int cib_server_process_diff(const char *op, int options, const char *section, xmlNode * req, xmlNode * input, xmlNode * existing_cib, xmlNode ** result_cib, xmlNode ** answer) { int rc = pcmk_ok; if (cib_is_master) { /* the master is never waiting for a resync */ sync_in_progress = 0; } if (sync_in_progress > MAX_DIFF_RETRY) { /* request another full-sync, * the last request may have been lost */ sync_in_progress = 0; } if (sync_in_progress) { int diff_add_updates = 0; int diff_add_epoch = 0; int diff_add_admin_epoch = 0; int diff_del_updates = 0; int diff_del_epoch = 0; int diff_del_admin_epoch = 0; cib_diff_version_details(input, &diff_add_admin_epoch, &diff_add_epoch, &diff_add_updates, &diff_del_admin_epoch, &diff_del_epoch, &diff_del_updates); sync_in_progress++; crm_notice("Not applying diff %d.%d.%d -> %d.%d.%d (sync in progress)", diff_del_admin_epoch, diff_del_epoch, diff_del_updates, diff_add_admin_epoch, diff_add_epoch, diff_add_updates); return -pcmk_err_diff_resync; } rc = cib_process_diff(op, options, section, req, input, existing_cib, result_cib, answer); if (rc == -pcmk_err_diff_resync && cib_is_master == FALSE) { xmlNode *sync_me = create_xml_node(NULL, "sync-me"); free_xml(*result_cib); *result_cib = NULL; crm_info("Requesting re-sync from peer"); sync_in_progress++; crm_xml_add(sync_me, F_TYPE, "cib"); crm_xml_add(sync_me, F_CIB_OPERATION, CIB_OP_SYNC_ONE); crm_xml_add(sync_me, F_CIB_DELEGATED, cib_our_uname); if (send_cluster_message(NULL, crm_msg_cib, sync_me, FALSE) == FALSE) { rc = -ENOTCONN; } free_xml(sync_me); } else if (rc == -pcmk_err_diff_resync) { rc = -pcmk_err_diff_failed; if (options & cib_force_diff) { crm_warn("Not requesting full refresh in R/W mode"); } } return rc; }
enum crmd_fsa_input handle_request(xmlNode * stored_msg, enum crmd_fsa_cause cause) { xmlNode *msg = NULL; const char *op = crm_element_value(stored_msg, F_CRM_TASK); /* Optimize this for the DC - it has the most to do */ if (op == NULL) { crm_log_xml_err(stored_msg, "Bad message"); return I_NULL; } if (strcmp(op, CRM_OP_SHUTDOWN_REQ) == 0) { const char *from = crm_element_value(stored_msg, F_CRM_HOST_FROM); crm_node_t *node = crm_find_peer(0, from); crm_update_peer_expected(__FUNCTION__, node, CRMD_JOINSTATE_DOWN); if(AM_I_DC == FALSE) { return I_NULL; /* Done */ } } /*========== DC-Only Actions ==========*/ if (AM_I_DC) { if (strcmp(op, CRM_OP_JOIN_ANNOUNCE) == 0) { return I_NODE_JOIN; } else if (strcmp(op, CRM_OP_JOIN_REQUEST) == 0) { return I_JOIN_REQUEST; } else if (strcmp(op, CRM_OP_JOIN_CONFIRM) == 0) { return I_JOIN_RESULT; } else if (strcmp(op, CRM_OP_SHUTDOWN) == 0) { const char *host_from = crm_element_value(stored_msg, F_CRM_HOST_FROM); gboolean dc_match = safe_str_eq(host_from, fsa_our_dc); if (is_set(fsa_input_register, R_SHUTDOWN)) { crm_info("Shutting ourselves down (DC)"); return I_STOP; } else if (dc_match) { crm_err("We didn't ask to be shut down, yet our" " TE is telling us to. Better get out now!"); return I_TERMINATE; } else if (fsa_state != S_STOPPING) { crm_err("Another node is asking us to shutdown" " but we think we're ok."); return I_ELECTION; } } else if (strcmp(op, CRM_OP_SHUTDOWN_REQ) == 0) { /* a slave wants to shut down */ /* create cib fragment and add to message */ return handle_shutdown_request(stored_msg); } } /*========== common actions ==========*/ if (strcmp(op, CRM_OP_NOVOTE) == 0) { ha_msg_input_t fsa_input; fsa_input.msg = stored_msg; register_fsa_input_adv(C_HA_MESSAGE, I_NULL, &fsa_input, A_ELECTION_COUNT | A_ELECTION_CHECK, FALSE, __FUNCTION__); } else if (strcmp(op, CRM_OP_THROTTLE) == 0) { throttle_update(stored_msg); return I_NULL; } else if (strcmp(op, CRM_OP_CLEAR_FAILCOUNT) == 0) { return handle_failcount_op(stored_msg); } else if (strcmp(op, CRM_OP_VOTE) == 0) { /* count the vote and decide what to do after that */ ha_msg_input_t fsa_input; fsa_input.msg = stored_msg; register_fsa_input_adv(C_HA_MESSAGE, I_NULL, &fsa_input, A_ELECTION_COUNT | A_ELECTION_CHECK, FALSE, __FUNCTION__); /* Sometimes we _must_ go into S_ELECTION */ if (fsa_state == S_HALT) { crm_debug("Forcing an election from S_HALT"); return I_ELECTION; #if 0 } else if (AM_I_DC) { /* This is the old way of doing things but what is gained? */ return I_ELECTION; #endif } } else if (strcmp(op, CRM_OP_JOIN_OFFER) == 0) { crm_debug("Raising I_JOIN_OFFER: join-%s", crm_element_value(stored_msg, F_CRM_JOIN_ID)); return I_JOIN_OFFER; } else if (strcmp(op, CRM_OP_JOIN_ACKNAK) == 0) { crm_debug("Raising I_JOIN_RESULT: join-%s", crm_element_value(stored_msg, F_CRM_JOIN_ID)); return I_JOIN_RESULT; } else if (strcmp(op, CRM_OP_LRM_DELETE) == 0 || strcmp(op, CRM_OP_LRM_FAIL) == 0 || strcmp(op, CRM_OP_LRM_REFRESH) == 0 || strcmp(op, CRM_OP_REPROBE) == 0) { crm_xml_add(stored_msg, F_CRM_SYS_TO, CRM_SYSTEM_LRMD); return I_ROUTER; } else if (strcmp(op, CRM_OP_NOOP) == 0) { return I_NULL; } else if (strcmp(op, CRM_OP_LOCAL_SHUTDOWN) == 0) { crm_shutdown(SIGTERM); /*return I_SHUTDOWN; */ return I_NULL; /*========== (NOT_DC)-Only Actions ==========*/ } else if (AM_I_DC == FALSE && strcmp(op, CRM_OP_SHUTDOWN) == 0) { const char *host_from = crm_element_value(stored_msg, F_CRM_HOST_FROM); gboolean dc_match = safe_str_eq(host_from, fsa_our_dc); if (dc_match || fsa_our_dc == NULL) { if (is_set(fsa_input_register, R_SHUTDOWN) == FALSE) { crm_err("We didn't ask to be shut down, yet our DC is telling us to."); set_bit(fsa_input_register, R_STAYDOWN); return I_STOP; } crm_info("Shutting down"); return I_STOP; } else { crm_warn("Discarding %s op from %s", op, host_from); } } else if (strcmp(op, CRM_OP_PING) == 0) { /* eventually do some stuff to figure out * if we /are/ ok */ const char *sys_to = crm_element_value(stored_msg, F_CRM_SYS_TO); xmlNode *ping = create_xml_node(NULL, XML_CRM_TAG_PING); crm_xml_add(ping, XML_PING_ATTR_STATUS, "ok"); crm_xml_add(ping, XML_PING_ATTR_SYSFROM, sys_to); crm_xml_add(ping, "crmd_state", fsa_state2string(fsa_state)); /* Ok, so technically not so interesting, but CTS needs to see this */ crm_notice("Current ping state: %s", fsa_state2string(fsa_state)); msg = create_reply(stored_msg, ping); if (msg) { (void)relay_message(msg, TRUE); } free_xml(ping); free_xml(msg); } else if (strcmp(op, CRM_OP_RM_NODE_CACHE) == 0) { int id = 0; const char *name = NULL; crm_element_value_int(stored_msg, XML_ATTR_ID, &id); name = crm_element_value(stored_msg, XML_ATTR_UNAME); if(cause == C_IPC_MESSAGE) { msg = create_request(CRM_OP_RM_NODE_CACHE, NULL, NULL, CRM_SYSTEM_CRMD, CRM_SYSTEM_CRMD, NULL); if (send_cluster_message(NULL, crm_msg_crmd, msg, TRUE) == FALSE) { crm_err("Could not instruct peers to remove references to node %s/%u", name, id); } else { crm_notice("Instructing peers to remove references to node %s/%u", name, id); } free_xml(msg); } else { reap_crm_member(id, name); if(attrd_ipc) { int rc = 0; xmlNode *update = create_xml_node(NULL, __FUNCTION__); crm_xml_add(update, F_TYPE, T_ATTRD); crm_xml_add(update, F_ORIG, crm_system_name); crm_xml_add(update, F_ATTRD_TASK, "peer-remove"); crm_xml_add(update, F_ATTRD_HOST, name); crm_xml_add_int(update, F_ATTRD_HOST_ID, id); rc = crm_ipc_send(attrd_ipc, update, crm_ipc_client_response, 0, NULL); if (rc > 0) { rc = pcmk_ok; } crm_debug("Peer cache cleanup for %s (%d): %s (%d)", name, id, pcmk_strerror(rc), rc); free_xml(update); } } } else { crm_err("Unexpected request (%s) sent to %s", op, AM_I_DC ? "the DC" : "non-DC node"); crm_log_xml_err(stored_msg, "Unexpected"); } return I_NULL; }
static void pcmk_quorum_notification(quorum_handle_t handle, uint32_t quorate, uint64_t ring_id, uint32_t view_list_entries, uint32_t * view_list) { int i; GHashTableIter iter; crm_node_t *node = NULL; static gboolean init_phase = TRUE; if (quorate != crm_have_quorum) { if (quorate) { crm_notice("Quorum acquired " CRM_XS " membership=%" U64T " members=%lu", ring_id, (long unsigned int)view_list_entries); } else { crm_warn("Quorum lost " CRM_XS " membership=%" U64T " members=%lu", ring_id, (long unsigned int)view_list_entries); } crm_have_quorum = quorate; } else { crm_info("Quorum %s " CRM_XS " membership=%" U64T " members=%lu", (quorate? "retained" : "still lost"), ring_id, (long unsigned int)view_list_entries); } if (view_list_entries == 0 && init_phase) { crm_info("Corosync membership is still forming, ignoring"); return; } init_phase = FALSE; /* Reset last_seen for all cached nodes so we can tell which ones aren't * in the view list */ g_hash_table_iter_init(&iter, crm_peer_cache); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) { node->last_seen = 0; } /* Update the peer cache for each node in view list */ for (i = 0; i < view_list_entries; i++) { uint32_t id = view_list[i]; crm_debug("Member[%d] %u ", i, id); /* Get this node's peer cache entry (adding one if not already there) */ node = crm_get_peer(id, NULL); if (node->uname == NULL) { char *name = corosync_node_name(0, id); crm_info("Obtaining name for new node %u", id); node = crm_get_peer(id, name); free(name); } /* Update the node state (including updating last_seen to ring_id) */ crm_update_peer_state(__FUNCTION__, node, CRM_NODE_MEMBER, ring_id); } /* Remove any peer cache entries we didn't update */ crm_reap_unseen_nodes(ring_id); if (quorum_app_callback) { quorum_app_callback(ring_id, quorate); } }
