diff --git a/daemons/controld/controld_membership.c b/daemons/controld/controld_membership.c index 866741d93d..b17f781660 100644 --- a/daemons/controld/controld_membership.c +++ b/daemons/controld/controld_membership.c @@ -1,468 +1,468 @@ /* * Copyright 2004-2025 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ /* put these first so that uuid_t is defined without conflicts */ #include #include #include #include #include #include #include void post_cache_update(int instance); extern gboolean check_join_state(enum crmd_fsa_state cur_state, const char *source); static void reap_dead_nodes(gpointer key, gpointer value, gpointer user_data) { pcmk__node_status_t *node = value; if (pcmk__cluster_is_node_active(node)) { return; } crm_update_peer_join(__func__, node, controld_join_none); if ((node != NULL) && (node->name != NULL)) { if (controld_is_local_node(node->name)) { crm_err("We're not part of the cluster anymore"); register_fsa_input(C_FSA_INTERNAL, I_ERROR, NULL); } else if (!AM_I_DC && pcmk__str_eq(node->name, controld_globals.dc_name, pcmk__str_casei)) { crm_warn("Our DC node (%s) left the cluster", node->name); register_fsa_input(C_FSA_INTERNAL, I_ELECTION, NULL); } } if ((controld_globals.fsa_state == S_INTEGRATION) || (controld_globals.fsa_state == S_FINALIZE_JOIN)) { check_join_state(controld_globals.fsa_state, __func__); } if ((node != NULL) && (node->xml_id != NULL)) { fail_incompletable_actions(controld_globals.transition_graph, node->xml_id); } } void post_cache_update(int instance) { xmlNode *no_op = NULL; controld_globals.peer_seq = instance; crm_debug("Updated cache after membership event %d.", instance); g_hash_table_foreach(pcmk__peer_cache, reap_dead_nodes, NULL); controld_set_fsa_input_flags(R_MEMBERSHIP); if (AM_I_DC) { populate_cib_nodes(node_update_quick | node_update_cluster | node_update_peer | node_update_expected, __func__); } /* * If we lost nodes, we should re-check the election status * Safe to call outside of an election */ controld_set_fsa_action_flags(A_ELECTION_CHECK); controld_trigger_fsa(); /* Membership changed, remind everyone we're here. * This will aid detection of duplicate DCs */ no_op = pcmk__new_request(pcmk_ipc_controld, (AM_I_DC? CRM_SYSTEM_DC : CRM_SYSTEM_CRMD), NULL, CRM_SYSTEM_CRMD, CRM_OP_NOOP, NULL); pcmk__cluster_send_message(NULL, pcmk_ipc_controld, no_op); pcmk__xml_free(no_op); } static void crmd_node_update_complete(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data) { fsa_data_t *msg_data = NULL; if (rc == pcmk_ok) { crm_trace("Node update %d complete", call_id); } else if(call_id < pcmk_ok) { crm_err("Node update failed: %s (%d)", pcmk_strerror(call_id), call_id); crm_log_xml_debug(msg, "failed"); register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); } else { crm_err("Node update %d failed: %s (%d)", call_id, pcmk_strerror(rc), rc); crm_log_xml_debug(msg, "failed"); register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); } } /*! * \internal * \brief Create an XML node state tag with updates * * \param[in,out] node Node whose state will be used for update * \param[in] flags Bitmask of node_update_flags indicating what to update * \param[in,out] parent XML node to contain update (or NULL) * \param[in] source Who requested the update (only used for logging) * * \return Pointer to created node state tag */ xmlNode * create_node_state_update(pcmk__node_status_t *node, int flags, xmlNode *parent, const char *source) { // @TODO Ensure all callers handle NULL returns const char *value = NULL; xmlNode *node_state; if (!node->state) { crm_info("Node update for %s cancelled: no state, not seen yet", node->name); return NULL; } node_state = pcmk__xe_create(parent, PCMK__XE_NODE_STATE); if (pcmk_is_set(node->flags, pcmk__node_status_remote)) { pcmk__xe_set_bool_attr(node_state, PCMK_XA_REMOTE_NODE, true); } if (crm_xml_add(node_state, PCMK_XA_ID, pcmk__cluster_get_xml_id(node)) == NULL) { crm_info("Node update for %s cancelled: no ID", node->name); pcmk__xml_free(node_state); return NULL; } crm_xml_add(node_state, PCMK_XA_UNAME, node->name); if ((flags & node_update_cluster) && node->state) { if (compare_version(controld_globals.dc_version, "3.18.0") >= 0) { // A value 0 means the node is not a cluster member. - crm_xml_add_ll(node_state, PCMK__XA_IN_CCM, node->when_member); + pcmk__xe_set_ll(node_state, PCMK__XA_IN_CCM, node->when_member); } else { pcmk__xe_set_bool_attr(node_state, PCMK__XA_IN_CCM, pcmk__str_eq(node->state, PCMK_VALUE_MEMBER, pcmk__str_none)); } } if (!pcmk_is_set(node->flags, pcmk__node_status_remote)) { if (flags & node_update_peer) { if (compare_version(controld_globals.dc_version, "3.18.0") >= 0) { // A value 0 means the peer is offline in CPG. - crm_xml_add_ll(node_state, PCMK_XA_CRMD, node->when_online); + pcmk__xe_set_ll(node_state, PCMK_XA_CRMD, node->when_online); } else { // @COMPAT DCs < 2.1.7 use online/offline rather than timestamp value = PCMK_VALUE_OFFLINE; if (pcmk_is_set(node->processes, crm_get_cluster_proc())) { value = PCMK_VALUE_ONLINE; } crm_xml_add(node_state, PCMK_XA_CRMD, value); } } if (flags & node_update_join) { if (controld_get_join_phase(node) <= controld_join_none) { value = CRMD_JOINSTATE_DOWN; } else { value = CRMD_JOINSTATE_MEMBER; } crm_xml_add(node_state, PCMK__XA_JOIN, value); } if (flags & node_update_expected) { crm_xml_add(node_state, PCMK_XA_EXPECTED, node->expected); } } crm_xml_add(node_state, PCMK_XA_CRM_DEBUG_ORIGIN, source); return node_state; } static void remove_conflicting_node_callback(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data) { char *node_uuid = user_data; do_crm_log_unlikely(rc == 0 ? LOG_DEBUG : LOG_NOTICE, "Deletion of the unknown conflicting node \"%s\": %s (rc=%d)", node_uuid, pcmk_strerror(rc), rc); } static void search_conflicting_node_callback(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data) { char *new_node_uuid = user_data; xmlNode *node_xml = NULL; if (rc != pcmk_ok) { if (rc != -ENXIO) { crm_notice("Searching conflicting nodes for %s failed: %s (%d)", new_node_uuid, pcmk_strerror(rc), rc); } return; } else if (output == NULL) { return; } if (pcmk__xe_is(output, PCMK_XE_NODE)) { node_xml = output; } else { node_xml = pcmk__xe_first_child(output, PCMK_XE_NODE, NULL, NULL); } for (; node_xml != NULL; node_xml = pcmk__xe_next(node_xml, PCMK_XE_NODE)) { const char *node_uuid = NULL; const char *node_uname = NULL; GHashTableIter iter; pcmk__node_status_t *node = NULL; gboolean known = FALSE; node_uuid = pcmk__xe_get(node_xml, PCMK_XA_ID); node_uname = pcmk__xe_get(node_xml, PCMK_XA_UNAME); if (node_uuid == NULL || node_uname == NULL) { continue; } g_hash_table_iter_init(&iter, pcmk__peer_cache); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) { if ((node != NULL) && pcmk__str_eq(node->xml_id, node_uuid, pcmk__str_casei) && pcmk__str_eq(node->name, node_uname, pcmk__str_casei)) { known = TRUE; break; } } if (known == FALSE) { cib_t *cib_conn = controld_globals.cib_conn; int delete_call_id = 0; xmlNode *node_state_xml = NULL; crm_notice("Deleting unknown node %s/%s which has conflicting uname with %s", node_uuid, node_uname, new_node_uuid); delete_call_id = cib_conn->cmds->remove(cib_conn, PCMK_XE_NODES, node_xml, cib_none); fsa_register_cib_callback(delete_call_id, pcmk__str_copy(node_uuid), remove_conflicting_node_callback); node_state_xml = pcmk__xe_create(NULL, PCMK__XE_NODE_STATE); crm_xml_add(node_state_xml, PCMK_XA_ID, node_uuid); crm_xml_add(node_state_xml, PCMK_XA_UNAME, node_uname); delete_call_id = cib_conn->cmds->remove(cib_conn, PCMK_XE_STATUS, node_state_xml, cib_none); fsa_register_cib_callback(delete_call_id, pcmk__str_copy(node_uuid), remove_conflicting_node_callback); pcmk__xml_free(node_state_xml); } } } static void node_list_update_callback(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data) { fsa_data_t *msg_data = NULL; if(call_id < pcmk_ok) { crm_err("Node list update failed: %s (%d)", pcmk_strerror(call_id), call_id); crm_log_xml_debug(msg, "update:failed"); register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); } else if(rc < pcmk_ok) { crm_err("Node update %d failed: %s (%d)", call_id, pcmk_strerror(rc), rc); crm_log_xml_debug(msg, "update:failed"); register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); } } void populate_cib_nodes(enum node_update_flags flags, const char *source) { cib_t *cib_conn = controld_globals.cib_conn; int call_id = 0; gboolean from_hashtable = TRUE; xmlNode *node_list = pcmk__xe_create(NULL, PCMK_XE_NODES); #if SUPPORT_COROSYNC if (!pcmk_is_set(flags, node_update_quick) && (pcmk_get_cluster_layer() == pcmk_cluster_layer_corosync)) { from_hashtable = pcmk__corosync_add_nodes(node_list); } #endif if (from_hashtable) { GHashTableIter iter; pcmk__node_status_t *node = NULL; GString *xpath = NULL; g_hash_table_iter_init(&iter, pcmk__peer_cache); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) { xmlNode *new_node = NULL; if ((node->xml_id != NULL) && (node->name != NULL)) { crm_trace("Creating node entry for %s/%s", node->name, node->xml_id); if (xpath == NULL) { xpath = g_string_sized_new(512); } else { g_string_truncate(xpath, 0); } /* We need both to be valid */ new_node = pcmk__xe_create(node_list, PCMK_XE_NODE); crm_xml_add(new_node, PCMK_XA_ID, node->xml_id); crm_xml_add(new_node, PCMK_XA_UNAME, node->name); /* Search and remove unknown nodes with the conflicting uname from CIB */ pcmk__g_strcat(xpath, "/" PCMK_XE_CIB "/" PCMK_XE_CONFIGURATION "/" PCMK_XE_NODES "/" PCMK_XE_NODE "[@" PCMK_XA_UNAME "='", node->name, "']" "[@" PCMK_XA_ID "!='", node->xml_id, "']", NULL); call_id = cib_conn->cmds->query(cib_conn, (const char *) xpath->str, NULL, cib_xpath); fsa_register_cib_callback(call_id, pcmk__str_copy(node->xml_id), search_conflicting_node_callback); } } if (xpath != NULL) { g_string_free(xpath, TRUE); } } crm_trace("Populating section from %s", from_hashtable ? "hashtable" : "cluster"); if ((controld_update_cib(PCMK_XE_NODES, node_list, cib_none, node_list_update_callback) == pcmk_rc_ok) && (pcmk__peer_cache != NULL) && AM_I_DC) { /* * There is no need to update the local CIB with our values if * we've not seen valid membership data */ GHashTableIter iter; pcmk__node_status_t *node = NULL; pcmk__xml_free(node_list); node_list = pcmk__xe_create(NULL, PCMK_XE_STATUS); g_hash_table_iter_init(&iter, pcmk__peer_cache); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) { create_node_state_update(node, flags, node_list, source); } if (pcmk__remote_peer_cache != NULL) { g_hash_table_iter_init(&iter, pcmk__remote_peer_cache); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) { create_node_state_update(node, flags, node_list, source); } } controld_update_cib(PCMK_XE_STATUS, node_list, cib_none, crmd_node_update_complete); } pcmk__xml_free(node_list); } static void cib_quorum_update_complete(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data) { fsa_data_t *msg_data = NULL; if (rc == pcmk_ok) { crm_trace("Quorum update %d complete", call_id); } else { crm_err("Quorum update %d failed: %s (%d)", call_id, pcmk_strerror(rc), rc); crm_log_xml_debug(msg, "failed"); register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); } } void crm_update_quorum(gboolean quorum, gboolean force_update) { bool has_quorum = pcmk_is_set(controld_globals.flags, controld_has_quorum); if (quorum) { controld_set_global_flags(controld_ever_had_quorum); } else if (pcmk_all_flags_set(controld_globals.flags, controld_ever_had_quorum |controld_no_quorum_panic)) { pcmk__panic("Quorum lost"); } if (AM_I_DC && ((has_quorum && !quorum) || (!has_quorum && quorum) || force_update)) { xmlNode *update = NULL; update = pcmk__xe_create(NULL, PCMK_XE_CIB); crm_xml_add_int(update, PCMK_XA_HAVE_QUORUM, quorum); crm_xml_add(update, PCMK_XA_DC_UUID, controld_globals.our_uuid); crm_debug("Updating quorum status to %s", pcmk__btoa(quorum)); controld_update_cib(PCMK_XE_CIB, update, cib_none, cib_quorum_update_complete); pcmk__xml_free(update); /* Quorum changes usually cause a new transition via other activity: * quorum gained via a node joining will abort via the node join, * and quorum lost via a node leaving will usually abort via resource * activity and/or fencing. * * However, it is possible that nothing else causes a transition (e.g. * someone forces quorum via corosync-cmaptcl, or quorum is lost due to * a node in standby shutting down cleanly), so here ensure a new * transition is triggered. */ if (quorum) { /* If quorum was gained, abort after a short delay, in case multiple * nodes are joining around the same time, so the one that brings us * to quorum doesn't cause all the remaining ones to be fenced. */ abort_after_delay(PCMK_SCORE_INFINITY, pcmk__graph_restart, "Quorum gained", 5000); } else { abort_transition(PCMK_SCORE_INFINITY, pcmk__graph_restart, "Quorum lost", NULL); } } if (quorum) { controld_set_global_flags(controld_has_quorum); } else { controld_clear_global_flags(controld_has_quorum); } } diff --git a/daemons/controld/controld_messages.c b/daemons/controld/controld_messages.c index 7398de55b5..c88b95e948 100644 --- a/daemons/controld/controld_messages.c +++ b/daemons/controld/controld_messages.c @@ -1,1384 +1,1384 @@ /* * Copyright 2004-2025 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include // PRIx64 #include // uint64_t #include #include #include #include #include #include #include #include #include static enum crmd_fsa_input handle_message(xmlNode *msg, enum crmd_fsa_cause cause); static xmlNode* create_ping_reply(const xmlNode *msg); static void handle_response(xmlNode *stored_msg); static enum crmd_fsa_input handle_request(xmlNode *stored_msg, enum crmd_fsa_cause cause); static enum crmd_fsa_input handle_shutdown_request(xmlNode *stored_msg); static void send_msg_via_ipc(xmlNode * msg, const char *sys, const char *src); /* debug only, can wrap all it likes */ static int last_data_id = 0; void register_fsa_error_adv(enum crmd_fsa_cause cause, enum crmd_fsa_input input, fsa_data_t * cur_data, void *new_data, const char *raised_from) { /* save the current actions if any */ if (controld_globals.fsa_actions != A_NOTHING) { register_fsa_input_adv(cur_data ? cur_data->fsa_cause : C_FSA_INTERNAL, I_NULL, cur_data ? cur_data->data : NULL, controld_globals.fsa_actions, TRUE, __func__); } /* reset the action list */ crm_info("Resetting the current action list"); fsa_dump_actions(controld_globals.fsa_actions, "Drop"); controld_globals.fsa_actions = A_NOTHING; /* register the error */ register_fsa_input_adv(cause, input, new_data, A_NOTHING, TRUE, raised_from); } void register_fsa_input_adv(enum crmd_fsa_cause cause, enum crmd_fsa_input input, void *data, uint64_t with_actions, gboolean prepend, const char *raised_from) { unsigned old_len = g_list_length(controld_globals.fsa_message_queue); fsa_data_t *fsa_data = NULL; if (raised_from == NULL) { raised_from = ""; } if (input == I_NULL && with_actions == A_NOTHING /* && data == NULL */ ) { /* no point doing anything */ crm_err("Cannot add entry to queue: no input and no action"); return; } if (input == I_WAIT_FOR_EVENT) { controld_set_global_flags(controld_fsa_is_stalled); crm_debug("Stalling the FSA pending further input: source=%s cause=%s data=%p queue=%d", raised_from, fsa_cause2string(cause), data, old_len); if (old_len > 0) { fsa_dump_queue(LOG_TRACE); prepend = FALSE; } if (data == NULL) { controld_set_fsa_action_flags(with_actions); fsa_dump_actions(with_actions, "Restored"); return; } /* Store everything in the new event and reset * controld_globals.fsa_actions */ with_actions |= controld_globals.fsa_actions; controld_globals.fsa_actions = A_NOTHING; } last_data_id++; crm_trace("%s %s FSA input %d (%s) due to %s, %s data", raised_from, (prepend? "prepended" : "appended"), last_data_id, fsa_input2string(input), fsa_cause2string(cause), (data? "with" : "without")); fsa_data = pcmk__assert_alloc(1, sizeof(fsa_data_t)); fsa_data->id = last_data_id; fsa_data->fsa_input = input; fsa_data->fsa_cause = cause; fsa_data->origin = raised_from; fsa_data->data = NULL; fsa_data->data_type = fsa_dt_none; fsa_data->actions = with_actions; if (with_actions != A_NOTHING) { crm_trace("Adding actions %.16" PRIx64 " to input", with_actions); } if (data != NULL) { switch (cause) { case C_FSA_INTERNAL: case C_CRMD_STATUS_CALLBACK: case C_IPC_MESSAGE: case C_HA_MESSAGE: CRM_CHECK(((ha_msg_input_t *) data)->msg != NULL, crm_err("Bogus data from %s", raised_from)); crm_trace("Copying %s data from %s as cluster message data", fsa_cause2string(cause), raised_from); fsa_data->data = copy_ha_msg_input(data); fsa_data->data_type = fsa_dt_ha_msg; break; case C_LRM_OP_CALLBACK: crm_trace("Copying %s data from %s as lrmd_event_data_t", fsa_cause2string(cause), raised_from); fsa_data->data = lrmd_copy_event((lrmd_event_data_t *) data); fsa_data->data_type = fsa_dt_lrm; break; case C_TIMER_POPPED: case C_SHUTDOWN: case C_UNKNOWN: case C_STARTUP: crm_crit("Copying %s data (from %s) is not yet implemented", fsa_cause2string(cause), raised_from); crmd_exit(CRM_EX_SOFTWARE); break; } } /* make sure to free it properly later */ if (prepend) { controld_globals.fsa_message_queue = g_list_prepend(controld_globals.fsa_message_queue, fsa_data); } else { controld_globals.fsa_message_queue = g_list_append(controld_globals.fsa_message_queue, fsa_data); } crm_trace("FSA message queue length is %d", g_list_length(controld_globals.fsa_message_queue)); /* fsa_dump_queue(LOG_TRACE); */ if (old_len == g_list_length(controld_globals.fsa_message_queue)) { crm_err("Couldn't add message to the queue"); } if (input != I_WAIT_FOR_EVENT) { controld_trigger_fsa(); } } void fsa_dump_queue(int log_level) { int offset = 0; for (GList *iter = controld_globals.fsa_message_queue; iter != NULL; iter = iter->next) { fsa_data_t *data = (fsa_data_t *) iter->data; do_crm_log_unlikely(log_level, "queue[%d.%d]: input %s raised by %s(%p.