diff --git a/daemons/based/based_callbacks.c b/daemons/based/based_callbacks.c index d2ff9503d1..7af52ebee1 100644 --- a/daemons/based/based_callbacks.c +++ b/daemons/based/based_callbacks.c @@ -1,1761 +1,1761 @@ /* * 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 static unsigned long cib_local_bcast_num = 0; typedef struct cib_local_notify_s { xmlNode *notify_src; char *client_id; gboolean from_peer; gboolean sync_reply; } cib_local_notify_t; int next_client_id = 0; gboolean legacy_mode = FALSE; 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); gboolean cib_legacy_mode(void) { return legacy_mode; } 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 * * \note The caller is responsible for freeing the return value using * \p free_xml(). */ 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 rid = 0; int call_id = 0; pcmk__client_t *client_obj = NULL; CRM_ASSERT(notify_src && client_id); crm_element_value_int(notify_src, PCMK__XA_CIB_CALLID, &call_id); client_obj = pcmk__find_client_by_id(client_id); if (client_obj == NULL) { crm_debug("Could not send response %d: client %s not found", call_id, client_id); return; } if (sync_reply) { if (client_obj->ipcs) { CRM_LOG_ASSERT(client_obj->request_id); rid = client_obj->request_id; client_obj->request_id = 0; crm_trace("Sending response %d to client %s%s", rid, pcmk__client_name(client_obj), (from_peer? " (originator of delegated request)" : "")); } else { crm_trace("Sending response (call %d) to client %s%s", call_id, pcmk__client_name(client_obj), (from_peer? " (originator of delegated request)" : "")); } } else { crm_trace("Sending event %d to client %s%s", call_id, pcmk__client_name(client_obj), (from_peer? " (originator of delegated request)" : "")); } switch (PCMK__CLIENT_TYPE(client_obj)) { case pcmk__client_ipc: { int rc = pcmk__ipc_send_xml(client_obj, rid, notify_src, (sync_reply? crm_ipc_flags_none : crm_ipc_server_event)); if (rc != pcmk_rc_ok) { crm_warn("%s reply to client %s failed: %s " CRM_XS " rc=%d", (sync_reply? "Synchronous" : "Asynchronous"), pcmk__client_name(client_obj), pcmk_rc_str(rc), rc); } } break; #ifdef HAVE_GNUTLS_GNUTLS_H case pcmk__client_tls: #endif case pcmk__client_tcp: pcmk__remote_send_xml(client_obj->remote, notify_src); break; default: crm_err("Unknown transport for client %s " CRM_XS " flags=%#016" PRIx64, pcmk__client_name(client_obj), client_obj->flags); } } 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; free_xml(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); free_xml(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); send_cluster_message(NULL, crm_msg_cib, ping, TRUE); free_xml(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 { const char *version = crm_element_value(pong, PCMK_XA_CRM_FEATURE_SET); if(ping_digest == NULL) { crm_trace("Calculating new digest"); ping_digest = calculate_xml_versioned_digest(the_cib, FALSE, TRUE, version); } 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"); } free_xml(remote_cib); sync_our_cib(reply, FALSE); } } } static void local_notify_destroy_callback(gpointer data) { cib_local_notify_t *notify = data; free_xml(notify->notify_src); free(notify->client_id); free(notify); } static void check_local_notify(int bcast_id) { const cib_local_notify_t *notify = NULL; if (!local_notify_queue) { return; } notify = pcmk__intkey_table_lookup(local_notify_queue, bcast_id); if (notify) { do_local_notify(notify->notify_src, notify->client_id, notify->sync_reply, notify->from_peer); pcmk__intkey_table_remove(local_notify_queue, bcast_id); } } static void queue_local_notify(xmlNode * notify_src, const char *client_id, gboolean sync_reply, gboolean from_peer) { cib_local_notify_t *notify = pcmk__assert_alloc(1, sizeof(cib_local_notify_t)); notify->notify_src = notify_src; notify->client_id = pcmk__str_copy(client_id); notify->sync_reply = sync_reply; notify->from_peer = from_peer; if (!local_notify_queue) { local_notify_queue = pcmk__intkey_table(local_notify_destroy_callback); } pcmk__intkey_table_insert(local_notify_queue, cib_local_bcast_num, notify); // cppcheck doesn't know notify will get freed when hash table is destroyed // cppcheck-suppress memleak } static void parse_local_options_v1(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) { if (pcmk_is_set(operation->flags, cib__op_attr_modifies) && !pcmk_is_set(call_options, cib_inhibit_bcast)) { /* we need to send an update anyway */ *needs_reply = TRUE; } else { *needs_reply = FALSE; } if (host == NULL && (call_options & cib_scope_local)) { crm_trace("Processing locally scoped %s op from client %s", op, pcmk__client_name(cib_client)); *local_notify = TRUE; } else if ((host == NULL) && based_is_primary) { crm_trace("Processing %s op locally from client %s as primary", op, pcmk__client_name(cib_client)); *local_notify = TRUE; } else if (pcmk__str_eq(host, OUR_NODENAME, pcmk__str_casei)) { crm_trace("Processing locally addressed %s op from client %s", op, pcmk__client_name(cib_client)); *local_notify = TRUE; } else if (stand_alone) { *needs_forward = FALSE; *local_notify = TRUE; *process = TRUE; } else { crm_trace("%s op from %s needs to be forwarded to client %s", op, pcmk__client_name(cib_client), pcmk__s(host, "the primary instance")); *needs_forward = TRUE; *process = FALSE; } } static void parse_local_options_v2(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 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) { if(cib_legacy_mode()) { parse_local_options_v1(cib_client, operation, call_options, host, op, local_notify, needs_reply, process, needs_forward); } else { parse_local_options_v2(cib_client, operation, call_options, host, op, local_notify, needs_reply, process, needs_forward); } } static gboolean parse_peer_options_v1(const cib__operation_t *operation, xmlNode *request, gboolean *local_notify, gboolean *needs_reply, gboolean *process) { const char *op = NULL; const char *host = NULL; const char *delegated = NULL; 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 (pcmk__xe_attr_is_true(request, PCMK__XA_CIB_UPDATE)) { *needs_reply = FALSE; if (is_reply) { *local_notify = TRUE; crm_trace("Processing global/peer update from %s" " that originated from us", originator); } else { crm_trace("Processing global/peer update from %s", originator); } return TRUE; } op = crm_element_value(request, PCMK__XA_CIB_OP); crm_trace("Processing legacy %s request sent by %s", op, originator); if (pcmk__str_eq(op, PCMK__CIB_REQUEST_SHUTDOWN, pcmk__str_none)) { /* Always process these */ *local_notify = FALSE; if (reply_to == NULL || is_reply) { *process = TRUE; } if (is_reply) { *needs_reply = FALSE; } return *process; } if (is_reply && pcmk__str_eq(op, CRM_OP_PING, pcmk__str_casei)) { process_ping_reply(request); return FALSE; } if (is_reply) { crm_trace("Forward reply sent from %s to local clients", originator); *process = FALSE; *needs_reply = FALSE; *local_notify = TRUE; return TRUE; } 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); return TRUE; } else if(is_reply == FALSE && pcmk__str_eq(op, CRM_OP_PING, pcmk__str_casei)) { crm_trace("Processing %s request sent to %s by %s", op, host?host:"everyone", originator); *needs_reply = TRUE; return TRUE; } else if ((host == NULL) && based_is_primary) { crm_trace("Processing %s request sent to primary instance from %s", op, originator); return TRUE; } delegated = crm_element_value(request, PCMK__XA_CIB_DELEGATED_FROM); if (delegated != NULL) { crm_trace("Ignoring message for primary instance"); } else if (host != NULL) { /* this is for a specific instance and we're not it */ crm_trace("Ignoring msg for instance on %s", host); } else if ((reply_to == NULL) && !based_is_primary) { // This is for the primary instance, and we're not it crm_trace("Ignoring reply for primary instance"); } else if (pcmk__str_eq(op, PCMK__CIB_REQUEST_SHUTDOWN, pcmk__str_none)) { if (reply_to != NULL) { crm_debug("Processing %s from %s", op, originator); *needs_reply = FALSE; } else { crm_debug("Processing %s reply from %s", op, originator); } return TRUE; } else { crm_err("Nothing for us to do?"); crm_log_xml_err(request, "Peer[inbound]"); } return FALSE; } static gboolean parse_peer_options_v2(const cib__operation_t *operation, xmlNode *request, gboolean *local_notify, gboolean *needs_reply, gboolean *process) { 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); legacy_mode = TRUE; 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; } 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) */ if(cib_legacy_mode()) { return parse_peer_options_v1(operation, request, local_notify, needs_reply, process); } else { return parse_peer_options_v2(operation, request, local_notify, needs_reply, process); } } /*! * \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); crm_node_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, (cib_legacy_mode()? "primary" : "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); } send_cluster_message(peer, crm_msg_cib, request, FALSE); // Return the request to its original state pcmk__xe_remove_attr(request, PCMK__XA_CIB_DELEGATED_FROM); } static gboolean send_peer_reply(xmlNode * msg, xmlNode * result_diff, const char *originator, gboolean broadcast) { CRM_ASSERT(msg != NULL); if (broadcast) { /* @COMPAT: Legacy code * * This successful call modified the CIB, and the change needs to be * broadcast (sent via cluster to all nodes). */ 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; const char *digest = NULL; int format = 1; xmlNode *wrapper = NULL; CRM_LOG_ASSERT(result_diff != NULL); digest = crm_element_value(result_diff, PCMK__XA_DIGEST); crm_element_value_int(result_diff, PCMK_XA_FORMAT, &format); cib_diff_version_details(result_diff, &diff_add_admin_epoch, &diff_add_epoch, &diff_add_updates, &diff_del_admin_epoch, &diff_del_epoch, &diff_del_updates); crm_trace("Sending update diff %d.%d.%d -> %d.%d.%d %s", diff_del_admin_epoch, diff_del_epoch, diff_del_updates, diff_add_admin_epoch, diff_add_epoch, diff_add_updates, digest); crm_xml_add(msg, PCMK__XA_CIB_ISREPLYTO, originator); pcmk__xe_set_bool_attr(msg, PCMK__XA_CIB_UPDATE, true); crm_xml_add(msg, PCMK__XA_CIB_OP, PCMK__CIB_REQUEST_APPLY_PATCH); crm_xml_add(msg, PCMK__XA_CIB_USER, CRM_DAEMON_USER); if (format == 1) { CRM_ASSERT(digest != NULL); } wrapper = pcmk__xe_create(msg, PCMK__XA_CIB_UPDATE_DIFF); pcmk__xml_copy(wrapper, result_diff); crm_log_xml_explicit(msg, "copy"); return send_cluster_message(NULL, crm_msg_cib, msg, TRUE); } else if (originator != NULL) { /* send reply via HA to originating node */ crm_trace("Sending request result to %s only", originator); crm_xml_add(msg, PCMK__XA_CIB_ISREPLYTO, originator); return send_cluster_message(pcmk__get_node(0, originator, NULL, pcmk__node_search_cluster), crm_msg_cib, msg, FALSE); } return FALSE; } /*! * \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 *target = 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 *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; } // @TODO: Improve trace messages. Target is accurate only for legacy mode. if (host) { target = host; } else if (call_options & cib_scope_local) { target = "local host"; } else { target = "primary"; } 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, target, 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, target); } 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 = is_update && cib_legacy_mode(); 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 && !cib_legacy_mode()) { 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_legacy_mode() && rc == pcmk_ok && result_diff != NULL && !(call_options & cib_inhibit_bcast)) { gboolean broadcast = FALSE; cib_local_bcast_num++; crm_xml_add_int(request, PCMK__XA_CIB_LOCAL_NOTIFY_ID, cib_local_bcast_num); broadcast = send_peer_reply(request, result_diff, originator, TRUE); if (broadcast && client_id && local_notify && op_reply) { /* If we have been asked to sync the reply, * and a bcast msg has gone out, we queue the local notify * until we know the bcast message has been received */ local_notify = FALSE; crm_trace("Queuing local %ssync notification for %s", (call_options & cib_sync_call) ? "" : "a-", client_id); queue_local_notify(op_reply, client_id, pcmk_is_set(call_options, cib_sync_call), (cib_client == NULL)); op_reply = NULL; /* the reply is queued, so don't free here */ } } 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 (call_options & cib_inhibit_bcast) { crm_trace("Request not broadcast: inhibited"); } 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, result_diff, originator, FALSE); } 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)); } } free_xml(op_reply); free_xml(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 = NULL; xmlNode *input = NULL; *section = NULL; switch (type) { case cib__op_apply_patch: { const char *wrapper_name = PCMK__XE_CIB_CALLDATA; if (pcmk__xe_attr_is_true(request, PCMK__XA_CIB_UPDATE)) { wrapper_name = PCMK__XA_CIB_UPDATE_DIFF; } wrapper = pcmk__xe_first_child(request, wrapper_name, NULL, NULL); input = pcmk__xe_first_child(wrapper, NULL, NULL, NULL); } break; default: wrapper = pcmk__xe_first_child(request, PCMK__XE_CIB_CALLDATA, NULL, NULL); input = pcmk__xe_first_child(wrapper, NULL, NULL, NULL); *section = crm_element_value(request, PCMK__XA_CIB_SECTION); break; } // Grab the specified section if ((*section != NULL) && pcmk__xe_is(input, PCMK_XE_CIB)) { input = pcmk_find_cib_element(input, *section); } return input; } // v1 and v2 patch formats #define XPATH_CONFIG_CHANGE \ "//" PCMK_XE_CRM_CONFIG " | " \ "//" 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, free_xml(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; if (pcmk_is_set(call_options, cib_inhibit_bcast)) { crm_trace("Skipping update: inhibit broadcast"); manage_counters = false; } // 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); // @COMPAT: Legacy code if (!manage_counters) { int format = 1; // If the diff is NULL at this point, it's because nothing changed if (*cib_diff != NULL) { crm_element_value_int(*cib_diff, PCMK_XA_FORMAT, &format); } if (format == 1) { config_changed = cib__config_changed_v1(NULL, NULL, cib_diff); } } /* 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"); free_xml(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) { free_xml(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) || cib_legacy_mode()) { *reply = create_cib_reply(op, call_id, client_id, call_options, rc, output); } if (output != the_cib) { free_xml(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 (cib_legacy_mode() && pcmk__str_eq(originator, OUR_NODENAME, pcmk__str_casei|pcmk__str_null_matches)) { /* message