void attrd_peer_message(crm_node_t *peer, xmlNode *xml) { int peer_state = 0; const char *v = crm_element_value(xml, F_ATTRD_VERSION); const char *op = crm_element_value(xml, F_ATTRD_TASK); const char *election_op = crm_element_value(xml, F_CRM_TASK); const char *host = crm_element_value(xml, F_ATTRD_HOST); if(election_op) { enum election_result rc = 0; crm_xml_add(xml, F_CRM_HOST_FROM, peer->uname); rc = election_count_vote(writer, xml, TRUE); switch(rc) { case election_start: free(peer_writer); peer_writer = NULL; election_vote(writer); break; case election_lost: free(peer_writer); peer_writer = strdup(peer->uname); break; default: election_check(writer); break; } return; } else if(v == NULL) { /* From the non-atomic version */ if (safe_str_eq(op, ATTRD_OP_UPDATE)) { const char *name = crm_element_value(xml, F_ATTRD_ATTRIBUTE); crm_trace("Compatibility update of %s from %s", name, peer->uname); attrd_peer_update(peer, xml, host, FALSE); } else if (safe_str_eq(op, ATTRD_OP_FLUSH)) { const char *name = crm_element_value(xml, F_ATTRD_ATTRIBUTE); attribute_t *a = g_hash_table_lookup(attributes, name); if(a) { crm_trace("Compatibility write-out of %s for %s from %s", a->id, op, peer->uname); write_or_elect_attribute(a); } } else if (safe_str_eq(op, ATTRD_OP_REFRESH)) { GHashTableIter aIter; attribute_t *a = NULL; g_hash_table_iter_init(&aIter, attributes); while (g_hash_table_iter_next(&aIter, NULL, (gpointer *) & a)) { crm_trace("Compatibility write-out of %s for %s from %s", a->id, op, peer->uname); write_or_elect_attribute(a); } } } crm_element_value_int(xml, F_ATTRD_WRITER, &peer_state); if(election_state(writer) == election_won && peer_state == election_won && safe_str_neq(peer->uname, attrd_cluster->uname)) { crm_notice("Detected another attribute writer: %s", peer->uname); election_vote(writer); } else if(peer_state == election_won) { if(peer_writer == NULL) { peer_writer = strdup(peer->uname); crm_notice("Recorded attribute writer: %s", peer->uname); } else if(safe_str_neq(peer->uname, peer_writer)) { crm_notice("Recorded new attribute writer: %s (was %s)", peer->uname, peer_writer); free(peer_writer); peer_writer = strdup(peer->uname); } } if (safe_str_eq(op, ATTRD_OP_UPDATE)) { attrd_peer_update(peer, xml, host, FALSE); } else if (safe_str_eq(op, ATTRD_OP_SYNC)) { attrd_peer_sync(peer, xml); } else if (safe_str_eq(op, ATTRD_OP_PEER_REMOVE)) { int host_id = 0; char *endptr = NULL; host_id = strtol(host, &endptr, 10); if (errno != 0 || endptr == host || *endptr != '\0') { host_id = 0; } else { host = NULL; } attrd_peer_remove(host_id, host, TRUE, peer->uname); } else if (safe_str_eq(op, ATTRD_OP_SYNC_RESPONSE) && safe_str_neq(peer->uname, attrd_cluster->uname)) { xmlNode *child = NULL; crm_notice("Processing %s from %s", op, peer->uname); for (child = __xml_first_child(xml); child != NULL; child = __xml_next(child)) { host = crm_element_value(child, F_ATTRD_HOST); attrd_peer_update(peer, child, host, TRUE); } } }
void write_attribute(attribute_t *a) { int private_updates = 0, cib_updates = 0; xmlNode *xml_top = NULL; attribute_value_t *v = NULL; GHashTableIter iter; enum cib_call_options flags = cib_quorum_override; if (a == NULL) { return; } /* If this attribute will be written to the CIB ... */ if (!a->is_private) { /* Defer the write if now's not a good time */ if (the_cib == NULL) { crm_info("Write out of '%s' delayed: cib not connected", a->id); return; } else if (a->update && (a->update < last_cib_op_done)) { crm_info("Write out of '%s' continuing: update %d considered lost", a->id, a->update); } else if (a->update) { crm_info("Write out of '%s' delayed: update %d in progress", a->id, a->update); return; } else if (mainloop_timer_running(a->timer)) { crm_info("Write out of '%s' delayed: timer is running", a->id); return; } /* Initialize the status update XML */ xml_top = create_xml_node(NULL, XML_CIB_TAG_STATUS); } /* Attribute will be written shortly, so clear changed flag */ a->changed = FALSE; /* We will check all peers' uuids shortly, so initialize this to false */ a->unknown_peer_uuids = FALSE; /* Iterate over each peer value of this attribute */ g_hash_table_iter_init(&iter, a->values); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) & v)) { crm_node_t *peer = crm_get_peer_full(v->nodeid, v->nodename, CRM_GET_PEER_REMOTE|CRM_GET_PEER_CLUSTER); /* If the value's peer info does not correspond to a peer, ignore it */ if (peer == NULL) { crm_notice("Update error (peer not found): %s[%s]=%s failed (host=%p)", v->nodename, a->id, v->current, peer); continue; } /* If we're just learning the peer's node id, remember it */ if (peer->id && (v->nodeid == 0)) { crm_trace("Updating value's nodeid"); v->nodeid = peer->id; } /* If this is a private attribute, no update needs to be sent */ if (a->is_private) { private_updates++; continue; } /* If the peer is found, but its uuid is unknown, defer write */ if (peer->uuid == NULL) { a->unknown_peer_uuids = TRUE; crm_notice("Update error (unknown peer uuid, retry will be attempted once uuid is discovered): %s[%s]=%s failed (host=%p)", v->nodename, a->id, v->current, peer); continue; } /* Add this value to status update XML */ crm_debug("Update: %s[%s]=%s (%s %u %u %s)", v->nodename, a->id, v->current, peer->uuid, peer->id, v->nodeid, peer->uname); build_update_element(xml_top, a, peer->uuid, v->current); cib_updates++; free(v->requested); v->requested = NULL; if (v->current) { v->requested = strdup(v->current); } else { /* Older attrd versions don't know about the cib_mixed_update * flag so make sure it goes to the local cib which does */ flags |= cib_mixed_update|cib_scope_local; } } if (private_updates) { crm_info("Processed %d private change%s for %s, id=%s, set=%s", private_updates, ((private_updates == 1)? "" : "s"), a->id, (a->uuid? a->uuid : "<n/a>"), a->set); } if (cib_updates) { crm_log_xml_trace(xml_top, __FUNCTION__); a->update = cib_internal_op(the_cib, CIB_OP_MODIFY, NULL, XML_CIB_TAG_STATUS, xml_top, NULL, flags, a->user); crm_info("Sent update %d with %d changes for %s, id=%s, set=%s", a->update, cib_updates, a->id, (a->uuid? a->uuid : "<n/a>"), a->set); the_cib->cmds->register_callback( the_cib, a->update, 120, FALSE, strdup(a->id), "attrd_cib_callback", attrd_cib_callback); } free_xml(xml_top); }
gboolean cluster_connect_quorum(gboolean(*dispatch) (unsigned long long, gboolean), void (*destroy) (gpointer)) { cs_error_t rc; int fd = 0; int quorate = 0; uint32_t quorum_type = 0; struct mainloop_fd_callbacks quorum_fd_callbacks; uid_t found_uid = 0; gid_t found_gid = 0; pid_t found_pid = 0; int rv; quorum_fd_callbacks.dispatch = pcmk_quorum_dispatch; quorum_fd_callbacks.destroy = destroy; crm_debug("Configuring Pacemaker to obtain quorum from Corosync"); rc = quorum_initialize(&pcmk_quorum_handle, &quorum_callbacks, &quorum_type); if (rc != CS_OK) { crm_err("Could not connect to the Quorum API: %s (%d)", cs_strerror(rc), rc); goto bail; } else if (quorum_type != QUORUM_SET) { crm_err("Corosync quorum is not configured"); goto bail; } rc = quorum_fd_get(pcmk_quorum_handle, &fd); if (rc != CS_OK) { crm_err("Could not obtain the Quorum API connection: %s (%d)", strerror(rc), rc); goto bail; } /* Quorum provider run as root (in given user namespace, anyway)? */ if (!(rv = crm_ipc_is_authentic_process(fd, (uid_t) 0,(gid_t) 0, &found_pid, &found_uid, &found_gid))) { crm_err("Quorum provider is not authentic:" " process %lld (uid: %lld, gid: %lld)", (long long) PCMK__SPECIAL_PID_AS_0(found_pid), (long long) found_uid, (long long) found_gid); rc = CS_ERR_ACCESS; goto bail; } else if (rv < 0) { crm_err("Could not verify authenticity of Quorum provider: %s (%d)", strerror(-rv), -rv); rc = CS_ERR_ACCESS; goto bail; } rc = quorum_getquorate(pcmk_quorum_handle, &quorate); if (rc != CS_OK) { crm_err("Could not obtain the current Quorum API state: %d", rc); goto bail; } if (quorate) { crm_notice("Quorum acquired"); } else { crm_warn("Quorum lost"); } quorum_app_callback = dispatch; crm_have_quorum = quorate; rc = quorum_trackstart(pcmk_quorum_handle, CS_TRACK_CHANGES | CS_TRACK_CURRENT); if (rc != CS_OK) { crm_err("Could not setup Quorum API notifications: %d", rc); goto bail; } mainloop_add_fd("quorum", G_PRIORITY_HIGH, fd, dispatch, &quorum_fd_callbacks); corosync_initialize_nodelist(NULL, FALSE, NULL); bail: if (rc != CS_OK) { quorum_finalize(pcmk_quorum_handle); return FALSE; } return TRUE; }
static node_t * can_be_master(resource_t * rsc) { node_t *node = NULL; node_t *local_node = NULL; resource_t *parent = uber_parent(rsc); clone_variant_data_t *clone_data = NULL; #if 0 enum rsc_role_e role = RSC_ROLE_UNKNOWN; role = rsc->fns->state(rsc, FALSE); crm_info("%s role: %s", rsc->id, role2text(role)); #endif if (rsc->children) { GListPtr gIter = rsc->children; for (; gIter != NULL; gIter = gIter->next) { resource_t *child = (resource_t *) gIter->data; if (can_be_master(child) == NULL) { pe_rsc_trace(rsc, "Child %s of %s can't be promoted", child->id, rsc->id); return NULL; } } } node = rsc->fns->location(rsc, NULL, FALSE); if (node == NULL) { pe_rsc_trace(rsc, "%s cannot be master: not allocated", rsc->id); return NULL; } else if (is_not_set(rsc->flags, pe_rsc_managed)) { if (rsc->fns->state(rsc, TRUE) == RSC_ROLE_MASTER) { crm_notice("Forcing unmanaged master %s to remain promoted on %s", rsc->id, node->details->uname); } else { return NULL; } } else if (rsc->priority < 0) { pe_rsc_trace(rsc, "%s cannot be master: preference: %d", rsc->id, rsc->priority); return NULL; } else if (can_run_resources(node) == FALSE) { crm_trace("Node cant run any resources: %s", node->details->uname); return NULL; } get_clone_variant_data(clone_data, parent); local_node = pe_hash_table_lookup(parent->allowed_nodes, node->details->id); if (local_node == NULL) { crm_err("%s cannot run on %s: node not allowed", rsc->id, node->details->uname); return NULL; } else if (local_node->count < clone_data->master_node_max || is_not_set(rsc->flags, pe_rsc_managed)) { return local_node; } else { pe_rsc_trace(rsc, "%s cannot be master on %s: node full", rsc->id, node->details->uname); } return NULL; }
