crm_node_t *crm_update_ccm_node( const oc_ev_membership_t *oc, int offset, const char *state, uint64_t seq) { crm_node_t *node = NULL; const char *uuid = NULL; CRM_CHECK(oc->m_array[offset].node_uname != NULL, return NULL); uuid = get_uuid(oc->m_array[offset].node_uname); node = crm_update_peer(oc->m_array[offset].node_id, oc->m_array[offset].node_born_on, seq, -1, 0, uuid, oc->m_array[offset].node_uname, NULL, state); if(safe_str_eq(CRM_NODE_ACTIVE, state)) { /* Heartbeat doesn't send status notifications for nodes that were already part of the cluster */ crm_update_peer_proc( oc->m_array[offset].node_uname, crm_proc_ais, ONLINESTATUS); /* Nor does it send status notifications for processes that were already active */ crm_update_peer_proc( oc->m_array[offset].node_uname, crm_proc_crmd, ONLINESTATUS); } return node; }
void send_stonith_update(crm_action_t * action, const char *target, const char *uuid) { int rc = pcmk_ok; crm_node_t *peer = NULL; /* zero out the node-status & remove all LRM status info */ xmlNode *node_state = NULL; CRM_CHECK(target != NULL, return); CRM_CHECK(uuid != NULL, return); if (get_node_uuid(0, target) == NULL) { set_node_uuid(target, uuid); } /* Make sure the membership and join caches are accurate */ peer = crm_get_peer(0, target); if (peer->uuid == NULL) { crm_info("Recording uuid '%s' for node '%s'", uuid, target); peer->uuid = strdup(uuid); } crm_update_peer_proc(__FUNCTION__, peer, crm_proc_none, NULL); crm_update_peer_state(__FUNCTION__, peer, CRM_NODE_LOST, 0); crm_update_peer_expected(__FUNCTION__, peer, CRMD_JOINSTATE_DOWN); erase_node_from_join(target); node_state = do_update_node_cib(peer, node_update_cluster | node_update_peer | node_update_join | node_update_expected, NULL, __FUNCTION__); /* Force our known ID */ crm_xml_add(node_state, XML_ATTR_UUID, uuid); rc = fsa_cib_conn->cmds->update(fsa_cib_conn, XML_CIB_TAG_STATUS, node_state, cib_quorum_override | cib_scope_local | cib_can_create); /* Delay processing the trigger until the update completes */ crm_debug("Sending fencing update %d for %s", rc, target); fsa_register_cib_callback(rc, FALSE, strdup(target), cib_fencing_updated); /* Make sure it sticks */ /* fsa_cib_conn->cmds->bump_epoch(fsa_cib_conn, cib_quorum_override|cib_scope_local); */ erase_status_tag(target, XML_CIB_TAG_LRM, cib_scope_local); erase_status_tag(target, XML_TAG_TRANSIENT_NODEATTRS, cib_scope_local); free_xml(node_state); return; }
gboolean cluster_connect_cpg(crm_cluster_t *cluster) { int rc = -1; int fd = 0; int retries = 0; uint32_t id = 0; crm_node_t *peer = NULL; cpg_handle_t handle = 0; struct mainloop_fd_callbacks cpg_fd_callbacks = { .dispatch = pcmk_cpg_dispatch, .destroy = cluster->destroy, }; cpg_callbacks_t cpg_callbacks = { .cpg_deliver_fn = cluster->cpg.cpg_deliver_fn, .cpg_confchg_fn = cluster->cpg.cpg_confchg_fn, /* .cpg_deliver_fn = pcmk_cpg_deliver, */ /* .cpg_confchg_fn = pcmk_cpg_membership, */ }; cpg_evicted = FALSE; cluster->group.length = 0; cluster->group.value[0] = 0; /* group.value is char[128] */ strncpy(cluster->group.value, crm_system_name?crm_system_name:"unknown", 127); cluster->group.value[127] = 0; cluster->group.length = 1 + QB_MIN(127, strlen(cluster->group.value)); cs_repeat(retries, 30, rc = cpg_initialize(&handle, &cpg_callbacks)); if (rc != CS_OK) { crm_err("Could not connect to the Cluster Process Group API: %d", rc); goto bail; } id = get_local_nodeid(handle); if (id == 0) { crm_err("Could not get local node id from the CPG API"); goto bail; } cluster->nodeid = id; retries = 0; cs_repeat(retries, 30, rc = cpg_join(handle, &cluster->group)); if (rc != CS_OK) { crm_err("Could not join the CPG group '%s': %d", crm_system_name, rc); goto bail; } rc = cpg_fd_get(handle, &fd); if (rc != CS_OK) { crm_err("Could not obtain the CPG API connection: %d", rc); goto bail; } pcmk_cpg_handle = handle; cluster->cpg_handle = handle; mainloop_add_fd("corosync-cpg", G_PRIORITY_MEDIUM, fd, cluster, &cpg_fd_callbacks); bail: if (rc != CS_OK) { cpg_finalize(handle); return FALSE; } peer = crm_get_peer(id, NULL); crm_update_peer_proc(__FUNCTION__, peer, crm_proc_cpg, ONLINESTATUS); return TRUE; } gboolean send_cluster_message_cs(xmlNode * msg, gboolean local, crm_node_t * node, enum crm_ais_msg_types dest) { gboolean rc = TRUE; char *data = NULL; data = dump_xml_unformatted(msg); rc = send_cluster_text(crm_class_cluster, data, local, node, dest); free(data); return rc; } gboolean send_cluster_text(int class, const char *data, gboolean local, crm_node_t * node, enum crm_ais_msg_types dest) { static int msg_id = 0; static int local_pid = 0; static int local_name_len = 0; static const char *local_name = NULL; char *target = NULL; struct iovec *iov; AIS_Message *msg = NULL; enum crm_ais_msg_types sender = text2msg_type(crm_system_name); /* There are only 6 handlers registered to crm_lib_service in plugin.c */ CRM_CHECK(class < 6, crm_err("Invalid message class: %d", class); return FALSE); #if !SUPPORT_PLUGIN CRM_CHECK(dest != crm_msg_ais, return FALSE); #endif if(local_name == NULL) { local_name = get_local_node_name(); } if(local_name_len == 0 && local_name) { local_name_len = strlen(local_name); } if (data == NULL) { data = ""; } if (local_pid == 0) { local_pid = getpid(); } if (sender == crm_msg_none) { sender = local_pid; } msg = calloc(1, sizeof(AIS_Message)); msg_id++; msg->id = msg_id; msg->header.id = class; msg->header.error = CS_OK; msg->host.type = dest; msg->host.local = local; if (node) { if (node->uname) { target = strdup(node->uname); msg->host.size = strlen(node->uname); memset(msg->host.uname, 0, MAX_NAME); memcpy(msg->host.uname, node->uname, msg->host.size); } else { target = crm_strdup_printf("%u", node->id); } msg->host.id = node->id; } else { target = strdup("all"); } msg->sender.id = 0; msg->sender.type = sender; msg->sender.pid = local_pid; msg->sender.size = local_name_len; memset(msg->sender.uname, 0, MAX_NAME); if(local_name && msg->sender.size) { memcpy(msg->sender.uname, local_name, msg->sender.size); } msg->size = 1 + strlen(data); msg->header.size = sizeof(AIS_Message) + msg->size; if (msg->size < CRM_BZ2_THRESHOLD) { msg = realloc_safe(msg, msg->header.size); memcpy(msg->data, data, msg->size); } else { char *compressed = NULL; unsigned int new_size = 0; char *uncompressed = strdup(data); if (crm_compress_string(uncompressed, msg->size, 0, &compressed, &new_size)) { msg->header.size = sizeof(AIS_Message) + new_size; msg = realloc_safe(msg, msg->header.size); memcpy(msg->data, compressed, new_size); msg->is_compressed = TRUE; msg->compressed_size = new_size; } else { msg = realloc_safe(msg, msg->header.size); memcpy(msg->data, data, msg->size); } free(uncompressed); free(compressed); } iov = calloc(1, sizeof(struct iovec)); iov->iov_base = msg; iov->iov_len = msg->header.size; if (msg->compressed_size) { crm_trace("Queueing CPG message %u to %s (%llu bytes, %d bytes compressed payload): %.200s", msg->id, target, (unsigned long long) iov->iov_len, msg->compressed_size, data); } else { crm_trace("Queueing CPG message %u to %s (%llu bytes, %d bytes payload): %.200s", msg->id, target, (unsigned long long) iov->iov_len, msg->size, data); } free(target); #if SUPPORT_PLUGIN /* The plugin is the only time we dont use CPG messaging */ if(get_cluster_type() == pcmk_cluster_classic_ais) { return send_plugin_text(class, iov); } #endif