg_input_t fsa_input; xmlNode *wrapper = NULL; fsa_input.msg = msg; wrapper = pcmk__xe_first_child(msg, PCMK__XE_CRM_XML, NULL, NULL); fsa_input.xml = pcmk__xe_first_child(wrapper, NULL, NULL, NULL); fsa_data.id = 0; fsa_data.actions = 0; fsa_data.data = &fsa_input; fsa_data.fsa_input = I_MESSAGE; fsa_data.fsa_cause = C_IPC_MESSAGE; fsa_data.origin = __func__; fsa_data.data_type = fsa_dt_ha_msg; do_lrm_invoke(A_LRM_INVOKE, C_IPC_MESSAGE, controld_globals.fsa_state, I_MESSAGE, &fsa_data); } else if (crmd_is_proxy_session(sys)) { crmd_proxy_send(sys, msg); } else { crm_info("Received invalid request: unknown subsystem '%s'", sys); } } void delete_ha_msg_input(ha_msg_input_t * orig) { if (orig == NULL) { return; } pcmk__xml_free(orig->msg); free(orig); } /*! * \internal * \brief Notify the cluster of a remote node state change * * \param[in] node_name Node's name * \param[in] node_up true if node is up, false if down */ void broadcast_remote_state_message(const char *node_name, bool node_up) { xmlNode *msg = create_request(CRM_OP_REMOTE_STATE, NULL, NULL, CRM_SYSTEM_CRMD, CRM_SYSTEM_CRMD, NULL); crm_info("Notifying cluster of Pacemaker Remote node %s %s", node_name, node_up? "coming up" : "going down"); crm_xml_add(msg, PCMK_XA_ID, node_name); pcmk__xe_set_bool_attr(msg, PCMK__XA_IN_CCM, node_up); if (node_up) { crm_xml_add(msg, PCMK__XA_CONNECTION_HOST, controld_globals.our_nodename); } - pcmk__cluster_send_message(NULL, crm_msg_crmd, msg); + pcmk__cluster_send_message(NULL, pcmk__cluster_msg_controld, msg); pcmk__xml_free(msg); } diff --git a/daemons/controld/controld_te_actions.c b/daemons/controld/controld_te_actions.c index 510866001d..3d4a8ed68d 100644 --- a/daemons/controld/controld_te_actions.c +++ b/daemons/controld/controld_te_actions.c @@ -1,765 +1,765 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include // lrmd_event_data_t, lrmd_free_event() #include #include #include #include static GHashTable *te_targets = NULL; void send_rsc_command(pcmk__graph_action_t *action); static void te_update_job_count(pcmk__graph_action_t *action, int offset); static void te_start_action_timer(const pcmk__graph_t *graph, pcmk__graph_action_t *action) { action->timer = g_timeout_add(action->timeout + graph->network_delay, action_timer_callback, (void *) action); CRM_ASSERT(action->timer != 0); } /*! * \internal * \brief Execute a graph pseudo-action * * \param[in,out] graph Transition graph being executed * \param[in,out] pseudo Pseudo-action to execute * * \return Standard Pacemaker return code */ static int execute_pseudo_action(pcmk__graph_t *graph, pcmk__graph_action_t *pseudo) { const char *task = crm_element_value(pseudo->xml, PCMK_XA_OPERATION); /* send to peers as well? */ if (pcmk__str_eq(task, PCMK_ACTION_MAINTENANCE_NODES, pcmk__str_casei)) { GHashTableIter iter; pcmk__node_status_t *node = NULL; g_hash_table_iter_init(&iter, crm_peer_cache); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) { xmlNode *cmd = NULL; if (pcmk__str_eq(controld_globals.our_nodename, node->name, pcmk__str_casei)) { continue; } cmd = create_request(task, pseudo->xml, node->name, CRM_SYSTEM_CRMD, CRM_SYSTEM_TENGINE, NULL); - pcmk__cluster_send_message(node, crm_msg_crmd, cmd); + pcmk__cluster_send_message(node, pcmk__cluster_msg_controld, cmd); pcmk__xml_free(cmd); } remote_ra_process_maintenance_nodes(pseudo->xml); } else { /* Check action for Pacemaker Remote node side effects */ remote_ra_process_pseudo(pseudo->xml); } crm_debug("Pseudo-action %d (%s) fired and confirmed", pseudo->id, crm_element_value(pseudo->xml, PCMK__XA_OPERATION_KEY)); te_action_confirmed(pseudo, graph); return pcmk_rc_ok; } static int get_target_rc(pcmk__graph_action_t *action) { int exit_status; pcmk__scan_min_int(crm_meta_value(action->params, PCMK__META_OP_TARGET_RC), &exit_status, 0); return exit_status; } /*! * \internal * \brief Execute a cluster action from a transition graph * * \param[in,out] graph Transition graph being executed * \param[in,out] action Cluster action to execute * * \return Standard Pacemaker return code */ static int execute_cluster_action(pcmk__graph_t *graph, pcmk__graph_action_t *action) { char *counter = NULL; xmlNode *cmd = NULL; gboolean is_local = FALSE; const char *id = NULL; const char *task = NULL; const char *value = NULL; const char *on_node = NULL; const char *router_node = NULL; gboolean rc = TRUE; gboolean no_wait = FALSE; const pcmk__node_status_t *node = NULL; id = pcmk__xe_id(action->xml); CRM_CHECK(!pcmk__str_empty(id), return EPROTO); task = crm_element_value(action->xml, PCMK_XA_OPERATION); CRM_CHECK(!pcmk__str_empty(task), return EPROTO); on_node = crm_element_value(action->xml, PCMK__META_ON_NODE); CRM_CHECK(!pcmk__str_empty(on_node), return pcmk_rc_node_unknown); router_node = crm_element_value(action->xml, PCMK__XA_ROUTER_NODE); if (router_node == NULL) { router_node = on_node; if (pcmk__str_eq(task, PCMK_ACTION_LRM_DELETE, pcmk__str_none)) { const char *mode = crm_element_value(action->xml, PCMK__XA_MODE); if (pcmk__str_eq(mode, PCMK__VALUE_CIB, pcmk__str_none)) { router_node = controld_globals.our_nodename; } } } if (pcmk__str_eq(router_node, controld_globals.our_nodename, pcmk__str_casei)) { is_local = TRUE; } value = crm_meta_value(action->params, PCMK__META_OP_NO_WAIT); if (crm_is_true(value)) { no_wait = TRUE; } crm_info("Handling controller request '%s' (%s on %s)%s%s", id, task, on_node, (is_local? " locally" : ""), (no_wait? " without waiting" : "")); if (is_local && pcmk__str_eq(task, PCMK_ACTION_DO_SHUTDOWN, pcmk__str_none)) { /* defer until everything else completes */ crm_info("Controller request '%s' is a local shutdown", id); graph->completion_action = pcmk__graph_shutdown; graph->abort_reason = "local shutdown"; te_action_confirmed(action, graph); return pcmk_rc_ok; } else if (pcmk__str_eq(task, PCMK_ACTION_DO_SHUTDOWN, pcmk__str_none)) { pcmk__node_status_t *peer = pcmk__get_node(0, router_node, NULL, pcmk__node_search_cluster_member); pcmk__update_peer_expected(__func__, peer, CRMD_JOINSTATE_DOWN); } cmd = create_request(task, action->xml, router_node, CRM_SYSTEM_CRMD, CRM_SYSTEM_TENGINE, NULL); counter = pcmk__transition_key(controld_globals.transition_graph->id, action->id, get_target_rc(action), controld_globals.te_uuid); crm_xml_add(cmd, PCMK__XA_TRANSITION_KEY, counter); node = pcmk__get_node(0, router_node, NULL, pcmk__node_search_cluster_member); - rc = pcmk__cluster_send_message(node, crm_msg_crmd, cmd); + rc = pcmk__cluster_send_message(node, pcmk__cluster_msg_controld, cmd); free(counter); pcmk__xml_free(cmd); if (rc == FALSE) { crm_err("Action %d failed: send", action->id); return ECOMM; } else if (no_wait) { te_action_confirmed(action, graph); } else { if (action->timeout <= 0) { crm_err("Action %d: %s on %s had an invalid timeout (%dms). Using %ums instead", action->id, task, on_node, action->timeout, graph->network_delay); action->timeout = (int) graph->network_delay; } te_start_action_timer(graph, action); } return pcmk_rc_ok; } /*! * \internal * \brief Synthesize an executor event for a resource action timeout * * \param[in] action Resource action that timed out * \param[in] target_rc Expected result of action that timed out * * Synthesize an executor event for a resource action timeout. (If the executor * gets a timeout while waiting for a resource action to complete, that will be * reported via the usual callback. This timeout means we didn't hear from the * executor itself or the controller that relayed the action to the executor.) * * \return Newly created executor event for result of \p action * \note The caller is responsible for freeing the return value using * lrmd_free_event(). */ static lrmd_event_data_t * synthesize_timeout_event(const pcmk__graph_action_t *action, int target_rc) { lrmd_event_data_t *op = NULL; const char *target = crm_element_value(action->xml, PCMK__META_ON_NODE); const char *reason = NULL; char *dynamic_reason = NULL; if (pcmk__str_eq(target, pcmk__cluster_local_node_name(), pcmk__str_casei)) { reason = "Local executor did not return result in time"; } else { const char *router_node = NULL; router_node = crm_element_value(action->xml, PCMK__XA_ROUTER_NODE); if (router_node == NULL) { router_node = target; } dynamic_reason = crm_strdup_printf("Controller on %s did not return " "result in time", router_node); reason = dynamic_reason; } op = pcmk__event_from_graph_action(NULL, action, PCMK_EXEC_TIMEOUT, PCMK_OCF_UNKNOWN_ERROR, reason); op->call_id = -1; op->user_data = pcmk__transition_key(controld_globals.transition_graph->id, action->id, target_rc, controld_globals.te_uuid); free(dynamic_reason); return op; } static void controld_record_action_event(pcmk__graph_action_t *action, lrmd_event_data_t *op) { cib_t *cib_conn = controld_globals.cib_conn; xmlNode *state = NULL; xmlNode *rsc = NULL; xmlNode *action_rsc = NULL; int rc = pcmk_ok; const char *rsc_id = NULL; const char *target = crm_element_value(action->xml, PCMK__META_ON_NODE); const char *task_uuid = crm_element_value(action->xml, PCMK__XA_OPERATION_KEY); const char *target_uuid = crm_element_value(action->xml, PCMK__META_ON_NODE_UUID); int target_rc = get_target_rc(action); action_rsc = pcmk__xe_first_child(action->xml, PCMK_XE_PRIMITIVE, NULL, NULL); if (action_rsc == NULL) { return; } rsc_id = pcmk__xe_id(action_rsc); CRM_CHECK(rsc_id != NULL, crm_log_xml_err(action->xml, "Bad:action"); return); /* update the CIB */ state = pcmk__xe_create(NULL, PCMK__XE_NODE_STATE); crm_xml_add(state, PCMK_XA_ID, target_uuid); crm_xml_add(state, PCMK_XA_UNAME, target); rsc = pcmk__xe_create(state, PCMK__XE_LRM); crm_xml_add(rsc, PCMK_XA_ID, target_uuid); rsc = pcmk__xe_create(rsc, PCMK__XE_LRM_RESOURCES); rsc = pcmk__xe_create(rsc, PCMK__XE_LRM_RESOURCE); crm_xml_add(rsc, PCMK_XA_ID, rsc_id); crm_copy_xml_element(action_rsc, rsc, PCMK_XA_TYPE); crm_copy_xml_element(action_rsc, rsc, PCMK_XA_CLASS); crm_copy_xml_element(action_rsc, rsc, PCMK_XA_PROVIDER); pcmk__create_history_xml(rsc, op, CRM_FEATURE_SET, target_rc, target, __func__); rc = cib_conn->cmds->modify(cib_conn, PCMK_XE_STATUS, state, cib_none); fsa_register_cib_callback(rc, NULL, cib_action_updated); pcmk__xml_free(state); crm_trace("Sent CIB update (call ID %d) for synthesized event of action %d (%s on %s)", rc, action->id, task_uuid, target); pcmk__set_graph_action_flags(action, pcmk__graph_action_sent_update); } void controld_record_action_timeout(pcmk__graph_action_t *action) { lrmd_event_data_t *op = NULL; const char *target = crm_element_value(action->xml, PCMK__META_ON_NODE); const char *task_uuid = crm_element_value(action->xml, PCMK__XA_OPERATION_KEY); int target_rc = get_target_rc(action); crm_warn("%s %d: %s on %s timed out", action->xml->name, action->id, task_uuid, target); op = synthesize_timeout_event(action, target_rc); controld_record_action_event(action, op); lrmd_free_event(op); } /*! * \internal * \brief Execute a resource action from a transition graph * * \param[in,out] graph Transition graph being executed * \param[in,out] action Resource action to execute * * \return Standard Pacemaker return code */ static int execute_rsc_action(pcmk__graph_t *graph, pcmk__graph_action_t *action) { /* never overwrite stop actions in the CIB with * anything other than completed results * * Writing pending stops makes it look like the * resource is running again */ xmlNode *cmd = NULL; xmlNode *rsc_op = NULL; gboolean rc = TRUE; gboolean no_wait = FALSE; gboolean is_local = FALSE; char *counter = NULL; const char *task = NULL; const char *value = NULL; const char *on_node = NULL; const char *router_node = NULL; const char *task_uuid = NULL; CRM_ASSERT(action != NULL); CRM_ASSERT(action->xml != NULL); pcmk__clear_graph_action_flags(action, pcmk__graph_action_executed); on_node = crm_element_value(action->xml, PCMK__META_ON_NODE); CRM_CHECK(!pcmk__str_empty(on_node), crm_err("Corrupted command(id=%s) %s: no node", pcmk__xe_id(action->xml), pcmk__s(task, "without task")); return pcmk_rc_node_unknown); rsc_op = action->xml; task = crm_element_value(rsc_op, PCMK_XA_OPERATION); task_uuid = crm_element_value(action->xml, PCMK__XA_OPERATION_KEY); router_node = crm_element_value(rsc_op, PCMK__XA_ROUTER_NODE); if (!router_node) { router_node = on_node; } counter = pcmk__transition_key(controld_globals.transition_graph->id, action->id, get_target_rc(action), controld_globals.te_uuid); crm_xml_add(rsc_op, PCMK__XA_TRANSITION_KEY, counter); if (pcmk__str_eq(router_node, controld_globals.our_nodename, pcmk__str_casei)) { is_local = TRUE; } value = crm_meta_value(action->params, PCMK__META_OP_NO_WAIT); if (crm_is_true(value)) { no_wait = TRUE; } crm_notice("Initiating %s operation %s%s on %s%s " QB_XS " action %d", task, task_uuid, (is_local? " locally" : ""), on_node, (no_wait? " without waiting" : ""), action->id); cmd = create_request(CRM_OP_INVOKE_LRM, rsc_op, router_node, CRM_SYSTEM_LRMD, CRM_SYSTEM_TENGINE, NULL); if (is_local) { /* shortcut local resource commands */ ha_msg_input_t data = { .msg = cmd, .xml = rsc_op, }; fsa_data_t msg = { .id = 0, .data = &data, .data_type = fsa_dt_ha_msg, .fsa_input = I_NULL, .fsa_cause = C_FSA_INTERNAL, .actions = A_LRM_INVOKE, .origin = __func__, }; do_lrm_invoke(A_LRM_INVOKE, C_FSA_INTERNAL, controld_globals.fsa_state, I_NULL, &msg); } else { const pcmk__node_status_t *node = pcmk__get_node(0, router_node, NULL, pcmk__node_search_cluster_member); - rc = pcmk__cluster_send_message(node, crm_msg_lrmd, cmd); + rc = pcmk__cluster_send_message(node, pcmk__cluster_msg_execd, cmd); } free(counter); pcmk__xml_free(cmd); pcmk__set_graph_action_flags(action, pcmk__graph_action_executed); if (rc == FALSE) { crm_err("Action %d failed: send", action->id); return ECOMM; } else if (no_wait) { /* Just mark confirmed. Don't bump the job count only to immediately * decrement it. */ crm_info("Action %d confirmed - no wait", action->id); pcmk__set_graph_action_flags(action, pcmk__graph_action_confirmed); pcmk__update_graph(controld_globals.transition_graph, action); trigger_graph(); } else if (pcmk_is_set(action->flags, pcmk__graph_action_confirmed)) { crm_debug("Action %d: %s %s on %s(timeout %dms) was already confirmed.", action->id, task, task_uuid, on_node, action->timeout); } else { if (action->timeout <= 0) { crm_err("Action %d: %s %s on %s had an invalid timeout (%dms). Using %ums instead", action->id, task, task_uuid, on_node, action->timeout, graph->network_delay); action->timeout = (int) graph->network_delay; } te_update_job_count(action, 1); te_start_action_timer(graph, action); } return pcmk_rc_ok; } struct te_peer_s { char *name; int jobs; int migrate_jobs; }; static void te_peer_free(gpointer p) { struct te_peer_s *peer = p; free(peer->name); free(peer); } void te_reset_job_counts(void) { GHashTableIter iter; struct te_peer_s *peer = NULL; if(te_targets == NULL) { te_targets = pcmk__strkey_table(NULL, te_peer_free); } g_hash_table_iter_init(&iter, te_targets); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) & peer)) { peer->jobs = 0; peer->migrate_jobs = 0; } } static void te_update_job_count_on(const char *target, int offset, bool migrate) { struct te_peer_s *r = NULL; if(target == NULL || te_targets == NULL) { return; } r = g_hash_table_lookup(te_targets, target); if(r == NULL) { r = pcmk__assert_alloc(1, sizeof(struct te_peer_s)); r->name = pcmk__str_copy(target); g_hash_table_insert(te_targets, r->name, r); } r->jobs += offset; if(migrate) { r->migrate_jobs += offset; } crm_trace("jobs[%s] = %d", target, r->jobs); } static void te_update_job_count(pcmk__graph_action_t *action, int offset) { const char *task = crm_element_value(action->xml, PCMK_XA_OPERATION); const char *target = crm_element_value(action->xml, PCMK__META_ON_NODE); if ((action->type != pcmk__rsc_graph_action) || (target == NULL)) { /* No limit on these */ return; } /* if we have a router node, this means the action is performing * on a remote node. For now, we count all actions occurring on a * remote node against the job list on the cluster node hosting * the connection resources */ target = crm_element_value(action->xml, PCMK__XA_ROUTER_NODE); if ((target == NULL) && pcmk__strcase_any_of(task, PCMK_ACTION_MIGRATE_TO, PCMK_ACTION_MIGRATE_FROM, NULL)) { const char *t1 = crm_meta_value(action->params, PCMK__META_MIGRATE_SOURCE); const char *t2 = crm_meta_value(action->params, PCMK__META_MIGRATE_TARGET); te_update_job_count_on(t1, offset, TRUE); te_update_job_count_on(t2, offset, TRUE); return; } else if (target == NULL) { target = crm_element_value(action->xml, PCMK__META_ON_NODE); } te_update_job_count_on(target, offset, FALSE); } /*! * \internal * \brief Check whether a graph action is allowed to be executed on a node * * \param[in] graph Transition graph being executed * \param[in] action Graph action being executed * \param[in] target Name of node where action should be executed * * \return true if action is allowed, otherwise false */ static bool allowed_on_node(const pcmk__graph_t *graph, const pcmk__graph_action_t *action, const char *target) { int limit = 0; struct te_peer_s *r = NULL; const char *task = crm_element_value(action->xml, PCMK_XA_OPERATION); const char *id = crm_element_value(action->xml, PCMK__XA_OPERATION_KEY); if(target == NULL) { /* No limit on these */ return true; } else if(te_targets == NULL) { return false; } r = g_hash_table_lookup(te_targets, target); limit = throttle_get_job_limit(target); if(r == NULL) { r = pcmk__assert_alloc(1, sizeof(struct te_peer_s)); r->name = pcmk__str_copy(target); g_hash_table_insert(te_targets, r->name, r); } if(limit <= r->jobs) { crm_trace("Peer %s is over their job limit of %d (%d): deferring %s", target, limit, r->jobs, id); return false; } else if(graph->migration_limit > 0 && r->migrate_jobs >= graph->migration_limit) { if (pcmk__strcase_any_of(task, PCMK_ACTION_MIGRATE_TO, PCMK_ACTION_MIGRATE_FROM, NULL)) { crm_trace("Peer %s is over their migration job limit of %d (%d): deferring %s", target, graph->migration_limit, r->migrate_jobs, id); return false; } } crm_trace("Peer %s has not hit their limit yet. current jobs = %d limit= %d limit", target, r->jobs, limit); return true; } /*! * \internal * \brief Check whether a graph action is allowed to be executed * * \param[in] graph Transition graph being executed * \param[in] action Graph action being executed * * \return true if action is allowed, otherwise false */ static bool graph_action_allowed(pcmk__graph_t *graph, pcmk__graph_action_t *action) { const char *target = NULL; const char *task = crm_element_value(action->xml, PCMK_XA_OPERATION); if (action->type != pcmk__rsc_graph_action) { /* No limit on these */ return true; } /* if we have a router node, this means the action is performing * on a remote node. For now, we count all actions occurring on a * remote node against the job list on the cluster node hosting * the connection resources */ target = crm_element_value(action->xml, PCMK__XA_ROUTER_NODE); if ((target == NULL) && pcmk__strcase_any_of(task, PCMK_ACTION_MIGRATE_TO, PCMK_ACTION_MIGRATE_FROM, NULL)) { target = crm_meta_value(action->params, PCMK__META_MIGRATE_SOURCE); if (!allowed_on_node(graph, action, target)) { return false; } target = crm_meta_value(action->params, PCMK__META_MIGRATE_TARGET); } else if (target == NULL) { target = crm_element_value(action->xml, PCMK__META_ON_NODE); } return allowed_on_node(graph, action, target); } /*! * \brief Confirm a graph action (and optionally update graph) * * \param[in,out] action Action to confirm * \param[in,out] graph Update and trigger this graph (if non-NULL) */ void te_action_confirmed(pcmk__graph_action_t *action, pcmk__graph_t *graph) { if (!pcmk_is_set(action->flags, pcmk__graph_action_confirmed)) { if ((action->type == pcmk__rsc_graph_action) && (crm_element_value(action->xml, PCMK__META_ON_NODE) != NULL)) { te_update_job_count(action, -1); } pcmk__set_graph_action_flags(action, pcmk__graph_action_confirmed); } if (graph) { pcmk__update_graph(graph, action); trigger_graph(); } } static pcmk__graph_functions_t te_graph_fns = { execute_pseudo_action, execute_rsc_action, execute_cluster_action, controld_execute_fence_action, graph_action_allowed, }; /* * \internal * \brief Register the transitioner's graph functions with \p libpacemaker */ void controld_register_graph_functions(void) { pcmk__set_graph_functions(&te_graph_fns); } void notify_crmd(pcmk__graph_t *graph) { const char *type = "unknown"; enum crmd_fsa_input event = I_NULL; crm_debug("Processing transition completion in state %s", fsa_state2string(controld_globals.fsa_state)); CRM_CHECK(graph->complete, graph->complete = true); switch (graph->completion_action) { case pcmk__graph_wait: type = "stop"; if (controld_globals.fsa_state == S_TRANSITION_ENGINE) { event = I_TE_SUCCESS; } break; case pcmk__graph_done: type = "done"; if (controld_globals.fsa_state == S_TRANSITION_ENGINE) { event = I_TE_SUCCESS; } break; case pcmk__graph_restart: type = "restart"; if (controld_globals.fsa_state == S_TRANSITION_ENGINE) { if (controld_get_period_transition_timer() > 0) { controld_stop_transition_timer(); controld_start_transition_timer(); } else { event = I_PE_CALC; } } else if (controld_globals.fsa_state == S_POLICY_ENGINE) { controld_set_fsa_action_flags(A_PE_INVOKE); controld_trigger_fsa(); } break; case pcmk__graph_shutdown: type = "shutdown"; if (pcmk_is_set(controld_globals.fsa_input_register, R_SHUTDOWN)) { event = I_STOP; } else { crm_err("We didn't ask to be shut down, yet the scheduler is telling us to"); event = I_TERMINATE; } } crm_debug("Transition %d status: %s - %s", graph->id, type, pcmk__s(graph->abort_reason, "unspecified reason")); graph->abort_reason = NULL; graph->completion_action = pcmk__graph_done; if (event != I_NULL) { register_fsa_input(C_FSA_INTERNAL, event, NULL); } else { controld_trigger_fsa(); } } diff --git a/daemons/controld/controld_throttle.c b/daemons/controld/controld_throttle.c index 415b89ca75..995d33b5ae 100644 --- a/daemons/controld/controld_throttle.c +++ b/daemons/controld/controld_throttle.c @@ -1,574 +1,574 @@ /* * Copyright 2013-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include #include /* These values don't need to be bits, but these particular values must be kept * for backward compatibility during rolling upgrades. */ enum throttle_state_e { throttle_none = 0x0000, throttle_low = 0x0001, throttle_med = 0x0010, throttle_high = 0x0100, throttle_extreme = 0x1000, }; struct throttle_record_s { int max; enum throttle_state_e mode; char *node; }; static int throttle_job_max = 0; static float throttle_load_target = 0.0; #define THROTTLE_FACTOR_LOW 1.2 #define THROTTLE_FACTOR_MEDIUM 1.6 #define THROTTLE_FACTOR_HIGH 2.0 static GHashTable *throttle_records = NULL; static mainloop_timer_t *throttle_timer = NULL; static const char * load2str(enum throttle_state_e mode) { switch (mode) { case throttle_extreme: return "extreme"; case throttle_high: return "high"; case throttle_med: return "medium"; case throttle_low: return "low"; case throttle_none: return "negligible"; default: return "undetermined"; } } #if HAVE_LINUX_PROCFS /*! * \internal * \brief Return name of /proc file containing the CIB daemon's load statistics * * \return Newly allocated memory with file name on success, NULL otherwise * * \note It is the caller's responsibility to free the return value. * This will return NULL if the daemon is being run via valgrind. * This should be called only on Linux systems. */ static char * find_cib_loadfile(void) { pid_t pid = pcmk__procfs_pid_of("pacemaker-based"); return pid? crm_strdup_printf("/proc/%lld/stat", (long long) pid) : NULL; } static bool throttle_cib_load(float *load) { /* /proc/[pid]/stat Status information about the process. This is used by ps(1). It is defined in /usr/src/linux/fs/proc/array.c. The fields, in order, with their proper scanf(3) format specifiers, are: pid %d (1) The process ID. comm %s (2) The filename of the executable, in parentheses. This is visible whether or not the executable is swapped out. state %c (3) One character from the string "RSDZTW" where R is running, S is sleeping in an interruptible wait, D is waiting in uninterruptible disk sleep, Z is zombie, T is traced or stopped (on a signal), and W is paging. ppid %d (4) The PID of the parent. pgrp %d (5) The process group ID of the process. session %d (6) The session ID of the process. tty_nr %d (7) The controlling terminal of the process. (The minor device number is contained in the combination of bits 31 to 20 and 7 to 0; the major device number is in bits 15 to 8.) tpgid %d (8) The ID of the foreground process group of the controlling terminal of the process. flags %u (%lu before Linux 2.6.22) (9) The kernel flags word of the process. For bit meanings, see the PF_* defines in the Linux kernel source file include/linux/sched.h. Details depend on the kernel version. minflt %lu (10) The number of minor faults the process has made which have not required loading a memory page from disk. cminflt %lu (11) The number of minor faults that the process's waited-for children have made. majflt %lu (12) The number of major faults the process has made which have required loading a memory page from disk. cmajflt %lu (13) The number of major faults that the process's waited-for children have made. utime %lu (14) Amount of time that this process has been scheduled in user mode, measured in clock ticks (divide by sysconf(_SC_CLK_TCK)). This includes guest time, guest_time (time spent running a virtual CPU, see below), so that applications that are not aware of the guest time field do not lose that time from their calculations. stime %lu (15) Amount of time that this process has been scheduled in kernel mode, measured in clock ticks (divide by sysconf(_SC_CLK_TCK)). */ static char *loadfile = NULL; static time_t last_call = 0; static long ticks_per_s = 0; static unsigned long last_utime, last_stime; char buffer[64*1024]; FILE *stream = NULL; time_t now = time(NULL); if(load == NULL) { return FALSE; } else { *load = 0.0; } if(loadfile == NULL) { last_call = 0; last_utime = 0; last_stime = 0; loadfile = find_cib_loadfile(); if (loadfile == NULL) { crm_warn("Couldn't find CIB load file"); return FALSE; } ticks_per_s = sysconf(_SC_CLK_TCK); crm_trace("Found %s", loadfile); } stream = fopen(loadfile, "r"); if(stream == NULL) { int rc = errno; crm_warn("Couldn't read %s: %s (%d)", loadfile, pcmk_rc_str(rc), rc); free(loadfile); loadfile = NULL; return FALSE; } if(fgets(buffer, sizeof(buffer), stream)) { char *comm = pcmk__assert_alloc(1, 256); char state = 0; int rc = 0, pid = 0, ppid = 0, pgrp = 0, session = 0, tty_nr = 0, tpgid = 0; unsigned long flags = 0, minflt = 0, cminflt = 0, majflt = 0, cmajflt = 0, utime = 0, stime = 0; rc = sscanf(buffer, "%d %[^ ] %c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu", &pid, comm, &state, &ppid, &pgrp, &session, &tty_nr, &tpgid, &flags, &minflt, &cminflt, &majflt, &cmajflt, &utime, &stime); free(comm); if(rc != 15) { crm_err("Only %d of 15 fields found in %s", rc, loadfile); fclose(stream); return FALSE; } else if(last_call > 0 && last_call < now && last_utime <= utime && last_stime <= stime) { time_t elapsed = now - last_call; unsigned long delta_utime = utime - last_utime; unsigned long delta_stime = stime - last_stime; *load = (delta_utime + delta_stime); /* Cast to a float before division */ *load /= ticks_per_s; *load /= elapsed; crm_debug("cib load: %f (%lu ticks in %lds)", *load, delta_utime + delta_stime, (long)elapsed); } else { crm_debug("Init %lu + %lu ticks at %ld (%lu tps)", utime, stime, (long)now, ticks_per_s); } last_call = now; last_utime = utime; last_stime = stime; fclose(stream); return TRUE; } fclose(stream); return FALSE; } static bool throttle_load_avg(float *load) { char buffer[256]; FILE *stream = NULL; const char *loadfile = "/proc/loadavg"; if(load == NULL) { return FALSE; } stream = fopen(loadfile, "r"); if(stream == NULL) { int rc = errno; crm_warn("Couldn't read %s: %s (%d)", loadfile, pcmk_rc_str(rc), rc); return FALSE; } if(fgets(buffer, sizeof(buffer), stream)) { char *nl = strstr(buffer, "\n"); /* Grab the 1-minute average, ignore the rest */ *load = strtof(buffer, NULL); if(nl) { nl[0] = 0; } fclose(stream); return TRUE; } fclose(stream); return FALSE; } /*! * \internal * \brief Check a load value against throttling thresholds * * \param[in] load Load value to check * \param[in] desc Description of metric (for logging) * \param[in] thresholds Low/medium/high/extreme thresholds * * \return Throttle mode corresponding to load value */ static enum throttle_state_e throttle_check_thresholds(float load, const char *desc, const float thresholds[4]) { if (load > thresholds[3]) { crm_notice("Extreme %s detected: %f", desc, load); return throttle_extreme; } else if (load > thresholds[2]) { crm_notice("High %s detected: %f", desc, load); return throttle_high; } else if (load > thresholds[1]) { crm_info("Moderate %s detected: %f", desc, load); return throttle_med; } else if (load > thresholds[0]) { crm_debug("Noticeable %s detected: %f", desc, load); return throttle_low; } crm_trace("Negligible %s detected: %f", desc, load); return throttle_none; } static enum throttle_state_e throttle_handle_load(float load, const char *desc, int cores) { float normalize; float thresholds[4]; if (cores == 1) { /* On a single core machine, a load of 1.0 is already too high */ normalize = 0.6; } else { /* Normalize the load to be per-core */ normalize = cores; } thresholds[0] = throttle_load_target * normalize * THROTTLE_FACTOR_LOW; thresholds[1] = throttle_load_target * normalize * THROTTLE_FACTOR_MEDIUM; thresholds[2] = throttle_load_target * normalize * THROTTLE_FACTOR_HIGH; thresholds[3] = load + 1.0; /* never extreme */ return throttle_check_thresholds(load, desc, thresholds); } #endif // HAVE_LINUX_PROCFS static enum throttle_state_e throttle_mode(void) { enum throttle_state_e mode = throttle_none; #if HAVE_LINUX_PROCFS unsigned int cores; float load; float thresholds[4]; cores = pcmk__procfs_num_cores(); if(throttle_cib_load(&load)) { float cib_max_cpu = 0.95; /* The CIB is a single-threaded task and thus cannot consume * more than 100% of a CPU (and 1/cores of the overall system * load). * * On a many-cored system, the CIB might therefore be maxed out * (causing operations to fail or appear to fail) even though * the overall system load is still reasonable. * * Therefore, the 'normal' thresholds can not apply here, and we * need a special case. */ if(cores == 1) { cib_max_cpu = 0.4; } if(throttle_load_target > 0.0 && throttle_load_target < cib_max_cpu) { cib_max_cpu = throttle_load_target; } thresholds[0] = cib_max_cpu * 0.8; thresholds[1] = cib_max_cpu * 0.9; thresholds[2] = cib_max_cpu; /* Can only happen on machines with a low number of cores */ thresholds[3] = cib_max_cpu * 1.5; mode = throttle_check_thresholds(load, "CIB load", thresholds); } if(throttle_load_target <= 0) { /* If we ever make this a valid value, the cluster will at least behave as expected */ return mode; } if(throttle_load_avg(&load)) { enum throttle_state_e cpu_load; cpu_load = throttle_handle_load(load, "CPU load", cores); if (cpu_load > mode) { mode = cpu_load; } crm_debug("Current load is %f across %u core(s)", load, cores); } #endif // HAVE_LINUX_PROCFS return mode; } static void throttle_send_command(enum throttle_state_e mode) { xmlNode *xml = NULL; static enum throttle_state_e last = -1; if(mode != last) { crm_info("New throttle mode: %s load (was %s)", load2str(mode), load2str(last)); last = mode; xml = create_request(CRM_OP_THROTTLE, NULL, NULL, CRM_SYSTEM_CRMD, CRM_SYSTEM_CRMD, NULL); crm_xml_add_int(xml, PCMK__XA_CRM_LIMIT_MODE, mode); crm_xml_add_int(xml, PCMK__XA_CRM_LIMIT_MAX, throttle_job_max); - pcmk__cluster_send_message(NULL, crm_msg_crmd, xml); + pcmk__cluster_send_message(NULL, pcmk__cluster_msg_controld, xml); pcmk__xml_free(xml); } } static gboolean throttle_timer_cb(gpointer data) { throttle_send_command(throttle_mode()); return TRUE; } static void throttle_record_free(gpointer p) { struct throttle_record_s *r = p; free(r->node); free(r); } static void throttle_set_load_target(float target) { throttle_load_target = target; } /*! * \internal * \brief Update the maximum number of simultaneous jobs * * \param[in] preference Cluster-wide \c PCMK_OPT_NODE_ACTION_LIMIT from the * CIB */ static void throttle_update_job_max(const char *preference) { long long max = 0LL; const char *env_limit = pcmk__env_option(PCMK__ENV_NODE_ACTION_LIMIT); if (env_limit != NULL) { preference = env_limit; // Per-node override } if (preference != NULL) { pcmk__scan_ll(preference, &max, 0LL); } if (max > 0) { throttle_job_max = (max >= INT_MAX)? INT_MAX : (int) max; } else { // Default is based on the number of cores detected throttle_job_max = 2 * pcmk__procfs_num_cores(); } } void throttle_init(void) { if(throttle_records == NULL) { throttle_records = pcmk__strkey_table(NULL, throttle_record_free); throttle_timer = mainloop_timer_add("throttle", 30 * 1000, TRUE, throttle_timer_cb, NULL); } throttle_update_job_max(NULL); mainloop_timer_start(throttle_timer); } /*! * \internal * \brief Configure throttle options based on the CIB * * \param[in,out] options Name/value pairs for configured options */ void controld_configure_throttle(GHashTable *options) { const char *value = g_hash_table_lookup(options, PCMK_OPT_LOAD_THRESHOLD); if (value != NULL) { throttle_set_load_target(strtof(value, NULL) / 100.0); } value = g_hash_table_lookup(options, PCMK_OPT_NODE_ACTION_LIMIT); throttle_update_job_max(value); } void throttle_fini(void) { if (throttle_timer != NULL) { mainloop_timer_del(throttle_timer); throttle_timer = NULL; } if (throttle_records != NULL) { g_hash_table_destroy(throttle_records); throttle_records = NULL; } } int throttle_get_total_job_limit(int l) { /* Cluster-wide limit */ GHashTableIter iter; int limit = l; int peers = pcmk__cluster_num_active_nodes(); struct throttle_record_s *r = NULL; g_hash_table_iter_init(&iter, throttle_records); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &r)) { switch(r->mode) { case throttle_extreme: if(limit == 0 || limit > peers/4) { limit = QB_MAX(1, peers/4); } break; case throttle_high: if(limit == 0 || limit > peers/2) { limit = QB_MAX(1, peers/2); } break; default: break; } } if(limit == l) { } else if(l == 0) { crm_trace("Using " PCMK_OPT_BATCH_LIMIT "=%d", limit); } else { crm_trace("Using " PCMK_OPT_BATCH_LIMIT "=%d instead of %d", limit, l); } return limit; } int throttle_get_job_limit(const char *node) { int jobs = 1; struct throttle_record_s *r = NULL; r = g_hash_table_lookup(throttle_records, node); if(r == NULL) { r = pcmk__assert_alloc(1, sizeof(struct throttle_record_s)); r->node = pcmk__str_copy(node); r->mode = throttle_low; r->max = throttle_job_max; crm_trace("Defaulting to local values for unknown node %s", node); g_hash_table_insert(throttle_records, r->node, r); } switch(r->mode) { case throttle_extreme: case throttle_high: jobs = 1; /* At least one job must always be allowed */ break; case throttle_med: jobs = QB_MAX(1, r->max / 4); break; case throttle_low: jobs = QB_MAX(1, r->max / 2); break; case throttle_none: jobs = QB_MAX(1, r->max); break; default: crm_err("Unknown throttle mode %.4x on %s", r->mode, node); break; } return jobs; } void throttle_update(xmlNode *xml) { int max = 0; int mode = 0; struct throttle_record_s *r = NULL; const char *from = crm_element_value(xml, PCMK__XA_SRC); crm_element_value_int(xml, PCMK__XA_CRM_LIMIT_MODE, &mode); crm_element_value_int(xml, PCMK__XA_CRM_LIMIT_MAX, &max); r = g_hash_table_lookup(throttle_records, from); if(r == NULL) { r = pcmk__assert_alloc(1, sizeof(struct throttle_record_s)); r->node = pcmk__str_copy(from); g_hash_table_insert(throttle_records, r->node, r); } r->max = max; r->mode = (enum throttle_state_e) mode; crm_debug("Node %s has %s load and supports at most %d jobs; new job limit %d", from, load2str((enum throttle_state_e) mode), max, throttle_get_job_limit(from)); } diff --git a/daemons/fenced/fenced_commands.c b/daemons/fenced/fenced_commands.c index e8d9a9f37a..067a379692 100644 --- a/daemons/fenced/fenced_commands.c +++ b/daemons/fenced/fenced_commands.c @@ -1,3638 +1,3639 @@ /* * Copyright 2009-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include GHashTable *device_list = NULL; GHashTable *topology = NULL; static GList *cmd_list = NULL; static GHashTable *fenced_handlers = NULL; struct device_search_s { /* target of fence action */ char *host; /* requested fence action */ char *action; /* timeout to use if a device is queried dynamically for possible targets */ int per_device_timeout; /* number of registered fencing devices at time of request */ int replies_needed; /* number of device replies received so far */ int replies_received; /* whether the target is eligible to perform requested action (or off) */ bool allow_suicide; /* private data to pass to search callback function */ void *user_data; /* function to call when all replies have been received */ void (*callback) (GList * devices, void *user_data); /* devices capable of performing requested action (or off if remapping) */ GList *capable; /* Whether to perform searches that support the action */ uint32_t support_action_only; }; static gboolean stonith_device_dispatch(gpointer user_data); static void st_child_done(int pid, const pcmk__action_result_t *result, void *user_data); static void search_devices_record_result(struct device_search_s *search, const char *device, gboolean can_fence); static int get_agent_metadata(const char *agent, xmlNode **metadata); static void read_action_metadata(stonith_device_t *device); static enum fenced_target_by unpack_level_kind(const xmlNode *level); typedef struct async_command_s { int id; int pid; int fd_stdout; int options; int default_timeout; /* seconds */ int timeout; /* seconds */ int start_delay; // seconds (-1 means disable static/random fencing delays) int delay_id; char *op; char *origin; char *client; char *client_name; char *remote_op_id; char *target; uint32_t target_nodeid; char *action; char *device; GList *device_list; GList *next_device_iter; // device_list entry for next device to execute void *internal_user_data; void (*done_cb) (int pid, const pcmk__action_result_t *result, void *user_data); guint timer_sigterm; guint timer_sigkill; /*! If the operation timed out, this is the last signal * we sent to the process to get it to terminate */ int last_timeout_signo; stonith_device_t *active_on; stonith_device_t *activating_on; } async_command_t; static xmlNode *construct_async_reply(const async_command_t *cmd, const pcmk__action_result_t *result); static gboolean is_action_required(const char *action, const stonith_device_t *device) { return (device != NULL) && device->automatic_unfencing && pcmk__str_eq(action, PCMK_ACTION_ON, pcmk__str_none); } static int get_action_delay_max(const stonith_device_t *device, const char *action) { const char *value = NULL; guint delay_max = 0U; if (!pcmk__is_fencing_action(action)) { return 0; } value = g_hash_table_lookup(device->params, PCMK_STONITH_DELAY_MAX); if (value) { pcmk_parse_interval_spec(value, &delay_max); delay_max /= 1000; } return (int) delay_max; } static int get_action_delay_base(const stonith_device_t *device, const char *action, const char *target) { char *hash_value = NULL; guint delay_base = 0U; if (!pcmk__is_fencing_action(action)) { return 0; } hash_value = g_hash_table_lookup(device->params, PCMK_STONITH_DELAY_BASE); if (hash_value) { char *value = pcmk__str_copy(hash_value); char *valptr = value; if (target != NULL) { for (char *val = strtok(value, "; \t"); val != NULL; val = strtok(NULL, "; \t")) { char *mapval = strchr(val, ':'); if (mapval == NULL || mapval[1] == 0) { crm_err("pcmk_delay_base: empty value in mapping", val); continue; } if (mapval != val && strncasecmp(target, val, (size_t)(mapval - val)) == 0) { value = mapval + 1; crm_debug("pcmk_delay_base mapped to %s for %s", value, target); break; } } } if (strchr(value, ':') == 0) { pcmk_parse_interval_spec(value, &delay_base); delay_base /= 1000; } free(valptr); } return (int) delay_base; } /*! * \internal * \brief Override STONITH timeout with pcmk_*_timeout if available * * \param[in] device STONITH device to use * \param[in] action STONITH action name * \param[in] default_timeout Timeout to use if device does not have * a pcmk_*_timeout parameter for action * * \return Value of pcmk_(action)_timeout if available, otherwise default_timeout * \note For consistency, it would be nice if reboot/off/on timeouts could be * set the same way as start/stop/monitor timeouts, i.e. with an * entry in the fencing resource configuration. However that * is insufficient because fencing devices may be registered directly via * the fencer's register_device() API instead of going through the CIB * (e.g. stonith_admin uses it for its -R option, and the executor uses it * to ensure a device is registered when a command is issued). As device * properties, pcmk_*_timeout parameters can be grabbed by the fencer when * the device is registered, whether by CIB change or API call. */ static int get_action_timeout(const stonith_device_t *device, const char *action, int default_timeout) { if (action && device && device->params) { char buffer[64] = { 0, }; const char *value = NULL; /* If "reboot" was requested but the device does not support it, * we will remap to "off", so check timeout for "off" instead */ if (pcmk__str_eq(action, PCMK_ACTION_REBOOT, pcmk__str_none) && !pcmk_is_set(device->flags, st_device_supports_reboot)) { crm_trace("%s doesn't support reboot, using timeout for off instead", device->id); action = PCMK_ACTION_OFF; } /* If the device config specified an action-specific timeout, use it */ snprintf(buffer, sizeof(buffer), "pcmk_%s_timeout", action); value = g_hash_table_lookup(device->params, buffer); if (value) { long long timeout_ms = crm_get_msec(value); return (int) QB_MIN(timeout_ms / 1000, INT_MAX); } } return default_timeout; } /*! * \internal * \brief Get the currently executing device for a fencing operation * * \param[in] cmd Fencing operation to check * * \return Currently executing device for \p cmd if any, otherwise NULL */ static stonith_device_t * cmd_device(const async_command_t *cmd) { if ((cmd == NULL) || (cmd->device == NULL) || (device_list == NULL)) { return NULL; } return g_hash_table_lookup(device_list, cmd->device); } /*! * \internal * \brief Return the configured reboot action for a given device * * \param[in] device_id Device ID * * \return Configured reboot action for \p device_id */ const char * fenced_device_reboot_action(const char *device_id) { const char *action = NULL; if ((device_list != NULL) && (device_id != NULL)) { stonith_device_t *device = g_hash_table_lookup(device_list, device_id); if ((device != NULL) && (device->params != NULL)) { action = g_hash_table_lookup(device->params, "pcmk_reboot_action"); } } return pcmk__s(action, PCMK_ACTION_REBOOT); } /*! * \internal * \brief Check whether a given device supports the "on" action * * \param[in] device_id Device ID * * \return true if \p device_id supports "on", otherwise false */ bool fenced_device_supports_on(const char *device_id) { if ((device_list != NULL) && (device_id != NULL)) { stonith_device_t *device = g_hash_table_lookup(device_list, device_id); if (device != NULL) { return pcmk_is_set(device->flags, st_device_supports_on); } } return false; } static void free_async_command(async_command_t * cmd) { if (!cmd) { return; } if (cmd->delay_id) { g_source_remove(cmd->delay_id); } cmd_list = g_list_remove(cmd_list, cmd); g_list_free_full(cmd->device_list, free); free(cmd->device); free(cmd->action); free(cmd->target); free(cmd->remote_op_id); free(cmd->client); free(cmd->client_name); free(cmd->origin); free(cmd->op); free(cmd); } /*! * \internal * \brief Create a new asynchronous fencing operation from request XML * * \param[in] msg Fencing request XML (from IPC or CPG) * * \return Newly allocated fencing operation on success, otherwise NULL * * \note This asserts on memory errors, so a NULL return indicates an * unparseable message. */ static async_command_t * create_async_command(xmlNode *msg) { xmlNode *op = NULL; async_command_t *cmd = NULL; if (msg == NULL) { return NULL; } op = get_xpath_object("//@" PCMK__XE_ST_DEVICE_ACTION, msg, LOG_ERR); if (op == NULL) { return NULL; } cmd = pcmk__assert_alloc(1, sizeof(async_command_t)); // All messages must include these cmd->action = crm_element_value_copy(op, PCMK__XA_ST_DEVICE_ACTION); cmd->op = crm_element_value_copy(msg, PCMK__XA_ST_OP); cmd->client = crm_element_value_copy(msg, PCMK__XA_ST_CLIENTID); if ((cmd->action == NULL) || (cmd->op == NULL) || (cmd->client == NULL)) { free_async_command(cmd); return NULL; } crm_element_value_int(msg, PCMK__XA_ST_CALLID, &(cmd->id)); crm_element_value_int(msg, PCMK__XA_ST_CALLOPT, &(cmd->options)); crm_element_value_int(msg, PCMK__XA_ST_DELAY, &(cmd->start_delay)); crm_element_value_int(msg, PCMK__XA_ST_TIMEOUT, &(cmd->default_timeout)); cmd->timeout = cmd->default_timeout; cmd->origin = crm_element_value_copy(msg, PCMK__XA_SRC); cmd->remote_op_id = crm_element_value_copy(msg, PCMK__XA_ST_REMOTE_OP); cmd->client_name = crm_element_value_copy(msg, PCMK__XA_ST_CLIENTNAME); cmd->target = crm_element_value_copy(op, PCMK__XA_ST_TARGET); cmd->device = crm_element_value_copy(op, PCMK__XA_ST_DEVICE_ID); cmd->done_cb = st_child_done; // Track in global command list cmd_list = g_list_append(cmd_list, cmd); return cmd; } static int get_action_limit(stonith_device_t * device) { const char *value = NULL; int action_limit = 1; value = g_hash_table_lookup(device->params, PCMK_STONITH_ACTION_LIMIT); if ((value == NULL) || (pcmk__scan_min_int(value, &action_limit, INT_MIN) != pcmk_rc_ok) || (action_limit == 0)) { action_limit = 1; } return action_limit; } static int get_active_cmds(stonith_device_t * device) { int counter = 0; GList *gIter = NULL; GList *gIterNext = NULL; CRM_CHECK(device != NULL, return 0); for (gIter = cmd_list; gIter != NULL; gIter = gIterNext) { async_command_t *cmd = gIter->data; gIterNext = gIter->next; if (cmd->active_on == device) { counter++; } } return counter; } static void fork_cb(int pid, void *user_data) { async_command_t *cmd = (async_command_t *) user_data; stonith_device_t * device = /* in case of a retry we've done the move from activating_on to active_on already */ cmd->activating_on?cmd->activating_on:cmd->active_on; CRM_ASSERT(device); crm_debug("Operation '%s' [%d]%s%s using %s now running with %ds timeout", cmd->action, pid, ((cmd->target == NULL)? "" : " targeting "), pcmk__s(cmd->target, ""), device->id, cmd->timeout); cmd->active_on = device; cmd->activating_on = NULL; } static int get_agent_metadata_cb(gpointer data) { stonith_device_t *device = data; guint period_ms; switch (get_agent_metadata(device->agent, &device->agent_metadata)) { case pcmk_rc_ok: if (device->agent_metadata) { read_action_metadata(device); stonith__device_parameter_flags(&(device->flags), device->id, device->agent_metadata); } return G_SOURCE_REMOVE; case EAGAIN: period_ms = pcmk__mainloop_timer_get_period(device->timer); if (period_ms < 160 * 1000) { mainloop_timer_set_period(device->timer, 2 * period_ms); } return G_SOURCE_CONTINUE; default: return G_SOURCE_REMOVE; } } /*! * \internal * \brief Call a command's action callback for an internal (not library) result * * \param[in,out] cmd Command to report result for * \param[in] execution_status Execution status to use for result * \param[in] exit_status Exit status to use for result * \param[in] exit_reason Exit reason to use for result */ static void report_internal_result(async_command_t *cmd, int exit_status, int execution_status, const char *exit_reason) { pcmk__action_result_t result = PCMK__UNKNOWN_RESULT; pcmk__set_result(&result, exit_status, execution_status, exit_reason); cmd->done_cb(0, &result, cmd); pcmk__reset_result(&result); } static gboolean stonith_device_execute(stonith_device_t * device) { int exec_rc = 0; const char *action_str = NULL; const char *host_arg = NULL; async_command_t *cmd = NULL; stonith_action_t *action = NULL; int active_cmds = 0; int action_limit = 0; GList *gIter = NULL; GList *gIterNext = NULL; CRM_CHECK(device != NULL, return FALSE); active_cmds = get_active_cmds(device); action_limit = get_action_limit(device); if (action_limit > -1 && active_cmds >= action_limit) { crm_trace("%s is over its action limit of %d (%u active action%s)", device->id, action_limit, active_cmds, pcmk__plural_s(active_cmds)); return TRUE; } for (gIter = device->pending_ops; gIter != NULL; gIter = gIterNext) { async_command_t *pending_op = gIter->data; gIterNext = gIter->next; if (pending_op && pending_op->delay_id) { crm_trace("Operation '%s'%s%s using %s was asked to run too early, " "waiting for start delay of %ds", pending_op->action, ((pending_op->target == NULL)? "" : " targeting "), pcmk__s(pending_op->target, ""), device->id, pending_op->start_delay); continue; } device->pending_ops = g_list_remove_link(device->pending_ops, gIter); g_list_free_1(gIter); cmd = pending_op; break; } if (cmd == NULL) { crm_trace("No actions using %s are needed", device->id); return TRUE; } if (pcmk__str_any_of(device->agent, STONITH_WATCHDOG_AGENT, STONITH_WATCHDOG_AGENT_INTERNAL, NULL)) { if (pcmk__is_fencing_action(cmd->action)) { if (node_does_watchdog_fencing(stonith_our_uname)) { pcmk__panic(__func__); goto done; } } else { crm_info("Faking success for %s watchdog operation", cmd->action); report_internal_result(cmd, CRM_EX_OK, PCMK_EXEC_DONE, NULL); goto done; } } #if SUPPORT_CIBSECRETS exec_rc = pcmk__substitute_secrets(device->id, device->params); if (exec_rc != pcmk_rc_ok) { if (pcmk__str_eq(cmd->action, PCMK_ACTION_STOP, pcmk__str_none)) { crm_info("Proceeding with stop operation for %s " "despite being unable to load CIB secrets (%s)", device->id, pcmk_rc_str(exec_rc)); } else { crm_err("Considering %s unconfigured " "because unable to load CIB secrets: %s", device->id, pcmk_rc_str(exec_rc)); report_internal_result(cmd, CRM_EX_ERROR, PCMK_EXEC_NO_SECRETS, "Failed to get CIB secrets"); goto done; } } #endif action_str = cmd->action; if (pcmk__str_eq(cmd->action, PCMK_ACTION_REBOOT, pcmk__str_none) && !pcmk_is_set(device->flags, st_device_supports_reboot)) { crm_notice("Remapping 'reboot' action%s%s using %s to 'off' " "because agent '%s' does not support reboot", ((cmd->target == NULL)? "" : " targeting "), pcmk__s(cmd->target, ""), device->id, device->agent); action_str = PCMK_ACTION_OFF; } if (pcmk_is_set(device->flags, st_device_supports_parameter_port)) { host_arg = "port"; } else if (pcmk_is_set(device->flags, st_device_supports_parameter_plug)) { host_arg = "plug"; } action = stonith__action_create(device->agent, action_str, cmd->target, cmd->target_nodeid, cmd->timeout, device->params, device->aliases, host_arg); /* for async exec, exec_rc is negative for early error exit otherwise handling of success/errors is done via callbacks */ cmd->activating_on = device; exec_rc = stonith__execute_async(action, (void *)cmd, cmd->done_cb, fork_cb); if (exec_rc < 0) { cmd->activating_on = NULL; cmd->done_cb(0, stonith__action_result(action), cmd); stonith__destroy_action(action); } done: /* Device might get triggered to work by multiple fencing commands * simultaneously. Trigger the device again to make sure any * remaining concurrent commands get executed. */ if (device->pending_ops) { mainloop_set_trigger(device->work); } return TRUE; } static gboolean stonith_device_dispatch(gpointer user_data) { return stonith_device_execute(user_data); } static gboolean start_delay_helper(gpointer data) { async_command_t *cmd = data; stonith_device_t *device = cmd_device(cmd); cmd->delay_id = 0; if (device) { mainloop_set_trigger(device->work); } return FALSE; } static void schedule_stonith_command(async_command_t * cmd, stonith_device_t * device) { int delay_max = 0; int delay_base = 0; int requested_delay = cmd->start_delay; CRM_CHECK(cmd != NULL, return); CRM_CHECK(device != NULL, return); if (cmd->device) { free(cmd->device); } if (device->include_nodeid && (cmd->target != NULL)) { pcmk__node_status_t *node = pcmk__get_node(0, cmd->target, NULL, pcmk__node_search_cluster_member); cmd->target_nodeid = node->cluster_layer_id; } cmd->device = pcmk__str_copy(device->id); cmd->timeout = get_action_timeout(device, cmd->action, cmd->default_timeout); if (cmd->remote_op_id) { crm_debug("Scheduling '%s' action%s%s using %s for remote peer %s " "with op id %.8s and timeout %ds", cmd->action, (cmd->target == NULL)? "" : " targeting ", pcmk__s(cmd->target, ""), device->id, cmd->origin, cmd->remote_op_id, cmd->timeout); } else { crm_debug("Scheduling '%s' action%s%s using %s for %s with timeout %ds", cmd->action, (cmd->target == NULL)? "" : " targeting ", pcmk__s(cmd->target, ""), device->id, cmd->client, cmd->timeout); } device->pending_ops = g_list_append(device->pending_ops, cmd); mainloop_set_trigger(device->work); // Value -1 means disable any static/random fencing delays if (requested_delay < 0) { return; } delay_max = get_action_delay_max(device, cmd->action); delay_base = get_action_delay_base(device, cmd->action, cmd->target); if (delay_max == 0) { delay_max = delay_base; } if (delay_max < delay_base) { crm_warn(PCMK_STONITH_DELAY_BASE " (%ds) is larger than " PCMK_STONITH_DELAY_MAX " (%ds) for %s using %s " "(limiting to maximum delay)", delay_base, delay_max, cmd->action, device->id); delay_base = delay_max; } if (delay_max > 0) { // coverity[dontcall] It doesn't matter here if rand() is predictable cmd->start_delay += ((delay_max != delay_base)?(rand() % (delay_max - delay_base)):0) + delay_base; } if (cmd->start_delay > 0) { crm_notice("Delaying '%s' action%s%s using %s for %ds " QB_XS " timeout=%ds requested_delay=%ds base=%ds max=%ds", cmd->action, (cmd->target == NULL)? "" : " targeting ", pcmk__s(cmd->target, ""), device->id, cmd->start_delay, cmd->timeout, requested_delay, delay_base, delay_max); cmd->delay_id = g_timeout_add_seconds(cmd->start_delay, start_delay_helper, cmd); } } static void free_device(gpointer data) { GList *gIter = NULL; stonith_device_t *device = data; g_hash_table_destroy(device->params); g_hash_table_destroy(device->aliases); for (gIter = device->pending_ops; gIter != NULL; gIter = gIter->next) { async_command_t *cmd = gIter->data; crm_warn("Removal of device '%s' purged operation '%s'", device->id, cmd->action); report_internal_result(cmd, CRM_EX_ERROR, PCMK_EXEC_NO_FENCE_DEVICE, "Device was removed before action could be executed"); } g_list_free(device->pending_ops); g_list_free_full(device->targets, free); if (device->timer) { mainloop_timer_stop(device->timer); mainloop_timer_del(device->timer); } mainloop_destroy_trigger(device->work); pcmk__xml_free(device->agent_metadata); free(device->namespace); if (device->on_target_actions != NULL) { g_string_free(device->on_target_actions, TRUE); } free(device->agent); free(device->id); free(device); } void free_device_list(void) { if (device_list != NULL) { g_hash_table_destroy(device_list); device_list = NULL; } } void init_device_list(void) { if (device_list == NULL) { device_list = pcmk__strkey_table(NULL, free_device); } } static GHashTable * build_port_aliases(const char *hostmap, GList ** targets) { char *name = NULL; int last = 0, lpc = 0, max = 0, added = 0; GHashTable *aliases = pcmk__strikey_table(free, free); if (hostmap == NULL) { return aliases; } max = strlen(hostmap); for (; lpc <= max; lpc++) { switch (hostmap[lpc]) { /* Skip escaped chars */ case '\\': lpc++; break; /* Assignment chars */ case '=': case ':': if (lpc > last) { free(name); name = pcmk__assert_alloc(1, 1 + lpc - last); memcpy(name, hostmap + last, lpc - last); } last = lpc + 1; break; /* Delimeter chars */ /* case ',': Potentially used to specify multiple ports */ case 0: case ';': case ' ': case '\t': if (name) { char *value = NULL; int k = 0; value = pcmk__assert_alloc(1, 1 + lpc - last); memcpy(value, hostmap + last, lpc - last); for (int i = 0; value[i] != '\0'; i++) { if (value[i] != '\\') { value[k++] = value[i]; } } value[k] = '\0'; crm_debug("Adding alias '%s'='%s'", name, value); g_hash_table_replace(aliases, name, value); if (targets) { *targets = g_list_append(*targets, pcmk__str_copy(value)); } value = NULL; name = NULL; added++; } else if (lpc > last) { crm_debug("Parse error at offset %d near '%s'", lpc - last, hostmap + last); } last = lpc + 1; break; } if (hostmap[lpc] == 0) { break; } } if (added == 0) { crm_info("No host mappings detected in '%s'", hostmap); } free(name); return aliases; } GHashTable *metadata_cache = NULL; void free_metadata_cache(void) { if (metadata_cache != NULL) { g_hash_table_destroy(metadata_cache); metadata_cache = NULL; } } static void init_metadata_cache(void) { if (metadata_cache == NULL) { metadata_cache = pcmk__strkey_table(free, free); } } int get_agent_metadata(const char *agent, xmlNode ** metadata) { char *buffer = NULL; if (metadata == NULL) { return EINVAL; } *metadata = NULL; if (pcmk__str_eq(agent, STONITH_WATCHDOG_AGENT_INTERNAL, pcmk__str_none)) { return pcmk_rc_ok; } init_metadata_cache(); buffer = g_hash_table_lookup(metadata_cache, agent); if (buffer == NULL) { stonith_t *st = stonith_api_new(); int rc; if (st == NULL) { crm_warn("Could not get agent meta-data: " "API memory allocation failed"); return EAGAIN; } rc = st->cmds->metadata(st, st_opt_sync_call, agent, NULL, &buffer, 10); stonith_api_delete(st); if (rc || !buffer) { crm_err("Could not retrieve metadata for fencing agent %s", agent); return EAGAIN; } g_hash_table_replace(metadata_cache, pcmk__str_copy(agent), buffer); } *metadata = pcmk__xml_parse(buffer); return pcmk_rc_ok; } static gboolean is_nodeid_required(xmlNode * xml) { xmlXPathObjectPtr xpath = NULL; if (stand_alone) { return FALSE; } if (!xml) { return FALSE; } xpath = xpath_search(xml, "//" PCMK_XE_PARAMETER "[@" PCMK_XA_NAME "='nodeid']"); if (numXpathResults(xpath) <= 0) { freeXpathObject(xpath); return FALSE; } freeXpathObject(xpath); return TRUE; } static void read_action_metadata(stonith_device_t *device) { xmlXPathObjectPtr xpath = NULL; int max = 0; int lpc = 0; if (device->agent_metadata == NULL) { return; } xpath = xpath_search(device->agent_metadata, "//action"); max = numXpathResults(xpath); if (max <= 0) { freeXpathObject(xpath); return; } for (lpc = 0; lpc < max; lpc++) { const char *action = NULL; xmlNode *match = getXpathResult(xpath, lpc); CRM_LOG_ASSERT(match != NULL); if(match == NULL) { continue; }; action = crm_element_value(match, PCMK_XA_NAME); if (pcmk__str_eq(action, PCMK_ACTION_LIST, pcmk__str_none)) { stonith__set_device_flags(device->flags, device->id, st_device_supports_list); } else if (pcmk__str_eq(action, PCMK_ACTION_STATUS, pcmk__str_none)) { stonith__set_device_flags(device->flags, device->id, st_device_supports_status); } else if (pcmk__str_eq(action, PCMK_ACTION_REBOOT, pcmk__str_none)) { stonith__set_device_flags(device->flags, device->id, st_device_supports_reboot); } else if (pcmk__str_eq(action, PCMK_ACTION_ON, pcmk__str_none)) { /* PCMK_XA_AUTOMATIC means the cluster will unfence a node when it * joins. * * @COMPAT PCMK__XA_REQUIRED is a deprecated synonym for * PCMK_XA_AUTOMATIC. */ if (pcmk__xe_attr_is_true(match, PCMK_XA_AUTOMATIC) || pcmk__xe_attr_is_true(match, PCMK__XA_REQUIRED)) { device->automatic_unfencing = TRUE; } stonith__set_device_flags(device->flags, device->id, st_device_supports_on); } if ((action != NULL) && pcmk__xe_attr_is_true(match, PCMK_XA_ON_TARGET)) { pcmk__add_word(&(device->on_target_actions), 64, action); } } freeXpathObject(xpath); } static const char * target_list_type(stonith_device_t * dev) { const char *check_type = NULL; check_type = g_hash_table_lookup(dev->params, PCMK_STONITH_HOST_CHECK); if (check_type == NULL) { if (g_hash_table_lookup(dev->params, PCMK_STONITH_HOST_LIST)) { check_type = PCMK_VALUE_STATIC_LIST; } else if (g_hash_table_lookup(dev->params, PCMK_STONITH_HOST_MAP)) { check_type = PCMK_VALUE_STATIC_LIST; } else if (pcmk_is_set(dev->flags, st_device_supports_list)) { check_type = PCMK_VALUE_DYNAMIC_LIST; } else if (pcmk_is_set(dev->flags, st_device_supports_status)) { check_type = PCMK_VALUE_STATUS; } else { check_type = PCMK_VALUE_NONE; } } return check_type; } static stonith_device_t * build_device_from_xml(xmlNode *dev) { const char *value; stonith_device_t *device = NULL; char *agent = crm_element_value_copy(dev, PCMK_XA_AGENT); CRM_CHECK(agent != NULL, return device); device = pcmk__assert_alloc(1, sizeof(stonith_device_t)); device->id = crm_element_value_copy(dev, PCMK_XA_ID); device->agent = agent; device->namespace = crm_element_value_copy(dev, PCMK__XA_NAMESPACE); device->params = xml2list(dev); value = g_hash_table_lookup(device->params, PCMK_STONITH_HOST_LIST); if (value) { device->targets = stonith__parse_targets(value); } value = g_hash_table_lookup(device->params, PCMK_STONITH_HOST_MAP); device->aliases = build_port_aliases(value, &(device->targets)); value = target_list_type(device); if (!pcmk__str_eq(value, PCMK_VALUE_STATIC_LIST, pcmk__str_casei) && (device->targets != NULL)) { // device->targets is necessary only with PCMK_VALUE_STATIC_LIST g_list_free_full(device->targets, free); device->targets = NULL; } switch (get_agent_metadata(device->agent, &device->agent_metadata)) { case pcmk_rc_ok: if (device->agent_metadata) { read_action_metadata(device); stonith__device_parameter_flags(&(device->flags), device->id, device->agent_metadata); } break; case EAGAIN: if (device->timer == NULL) { device->timer = mainloop_timer_add("get_agent_metadata", 10 * 1000, TRUE, get_agent_metadata_cb, device); } if (!mainloop_timer_running(device->timer)) { mainloop_timer_start(device->timer); } break; default: break; } value = g_hash_table_lookup(device->params, "nodeid"); if (!value) { device->include_nodeid = is_nodeid_required(device->agent_metadata); } value = crm_element_value(dev, PCMK__XA_RSC_PROVIDES); if (pcmk__str_eq(value, PCMK_VALUE_UNFENCING, pcmk__str_casei)) { device->automatic_unfencing = TRUE; } if (is_action_required(PCMK_ACTION_ON, device)) { crm_info("Fencing device '%s' requires unfencing", device->id); } if (device->on_target_actions != NULL) { crm_info("Fencing device '%s' requires actions (%s) to be executed " "on target", device->id, (const char *) device->on_target_actions->str); } device->work = mainloop_add_trigger(G_PRIORITY_HIGH, stonith_device_dispatch, device); /* TODO: Hook up priority */ return device; } static void schedule_internal_command(const char *origin, stonith_device_t * device, const char *action, const char *target, int timeout, void *internal_user_data, void (*done_cb) (int pid, const pcmk__action_result_t *result, void *user_data)) { async_command_t *cmd = NULL; cmd = pcmk__assert_alloc(1, sizeof(async_command_t)); cmd->id = -1; cmd->default_timeout = timeout ? timeout : 60; cmd->timeout = cmd->default_timeout; cmd->action = pcmk__str_copy(action); cmd->target = pcmk__str_copy(target); cmd->device = pcmk__str_copy(device->id); cmd->origin = pcmk__str_copy(origin); cmd->client = pcmk__str_copy(crm_system_name); cmd->client_name = pcmk__str_copy(crm_system_name); cmd->internal_user_data = internal_user_data; cmd->done_cb = done_cb; /* cmd, not internal_user_data, is passed to 'done_cb' as the userdata */ schedule_stonith_command(cmd, device); } // Fence agent status commands use custom exit status codes enum fence_status_code { fence_status_invalid = -1, fence_status_active = 0, fence_status_unknown = 1, fence_status_inactive = 2, }; static void status_search_cb(int pid, const pcmk__action_result_t *result, void *user_data) { async_command_t *cmd = user_data; struct device_search_s *search = cmd->internal_user_data; stonith_device_t *dev = cmd_device(cmd); gboolean can = FALSE; free_async_command(cmd); if (!dev) { search_devices_record_result(search, NULL, FALSE); return; } mainloop_set_trigger(dev->work); if (result->execution_status != PCMK_EXEC_DONE) { crm_warn("Assuming %s cannot fence %s " "because status could not be executed: %s%s%s%s", dev->id, search->host, pcmk_exec_status_str(result->execution_status), ((result->exit_reason == NULL)? "" : " ("), ((result->exit_reason == NULL)? "" : result->exit_reason), ((result->exit_reason == NULL)? "" : ")")); search_devices_record_result(search, dev->id, FALSE); return; } switch (result->exit_status) { case fence_status_unknown: crm_trace("%s reported it cannot fence %s", dev->id, search->host); break; case fence_status_active: case fence_status_inactive: crm_trace("%s reported it can fence %s", dev->id, search->host); can = TRUE; break; default: crm_warn("Assuming %s cannot fence %s " "(status returned unknown code %d)", dev->id, search->host, result->exit_status); break; } search_devices_record_result(search, dev->id, can); } static void dynamic_list_search_cb(int pid, const pcmk__action_result_t *result, void *user_data) { async_command_t *cmd = user_data; struct device_search_s *search = cmd->internal_user_data; stonith_device_t *dev = cmd_device(cmd); gboolean can_fence = FALSE; free_async_command(cmd); /* Host/alias must be in the list output to be eligible to be fenced * * Will cause problems if down'd nodes aren't listed or (for virtual nodes) * if the guest is still listed despite being moved to another machine */ if (!dev) { search_devices_record_result(search, NULL, FALSE); return; } mainloop_set_trigger(dev->work); if (pcmk__result_ok(result)) { crm_info("Refreshing target list for %s", dev->id); g_list_free_full(dev->targets, free); dev->targets = stonith__parse_targets(result->action_stdout); dev->targets_age = time(NULL); } else if (dev->targets != NULL) { if (result->execution_status == PCMK_EXEC_DONE) { crm_info("Reusing most recent target list for %s " "because list returned error code %d", dev->id, result->exit_status); } else { crm_info("Reusing most recent target list for %s " "because list could not be executed: %s%s%s%s", dev->id, pcmk_exec_status_str(result->execution_status), ((result->exit_reason == NULL)? "" : " ("), ((result->exit_reason == NULL)? "" : result->exit_reason), ((result->exit_reason == NULL)? "" : ")")); } } else { // We have never successfully executed list if (result->execution_status == PCMK_EXEC_DONE) { crm_warn("Assuming %s cannot fence %s " "because list returned error code %d", dev->id, search->host, result->exit_status); } else { crm_warn("Assuming %s cannot fence %s " "because list could not be executed: %s%s%s%s", dev->id, search->host, pcmk_exec_status_str(result->execution_status), ((result->exit_reason == NULL)? "" : " ("), ((result->exit_reason == NULL)? "" : result->exit_reason), ((result->exit_reason == NULL)? "" : ")")); } /* Fall back to pcmk_host_check=PCMK_VALUE_STATUS if the user didn't * explicitly specify PCMK_VALUE_DYNAMIC_LIST */ if (g_hash_table_lookup(dev->params, PCMK_STONITH_HOST_CHECK) == NULL) { crm_notice("Switching to pcmk_host_check='status' for %s", dev->id); pcmk__insert_dup(dev->params, PCMK_STONITH_HOST_CHECK, PCMK_VALUE_STATUS); } } if (dev->targets) { const char *alias = g_hash_table_lookup(dev->aliases, search->host); if (!alias) { alias = search->host; } if (pcmk__str_in_list(alias, dev->targets, pcmk__str_casei)) { can_fence = TRUE; } } search_devices_record_result(search, dev->id, can_fence); } /*! * \internal * \brief Returns true if any key in first is not in second or second has a different value for key */ static int device_params_diff(GHashTable *first, GHashTable *second) { char *key = NULL; char *value = NULL; GHashTableIter gIter; g_hash_table_iter_init(&gIter, first); while (g_hash_table_iter_next(&gIter, (void **)&key, (void **)&value)) { if(strstr(key, "CRM_meta") == key) { continue; } else if (strcmp(key, PCMK_XA_CRM_FEATURE_SET) == 0) { continue; } else { char *other_value = g_hash_table_lookup(second, key); if (!other_value || !pcmk__str_eq(other_value, value, pcmk__str_casei)) { crm_trace("Different value for %s: %s != %s", key, other_value, value); return 1; } } } return 0; } /*! * \internal * \brief Checks to see if an identical device already exists in the device_list */ static stonith_device_t * device_has_duplicate(const stonith_device_t *device) { stonith_device_t *dup = g_hash_table_lookup(device_list, device->id); if (!dup) { crm_trace("No match for %s", device->id); return NULL; } else if (!pcmk__str_eq(dup->agent, device->agent, pcmk__str_casei)) { crm_trace("Different agent: %s != %s", dup->agent, device->agent); return NULL; } // Use pcmk__digest_operation() here? if (device_params_diff(device->params, dup->params) || device_params_diff(dup->params, device->params)) { return NULL; } crm_trace("Match"); return dup; } int stonith_device_register(xmlNode *dev, gboolean from_cib) { stonith_device_t *dup = NULL; stonith_device_t *device = build_device_from_xml(dev); guint ndevices = 0; int rv = pcmk_ok; CRM_CHECK(device != NULL, return -ENOMEM); /* do we have a watchdog-device? */ if (pcmk__str_eq(device->id, STONITH_WATCHDOG_ID, pcmk__str_none) || pcmk__str_any_of(device->agent, STONITH_WATCHDOG_AGENT, STONITH_WATCHDOG_AGENT_INTERNAL, NULL)) do { if (stonith_watchdog_timeout_ms <= 0) { crm_err("Ignoring watchdog fence device without " PCMK_OPT_STONITH_WATCHDOG_TIMEOUT " set."); rv = -ENODEV; /* fall through to cleanup & return */ } else if (!pcmk__str_any_of(device->agent, STONITH_WATCHDOG_AGENT, STONITH_WATCHDOG_AGENT_INTERNAL, NULL)) { crm_err("Ignoring watchdog fence device with unknown " "agent '%s' unequal '" STONITH_WATCHDOG_AGENT "'.", device->agent?device->agent:""); rv = -ENODEV; /* fall through to cleanup & return */ } else if (!pcmk__str_eq(device->id, STONITH_WATCHDOG_ID, pcmk__str_none)) { crm_err("Ignoring watchdog fence device " "named %s !='"STONITH_WATCHDOG_ID"'.", device->id?device->id:""); rv = -ENODEV; /* fall through to cleanup & return */ } else { if (pcmk__str_eq(device->agent, STONITH_WATCHDOG_AGENT, pcmk__str_none)) { /* this either has an empty list or the targets configured for watchdog-fencing */ g_list_free_full(stonith_watchdog_targets, free); stonith_watchdog_targets = device->targets; device->targets = NULL; } if (node_does_watchdog_fencing(stonith_our_uname)) { g_list_free_full(device->targets, free); device->targets = stonith__parse_targets(stonith_our_uname); pcmk__insert_dup(device->params, PCMK_STONITH_HOST_LIST, stonith_our_uname); /* proceed as with any other stonith-device */ break; } crm_debug("Skip registration of watchdog fence device on node not in host-list."); /* cleanup and fall through to more cleanup and return */ device->targets = NULL; stonith_device_remove(device->id, from_cib); } free_device(device); return rv; } while (0); dup = device_has_duplicate(device); if (dup) { ndevices = g_hash_table_size(device_list); crm_debug("Device '%s' already in device list (%d active device%s)", device->id, ndevices, pcmk__plural_s(ndevices)); free_device(device); device = dup; dup = g_hash_table_lookup(device_list, device->id); dup->dirty = FALSE; } else { stonith_device_t *old = g_hash_table_lookup(device_list, device->id); if (from_cib && old && old->api_registered) { /* If the cib is writing over an entry that is shared with a stonith client, * copy any pending ops that currently exist on the old entry to the new one. * Otherwise the pending ops will be reported as failures */ crm_info("Overwriting existing entry for %s from CIB", device->id); device->pending_ops = old->pending_ops; device->api_registered = TRUE; old->pending_ops = NULL; if (device->pending_ops) { mainloop_set_trigger(device->work); } } g_hash_table_replace(device_list, device->id, device); ndevices = g_hash_table_size(device_list); crm_notice("Added '%s' to device list (%d active device%s)", device->id, ndevices, pcmk__plural_s(ndevices)); } if (from_cib) { device->cib_registered = TRUE; } else { device->api_registered = TRUE; } return pcmk_ok; } void stonith_device_remove(const char *id, bool from_cib) { stonith_device_t *device = g_hash_table_lookup(device_list, id); guint ndevices = 0; if (!device) { ndevices = g_hash_table_size(device_list); crm_info("Device '%s' not found (%d active device%s)", id, ndevices, pcmk__plural_s(ndevices)); return; } if (from_cib) { device->cib_registered = FALSE; } else { device->verified = FALSE; device->api_registered = FALSE; } if (!device->cib_registered && !device->api_registered) { g_hash_table_remove(device_list, id); ndevices = g_hash_table_size(device_list); crm_info("Removed '%s' from device list (%d active device%s)", id, ndevices, pcmk__plural_s(ndevices)); } else { crm_trace("Not removing '%s' from device list (%d active) because " "still registered via:%s%s", id, g_hash_table_size(device_list), (device->cib_registered? " cib" : ""), (device->api_registered? " api" : "")); } } /*! * \internal * \brief Return the number of stonith levels registered for a node * * \param[in] tp Node's topology table entry * * \return Number of non-NULL levels in topology entry * \note This function is used only for log messages. */ static int count_active_levels(const stonith_topology_t *tp) { int lpc = 0; int count = 0; for (lpc = 0; lpc < ST_LEVEL_MAX; lpc++) { if (tp->levels[lpc] != NULL) { count++; } } return count; } static void free_topology_entry(gpointer data) { stonith_topology_t *tp = data; int lpc = 0; for (lpc = 0; lpc < ST_LEVEL_MAX; lpc++) { if (tp->levels[lpc] != NULL) { g_list_free_full(tp->levels[lpc], free); } } free(tp->target); free(tp->target_value); free(tp->target_pattern); free(tp->target_attribute); free(tp); } void free_topology_list(void) { if (topology != NULL) { g_hash_table_destroy(topology); topology = NULL; } } void init_topology_list(void) { if (topology == NULL) { topology = pcmk__strkey_table(NULL, free_topology_entry); } } char * stonith_level_key(const xmlNode *level, enum fenced_target_by mode) { if (mode == fenced_target_by_unknown) { mode = unpack_level_kind(level); } switch (mode) { case fenced_target_by_name: return crm_element_value_copy(level, PCMK_XA_TARGET); case fenced_target_by_pattern: return crm_element_value_copy(level, PCMK_XA_TARGET_PATTERN); case fenced_target_by_attribute: return crm_strdup_printf("%s=%s", crm_element_value(level, PCMK_XA_TARGET_ATTRIBUTE), crm_element_value(level, PCMK_XA_TARGET_VALUE)); default: return crm_strdup_printf("unknown-%s", pcmk__xe_id(level)); } } /*! * \internal * \brief Parse target identification from topology level XML * * \param[in] level Topology level XML to parse * * \return How to identify target of \p level */ static enum fenced_target_by unpack_level_kind(const xmlNode *level) { if (crm_element_value(level, PCMK_XA_TARGET) != NULL) { return fenced_target_by_name; } if (crm_element_value(level, PCMK_XA_TARGET_PATTERN) != NULL) { return fenced_target_by_pattern; } if (!stand_alone /* if standalone, there's no attribute manager */ && (crm_element_value(level, PCMK_XA_TARGET_ATTRIBUTE) != NULL) && (crm_element_value(level, PCMK_XA_TARGET_VALUE) != NULL)) { return fenced_target_by_attribute; } return fenced_target_by_unknown; } static stonith_key_value_t * parse_device_list(const char *devices) { int lpc = 0; int max = 0; int last = 0; stonith_key_value_t *output = NULL; if (devices == NULL) { return output; } max = strlen(devices); for (lpc = 0; lpc <= max; lpc++) { if (devices[lpc] == ',' || devices[lpc] == 0) { char *line = strndup(devices + last, lpc - last); output = stonith_key_value_add(output, NULL, line); free(line); last = lpc + 1; } } return output; } /*! * \internal * \brief Unpack essential information from topology request XML * * \param[in] xml Request XML to search * \param[out] mode If not NULL, where to store level kind * \param[out] target If not NULL, where to store representation of target * \param[out] id If not NULL, where to store level number * \param[out] desc If not NULL, where to store log-friendly level description * * \return Topology level XML from within \p xml, or NULL if not found * \note The caller is responsible for freeing \p *target and \p *desc if set. */ static xmlNode * unpack_level_request(xmlNode *xml, enum fenced_target_by *mode, char **target, int *id, char **desc) { enum fenced_target_by local_mode = fenced_target_by_unknown; char *local_target = NULL; int local_id = 0; /* The level element can be the top element or lower. If top level, don't * search by xpath, because it might give multiple hits if the XML is the * CIB. */ if ((xml != NULL) && !pcmk__xe_is(xml, PCMK_XE_FENCING_LEVEL)) { xml = get_xpath_object("//" PCMK_XE_FENCING_LEVEL, xml, LOG_WARNING); } if (xml == NULL) { if (desc != NULL) { *desc = crm_strdup_printf("missing"); } } else { local_mode = unpack_level_kind(xml); local_target = stonith_level_key(xml, local_mode); crm_element_value_int(xml, PCMK_XA_INDEX, &local_id); if (desc != NULL) { *desc = crm_strdup_printf("%s[%d]", local_target, local_id); } } if (mode != NULL) { *mode = local_mode; } if (id != NULL) { *id = local_id; } if (target != NULL) { *target = local_target; } else { free(local_target); } return xml; } /*! * \internal * \brief Register a fencing topology level for a target * * Given an XML request specifying the target name, level index, and device IDs * for the level, this will create an entry for the target in the global topology * table if one does not already exist, then append the specified device IDs to * the entry's device list for the specified level. * * \param[in] msg XML request for STONITH level registration * \param[out] desc If not NULL, set to string representation "TARGET[LEVEL]" * \param[out] result Where to set result of registration */ void fenced_register_level(xmlNode *msg, char **desc, pcmk__action_result_t *result) { int id = 0; xmlNode *level; enum fenced_target_by mode; char *target; stonith_topology_t *tp; stonith_key_value_t *dIter = NULL; stonith_key_value_t *devices = NULL; CRM_CHECK((msg != NULL) && (result != NULL), return); level = unpack_level_request(msg, &mode, &target, &id, desc); if (level == NULL) { fenced_set_protocol_error(result); return; } // Ensure an ID was given (even the client API adds an ID) if (pcmk__str_empty(pcmk__xe_id(level))) { crm_warn("Ignoring registration for topology level without ID"); free(target); crm_log_xml_trace(level, "Bad level"); pcmk__format_result(result, CRM_EX_INVALID_PARAM, PCMK_EXEC_INVALID, "Topology level is invalid without ID"); return; } // Ensure a valid target was specified if (mode == fenced_target_by_unknown) { crm_warn("Ignoring registration for topology level '%s' " "without valid target", pcmk__xe_id(level)); free(target); crm_log_xml_trace(level, "Bad level"); pcmk__format_result(result, CRM_EX_INVALID_PARAM, PCMK_EXEC_INVALID, "Invalid target for topology level '%s'", pcmk__xe_id(level)); return; } // Ensure level ID is in allowed range if ((id <= 0) || (id >= ST_LEVEL_MAX)) { crm_warn("Ignoring topology registration for %s with invalid level %d", target, id); free(target); crm_log_xml_trace(level, "Bad level"); pcmk__format_result(result, CRM_EX_INVALID_PARAM, PCMK_EXEC_INVALID, "Invalid level number '%s' for topology level '%s'", pcmk__s(crm_element_value(level, PCMK_XA_INDEX), ""), pcmk__xe_id(level)); return; } /* Find or create topology table entry */ tp = g_hash_table_lookup(topology, target); if (tp == NULL) { tp = pcmk__assert_alloc(1, sizeof(stonith_topology_t)); tp->kind = mode; tp->target = target; tp->target_value = crm_element_value_copy(level, PCMK_XA_TARGET_VALUE); tp->target_pattern = crm_element_value_copy(level, PCMK_XA_TARGET_PATTERN); tp->target_attribute = crm_element_value_copy(level, PCMK_XA_TARGET_ATTRIBUTE); g_hash_table_replace(topology, tp->target, tp); crm_trace("Added %s (%d) to the topology (%d active entries)", target, (int) mode, g_hash_table_size(topology)); } else { free(target); } if (tp->levels[id] != NULL) { crm_info("Adding to the existing %s[%d] topology entry", tp->target, id); } devices = parse_device_list(crm_element_value(level, PCMK_XA_DEVICES)); for (dIter = devices; dIter; dIter = dIter->next) { const char *device = dIter->value; crm_trace("Adding device '%s' for %s[%d]", device, tp->target, id); tp->levels[id] = g_list_append(tp->levels[id], pcmk__str_copy(device)); } stonith_key_value_freeall(devices, 1, 1); { int nlevels = count_active_levels(tp); crm_info("Target %s has %d active fencing level%s", tp->target, nlevels, pcmk__plural_s(nlevels)); } pcmk__set_result(result, CRM_EX_OK, PCMK_EXEC_DONE, NULL); } /*! * \internal * \brief Unregister a fencing topology level for a target * * Given an XML request specifying the target name and level index (or 0 for all * levels), this will remove any corresponding entry for the target from the * global topology table. * * \param[in] msg XML request for STONITH level registration * \param[out] desc If not NULL, set to string representation "TARGET[LEVEL]" * \param[out] result Where to set result of unregistration */ void fenced_unregister_level(xmlNode *msg, char **desc, pcmk__action_result_t *result) { int id = -1; stonith_topology_t *tp; char *target; xmlNode *level = NULL; CRM_CHECK(result != NULL, return); level = unpack_level_request(msg, NULL, &target, &id, desc); if (level == NULL) { fenced_set_protocol_error(result); return; } // Ensure level ID is in allowed range if ((id < 0) || (id >= ST_LEVEL_MAX)) { crm_warn("Ignoring topology unregistration for %s with invalid level %d", target, id); free(target); crm_log_xml_trace(level, "Bad level"); pcmk__format_result(result, CRM_EX_INVALID_PARAM, PCMK_EXEC_INVALID, "Invalid level number '%s' for topology level %s", pcmk__s(crm_element_value(level, PCMK_XA_INDEX), ""), // Client API doesn't add ID to unregistration XML pcmk__s(pcmk__xe_id(level), "")); return; } tp = g_hash_table_lookup(topology, target); if (tp == NULL) { guint nentries = g_hash_table_size(topology); crm_info("No fencing topology found for %s (%d active %s)", target, nentries, pcmk__plural_alt(nentries, "entry", "entries")); } else if (id == 0 && g_hash_table_remove(topology, target)) { guint nentries = g_hash_table_size(topology); crm_info("Removed all fencing topology entries related to %s " "(%d active %s remaining)", target, nentries, pcmk__plural_alt(nentries, "entry", "entries")); } else if (tp->levels[id] != NULL) { guint nlevels; g_list_free_full(tp->levels[id], free); tp->levels[id] = NULL; nlevels = count_active_levels(tp); crm_info("Removed level %d from fencing topology for %s " "(%d active level%s remaining)", id, target, nlevels, pcmk__plural_s(nlevels)); } free(target); pcmk__set_result(result, CRM_EX_OK, PCMK_EXEC_DONE, NULL); } static char * list_to_string(GList *list, const char *delim, gboolean terminate_with_delim) { int max = g_list_length(list); size_t delim_len = delim?strlen(delim):0; size_t alloc_size = 1 + (max?