static void st_global_callback(stonith_t * stonith, stonith_callback_data_t * data) { crm_notice("Call id %d completed with rc %d", data->call_id, data->rc); }
static enum throttle_state_e throttle_mode(void) { int cores; float load; unsigned int blocked = 0; enum throttle_state_e mode = throttle_none; #if defined(ON_BSD) || defined(ON_SOLARIS) return throttle_none; #endif cores = throttle_num_cores(); if(throttle_cib_load(&load)) { float cib_max_cpu = 0.95; const char *desc = "CIB load"; /* 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 therefor be maxed out * (causing operations to fail or appear to fail) even though * the overall system load is still reasonable. * * Therefor 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; } if(load > 1.5 * cib_max_cpu) { /* Can only happen on machines with a low number of cores */ crm_notice("Extreme %s detected: %f", desc, load); mode |= throttle_extreme; } else if(load > cib_max_cpu) { crm_notice("High %s detected: %f", desc, load); mode |= throttle_high; } else if(load > cib_max_cpu * 0.9) { crm_info("Moderate %s detected: %f", desc, load); mode |= throttle_med; } else if(load > cib_max_cpu * 0.8) { crm_debug("Noticeable %s detected: %f", desc, load); mode |= throttle_low; } else { crm_trace("Negligible %s detected: %f", desc, load); } } 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)) { mode |= throttle_handle_load(load, "CPU load", cores); } if(throttle_io_load(&load, &blocked)) { mode |= throttle_handle_load(load, "IO load", 0); mode |= throttle_handle_load(blocked, "blocked IO ratio", cores); } if(mode & throttle_extreme) { return throttle_extreme; } else if(mode & throttle_high) { return throttle_high; } else if(mode & throttle_med) { return throttle_med; } else if(mode & throttle_low) { return throttle_low; } return throttle_none; }
int cib_process_upgrade_server(const char *op, int options, const char *section, xmlNode * req, xmlNode * input, xmlNode * existing_cib, xmlNode ** result_cib, xmlNode ** answer) { #ifdef CIBPIPE return -EINVAL; #else int rc = pcmk_ok; *answer = NULL; if(crm_element_value(req, F_CIB_SCHEMA_MAX)) { return cib_process_upgrade( op, options, section, req, input, existing_cib, result_cib, answer); } else { int new_version = 0; int current_version = 0; xmlNode *scratch = copy_xml(existing_cib); const char *host = crm_element_value(req, F_ORIG); const char *value = crm_element_value(existing_cib, XML_ATTR_VALIDATION); crm_trace("Processing \"%s\" event", op); if (value != NULL) { current_version = get_schema_version(value); } rc = update_validation(&scratch, &new_version, 0, TRUE, TRUE); if (new_version > current_version) { xmlNode *up = create_xml_node(NULL, __FUNCTION__); rc = pcmk_ok; crm_notice("Upgrade request from %s verified", host); crm_xml_add(up, F_TYPE, "cib"); crm_xml_add(up, F_CIB_OPERATION, CIB_OP_UPGRADE); crm_xml_add(up, F_CIB_SCHEMA_MAX, get_schema_name(new_version)); crm_xml_add(up, F_CIB_DELEGATED, host); crm_xml_add(up, F_CIB_CLIENTID, crm_element_value(req, F_CIB_CLIENTID)); crm_xml_add(up, F_CIB_CALLOPTS, crm_element_value(req, F_CIB_CALLOPTS)); crm_xml_add(up, F_CIB_CALLID, crm_element_value(req, F_CIB_CALLID)); if (cib_legacy_mode() && cib_is_master) { rc = cib_process_upgrade( op, options, section, up, input, existing_cib, result_cib, answer); } else { send_cluster_message(NULL, crm_msg_cib, up, FALSE); } free_xml(up); } else if(rc == pcmk_ok) { rc = -pcmk_err_schema_unchanged; } free_xml(scratch); } return rc; #endif }
int cib_server_process_diff(const char *op, int options, const char *section, xmlNode * req, xmlNode * input, xmlNode * existing_cib, xmlNode ** result_cib, xmlNode ** answer) { int rc = pcmk_ok; if (cib_is_master) { /* the master is never waiting for a resync */ sync_in_progress = 0; } if (sync_in_progress > MAX_DIFF_RETRY) { /* request another full-sync, * the last request may have been lost */ sync_in_progress = 0; } if (sync_in_progress) { int diff_add_updates = 0; int diff_add_epoch = 0; int diff_add_admin_epoch = 0; int diff_del_updates = 0; int diff_del_epoch = 0; int diff_del_admin_epoch = 0; cib_diff_version_details(input, &diff_add_admin_epoch, &diff_add_epoch, &diff_add_updates, &diff_del_admin_epoch, &diff_del_epoch, &diff_del_updates); sync_in_progress++; crm_notice("Not applying diff %d.%d.%d -> %d.%d.%d (sync in progress)", diff_del_admin_epoch, diff_del_epoch, diff_del_updates, diff_add_admin_epoch, diff_add_epoch, diff_add_updates); return -pcmk_err_diff_resync; } rc = cib_process_diff(op, options, section, req, input, existing_cib, result_cib, answer); crm_trace("result: %s (%d), %s", pcmk_strerror(rc), rc, cib_is_master?"master":"slave"); if (rc == -pcmk_err_diff_resync && cib_is_master == FALSE) { free_xml(*result_cib); *result_cib = NULL; send_sync_request(NULL); } else if (rc == -pcmk_err_diff_resync) { rc = -pcmk_err_diff_failed; if (options & cib_force_diff) { crm_warn("Not requesting full refresh in R/W mode"); } } else if(rc != pcmk_ok && cib_legacy_mode()) { crm_warn("Something went wrong in compatibility mode, requesting full refresh"); xml_log_patchset(LOG_INFO, __FUNCTION__, input); free_xml(*result_cib); *result_cib = NULL; send_sync_request(NULL); } return rc; }
gboolean read_config(void) { int rc = CS_OK; int retries = 0; gboolean have_log = FALSE; const char *const_value = NULL; char *logging_debug = NULL; char *logging_logfile = NULL; char *logging_to_logfile = NULL; char *logging_to_syslog = NULL; char *logging_syslog_facility = NULL; #if HAVE_CONFDB char *value = NULL; confdb_handle_t config = 0; confdb_handle_t top_handle = 0; hdb_handle_t local_handle; static confdb_callbacks_t callbacks = { }; do { rc = confdb_initialize(&config, &callbacks); if (rc != CS_OK) { retries++; printf("confdb connection setup failed: %s. Retrying in %ds\n", ais_error2text(rc), retries); crm_info("confdb connection setup failed: %s. Retrying in %ds", ais_error2text(rc), retries); sleep(retries); } else { break; } } while (retries < 5); #elif HAVE_CMAP cmap_handle_t local_handle; uint64_t config = 0; /* There can be only one (possibility if confdb isn't around) */ do { rc = cmap_initialize(&local_handle); if (rc != CS_OK) { retries++; printf("cmap connection setup failed: %s. Retrying in %ds\n", cs_strerror(rc), retries); crm_info("cmap connection setup failed: %s. Retrying in %ds", cs_strerror(rc), retries); sleep(retries); } else { break; } } while (retries < 5); #endif if (rc != CS_OK) { printf("Could not connect to Cluster Configuration Database API, error %d\n", rc); crm_warn("Could not connect to Cluster Configuration Database API, error %d", rc); return FALSE; } stack = get_cluster_type(); crm_info("Reading configure for stack: %s", name_for_cluster_type(stack)); /* =::=::= Should we be here =::=::= */ if (stack == pcmk_cluster_corosync) { set_daemon_option("cluster_type", "corosync"); set_daemon_option("quorum_type", "corosync"); #if HAVE_CONFDB } else if (stack == pcmk_cluster_cman) { set_daemon_option("cluster_type", "cman"); set_daemon_option("quorum_type", "cman"); enable_crmd_as_root(TRUE); } else if (stack == pcmk_cluster_classic_ais) { set_daemon_option("cluster_type", "openais"); set_daemon_option("quorum_type", "pcmk"); /* Look for a service block to indicate our plugin is loaded */ top_handle = config_find_init(config); local_handle = config_find_next(config, "service", top_handle); while (local_handle) { get_config_opt(config, local_handle, "name", &value, NULL); if (safe_str_eq("pacemaker", value)) { get_config_opt(config, local_handle, "ver", &value, "0"); if (safe_str_eq(value, "1")) { get_config_opt(config, local_handle, "use_logd", &value, "no"); set_daemon_option("use_logd", value); set_daemon_option("LOGD", value); get_config_opt(config, local_handle, "use_mgmtd", &value, "no"); enable_mgmtd(crm_is_true(value)); } else { crm_err("We can only start Pacemaker from init if using version 1" " of the Pacemaker plugin for Corosync. Terminating."); crm_exit(DAEMON_RESPAWN_STOP); } break; } local_handle = config_find_next(config, "service", top_handle); } free(value); #endif } else { crm_err("Unsupported stack type: %s", name_for_cluster_type(stack)); return FALSE; } #if HAVE_CONFDB top_handle = config_find_init(config); local_handle = config_find_next(config, "logging", top_handle); #endif /* =::=::= Logging =::=::= */ get_config_opt(config, local_handle, KEY_PREFIX "debug", &logging_debug, "off"); const_value = daemon_option("debugfile"); if (const_value) { logging_to_logfile = strdup("on"); logging_logfile = strdup(const_value); crm_trace("Using debugfile setting from the environment: %s", logging_logfile); } else { get_config_opt(config, local_handle, KEY_PREFIX "to_logfile", &logging_to_logfile, "off"); get_config_opt(config, local_handle, KEY_PREFIX "logfile", &logging_logfile, "/var/log/pacemaker"); } const_value = daemon_option("logfacility"); if (const_value) { logging_syslog_facility = strdup(const_value); crm_trace("Using logfacility setting from the environment: %s", logging_syslog_facility); if (safe_str_eq(logging_syslog_facility, "none")) { logging_to_syslog = strdup("off"); } else { logging_to_syslog = strdup("on"); } } else { get_config_opt(config, local_handle, KEY_PREFIX "to_syslog", &logging_to_syslog, "on"); get_config_opt(config, local_handle, KEY_PREFIX "syslog_facility", &logging_syslog_facility, "daemon"); } #if HAVE_CONFDB confdb_finalize(config); #elif HAVE_CMAP if(local_handle){ gid_t gid = 0; if (crm_user_lookup(CRM_DAEMON_USER, NULL, &gid) < 0) { crm_warn("No group found for user %s", CRM_DAEMON_USER); } else { char key[PATH_MAX]; snprintf(key, PATH_MAX, "uidgid.gid.