%d)\t(cause=%s)", offset++, data->id, fsa_input2string(data->fsa_input), data->origin, data->data, data->data_type, fsa_cause2string(data->fsa_cause)); } } ha_msg_input_t * copy_ha_msg_input(ha_msg_input_t * orig) { xmlNode *wrapper = NULL; ha_msg_input_t *copy = pcmk__assert_alloc(1, sizeof(ha_msg_input_t)); copy->msg = (orig != NULL)? pcmk__xml_copy(NULL, orig->msg) : NULL; wrapper = pcmk__xe_first_child(copy->msg, PCMK__XE_CRM_XML, NULL, NULL); copy->xml = pcmk__xe_first_child(wrapper, NULL, NULL, NULL); return copy; } void delete_fsa_input(fsa_data_t * fsa_data) { lrmd_event_data_t *op = NULL; xmlNode *foo = NULL; if (fsa_data == NULL) { return; } crm_trace("About to free %s data", fsa_cause2string(fsa_data->fsa_cause)); if (fsa_data->data != NULL) { switch (fsa_data->data_type) { case fsa_dt_ha_msg: delete_ha_msg_input(fsa_data->data); break; case fsa_dt_xml: foo = fsa_data->data; pcmk__xml_free(foo); break; case fsa_dt_lrm: op = (lrmd_event_data_t *) fsa_data->data; lrmd_free_event(op); break; case fsa_dt_none: if (fsa_data->data != NULL) { crm_err("Don't know how to free %s data from %s", fsa_cause2string(fsa_data->fsa_cause), fsa_data->origin); crmd_exit(CRM_EX_SOFTWARE); } break; } crm_trace("%s data freed", fsa_cause2string(fsa_data->fsa_cause)); } free(fsa_data); } /* returns the next message */ fsa_data_t * get_message(void) { fsa_data_t *message = (fsa_data_t *) controld_globals.fsa_message_queue->data; controld_globals.fsa_message_queue = g_list_remove(controld_globals.fsa_message_queue, message); crm_trace("Processing input %d", message->id); return message; } void * fsa_typed_data_adv(fsa_data_t * fsa_data, enum fsa_data_type a_type, const char *caller) { void *ret_val = NULL; if (fsa_data == NULL) { crm_err("%s: No FSA data available", caller); } else if (fsa_data->data == NULL) { crm_err("%s: No message data available. Origin: %s", caller, fsa_data->origin); } else if (fsa_data->data_type != a_type) { crm_crit("%s: Message data was the wrong type! %d vs. requested=%d. Origin: %s", caller, fsa_data->data_type, a_type, fsa_data->origin); pcmk__assert(fsa_data->data_type == a_type); } else { ret_val = fsa_data->data; } return ret_val; } /* A_MSG_ROUTE */ void do_msg_route(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data) { ha_msg_input_t *input = fsa_typed_data(fsa_dt_ha_msg); route_message(msg_data->fsa_cause, input->msg); } void route_message(enum crmd_fsa_cause cause, xmlNode * input) { ha_msg_input_t fsa_input; enum crmd_fsa_input result = I_NULL; fsa_input.msg = input; CRM_CHECK(cause == C_IPC_MESSAGE || cause == C_HA_MESSAGE, return); /* try passing the buck first */ if (relay_message(input, cause == C_IPC_MESSAGE)) { return; } /* handle locally */ result = handle_message(input, cause); /* done or process later? */ switch (result) { case I_NULL: case I_ROUTER: case I_NODE_JOIN: case I_JOIN_REQUEST: case I_JOIN_RESULT: break; default: /* Defering local processing of message */ register_fsa_input_later(cause, result, &fsa_input); return; } if (result != I_NULL) { /* add to the front of the queue */ register_fsa_input(cause, result, &fsa_input); } } gboolean relay_message(xmlNode * msg, gboolean originated_locally) { enum pcmk_ipc_server dest = pcmk_ipc_unknown; bool is_for_dc = false; bool is_for_dcib = false; bool is_for_te = false; bool is_for_crm = false; bool is_for_cib = false; bool is_local = false; bool broadcast = false; const char *host_to = NULL; const char *sys_to = NULL; const char *sys_from = NULL; const char *type = NULL; const char *task = NULL; const char *ref = NULL; pcmk__node_status_t *node_to = NULL; CRM_CHECK(msg != NULL, return TRUE); host_to = pcmk__xe_get(msg, PCMK__XA_CRM_HOST_TO); sys_to = pcmk__xe_get(msg, PCMK__XA_CRM_SYS_TO); sys_from = pcmk__xe_get(msg, PCMK__XA_CRM_SYS_FROM); type = pcmk__xe_get(msg, PCMK__XA_T); task = pcmk__xe_get(msg, PCMK__XA_CRM_TASK); ref = pcmk__xe_get(msg, PCMK_XA_REFERENCE); broadcast = pcmk__str_empty(host_to); if (ref == NULL) { ref = "without reference ID"; } if (pcmk__str_eq(task, CRM_OP_HELLO, pcmk__str_casei)) { crm_trace("Received hello %s from %s (no processing needed)", ref, pcmk__s(sys_from, "unidentified source")); crm_log_xml_trace(msg, "hello"); return TRUE; } // Require message type (set by pcmk__new_request()) if (!pcmk__str_eq(type, PCMK__VALUE_CRMD, pcmk__str_none)) { crm_warn("Ignoring invalid message %s with type '%s' " "(not '" PCMK__VALUE_CRMD "')", ref, pcmk__s(type, "")); crm_log_xml_trace(msg, "ignored"); return TRUE; } // Require a destination subsystem (also set by pcmk__new_request()) if (sys_to == NULL) { crm_warn("Ignoring invalid message %s with no " PCMK__XA_CRM_SYS_TO, ref); crm_log_xml_trace(msg, "ignored"); return TRUE; } // Get the message type appropriate to the destination subsystem if (pcmk_get_cluster_layer() == pcmk_cluster_layer_corosync) { dest = pcmk__parse_server(sys_to); if (dest == pcmk_ipc_unknown) { /* Unrecognized value, use a sane default * * @TODO Maybe we should bail instead */ dest = pcmk_ipc_controld; } } is_for_dc = (strcasecmp(CRM_SYSTEM_DC, sys_to) == 0); is_for_dcib = (strcasecmp(CRM_SYSTEM_DCIB, sys_to) == 0); is_for_te = (strcasecmp(CRM_SYSTEM_TENGINE, sys_to) == 0); is_for_cib = (strcasecmp(CRM_SYSTEM_CIB, sys_to) == 0); is_for_crm = (strcasecmp(CRM_SYSTEM_CRMD, sys_to) == 0); // Check whether message should be processed locally is_local = false; if (broadcast) { if (is_for_dc || is_for_te) { is_local = false; } else if (is_for_crm) { if (pcmk__strcase_any_of(task, CRM_OP_NODE_INFO, PCMK__CONTROLD_CMD_NODES, NULL)) { /* Node info requests do not specify a host, which is normally * treated as "all hosts", because the whole point is that the * client may not know the local node name. Always handle these * requests locally. */ is_local = true; } else { is_local = !originated_locally; } } else { is_local = true; } } else if (controld_is_local_node(host_to)) { is_local = true; } else if (is_for_crm && pcmk__str_eq(task, CRM_OP_LRM_DELETE, pcmk__str_casei)) { xmlNode *wrapper = pcmk__xe_first_child(msg, PCMK__XE_CRM_XML, NULL, NULL); xmlNode *msg_data = pcmk__xe_first_child(wrapper, NULL, NULL, NULL); const char *mode = pcmk__xe_get(msg_data, PCMK__XA_MODE); if (pcmk__str_eq(mode, PCMK__VALUE_CIB, pcmk__str_none)) { // Local delete of an offline node's resource history is_local = true; } } // If is for DC and DC is not yet selected if (is_for_dc && pcmk__str_eq(task, CRM_OP_PING, pcmk__str_casei) && (controld_globals.dc_name == NULL)) { xmlNode *reply = create_ping_reply(msg); sys_to = pcmk__xe_get(reply, PCMK__XA_CRM_SYS_TO); // Explicitly leave src empty. It indicates that dc is "not yet selected" send_msg_via_ipc(reply, sys_to, NULL); pcmk__xml_free(reply); return TRUE; } // Check whether message should be relayed if (is_for_dc || is_for_dcib || is_for_te) { if (AM_I_DC) { if (is_for_te) { crm_trace("Route message %s locally as transition request", ref); crm_log_xml_trace(msg, sys_to); send_msg_via_ipc(msg, sys_to, controld_globals.cluster->priv->node_name); return TRUE; // No further processing of message is needed } crm_trace("Route message %s locally as DC request", ref); return FALSE; // More to be done by caller } if (originated_locally && !pcmk__strcase_any_of(sys_from, CRM_SYSTEM_PENGINE, CRM_SYSTEM_TENGINE, NULL)) { crm_trace("Relay message %s to DC (via %s)", ref, pcmk__s(host_to, "broadcast")); crm_log_xml_trace(msg, "relayed"); if (!broadcast) { node_to = pcmk__get_node(0, host_to, NULL, pcmk__node_search_cluster_member); } pcmk__cluster_send_message(node_to, dest, msg); return TRUE; } /* Transition engine and scheduler messages are sent only to the DC on * the same node. If we are no longer the DC, discard this message. */ crm_trace("Ignoring message %s because we are no longer DC", ref); crm_log_xml_trace(msg, "ignored"); return TRUE; // No further processing of message is needed } if (is_local) { if (is_for_crm || is_for_cib) { crm_trace("Route message %s locally as controller request", ref); return FALSE; // More to be done by caller } crm_trace("Relay message %s locally to %s", ref, sys_to); crm_log_xml_trace(msg, "IPC-relay"); send_msg_via_ipc(msg, sys_to, controld_globals.cluster->priv->node_name); return TRUE; } if (!broadcast) { node_to = pcmk__search_node_caches(0, host_to, NULL, pcmk__node_search_cluster_member); if (node_to == NULL) { crm_warn("Ignoring message %s because node %s is unknown", ref, host_to); crm_log_xml_trace(msg, "ignored"); return TRUE; } } crm_trace("Relay message %s to %s", ref, pcmk__s(host_to, "all peers")); crm_log_xml_trace(msg, "relayed"); pcmk__cluster_send_message(node_to, dest, msg); return TRUE; } // Return true if field contains a positive integer static bool authorize_version(xmlNode *message_data, const char *field, const char *client_name, const char *ref, const char *uuid) { const char *version = pcmk__xe_get(message_data, field); long long version_num; if ((pcmk__scan_ll(version, &version_num, -1LL) != pcmk_rc_ok) || (version_num < 0LL)) { crm_warn("Rejected IPC hello from %s: '%s' is not a valid protocol %s " QB_XS " ref=%s uuid=%s", client_name, ((version == NULL)? "" : version), field, (ref? ref : "none"), uuid); return false; } return true; } /*! * \internal * \brief Check whether a client IPC message is acceptable * * If a given client IPC message is a hello, "authorize" it by ensuring it has * valid information such as a protocol version, and return false indicating * that nothing further needs to be done with the message. If the message is not * a hello, just return true to indicate it needs further processing. * * \param[in] client_msg XML of IPC message * \param[in,out] curr_client If IPC is not proxied, client that sent message * \param[in] proxy_session If IPC is proxied, the session ID * * \return true if message needs further processing, false if it doesn't */ bool controld_authorize_ipc_message(const xmlNode *client_msg, pcmk__client_t *curr_client, const char *proxy_session) { xmlNode *wrapper = NULL; xmlNode *message_data = NULL; const char *client_name = NULL; const char *op = pcmk__xe_get(client_msg, PCMK__XA_CRM_TASK); const char *ref = pcmk__xe_get(client_msg, PCMK_XA_REFERENCE); const char *uuid = (curr_client? curr_client->id : proxy_session); if (uuid == NULL) { crm_warn("IPC message from client rejected: No client identifier " QB_XS " ref=%s", (ref? ref : "none")); goto rejected; } if (!pcmk__str_eq(CRM_OP_HELLO, op, pcmk__str_casei)) { // Only hello messages need to be authorized return true; } wrapper = pcmk__xe_first_child(client_msg, PCMK__XE_CRM_XML, NULL, NULL); message_data = pcmk__xe_first_child(wrapper, NULL, NULL, NULL); client_name = pcmk__xe_get(message_data, PCMK__XA_CLIENT_NAME); if (pcmk__str_empty(client_name)) { crm_warn("IPC hello from client rejected: No client name", QB_XS " ref=%s uuid=%s", (ref? ref : "none"), uuid); goto rejected; } if (!authorize_version(message_data, PCMK__XA_MAJOR_VERSION, client_name, ref, uuid)) { goto rejected; } if (!authorize_version(message_data, PCMK__XA_MINOR_VERSION, client_name, ref, uuid)) { goto rejected; } crm_trace("Validated IPC hello from client %s", client_name); crm_log_xml_trace(client_msg, "hello"); if (curr_client) { curr_client->userdata = pcmk__str_copy(client_name); } controld_trigger_fsa(); return false; rejected: crm_log_xml_trace(client_msg, "rejected"); if (curr_client) { qb_ipcs_disconnect(curr_client->ipcs); } return false; } static enum crmd_fsa_input handle_message(xmlNode *msg, enum crmd_fsa_cause cause) { const char *type = NULL; CRM_CHECK(msg != NULL, return I_NULL); type = pcmk__xe_get(msg, PCMK__XA_SUBT); if (pcmk__str_eq(type, PCMK__VALUE_REQUEST, pcmk__str_none)) { return handle_request(msg, cause); } if (pcmk__str_eq(type, PCMK__VALUE_RESPONSE, pcmk__str_none)) { handle_response(msg); return I_NULL; } crm_warn("Ignoring message with unknown " PCMK__XA_SUBT" '%s'", pcmk__s(type, "")); crm_log_xml_trace(msg, "bad"); return I_NULL; } static enum crmd_fsa_input handle_failcount_op(xmlNode * stored_msg) { const char *rsc = NULL; const char *uname = NULL; const char *op = NULL; char *interval_spec = NULL; guint interval_ms = 0; gboolean is_remote_node = FALSE; xmlNode *wrapper = pcmk__xe_first_child(stored_msg, PCMK__XE_CRM_XML, NULL, NULL); xmlNode *xml_op = pcmk__xe_first_child(wrapper, NULL, NULL, NULL); if (xml_op) { xmlNode *xml_rsc = pcmk__xe_first_child(xml_op, PCMK_XE_PRIMITIVE, NULL, NULL); xmlNode *xml_attrs = pcmk__xe_first_child(xml_op, PCMK__XE_ATTRIBUTES, NULL, NULL); if (xml_rsc) { rsc = pcmk__xe_id(xml_rsc); } if (xml_attrs) { op = pcmk__xe_get(xml_attrs, CRM_META "_" PCMK__META_CLEAR_FAILURE_OP); pcmk__xe_get_guint(xml_attrs, CRM_META "_" PCMK__META_CLEAR_FAILURE_INTERVAL, &interval_ms); } } uname = pcmk__xe_get(xml_op, PCMK__META_ON_NODE); if ((rsc == NULL) || (uname == NULL)) { crm_log_xml_warn(stored_msg, "invalid failcount op"); return I_NULL; } if (pcmk__xe_get(xml_op, PCMK__XA_ROUTER_NODE)) { is_remote_node = TRUE; } crm_debug("Clearing failures for %s-interval %s on %s " "from attribute manager, CIB, and executor state", pcmk__readable_interval(interval_ms), rsc, uname); if (interval_ms) { interval_spec = crm_strdup_printf("%ums", interval_ms); } update_attrd_clear_failures(uname, rsc, op, interval_spec, is_remote_node); free(interval_spec); controld_cib_delete_last_failure(rsc, uname, op, interval_ms); lrm_clear_last_failure(rsc, uname, op, interval_ms); return I_NULL; } static enum crmd_fsa_input handle_lrm_delete(xmlNode *stored_msg) { const char *mode = NULL; xmlNode *wrapper = pcmk__xe_first_child(stored_msg, PCMK__XE_CRM_XML, NULL, NULL); xmlNode *msg_data = pcmk__xe_first_child(wrapper, NULL, NULL, NULL); CRM_CHECK(msg_data != NULL, return I_NULL); /* CRM_OP_LRM_DELETE has two distinct modes. The default behavior is to * relay the operation to the affected node, which will unregister the * resource from the local executor, clear the resource's history from the * CIB, and do some bookkeeping in the controller. * * However, if the affected node is offline, the client will specify * mode=PCMK__VALUE_CIB which means the controller receiving the operation * should clear the resource's history from the CIB and nothing else. This * is used to clear shutdown locks. */ mode = pcmk__xe_get(msg_data, PCMK__XA_MODE); if (!pcmk__str_eq(mode, PCMK__VALUE_CIB, pcmk__str_none)) { // Relay to affected node crm_xml_add(stored_msg, PCMK__XA_CRM_SYS_TO, CRM_SYSTEM_LRMD); return I_ROUTER; } else { // Delete CIB history locally (compare with do_lrm_delete()) const char *from_sys = NULL; const char *user_name = NULL; const char *rsc_id = NULL; const char *node = NULL; xmlNode *rsc_xml = NULL; int rc = pcmk_rc_ok; rsc_xml = pcmk__xe_first_child(msg_data, PCMK_XE_PRIMITIVE, NULL, NULL); CRM_CHECK(rsc_xml != NULL, return I_NULL); rsc_id = pcmk__xe_id(rsc_xml); from_sys = pcmk__xe_get(stored_msg, PCMK__XA_CRM_SYS_FROM); node = pcmk__xe_get(msg_data, PCMK__META_ON_NODE); user_name = pcmk__update_acl_user(stored_msg, PCMK__XA_CRM_USER, NULL); crm_debug("Handling " CRM_OP_LRM_DELETE " for %s on %s locally%s%s " "(clearing CIB resource history only)", rsc_id, node, (user_name? " for user " : ""), (user_name? user_name : "")); rc = controld_delete_resource_history(rsc_id, node, user_name, cib_dryrun|cib_sync_call); if (rc == pcmk_rc_ok) { rc = controld_delete_resource_history(rsc_id, node, user_name, crmd_cib_smart_opt()); } /* Notify client. Also notify tengine if mode=PCMK__VALUE_CIB and * op=CRM_OP_LRM_DELETE. */ if (from_sys) { lrmd_event_data_t *op = NULL; const char *from_host = pcmk__xe_get(stored_msg, PCMK__XA_SRC); const char *transition; if (strcmp(from_sys, CRM_SYSTEM_TENGINE)) { transition = pcmk__xe_get(msg_data, PCMK__XA_TRANSITION_KEY); } else { transition = pcmk__xe_get(stored_msg, PCMK__XA_TRANSITION_KEY); } crm_info("Notifying %s on %s that %s was%s deleted", from_sys, (from_host? from_host : "local node"), rsc_id, ((rc == pcmk_rc_ok)? "" : " not")); op = lrmd_new_event(rsc_id, PCMK_ACTION_DELETE, 0); op->type = lrmd_event_exec_complete; op->user_data = pcmk__str_copy(pcmk__s(transition, FAKE_TE_ID)); op->params = pcmk__strkey_table(free, free); pcmk__insert_dup(op->params, PCMK_XA_CRM_FEATURE_SET, CRM_FEATURE_SET); controld_rc2event(op, rc); controld_ack_event_directly(from_host, from_sys, NULL, op, rsc_id); lrmd_free_event(op); controld_trigger_delete_refresh(from_sys, rsc_id); } return I_NULL; } } /*! * \brief Handle a CRM_OP_REMOTE_STATE message by updating remote peer cache * * \param[in] msg Message XML * * \return Next FSA input */ static enum crmd_fsa_input handle_remote_state(const xmlNode *msg) { const char *conn_host = NULL; const char *remote_uname = pcmk__xe_id(msg); pcmk__node_status_t *remote_peer; bool remote_is_up = false; int rc = pcmk_rc_ok; rc = pcmk__xe_get_bool_attr(msg, PCMK__XA_IN_CCM, &remote_is_up); CRM_CHECK(remote_uname && rc == pcmk_rc_ok, return I_NULL); remote_peer = pcmk__cluster_lookup_remote_node(remote_uname); CRM_CHECK(remote_peer, return I_NULL); pcmk__update_peer_state(__func__, remote_peer, remote_is_up ? PCMK_VALUE_MEMBER : PCMK__VALUE_LOST, 0); conn_host = pcmk__xe_get(msg, PCMK__XA_CONNECTION_HOST); if (conn_host) { pcmk__str_update(&remote_peer->conn_host, conn_host); } else if (remote_peer->conn_host) { free(remote_peer->conn_host); remote_peer->conn_host = NULL; } return I_NULL; } /*! * \brief Handle a CRM_OP_PING message * * \param[in] msg Message XML * * \return Next FSA input */ static xmlNode* create_ping_reply(const xmlNode *msg) { const char *value = NULL; xmlNode *ping = NULL; xmlNode *reply = NULL; // Build reply ping = pcmk__xe_create(NULL, PCMK__XE_PING_RESPONSE); value = pcmk__xe_get(msg, PCMK__XA_CRM_SYS_TO); crm_xml_add(ping, PCMK__XA_CRM_SUBSYSTEM, value); // Add controller state value = fsa_state2string(controld_globals.fsa_state); crm_xml_add(ping, PCMK__XA_CRMD_STATE, value); crm_notice("Current ping state: %s", value); // CTS needs this // Add controller health // @TODO maybe do some checks to determine meaningful status crm_xml_add(ping, PCMK_XA_RESULT, "ok"); reply = pcmk__new_reply(msg, ping); pcmk__xml_free(ping); return reply; } static enum crmd_fsa_input handle_ping(const xmlNode *msg) { xmlNode *reply = create_ping_reply(msg); if (reply != NULL) { (void) relay_message(reply, TRUE); pcmk__xml_free(reply); } // Nothing further to do return I_NULL; } /*! * \brief Handle a PCMK__CONTROLD_CMD_NODES message * * \param[in] request Message XML * * \return Next FSA input */ static enum crmd_fsa_input handle_node_list(const xmlNode *request) { GHashTableIter iter; pcmk__node_status_t *node = NULL; xmlNode *reply = NULL; xmlNode *reply_data = NULL; // Create message data for reply reply_data = pcmk__xe_create(NULL, PCMK_XE_NODES); g_hash_table_iter_init(&iter, pcmk__peer_cache); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) & node)) { xmlNode *xml = pcmk__xe_create(reply_data, PCMK_XE_NODE); - crm_xml_add_ll(xml, PCMK_XA_ID, - (long long) node->cluster_layer_id); // uint32_t + pcmk__xe_set_ll(xml, PCMK_XA_ID, + (long long) node->cluster_layer_id); // uint32_t crm_xml_add(xml, PCMK_XA_UNAME, node->name); crm_xml_add(xml, PCMK__XA_IN_CCM, node->state); } // Create and send reply reply = pcmk__new_reply(request, reply_data); pcmk__xml_free(reply_data); if (reply) { (void) relay_message(reply, TRUE); pcmk__xml_free(reply); } // Nothing further to do return I_NULL; } /*! * \brief Handle a CRM_OP_NODE_INFO request * * \param[in] msg Message XML * * \return Next FSA input */ static enum crmd_fsa_input handle_node_info_request(const xmlNode *msg) { const char *value = NULL; pcmk__node_status_t *node = NULL; int node_id = 0; xmlNode *reply = NULL; xmlNode *reply_data = NULL; // Build reply reply_data = pcmk__xe_create(NULL, PCMK_XE_NODE); crm_xml_add(reply_data, PCMK__XA_CRM_SUBSYSTEM, CRM_SYSTEM_CRMD); // Add whether current partition has quorum pcmk__xe_set_bool_attr(reply_data, PCMK_XA_HAVE_QUORUM, pcmk_is_set(controld_globals.flags, controld_has_quorum)); /* Check whether client requested node info by ID and/or name * * @TODO A Corosync-layer node ID is of type uint32_t. We should be able to * handle legitimate node IDs greater than INT_MAX, but currently we do not. */ pcmk__xe_get_int(msg, PCMK_XA_ID, &node_id); if (node_id < 0) { node_id = 0; } value = pcmk__xe_get(msg, PCMK_XA_UNAME); // Default to local node if none given if ((node_id == 0) && (value == NULL)) { value = controld_globals.cluster->priv->node_name; } node = pcmk__search_node_caches(node_id, value, NULL, pcmk__node_search_any); if (node) { crm_xml_add(reply_data, PCMK_XA_ID, node->xml_id); crm_xml_add(reply_data, PCMK_XA_UNAME, node->name); crm_xml_add(reply_data, PCMK_XA_CRMD, node->state); pcmk__xe_set_bool_attr(reply_data, PCMK_XA_REMOTE_NODE, pcmk_is_set(node->flags, pcmk__node_status_remote)); } // Send reply reply = pcmk__new_reply(msg, reply_data); pcmk__xml_free(reply_data); if (reply != NULL) { (void) relay_message(reply, TRUE); pcmk__xml_free(reply); } // Nothing further to do return I_NULL; } static void verify_feature_set(xmlNode *msg) { const char *dc_version = pcmk__xe_get(msg, PCMK_XA_CRM_FEATURE_SET); if (dc_version == NULL) { /* All we really know is that the DC feature set is older than 3.1.0, * but that's also all that really matters. */ dc_version = "3.0.14"; } if (feature_set_compatible(dc_version, CRM_FEATURE_SET)) { crm_trace("Local feature set (%s) is compatible with DC's (%s)", CRM_FEATURE_SET, dc_version); } else { crm_err("Local feature set (%s) is incompatible with DC's (%s)", CRM_FEATURE_SET, dc_version); // Nothing is likely to improve without administrator involvement controld_set_fsa_input_flags(R_STAYDOWN); crmd_exit(CRM_EX_FATAL); } } // DC gets own shutdown all-clear static enum crmd_fsa_input handle_shutdown_self_ack(xmlNode *stored_msg) { const char *host_from = pcmk__xe_get(stored_msg, PCMK__XA_SRC); if (pcmk_is_set(controld_globals.fsa_input_register, R_SHUTDOWN)) { // The expected case -- we initiated own shutdown sequence crm_info("Shutting down controller"); return I_STOP; } if (pcmk__str_eq(host_from, controld_globals.dc_name, pcmk__str_casei)) { // Must be logic error -- DC confirming its own unrequested shutdown crm_err("Shutting down controller immediately due to " "unexpected shutdown confirmation"); return I_TERMINATE; } if (controld_globals.fsa_state != S_STOPPING) { // Shouldn't happen -- non-DC confirming unrequested shutdown crm_err("Starting new DC election because %s is " "confirming shutdown we did not request", (host_from? host_from : "another node")); return I_ELECTION; } // Shouldn't happen, but we are already stopping anyway crm_debug("Ignoring unexpected shutdown confirmation from %s", (host_from? host_from : "another node")); return I_NULL; } // Non-DC gets shutdown all-clear from DC static enum crmd_fsa_input handle_shutdown_ack(xmlNode *stored_msg) { const char *host_from = pcmk__xe_get(stored_msg, PCMK__XA_SRC); if (host_from == NULL) { crm_warn("Ignoring shutdown request without origin specified"); return I_NULL; } if (pcmk__str_eq(host_from, controld_globals.dc_name, pcmk__str_null_matches|pcmk__str_casei)) { if (pcmk_is_set(controld_globals.fsa_input_register, R_SHUTDOWN)) { crm_info("Shutting down controller after confirmation from %s", host_from); } else { crm_err("Shutting down controller after unexpected " "shutdown request from %s", host_from); controld_set_fsa_input_flags(R_STAYDOWN); } return I_STOP; } crm_warn("Ignoring shutdown request from %s because DC is %s", host_from, controld_globals.dc_name); return I_NULL; } static enum crmd_fsa_input handle_request(xmlNode *stored_msg, enum crmd_fsa_cause cause) { xmlNode *msg = NULL; const char *op = pcmk__xe_get(stored_msg, PCMK__XA_CRM_TASK); /* Optimize this for the DC - it has the most to do */ crm_log_xml_trace(stored_msg, "request"); if (op == NULL) { crm_warn("Ignoring request without " PCMK__XA_CRM_TASK); return I_NULL; } if (strcmp(op, CRM_OP_SHUTDOWN_REQ) == 0) { const char *from = pcmk__xe_get(stored_msg, PCMK__XA_SRC); pcmk__node_status_t *node = pcmk__search_node_caches(0, from, NULL, pcmk__node_search_cluster_member); pcmk__update_peer_expected(__func__, node, CRMD_JOINSTATE_DOWN); if(AM_I_DC == FALSE) { return I_NULL; /* Done */ } } /*========== DC-Only Actions ==========*/ if (AM_I_DC) { if (strcmp(op, CRM_OP_JOIN_ANNOUNCE) == 0) { return I_NODE_JOIN; } else if (strcmp(op, CRM_OP_JOIN_REQUEST) == 0) { return I_JOIN_REQUEST; } else if (strcmp(op, CRM_OP_JOIN_CONFIRM) == 0) { return I_JOIN_RESULT; } else if (strcmp(op, CRM_OP_SHUTDOWN) == 0) { return handle_shutdown_self_ack(stored_msg); } else if (strcmp(op, CRM_OP_SHUTDOWN_REQ) == 0) { // Another controller wants to shut down its node return handle_shutdown_request(stored_msg); } } /*========== common actions ==========*/ if (strcmp(op, CRM_OP_NOVOTE) == 0) { ha_msg_input_t fsa_input; fsa_input.msg = stored_msg; register_fsa_input_adv(C_HA_MESSAGE, I_NULL, &fsa_input, A_ELECTION_COUNT | A_ELECTION_CHECK, FALSE, __func__); } else if (strcmp(op, CRM_OP_REMOTE_STATE) == 0) { /* a remote connection host is letting us know the node state */ return handle_remote_state(stored_msg); } else if (strcmp(op, CRM_OP_THROTTLE) == 0) { throttle_update(stored_msg); if (AM_I_DC && (controld_globals.transition_graph != NULL) && !controld_globals.transition_graph->complete) { crm_debug("The throttle changed. Trigger a graph."); trigger_graph(); } return I_NULL; } else if (strcmp(op, CRM_OP_CLEAR_FAILCOUNT) == 0) { return handle_failcount_op(stored_msg); } else if (strcmp(op, CRM_OP_VOTE) == 0) { /* count the vote and decide what to do after that */ ha_msg_input_t fsa_input; fsa_input.msg = stored_msg; register_fsa_input_adv(C_HA_MESSAGE, I_NULL, &fsa_input, A_ELECTION_COUNT | A_ELECTION_CHECK, FALSE, __func__); /* Sometimes we _must_ go into S_ELECTION */ if (controld_globals.fsa_state == S_HALT) { crm_debug("Forcing an election from S_HALT"); return I_ELECTION; } } else if (strcmp(op, CRM_OP_JOIN_OFFER) == 0) { verify_feature_set(stored_msg); crm_debug("Raising I_JOIN_OFFER: join-%s", pcmk__xe_get(stored_msg, PCMK__XA_JOIN_ID)); return I_JOIN_OFFER; } else if (strcmp(op, CRM_OP_JOIN_ACKNAK) == 0) { crm_debug("Raising I_JOIN_RESULT: join-%s", pcmk__xe_get(stored_msg, PCMK__XA_JOIN_ID)); return I_JOIN_RESULT; } else if (strcmp(op, CRM_OP_LRM_DELETE) == 0) { return handle_lrm_delete(stored_msg); } else if ((strcmp(op, CRM_OP_LRM_FAIL) == 0) || (strcmp(op, CRM_OP_REPROBE) == 0)) { crm_xml_add(stored_msg, PCMK__XA_CRM_SYS_TO, CRM_SYSTEM_LRMD); return I_ROUTER; } else if (strcmp(op, CRM_OP_NOOP) == 0) { return I_NULL; } else if (strcmp(op, CRM_OP_PING) == 0) { return handle_ping(stored_msg); } else if (strcmp(op, CRM_OP_NODE_INFO) == 0) { return handle_node_info_request(stored_msg); } else if (strcmp(op, CRM_OP_RM_NODE_CACHE) == 0) { int id = 0; const char *name = NULL; pcmk__xe_get_int(stored_msg, PCMK_XA_ID, &id); name = pcmk__xe_get(stored_msg, PCMK_XA_UNAME); if(cause == C_IPC_MESSAGE) { msg = pcmk__new_request(pcmk_ipc_controld, CRM_SYSTEM_CRMD, NULL, CRM_SYSTEM_CRMD, CRM_OP_RM_NODE_CACHE, NULL); if (!pcmk__cluster_send_message(NULL, pcmk_ipc_controld, msg)) { crm_err("Could not instruct peers to remove references to node %s/%u", name, id); } else { crm_notice("Instructing peers to remove references to node %s/%u", name, id); } pcmk__xml_free(msg); } else { pcmk__cluster_forget_cluster_node(id, name); /* If we're forgetting this node, also forget any failures to fence * it, so we don't carry that over to any node added later with the * same name. */ st_fail_count_reset(name); } } else if (strcmp(op, CRM_OP_MAINTENANCE_NODES) == 0) { xmlNode *wrapper = pcmk__xe_first_child(stored_msg, PCMK__XE_CRM_XML, NULL, NULL); xmlNode *xml = pcmk__xe_first_child(wrapper, NULL, NULL, NULL); remote_ra_process_maintenance_nodes(xml); } else if (strcmp(op, PCMK__CONTROLD_CMD_NODES) == 0) { return handle_node_list(stored_msg); /*========== (NOT_DC)-Only Actions ==========*/ } else if (!AM_I_DC) { if (strcmp(op, CRM_OP_SHUTDOWN) == 0) { return handle_shutdown_ack(stored_msg); } } else { crm_err("Unexpected request (%s) sent to %s", op, AM_I_DC ? "the DC" : "non-DC node"); crm_log_xml_err(stored_msg, "Unexpected"); } return I_NULL; } static void handle_response(xmlNode *stored_msg) { const char *op = pcmk__xe_get(stored_msg, PCMK__XA_CRM_TASK); crm_log_xml_trace(stored_msg, "reply"); if (op == NULL) { crm_warn("Ignoring reply without " PCMK__XA_CRM_TASK); } else if (AM_I_DC && strcmp(op, CRM_OP_PECALC) == 0) { // Check whether scheduler answer been superseded by subsequent request const char *msg_ref = pcmk__xe_get(stored_msg, PCMK_XA_REFERENCE); if (msg_ref == NULL) { crm_err("%s - Ignoring calculation with no reference", op); } else if (pcmk__str_eq(msg_ref, controld_globals.fsa_pe_ref, pcmk__str_none)) { ha_msg_input_t fsa_input; controld_stop_sched_timer(); fsa_input.msg = stored_msg; register_fsa_input_later(C_IPC_MESSAGE, I_PE_SUCCESS, &fsa_input); } else { crm_info("%s calculation %s is obsolete", op, msg_ref); } } else if (strcmp(op, CRM_OP_VOTE) == 0 || strcmp(op, CRM_OP_SHUTDOWN_REQ) == 0 || strcmp(op, CRM_OP_SHUTDOWN) == 0) { } else { const char *host_from = pcmk__xe_get(stored_msg, PCMK__XA_SRC); crm_err("Unexpected response (op=%s, src=%s) sent to the %s", op, host_from, AM_I_DC ? "DC" : "controller"); } } static enum crmd_fsa_input handle_shutdown_request(xmlNode * stored_msg) { /* handle here to avoid potential version issues * where the shutdown message/procedure may have * been changed in later versions. * * This way the DC is always in control of the shutdown */ char *now_s = NULL; const char *host_from = pcmk__xe_get(stored_msg, PCMK__XA_SRC); if (host_from == NULL) { /* we're shutting down and the DC */ host_from = controld_globals.cluster->priv->node_name; } crm_info("Creating shutdown request for %s (state=%s)", host_from, fsa_state2string(controld_globals.fsa_state)); crm_log_xml_trace(stored_msg, "message"); now_s = pcmk__ttoa(time(NULL)); update_attrd(host_from, PCMK__NODE_ATTR_SHUTDOWN, now_s, NULL, FALSE); free(now_s); /* will be picked up by the TE as long as its running */ return I_NULL; } static void send_msg_via_ipc(xmlNode * msg, const char *sys, const char *src) { pcmk__client_t *client_channel = NULL; CRM_CHECK(sys != NULL, return); client_channel = pcmk__find_client_by_id(sys); if (pcmk__xe_get(msg, PCMK__XA_SRC) == NULL) { crm_xml_add(msg, PCMK__XA_SRC, src); } if (client_channel != NULL) { /* Transient clients such as crmadmin */ pcmk__ipc_send_xml(client_channel, 0, msg, crm_ipc_server_event); } else if (pcmk__str_eq(sys, CRM_SYSTEM_TENGINE, pcmk__str_none)) { xmlNode *wrapper = pcmk__xe_first_child(msg, PCMK__XE_CRM_XML, NULL, NULL); xmlNode *data = pcmk__xe_first_child(wrapper, NULL, NULL, NULL); process_te_message(msg, data); } else if (pcmk__str_eq(sys, CRM_SYSTEM_LRMD, pcmk__str_none)) { fsa_data_t fsa_data; ha_msg_input_t fsa_input; xmlNode *wrapper = NULL; fsa_input.msg = msg; wrapper = pcmk__xe_first_child(msg, PCMK__XE_CRM_XML, NULL, NULL); fsa_input.xml = pcmk__xe_first_child(wrapper, NULL, NULL, NULL); fsa_data.id = 0; fsa_data.actions = 0; fsa_data.data = &fsa_input; fsa_data.fsa_input = I_MESSAGE; fsa_data.fsa_cause = C_IPC_MESSAGE; fsa_data.origin = __func__; fsa_data.data_type = fsa_dt_ha_msg; do_lrm_invoke(A_LRM_INVOKE, C_IPC_MESSAGE, controld_globals.fsa_state, I_MESSAGE, &fsa_data); } else if (crmd_is_proxy_session(sys)) { crmd_proxy_send(sys, msg); } else { crm_info("Received invalid request: unknown subsystem '%s'", sys); } } void delete_ha_msg_input(ha_msg_input_t * orig) { if (orig == NULL) { return; } pcmk__xml_free(orig->msg); free(orig); } /*! * \internal * \brief Notify the cluster of a remote node state change * * \param[in] node_name Node's name * \param[in] node_up true if node is up, false if down */ void broadcast_remote_state_message(const char *node_name, bool node_up) { xmlNode *msg = pcmk__new_request(pcmk_ipc_controld, CRM_SYSTEM_CRMD, NULL, CRM_SYSTEM_CRMD, CRM_OP_REMOTE_STATE, NULL); crm_info("Notifying cluster of Pacemaker Remote node %s %s", node_name, node_up? "coming up" : "going down"); crm_xml_add(msg, PCMK_XA_ID, node_name); pcmk__xe_set_bool_attr(msg, PCMK__XA_IN_CCM, node_up); if (node_up) { crm_xml_add(msg, PCMK__XA_CONNECTION_HOST, controld_globals.cluster->priv->node_name); } pcmk__cluster_send_message(NULL, pcmk_ipc_controld, msg); pcmk__xml_free(msg); } diff --git a/daemons/fenced/fenced_history.c b/daemons/fenced/fenced_history.c index a555b75386..722c9f4407 100644 --- a/daemons/fenced/fenced_history.c +++ b/daemons/fenced/fenced_history.c @@ -1,574 +1,574 @@ /* * Copyright 2009-2025 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include // xmlNode #include #include #include #include #include #include #include #include #include #define MAX_STONITH_HISTORY 500 /*! * \internal * \brief Send a broadcast to all nodes to trigger cleanup or * history synchronisation * * \param[in] history Optional history to be attached * \param[in] callopts We control cleanup via a flag in the callopts * \param[in] target Cleanup can be limited to certain fence-targets */ static void stonith_send_broadcast_history(xmlNode *history, int callopts, const char *target) { xmlNode *bcast = pcmk__xe_create(NULL, PCMK__XE_STONITH_COMMAND); xmlNode *wrapper = pcmk__xe_create(bcast, PCMK__XE_ST_CALLDATA); xmlNode *call_data = pcmk__xe_create(wrapper, __func__); crm_xml_add(bcast, PCMK__XA_T, PCMK__VALUE_STONITH_NG); crm_xml_add(bcast, PCMK__XA_SUBT, PCMK__VALUE_BROADCAST); crm_xml_add(bcast, PCMK__XA_ST_OP, STONITH_OP_FENCE_HISTORY); crm_xml_add_int(bcast, PCMK__XA_ST_CALLOPT, callopts); pcmk__xml_copy(call_data, history); if (target != NULL) { crm_xml_add(call_data, PCMK__XA_ST_TARGET, target); } pcmk__cluster_send_message(NULL, pcmk_ipc_fenced, bcast); pcmk__xml_free(bcast); } static gboolean stonith_remove_history_entry (gpointer key, gpointer value, gpointer user_data) { remote_fencing_op_t *op = value; const char *target = (const char *) user_data; if ((op->state == st_failed) || (op->state == st_done)) { if ((target) && (strcmp(op->target, target) != 0)) { return FALSE; } return TRUE; } return FALSE; /* don't clean pending operations */ } /*! * \internal * \brief Send out a cleanup broadcast or do a local history-cleanup * * \param[in] target Cleanup can be limited to certain fence-targets * \param[in] broadcast Send out a cleanup broadcast */ static void stonith_fence_history_cleanup(const char *target, gboolean broadcast) { if (broadcast) { stonith_send_broadcast_history(NULL, st_opt_cleanup | st_opt_discard_reply, target); /* we'll do the local clean when we receive back our own broadcast */ } else if (stonith_remote_op_list) { g_hash_table_foreach_remove(stonith_remote_op_list, stonith_remove_history_entry, (gpointer) target); fenced_send_notification(PCMK__VALUE_ST_NOTIFY_HISTORY, NULL, NULL); } } /* keeping the length of fence-history within bounds * ================================================= * * If things are really running wild a lot of fencing-attempts * might fill up the hash-map, eventually using up a lot * of memory and creating huge history-sync messages. * Before the history being synced across nodes at least * the reboot of a cluster-node helped keeping the * history within bounds even though not in a reliable * manner. * * stonith_remote_op_list isn't sorted for time-stamps * thus it would be kind of expensive to delete e.g. * the oldest entry if it would grow past MAX_STONITH_HISTORY * entries. * It is more efficient to purge MAX_STONITH_HISTORY/2 * entries whenever the list grows beyond MAX_STONITH_HISTORY. * (sort for age + purge the MAX_STONITH_HISTORY/2 oldest) * That done on a per-node-base might raise the * probability of large syncs to occur. * Things like introducing a broadcast to purge * MAX_STONITH_HISTORY/2 entries or not sync above a certain * threshold coming to mind ... * Simplest thing though is to purge the full history * throughout the cluster once MAX_STONITH_HISTORY