((max-1+(terminate_with_delim?1:0))*delim_len):0); char *rv; GList *gIter; char *pos = NULL; const char *lead_delim = ""; for (gIter = list; gIter != NULL; gIter = gIter->next) { const char *value = (const char *) gIter->data; alloc_size += strlen(value); } rv = pcmk__assert_alloc(alloc_size, sizeof(char)); pos = rv; for (gIter = list; gIter != NULL; gIter = gIter->next) { const char *value = (const char *) gIter->data; pos = &pos[sprintf(pos, "%s%s", lead_delim, value)]; lead_delim = delim; } if (max && terminate_with_delim) { sprintf(pos, "%s", delim); } return rv; } /*! * \internal * \brief Execute a fence agent action directly (and asynchronously) * * Handle a STONITH_OP_EXEC API message by scheduling a requested agent action * directly on a specified device. Only list, monitor, and status actions are * expected to use this call, though it should work with any agent command. * * \param[in] msg Request XML specifying action * \param[out] result Where to store result of action * * \note If the action is monitor, the device must be registered via the API * (CIB registration is not sufficient), because monitor should not be * possible unless the device is "started" (API registered). */ static void execute_agent_action(xmlNode *msg, pcmk__action_result_t *result) { xmlNode *dev = get_xpath_object("//" PCMK__XE_ST_DEVICE_ID, msg, LOG_ERR); xmlNode *op = get_xpath_object("//@" PCMK__XE_ST_DEVICE_ACTION, msg, LOG_ERR); const char *id = crm_element_value(dev, PCMK__XA_ST_DEVICE_ID); const char *action = crm_element_value(op, PCMK__XA_ST_DEVICE_ACTION); async_command_t *cmd = NULL; stonith_device_t *device = NULL; if ((id == NULL) || (action == NULL)) { crm_info("Malformed API action request: device %s, action %s", (id? id : "not specified"), (action? action : "not specified")); fenced_set_protocol_error(result); return; } if (pcmk__str_eq(id, STONITH_WATCHDOG_ID, pcmk__str_none)) { // Watchdog agent actions are implemented internally if (stonith_watchdog_timeout_ms <= 0) { pcmk__set_result(result, CRM_EX_ERROR, PCMK_EXEC_NO_FENCE_DEVICE, "Watchdog fence device not configured"); return; } else if (pcmk__str_eq(action, PCMK_ACTION_LIST, pcmk__str_none)) { pcmk__set_result(result, CRM_EX_OK, PCMK_EXEC_DONE, NULL); pcmk__set_result_output(result, list_to_string(stonith_watchdog_targets, "\n", TRUE), NULL); return; } else if (pcmk__str_eq(action, PCMK_ACTION_MONITOR, pcmk__str_none)) { pcmk__set_result(result, CRM_EX_OK, PCMK_EXEC_DONE, NULL); return; } } device = g_hash_table_lookup(device_list, id); if (device == NULL) { crm_info("Ignoring API '%s' action request because device %s not found", action, id); pcmk__format_result(result, CRM_EX_ERROR, PCMK_EXEC_NO_FENCE_DEVICE, "'%s' not found", id); return; } else if (!device->api_registered && (strcmp(action, PCMK_ACTION_MONITOR) == 0)) { // Monitors may run only on "started" (API-registered) devices crm_info("Ignoring API '%s' action request because device %s not active", action, id); pcmk__format_result(result, CRM_EX_ERROR, PCMK_EXEC_NO_FENCE_DEVICE, "'%s' not active", id); return; } cmd = create_async_command(msg); if (cmd == NULL) { crm_log_xml_warn(msg, "invalid"); fenced_set_protocol_error(result); return; } schedule_stonith_command(cmd, device); pcmk__set_result(result, CRM_EX_OK, PCMK_EXEC_PENDING, NULL); } static void search_devices_record_result(struct device_search_s *search, const char *device, gboolean can_fence) { search->replies_received++; if (can_fence && device) { if (search->support_action_only != st_device_supports_none) { stonith_device_t *dev = g_hash_table_lookup(device_list, device); if (dev && !pcmk_is_set(dev->flags, search->support_action_only)) { return; } } search->capable = g_list_append(search->capable, pcmk__str_copy(device)); } if (search->replies_needed == search->replies_received) { guint ndevices = g_list_length(search->capable); crm_debug("Search found %d device%s that can perform '%s' targeting %s", ndevices, pcmk__plural_s(ndevices), (search->action? search->action : "unknown action"), (search->host? search->host : "any node")); search->callback(search->capable, search->user_data); free(search->host); free(search->action); free(search); } } /*! * \internal * \brief Check whether the local host is allowed to execute a fencing action * * \param[in] device Fence device to check * \param[in] action Fence action to check * \param[in] target Hostname of fence target * \param[in] allow_suicide Whether self-fencing is allowed for this operation * * \return TRUE if local host is allowed to execute action, FALSE otherwise */ static gboolean localhost_is_eligible(const stonith_device_t *device, const char *action, const char *target, gboolean allow_suicide) { gboolean localhost_is_target = pcmk__str_eq(target, stonith_our_uname, pcmk__str_casei); if ((device != NULL) && (action != NULL) && (device->on_target_actions != NULL) && (strstr((const char*) device->on_target_actions->str, action) != NULL)) { if (!localhost_is_target) { crm_trace("Operation '%s' using %s can only be executed for local " "host, not %s", action, device->id, target); return FALSE; } } else if (localhost_is_target && !allow_suicide) { crm_trace("'%s' operation does not support self-fencing", action); return FALSE; } return TRUE; } /*! * \internal * \brief Check if local node is allowed to execute (possibly remapped) action * * \param[in] device Fence device to check * \param[in] action Fence action to check * \param[in] target Node name of fence target * \param[in] allow_self Whether self-fencing is allowed for this operation * * \return true if local node is allowed to execute \p action or any actions it * might be remapped to, otherwise false */ static bool localhost_is_eligible_with_remap(const stonith_device_t *device, const char *action, const char *target, gboolean allow_self) { // Check exact action if (localhost_is_eligible(device, action, target, allow_self)) { return true; } // Check potential remaps if (pcmk__str_eq(action, PCMK_ACTION_REBOOT, pcmk__str_none)) { /* "reboot" might get remapped to "off" then "on", so even if reboot is * disallowed, return true if either of those is allowed. We'll report * the disallowed actions with the results. We never allow self-fencing * for remapped "on" actions because the target is off at that point. */ if (localhost_is_eligible(device, PCMK_ACTION_OFF, target, allow_self) || localhost_is_eligible(device, PCMK_ACTION_ON, target, FALSE)) { return true; } } return false; } static void can_fence_host_with_device(stonith_device_t *dev, struct device_search_s *search) { gboolean can = FALSE; const char *check_type = "Internal bug"; const char *target = NULL; const char *alias = NULL; const char *dev_id = "Unspecified device"; const char *action = (search == NULL)? NULL : search->action; CRM_CHECK((dev != NULL) && (action != NULL), goto search_report_results); if (dev->id != NULL) { dev_id = dev->id; } target = search->host; if (target == NULL) { can = TRUE; check_type = "No target"; goto search_report_results; } /* Answer immediately if the device does not support the action * or the local node is not allowed to perform it */ if (pcmk__str_eq(action, PCMK_ACTION_ON, pcmk__str_none) && !pcmk_is_set(dev->flags, st_device_supports_on)) { check_type = "Agent does not support 'on'"; goto search_report_results; } else if (!localhost_is_eligible_with_remap(dev, action, target, search->allow_suicide)) { check_type = "This node is not allowed to execute action"; goto search_report_results; } // Check eligibility as specified by pcmk_host_check check_type = target_list_type(dev); alias = g_hash_table_lookup(dev->aliases, target); if (pcmk__str_eq(check_type, PCMK_VALUE_NONE, pcmk__str_casei)) { can = TRUE; } else if (pcmk__str_eq(check_type, PCMK_VALUE_STATIC_LIST, pcmk__str_casei)) { if (pcmk__str_in_list(target, dev->targets, pcmk__str_casei)) { can = TRUE; } else if (g_hash_table_lookup(dev->params, PCMK_STONITH_HOST_MAP) && g_hash_table_lookup(dev->aliases, target)) { can = TRUE; } } else if (pcmk__str_eq(check_type, PCMK_VALUE_DYNAMIC_LIST, pcmk__str_casei)) { time_t now = time(NULL); if (dev->targets == NULL || dev->targets_age + 60 < now) { int device_timeout = get_action_timeout(dev, PCMK_ACTION_LIST, search->per_device_timeout); if (device_timeout > search->per_device_timeout) { crm_notice("Since the pcmk_list_timeout (%ds) parameter of %s " "is larger than " PCMK_OPT_STONITH_TIMEOUT " (%ds), timeout may occur", device_timeout, dev_id, search->per_device_timeout); } crm_trace("Running '%s' to check whether %s is eligible to fence %s (%s)", check_type, dev_id, target, action); schedule_internal_command(__func__, dev, PCMK_ACTION_LIST, NULL, search->per_device_timeout, search, dynamic_list_search_cb); /* we'll respond to this search request async in the cb */ return; } if (pcmk__str_in_list(((alias == NULL)? target : alias), dev->targets, pcmk__str_casei)) { can = TRUE; } } else if (pcmk__str_eq(check_type, PCMK_VALUE_STATUS, pcmk__str_casei)) { int device_timeout = get_action_timeout(dev, check_type, search->per_device_timeout); if (device_timeout > search->per_device_timeout) { crm_notice("Since the pcmk_status_timeout (%ds) parameter of %s is " "larger than " PCMK_OPT_STONITH_TIMEOUT " (%ds), " "timeout may occur", device_timeout, dev_id, search->per_device_timeout); } crm_trace("Running '%s' to check whether %s is eligible to fence %s (%s)", check_type, dev_id, target, action); schedule_internal_command(__func__, dev, PCMK_ACTION_STATUS, target, search->per_device_timeout, search, status_search_cb); /* we'll respond to this search request async in the cb */ return; } else { crm_err("Invalid value for " PCMK_STONITH_HOST_CHECK ": %s", check_type); check_type = "Invalid " PCMK_STONITH_HOST_CHECK; } search_report_results: crm_info("%s is%s eligible to fence (%s) %s%s%s%s: %s", dev_id, (can? "" : " not"), pcmk__s(action, "unspecified action"), pcmk__s(target, "unspecified target"), (alias == NULL)? "" : " (as '", pcmk__s(alias, ""), (alias == NULL)? "" : "')", check_type); search_devices_record_result(search, ((dev == NULL)? NULL : dev_id), can); } static void search_devices(gpointer key, gpointer value, gpointer user_data) { stonith_device_t *dev = value; struct device_search_s *search = user_data; can_fence_host_with_device(dev, search); } #define DEFAULT_QUERY_TIMEOUT 20 static void get_capable_devices(const char *host, const char *action, int timeout, bool suicide, void *user_data, void (*callback) (GList * devices, void *user_data), uint32_t support_action_only) { struct device_search_s *search; guint ndevices = g_hash_table_size(device_list); if (ndevices == 0) { callback(NULL, user_data); return; } search = pcmk__assert_alloc(1, sizeof(struct device_search_s)); search->host = pcmk__str_copy(host); search->action = pcmk__str_copy(action); search->per_device_timeout = timeout; search->allow_suicide = suicide; search->callback = callback; search->user_data = user_data; search->support_action_only = support_action_only; /* We are guaranteed this many replies, even if a device is * unregistered while the search is in progress. */ search->replies_needed = ndevices; crm_debug("Searching %d device%s to see which can execute '%s' targeting %s", ndevices, pcmk__plural_s(ndevices), (search->action? search->action : "unknown action"), (search->host? search->host : "any node")); g_hash_table_foreach(device_list, search_devices, search); } struct st_query_data { xmlNode *reply; char *remote_peer; char *client_id; char *target; char *action; int call_options; }; /*! * \internal * \brief Add action-specific attributes to query reply XML * * \param[in,out] xml XML to add attributes to * \param[in] action Fence action * \param[in] device Fence device * \param[in] target Fence target */ static void add_action_specific_attributes(xmlNode *xml, const char *action, const stonith_device_t *device, const char *target) { int action_specific_timeout; int delay_max; int delay_base; CRM_CHECK(xml && action && device, return); // PCMK__XA_ST_REQUIRED is currently used only for unfencing if (is_action_required(action, device)) { crm_trace("Action '%s' is required using %s", action, device->id); crm_xml_add_int(xml, PCMK__XA_ST_REQUIRED, 1); } // pcmk__timeout if configured action_specific_timeout = get_action_timeout(device, action, 0); if (action_specific_timeout) { crm_trace("Action '%s' has timeout %ds using %s", action, action_specific_timeout, device->id); crm_xml_add_int(xml, PCMK__XA_ST_ACTION_TIMEOUT, action_specific_timeout); } delay_max = get_action_delay_max(device, action); if (delay_max > 0) { crm_trace("Action '%s' has maximum random delay %ds using %s", action, delay_max, device->id); crm_xml_add_int(xml, PCMK__XA_ST_DELAY_MAX, delay_max); } delay_base = get_action_delay_base(device, action, target); if (delay_base > 0) { crm_xml_add_int(xml, PCMK__XA_ST_DELAY_BASE, delay_base); } if ((delay_max > 0) && (delay_base == 0)) { crm_trace("Action '%s' has maximum random delay %ds using %s", action, delay_max, device->id); } else if ((delay_max == 0) && (delay_base > 0)) { crm_trace("Action '%s' has a static delay of %ds using %s", action, delay_base, device->id); } else if ((delay_max > 0) && (delay_base > 0)) { crm_trace("Action '%s' has a minimum delay of %ds and a randomly chosen " "maximum delay of %ds using %s", action, delay_base, delay_max, device->id); } } /*! * \internal * \brief Add "disallowed" attribute to query reply XML if appropriate * * \param[in,out] xml XML to add attribute to * \param[in] action Fence action * \param[in] device Fence device * \param[in] target Fence target * \param[in] allow_suicide Whether self-fencing is allowed */ static void add_disallowed(xmlNode *xml, const char *action, const stonith_device_t *device, const char *target, gboolean allow_suicide) { if (!localhost_is_eligible(device, action, target, allow_suicide)) { crm_trace("Action '%s' using %s is disallowed for local host", action, device->id); pcmk__xe_set_bool_attr(xml, PCMK__XA_ST_ACTION_DISALLOWED, true); } } /*! * \internal * \brief Add child element with action-specific values to query reply XML * * \param[in,out] xml XML to add attribute to * \param[in] action Fence action * \param[in] device Fence device * \param[in] target Fence target * \param[in] allow_suicide Whether self-fencing is allowed */ static void add_action_reply(xmlNode *xml, const char *action, const stonith_device_t *device, const char *target, gboolean allow_suicide) { xmlNode *child = pcmk__xe_create(xml, PCMK__XE_ST_DEVICE_ACTION); crm_xml_add(child, PCMK_XA_ID, action); add_action_specific_attributes(child, action, device, target); add_disallowed(child, action, device, target, allow_suicide); } /*! * \internal * \brief Send a reply to a CPG peer or IPC client * * \param[in] reply XML reply to send * \param[in] call_options Send synchronously if st_opt_sync_call is set * \param[in] remote_peer If not NULL, name of peer node to send CPG reply * \param[in,out] client If not NULL, client to send IPC reply */ static void stonith_send_reply(const xmlNode *reply, int call_options, const char *remote_peer, pcmk__client_t *client) { CRM_CHECK((reply != NULL) && ((remote_peer != NULL) || (client != NULL)), return); if (remote_peer == NULL) { do_local_reply(reply, client, call_options); } else { const pcmk__node_status_t *node = pcmk__get_node(0, remote_peer, NULL, pcmk__node_search_cluster_member); - pcmk__cluster_send_message(node, crm_msg_stonith_ng, reply); + pcmk__cluster_send_message(node, pcmk__cluster_msg_fenced, reply); } } static void stonith_query_capable_device_cb(GList * devices, void *user_data) { struct st_query_data *query = user_data; int available_devices = 0; xmlNode *wrapper = NULL; xmlNode *list = NULL; GList *lpc = NULL; pcmk__client_t *client = NULL; if (query->client_id != NULL) { client = pcmk__find_client_by_id(query->client_id); if ((client == NULL) && (query->remote_peer == NULL)) { crm_trace("Skipping reply to %s: no longer a client", query->client_id); goto done; } } // Pack the results into XML wrapper = pcmk__xe_create(query->reply, PCMK__XE_ST_CALLDATA); list = pcmk__xe_create(wrapper, __func__); crm_xml_add(list, PCMK__XA_ST_TARGET, query->target); for (lpc = devices; lpc != NULL; lpc = lpc->next) { stonith_device_t *device = g_hash_table_lookup(device_list, lpc->data); const char *action = query->action; xmlNode *dev = NULL; if (!device) { /* It is possible the device got unregistered while * determining who can fence the target */ continue; } available_devices++; dev = pcmk__xe_create(list, PCMK__XE_ST_DEVICE_ID); crm_xml_add(dev, PCMK_XA_ID, device->id); crm_xml_add(dev, PCMK__XA_NAMESPACE, device->namespace); crm_xml_add(dev, PCMK_XA_AGENT, device->agent); // Has had successful monitor, list, or status on this node crm_xml_add_int(dev, PCMK__XA_ST_MONITOR_VERIFIED, device->verified); crm_xml_add_int(dev, PCMK__XA_ST_DEVICE_SUPPORT_FLAGS, device->flags); /* If the originating fencer wants to reboot the node, and we have a * capable device that doesn't support "reboot", remap to "off" instead. */ if (!pcmk_is_set(device->flags, st_device_supports_reboot) && pcmk__str_eq(query->action, PCMK_ACTION_REBOOT, pcmk__str_none)) { crm_trace("%s doesn't support reboot, using values for off instead", device->id); action = PCMK_ACTION_OFF; } /* Add action-specific values if available */ add_action_specific_attributes(dev, action, device, query->target); if (pcmk__str_eq(query->action, PCMK_ACTION_REBOOT, pcmk__str_none)) { /* A "reboot" *might* get remapped to "off" then "on", so after * sending the "reboot"-specific values in the main element, we add * sub-elements for "off" and "on" values. * * We short-circuited earlier if "reboot", "off" and "on" are all * disallowed for the local host. However if only one or two are * disallowed, we send back the results and mark which ones are * disallowed. If "reboot" is disallowed, this might cause problems * with older fencer versions, which won't check for it. Older * versions will ignore "off" and "on", so they are not a problem. */ add_disallowed(dev, action, device, query->target, pcmk_is_set(query->call_options, st_opt_allow_suicide)); add_action_reply(dev, PCMK_ACTION_OFF, device, query->target, pcmk_is_set(query->call_options, st_opt_allow_suicide)); add_action_reply(dev, PCMK_ACTION_ON, device, query->target, FALSE); } /* A query without a target wants device parameters */ if (query->target == NULL) { xmlNode *attrs = pcmk__xe_create(dev, PCMK__XE_ATTRIBUTES); g_hash_table_foreach(device->params, hash2field, attrs); } } crm_xml_add_int(list, PCMK__XA_ST_AVAILABLE_DEVICES, available_devices); if (query->target) { crm_debug("Found %d matching device%s for target '%s'", available_devices, pcmk__plural_s(available_devices), query->target); } else { crm_debug("%d device%s installed", available_devices, pcmk__plural_s(available_devices)); } crm_log_xml_trace(list, "query-result"); stonith_send_reply(query->reply, query->call_options, query->remote_peer, client); done: pcmk__xml_free(query->reply); free(query->remote_peer); free(query->client_id); free(query->target); free(query->action); free(query); g_list_free_full(devices, free); } /*! * \internal * \brief Log the result of an asynchronous command * * \param[in] cmd Command the result is for * \param[in] result Result of command * \param[in] pid Process ID of command, if available * \param[in] next Alternate device that will be tried if command failed * \param[in] op_merged Whether this command was merged with an earlier one */ static void log_async_result(const async_command_t *cmd, const pcmk__action_result_t *result, int pid, const char *next, bool op_merged) { int log_level = LOG_ERR; int output_log_level = LOG_NEVER; guint devices_remaining = g_list_length(cmd->next_device_iter); GString *msg = g_string_sized_new(80); // Reasonable starting size // Choose log levels appropriately if we have a result if (pcmk__result_ok(result)) { log_level = (cmd->target == NULL)? LOG_DEBUG : LOG_NOTICE; if ((result->action_stdout != NULL) && !pcmk__str_eq(cmd->action, PCMK_ACTION_METADATA, pcmk__str_none)) { output_log_level = LOG_DEBUG; } next = NULL; } else { log_level = (cmd->target == NULL)? LOG_NOTICE : LOG_ERR; if ((result->action_stdout != NULL) && !pcmk__str_eq(cmd->action, PCMK_ACTION_METADATA, pcmk__str_none)) { output_log_level = LOG_WARNING; } } // Build the log message piece by piece pcmk__g_strcat(msg, "Operation '", cmd->action, "' ", NULL); if (pid != 0) { g_string_append_printf(msg, "[%d] ", pid); } if (cmd->target != NULL) { pcmk__g_strcat(msg, "targeting ", cmd->target, " ", NULL); } if (cmd->device != NULL) { pcmk__g_strcat(msg, "using ", cmd->device, " ", NULL); } // Add exit status or execution status as appropriate if (result->execution_status == PCMK_EXEC_DONE) { g_string_append_printf(msg, "returned %d", result->exit_status); } else { pcmk__g_strcat(msg, "could not be executed: ", pcmk_exec_status_str(result->execution_status), NULL); } // Add exit reason and next device if appropriate if (result->exit_reason != NULL) { pcmk__g_strcat(msg, " (", result->exit_reason, ")", NULL); } if (next != NULL) { pcmk__g_strcat(msg, ", retrying with ", next, NULL); } if (devices_remaining > 0) { g_string_append_printf(msg, " (%u device%s remaining)", (unsigned int) devices_remaining, pcmk__plural_s(devices_remaining)); } g_string_append_printf(msg, " " QB_XS " %scall %d from %s", (op_merged? "merged " : ""), cmd->id, cmd->client_name); // Log the result do_crm_log(log_level, "%s", msg->str); g_string_free(msg, TRUE); // Log the output (which may have multiple lines), if appropriate if (output_log_level != LOG_NEVER) { char *prefix = crm_strdup_printf("%s[%d]", cmd->device, pid); crm_log_output(output_log_level, prefix, result->action_stdout); free(prefix); } } /*! * \internal * \brief Reply to requester after asynchronous command completion * * \param[in] cmd Command that completed * \param[in] result Result of command * \param[in] pid Process ID of command, if available * \param[in] merged If true, command was merged with another, not executed */ static void send_async_reply(const async_command_t *cmd, const pcmk__action_result_t *result, int pid, bool merged) { xmlNode *reply = NULL; pcmk__client_t *client = NULL; CRM_CHECK((cmd != NULL) && (result != NULL), return); log_async_result(cmd, result, pid, NULL, merged); if (cmd->client != NULL) { client = pcmk__find_client_by_id(cmd->client); if ((client == NULL) && (cmd->origin == NULL)) { crm_trace("Skipping reply to %s: no longer a client", cmd->client); return; } } reply = construct_async_reply(cmd, result); if (merged) { pcmk__xe_set_bool_attr(reply, PCMK__XA_ST_OP_MERGED, true); } if (!stand_alone && pcmk__is_fencing_action(cmd->action) && pcmk__str_eq(cmd->origin, cmd->target, pcmk__str_casei)) { /* The target was also the originator, so broadcast the result on its * behalf (since it will be unable to). */ crm_trace("Broadcast '%s' result for %s (target was also originator)", cmd->action, cmd->target); crm_xml_add(reply, PCMK__XA_SUBT, PCMK__VALUE_BROADCAST); crm_xml_add(reply, PCMK__XA_ST_OP, STONITH_OP_NOTIFY); - pcmk__cluster_send_message(NULL, crm_msg_stonith_ng, reply); + pcmk__cluster_send_message(NULL, pcmk__cluster_msg_fenced, reply); } else { // Reply only to the originator stonith_send_reply(reply, cmd->options, cmd->origin, client); } crm_log_xml_trace(reply, "Reply"); pcmk__xml_free(reply); if (stand_alone) { /* Do notification with a clean data object */ xmlNode *notify_data = pcmk__xe_create(NULL, PCMK__XE_ST_NOTIFY_FENCE); stonith__xe_set_result(notify_data, result); crm_xml_add(notify_data, PCMK__XA_ST_TARGET, cmd->target); crm_xml_add(notify_data, PCMK__XA_ST_OP, cmd->op); crm_xml_add(notify_data, PCMK__XA_ST_DELEGATE, "localhost"); crm_xml_add(notify_data, PCMK__XA_ST_DEVICE_ID, cmd->device); crm_xml_add(notify_data, PCMK__XA_ST_REMOTE_OP, cmd->remote_op_id); crm_xml_add(notify_data, PCMK__XA_ST_ORIGIN, cmd->client); fenced_send_notification(PCMK__VALUE_ST_NOTIFY_FENCE, result, notify_data); fenced_send_notification(PCMK__VALUE_ST_NOTIFY_HISTORY, NULL, NULL); } } static void cancel_stonith_command(async_command_t * cmd) { stonith_device_t *device = cmd_device(cmd); if (device) { crm_trace("Cancel scheduled '%s' action using %s", cmd->action, device->id); device->pending_ops = g_list_remove(device->pending_ops, cmd); } } /*! * \internal * \brief Cancel and reply to any duplicates of a just-completed operation * * Check whether any fencing operations are scheduled to do the same thing as * one that just succeeded. If so, rather than performing the same operation * twice, return the result of this operation for all matching pending commands. * * \param[in,out] cmd Fencing operation that just succeeded * \param[in] result Result of \p cmd * \param[in] pid If nonzero, process ID of agent invocation (for logs) * * \note Duplicate merging will do the right thing for either type of remapped * reboot. If the executing fencer remapped an unsupported reboot to off, * then cmd->action will be "reboot" and will be merged with any other * reboot requests. If the originating fencer remapped a topology reboot * to off then on, we will get here once with cmd->action "off" and once * with "on", and they will be merged separately with similar requests. */ static void reply_to_duplicates(async_command_t *cmd, const pcmk__action_result_t *result, int pid) { GList *next = NULL; for (GList *iter = cmd_list; iter != NULL; iter = next) { async_command_t *cmd_other = iter->data; next = iter->next; // We might delete this entry, so grab next now if (cmd == cmd_other) { continue; } /* A pending operation matches if: * 1. The client connections are different. * 2. The target is the same. * 3. The fencing action is the same. * 4. The device scheduled to execute the action is the same. */ if (pcmk__str_eq(cmd->client, cmd_other->client, pcmk__str_casei) || !pcmk__str_eq(cmd->target, cmd_other->target, pcmk__str_casei) || !pcmk__str_eq(cmd->action, cmd_other->action, pcmk__str_none) || !pcmk__str_eq(cmd->device, cmd_other->device, pcmk__str_casei)) { continue; } crm_notice("Merging fencing action '%s'%s%s originating from " "client %s with identical fencing request from client %s", cmd_other->action, (cmd_other->target == NULL)? "" : " targeting ", pcmk__s(cmd_other->target, ""), cmd_other->client_name, cmd->client_name); // Stop tracking the duplicate, send its result, and cancel it cmd_list = g_list_remove_link(cmd_list, iter); send_async_reply(cmd_other, result, pid, true); cancel_stonith_command(cmd_other); free_async_command(cmd_other); g_list_free_1(iter); } } /*! * \internal * \brief Return the next required device (if any) for an operation * * \param[in,out] cmd Fencing operation that just succeeded * * \return Next device required for action if any, otherwise NULL */ static stonith_device_t * next_required_device(async_command_t *cmd) { for (GList *iter = cmd->next_device_iter; iter != NULL; iter = iter->next) { stonith_device_t *next_device = g_hash_table_lookup(device_list, iter->data); if (is_action_required(cmd->action, next_device)) { /* This is only called for successful actions, so it's OK to skip * non-required devices. */ cmd->next_device_iter = iter->next; return next_device; } } return NULL; } static void st_child_done(int pid, const pcmk__action_result_t *result, void *user_data) { async_command_t *cmd = user_data; stonith_device_t *device = NULL; stonith_device_t *next_device = NULL; CRM_CHECK(cmd != NULL, return); device = cmd_device(cmd); cmd->active_on = NULL; /* The device is ready to do something else now */ if (device) { if (!device->verified && pcmk__result_ok(result) && pcmk__strcase_any_of(cmd->action, PCMK_ACTION_LIST, PCMK_ACTION_MONITOR, PCMK_ACTION_STATUS, NULL)) { device->verified = TRUE; } mainloop_set_trigger(device->work); } if (pcmk__result_ok(result)) { next_device = next_required_device(cmd); } else if ((cmd->next_device_iter != NULL) && !is_action_required(cmd->action, device)) { /* if this device didn't work out, see if there are any others we can try. * if the failed device was 'required', we can't pick another device. */ next_device = g_hash_table_lookup(device_list, cmd->next_device_iter->data); cmd->next_device_iter = cmd->next_device_iter->next; } if (next_device == NULL) { send_async_reply(cmd, result, pid, false); if (pcmk__result_ok(result)) { reply_to_duplicates(cmd, result, pid); } free_async_command(cmd); } else { // This operation requires more fencing log_async_result(cmd, result, pid, next_device->id, false); schedule_stonith_command(cmd, next_device); } } static gint sort_device_priority(gconstpointer a, gconstpointer b) { const stonith_device_t *dev_a = a; const stonith_device_t *dev_b = b; if (dev_a->priority > dev_b->priority) { return -1; } else if (dev_a->priority < dev_b->priority) { return 1; } return 0; } static void stonith_fence_get_devices_cb(GList * devices, void *user_data) { async_command_t *cmd = user_data; stonith_device_t *device = NULL; guint ndevices = g_list_length(devices); crm_info("Found %d matching device%s for target '%s'", ndevices, pcmk__plural_s(ndevices), cmd->target); if (devices != NULL) { /* Order based on priority */ devices = g_list_sort(devices, sort_device_priority); device = g_hash_table_lookup(device_list, devices->data); } if (device == NULL) { // No device found pcmk__action_result_t result = PCMK__UNKNOWN_RESULT; pcmk__format_result(&result, CRM_EX_ERROR, PCMK_EXEC_NO_FENCE_DEVICE, "No device configured for target '%s'", cmd->target); send_async_reply(cmd, &result, 0, false); pcmk__reset_result(&result); free_async_command(cmd); g_list_free_full(devices, free); } else { // Device found, schedule it for fencing cmd->device_list = devices; cmd->next_device_iter = devices->next; schedule_stonith_command(cmd, device); } } /*! * \internal * \brief Execute a fence action via the local node * * \param[in] msg Fencing request * \param[out] result Where to store result of fence action */ static void fence_locally(xmlNode *msg, pcmk__action_result_t *result) { const char *device_id = NULL; stonith_device_t *device = NULL; async_command_t *cmd = NULL; xmlNode *dev = NULL; CRM_CHECK((msg != NULL) && (result != NULL), return); dev = get_xpath_object("//@" PCMK__XA_ST_TARGET, msg, LOG_ERR); cmd = create_async_command(msg); if (cmd == NULL) { crm_log_xml_warn(msg, "invalid"); fenced_set_protocol_error(result); return; } device_id = crm_element_value(dev, PCMK__XA_ST_DEVICE_ID); if (device_id != NULL) { device = g_hash_table_lookup(device_list, device_id); if (device == NULL) { crm_err("Requested device '%s' is not available", device_id); pcmk__format_result(result, CRM_EX_ERROR, PCMK_EXEC_NO_FENCE_DEVICE, "Requested device '%s' not found", device_id); return; } schedule_stonith_command(cmd, device); } else { const char *host = crm_element_value(dev, PCMK__XA_ST_TARGET); if (pcmk_is_set(cmd->options, st_opt_cs_nodeid)) { int nodeid = 0; pcmk__node_status_t *node = NULL; pcmk__scan_min_int(host, &nodeid, 0); node = pcmk__search_node_caches(nodeid, NULL, pcmk__node_search_any |pcmk__node_search_cluster_cib); if (node != NULL) { host = node->name; } } /* If we get to here, then self-fencing is implicitly allowed */ get_capable_devices(host, cmd->action, cmd->default_timeout, TRUE, cmd, stonith_fence_get_devices_cb, fenced_support_flag(cmd->action)); } pcmk__set_result(result, CRM_EX_OK, PCMK_EXEC_PENDING, NULL); } /*! * \internal * \brief Build an XML reply for a fencing operation * * \param[in] request Request that reply is for * \param[in] data If not NULL, add to reply as call data * \param[in] result Full result of fencing operation * * \return Newly created XML reply * \note The caller is responsible for freeing the result. * \note This has some overlap with construct_async_reply(), but that copies * values from an async_command_t, whereas this one copies them from the * request. */ xmlNode * fenced_construct_reply(const xmlNode *request, xmlNode *data, const pcmk__action_result_t *result) { xmlNode *reply = NULL; reply = pcmk__xe_create(NULL, PCMK__XE_ST_REPLY); crm_xml_add(reply, PCMK__XA_ST_ORIGIN, __func__); crm_xml_add(reply, PCMK__XA_T, PCMK__VALUE_STONITH_NG); stonith__xe_set_result(reply, result); if (request == NULL) { /* Most likely, this is the result of a stonith operation that was * initiated before we came up. Unfortunately that means we lack enough * information to provide clients with a full result. * * @TODO Maybe synchronize this information at start-up? */ crm_warn("Missing request information for client notifications for " "operation with result '%s' (initiated before we came up?)", pcmk_exec_status_str(result->execution_status)); } else { const char *name = NULL; const char *value = NULL; // Attributes to copy from request to reply const char *names[] = { PCMK__XA_ST_OP, PCMK__XA_ST_CALLID, PCMK__XA_ST_CLIENTID, PCMK__XA_ST_CLIENTNAME, PCMK__XA_ST_REMOTE_OP, PCMK__XA_ST_CALLOPT, }; for (int lpc = 0; lpc < PCMK__NELEM(names); lpc++) { name = names[lpc]; value = crm_element_value(request, name); crm_xml_add(reply, name, value); } if (data != NULL) { xmlNode *wrapper = pcmk__xe_create(reply, PCMK__XE_ST_CALLDATA); pcmk__xml_copy(wrapper, data); } } return reply; } /*! * \internal * \brief Build an XML reply to an asynchronous fencing command * * \param[in] cmd Fencing command that reply is for * \param[in] result Command result */ static xmlNode * construct_async_reply(const async_command_t *cmd, const pcmk__action_result_t *result) { xmlNode *reply = pcmk__xe_create(NULL, PCMK__XE_ST_REPLY); crm_xml_add(reply, PCMK__XA_ST_ORIGIN, __func__); crm_xml_add(reply, PCMK__XA_T, PCMK__VALUE_STONITH_NG); crm_xml_add(reply, PCMK__XA_ST_OP, cmd->op); crm_xml_add(reply, PCMK__XA_ST_DEVICE_ID, cmd->device); crm_xml_add(reply, PCMK__XA_ST_REMOTE_OP, cmd->remote_op_id); crm_xml_add(reply, PCMK__XA_ST_CLIENTID, cmd->client); crm_xml_add(reply, PCMK__XA_ST_CLIENTNAME, cmd->client_name); crm_xml_add(reply, PCMK__XA_ST_TARGET, cmd->target); crm_xml_add(reply, PCMK__XA_ST_DEVICE_ACTION, cmd->op); crm_xml_add(reply, PCMK__XA_ST_ORIGIN, cmd->origin); crm_xml_add_int(reply, PCMK__XA_ST_CALLID, cmd->id); crm_xml_add_int(reply, PCMK__XA_ST_CALLOPT, cmd->options); stonith__xe_set_result(reply, result); return reply; } bool fencing_peer_active(pcmk__node_status_t *peer) { return (peer != NULL) && (peer->name != NULL) && pcmk_is_set(peer->processes, crm_get_cluster_proc()); } void set_fencing_completed(remote_fencing_op_t *op) { struct timespec tv; qb_util_timespec_from_epoch_get(&tv); op->completed = tv.tv_sec; op->completed_nsec = tv.tv_nsec; } /*! * \internal * \brief Look for alternate node needed if local node shouldn't fence target * * \param[in] target Node that must be fenced * * \return Name of an alternate node that should fence \p target if any, * or NULL otherwise */ static const char * check_alternate_host(const char *target) { if (pcmk__str_eq(target, stonith_our_uname, pcmk__str_casei)) { GHashTableIter gIter; pcmk__node_status_t *entry = NULL; g_hash_table_iter_init(&gIter, crm_peer_cache); while (g_hash_table_iter_next(&gIter, NULL, (void **)&entry)) { if (fencing_peer_active(entry) && !pcmk__str_eq(entry->name, target, pcmk__str_casei)) { crm_notice("Forwarding self-fencing request to %s", entry->name); return entry->name; } } crm_warn("Will handle own fencing because no peer can"); } return NULL; } static void remove_relay_op(xmlNode * request) { xmlNode *dev = get_xpath_object("//@" PCMK__XE_ST_DEVICE_ACTION, request, LOG_TRACE); const char *relay_op_id = NULL; const char *op_id = NULL; const char *client_name = NULL; const char *target = NULL; remote_fencing_op_t *relay_op = NULL; if (dev) { target = crm_element_value(dev, PCMK__XA_ST_TARGET); } relay_op_id = crm_element_value(request, PCMK__XA_ST_REMOTE_OP_RELAY); op_id = crm_element_value(request, PCMK__XA_ST_REMOTE_OP); client_name = crm_element_value(request, PCMK__XA_ST_CLIENTNAME); /* Delete RELAY operation. */ if (relay_op_id && target && pcmk__str_eq(target, stonith_our_uname, pcmk__str_casei)) { relay_op = g_hash_table_lookup(stonith_remote_op_list, relay_op_id); if (relay_op) { GHashTableIter iter; remote_fencing_op_t *list_op = NULL; g_hash_table_iter_init(&iter, stonith_remote_op_list); /* If the operation to be deleted is registered as a duplicate, delete the registration. */ while (g_hash_table_iter_next(&iter, NULL, (void **)&list_op)) { GList *dup_iter = NULL; if (list_op != relay_op) { for (dup_iter = list_op->duplicates; dup_iter != NULL; dup_iter = dup_iter->next) { remote_fencing_op_t *other = dup_iter->data; if (other == relay_op) { other->duplicates = g_list_remove(other->duplicates, relay_op); break; } } } } crm_debug("Deleting relay op %s ('%s'%s%s for %s), " "replaced by op %s ('%s'%s%s for %s)", relay_op->id, relay_op->action, (relay_op->target == NULL)? "" : " targeting ", pcmk__s(relay_op->target, ""), relay_op->client_name, op_id, relay_op->action, (target == NULL)? "" : " targeting ", pcmk__s(target, ""), client_name); g_hash_table_remove(stonith_remote_op_list, relay_op_id); } } } /*! * \internal * \brief Check whether an API request was sent by a privileged user * * API commands related to fencing configuration may be done only by privileged * IPC users (i.e. root or hacluster), because all other users should go through * the CIB to have ACLs applied. If no client was given, this is a peer request, * which is always allowed. * * \param[in] c IPC client that sent request (or NULL if sent by CPG peer) * \param[in] op Requested API operation (for logging only) * * \return true if sender is peer or privileged client, otherwise false */ static inline bool is_privileged(const pcmk__client_t *c, const char *op) { if ((c == NULL) || pcmk_is_set(c->flags, pcmk__client_privileged)) { return true; } else { crm_warn("Rejecting IPC request '%s' from unprivileged client %s", pcmk__s(op, ""), pcmk__client_name(c)); return false; } } // CRM_OP_REGISTER static xmlNode * handle_register_request(pcmk__request_t *request) { xmlNode *reply = pcmk__xe_create(NULL, "reply"); CRM_ASSERT(request->ipc_client != NULL); crm_xml_add(reply, PCMK__XA_ST_OP, CRM_OP_REGISTER); crm_xml_add(reply, PCMK__XA_ST_CLIENTID, request->ipc_client->id); pcmk__set_result(&request->result, CRM_EX_OK, PCMK_EXEC_DONE, NULL); pcmk__set_request_flags(request, pcmk__request_reuse_options); return reply; } // STONITH_OP_EXEC static xmlNode * handle_agent_request(pcmk__request_t *request) { execute_agent_action(request->xml, &request->result); if (request->result.execution_status == PCMK_EXEC_PENDING) { return NULL; } return fenced_construct_reply(request->xml, NULL, &request->result); } // STONITH_OP_TIMEOUT_UPDATE static xmlNode * handle_update_timeout_request(pcmk__request_t *request) { const char *call_id = crm_element_value(request->xml, PCMK__XA_ST_CALLID); const char *client_id = crm_element_value(request->xml, PCMK__XA_ST_CLIENTID); int op_timeout = 0; crm_element_value_int(request->xml, PCMK__XA_ST_TIMEOUT, &op_timeout); do_stonith_async_timeout_update(client_id, call_id, op_timeout); pcmk__set_result(&request->result, CRM_EX_OK, PCMK_EXEC_DONE, NULL); return NULL; } // STONITH_OP_QUERY static xmlNode * handle_query_request(pcmk__request_t *request) { int timeout = 0; xmlNode *dev = NULL; const char *action = NULL; const char *target = NULL; const char *client_id = crm_element_value(request->xml, PCMK__XA_ST_CLIENTID); struct st_query_data *query = NULL; if (request->peer != NULL) { // Record it for the future notification create_remote_stonith_op(client_id, request->xml, TRUE); } /* Delete the DC node RELAY operation. */ remove_relay_op(request->xml); pcmk__set_result(&request->result, CRM_EX_OK, PCMK_EXEC_DONE, NULL); dev = get_xpath_object("//@" PCMK__XE_ST_DEVICE_ACTION, request->xml, LOG_NEVER); if (dev != NULL) { const char *device = crm_element_value(dev, PCMK__XA_ST_DEVICE_ID); if (pcmk__str_eq(device, "manual_ack", pcmk__str_casei)) { return NULL; // No query or reply necessary } target = crm_element_value(dev, PCMK__XA_ST_TARGET); action = crm_element_value(dev, PCMK__XA_ST_DEVICE_ACTION); } crm_log_xml_trace(request->xml, "Query"); query = pcmk__assert_alloc(1, sizeof(struct st_query_data)); query->reply = fenced_construct_reply(request->xml, NULL, &request->result); query->remote_peer = pcmk__str_copy(request->peer); query->client_id = pcmk__str_copy(client_id); query->target = pcmk__str_copy(target); query->action = pcmk__str_copy(action); query->call_options = request->call_options; crm_element_value_int(request->xml, PCMK__XA_ST_TIMEOUT, &timeout); get_capable_devices(target, action, timeout, pcmk_is_set(query->call_options, st_opt_allow_suicide), query, stonith_query_capable_device_cb, st_device_supports_none); return NULL; } // STONITH_OP_NOTIFY static xmlNode * handle_notify_request(pcmk__request_t *request) { const char *flag_name = NULL; CRM_ASSERT(request->ipc_client != NULL); flag_name = crm_element_value(request->xml, PCMK__XA_ST_NOTIFY_ACTIVATE); if (flag_name != NULL) { crm_debug("Enabling %s callbacks for client %s", flag_name, pcmk__request_origin(request)); pcmk__set_client_flags(request->ipc_client, get_stonith_flag(flag_name)); } flag_name = crm_element_value(request->xml, PCMK__XA_ST_NOTIFY_DEACTIVATE); if (flag_name != NULL) { crm_debug("Disabling %s callbacks for client %s", flag_name, pcmk__request_origin(request)); pcmk__clear_client_flags(request->ipc_client, get_stonith_flag(flag_name)); } pcmk__set_result(&request->result, CRM_EX_OK, PCMK_EXEC_DONE, NULL); pcmk__set_request_flags(request, pcmk__request_reuse_options); return pcmk__ipc_create_ack(request->ipc_flags, PCMK__XE_ACK, NULL, CRM_EX_OK); } // STONITH_OP_RELAY static xmlNode * handle_relay_request(pcmk__request_t *request) { xmlNode *dev = get_xpath_object("//@" PCMK__XA_ST_TARGET, request->xml, LOG_TRACE); crm_notice("Received forwarded fencing request from " "%s %s to fence (%s) peer %s", pcmk__request_origin_type(request), pcmk__request_origin(request), crm_element_value(dev, PCMK__XA_ST_DEVICE_ACTION), crm_element_value(dev, PCMK__XA_ST_TARGET)); if (initiate_remote_stonith_op(NULL, request->xml, FALSE) == NULL) { fenced_set_protocol_error(&request->result); return fenced_construct_reply(request->xml, NULL, &request->result); } pcmk__set_result(&request->result, CRM_EX_OK, PCMK_EXEC_PENDING, NULL); return NULL; } // STONITH_OP_FENCE static xmlNode * handle_fence_request(pcmk__request_t *request) { if ((request->peer != NULL) || stand_alone) { fence_locally(request->xml, &request->result); } else if (pcmk_is_set(request->call_options, st_opt_manual_ack)) { switch (fenced_handle_manual_confirmation(request->ipc_client, request->xml)) { case pcmk_rc_ok: pcmk__set_result(&request->result, CRM_EX_OK, PCMK_EXEC_DONE, NULL); break; case EINPROGRESS: pcmk__set_result(&request->result, CRM_EX_OK, PCMK_EXEC_PENDING, NULL); break; default: fenced_set_protocol_error(&request->result); break; } } else { const char *alternate_host = NULL; xmlNode *dev = get_xpath_object("//@" PCMK__XA_ST_TARGET, request->xml, LOG_TRACE); const char *target = crm_element_value(dev, PCMK__XA_ST_TARGET); const char *action = crm_element_value(dev, PCMK__XA_ST_DEVICE_ACTION); const char *device = crm_element_value(dev, PCMK__XA_ST_DEVICE_ID); if (request->ipc_client != NULL) { int tolerance = 0; crm_notice("Client %s wants to fence (%s) %s using %s", pcmk__request_origin(request), action, target, (device? device : "any device")); crm_element_value_int(dev, PCMK__XA_ST_TOLERANCE, &tolerance); if (stonith_check_fence_tolerance(tolerance, target, action)) { pcmk__set_result(&request->result, CRM_EX_OK, PCMK_EXEC_DONE, NULL); return fenced_construct_reply(request->xml, NULL, &request->result); } alternate_host = check_alternate_host(target); } else { crm_notice("Peer %s wants to fence (%s) '%s' with device '%s'", request->peer, action, target, (device == NULL)? "(any)" : device); } if (alternate_host != NULL) { const char *client_id = NULL; remote_fencing_op_t *op = NULL; pcmk__node_status_t *node = pcmk__get_node(0, alternate_host, NULL, pcmk__node_search_cluster_member); if (request->ipc_client->id == 0) { client_id = crm_element_value(request->xml, PCMK__XA_ST_CLIENTID); } else { client_id = request->ipc_client->id; } /* Create a duplicate fencing operation to relay with the client ID. * When a query response is received, this operation should be * deleted to avoid keeping the duplicate around. */ op = create_remote_stonith_op(client_id, request->xml, FALSE); crm_xml_add(request->xml, PCMK__XA_ST_OP, STONITH_OP_RELAY); crm_xml_add(request->xml, PCMK__XA_ST_CLIENTID, request->ipc_client->id); crm_xml_add(request->xml, PCMK__XA_ST_REMOTE_OP, op->id); - pcmk__cluster_send_message(node, crm_msg_stonith_ng, request->xml); + pcmk__cluster_send_message(node, pcmk__cluster_msg_fenced, + request->xml); pcmk__set_result(&request->result, CRM_EX_OK, PCMK_EXEC_PENDING, NULL); } else if (initiate_remote_stonith_op(request->ipc_client, request->xml, FALSE) == NULL) { fenced_set_protocol_error(&request->result); } else { pcmk__set_result(&request->result, CRM_EX_OK, PCMK_EXEC_PENDING, NULL); } } if (request->result.execution_status == PCMK_EXEC_PENDING) { return NULL; } return fenced_construct_reply(request->xml, NULL, &request->result); } // STONITH_OP_FENCE_HISTORY static xmlNode * handle_history_request(pcmk__request_t *request) { xmlNode *reply = NULL; xmlNode *data = NULL; stonith_fence_history(request->xml, &data, request->peer, request->call_options); pcmk__set_result(&request->result, CRM_EX_OK, PCMK_EXEC_DONE, NULL); if (!pcmk_is_set(request->call_options, st_opt_discard_reply)) { /* When the local node broadcasts its history, it sets * st_opt_discard_reply and doesn't need a reply. */ reply = fenced_construct_reply(request->xml, data, &request->result); } pcmk__xml_free(data); return reply; } // STONITH_OP_DEVICE_ADD static xmlNode * handle_device_add_request(pcmk__request_t *request) { const char *op = crm_element_value(request->xml, PCMK__XA_ST_OP); xmlNode *dev = get_xpath_object("//" PCMK__XE_ST_DEVICE_ID, request->xml, LOG_ERR); if (is_privileged(request->ipc_client, op)) { int rc = stonith_device_register(dev, FALSE); pcmk__set_result(&request->result, ((rc == pcmk_ok)? CRM_EX_OK : CRM_EX_ERROR), stonith__legacy2status(rc), ((rc == pcmk_ok)? NULL : pcmk_strerror(rc))); } else { pcmk__set_result(&request->result, CRM_EX_INSUFFICIENT_PRIV, PCMK_EXEC_INVALID, "Unprivileged users must register device via CIB"); } fenced_send_config_notification(op, &request->result, (dev == NULL)? NULL : pcmk__xe_id(dev)); return fenced_construct_reply(request->xml, NULL, &request->result); } // STONITH_OP_DEVICE_DEL static xmlNode * handle_device_delete_request(pcmk__request_t *request) { xmlNode *dev = get_xpath_object("//" PCMK__XE_ST_DEVICE_ID, request->xml, LOG_ERR); const char *device_id = crm_element_value(dev, PCMK_XA_ID); const char *op = crm_element_value(request->xml, PCMK__XA_ST_OP); if (is_privileged(request->ipc_client, op)) { stonith_device_remove(device_id, false); pcmk__set_result(&request->result, CRM_EX_OK, PCMK_EXEC_DONE, NULL); } else { pcmk__set_result(&request->result, CRM_EX_INSUFFICIENT_PRIV, PCMK_EXEC_INVALID, "Unprivileged users must delete device via CIB"); } fenced_send_config_notification(op, &request->result, device_id); return fenced_construct_reply(request->xml, NULL, &request->result); } // STONITH_OP_LEVEL_ADD static xmlNode * handle_level_add_request(pcmk__request_t *request) { char *desc = NULL; const char *op = crm_element_value(request->xml, PCMK__XA_ST_OP); if (is_privileged(request->ipc_client, op)) { fenced_register_level(request->xml, &desc, &request->result); } else { unpack_level_request(request->xml, NULL, NULL, NULL, &desc); pcmk__set_result(&request->result, CRM_EX_INSUFFICIENT_PRIV, PCMK_EXEC_INVALID, "Unprivileged users must add level via CIB"); } fenced_send_config_notification(op, &request->result, desc); free(desc); return fenced_construct_reply(request->xml, NULL, &request->result); } // STONITH_OP_LEVEL_DEL static xmlNode * handle_level_delete_request(pcmk__request_t *request) { char *desc = NULL; const char *op = crm_element_value(request->xml, PCMK__XA_ST_OP); if (is_privileged(request->ipc_client, op)) { fenced_unregister_level(request->xml, &desc, &request->result); } else { unpack_level_request(request->xml, NULL, NULL, NULL, &desc); pcmk__set_result(&request->result, CRM_EX_INSUFFICIENT_PRIV, PCMK_EXEC_INVALID, "Unprivileged users must delete level via CIB"); } fenced_send_config_notification(op, &request->result, desc); free(desc); return fenced_construct_reply(request->xml, NULL, &request->result); } // CRM_OP_RM_NODE_CACHE static xmlNode * handle_cache_request(pcmk__request_t *request) { int node_id = 0; const char *name = NULL; crm_element_value_int(request->xml, PCMK_XA_ID, &node_id); name = crm_element_value(request->xml, PCMK_XA_UNAME); pcmk__cluster_forget_cluster_node(node_id, name); pcmk__set_result(&request->result, CRM_EX_OK, PCMK_EXEC_DONE, NULL); return NULL; } static xmlNode * handle_unknown_request(pcmk__request_t *request) { crm_err("Unknown IPC request %s from %s %s", request->op, pcmk__request_origin_type(request), pcmk__request_origin(request)); pcmk__format_result(&request->result, CRM_EX_PROTOCOL, PCMK_EXEC_INVALID, "Unknown IPC request type '%s' (bug?)", request->op); return fenced_construct_reply(request->xml, NULL, &request->result); } static void fenced_register_handlers(void) { pcmk__server_command_t handlers[] = { { CRM_OP_REGISTER, handle_register_request }, { STONITH_OP_EXEC, handle_agent_request }, { STONITH_OP_TIMEOUT_UPDATE, handle_update_timeout_request }, { STONITH_OP_QUERY, handle_query_request }, { STONITH_OP_NOTIFY, handle_notify_request }, { STONITH_OP_RELAY, handle_relay_request }, { STONITH_OP_FENCE, handle_fence_request }, { STONITH_OP_FENCE_HISTORY, handle_history_request }, { STONITH_OP_DEVICE_ADD, handle_device_add_request }, { STONITH_OP_DEVICE_DEL, handle_device_delete_request }, { STONITH_OP_LEVEL_ADD, handle_level_add_request }, { STONITH_OP_LEVEL_DEL, handle_level_delete_request }, { CRM_OP_RM_NODE_CACHE, handle_cache_request }, { NULL, handle_unknown_request }, }; fenced_handlers = pcmk__register_handlers(handlers); } void fenced_unregister_handlers(void) { if (fenced_handlers != NULL) { g_hash_table_destroy(fenced_handlers); fenced_handlers = NULL; } } static void handle_request(pcmk__request_t *request) { xmlNode *reply = NULL; const char *reason = NULL; if (fenced_handlers == NULL) { fenced_register_handlers(); } reply = pcmk__process_request(request, fenced_handlers); if (reply != NULL) { if (pcmk_is_set(request->flags, pcmk__request_reuse_options) && (request->ipc_client != NULL)) { /* Certain IPC-only commands must reuse the call options from the * original request rather than the ones set by stonith_send_reply() * -> do_local_reply(). */ pcmk__ipc_send_xml(request->ipc_client, request->ipc_id, reply, request->ipc_flags); request->ipc_client->request_id = 0; } else { stonith_send_reply(reply, request->call_options, request->peer, request->ipc_client); } pcmk__xml_free(reply); } reason = request->result.exit_reason; crm_debug("Processed %s request from %s %s: %s%s%s%s", request->op, pcmk__request_origin_type(request), pcmk__request_origin(request), pcmk_exec_status_str(request->result.execution_status), (reason == NULL)? "" : " (", (reason == NULL)? "" : reason, (reason == NULL)? "" : ")"); } static void handle_reply(pcmk__client_t *client, xmlNode *request, const char *remote_peer) { // Copy, because request might be freed before we want to log this char *op = crm_element_value_copy(request, PCMK__XA_ST_OP); if (pcmk__str_eq(op, STONITH_OP_QUERY, pcmk__str_none)) { process_remote_stonith_query(request); } else if (pcmk__str_any_of(op, STONITH_OP_NOTIFY, STONITH_OP_FENCE, NULL)) { fenced_process_fencing_reply(request); } else { crm_err("Ignoring unknown %s reply from %s %s", pcmk__s(op, "untyped"), ((client == NULL)? "peer" : "client"), ((client == NULL)? remote_peer : pcmk__client_name(client))); crm_log_xml_warn(request, "UnknownOp"); free(op); return; } crm_debug("Processed %s reply from %s %s", op, ((client == NULL)? "peer" : "client"), ((client == NULL)? remote_peer : pcmk__client_name(client))); free(op); } /*! * \internal * \brief Handle a message from an IPC client or CPG peer * * \param[in,out] client If not NULL, IPC client that sent message * \param[in] id If from IPC client, IPC message ID * \param[in] flags Message flags * \param[in,out] message Message XML * \param[in] remote_peer If not NULL, CPG peer that sent message */ void stonith_command(pcmk__client_t *client, uint32_t id, uint32_t flags, xmlNode *message, const char *remote_peer) { int call_options = st_opt_none; bool is_reply = false; CRM_CHECK(message != NULL, return); if (get_xpath_object("//" PCMK__XE_ST_REPLY, message, LOG_NEVER) != NULL) { is_reply = true; } crm_element_value_int(message, PCMK__XA_ST_CALLOPT, &call_options); crm_debug("Processing %ssynchronous %s %s %u from %s %s", pcmk_is_set(call_options, st_opt_sync_call)? "" : "a", crm_element_value(message, PCMK__XA_ST_OP), (is_reply? "reply" : "request"), id, ((client == NULL)? "peer" : "client"), ((client == NULL)? remote_peer : pcmk__client_name(client))); if (pcmk_is_set(call_options, st_opt_sync_call)) { CRM_ASSERT(client == NULL || client->request_id == id); } if (is_reply) { handle_reply(client, message, remote_peer); } else { pcmk__request_t request = { .ipc_client = client, .ipc_id = id, .ipc_flags = flags, .peer = remote_peer, .xml = message, .call_options = call_options, .result = PCMK__UNKNOWN_RESULT, }; request.op = crm_element_value_copy(request.xml, PCMK__XA_ST_OP); CRM_CHECK(request.op != NULL, return); if (pcmk_is_set(request.call_options, st_opt_sync_call)) { pcmk__set_request_flags(&request, pcmk__request_sync); } handle_request(&request); pcmk__reset_request(&request); } } diff --git a/daemons/fenced/fenced_history.c b/daemons/fenced/fenced_history.c index b8e51912bc..e0b5729707 100644 --- a/daemons/fenced/fenced_history.c +++ b/daemons/fenced/fenced_history.c @@ -1,572 +1,572 @@ /* * Copyright 2009-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include #include #include #include #include #define MAX_STONITH_HISTORY 500 /*! * \internal * \brief Send a broadcast to all nodes to trigger cleanup or * history synchronisation * * \param[in] history Optional history to be attached * \param[in] callopts We control cleanup via a flag in the callopts * \param[in] target Cleanup can be limited to certain fence-targets */ static void stonith_send_broadcast_history(xmlNode *history, int callopts, const char *target) { xmlNode *bcast = pcmk__xe_create(NULL, PCMK__XE_STONITH_COMMAND); xmlNode *wrapper = pcmk__xe_create(bcast, PCMK__XE_ST_CALLDATA); xmlNode *call_data = pcmk__xe_create(wrapper, __func__); crm_xml_add(bcast, PCMK__XA_T, PCMK__VALUE_STONITH_NG); crm_xml_add(bcast, PCMK__XA_SUBT, PCMK__VALUE_BROADCAST); crm_xml_add(bcast, PCMK__XA_ST_OP, STONITH_OP_FENCE_HISTORY); crm_xml_add_int(bcast, PCMK__XA_ST_CALLOPT, callopts); pcmk__xml_copy(call_data, history); if (target != NULL) { crm_xml_add(call_data, PCMK__XA_ST_TARGET, target); } - pcmk__cluster_send_message(NULL, crm_msg_stonith_ng, bcast); + pcmk__cluster_send_message(NULL, pcmk__cluster_msg_fenced, bcast); pcmk__xml_free(bcast); } static gboolean stonith_remove_history_entry (gpointer key, gpointer value, gpointer user_data) { remote_fencing_op_t *op = value; const char *target = (const char *) user_data; if ((op->state == st_failed) || (op->state == st_done)) { if ((target) && (strcmp(op->target, target) != 0)) { return FALSE; } return TRUE; } return FALSE; /* don't clean pending operations */ } /*! * \internal * \brief Send out a cleanup broadcast or do a local history-cleanup * * \param[in] target Cleanup can be limited to certain fence-targets * \param[in] broadcast Send out a cleanup broadcast */ static void stonith_fence_history_cleanup(const char *target, gboolean broadcast) { if (broadcast) { stonith_send_broadcast_history(NULL, st_opt_cleanup | st_opt_discard_reply, target); /* we'll do the local clean when we receive back our own broadcast */ } else if (stonith_remote_op_list) { g_hash_table_foreach_remove(stonith_remote_op_list, stonith_remove_history_entry, (gpointer) target); fenced_send_notification(PCMK__VALUE_ST_NOTIFY_HISTORY, NULL, NULL); } } /* keeping the length of fence-history within bounds * ================================================= * * If things are really running wild a lot of fencing-attempts * might fill up the hash-map, eventually using up a lot * of memory and creating huge history-sync messages. * Before the history being synced across nodes at least * the reboot of a cluster-node helped keeping the * history within bounds even though not in a reliable * manner. * * stonith_remote_op_list isn't sorted for time-stamps * thus it would be kind of expensive to delete e.g. * the oldest entry if it would grow past MAX_STONITH_HISTORY * entries. * It is more efficient to purge MAX_STONITH_HISTORY/2 * entries whenever the list grows beyond MAX_STONITH_HISTORY. * (sort for age + purge the MAX_STONITH_HISTORY/2 oldest) * That done on a per-node-base might raise the * probability of large syncs to occur. * Things like introducing a broadcast to purge * MAX_STONITH_HISTORY/2 entries or not sync above a certain * threshold coming to mind ... * Simplest thing though is to purge the full history * throughout the cluster once MAX_STONITH_HISTORY is reached. * On the other hand this leads to purging the history in * situations where it would be handy to have it probably. */ /*! * \internal * \brief Compare two remote fencing operations by status and completion time * * A pending operation is ordered before a completed operation. If both * operations have completed, then the more recently completed operation is * ordered first. Two pending operations are considered equal. * * \param[in] a First \c remote_fencing_op_t to compare * \param[in] b Second \c remote_fencing_op_t to compare * * \return Standard comparison result (a negative integer if \p a is lesser, * 0 if the values are equal, and a positive integer if \p a is greater) */ static gint cmp_op_by_completion(gconstpointer a, gconstpointer b) { const remote_fencing_op_t *op1 = a; const remote_fencing_op_t *op2 = b; bool op1_pending = stonith__op_state_pending(op1->state); bool op2_pending = stonith__op_state_pending(op2->state); if (op1_pending && op2_pending) { return 0; } if (op1_pending) { return -1; } if (op2_pending) { return 1; } if (op1->completed > op2->completed) { return -1; } if (op1->completed < op2->completed) { return 1; } if (op1->completed_nsec > op2->completed_nsec) { return -1; } if (op1->completed_nsec < op2->completed_nsec) { return 1; } return 0; } /*! * \internal * \brief Remove a completed operation from \c stonith_remote_op_list * * \param[in] data \c remote_fencing_op_t to remove * \param[in] user_data Ignored */ static void remove_completed_remote_op(gpointer data, gpointer user_data) { const remote_fencing_op_t *op = data; if (!stonith__op_state_pending(op->state)) { g_hash_table_remove(stonith_remote_op_list, op->id); } } /*! * \internal * \brief Do a local history-trim to MAX_STONITH_HISTORY / 2 entries * once over MAX_STONITH_HISTORY */ void stonith_fence_history_trim(void) { if (stonith_remote_op_list == NULL) { return; } if (g_hash_table_size(stonith_remote_op_list) > MAX_STONITH_HISTORY) { GList *ops = g_hash_table_get_values(stonith_remote_op_list); crm_trace("More than %d entries in fencing history, purging oldest " "completed operations", MAX_STONITH_HISTORY); ops = g_list_sort(ops, cmp_op_by_completion); // Always keep pending ops regardless of number of entries g_list_foreach(g_list_nth(ops, MAX_STONITH_HISTORY / 2), remove_completed_remote_op, NULL); // No need for a notification after purging old data g_list_free(ops); } } /*! * \internal * \brief Convert xml fence-history to a hash-table like stonith_remote_op_list * * \param[in] history Fence-history in xml * * \return Fence-history as hash-table */ static GHashTable * stonith_xml_history_to_list(const xmlNode *history) { xmlNode *xml_op = NULL; GHashTable *rv = NULL; init_stonith_remote_op_hash_table(&rv); CRM_LOG_ASSERT(rv != NULL); for (xml_op = pcmk__xe_first_child(history, NULL, NULL, NULL); xml_op != NULL; xml_op = pcmk__xe_next(xml_op)) { remote_fencing_op_t *op = NULL; char *id = crm_element_value_copy(xml_op, PCMK__XA_ST_REMOTE_OP); int state; int exit_status = CRM_EX_OK; int execution_status = PCMK_EXEC_DONE; long long completed; long long completed_nsec = 0L; if (!id) { crm_warn("Malformed fencing history received from peer"); continue; } crm_trace("Attaching op %s to hashtable", id); op = pcmk__assert_alloc(1, sizeof(remote_fencing_op_t)); op->id = id; op->target = crm_element_value_copy(xml_op, PCMK__XA_ST_TARGET); op->action = crm_element_value_copy(xml_op, PCMK__XA_ST_DEVICE_ACTION); op->originator = crm_element_value_copy(xml_op, PCMK__XA_ST_ORIGIN); op->delegate = crm_element_value_copy(xml_op, PCMK__XA_ST_DELEGATE); op->client_name = crm_element_value_copy(xml_op, PCMK__XA_ST_CLIENTNAME); crm_element_value_ll(xml_op, PCMK__XA_ST_DATE, &completed); op->completed = (time_t) completed; crm_element_value_ll(xml_op, PCMK__XA_ST_DATE_NSEC, &completed_nsec); op->completed_nsec = completed_nsec; crm_element_value_int(xml_op, PCMK__XA_ST_STATE, &state); op->state = (enum op_state) state; /* @COMPAT We can't use stonith__xe_get_result() here because * fencers <2.1.3 didn't include results, leading it to assume an error * status. Instead, set an unknown status in that case. */ if ((crm_element_value_int(xml_op, PCMK__XA_RC_CODE, &exit_status) < 0) || (crm_element_value_int(xml_op, PCMK__XA_OP_STATUS, &execution_status) < 0)) { exit_status = CRM_EX_INDETERMINATE; execution_status = PCMK_EXEC_UNKNOWN; } pcmk__set_result(&op->result, exit_status, execution_status, crm_element_value(xml_op, PCMK_XA_EXIT_REASON)); pcmk__set_result_output(&op->result, crm_element_value_copy(xml_op, PCMK__XA_ST_OUTPUT), NULL); g_hash_table_replace(rv, id, op); CRM_LOG_ASSERT(g_hash_table_lookup(rv, id) != NULL); } return rv; } /*! * \internal * \brief Craft xml difference between local fence-history and a history * coming from remote, and merge the remote history into the local * * \param[in,out] remote_history Fence-history as hash-table (may be NULL) * \param[in] add_id If crafting the answer for an API * history-request there is no need for the id * \param[in] target Optionally limit to certain fence-target * * \return The fence-history as xml */ static xmlNode * stonith_local_history_diff_and_merge(GHashTable *remote_history, gboolean add_id, const char *target) { xmlNode *history = NULL; GHashTableIter iter; remote_fencing_op_t *op = NULL; gboolean updated = FALSE; int cnt = 0; if (stonith_remote_op_list) { char *id = NULL; history = pcmk__xe_create(NULL, PCMK__XE_ST_HISTORY); g_hash_table_iter_init(&iter, stonith_remote_op_list); while (g_hash_table_iter_next(&iter, (void **)&id, (void **)&op)) { xmlNode *entry = NULL; if (remote_history) { remote_fencing_op_t *remote_op = g_hash_table_lookup(remote_history, op->id); if (remote_op) { if (stonith__op_state_pending(op->state) && !stonith__op_state_pending(remote_op->state)) { crm_debug("Updating outdated pending operation %.8s " "(state=%s) according to the one (state=%s) from " "remote peer history", op->id, stonith_op_state_str(op->state), stonith_op_state_str(remote_op->state)); g_hash_table_steal(remote_history, op->id); op->id = remote_op->id; remote_op->id = id; g_hash_table_iter_replace(&iter, remote_op); updated = TRUE; continue; /* skip outdated entries */ } else if (!stonith__op_state_pending(op->state) && stonith__op_state_pending(remote_op->state)) { crm_debug("Broadcasting operation %.8s (state=%s) to " "update the outdated pending one " "(state=%s) in remote peer history", op->id, stonith_op_state_str(op->state), stonith_op_state_str(remote_op->state)); g_hash_table_remove(remote_history, op->id); } else { g_hash_table_remove(remote_history, op->id); continue; /* skip entries broadcasted already */ } } } if (!pcmk__str_eq(target, op->target, pcmk__str_null_matches)) { continue; } cnt++; crm_trace("Attaching op %s", op->id); entry = pcmk__xe_create(history, STONITH_OP_EXEC); if (add_id) { crm_xml_add(entry, PCMK__XA_ST_REMOTE_OP, op->id); } crm_xml_add(entry, PCMK__XA_ST_TARGET, op->target); crm_xml_add(entry, PCMK__XA_ST_DEVICE_ACTION, op->action); crm_xml_add(entry, PCMK__XA_ST_ORIGIN, op->originator); crm_xml_add(entry, PCMK__XA_ST_DELEGATE, op->delegate); crm_xml_add(entry, PCMK__XA_ST_CLIENTNAME, op->client_name); crm_xml_add_ll(entry, PCMK__XA_ST_DATE, op->completed); crm_xml_add_ll(entry, PCMK__XA_ST_DATE_NSEC, op->completed_nsec); crm_xml_add_int(entry, PCMK__XA_ST_STATE, op->state); stonith__xe_set_result(entry, &op->result); } } if (remote_history) { init_stonith_remote_op_hash_table(&stonith_remote_op_list); updated |= g_hash_table_size(remote_history); g_hash_table_iter_init(&iter, remote_history); while (g_hash_table_iter_next(&iter, NULL, (void **)&op)) { if (stonith__op_state_pending(op->state) && pcmk__str_eq(op->originator, stonith_our_uname, pcmk__str_casei)) { crm_warn("Failing pending operation %.8s originated by us but " "known only from peer history", op->id); op->state = st_failed; set_fencing_completed(op); /* CRM_EX_EXPIRED + PCMK_EXEC_INVALID prevents finalize_op() * from setting a delegate */ pcmk__set_result(&op->result, CRM_EX_EXPIRED, PCMK_EXEC_INVALID, "Initiated by earlier fencer " "process and presumed failed"); fenced_broadcast_op_result(op, false); } g_hash_table_iter_steal(&iter); g_hash_table_replace(stonith_remote_op_list, op->id, op); /* we could trim the history here but if we bail * out after trim we might miss more recent entries * of those that might still be in the list * if we don't bail out trimming once is more * efficient and memory overhead is minimal as * we are just moving pointers from one hash to * another */ } g_hash_table_destroy(remote_history); /* remove what is left */ } if (updated) { stonith_fence_history_trim(); fenced_send_notification(PCMK__VALUE_ST_NOTIFY_HISTORY, NULL, NULL); } if (cnt == 0) { pcmk__xml_free(history); return NULL; } else { return history; } } /*! * \internal * \brief Craft xml from the local fence-history * * \param[in] add_id If crafting the answer for an API * history-request there is no need for the id * \param[in] target Optionally limit to certain fence-target * * \return The fence-history as xml */ static xmlNode * stonith_local_history(gboolean add_id, const char *target) { return stonith_local_history_diff_and_merge(NULL, add_id, target); } /*! * \internal * \brief Handle fence-history messages (from API or coming in as broadcasts) * * \param[in,out] msg Request XML * \param[out] output Where to set local history, if requested * \param[in] remote_peer If broadcast, peer that sent it * \param[in] options Call options from the request */ void stonith_fence_history(xmlNode *msg, xmlNode **output, const char *remote_peer, int options) { const char *target = NULL; xmlNode *dev = get_xpath_object("//@" PCMK__XA_ST_TARGET, msg, LOG_NEVER); xmlNode *out_history = NULL; if (dev) { target = crm_element_value(dev, PCMK__XA_ST_TARGET); if (target && (options & st_opt_cs_nodeid)) { int nodeid; pcmk__node_status_t *node = NULL; pcmk__scan_min_int(target, &nodeid, 0); node = pcmk__search_node_caches(nodeid, NULL, pcmk__node_search_any |pcmk__node_search_cluster_cib); if (node != NULL) { target = node->name; } } } if (options & st_opt_cleanup) { const char *call_id = crm_element_value(msg, PCMK__XA_ST_CALLID); crm_trace("Cleaning up operations on %s in %p", target, stonith_remote_op_list); stonith_fence_history_cleanup(target, (call_id != NULL)); } else if (options & st_opt_broadcast) { /* there is no clear sign atm for when a history sync is done so send a notification for anything that smells like history-sync */ fenced_send_notification(PCMK__VALUE_ST_NOTIFY_HISTORY_SYNCED, NULL, NULL); if (crm_element_value(msg, PCMK__XA_ST_CALLID) != NULL) { /* this is coming from the stonith-API * * craft a broadcast with node's history * so that every node can merge and broadcast * what it has on top */ out_history = stonith_local_history(TRUE, NULL); crm_trace("Broadcasting history to peers"); stonith_send_broadcast_history(out_history, st_opt_broadcast | st_opt_discard_reply, NULL); } else if (remote_peer && !pcmk__str_eq(remote_peer, stonith_our_uname, pcmk__str_casei)) { xmlNode *history = get_xpath_object("//" PCMK__XE_ST_HISTORY, msg, LOG_NEVER); /* either a broadcast created directly upon stonith-API request * or a diff as response to such a thing * * in both cases it may have a history or not * if we have differential data * merge in what we've received and stop * otherwise broadcast what we have on top * marking as differential and merge in afterwards */ if (!history || !pcmk__xe_attr_is_true(history, PCMK__XA_ST_DIFFERENTIAL)) { GHashTable *received_history = NULL; if (history != NULL) { received_history = stonith_xml_history_to_list(history); } out_history = stonith_local_history_diff_and_merge(received_history, TRUE, NULL); if (out_history) { crm_trace("Broadcasting history-diff to peers"); pcmk__xe_set_bool_attr(out_history, PCMK__XA_ST_DIFFERENTIAL, true); stonith_send_broadcast_history(out_history, st_opt_broadcast | st_opt_discard_reply, NULL); } else { crm_trace("History-diff is empty - skip broadcast"); } } } else { crm_trace("Skipping history-query-broadcast (%s%s)" " we sent ourselves", remote_peer?"remote-peer=":"local-ipc", remote_peer?remote_peer:""); } } else { /* plain history request */ crm_trace("Looking for operations on %s in %p", target, stonith_remote_op_list); *output = stonith_local_history(FALSE, target); } pcmk__xml_free(out_history); } diff --git a/daemons/fenced/fenced_remote.c b/daemons/fenced/fenced_remote.c index a02c30b5d6..5ae1ffe384 100644 --- a/daemons/fenced/fenced_remote.c +++ b/daemons/fenced/fenced_remote.c @@ -1,2597 +1,2598 @@ /* * Copyright 2009-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #define TIMEOUT_MULTIPLY_FACTOR 1.2 /* When one fencer queries its peers for devices able to handle a fencing * request, each peer will reply with a list of such devices available to it. * Each reply will be parsed into a peer_device_info_t, with each device's * information kept in a device_properties_t. */ typedef struct device_properties_s { /* Whether access to this device has been verified */ gboolean verified; /* The remaining members are indexed by the operation's "phase" */ /* Whether this device has been executed in each phase */ gboolean executed[st_phase_max]; /* Whether this device is disallowed from executing in each phase */ gboolean disallowed[st_phase_max]; /* Action-specific timeout for each phase */ int custom_action_timeout[st_phase_max]; /* Action-specific maximum random delay for each phase */ int delay_max[st_phase_max]; /* Action-specific base delay for each phase */ int delay_base[st_phase_max]; /* Group of enum st_device_flags */ uint32_t device_support_flags; } device_properties_t; typedef struct { /* Name of peer that sent this result */ char *host; /* Only try peers for non-topology based operations once */ gboolean tried; /* Number of entries in the devices table */ int ndevices; /* Devices available to this host that are capable of fencing the target */ GHashTable *devices; } peer_device_info_t; GHashTable *stonith_remote_op_list = NULL; extern xmlNode *stonith_create_op(int call_id, const char *token, const char *op, xmlNode * data, int call_options); static void request_peer_fencing(remote_fencing_op_t *op, peer_device_info_t *peer); static void finalize_op(remote_fencing_op_t *op, xmlNode *data, bool dup); static void report_timeout_period(remote_fencing_op_t * op, int op_timeout); static int get_op_total_timeout(const remote_fencing_op_t *op, const peer_device_info_t *chosen_peer); static gint sort_strings(gconstpointer a, gconstpointer b) { return strcmp(a, b); } static void free_remote_query(gpointer data) { if (data != NULL) { peer_device_info_t *peer = data; g_hash_table_destroy(peer->devices); free(peer->host); free(peer); } } void free_stonith_remote_op_list(void) { if (stonith_remote_op_list != NULL) { g_hash_table_destroy(stonith_remote_op_list); stonith_remote_op_list = NULL; } } struct peer_count_data { const remote_fencing_op_t *op; gboolean verified_only; uint32_t support_action_only; int count; }; /*! * \internal * \brief Increment a counter if a device has not been executed yet * * \param[in] key Device ID (ignored) * \param[in] value Device properties * \param[in,out] user_data Peer count data */ static void count_peer_device(gpointer key, gpointer value, gpointer user_data) { device_properties_t *props = (device_properties_t*)value; struct peer_count_data *data = user_data; if (!props->executed[data->op->phase] && (!data->verified_only || props->verified) && ((data->support_action_only == st_device_supports_none) || pcmk_is_set(props->device_support_flags, data->support_action_only))) { ++(data->count); } } /*! * \internal * \brief Check the number of available devices in a peer's query results * * \param[in] op Operation that results are for * \param[in] peer Peer to count * \param[in] verified_only Whether to count only verified devices * \param[in] support_action_only Whether to count only devices that support action * * \return Number of devices available to peer that were not already executed */ static int count_peer_devices(const remote_fencing_op_t *op, const peer_device_info_t *peer, gboolean verified_only, uint32_t support_on_action_only) { struct peer_count_data data; data.op = op; data.verified_only = verified_only; data.support_action_only = support_on_action_only; data.count = 0; if (peer) { g_hash_table_foreach(peer->devices, count_peer_device, &data); } return data.count; } /*! * \internal * \brief Search for a device in a query result * * \param[in] op Operation that result is for * \param[in] peer Query result for a peer * \param[in] device Device ID to search for * * \return Device properties if found, NULL otherwise */ static device_properties_t * find_peer_device(const remote_fencing_op_t *op, const peer_device_info_t *peer, const char *device, uint32_t support_action_only) { device_properties_t *props = g_hash_table_lookup(peer->devices, device); if (props && support_action_only != st_device_supports_none && !pcmk_is_set(props->device_support_flags, support_action_only)) { return NULL; } return (props && !props->executed[op->phase] && !props->disallowed[op->phase])? props : NULL; } /*! * \internal * \brief Find a device in a peer's device list and mark it as executed * * \param[in] op Operation that peer result is for * \param[in,out] peer Peer with results to search * \param[in] device ID of device to mark as done * \param[in] verified_devices_only Only consider verified devices * * \return TRUE if device was found and marked, FALSE otherwise */ static gboolean grab_peer_device(const remote_fencing_op_t *op, peer_device_info_t *peer, const char *device, gboolean verified_devices_only) { device_properties_t *props = find_peer_device(op, peer, device, fenced_support_flag(op->action)); if ((props == NULL) || (verified_devices_only && !props->verified)) { return FALSE; } crm_trace("Removing %s from %s (%d remaining)", device, peer->host, count_peer_devices(op, peer, FALSE, st_device_supports_none)); props->executed[op->phase] = TRUE; return TRUE; } static void clear_remote_op_timers(remote_fencing_op_t * op) { if (op->query_timer) { g_source_remove(op->query_timer); op->query_timer = 0; } if (op->op_timer_total) { g_source_remove(op->op_timer_total); op->op_timer_total = 0; } if (op->op_timer_one) { g_source_remove(op->op_timer_one); op->op_timer_one = 0; } } static void free_remote_op(gpointer data) { remote_fencing_op_t *op = data; crm_log_xml_debug(op->request, "Destroying"); clear_remote_op_timers(op); free(op->id); free(op->action); free(op->delegate); free(op->target); free(op->client_id); free(op->client_name); free(op->originator); if (op->query_results) { g_list_free_full(op->query_results, free_remote_query); } if (op->request) { pcmk__xml_free(op->request); op->request = NULL; } if (op->devices_list) { g_list_free_full(op->devices_list, free); op->devices_list = NULL; } g_list_free_full(op->automatic_list, free); g_list_free(op->duplicates); pcmk__reset_result(&op->result); free(op); } void init_stonith_remote_op_hash_table(GHashTable **table) { if (*table == NULL) { *table = pcmk__strkey_table(NULL, free_remote_op); } } /*! * \internal * \brief Return an operation's originally requested action (before any remap) * * \param[in] op Operation to check * * \return Operation's original action */ static const char * op_requested_action(const remote_fencing_op_t *op) { return ((op->phase > st_phase_requested)? PCMK_ACTION_REBOOT : op->action); } /*! * \internal * \brief Remap a "reboot" operation to the "off" phase * * \param[in,out] op Operation to remap */ static void op_phase_off(remote_fencing_op_t *op) { crm_info("Remapping multiple-device reboot targeting %s to 'off' " QB_XS " id=%.8s", op->target, op->id); op->phase = st_phase_off; /* Happily, "off" and "on" are shorter than "reboot", so we can reuse the * memory allocation at each phase. */ strcpy(op->action, PCMK_ACTION_OFF); } /*! * \internal * \brief Advance a remapped reboot operation to the "on" phase * * \param[in,out] op Operation to remap */ static void op_phase_on(remote_fencing_op_t *op) { GList *iter = NULL; crm_info("Remapped 'off' targeting %s complete, " "remapping to 'on' for %s " QB_XS " id=%.8s", op->target, op->client_name, op->id); op->phase = st_phase_on; strcpy(op->action, PCMK_ACTION_ON); /* Skip devices with automatic unfencing, because the cluster will handle it * when the node rejoins. */ for (iter = op->automatic_list; iter != NULL; iter = iter->next) { GList *match = g_list_find_custom(op->devices_list, iter->data, sort_strings); if (match) { op->devices_list = g_list_remove(op->devices_list, match->data); } } g_list_free_full(op->automatic_list, free); op->automatic_list = NULL; /* Rewind device list pointer */ op->devices = op->devices_list; } /*! * \internal * \brief Reset a remapped reboot operation * * \param[in,out] op Operation to reset */ static void undo_op_remap(remote_fencing_op_t *op) { if (op->phase > 0) { crm_info("Undoing remap of reboot targeting %s for %s " QB_XS " id=%.8s", op->target, op->client_name, op->id); op->phase = st_phase_requested; strcpy(op->action, PCMK_ACTION_REBOOT); } } /*! * \internal * \brief Create notification data XML for a fencing operation result * * \param[in,out] parent Parent XML element for newly created element * \param[in] op Fencer operation that completed * * \return Newly created XML to add as notification data * \note The caller is responsible for freeing the result. */ static xmlNode * fencing_result2xml(xmlNode *parent, const remote_fencing_op_t *op) { xmlNode *notify_data = pcmk__xe_create(parent, PCMK__XE_ST_NOTIFY_FENCE); crm_xml_add_int(notify_data, PCMK_XA_STATE, op->state); crm_xml_add(notify_data, PCMK__XA_ST_TARGET, op->target); crm_xml_add(notify_data, PCMK__XA_ST_DEVICE_ACTION, op->action); crm_xml_add(notify_data, PCMK__XA_ST_DELEGATE, op->delegate); crm_xml_add(notify_data, PCMK__XA_ST_REMOTE_OP, op->id); crm_xml_add(notify_data, PCMK__XA_ST_ORIGIN, op->originator); crm_xml_add(notify_data, PCMK__XA_ST_CLIENTID, op->client_id); crm_xml_add(notify_data, PCMK__XA_ST_CLIENTNAME, op->client_name); return notify_data; } /*! * \internal * \brief Broadcast a fence result notification to all CPG peers * * \param[in] op Fencer operation that completed * \param[in] op_merged Whether this operation is a duplicate of another */ void fenced_broadcast_op_result(const remote_fencing_op_t *op, bool op_merged) { static int count = 0; xmlNode *bcast = pcmk__xe_create(NULL, PCMK__XE_ST_REPLY); xmlNode *wrapper = NULL; xmlNode *notify_data = NULL; count++; crm_trace("Broadcasting result to peers"); crm_xml_add(bcast, PCMK__XA_T, PCMK__VALUE_ST_NOTIFY); crm_xml_add(bcast, PCMK__XA_SUBT, PCMK__VALUE_BROADCAST); crm_xml_add(bcast, PCMK__XA_ST_OP, STONITH_OP_NOTIFY); crm_xml_add_int(bcast, PCMK_XA_COUNT, count); if (op_merged) { pcmk__xe_set_bool_attr(bcast, PCMK__XA_ST_OP_MERGED, true); } wrapper = pcmk__xe_create(bcast, PCMK__XE_ST_CALLDATA); notify_data = fencing_result2xml(wrapper, op); stonith__xe_set_result(notify_data, &op->result); - pcmk__cluster_send_message(NULL, crm_msg_stonith_ng, bcast); + pcmk__cluster_send_message(NULL, pcmk__cluster_msg_fenced, bcast); pcmk__xml_free(bcast); return; } /*! * \internal * \brief Reply to a local request originator and notify all subscribed clients * * \param[in,out] op Fencer operation that completed * \param[in,out] data Top-level XML to add notification to */ static void handle_local_reply_and_notify(remote_fencing_op_t *op, xmlNode *data) { xmlNode *notify_data = NULL; xmlNode *reply = NULL; pcmk__client_t *client = NULL; if (op->notify_sent == TRUE) { /* nothing to do */ return; } /* Do notification with a clean data object */ crm_xml_add_int(data, PCMK_XA_STATE, op->state); crm_xml_add(data, PCMK__XA_ST_TARGET, op->target); crm_xml_add(data, PCMK__XA_ST_OP, op->action); reply = fenced_construct_reply(op->request, data, &op->result); crm_xml_add(reply, PCMK__XA_ST_DELEGATE, op->delegate); /* Send fencing OP reply to local client that initiated fencing */ client = pcmk__find_client_by_id(op->client_id); if (client == NULL) { crm_trace("Skipping reply to %s: no longer a client", op->client_id); } else { do_local_reply(reply, client, op->call_options); } /* bcast to all local clients that the fencing operation happend */ notify_data = fencing_result2xml(NULL, op); fenced_send_notification(PCMK__VALUE_ST_NOTIFY_FENCE, &op->result, notify_data); pcmk__xml_free(notify_data); fenced_send_notification(PCMK__VALUE_ST_NOTIFY_HISTORY, NULL, NULL); /* mark this op as having notify's already sent */ op->notify_sent = TRUE; pcmk__xml_free(reply); } /*! * \internal * \brief Finalize all duplicates of a given fencer operation * * \param[in,out] op Fencer operation that completed * \param[in,out] data Top-level XML to add notification to */ static void finalize_op_duplicates(remote_fencing_op_t *op, xmlNode *data) { for (GList *iter = op->duplicates; iter != NULL; iter = iter->next) { remote_fencing_op_t *other = iter->data; if (other->state == st_duplicate) { other->state = op->state; crm_debug("Performing duplicate notification for %s@%s: %s " QB_XS " id=%.8s", other->client_name, other->originator, pcmk_exec_status_str(op->result.execution_status), other->id); pcmk__copy_result(&op->result, &other->result); finalize_op(other, data, true); } else { // Possible if (for example) it timed out already crm_err("Skipping duplicate notification for %s@%s " QB_XS " state=%s id=%.8s", other->client_name, other->originator, stonith_op_state_str(other->state), other->id); } } } static char * delegate_from_xml(xmlNode *xml) { xmlNode *match = get_xpath_object("//@" PCMK__XA_ST_DELEGATE, xml, LOG_NEVER); if (match == NULL) { return crm_element_value_copy(xml, PCMK__XA_SRC); } else { return crm_element_value_copy(match, PCMK__XA_ST_DELEGATE); } } /*! * \internal * \brief Finalize a peer fencing operation * * Clean up after a fencing operation completes. This function has two code * paths: the executioner uses it to broadcast the result to CPG peers, and then * each peer (including the executioner) uses it to process that broadcast and * notify its IPC clients of the result. * * \param[in,out] op Fencer operation that completed * \param[in,out] data If not NULL, XML reply of last delegated operation * \param[in] dup Whether this operation is a duplicate of another * (in which case, do not broadcast the result) * * \note The operation result should be set before calling this function. */ static void finalize_op(remote_fencing_op_t *op, xmlNode *data, bool dup) { int level = LOG_ERR; const char *subt = NULL; xmlNode *local_data = NULL; gboolean op_merged = FALSE; CRM_CHECK((op != NULL), return); // This is a no-op if timers have already been cleared clear_remote_op_timers(op); if (op->notify_sent) { // Most likely, this is a timed-out action that eventually completed crm_notice("Operation '%s'%s%s by %s for %s@%s%s: " "Result arrived too late " QB_XS " id=%.8s", op->action, (op->target? " targeting " : ""), (op->target? op->target : ""), (op->delegate? op->delegate : "unknown node"), op->client_name, op->originator, (op_merged? " (merged)" : ""), op->id); return; } set_fencing_completed(op); undo_op_remap(op); if (data == NULL) { data = pcmk__xe_create(NULL, "remote-op"); local_data = data; } else if (op->delegate == NULL) { switch (op->result.execution_status) { case PCMK_EXEC_NO_FENCE_DEVICE: break; case PCMK_EXEC_INVALID: if (op->result.exit_status != CRM_EX_EXPIRED) { op->delegate = delegate_from_xml(data); } break; default: op->delegate = delegate_from_xml(data); break; } } if (dup || (crm_element_value(data, PCMK__XA_ST_OP_MERGED) != NULL)) { op_merged = true; } /* Tell everyone the operation is done, we will continue * with doing the local notifications once we receive * the broadcast back. */ subt = crm_element_value(data, PCMK__XA_SUBT); if (!dup && !pcmk__str_eq(subt, PCMK__VALUE_BROADCAST, pcmk__str_none)) { /* Defer notification until the bcast message arrives */ fenced_broadcast_op_result(op, op_merged); pcmk__xml_free(local_data); return; } if (pcmk__result_ok(&op->result) || dup || !pcmk__str_eq(op->originator, stonith_our_uname, pcmk__str_casei)) { level = LOG_NOTICE; } do_crm_log(level, "Operation '%s'%s%s by %s for %s@%s%s: %s (%s%s%s) " QB_XS " id=%.8s", op->action, (op->target? " targeting " : ""), (op->target? op->target : ""), (op->delegate? op->delegate : "unknown node"), op->client_name, op->originator, (op_merged? " (merged)" : ""), crm_exit_str(op->result.exit_status), pcmk_exec_status_str(op->result.execution_status), ((op->result.exit_reason == NULL)? "" : ": "), ((op->result.exit_reason == NULL)? "" : op->result.exit_reason), op->id); handle_local_reply_and_notify(op, data); if (!dup) { finalize_op_duplicates(op, data); } /* Free non-essential parts of the record * Keep the record around so we can query the history */ if (op->query_results) { g_list_free_full(op->query_results, free_remote_query); op->query_results = NULL; } if (op->request) { pcmk__xml_free(op->request); op->request = NULL; } pcmk__xml_free(local_data); } /*! * \internal * \brief Finalize a watchdog fencer op after the waiting time expires * * \param[in,out] userdata Fencer operation that completed * * \return G_SOURCE_REMOVE (which tells glib not to restart timer) */ static gboolean remote_op_watchdog_done(gpointer userdata) { remote_fencing_op_t *op = userdata; op->op_timer_one = 0; crm_notice("Self-fencing (%s) by %s for %s assumed complete " QB_XS " id=%.8s", op->action, op->target, op->client_name, op->id); op->state = st_done; pcmk__set_result(&op->result, CRM_EX_OK, PCMK_EXEC_DONE, NULL); finalize_op(op, NULL, false); return G_SOURCE_REMOVE; } static gboolean remote_op_timeout_one(gpointer userdata) { remote_fencing_op_t *op = userdata; op->op_timer_one = 0; crm_notice("Peer's '%s' action targeting %s for client %s timed out " QB_XS " id=%.8s", op->action, op->target, op->client_name, op->id); pcmk__set_result(&op->result, CRM_EX_ERROR, PCMK_EXEC_TIMEOUT, "Peer did not return fence result within timeout"); // The requested delay has been applied for the first device if (op->client_delay > 0) { op->client_delay = 0; crm_trace("Try another device for '%s' action targeting %s " "for client %s without delay " QB_XS " id=%.8s", op->action, op->target, op->client_name, op->id); } // Try another device, if appropriate request_peer_fencing(op, NULL); return G_SOURCE_REMOVE; } /*! * \internal * \brief Finalize a remote fencer operation that timed out * * \param[in,out] op Fencer operation that timed out * \param[in] reason Readable description of what step timed out */ static void finalize_timed_out_op(remote_fencing_op_t *op, const char *reason) { crm_debug("Action '%s' targeting %s for client %s timed out " QB_XS " id=%.8s", op->action, op->target, op->client_name, op->id); if (op->phase == st_phase_on) { /* A remapped reboot operation timed out in the "on" phase, but the * "off" phase completed successfully, so quit trying any further * devices, and return success. */ op->state = st_done; pcmk__set_result(&op->result, CRM_EX_OK, PCMK_EXEC_DONE, NULL); } else { op->state = st_failed; pcmk__set_result(&op->result, CRM_EX_ERROR, PCMK_EXEC_TIMEOUT, reason); } finalize_op(op, NULL, false); } /*! * \internal * \brief Finalize a remote fencer operation that timed out * * \param[in,out] userdata Fencer operation that timed out * * \return G_SOURCE_REMOVE (which tells glib not to restart timer) */ static gboolean remote_op_timeout(gpointer userdata) { remote_fencing_op_t *op = userdata; op->op_timer_total = 0; if (op->state == st_done) { crm_debug("Action '%s' targeting %s for client %s already completed " QB_XS " id=%.8s", op->action, op->target, op->client_name, op->id); } else { finalize_timed_out_op(userdata, "Fencing did not complete within a " "total timeout based on the " "configured timeout and retries for " "any devices attempted"); } return G_SOURCE_REMOVE; } static gboolean remote_op_query_timeout(gpointer data) { remote_fencing_op_t *op = data; op->query_timer = 0; if (op->state == st_done) { crm_debug("Operation %.8s targeting %s already completed", op->id, op->target); } else if (op->state == st_exec) { crm_debug("Operation %.8s targeting %s already in progress", op->id, op->target); } else if (op->query_results) { // Query succeeded, so attempt the actual fencing crm_debug("Query %.8s targeting %s complete (state=%s)", op->id, op->target, stonith_op_state_str(op->state)); request_peer_fencing(op, NULL); } else { crm_debug("Query %.8s targeting %s timed out (state=%s)", op->id, op->target, stonith_op_state_str(op->state)); finalize_timed_out_op(op, "No capable peers replied to device query " "within timeout"); } return G_SOURCE_REMOVE; } static gboolean topology_is_empty(stonith_topology_t *tp) { int i; if (tp == NULL) { return TRUE; } for (i = 0; i < ST_LEVEL_MAX; i++) { if (tp->levels[i] != NULL) { return FALSE; } } return TRUE; } /*! * \internal * \brief Add a device to an operation's automatic unfencing list * * \param[in,out] op Operation to modify * \param[in] device Device ID to add */ static void add_required_device(remote_fencing_op_t *op, const char *device) { GList *match = g_list_find_custom(op->automatic_list, device, sort_strings); if (!match) { op->automatic_list = g_list_prepend(op->automatic_list, pcmk__str_copy(device)); } } /*! * \internal * \brief Remove a device from the automatic unfencing list * * \param[in,out] op Operation to modify * \param[in] device Device ID to remove */ static void remove_required_device(remote_fencing_op_t *op, const char *device) { GList *match = g_list_find_custom(op->automatic_list, device, sort_strings); if (match) { op->automatic_list = g_list_remove(op->automatic_list, match->data); } } /* deep copy the device list */ static void set_op_device_list(remote_fencing_op_t * op, GList *devices) { GList *lpc = NULL; if (op->devices_list) { g_list_free_full(op->devices_list, free); op->devices_list = NULL; } for (lpc = devices; lpc != NULL; lpc = lpc->next) { const char *device = lpc->data; op->devices_list = g_list_append(op->devices_list, pcmk__str_copy(device)); } op->devices = op->devices_list; } /*! * \internal * \brief Check whether a node matches a topology target * * \param[in] tp Topology table entry to check * \param[in] node Name of node to check * * \return TRUE if node matches topology target */ static gboolean topology_matches(const stonith_topology_t *tp, const char *node) { regex_t r_patt; CRM_CHECK(node && tp && tp->target, return FALSE); switch (tp->kind) { case fenced_target_by_attribute: /* This level targets by attribute, so tp->target is a NAME=VALUE pair * of a permanent attribute applied to targeted nodes. The test below * relies on the locally cached copy of the CIB, so if fencing needs to * be done before the initial CIB is received or after a malformed CIB * is received, then the topology will be unable to be used. */ if (node_has_attr(node, tp->target_attribute, tp->target_value)) { crm_notice("Matched %s with %s by attribute", node, tp->target); return TRUE; } break; case fenced_target_by_pattern: /* This level targets node names matching a pattern, so tp->target * (and tp->target_pattern) is a regular expression. */ if (regcomp(&r_patt, tp->target_pattern, REG_EXTENDED|REG_NOSUB)) { crm_info("Bad regex '%s' for fencing level", tp->target); } else { int status = regexec(&r_patt, node, 0, NULL, 0); regfree(&r_patt); if (status == 0) { crm_notice("Matched %s with %s by name", node, tp->target); return TRUE; } } break; case fenced_target_by_name: crm_trace("Testing %s against %s", node, tp->target); return pcmk__str_eq(tp->target, node, pcmk__str_casei); default: break; } crm_trace("No match for %s with %s", node, tp->target); return FALSE; } stonith_topology_t * find_topology_for_host(const char *host) { GHashTableIter tIter; stonith_topology_t *tp = g_hash_table_lookup(topology, host); if(tp != NULL) { crm_trace("Found %s for %s in %d entries", tp->target, host, g_hash_table_size(topology)); return tp; } g_hash_table_iter_init(&tIter, topology); while (g_hash_table_iter_next(&tIter, NULL, (gpointer *) & tp)) { if (topology_matches(tp, host)) { crm_trace("Found %s for %s in %d entries", tp->target, host, g_hash_table_size(topology)); return tp; } } crm_trace("No matches for %s in %d topology entries", host, g_hash_table_size(topology)); return NULL; } /*! * \internal * \brief Set fencing operation's device list to target's next topology level * * \param[in,out] op Remote fencing operation to modify * \param[in] empty_ok If true, an operation without a target (i.e. * queries) or a target without a topology will get a * pcmk_rc_ok return value instead of ENODEV * * \return Standard Pacemaker return value */ static int advance_topology_level(remote_fencing_op_t *op, bool empty_ok) { stonith_topology_t *tp = NULL; if (op->target) { tp = find_topology_for_host(op->target); } if (topology_is_empty(tp)) { return empty_ok? pcmk_rc_ok : ENODEV; } CRM_ASSERT(tp->levels != NULL); stonith__set_call_options(op->call_options, op->id, st_opt_topology); /* This is a new level, so undo any remapping left over from previous */ undo_op_remap(op); do { op->level++; } while (op->level < ST_LEVEL_MAX && tp->levels[op->level] == NULL); if (op->level < ST_LEVEL_MAX) { crm_trace("Attempting fencing level %d targeting %s (%d devices) " "for client %s@%s (id=%.8s)", op->level, op->target, g_list_length(tp->levels[op->level]), op->client_name, op->originator, op->id); set_op_device_list(op, tp->levels[op->level]); // The requested delay has been applied for the first fencing level if ((op->level > 1) && (op->client_delay > 0)) { op->client_delay = 0; } if ((g_list_next(op->devices_list) != NULL) && pcmk__str_eq(op->action, PCMK_ACTION_REBOOT, pcmk__str_none)) { /* A reboot has been requested for a topology level with multiple * devices. Instead of rebooting the devices sequentially, we will * turn them all off, then turn them all on again. (Think about * switched power outlets for redundant power supplies.) */ op_phase_off(op); } return pcmk_rc_ok; } crm_info("All %sfencing options targeting %s for client %s@%s failed " QB_XS " id=%.8s", (stonith_watchdog_timeout_ms > 0)?"non-watchdog ":"", op->target, op->client_name, op->originator, op->id); return ENODEV; } /*! * \internal * \brief If fencing operation is a duplicate, merge it into the other one * * \param[in,out] op Fencing operation to check */ static void merge_duplicates(remote_fencing_op_t *op) { GHashTableIter iter; remote_fencing_op_t *other = NULL; time_t now = time(NULL); g_hash_table_iter_init(&iter, stonith_remote_op_list); while (g_hash_table_iter_next(&iter, NULL, (void **)&other)) { const char *other_action = op_requested_action(other); pcmk__node_status_t *node = NULL; if (!strcmp(op->id, other->id)) { continue; // Don't compare against self } if (other->state > st_exec) { crm_trace("%.8s not duplicate of %.8s: not in progress", op->id, other->id); continue; } if (!pcmk__str_eq(op->target, other->target, pcmk__str_casei)) { crm_trace("%.8s not duplicate of %.8s: node %s vs. %s", op->id, other->id, op->target, other->target); continue; } if (!pcmk__str_eq(op->action, other_action, pcmk__str_none)) { crm_trace("%.8s not duplicate of %.8s: action %s vs. %s", op->id, other->id, op->action, other_action); continue; } if (pcmk__str_eq(op->client_name, other->client_name, pcmk__str_casei)) { crm_trace("%.8s not duplicate of %.8s: same client %s", op->id, other->id, op->client_name); continue; } if (pcmk__str_eq(other->target, other->originator, pcmk__str_casei)) { crm_trace("%.8s not duplicate of %.8s: suicide for %s", op->id, other->id, other->target); continue; } node = pcmk__get_node(0, other->originator, NULL, pcmk__node_search_cluster_member); if (!fencing_peer_active(node)) { crm_notice("Failing action '%s' targeting %s originating from " "client %s@%s: Originator is dead " QB_XS " id=%.8s", other->action, other->target, other->client_name, other->originator, other->id); crm_trace("%.8s not duplicate of %.8s: originator dead", op->id, other->id); other->state = st_failed; continue; } if ((other->total_timeout > 0) && (now > (other->total_timeout + other->created))) { crm_trace("%.8s not duplicate of %.8s: old (%lld vs. %lld + %ds)", op->id, other->id, (long long)now, (long long)other->created, other->total_timeout); continue; } /* There is another in-flight request to fence the same host * Piggyback on that instead. If it fails, so do we. */ other->duplicates = g_list_append(other->duplicates, op); if (other->total_timeout == 0) { other->total_timeout = op->total_timeout = TIMEOUT_MULTIPLY_FACTOR * get_op_total_timeout(op, NULL); crm_trace("Best guess as to timeout used for %.8s: %ds", other->id, other->total_timeout); } crm_notice("Merging fencing action '%s' targeting %s originating from " "client %s with identical request from %s@%s " QB_XS " original=%.8s duplicate=%.8s total_timeout=%ds", op->action, op->target, op->client_name, other->client_name, other->originator, op->id, other->id, other->total_timeout); report_timeout_period(op, other->total_timeout); op->state = st_duplicate; } } static uint32_t fencing_active_peers(void) { uint32_t count = 0; pcmk__node_status_t *entry = NULL; GHashTableIter gIter; g_hash_table_iter_init(&gIter, crm_peer_cache); while (g_hash_table_iter_next(&gIter, NULL, (void **)&entry)) { if(fencing_peer_active(entry)) { count++; } } return count; } /*! * \internal * \brief Process a manual confirmation of a pending fence action * * \param[in] client IPC client that sent confirmation * \param[in,out] msg Request XML with manual confirmation * * \return Standard Pacemaker return code */ int fenced_handle_manual_confirmation(const pcmk__client_t *client, xmlNode *msg) { remote_fencing_op_t *op = NULL; xmlNode *dev = get_xpath_object("//@" PCMK__XA_ST_TARGET, msg, LOG_ERR); CRM_CHECK(dev != NULL, return EPROTO); crm_notice("Received manual confirmation that %s has been fenced", pcmk__s(crm_element_value(dev, PCMK__XA_ST_TARGET), "unknown target")); op = initiate_remote_stonith_op(client, msg, TRUE); if (op == NULL) { return EPROTO; } op->state = st_done; set_fencing_completed(op); op->delegate = pcmk__str_copy("a human"); // For the fencer's purposes, the fencing operation is done pcmk__set_result(&op->result, CRM_EX_OK, PCMK_EXEC_DONE, NULL); finalize_op(op, msg, false); /* For the requester's purposes, the operation is still pending. The * actual result will be sent asynchronously via the operation's done_cb(). */ return EINPROGRESS; } /*! * \internal * \brief Create a new remote stonith operation * * \param[in] client ID of local stonith client that initiated the operation * \param[in] request The request from the client that started the operation * \param[in] peer TRUE if this operation is owned by another stonith peer * (an operation owned by one peer is stored on all peers, * but only the owner executes it; all nodes get the results * once the owner finishes execution) */ void * create_remote_stonith_op(const char *client, xmlNode *request, gboolean peer) { remote_fencing_op_t *op = NULL; xmlNode *dev = get_xpath_object("//@" PCMK__XA_ST_TARGET, request, LOG_NEVER); int call_options = 0; const char *operation = NULL; init_stonith_remote_op_hash_table(&stonith_remote_op_list); /* If this operation is owned by another node, check to make * sure we haven't already created this operation. */ if (peer && dev) { const char *op_id = crm_element_value(dev, PCMK__XA_ST_REMOTE_OP); CRM_CHECK(op_id != NULL, return NULL); op = g_hash_table_lookup(stonith_remote_op_list, op_id); if (op) { crm_debug("Reusing existing remote fencing op %.8s for %s", op_id, ((client == NULL)? "unknown client" : client)); return op; } } op = pcmk__assert_alloc(1, sizeof(remote_fencing_op_t)); crm_element_value_int(request, PCMK__XA_ST_TIMEOUT, &(op->base_timeout)); // Value -1 means disable any static/random fencing delays crm_element_value_int(request, PCMK__XA_ST_DELAY, &(op->client_delay)); if (peer && dev) { op->id = crm_element_value_copy(dev, PCMK__XA_ST_REMOTE_OP); } else { op->id = crm_generate_uuid(); } g_hash_table_replace(stonith_remote_op_list, op->id, op); op->state = st_query; op->replies_expected = fencing_active_peers(); op->action = crm_element_value_copy(dev, PCMK__XA_ST_DEVICE_ACTION); /* The node initiating the stonith operation. If an operation is relayed, * this is the last node the operation lands on. When in standalone mode, * origin is the ID of the client that originated the operation. * * Or may be the name of the function that created the operation. */ op->originator = crm_element_value_copy(dev, PCMK__XA_ST_ORIGIN); if (op->originator == NULL) { /* Local or relayed request */ op->originator = pcmk__str_copy(stonith_our_uname); } // Delegate may not be set op->delegate = crm_element_value_copy(dev, PCMK__XA_ST_DELEGATE); op->created = time(NULL); CRM_LOG_ASSERT(client != NULL); op->client_id = pcmk__str_copy(client); /* For a RELAY operation, set fenced on the client. */ operation = crm_element_value(request, PCMK__XA_ST_OP); if (pcmk__str_eq(operation, STONITH_OP_RELAY, pcmk__str_none)) { op->client_name = crm_strdup_printf("%s.%lu", crm_system_name, (unsigned long) getpid()); } else { op->client_name = crm_element_value_copy(request, PCMK__XA_ST_CLIENTNAME); } op->target = crm_element_value_copy(dev, PCMK__XA_ST_TARGET); // @TODO Figure out how to avoid copying XML here op->request = pcmk__xml_copy(NULL, request); crm_element_value_int(request, PCMK__XA_ST_CALLOPT, &call_options); op->call_options = call_options; crm_element_value_int(request, PCMK__XA_ST_CALLID, &(op->client_callid)); crm_trace("%s new fencing op %s ('%s' targeting %s for client %s, " "base timeout %ds, %u %s expected)", (peer && dev)? "Recorded" : "Generated", op->id, op->action, op->target, op->client_name, op->base_timeout, op->replies_expected, pcmk__plural_alt(op->replies_expected, "reply", "replies")); if (op->call_options & st_opt_cs_nodeid) { int nodeid; pcmk__node_status_t *node = NULL; pcmk__scan_min_int(op->target, &nodeid, 0); node = pcmk__search_node_caches(nodeid, NULL, pcmk__node_search_any |pcmk__node_search_cluster_cib); /* Ensure the conversion only happens once */ stonith__clear_call_options(op->call_options, op->id, st_opt_cs_nodeid); if ((node != NULL) && (node->name != NULL)) { pcmk__str_update(&(op->target), node->name); } else { crm_warn("Could not expand nodeid '%s' into a host name", op->target); } } /* check to see if this is a duplicate operation of another in-flight operation */ merge_duplicates(op); if (op->state != st_duplicate) { /* kick history readers */ fenced_send_notification(PCMK__VALUE_ST_NOTIFY_HISTORY, NULL, NULL); } /* safe to trim as long as that doesn't touch pending ops */ stonith_fence_history_trim(); return op; } /*! * \internal * \brief Create a peer fencing operation from a request, and initiate it * * \param[in] client IPC client that made request (NULL to get from request) * \param[in] request Request XML * \param[in] manual_ack Whether this is a manual action confirmation * * \return Newly created operation on success, otherwise NULL */ remote_fencing_op_t * initiate_remote_stonith_op(const pcmk__client_t *client, xmlNode *request, gboolean manual_ack) { int query_timeout = 0; xmlNode *query = NULL; const char *client_id = NULL; remote_fencing_op_t *op = NULL; const char *relay_op_id = NULL; const char *operation = NULL; if (client) { client_id = client->id; } else { client_id = crm_element_value(request, PCMK__XA_ST_CLIENTID); } CRM_LOG_ASSERT(client_id != NULL); op = create_remote_stonith_op(client_id, request, FALSE); op->owner = TRUE; if (manual_ack) { return op; } CRM_CHECK(op->action, return NULL); if (advance_topology_level(op, true) != pcmk_rc_ok) { op->state = st_failed; } switch (op->state) { case st_failed: // advance_topology_level() exhausted levels pcmk__set_result(&op->result, CRM_EX_ERROR, PCMK_EXEC_ERROR, "All topology levels failed"); crm_warn("Could not request peer fencing (%s) targeting %s " QB_XS " id=%.8s", op->action, op->target, op->id); finalize_op(op, NULL, false); return op; case st_duplicate: crm_info("Requesting peer fencing (%s) targeting %s (duplicate) " QB_XS " id=%.8s", op->action, op->target, op->id); return op; default: crm_notice("Requesting peer fencing (%s) targeting %s " QB_XS " id=%.8s state=%s base_timeout=%ds", op->action, op->target, op->id, stonith_op_state_str(op->state), op->base_timeout); } query = stonith_create_op(op->client_callid, op->id, STONITH_OP_QUERY, NULL, op->call_options); crm_xml_add(query, PCMK__XA_ST_REMOTE_OP, op->id); crm_xml_add(query, PCMK__XA_ST_TARGET, op->target); crm_xml_add(query, PCMK__XA_ST_DEVICE_ACTION, op_requested_action(op)); crm_xml_add(query, PCMK__XA_ST_ORIGIN, op->originator); crm_xml_add(query, PCMK__XA_ST_CLIENTID, op->client_id); crm_xml_add(query, PCMK__XA_ST_CLIENTNAME, op->client_name); crm_xml_add_int(query, PCMK__XA_ST_TIMEOUT, op->base_timeout); /* In case of RELAY operation, RELAY information is added to the query to delete the original operation of RELAY. */ operation = crm_element_value(request, PCMK__XA_ST_OP); if (pcmk__str_eq(operation, STONITH_OP_RELAY, pcmk__str_none)) { relay_op_id = crm_element_value(request, PCMK__XA_ST_REMOTE_OP); if (relay_op_id) { crm_xml_add(query, PCMK__XA_ST_REMOTE_OP_RELAY, relay_op_id); } } - pcmk__cluster_send_message(NULL, crm_msg_stonith_ng, query); + pcmk__cluster_send_message(NULL, pcmk__cluster_msg_fenced, query); pcmk__xml_free(query); query_timeout = op->base_timeout * TIMEOUT_MULTIPLY_FACTOR; op->query_timer = g_timeout_add((1000 * query_timeout), remote_op_query_timeout, op); return op; } enum find_best_peer_options { /*! Skip checking the target peer for capable fencing devices */ FIND_PEER_SKIP_TARGET = 0x0001, /*! Only check the target peer for capable fencing devices */ FIND_PEER_TARGET_ONLY = 0x0002, /*! Skip peers and devices that are not verified */ FIND_PEER_VERIFIED_ONLY = 0x0004, }; static bool is_watchdog_fencing(const remote_fencing_op_t *op, const char *device) { return (stonith_watchdog_timeout_ms > 0 // Only an explicit mismatch is considered not a watchdog fencing. && pcmk__str_eq(device, STONITH_WATCHDOG_ID, pcmk__str_null_matches) && pcmk__is_fencing_action(op->action) && node_does_watchdog_fencing(op->target)); } static peer_device_info_t * find_best_peer(const char *device, remote_fencing_op_t * op, enum find_best_peer_options options) { GList *iter = NULL; gboolean verified_devices_only = (options & FIND_PEER_VERIFIED_ONLY) ? TRUE : FALSE; if (!device && pcmk_is_set(op->call_options, st_opt_topology)) { return NULL; } for (iter = op->query_results; iter != NULL; iter = iter->next) { peer_device_info_t *peer = iter->data; crm_trace("Testing result from %s targeting %s with %d device%s: %d %x", peer->host, op->target, peer->ndevices, pcmk__plural_s(peer->ndevices), peer->tried, options); if ((options & FIND_PEER_SKIP_TARGET) && pcmk__str_eq(peer->host, op->target, pcmk__str_casei)) { continue; } if ((options & FIND_PEER_TARGET_ONLY) && !pcmk__str_eq(peer->host, op->target, pcmk__str_casei)) { continue; } if (pcmk_is_set(op->call_options, st_opt_topology)) { if (grab_peer_device(op, peer, device, verified_devices_only)) { return peer; } } else if (!peer->tried && count_peer_devices(op, peer, verified_devices_only, fenced_support_flag(op->action))) { /* No topology: Use the current best peer */ crm_trace("Simple fencing"); return peer; } } return NULL; } static peer_device_info_t * stonith_choose_peer(remote_fencing_op_t * op) { const char *device = NULL; peer_device_info_t *peer = NULL; uint32_t active = fencing_active_peers(); do { if (op->devices) { device = op->devices->data; crm_trace("Checking for someone to fence (%s) %s using %s", op->action, op->target, device); } else { crm_trace("Checking for someone to fence (%s) %s", op->action, op->target); } /* Best choice is a peer other than the target with verified access */ peer = find_best_peer(device, op, FIND_PEER_SKIP_TARGET|FIND_PEER_VERIFIED_ONLY); if (peer) { crm_trace("Found verified peer %s for %s", peer->host, device?device:""); return peer; } if(op->query_timer != 0 && op->replies < QB_MIN(op->replies_expected, active)) { crm_trace("Waiting before looking for unverified devices to fence %s", op->target); return NULL; } /* If no other peer has verified access, next best is unverified access */ peer = find_best_peer(device, op, FIND_PEER_SKIP_TARGET); if (peer) { crm_trace("Found best unverified peer %s", peer->host); return peer; } /* If no other peer can do it, last option is self-fencing * (which is never allowed for the "on" phase of a remapped reboot) */ if (op->phase != st_phase_on) { peer = find_best_peer(device, op, FIND_PEER_TARGET_ONLY); if (peer) { crm_trace("%s will fence itself", peer->host); return peer; } } /* Try the next fencing level if there is one (unless we're in the "on" * phase of a remapped "reboot", because we ignore errors in that case) */ } while ((op->phase != st_phase_on) && pcmk_is_set(op->call_options, st_opt_topology) && (advance_topology_level(op, false) == pcmk_rc_ok)); /* With a simple watchdog fencing configuration without a topology, * "device" is NULL here. Consider it should be done with watchdog fencing. */ if (is_watchdog_fencing(op, device)) { crm_info("Couldn't contact watchdog-fencing target-node (%s)", op->target); /* check_watchdog_fencing_and_wait will log additional info */ } else { crm_notice("Couldn't find anyone to fence (%s) %s using %s", op->action, op->target, (device? device : "any device")); } return NULL; } static int valid_fencing_timeout(int specified_timeout, bool action_specific, const remote_fencing_op_t *op, const char *device) { int timeout = specified_timeout; if (!is_watchdog_fencing(op, device)) { return timeout; } timeout = (int) QB_MIN(QB_MAX(specified_timeout, stonith_watchdog_timeout_ms / 1000), INT_MAX); if (timeout > specified_timeout) { if (action_specific) { crm_warn("pcmk_%s_timeout %ds for %s is too short (must be >= " PCMK_OPT_STONITH_WATCHDOG_TIMEOUT " %ds), using %ds " "instead", op->action, specified_timeout, device? device : "watchdog", timeout, timeout); } else { crm_warn("Fencing timeout %ds is too short (must be >= " PCMK_OPT_STONITH_WATCHDOG_TIMEOUT " %ds), using %ds " "instead", specified_timeout, timeout, timeout); } } return timeout; } static int get_device_timeout(const remote_fencing_op_t *op, const peer_device_info_t *peer, const char *device, bool with_delay) { int timeout = op->base_timeout; device_properties_t *props; timeout = valid_fencing_timeout(op->base_timeout, false, op, device); if (!peer || !device) { return timeout; } props = g_hash_table_lookup(peer->devices, device); if (!props) { return timeout; } if (props->custom_action_timeout[op->phase]) { timeout = valid_fencing_timeout(props->custom_action_timeout[op->phase], true, op, device); } // op->client_delay < 0 means disable any static/random fencing delays if (with_delay && (op->client_delay >= 0)) { // delay_base is eventually limited by delay_max timeout += (props->delay_max[op->phase] > 0 ? props->delay_max[op->phase] : props->delay_base[op->phase]); } return timeout; } struct timeout_data { const remote_fencing_op_t *op; const peer_device_info_t *peer; int total_timeout; }; /*! * \internal * \brief Add timeout to a total if device has not been executed yet * * \param[in] key GHashTable key (device ID) * \param[in] value GHashTable value (device properties) * \param[in,out] user_data Timeout data */ static void add_device_timeout(gpointer key, gpointer value, gpointer user_data) { const char *device_id = key; device_properties_t *props = value; struct timeout_data *timeout = user_data; if (!props->executed[timeout->op->phase] && !props->disallowed[timeout->op->phase]) { timeout->total_timeout += get_device_timeout(timeout->op, timeout->peer, device_id, true); } } static int get_peer_timeout(const remote_fencing_op_t *op, const peer_device_info_t *peer) { struct timeout_data timeout; timeout.op = op; timeout.peer = peer; timeout.total_timeout = 0; g_hash_table_foreach(peer->devices, add_device_timeout, &timeout); return (timeout.total_timeout? timeout.total_timeout : op->base_timeout); } static int get_op_total_timeout(const remote_fencing_op_t *op, const peer_device_info_t *chosen_peer) { long long total_timeout = 0; stonith_topology_t *tp = find_topology_for_host(op->target); if (pcmk_is_set(op->call_options, st_opt_topology) && tp) { int i; GList *device_list = NULL; GList *iter = NULL; GList *auto_list = NULL; if (pcmk__str_eq(op->action, PCMK_ACTION_ON, pcmk__str_none) && (op->automatic_list != NULL)) { auto_list = g_list_copy(op->automatic_list); } /* Yep, this looks scary, nested loops all over the place. * Here is what is going on. * Loop1: Iterate through fencing levels. * Loop2: If a fencing level has devices, loop through each device * Loop3: For each device in a fencing level, see what peer owns it * and what that peer has reported the timeout is for the device. */ for (i = 0; i < ST_LEVEL_MAX; i++) { if (!tp->levels[i]) { continue; } for (device_list = tp->levels[i]; device_list; device_list = device_list->next) { bool found = false; for (iter = op->query_results; iter != NULL; iter = iter->next) { const peer_device_info_t *peer = iter->data; if (auto_list) { GList *match = g_list_find_custom(auto_list, device_list->data, sort_strings); if (match) { auto_list = g_list_remove(auto_list, match->data); } } if (find_peer_device(op, peer, device_list->data, fenced_support_flag(op->action))) { total_timeout += get_device_timeout(op, peer, device_list->data, true); found = true; break; } } /* End Loop3: match device with peer that owns device, find device's timeout period */ /* in case of watchdog-device we add the timeout to the budget if didn't get a reply */ if (!found && is_watchdog_fencing(op, device_list->data)) { total_timeout += stonith_watchdog_timeout_ms / 1000; } } /* End Loop2: iterate through devices at a specific level */ } /*End Loop1: iterate through fencing levels */ //Add only exists automatic_list device timeout if (auto_list) { for (iter = auto_list; iter != NULL; iter = iter->next) { GList *iter2 = NULL; for (iter2 = op->query_results; iter2 != NULL; iter = iter2->next) { peer_device_info_t *peer = iter2->data; if (find_peer_device(op, peer, iter->data, st_device_supports_on)) { total_timeout += get_device_timeout(op, peer, iter->data, true); break; } } } } g_list_free(auto_list); } else if (chosen_peer) { total_timeout = get_peer_timeout(op, chosen_peer); } else { total_timeout = valid_fencing_timeout(op->base_timeout, false, op, NULL); } if (total_timeout <= 0) { total_timeout = op->base_timeout; } /* Take any requested fencing delay into account to prevent it from eating * up the total timeout. */ if (op->client_delay > 0) { total_timeout += op->client_delay; } return (int) QB_MIN(total_timeout, INT_MAX); } static void report_timeout_period(remote_fencing_op_t * op, int op_timeout) { GList *iter = NULL; xmlNode *update = NULL; const char *client_node = NULL; const char *client_id = NULL; const char *call_id = NULL; if (op->call_options & st_opt_sync_call) { /* There is no reason to report the timeout for a synchronous call. It * is impossible to use the reported timeout to do anything when the client * is blocking for the response. This update is only important for * async calls that require a callback to report the results in. */ return; } else if (!op->request) { return; } crm_trace("Reporting timeout for %s (id=%.8s)", op->client_name, op->id); client_node = crm_element_value(op->request, PCMK__XA_ST_CLIENTNODE); call_id = crm_element_value(op->request, PCMK__XA_ST_CALLID); client_id = crm_element_value(op->request, PCMK__XA_ST_CLIENTID); if (!client_node || !call_id || !client_id) { return; } if (pcmk__str_eq(client_node, stonith_our_uname, pcmk__str_casei)) { // Client is connected to this node, so send update directly to them do_stonith_async_timeout_update(client_id, call_id, op_timeout); return; } /* The client is connected to another node, relay this update to them */ update = stonith_create_op(op->client_callid, op->id, STONITH_OP_TIMEOUT_UPDATE, NULL, 0); crm_xml_add(update, PCMK__XA_ST_REMOTE_OP, op->id); crm_xml_add(update, PCMK__XA_ST_CLIENTID, client_id); crm_xml_add(update, PCMK__XA_ST_CALLID, call_id); crm_xml_add_int(update, PCMK__XA_ST_TIMEOUT, op_timeout); pcmk__cluster_send_message(pcmk__get_node(0, client_node, NULL, pcmk__node_search_cluster_member), - crm_msg_stonith_ng, update); + pcmk__cluster_msg_fenced, update); pcmk__xml_free(update); for (iter = op->duplicates; iter != NULL; iter = iter->next) { remote_fencing_op_t *dup = iter->data; crm_trace("Reporting timeout for duplicate %.8s to client %s", dup->id, dup->client_name); report_timeout_period(iter->data, op_timeout); } } /*! * \internal * \brief Advance an operation to the next device in its topology * * \param[in,out] op Fencer operation to advance * \param[in] device ID of device that just completed * \param[in,out] msg If not NULL, XML reply of last delegated operation */ static void advance_topology_device_in_level(remote_fencing_op_t *op, const char *device, xmlNode *msg) { /* Advance to the next device at this topology level, if any */ if (op->devices) { op->devices = op->devices->next; } /* Handle automatic unfencing if an "on" action was requested */ if ((op->phase == st_phase_requested) && pcmk__str_eq(op->action, PCMK_ACTION_ON, pcmk__str_none)) { /* If the device we just executed was required, it's not anymore */ remove_required_device(op, device); /* If there are no more devices at this topology level, run through any * remaining devices with automatic unfencing */ if (op->devices == NULL) { op->devices = op->automatic_list; } } if ((op->devices == NULL) && (op->phase == st_phase_off)) { /* We're done with this level and with required devices, but we had * remapped "reboot" to "off", so start over with "on". If any devices * need to be turned back on, op->devices will be non-NULL after this. */ op_phase_on(op); } // This function is only called if the previous device succeeded pcmk__set_result(&op->result, CRM_EX_OK, PCMK_EXEC_DONE, NULL); if (op->devices) { /* Necessary devices remain, so execute the next one */ crm_trace("Next targeting %s on behalf of %s@%s", op->target, op->client_name, op->originator); // The requested delay has been applied for the first device if (op->client_delay > 0) { op->client_delay = 0; } request_peer_fencing(op, NULL); } else { /* We're done with all devices and phases, so finalize operation */ crm_trace("Marking complex fencing op targeting %s as complete", op->target); op->state = st_done; finalize_op(op, msg, false); } } static gboolean check_watchdog_fencing_and_wait(remote_fencing_op_t * op) { if (node_does_watchdog_fencing(op->target)) { guint timeout_ms = QB_MIN(stonith_watchdog_timeout_ms, UINT_MAX); crm_notice("Waiting %s for %s to self-fence (%s) for " "client %s " QB_XS " id=%.8s", pcmk__readable_interval(timeout_ms), op->target, op->action, op->client_name, op->id); if (op->op_timer_one) { g_source_remove(op->op_timer_one); } op->op_timer_one = g_timeout_add(timeout_ms, remote_op_watchdog_done, op); return TRUE; } else { crm_debug("Skipping fallback to watchdog-fencing as %s is " "not in host-list", op->target); } return FALSE; } /*! * \internal * \brief Ask a peer to execute a fencing operation * * \param[in,out] op Fencing operation to be executed * \param[in,out] peer If NULL or topology is in use, choose best peer to * execute the fencing, otherwise use this peer */ static void request_peer_fencing(remote_fencing_op_t *op, peer_device_info_t *peer) { const char *device = NULL; int timeout; CRM_CHECK(op != NULL, return); crm_trace("Action %.8s targeting %s for %s is %s", op->id, op->target, op->client_name, stonith_op_state_str(op->state)); if ((op->phase == st_phase_on) && (op->devices != NULL)) { /* We are in the "on" phase of a remapped topology reboot. If this * device has pcmk_reboot_action="off", or doesn't support the "on" * action, skip it. * * We can't check device properties at this point because we haven't * chosen a peer for this stage yet. Instead, we check the local node's * knowledge about the device. If different versions of the fence agent * are installed on different nodes, there's a chance this could be * mistaken, but the worst that could happen is we don't try turning the * node back on when we should. */ device = op->devices->data; if (pcmk__str_eq(fenced_device_reboot_action(device), PCMK_ACTION_OFF, pcmk__str_none)) { crm_info("Not turning %s back on using %s because the device is " "configured to stay off (pcmk_reboot_action='off')", op->target, device); advance_topology_device_in_level(op, device, NULL); return; } if (!fenced_device_supports_on(device)) { crm_info("Not turning %s back on using %s because the agent " "doesn't support 'on'", op->target, device); advance_topology_device_in_level(op, device, NULL); return; } } timeout = op->base_timeout; if ((peer == NULL) && !pcmk_is_set(op->call_options, st_opt_topology)) { peer = stonith_choose_peer(op); } if (!op->op_timer_total) { op->total_timeout = TIMEOUT_MULTIPLY_FACTOR * get_op_total_timeout(op, peer); op->op_timer_total = g_timeout_add(1000 * op->total_timeout, remote_op_timeout, op); report_timeout_period(op, op->total_timeout); crm_info("Total timeout set to %ds for peer's fencing targeting %s for %s" QB_XS "id=%.8s", op->total_timeout, op->target, op->client_name, op->id); } if (pcmk_is_set(op->call_options, st_opt_topology) && op->devices) { /* Ignore the caller's peer preference if topology is in use, because * that peer might not have access to the required device. With * topology, stonith_choose_peer() removes the device from further * consideration, so the timeout must be calculated beforehand. * * @TODO Basing the total timeout on the caller's preferred peer (above) * is less than ideal. */ peer = stonith_choose_peer(op); device = op->devices->data; /* Fencing timeout sent to peer takes no delay into account. * The peer will add a dedicated timer for any delay upon * schedule_stonith_command(). */ timeout = get_device_timeout(op, peer, device, false); } if (peer) { int timeout_one = 0; xmlNode *remote_op = stonith_create_op(op->client_callid, op->id, STONITH_OP_FENCE, NULL, 0); const pcmk__node_status_t *peer_node = pcmk__get_node(0, peer->host, NULL, pcmk__node_search_cluster_member); if (op->client_delay > 0) { /* Take requested fencing delay into account to prevent it from * eating up the timeout. */ timeout_one = TIMEOUT_MULTIPLY_FACTOR * op->client_delay; } crm_xml_add(remote_op, PCMK__XA_ST_REMOTE_OP, op->id); crm_xml_add(remote_op, PCMK__XA_ST_TARGET, op->target); crm_xml_add(remote_op, PCMK__XA_ST_DEVICE_ACTION, op->action); crm_xml_add(remote_op, PCMK__XA_ST_ORIGIN, op->originator); crm_xml_add(remote_op, PCMK__XA_ST_CLIENTID, op->client_id); crm_xml_add(remote_op, PCMK__XA_ST_CLIENTNAME, op->client_name); crm_xml_add_int(remote_op, PCMK__XA_ST_TIMEOUT, timeout); crm_xml_add_int(remote_op, PCMK__XA_ST_CALLOPT, op->call_options); crm_xml_add_int(remote_op, PCMK__XA_ST_DELAY, op->client_delay); if (device) { timeout_one += TIMEOUT_MULTIPLY_FACTOR * get_device_timeout(op, peer, device, true); crm_notice("Requesting that %s perform '%s' action targeting %s " "using %s " QB_XS " for client %s (%ds)", peer->host, op->action, op->target, device, op->client_name, timeout_one); crm_xml_add(remote_op, PCMK__XA_ST_DEVICE_ID, device); } else { timeout_one += TIMEOUT_MULTIPLY_FACTOR * get_peer_timeout(op, peer); crm_notice("Requesting that %s perform '%s' action targeting %s " QB_XS " for client %s (%ds, %s)", peer->host, op->action, op->target, op->client_name, timeout_one, pcmk__readable_interval(stonith_watchdog_timeout_ms)); } op->state = st_exec; if (op->op_timer_one) { g_source_remove(op->op_timer_one); op->op_timer_one = 0; } if (!is_watchdog_fencing(op, device) || !check_watchdog_fencing_and_wait(op)) { /* Some thoughts about self-fencing cases reaching this point: - Actually check in check_watchdog_fencing_and_wait shouldn't fail if STONITH_WATCHDOG_ID is chosen as fencing-device and it being present implies watchdog-fencing is enabled anyway - If watchdog-fencing is disabled either in general or for a specific target - detected in check_watchdog_fencing_and_wait - for some other kind of self-fencing we can't expect a success answer but timeout is fine if the node doesn't come back in between - Delicate might be the case where we have watchdog-fencing enabled for a node but the watchdog-fencing-device isn't explicitly chosen for suicide. Local pe-execution in sbd may detect the node as unclean and lead to timely suicide. Otherwise the selection of PCMK_OPT_STONITH_WATCHDOG_TIMEOUT at least is questionable. */ /* coming here we're not waiting for watchdog timeout - thus engage timer with timout evaluated before */ op->op_timer_one = g_timeout_add((1000 * timeout_one), remote_op_timeout_one, op); } - pcmk__cluster_send_message(peer_node, crm_msg_stonith_ng, remote_op); + pcmk__cluster_send_message(peer_node, pcmk__cluster_msg_fenced, + remote_op); peer->tried = TRUE; pcmk__xml_free(remote_op); return; } else if (op->phase == st_phase_on) { /* A remapped "on" cannot be executed, but the node was already * turned off successfully, so ignore the error and continue. */ crm_warn("Ignoring %s 'on' failure (no capable peers) targeting %s " "after successful 'off'", device, op->target); advance_topology_device_in_level(op, device, NULL); return; } else if (op->owner == FALSE) { crm_err("Fencing (%s) targeting %s for client %s is not ours to control", op->action, op->target, op->client_name); } else if (op->query_timer == 0) { /* We've exhausted all available peers */ crm_info("No remaining peers capable of fencing (%s) %s for client %s " QB_XS " state=%s", op->action, op->target, op->client_name, stonith_op_state_str(op->state)); CRM_CHECK(op->state < st_done, return); finalize_timed_out_op(op, "All nodes failed, or are unable, to " "fence target"); } else if(op->replies >= op->replies_expected || op->replies >= fencing_active_peers()) { /* if the operation never left the query state, * but we have all the expected replies, then no devices * are available to execute the fencing operation. */ if (is_watchdog_fencing(op, device) && check_watchdog_fencing_and_wait(op)) { /* Consider a watchdog fencing targeting an offline node executing * once it starts waiting for the target to self-fence. So that when * the query timer pops, remote_op_query_timeout() considers the * fencing already in progress. */ op->state = st_exec; return; } if (op->state == st_query) { crm_info("No peers (out of %d) have devices capable of fencing " "(%s) %s for client %s " QB_XS " state=%s", op->replies, op->action, op->target, op->client_name, stonith_op_state_str(op->state)); pcmk__reset_result(&op->result); pcmk__set_result(&op->result, CRM_EX_ERROR, PCMK_EXEC_NO_FENCE_DEVICE, NULL); } else { if (pcmk_is_set(op->call_options, st_opt_topology)) { pcmk__reset_result(&op->result); pcmk__set_result(&op->result, CRM_EX_ERROR, PCMK_EXEC_NO_FENCE_DEVICE, NULL); } /* ... else use existing result from previous failed attempt * (topology is not in use, and no devices remain to be attempted). * Overwriting the result with PCMK_EXEC_NO_FENCE_DEVICE would * prevent finalize_op() from setting the correct delegate if * needed. */ crm_info("No peers (out of %d) are capable of fencing (%s) %s " "for client %s " QB_XS " state=%s", op->replies, op->action, op->target, op->client_name, stonith_op_state_str(op->state)); } op->state = st_failed; finalize_op(op, NULL, false); } else { crm_info("Waiting for additional peers capable of fencing (%s) %s%s%s " "for client %s " QB_XS " id=%.8s", op->action, op->target, (device? " using " : ""), (device? device : ""), op->client_name, op->id); } } /*! * \internal * \brief Comparison function for sorting query results * * \param[in] a GList item to compare * \param[in] b GList item to compare * * \return Per the glib documentation, "a negative integer if the first value * comes before the second, 0 if they are equal, or a positive integer * if the first value comes after the second." */ static gint sort_peers(gconstpointer a, gconstpointer b) { const peer_device_info_t *peer_a = a; const peer_device_info_t *peer_b = b; return (peer_b->ndevices - peer_a->ndevices); } /*! * \internal * \brief Determine if all the devices in the topology are found or not * * \param[in] op Fencing operation with topology to check */ static gboolean all_topology_devices_found(const remote_fencing_op_t *op) { GList *device = NULL; GList *iter = NULL; device_properties_t *match = NULL; stonith_topology_t *tp = NULL; gboolean skip_target = FALSE; int i; tp = find_topology_for_host(op->target); if (!tp) { return FALSE; } if (pcmk__is_fencing_action(op->action)) { /* Don't count the devices on the target node if we are killing * the target node. */ skip_target = TRUE; } for (i = 0; i < ST_LEVEL_MAX; i++) { for (device = tp->levels[i]; device; device = device->next) { match = NULL; for (iter = op->query_results; iter && !match; iter = iter->next) { peer_device_info_t *peer = iter->data; if (skip_target && pcmk__str_eq(peer->host, op->target, pcmk__str_casei)) { continue; } match = find_peer_device(op, peer, device->data, st_device_supports_none); } if (!match) { return FALSE; } } } return TRUE; } /*! * \internal * \brief Parse action-specific device properties from XML * * \param[in] xml XML element containing the properties * \param[in] peer Name of peer that sent XML (for logs) * \param[in] device Device ID (for logs) * \param[in] action Action the properties relate to (for logs) * \param[in,out] op Fencing operation that properties are being parsed for * \param[in] phase Phase the properties relate to * \param[in,out] props Device properties to update */ static void parse_action_specific(const xmlNode *xml, const char *peer, const char *device, const char *action, remote_fencing_op_t *op, enum st_remap_phase phase, device_properties_t *props) { props->custom_action_timeout[phase] = 0; crm_element_value_int(xml, PCMK__XA_ST_ACTION_TIMEOUT, &props->custom_action_timeout[phase]); if (props->custom_action_timeout[phase]) { crm_trace("Peer %s with device %s returned %s action timeout %ds", peer, device, action, props->custom_action_timeout[phase]); } props->delay_max[phase] = 0; crm_element_value_int(xml, PCMK__XA_ST_DELAY_MAX, &props->delay_max[phase]); if (props->delay_max[phase]) { crm_trace("Peer %s with device %s returned maximum of random delay %ds for %s", peer, device, props->delay_max[phase], action); } props->delay_base[phase] = 0; crm_element_value_int(xml, PCMK__XA_ST_DELAY_BASE, &props->delay_base[phase]); if (props->delay_base[phase]) { crm_trace("Peer %s with device %s returned base delay %ds for %s", peer, device, props->delay_base[phase], action); } /* Handle devices with automatic unfencing */ if (pcmk__str_eq(action, PCMK_ACTION_ON, pcmk__str_none)) { int required = 0; crm_element_value_int(xml, PCMK__XA_ST_REQUIRED, &required); if (required) { crm_trace("Peer %s requires device %s to execute for action %s", peer, device, action); add_required_device(op, device); } } /* If a reboot is remapped to off+on, it's possible that a node is allowed * to perform one action but not another. */ if (pcmk__xe_attr_is_true(xml, PCMK__XA_ST_ACTION_DISALLOWED)) { props->disallowed[phase] = TRUE; crm_trace("Peer %s is disallowed from executing %s for device %s", peer, action, device); } } /*! * \internal * \brief Parse one device's properties from peer's XML query reply * * \param[in] xml XML node containing device properties * \param[in,out] op Operation that query and reply relate to * \param[in,out] peer Peer's device information * \param[in] device ID of device being parsed */ static void add_device_properties(const xmlNode *xml, remote_fencing_op_t *op, peer_device_info_t *peer, const char *device) { xmlNode *child; int verified = 0; device_properties_t *props = pcmk__assert_alloc(1, sizeof(device_properties_t)); int flags = st_device_supports_on; /* Old nodes that don't set the flag assume they support the on action */ /* Add a new entry to this peer's devices list */ g_hash_table_insert(peer->devices, pcmk__str_copy(device), props); /* Peers with verified (monitored) access will be preferred */ crm_element_value_int(xml, PCMK__XA_ST_MONITOR_VERIFIED, &verified); if (verified) { crm_trace("Peer %s has confirmed a verified device %s", peer->host, device); props->verified = TRUE; } crm_element_value_int(xml, PCMK__XA_ST_DEVICE_SUPPORT_FLAGS, &flags); props->device_support_flags = flags; /* Parse action-specific device properties */ parse_action_specific(xml, peer->host, device, op_requested_action(op), op, st_phase_requested, props); for (child = pcmk__xe_first_child(xml, NULL, NULL, NULL); child != NULL; child = pcmk__xe_next(child)) { /* Replies for "reboot" operations will include the action-specific * values for "off" and "on" in child elements, just in case the reboot * winds up getting remapped. */ if (pcmk__str_eq(pcmk__xe_id(child), PCMK_ACTION_OFF, pcmk__str_none)) { parse_action_specific(child, peer->host, device, PCMK_ACTION_OFF, op, st_phase_off, props); } else if (pcmk__str_eq(pcmk__xe_id(child), PCMK_ACTION_ON, pcmk__str_none)) { parse_action_specific(child, peer->host, device, PCMK_ACTION_ON, op, st_phase_on, props); } } } /*! * \internal * \brief Parse a peer's XML query reply and add it to operation's results * * \param[in,out] op Operation that query and reply relate to * \param[in] host Name of peer that sent this reply * \param[in] ndevices Number of devices expected in reply * \param[in] xml XML node containing device list * * \return Newly allocated result structure with parsed reply */ static peer_device_info_t * add_result(remote_fencing_op_t *op, const char *host, int ndevices, const xmlNode *xml) { peer_device_info_t *peer = pcmk__assert_alloc(1, sizeof(peer_device_info_t)); xmlNode *child; peer->host = pcmk__str_copy(host); peer->devices = pcmk__strkey_table(free, free); /* Each child element describes one capable device available to the peer */ for (child = pcmk__xe_first_child(xml, NULL, NULL, NULL); child != NULL; child = pcmk__xe_next(child)) { const char *device = pcmk__xe_id(child); if (device) { add_device_properties(child, op, peer, device); } } peer->ndevices = g_hash_table_size(peer->devices); CRM_CHECK(ndevices == peer->ndevices, crm_err("Query claimed to have %d device%s but %d found", ndevices, pcmk__plural_s(ndevices), peer->ndevices)); op->query_results = g_list_insert_sorted(op->query_results, peer, sort_peers); return peer; } /*! * \internal * \brief Handle a peer's reply to our fencing query * * Parse a query result from XML and store it in the remote operation * table, and when enough replies have been received, issue a fencing request. * * \param[in] msg XML reply received * * \return pcmk_ok on success, -errno on error * * \note See initiate_remote_stonith_op() for how the XML query was initially * formed, and stonith_query() for how the peer formed its XML reply. */ int process_remote_stonith_query(xmlNode *msg) { int ndevices = 0; gboolean host_is_target = FALSE; gboolean have_all_replies = FALSE; const char *id = NULL; const char *host = NULL; remote_fencing_op_t *op = NULL; peer_device_info_t *peer = NULL; uint32_t replies_expected; xmlNode *dev = get_xpath_object("//@" PCMK__XA_ST_REMOTE_OP, msg, LOG_ERR); CRM_CHECK(dev != NULL, return -EPROTO); id = crm_element_value(dev, PCMK__XA_ST_REMOTE_OP); CRM_CHECK(id != NULL, return -EPROTO); dev = get_xpath_object("//@" PCMK__XA_ST_AVAILABLE_DEVICES, msg, LOG_ERR); CRM_CHECK(dev != NULL, return -EPROTO); crm_element_value_int(dev, PCMK__XA_ST_AVAILABLE_DEVICES, &ndevices); op = g_hash_table_lookup(stonith_remote_op_list, id); if (op == NULL) { crm_debug("Received query reply for unknown or expired operation %s", id); return -EOPNOTSUPP; } replies_expected = fencing_active_peers(); if (op->replies_expected < replies_expected) { replies_expected = op->replies_expected; } if ((++op->replies >= replies_expected) && (op->state == st_query)) { have_all_replies = TRUE; } host = crm_element_value(msg, PCMK__XA_SRC); host_is_target = pcmk__str_eq(host, op->target, pcmk__str_casei); crm_info("Query result %d of %d from %s for %s/%s (%d device%s) %s", op->replies, replies_expected, host, op->target, op->action, ndevices, pcmk__plural_s(ndevices), id); if (ndevices > 0) { peer = add_result(op, host, ndevices, dev); } pcmk__set_result(&op->result, CRM_EX_OK, PCMK_EXEC_DONE, NULL); if (pcmk_is_set(op->call_options, st_opt_topology)) { /* If we start the fencing before all the topology results are in, * it is possible fencing levels will be skipped because of the missing * query results. */ if (op->state == st_query && all_topology_devices_found(op)) { /* All the query results are in for the topology, start the fencing ops. */ crm_trace("All topology devices found"); request_peer_fencing(op, peer); } else if (have_all_replies) { crm_info("All topology query replies have arrived, continuing (%d expected/%d received) ", replies_expected, op->replies); request_peer_fencing(op, NULL); } } else if (op->state == st_query) { int nverified = count_peer_devices(op, peer, TRUE, fenced_support_flag(op->action)); /* We have a result for a non-topology fencing op that looks promising, * go ahead and start fencing before query timeout */ if ((peer != NULL) && !host_is_target && nverified) { /* we have a verified device living on a peer that is not the target */ crm_trace("Found %d verified device%s", nverified, pcmk__plural_s(nverified)); request_peer_fencing(op, peer); } else if (have_all_replies) { crm_info("All query replies have arrived, continuing (%d expected/%d received) ", replies_expected, op->replies); request_peer_fencing(op, NULL); } else { crm_trace("Waiting for more peer results before launching fencing operation"); } } else if ((peer != NULL) && (op->state == st_done)) { crm_info("Discarding query result from %s (%d device%s): " "Operation is %s", peer->host, peer->ndevices, pcmk__plural_s(peer->ndevices), stonith_op_state_str(op->state)); } return pcmk_ok; } /*! * \internal * \brief Handle a peer's reply to a fencing request * * Parse a fencing reply from XML, and either finalize the operation * or attempt another device as appropriate. * * \param[in] msg XML reply received */ void fenced_process_fencing_reply(xmlNode *msg) { const char *id = NULL; const char *device = NULL; remote_fencing_op_t *op = NULL; xmlNode *dev = get_xpath_object("//@" PCMK__XA_ST_REMOTE_OP, msg, LOG_ERR); pcmk__action_result_t result = PCMK__UNKNOWN_RESULT; CRM_CHECK(dev != NULL, return); id = crm_element_value(dev, PCMK__XA_ST_REMOTE_OP); CRM_CHECK(id != NULL, return); dev = stonith__find_xe_with_result(msg); CRM_CHECK(dev != NULL, return); stonith__xe_get_result(dev, &result); device = crm_element_value(dev, PCMK__XA_ST_DEVICE_ID); if (stonith_remote_op_list) { op = g_hash_table_lookup(stonith_remote_op_list, id); } if ((op == NULL) && pcmk__result_ok(&result)) { /* Record successful fencing operations */ const char *client_id = crm_element_value(dev, PCMK__XA_ST_CLIENTID); op = create_remote_stonith_op(client_id, dev, TRUE); } if (op == NULL) { /* Could be for an event that began before we started */ /* TODO: Record the op for later querying */ crm_info("Received peer result of unknown or expired operation %s", id); pcmk__reset_result(&result); return; } pcmk__reset_result(&op->result); op->result = result; // The operation takes ownership of the result if (op->devices && device && !pcmk__str_eq(op->devices->data, device, pcmk__str_casei)) { crm_err("Received outdated reply for device %s (instead of %s) to " "fence (%s) %s. Operation already timed out at peer level.", device, (const char *) op->devices->data, op->action, op->target); return; } if (pcmk__str_eq(crm_element_value(msg, PCMK__XA_SUBT), PCMK__VALUE_BROADCAST, pcmk__str_none)) { if (pcmk__result_ok(&op->result)) { op->state = st_done; } else { op->state = st_failed; } finalize_op(op, msg, false); return; } else if (!pcmk__str_eq(op->originator, stonith_our_uname, pcmk__str_casei)) { /* If this isn't a remote level broadcast, and we are not the * originator of the operation, we should not be receiving this msg. */ crm_err("Received non-broadcast fencing result for operation %.8s " "we do not own (device %s targeting %s)", op->id, device, op->target); return; } if (pcmk_is_set(op->call_options, st_opt_topology)) { const char *device = NULL; const char *reason = op->result.exit_reason; /* We own the op, and it is complete. broadcast the result to all nodes * and notify our local clients. */ if (op->state == st_done) { finalize_op(op, msg, false); return; } device = crm_element_value(msg, PCMK__XA_ST_DEVICE_ID); if ((op->phase == 2) && !pcmk__result_ok(&op->result)) { /* A remapped "on" failed, but the node was already turned off * successfully, so ignore the error and continue. */ crm_warn("Ignoring %s 'on' failure (%s%s%s) targeting %s " "after successful 'off'", device, pcmk_exec_status_str(op->result.execution_status), (reason == NULL)? "" : ": ", (reason == NULL)? "" : reason, op->target); pcmk__set_result(&op->result, CRM_EX_OK, PCMK_EXEC_DONE, NULL); } else { crm_notice("Action '%s' targeting %s%s%s on behalf of %s@%s: " "%s%s%s%s", op->action, op->target, ((device == NULL)? "" : " using "), ((device == NULL)? "" : device), op->client_name, op->originator, pcmk_exec_status_str(op->result.execution_status), (reason == NULL)? "" : " (", (reason == NULL)? "" : reason, (reason == NULL)? "" : ")"); } if (pcmk__result_ok(&op->result)) { /* An operation completed successfully. Try another device if * necessary, otherwise mark the operation as done. */ advance_topology_device_in_level(op, device, msg); return; } else { /* This device failed, time to try another topology level. If no other * levels are available, mark this operation as failed and report results. */ if (advance_topology_level(op, false) != pcmk_rc_ok) { op->state = st_failed; finalize_op(op, msg, false); return; } } } else if (pcmk__result_ok(&op->result) && (op->devices == NULL)) { op->state = st_done; finalize_op(op, msg, false); return; } else if ((op->result.execution_status == PCMK_EXEC_TIMEOUT) && (op->devices == NULL)) { /* If the operation timed out don't bother retrying other peers. */ op->state = st_failed; finalize_op(op, msg, false); return; } else { /* fall-through and attempt other fencing action using another peer */ } /* Retry on failure */ crm_trace("Next for %s on behalf of %s@%s (result was: %s)", op->target, op->originator, op->client_name, pcmk_exec_status_str(op->result.execution_status)); request_peer_fencing(op, NULL); } gboolean stonith_check_fence_tolerance(int tolerance, const char *target, const char *action) { GHashTableIter iter; time_t now = time(NULL); remote_fencing_op_t *rop = NULL; if (tolerance <= 0 || !stonith_remote_op_list || target == NULL || action == NULL) { return FALSE; } g_hash_table_iter_init(&iter, stonith_remote_op_list); while (g_hash_table_iter_next(&iter, NULL, (void **)&rop)) { if (strcmp(rop->target, target) != 0) { continue; } else if (rop->state != st_done) { continue; /* We don't have to worry about remapped reboots here * because if state is done, any remapping has been undone */ } else if (strcmp(rop->action, action) != 0) { continue; } else if ((rop->completed + tolerance) < now) { continue; } crm_notice("Target %s was fenced (%s) less than %ds ago by %s on behalf of %s", target, action, tolerance, rop->delegate, rop->originator); return TRUE; } return FALSE; } diff --git a/daemons/fenced/pacemaker-fenced.c b/daemons/fenced/pacemaker-fenced.c index 990d3cdbb5..842f1c54e4 100644 --- a/daemons/fenced/pacemaker-fenced.c +++ b/daemons/fenced/pacemaker-fenced.c @@ -1,681 +1,681 @@ /* * Copyright 2009-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include #include #include // PRIu32, PRIx32 #include #include #include #include #include #include #include #include #include #include #include #include #define SUMMARY "daemon for executing fencing devices in a Pacemaker cluster" char *stonith_our_uname = NULL; long long stonith_watchdog_timeout_ms = 0; GList *stonith_watchdog_targets = NULL; static GMainLoop *mainloop = NULL; gboolean stand_alone = FALSE; gboolean stonith_shutdown_flag = FALSE; static qb_ipcs_service_t *ipcs = NULL; static pcmk__output_t *out = NULL; pcmk__supported_format_t formats[] = { PCMK__SUPPORTED_FORMAT_NONE, PCMK__SUPPORTED_FORMAT_TEXT, PCMK__SUPPORTED_FORMAT_XML, { NULL, NULL, NULL } }; static struct { bool no_cib_connect; gchar **log_files; } options; crm_exit_t exit_code = CRM_EX_OK; static void stonith_cleanup(void); static int32_t st_ipc_accept(qb_ipcs_connection_t * c, uid_t uid, gid_t gid) { if (stonith_shutdown_flag) { crm_info("Ignoring new client [%d] during shutdown", pcmk__client_pid(c)); return -ECONNREFUSED; } if (pcmk__new_client(c, uid, gid) == NULL) { return -ENOMEM; } return 0; } /* Exit code means? */ static int32_t st_ipc_dispatch(qb_ipcs_connection_t * qbc, void *data, size_t size) { uint32_t id = 0; uint32_t flags = 0; int call_options = 0; xmlNode *request = NULL; pcmk__client_t *c = pcmk__find_client(qbc); const char *op = NULL; if (c == NULL) { crm_info("Invalid client: %p", qbc); return 0; } request = pcmk__client_data2xml(c, data, &id, &flags); if (request == NULL) { pcmk__ipc_send_ack(c, id, flags, PCMK__XE_NACK, NULL, CRM_EX_PROTOCOL); return 0; } op = crm_element_value(request, PCMK__XA_CRM_TASK); if(pcmk__str_eq(op, CRM_OP_RM_NODE_CACHE, pcmk__str_casei)) { crm_xml_add(request, PCMK__XA_T, PCMK__VALUE_STONITH_NG); crm_xml_add(request, PCMK__XA_ST_OP, op); crm_xml_add(request, PCMK__XA_ST_CLIENTID, c->id); crm_xml_add(request, PCMK__XA_ST_CLIENTNAME, pcmk__client_name(c)); crm_xml_add(request, PCMK__XA_ST_CLIENTNODE, stonith_our_uname); - pcmk__cluster_send_message(NULL, crm_msg_stonith_ng, request); + pcmk__cluster_send_message(NULL, pcmk__cluster_msg_fenced, request); pcmk__xml_free(request); return 0; } if (c->name == NULL) { const char *value = crm_element_value(request, PCMK__XA_ST_CLIENTNAME); c->name = crm_strdup_printf("%s.%u", pcmk__s(value, "unknown"), c->pid); } crm_element_value_int(request, PCMK__XA_ST_CALLOPT, &call_options); crm_trace("Flags %#08" PRIx32 "/%#08x for command %" PRIu32 " from client %s", flags, call_options, id, pcmk__client_name(c)); if (pcmk_is_set(call_options, st_opt_sync_call)) { CRM_ASSERT(flags & crm_ipc_client_response); CRM_LOG_ASSERT(c->request_id == 0); /* This means the client has two synchronous events in-flight */ c->request_id = id; /* Reply only to the last one */ } crm_xml_add(request, PCMK__XA_ST_CLIENTID, c->id); crm_xml_add(request, PCMK__XA_ST_CLIENTNAME, pcmk__client_name(c)); crm_xml_add(request, PCMK__XA_ST_CLIENTNODE, stonith_our_uname); crm_log_xml_trace(request, "ipc-received"); stonith_command(c, id, flags, request, NULL); pcmk__xml_free(request); return 0; } /* Error code means? */ static int32_t st_ipc_closed(qb_ipcs_connection_t * c) { pcmk__client_t *client = pcmk__find_client(c); if (client == NULL) { return 0; } crm_trace("Connection %p closed", c); pcmk__free_client(client); /* 0 means: yes, go ahead and destroy the connection */ return 0; } static void st_ipc_destroy(qb_ipcs_connection_t * c) { crm_trace("Connection %p destroyed", c); st_ipc_closed(c); } static void stonith_peer_callback(xmlNode * msg, void *private_data) { const char *remote_peer = crm_element_value(msg, PCMK__XA_SRC); const char *op = crm_element_value(msg, PCMK__XA_ST_OP); if (pcmk__str_eq(op, STONITH_OP_POKE, pcmk__str_none)) { return; } crm_log_xml_trace(msg, "Peer[inbound]"); stonith_command(NULL, 0, 0, msg, remote_peer); } #if SUPPORT_COROSYNC static void stonith_peer_ais_callback(cpg_handle_t handle, const struct cpg_name *groupName, uint32_t nodeid, uint32_t pid, void *msg, size_t msg_len) { uint32_t kind = 0; xmlNode *xml = NULL; const char *from = NULL; char *data = pcmk__cpg_message_data(handle, nodeid, pid, msg, &kind, &from); if(data == NULL) { return; } if (kind == crm_class_cluster) { xml = pcmk__xml_parse(data); if (xml == NULL) { crm_err("Invalid XML: '%.120s'", data); free(data); return; } crm_xml_add(xml, PCMK__XA_SRC, from); stonith_peer_callback(xml, NULL); } pcmk__xml_free(xml); free(data); return; } static void stonith_peer_cs_destroy(gpointer user_data) { crm_crit("Lost connection to cluster layer, shutting down"); stonith_shutdown(0); } #endif void do_local_reply(const xmlNode *notify_src, pcmk__client_t *client, int call_options) { /* send callback to originating child */ int local_rc = pcmk_rc_ok; int rid = 0; uint32_t ipc_flags = crm_ipc_server_event; if (pcmk_is_set(call_options, st_opt_sync_call)) { CRM_LOG_ASSERT(client->request_id); rid = client->request_id; client->request_id = 0; ipc_flags = crm_ipc_flags_none; } local_rc = pcmk__ipc_send_xml(client, rid, notify_src, ipc_flags); if (local_rc == pcmk_rc_ok) { crm_trace("Sent response %d to client %s", rid, pcmk__client_name(client)); } else { crm_warn("%synchronous reply to client %s failed: %s", (pcmk_is_set(call_options, st_opt_sync_call)? "S" : "As"), pcmk__client_name(client), pcmk_rc_str(local_rc)); } } uint64_t get_stonith_flag(const char *name) { if (pcmk__str_eq(name, PCMK__VALUE_ST_NOTIFY_FENCE, pcmk__str_none)) { return st_callback_notify_fence; } else if (pcmk__str_eq(name, STONITH_OP_DEVICE_ADD, pcmk__str_casei)) { return st_callback_device_add; } else if (pcmk__str_eq(name, STONITH_OP_DEVICE_DEL, pcmk__str_casei)) { return st_callback_device_del; } else if (pcmk__str_eq(name, PCMK__VALUE_ST_NOTIFY_HISTORY, pcmk__str_none)) { return st_callback_notify_history; } else if (pcmk__str_eq(name, PCMK__VALUE_ST_NOTIFY_HISTORY_SYNCED, pcmk__str_none)) { return st_callback_notify_history_synced; } return st_callback_unknown; } static void stonith_notify_client(gpointer key, gpointer value, gpointer user_data) { const xmlNode *update_msg = user_data; pcmk__client_t *client = value; const char *type = NULL; CRM_CHECK(client != NULL, return); CRM_CHECK(update_msg != NULL, return); type = crm_element_value(update_msg, PCMK__XA_SUBT); CRM_CHECK(type != NULL, crm_log_xml_err(update_msg, "notify"); return); if (client->ipcs == NULL) { crm_trace("Skipping client with NULL channel"); return; } if (pcmk_is_set(client->flags, get_stonith_flag(type))) { int rc = pcmk__ipc_send_xml(client, 0, update_msg, crm_ipc_server_event); if (rc != pcmk_rc_ok) { crm_warn("%s notification of client %s failed: %s " QB_XS " id=%.8s rc=%d", type, pcmk__client_name(client), pcmk_rc_str(rc), client->id, rc); } else { crm_trace("Sent %s notification to client %s", type, pcmk__client_name(client)); } } } void do_stonith_async_timeout_update(const char *client_id, const char *call_id, int timeout) { pcmk__client_t *client = NULL; xmlNode *notify_data = NULL; if (!timeout || !call_id || !client_id) { return; } client = pcmk__find_client_by_id(client_id); if (!client) { return; } notify_data = pcmk__xe_create(NULL, PCMK__XE_ST_ASYNC_TIMEOUT_VALUE); crm_xml_add(notify_data, PCMK__XA_T, PCMK__VALUE_ST_ASYNC_TIMEOUT_VALUE); crm_xml_add(notify_data, PCMK__XA_ST_CALLID, call_id); crm_xml_add_int(notify_data, PCMK__XA_ST_TIMEOUT, timeout); crm_trace("timeout update is %d for client %s and call id %s", timeout, client_id, call_id); if (client) { pcmk__ipc_send_xml(client, 0, notify_data, crm_ipc_server_event); } pcmk__xml_free(notify_data); } /*! * \internal * \brief Notify relevant IPC clients of a fencing operation result * * \param[in] type Notification type * \param[in] result Result of fencing operation (assume success if NULL) * \param[in] data If not NULL, add to notification as call data */ void fenced_send_notification(const char *type, const pcmk__action_result_t *result, xmlNode *data) { /* TODO: Standardize the contents of data */ xmlNode *update_msg = pcmk__xe_create(NULL, PCMK__XE_NOTIFY); CRM_LOG_ASSERT(type != NULL); crm_xml_add(update_msg, PCMK__XA_T, PCMK__VALUE_ST_NOTIFY); crm_xml_add(update_msg, PCMK__XA_SUBT, type); crm_xml_add(update_msg, PCMK__XA_ST_OP, type); stonith__xe_set_result(update_msg, result); if (data != NULL) { xmlNode *wrapper = pcmk__xe_create(update_msg, PCMK__XE_ST_CALLDATA); pcmk__xml_copy(wrapper, data); } crm_trace("Notifying clients"); pcmk__foreach_ipc_client(stonith_notify_client, update_msg); pcmk__xml_free(update_msg); crm_trace("Notify complete"); } /*! * \internal * \brief Send notifications for a configuration change to subscribed clients * * \param[in] op Notification type (\c STONITH_OP_DEVICE_ADD, * \c STONITH_OP_DEVICE_DEL, \c STONITH_OP_LEVEL_ADD, or * \c STONITH_OP_LEVEL_DEL) * \param[in] result Operation result * \param[in] desc Description of what changed (either device ID or string * representation of level * ([])) */ void fenced_send_config_notification(const char *op, const pcmk__action_result_t *result, const char *desc) { xmlNode *notify_data = pcmk__xe_create(NULL, op); crm_xml_add(notify_data, PCMK__XA_ST_DEVICE_ID, desc); fenced_send_notification(op, result, notify_data); pcmk__xml_free(notify_data); } /*! * \internal * \brief Check whether a node does watchdog-fencing * * \param[in] node Name of node to check * * \return TRUE if node found in stonith_watchdog_targets * or stonith_watchdog_targets is empty indicating * all nodes are doing watchdog-fencing */ gboolean node_does_watchdog_fencing(const char *node) { return ((stonith_watchdog_targets == NULL) || pcmk__str_in_list(node, stonith_watchdog_targets, pcmk__str_casei)); } void stonith_shutdown(int nsig) { crm_info("Terminating with %d clients", pcmk__ipc_client_count()); stonith_shutdown_flag = TRUE; if (mainloop != NULL && g_main_loop_is_running(mainloop)) { g_main_loop_quit(mainloop); } } static void stonith_cleanup(void) { fenced_cib_cleanup(); if (ipcs) { qb_ipcs_destroy(ipcs); } pcmk__cluster_destroy_node_caches(); pcmk__client_cleanup(); free_stonith_remote_op_list(); free_topology_list(); free_device_list(); free_metadata_cache(); fenced_unregister_handlers(); free(stonith_our_uname); stonith_our_uname = NULL; } static gboolean stand_alone_cpg_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error) { stand_alone = FALSE; options.no_cib_connect = true; return TRUE; } struct qb_ipcs_service_handlers ipc_callbacks = { .connection_accept = st_ipc_accept, .connection_created = NULL, .msg_process = st_ipc_dispatch, .connection_closed = st_ipc_closed, .connection_destroyed = st_ipc_destroy }; /*! * \internal * \brief Callback for peer status changes * * \param[in] type What changed * \param[in] node What peer had the change * \param[in] data Previous value of what changed */ static void st_peer_update_callback(enum pcmk__node_update type, pcmk__node_status_t *node, const void *data) { if ((type != pcmk__node_update_processes) && !pcmk_is_set(node->flags, pcmk__node_status_remote)) { /* * This is a hack until we can send to a nodeid and/or we fix node name lookups * These messages are ignored in stonith_peer_callback() */ xmlNode *query = pcmk__xe_create(NULL, PCMK__XE_STONITH_COMMAND); crm_xml_add(query, PCMK__XA_T, PCMK__VALUE_STONITH_NG); crm_xml_add(query, PCMK__XA_ST_OP, STONITH_OP_POKE); crm_debug("Broadcasting our uname because of node %" PRIu32, node->cluster_layer_id); - pcmk__cluster_send_message(NULL, crm_msg_stonith_ng, query); + pcmk__cluster_send_message(NULL, pcmk__cluster_msg_fenced, query); pcmk__xml_free(query); } } /* @COMPAT Deprecated since 2.1.8. Use pcmk_list_fence_attrs() or * crm_resource --list-options=fencing instead of querying daemon metadata. */ static int fencer_metadata(void) { const char *name = "pacemaker-fenced"; const char *desc_short = N_("Instance attributes available for all " "\"stonith\"-class resources"); const char *desc_long = N_("Instance attributes available for all " "\"stonith\"-class resources and used by " "Pacemaker's fence daemon"); return pcmk__daemon_metadata(out, name, desc_short, desc_long, pcmk__opt_fencing); } static GOptionEntry entries[] = { { "stand-alone", 's', G_OPTION_FLAG_NONE, G_OPTION_ARG_NONE, &stand_alone, N_("Deprecated (will be removed in a future release)"), NULL }, { "stand-alone-w-cpg", 'c', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, stand_alone_cpg_cb, N_("Intended for use in regression testing only"), NULL }, { "logfile", 'l', G_OPTION_FLAG_NONE, G_OPTION_ARG_FILENAME_ARRAY, &options.log_files, N_("Send logs to the additional named logfile"), 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) { int rc = pcmk_rc_ok; pcmk_cluster_t *cluster = NULL; crm_ipc_t *old_instance = NULL; GError *error = NULL; GOptionGroup *output_group = NULL; pcmk__common_args_t *args = pcmk__new_common_args(SUMMARY); gchar **processed_args = pcmk__cmdline_preproc(argv, "l"); GOptionContext *context = build_arg_context(args, &output_group); crm_log_preinit(NULL, 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; } if ((g_strv_length(processed_args) >= 2) && pcmk__str_eq(processed_args[1], "metadata", pcmk__str_none)) { rc = fencer_metadata(); if (rc != pcmk_rc_ok) { exit_code = CRM_EX_FATAL; g_set_error(&error, PCMK__EXITC_ERROR, exit_code, "Unable to display metadata: %s", pcmk_rc_str(rc)); } goto done; } // Open additional log files pcmk__add_logfiles(options.log_files, out); crm_log_init(NULL, LOG_INFO + args->verbosity, TRUE, (args->verbosity > 0), argc, argv, FALSE); crm_notice("Starting Pacemaker fencer"); old_instance = crm_ipc_new("stonith-ng", 0); if (old_instance == NULL) { /* crm_ipc_new() will have already logged an error message with * crm_err() */ exit_code = CRM_EX_FATAL; goto done; } if (pcmk__connect_generic_ipc(old_instance) == pcmk_rc_ok) { // IPC endpoint already up crm_ipc_close(old_instance); crm_ipc_destroy(old_instance); crm_err("pacemaker-fenced is already active, aborting startup"); goto done; } else { // Not up or not authentic, we'll proceed either way crm_ipc_destroy(old_instance); old_instance = NULL; } mainloop_add_signal(SIGTERM, stonith_shutdown); pcmk__cluster_init_node_caches(); rc = fenced_scheduler_init(); if (rc != pcmk_rc_ok) { exit_code = CRM_EX_FATAL; g_set_error(&error, PCMK__EXITC_ERROR, exit_code, "Error initializing scheduler data: %s", pcmk_rc_str(rc)); goto done; } cluster = pcmk_cluster_new(); if (!stand_alone) { #if SUPPORT_COROSYNC if (pcmk_get_cluster_layer() == pcmk_cluster_layer_corosync) { pcmk_cluster_set_destroy_fn(cluster, stonith_peer_cs_destroy); pcmk_cpg_set_deliver_fn(cluster, stonith_peer_ais_callback); pcmk_cpg_set_confchg_fn(cluster, pcmk__cpg_confchg_cb); } #endif // SUPPORT_COROSYNC pcmk__cluster_set_status_callback(&st_peer_update_callback); if (pcmk_cluster_connect(cluster) != pcmk_rc_ok) { exit_code = CRM_EX_FATAL; crm_crit("Cannot sign in to the cluster... terminating"); goto done; } pcmk__str_update(&stonith_our_uname, cluster->uname); if (!options.no_cib_connect) { setup_cib(); } } else { pcmk__str_update(&stonith_our_uname, "localhost"); crm_warn("Stand-alone mode is deprecated and will be removed " "in a future release"); } init_device_list(); init_topology_list(); pcmk__serve_fenced_ipc(&ipcs, &ipc_callbacks); // Create the mainloop and run it... mainloop = g_main_loop_new(NULL, FALSE); crm_notice("Pacemaker fencer successfully started and accepting connections"); g_main_loop_run(mainloop); done: g_strfreev(processed_args); pcmk__free_arg_context(context); g_strfreev(options.log_files); stonith_cleanup(); pcmk_cluster_free(cluster); fenced_scheduler_cleanup(); 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/include/crm/cluster.h b/include/crm/cluster.h index f9822cece5..81d36fae97 100644 --- a/include/crm/cluster.h +++ b/include/crm/cluster.h @@ -1,163 +1,146 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #ifndef PCMK__CRM_CLUSTER__H # define PCMK__CRM_CLUSTER__H # include // uint32_t, uint64_t # include // gboolean, GHashTable # include // xmlNode # include # include #ifdef __cplusplus extern "C" { #endif # if SUPPORT_COROSYNC # include # endif // @COMPAT Make this internal when we can break API backward compatibility //! \deprecated Do not use (public access will be removed in a future release) extern GHashTable *crm_peer_cache; // @COMPAT Make this internal when we can break API backward compatibility //! \deprecated Do not use (public access will be removed in a future release) extern GHashTable *crm_remote_peer_cache; // @COMPAT Make this internal when we can break API backward compatibility //! \deprecated Do not use (public access will be removed in a future release) extern unsigned long long crm_peer_seq; // @COMPAT Make this internal when we can break API backward compatibility //! \deprecated Do not use (public access will be removed in a future release) #define CRM_NODE_LOST "lost" // @COMPAT Make this internal when we can break API backward compatibility //! \deprecated Do not use (public access will be removed in a future release) #define CRM_NODE_MEMBER "member" // @COMPAT Make this internal when we can break API backward compatibility //!@{ //! \deprecated Do not use (public access will be removed in a future release) enum crm_join_phase { /* @COMPAT: crm_join_nack_quiet can be replaced by * pcmk__node_status_t:user_data at a compatibility break */ //! Not allowed to join, but don't send a nack message crm_join_nack_quiet = -2, crm_join_nack = -1, crm_join_none = 0, crm_join_welcomed = 1, crm_join_integrated = 2, crm_join_finalized = 3, crm_join_confirmed = 4, }; //!@} // Implementation of pcmk_cluster_t // @COMPAT Make this internal when we can break API backward compatibility //!@{ //! \deprecated Do not use (public access will be removed in a future release) struct crm_cluster_s { char *uuid; char *uname; uint32_t nodeid; // NOTE: sbd (as of at least 1.5.2) uses this //! \deprecated Call pcmk_cluster_set_destroy_fn() to set this void (*destroy) (gpointer); # if SUPPORT_COROSYNC /* @TODO When we can break public API compatibility, make these members a * separate struct and use void *cluster_data here instead, to abstract the * cluster layer further. */ struct cpg_name group; // NOTE: sbd (as of at least 1.5.2) uses this /*! * \deprecated Call pcmk_cpg_set_deliver_fn() and pcmk_cpg_set_confchg_fn() * to set these */ cpg_callbacks_t cpg; cpg_handle_t cpg_handle; # endif }; //!@} //! Connection to a cluster layer typedef struct crm_cluster_s pcmk_cluster_t; int pcmk_cluster_connect(pcmk_cluster_t *cluster); int pcmk_cluster_disconnect(pcmk_cluster_t *cluster); pcmk_cluster_t *pcmk_cluster_new(void); void pcmk_cluster_free(pcmk_cluster_t *cluster); int pcmk_cluster_set_destroy_fn(pcmk_cluster_t *cluster, void (*fn)(gpointer)); #if SUPPORT_COROSYNC int pcmk_cpg_set_deliver_fn(pcmk_cluster_t *cluster, cpg_deliver_fn_t fn); int pcmk_cpg_set_confchg_fn(pcmk_cluster_t *cluster, cpg_confchg_fn_t fn); #endif // SUPPORT_COROSYNC /* @COMPAT Make this internal when we can break API backward compatibility. Also * evaluate whether we can drop this entirely. Since 2.0.0, we have sent only * messages with crm_class_cluster. */ //!@{ //! \deprecated Do not use (public access will be removed in a future release) enum crm_ais_msg_class { crm_class_cluster = 0, }; //!@} -// @COMPAT Make this internal when we can break API backward compatibility -//!@{ -//! \deprecated Do not use (public access will be removed in a future release) -enum crm_ais_msg_types { - crm_msg_none = 0, - crm_msg_ais = 1, // Unused - crm_msg_lrmd = 2, - crm_msg_cib = 3, - crm_msg_crmd = 4, - crm_msg_attrd = 5, - crm_msg_stonithd = 6, // Unused - crm_msg_te = 7, // Unused - crm_msg_pe = 8, // Unused - crm_msg_stonith_ng = 9, -}; -//!@} - /*! * \enum pcmk_cluster_layer * \brief Types of cluster layer */ enum pcmk_cluster_layer { pcmk_cluster_layer_unknown = 1, //!< Unknown cluster layer pcmk_cluster_layer_invalid = 2, //!< Invalid cluster layer pcmk_cluster_layer_corosync = 32, //!< Corosync Cluster Engine }; enum pcmk_cluster_layer pcmk_get_cluster_layer(void); const char *pcmk_cluster_layer_text(enum pcmk_cluster_layer layer); #ifdef __cplusplus } #endif #if !defined(PCMK_ALLOW_DEPRECATED) || (PCMK_ALLOW_DEPRECATED == 1) #include #endif #endif diff --git a/include/crm/cluster/internal.h b/include/crm/cluster/internal.h index b6b9d0e40b..d5904d3ba1 100644 --- a/include/crm/cluster/internal.h +++ b/include/crm/cluster/internal.h @@ -1,296 +1,310 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #ifndef PCMK__CRM_CLUSTER_INTERNAL__H #define PCMK__CRM_CLUSTER_INTERNAL__H #include #include // uint32_t, uint64_t #include // gboolean #include #ifdef __cplusplus extern "C" { #endif +/*! + * \internal + * \enum pcmk__cluster_msg + * \brief Types of message sent via the cluster layer + */ +enum pcmk__cluster_msg { + pcmk__cluster_msg_unknown, + pcmk__cluster_msg_attrd, + pcmk__cluster_msg_based, + pcmk__cluster_msg_controld, + pcmk__cluster_msg_execd, + pcmk__cluster_msg_fenced, +}; + enum crm_proc_flag { /* @COMPAT When pcmk__node_status_t:processes is made internal, we can merge * this into node flags or turn it into a boolean. Until then, in theory * something could depend on these particular numeric values. */ crm_proc_none = 0x00000001, // Cluster layers crm_proc_cpg = 0x04000000, }; /*! * \internal * \enum pcmk__node_status_flags * \brief Boolean flags for a \c pcmk__node_status_t object * * Some flags may not be related to status specifically. However, we keep these * separate from enum pcmk__node_flags because they're used with * different object types. */ enum pcmk__node_status_flags { /*! * Node is a Pacemaker Remote node and should not be considered for cluster * membership */ pcmk__node_status_remote = (UINT32_C(1) << 0), //! Node's cache entry is dirty pcmk__node_status_dirty = (UINT32_C(1) << 1), }; // Used with node cache search functions enum pcmk__node_search_flags { //! Does not affect search pcmk__node_search_none = 0, //! Search for cluster nodes from membership cache pcmk__node_search_cluster_member = (1 << 0), //! Search for remote nodes pcmk__node_search_remote = (1 << 1), //! Search for cluster member nodes and remote nodes pcmk__node_search_any = pcmk__node_search_cluster_member |pcmk__node_search_remote, //! Search for cluster nodes from CIB (as of last cache refresh) pcmk__node_search_cluster_cib = (1 << 2), }; /*! * \internal * \enum pcmk__node_update * \brief Type of update to a \c pcmk__node_status_t object */ enum pcmk__node_update { pcmk__node_update_name, //!< Node name updated pcmk__node_update_state, //!< Node connection state updated pcmk__node_update_processes, //!< Node process group membership updated }; //! Node status data (may be a cluster node or a Pacemaker Remote node) typedef struct pcmk__node_status { //! Node name as known to cluster layer, or Pacemaker Remote node name char *name; /* @COMPAT This is less than ideal since the value is not a valid XML ID * (for Corosync, it's the string equivalent of the node's numeric node ID, * but XML IDs can't start with a number) and the three elements should have * different IDs. * * Ideally, we would use something like node-NODEID, node_state-NODEID, and * transient_attributes-NODEID as the element IDs. Unfortunately changing it * would be impractical due to backward compatibility; older nodes in a * rolling upgrade will always write and expect the value in the old format. */ /*! * Value of the PCMK_XA_ID XML attribute to use with the node's * PCMK_XE_NODE, PCMK_XE_NODE_STATE, and PCMK_XE_TRANSIENT_ATTRIBUTES * XML elements in the CIB */ char *xml_id; char *state; // @TODO change to enum //! Group of enum pcmk__node_status_flags uint32_t flags; /*! * Most recent cluster membership in which node was seen (0 for Pacemaker * Remote nodes) */ uint64_t membership_id; uint32_t processes; // @TODO most not needed, merge into flags /* @TODO When we can break public API compatibility, we can make the rest of * these members separate structs and use void *cluster_data and * void *user_data here instead, to abstract the cluster layer further. */ // Only used by controller enum crm_join_phase join; char *expected; time_t peer_lost; char *conn_host; time_t when_member; // Since when node has been a cluster member time_t when_online; // Since when peer has been online in CPG /* @TODO The following are currently needed only by the Corosync stack. * Eventually consider moving them to a cluster-layer-specific data object. */ uint32_t cluster_layer_id; //!< Cluster-layer numeric node ID time_t when_lost; //!< When CPG membership was last lost } pcmk__node_status_t; /*! * \internal * \brief Return the process bit corresponding to the current cluster stack * * \return Process flag if detectable, otherwise 0 */ static inline uint32_t crm_get_cluster_proc(void) { switch (pcmk_get_cluster_layer()) { case pcmk_cluster_layer_corosync: return crm_proc_cpg; default: break; } return crm_proc_none; } /*! * \internal * \brief Get log-friendly string description of a Corosync return code * * \param[in] error Corosync return code * * \return Log-friendly string description corresponding to \p error */ static inline const char * pcmk__cs_err_str(int error) { # if SUPPORT_COROSYNC switch (error) { case CS_OK: return "OK"; case CS_ERR_LIBRARY: return "Library error"; case CS_ERR_VERSION: return "Version error"; case CS_ERR_INIT: return "Initialization error"; case CS_ERR_TIMEOUT: return "Timeout"; case CS_ERR_TRY_AGAIN: return "Try again"; case CS_ERR_INVALID_PARAM: return "Invalid parameter"; case CS_ERR_NO_MEMORY: return "No memory"; case CS_ERR_BAD_HANDLE: return "Bad handle"; case CS_ERR_BUSY: return "Busy"; case CS_ERR_ACCESS: return "Access error"; case CS_ERR_NOT_EXIST: return "Doesn't exist"; case CS_ERR_NAME_TOO_LONG: return "Name too long"; case CS_ERR_EXIST: return "Exists"; case CS_ERR_NO_SPACE: return "No space"; case CS_ERR_INTERRUPT: return "Interrupt"; case CS_ERR_NAME_NOT_FOUND: return "Name not found"; case CS_ERR_NO_RESOURCES: return "No resources"; case CS_ERR_NOT_SUPPORTED: return "Not supported"; case CS_ERR_BAD_OPERATION: return "Bad operation"; case CS_ERR_FAILED_OPERATION: return "Failed operation"; case CS_ERR_MESSAGE_ERROR: return "Message error"; case CS_ERR_QUEUE_FULL: return "Queue full"; case CS_ERR_QUEUE_NOT_AVAILABLE: return "Queue not available"; case CS_ERR_BAD_FLAGS: return "Bad flags"; case CS_ERR_TOO_BIG: return "Too big"; case CS_ERR_NO_SECTIONS: return "No sections"; } # endif return "Corosync error"; } # if SUPPORT_COROSYNC #if 0 /* This is the new way to do it, but we still support all Corosync 2 versions, * and this isn't always available. A better alternative here would be to check * for support in the configure script and enable this conditionally. */ #define pcmk__init_cmap(handle) cmap_initialize_map((handle), CMAP_MAP_ICMAP) #else #define pcmk__init_cmap(handle) cmap_initialize(handle) #endif char *pcmk__corosync_cluster_name(void); bool pcmk__corosync_add_nodes(xmlNode *xml_parent); void pcmk__cpg_confchg_cb(cpg_handle_t handle, const struct cpg_name *group_name, const struct cpg_address *member_list, size_t member_list_entries, const struct cpg_address *left_list, size_t left_list_entries, const struct cpg_address *joined_list, size_t joined_list_entries); char *pcmk__cpg_message_data(cpg_handle_t handle, uint32_t sender_id, uint32_t pid, void *content, uint32_t *kind, const char **from); # endif const char *pcmk__cluster_node_uuid(pcmk__node_status_t *node); char *pcmk__cluster_node_name(uint32_t nodeid); const char *pcmk__cluster_local_node_name(void); const char *pcmk__node_name_from_uuid(const char *uuid); pcmk__node_status_t *crm_update_peer_proc(const char *source, pcmk__node_status_t *peer, uint32_t flag, const char *status); pcmk__node_status_t *pcmk__update_peer_state(const char *source, pcmk__node_status_t *node, const char *state, uint64_t membership); void pcmk__update_peer_expected(const char *source, pcmk__node_status_t *node, const char *expected); void pcmk__reap_unseen_nodes(uint64_t ring_id); void pcmk__corosync_quorum_connect(gboolean (*dispatch)(unsigned long long, gboolean), void (*destroy) (gpointer)); -enum crm_ais_msg_types pcmk__cluster_parse_msg_type(const char *text); +enum pcmk__cluster_msg pcmk__cluster_parse_msg_type(const char *text); bool pcmk__cluster_send_message(const pcmk__node_status_t *node, - enum crm_ais_msg_types service, + enum pcmk__cluster_msg service, const xmlNode *data); // Membership bool pcmk__cluster_has_quorum(void); void pcmk__cluster_init_node_caches(void); void pcmk__cluster_destroy_node_caches(void); void pcmk__cluster_set_autoreap(bool enable); void pcmk__cluster_set_status_callback(void (*dispatch)(enum pcmk__node_update, pcmk__node_status_t *, const void *)); bool pcmk__cluster_is_node_active(const pcmk__node_status_t *node); unsigned int pcmk__cluster_num_active_nodes(void); unsigned int pcmk__cluster_num_remote_nodes(void); pcmk__node_status_t *pcmk__cluster_lookup_remote_node(const char *node_name); void pcmk__cluster_forget_cluster_node(uint32_t id, const char *node_name); void pcmk__cluster_forget_remote_node(const char *node_name); pcmk__node_status_t *pcmk__search_node_caches(unsigned int id, const char *uname, uint32_t flags); void pcmk__purge_node_from_cache(const char *node_name, uint32_t node_id); void pcmk__refresh_node_caches_from_cib(xmlNode *cib); pcmk__node_status_t *pcmk__get_node(unsigned int id, const char *uname, const char *uuid, uint32_t flags); #ifdef __cplusplus } #endif #endif // PCMK__CRM_CLUSTER_INTERNAL__H diff --git a/lib/cluster/cluster.c b/lib/cluster/cluster.c index 5cd6263d40..8b250c379b 100644 --- a/lib/cluster/cluster.c +++ b/lib/cluster/cluster.c @@ -1,495 +1,485 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #include #include #include // PRIu32 #include #include #include #include #include #include #include #include #include // uname() #include // gboolean #include #include #include #include #include "crmcluster_private.h" CRM_TRACE_INIT_DATA(cluster); /*! * \internal * \brief Get the message type equivalent of a string * * \param[in] text String of message type * * \return Message type equivalent of \p text */ -enum crm_ais_msg_types +enum pcmk__cluster_msg pcmk__cluster_parse_msg_type(const char *text) { - CRM_CHECK(text != NULL, return crm_msg_none); + CRM_CHECK(text != NULL, return pcmk__cluster_msg_unknown); text = pcmk__message_name(text); - if (pcmk__str_eq(text, "ais", pcmk__str_none)) { - return crm_msg_ais; - } - if (pcmk__str_eq(text, CRM_SYSTEM_CIB, pcmk__str_none)) { - return crm_msg_cib; - } - if (pcmk__str_any_of(text, CRM_SYSTEM_CRMD, CRM_SYSTEM_DC, NULL)) { - return crm_msg_crmd; - } - if (pcmk__str_eq(text, CRM_SYSTEM_TENGINE, pcmk__str_none)) { - return crm_msg_te; - } - if (pcmk__str_eq(text, CRM_SYSTEM_PENGINE, pcmk__str_none)) { - return crm_msg_pe; - } - if (pcmk__str_eq(text, CRM_SYSTEM_LRMD, pcmk__str_none)) { - return crm_msg_lrmd; - } - if (pcmk__str_eq(text, CRM_SYSTEM_STONITHD, pcmk__str_none)) { - return crm_msg_stonithd; - } - if (pcmk__str_eq(text, "stonith-ng", pcmk__str_none)) { - return crm_msg_stonith_ng; - } if (pcmk__str_eq(text, "attrd", pcmk__str_none)) { - return crm_msg_attrd; + return pcmk__cluster_msg_attrd; + + } else if (pcmk__str_eq(text, CRM_SYSTEM_CIB, pcmk__str_none)) { + return pcmk__cluster_msg_based; + + } else if (pcmk__str_any_of(text, CRM_SYSTEM_CRMD, CRM_SYSTEM_DC, NULL)) { + return pcmk__cluster_msg_controld; + + } else if (pcmk__str_eq(text, CRM_SYSTEM_LRMD, pcmk__str_none)) { + return pcmk__cluster_msg_execd; + + } else if (pcmk__str_eq(text, "stonith-ng", pcmk__str_none)) { + return pcmk__cluster_msg_fenced; + + } else { + return pcmk__cluster_msg_unknown; } - return crm_msg_none; } /*! * \internal * \brief Get a node's cluster-layer UUID, setting it if not already set * * \param[in,out] node Node to check * * \return Cluster-layer node UUID of \p node, or \c NULL if unknown */ const char * pcmk__cluster_node_uuid(pcmk__node_status_t *node) { const enum pcmk_cluster_layer cluster_layer = pcmk_get_cluster_layer(); if (node == NULL) { return NULL; } if (node->xml_id != NULL) { return node->xml_id; } switch (cluster_layer) { #if SUPPORT_COROSYNC case pcmk_cluster_layer_corosync: node->xml_id = pcmk__corosync_uuid(node); return node->xml_id; #endif // SUPPORT_COROSYNC default: crm_err("Unsupported cluster layer %s", pcmk_cluster_layer_text(cluster_layer)); return NULL; } } /*! * \internal * \brief Connect to the cluster layer * * \param[in,out] cluster Initialized cluster object to connect * * \return Standard Pacemaker return code */ int pcmk_cluster_connect(pcmk_cluster_t *cluster) { const enum pcmk_cluster_layer cluster_layer = pcmk_get_cluster_layer(); const char *cluster_layer_s = pcmk_cluster_layer_text(cluster_layer); // cts-lab looks for this message crm_notice("Connecting to %s cluster layer", cluster_layer_s); switch (cluster_layer) { #if SUPPORT_COROSYNC case pcmk_cluster_layer_corosync: return pcmk__corosync_connect(cluster); #endif // SUPPORT_COROSYNC default: break; } crm_err("Failed to connect to unsupported cluster layer %s", cluster_layer_s); return EPROTONOSUPPORT; } /*! * \brief Disconnect from the cluster layer * * \param[in,out] cluster Cluster object to disconnect * * \return Standard Pacemaker return code */ int pcmk_cluster_disconnect(pcmk_cluster_t *cluster) { const enum pcmk_cluster_layer cluster_layer = pcmk_get_cluster_layer(); const char *cluster_layer_s = pcmk_cluster_layer_text(cluster_layer); crm_info("Disconnecting from %s cluster layer", cluster_layer_s); switch (cluster_layer) { #if SUPPORT_COROSYNC case pcmk_cluster_layer_corosync: pcmk__corosync_disconnect(cluster); pcmk__cluster_destroy_node_caches(); return pcmk_rc_ok; #endif // SUPPORT_COROSYNC default: break; } crm_err("Failed to disconnect from unsupported cluster layer %s", cluster_layer_s); return EPROTONOSUPPORT; } /*! * \brief Allocate a new \p pcmk_cluster_t object * * \return A newly allocated \p pcmk_cluster_t object (guaranteed not \c NULL) * \note The caller is responsible for freeing the return value using * \p pcmk_cluster_free(). */ pcmk_cluster_t * pcmk_cluster_new(void) { return (pcmk_cluster_t *) pcmk__assert_alloc(1, sizeof(pcmk_cluster_t)); } /*! * \brief Free a \p pcmk_cluster_t object and its dynamically allocated members * * \param[in,out] cluster Cluster object to free */ void pcmk_cluster_free(pcmk_cluster_t *cluster) { if (cluster == NULL) { return; } free(cluster->uuid); free(cluster->uname); free(cluster); } /*! * \brief Set the destroy function for a cluster object * * \param[in,out] cluster Cluster object * \param[in] fn Destroy function to set * * \return Standard Pacemaker return code */ int pcmk_cluster_set_destroy_fn(pcmk_cluster_t *cluster, void (*fn)(gpointer)) { if (cluster == NULL) { return EINVAL; } cluster->destroy = fn; return pcmk_rc_ok; } /*! * \internal * \brief Send an XML message via the cluster messaging layer * * \param[in] node Cluster node to send message to * \param[in] service Message type to use in message host info * \param[in] data XML message to send * * \return \c true on success, or \c false otherwise */ bool pcmk__cluster_send_message(const pcmk__node_status_t *node, - enum crm_ais_msg_types service, const xmlNode *data) + enum pcmk__cluster_msg service, const xmlNode *data) { // @TODO Return standard Pacemaker return code switch (pcmk_get_cluster_layer()) { #if SUPPORT_COROSYNC case pcmk_cluster_layer_corosync: return pcmk__cpg_send_xml(data, node, service); #endif // SUPPORT_COROSYNC default: break; } return false; } /*! * \internal * \brief Get the node name corresponding to a cluster-layer node ID * * Get the node name from the cluster layer if possible. Otherwise, if for the * local node, call \c uname() and get the \c nodename member from the * struct utsname object. * * \param[in] nodeid Node ID to check (or 0 for the local node) * * \return Node name corresponding to \p nodeid * * \note This will fatally exit if \c uname() fails to get the local node name * or we run out of memory. * \note The caller is responsible for freeing the return value using \c free(). */ char * pcmk__cluster_node_name(uint32_t nodeid) { const enum pcmk_cluster_layer cluster_layer = pcmk_get_cluster_layer(); const char *cluster_layer_s = pcmk_cluster_layer_text(cluster_layer); switch (cluster_layer) { #if SUPPORT_COROSYNC case pcmk_cluster_layer_corosync: return pcmk__corosync_name(0, nodeid); #else break; #endif // SUPPORT_COROSYNC default: crm_err("Unsupported cluster layer: %s", cluster_layer_s); break; } if (nodeid == 0) { struct utsname hostinfo; crm_notice("Could not get local node name from %s cluster layer, " "defaulting to local hostname", cluster_layer_s); if (uname(&hostinfo) < 0) { // @TODO Maybe let the caller decide what to do crm_err("Failed to get the local hostname"); crm_exit(CRM_EX_FATAL); } return pcmk__str_copy(hostinfo.nodename); } crm_notice("Could not obtain a node name for node with " PCMK_XA_ID "=" PRIu32, nodeid); return NULL; } /*! * \internal * \brief Get the local node's cluster-layer node name * * If getting the node name from the cluster layer is impossible, call * \c uname() and get the \c nodename member from the struct utsname * object. * * \return Local node's name * * \note This will fatally exit if \c uname() fails to get the local node name * or we run out of memory. */ const char * pcmk__cluster_local_node_name(void) { // @TODO Refactor to avoid trivially leaking name at exit static char *name = NULL; if (name == NULL) { name = pcmk__cluster_node_name(0); } return name; } /*! * \internal * \brief Get the node name corresonding to a node UUID * * Look for the UUID in both the remote node cache and the cluster member cache. * * \param[in] uuid UUID to search for * * \return Node name corresponding to \p uuid if found, or \c NULL otherwise */ const char * pcmk__node_name_from_uuid(const char *uuid) { /* @TODO There are too many functions in libcrmcluster that look up a node * from the node caches (possibly creating a cache entry if none exists). * There are at least the following: * * pcmk__cluster_lookup_remote_node() * * pcmk__get_node() * * pcmk__node_name_from_uuid() * * pcmk__search_node_caches() * * There's a lot of duplication among them, but they all do slightly * different things. We should try to clean them up and consolidate them to * the extent possible, likely with new helper functions. */ GHashTableIter iter; pcmk__node_status_t *node = NULL; CRM_CHECK(uuid != NULL, return NULL); // Remote nodes have the same uname and uuid if (g_hash_table_lookup(crm_remote_peer_cache, uuid)) { return uuid; } g_hash_table_iter_init(&iter, crm_peer_cache); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) { if (pcmk__str_eq(node->xml_id, uuid, pcmk__str_casei)) { return node->name; } } return NULL; } /*! * \brief Get a log-friendly string equivalent of a cluster layer * * \param[in] layer Cluster layer * * \return Log-friendly string corresponding to \p layer */ const char * pcmk_cluster_layer_text(enum pcmk_cluster_layer layer) { switch (layer) { case pcmk_cluster_layer_corosync: return "corosync"; case pcmk_cluster_layer_unknown: return "unknown"; case pcmk_cluster_layer_invalid: return "invalid"; default: crm_err("Invalid cluster layer: %d", layer); return "invalid"; } } /*! * \brief Get and validate the local cluster layer * * If a cluster layer is not configured via the \c PCMK__ENV_CLUSTER_TYPE local * option, this will try to detect an active cluster from among the supported * cluster layers. * * \return Local cluster layer * * \note This will fatally exit if the configured cluster layer is invalid. */ enum pcmk_cluster_layer pcmk_get_cluster_layer(void) { static enum pcmk_cluster_layer cluster_layer = pcmk_cluster_layer_unknown; const char *cluster = NULL; // Cluster layer is stable once set if (cluster_layer != pcmk_cluster_layer_unknown) { return cluster_layer; } cluster = pcmk__env_option(PCMK__ENV_CLUSTER_TYPE); if (cluster != NULL) { crm_info("Verifying configured cluster layer '%s'", cluster); cluster_layer = pcmk_cluster_layer_invalid; #if SUPPORT_COROSYNC if (pcmk__str_eq(cluster, PCMK_VALUE_COROSYNC, pcmk__str_casei)) { cluster_layer = pcmk_cluster_layer_corosync; } #endif // SUPPORT_COROSYNC if (cluster_layer == pcmk_cluster_layer_invalid) { crm_notice("This installation does not support the '%s' cluster " "infrastructure: terminating", cluster); crm_exit(CRM_EX_FATAL); } crm_info("Assuming an active '%s' cluster", cluster); } else { // Nothing configured, so test supported cluster layers #if SUPPORT_COROSYNC crm_debug("Testing with Corosync"); if (pcmk__corosync_is_active()) { cluster_layer = pcmk_cluster_layer_corosync; } #endif // SUPPORT_COROSYNC if (cluster_layer == pcmk_cluster_layer_unknown) { crm_notice("Could not determine the current cluster layer"); } else { crm_info("Detected an active '%s' cluster", pcmk_cluster_layer_text(cluster_layer)); } } return cluster_layer; } // Deprecated functions kept only for backward API compatibility // LCOV_EXCL_START #include gboolean crm_cluster_connect(pcmk_cluster_t *cluster) { return pcmk_cluster_connect(cluster) == pcmk_rc_ok; } const char * name_for_cluster_type(enum cluster_type_e type) { switch (type) { case pcmk_cluster_corosync: return "corosync"; case pcmk_cluster_unknown: return "unknown"; case pcmk_cluster_invalid: return "invalid"; } crm_err("Invalid cluster type: %d", type); return "invalid"; } enum cluster_type_e get_cluster_type(void) { return (enum cluster_type_e) pcmk_get_cluster_layer(); } // LCOV_EXCL_STOP // End deprecated API diff --git a/lib/cluster/cpg.c b/lib/cluster/cpg.c index 2b14909d15..726e136831 100644 --- a/lib/cluster/cpg.c +++ b/lib/cluster/cpg.c @@ -1,1088 +1,1065 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #include #include #include // PRIu32 #include #include #include #include // uint32_t #include #include // size_t #include #include #include #include #include #include #include #include #include #include #include #include // PCMK__SPECIAL_PID #include #include #include "crmcluster_private.h" /* @TODO Once we can update the public API to require pcmk_cluster_t* in more * functions, we can ditch this in favor of cluster->cpg_handle. */ static cpg_handle_t pcmk_cpg_handle = 0; // @TODO These could be moved to pcmk_cluster_t* at that time as well static bool cpg_evicted = false; static GList *cs_message_queue = NULL; static int cs_message_timer = 0; struct pcmk__cpg_host_s { uint32_t id; uint32_t pid; - enum crm_ais_msg_types type; + enum pcmk__cluster_msg type; uint32_t size; char uname[MAX_NAME]; } __attribute__ ((packed)); typedef struct pcmk__cpg_host_s pcmk__cpg_host_t; struct pcmk__cpg_msg_s { struct qb_ipc_response_header header __attribute__ ((aligned(8))); uint32_t id; gboolean is_compressed; pcmk__cpg_host_t host; pcmk__cpg_host_t sender; uint32_t size; uint32_t compressed_size; /* 584 bytes */ char data[0]; } __attribute__ ((packed)); typedef struct pcmk__cpg_msg_s pcmk__cpg_msg_t; static void crm_cs_flush(gpointer data); #define msg_data_len(msg) (msg->is_compressed?msg->compressed_size:msg->size) #define cs_repeat(rc, counter, max, code) do { \ rc = code; \ if ((rc == CS_ERR_TRY_AGAIN) || (rc == CS_ERR_QUEUE_FULL)) { \ counter++; \ crm_debug("Retrying operation after %ds", counter); \ sleep(counter); \ } else { \ break; \ } \ } while (counter < max) /*! * \internal * \brief Get the local Corosync node ID (via CPG) * * \param[in] handle CPG connection to use (or 0 to use new connection) * * \return Corosync ID of local node (or 0 if not known) */ uint32_t pcmk__cpg_local_nodeid(cpg_handle_t handle) { cs_error_t rc = CS_OK; int retries = 0; static uint32_t local_nodeid = 0; cpg_handle_t local_handle = handle; cpg_model_v1_data_t cpg_model_info = {CPG_MODEL_V1, NULL, NULL, NULL, 0}; int fd = -1; uid_t found_uid = 0; gid_t found_gid = 0; pid_t found_pid = 0; int rv = 0; if (local_nodeid != 0) { return local_nodeid; } if (handle == 0) { crm_trace("Creating connection"); cs_repeat(rc, retries, 5, cpg_model_initialize(&local_handle, CPG_MODEL_V1, (cpg_model_data_t *) &cpg_model_info, NULL)); if (rc != CS_OK) { crm_err("Could not connect to the CPG API: %s (%d)", cs_strerror(rc), rc); return 0; } rc = cpg_fd_get(local_handle, &fd); if (rc != CS_OK) { crm_err("Could not obtain the CPG API connection: %s (%d)", cs_strerror(rc), rc); goto bail; } // CPG provider run as root (at least in given user namespace)? rv = crm_ipc_is_authentic_process(fd, (uid_t) 0, (gid_t) 0, &found_pid, &found_uid, &found_gid); if (rv == 0) { crm_err("CPG provider is not authentic:" " process %lld (uid: %lld, gid: %lld)", (long long) PCMK__SPECIAL_PID_AS_0(found_pid), (long long) found_uid, (long long) found_gid); goto bail; } else if (rv < 0) { crm_err("Could not verify authenticity of CPG provider: %s (%d)", strerror(-rv), -rv); goto bail; } } if (rc == CS_OK) { retries = 0; crm_trace("Performing lookup"); cs_repeat(rc, retries, 5, cpg_local_get(local_handle, &local_nodeid)); } if (rc != CS_OK) { crm_err("Could not get local node id from the CPG API: %s (%d)", pcmk__cs_err_str(rc), rc); } bail: if (handle == 0) { crm_trace("Closing connection"); cpg_finalize(local_handle); } crm_debug("Local nodeid is %u", local_nodeid); return local_nodeid; } /*! * \internal * \brief Callback function for Corosync message queue timer * * \param[in] data CPG handle * * \return FALSE (to indicate to glib that timer should not be removed) */ static gboolean crm_cs_flush_cb(gpointer data) { cs_message_timer = 0; crm_cs_flush(data); return FALSE; } // Send no more than this many CPG messages in one flush #define CS_SEND_MAX 200 /*! * \internal * \brief Send messages in Corosync CPG message queue * * \param[in] data CPG handle */ static void crm_cs_flush(gpointer data) { unsigned int sent = 0; guint queue_len = 0; cs_error_t rc = 0; cpg_handle_t *handle = (cpg_handle_t *) data; if (*handle == 0) { crm_trace("Connection is dead"); return; } queue_len = g_list_length(cs_message_queue); if (((queue_len % 1000) == 0) && (queue_len > 1)) { crm_err("CPG queue has grown to %d", queue_len); } else if (queue_len == CS_SEND_MAX) { crm_warn("CPG queue has grown to %d", queue_len); } if (cs_message_timer != 0) { /* There is already a timer, wait until it goes off */ crm_trace("Timer active %d", cs_message_timer); return; } while ((cs_message_queue != NULL) && (sent < CS_SEND_MAX)) { struct iovec *iov = cs_message_queue->data; rc = cpg_mcast_joined(*handle, CPG_TYPE_AGREED, iov, 1); if (rc != CS_OK) { break; } sent++; crm_trace("CPG message sent, size=%llu", (unsigned long long) iov->iov_len); cs_message_queue = g_list_remove(cs_message_queue, iov); free(iov->iov_base); free(iov); } queue_len -= sent; do_crm_log((queue_len > 5)? LOG_INFO : LOG_TRACE, "Sent %u CPG message%s (%d still queued): %s (rc=%d)", sent, pcmk__plural_s(sent), queue_len, pcmk__cs_err_str(rc), (int) rc); if (cs_message_queue) { uint32_t delay_ms = 100; if (rc != CS_OK) { /* Proportionally more if sending failed but cap at 1s */ delay_ms = QB_MIN(1000, CS_SEND_MAX + (10 * queue_len)); } cs_message_timer = g_timeout_add(delay_ms, crm_cs_flush_cb, data); } } /*! * \internal * \brief Dispatch function for CPG handle * * \param[in,out] user_data Cluster object * * \return 0 on success, -1 on error (per mainloop_io_t interface) */ static int pcmk_cpg_dispatch(gpointer user_data) { cs_error_t rc = CS_OK; pcmk_cluster_t *cluster = (pcmk_cluster_t *) user_data; rc = cpg_dispatch(cluster->cpg_handle, CS_DISPATCH_ONE); if (rc != CS_OK) { crm_err("Connection to the CPG API failed: %s (%d)", pcmk__cs_err_str(rc), rc); cpg_finalize(cluster->cpg_handle); cluster->cpg_handle = 0; return -1; } else if (cpg_evicted) { crm_err("Evicted from CPG membership"); return -1; } return 0; } static inline const char * ais_dest(const pcmk__cpg_host_t *host) { return (host->size > 0)? host->uname : ""; } static inline const char * -msg_type2text(enum crm_ais_msg_types type) +msg_type2text(enum pcmk__cluster_msg type) { - const char *text = "unknown"; - switch (type) { - case crm_msg_none: - text = "unknown"; - break; - case crm_msg_ais: - text = "ais"; - break; - case crm_msg_cib: - text = "cib"; - break; - case crm_msg_crmd: - text = "crmd"; - break; - case crm_msg_pe: - text = "pengine"; - break; - case crm_msg_te: - text = "tengine"; - break; - case crm_msg_lrmd: - text = "lrmd"; - break; - case crm_msg_attrd: - text = "attrd"; - break; - case crm_msg_stonithd: - text = "stonithd"; - break; - case crm_msg_stonith_ng: - text = "stonith-ng"; - break; + case pcmk__cluster_msg_attrd: + return "attrd"; + case pcmk__cluster_msg_based: + return "cib"; + case pcmk__cluster_msg_controld: + return "crmd"; + case pcmk__cluster_msg_execd: + return "lrmd"; + case pcmk__cluster_msg_fenced: + return "stonith-ng"; + default: + return "unknown"; } - return text; } /*! * \internal * \brief Check whether a Corosync CPG message is valid * * \param[in] msg Corosync CPG message to check * * \return true if \p msg is valid, otherwise false */ static bool check_message_sanity(const pcmk__cpg_msg_t *msg) { int32_t payload_size = msg->header.size - sizeof(pcmk__cpg_msg_t); if (payload_size < 1) { crm_err("%sCPG message %d from %s invalid: " "Claimed size of %d bytes is too small " QB_XS " from %s[%u] to %s@%s", (msg->is_compressed? "Compressed " : ""), msg->id, ais_dest(&(msg->sender)), (int) msg->header.size, msg_type2text(msg->sender.type), msg->sender.pid, msg_type2text(msg->host.type), ais_dest(&(msg->host))); return false; } if (msg->header.error != CS_OK) { crm_err("%sCPG message %d from %s invalid: " "Sender indicated error %d " QB_XS " from %s[%u] to %s@%s", (msg->is_compressed? "Compressed " : ""), msg->id, ais_dest(&(msg->sender)), msg->header.error, msg_type2text(msg->sender.type), msg->sender.pid, msg_type2text(msg->host.type), ais_dest(&(msg->host))); return false; } if (msg_data_len(msg) != payload_size) { crm_err("%sCPG message %d from %s invalid: " "Total size %d inconsistent with payload size %d " QB_XS " from %s[%u] to %s@%s", (msg->is_compressed? "Compressed " : ""), msg->id, ais_dest(&(msg->sender)), (int) msg->header.size, (int) msg_data_len(msg), msg_type2text(msg->sender.type), msg->sender.pid, msg_type2text(msg->host.type), ais_dest(&(msg->host))); return false; } if (!msg->is_compressed && /* msg->size != (strlen(msg->data) + 1) would be a stronger check, * but checking the last byte or two should be quick */ (((msg->size > 1) && (msg->data[msg->size - 2] == '\0')) || (msg->data[msg->size - 1] != '\0'))) { crm_err("CPG message %d from %s invalid: " "Payload does not end at byte %llu " QB_XS " from %s[%u] to %s@%s", msg->id, ais_dest(&(msg->sender)), (unsigned long long) msg->size, msg_type2text(msg->sender.type), msg->sender.pid, msg_type2text(msg->host.type), ais_dest(&(msg->host))); return false; } crm_trace("Verified %d-byte %sCPG message %d from %s[%u]@%s to %s@%s", (int) msg->header.size, (msg->is_compressed? "compressed " : ""), msg->id, msg_type2text(msg->sender.type), msg->sender.pid, ais_dest(&(msg->sender)), msg_type2text(msg->host.type), ais_dest(&(msg->host))); return true; } /*! * \internal * \brief Extract text data from a Corosync CPG message * * \param[in] handle CPG connection (to get local node ID if not known) * \param[in] sender_id Corosync ID of node that sent message * \param[in] pid Process ID of message sender (for logging only) * \param[in,out] content CPG message * \param[out] kind If not \c NULL, will be set to CPG header ID * (which should be an enum crm_ais_msg_class * value, currently always \c crm_class_cluster) * \param[out] from If not \c NULL, will be set to sender uname * (valid for the lifetime of \p content) * * \return Newly allocated string with message data * * \note The caller is responsible for freeing the return value using \c free(). */ char * pcmk__cpg_message_data(cpg_handle_t handle, uint32_t sender_id, uint32_t pid, void *content, uint32_t *kind, const char **from) { char *data = NULL; pcmk__cpg_msg_t *msg = content; if (handle != 0) { // Do filtering and field massaging uint32_t local_nodeid = pcmk__cpg_local_nodeid(handle); const char *local_name = pcmk__cluster_local_node_name(); if ((msg->sender.id != 0) && (msg->sender.id != sender_id)) { crm_err("Nodeid mismatch from %" PRIu32 ".