%u", gid); rc = cmap_set_uint8(local_handle, key, 1); crm_notice("Configured corosync to accept connections from group %u: %s (%d)", gid, ais_error2text(rc), rc); } } cmap_finalize(local_handle); #endif if (daemon_option("debug")) { crm_trace("Using debug setting from the environment: %s", daemon_option("debug")); if (get_crm_log_level() < LOG_DEBUG && daemon_option_enabled("pacemakerd", "debug")) { set_crm_log_level(LOG_DEBUG); } } else if (crm_is_true(logging_debug)) { set_daemon_option("debug", "1"); if (get_crm_log_level() < LOG_DEBUG) { set_crm_log_level(LOG_DEBUG); } } else { set_daemon_option("debug", "0"); } if (crm_is_true(logging_to_logfile)) { if (crm_add_logfile(logging_logfile)) { /* What a cluster fsck, eventually we need to mandate /one/ */ set_daemon_option("debugfile", logging_logfile); set_daemon_option("DEBUGLOG", logging_logfile); have_log = TRUE; } else { crm_err("Couldn't create logfile: %s", logging_logfile); } } if (have_log && crm_is_true(logging_to_syslog) == FALSE) { qb_log_ctl(QB_LOG_SYSLOG, QB_LOG_CONF_ENABLED, QB_FALSE); free(logging_syslog_facility); logging_syslog_facility = strdup("none"); crm_info("User configured file based logging and explicitly disabled syslog."); } else if (crm_is_true(logging_to_syslog) == FALSE) { crm_err("Please enable some sort of logging, either 'to_logfile: on' or 'to_syslog: on'."); crm_err("If you use file logging, be sure to also define a value for 'logfile'"); } set_daemon_option("logfacility", logging_syslog_facility); setenv("HA_LOGFACILITY", logging_syslog_facility, 1); free(logging_debug); free(logging_logfile); free(logging_to_logfile); free(logging_to_syslog); free(logging_syslog_facility); return TRUE; }
void tengine_stonith_callback(stonith_t * stonith, stonith_callback_data_t * data) { char *uuid = NULL; int stonith_id = -1; int transition_id = -1; crm_action_t *action = NULL; int call_id = data->call_id; int rc = data->rc; char *userdata = data->userdata; CRM_CHECK(userdata != NULL, return); crm_notice("Stonith operation %d/%s: %s (%d)", call_id, (char *)userdata, pcmk_strerror(rc), rc); if (AM_I_DC == FALSE) { return; } /* crm_info("call=%d, optype=%d, node_name=%s, result=%d, node_list=%s, action=%s", */ /* op->call_id, op->optype, op->node_name, op->op_result, */ /* (char *)op->node_list, op->private_data); */ /* filter out old STONITH actions */ CRM_CHECK(decode_transition_key(userdata, &uuid, &transition_id, &stonith_id, NULL), goto bail); if (transition_graph->complete || stonith_id < 0 || safe_str_neq(uuid, te_uuid) || transition_graph->id != transition_id) { crm_info("Ignoring STONITH action initiated outside of the current transition"); goto bail; } action = get_action(stonith_id, FALSE); if (action == NULL) { crm_err("Stonith action not matched"); goto bail; } stop_te_timer(action->timer); if (rc == pcmk_ok) { const char *target = crm_element_value(action->xml, XML_LRM_ATTR_TARGET); const char *uuid = crm_element_value(action->xml, XML_LRM_ATTR_TARGET_UUID); const char *op = crm_meta_value(action->params, "stonith_action"); crm_info("Stonith operation %d for %s passed", call_id, target); if (action->confirmed == FALSE) { te_action_confirmed(action); if (safe_str_eq("on", op)) { const char *value = NULL; char *now = crm_itoa(time(NULL)); update_attrd(target, CRM_ATTR_UNFENCED, now, NULL, FALSE); free(now); value = crm_meta_value(action->params, XML_OP_ATTR_DIGESTS_ALL); update_attrd(target, CRM_ATTR_DIGESTS_ALL, value, NULL, FALSE); value = crm_meta_value(action->params, XML_OP_ATTR_DIGESTS_SECURE); update_attrd(target, CRM_ATTR_DIGESTS_SECURE, value, NULL, FALSE); } else if (action->sent_update == FALSE) { send_stonith_update(action, target, uuid); action->sent_update = TRUE; } } st_fail_count_reset(target); } else { const char *target = crm_element_value(action->xml, XML_LRM_ATTR_TARGET); enum transition_action abort_action = tg_restart; action->failed = TRUE; crm_notice("Stonith operation %d for %s failed (%s): aborting transition.", call_id, target, pcmk_strerror(rc)); /* If no fence devices were available, there's no use in immediately * checking again, so don't start a new transition in that case. */ if (rc == -ENODEV) { crm_warn("No devices found in cluster to fence %s, giving up", target); abort_action = tg_stop; } /* Increment the fail count now, so abort_for_stonith_failure() can * check it. Non-DC nodes will increment it in tengine_stonith_notify(). */ st_fail_count_increment(target); abort_for_stonith_failure(abort_action, target, NULL); } update_graph(transition_graph, action); trigger_graph(); bail: free(userdata); free(uuid); return; }
int main(int argc, char **argv) { int flag = 0; int index = 0; int argerr = 0; crm_ipc_t *old_instance = NULL; attrd_init_mainloop(); crm_log_preinit(NULL, argc, argv); crm_set_options(NULL, "[options]", long_options, "Daemon for aggregating and atomically storing node attribute updates into the CIB"); mainloop_add_signal(SIGTERM, attrd_shutdown); while (1) { flag = crm_get_option(argc, argv, &index); if (flag == -1) break; switch (flag) { case 'V': crm_bump_log_level(argc, argv); break; case 'h': /* Help message */ crm_help(flag, CRM_EX_OK); break; default: ++argerr; break; } } if (optind > argc) { ++argerr; } if (argerr) { crm_help('?', CRM_EX_USAGE); } crm_log_init(T_ATTRD, LOG_INFO, TRUE, FALSE, argc, argv, FALSE); crm_notice("Starting Pacemaker node attribute manager"); old_instance = crm_ipc_new(T_ATTRD, 0); if (crm_ipc_connect(old_instance)) { /* IPC end-point already up */ crm_ipc_close(old_instance); crm_ipc_destroy(old_instance); crm_err("pacemaker-attrd is already active, aborting startup"); crm_exit(CRM_EX_OK); } else { /* not up or not authentic, we'll proceed either way */ crm_ipc_destroy(old_instance); old_instance = NULL; } attributes = g_hash_table_new_full(crm_str_hash, g_str_equal, NULL, free_attribute); /* Connect to the CIB before connecting to the cluster or listening for IPC. * This allows us to assume the CIB is connected whenever we process a * cluster or IPC message (which also avoids start-up race conditions). */ if (attrd_cib_connect(10) != pcmk_ok) { attrd_exit_status = CRM_EX_FATAL; goto done; } crm_info("CIB connection active"); if (attrd_cluster_connect() != pcmk_ok) { attrd_exit_status = CRM_EX_FATAL; goto done; } crm_info("Cluster connection active"); // Initialization that requires the cluster to be connected attrd_election_init(); attrd_cib_init(); /* Set a private attribute for ourselves with the protocol version we * support. This lets all nodes determine the minimum supported version * across all nodes. It also ensures that the writer learns our node name, * so it can send our attributes to the CIB. */ attrd_broadcast_protocol(); attrd_init_ipc(&ipcs, attrd_ipc_dispatch); crm_notice("Pacemaker node attribute manager successfully started and accepting connections"); attrd_run_mainloop(); done: crm_info("Shutting down attribute manager"); attrd_election_fini(); attrd_ipc_fini(); attrd_lrmd_disconnect(); attrd_cib_disconnect(); g_hash_table_destroy(attributes); crm_exit(attrd_exit_status); }
ssize_t crm_ipcs_sendv(crm_client_t * c, struct iovec * iov, enum crm_ipc_flags flags) { ssize_t rc; static uint32_t id = 1; struct crm_ipc_response_header *header = iov[0].iov_base; if (c->flags & crm_client_flag_ipc_proxied) { /* _ALL_ replies to proxied connections need to be sent as events */ if (is_not_set(flags, crm_ipc_server_event)) { flags |= crm_ipc_server_event; /* this flag lets us know this was originally meant to be a response. * even though we're sending it over the event channel. */ flags |= crm_ipc_proxied_relay_response; } } header->flags |= flags; if (flags & crm_ipc_server_event) { header->qb.id = id++; /* We don't really use it, but doesn't hurt to set one */ if (flags & crm_ipc_server_free) { crm_trace("Sending the original to %p[%d]", c->ipcs, c->pid); c->event_queue = g_list_append(c->event_queue, iov); } else { struct iovec *iov_copy = calloc(2, sizeof(struct iovec)); crm_trace("Sending a copy to %p[%d]", c->ipcs, c->pid); iov_copy[0].iov_len = iov[0].iov_len; iov_copy[0].iov_base = malloc(iov[0].iov_len); memcpy(iov_copy[0].iov_base, iov[0].iov_base, iov[0].iov_len); iov_copy[1].iov_len = iov[1].iov_len; iov_copy[1].iov_base = malloc(iov[1].iov_len); memcpy(iov_copy[1].iov_base, iov[1].iov_base, iov[1].iov_len); c->event_queue = g_list_append(c->event_queue, iov_copy); } } else { CRM_LOG_ASSERT(header->qb.id != 0); /* Replying to a specific request */ rc = qb_ipcs_response_sendv(c->ipcs, iov, 2); if (rc < header->qb.size) { crm_notice("Response %d to %p[%d] (%d bytes) failed: %s (%d)", header->qb.id, c->ipcs, c->pid, header->qb.size, pcmk_strerror(rc), rc); } else { crm_trace("Response %d sent, %d bytes to %p[%d]", header->qb.id, rc, c->ipcs, c->pid); } if (flags & crm_ipc_server_free) { free(iov[0].iov_base); free(iov[1].iov_base); free(iov); } } if (flags & crm_ipc_server_event) { rc = crm_ipcs_flush_events(c); } else { crm_ipcs_flush_events(c); } if (rc == -EPIPE || rc == -ENOTCONN) { crm_trace("Client %p disconnected", c->ipcs); } return rc; }