send_cpg_iov(iov); return TRUE; } enum crm_ais_msg_types text2msg_type(const char *text) { int type = crm_msg_none; CRM_CHECK(text != NULL, return type); if (safe_str_eq(text, "ais")) { type = crm_msg_ais; } else if (safe_str_eq(text, "crm_plugin")) { type = crm_msg_ais; } else if (safe_str_eq(text, CRM_SYSTEM_CIB)) { type = crm_msg_cib; } else if (safe_str_eq(text, CRM_SYSTEM_CRMD)) { type = crm_msg_crmd; } else if (safe_str_eq(text, CRM_SYSTEM_DC)) { type = crm_msg_crmd; } else if (safe_str_eq(text, CRM_SYSTEM_TENGINE)) { type = crm_msg_te; } else if (safe_str_eq(text, CRM_SYSTEM_PENGINE)) { type = crm_msg_pe; } else if (safe_str_eq(text, CRM_SYSTEM_LRMD)) { type = crm_msg_lrmd; } else if (safe_str_eq(text, CRM_SYSTEM_STONITHD)) { type = crm_msg_stonithd; } else if (safe_str_eq(text, "stonith-ng")) { type = crm_msg_stonith_ng; } else if (safe_str_eq(text, "attrd")) { type = crm_msg_attrd; } else { /* This will normally be a transient client rather than * a cluster daemon. Set the type to the pid of the client */ int scan_rc = sscanf(text, "%d", &type); if (scan_rc != 1 || type <= crm_msg_stonith_ng) { /* Ensure it's sane */ type = crm_msg_none; } } return type; }
void pcmk_cpg_membership(cpg_handle_t handle, const struct cpg_name *groupName, const struct cpg_address *member_list, size_t member_list_entries, const struct cpg_address *left_list, size_t left_list_entries, const struct cpg_address *joined_list, size_t joined_list_entries) { int i; gboolean found = FALSE; static int counter = 0; uint32_t local_nodeid = get_local_nodeid(handle); for (i = 0; i < left_list_entries; i++) { crm_node_t *peer = crm_find_peer(left_list[i].nodeid, NULL); crm_info("Node %u left group %s (peer=%s, counter=%d.%d)", left_list[i].nodeid, groupName->value, (peer? peer->uname : "<none>"), counter, i); if (peer) { crm_update_peer_proc(__FUNCTION__, peer, crm_proc_cpg, OFFLINESTATUS); } } for (i = 0; i < joined_list_entries; i++) { crm_info("Node %u joined group %s (counter=%d.%d)", joined_list[i].nodeid, groupName->value, counter, i); } for (i = 0; i < member_list_entries; i++) { crm_node_t *peer = crm_get_peer(member_list[i].nodeid, NULL); crm_info("Node %u still member of group %s (peer=%s, counter=%d.%d)", member_list[i].nodeid, groupName->value, (peer? peer->uname : "<none>"), counter, i); /* Anyone that is sending us CPG messages must also be a _CPG_ member. * But it's _not_ safe to assume it's in the quorum membership. * We may have just found out it's dead and are processing the last couple of messages it sent */ peer = crm_update_peer_proc(__FUNCTION__, peer, crm_proc_cpg, ONLINESTATUS); if(peer && peer->state && crm_is_peer_active(peer) == FALSE) { time_t now = time(NULL); /* Co-opt the otherwise unused votes field */ if(peer->votes == 0) { peer->votes = now; } else if(now > (60 + peer->votes)) { /* On the otherhand, if we're still getting messages, at a certain point * we need to acknowledge our internal cache is probably wrong * * Set the threshold to 1 minute */ crm_err("Node %s[%u] appears to be online even though we think it is dead", peer->uname, peer->id); if (crm_update_peer_state(__FUNCTION__, peer, CRM_NODE_MEMBER, 0)) { peer->votes = 0; } } } if (local_nodeid == member_list[i].nodeid) { found = TRUE; } } if (!found) { crm_err("We're not part of CPG group '%s' anymore!", groupName->value); cpg_evicted = TRUE; } counter++; }