is reached. * On the other hand this leads to purging the history in * situations where it would be handy to have it probably. */ /*! * \internal * \brief Compare two remote fencing operations by status and completion time * * A pending operation is ordered before a completed operation. If both * operations have completed, then the more recently completed operation is * ordered first. Two pending operations are considered equal. * * \param[in] a First \c remote_fencing_op_t to compare * \param[in] b Second \c remote_fencing_op_t to compare * * \return Standard comparison result (a negative integer if \p a is lesser, * 0 if the values are equal, and a positive integer if \p a is greater) */ static gint cmp_op_by_completion(gconstpointer a, gconstpointer b) { const remote_fencing_op_t *op1 = a; const remote_fencing_op_t *op2 = b; bool op1_pending = stonith__op_state_pending(op1->state); bool op2_pending = stonith__op_state_pending(op2->state); if (op1_pending && op2_pending) { return 0; } if (op1_pending) { return -1; } if (op2_pending) { return 1; } if (op1->completed > op2->completed) { return -1; } if (op1->completed < op2->completed) { return 1; } if (op1->completed_nsec > op2->completed_nsec) { return -1; } if (op1->completed_nsec < op2->completed_nsec) { return 1; } return 0; } /*! * \internal * \brief Remove a completed operation from \c stonith_remote_op_list * * \param[in] data \c remote_fencing_op_t to remove * \param[in] user_data Ignored */ static void remove_completed_remote_op(gpointer data, gpointer user_data) { const remote_fencing_op_t *op = data; if (!stonith__op_state_pending(op->state)) { g_hash_table_remove(stonith_remote_op_list, op->id); } } /*! * \internal * \brief Do a local history-trim to MAX_STONITH_HISTORY / 2 entries * once over MAX_STONITH_HISTORY */ void stonith_fence_history_trim(void) { if (stonith_remote_op_list == NULL) { return; } if (g_hash_table_size(stonith_remote_op_list) > MAX_STONITH_HISTORY) { GList *ops = g_hash_table_get_values(stonith_remote_op_list); crm_trace("More than %d entries in fencing history, purging oldest " "completed operations", MAX_STONITH_HISTORY); ops = g_list_sort(ops, cmp_op_by_completion); // Always keep pending ops regardless of number of entries g_list_foreach(g_list_nth(ops, MAX_STONITH_HISTORY / 2), remove_completed_remote_op, NULL); // No need for a notification after purging old data g_list_free(ops); } } /*! * \internal * \brief Convert xml fence-history to a hash-table like stonith_remote_op_list * * \param[in] history Fence-history in xml * * \return Fence-history as hash-table */ static GHashTable * stonith_xml_history_to_list(const xmlNode *history) { xmlNode *xml_op = NULL; GHashTable *rv = NULL; init_stonith_remote_op_hash_table(&rv); CRM_LOG_ASSERT(rv != NULL); for (xml_op = pcmk__xe_first_child(history, NULL, NULL, NULL); xml_op != NULL; xml_op = pcmk__xe_next(xml_op, NULL)) { remote_fencing_op_t *op = NULL; char *id = pcmk__xe_get_copy(xml_op, PCMK__XA_ST_REMOTE_OP); int state; int exit_status = CRM_EX_OK; int execution_status = PCMK_EXEC_DONE; if (!id) { crm_warn("Malformed fencing history received from peer"); continue; } crm_trace("Attaching op %s to hashtable", id); op = pcmk__assert_alloc(1, sizeof(remote_fencing_op_t)); op->id = id; op->target = pcmk__xe_get_copy(xml_op, PCMK__XA_ST_TARGET); op->action = pcmk__xe_get_copy(xml_op, PCMK__XA_ST_DEVICE_ACTION); op->originator = pcmk__xe_get_copy(xml_op, PCMK__XA_ST_ORIGIN); op->delegate = pcmk__xe_get_copy(xml_op, PCMK__XA_ST_DELEGATE); op->client_name = pcmk__xe_get_copy(xml_op, PCMK__XA_ST_CLIENTNAME); pcmk__xe_get_time(xml_op, PCMK__XA_ST_DATE, &op->completed); pcmk__xe_get_ll(xml_op, PCMK__XA_ST_DATE_NSEC, &op->completed_nsec); pcmk__xe_get_int(xml_op, PCMK__XA_ST_STATE, &state); op->state = (enum op_state) state; /* @COMPAT We can't use stonith__xe_get_result() here because * fencers <2.1.3 didn't include results, leading it to assume an error * status. Instead, set an unknown status in that case. */ if ((pcmk__xe_get_int(xml_op, PCMK__XA_RC_CODE, &exit_status) != pcmk_rc_ok) || (pcmk__xe_get_int(xml_op, PCMK__XA_OP_STATUS, &execution_status) != pcmk_rc_ok)) { exit_status = CRM_EX_INDETERMINATE; execution_status = PCMK_EXEC_UNKNOWN; } pcmk__set_result(&op->result, exit_status, execution_status, pcmk__xe_get(xml_op, PCMK_XA_EXIT_REASON)); pcmk__set_result_output(&op->result, pcmk__xe_get_copy(xml_op, PCMK__XA_ST_OUTPUT), NULL); g_hash_table_replace(rv, id, op); CRM_LOG_ASSERT(g_hash_table_lookup(rv, id) != NULL); } return rv; } /*! * \internal * \brief Craft xml difference between local fence-history and a history * coming from remote, and merge the remote history into the local * * \param[in,out] remote_history Fence-history as hash-table (may be NULL) * \param[in] add_id If crafting the answer for an API * history-request there is no need for the id * \param[in] target Optionally limit to certain fence-target * * \return The fence-history as xml */ static xmlNode * stonith_local_history_diff_and_merge(GHashTable *remote_history, gboolean add_id, const char *target) { xmlNode *history = NULL; GHashTableIter iter; remote_fencing_op_t *op = NULL; gboolean updated = FALSE; int cnt = 0; if (stonith_remote_op_list) { char *id = NULL; history = pcmk__xe_create(NULL, PCMK__XE_ST_HISTORY); g_hash_table_iter_init(&iter, stonith_remote_op_list); while (g_hash_table_iter_next(&iter, (void **)&id, (void **)&op)) { xmlNode *entry = NULL; if (remote_history) { remote_fencing_op_t *remote_op = g_hash_table_lookup(remote_history, op->id); if (remote_op) { if (stonith__op_state_pending(op->state) && !stonith__op_state_pending(remote_op->state)) { crm_debug("Updating outdated pending operation %.8s " "(state=%s) according to the one (state=%s) from " "remote peer history", op->id, stonith_op_state_str(op->state), stonith_op_state_str(remote_op->state)); g_hash_table_steal(remote_history, op->id); op->id = remote_op->id; remote_op->id = id; g_hash_table_iter_replace(&iter, remote_op); updated = TRUE; continue; /* skip outdated entries */ } else if (!stonith__op_state_pending(op->state) && stonith__op_state_pending(remote_op->state)) { crm_debug("Broadcasting operation %.8s (state=%s) to " "update the outdated pending one " "(state=%s) in remote peer history", op->id, stonith_op_state_str(op->state), stonith_op_state_str(remote_op->state)); g_hash_table_remove(remote_history, op->id); } else { g_hash_table_remove(remote_history, op->id); continue; /* skip entries broadcasted already */ } } } if (!pcmk__str_eq(target, op->target, pcmk__str_null_matches)) { continue; } cnt++; crm_trace("Attaching op %s", op->id); entry = pcmk__xe_create(history, STONITH_OP_EXEC); if (add_id) { crm_xml_add(entry, PCMK__XA_ST_REMOTE_OP, op->id); } crm_xml_add(entry, PCMK__XA_ST_TARGET, op->target); crm_xml_add(entry, PCMK__XA_ST_DEVICE_ACTION, op->action); crm_xml_add(entry, PCMK__XA_ST_ORIGIN, op->originator); crm_xml_add(entry, PCMK__XA_ST_DELEGATE, op->delegate); crm_xml_add(entry, PCMK__XA_ST_CLIENTNAME, op->client_name); pcmk__xe_set_time(entry, PCMK__XA_ST_DATE, op->completed); - crm_xml_add_ll(entry, PCMK__XA_ST_DATE_NSEC, - op->completed_nsec); + pcmk__xe_set_ll(entry, PCMK__XA_ST_DATE_NSEC, + op->completed_nsec); crm_xml_add_int(entry, PCMK__XA_ST_STATE, op->state); stonith__xe_set_result(entry, &op->result); } } if (remote_history) { init_stonith_remote_op_hash_table(&stonith_remote_op_list); updated |= g_hash_table_size(remote_history); g_hash_table_iter_init(&iter, remote_history); while (g_hash_table_iter_next(&iter, NULL, (void **)&op)) { if (stonith__op_state_pending(op->state) && pcmk__str_eq(op->originator, fenced_get_local_node(), pcmk__str_casei)) { crm_warn("Failing pending operation %.8s originated by us but " "known only from peer history", op->id); op->state = st_failed; set_fencing_completed(op); /* CRM_EX_EXPIRED + PCMK_EXEC_INVALID prevents finalize_op() * from setting a delegate */ pcmk__set_result(&op->result, CRM_EX_EXPIRED, PCMK_EXEC_INVALID, "Initiated by earlier fencer " "process and presumed failed"); fenced_broadcast_op_result(op, false); } g_hash_table_iter_steal(&iter); g_hash_table_replace(stonith_remote_op_list, op->id, op); /* we could trim the history here but if we bail * out after trim we might miss more recent entries * of those that might still be in the list * if we don't bail out trimming once is more * efficient and memory overhead is minimal as * we are just moving pointers from one hash to * another */ } g_hash_table_destroy(remote_history); /* remove what is left */ } if (updated) { stonith_fence_history_trim(); fenced_send_notification(PCMK__VALUE_ST_NOTIFY_HISTORY, NULL, NULL); } if (cnt == 0) { pcmk__xml_free(history); return NULL; } else { return history; } } /*! * \internal * \brief Craft xml from the local fence-history * * \param[in] add_id If crafting the answer for an API * history-request there is no need for the id * \param[in] target Optionally limit to certain fence-target * * \return The fence-history as xml */ static xmlNode * stonith_local_history(gboolean add_id, const char *target) { return stonith_local_history_diff_and_merge(NULL, add_id, target); } /*! * \internal * \brief Handle fence-history messages (from API or coming in as broadcasts) * * \param[in,out] msg Request XML * \param[out] output Where to set local history, if requested * \param[in] remote_peer If broadcast, peer that sent it * \param[in] options Call options from the request */ void stonith_fence_history(xmlNode *msg, xmlNode **output, const char *remote_peer, int options) { const char *target = NULL; xmlNode *dev = pcmk__xpath_find_one(msg->doc, "//*[@" PCMK__XA_ST_TARGET "]", LOG_NEVER); xmlNode *out_history = NULL; if (dev) { target = pcmk__xe_get(dev, PCMK__XA_ST_TARGET); if (target && (options & st_opt_cs_nodeid)) { int nodeid; pcmk__node_status_t *node = NULL; pcmk__scan_min_int(target, &nodeid, 0); node = pcmk__search_node_caches(nodeid, NULL, NULL, pcmk__node_search_any |pcmk__node_search_cluster_cib); if (node != NULL) { target = node->name; } } } if (options & st_opt_cleanup) { const char *call_id = pcmk__xe_get(msg, PCMK__XA_ST_CALLID); crm_trace("Cleaning up operations on %s in %p", target, stonith_remote_op_list); stonith_fence_history_cleanup(target, (call_id != NULL)); } else if (options & st_opt_broadcast) { /* there is no clear sign atm for when a history sync is done so send a notification for anything that smells like history-sync */ fenced_send_notification(PCMK__VALUE_ST_NOTIFY_HISTORY_SYNCED, NULL, NULL); if (pcmk__xe_get(msg, PCMK__XA_ST_CALLID) != NULL) { /* this is coming from the stonith-API * * craft a broadcast with node's history * so that every node can merge and broadcast * what it has on top */ out_history = stonith_local_history(TRUE, NULL); crm_trace("Broadcasting history to peers"); stonith_send_broadcast_history(out_history, st_opt_broadcast | st_opt_discard_reply, NULL); } else if (remote_peer && !pcmk__str_eq(remote_peer, fenced_get_local_node(), pcmk__str_casei)) { xmlNode *history = pcmk__xpath_find_one(msg->doc, "//" PCMK__XE_ST_HISTORY, LOG_NEVER); /* either a broadcast created directly upon stonith-API request * or a diff as response to such a thing * * in both cases it may have a history or not * if we have differential data * merge in what we've received and stop * otherwise broadcast what we have on top * marking as differential and merge in afterwards */ if (!history || !pcmk__xe_attr_is_true(history, PCMK__XA_ST_DIFFERENTIAL)) { GHashTable *received_history = NULL; if (history != NULL) { received_history = stonith_xml_history_to_list(history); } out_history = stonith_local_history_diff_and_merge(received_history, TRUE, NULL); if (out_history) { crm_trace("Broadcasting history-diff to peers"); pcmk__xe_set_bool_attr(out_history, PCMK__XA_ST_DIFFERENTIAL, true); stonith_send_broadcast_history(out_history, st_opt_broadcast | st_opt_discard_reply, NULL); } else { crm_trace("History-diff is empty - skip broadcast"); } } } else { crm_trace("Skipping history-query-broadcast (%s%s)" " we sent ourselves", remote_peer?"remote-peer=":"local-ipc", remote_peer?remote_peer:""); } } else { /* plain history request */ crm_trace("Looking for operations on %s in %p", target, stonith_remote_op_list); *output = stonith_local_history(FALSE, target); } pcmk__xml_free(out_history); } diff --git a/include/crm/common/xml_element_internal.h b/include/crm/common/xml_element_internal.h index dfe66625e2..b9ba2b1323 100644 --- a/include/crm/common/xml_element_internal.h +++ b/include/crm/common/xml_element_internal.h @@ -1,203 +1,205 @@ /* * Copyright 2017-2025 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #ifndef PCMK__CRM_COMMON_XML_ELEMENT_INTERNAL__H #define PCMK__CRM_COMMON_XML_ELEMENT_INTERNAL__H /* * Internal-only wrappers for and extensions to libxml2 for processing XML * elements */ #include // bool #include // uint32_t #include // NULL #include // strcmp() #include // guint #include // xmlNode, etc. #include // crm_time_t #include // pcmk__str_copy() #include // PCMK_XA_ID #ifdef __cplusplus extern "C" { #endif const char *pcmk__xe_add_last_written(xmlNode *xe); xmlNode *pcmk__xe_first_child(const xmlNode *parent, const char *node_name, const char *attr_n, const char *attr_v); void pcmk__xe_remove_attr(xmlNode *element, const char *name); bool pcmk__xe_remove_attr_cb(xmlNode *xml, void *user_data); void pcmk__xe_remove_matching_attrs(xmlNode *element, bool force, bool (*match)(xmlAttrPtr, void *), void *user_data); int pcmk__xe_delete_match(xmlNode *xml, xmlNode *search); int pcmk__xe_replace_match(xmlNode *xml, xmlNode *replace); int pcmk__xe_update_match(xmlNode *xml, xmlNode *update, uint32_t flags); /*! * \internal * \brief Check whether an XML element is of a particular type * * \param[in] xml XML element to compare * \param[in] name XML element name to compare * * \return \c true if \p xml is of type \p name, otherwise \c false */ static inline bool pcmk__xe_is(const xmlNode *xml, const char *name) { return (xml != NULL) && (xml->name != NULL) && (name != NULL) && (strcmp((const char *) xml->name, name) == 0); } xmlNode *pcmk__xe_create(xmlNode *parent, const char *name); xmlNode *pcmk__xe_next(const xmlNode *node, const char *element_name); void pcmk__xe_set_content(xmlNode *node, const char *format, ...) G_GNUC_PRINTF(2, 3); int pcmk__xe_get_score(const xmlNode *xml, const char *name, int *score, int default_score); int pcmk__xe_copy_attrs(xmlNode *target, const xmlNode *src, uint32_t flags); void pcmk__xe_sort_attrs(xmlNode *xml); void pcmk__xe_set_id(xmlNode *xml, const char *format, ...) G_GNUC_PRINTF(2, 3); /*! * \internal * \brief Like pcmk__xe_set_props, but takes a va_list instead of * arguments directly. * * \param[in,out] node XML to add attributes to * \param[in] pairs NULL-terminated list of name/value pairs to add */ void pcmk__xe_set_propv(xmlNodePtr node, va_list pairs); /*! * \internal * \brief Add a NULL-terminated list of name/value pairs to the given * XML node as properties. * * \param[in,out] node XML node to add properties to * \param[in] ... NULL-terminated list of name/value pairs * * \note A NULL name terminates the arguments; a NULL value will be skipped. */ void pcmk__xe_set_props(xmlNodePtr node, ...) G_GNUC_NULL_TERMINATED; /*! * \internal * \brief Get first attribute of an XML element * * \param[in] xe XML element to check * * \return First attribute of \p xe (or NULL if \p xe is NULL or has none) */ static inline xmlAttr * pcmk__xe_first_attr(const xmlNode *xe) { return (xe == NULL)? NULL : xe->properties; } /*! * \internal * \brief Iterate over child elements of \p xml * * This function iterates over the children of \p xml, performing the * callback function \p handler on each node. If the callback returns * a value other than pcmk_rc_ok, the iteration stops and the value is * returned. It is therefore possible that not all children will be * visited. * * \param[in,out] xml The starting XML node. Can be NULL. * \param[in] child_element_name The name that the node must match in order * for \p handler to be run. If NULL, all * child elements will match. * \param[in] handler The callback function. * \param[in,out] userdata User data to pass to the callback function. * Can be NULL. * * \return Standard Pacemaker return code */ int pcmk__xe_foreach_child(xmlNode *xml, const char *child_element_name, int (*handler)(xmlNode *xml, void *userdata), void *userdata); const char *pcmk__xe_get(const xmlNode *xml, const char *attr_name); int pcmk__xe_get_datetime(const xmlNode *xml, const char *attr, crm_time_t **t); int pcmk__xe_get_flags(const xmlNode *xml, const char *name, uint32_t *dest, uint32_t default_value); int pcmk__xe_get_guint(const xmlNode *xml, const char *attr, guint *dest); void pcmk__xe_set_guint(xmlNode *xml, const char *attr, guint value); int pcmk__xe_get_int(const xmlNode *xml, const char *name, int *dest); + int pcmk__xe_get_ll(const xmlNode *xml, const char *name, long long *dest); +int pcmk__xe_set_ll(xmlNode *xml, const char *attr, long long value); int pcmk__xe_get_time(const xmlNode *xml, const char *attr, time_t *dest); void pcmk__xe_set_time(xmlNode *xml, const char *attr, time_t value); int pcmk__xe_get_timeval(const xmlNode *xml, const char *sec_attr, const char *usec_attr, struct timeval *dest); void pcmk__xe_set_timeval(xmlNode *xml, const char *sec_attr, const char *usec_attr, const struct timeval *value); void pcmk__xe_set_bool_attr(xmlNodePtr node, const char *name, bool value); int pcmk__xe_get_bool_attr(const xmlNode *node, const char *name, bool *value); bool pcmk__xe_attr_is_true(const xmlNode *node, const char *name); /*! * \internal * \brief Retrieve a copy of the value of an XML attribute * * This is like \c pcmk__xe_get() but allocates new memory for the result. * * \param[in] xml XML element whose attribute to get * \param[in] attr Attribute name * * \return Value of specified attribute (or \c NULL if not set) * * \note The caller is responsible for freeing the result using \c free(). */ static inline char * pcmk__xe_get_copy(const xmlNode *xml, const char *attr) { return pcmk__str_copy(pcmk__xe_get(xml, attr)); } /*! * \internal * \brief Retrieve the value of the \c PCMK_XA_ID XML attribute * * \param[in] xml XML element to check * * \return Value of the \c PCMK_XA_ID attribute (may be \c NULL) */ static inline const char * pcmk__xe_id(const xmlNode *xml) { return pcmk__xe_get(xml, PCMK_XA_ID); } #ifdef __cplusplus } #endif #endif // PCMK__CRM_COMMON_XML_ELEMENT_INTERNAL__H diff --git a/lib/cluster/corosync.c b/lib/cluster/corosync.c index 2782b10067..a0e7e6b38d 100644 --- a/lib/cluster/corosync.c +++ b/lib/cluster/corosync.c @@ -1,821 +1,821 @@ /* - * Copyright 2004-2024 the Pacemaker project contributors + * Copyright 2004-2025 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #include #include #include // PRIu64, etc. #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include // PCMK__SPECIAL_PID #include #include #include "crmcluster_private.h" static quorum_handle_t pcmk_quorum_handle = 0; static gboolean (*quorum_app_callback)(unsigned long long seq, gboolean quorate) = NULL; /*! * \internal * \brief Get the Corosync UUID associated with a Pacemaker node * * \param[in] node Pacemaker node * * \return Newly allocated string with node's Corosync UUID, or NULL if unknown * \note It is the caller's responsibility to free the result with free(). */ char * pcmk__corosync_uuid(const pcmk__node_status_t *node) { pcmk__assert(pcmk_get_cluster_layer() == pcmk_cluster_layer_corosync); if (node != NULL) { if (node->cluster_layer_id > 0) { return crm_strdup_printf("%" PRIu32, node->cluster_layer_id); } else { crm_info("Node %s is not yet known by Corosync", node->name); } } return NULL; } static bool node_name_is_valid(const char *key, const char *name) { int octet; if (name == NULL) { crm_trace("%s is empty", key); return false; } else if (sscanf(name, "%d.