static void operation_finished(mainloop_child_t *p, int status, int signo, int exitcode) { char *next = NULL; char *offset = NULL; svc_action_t *op = mainloop_get_child_userdata(p); pid_t pid = mainloop_get_child_pid(p); mainloop_clear_child_userdata(p); op->status = PCMK_LRM_OP_DONE; CRM_ASSERT(op->pid == pid); if (op->opaque->stderr_gsource) { /* Make sure we have read everything from the buffer. * Depending on the priority mainloop gives the fd, operation_finished * could occur before all the reads are done. Force the read now.*/ dispatch_stderr(op); } if (op->opaque->stdout_gsource) { /* Make sure we have read everything from the buffer. * Depending on the priority mainloop gives the fd, operation_finished * could occur before all the reads are done. Force the read now.*/ dispatch_stdout(op); } if (signo) { if (mainloop_get_child_timeout(p)) { crm_warn("%s:%d - timed out after %dms", op->id, op->pid, op->timeout); op->status = PCMK_LRM_OP_TIMEOUT; op->rc = PCMK_OCF_TIMEOUT; } else { crm_warn("%s:%d - terminated with signal %d", op->id, op->pid, signo); op->status = PCMK_LRM_OP_ERROR; op->rc = PCMK_OCF_SIGNAL; } } else { op->rc = exitcode; crm_debug("%s:%d - exited with rc=%d", op->id, op->pid, exitcode); if (op->stdout_data) { next = op->stdout_data; do { offset = next; next = strchrnul(offset, '\n'); crm_debug("%s:%d [ %.*s ]", op->id, op->pid, (int) (next - offset), offset); if (next[0] != 0) { next++; } } while (next != NULL && next[0] != 0); } if (op->stderr_data) { next = op->stderr_data; do { offset = next; next = strchrnul(offset, '\n'); crm_notice("%s:%d [ %.*s ]", op->id, op->pid, (int) (next - offset), offset); if (next[0] != 0) { next++; } } while (next != NULL && next[0] != 0); } } op->pid = 0; operation_finalize(op); }
static void tengine_stonith_notify(stonith_t * st, stonith_event_t *st_event) { 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_err("We were alegedly just fenced by %s for %s!", st_event->executioner, st_event->origin); register_fsa_error_adv(C_FSA_INTERNAL, I_ERROR, NULL, NULL, __FUNCTION__); return; } 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->operation, 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; int confirm = 0; char *target_copy = strdup(st_event->target); /* In case fenced hasn't noticed yet */ 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); } /* In case fenced is already trying to shoot it */ confirm = open("/var/run/cluster/fenced_override", O_NONBLOCK|O_WRONLY); if (confirm) { int ignore = 0; int len = strlen(target_copy); errno = 0; local_rc = write(confirm, target_copy, len); ignore = write(confirm, "\n", 1); if(errno == EBADF) { crm_trace("CMAN not expecting %s to be fenced (yet)", st_event->target); } else if (local_rc < len) { crm_perror(LOG_ERR, "Confirmation of CMAN fencing event for '%s' failed: %d", st_event->target, local_rc); } else { fsync(confirm); crm_notice("Confirmed CMAN fencing event for '%s'", st_event->target); } close(confirm); } } #endif if (st_event->result == pcmk_ok) { 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 (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) { const char *uuid = get_uuid(st_event->target); send_stonith_update(NULL, st_event->target, uuid); } stonith_cleanup_list = g_list_append(stonith_cleanup_list, strdup(st_event->target)); } else if (AM_I_DC && st_event->client_origin && safe_str_neq(st_event->client_origin, crm_system_name)) { const char *uuid = get_uuid(st_event->target); /* If a remote process outside of pacemaker invoked stonith to * fence someone, report the fencing result to the cib * and abort the transition graph. */ crm_info("External fencing operation from %s fenced %s", st_event->client_origin, st_event->target); send_stonith_update(NULL, st_event->target, uuid); abort_transition(INFINITY, tg_restart, "External Fencing Operation", NULL); } } }
gboolean fail_incompletable_actions(crm_graph_t * graph, const char *down_node) { const char *target = NULL; xmlNode *last_action = NULL; GListPtr gIter = NULL; GListPtr gIter2 = NULL; if (graph == NULL || graph->complete) { return FALSE; } gIter = graph->synapses; for (; gIter != NULL; gIter = gIter->next) { synapse_t *synapse = (synapse_t *) gIter->data; if (synapse->confirmed) { continue; } gIter2 = synapse->actions; for (; gIter2 != NULL; gIter2 = gIter2->next) { crm_action_t *action = (crm_action_t *) gIter2->data; if (action->type == action_type_pseudo || action->confirmed) { continue; } else if (action->type == action_type_crm) { const char *task = crm_element_value(action->xml, XML_LRM_ATTR_TASK); if (safe_str_eq(task, CRM_OP_FENCE)) { continue; } } target = crm_element_value(action->xml, XML_LRM_ATTR_TARGET_UUID); if (safe_str_eq(target, down_node)) { action->failed = TRUE; synapse->failed = TRUE; last_action = action->xml; stop_te_timer(action->timer); update_graph(graph, action); if (synapse->executed) { crm_notice("Action %d (%s) was pending on %s (offline)", action->id, ID(action->xml), down_node); } else { crm_notice("Action %d (%s) is scheduled for %s (offline)", action->id, ID(action->xml), down_node); } } } } if (last_action != NULL) { crm_warn("Node %s shutdown resulted in un-runnable actions", down_node); abort_transition(INFINITY, tg_restart, "Node failure", last_action); return TRUE; } return FALSE; }
gboolean process_pe_message(xmlNode * msg, xmlNode * xml_data, qb_ipcs_connection_t* sender) { static char *filename = NULL; static char *last_digest = NULL; const char *sys_to = crm_element_value(msg, F_CRM_SYS_TO); const char *op = crm_element_value(msg, F_CRM_TASK); const char *ref = crm_element_value(msg, XML_ATTR_REFERENCE); crm_trace("Processing %s op (ref=%s)...", op, ref); if (op == NULL) { /* error */ } else if (strcasecmp(op, CRM_OP_HELLO) == 0) { /* ignore */ } else if (safe_str_eq(crm_element_value(msg, F_CRM_MSG_TYPE), XML_ATTR_RESPONSE)) { /* ignore */ } else if (sys_to == NULL || strcasecmp(sys_to, CRM_SYSTEM_PENGINE) != 0) { crm_trace("Bad sys-to %s", crm_str(sys_to)); return FALSE; } else if (strcasecmp(op, CRM_OP_PECALC) == 0) { int seq = -1; int series_id = 0; int series_wrap = 0; char *digest = NULL; char *graph_file = NULL; const char *value = NULL; pe_working_set_t data_set; xmlNode *converted = NULL; xmlNode *reply = NULL; gboolean is_repoke = FALSE; gboolean process = TRUE; #if HAVE_BZLIB_H gboolean compress = TRUE; #else gboolean compress = FALSE; #endif crm_config_error = FALSE; crm_config_warning = FALSE; was_processing_error = FALSE; was_processing_warning = FALSE; graph_file = strdup(CRM_STATE_DIR "/graph.XXXXXX"); graph_file = mktemp(graph_file); set_working_set_defaults(&data_set); digest = calculate_xml_versioned_digest(xml_data, FALSE, FALSE, CRM_FEATURE_SET); converted = copy_xml(xml_data); if (cli_config_update(&converted, NULL, TRUE) == FALSE) { data_set.graph = create_xml_node(NULL, XML_TAG_GRAPH); crm_xml_add_int(data_set.graph, "transition_id", 0); crm_xml_add_int(data_set.graph, "cluster-delay", 0); process = FALSE; } else if (safe_str_eq(digest, last_digest)) { crm_trace("Input has not changed since last time, not saving to disk"); is_repoke = TRUE; } else { free(last_digest); last_digest = digest; } if (process) { do_calculations(&data_set, converted, NULL); } series_id = get_series(); series_wrap = series[series_id].wrap; value = pe_pref(data_set.config_hash, series[series_id].param); if (value != NULL) { series_wrap = crm_int_helper(value, NULL); if (errno != 0) { series_wrap = series[series_id].wrap; } } else { crm_config_warn("No value specified for cluster" " preference: %s", series[series_id].param); } seq = get_last_sequence(PE_STATE_DIR, series[series_id].name); data_set.input = NULL; reply = create_reply(msg, data_set.graph); CRM_ASSERT(reply != NULL); if (is_repoke == FALSE) { free(filename); filename = generate_series_filename(PE_STATE_DIR, series[series_id].name, seq, compress); } crm_xml_add(reply, F_CRM_TGRAPH_INPUT, filename); crm_xml_add_int(reply, "graph-errors", was_processing_error); crm_xml_add_int(reply, "graph-warnings", was_processing_warning); crm_xml_add_int(reply, "config-errors", crm_config_error); crm_xml_add_int(reply, "config-warnings", crm_config_warning); if (crm_ipcs_send(sender, reply, TRUE) == FALSE) { crm_err("Couldn't send transition graph to peer, discarding"); } free_xml(reply); cleanup_alloc_calculations(&data_set); if (is_repoke == FALSE && series_wrap != 0) { write_xml_file(xml_data, filename, compress); write_last_sequence(PE_STATE_DIR, series[series_id].name, seq + 1, series_wrap); } if (was_processing_error) { crm_err("Transition %d:" " ERRORs found during PE processing." " PEngine Input stored in: %s", transition_id, filename); } else if (was_processing_warning) { crm_warn("Transition %d:" " WARNINGs found during PE processing." " PEngine Input stored in: %s", transition_id, filename); } else { crm_notice("Transition %d: PEngine Input stored in: %s", transition_id, filename); } if (crm_config_error) { crm_notice("Configuration ERRORs found during PE processing." " Please run \"crm_verify -L\" to identify issues."); } else if (crm_config_warning) { crm_notice("Configuration WARNINGs found during PE processing." " Please run \"crm_verify -L\" to identify issues."); } free_xml(converted); free(graph_file); } else if (strcasecmp(op, CRM_OP_QUIT) == 0) { crm_warn("Received quit message, terminating"); exit(0); } return TRUE; }