crm_node_t * crm_update_peer(const char *source, unsigned int id, uint64_t born, uint64_t seen, int32_t votes, uint32_t children, const char *uuid, const char *uname, const char *addr, const char *state) { #if SUPPORT_PLUGIN gboolean addr_changed = FALSE; gboolean votes_changed = FALSE; #endif crm_node_t *node = NULL; id = get_corosync_id(id, uuid); node = crm_get_peer(id, uname); CRM_ASSERT(node != NULL); if (node->uuid == NULL) { if (is_openais_cluster()) { /* Yes, overrule whatever was passed in */ crm_peer_uuid(node); } else if (uuid != NULL) { node->uuid = strdup(uuid); } } if (children > 0) { crm_update_peer_proc(source, node, children, state); } if (state != NULL) { crm_update_peer_state(source, node, state, seen); } #if SUPPORT_HEARTBEAT if (born != 0) { node->born = born; } #endif #if SUPPORT_PLUGIN /* These were only used by the plugin */ if (born != 0) { node->born = born; } if (votes > 0 && node->votes != votes) { votes_changed = TRUE; node->votes = votes; } if (addr != NULL) { if (node->addr == NULL || crm_str_eq(node->addr, addr, FALSE) == FALSE) { addr_changed = TRUE; free(node->addr); node->addr = strdup(addr); } } if (addr_changed || votes_changed) { crm_info("%s: Node %s: id=%u state=%s addr=%s%s votes=%d%s born=" U64T " seen=" U64T " proc=%.32x", source, node->uname, node->id, node->state, node->addr, addr_changed ? " (new)" : "", node->votes, votes_changed ? " (new)" : "", node->born, node->last_seen, node->processes); } #endif return node; }
static void tengine_stonith_notify(stonith_t * st, stonith_event_t * st_event) { if(te_client_id == NULL) { te_client_id = g_strdup_printf("%s.%d", crm_system_name, getpid()); } if (st_event == NULL) { crm_err("Notify data not found"); return; } if (st_event->result == pcmk_ok && crm_str_eq(st_event->target, fsa_our_uname, TRUE)) { crm_crit("We were alegedly just fenced by %s for %s with %s!", st_event->executioner, st_event->origin, st_event->device); /* Dumps blackbox if enabled */ qb_log_fini(); /* Try to get the above log message to disk - somehow */ /* Get out ASAP and do not come back up. * * Triggering a reboot is also not the worst idea either since * the rest of the cluster thinks we're safely down */ #ifdef RB_HALT_SYSTEM reboot(RB_HALT_SYSTEM); #endif /* * If reboot() fails or is not supported, coming back up will * probably lead to a situation where the other nodes set our * status to 'lost' because of the fencing callback and will * discard subsequent election votes with: * * Election 87 (current: 5171, owner: 103): Processed vote from east-03 (Peer is not part of our cluster) * * So just stay dead, something is seriously messed up anyway. * */ exit(100); /* None of our wrappers since we already called qb_log_fini() */ return; } if (st_event->result == pcmk_ok && safe_str_eq(st_event->operation, T_STONITH_NOTIFY_FENCE)) { st_fail_count_reset(st_event->target); } crm_notice("Peer %s was%s terminated (%s) by %s for %s: %s (ref=%s) by client %s", st_event->target, st_event->result == pcmk_ok ? "" : " not", st_event->action, st_event->executioner ? st_event->executioner : "<anyone>", st_event->origin, pcmk_strerror(st_event->result), st_event->id, st_event->client_origin ? st_event->client_origin : "<unknown>"); #if SUPPORT_CMAN if (st_event->result == pcmk_ok && is_cman_cluster()) { int local_rc = 0; char *target_copy = strdup(st_event->target); /* In case fenced hasn't noticed yet * * Any fencing that has been inititated will be completed by way of the fence_pcmk redirect */ local_rc = fenced_external(target_copy); if (local_rc != 0) { crm_err("Could not notify CMAN that '%s' is now fenced: %d", st_event->target, local_rc); } else { crm_notice("Notified CMAN that '%s' is now fenced", st_event->target); } free(target_copy); } #endif if (st_event->result == pcmk_ok) { crm_node_t *peer = crm_get_peer_full(0, st_event->target, CRM_GET_PEER_REMOTE | CRM_GET_PEER_CLUSTER); const char *uuid = crm_peer_uuid(peer); gboolean