%d.%d.%d", &octet, &octet, &octet, &octet) == 4) { crm_trace("%s contains an IPv4 address (%s), ignoring", key, name); return false; } else if (strstr(name, ":") != NULL) { crm_trace("%s contains an IPv6 address (%s), ignoring", key, name); return false; } crm_trace("'%s: %s' is valid", key, name); return true; } /* * \internal * \brief Get Corosync node name corresponding to a node ID * * \param[in] cmap_handle Connection to Corosync CMAP * \param[in] nodeid Node ID to check * * \return Newly allocated string with name or (if no name) IP address * associated with first address assigned to a Corosync node ID (or NULL * if unknown) * \note It is the caller's responsibility to free the result with free(). */ char * pcmk__corosync_name(uint64_t /*cmap_handle_t */ cmap_handle, uint32_t nodeid) { // Originally based on corosync-quorumtool.c:node_name() int lpc = 0; cs_error_t rc = CS_OK; int retries = 0; char *name = NULL; cmap_handle_t local_handle = 0; int fd = -1; uid_t found_uid = 0; gid_t found_gid = 0; pid_t found_pid = 0; int rv; if (nodeid == 0) { nodeid = pcmk__cpg_local_nodeid(0); } if (cmap_handle == 0 && local_handle == 0) { retries = 0; crm_trace("Initializing CMAP connection"); do { rc = pcmk__init_cmap(&local_handle); if (rc != CS_OK) { retries++; crm_debug("API connection setup failed: %s. Retrying in %ds", cs_strerror(rc), retries); sleep(retries); } } while (retries < 5 && rc != CS_OK); if (rc != CS_OK) { crm_warn("Could not connect to Cluster Configuration Database API, error %s", cs_strerror(rc)); local_handle = 0; } } if (cmap_handle == 0) { cmap_handle = local_handle; rc = cmap_fd_get(cmap_handle, &fd); if (rc != CS_OK) { crm_err("Could not obtain the CMAP API connection: %s (%d)", cs_strerror(rc), rc); goto bail; } /* CMAP 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("CMAP provider is not authentic:" " process %lld (uid: %lld, gid: %lld)", (long long) PCMK__SPECIAL_PID_AS_0(found_pid), (long long) found_uid, (long long) found_gid); goto bail; } else if (rv < 0) { crm_err("Could not verify authenticity of CMAP provider: %s (%d)", strerror(-rv), -rv); goto bail; } } while (name == NULL && cmap_handle != 0) { uint32_t id = 0; char *key = NULL; key = crm_strdup_printf("nodelist.node.%d.nodeid", lpc); rc = cmap_get_uint32(cmap_handle, key, &id); crm_trace("Checking %u vs %u from %s", nodeid, id, key); free(key); if (rc != CS_OK) { break; } if (nodeid == id) { crm_trace("Searching for node name for %u in nodelist.node.%d %s", nodeid, lpc, pcmk__s(name, "")); if (name == NULL) { key = crm_strdup_printf("nodelist.node.%d.name", lpc); cmap_get_string(cmap_handle, key, &name); crm_trace("%s = %s", key, pcmk__s(name, "")); free(key); } if (name == NULL) { key = crm_strdup_printf("nodelist.node.%d.ring0_addr", lpc); cmap_get_string(cmap_handle, key, &name); crm_trace("%s = %s", key, pcmk__s(name, "")); if (!node_name_is_valid(key, name)) { free(name); name = NULL; } free(key); } break; } lpc++; } bail: if(local_handle) { cmap_finalize(local_handle); } if (name == NULL) { crm_info("Unable to get node name for nodeid %u", nodeid); } return name; } /*! * \internal * \brief Disconnect from Corosync cluster * * \param[in,out] cluster Cluster object to disconnect */ void pcmk__corosync_disconnect(pcmk_cluster_t *cluster) { pcmk__cpg_disconnect(cluster); if (pcmk_quorum_handle != 0) { quorum_finalize(pcmk_quorum_handle); pcmk_quorum_handle = 0; } crm_notice("Disconnected from Corosync"); } /*! * \internal * \brief Dispatch function for quorum connection file descriptor * * \param[in] user_data Ignored * * \return 0 on success, -1 on error (per mainloop_io_t interface) */ static int quorum_dispatch_cb(gpointer user_data) { int rc = quorum_dispatch(pcmk_quorum_handle, CS_DISPATCH_ALL); if (rc < 0) { crm_err("Connection to the Quorum API failed: %d", rc); quorum_finalize(pcmk_quorum_handle); pcmk_quorum_handle = 0; return -1; } return 0; } /*! * \internal * \brief Notification callback for Corosync quorum connection * * \param[in] handle Corosync quorum connection * \param[in] quorate Whether cluster is quorate * \param[in] ring_id Corosync ring ID * \param[in] view_list_entries Number of entries in \p view_list * \param[in] view_list Corosync node IDs in membership */ static void quorum_notification_cb(quorum_handle_t handle, uint32_t quorate, uint64_t ring_id, uint32_t view_list_entries, uint32_t *view_list) { int i; GHashTableIter iter; pcmk__node_status_t *node = NULL; static gboolean init_phase = TRUE; bool is_quorate = (quorate != 0); bool was_quorate = pcmk__cluster_has_quorum(); if (is_quorate && !was_quorate) { crm_notice("Quorum acquired " QB_XS " membership=%" PRIu64 " members=%" PRIu32, ring_id, view_list_entries); pcmk__cluster_set_quorum(true); } else if (!is_quorate && was_quorate) { crm_warn("Quorum lost " QB_XS " membership=%" PRIu64 " members=" PRIu32, ring_id, view_list_entries); pcmk__cluster_set_quorum(false); } else { crm_info("Quorum %s " QB_XS " membership=%" PRIu64 " members=%" PRIu32, (is_quorate? "retained" : "still lost"), ring_id, view_list_entries); } if (view_list_entries == 0 && init_phase) { crm_info("Corosync membership is still forming, ignoring"); return; } init_phase = FALSE; /* Reset membership_id for all cached nodes so we can tell which ones aren't * in the view list */ g_hash_table_iter_init(&iter, pcmk__peer_cache); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) { node->membership_id = 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 = pcmk__get_node(id, NULL, NULL, pcmk__node_search_cluster_member); if (node->name == NULL) { char *name = pcmk__corosync_name(0, id); crm_info("Obtaining name for new node %u", id); node = pcmk__get_node(id, name, NULL, pcmk__node_search_cluster_member); free(name); } // Update the node state (including updating membership_id to ring_id) pcmk__update_peer_state(__func__, node, PCMK_VALUE_MEMBER, ring_id); } /* Remove any peer cache entries we didn't update */ pcmk__reap_unseen_nodes(ring_id); if (quorum_app_callback) { quorum_app_callback(ring_id, is_quorate); } } /*! * \internal * \brief Connect to Corosync quorum service * * \param[in] dispatch Connection dispatch callback * \param[in] destroy Connection destroy callback */ void pcmk__corosync_quorum_connect(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 = quorum_dispatch_cb; quorum_fd_callbacks.destroy = destroy; crm_debug("Configuring Pacemaker to obtain quorum from Corosync"); { #if 0 // New way but not supported by all Corosync 2 versions quorum_model_v0_data_t quorum_model_data = { .model = QUORUM_MODEL_V0, .quorum_notify_fn = quorum_notification_cb, }; rc = quorum_model_initialize(&pcmk_quorum_handle, QUORUM_MODEL_V0, (quorum_model_data_t *) &quorum_model_data, &quorum_type, NULL); #else quorum_callbacks_t quorum_callbacks = { .quorum_notify_fn = quorum_notification_cb, }; rc = quorum_initialize(&pcmk_quorum_handle, &quorum_callbacks, &quorum_type); #endif } 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("No quorum"); } quorum_app_callback = dispatch; pcmk__cluster_set_quorum(quorate != 0); 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); pcmk__corosync_add_nodes(NULL); bail: if (rc != CS_OK) { quorum_finalize(pcmk_quorum_handle); } } /*! * \internal * \brief Connect to Corosync cluster layer * * \param[in,out] cluster Initialized cluster object to connect * * \return Standard Pacemaker return code */ int pcmk__corosync_connect(pcmk_cluster_t *cluster) { const enum pcmk_cluster_layer cluster_layer = pcmk_get_cluster_layer(); const char *cluster_layer_s = pcmk_cluster_layer_text(cluster_layer); pcmk__node_status_t *local_node = NULL; int rc = pcmk_rc_ok; pcmk__cluster_init_node_caches(); if (cluster_layer != pcmk_cluster_layer_corosync) { crm_err("Invalid cluster layer: %s " QB_XS " cluster_layer=%d", cluster_layer_s, cluster_layer); return EINVAL; } rc = pcmk__cpg_connect(cluster); if (rc != pcmk_rc_ok) { // Error message was logged by pcmk__cpg_connect() return rc; } crm_info("Connection to %s established", cluster_layer_s); cluster->priv->node_id = pcmk__cpg_local_nodeid(0); if (cluster->priv->node_id == 0) { crm_err("Could not determine local node ID"); return ENXIO; } cluster->priv->node_name = pcmk__cluster_node_name(0); if (cluster->priv->node_name == NULL) { crm_err("Could not determine local node name"); return ENXIO; } // Ensure local node always exists in peer cache local_node = pcmk__get_node(cluster->priv->node_id, cluster->priv->node_name, NULL, pcmk__node_search_cluster_member); cluster->priv->node_xml_id = pcmk__corosync_uuid(local_node); CRM_LOG_ASSERT(cluster->priv->node_xml_id != NULL); return pcmk_rc_ok; } /*! * \internal * \brief Check whether a Corosync cluster is active * * \return \c true if Corosync is found active, or \c false otherwise */ bool pcmk__corosync_is_active(void) { cmap_handle_t handle; int rc = pcmk__init_cmap(&handle); if (rc == CS_OK) { cmap_finalize(handle); return true; } crm_info("Failed to initialize the cmap API: %s (%d)", pcmk__cs_err_str(rc), rc); return false; } /*! * \internal * \brief Check whether a Corosync cluster peer is active * * \param[in] node Node to check * * \return \c true if \p node is an active Corosync peer, or \c false otherwise */ bool pcmk__corosync_is_peer_active(const pcmk__node_status_t *node) { if (node == NULL) { crm_trace("Corosync peer inactive: NULL"); return false; } if (!pcmk__str_eq(node->state, PCMK_VALUE_MEMBER, pcmk__str_none)) { crm_trace("Corosync peer %s inactive: state=%s", node->name, node->state); return false; } if (!pcmk_is_set(node->processes, crm_proc_cpg)) { crm_trace("Corosync peer %s inactive " QB_XS " processes=%.16" PRIx32, node->name, node->processes); return false; } return true; } /*! * \internal * \brief Load Corosync node list (via CMAP) into peer cache and optionally XML * * \param[in,out] xml_parent If not NULL, add entry here for each node * * \return true if any nodes were found, false otherwise */ bool pcmk__corosync_add_nodes(xmlNode *xml_parent) { int lpc = 0; cs_error_t rc = CS_OK; int retries = 0; bool any = false; cmap_handle_t cmap_handle; int fd = -1; uid_t found_uid = 0; gid_t found_gid = 0; pid_t found_pid = 0; int rv; do { rc = pcmk__init_cmap(&cmap_handle); if (rc != CS_OK) { retries++; crm_debug("API connection setup failed: %s. Retrying in %ds", cs_strerror(rc), retries); sleep(retries); } } while (retries < 5 && rc != CS_OK); if (rc != CS_OK) { crm_warn("Could not connect to Cluster Configuration Database API, error %d", rc); return false; } rc = cmap_fd_get(cmap_handle, &fd); if (rc != CS_OK) { crm_err("Could not obtain the CMAP API connection: %s (%d)", cs_strerror(rc), rc); goto bail; } /* CMAP 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("CMAP provider is not authentic:" " process %lld (uid: %lld, gid: %lld)", (long long) PCMK__SPECIAL_PID_AS_0(found_pid), (long long) found_uid, (long long) found_gid); goto bail; } else if (rv < 0) { crm_err("Could not verify authenticity of CMAP provider: %s (%d)", strerror(-rv), -rv); goto bail; } pcmk__cluster_init_node_caches(); crm_trace("Initializing Corosync node list"); for (lpc = 0; TRUE; lpc++) { uint32_t nodeid = 0; char *name = NULL; char *key = NULL; key = crm_strdup_printf("nodelist.node.%d.nodeid", lpc); rc = cmap_get_uint32(cmap_handle, key, &nodeid); free(key); if (rc != CS_OK) { break; } name = pcmk__corosync_name(cmap_handle, nodeid); if (name != NULL) { GHashTableIter iter; pcmk__node_status_t *node = NULL; g_hash_table_iter_init(&iter, pcmk__peer_cache); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) { if ((node != NULL) && (node->cluster_layer_id > 0) && (node->cluster_layer_id != nodeid) && pcmk__str_eq(node->name, name, pcmk__str_casei)) { crm_crit("Nodes %" PRIu32 " and %" PRIu32 " share the " "same name '%s': shutting down", node->cluster_layer_id, nodeid, name); crm_exit(CRM_EX_FATAL); } } } if (nodeid > 0 || name != NULL) { crm_trace("Initializing node[%d] %u = %s", lpc, nodeid, name); pcmk__get_node(nodeid, name, NULL, pcmk__node_search_cluster_member); } if (nodeid > 0 && name != NULL) { any = true; if (xml_parent) { xmlNode *node = pcmk__xe_create(xml_parent, PCMK_XE_NODE); - crm_xml_add_ll(node, PCMK_XA_ID, (long long) nodeid); + pcmk__xe_set_ll(node, PCMK_XA_ID, (long long) nodeid); crm_xml_add(node, PCMK_XA_UNAME, name); } } free(name); } bail: cmap_finalize(cmap_handle); return any; } /*! * \internal * \brief Get cluster name from Corosync configuration (via CMAP) * * \return Newly allocated string with cluster name if configured, or NULL */ char * pcmk__corosync_cluster_name(void) { cmap_handle_t handle; char *cluster_name = NULL; cs_error_t rc = CS_OK; int fd = -1; uid_t found_uid = 0; gid_t found_gid = 0; pid_t found_pid = 0; int rv; rc = pcmk__init_cmap(&handle); if (rc != CS_OK) { crm_info("Failed to initialize the cmap API: %s (%d)", cs_strerror(rc), rc); return NULL; } rc = cmap_fd_get(handle, &fd); if (rc != CS_OK) { crm_err("Could not obtain the CMAP API connection: %s (%d)", cs_strerror(rc), rc); goto bail; } /* CMAP 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("CMAP provider is not authentic:" " process %lld (uid: %lld, gid: %lld)", (long long) PCMK__SPECIAL_PID_AS_0(found_pid), (long long) found_uid, (long long) found_gid); goto bail; } else if (rv < 0) { crm_err("Could not verify authenticity of CMAP provider: %s (%d)", strerror(-rv), -rv); goto bail; } rc = cmap_get_string(handle, "totem.cluster_name", &cluster_name); if (rc != CS_OK) { crm_info("Cannot get totem.cluster_name: %s (%d)", cs_strerror(rc), rc); } else { crm_debug("cmap totem.cluster_name = '%s'", cluster_name); } bail: cmap_finalize(handle); return cluster_name; } /*! * \internal * \brief Check (via CMAP) whether Corosync configuration has a node list * * \return true if Corosync has node list, otherwise false */ bool pcmk__corosync_has_nodelist(void) { cs_error_t cs_rc = CS_OK; int retries = 0; cmap_handle_t cmap_handle; cmap_iter_handle_t iter_handle; char key_name[CMAP_KEYNAME_MAXLEN + 1]; int fd = -1; uid_t found_uid = 0; gid_t found_gid = 0; pid_t found_pid = 0; int rc = pcmk_ok; static bool got_result = false; static bool result = false; if (got_result) { return result; } // Connect to CMAP do { cs_rc = pcmk__init_cmap(&cmap_handle); if (cs_rc != CS_OK) { retries++; crm_debug("CMAP connection failed: %s (rc=%d, retrying in %ds)", cs_strerror(cs_rc), cs_rc, retries); sleep(retries); } } while ((retries < 5) && (cs_rc != CS_OK)); if (cs_rc != CS_OK) { crm_warn("Assuming Corosync does not have node list: " "CMAP connection failed (%s) " QB_XS " rc=%d", cs_strerror(cs_rc), cs_rc); return false; } // Get CMAP connection file descriptor cs_rc = cmap_fd_get(cmap_handle, &fd); if (cs_rc != CS_OK) { crm_warn("Assuming Corosync does not have node list: " "CMAP unusable (%s) " QB_XS " rc=%d", cs_strerror(cs_rc), cs_rc); goto bail; } // Check whether CMAP connection is authentic (i.e. provided by root) rc = crm_ipc_is_authentic_process(fd, (uid_t) 0, (gid_t) 0, &found_pid, &found_uid, &found_gid); if (rc == 0) { crm_warn("Assuming Corosync does not have node list: " "CMAP provider is inauthentic " QB_XS " pid=%lld uid=%lld gid=%lld", (long long) PCMK__SPECIAL_PID_AS_0(found_pid), (long long) found_uid, (long long) found_gid); goto bail; } else if (rc < 0) { crm_warn("Assuming Corosync does not have node list: " "Could not verify CMAP authenticity (%s) " QB_XS " rc=%d", pcmk_strerror(rc), rc); goto bail; } // Check whether nodelist section is presetn cs_rc = cmap_iter_init(cmap_handle, "nodelist", &iter_handle); if (cs_rc != CS_OK) { crm_warn("Assuming Corosync does not have node list: " "CMAP not readable (%s) " QB_XS " rc=%d", cs_strerror(cs_rc), cs_rc); goto bail; } cs_rc = cmap_iter_next(cmap_handle, iter_handle, key_name, NULL, NULL); if (cs_rc == CS_OK) { result = true; } cmap_iter_finalize(cmap_handle, iter_handle); got_result = true; crm_debug("Corosync %s node list", (result? "has" : "does not have")); bail: cmap_finalize(cmap_handle); return result; } diff --git a/lib/common/ipc_client.c b/lib/common/ipc_client.c index 63483e7b2d..6cc04543e1 100644 --- a/lib/common/ipc_client.c +++ b/lib/common/ipc_client.c @@ -1,1694 +1,1694 @@ /* * Copyright 2004-2025 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #include #if defined(HAVE_UCRED) || defined(HAVE_SOCKPEERCRED) #include #elif defined(HAVE_GETPEERUCRED) #include #endif #include #include #include #include #include /* indirectly: pcmk_err_generic */ #include #include #include #include "crmcommon_private.h" static int is_ipc_provider_expected(qb_ipcc_connection_t *qb_ipc, int sock, uid_t refuid, gid_t refgid, pid_t *gotpid, uid_t *gotuid, gid_t *gotgid); /*! * \brief Create a new object for using Pacemaker daemon IPC * * \param[out] api Where to store new IPC object * \param[in] server Which Pacemaker daemon the object is for * * \return Standard Pacemaker result code * * \note The caller is responsible for freeing *api using pcmk_free_ipc_api(). * \note This is intended to supersede crm_ipc_new() but currently only * supports the controller, pacemakerd, and schedulerd IPC API. */ int pcmk_new_ipc_api(pcmk_ipc_api_t **api, enum pcmk_ipc_server server) { if (api == NULL) { return EINVAL; } *api = calloc(1, sizeof(pcmk_ipc_api_t)); if (*api == NULL) { return errno; } (*api)->server = server; if (pcmk_ipc_name(*api, false) == NULL) { pcmk_free_ipc_api(*api); *api = NULL; return EOPNOTSUPP; } (*api)->ipc_size_max = 0; // Set server methods and max_size (if not default) switch (server) { case pcmk_ipc_attrd: (*api)->cmds = pcmk__attrd_api_methods(); break; case pcmk_ipc_based: (*api)->ipc_size_max = 512 * 1024; // 512KB break; case pcmk_ipc_controld: (*api)->cmds = pcmk__controld_api_methods(); break; case pcmk_ipc_execd: break; case pcmk_ipc_fenced: break; case pcmk_ipc_pacemakerd: (*api)->cmds = pcmk__pacemakerd_api_methods(); break; case pcmk_ipc_schedulerd: (*api)->cmds = pcmk__schedulerd_api_methods(); // @TODO max_size could vary by client, maybe take as argument? (*api)->ipc_size_max = 5 * 1024 * 1024; // 5MB break; default: // pcmk_ipc_unknown pcmk_free_ipc_api(*api); *api = NULL; return EINVAL; } if ((*api)->cmds == NULL) { pcmk_free_ipc_api(*api); *api = NULL; return ENOMEM; } (*api)->ipc = crm_ipc_new(pcmk_ipc_name(*api, false), (*api)->ipc_size_max); if ((*api)->ipc == NULL) { pcmk_free_ipc_api(*api); *api = NULL; return ENOMEM; } // If daemon API has its own data to track, allocate it if ((*api)->cmds->new_data != NULL) { if ((*api)->cmds->new_data(*api) != pcmk_rc_ok) { pcmk_free_ipc_api(*api); *api = NULL; return ENOMEM; } } crm_trace("Created %s API IPC object", pcmk_ipc_name(*api, true)); return pcmk_rc_ok; } static void free_daemon_specific_data(pcmk_ipc_api_t *api) { if ((api != NULL) && (api->cmds != NULL)) { if ((api->cmds->free_data != NULL) && (api->api_data != NULL)) { api->cmds->free_data(api->api_data); api->api_data = NULL; } free(api->cmds); api->cmds = NULL; } } /*! * \internal * \brief Call an IPC API event callback, if one is registed * * \param[in,out] api IPC API connection * \param[in] event_type The type of event that occurred * \param[in] status Event status * \param[in,out] event_data Event-specific data */ void pcmk__call_ipc_callback(pcmk_ipc_api_t *api, enum pcmk_ipc_event event_type, crm_exit_t status, void *event_data) { if ((api != NULL) && (api->cb != NULL)) { api->cb(api, event_type, status, event_data, api->user_data); } } /*! * \internal * \brief Clean up after an IPC disconnect * * \param[in,out] user_data IPC API connection that disconnected * * \note This function can be used as a main loop IPC destroy callback. */ static void ipc_post_disconnect(gpointer user_data) { pcmk_ipc_api_t *api = user_data; crm_info("Disconnected from %s", pcmk_ipc_name(api, true)); // Perform any daemon-specific handling needed if ((api->cmds != NULL) && (api->cmds->post_disconnect != NULL)) { api->cmds->post_disconnect(api); } // Call client's registered event callback pcmk__call_ipc_callback(api, pcmk_ipc_event_disconnect, CRM_EX_DISCONNECT, NULL); /* If this is being called from a running main loop, mainloop_gio_destroy() * will free ipc and mainloop_io immediately after calling this function. * If this is called from a stopped main loop, these will leak, so the best * practice is to close the connection before stopping the main loop. */ api->ipc = NULL; api->mainloop_io = NULL; if (api->free_on_disconnect) { /* pcmk_free_ipc_api() has already been called, but did not free api * or api->cmds because this function needed them. Do that now. */ free_daemon_specific_data(api); crm_trace("Freeing IPC API object after disconnect"); free(api); } } /*! * \brief Free the contents of an IPC API object * * \param[in,out] api IPC API object to free */ void pcmk_free_ipc_api(pcmk_ipc_api_t *api) { bool free_on_disconnect = false; if (api == NULL) { return; } crm_debug("Releasing %s IPC API", pcmk_ipc_name(api, true)); if (api->ipc != NULL) { if (api->mainloop_io != NULL) { /* We need to keep the api pointer itself around, because it is the * user data for the IPC client destroy callback. That will be * triggered by the pcmk_disconnect_ipc() call below, but it might * happen later in the main loop (if still running). * * This flag tells the destroy callback to free the object. It can't * do that unconditionally, because the application might call this * function after a disconnect that happened by other means. */ free_on_disconnect = api->free_on_disconnect = true; } pcmk_disconnect_ipc(api); // Frees api if free_on_disconnect is true } if (!free_on_disconnect) { free_daemon_specific_data(api); crm_trace("Freeing IPC API object"); free(api); } } /*! * \brief Get the IPC name used with an IPC API connection * * \param[in] api IPC API connection * \param[in] for_log If true, return human-friendly name instead of IPC name * * \return IPC API's human-friendly or connection name, or if none is available, * "Pacemaker" if for_log is true and NULL if for_log is false */ const char * pcmk_ipc_name(const pcmk_ipc_api_t *api, bool for_log) { if (api == NULL) { return for_log? "Pacemaker" : NULL; } if (for_log) { const char *name = pcmk__server_log_name(api->server); return pcmk__s(name, "Pacemaker"); } switch (api->server) { // These servers do not have pcmk_ipc_api_t implementations yet case pcmk_ipc_based: case pcmk_ipc_execd: case pcmk_ipc_fenced: return NULL; default: return pcmk__server_ipc_name(api->server); } } /*! * \brief Check whether an IPC API connection is active * * \param[in,out] api IPC API connection * * \return true if IPC is connected, false otherwise */ bool pcmk_ipc_is_connected(pcmk_ipc_api_t *api) { return (api != NULL) && crm_ipc_connected(api->ipc); } /*! * \internal * \brief Call the daemon-specific API's dispatch function * * Perform daemon-specific handling of IPC reply dispatch. It is the daemon * method's responsibility to call the client's registered event callback, as * well as allocate and free any event data. * * \param[in,out] api IPC API connection * \param[in,out] message IPC reply XML to dispatch */ static bool call_api_dispatch(pcmk_ipc_api_t *api, xmlNode *message) { crm_log_xml_trace(message, "ipc-received"); if ((api->cmds != NULL) && (api->cmds->dispatch != NULL)) { return api->cmds->dispatch(api, message); } return false; } /*! * \internal * \brief Dispatch previously read IPC data * * \param[in] buffer Data read from IPC * \param[in,out] api IPC object * * \return Standard Pacemaker return code. In particular: * * pcmk_rc_ok: There are no more messages expected from the server. Quit * reading. * EINPROGRESS: There are more messages expected from the server. Keep reading. * * All other values indicate an error. */ static int dispatch_ipc_data(const char *buffer, pcmk_ipc_api_t *api) { bool more = false; xmlNode *msg; if (buffer == NULL) { crm_warn("Empty message received from %s IPC", pcmk_ipc_name(api, true)); return ENOMSG; } msg = pcmk__xml_parse(buffer); if (msg == NULL) { crm_warn("Malformed message received from %s IPC", pcmk_ipc_name(api, true)); return EPROTO; } more = call_api_dispatch(api, msg); pcmk__xml_free(msg); if (more) { return EINPROGRESS; } else { return pcmk_rc_ok; } } /*! * \internal * \brief Dispatch data read from IPC source * * \param[in] buffer Data read from IPC * \param[in] length Number of bytes of data in buffer (ignored) * \param[in,out] user_data IPC object * * \return Always 0 (meaning connection is still required) * * \note This function can be used as a main loop IPC dispatch callback. */ static int dispatch_ipc_source_data(const char *buffer, ssize_t length, gpointer user_data) { pcmk_ipc_api_t *api = user_data; CRM_CHECK(api != NULL, return 0); dispatch_ipc_data(buffer, api); return 0; } /*! * \brief Check whether an IPC connection has data available (without main loop) * * \param[in] api IPC API connection * \param[in] timeout_ms If less than 0, poll indefinitely; if 0, poll once * and return immediately; otherwise, poll for up to * this many milliseconds * * \return Standard Pacemaker return code * * \note Callers of pcmk_connect_ipc() using pcmk_ipc_dispatch_poll should call * this function to check whether IPC data is available. Return values of * interest include pcmk_rc_ok meaning data is available, and EAGAIN * meaning no data is available; all other values indicate errors. * \todo This does not allow the caller to poll multiple file descriptors at * once. If there is demand for that, we could add a wrapper for * pcmk__ipc_fd(api->ipc), so the caller can call poll() themselves. */ int pcmk_poll_ipc(const pcmk_ipc_api_t *api, int timeout_ms) { int rc; struct pollfd pollfd = { 0, }; if ((api == NULL) || (api->dispatch_type != pcmk_ipc_dispatch_poll)) { return EINVAL; } rc = pcmk__ipc_fd(api->ipc, &(pollfd.fd)); if (rc != pcmk_rc_ok) { crm_debug("Could not obtain file descriptor for %s IPC: %s", pcmk_ipc_name(api, true), pcmk_rc_str(rc)); return rc; } pollfd.events = POLLIN; rc = poll(&pollfd, 1, timeout_ms); if (rc < 0) { /* Some UNIX systems return negative and set EAGAIN for failure to * allocate memory; standardize the return code in that case */ return (errno == EAGAIN)? ENOMEM : errno; } else if (rc == 0) { return EAGAIN; } return pcmk_rc_ok; } /*! * \brief Dispatch available messages on an IPC connection (without main loop) * * \param[in,out] api IPC API connection * * \return Standard Pacemaker return code * * \note Callers of pcmk_connect_ipc() using pcmk_ipc_dispatch_poll should call * this function when IPC data is available. */ void pcmk_dispatch_ipc(pcmk_ipc_api_t *api) { if (api == NULL) { return; } while (crm_ipc_ready(api->ipc) > 0) { if (crm_ipc_read(api->ipc) > 0) { dispatch_ipc_data(crm_ipc_buffer(api->ipc), api); } } } // \return Standard Pacemaker return code static int connect_with_main_loop(pcmk_ipc_api_t *api) { int rc; struct ipc_client_callbacks callbacks = { .dispatch = dispatch_ipc_source_data, .destroy = ipc_post_disconnect, }; rc = pcmk__add_mainloop_ipc(api->ipc, G_PRIORITY_DEFAULT, api, &callbacks, &(api->mainloop_io)); if (rc != pcmk_rc_ok) { return rc; } crm_debug("Connected to %s IPC (attached to main loop)", pcmk_ipc_name(api, true)); /* After this point, api->mainloop_io owns api->ipc, so api->ipc * should not be explicitly freed. */ return pcmk_rc_ok; } // \return Standard Pacemaker return code static int connect_without_main_loop(pcmk_ipc_api_t *api) { int rc = pcmk__connect_generic_ipc(api->ipc); if (rc != pcmk_rc_ok) { crm_ipc_close(api->ipc); } else { crm_debug("Connected to %s IPC (without main loop)", pcmk_ipc_name(api, true)); } return rc; } /*! * \internal * \brief Connect to a Pacemaker daemon via IPC (retrying after soft errors) * * \param[in,out] api IPC API instance * \param[in] dispatch_type How IPC replies should be dispatched * \param[in] attempts How many times to try (in case of soft error) * * \return Standard Pacemaker return code */ int pcmk__connect_ipc(pcmk_ipc_api_t *api, enum pcmk_ipc_dispatch dispatch_type, int attempts) { int rc = pcmk_rc_ok; if ((api == NULL) || (attempts < 1)) { return EINVAL; } if (api->ipc == NULL) { api->ipc = crm_ipc_new(pcmk_ipc_name(api, false), api->ipc_size_max); if (api->ipc == NULL) { return ENOMEM; } } if (crm_ipc_connected(api->ipc)) { crm_trace("Already connected to %s", pcmk_ipc_name(api, true)); return pcmk_rc_ok; } api->dispatch_type = dispatch_type; crm_debug("Attempting connection to %s (up to %d time%s)", pcmk_ipc_name(api, true), attempts, pcmk__plural_s(attempts)); for (int remaining = attempts - 1; remaining >= 0; --remaining) { switch (dispatch_type) { case pcmk_ipc_dispatch_main: rc = connect_with_main_loop(api); break; case pcmk_ipc_dispatch_sync: case pcmk_ipc_dispatch_poll: rc = connect_without_main_loop(api); break; } if ((remaining == 0) || ((rc != EAGAIN) && (rc != EALREADY))) { break; // Result is final } // Retry after soft error (interrupted by signal, etc.) pcmk__sleep_ms((attempts - remaining) * 500); crm_debug("Re-attempting connection to %s (%d attempt%s remaining)", pcmk_ipc_name(api, true), remaining, pcmk__plural_s(remaining)); } if (rc != pcmk_rc_ok) { return rc; } if ((api->cmds != NULL) && (api->cmds->post_connect != NULL)) { rc = api->cmds->post_connect(api); if (rc != pcmk_rc_ok) { crm_ipc_close(api->ipc); } } return rc; } /*! * \brief Connect to a Pacemaker daemon via IPC * * \param[in,out] api IPC API instance * \param[in] dispatch_type How IPC replies should be dispatched * * \return Standard Pacemaker return code */ int pcmk_connect_ipc(pcmk_ipc_api_t *api, enum pcmk_ipc_dispatch dispatch_type) { int rc = pcmk__connect_ipc(api, dispatch_type, 2); if (rc != pcmk_rc_ok) { crm_err("Connection to %s failed: %s", pcmk_ipc_name(api, true), pcmk_rc_str(rc)); } return rc; } /*! * \brief Disconnect an IPC API instance * * \param[in,out] api IPC API connection * * \return Standard Pacemaker return code * * \note If the connection is attached to a main loop, this function should be * called before quitting the main loop, to ensure that all memory is * freed. */ void pcmk_disconnect_ipc(pcmk_ipc_api_t *api) { if ((api == NULL) || (api->ipc == NULL)) { return; } switch (api->dispatch_type) { case pcmk_ipc_dispatch_main: { mainloop_io_t *mainloop_io = api->mainloop_io; // Make sure no code with access to api can use these again api->mainloop_io = NULL; api->ipc = NULL; mainloop_del_ipc_client(mainloop_io); // After this point api might have already been freed } break; case pcmk_ipc_dispatch_poll: case pcmk_ipc_dispatch_sync: { crm_ipc_t *ipc = api->ipc; // Make sure no code with access to api can use ipc again api->ipc = NULL; // This should always be the case already, but to be safe api->free_on_disconnect = false; crm_ipc_close(ipc); crm_ipc_destroy(ipc); ipc_post_disconnect(api); } break; } } /*! * \brief Register a callback for IPC API events * * \param[in,out] api IPC API connection * \param[in] callback Callback to register * \param[in] userdata Caller data to pass to callback * * \note This function may be called multiple times to update the callback * and/or user data. The caller remains responsible for freeing * userdata in any case (after the IPC is disconnected, if the * user data is still registered with the IPC). */ void pcmk_register_ipc_callback(pcmk_ipc_api_t *api, pcmk_ipc_callback_t cb, void *user_data) { if (api == NULL) { return; } api->cb = cb; api->user_data = user_data; } /*! * \internal * \brief Send an XML request across an IPC API connection * * \param[in,out] api IPC API connection * \param[in] request XML request to send * * \return Standard Pacemaker return code * * \note Daemon-specific IPC API functions should call this function to send * requests, because it handles different dispatch types appropriately. */ int pcmk__send_ipc_request(pcmk_ipc_api_t *api, const xmlNode *request) { int rc; xmlNode *reply = NULL; enum crm_ipc_flags flags = crm_ipc_flags_none; if ((api == NULL) || (api->ipc == NULL) || (request == NULL)) { return EINVAL; } crm_log_xml_trace(request, "ipc-sent"); // Synchronous dispatch requires waiting for a reply if ((api->dispatch_type == pcmk_ipc_dispatch_sync) && (api->cmds != NULL) && (api->cmds->reply_expected != NULL) && (api->cmds->reply_expected(api, request))) { flags = crm_ipc_client_response; } /* The 0 here means a default timeout of 5 seconds * * @TODO Maybe add a timeout_ms member to pcmk_ipc_api_t and a * pcmk_set_ipc_timeout() setter for it, then use it here. */ rc = crm_ipc_send(api->ipc, request, flags, 0, &reply); if (rc < 0) { return pcmk_legacy2rc(rc); } else if (rc == 0) { return ENODATA; } // With synchronous dispatch, we dispatch any reply now if (reply != NULL) { bool more = call_api_dispatch(api, reply); pcmk__xml_free(reply); while (more) { rc = crm_ipc_read(api->ipc); if (rc == -EAGAIN) { continue; } else if (rc == -ENOMSG || rc == pcmk_ok) { return pcmk_rc_ok; } else if (rc < 0) { return -rc; } rc = dispatch_ipc_data(crm_ipc_buffer(api->ipc), api); if (rc == pcmk_rc_ok) { more = false; } else if (rc == EINPROGRESS) { more = true; } else { continue; } } } return pcmk_rc_ok; } /*! * \internal * \brief Create the XML for an IPC request to purge a node from the peer cache * * \param[in] api IPC API connection * \param[in] node_name If not NULL, name of node to purge * \param[in] nodeid If not 0, node ID of node to purge * * \return Newly allocated IPC request XML * * \note The controller, fencer, and pacemakerd use the same request syntax, but * the attribute manager uses a different one. The CIB manager doesn't * have any syntax for it. The executor and scheduler don't connect to the * cluster layer and thus don't have or need any syntax for it. * * \todo Modify the attribute manager to accept the common syntax (as well * as its current one, for compatibility with older clients). Modify * the CIB manager to accept and honor the common syntax. Modify the * executor and scheduler to accept the syntax (immediately returning * success), just