is from ourselves */ int bcast_id = 0; if (crm_element_value_int(msg, PCMK__XA_CIB_LOCAL_NOTIFY_ID, &bcast_id) == 0) { check_local_notify(bcast_id); } return; } else 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 = crm_active_peers(); + 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); send_cluster_message(NULL, crm_msg_cib, leaving, TRUE); free_xml(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/pacemaker-based.c b/daemons/based/pacemaker-based.c index a9e7e2af6c..b42dbc08fc 100644 --- a/daemons/based/pacemaker-based.c +++ b/daemons/based/pacemaker-based.c @@ -1,450 +1,450 @@ /* * 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 #define SUMMARY "daemon for managing the configuration of a Pacemaker cluster" extern int init_remote_listener(int port, gboolean encrypted); gboolean cib_shutdown_flag = FALSE; int cib_status = pcmk_ok; crm_cluster_t *crm_cluster = NULL; GMainLoop *mainloop = NULL; gchar *cib_root = NULL; static gboolean preserve_status = FALSE; gboolean cib_writes_enabled = TRUE; gboolean stand_alone = FALSE; int remote_fd = 0; int remote_tls_fd = 0; GHashTable *config_hash = NULL; GHashTable *local_notify_queue = NULL; static void cib_init(void); void cib_shutdown(int nsig); static bool startCib(const char *filename); extern int write_cib_contents(gpointer p); static crm_exit_t exit_code = CRM_EX_OK; static void cib_enable_writes(int nsig) { crm_info("(Re)enabling disk writes"); cib_writes_enabled = TRUE; } /*! * \internal * \brief Set up options, users, and groups for stand-alone mode * * \param[out] error GLib error object * * \return Standard Pacemaker return code */ static int setup_stand_alone(GError **error) { int rc = 0; struct passwd *pwentry = NULL; preserve_status = TRUE; cib_writes_enabled = FALSE; errno = 0; pwentry = getpwnam(CRM_DAEMON_USER); if (pwentry == NULL) { exit_code = CRM_EX_FATAL; if (errno != 0) { g_set_error(error, PCMK__EXITC_ERROR, exit_code, "Error getting password DB entry for %s: %s", CRM_DAEMON_USER, strerror(errno)); return errno; } g_set_error(error, PCMK__EXITC_ERROR, exit_code, "Password DB entry for '%s' not found", CRM_DAEMON_USER); return ENXIO; } rc = setgid(pwentry->pw_gid); if (rc < 0) { exit_code = CRM_EX_FATAL; g_set_error(error, PCMK__EXITC_ERROR, exit_code, "Could not set group to %d: %s", pwentry->pw_gid, strerror(errno)); return errno; } rc = initgroups(CRM_DAEMON_USER, pwentry->pw_gid); if (rc < 0) { exit_code = CRM_EX_FATAL; g_set_error(error, PCMK__EXITC_ERROR, exit_code, "Could not setup groups for user %d: %s", pwentry->pw_uid, strerror(errno)); return errno; } rc = setuid(pwentry->pw_uid); if (rc < 0) { exit_code = CRM_EX_FATAL; g_set_error(error, PCMK__EXITC_ERROR, exit_code, "Could not set user to %d: %s", pwentry->pw_uid, strerror(errno)); return errno; } return pcmk_rc_ok; } /* @COMPAT Deprecated since 2.1.8. Use pcmk_list_cluster_options() or * crm_attribute --list-options=cluster instead of querying daemon metadata. */ static int based_metadata(pcmk__output_t *out) { return pcmk__daemon_metadata(out, "pacemaker-based", "Cluster Information Base manager options", "Cluster options used by Pacemaker's Cluster " "Information Base manager", pcmk__opt_based); } static GOptionEntry entries[] = { { "stand-alone", 's', G_OPTION_FLAG_NONE, G_OPTION_ARG_NONE, &stand_alone, "(Advanced use only) Run in stand-alone mode", NULL }, { "disk-writes", 'w', G_OPTION_FLAG_NONE, G_OPTION_ARG_NONE, &cib_writes_enabled, "(Advanced use only) Enable disk writes (enabled by default unless in " "stand-alone mode)", NULL }, { "cib-root", 'r', G_OPTION_FLAG_NONE, G_OPTION_ARG_FILENAME, &cib_root, "(Advanced use only) Directory where the CIB XML file should be located " "(default: " CRM_CONFIG_DIR ")", NULL }, { NULL } }; static pcmk__supported_format_t formats[] = { PCMK__SUPPORTED_FORMAT_NONE, PCMK__SUPPORTED_FORMAT_TEXT, PCMK__SUPPORTED_FORMAT_XML, { NULL, NULL, NULL } }; static GOptionContext * build_arg_context(pcmk__common_args_t *args, GOptionGroup **group) { GOptionContext *context = NULL; context = pcmk__build_arg_context(args, "text (default), xml", group, NULL); pcmk__add_main_args(context, entries); return context; } int main(int argc, char **argv) { int rc = pcmk_rc_ok; crm_ipc_t *old_instance = NULL; pcmk__output_t *out = 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, "r"); 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; } mainloop_add_signal(SIGTERM, cib_shutdown); mainloop_add_signal(SIGPIPE, cib_enable_writes); cib_writer = mainloop_add_trigger(G_PRIORITY_LOW, write_cib_contents, NULL); if ((g_strv_length(processed_args) >= 2) && pcmk__str_eq(processed_args[1], "metadata", pcmk__str_none)) { rc = based_metadata(out); 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; } pcmk__cli_init_logging("pacemaker-based", args->verbosity); crm_log_init(NULL, LOG_INFO, TRUE, FALSE, argc, argv, FALSE); crm_notice("Starting Pacemaker CIB manager"); old_instance = crm_ipc_new(PCMK__SERVER_BASED_RO, 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 end-point already up */ crm_ipc_close(old_instance); crm_ipc_destroy(old_instance); crm_err("pacemaker-based 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; } if (stand_alone) { rc = setup_stand_alone(&error); if (rc != pcmk_rc_ok) { goto done; } } if (cib_root == NULL) { cib_root = g_strdup(CRM_CONFIG_DIR); } else { crm_notice("Using custom config location: %s", cib_root); } if (!pcmk__daemon_can_write(cib_root, NULL)) { exit_code = CRM_EX_FATAL; crm_err("Terminating due to bad permissions on %s", cib_root); g_set_error(&error, PCMK__EXITC_ERROR, exit_code, "Bad permissions on %s (see logs for details)", cib_root); goto done; } crm_peer_init(); // Read initial CIB, connect to cluster, and start IPC servers cib_init(); // Run the main loop mainloop = g_main_loop_new(NULL, FALSE); crm_notice("Pacemaker CIB manager successfully started and accepting connections"); g_main_loop_run(mainloop); /* If main loop returned, clean up and exit. We disconnect in case * terminate_cib() was called with fast=-1. */ pcmk_cluster_disconnect(crm_cluster); pcmk__stop_based_ipc(ipcs_ro, ipcs_rw, ipcs_shm); done: g_strfreev(processed_args); pcmk__free_arg_context(context); crm_peer_destroy(); if (local_notify_queue != NULL) { g_hash_table_destroy(local_notify_queue); } if (config_hash != NULL) { g_hash_table_destroy(config_hash); } pcmk__client_cleanup(); pcmk_cluster_free(crm_cluster); g_free(cib_root); 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); } #if SUPPORT_COROSYNC static void cib_cs_dispatch(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_message_common_cs(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); cib_peer_callback(xml, NULL); } free_xml(xml); free(data); } static void cib_cs_destroy(gpointer user_data) { if (cib_shutdown_flag) { crm_info("Corosync disconnection complete"); } else { crm_crit("Lost connection to cluster layer, shutting down"); terminate_cib(__func__, CRM_EX_DISCONNECT); } } #endif static void cib_peer_update_callback(enum crm_status_type type, crm_node_t * node, const void *data) { switch (type) { case crm_status_processes: if (cib_legacy_mode() && !pcmk_is_set(node->processes, crm_get_cluster_proc())) { uint32_t old = data? *(const uint32_t *)data : 0; if ((node->processes ^ old) & crm_proc_cpg) { crm_info("Attempting to disable legacy mode after %s left the cluster", node->uname); legacy_mode = FALSE; } } break; case crm_status_uname: case crm_status_nstate: - if (cib_shutdown_flag && (crm_active_peers() < 2) + if (cib_shutdown_flag && (pcmk__cluster_num_active_nodes() < 2) && (pcmk__ipc_client_count() == 0)) { crm_info("No more peers"); terminate_cib(__func__, -1); } break; } } static void cib_init(void) { crm_cluster = pcmk_cluster_new(); #if SUPPORT_COROSYNC if (is_corosync_cluster()) { crm_cluster->destroy = cib_cs_destroy; crm_cluster->cpg.cpg_deliver_fn = cib_cs_dispatch; crm_cluster->cpg.cpg_confchg_fn = pcmk_cpg_membership; } #endif // SUPPORT_COROSYNC config_hash = pcmk__strkey_table(free, free); if (startCib("cib.xml") == FALSE) { crm_crit("Cannot start CIB... terminating"); crm_exit(CRM_EX_NOINPUT); } if (!stand_alone) { crm_set_status_callback(&cib_peer_update_callback); if (pcmk_cluster_connect(crm_cluster) != pcmk_rc_ok) { crm_crit("Cannot sign in to the cluster... terminating"); crm_exit(CRM_EX_FATAL); } } pcmk__serve_based_ipc(&ipcs_ro, &ipcs_rw, &ipcs_shm, &ipc_ro_callbacks, &ipc_rw_callbacks); if (stand_alone) { based_is_primary = true; } } static bool startCib(const char *filename) { gboolean active = FALSE; xmlNode *cib = readCibXmlFile(cib_root, filename, !preserve_status); if (activateCibXml(cib, TRUE, "start") == 0) { int port = 0; active = TRUE; cib_read_config(config_hash, cib); pcmk__scan_port(crm_element_value(cib, PCMK_XA_REMOTE_TLS_PORT), &port); if (port >= 0) { remote_tls_fd = init_remote_listener(port, TRUE); } pcmk__scan_port(crm_element_value(cib, PCMK_XA_REMOTE_CLEAR_PORT), &port); if (port >= 0) { remote_fd = init_remote_listener(port, FALSE); } } return active; } diff --git a/daemons/controld/controld_cib.c b/daemons/controld/controld_cib.c index 47fb5c6e69..7acff3064f 100644 --- a/daemons/controld/controld_cib.c +++ b/daemons/controld/controld_cib.c @@ -1,1066 +1,1066 @@ /* * 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 /* sleep */ #include #include #include #include #include // Call ID of the most recent in-progress CIB resource update (or 0 if none) static int pending_rsc_update = 0; /*! * \internal * \brief Respond to a dropped CIB connection * * \param[in] user_data CIB connection that dropped */ static void handle_cib_disconnect(gpointer user_data) { CRM_LOG_ASSERT(user_data == controld_globals.cib_conn); controld_trigger_fsa(); controld_globals.cib_conn->state = cib_disconnected; if (pcmk_is_set(controld_globals.fsa_input_register, R_CIB_CONNECTED)) { // @TODO This should trigger a reconnect, not a shutdown crm_crit("Lost connection to the CIB manager, shutting down"); register_fsa_input(C_FSA_INTERNAL, I_ERROR, NULL); controld_clear_fsa_input_flags(R_CIB_CONNECTED); } else { // Expected crm_info("Disconnected from the CIB manager"); } } static void do_cib_updated(const char *event, xmlNode * msg) { const xmlNode *patchset = NULL; const char *client_name = NULL; crm_debug("Received CIB diff notification: DC=%s", pcmk__btoa(AM_I_DC)); if (cib__get_notify_patchset(msg, &patchset) != pcmk_rc_ok) { return; } if (cib__element_in_patchset(patchset, PCMK_XE_ALERTS) || cib__element_in_patchset(patchset, PCMK_XE_CRM_CONFIG)) { controld_trigger_config(); } if (!AM_I_DC) { // We're not in control of the join sequence return; } client_name = crm_element_value(msg, PCMK__XA_CIB_CLIENTNAME); if (!cib__client_triggers_refresh(client_name)) { // The CIB is still accurate return; } if (cib__element_in_patchset(patchset, PCMK_XE_NODES) || cib__element_in_patchset(patchset, PCMK_XE_STATUS)) { /* An unsafe client modified the PCMK_XE_NODES or PCMK_XE_STATUS * section. Ensure the node list is up-to-date, and start the join * process again so we get everyone's current resource history. */ if (client_name == NULL) { client_name = crm_element_value(msg, PCMK__XA_CIB_CLIENTID); } crm_notice("Populating nodes and starting an election after %s event " "triggered by %s", event, pcmk__s(client_name, "(unidentified client)")); populate_cib_nodes(node_update_quick|node_update_all, __func__); register_fsa_input(C_FSA_INTERNAL, I_ELECTION, NULL); } } void controld_disconnect_cib_manager(void) { cib_t *cib_conn = controld_globals.cib_conn; CRM_ASSERT(cib_conn != NULL); crm_debug("Disconnecting from the CIB manager"); controld_clear_fsa_input_flags(R_CIB_CONNECTED); cib_conn->cmds->del_notify_callback(cib_conn, PCMK__VALUE_CIB_DIFF_NOTIFY, do_cib_updated); cib_free_callbacks(cib_conn); if (cib_conn->state != cib_disconnected) { cib_conn->cmds->set_secondary(cib_conn, cib_scope_local|cib_discard_reply); cib_conn->cmds->signoff(cib_conn); } } /* A_CIB_STOP, A_CIB_START, O_CIB_RESTART */ void do_cib_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) { static int cib_retries = 0; cib_t *cib_conn = controld_globals.cib_conn; void (*dnotify_fn) (gpointer user_data) = handle_cib_disconnect; void (*update_cb) (const char *event, xmlNodePtr msg) = do_cib_updated; int rc = pcmk_ok; CRM_ASSERT(cib_conn != NULL); if (pcmk_is_set(action, A_CIB_STOP)) { if ((cib_conn->state != cib_disconnected) && (pending_rsc_update != 0)) { crm_info("Waiting for resource update %d to complete", pending_rsc_update); crmd_fsa_stall(FALSE); return; } controld_disconnect_cib_manager(); } if (!pcmk_is_set(action, A_CIB_START)) { return; } if (cur_state == S_STOPPING) { crm_err("Ignoring request to connect to the CIB manager after " "shutdown"); return; } rc = cib_conn->cmds->signon(cib_conn, CRM_SYSTEM_CRMD, cib_command_nonblocking); if (rc != pcmk_ok) { // A short wait that usually avoids stalling the FSA sleep(1); rc = cib_conn->cmds->signon(cib_conn, CRM_SYSTEM_CRMD, cib_command_nonblocking); } if (rc != pcmk_ok) { crm_info("Could not connect to the CIB manager: %s", pcmk_strerror(rc)); } else if (cib_conn->cmds->set_connection_dnotify(cib_conn, dnotify_fn) != pcmk_ok) { crm_err("Could not set dnotify callback"); } else if (cib_conn->cmds->add_notify_callback(cib_conn, PCMK__VALUE_CIB_DIFF_NOTIFY, update_cb) != pcmk_ok) { crm_err("Could not set CIB notification callback (update)"); } else { controld_set_fsa_input_flags(R_CIB_CONNECTED); cib_retries = 0; } if (!pcmk_is_set(controld_globals.fsa_input_register, R_CIB_CONNECTED)) { cib_retries++; if (cib_retries < 30) { crm_warn("Couldn't complete CIB registration %d times... " "pause and retry", cib_retries); controld_start_wait_timer(); crmd_fsa_stall(FALSE); } else { crm_err("Could not complete CIB registration %d times... " "hard error", cib_retries); register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); } } } #define MIN_CIB_OP_TIMEOUT (30) /*! * \internal * \brief Get the timeout (in seconds) that should be used with CIB operations * * \return The maximum of 30 seconds, the value of the PCMK_cib_timeout * environment variable, or 10 seconds times one more than the number of * nodes in the cluster. */ unsigned int cib_op_timeout(void) { // @COMPAT: Drop env_timeout at 3.0.0 static int env_timeout = -1; unsigned int calculated_timeout = 0; if (env_timeout == -1) { const char *env = pcmk__env_option(PCMK__ENV_CIB_TIMEOUT); pcmk__scan_min_int(env, &env_timeout, MIN_CIB_OP_TIMEOUT); crm_trace("Minimum CIB op timeout: %ds (environment: %s)", env_timeout, (env? env : "none")); } calculated_timeout = 10U * (1U - + crm_active_peers() + + pcmk__cluster_num_active_nodes() + pcmk__cluster_num_remote_nodes()); calculated_timeout = QB_MAX(calculated_timeout, env_timeout); crm_trace("Calculated timeout: %us", calculated_timeout); if (controld_globals.cib_conn) { controld_globals.cib_conn->call_timeout = calculated_timeout; } return calculated_timeout; } /*! * \internal * \brief Get CIB call options to use local scope if primary is unavailable * * \return CIB call options */ int crmd_cib_smart_opt(void) { int call_opt = cib_none; if ((controld_globals.fsa_state == S_ELECTION) || (controld_globals.fsa_state == S_PENDING)) { crm_info("Sending update to local CIB in state: %s", fsa_state2string(controld_globals.fsa_state)); cib__set_call_options(call_opt, "update", cib_scope_local); } return call_opt; } static void cib_delete_callback(xmlNode *msg, int call_id, int rc, xmlNode *output, void *user_data) { char *desc = user_data; if (rc == 0) { crm_debug("Deletion of %s (via CIB call %d) succeeded", desc, call_id); } else { crm_warn("Deletion of %s (via CIB call %d) failed: %s " CRM_XS " rc=%d", desc, call_id, pcmk_strerror(rc), rc); } } // Searches for various portions of PCMK__XE_NODE_STATE to delete // Match a particular node's PCMK__XE_NODE_STATE (takes node name 1x) #define XPATH_NODE_STATE "//" PCMK__XE_NODE_STATE "[@" PCMK_XA_UNAME "='%s']" // Node's lrm section (name 1x) #define XPATH_NODE_LRM XPATH_NODE_STATE "/" PCMK__XE_LRM /* Node's PCMK__XE_LRM_RSC_OP entries and PCMK__XE_LRM_RESOURCE entries without * unexpired lock * (name 2x, (seconds_since_epoch - PCMK_OPT_SHUTDOWN_LOCK_LIMIT) 1x) */ #define XPATH_NODE_LRM_UNLOCKED XPATH_NODE_STATE "//" PCMK__XE_LRM_RSC_OP \ "|" XPATH_NODE_STATE \ "//" PCMK__XE_LRM_RESOURCE \ "[not(@" PCMK_OPT_SHUTDOWN_LOCK ") " \ "or " PCMK_OPT_SHUTDOWN_LOCK "<%lld]" // Node's PCMK__XE_TRANSIENT_ATTRIBUTES section (name 1x) #define XPATH_NODE_ATTRS XPATH_NODE_STATE "/" PCMK__XE_TRANSIENT_ATTRIBUTES // Everything under PCMK__XE_NODE_STATE (name 1x) #define XPATH_NODE_ALL XPATH_NODE_STATE "/*" /* Unlocked history + transient attributes * (name 2x, (seconds_since_epoch - PCMK_OPT_SHUTDOWN_LOCK_LIMIT) 1x, name 1x) */ #define XPATH_NODE_ALL_UNLOCKED XPATH_NODE_LRM_UNLOCKED "|" XPATH_NODE_ATTRS /*! * \internal * \brief Get the XPath and description of a node state section to be deleted * * \param[in] uname Desired node * \param[in] section Subsection of \c PCMK__XE_NODE_STATE to be deleted * \param[out] xpath Where to store XPath of \p section * \param[out] desc If not \c NULL, where to store description of \p section */ void controld_node_state_deletion_strings(const char *uname, enum controld_section_e section, char **xpath, char **desc) { const char *desc_pre = NULL; // Shutdown locks that started before this time are expired long long expire = (long long) time(NULL) - controld_globals.shutdown_lock_limit; switch (section) { case controld_section_lrm: *xpath = crm_strdup_printf(XPATH_NODE_LRM, uname); desc_pre = "resource history"; break; case controld_section_lrm_unlocked: *xpath = crm_strdup_printf(XPATH_NODE_LRM_UNLOCKED, uname, uname, expire); desc_pre = "resource history (other than shutdown locks)"; break; case controld_section_attrs: *xpath = crm_strdup_printf(XPATH_NODE_ATTRS, uname); desc_pre = "transient attributes"; break; case controld_section_all: *xpath = crm_strdup_printf(XPATH_NODE_ALL, uname); desc_pre = "all state"; break; case controld_section_all_unlocked: *xpath = crm_strdup_printf(XPATH_NODE_ALL_UNLOCKED, uname, uname, expire, uname); desc_pre = "all state (other than shutdown locks)"; break; default: // We called this function incorrectly CRM_ASSERT(false); break; } if (desc != NULL) { *desc = crm_strdup_printf("%s for node %s", desc_pre, uname); } } /*! * \internal * \brief Delete subsection of a node's CIB \c PCMK__XE_NODE_STATE * * \param[in] uname Desired node * \param[in] section Subsection of \c PCMK__XE_NODE_STATE to delete * \param[in] options CIB call options to use */ void controld_delete_node_state(const char *uname, enum controld_section_e section, int options) { cib_t *cib = controld_globals.cib_conn; char *xpath = NULL; char *desc = NULL; int cib_rc = pcmk_ok; CRM_ASSERT((uname != NULL) && (cib != NULL)); controld_node_state_deletion_strings(uname, section, &xpath, &desc); cib__set_call_options(options, "node state deletion", cib_xpath|cib_multiple); cib_rc = cib->cmds->remove(cib, xpath, NULL, options); fsa_register_cib_callback(cib_rc, desc, cib_delete_callback); crm_info("Deleting %s (via CIB call %d) " CRM_XS " xpath=%s", desc, cib_rc, xpath); // CIB library handles freeing desc free(xpath); } // Takes node name and resource ID #define XPATH_RESOURCE_HISTORY "//" PCMK__XE_NODE_STATE \ "[@" PCMK_XA_UNAME "='%s']/" \ PCMK__XE_LRM "/" PCMK__XE_LRM_RESOURCES \ "/" PCMK__XE_LRM_RESOURCE \ "[@" PCMK_XA_ID "='%s']" // @TODO could add "and @PCMK_OPT_SHUTDOWN_LOCK" to limit to locks /*! * \internal * \brief Clear resource history from CIB for a given resource and node * * \param[in] rsc_id ID of resource to be cleared * \param[in] node Node whose resource history should be cleared * \param[in] user_name ACL user name to use * \param[in] call_options CIB call options * * \return Standard Pacemaker return code */ int controld_delete_resource_history(const char *rsc_id, const char *node, const char *user_name, int call_options) { char *desc = NULL; char *xpath = NULL; int rc = pcmk_rc_ok; cib_t *cib = controld_globals.cib_conn; CRM_CHECK((rsc_id != NULL) && (node != NULL), return EINVAL); desc = crm_strdup_printf("resource history for %s on %s", rsc_id, node); if (cib == NULL) { crm_err("Unable to clear %s: no CIB connection", desc); free(desc); return ENOTCONN; } // Ask CIB to delete the entry xpath = crm_strdup_printf(XPATH_RESOURCE_HISTORY, node, rsc_id); cib->cmds->set_user(cib, user_name); rc = cib->cmds->remove(cib, xpath, NULL, call_options|cib_xpath); cib->cmds->set_user(cib, NULL); if (rc < 0) { rc = pcmk_legacy2rc(rc); crm_err("Could not delete resource status of %s on %s%s%s: %s " CRM_XS " rc=%d", rsc_id, node, (user_name? " for user " : ""), (user_name? user_name : ""), pcmk_rc_str(rc), rc); free(desc); free(xpath); return rc; } if (pcmk_is_set(call_options, cib_sync_call)) { if (pcmk_is_set(call_options, cib_dryrun)) { crm_debug("Deletion of %s would succeed", desc); } else { crm_debug("Deletion of %s succeeded", desc); } free(desc); } else { crm_info("Clearing %s (via CIB call %d) " CRM_XS " xpath=%s", desc, rc, xpath); fsa_register_cib_callback(rc, desc, cib_delete_callback); // CIB library handles freeing desc } free(xpath); return pcmk_rc_ok; } /*! * \internal * \brief Build XML and string of parameters meeting some criteria, for digest * * \param[in] op Executor event with parameter table to use * \param[in] metadata Parsed meta-data for executed resource agent * \param[in] param_type Flag used for selection criteria * \param[out] result Will be set to newly created XML with selected * parameters as attributes * * \return Newly allocated space-separated string of parameter names * \note Selection criteria varies by param_type: for the restart digest, we * want parameters that are *not* marked reloadable (OCF 1.1) or that * *are* marked unique (pre-1.1), for both string and XML results; for the * secure digest, we want parameters that *are* marked private for the * string, but parameters that are *not* marked private for the XML. * \note It is the caller's responsibility to free the string return value with * \p g_string_free() and the XML result with \p free_xml(). */ static GString * build_parameter_list(const lrmd_event_data_t *op, const struct ra_metadata_s *metadata, enum ra_param_flags_e param_type, xmlNode **result) { GString *list = NULL; *result = pcmk__xe_create(NULL, PCMK_XE_PARAMETERS); /* Consider all parameters only except private ones to be consistent with * what scheduler does with calculate_secure_digest(). */ if (param_type == ra_param_private && compare_version(controld_globals.dc_version, "3.16.0") >= 0) { g_hash_table_foreach(op->params, hash2field, *result); pcmk__filter_op_for_digest(*result); } for (GList *iter = metadata->ra_params; iter != NULL; iter = iter->next) { struct ra_param_s *param = (struct ra_param_s *) iter->data; bool accept_for_list = false; bool accept_for_xml = false; switch (param_type) { case ra_param_reloadable: accept_for_list = !pcmk_is_set(param->rap_flags, param_type); accept_for_xml = accept_for_list; break; case ra_param_unique: accept_for_list = pcmk_is_set(param->rap_flags, param_type); accept_for_xml = accept_for_list; break; case ra_param_private: accept_for_list = pcmk_is_set(param->rap_flags, param_type); accept_for_xml = !accept_for_list; break; } if (accept_for_list) { crm_trace("Attr %s is %s", param->rap_name, ra_param_flag2text(param_type)); if (list == NULL) { // We will later search for " WORD ", so start list with a space pcmk__add_word(&list, 256, " "); } pcmk__add_word(&list, 0, param->rap_name); } else { crm_trace("Rejecting %s for %s", param->rap_name, ra_param_flag2text(param_type)); } if (accept_for_xml) { const char *v = g_hash_table_lookup(op->params, param->rap_name); if (v != NULL) { crm_trace("Adding attr %s=%s to the xml result", param->rap_name, v); crm_xml_add(*result, param->rap_name, v); } } else { crm_trace("Removing attr %s from the xml result", param->rap_name); pcmk__xe_remove_attr(*result, param->rap_name); } } if (list != NULL) { // We will later search for " WORD ", so end list with a space pcmk__add_word(&list, 0, " "); } return list; } static void append_restart_list(lrmd_event_data_t *op, struct ra_metadata_s *metadata, xmlNode *update, const char *version) { GString *list = NULL; char *digest = NULL; xmlNode *restart = NULL; CRM_LOG_ASSERT(op->params != NULL); if (op->interval_ms > 0) { /* monitors are not reloadable */ return; } if (pcmk_is_set(metadata->ra_flags, ra_supports_reload_agent)) { /* Add parameters not marked reloadable to the PCMK__XA_OP_FORCE_RESTART * list */ list = build_parameter_list(op, metadata, ra_param_reloadable, &restart); } else if (pcmk_is_set(metadata->ra_flags, ra_supports_legacy_reload)) { /* @COMPAT pre-OCF-1.1 resource agents * * Before OCF 1.1, Pacemaker abused "unique=0" to indicate * reloadability. Add any parameters with unique="1" to the * PCMK__XA_OP_FORCE_RESTART list. */ list = build_parameter_list(op, metadata, ra_param_unique, &restart); } else { // Resource does not support agent reloads return; } digest = calculate_operation_digest(restart, version); /* Add PCMK__XA_OP_FORCE_RESTART and PCMK__XA_OP_RESTART_DIGEST to indicate * the resource supports reload, no matter if it actually supports any * reloadable parameters */ crm_xml_add(update, PCMK__XA_OP_FORCE_RESTART, (list == NULL)? "" : (const char *) list->str); crm_xml_add(update, PCMK__XA_OP_RESTART_DIGEST, digest); if ((list != NULL) && (list->len > 0)) { crm_trace("%s: %s, %s", op->rsc_id, digest, (const char *) list->str); } else { crm_trace("%s: %s", op->rsc_id, digest); } if (list != NULL) { g_string_free(list, TRUE); } free_xml(restart); free(digest); } static void append_secure_list(lrmd_event_data_t *op, struct ra_metadata_s *metadata, xmlNode *update, const char *version) { GString *list = NULL; char *digest = NULL; xmlNode *secure = NULL; CRM_LOG_ASSERT(op->params != NULL); /* To keep PCMK__XA_OP_SECURE_PARAMS short, we want it to contain the secure * parameters but PCMK__XA_OP_SECURE_DIGEST to be based on the insecure ones */ list = build_parameter_list(op, metadata, ra_param_private, &secure); if (list != NULL) { digest = calculate_operation_digest(secure, version); crm_xml_add(update, PCMK__XA_OP_SECURE_PARAMS, (const char *) list->str); crm_xml_add(update, PCMK__XA_OP_SECURE_DIGEST, digest); crm_trace("%s: %s, %s", op->rsc_id, digest, (const char *) list->str); g_string_free(list, TRUE); } else { crm_trace("%s: no secure parameters", op->rsc_id); } free_xml(secure); free(digest); } /*! * \internal * \brief Create XML for a resource history entry * * \param[in] func Function name of caller * \param[in,out] parent XML to add entry to * \param[in] rsc Affected resource * \param[in,out] op Action to add an entry for (or NULL to do nothing) * \param[in] node_name Node where action occurred */ void controld_add_resource_history_xml_as(const char *func, xmlNode *parent, const lrmd_rsc_info_t *rsc, lrmd_event_data_t *op, const char *node_name) { int target_rc = 0; xmlNode *xml_op = NULL; struct ra_metadata_s *metadata = NULL; const char *caller_version = NULL; lrm_state_t *lrm_state = NULL; if (op == NULL) { return; } target_rc = rsc_op_expected_rc(op); caller_version = g_hash_table_lookup(op->params, PCMK_XA_CRM_FEATURE_SET); CRM_CHECK(caller_version != NULL, caller_version = CRM_FEATURE_SET); xml_op = pcmk__create_history_xml(parent, op, caller_version, target_rc, controld_globals.our_nodename, func); if (xml_op == NULL) { return; } if ((rsc == NULL) || (op->params == NULL) || !crm_op_needs_metadata(rsc->standard, op->op_type)) { crm_trace("No digests needed for %s action on %s (params=%p rsc=%p)", op->op_type, op->rsc_id, op->params, rsc); return; } lrm_state = lrm_state_find(node_name); if (lrm_state == NULL) { crm_warn("Cannot calculate digests for operation " PCMK__OP_FMT " because we have no connection to executor for %s", op->rsc_id, op->op_type, op->interval_ms, node_name); return; } /* Ideally the metadata is cached, and the agent is just a fallback. * * @TODO Go through all callers and ensure they get metadata asynchronously * first. */ metadata = controld_get_rsc_metadata(lrm_state, rsc, controld_metadata_from_agent |controld_metadata_from_cache); if (metadata == NULL) { return; } crm_trace("Including additional digests for %s:%s:%s", rsc->standard, rsc->provider, rsc->type); append_restart_list(op, metadata, xml_op, caller_version); append_secure_list(op, metadata, xml_op, caller_version); return; } /*! * \internal * \brief Record an action as pending in the CIB, if appropriate * * \param[in] node_name Node where the action is pending * \param[in] rsc Resource that action is for * \param[in,out] op Pending action * * \return true if action was recorded in CIB, otherwise false */ bool controld_record_pending_op(const char *node_name, const lrmd_rsc_info_t *rsc, lrmd_event_data_t *op) { const char *record_pending = NULL; CRM_CHECK((node_name != NULL) && (rsc != NULL) && (op != NULL), return false); // Never record certain operation types as pending if ((op->op_type == NULL) || (op->params == NULL) || !controld_action_is_recordable(op->op_type)) { return false; } // Check action's PCMK_META_RECORD_PENDING meta-attribute (defaults to true) record_pending = crm_meta_value(op->params, PCMK_META_RECORD_PENDING); if ((record_pending != NULL) && !crm_is_true(record_pending)) { return false; } op->call_id = -1; op->t_run = time(NULL); op->t_rcchange = op->t_run; lrmd__set_result(op, PCMK_OCF_UNKNOWN, PCMK_EXEC_PENDING, NULL); crm_debug("Recording pending %s-interval %s for %s on %s in the CIB", pcmk__readable_interval(op->interval_ms), op->op_type, op->rsc_id, node_name); controld_update_resource_history(node_name, rsc, op, 0); return true; } static void cib_rsc_callback(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data) { switch (rc) { case pcmk_ok: case -pcmk_err_diff_failed: case -pcmk_err_diff_resync: crm_trace("Resource history update completed (call=%d rc=%d)", call_id, rc); break; default: if (call_id > 0) { crm_warn("Resource history update %d failed: %s " CRM_XS " rc=%d", call_id, pcmk_strerror(rc), rc); } else { crm_warn("Resource history update failed: %s " CRM_XS " rc=%d", pcmk_strerror(rc), rc); } } if (call_id == pending_rsc_update) { pending_rsc_update = 0; controld_trigger_fsa(); } } /* Only successful stops, and probes that found the resource inactive, get locks * recorded in the history. This ensures the resource stays locked to the node * until it is active there again after the node comes back up. */ static bool should_preserve_lock(lrmd_event_data_t *op) { if (!pcmk_is_set(controld_globals.flags, controld_shutdown_lock_enabled)) { return false; } if (!strcmp(op->op_type, PCMK_ACTION_STOP) && (op->rc == PCMK_OCF_OK)) { return true; } if (!strcmp(op->op_type, PCMK_ACTION_MONITOR) && (op->rc == PCMK_OCF_NOT_RUNNING)) { return true; } return false; } /*! * \internal * \brief Request a CIB update * * \param[in] section Section of CIB to update * \param[in] data New XML of CIB section to update * \param[in] options CIB call options * \param[in] callback If not \c NULL, set this as the operation callback * * \return Standard Pacemaker return code * * \note If \p callback is \p cib_rsc_callback(), the CIB update's call ID is * stored in \p pending_rsc_update on success. */ int controld_update_cib(const char *section, xmlNode *data, int options, void (*callback)(xmlNode *, int, int, xmlNode *, void *)) { cib_t *cib = controld_globals.cib_conn; int cib_rc = -ENOTCONN; CRM_ASSERT(data != NULL); if (cib != NULL) { cib_rc = cib->cmds->modify(cib, section, data, options); if (cib_rc >= 0) { crm_debug("Submitted CIB update %d for %s section", cib_rc, section); } } if (callback == NULL) { if (cib_rc < 0) { crm_err("Failed to update CIB %s section: %s", section, pcmk_rc_str(pcmk_legacy2rc(cib_rc))); } } else { if ((cib_rc >= 0) && (callback == cib_rsc_callback)) { /* Checking for a particular callback is a little hacky, but it * didn't seem worth adding an output argument for cib_rc for just * one use case. */ pending_rsc_update = cib_rc; } fsa_register_cib_callback(cib_rc, NULL, callback); } return (cib_rc >= 0)? pcmk_rc_ok : pcmk_legacy2rc(cib_rc); } /*! * \internal * \brief Update resource history entry in CIB * * \param[in] node_name Node where action occurred * \param[in] rsc Resource that action is for * \param[in,out] op Action to record * \param[in] lock_time If nonzero, when resource was locked to node * * \note On success, the CIB update's call ID will be stored in * pending_rsc_update. */ void controld_update_resource_history(const char *node_name, const lrmd_rsc_info_t *rsc, lrmd_event_data_t *op, time_t lock_time) { xmlNode *update = NULL; xmlNode *xml = NULL; int call_opt = crmd_cib_smart_opt(); const char *node_id = NULL; const char *container = NULL; CRM_CHECK((node_name != NULL) && (op != NULL), return); if (rsc == NULL) { crm_warn("Resource %s no longer exists in the executor", op->rsc_id); controld_ack_event_directly(NULL, NULL, rsc, op, op->rsc_id); return; } // update = pcmk__xe_create(NULL, PCMK_XE_STATUS); // xml = pcmk__xe_create(update, PCMK__XE_NODE_STATE); if (pcmk__str_eq(node_name, controld_globals.our_nodename, pcmk__str_casei)) { node_id = controld_globals.our_uuid; } else { node_id = node_name; pcmk__xe_set_bool_attr(xml, PCMK_XA_REMOTE_NODE, true); } crm_xml_add(xml, PCMK_XA_ID, node_id); crm_xml_add(xml, PCMK_XA_UNAME, node_name); crm_xml_add(xml, PCMK_XA_CRM_DEBUG_ORIGIN, __func__); // xml = pcmk__xe_create(xml, PCMK__XE_LRM); crm_xml_add(xml, PCMK_XA_ID, node_id); // xml = pcmk__xe_create(xml, PCMK__XE_LRM_RESOURCES); // xml = pcmk__xe_create(xml, PCMK__XE_LRM_RESOURCE); crm_xml_add(xml, PCMK_XA_ID, op->rsc_id); crm_xml_add(xml, PCMK_XA_CLASS, rsc->standard); crm_xml_add(xml, PCMK_XA_PROVIDER, rsc->provider); crm_xml_add(xml, PCMK_XA_TYPE, rsc->type); if (lock_time != 0) { /* Actions on a locked resource should either preserve the lock by * recording it with the action result, or clear it. */ if (!should_preserve_lock(op)) { lock_time = 0; } crm_xml_add_ll(xml, PCMK_OPT_SHUTDOWN_LOCK, (long long) lock_time); } if (op->params != NULL) { container = g_hash_table_lookup(op->params, CRM_META "_" PCMK__META_CONTAINER); if (container != NULL) { crm_trace("Resource %s is a part of container resource %s", op->rsc_id, container); crm_xml_add(xml, PCMK__META_CONTAINER, container); } } // (possibly more than one) controld_add_resource_history_xml(xml, rsc, op, node_name); /* Update CIB asynchronously. Even if it fails, the resource state should be * discovered during the next election. Worst case, the node is wrongly * fenced for running a resource it isn't. */ crm_log_xml_trace(update, __func__); controld_update_cib(PCMK_XE_STATUS, update, call_opt, cib_rsc_callback); free_xml(update); } /*! * \internal * \brief Erase an LRM history entry from the CIB, given the operation data * * \param[in] op Operation whose history should be deleted */ void controld_delete_action_history(const lrmd_event_data_t *op) { xmlNode *xml_top = NULL; CRM_CHECK(op != NULL, return); xml_top = pcmk__xe_create(NULL, PCMK__XE_LRM_RSC_OP); crm_xml_add_int(xml_top, PCMK__XA_CALL_ID, op->call_id); crm_xml_add(xml_top, PCMK__XA_TRANSITION_KEY, op->user_data); if (op->interval_ms > 0) { char *op_id = pcmk__op_key(op->rsc_id, op->op_type, op->interval_ms); /* Avoid deleting last_failure too (if it was a result of this recurring op failing) */ crm_xml_add(xml_top, PCMK_XA_ID, op_id); free(op_id); } crm_debug("Erasing resource operation history for " PCMK__OP_FMT " (call=%d)", op->rsc_id, op->op_type, op->interval_ms, op->call_id); controld_globals.cib_conn->cmds->remove(controld_globals.cib_conn, PCMK_XE_STATUS, xml_top, cib_none); crm_log_xml_trace(xml_top, "op:cancel"); free_xml(xml_top); } /* Define xpath to find LRM resource history entry by node and resource */ #define XPATH_HISTORY \ "/" PCMK_XE_CIB "/" PCMK_XE_STATUS \ "/" PCMK__XE_NODE_STATE "[@" PCMK_XA_UNAME "='%s']" \ "/" PCMK__XE_LRM "/" PCMK__XE_LRM_RESOURCES \ "/" PCMK__XE_LRM_RESOURCE "[@" PCMK_XA_ID "='%s']" \ "/" PCMK__XE_LRM_RSC_OP /* ... and also by operation key */ #define XPATH_HISTORY_ID XPATH_HISTORY "[@" PCMK_XA_ID "='%s']" /* ... and also by operation key and operation call ID */ #define XPATH_HISTORY_CALL XPATH_HISTORY \ "[@" PCMK_XA_ID "='%s' and @" PCMK__XA_CALL_ID "='%d']" /* ... and also by operation key and original operation key */ #define XPATH_HISTORY_ORIG XPATH_HISTORY \ "[@" PCMK_XA_ID "='%s' and @" PCMK__XA_OPERATION_KEY "='%s']" /*! * \internal * \brief Delete a last_failure resource history entry from the CIB * * \param[in] rsc_id Name of resource to clear history for * \param[in] node Name of node to clear history for * \param[in] action If specified, delete only if this was failed action * \param[in] interval_ms If \p action is specified, it has this interval */ void controld_cib_delete_last_failure(const char *rsc_id, const char *node, const char *action, guint interval_ms) { char *xpath = NULL; char *last_failure_key = NULL; CRM_CHECK((rsc_id != NULL) && (node != NULL), return); // Generate XPath to match desired entry last_failure_key = pcmk__op_key(rsc_id, "last_failure", 0); if (action == NULL) { xpath = crm_strdup_printf(XPATH_HISTORY_ID, node, rsc_id, last_failure_key); } else { char *action_key = pcmk__op_key(rsc_id, action, interval_ms); xpath = crm_strdup_printf(XPATH_HISTORY_ORIG, node, rsc_id, last_failure_key, action_key); free(action_key); } free(last_failure_key); controld_globals.cib_conn->cmds->remove(controld_globals.cib_conn, xpath, NULL, cib_xpath); free(xpath); } /*! * \internal * \brief Delete resource history entry from the CIB, given operation key * * \param[in] rsc_id Name of resource to clear history for * \param[in] node Name of node to clear history for * \param[in] key Operation key of operation to clear history for * \param[in] call_id If specified, delete entry only if it has this call ID */ void controld_delete_action_history_by_key(const char *rsc_id, const char *node, const char *key, int call_id) { char *xpath = NULL; CRM_CHECK((rsc_id != NULL) && (node != NULL) && (key != NULL), return); if (call_id > 0) { xpath = crm_strdup_printf(XPATH_HISTORY_CALL, node, rsc_id, key, call_id); } else { xpath = crm_strdup_printf(XPATH_HISTORY_ID, node, rsc_id, key); } controld_globals.cib_conn->cmds->remove(controld_globals.cib_conn, xpath, NULL, cib_xpath); free(xpath); } diff --git a/daemons/controld/controld_fsa.c b/daemons/controld/controld_fsa.c index 2840349815..79b3507ce5 100644 --- a/daemons/controld/controld_fsa.c +++ b/daemons/controld/controld_fsa.c @@ -1,733 +1,733 @@ /* * 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 // uint64_t #include #include #include #include #include #include #include #include #include //! Triggers an FSA invocation static crm_trigger_t *fsa_trigger = NULL; #define DOT_PREFIX "actions:trace: " #define do_dot_log(fmt, args...) crm_trace( fmt, ##args) static void do_state_transition(enum crmd_fsa_state cur_state, enum crmd_fsa_state next_state, fsa_data_t *msg_data); void s_crmd_fsa_actions(fsa_data_t * fsa_data); void log_fsa_input(fsa_data_t * stored_msg); void init_dotfile(void); void init_dotfile(void) { do_dot_log(DOT_PREFIX "digraph \"g\" {"); do_dot_log(DOT_PREFIX " size = \"30,30\""); do_dot_log(DOT_PREFIX " graph ["); do_dot_log(DOT_PREFIX " fontsize = \"12\""); do_dot_log(DOT_PREFIX " fontname = \"Times-Roman\""); do_dot_log(DOT_PREFIX " fontcolor = \"black\""); do_dot_log(DOT_PREFIX " bb = \"0,0,398.922306,478.927856\""); do_dot_log(DOT_PREFIX " color = \"black\""); do_dot_log(DOT_PREFIX " ]"); do_dot_log(DOT_PREFIX " node ["); do_dot_log(DOT_PREFIX " fontsize = \"12\""); do_dot_log(DOT_PREFIX " fontname = \"Times-Roman\""); do_dot_log(DOT_PREFIX " fontcolor = \"black\""); do_dot_log(DOT_PREFIX " shape = \"ellipse\""); do_dot_log(DOT_PREFIX " color = \"black\""); do_dot_log(DOT_PREFIX " ]"); do_dot_log(DOT_PREFIX " edge ["); do_dot_log(DOT_PREFIX " fontsize = \"12\""); do_dot_log(DOT_PREFIX " fontname = \"Times-Roman\""); do_dot_log(DOT_PREFIX " fontcolor = \"black\""); do_dot_log(DOT_PREFIX " color = \"black\""); do_dot_log(DOT_PREFIX " ]"); do_dot_log(DOT_PREFIX "// special nodes"); do_dot_log(DOT_PREFIX " \"S_PENDING\" "); do_dot_log(DOT_PREFIX " ["); do_dot_log(DOT_PREFIX " color = \"blue\""); do_dot_log(DOT_PREFIX " fontcolor = \"blue\""); do_dot_log(DOT_PREFIX " ]"); do_dot_log(DOT_PREFIX " \"S_TERMINATE\" "); do_dot_log(DOT_PREFIX " ["); do_dot_log(DOT_PREFIX " color = \"red\""); do_dot_log(DOT_PREFIX " fontcolor = \"red\""); do_dot_log(DOT_PREFIX " ]"); do_dot_log(DOT_PREFIX "// DC only nodes"); do_dot_log(DOT_PREFIX " \"S_INTEGRATION\" [ fontcolor = \"green\" ]"); do_dot_log(DOT_PREFIX " \"S_POLICY_ENGINE\" [ fontcolor = \"green\" ]"); do_dot_log(DOT_PREFIX " \"S_TRANSITION_ENGINE\" [ fontcolor = \"green\" ]"); do_dot_log(DOT_PREFIX " \"S_RELEASE_DC\" [ fontcolor = \"green\" ]"); do_dot_log(DOT_PREFIX " \"S_IDLE\" [ fontcolor = \"green\" ]"); } static void do_fsa_action(fsa_data_t * fsa_data, long long an_action, void (*function) (long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input cur_input, fsa_data_t * msg_data)) { controld_clear_fsa_action_flags(an_action); crm_trace(DOT_PREFIX "\t// %s", fsa_action2string(an_action)); function(an_action, fsa_data->fsa_cause, controld_globals.fsa_state, fsa_data->fsa_input, fsa_data); } static const uint64_t startup_actions = A_STARTUP | A_CIB_START | A_LRM_CONNECT | A_HA_CONNECT | A_READCONFIG | A_STARTED | A_CL_JOIN_QUERY; // A_LOG, A_WARN, A_ERROR void do_log(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) { unsigned log_type = LOG_TRACE; if (action & A_LOG) { log_type = LOG_INFO; } else if (action & A_WARN) { log_type = LOG_WARNING; } else if (action & A_ERROR) { log_type = LOG_ERR; } do_crm_log(log_type, "Input %s received in state %s from %s", fsa_input2string(msg_data->fsa_input), fsa_state2string(cur_state), msg_data->origin); if (msg_data->data_type == fsa_dt_ha_msg) { ha_msg_input_t *input = fsa_typed_data(msg_data->data_type); crm_log_xml_debug(input->msg, __func__); } else if (msg_data->data_type == fsa_dt_xml) { xmlNode *input = fsa_typed_data(msg_data->data_type); crm_log_xml_debug(input, __func__); } else if (msg_data->data_type == fsa_dt_lrm) { lrmd_event_data_t *input = fsa_typed_data(msg_data->data_type); do_crm_log(log_type, "Resource %s: Call ID %d returned %d (%d)." " New status if rc=0: %s", input->rsc_id, input->call_id, input->rc, input->op_status, (char *)input->user_data); } } /*! * \internal * \brief Initialize the FSA trigger */ void controld_init_fsa_trigger(void) { fsa_trigger = mainloop_add_trigger(G_PRIORITY_HIGH, crm_fsa_trigger, NULL); } /*! * \internal * \brief Destroy the FSA trigger */ void controld_destroy_fsa_trigger(void) { // This basically will not work, since mainloop has a reference to it mainloop_destroy_trigger(fsa_trigger); fsa_trigger = NULL; } /*! * \internal * \brief Trigger an FSA invocation * * \param[in] fn Calling function name * \param[in] line Line number where call occurred */ void controld_trigger_fsa_as(const char *fn, int line) { if (fsa_trigger != NULL) { crm_trace("%s:%d - Triggered FSA invocation", fn, line); mainloop_set_trigger(fsa_trigger); } } enum crmd_fsa_state s_crmd_fsa(enum crmd_fsa_cause cause) { controld_globals_t *globals = &controld_globals; fsa_data_t *fsa_data = NULL; uint64_t register_copy = controld_globals.fsa_input_register; uint64_t new_actions = A_NOTHING; enum crmd_fsa_state last_state; crm_trace("FSA invoked with Cause: %s\tState: %s", fsa_cause2string(cause), fsa_state2string(globals->fsa_state)); fsa_dump_actions(controld_globals.fsa_actions, "Initial"); controld_clear_global_flags(controld_fsa_is_stalled); if ((controld_globals.fsa_message_queue == NULL) && (controld_globals.fsa_actions != A_NOTHING)) { /* fake the first message so we can get into the loop */ fsa_data = pcmk__assert_alloc(1, sizeof(fsa_data_t)); fsa_data->fsa_input = I_NULL; fsa_data->fsa_cause = C_FSA_INTERNAL; fsa_data->origin = __func__; fsa_data->data_type = fsa_dt_none; controld_globals.fsa_message_queue = g_list_append(controld_globals.fsa_message_queue, fsa_data); } while ((controld_globals.fsa_message_queue != NULL) && !pcmk_is_set(controld_globals.flags, controld_fsa_is_stalled)) { crm_trace("Checking messages (%d remaining)", g_list_length(controld_globals.fsa_message_queue)); fsa_data = get_message(); if(fsa_data == NULL) { continue; } log_fsa_input(fsa_data); /* add any actions back to the queue */ controld_set_fsa_action_flags(fsa_data->actions); fsa_dump_actions(fsa_data->actions, "Restored actions"); /* get the next batch of actions */ new_actions = controld_fsa_get_action(fsa_data->fsa_input); controld_set_fsa_action_flags(new_actions); fsa_dump_actions(new_actions, "New actions"); if (fsa_data->fsa_input != I_NULL && fsa_data->fsa_input != I_ROUTER) { crm_debug("Processing %s: [ state=%s cause=%s origin=%s ]", fsa_input2string(fsa_data->fsa_input), fsa_state2string(globals->fsa_state), fsa_cause2string(fsa_data->fsa_cause), fsa_data->origin); } /* logging : *before* the state is changed */ if (pcmk_is_set(controld_globals.fsa_actions, A_ERROR)) { do_fsa_action(fsa_data, A_ERROR, do_log); } if (pcmk_is_set(controld_globals.fsa_actions, A_WARN)) { do_fsa_action(fsa_data, A_WARN, do_log); } if (pcmk_is_set(controld_globals.fsa_actions, A_LOG)) { do_fsa_action(fsa_data, A_LOG, do_log); } /* update state variables */ last_state = globals->fsa_state; globals->fsa_state = controld_fsa_get_next_state(fsa_data->fsa_input); /* * Remove certain actions during shutdown */ if ((globals->fsa_state == S_STOPPING) || pcmk_is_set(controld_globals.fsa_input_register, R_SHUTDOWN)) { controld_clear_fsa_action_flags(startup_actions); } /* * Hook for change of state. * Allows actions to be added or removed when entering a state */ if (last_state != globals->fsa_state) { do_state_transition(last_state, globals->fsa_state, fsa_data); } else { do_dot_log(DOT_PREFIX "\t// FSA input: State=%s \tCause=%s" " \tInput=%s \tOrigin=%s() \tid=%d", fsa_state2string(globals->fsa_state), fsa_cause2string(fsa_data->fsa_cause), fsa_input2string(fsa_data->fsa_input), fsa_data->origin, fsa_data->id); } /* start doing things... */ s_crmd_fsa_actions(fsa_data); delete_fsa_input(fsa_data); } if ((controld_globals.fsa_message_queue != NULL) || (controld_globals.fsa_actions != A_NOTHING) || pcmk_is_set(controld_globals.flags, controld_fsa_is_stalled)) { crm_debug("Exiting the FSA: queue=%d, fsa_actions=%#llx, stalled=%s", g_list_length(controld_globals.fsa_message_queue), (unsigned long long) controld_globals.fsa_actions, pcmk__flag_text(controld_globals.flags, controld_fsa_is_stalled)); } else { crm_trace("Exiting the FSA"); } /* cleanup inputs? */ if (register_copy != controld_globals.fsa_input_register) { uint64_t same = register_copy & controld_globals.fsa_input_register; fsa_dump_inputs(LOG_DEBUG, "Added", controld_globals.fsa_input_register ^ same); fsa_dump_inputs(LOG_DEBUG, "Removed", register_copy ^ same); } fsa_dump_actions(controld_globals.fsa_actions, "Remaining"); fsa_dump_queue(LOG_DEBUG); return globals->fsa_state; } void s_crmd_fsa_actions(fsa_data_t * fsa_data) { /* * Process actions in order of priority but do only one * action at a time to avoid complicating the ordering. */ CRM_CHECK(fsa_data != NULL, return); while ((controld_globals.fsa_actions != A_NOTHING) && !pcmk_is_set(controld_globals.flags, controld_fsa_is_stalled)) { /* regular action processing in order of action priority * * Make sure all actions that connect to required systems * are performed first */ if (pcmk_is_set(controld_globals.fsa_actions, A_ERROR)) { do_fsa_action(fsa_data, A_ERROR, do_log); } else if (pcmk_is_set(controld_globals.fsa_actions, A_WARN)) { do_fsa_action(fsa_data, A_WARN, do_log); } else if (pcmk_is_set(controld_globals.fsa_actions, A_LOG)) { do_fsa_action(fsa_data, A_LOG, do_log); /* get out of here NOW! before anything worse happens */ } else if (pcmk_is_set(controld_globals.fsa_actions, A_EXIT_1)) { do_fsa_action(fsa_data, A_EXIT_1, do_exit); /* sub-system restart */ } else if (pcmk_all_flags_set(controld_globals.fsa_actions, O_LRM_RECONNECT)) { do_fsa_action(fsa_data, O_LRM_RECONNECT, do_lrm_control); } else if (pcmk_all_flags_set(controld_globals.fsa_actions, O_CIB_RESTART)) { do_fsa_action(fsa_data, O_CIB_RESTART, do_cib_control); } else if (pcmk_all_flags_set(controld_globals.fsa_actions, O_PE_RESTART)) { do_fsa_action(fsa_data, O_PE_RESTART, do_pe_control); } else if (pcmk_all_flags_set(controld_globals.fsa_actions, O_TE_RESTART)) { do_fsa_action(fsa_data, O_TE_RESTART, do_te_control); /* essential start tasks */ } else if (pcmk_is_set(controld_globals.fsa_actions, A_STARTUP)) { do_fsa_action(fsa_data, A_STARTUP, do_startup); } else if (pcmk_is_set(controld_globals.fsa_actions, A_CIB_START)) { do_fsa_action(fsa_data, A_CIB_START, do_cib_control); } else if (pcmk_is_set(controld_globals.fsa_actions, A_HA_CONNECT)) { do_fsa_action(fsa_data, A_HA_CONNECT, do_ha_control); } else if (pcmk_is_set(controld_globals.fsa_actions, A_READCONFIG)) { do_fsa_action(fsa_data, A_READCONFIG, do_read_config); /* sub-system start/connect */ } else if (pcmk_is_set(controld_globals.fsa_actions, A_LRM_CONNECT)) { do_fsa_action(fsa_data, A_LRM_CONNECT, do_lrm_control); } else if (pcmk_is_set(controld_globals.fsa_actions, A_TE_START)) { do_fsa_action(fsa_data, A_TE_START, do_te_control); } else if (pcmk_is_set(controld_globals.fsa_actions, A_PE_START)) { do_fsa_action(fsa_data, A_PE_START, do_pe_control); /* Timers */ } else if (pcmk_is_set(controld_globals.fsa_actions, A_DC_TIMER_STOP)) { do_fsa_action(fsa_data, A_DC_TIMER_STOP, do_timer_control); } else if (pcmk_is_set(controld_globals.fsa_actions, A_INTEGRATE_TIMER_STOP)) { do_fsa_action(fsa_data, A_INTEGRATE_TIMER_STOP, do_timer_control); } else if (pcmk_is_set(controld_globals.fsa_actions, A_INTEGRATE_TIMER_START)) { do_fsa_action(fsa_data, A_INTEGRATE_TIMER_START, do_timer_control); } else if (pcmk_is_set(controld_globals.fsa_actions, A_FINALIZE_TIMER_STOP)) { do_fsa_action(fsa_data, A_FINALIZE_TIMER_STOP, do_timer_control); } else if (pcmk_is_set(controld_globals.fsa_actions, A_FINALIZE_TIMER_START)) { do_fsa_action(fsa_data, A_FINALIZE_TIMER_START, do_timer_control); /* * Highest priority actions */ } else if (pcmk_is_set(controld_globals.fsa_actions, A_MSG_ROUTE)) { do_fsa_action(fsa_data, A_MSG_ROUTE, do_msg_route); } else if (pcmk_is_set(controld_globals.fsa_actions, A_RECOVER)) { do_fsa_action(fsa_data, A_RECOVER, do_recover); } else if (pcmk_is_set(controld_globals.fsa_actions, A_CL_JOIN_RESULT)) { do_fsa_action(fsa_data, A_CL_JOIN_RESULT, do_cl_join_finalize_respond); } else if (pcmk_is_set(controld_globals.fsa_actions, A_CL_JOIN_REQUEST)) { do_fsa_action(fsa_data, A_CL_JOIN_REQUEST, do_cl_join_offer_respond); } else if (pcmk_is_set(controld_globals.fsa_actions, A_SHUTDOWN_REQ)) { do_fsa_action(fsa_data, A_SHUTDOWN_REQ, do_shutdown_req); } else if (pcmk_is_set(controld_globals.fsa_actions, A_ELECTION_VOTE)) { do_fsa_action(fsa_data, A_ELECTION_VOTE, do_election_vote); } else if (pcmk_is_set(controld_globals.fsa_actions, A_ELECTION_COUNT)) { do_fsa_action(fsa_data, A_ELECTION_COUNT, do_election_count_vote); } else if (pcmk_is_set(controld_globals.fsa_actions, A_LRM_EVENT)) { do_fsa_action(fsa_data, A_LRM_EVENT, do_lrm_event); /* * High priority actions */ } else if (pcmk_is_set(controld_globals.fsa_actions, A_STARTED)) { do_fsa_action(fsa_data, A_STARTED, do_started); } else if (pcmk_is_set(controld_globals.fsa_actions, A_CL_JOIN_QUERY)) { do_fsa_action(fsa_data, A_CL_JOIN_QUERY, do_cl_join_query); } else if (pcmk_is_set(controld_globals.fsa_actions, A_DC_TIMER_START)) { do_fsa_action(fsa_data, A_DC_TIMER_START, do_timer_control); /* * Medium priority actions * - Membership */ } else if (pcmk_is_set(controld_globals.fsa_actions, A_DC_TAKEOVER)) { do_fsa_action(fsa_data, A_DC_TAKEOVER, do_dc_takeover); } else if (pcmk_is_set(controld_globals.fsa_actions, A_DC_RELEASE)) { do_fsa_action(fsa_data, A_DC_RELEASE, do_dc_release); } else if (pcmk_is_set(controld_globals.fsa_actions, A_DC_JOIN_FINAL)) { do_fsa_action(fsa_data, A_DC_JOIN_FINAL, do_dc_join_final); } else if (pcmk_is_set(controld_globals.fsa_actions, A_ELECTION_CHECK)) { do_fsa_action(fsa_data, A_ELECTION_CHECK, do_election_check); } else if (pcmk_is_set(controld_globals.fsa_actions, A_ELECTION_START)) { do_fsa_action(fsa_data, A_ELECTION_START, do_election_vote); } else if (pcmk_is_set(controld_globals.fsa_actions, A_DC_JOIN_OFFER_ALL)) { do_fsa_action(fsa_data, A_DC_JOIN_OFFER_ALL, do_dc_join_offer_all); } else if (pcmk_is_set(controld_globals.fsa_actions, A_DC_JOIN_OFFER_ONE)) { do_fsa_action(fsa_data, A_DC_JOIN_OFFER_ONE, do_dc_join_offer_one); } else if (pcmk_is_set(controld_globals.fsa_actions, A_DC_JOIN_PROCESS_REQ)) { do_fsa_action(fsa_data, A_DC_JOIN_PROCESS_REQ, do_dc_join_filter_offer); } else if (pcmk_is_set(controld_globals.fsa_actions, A_DC_JOIN_PROCESS_ACK)) { do_fsa_action(fsa_data, A_DC_JOIN_PROCESS_ACK, do_dc_join_ack); } else if (pcmk_is_set(controld_globals.fsa_actions, A_DC_JOIN_FINALIZE)) { do_fsa_action(fsa_data, A_DC_JOIN_FINALIZE, do_dc_join_finalize); } else if (pcmk_is_set(controld_globals.fsa_actions, A_CL_JOIN_ANNOUNCE)) { do_fsa_action(fsa_data, A_CL_JOIN_ANNOUNCE, do_cl_join_announce); /* * Low(er) priority actions * Make sure the CIB is always updated before invoking the * scheduler, and the scheduler before the transition engine. */ } else if (pcmk_is_set(controld_globals.fsa_actions, A_TE_HALT)) { do_fsa_action(fsa_data, A_TE_HALT, do_te_invoke); } else if (pcmk_is_set(controld_globals.fsa_actions, A_TE_CANCEL)) { do_fsa_action(fsa_data, A_TE_CANCEL, do_te_invoke); } else if (pcmk_is_set(controld_globals.fsa_actions, A_LRM_INVOKE)) { do_fsa_action(fsa_data, A_LRM_INVOKE, do_lrm_invoke); } else if (pcmk_is_set(controld_globals.fsa_actions, A_PE_INVOKE)) { do_fsa_action(fsa_data, A_PE_INVOKE, do_pe_invoke); } else if (pcmk_is_set(controld_globals.fsa_actions, A_TE_INVOKE)) { do_fsa_action(fsa_data, A_TE_INVOKE, do_te_invoke); /* Shutdown actions */ } else if (pcmk_is_set(controld_globals.fsa_actions, A_DC_RELEASED)) { do_fsa_action(fsa_data, A_DC_RELEASED, do_dc_release); } else if (pcmk_is_set(controld_globals.fsa_actions, A_PE_STOP)) { do_fsa_action(fsa_data, A_PE_STOP, do_pe_control); } else if (pcmk_is_set(controld_globals.fsa_actions, A_TE_STOP)) { do_fsa_action(fsa_data, A_TE_STOP, do_te_control); } else if (pcmk_is_set(controld_globals.fsa_actions, A_SHUTDOWN)) { do_fsa_action(fsa_data, A_SHUTDOWN, do_shutdown); } else if (pcmk_is_set(controld_globals.fsa_actions, A_LRM_DISCONNECT)) { do_fsa_action(fsa_data, A_LRM_DISCONNECT, do_lrm_control); } else if (pcmk_is_set(controld_globals.fsa_actions, A_HA_DISCONNECT)) { do_fsa_action(fsa_data, A_HA_DISCONNECT, do_ha_control); } else if (pcmk_is_set(controld_globals.fsa_actions, A_CIB_STOP)) { do_fsa_action(fsa_data, A_CIB_STOP, do_cib_control); } else if (pcmk_is_set(controld_globals.fsa_actions, A_STOP)) { do_fsa_action(fsa_data, A_STOP, do_stop); /* exit gracefully */ } else if (pcmk_is_set(controld_globals.fsa_actions, A_EXIT_0)) { do_fsa_action(fsa_data, A_EXIT_0, do_exit); /* Error checking and reporting */ } else { crm_err("Action %s not supported "CRM_XS" %#llx", fsa_action2string(controld_globals.fsa_actions), (unsigned long long) controld_globals.fsa_actions); register_fsa_error_adv(C_FSA_INTERNAL, I_ERROR, fsa_data, NULL, __func__); } } } void log_fsa_input(fsa_data_t * stored_msg) { CRM_ASSERT(stored_msg); crm_trace("Processing queued input %d", stored_msg->id); if (stored_msg->fsa_cause == C_LRM_OP_CALLBACK) { crm_trace("FSA processing LRM callback from %s", stored_msg->origin); } else if (stored_msg->data == NULL) { crm_trace("FSA processing input from %s", stored_msg->origin); } else { ha_msg_input_t *ha_input = fsa_typed_data_adv(stored_msg, fsa_dt_ha_msg, __func__); crm_trace("FSA processing XML message from %s", stored_msg->origin); crm_log_xml_trace(ha_input->xml, "FSA message data"); } } static void check_join_counts(fsa_data_t *msg_data) { int count; guint npeers; count = crmd_join_phase_count(crm_join_finalized); if (count > 0) { crm_err("%d cluster node%s failed to confirm join", count, pcmk__plural_s(count)); crmd_join_phase_log(LOG_NOTICE); return; } - npeers = crm_active_peers(); + npeers = pcmk__cluster_num_active_nodes(); count = crmd_join_phase_count(crm_join_confirmed); if (count == npeers) { if (npeers == 1) { crm_debug("Sole active cluster node is fully joined"); } else { crm_debug("All %d active cluster nodes are fully joined", count); } } else if (count > npeers) { crm_err("New election needed because more nodes confirmed join " "than are in membership (%d > %u)", count, npeers); register_fsa_input(C_FSA_INTERNAL, I_ELECTION, NULL); } else if (controld_globals.membership_id != crm_peer_seq) { crm_info("New join needed because membership changed (%llu -> %llu)", controld_globals.membership_id, crm_peer_seq); register_fsa_input_before(C_FSA_INTERNAL, I_NODE_JOIN, NULL); } else { crm_warn("Only %d of %u active cluster nodes fully joined " "(%d did not respond to offer)", count, npeers, crmd_join_phase_count(crm_join_welcomed)); } } static void do_state_transition(enum crmd_fsa_state cur_state, enum crmd_fsa_state next_state, fsa_data_t *msg_data) { int level = LOG_INFO; int count = 0; gboolean clear_recovery_bit = TRUE; #if 0 uint64_t original_fsa_actions = controld_globals.fsa_actions; #endif enum crmd_fsa_cause cause = msg_data->fsa_cause; enum crmd_fsa_input current_input = msg_data->fsa_input; const char *state_from = fsa_state2string(cur_state); const char *state_to = fsa_state2string(next_state); const char *input = fsa_input2string(current_input); CRM_LOG_ASSERT(cur_state != next_state); do_dot_log(DOT_PREFIX "\t%s -> %s [ label=%s cause=%s origin=%s ]", state_from, state_to, input, fsa_cause2string(cause), msg_data->origin); if (cur_state == S_IDLE || next_state == S_IDLE) { level = LOG_NOTICE; } else if (cur_state == S_NOT_DC || next_state == S_NOT_DC) { level = LOG_NOTICE; } else if (cur_state == S_ELECTION) { level = LOG_NOTICE; } else if (cur_state == S_STARTING) { level = LOG_NOTICE; } else if (next_state == S_RECOVERY) { level = LOG_WARNING; } do_crm_log(level, "State transition %s -> %s " CRM_XS " input=%s cause=%s origin=%s", state_from, state_to, input, fsa_cause2string(cause), msg_data->origin); if (next_state != S_ELECTION && cur_state != S_RELEASE_DC) { controld_stop_current_election_timeout(); } if (next_state == S_INTEGRATION) { controld_set_fsa_action_flags(A_INTEGRATE_TIMER_START); } else { controld_set_fsa_action_flags(A_INTEGRATE_TIMER_STOP); } if (next_state == S_FINALIZE_JOIN) { controld_set_fsa_action_flags(A_FINALIZE_TIMER_START); } else { controld_set_fsa_action_flags(A_FINALIZE_TIMER_STOP); } if (next_state != S_PENDING) { controld_set_fsa_action_flags(A_DC_TIMER_STOP); } if (next_state != S_IDLE) { controld_stop_recheck_timer(); } if (cur_state == S_FINALIZE_JOIN && next_state == S_POLICY_ENGINE) { populate_cib_nodes(node_update_quick|node_update_all, __func__); } switch (next_state) { case S_PENDING: { cib_t *cib_conn = controld_globals.cib_conn; cib_conn->cmds->set_secondary(cib_conn, cib_scope_local); } update_dc(NULL); break; case S_ELECTION: update_dc(NULL); break; case S_NOT_DC: controld_reset_counter_election_timer(); purge_stonith_cleanup(); if (pcmk_is_set(controld_globals.fsa_input_register, R_SHUTDOWN)) { crm_info("(Re)Issuing shutdown request now" " that we have a new DC"); controld_set_fsa_action_flags(A_SHUTDOWN_REQ); } CRM_LOG_ASSERT(controld_globals.dc_name != NULL); if (controld_globals.dc_name == NULL) { crm_err("Reached S_NOT_DC without a DC" " being recorded"); } break; case S_RECOVERY: clear_recovery_bit = FALSE; break; case S_FINALIZE_JOIN: CRM_LOG_ASSERT(AM_I_DC); if (cause == C_TIMER_POPPED) { crm_warn("Progressed to state %s after %s", fsa_state2string(next_state), fsa_cause2string(cause)); } count = crmd_join_phase_count(crm_join_welcomed); if (count > 0) { crm_warn("%d cluster node%s failed to respond to join offer", count, pcmk__plural_s(count)); crmd_join_phase_log(LOG_NOTICE); } else { crm_debug("All cluster nodes (%d) responded to join offer", crmd_join_phase_count(crm_join_integrated)); } break; case S_POLICY_ENGINE: controld_reset_counter_election_timer(); CRM_LOG_ASSERT(AM_I_DC); if (cause == C_TIMER_POPPED) { crm_info("Progressed to state %s after %s", fsa_state2string(next_state), fsa_cause2string(cause)); } check_join_counts(msg_data); break; case S_STOPPING: case S_TERMINATE: /* possibly redundant */ controld_set_fsa_input_flags(R_SHUTDOWN); break; case S_IDLE: CRM_LOG_ASSERT(AM_I_DC); if (pcmk_is_set(controld_globals.fsa_input_register, R_SHUTDOWN)) { crm_info("(Re)Issuing shutdown request now" " that we are the DC"); controld_set_fsa_action_flags(A_SHUTDOWN_REQ); } controld_start_recheck_timer(); break; default: break; } if (clear_recovery_bit && next_state != S_PENDING) { controld_clear_fsa_action_flags(A_RECOVER); } else if (clear_recovery_bit == FALSE) { controld_set_fsa_action_flags(A_RECOVER); } #if 0 if (original_fsa_actions != controld_globals.fsa_actions) { fsa_dump_actions(original_fsa_actions ^ controld_globals.fsa_actions, "New actions"); } #endif } diff --git a/daemons/controld/controld_throttle.c b/daemons/controld/controld_throttle.c index 95045e725f..0a1142d0ea 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); send_cluster_message(NULL, crm_msg_crmd, xml, TRUE); free_xml(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 = crm_active_peers(); + 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/include/crm/cluster/internal.h b/include/crm/cluster/internal.h index c46377e6e3..abc9afa941 100644 --- a/include/crm/cluster/internal.h +++ b/include/crm/cluster/internal.h @@ -1,159 +1,160 @@ /* * 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 // uint32_t, uint64_t # include // gboolean # include /* *INDENT-OFF* */ enum crm_proc_flag { crm_proc_none = 0x00000001, // Cluster layers crm_proc_cpg = 0x04000000, // Daemons crm_proc_execd = 0x00000010, crm_proc_based = 0x00000100, crm_proc_controld = 0x00000200, crm_proc_attrd = 0x00001000, crm_proc_schedulerd = 0x00010000, crm_proc_fenced = 0x00100000, }; /* *INDENT-ON* */ // Used with node cache search functions enum pcmk__node_search_flags { pcmk__node_search_none = 0, pcmk__node_search_cluster = (1 << 0), // Search for cluster nodes pcmk__node_search_remote = (1 << 1), // Search for remote nodes pcmk__node_search_any = pcmk__node_search_cluster |pcmk__node_search_remote, /* @COMPAT The values before this must stay the same until we can drop * support for enum crm_get_peer_flags */ pcmk__node_search_known = (1 << 2), // Search previously known nodes }; /*! * \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 (get_cluster_type()) { case pcmk_cluster_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); # endif crm_node_t *crm_update_peer_proc(const char *source, crm_node_t * peer, uint32_t flag, const char *status); crm_node_t *pcmk__update_peer_state(const char *source, crm_node_t *node, const char *state, uint64_t membership); void pcmk__update_peer_expected(const char *source, crm_node_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)); bool pcmk__cluster_is_node_active(const crm_node_t *node); +unsigned int pcmk__cluster_num_active_nodes(void); unsigned int pcmk__cluster_num_remote_nodes(void); crm_node_t *pcmk__cluster_lookup_remote_node(const char *node_name); void pcmk__cluster_forget_remote_node(const char *node_name); crm_node_t *pcmk__search_node_caches(unsigned int id, const char *uname, uint32_t flags); crm_node_t *pcmk__search_cluster_node_cache(unsigned int id, const char *uname, const char *uuid); void pcmk__purge_node_from_cache(const char *node_name, uint32_t node_id); void pcmk__refresh_node_caches_from_cib(xmlNode *cib); crm_node_t *pcmk__get_node(unsigned int id, const char *uname, const char *uuid, uint32_t flags); #endif // PCMK__CRM_CLUSTER_INTERNAL__H diff --git a/lib/cluster/election.c b/lib/cluster/election.c index ec9a6f6f56..02bf024e2c 100644 --- a/lib/cluster/election.c +++ b/lib/cluster/election.c @@ -1,723 +1,723 @@ /* * 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; crm_node_t *our_node; 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); 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->uuid); 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); send_cluster_message(NULL, crm_msg_crmd, vote, TRUE); free_xml(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 = crm_active_peers(); + 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 crm_node_t *node; 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->uname); } } 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(crm_node_t *peer, struct vote *vote) { // @TODO probably shouldn't hardcode CRM_SYSTEM_CRMD and crm_msg_crmd 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); send_cluster_message(peer, crm_msg_crmd, novote, TRUE); free_xml(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; crm_node_t *our_node = NULL, *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); our_node = pcmk__get_node(0, e->uname, NULL, pcmk__node_search_cluster); we_are_owner = (our_node != NULL) && pcmk__str_eq(our_node->uuid, 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; } diff --git a/lib/cluster/membership.c b/lib/cluster/membership.c index 63fa0c10f4..d900563c28 100644 --- a/lib/cluster/membership.c +++ b/lib/cluster/membership.c @@ -1,1441 +1,1456 @@ /* * 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 #ifndef _GNU_SOURCE # define _GNU_SOURCE #endif #include #include #include #include #include #include #include #include #include #include #include #include "crmcluster_private.h" /* The peer cache remembers cluster nodes that have been seen. * This is managed mostly automatically by libcluster, based on * cluster membership events. * * Because cluster nodes can have conflicting names or UUIDs, * the hash table key is a uniquely generated ID. */ GHashTable *crm_peer_cache = NULL; /* * The remote peer cache tracks pacemaker_remote nodes. While the * value has the same type as the peer cache's, it is tracked separately for * three reasons: pacemaker_remote nodes can't have conflicting names or UUIDs, * so the name (which is also the UUID) is used as the hash table key; there * is no equivalent of membership events, so management is not automatic; and * most users of the peer cache need to exclude pacemaker_remote nodes. * * That said, using a single cache would be more logical and less error-prone, * so it would be a good idea to merge them one day. * * libcluster provides two avenues for populating the cache: * pcmk__cluster_lookup_remote_node() and pcmk__cluster_forget_remote_node() * directly manage it, while refresh_remote_nodes() populates it via the CIB. */ GHashTable *crm_remote_peer_cache = NULL; /* * The known node cache tracks cluster and remote nodes that have been seen in * the CIB. It is useful mainly when a caller needs to know about a node that * may no longer be in the membership, but doesn't want to add the node to the * main peer cache tables. */ static GHashTable *known_node_cache = NULL; unsigned long long crm_peer_seq = 0; gboolean crm_have_quorum = FALSE; static gboolean crm_autoreap = TRUE; // Flag setting and clearing for crm_node_t:flags #define set_peer_flags(peer, flags_to_set) do { \ (peer)->flags = pcmk__set_flags_as(__func__, __LINE__, LOG_TRACE, \ "Peer", (peer)->uname, \ (peer)->flags, (flags_to_set), \ #flags_to_set); \ } while (0) #define clear_peer_flags(peer, flags_to_clear) do { \ (peer)->flags = pcmk__clear_flags_as(__func__, __LINE__, \ LOG_TRACE, \ "Peer", (peer)->uname, \ (peer)->flags, (flags_to_clear), \ #flags_to_clear); \ } while (0) static void update_peer_uname(crm_node_t *node, const char *uname); static crm_node_t *find_known_node(const char *id, const char *uname); /*! * \internal * \brief Get the number of Pacemaker Remote nodes that have been seen * * \return Number of cached Pacemaker Remote nodes */ unsigned int pcmk__cluster_num_remote_nodes(void) { if (crm_remote_peer_cache == NULL) { return 0U; } return g_hash_table_size(crm_remote_peer_cache); } /*! * \internal * \brief Get a remote node cache entry, creating it if necessary * * \param[in] node_name Name of remote node * * \return Cache entry for node on success, or \c NULL (and set \c errno) * otherwise * * \note When creating a new entry, this will leave the node state undetermined. * The caller should also call \c pcmk__update_peer_state() if the state * is known. * \note Because this can add and remove cache entries, callers should not * assume any previously obtained cache entry pointers remain valid. */ crm_node_t * pcmk__cluster_lookup_remote_node(const char *node_name) { crm_node_t *node; char *node_name_copy = NULL; if (node_name == NULL) { errno = EINVAL; return NULL; } /* It's theoretically possible that the node was added to the cluster peer * cache before it was known to be a Pacemaker Remote node. Remove that * entry unless it has a node ID, which means the name actually is * associated with a cluster node. (@TODO return an error in that case?) */ node = pcmk__search_node_caches(0, node_name, pcmk__node_search_cluster); if ((node != NULL) && (node->uuid == NULL)) { /* node_name could be a pointer into the cache entry being removed, so * reassign it to a copy before the original gets freed */ node_name_copy = strdup(node_name); if (node_name_copy == NULL) { errno = ENOMEM; return NULL; } node_name = node_name_copy; reap_crm_member(0, node_name); } /* Return existing cache entry if one exists */ node = g_hash_table_lookup(crm_remote_peer_cache, node_name); if (node) { free(node_name_copy); return node; } /* Allocate a new entry */ node = calloc(1, sizeof(crm_node_t)); if (node == NULL) { free(node_name_copy); return NULL; } /* Populate the essential information */ set_peer_flags(node, crm_remote_node); node->uuid = strdup(node_name); if (node->uuid == NULL) { free(node); errno = ENOMEM; free(node_name_copy); return NULL; } /* Add the new entry to the cache */ g_hash_table_replace(crm_remote_peer_cache, node->uuid, node); crm_trace("added %s to remote cache", node_name); /* Update the entry's uname, ensuring peer status callbacks are called */ update_peer_uname(node, node_name); free(node_name_copy); return node; } /*! * \internal * \brief Remove a node from the Pacemaker Remote node cache * * \param[in] node_name Name of node to remove from cache * * \note The caller must be careful not to use \p node_name after calling this * function if it might be a pointer into the cache entry being removed. */ void pcmk__cluster_forget_remote_node(const char *node_name) { /* Do a lookup first, because node_name could be a pointer within the entry * being removed -- we can't log it *after* removing it. */ if (g_hash_table_lookup(crm_remote_peer_cache, node_name) != NULL) { crm_trace("Removing %s from Pacemaker Remote node cache", node_name); g_hash_table_remove(crm_remote_peer_cache, node_name); } } /*! * \internal * \brief Return node status based on a CIB status entry * * \param[in] node_state XML of node state * * \return \c CRM_NODE_LOST if \c PCMK__XA_IN_CCM is false in * \c PCMK__XE_NODE_STATE, \c CRM_NODE_MEMBER otherwise * \note Unlike most boolean XML attributes, this one defaults to true, for * backward compatibility with older controllers that don't set it. */ static const char * remote_state_from_cib(const xmlNode *node_state) { bool status = false; if ((pcmk__xe_get_bool_attr(node_state, PCMK__XA_IN_CCM, &status) == pcmk_rc_ok) && !status) { return CRM_NODE_LOST; } else { return CRM_NODE_MEMBER; } } /* user data for looping through remote node xpath searches */ struct refresh_data { const char *field; /* XML attribute to check for node name */ gboolean has_state; /* whether to update node state based on XML */ }; /*! * \internal * \brief Process one pacemaker_remote node xpath search result * * \param[in] result XML search result * \param[in] user_data what to look for in the XML */ static void remote_cache_refresh_helper(xmlNode *result, void *user_data) { const struct refresh_data *data = user_data; const char *remote = crm_element_value(result, data->field); const char *state = NULL; crm_node_t *node; CRM_CHECK(remote != NULL, return); /* Determine node's state, if the result has it */ if (data->has_state) { state = remote_state_from_cib(result); } /* Check whether cache already has entry for node */ node = g_hash_table_lookup(crm_remote_peer_cache, remote); if (node == NULL) { /* Node is not in cache, so add a new entry for it */ node = pcmk__cluster_lookup_remote_node(remote); CRM_ASSERT(node); if (state) { pcmk__update_peer_state(__func__, node, state, 0); } } else if (pcmk_is_set(node->flags, crm_node_dirty)) { /* Node is in cache and hasn't been updated already, so mark it clean */ clear_peer_flags(node, crm_node_dirty); if (state) { pcmk__update_peer_state(__func__, node, state, 0); } } } static void mark_dirty(gpointer key, gpointer value, gpointer user_data) { set_peer_flags((crm_node_t *) value, crm_node_dirty); } static gboolean is_dirty(gpointer key, gpointer value, gpointer user_data) { return pcmk_is_set(((crm_node_t*)value)->flags, crm_node_dirty); } /*! * \internal * \brief Repopulate the remote node cache based on CIB XML * * \param[in] cib CIB XML to parse */ static void refresh_remote_nodes(xmlNode *cib) { struct refresh_data data; crm_peer_init(); /* First, we mark all existing cache entries as dirty, * so that later we can remove any that weren't in the CIB. * We don't empty the cache, because we need to detect changes in state. */ g_hash_table_foreach(crm_remote_peer_cache, mark_dirty, NULL); /* Look for guest nodes and remote nodes in the status section */ data.field = PCMK_XA_ID; data.has_state = TRUE; crm_foreach_xpath_result(cib, PCMK__XP_REMOTE_NODE_STATUS, remote_cache_refresh_helper, &data); /* Look for guest nodes and remote nodes in the configuration section, * because they may have just been added and not have a status entry yet. * In that case, the cached node state will be left NULL, so that the * peer status callback isn't called until we're sure the node started * successfully. */ data.field = PCMK_XA_VALUE; data.has_state = FALSE; crm_foreach_xpath_result(cib, PCMK__XP_GUEST_NODE_CONFIG, remote_cache_refresh_helper, &data); data.field = PCMK_XA_ID; data.has_state = FALSE; crm_foreach_xpath_result(cib, PCMK__XP_REMOTE_NODE_CONFIG, remote_cache_refresh_helper, &data); /* Remove all old cache entries that weren't seen in the CIB */ g_hash_table_foreach_remove(crm_remote_peer_cache, is_dirty, NULL); } /*! * \internal * \brief Check whether a node is an active cluster node * * Remote nodes are never considered active. This guarantees that they can never * become DC. * * \param[in] node Node to check * * \return \c true if the node is an active cluster node, or \c false otherwise */ bool pcmk__cluster_is_node_active(const crm_node_t *node) { const enum cluster_type_e type = get_cluster_type(); if ((node == NULL) || pcmk_is_set(node->flags, crm_remote_node)) { return false; } switch (type) { case pcmk_cluster_corosync: #if SUPPORT_COROSYNC return crm_is_corosync_peer_active(node); #else break; #endif // SUPPORT_COROSYNC default: break; } crm_err("Unhandled cluster type: %s", name_for_cluster_type(type)); return false; } static gboolean crm_reap_dead_member(gpointer key, gpointer value, gpointer user_data) { crm_node_t *node = value; crm_node_t *search = user_data; if (search == NULL) { return FALSE; } else if (search->id && node->id != search->id) { return FALSE; } else if (search->id == 0 && !pcmk__str_eq(node->uname, search->uname, pcmk__str_casei)) { return FALSE; } else if (!pcmk__cluster_is_node_active(value)) { crm_info("Removing node with name %s and " PCMK_XA_ID " %u from membership cache", (node->uname? node->uname : "unknown"), node->id); return TRUE; } return FALSE; } /*! * \brief Remove all peer cache entries matching a node ID and/or uname * * \param[in] id ID of node to remove (or 0 to ignore) * \param[in] name Uname of node to remove (or NULL to ignore) * * \return Number of cache entries removed * * \note The caller must be careful not to use \p name after calling this * function if it might be a pointer into the cache entry being removed. */ guint reap_crm_member(uint32_t id, const char *name) { int matches = 0; crm_node_t search = { 0, }; if (crm_peer_cache == NULL) { crm_trace("Membership cache not initialized, ignoring purge request"); return 0; } search.id = id; search.uname = pcmk__str_copy(name); matches = g_hash_table_foreach_remove(crm_peer_cache, crm_reap_dead_member, &search); if(matches) { crm_notice("Purged %d peer%s with " PCMK_XA_ID "=%u%s%s from the membership cache", matches, pcmk__plural_s(matches), search.id, (search.uname? " and/or uname=" : ""), (search.uname? search.uname : "")); } else { crm_info("No peers with " PCMK_XA_ID "=%u%s%s to purge from the membership cache", search.id, (search.uname? " and/or uname=" : ""), (search.uname? search.uname : "")); } free(search.uname); return matches; } static void count_peer(gpointer key, gpointer value, gpointer user_data) { - guint *count = user_data; + unsigned int *count = user_data; crm_node_t *node = value; if (pcmk__cluster_is_node_active(node)) { *count = *count + 1; } } -guint -crm_active_peers(void) +/*! + * \internal + * \brief Get the number of active cluster nodes that have been seen + * + * Remote nodes are never considered active. This guarantees that they can never + * become DC. + * + * \return Number of active nodes in the cluster node cache + */ +unsigned int +pcmk__cluster_num_active_nodes(void) { - guint count = 0; + unsigned int count = 0; - if (crm_peer_cache) { + if (crm_peer_cache != NULL) { g_hash_table_foreach(crm_peer_cache, count_peer, &count); } return count; } +guint +crm_active_peers(void) +{ + return pcmk__cluster_num_active_nodes(); +} + static void destroy_crm_node(gpointer data) { crm_node_t *node = data; crm_trace("Destroying entry for node %u: %s", node->id, node->uname); free(node->uname); free(node->state); free(node->uuid); free(node->expected); free(node->conn_host); free(node); } void crm_peer_init(void) { if (crm_peer_cache == NULL) { crm_peer_cache = pcmk__strikey_table(free, destroy_crm_node); } if (crm_remote_peer_cache == NULL) { crm_remote_peer_cache = pcmk__strikey_table(NULL, destroy_crm_node); } if (known_node_cache == NULL) { known_node_cache = pcmk__strikey_table(free, destroy_crm_node); } } void crm_peer_destroy(void) { if (crm_peer_cache != NULL) { crm_trace("Destroying peer cache with %d members", g_hash_table_size(crm_peer_cache)); g_hash_table_destroy(crm_peer_cache); crm_peer_cache = NULL; } if (crm_remote_peer_cache != NULL) { crm_trace("Destroying remote peer cache with %d members", pcmk__cluster_num_remote_nodes()); g_hash_table_destroy(crm_remote_peer_cache); crm_remote_peer_cache = NULL; } if (known_node_cache != NULL) { crm_trace("Destroying known node cache with %d members", g_hash_table_size(known_node_cache)); g_hash_table_destroy(known_node_cache); known_node_cache = NULL; } } static void (*peer_status_callback)(enum crm_status_type, crm_node_t *, const void *) = NULL; /*! * \brief Set a client function that will be called after peer status changes * * \param[in] dispatch Pointer to function to use as callback * * \note Previously, client callbacks were responsible for peer cache * management. This is no longer the case, and client callbacks should do * only client-specific handling. Callbacks MUST NOT add or remove entries * in the peer caches. */ void crm_set_status_callback(void (*dispatch) (enum crm_status_type, crm_node_t *, const void *)) { peer_status_callback = dispatch; } /*! * \brief Tell the library whether to automatically reap lost nodes * * If TRUE (the default), calling crm_update_peer_proc() will also update the * peer state to CRM_NODE_MEMBER or CRM_NODE_LOST, and pcmk__update_peer_state() * will reap peers whose state changes to anything other than CRM_NODE_MEMBER. * Callers should leave this enabled unless they plan to manage the cache * separately on their own. * * \param[in] autoreap TRUE to enable automatic reaping, FALSE to disable */ void crm_set_autoreap(gboolean autoreap) { crm_autoreap = autoreap; } static void dump_peer_hash(int level, const char *caller) { GHashTableIter iter; const char *id = NULL; crm_node_t *node = NULL; g_hash_table_iter_init(&iter, crm_peer_cache); while (g_hash_table_iter_next(&iter, (gpointer *) &id, (gpointer *) &node)) { do_crm_log(level, "%s: Node %u/%s = %p - %s", caller, node->id, node->uname, node, id); } } static gboolean hash_find_by_data(gpointer key, gpointer value, gpointer user_data) { return value == user_data; } /*! * \internal * \brief Search caches for a node (cluster or Pacemaker Remote) * * \param[in] id If not 0, cluster node ID to search for * \param[in] uname If not NULL, node name to search for * \param[in] flags Group of enum pcmk__node_search_flags * * \return Node cache entry if found, otherwise NULL */ crm_node_t * pcmk__search_node_caches(unsigned int id, const char *uname, uint32_t flags) { crm_node_t *node = NULL; CRM_ASSERT(id > 0 || uname != NULL); crm_peer_init(); if ((uname != NULL) && pcmk_is_set(flags, pcmk__node_search_remote)) { node = g_hash_table_lookup(crm_remote_peer_cache, uname); } if ((node == NULL) && pcmk_is_set(flags, pcmk__node_search_cluster)) { node = pcmk__search_cluster_node_cache(id, uname, NULL); } if ((node == NULL) && pcmk_is_set(flags, pcmk__node_search_known)) { char *id_str = (id == 0)? NULL : crm_strdup_printf("%u", id); node = find_known_node(id_str, uname); free(id_str); } return node; } /*! * \internal * \brief Purge a node from cache (both cluster and Pacemaker Remote) * * \param[in] node_name If not NULL, purge only nodes with this name * \param[in] node_id If not 0, purge cluster nodes only if they have this ID * * \note If \p node_name is NULL and \p node_id is 0, no nodes will be purged. * If \p node_name is not NULL and \p node_id is not 0, Pacemaker Remote * nodes that match \p node_name will be purged, and cluster nodes that * match both \p node_name and \p node_id will be purged. * \note The caller must be careful not to use \p node_name after calling this * function if it might be a pointer into a cache entry being removed. */ void pcmk__purge_node_from_cache(const char *node_name, uint32_t node_id) { char *node_name_copy = NULL; if ((node_name == NULL) && (node_id == 0U)) { return; } // Purge from Pacemaker Remote node cache if ((node_name != NULL) && (g_hash_table_lookup(crm_remote_peer_cache, node_name) != NULL)) { /* node_name could be a pointer into the cache entry being purged, * so reassign it to a copy before the original gets freed */ node_name_copy = pcmk__str_copy(node_name); node_name = node_name_copy; crm_trace("Purging %s from Pacemaker Remote node cache", node_name); g_hash_table_remove(crm_remote_peer_cache, node_name); } reap_crm_member(node_id, node_name); free(node_name_copy); } /*! * \internal * \brief Search cluster node cache * * \param[in] id If not 0, cluster node ID to search for * \param[in] uname If not NULL, node name to search for * \param[in] uuid If not NULL while id is 0, node UUID instead of cluster * node ID to search for * * \return Cluster node cache entry if found, otherwise NULL */ crm_node_t * pcmk__search_cluster_node_cache(unsigned int id, const char *uname, const char *uuid) { GHashTableIter iter; crm_node_t *node = NULL; crm_node_t *by_id = NULL; crm_node_t *by_name = NULL; CRM_ASSERT(id > 0 || uname != NULL); crm_peer_init(); if (uname != NULL) { g_hash_table_iter_init(&iter, crm_peer_cache); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) { if(node->uname && strcasecmp(node->uname, uname) == 0) { crm_trace("Name match: %s = %p", node->uname, node); by_name = node; break; } } } if (id > 0) { g_hash_table_iter_init(&iter, crm_peer_cache); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) { if(node->id == id) { crm_trace("ID match: %u = %p", node->id, node); by_id = node; break; } } } else if (uuid != NULL) { g_hash_table_iter_init(&iter, crm_peer_cache); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) { if (pcmk__str_eq(node->uuid, uuid, pcmk__str_casei)) { crm_trace("UUID match: %s = %p", node->uuid, node); by_id = node; break; } } } node = by_id; /* Good default */ if(by_id == by_name) { /* Nothing to do if they match (both NULL counts) */ crm_trace("Consistent: %p for %u/%s", by_id, id, uname); } else if(by_id == NULL && by_name) { crm_trace("Only one: %p for %u/%s", by_name, id, uname); if(id && by_name->id) { dump_peer_hash(LOG_WARNING, __func__); crm_crit("Node %u and %u share the same name '%s'", id, by_name->id, uname); node = NULL; /* Create a new one */ } else { node = by_name; } } else if(by_name == NULL && by_id) { crm_trace("Only one: %p for %u/%s", by_id, id, uname); if(uname && by_id->uname) { dump_peer_hash(LOG_WARNING, __func__); crm_crit("Node '%s' and '%s' share the same cluster nodeid %u: assuming '%s' is correct", uname, by_id->uname, id, uname); } } else if(uname && by_id->uname) { if(pcmk__str_eq(uname, by_id->uname, pcmk__str_casei)) { crm_notice("Node '%s' has changed its ID from %u to %u", by_id->uname, by_name->id, by_id->id); g_hash_table_foreach_remove(crm_peer_cache, hash_find_by_data, by_name); } else { crm_warn("Node '%s' and '%s' share the same cluster nodeid: %u %s", by_id->uname, by_name->uname, id, uname); dump_peer_hash(LOG_INFO, __func__); crm_abort(__FILE__, __func__, __LINE__, "member weirdness", TRUE, TRUE); } } else if(id && by_name->id) { crm_warn("Node %u and %u share the same name: '%s'", by_id->id, by_name->id, uname); } else { /* Simple merge */ /* Only corosync-based clusters use node IDs. The functions that call * pcmk__update_peer_state() and crm_update_peer_proc() only know * nodeid, so 'by_id' is authoritative when merging. */ dump_peer_hash(LOG_DEBUG, __func__); crm_info("Merging %p into %p", by_name, by_id); g_hash_table_foreach_remove(crm_peer_cache, hash_find_by_data, by_name); } return node; } #if SUPPORT_COROSYNC static guint remove_conflicting_peer(crm_node_t *node) { int matches = 0; GHashTableIter iter; crm_node_t *existing_node = NULL; if (node->id == 0 || node->uname == NULL) { return 0; } if (!pcmk__corosync_has_nodelist()) { return 0; } g_hash_table_iter_init(&iter, crm_peer_cache); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &existing_node)) { if (existing_node->id > 0 && existing_node->id != node->id && existing_node->uname != NULL && strcasecmp(existing_node->uname, node->uname) == 0) { if (pcmk__cluster_is_node_active(existing_node)) { continue; } crm_warn("Removing cached offline node %u/%s which has conflicting uname with %u", existing_node->id, existing_node->uname, node->id); g_hash_table_iter_remove(&iter); matches++; } } return matches; } #endif /*! * \brief Get a cluster node cache entry * * \param[in] id If not 0, cluster node ID to search for * \param[in] uname If not NULL, node name to search for * \param[in] uuid If not NULL while id is 0, node UUID instead of cluster * node ID to search for * \param[in] flags Group of enum pcmk__node_search_flags * * \return (Possibly newly created) cluster node cache entry */ /* coverity[-alloc] Memory is referenced in one or both hashtables */ crm_node_t * pcmk__get_node(unsigned int id, const char *uname, const char *uuid, uint32_t flags) { crm_node_t *node = NULL; char *uname_lookup = NULL; CRM_ASSERT(id > 0 || uname != NULL); crm_peer_init(); // Check the Pacemaker Remote node cache first if (pcmk_is_set(flags, pcmk__node_search_remote)) { node = g_hash_table_lookup(crm_remote_peer_cache, uname); if (node != NULL) { return node; } } if (!pcmk_is_set(flags, pcmk__node_search_cluster)) { return NULL; } node = pcmk__search_cluster_node_cache(id, uname, uuid); /* if uname wasn't provided, and find_peer did not turn up a uname based on id. * we need to do a lookup of the node name using the id in the cluster membership. */ if ((node == NULL || node->uname == NULL) && (uname == NULL)) { uname_lookup = get_node_name(id); } if (uname_lookup) { uname = uname_lookup; crm_trace("Inferred a name of '%s' for node %u", uname, id); /* try to turn up the node one more time now that we know the uname. */ if (node == NULL) { node = pcmk__search_cluster_node_cache(id, uname, uuid); } } if (node == NULL) { char *uniqueid = crm_generate_uuid(); node = pcmk__assert_alloc(1, sizeof(crm_node_t)); crm_info("Created entry %s/%p for node %s/%u (%d total)", uniqueid, node, uname, id, 1 + g_hash_table_size(crm_peer_cache)); g_hash_table_replace(crm_peer_cache, uniqueid, node); } if(id > 0 && uname && (node->id == 0 || node->uname == NULL)) { crm_info("Node %u is now known as %s", id, uname); } if(id > 0 && node->id == 0) { node->id = id; } if (uname && (node->uname == NULL)) { update_peer_uname(node, uname); } if(node->uuid == NULL) { if (uuid == NULL) { uuid = crm_peer_uuid(node); } if (uuid) { crm_info("Node %u has uuid %s", id, uuid); } else { crm_info("Cannot obtain a UUID for node %u/%s", id, node->uname); } } free(uname_lookup); return node; } /*! * \internal * \brief Update a node's uname * * \param[in,out] node Node object to update * \param[in] uname New name to set * * \note This function should not be called within a peer cache iteration, * because in some cases it can remove conflicting cache entries, * which would invalidate the iterator. */ static void update_peer_uname(crm_node_t *node, const char *uname) { CRM_CHECK(uname != NULL, crm_err("Bug: can't update node name without name"); return); CRM_CHECK(node != NULL, crm_err("Bug: can't update node name to %s without node", uname); return); if (pcmk__str_eq(uname, node->uname, pcmk__str_casei)) { crm_debug("Node uname '%s' did not change", uname); return; } for (const char *c = uname; *c; ++c) { if ((*c >= 'A') && (*c <= 'Z')) { crm_warn("Node names with capitals are discouraged, consider changing '%s'", uname); break; } } pcmk__str_update(&node->uname, uname); if (peer_status_callback != NULL) { peer_status_callback(crm_status_uname, node, NULL); } #if SUPPORT_COROSYNC if (is_corosync_cluster() && !pcmk_is_set(node->flags, crm_remote_node)) { remove_conflicting_peer(node); } #endif } /*! * \internal * \brief Get log-friendly string equivalent of a process flag * * \param[in] proc Process flag * * \return Log-friendly string equivalent of \p proc */ static inline const char * proc2text(enum crm_proc_flag proc) { const char *text = "unknown"; switch (proc) { case crm_proc_none: text = "none"; break; case crm_proc_based: text = "pacemaker-based"; break; case crm_proc_controld: text = "pacemaker-controld"; break; case crm_proc_schedulerd: text = "pacemaker-schedulerd"; break; case crm_proc_execd: text = "pacemaker-execd"; break; case crm_proc_attrd: text = "pacemaker-attrd"; break; case crm_proc_fenced: text = "pacemaker-fenced"; break; case crm_proc_cpg: text = "corosync-cpg"; break; } return text; } /*! * \internal * \brief Update a node's process information (and potentially state) * * \param[in] source Caller's function name (for log messages) * \param[in,out] node Node object to update * \param[in] flag Bitmask of new process information * \param[in] status node status (online, offline, etc.) * * \return NULL if any node was reaped from peer caches, value of node otherwise * * \note If this function returns NULL, the supplied node object was likely * freed and should not be used again. This function should not be * called within a cache iteration if reaping is possible, otherwise * reaping could invalidate the iterator. */ crm_node_t * crm_update_peer_proc(const char *source, crm_node_t * node, uint32_t flag, const char *status) { uint32_t last = 0; gboolean changed = FALSE; CRM_CHECK(node != NULL, crm_err("%s: Could not set %s to %s for NULL", source, proc2text(flag), status); return NULL); /* Pacemaker doesn't spawn processes on remote nodes */ if (pcmk_is_set(node->flags, crm_remote_node)) { return node; } last = node->processes; if (status == NULL) { node->processes = flag; if (node->processes != last) { changed = TRUE; } } else if (pcmk__str_eq(status, PCMK_VALUE_ONLINE, pcmk__str_casei)) { if ((node->processes & flag) != flag) { node->processes = pcmk__set_flags_as(__func__, __LINE__, LOG_TRACE, "Peer process", node->uname, node->processes, flag, "processes"); changed = TRUE; } } else if (node->processes & flag) { node->processes = pcmk__clear_flags_as(__func__, __LINE__, LOG_TRACE, "Peer process", node->uname, node->processes, flag, "processes"); changed = TRUE; } if (changed) { if (status == NULL && flag <= crm_proc_none) { crm_info("%s: Node %s[%u] - all processes are now offline", source, node->uname, node->id); } else { crm_info("%s: Node %s[%u] - %s is now %s", source, node->uname, node->id, proc2text(flag), status); } if (pcmk_is_set(node->processes, crm_get_cluster_proc())) { node->when_online = time(NULL); } else { node->when_online = 0; } /* Call the client callback first, then update the peer state, * in case the node will be reaped */ if (peer_status_callback != NULL) { peer_status_callback(crm_status_processes, node, &last); } /* The client callback shouldn't touch the peer caches, * but as a safety net, bail if the peer cache was destroyed. */ if (crm_peer_cache == NULL) { return NULL; } if (crm_autoreap) { const char *peer_state = NULL; if (pcmk_is_set(node->processes, crm_get_cluster_proc())) { peer_state = CRM_NODE_MEMBER; } else { peer_state = CRM_NODE_LOST; } node = pcmk__update_peer_state(__func__, node, peer_state, 0); } } else { crm_trace("%s: Node %s[%u] - %s is unchanged (%s)", source, node->uname, node->id, proc2text(flag), status); } return node; } /*! * \internal * \brief Update a cluster node cache entry's expected join state * * \param[in] source Caller's function name (for logging) * \param[in,out] node Node to update * \param[in] expected Node's new join state */ void pcmk__update_peer_expected(const char *source, crm_node_t *node, const char *expected) { char *last = NULL; gboolean changed = FALSE; CRM_CHECK(node != NULL, crm_err("%s: Could not set 'expected' to %s", source, expected); return); /* Remote nodes don't participate in joins */ if (pcmk_is_set(node->flags, crm_remote_node)) { return; } last = node->expected; if (expected != NULL && !pcmk__str_eq(node->expected, expected, pcmk__str_casei)) { node->expected = strdup(expected); changed = TRUE; } if (changed) { crm_info("%s: Node %s[%u] - expected state is now %s (was %s)", source, node->uname, node->id, expected, last); free(last); } else { crm_trace("%s: Node %s[%u] - expected state is unchanged (%s)", source, node->uname, node->id, expected); } } /*! * \internal * \brief Update a node's state and membership information * * \param[in] source Caller's function name (for log messages) * \param[in,out] node Node object to update * \param[in] state Node's new state * \param[in] membership Node's new membership ID * \param[in,out] iter If not NULL, pointer to node's peer cache iterator * * \return NULL if any node was reaped, value of node otherwise * * \note If this function returns NULL, the supplied node object was likely * freed and should not be used again. This function may be called from * within a peer cache iteration if the iterator is supplied. */ static crm_node_t * update_peer_state_iter(const char *source, crm_node_t *node, const char *state, uint64_t membership, GHashTableIter *iter) { gboolean is_member; CRM_CHECK(node != NULL, crm_err("Could not set state for unknown host to %s" CRM_XS " source=%s", state, source); return NULL); is_member = pcmk__str_eq(state, CRM_NODE_MEMBER, pcmk__str_casei); if (is_member) { node->when_lost = 0; if (membership) { node->last_seen = membership; } } if (state && !pcmk__str_eq(node->state, state, pcmk__str_casei)) { char *last = node->state; if (is_member) { node->when_member = time(NULL); } else { node->when_member = 0; } node->state = strdup(state); crm_notice("Node %s state is now %s " CRM_XS " nodeid=%u previous=%s source=%s", node->uname, state, node->id, (last? last : "unknown"), source); if (peer_status_callback != NULL) { peer_status_callback(crm_status_nstate, node, last); } free(last); if (crm_autoreap && !is_member && !pcmk_is_set(node->flags, crm_remote_node)) { /* We only autoreap from the peer cache, not the remote peer cache, * because the latter should be managed only by * refresh_remote_nodes(). */ if(iter) { crm_notice("Purged 1 peer with " PCMK_XA_ID "=%u and/or uname=%s from the membership cache", node->id, node->uname); g_hash_table_iter_remove(iter); } else { reap_crm_member(node->id, node->uname); } node = NULL; } } else { crm_trace("Node %s state is unchanged (%s) " CRM_XS " nodeid=%u source=%s", node->uname, state, node->id, source); } return node; } /*! * \brief Update a node's state and membership information * * \param[in] source Caller's function name (for log messages) * \param[in,out] node Node object to update * \param[in] state Node's new state * \param[in] membership Node's new membership ID * * \return NULL if any node was reaped, value of node otherwise * * \note If this function returns NULL, the supplied node object was likely * freed and should not be used again. This function should not be * called within a cache iteration if reaping is possible, * otherwise reaping could invalidate the iterator. */ crm_node_t * pcmk__update_peer_state(const char *source, crm_node_t *node, const char *state, uint64_t membership) { return update_peer_state_iter(source, node, state, membership, NULL); } /*! * \internal * \brief Reap all nodes from cache whose membership information does not match * * \param[in] membership Membership ID of nodes to keep */ void pcmk__reap_unseen_nodes(uint64_t membership) { GHashTableIter iter; crm_node_t *node = NULL; crm_trace("Reaping unseen nodes..."); g_hash_table_iter_init(&iter, crm_peer_cache); while (g_hash_table_iter_next(&iter, NULL, (gpointer *)&node)) { if (node->last_seen != membership) { if (node->state) { /* * Calling update_peer_state_iter() allows us to * remove the node from crm_peer_cache without * invalidating our iterator */ update_peer_state_iter(__func__, node, CRM_NODE_LOST, membership, &iter); } else { crm_info("State of node %s[%u] is still unknown", node->uname, node->id); } } } } static crm_node_t * find_known_node(const char *id, const char *uname) { GHashTableIter iter; crm_node_t *node = NULL; crm_node_t *by_id = NULL; crm_node_t *by_name = NULL; if (uname) { g_hash_table_iter_init(&iter, known_node_cache); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) { if (node->uname && strcasecmp(node->uname, uname) == 0) { crm_trace("Name match: %s = %p", node->uname, node); by_name = node; break; } } } if (id) { g_hash_table_iter_init(&iter, known_node_cache); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) { if(strcasecmp(node->uuid, id) == 0) { crm_trace("ID match: %s= %p", id, node); by_id = node; break; } } } node = by_id; /* Good default */ if (by_id == by_name) { /* Nothing to do if they match (both NULL counts) */ crm_trace("Consistent: %p for %s/%s", by_id, id, uname); } else if (by_id == NULL && by_name) { crm_trace("Only one: %p for %s/%s", by_name, id, uname); if (id) { node = NULL; } else { node = by_name; } } else if (by_name == NULL && by_id) { crm_trace("Only one: %p for %s/%s", by_id, id, uname); if (uname) { node = NULL; } } else if (uname && by_id->uname && pcmk__str_eq(uname, by_id->uname, pcmk__str_casei)) { /* Multiple nodes have the same uname in the CIB. * Return by_id. */ } else if (id && by_name->uuid && pcmk__str_eq(id, by_name->uuid, pcmk__str_casei)) { /* Multiple nodes have the same id in the CIB. * Return by_name. */ node = by_name; } else { node = NULL; } if (node == NULL) { crm_debug("Couldn't find node%s%s%s%s", id? " " : "", id? id : "", uname? " with name " : "", uname? uname : ""); } return node; } static void known_node_cache_refresh_helper(xmlNode *xml_node, void *user_data) { const char *id = crm_element_value(xml_node, PCMK_XA_ID); const char *uname = crm_element_value(xml_node, PCMK_XA_UNAME); crm_node_t * node = NULL; CRM_CHECK(id != NULL && uname !=NULL, return); node = find_known_node(id, uname); if (node == NULL) { char *uniqueid = crm_generate_uuid(); node = pcmk__assert_alloc(1, sizeof(crm_node_t)); node->uname = pcmk__str_copy(uname); node->uuid = pcmk__str_copy(id); g_hash_table_replace(known_node_cache, uniqueid, node); } else if (pcmk_is_set(node->flags, crm_node_dirty)) { pcmk__str_update(&node->uname, uname); /* Node is in cache and hasn't been updated already, so mark it clean */ clear_peer_flags(node, crm_node_dirty); } } static void refresh_known_node_cache(xmlNode *cib) { crm_peer_init(); g_hash_table_foreach(known_node_cache, mark_dirty, NULL); crm_foreach_xpath_result(cib, PCMK__XP_MEMBER_NODE_CONFIG, known_node_cache_refresh_helper, NULL); /* Remove all old cache entries that weren't seen in the CIB */ g_hash_table_foreach_remove(known_node_cache, is_dirty, NULL); } void pcmk__refresh_node_caches_from_cib(xmlNode *cib) { refresh_remote_nodes(cib); refresh_known_node_cache(cib); } // Deprecated functions kept only for backward API compatibility // LCOV_EXCL_START #include int crm_terminate_member(int nodeid, const char *uname, void *unused) { return stonith_api_kick(nodeid, uname, 120, TRUE); } int crm_terminate_member_no_mainloop(int nodeid, const char *uname, int *connection) { return stonith_api_kick(nodeid, uname, 120, TRUE); } crm_node_t * crm_get_peer(unsigned int id, const char *uname) { return pcmk__get_node(id, uname, NULL, pcmk__node_search_cluster); } crm_node_t * crm_get_peer_full(unsigned int id, const char *uname, int flags) { return pcmk__get_node(id, uname, NULL, flags); } int crm_remote_peer_cache_size(void) { unsigned int count = pcmk__cluster_num_remote_nodes(); return QB_MIN(count, INT_MAX); } void crm_remote_peer_cache_refresh(xmlNode *cib) { refresh_remote_nodes(cib); } crm_node_t * crm_remote_peer_get(const char *node_name) { return pcmk__cluster_lookup_remote_node(node_name); } void crm_remote_peer_cache_remove(const char *node_name) { pcmk__cluster_forget_remote_node(node_name); } gboolean crm_is_peer_active(const crm_node_t * node) { return pcmk__cluster_is_node_active(node); } // LCOV_EXCL_STOP // End deprecated API