we_are_executioner = safe_str_eq(st_event->executioner, fsa_our_uname); crm_trace("target=%s dc=%s", st_event->target, fsa_our_dc); if(AM_I_DC) { /* The DC always sends updates */ send_stonith_update(NULL, st_event->target, uuid); if (st_event->client_origin && safe_str_neq(st_event->client_origin, te_client_id)) { /* Abort the current transition graph if it wasn't us * that invoked stonith to fence someone */ crm_info("External fencing operation from %s fenced %s", st_event->client_origin, st_event->target); abort_transition(INFINITY, tg_restart, "External Fencing Operation", NULL); } /* Assume it was our leader if we dont currently have one */ } else if (fsa_our_dc == NULL || safe_str_eq(fsa_our_dc, st_event->target)) { crm_notice("Target %s our leader %s (recorded: %s)", fsa_our_dc ? "was" : "may have been", st_event->target, fsa_our_dc ? fsa_our_dc : "<unset>"); /* Given the CIB resyncing that occurs around elections, * have one node update the CIB now and, if the new DC is different, * have them do so too after the election */ if (we_are_executioner) { send_stonith_update(NULL, st_event->target, uuid); } stonith_cleanup_list = g_list_append(stonith_cleanup_list, strdup(st_event->target)); } crm_update_peer_proc(__FUNCTION__, peer, crm_proc_none, NULL); crm_update_peer_state(__FUNCTION__, peer, CRM_NODE_LOST, 0); crm_update_peer_expected(__FUNCTION__, peer, CRMD_JOINSTATE_DOWN); crm_update_peer_join(__FUNCTION__, peer, crm_join_none); } }
static void crmd_cs_dispatch(cpg_handle_t handle, const struct cpg_name *groupName, uint32_t nodeid, uint32_t pid, void *msg, size_t msg_len) { int seq = 0; xmlNode *xml = NULL; const char *seq_s = NULL; crm_node_t *peer = NULL; enum crm_proc_flag flag = crm_proc_cpg; uint32_t kind = 0; const char *from = NULL; char *data = pcmk_message_common_cs(handle, nodeid, pid, msg, &kind, &from); if(data == NULL) { return; } xml = string2xml(data); if (xml == NULL) { crm_err("Could not parse message content (%d): %.100s", kind, data); free(data); return; } switch (kind) { case crm_class_members: seq_s = crm_element_value(xml, "id"); seq = crm_int_helper(seq_s, NULL); set_bit(fsa_input_register, R_PEER_DATA); post_cache_update(seq); /* fall through */ case crm_class_quorum: crm_update_quorum(crm_have_quorum, FALSE); if (AM_I_DC) { const char *votes = crm_element_value(xml, "expected"); if (votes == NULL || check_number(votes) == FALSE) { crm_log_xml_err(xml, "Invalid quorum/membership update"); } else { int rc = update_attr_delegate(fsa_cib_conn, cib_quorum_override | cib_scope_local | cib_inhibit_notify, XML_CIB_TAG_CRMCONFIG, NULL, NULL, NULL, NULL, XML_ATTR_EXPECTED_VOTES, votes, FALSE, NULL); crm_info("Setting expected votes to %s", votes); if (pcmk_ok > rc) { crm_err("Quorum update failed: %s", pcmk_strerror(rc)); } } } break; case crm_class_cluster: crm_xml_add(xml, F_ORIG, from); /* crm_xml_add_int(xml, F_SEQ, wrapper->id); Fake? */ if (is_heartbeat_cluster()) { flag = crm_proc_heartbeat; } else if (is_classic_ais_cluster()) { flag = crm_proc_plugin; } peer = crm_get_peer(0, from); if (is_not_set(peer->processes, flag)) { /* If we can still talk to our peer process on that node, * then its also part of the corosync membership */ crm_warn("Receiving messages from a node we think is dead: %s[%d]", peer->uname, peer->id); crm_update_peer_proc(__FUNCTION__, peer, flag, ONLINESTATUS); } crmd_ha_msg_filter(xml); break; case crm_class_rmpeer: /* Ignore */ break; case crm_class_notify: case crm_class_nodeid: crm_err("Unexpected message class (%d): %.100s", kind, data); break; default: crm_err("Invalid message class (%d): %.100s", kind, data); } free(data); free_xml(xml); }