static void tengine_stonith_notify(stonith_t * st, stonith_event_t * st_event) { static char *client_id = NULL; if(client_id == NULL) { 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_err("We were alegedly just fenced by %s for %s!", st_event->executioner, st_event->origin); register_fsa_error_adv(C_FSA_INTERNAL, I_ERROR, NULL, NULL, __FUNCTION__); return; } 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->operation, 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) { 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 (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) { const char *uuid = get_uuid(st_event->target); send_stonith_update(NULL, st_event->target, uuid); } stonith_cleanup_list = g_list_append(stonith_cleanup_list, strdup(st_event->target)); } else if (AM_I_DC && st_event->client_origin && safe_str_neq(st_event->client_origin, client_id)) { const char *uuid = get_uuid(st_event->target); /* If a remote process outside of pacemaker invoked stonith to * fence someone, report the fencing result to the cib * and abort the transition graph. */ crm_info("External fencing operation from %s fenced %s", st_event->client_origin, st_event->target); send_stonith_update(NULL, st_event->target, uuid); abort_transition(INFINITY, tg_restart, "External Fencing Operation", NULL); } } }
gboolean crm_add_logfile(const char *filename) { bool is_default = false; static int default_fd = -1; static gboolean have_logfile = FALSE; const char *default_logfile = "/var/log/pacemaker.log"; struct stat parent; int fd = 0, rc = 0; FILE *logfile = NULL; char *parent_dir = NULL; char *filename_cp; if (filename == NULL && have_logfile == FALSE) { filename = default_logfile; } if (filename == NULL) { return FALSE; /* Nothing to do */ } else if(safe_str_eq(filename, "none")) { return FALSE; /* Nothing to do */ } else if(safe_str_eq(filename, "/dev/null")) { return FALSE; /* Nothing to do */ } else if(safe_str_eq(filename, default_logfile)) { is_default = TRUE; } if(is_default && default_fd >= 0) { return FALSE; /* Nothing to do */ } /* Check the parent directory */ filename_cp = strdup(filename); parent_dir = dirname(filename_cp); rc = stat(parent_dir, &parent); if (rc != 0) { crm_err("Directory '%s' does not exist: logging to '%s' is disabled", parent_dir, filename); free(filename_cp); return FALSE; } free(filename_cp); errno = 0; logfile = fopen(filename, "a"); if(logfile == NULL) { crm_err("%s (%d): Logging to '%s' as uid=%u, gid=%u is disabled", pcmk_strerror(errno), errno, filename, geteuid(), getegid()); return FALSE; } /* Check/Set permissions if we're root */ if (geteuid() == 0) { struct stat st; uid_t pcmk_uid = 0; gid_t pcmk_gid = 0; gboolean fix = FALSE; int logfd = fileno(logfile); rc = fstat(logfd, &st); if (rc < 0) { crm_perror(LOG_WARNING, "Cannot stat %s", filename); fclose(logfile); return FALSE; } if(crm_user_lookup(CRM_DAEMON_USER, &pcmk_uid, &pcmk_gid) == 0) { if (st.st_gid != pcmk_gid) { /* Wrong group */ fix = TRUE; } else if ((st.st_mode & S_IRWXG) != (S_IRGRP | S_IWGRP)) { /* Not read/writable by the correct group */ fix = TRUE; } } if (fix) { rc = fchown(logfd, pcmk_uid, pcmk_gid); if (rc < 0) { crm_warn("Cannot change the ownership of %s to user %s and gid %d", filename, CRM_DAEMON_USER, pcmk_gid); } rc = fchmod(logfd, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP); if (rc < 0) { crm_warn("Cannot change the mode of %s to rw-rw----", filename); } fprintf(logfile, "Set r/w permissions for uid=%d, gid=%d on %s\n", pcmk_uid, pcmk_gid, filename); if (fflush(logfile) < 0 || fsync(logfd) < 0) { crm_err("Couldn't write out logfile: %s", filename); } } } /* Close and reopen with libqb */ fclose(logfile); fd = qb_log_file_open(filename); if (fd < 0) { crm_perror(LOG_WARNING, "Couldn't send additional logging to %s", filename); return FALSE; } if(is_default) { default_fd = fd; } else if(default_fd >= 0) { crm_notice("Switching to %s", filename); qb_log_ctl(default_fd, QB_LOG_CONF_ENABLED, QB_FALSE); } crm_notice("Additional logging available in %s", filename); qb_log_ctl(fd, QB_LOG_CONF_ENABLED, QB_TRUE); /* qb_log_ctl(fd, QB_LOG_CONF_FILE_SYNC, 1); Turn on synchronous writes */ /* Enable callsites */ crm_update_callsites(); have_logfile = TRUE; return TRUE; }
static void tengine_stonith_notify(stonith_t * st, const char *event, xmlNode * msg) { int rc = -99; const char *origin = NULL; const char *target = NULL; const char *executioner = NULL; xmlNode *action = get_xpath_object("//st-data", msg, LOG_ERR); if (action == NULL) { crm_log_xml_err(msg, "Notify data not found"); return; } crm_log_xml_debug(msg, "stonith_notify"); crm_element_value_int(msg, F_STONITH_RC, &rc); origin = crm_element_value(action, F_STONITH_ORIGIN); target = crm_element_value(action, F_STONITH_TARGET); executioner = crm_element_value(action, F_STONITH_DELEGATE); if (rc == stonith_ok && crm_str_eq(target, fsa_our_uname, TRUE)) { crm_err("We were alegedly just fenced by %s for %s!", executioner, origin); register_fsa_error_adv(C_FSA_INTERNAL, I_ERROR, NULL, NULL, __FUNCTION__); return; } crm_notice("Peer %s was%s terminated (%s) by %s for %s: %s (ref=%s)", target, rc == stonith_ok?"":" not", crm_element_value(action, F_STONITH_OPERATION), executioner ? executioner : "<anyone>", origin, stonith_error2string(rc), crm_element_value(action, F_STONITH_REMOTE)); #if SUPPORT_CMAN if (rc == stonith_ok && is_cman_cluster()) { int local_rc = 0; int confirm = 0; char *target_copy = crm_strdup(target); /* In case fenced hasn't noticed yet */ local_rc = fenced_external(target_copy); if (local_rc != 0) { crm_err("Could not notify CMAN that '%s' is now fenced: %d", target, local_rc); } else { crm_notice("Notified CMAN that '%s' is now fenced", target); } /* In case fenced is already trying to shoot it */ confirm = open("/var/run/cluster/fenced_override", O_NONBLOCK|O_WRONLY); if (confirm) { int len = strlen(target_copy); errno = 0; local_rc = write(confirm, target_copy, len); write(confirm, "\n", 1); if(errno == EBADF) { crm_trace("CMAN not expecting %s to be fenced (yet)", target); } else if (local_rc < len) { crm_perror(LOG_ERR, "Confirmation of CMAN fencing event for '%s' failed: %d", target, local_rc); } else { fsync(confirm); crm_notice("Confirmed CMAN fencing event for '%s'", target); } close(confirm); } } #endif if (rc == stonith_ok && safe_str_eq(target, origin)) { if (fsa_our_dc == NULL || safe_str_eq(fsa_our_dc, target)) { crm_notice("Target was our leader %s (recorded: %s)", 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 (safe_str_eq(executioner, fsa_our_uname)) { const char *uuid = get_uuid(target); send_stonith_update(NULL, target, uuid); } else { stonith_cleanup_list = g_list_append(stonith_cleanup_list, crm_strdup(target)); } } } }
void attrd_peer_update(crm_node_t *peer, xmlNode *xml, const char *host, bool filter) { bool changed = FALSE; attribute_value_t *v = NULL; const char *attr = crm_element_value(xml, F_ATTRD_ATTRIBUTE); const char *value = crm_element_value(xml, F_ATTRD_VALUE); attribute_t *a = g_hash_table_lookup(attributes, attr); if(a == NULL) { a = create_attribute(xml); } if(host == NULL) { GHashTableIter vIter; g_hash_table_iter_init(&vIter, a->values); crm_debug("Setting %s for all hosts to %s", attr, value); xml_remove_prop(xml, F_ATTRD_HOST_ID); while (g_hash_table_iter_next(&vIter, (gpointer *) & host, NULL)) { attrd_peer_update(peer, xml, host, filter); } return; } v = attrd_lookup_or_create_value(a->values, host, xml); if(filter && safe_str_neq(v->current, value) && safe_str_eq(host, attrd_cluster->uname)) { xmlNode *sync = create_xml_node(NULL, __FUNCTION__); crm_notice("%s[%s]: local value '%s' takes priority over '%s' from %s", a->id, host, v->current, value, peer->uname); crm_xml_add(sync, F_ATTRD_TASK, ATTRD_OP_SYNC_RESPONSE); v = g_hash_table_lookup(a->values, host); build_attribute_xml(sync, a->id, a->set, a->uuid, a->timeout_ms, a->user, a->is_private, v->nodename, v->nodeid, v->current); crm_xml_add_int(sync, F_ATTRD_WRITER, election_state(writer)); send_attrd_message(peer, sync); free_xml(sync); } else if(safe_str_neq(v->current, value)) { crm_info("Setting %s[%s]: %s -> %s from %s", attr, host, v->current, value, peer->uname); free(v->current); if(value) { v->current = strdup(value); } else { v->current = NULL; } changed = TRUE; } else { crm_trace("Unchanged %s[%s] from %s is %s", attr, host, peer->uname, value); } a->changed |= changed; if(changed) { if(a->timer) { crm_trace("Delayed write out (%dms) for %s", a->timeout_ms, a->id); mainloop_timer_start(a->timer); } else { write_or_elect_attribute(a); } } /* this only involves cluster nodes. */ if(v->nodeid == 0 && (v->is_remote == FALSE)) { if(crm_element_value_int(xml, F_ATTRD_HOST_ID, (int*)&v->nodeid) == 0) { /* Create the name/id association */ crm_node_t *peer = crm_get_peer(v->nodeid, host); crm_trace("We know %s's node id now: %s", peer->uname, peer->uuid); if(election_state(writer) == election_won) { write_attributes(FALSE, TRUE); return; } } } }