%" PRIu32 ": claimed nodeid=%" PRIu32, sender_id, pid, msg->sender.id); return NULL; } if ((msg->host.id != 0) && (local_nodeid != msg->host.id)) { crm_trace("Not for us: %" PRIu32" != %" PRIu32, msg->host.id, local_nodeid); return NULL; } if ((msg->host.size > 0) && !pcmk__str_eq(msg->host.uname, local_name, pcmk__str_casei)) { crm_trace("Not for us: %s != %s", msg->host.uname, local_name); return NULL; } msg->sender.id = sender_id; if (msg->sender.size == 0) { const pcmk__node_status_t *peer = pcmk__get_node(sender_id, NULL, NULL, pcmk__node_search_cluster_member); if (peer->name == NULL) { crm_err("No node name for peer with nodeid=%u", sender_id); } else { crm_notice("Fixing node name for peer with nodeid=%u", sender_id); msg->sender.size = strlen(peer->name); memset(msg->sender.uname, 0, MAX_NAME); memcpy(msg->sender.uname, peer->name, msg->sender.size); } } } crm_trace("Got new%s message (size=%d, %d, %d)", msg->is_compressed ? " compressed" : "", msg_data_len(msg), msg->size, msg->compressed_size); if (kind != NULL) { *kind = msg->header.id; } if (from != NULL) { *from = msg->sender.uname; } if (msg->is_compressed && (msg->size > 0)) { int rc = BZ_OK; char *uncompressed = NULL; unsigned int new_size = msg->size + 1; if (!check_message_sanity(msg)) { goto badmsg; } crm_trace("Decompressing message data"); uncompressed = pcmk__assert_alloc(1, new_size); rc = BZ2_bzBuffToBuffDecompress(uncompressed, &new_size, msg->data, msg->compressed_size, 1, 0); rc = pcmk__bzlib2rc(rc); if (rc != pcmk_rc_ok) { crm_err("Decompression failed: %s " QB_XS " rc=%d", pcmk_rc_str(rc), rc); free(uncompressed); goto badmsg; } CRM_ASSERT(new_size == msg->size); data = uncompressed; } else if (!check_message_sanity(msg)) { goto badmsg; } else { data = strdup(msg->data); } // Is this necessary? pcmk__get_node(msg->sender.id, msg->sender.uname, NULL, pcmk__node_search_cluster_member); crm_trace("Payload: %.200s", data); return data; badmsg: crm_err("Invalid message (id=%d, dest=%s:%s, from=%s:%s.%d):" " min=%d, total=%d, size=%d, bz2_size=%d", msg->id, ais_dest(&(msg->host)), msg_type2text(msg->host.type), ais_dest(&(msg->sender)), msg_type2text(msg->sender.type), msg->sender.pid, (int)sizeof(pcmk__cpg_msg_t), msg->header.size, msg->size, msg->compressed_size); free(data); return NULL; } /*! * \internal * \brief Compare cpg_address objects by node ID * * \param[in] first First cpg_address structure to compare * \param[in] second Second cpg_address structure to compare * * \return Negative number if first's node ID is lower, * positive number if first's node ID is greater, * or 0 if both node IDs are equal */ static int cmp_member_list_nodeid(const void *first, const void *second) { const struct cpg_address *const a = *((const struct cpg_address **) first), *const b = *((const struct cpg_address **) second); if (a->nodeid < b->nodeid) { return -1; } else if (a->nodeid > b->nodeid) { return 1; } /* don't bother with "reason" nor "pid" */ return 0; } /*! * \internal * \brief Get a readable string equivalent of a cpg_reason_t value * * \param[in] reason CPG reason value * * \return Readable string suitable for logging */ static const char * cpgreason2str(cpg_reason_t reason) { switch (reason) { case CPG_REASON_JOIN: return " via cpg_join"; case CPG_REASON_LEAVE: return " via cpg_leave"; case CPG_REASON_NODEDOWN: return " via cluster exit"; case CPG_REASON_NODEUP: return " via cluster join"; case CPG_REASON_PROCDOWN: return " for unknown reason"; default: break; } return ""; } /*! * \internal * \brief Get a log-friendly node name * * \param[in] peer Node to check * * \return Node's uname, or readable string if not known */ static inline const char * peer_name(const pcmk__node_status_t *peer) { return (peer != NULL)? pcmk__s(peer->name, "peer node") : "unknown node"; } /*! * \internal * \brief Process a CPG peer's leaving the cluster * * \param[in] cpg_group_name CPG group name (for logging) * \param[in] event_counter Event number (for logging) * \param[in] local_nodeid Node ID of local node * \param[in] cpg_peer CPG peer that left * \param[in] sorted_member_list List of remaining members, qsort()-ed by ID * \param[in] member_list_entries Number of entries in \p sorted_member_list */ static void node_left(const char *cpg_group_name, int event_counter, uint32_t local_nodeid, const struct cpg_address *cpg_peer, const struct cpg_address **sorted_member_list, size_t member_list_entries) { pcmk__node_status_t *peer = pcmk__search_node_caches(cpg_peer->nodeid, NULL, pcmk__node_search_cluster_member); const struct cpg_address **rival = NULL; /* Most CPG-related Pacemaker code assumes that only one process on a node * can be in the process group, but Corosync does not impose this * limitation, and more than one can be a member in practice due to a * daemon attempting to start while another instance is already running. * * Check for any such duplicate instances, because we don't want to process * their leaving as if our actual peer left. If the peer that left still has * an entry in sorted_member_list (with a different PID), we will ignore the * leaving. * * @TODO Track CPG members' PIDs so we can tell exactly who left. */ if (peer != NULL) { rival = bsearch(&cpg_peer, sorted_member_list, member_list_entries, sizeof(const struct cpg_address *), cmp_member_list_nodeid); } if (rival == NULL) { crm_info("Group %s event %d: %s (node %u pid %u) left%s", cpg_group_name, event_counter, peer_name(peer), cpg_peer->nodeid, cpg_peer->pid, cpgreason2str(cpg_peer->reason)); if (peer != NULL) { crm_update_peer_proc(__func__, peer, crm_proc_cpg, PCMK_VALUE_OFFLINE); } } else if (cpg_peer->nodeid == local_nodeid) { crm_warn("Group %s event %d: duplicate local pid %u left%s", cpg_group_name, event_counter, cpg_peer->pid, cpgreason2str(cpg_peer->reason)); } else { crm_warn("Group %s event %d: " "%s (node %u) duplicate pid %u left%s (%u remains)", cpg_group_name, event_counter, peer_name(peer), cpg_peer->nodeid, cpg_peer->pid, cpgreason2str(cpg_peer->reason), (*rival)->pid); } } /*! * \internal * \brief Handle a CPG configuration change event * * \param[in] handle CPG connection * \param[in] group_name CPG group name * \param[in] member_list List of current CPG members * \param[in] member_list_entries Number of entries in \p member_list * \param[in] left_list List of CPG members that left * \param[in] left_list_entries Number of entries in \p left_list * \param[in] joined_list List of CPG members that joined * \param[in] joined_list_entries Number of entries in \p joined_list * * \note This is of type \c cpg_confchg_fn_t, intended to be used in a * \c cpg_callbacks_t object. */ void pcmk__cpg_confchg_cb(cpg_handle_t handle, const struct cpg_name *group_name, const struct cpg_address *member_list, size_t member_list_entries, const struct cpg_address *left_list, size_t left_list_entries, const struct cpg_address *joined_list, size_t joined_list_entries) { static int counter = 0; bool found = false; uint32_t local_nodeid = pcmk__cpg_local_nodeid(handle); const struct cpg_address **sorted = NULL; sorted = pcmk__assert_alloc(member_list_entries, sizeof(const struct cpg_address *)); for (size_t iter = 0; iter < member_list_entries; iter++) { sorted[iter] = member_list + iter; } // So that the cross-matching of multiply-subscribed nodes is then cheap qsort(sorted, member_list_entries, sizeof(const struct cpg_address *), cmp_member_list_nodeid); for (int i = 0; i < left_list_entries; i++) { node_left(group_name->value, counter, local_nodeid, &left_list[i], sorted, member_list_entries); } free(sorted); sorted = NULL; for (int i = 0; i < joined_list_entries; i++) { crm_info("Group %s event %d: node %u pid %u joined%s", group_name->value, counter, joined_list[i].nodeid, joined_list[i].pid, cpgreason2str(joined_list[i].reason)); } for (int i = 0; i < member_list_entries; i++) { pcmk__node_status_t *peer = pcmk__get_node(member_list[i].nodeid, NULL, NULL, pcmk__node_search_cluster_member); if (member_list[i].nodeid == local_nodeid && member_list[i].pid != getpid()) { // See the note in node_left() crm_warn("Group %s event %d: detected duplicate local pid %u", group_name->value, counter, member_list[i].pid); continue; } crm_info("Group %s event %d: %s (node %u pid %u) is member", group_name->value, counter, peer_name(peer), member_list[i].nodeid, member_list[i].pid); /* If the caller left auto-reaping enabled, this will also update the * state to member. */ peer = crm_update_peer_proc(__func__, peer, crm_proc_cpg, PCMK_VALUE_ONLINE); if (peer && peer->state && strcmp(peer->state, CRM_NODE_MEMBER)) { /* The node is a CPG member, but we currently think it's not a * cluster member. This is possible only if auto-reaping was * disabled. The node may be joining, and we happened to get the CPG * notification before the quorum notification; or the node may have * just died, and we are processing its final messages; or a bug * has affected the peer cache. */ time_t now = time(NULL); if (peer->when_lost == 0) { // Track when we first got into this contradictory state peer->when_lost = now; } else if (now > (peer->when_lost + 60)) { // If it persists for more than a minute, update the state crm_warn("Node %u is member of group %s but was believed " "offline", member_list[i].nodeid, group_name->value); pcmk__update_peer_state(__func__, peer, CRM_NODE_MEMBER, 0); } } if (local_nodeid == member_list[i].nodeid) { found = true; } } if (!found) { crm_err("Local node was evicted from group %s", group_name->value); cpg_evicted = true; } counter++; } /*! * \brief Set the CPG deliver callback function for a cluster object * * \param[in,out] cluster Cluster object * \param[in] fn Deliver callback function to set * * \return Standard Pacemaker return code */ int pcmk_cpg_set_deliver_fn(pcmk_cluster_t *cluster, cpg_deliver_fn_t fn) { if (cluster == NULL) { return EINVAL; } cluster->cpg.cpg_deliver_fn = fn; return pcmk_rc_ok; } /*! * \brief Set the CPG config change callback function for a cluster object * * \param[in,out] cluster Cluster object * \param[in] fn Configuration change callback function to set * * \return Standard Pacemaker return code */ int pcmk_cpg_set_confchg_fn(pcmk_cluster_t *cluster, cpg_confchg_fn_t fn) { if (cluster == NULL) { return EINVAL; } cluster->cpg.cpg_confchg_fn = fn; return pcmk_rc_ok; } /*! * \brief Connect to Corosync CPG * * \param[in,out] cluster Initialized cluster object to connect * * \return Standard Pacemaker return code */ int pcmk__cpg_connect(pcmk_cluster_t *cluster) { cs_error_t rc; int fd = -1; int retries = 0; uint32_t id = 0; pcmk__node_status_t *peer = NULL; cpg_handle_t handle = 0; const char *message_name = pcmk__message_name(crm_system_name); uid_t found_uid = 0; gid_t found_gid = 0; pid_t found_pid = 0; int rv; struct mainloop_fd_callbacks cpg_fd_callbacks = { .dispatch = pcmk_cpg_dispatch, .destroy = cluster->destroy, }; cpg_model_v1_data_t cpg_model_info = { .model = CPG_MODEL_V1, .cpg_deliver_fn = cluster->cpg.cpg_deliver_fn, .cpg_confchg_fn = cluster->cpg.cpg_confchg_fn, .cpg_totem_confchg_fn = NULL, .flags = 0, }; cpg_evicted = false; cluster->group.length = 0; cluster->group.value[0] = 0; /* group.value is char[128] */ strncpy(cluster->group.value, message_name, 127); cluster->group.value[127] = 0; cluster->group.length = 1 + QB_MIN(127, strlen(cluster->group.value)); cs_repeat(rc, retries, 30, cpg_model_initialize(&handle, CPG_MODEL_V1, (cpg_model_data_t *)&cpg_model_info, NULL)); if (rc != CS_OK) { crm_err("Could not connect to the CPG API: %s (%d)", cs_strerror(rc), rc); goto bail; } rc = cpg_fd_get(handle, &fd); if (rc != CS_OK) { crm_err("Could not obtain the CPG API connection: %s (%d)", cs_strerror(rc), rc); goto bail; } /* CPG provider run as root (in given user namespace, anyway)? */ if (!(rv = crm_ipc_is_authentic_process(fd, (uid_t) 0,(gid_t) 0, &found_pid, &found_uid, &found_gid))) { crm_err("CPG provider is not authentic:" " process %lld (uid: %lld, gid: %lld)", (long long) PCMK__SPECIAL_PID_AS_0(found_pid), (long long) found_uid, (long long) found_gid); rc = CS_ERR_ACCESS; goto bail; } else if (rv < 0) { crm_err("Could not verify authenticity of CPG provider: %s (%d)", strerror(-rv), -rv); rc = CS_ERR_ACCESS; goto bail; } id = pcmk__cpg_local_nodeid(handle); if (id == 0) { crm_err("Could not get local node id from the CPG API"); goto bail; } cluster->nodeid = id; retries = 0; cs_repeat(rc, retries, 30, cpg_join(handle, &cluster->group)); if (rc != CS_OK) { crm_err("Could not join the CPG group '%s': %d", message_name, rc); goto bail; } pcmk_cpg_handle = handle; cluster->cpg_handle = handle; mainloop_add_fd("corosync-cpg", G_PRIORITY_MEDIUM, fd, cluster, &cpg_fd_callbacks); bail: if (rc != CS_OK) { cpg_finalize(handle); // @TODO Map rc to more specific Pacemaker return code return ENOTCONN; } peer = pcmk__get_node(id, NULL, NULL, pcmk__node_search_cluster_member); crm_update_peer_proc(__func__, peer, crm_proc_cpg, PCMK_VALUE_ONLINE); return pcmk_rc_ok; } /*! * \internal * \brief Disconnect from Corosync CPG * * \param[in,out] cluster Cluster object to disconnect */ void pcmk__cpg_disconnect(pcmk_cluster_t *cluster) { pcmk_cpg_handle = 0; if (cluster->cpg_handle != 0) { crm_trace("Disconnecting CPG"); cpg_leave(cluster->cpg_handle, &cluster->group); cpg_finalize(cluster->cpg_handle); cluster->cpg_handle = 0; } else { crm_info("No CPG connection"); } } /*! * \internal * \brief Send string data via Corosync CPG * * \param[in] data Data to send * \param[in] node Cluster node to send message to * \param[in] dest Type of message to send * * \return \c true on success, or \c false otherwise */ static bool send_cpg_text(const char *data, const pcmk__node_status_t *node, - enum crm_ais_msg_types dest) + enum pcmk__cluster_msg dest) { static int msg_id = 0; static int local_pid = 0; static int local_name_len = 0; static const char *local_name = NULL; char *target = NULL; struct iovec *iov; pcmk__cpg_msg_t *msg = NULL; - CRM_CHECK(dest != crm_msg_ais, return false); - if (local_name == NULL) { local_name = pcmk__cluster_local_node_name(); } if ((local_name_len == 0) && (local_name != NULL)) { local_name_len = strlen(local_name); } if (data == NULL) { data = ""; } if (local_pid == 0) { local_pid = getpid(); } msg = pcmk__assert_alloc(1, sizeof(pcmk__cpg_msg_t)); msg_id++; msg->id = msg_id; msg->header.id = crm_class_cluster; msg->header.error = CS_OK; msg->host.type = dest; if (node != NULL) { if (node->name != NULL) { target = pcmk__str_copy(node->name); msg->host.size = strlen(node->name); memset(msg->host.uname, 0, MAX_NAME); memcpy(msg->host.uname, node->name, msg->host.size); } else { target = crm_strdup_printf("%" PRIu32, node->cluster_layer_id); } msg->host.id = node->cluster_layer_id; } else { target = pcmk__str_copy("all"); } msg->sender.id = 0; msg->sender.type = pcmk__cluster_parse_msg_type(crm_system_name); msg->sender.pid = local_pid; msg->sender.size = local_name_len; memset(msg->sender.uname, 0, MAX_NAME); if ((local_name != NULL) && (msg->sender.size != 0)) { memcpy(msg->sender.uname, local_name, msg->sender.size); } msg->size = 1 + strlen(data); msg->header.size = sizeof(pcmk__cpg_msg_t) + msg->size; if (msg->size < CRM_BZ2_THRESHOLD) { msg = pcmk__realloc(msg, msg->header.size); memcpy(msg->data, data, msg->size); } else { char *compressed = NULL; unsigned int new_size = 0; if (pcmk__compress(data, (unsigned int) msg->size, 0, &compressed, &new_size) == pcmk_rc_ok) { msg->header.size = sizeof(pcmk__cpg_msg_t) + new_size; msg = pcmk__realloc(msg, msg->header.size); memcpy(msg->data, compressed, new_size); msg->is_compressed = TRUE; msg->compressed_size = new_size; } else { // cppcheck seems not to understand the abort logic in pcmk__realloc // cppcheck-suppress memleak msg = pcmk__realloc(msg, msg->header.size); memcpy(msg->data, data, msg->size); } free(compressed); } iov = pcmk__assert_alloc(1, sizeof(struct iovec)); iov->iov_base = msg; iov->iov_len = msg->header.size; if (msg->compressed_size > 0) { crm_trace("Queueing CPG message %u to %s " "(%llu bytes, %d bytes compressed payload): %.200s", msg->id, target, (unsigned long long) iov->iov_len, msg->compressed_size, data); } else { crm_trace("Queueing CPG message %u to %s " "(%llu bytes, %d bytes payload): %.200s", msg->id, target, (unsigned long long) iov->iov_len, msg->size, data); } free(target); cs_message_queue = g_list_append(cs_message_queue, iov); crm_cs_flush(&pcmk_cpg_handle); return true; } /*! * \internal * \brief Send an XML message via Corosync CPG * * \param[in] msg XML message to send * \param[in] node Cluster node to send message to * \param[in] dest Type of message to send * * \return TRUE on success, otherwise FALSE */ bool pcmk__cpg_send_xml(const xmlNode *msg, const pcmk__node_status_t *node, - enum crm_ais_msg_types dest) + enum pcmk__cluster_msg dest) { bool rc = true; GString *data = g_string_sized_new(1024); pcmk__xml_string(msg, 0, data, 0); rc = send_cpg_text(data->str, node, dest); g_string_free(data, TRUE); return rc; } diff --git a/lib/cluster/crmcluster_private.h b/lib/cluster/crmcluster_private.h index 9fda73367c..ff6487b376 100644 --- a/lib/cluster/crmcluster_private.h +++ b/lib/cluster/crmcluster_private.h @@ -1,80 +1,80 @@ /* * Copyright 2020-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #ifndef PCMK__CLUSTER_CRMCLUSTER_PRIVATE__H #define PCMK__CLUSTER_CRMCLUSTER_PRIVATE__H /* This header is for the sole use of libcrmcluster, so that functions can be * declared with G_GNUC_INTERNAL for efficiency. */ #include // bool #include // uint32_t, uint64_t #include // G_GNUC_INTERNAL, gboolean #include // xmlNode #if SUPPORT_COROSYNC #include // cpg_handle_t #endif // SUPPORT_COROSYNC #include // pcmk_cluster_t #include // pcmk__node_status_t #ifdef __cplusplus extern "C" { #endif G_GNUC_INTERNAL void pcmk__cluster_set_quorum(bool quorate); #if SUPPORT_COROSYNC G_GNUC_INTERNAL bool pcmk__corosync_is_active(void); G_GNUC_INTERNAL bool pcmk__corosync_has_nodelist(void); G_GNUC_INTERNAL char *pcmk__corosync_uuid(const pcmk__node_status_t *peer); G_GNUC_INTERNAL char *pcmk__corosync_name(uint64_t /*cmap_handle_t */ cmap_handle, uint32_t nodeid); G_GNUC_INTERNAL int pcmk__corosync_connect(pcmk_cluster_t *cluster); G_GNUC_INTERNAL void pcmk__corosync_disconnect(pcmk_cluster_t *cluster); G_GNUC_INTERNAL bool pcmk__corosync_is_peer_active(const pcmk__node_status_t *node); G_GNUC_INTERNAL int pcmk__cpg_connect(pcmk_cluster_t *cluster); G_GNUC_INTERNAL void pcmk__cpg_disconnect(pcmk_cluster_t *cluster); G_GNUC_INTERNAL uint32_t pcmk__cpg_local_nodeid(cpg_handle_t handle); G_GNUC_INTERNAL bool pcmk__cpg_send_xml(const xmlNode *msg, const pcmk__node_status_t *node, - enum crm_ais_msg_types dest); + enum pcmk__cluster_msg dest); #endif // SUPPORT_COROSYNC #ifdef __cplusplus } #endif #endif // PCMK__CLUSTER_CRMCLUSTER_PRIVATE__H diff --git a/lib/cluster/election.c b/lib/cluster/election.c index 9ebe7c7e9f..8aeea1f7cb 100644 --- a/lib/cluster/election.c +++ b/lib/cluster/election.c @@ -1,727 +1,729 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #define STORM_INTERVAL 2 /* in seconds */ struct election_s { enum election_result state; guint count; // How many times local node has voted char *name; // Descriptive name for this election char *uname; // Local node's name GSourceFunc cb; // Function to call if election is won GHashTable *voted; // Key = node name, value = how node voted mainloop_timer_t *timeout; // When to abort if all votes not received int election_wins; // Track wins, for storm detection bool wrote_blackbox; // Write a storm blackbox at most once time_t expires; // When storm detection period ends time_t last_election_loss; // When dampening period ends }; static void election_complete(election_t *e) { e->state = election_won; if (e->cb != NULL) { e->cb(e); } election_reset(e); } static gboolean election_timer_cb(gpointer user_data) { election_t *e = user_data; crm_info("%s timed out, declaring local node as winner", e->name); election_complete(e); return FALSE; } /*! * \brief Get current state of an election * * \param[in] e Election object * * \return Current state of \e */ enum election_result election_state(const election_t *e) { return (e == NULL)? election_error : e->state; } /*! * \brief Create a new election object * * Every node that wishes to participate in an election must create an election * object. Typically, this should be done once, at start-up. A caller should * only create a single election object. * * \param[in] name Label for election (for logging) * \param[in] uname Local node's name * \param[in] period_ms How long to wait for all peers to vote * \param[in] cb Function to call if local node wins election * * \return Newly allocated election object on success, NULL on error * \note The caller is responsible for freeing the returned value using * election_fini(). */ election_t * election_init(const char *name, const char *uname, guint period_ms, GSourceFunc cb) { election_t *e = NULL; static guint count = 0; CRM_CHECK(uname != NULL, return NULL); e = calloc(1, sizeof(election_t)); if (e == NULL) { crm_perror(LOG_CRIT, "Cannot create election"); return NULL; } e->uname = strdup(uname); if (e->uname == NULL) { crm_perror(LOG_CRIT, "Cannot create election"); free(e); return NULL; } e->name = name? crm_strdup_printf("election-%s", name) : crm_strdup_printf("election-%u", count++); e->cb = cb; e->timeout = mainloop_timer_add(e->name, period_ms, FALSE, election_timer_cb, e); crm_trace("Created %s", e->name); return e; } /*! * \brief Disregard any previous vote by specified peer * * This discards any recorded vote from a specified peer. Election users should * call this whenever a voting peer becomes inactive. * * \param[in,out] e Election object * \param[in] uname Name of peer to disregard */ void election_remove(election_t *e, const char *uname) { if ((e != NULL) && (uname != NULL) && (e->voted != NULL)) { crm_trace("Discarding %s (no-)vote from lost peer %s", e->name, uname); g_hash_table_remove(e->voted, uname); } } /*! * \brief Stop election timer and disregard all votes * * \param[in,out] e Election object */ void election_reset(election_t *e) { if (e != NULL) { crm_trace("Resetting election %s", e->name); mainloop_timer_stop(e->timeout); if (e->voted) { crm_trace("Destroying voted cache with %d members", g_hash_table_size(e->voted)); g_hash_table_destroy(e->voted); e->voted = NULL; } } } /*! * \brief Free an election object * * Free all memory associated with an election object, stopping its * election timer (if running). * * \param[in,out] e Election object */ void election_fini(election_t *e) { if (e != NULL) { election_reset(e); crm_trace("Destroying %s", e->name); mainloop_timer_del(e->timeout); free(e->uname); free(e->name); free(e); } } static void election_timeout_start(election_t *e) { if (e != NULL) { mainloop_timer_start(e->timeout); } } /*! * \brief Stop an election's timer, if running * * \param[in,out] e Election object */ void election_timeout_stop(election_t *e) { if (e != NULL) { mainloop_timer_stop(e->timeout); } } /*! * \brief Change an election's timeout (restarting timer if running) * * \param[in,out] e Election object * \param[in] period New timeout */ void election_timeout_set_period(election_t *e, guint period) { if (e != NULL) { mainloop_timer_set_period(e->timeout, period); } else { crm_err("No election defined"); } } static int get_uptime(struct timeval *output) { static time_t expires = 0; static struct rusage info; time_t tm_now = time(NULL); if (expires < tm_now) { int rc = 0; info.ru_utime.tv_sec = 0; info.ru_utime.tv_usec = 0; rc = getrusage(RUSAGE_SELF, &info); output->tv_sec = 0; output->tv_usec = 0; if (rc < 0) { crm_perror(LOG_ERR, "Could not calculate the current uptime"); expires = 0; return -1; } crm_debug("Current CPU usage is: %lds, %ldus", (long)info.ru_utime.tv_sec, (long)info.ru_utime.tv_usec); } expires = tm_now + STORM_INTERVAL; /* N seconds after the last _access_ */ output->tv_sec = info.ru_utime.tv_sec; output->tv_usec = info.ru_utime.tv_usec; return 1; } static int compare_age(struct timeval your_age) { struct timeval our_age; get_uptime(&our_age); /* If an error occurred, our_age will be compared as {0,0} */ if (our_age.tv_sec > your_age.tv_sec) { crm_debug("Win: %ld vs %ld (seconds)", (long)our_age.tv_sec, (long)your_age.tv_sec); return 1; } else if (our_age.tv_sec < your_age.tv_sec) { crm_debug("Lose: %ld vs %ld (seconds)", (long)our_age.tv_sec, (long)your_age.tv_sec); return -1; } else if (our_age.tv_usec > your_age.tv_usec) { crm_debug("Win: %ld.%06ld vs %ld.%06ld (usec)", (long)our_age.tv_sec, (long)our_age.tv_usec, (long)your_age.tv_sec, (long)your_age.tv_usec); return 1; } else if (our_age.tv_usec < your_age.tv_usec) { crm_debug("Lose: %ld.%06ld vs %ld.%06ld (usec)", (long)our_age.tv_sec, (long)our_age.tv_usec, (long)your_age.tv_sec, (long)your_age.tv_usec); return -1; } return 0; } /*! * \brief Start a new election by offering local node's candidacy * * Broadcast a "vote" election message containing the local node's ID, * (incremented) election counter, and uptime, and start the election timer. * * \param[in,out] e Election object * * \note Any nodes agreeing to the candidacy will send a "no-vote" reply, and if * all active peers do so, or if the election times out, the local node * wins the election. (If we lose to any peer vote, we will stop the * timer, so a timeout means we did not lose -- either some peer did not * vote, or we did not call election_check() in time.) */ void election_vote(election_t *e) { struct timeval age; xmlNode *vote = NULL; pcmk__node_status_t *our_node = NULL; if (e == NULL) { crm_trace("Election vote requested, but no election available"); return; } our_node = pcmk__get_node(0, e->uname, NULL, pcmk__node_search_cluster_member); if (!pcmk__cluster_is_node_active(our_node)) { crm_trace("Cannot vote in %s yet: local node not connected to cluster", e->name); return; } election_reset(e); e->state = election_in_progress; vote = create_request(CRM_OP_VOTE, NULL, NULL, CRM_SYSTEM_CRMD, CRM_SYSTEM_CRMD, NULL); e->count++; crm_xml_add(vote, PCMK__XA_ELECTION_OWNER, our_node->xml_id); crm_xml_add_int(vote, PCMK__XA_ELECTION_ID, e->count); // Warning: PCMK__XA_ELECTION_AGE_NANO_SEC value is actually microseconds get_uptime(&age); crm_xml_add_timeval(vote, PCMK__XA_ELECTION_AGE_SEC, PCMK__XA_ELECTION_AGE_NANO_SEC, &age); - pcmk__cluster_send_message(NULL, crm_msg_crmd, vote); + pcmk__cluster_send_message(NULL, pcmk__cluster_msg_controld, vote); pcmk__xml_free(vote); crm_debug("Started %s round %d", e->name, e->count); election_timeout_start(e); return; } /*! * \brief Check whether local node has won an election * * If all known peers have sent no-vote messages, stop the election timer, set * the election state to won, and call any registered win callback. * * \param[in,out] e Election object * * \return TRUE if local node has won, FALSE otherwise * \note If all known peers have sent no-vote messages, but the election owner * does not call this function, the election will not be won (and the * callback will not be called) until the election times out. * \note This should be called when election_count_vote() returns * \c election_in_progress. */ bool election_check(election_t *e) { int voted_size = 0; int num_members = 0; if (e == NULL) { crm_trace("Election check requested, but no election available"); return FALSE; } if (e->voted == NULL) { crm_trace("%s check requested, but no votes received yet", e->name); return FALSE; } voted_size = g_hash_table_size(e->voted); num_members = pcmk__cluster_num_active_nodes(); /* in the case of #voted > #members, it is better to * wait for the timeout and give the cluster time to * stabilize */ if (voted_size >= num_members) { /* we won and everyone has voted */ election_timeout_stop(e); if (voted_size > num_members) { GHashTableIter gIter; const pcmk__node_status_t *node = NULL; char *key = NULL; crm_warn("Received too many votes in %s", e->name); g_hash_table_iter_init(&gIter, crm_peer_cache); while (g_hash_table_iter_next(&gIter, NULL, (gpointer *) & node)) { if (pcmk__cluster_is_node_active(node)) { crm_warn("* expected vote: %s", node->name); } } g_hash_table_iter_init(&gIter, e->voted); while (g_hash_table_iter_next(&gIter, (gpointer *) & key, NULL)) { crm_warn("* actual vote: %s", key); } } crm_info("%s won by local node", e->name); election_complete(e); return TRUE; } else { crm_debug("%s still waiting on %d of %d votes", e->name, num_members - voted_size, num_members); } return FALSE; } #define LOSS_DAMPEN 2 /* in seconds */ struct vote { const char *op; const char *from; const char *version; const char *election_owner; int election_id; struct timeval age; }; /*! * \brief Unpack an election message * * \param[in] e Election object (for logging only) * \param[in] message Election message XML * \param[out] vote Parsed fields from message * * \return TRUE if election message and election are valid, FALSE otherwise * \note The parsed struct's pointer members are valid only for the lifetime of * the message argument. */ static bool parse_election_message(const election_t *e, const xmlNode *message, struct vote *vote) { CRM_CHECK(message && vote, return FALSE); vote->election_id = -1; vote->age.tv_sec = -1; vote->age.tv_usec = -1; vote->op = crm_element_value(message, PCMK__XA_CRM_TASK); vote->from = crm_element_value(message, PCMK__XA_SRC); vote->version = crm_element_value(message, PCMK_XA_VERSION); vote->election_owner = crm_element_value(message, PCMK__XA_ELECTION_OWNER); crm_element_value_int(message, PCMK__XA_ELECTION_ID, &(vote->election_id)); if ((vote->op == NULL) || (vote->from == NULL) || (vote->version == NULL) || (vote->election_owner == NULL) || (vote->election_id < 0)) { crm_warn("Invalid %s message from %s in %s ", (vote->op? vote->op : "election"), (vote->from? vote->from : "unspecified node"), (e? e->name : "election")); return FALSE; } // Op-specific validation if (pcmk__str_eq(vote->op, CRM_OP_VOTE, pcmk__str_none)) { /* Only vote ops have uptime. Warning: PCMK__XA_ELECTION_AGE_NANO_SEC value is in microseconds. */ crm_element_value_timeval(message, PCMK__XA_ELECTION_AGE_SEC, PCMK__XA_ELECTION_AGE_NANO_SEC, &(vote->age)); if ((vote->age.tv_sec < 0) || (vote->age.tv_usec < 0)) { crm_warn("Cannot count %s %s from %s because it is missing uptime", (e? e->name : "election"), vote->op, vote->from); return FALSE; } } else if (!pcmk__str_eq(vote->op, CRM_OP_NOVOTE, pcmk__str_none)) { crm_info("Cannot process %s message from %s because %s is not a known election op", (e? e->name : "election"), vote->from, vote->op); return FALSE; } // Election validation if (e == NULL) { crm_info("Cannot count %s from %s because no election available", vote->op, vote->from); return FALSE; } /* If the membership cache is NULL, we REALLY shouldn't be voting -- * the question is how we managed to get here. */ if (crm_peer_cache == NULL) { crm_info("Cannot count %s %s from %s because no peer information available", e->name, vote->op, vote->from); return FALSE; } return TRUE; } static void record_vote(election_t *e, struct vote *vote) { CRM_ASSERT(e && vote && vote->from && vote->op); if (e->voted == NULL) { e->voted = pcmk__strkey_table(free, free); } pcmk__insert_dup(e->voted, vote->from, vote->op); } static void send_no_vote(pcmk__node_status_t *peer, struct vote *vote) { - // @TODO probably shouldn't hardcode CRM_SYSTEM_CRMD and crm_msg_crmd + /* @TODO probably shouldn't hardcode CRM_SYSTEM_CRMD and + * pcmk__cluster_msg_controld + */ xmlNode *novote = create_request(CRM_OP_NOVOTE, NULL, vote->from, CRM_SYSTEM_CRMD, CRM_SYSTEM_CRMD, NULL); crm_xml_add(novote, PCMK__XA_ELECTION_OWNER, vote->election_owner); crm_xml_add_int(novote, PCMK__XA_ELECTION_ID, vote->election_id); - pcmk__cluster_send_message(peer, crm_msg_crmd, novote); + pcmk__cluster_send_message(peer, pcmk__cluster_msg_controld, novote); pcmk__xml_free(novote); } /*! * \brief Process an election message (vote or no-vote) from a peer * * \param[in,out] e Election object * \param[in] message Election message XML from peer * \param[in] can_win Whether local node is eligible to win * * \return Election state after new vote is considered * \note If the peer message is a vote, and we prefer the peer to win, this will * send a no-vote reply to the peer. * \note The situations "we lost to this vote" from "this is a late no-vote * after we've already lost" both return election_lost. If a caller needs * to distinguish them, it should save the current state before calling * this function, and then compare the result. */ enum election_result election_count_vote(election_t *e, const xmlNode *message, bool can_win) { int log_level = LOG_INFO; gboolean done = FALSE; gboolean we_lose = FALSE; const char *reason = "unknown"; bool we_are_owner = FALSE; pcmk__node_status_t *our_node = NULL; pcmk__node_status_t *your_node = NULL; time_t tm_now = time(NULL); struct vote vote; CRM_CHECK(message != NULL, return election_error); if (parse_election_message(e, message, &vote) == FALSE) { return election_error; } your_node = pcmk__get_node(0, vote.from, NULL, pcmk__node_search_cluster_member); our_node = pcmk__get_node(0, e->uname, NULL, pcmk__node_search_cluster_member); we_are_owner = (our_node != NULL) && pcmk__str_eq(our_node->xml_id, vote.election_owner, pcmk__str_none); if (!can_win) { reason = "Not eligible"; we_lose = TRUE; } else if (!pcmk__cluster_is_node_active(our_node)) { reason = "We are not part of the cluster"; log_level = LOG_ERR; we_lose = TRUE; } else if (we_are_owner && (vote.election_id != e->count)) { log_level = LOG_TRACE; reason = "Superseded"; done = TRUE; } else if (!pcmk__cluster_is_node_active(your_node)) { /* Possibly we cached the message in the FSA queue at a point that it wasn't */ reason = "Peer is not part of our cluster"; log_level = LOG_WARNING; done = TRUE; } else if (pcmk__str_eq(vote.op, CRM_OP_NOVOTE, pcmk__str_none) || pcmk__str_eq(vote.from, e->uname, pcmk__str_none)) { /* Receiving our own broadcast vote, or a no-vote from peer, is a vote * for us to win */ if (!we_are_owner) { crm_warn("Cannot count %s round %d %s from %s because we are not election owner (%s)", e->name, vote.election_id, vote.op, vote.from, vote.election_owner); return election_error; } if (e->state != election_in_progress) { // Should only happen if we already lost crm_debug("Not counting %s round %d %s from %s because no election in progress", e->name, vote.election_id, vote.op, vote.from); return e->state; } record_vote(e, &vote); reason = "Recorded"; done = TRUE; } else { // A peer vote requires a comparison to determine which node is better int age_result = compare_age(vote.age); int version_result = compare_version(vote.version, CRM_FEATURE_SET); if (version_result < 0) { reason = "Version"; we_lose = TRUE; } else if (version_result > 0) { reason = "Version"; } else if (age_result < 0) { reason = "Uptime"; we_lose = TRUE; } else if (age_result > 0) { reason = "Uptime"; } else if (strcasecmp(e->uname, vote.from) > 0) { reason = "Host name"; we_lose = TRUE; } else { reason = "Host name"; } } if (e->expires < tm_now) { e->election_wins = 0; e->expires = tm_now + STORM_INTERVAL; } else if (done == FALSE && we_lose == FALSE) { int peers = 1 + g_hash_table_size(crm_peer_cache); /* If every node has to vote down every other node, thats N*(N-1) total elections * Allow some leeway before _really_ complaining */ e->election_wins++; if (e->election_wins > (peers * peers)) { crm_warn("%s election storm detected: %d wins in %d seconds", e->name, e->election_wins, STORM_INTERVAL); e->election_wins = 0; e->expires = tm_now + STORM_INTERVAL; if (e->wrote_blackbox == FALSE) { /* It's questionable whether a black box (from every node in the * cluster) would be truly helpful in diagnosing an election * storm. It's also highly doubtful a production environment * would get multiple election storms from distinct causes, so * saving one blackbox per process lifetime should be * sufficient. Alternatives would be to save a timestamp of the * last blackbox write instead of a boolean, and write a new one * if some amount of time has passed; or to save a storm count, * write a blackbox on every Nth occurrence. */ crm_write_blackbox(0, NULL); e->wrote_blackbox = TRUE; } } } if (done) { do_crm_log(log_level + 1, "Processed %s round %d %s (current round %d) from %s (%s)", e->name, vote.election_id, vote.op, e->count, vote.from, reason); return e->state; } else if (we_lose == FALSE) { /* We track the time of the last election loss to implement an election * dampening period, reducing the likelihood of an election storm. If * this node has lost within the dampening period, don't start a new * election, even if we win against a peer's vote -- the peer we lost to * should win again. * * @TODO This has a problem case: if an election winner immediately * leaves the cluster, and a new election is immediately called, all * nodes could lose, with no new winner elected. The ideal solution * would be to tie the election structure with the peer caches, which * would allow us to clear the dampening when the previous winner * leaves (and would allow other improvements as well). */ if ((e->last_election_loss == 0) || ((tm_now - e->last_election_loss) > (time_t) LOSS_DAMPEN)) { do_crm_log(log_level, "%s round %d (owner node ID %s) pass: %s from %s (%s)", e->name, vote.election_id, vote.election_owner, vote.op, vote.from, reason); e->last_election_loss = 0; election_timeout_stop(e); /* Start a new election by voting down this, and other, peers */ e->state = election_start; return e->state; } else { char *loss_time = ctime(&e->last_election_loss); if (loss_time) { // Show only HH:MM:SS loss_time += 11; loss_time[8] = '\0'; } crm_info("Ignoring %s round %d (owner node ID %s) pass vs %s because we lost less than %ds ago at %s", e->name, vote.election_id, vote.election_owner, vote.from, LOSS_DAMPEN, (loss_time? loss_time : "unknown")); } } e->last_election_loss = tm_now; do_crm_log(log_level, "%s round %d (owner node ID %s) lost: %s from %s (%s)", e->name, vote.election_id, vote.election_owner, vote.op, vote.from, reason); election_reset(e); send_no_vote(your_node, &vote); e->state = election_lost; return e->state; } /*! * \brief Reset any election dampening currently in effect * * \param[in,out] e Election object to clear */ void election_clear_dampening(election_t *e) { e->last_election_loss = 0; }