for consistency. Modify this function to use the * common syntax with all daemons if their version supports it. */ static xmlNode * create_purge_node_request(const pcmk_ipc_api_t *api, const char *node_name, uint32_t nodeid) { xmlNode *request = NULL; const char *client = crm_system_name? crm_system_name : "client"; switch (api->server) { case pcmk_ipc_attrd: request = pcmk__xe_create(NULL, __func__); crm_xml_add(request, PCMK__XA_T, PCMK__VALUE_ATTRD); crm_xml_add(request, PCMK__XA_SRC, crm_system_name); crm_xml_add(request, PCMK_XA_TASK, PCMK__ATTRD_CMD_PEER_REMOVE); pcmk__xe_set_bool_attr(request, PCMK__XA_REAP, true); crm_xml_add(request, PCMK__XA_ATTR_HOST, node_name); if (nodeid > 0) { crm_xml_add_int(request, PCMK__XA_ATTR_HOST_ID, nodeid); } break; case pcmk_ipc_controld: case pcmk_ipc_fenced: case pcmk_ipc_pacemakerd: request = pcmk__new_request(api->server, client, NULL, pcmk_ipc_name(api, false), CRM_OP_RM_NODE_CACHE, NULL); if (nodeid > 0) { - crm_xml_add_ll(request, PCMK_XA_ID, (long long) nodeid); + pcmk__xe_set_ll(request, PCMK_XA_ID, (long long) nodeid); } crm_xml_add(request, PCMK_XA_UNAME, node_name); break; case pcmk_ipc_based: case pcmk_ipc_execd: case pcmk_ipc_schedulerd: break; default: // pcmk_ipc_unknown (shouldn't be possible) return NULL; } return request; } /*! * \brief Ask a Pacemaker daemon to purge a node from its peer cache * * \param[in,out] api IPC API connection * \param[in] node_name If not NULL, name of node to purge * \param[in] nodeid If not 0, node ID of node to purge * * \return Standard Pacemaker return code * * \note At least one of node_name or nodeid must be specified. */ int pcmk_ipc_purge_node(pcmk_ipc_api_t *api, const char *node_name, uint32_t nodeid) { int rc = 0; xmlNode *request = NULL; if (api == NULL) { return EINVAL; } if ((node_name == NULL) && (nodeid == 0)) { return EINVAL; } request = create_purge_node_request(api, node_name, nodeid); if (request == NULL) { return EOPNOTSUPP; } rc = pcmk__send_ipc_request(api, request); pcmk__xml_free(request); crm_debug("%s peer cache purge of node %s[%lu]: rc=%d", pcmk_ipc_name(api, true), node_name, (unsigned long) nodeid, rc); return rc; } /* * Generic IPC API (to eventually be deprecated as public API and made internal) */ struct crm_ipc_s { struct pollfd pfd; unsigned int max_buf_size; // maximum bytes we can send or receive over IPC unsigned int buf_size; // size of allocated buffer int msg_size; int need_reply; char *buffer; char *server_name; // server IPC name being connected to qb_ipcc_connection_t *ipc; }; /*! * \brief Create a new (legacy) object for using Pacemaker daemon IPC * * \param[in] name IPC system name to connect to * \param[in] max_size Use a maximum IPC buffer size of at least this size * * \return Newly allocated IPC object on success, NULL otherwise * * \note The caller is responsible for freeing the result using * crm_ipc_destroy(). * \note This should be considered deprecated for use with daemons supported by * pcmk_new_ipc_api(). */ crm_ipc_t * crm_ipc_new(const char *name, size_t max_size) { crm_ipc_t *client = NULL; client = calloc(1, sizeof(crm_ipc_t)); if (client == NULL) { crm_err("Could not create IPC connection: %s", strerror(errno)); return NULL; } client->server_name = strdup(name); if (client->server_name == NULL) { crm_err("Could not create %s IPC connection: %s", name, strerror(errno)); free(client); return NULL; } client->buf_size = pcmk__ipc_buffer_size(max_size); client->buffer = malloc(client->buf_size); if (client->buffer == NULL) { crm_err("Could not create %s IPC connection: %s", name, strerror(errno)); free(client->server_name); free(client); return NULL; } /* Clients initiating connection pick the max buf size */ client->max_buf_size = client->buf_size; client->pfd.fd = -1; client->pfd.events = POLLIN; client->pfd.revents = 0; return client; } /*! * \internal * \brief Connect a generic (not daemon-specific) IPC object * * \param[in,out] ipc Generic IPC object to connect * * \return Standard Pacemaker return code */ int pcmk__connect_generic_ipc(crm_ipc_t *ipc) { uid_t cl_uid = 0; gid_t cl_gid = 0; pid_t found_pid = 0; uid_t found_uid = 0; gid_t found_gid = 0; int rc = pcmk_rc_ok; if (ipc == NULL) { return EINVAL; } ipc->need_reply = FALSE; ipc->ipc = qb_ipcc_connect(ipc->server_name, ipc->buf_size); if (ipc->ipc == NULL) { return errno; } rc = qb_ipcc_fd_get(ipc->ipc, &ipc->pfd.fd); if (rc < 0) { // -errno crm_ipc_close(ipc); return -rc; } rc = pcmk_daemon_user(&cl_uid, &cl_gid); rc = pcmk_legacy2rc(rc); if (rc != pcmk_rc_ok) { crm_ipc_close(ipc); return rc; } rc = is_ipc_provider_expected(ipc->ipc, ipc->pfd.fd, cl_uid, cl_gid, &found_pid, &found_uid, &found_gid); if (rc != pcmk_rc_ok) { if (rc == pcmk_rc_ipc_unauthorized) { crm_info("%s IPC provider authentication failed: process %lld has " "uid %lld (expected %lld) and gid %lld (expected %lld)", ipc->server_name, (long long) PCMK__SPECIAL_PID_AS_0(found_pid), (long long) found_uid, (long long) cl_uid, (long long) found_gid, (long long) cl_gid); } crm_ipc_close(ipc); return rc; } ipc->max_buf_size = qb_ipcc_get_buffer_size(ipc->ipc); if (ipc->max_buf_size > ipc->buf_size) { free(ipc->buffer); ipc->buffer = calloc(ipc->max_buf_size, sizeof(char)); if (ipc->buffer == NULL) { rc = errno; crm_ipc_close(ipc); return rc; } ipc->buf_size = ipc->max_buf_size; } return pcmk_rc_ok; } void crm_ipc_close(crm_ipc_t * client) { if (client) { if (client->ipc) { qb_ipcc_connection_t *ipc = client->ipc; client->ipc = NULL; qb_ipcc_disconnect(ipc); } } } void crm_ipc_destroy(crm_ipc_t * client) { if (client) { if (client->ipc && qb_ipcc_is_connected(client->ipc)) { crm_notice("Destroying active %s IPC connection", client->server_name); /* The next line is basically unsafe * * If this connection was attached to mainloop and mainloop is active, * the 'disconnected' callback will end up back here and we'll end * up free'ing the memory twice - something that can still happen * even without this if we destroy a connection and it closes before * we call exit */ /* crm_ipc_close(client); */ } else { crm_trace("Destroying inactive %s IPC connection", client->server_name); } free(client->buffer); free(client->server_name); free(client); } } /*! * \internal * \brief Get the file descriptor for a generic IPC object * * \param[in,out] ipc Generic IPC object to get file descriptor for * \param[out] fd Where to store file descriptor * * \return Standard Pacemaker return code */ int pcmk__ipc_fd(crm_ipc_t *ipc, int *fd) { if ((ipc == NULL) || (fd == NULL)) { return EINVAL; } if ((ipc->ipc == NULL) || (ipc->pfd.fd < 0)) { return ENOTCONN; } *fd = ipc->pfd.fd; return pcmk_rc_ok; } int crm_ipc_get_fd(crm_ipc_t * client) { int fd = -1; if (pcmk__ipc_fd(client, &fd) != pcmk_rc_ok) { crm_err("Could not obtain file descriptor for %s IPC", ((client == NULL)? "unspecified" : client->server_name)); errno = EINVAL; return -EINVAL; } return fd; } bool crm_ipc_connected(crm_ipc_t * client) { bool rc = FALSE; if (client == NULL) { crm_trace("No client"); return FALSE; } else if (client->ipc == NULL) { crm_trace("No connection"); return FALSE; } else if (client->pfd.fd < 0) { crm_trace("Bad descriptor"); return FALSE; } rc = qb_ipcc_is_connected(client->ipc); if (rc == FALSE) { client->pfd.fd = -EINVAL; } return rc; } /*! * \brief Check whether an IPC connection is ready to be read * * \param[in,out] client Connection to check * * \return Positive value if ready to be read, 0 if not ready, -errno on error */ int crm_ipc_ready(crm_ipc_t *client) { int rc; pcmk__assert(client != NULL); if (!crm_ipc_connected(client)) { return -ENOTCONN; } client->pfd.revents = 0; rc = poll(&(client->pfd), 1, 0); return (rc < 0)? -errno : rc; } // \return Standard Pacemaker return code static int crm_ipc_decompress(crm_ipc_t * client) { pcmk__ipc_header_t *header = (pcmk__ipc_header_t *)(void*)client->buffer; if (header->size_compressed) { int rc = 0; unsigned int size_u = 1 + header->size_uncompressed; /* never let buf size fall below our max size required for ipc reads. */ unsigned int new_buf_size = QB_MAX((sizeof(pcmk__ipc_header_t) + size_u), client->max_buf_size); char *uncompressed = pcmk__assert_alloc(1, new_buf_size); crm_trace("Decompressing message data %u bytes into %u bytes", header->size_compressed, size_u); rc = BZ2_bzBuffToBuffDecompress(uncompressed + sizeof(pcmk__ipc_header_t), &size_u, client->buffer + sizeof(pcmk__ipc_header_t), header->size_compressed, 1, 0); rc = pcmk__bzlib2rc(rc); if (rc != pcmk_rc_ok) { crm_err("Decompression failed: %s " QB_XS " rc=%d", pcmk_rc_str(rc), rc); free(uncompressed); return rc; } pcmk__assert(size_u == header->size_uncompressed); memcpy(uncompressed, client->buffer, sizeof(pcmk__ipc_header_t)); /* Preserve the header */ header = (pcmk__ipc_header_t *)(void*)uncompressed; free(client->buffer); client->buf_size = new_buf_size; client->buffer = uncompressed; } pcmk__assert(client->buffer[sizeof(pcmk__ipc_header_t) + header->size_uncompressed - 1] == 0); return pcmk_rc_ok; } long crm_ipc_read(crm_ipc_t * client) { pcmk__ipc_header_t *header = NULL; pcmk__assert((client != NULL) && (client->ipc != NULL) && (client->buffer != NULL)); client->buffer[0] = 0; client->msg_size = qb_ipcc_event_recv(client->ipc, client->buffer, client->buf_size, 0); if (client->msg_size >= 0) { int rc = crm_ipc_decompress(client); if (rc != pcmk_rc_ok) { return pcmk_rc2legacy(rc); } header = (pcmk__ipc_header_t *)(void*)client->buffer; if (!pcmk__valid_ipc_header(header)) { return -EBADMSG; } crm_trace("Received %s IPC event %d size=%u rc=%d text='%.100s'", client->server_name, header->qb.id, header->qb.size, client->msg_size, client->buffer + sizeof(pcmk__ipc_header_t)); } else { crm_trace("No message received from %s IPC: %s", client->server_name, pcmk_strerror(client->msg_size)); if (client->msg_size == -EAGAIN) { return -EAGAIN; } } if (!crm_ipc_connected(client) || client->msg_size == -ENOTCONN) { crm_err("Connection to %s IPC failed", client->server_name); } if (header) { /* Data excluding the header */ return header->size_uncompressed; } return -ENOMSG; } const char * crm_ipc_buffer(crm_ipc_t * client) { pcmk__assert(client != NULL); return client->buffer + sizeof(pcmk__ipc_header_t); } uint32_t crm_ipc_buffer_flags(crm_ipc_t * client) { pcmk__ipc_header_t *header = NULL; pcmk__assert(client != NULL); if (client->buffer == NULL) { return 0; } header = (pcmk__ipc_header_t *)(void*)client->buffer; return header->flags; } const char * crm_ipc_name(crm_ipc_t * client) { pcmk__assert(client != NULL); return client->server_name; } // \return Standard Pacemaker return code static int internal_ipc_get_reply(crm_ipc_t *client, int request_id, int ms_timeout, ssize_t *bytes, xmlNode **reply) { pcmk__ipc_header_t *hdr = NULL; time_t timeout = 0; int32_t qb_timeout = -1; int rc = pcmk_rc_ok; if (ms_timeout > 0) { timeout = time(NULL) + 1 + pcmk__timeout_ms2s(ms_timeout); qb_timeout = 1000; } /* get the reply */ crm_trace("Expecting reply to %s IPC message %d", client->server_name, request_id); do { xmlNode *xml = NULL; *bytes = qb_ipcc_recv(client->ipc, client->buffer, client->buf_size, qb_timeout); if (*bytes <= 0) { if (!crm_ipc_connected(client)) { crm_err("%s IPC provider disconnected while waiting for message %d", client->server_name, request_id); break; } continue; } rc = crm_ipc_decompress(client); if (rc != pcmk_rc_ok) { return rc; } hdr = (pcmk__ipc_header_t *)(void*) client->buffer; if (hdr->qb.id == request_id) { /* Got the reply we were expecting. */ break; } xml = pcmk__xml_parse(crm_ipc_buffer(client)); if (hdr->qb.id < request_id) { crm_err("Discarding old reply %d (need %d)", hdr->qb.id, request_id); crm_log_xml_notice(xml, "OldIpcReply"); } else if (hdr->qb.id > request_id) { crm_err("Discarding newer reply %d (need %d)", hdr->qb.id, request_id); crm_log_xml_notice(xml, "ImpossibleReply"); pcmk__assert(hdr->qb.id <= request_id); } } while (time(NULL) < timeout || (timeout == 0 && *bytes == -EAGAIN)); if (*bytes > 0) { crm_trace("Received %zd-byte reply %" PRId32 " to %s IPC %d: %.100s", *bytes, hdr->qb.id, client->server_name, request_id, crm_ipc_buffer(client)); if (reply != NULL) { *reply = pcmk__xml_parse(crm_ipc_buffer(client)); } } else if (*bytes < 0) { rc = (int) -*bytes; // System errno crm_trace("No reply to %s IPC %d: %s " QB_XS " rc=%d", client->server_name, request_id, pcmk_rc_str(rc), rc); } /* If bytes == 0, we'll return that to crm_ipc_send which will interpret * that as pcmk_rc_ok, log that the IPC request failed (since we did not * give it a valid reply), and return that 0 to its callers. It's up to * the callers to take appropriate action after that. */ return rc; } /*! * \brief Send an IPC XML message * * \param[in,out] client Connection to IPC server * \param[in] message XML message to send * \param[in] flags Bitmask of crm_ipc_flags * \param[in] ms_timeout Give up if not sent within this much time * (5 seconds if 0, or no timeout if negative) * \param[out] reply Reply from server (or NULL if none) * * \return Negative errno on error, otherwise size of reply received in bytes * if reply was needed, otherwise number of bytes sent */ int crm_ipc_send(crm_ipc_t *client, const xmlNode *message, enum crm_ipc_flags flags, int32_t ms_timeout, xmlNode **reply) { int rc = 0; time_t timeout = 0; ssize_t qb_rc = 0; ssize_t bytes = 0; struct iovec *iov; static uint32_t id = 0; static int factor = 8; pcmk__ipc_header_t *header; if (client == NULL) { crm_notice("Can't send IPC request without connection (bug?): %.100s", message); return -ENOTCONN; } else if (!crm_ipc_connected(client)) { /* Don't even bother */ crm_notice("Can't send %s IPC requests: Connection closed", client->server_name); return -ENOTCONN; } if (ms_timeout == 0) { ms_timeout = 5000; } if (client->need_reply) { qb_rc = qb_ipcc_recv(client->ipc, client->buffer, client->buf_size, ms_timeout); if (qb_rc < 0) { crm_warn("Sending %s IPC disabled until pending reply received", client->server_name); return -EALREADY; } else { crm_notice("Sending %s IPC re-enabled after pending reply received", client->server_name); client->need_reply = FALSE; } } id++; CRM_LOG_ASSERT(id != 0); /* Crude wrap-around detection */ rc = pcmk__ipc_prepare_iov(id, message, client->max_buf_size, &iov, &bytes); if (rc != pcmk_rc_ok) { crm_warn("Couldn't prepare %s IPC request: %s " QB_XS " rc=%d", client->server_name, pcmk_rc_str(rc), rc); return pcmk_rc2legacy(rc); } header = iov[0].iov_base; pcmk__set_ipc_flags(header->flags, client->server_name, flags); if (pcmk_is_set(flags, crm_ipc_proxied)) { /* Don't look for a synchronous response */ pcmk__clear_ipc_flags(flags, "client", crm_ipc_client_response); } if(header->size_compressed) { if(factor < 10 && (client->max_buf_size / 10) < (bytes / factor)) { crm_notice("Compressed message exceeds %d0%% of configured IPC " "limit (%u bytes); consider setting PCMK_ipc_buffer to " "%u or higher", factor, client->max_buf_size, 2 * client->max_buf_size); factor++; } } crm_trace("Sending %s IPC request %d of %u bytes using %dms timeout", client->server_name, header->qb.id, header->qb.size, ms_timeout); /* Send the IPC request, respecting any timeout we were passed */ if (ms_timeout > 0) { timeout = time(NULL) + 1 + pcmk__timeout_ms2s(ms_timeout); } do { qb_rc = qb_ipcc_sendv(client->ipc, iov, 2); } while ((qb_rc == -EAGAIN) && ((timeout == 0) || (time(NULL) < timeout))); rc = (int) qb_rc; // Negative of system errno, or bytes sent if (qb_rc <= 0) { goto send_cleanup; } /* If we should not wait for a response, bail now */ if (!pcmk_is_set(flags, crm_ipc_client_response)) { crm_trace("Not waiting for reply to %s IPC request %d", client->server_name, header->qb.id); goto send_cleanup; } rc = internal_ipc_get_reply(client, header->qb.id, ms_timeout, &bytes, reply); if (rc == pcmk_rc_ok) { rc = (int) bytes; // Size of reply received } else { /* rc is either a positive system errno or a negative standard Pacemaker * return code. If it's an errno, we need to convert it back to a * negative number for comparison and return at the end of this function. */ rc = pcmk_rc2legacy(rc); if (ms_timeout > 0) { /* We didn't get the reply in time, so disable future sends for now. * The only alternative would be to close the connection since we * don't know how to detect and discard out-of-sequence replies. * * @TODO Implement out-of-sequence detection */ client->need_reply = TRUE; } } send_cleanup: if (!crm_ipc_connected(client)) { crm_notice("Couldn't send %s IPC request %d: Connection closed " QB_XS " rc=%d", client->server_name, header->qb.id, rc); } else if (rc == -ETIMEDOUT) { crm_warn("%s IPC request %d failed: %s after %dms " QB_XS " rc=%d", client->server_name, header->qb.id, pcmk_strerror(rc), ms_timeout, rc); crm_write_blackbox(0, NULL); } else if (rc <= 0) { crm_warn("%s IPC request %d failed: %s " QB_XS " rc=%d", client->server_name, header->qb.id, ((rc == 0)? "No bytes sent" : pcmk_strerror(rc)), rc); } pcmk_free_ipc_event(iov); return rc; } /*! * \brief Ensure an IPC provider has expected user or group * * \param[in] qb_ipc libqb client connection if available * \param[in] sock Connected Unix socket for IPC * \param[in] refuid Expected user ID * \param[in] refgid Expected group ID * \param[out] gotpid If not NULL, where to store provider's actual process ID * (or 1 on platforms where ID is not available) * \param[out] gotuid If not NULL, where to store provider's actual user ID * \param[out] gotgid If not NULL, where to store provider's actual group ID * * \return Standard Pacemaker return code * \note An actual user ID of 0 (root) will always be considered authorized, * regardless of the expected values provided. The caller can use the * output arguments to be stricter than this function. */ static int is_ipc_provider_expected(qb_ipcc_connection_t *qb_ipc, int sock, uid_t refuid, gid_t refgid, pid_t *gotpid, uid_t *gotuid, gid_t *gotgid) { int rc = EOPNOTSUPP; pid_t found_pid = 0; uid_t found_uid = 0; gid_t