int process_remote_stonith_exec(xmlNode *msg) { int rc = 0; const char *id = NULL; remote_fencing_op_t *op = NULL; xmlNode *dev = get_xpath_object("//@"F_STONITH_REMOTE, msg, LOG_ERR); CRM_CHECK(dev != NULL, return -EPROTO); id = crm_element_value(dev, F_STONITH_REMOTE); CRM_CHECK(id != NULL, return -EPROTO); dev = get_xpath_object("//@"F_STONITH_RC, msg, LOG_ERR); CRM_CHECK(dev != NULL, return -EPROTO); crm_element_value_int(dev, F_STONITH_RC, &rc); if(remote_op_list) { op = g_hash_table_lookup(remote_op_list, id); } if(op == NULL && rc == pcmk_ok) { /* Record successful fencing operations */ const char *client_id = crm_element_value(msg, F_STONITH_CLIENTID); op = create_remote_stonith_op(client_id, msg, TRUE); } if(op == NULL) { /* Could be for an event that began before we started */ /* TODO: Record the op for later querying */ crm_info("Unknown or expired remote op: %s", id); return -EOPNOTSUPP; } if(is_set(op->call_options, st_opt_topology)) { const char *device = crm_element_value(msg, F_STONITH_DEVICE); crm_notice("Call to %s for %s on behalf of %s: %s (%d)", device, op->target, op->originator, rc == pcmk_ok?"passed":"failed", rc); if(safe_str_eq(op->originator, stonith_our_uname)) { if(op->state == st_done) { remote_op_done(op, msg, rc); return rc; } else if(rc == pcmk_ok && op->devices) { /* Success, are there any more? */ op->devices = op->devices->next; } if(op->devices == NULL) { crm_trace("Broadcasting completion of complex fencing op for %s", op->target); send_cluster_message(NULL, crm_msg_stonith_ng, msg, FALSE); op->state = st_done; return rc; } } else { op->state = st_done; remote_op_done(op, msg, rc); } } else if(rc == pcmk_ok && op->devices == NULL) { crm_trace("All done for %s", op->target); op->state = st_done; remote_op_done(op, msg, rc); return rc; } /* Retry on failure or execute the rest of the topology */ crm_trace("Next for %s (rc was %d)", op->target, rc); call_remote_stonith(op, NULL); return rc; }
/* A_TE_INVOKE, A_TE_CANCEL */ void do_te_invoke(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) { if (AM_I_DC == FALSE || (fsa_state != S_TRANSITION_ENGINE && (action & A_TE_INVOKE))) { crm_notice("No need to invoke the TE (%s) in state %s", fsa_action2string(action), fsa_state2string(fsa_state)); return; } if (action & A_TE_CANCEL) { crm_debug("Cancelling the transition: %s", transition_graph->complete ? "inactive" : "active"); abort_transition(INFINITY, tg_restart, "Peer Cancelled", NULL); if (transition_graph->complete == FALSE) { crmd_fsa_stall(NULL); } } else if (action & A_TE_HALT) { crm_debug("Halting the transition: %s", transition_graph->complete ? "inactive" : "active"); abort_transition(INFINITY, tg_stop, "Peer Halt", NULL); if (transition_graph->complete == FALSE) { crmd_fsa_stall(NULL); } } else if (action & A_TE_INVOKE) { const char *value = NULL; xmlNode *graph_data = NULL; ha_msg_input_t *input = fsa_typed_data(fsa_dt_ha_msg); const char *ref = crm_element_value(input->msg, XML_ATTR_REFERENCE); const char *graph_file = crm_element_value(input->msg, F_CRM_TGRAPH); const char *graph_input = crm_element_value(input->msg, F_CRM_TGRAPH_INPUT); if (graph_file == NULL && input->xml == NULL) { crm_log_xml_err(input->msg, "Bad command"); register_fsa_error(C_FSA_INTERNAL, I_FAIL, NULL); return; } if (transition_graph->complete == FALSE) { crm_info("Another transition is already active"); abort_transition(INFINITY, tg_restart, "Transition Active", NULL); return; } if (fsa_pe_ref == NULL || safe_str_neq(fsa_pe_ref, ref)) { crm_info("Transition is redundant: %s vs. %s", crm_str(fsa_pe_ref), crm_str(ref)); abort_transition(INFINITY, tg_restart, "Transition Redundant", NULL); } graph_data = input->xml; if (graph_data == NULL && graph_file != NULL) { graph_data = filename2xml(graph_file); } if (is_timer_started(transition_timer)) { crm_debug("The transitioner wait for a transition timer"); return; } CRM_CHECK(graph_data != NULL, crm_err("Input raised by %s is invalid", msg_data->origin); crm_log_xml_err(input->msg, "Bad command"); return); destroy_graph(transition_graph); transition_graph = unpack_graph(graph_data, graph_input); CRM_CHECK(transition_graph != NULL, transition_graph = create_blank_graph(); return); crm_info("Processing graph %d (ref=%s) derived from %s", transition_graph->id, ref, graph_input); value = crm_element_value(graph_data, "failed-stop-offset"); if (value) { free(failed_stop_offset); failed_stop_offset = strdup(value); } value = crm_element_value(graph_data, "failed-start-offset"); if (value) { free(failed_start_offset); failed_start_offset = strdup(value); } trigger_graph(); print_graph(LOG_DEBUG_2, transition_graph); if (graph_data != input->xml) { free_xml(graph_data); } } }
void attrd_peer_update(crm_node_t *peer, xmlNode *xml, const char *host, bool filter) { bool update_both = FALSE; attribute_t *a; attribute_value_t *v = NULL; gboolean is_force_write = FALSE; const char *op = crm_element_value(xml, F_ATTRD_TASK); const char *attr = crm_element_value(xml, F_ATTRD_ATTRIBUTE); const char *value = crm_element_value(xml, F_ATTRD_VALUE); crm_element_value_int(xml, F_ATTRD_IS_FORCE_WRITE, &is_force_write); if (attr == NULL) { crm_warn("Could not update attribute: peer did not specify name"); return; } update_both = ((op == NULL) // ATTRD_OP_SYNC_RESPONSE has no F_ATTRD_TASK || safe_str_eq(op, ATTRD_OP_UPDATE_BOTH)); // Look up or create attribute entry a = g_hash_table_lookup(attributes, attr); if (a == NULL) { if (update_both || safe_str_eq(op, ATTRD_OP_UPDATE)) { a = create_attribute(xml); } else { crm_warn("Could not update %s: attribute not found", attr); return; } } // Update attribute dampening if (update_both || safe_str_eq(op, ATTRD_OP_UPDATE_DELAY)) { const char *dvalue = crm_element_value(xml, F_ATTRD_DAMPEN); int dampen = 0; if (dvalue == NULL) { crm_warn("Could not update %s: peer did not specify value for delay", attr); return; } dampen = crm_get_msec(dvalue); if (dampen < 0) { crm_warn("Could not update %s: invalid delay value %dms (%s)", attr, dampen, dvalue); return; } if (a->timeout_ms != dampen) { mainloop_timer_del(a->timer); a->timeout_ms = dampen; if (dampen > 0) { a->timer = mainloop_timer_add(attr, a->timeout_ms, FALSE, attribute_timer_cb, a); crm_info("Update attribute %s delay to %dms (%s)", attr, dampen, dvalue); } else { a->timer = NULL; crm_info("Update attribute %s to remove delay", attr); } /* If dampening changed, do an immediate write-out, * otherwise repeated dampening changes would prevent write-outs */ write_or_elect_attribute(a); } if (!update_both) { return; } } // If no host was specified, update all hosts recursively if (host == NULL) { GHashTableIter vIter; crm_debug("Setting %s for all hosts to %s", attr, value); xml_remove_prop(xml, F_ATTRD_HOST_ID); g_hash_table_iter_init(&vIter, a->values); while (g_hash_table_iter_next(&vIter, (gpointer *) & host, NULL)) { attrd_peer_update(peer, xml, host, filter); } return; } // Update attribute value for one host v = attrd_lookup_or_create_value(a->values, host, xml); if (filter && safe_str_neq(v->current, value) && safe_str_eq(host, attrd_cluster->uname)) { xmlNode *sync = create_xml_node(NULL, __FUNCTION__); crm_notice("%s[%s]: local value '%s' takes priority over '%s' from %s", attr, host, v->current, value, peer->uname); crm_xml_add(sync, F_ATTRD_TASK, ATTRD_OP_SYNC_RESPONSE); v = g_hash_table_lookup(a->values, host); build_attribute_xml(sync, attr, a->set, a->uuid, a->timeout_ms, a->user, a->is_private, v->nodename, v->nodeid, v->current, FALSE); attrd_xml_add_writer(sync); /* Broadcast in case any other nodes had the inconsistent value */ send_attrd_message(NULL, sync); free_xml(sync); } else if (safe_str_neq(v->current, value)) { crm_info("Setting %s[%s]: %s -> %s from %s", attr, host, v->current, value, peer->uname); free(v->current); v->current = (value? strdup(value) : NULL); a->changed = TRUE; // Write out new value or start dampening timer if (a->timeout_ms && a->timer) { crm_trace("Delayed write out (%dms) for %s", a->timeout_ms, attr); mainloop_timer_start(a->timer); } else { write_or_elect_attribute(a); } } else { if (is_force_write && a->timeout_ms && a->timer) { /* Save forced writing and set change flag. */ /* The actual attribute is written by Writer after election. */ crm_trace("Unchanged %s[%s] from %s is %s(Set the forced write flag)", attr, host, peer->uname, value); a->force_write = TRUE; } else { crm_trace("Unchanged %s[%s] from %s is %s", attr, host, peer->uname, value); } } /* Set the seen flag for attribute processing held only in the own node. */ v->seen = TRUE; /* If this is a cluster node whose node ID we are learning, remember it */ if ((v->nodeid == 0) && (v->is_remote == FALSE) && (crm_element_value_int(xml, F_ATTRD_HOST_ID, (int*)&v->nodeid) == 0)) { crm_node_t *known_peer = crm_get_peer(v->nodeid, host); crm_trace("Learned %s has node id %s", known_peer->uname, known_peer->uuid); if (attrd_election_won()) { write_attributes(FALSE, FALSE); } } }
static gboolean te_rsc_command(crm_graph_t * graph, crm_action_t * action) { /* never overwrite stop actions in the CIB with * anything other than completed results * * Writing pending stops makes it look like the * resource is running again */ xmlNode *cmd = NULL; xmlNode *rsc_op = NULL; gboolean rc = TRUE; gboolean no_wait = FALSE; gboolean is_local = FALSE; char *counter = NULL; const char *task = NULL; const char *value = NULL; const char *on_node = NULL; const char *router_node = NULL; const char *task_uuid = NULL; CRM_ASSERT(action != NULL); CRM_ASSERT(action->xml != NULL); action->executed = FALSE; on_node = crm_element_value(action->xml, XML_LRM_ATTR_TARGET); CRM_CHECK(on_node != NULL && strlen(on_node) != 0, crm_err("Corrupted command(id=%s) %s: no node", ID(action->xml), crm_str(task)); return FALSE); rsc_op = action->xml; task = crm_element_value(rsc_op, XML_LRM_ATTR_TASK); task_uuid = crm_element_value(action->xml, XML_LRM_ATTR_TASK_KEY); router_node = crm_element_value(rsc_op, XML_LRM_ATTR_ROUTER_NODE); if (!router_node) { router_node = on_node; } counter = generate_transition_key(transition_graph->id, action->id, get_target_rc(action), te_uuid); crm_xml_add(rsc_op, XML_ATTR_TRANSITION_KEY, counter); if (safe_str_eq(router_node, fsa_our_uname)) { is_local = TRUE; } value = crm_meta_value(action->params, XML_ATTR_TE_NOWAIT); if (crm_is_true(value)) { no_wait = TRUE; } crm_notice("Initiating %s operation %s%s on %s%s "CRM_XS" action %d", task, task_uuid, (is_local? " locally" : ""), on_node, (no_wait? " without waiting" : ""), action->id); cmd = create_request(CRM_OP_INVOKE_LRM, rsc_op, router_node, CRM_SYSTEM_LRMD, CRM_SYSTEM_TENGINE, NULL); if (is_local) { /* shortcut local resource commands */ ha_msg_input_t data = { .msg = cmd, .xml = rsc_op, }; fsa_data_t msg = { .id = 0, .data = &data, .data_type = fsa_dt_ha_msg, .fsa_input = I_NULL, .fsa_cause = C_FSA_INTERNAL, .actions = A_LRM_INVOKE, .origin = __FUNCTION__, }; do_lrm_invoke(A_LRM_INVOKE, C_FSA_INTERNAL, fsa_state, I_NULL, &msg); } else { rc = send_cluster_message(crm_get_peer(0, router_node), crm_msg_lrmd, cmd, TRUE); } free(counter); free_xml(cmd); action->executed = TRUE; if (rc == FALSE) { crm_err("Action %d failed: send", action->id); return FALSE; } else if (no_wait) { crm_info("Action %d confirmed - no wait", action->id); action->confirmed = TRUE; /* Just mark confirmed. * Don't bump the job count only to immediately decrement it */ update_graph(transition_graph, action); trigger_graph(); } else if (action->confirmed == TRUE) { crm_debug("Action %d: %s %s on %s(timeout %dms) was already confirmed.", action->id, task, task_uuid, on_node, action->timeout); } else { if (action->timeout <= 0) { crm_err("Action %d: %s %s on %s had an invalid timeout (%dms). Using %dms instead", action->id, task, task_uuid, on_node, action->timeout, graph->network_delay); action->timeout = graph->network_delay; } te_update_job_count(action, 1); te_start_action_timer(graph, action); } return TRUE; }
int crm_ipc_send(crm_ipc_t * client, xmlNode * message, enum crm_ipc_flags flags, int32_t ms_timeout, xmlNode ** reply) { long rc = 0; struct iovec *iov; static uint32_t id = 0; static int factor = 8; struct crm_ipc_response_header *header; crm_ipc_init(); if (client == NULL) { crm_notice("Invalid connection"); return -ENOTCONN; } else if (crm_ipc_connected(client) == FALSE) { /* Don't even bother */ crm_notice("Connection to %s closed", client->name); return -ENOTCONN; } if (ms_timeout == 0) { ms_timeout = 5000; } if (client->need_reply) { crm_trace("Trying again to obtain pending reply from %s", client->name); rc = qb_ipcc_recv(client->ipc, client->buffer, client->buf_size, ms_timeout); if (rc < 0) { crm_warn("Sending to %s (%p) is disabled until pending reply is received", client->name, client->ipc); return -EALREADY; } else { crm_notice("Lost reply from %s (%p) finally arrived, sending re-enabled", client->name, client->ipc); client->need_reply = FALSE; } } id++; CRM_LOG_ASSERT(id != 0); /* Crude wrap-around detection */ rc = crm_ipc_prepare(id, message, &iov, client->max_buf_size); if(rc < 0) { return rc; } header = iov[0].iov_base; header->flags |= flags; if(is_set(flags, crm_ipc_proxied)) { /* Don't look for a synchronous response */ clear_bit(flags, crm_ipc_client_response); } if(header->size_compressed) { if(factor < 10 && (client->max_buf_size / 10) < (rc / factor)) { crm_notice("Compressed message exceeds %d0%% of the configured ipc limit (%d bytes), " "consider setting PCMK_ipc_buffer to %d or higher", factor, client->max_buf_size, 2*client->max_buf_size); factor++; } } crm_trace("Sending from client: %s request id: %d bytes: %u timeout:%d msg...", client->name, header->qb.id, header->qb.size, ms_timeout); if (ms_timeout > 0 || is_not_set(flags, crm_ipc_client_response)) { rc = internal_ipc_send_request(client, iov, ms_timeout); if (rc <= 0) { crm_trace("Failed to send from client %s request %d with %u bytes...", client->name, header->qb.id, header->qb.size); goto send_cleanup; } else if (is_not_set(flags, crm_ipc_client_response)) { crm_trace("Message sent, not waiting for reply to %d from %s to %u bytes...", header->qb.id, client->name, header->qb.size); goto send_cleanup; } rc = internal_ipc_get_reply(client, header->qb.id, ms_timeout); if (rc < 0) { /* No reply, for now, disable sending * * The alternative is to close the connection since we don't know * how to detect and discard out-of-sequence replies * * TODO - implement the above */ client->need_reply = TRUE; } } else { rc = internal_ipc_send_recv(client, iov); } if (rc > 0) { struct crm_ipc_response_header *hdr = (struct crm_ipc_response_header *)(void*)client->buffer; crm_trace("Received response %d, size=%d, rc=%ld, text: %.200s", hdr->qb.id, hdr->qb.size, rc, crm_ipc_buffer(client)); if (reply) { *reply = string2xml(crm_ipc_buffer(client)); } } else { crm_trace("Response not received: rc=%ld, errno=%d", rc, errno); } send_cleanup: if (crm_ipc_connected(client) == FALSE) { crm_notice("Connection to %s closed: %s (%ld)", client->name, pcmk_strerror(rc), rc); } else if (rc == -ETIMEDOUT) { crm_warn("Request %d to %s (%p) failed: %s (%ld) after %dms", header->qb.id, client->name, client->ipc, pcmk_strerror(rc), rc, ms_timeout); crm_write_blackbox(0, NULL); } else if (rc <= 0) { crm_warn("Request %d to %s (%p) failed: %s (%ld)", header->qb.id, client->name, client->ipc, pcmk_strerror(rc), rc); } free(header); free(iov[1].iov_base); free(iov); return rc; }
gboolean cluster_connect_quorum(gboolean(*dispatch) (unsigned long long, gboolean), void (*destroy) (gpointer)) { int rc = -1; int fd = 0; int quorate = 0; uint32_t quorum_type = 0; struct mainloop_fd_callbacks quorum_fd_callbacks; quorum_fd_callbacks.dispatch = pcmk_quorum_dispatch; quorum_fd_callbacks.destroy = destroy; crm_debug("Configuring Pacemaker to obtain quorum from Corosync"); rc = quorum_initialize(&pcmk_quorum_handle, &quorum_callbacks, &quorum_type); if (rc != CS_OK) { crm_err("Could not connect to the Quorum API: %d", rc); goto bail; } else if (quorum_type != QUORUM_SET) { crm_err("Corosync quorum is not configured"); goto bail; } rc = quorum_getquorate(pcmk_quorum_handle, &quorate); if (rc != CS_OK) { crm_err("Could not obtain the current Quorum API state: %d", rc); goto bail; } if (quorate) { crm_notice("Quorum acquired"); } else { crm_warn("Quorum lost"); } quorum_app_callback = dispatch; crm_have_quorum = quorate; rc = quorum_trackstart(pcmk_quorum_handle, CS_TRACK_CHANGES | CS_TRACK_CURRENT); if (rc != CS_OK) { crm_err("Could not setup Quorum API notifications: %d", rc); goto bail; } rc = quorum_fd_get(pcmk_quorum_handle, &fd); if (rc != CS_OK) { crm_err("Could not obtain the Quorum API connection: %d", rc); goto bail; } mainloop_add_fd("quorum", G_PRIORITY_HIGH, fd, dispatch, &quorum_fd_callbacks); corosync_initialize_nodelist(NULL, FALSE, NULL); bail: if (rc != CS_OK) { quorum_finalize(pcmk_quorum_handle); return FALSE; } return TRUE; }
static char *find_cib_loadfile(void) { DIR *dp; struct dirent *entry; struct stat statbuf; char *match = NULL; dp = opendir("/proc"); if (!dp) { /* no proc directory to search through */ crm_notice("Can not read /proc directory to track existing components"); return FALSE; } while ((entry = readdir(dp)) != NULL) { char procpath[128]; char value[64]; char key[16]; FILE *file; int pid; strcpy(procpath, "/proc/"); /* strlen("/proc/") + strlen("/status") + 1 = 14 * 128 - 14 = 114 */ strncat(procpath, entry->d_name, 114); if (lstat(procpath, &statbuf)) { continue; } if (!S_ISDIR(statbuf.st_mode) || !isdigit(entry->d_name[0])) { continue; } strcat(procpath, "/status"); file = fopen(procpath, "r"); if (!file) { continue; } if (fscanf(file, "%15s%63s", key, value) != 2) { fclose(file); continue; } fclose(file); if (safe_str_neq("cib", value)) { continue; } pid = atoi(entry->d_name); if (pid <= 0) { continue; } match = g_strdup_printf("/proc/%d/stat", pid); break; } closedir(dp); return match; }