found_gid = 0; #ifdef HAVE_QB_IPCC_AUTH_GET if (qb_ipc != NULL) { rc = qb_ipcc_auth_get(qb_ipc, &found_pid, &found_uid, &found_gid); rc = -rc; // libqb returns 0 or -errno if (rc == pcmk_rc_ok) { goto found; } } #endif #ifdef HAVE_UCRED { struct ucred ucred; socklen_t ucred_len = sizeof(ucred); if (getsockopt(sock, SOL_SOCKET, SO_PEERCRED, &ucred, &ucred_len) < 0) { rc = errno; } else if (ucred_len != sizeof(ucred)) { rc = EOPNOTSUPP; } else { found_pid = ucred.pid; found_uid = ucred.uid; found_gid = ucred.gid; goto found; } } #endif #ifdef HAVE_SOCKPEERCRED { struct sockpeercred sockpeercred; socklen_t sockpeercred_len = sizeof(sockpeercred); if (getsockopt(sock, SOL_SOCKET, SO_PEERCRED, &sockpeercred, &sockpeercred_len) < 0) { rc = errno; } else if (sockpeercred_len != sizeof(sockpeercred)) { rc = EOPNOTSUPP; } else { found_pid = sockpeercred.pid; found_uid = sockpeercred.uid; found_gid = sockpeercred.gid; goto found; } } #endif #ifdef HAVE_GETPEEREID // For example, FreeBSD if (getpeereid(sock, &found_uid, &found_gid) < 0) { rc = errno; } else { found_pid = PCMK__SPECIAL_PID; goto found; } #endif #ifdef HAVE_GETPEERUCRED { ucred_t *ucred = NULL; if (getpeerucred(sock, &ucred) < 0) { rc = errno; } else { found_pid = ucred_getpid(ucred); found_uid = ucred_geteuid(ucred); found_gid = ucred_getegid(ucred); ucred_free(ucred); goto found; } } #endif return rc; // If we get here, nothing succeeded found: if (gotpid != NULL) { *gotpid = found_pid; } if (gotuid != NULL) { *gotuid = found_uid; } if (gotgid != NULL) { *gotgid = found_gid; } if ((found_uid != 0) && (found_uid != refuid) && (found_gid != refgid)) { return pcmk_rc_ipc_unauthorized; } return pcmk_rc_ok; } int crm_ipc_is_authentic_process(int sock, uid_t refuid, gid_t refgid, pid_t *gotpid, uid_t *gotuid, gid_t *gotgid) { int ret = is_ipc_provider_expected(NULL, sock, refuid, refgid, gotpid, gotuid, gotgid); /* The old function had some very odd return codes*/ if (ret == 0) { return 1; } else if (ret == pcmk_rc_ipc_unauthorized) { return 0; } else { return pcmk_rc2legacy(ret); } } int pcmk__ipc_is_authentic_process_active(const char *name, uid_t refuid, gid_t refgid, pid_t *gotpid) { static char last_asked_name[PATH_MAX / 2] = ""; /* log spam prevention */ int fd; int rc = pcmk_rc_ipc_unresponsive; int auth_rc = 0; int32_t qb_rc; pid_t found_pid = 0; uid_t found_uid = 0; gid_t found_gid = 0; qb_ipcc_connection_t *c; #ifdef HAVE_QB_IPCC_CONNECT_ASYNC struct pollfd pollfd = { 0, }; int poll_rc; c = qb_ipcc_connect_async(name, 0, &(pollfd.fd)); #else c = qb_ipcc_connect(name, 0); #endif if (c == NULL) { crm_info("Could not connect to %s IPC: %s", name, strerror(errno)); rc = pcmk_rc_ipc_unresponsive; goto bail; } #ifdef HAVE_QB_IPCC_CONNECT_ASYNC pollfd.events = POLLIN; do { poll_rc = poll(&pollfd, 1, 2000); } while ((poll_rc == -1) && (errno == EINTR)); /* If poll() failed, given that disconnect function is not registered yet, * qb_ipcc_disconnect() won't clean up the socket. In any case, call * qb_ipcc_connect_continue() here so that it may fail and do the cleanup * for us. */ if (qb_ipcc_connect_continue(c) != 0) { crm_info("Could not connect to %s IPC: %s", name, (poll_rc == 0)?"timeout":strerror(errno)); rc = pcmk_rc_ipc_unresponsive; c = NULL; // qb_ipcc_connect_continue cleaned up for us goto bail; } #endif qb_rc = qb_ipcc_fd_get(c, &fd); if (qb_rc != 0) { rc = (int) -qb_rc; // System errno crm_err("Could not get fd from %s IPC: %s " QB_XS " rc=%d", name, pcmk_rc_str(rc), rc); goto bail; } auth_rc = is_ipc_provider_expected(c, fd, refuid, refgid, &found_pid, &found_uid, &found_gid); if (auth_rc == pcmk_rc_ipc_unauthorized) { crm_err("Daemon (IPC %s) effectively blocked with unauthorized" " process %lld (uid: %lld, gid: %lld)", name, (long long) PCMK__SPECIAL_PID_AS_0(found_pid), (long long) found_uid, (long long) found_gid); rc = pcmk_rc_ipc_unauthorized; goto bail; } if (auth_rc != pcmk_rc_ok) { rc = auth_rc; crm_err("Could not get peer credentials from %s IPC: %s " QB_XS " rc=%d", name, pcmk_rc_str(rc), rc); goto bail; } if (gotpid != NULL) { *gotpid = found_pid; } rc = pcmk_rc_ok; if ((found_uid != refuid || found_gid != refgid) && strncmp(last_asked_name, name, sizeof(last_asked_name))) { if ((found_uid == 0) && (refuid != 0)) { crm_warn("Daemon (IPC %s) runs as root, whereas the expected" " credentials are %lld:%lld, hazard of violating" " the least privilege principle", name, (long long) refuid, (long long) refgid); } else { crm_notice("Daemon (IPC %s) runs as %lld:%lld, whereas the" " expected credentials are %lld:%lld, which may" " mean a different set of privileges than expected", name, (long long) found_uid, (long long) found_gid, (long long) refuid, (long long) refgid); } memccpy(last_asked_name, name, '\0', sizeof(last_asked_name)); } bail: if (c != NULL) { qb_ipcc_disconnect(c); } return rc; } // Deprecated functions kept only for backward API compatibility // LCOV_EXCL_START #include bool crm_ipc_connect(crm_ipc_t *client) { int rc = pcmk__connect_generic_ipc(client); if (rc == pcmk_rc_ok) { return true; } if ((client != NULL) && (client->ipc == NULL)) { errno = (rc > 0)? rc : ENOTCONN; crm_debug("Could not establish %s IPC connection: %s (%d)", client->server_name, pcmk_rc_str(errno), errno); } else if (rc == pcmk_rc_ipc_unauthorized) { crm_err("%s IPC provider authentication failed", (client == NULL)? "Pacemaker" : client->server_name); errno = ECONNABORTED; } else { crm_err("Could not verify authenticity of %s IPC provider", (client == NULL)? "Pacemaker" : client->server_name); errno = ENOTCONN; } return false; } // LCOV_EXCL_STOP // End deprecated API diff --git a/lib/common/ipc_controld.c b/lib/common/ipc_controld.c index 6951c8fc7d..a20b4dfb20 100644 --- a/lib/common/ipc_controld.c +++ b/lib/common/ipc_controld.c @@ -1,672 +1,672 @@ /* * Copyright 2020-2025 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #include #include #include // PRIu32 #include #include // uint32_t #include #include #include #include #include #include #include #include "crmcommon_private.h" struct controld_api_private_s { char *client_uuid; unsigned int replies_expected; }; static xmlNode *create_hello_message(const char *uuid, const char *client_name, const char *major_version, const char *minor_version); /*! * \internal * \brief Get a string representation of a controller API reply type * * \param[in] reply Controller API reply type * * \return String representation of a controller API reply type */ const char * pcmk__controld_api_reply2str(enum pcmk_controld_api_reply reply) { switch (reply) { case pcmk_controld_reply_reprobe: return "reprobe"; case pcmk_controld_reply_info: return "info"; case pcmk_controld_reply_resource: return "resource"; case pcmk_controld_reply_ping: return "ping"; case pcmk_controld_reply_nodes: return "nodes"; default: return "unknown"; } } // \return Standard Pacemaker return code static int new_data(pcmk_ipc_api_t *api) { struct controld_api_private_s *private = NULL; api->api_data = calloc(1, sizeof(struct controld_api_private_s)); if (api->api_data == NULL) { return errno; } private = api->api_data; /* This is set to the PID because that's how it was always done, but PIDs * are not unique because clients can be remote. The value appears to be * unused other than as part of PCMK__XA_CRM_SYS_FROM in IPC requests, which * is only compared against the internal system names (CRM_SYSTEM_TENGINE, * etc.), so it shouldn't be a problem. */ private->client_uuid = pcmk__getpid_s(); /* @TODO Implement a call ID model similar to the CIB, executor, and fencer * IPC APIs, so that requests and replies can be matched, and * duplicate replies can be discarded. */ return pcmk_rc_ok; } static void free_data(void *data) { free(((struct controld_api_private_s *) data)->client_uuid); free(data); } // \return Standard Pacemaker return code static int post_connect(pcmk_ipc_api_t *api) { /* The controller currently requires clients to register via a hello * request, but does not reply back. */ struct controld_api_private_s *private = api->api_data; const char *client_name = crm_system_name? crm_system_name : "client"; xmlNode *hello; int rc; hello = create_hello_message(private->client_uuid, client_name, PCMK__CONTROLD_API_MAJOR, PCMK__CONTROLD_API_MINOR); rc = pcmk__send_ipc_request(api, hello); pcmk__xml_free(hello); if (rc != pcmk_rc_ok) { crm_info("Could not send IPC hello to %s: %s " QB_XS " rc=%s", pcmk_ipc_name(api, true), pcmk_rc_str(rc), rc); } else { crm_debug("Sent IPC hello to %s", pcmk_ipc_name(api, true)); } return rc; } static void set_node_info_data(pcmk_controld_api_reply_t *data, xmlNode *msg_data) { data->reply_type = pcmk_controld_reply_info; if (msg_data == NULL) { return; } data->data.node_info.have_quorum = pcmk__xe_attr_is_true(msg_data, PCMK_XA_HAVE_QUORUM); data->data.node_info.is_remote = pcmk__xe_attr_is_true(msg_data, PCMK_XA_REMOTE_NODE); /* Integer node_info.id is currently valid only for Corosync nodes. * * @TODO: Improve handling after pcmk__node_status_t is refactored to handle * layer-specific data better. */ pcmk__xe_get_int(msg_data, PCMK_XA_ID, &(data->data.node_info.id)); data->data.node_info.uuid = pcmk__xe_get(msg_data, PCMK_XA_ID); data->data.node_info.uname = pcmk__xe_get(msg_data, PCMK_XA_UNAME); data->data.node_info.state = pcmk__xe_get(msg_data, PCMK_XA_CRMD); } static void set_ping_data(pcmk_controld_api_reply_t *data, xmlNode *msg_data) { data->reply_type = pcmk_controld_reply_ping; if (msg_data == NULL) { return; } data->data.ping.sys_from = pcmk__xe_get(msg_data, PCMK__XA_CRM_SUBSYSTEM); data->data.ping.fsa_state = pcmk__xe_get(msg_data, PCMK__XA_CRMD_STATE); data->data.ping.result = pcmk__xe_get(msg_data, PCMK_XA_RESULT); } static void set_nodes_data(pcmk_controld_api_reply_t *data, xmlNode *msg_data) { pcmk_controld_api_node_t *node_info; data->reply_type = pcmk_controld_reply_nodes; for (xmlNode *node = pcmk__xe_first_child(msg_data, PCMK_XE_NODE, NULL, NULL); node != NULL; node = pcmk__xe_next(node, PCMK_XE_NODE)) { long long id_ll = 0; node_info = pcmk__assert_alloc(1, sizeof(pcmk_controld_api_node_t)); pcmk__xe_get_ll(node, PCMK_XA_ID, &id_ll); if (id_ll > 0) { node_info->id = id_ll; } node_info->uname = pcmk__xe_get(node, PCMK_XA_UNAME); node_info->state = pcmk__xe_get(node, PCMK__XA_IN_CCM); data->data.nodes = g_list_prepend(data->data.nodes, node_info); } } static bool reply_expected(pcmk_ipc_api_t *api, const xmlNode *request) { // We only need to handle commands that API functions can send return pcmk__str_any_of(pcmk__xe_get(request, PCMK__XA_CRM_TASK), PCMK__CONTROLD_CMD_NODES, CRM_OP_LRM_DELETE, CRM_OP_LRM_FAIL, CRM_OP_NODE_INFO, CRM_OP_PING, CRM_OP_REPROBE, CRM_OP_RM_NODE_CACHE, NULL); } static bool dispatch(pcmk_ipc_api_t *api, xmlNode *reply) { struct controld_api_private_s *private = api->api_data; crm_exit_t status = CRM_EX_OK; xmlNode *wrapper = NULL; xmlNode *msg_data = NULL; const char *value = NULL; pcmk_controld_api_reply_t reply_data = { pcmk_controld_reply_unknown, NULL, NULL, }; if (pcmk__xe_is(reply, PCMK__XE_ACK)) { /* ACKs are trivial responses that do not count toward expected replies, * and do not have all the fields that validation requires, so skip that * processing. */ return private->replies_expected > 0; } if (private->replies_expected > 0) { private->replies_expected--; } // Do some basic validation of the reply value = pcmk__xe_get(reply, PCMK__XA_SUBT); if (pcmk__str_eq(value, PCMK__VALUE_REQUEST, pcmk__str_none)) { /* @COMPAT Controllers <3.0.0 set PCMK__XA_SUBT to PCMK__VALUE_REQUEST * for certain replies. Once we no longer support Pacemaker Remote nodes * connecting to cluster nodes <3.0.0, or rolling upgrades from <3.0.0, * we can drop this check. */ crm_trace("Received a reply that was marked as a request " "(bug unless sent by a controller <3.0.0)"); } else if (!pcmk__str_eq(value, PCMK__VALUE_RESPONSE, pcmk__str_none)) { crm_info("Unrecognizable message from controller: " "invalid message type '%s'", pcmk__s(value, "")); status = CRM_EX_PROTOCOL; goto done; } if (pcmk__str_empty(pcmk__xe_get(reply, PCMK_XA_REFERENCE))) { crm_info("Unrecognizable message from controller: no reference"); status = CRM_EX_PROTOCOL; goto done; } value = pcmk__xe_get(reply, PCMK__XA_CRM_TASK); if (pcmk__str_empty(value)) { crm_info("Unrecognizable message from controller: no command name"); status = CRM_EX_PROTOCOL; goto done; } // Parse useful info from reply reply_data.feature_set = pcmk__xe_get(reply, PCMK_XA_VERSION); reply_data.host_from = pcmk__xe_get(reply, PCMK__XA_SRC); wrapper = pcmk__xe_first_child(reply, PCMK__XE_CRM_XML, NULL, NULL); msg_data = pcmk__xe_first_child(wrapper, NULL, NULL, NULL); if (!strcmp(value, CRM_OP_REPROBE)) { reply_data.reply_type = pcmk_controld_reply_reprobe; } else if (!strcmp(value, CRM_OP_NODE_INFO)) { set_node_info_data(&reply_data, msg_data); } else if (!strcmp(value, CRM_OP_INVOKE_LRM)) { reply_data.reply_type = pcmk_controld_reply_resource; reply_data.data.resource.node_state = msg_data; } else if (!strcmp(value, CRM_OP_PING)) { set_ping_data(&reply_data, msg_data); } else if (!strcmp(value, PCMK__CONTROLD_CMD_NODES)) { set_nodes_data(&reply_data, msg_data); } else { crm_info("Unrecognizable message from controller: unknown command '%s'", value); status = CRM_EX_PROTOCOL; } done: pcmk__call_ipc_callback(api, pcmk_ipc_event_reply, status, &reply_data); // Free any reply data that was allocated if (pcmk__str_eq(value, PCMK__CONTROLD_CMD_NODES, pcmk__str_casei)) { g_list_free_full(reply_data.data.nodes, free); } return false; // No further replies needed } pcmk__ipc_methods_t * pcmk__controld_api_methods(void) { pcmk__ipc_methods_t *cmds = calloc(1, sizeof(pcmk__ipc_methods_t)); if (cmds != NULL) { cmds->new_data = new_data; cmds->free_data = free_data; cmds->post_connect = post_connect; cmds->reply_expected = reply_expected; cmds->dispatch = dispatch; } return cmds; } /*! * \internal * \brief Create XML for a controller IPC request * * \param[in] api Controller connection * \param[in] op Controller IPC command name * \param[in] node Node name to set as destination host * \param[in] msg_data XML to attach to request as message data * * \return Newly allocated XML for request */ static xmlNode * create_controller_request(const pcmk_ipc_api_t *api, const char *op, const char *node, xmlNode *msg_data) { struct controld_api_private_s *private = NULL; const char *sys_to = NULL; char *sender_system = NULL; xmlNode *request = NULL; if (api == NULL) { return NULL; } private = api->api_data; if ((node == NULL) && !strcmp(op, CRM_OP_PING)) { sys_to = CRM_SYSTEM_DC; } else { sys_to = CRM_SYSTEM_CRMD; } sender_system = crm_strdup_printf("%s_%s", private->client_uuid, pcmk__s(crm_system_name, "client")); request = pcmk__new_request(pcmk_ipc_controld, sender_system, node, sys_to, op, msg_data); free(sender_system); return request; } // \return Standard Pacemaker return code static int send_controller_request(pcmk_ipc_api_t *api, const xmlNode *request, bool reply_is_expected) { if (pcmk__xe_get(request, PCMK_XA_REFERENCE) == NULL) { return EINVAL; } if (reply_is_expected) { struct controld_api_private_s *private = api->api_data; private->replies_expected++; } return pcmk__send_ipc_request(api, request); } static xmlNode * create_reprobe_message_data(const char *target_node, const char *router_node) { xmlNode *msg_data; msg_data = pcmk__xe_create(NULL, "data_for_" CRM_OP_REPROBE); crm_xml_add(msg_data, PCMK__META_ON_NODE, target_node); if ((router_node != NULL) && !pcmk__str_eq(router_node, target_node, pcmk__str_casei)) { crm_xml_add(msg_data, PCMK__XA_ROUTER_NODE, router_node); } return msg_data; } /*! * \brief Send a reprobe controller operation * * \param[in,out] api Controller connection * \param[in] target_node Name of node to reprobe * \param[in] router_node Router node for host * * \return Standard Pacemaker return code * \note Event callback will get a reply of type pcmk_controld_reply_reprobe. */ int pcmk_controld_api_reprobe(pcmk_ipc_api_t *api, const char *target_node, const char *router_node) { xmlNode *request; xmlNode *msg_data; int rc = pcmk_rc_ok; if (api == NULL) { return EINVAL; } if (router_node == NULL) { router_node = target_node; } crm_debug("Sending %s IPC request to reprobe %s via %s", pcmk_ipc_name(api, true), pcmk__s(target_node, "local node"), pcmk__s(router_node, "local node")); msg_data = create_reprobe_message_data(target_node, router_node); request = create_controller_request(api, CRM_OP_REPROBE, router_node, msg_data); rc = send_controller_request(api, request, true); pcmk__xml_free(msg_data); pcmk__xml_free(request); return rc; } /*! * \brief Send a "node info" controller operation * * \param[in,out] api Controller connection * \param[in] nodeid ID of node to get info for (or 0 for local node) * * \return Standard Pacemaker return code * \note Event callback will get a reply of type pcmk_controld_reply_info. */ int pcmk_controld_api_node_info(pcmk_ipc_api_t *api, uint32_t nodeid) { xmlNode *request; int rc = pcmk_rc_ok; request = create_controller_request(api, CRM_OP_NODE_INFO, NULL, NULL); if (request == NULL) { return EINVAL; } if (nodeid > 0) { - crm_xml_add_ll(request, PCMK_XA_ID, nodeid); + pcmk__xe_set_ll(request, PCMK_XA_ID, nodeid); } rc = send_controller_request(api, request, true); pcmk__xml_free(request); return rc; } /*! * \brief Ask the controller for status * * \param[in,out] api Controller connection * \param[in] node_name Name of node whose status is desired (NULL for DC) * * \return Standard Pacemaker return code * \note Event callback will get a reply of type pcmk_controld_reply_ping. */ int pcmk_controld_api_ping(pcmk_ipc_api_t *api, const char *node_name) { xmlNode *request; int rc = pcmk_rc_ok; request = create_controller_request(api, CRM_OP_PING, node_name, NULL); if (request == NULL) { return EINVAL; } rc = send_controller_request(api, request, true); pcmk__xml_free(request); return rc; } /*! * \brief Ask the controller for cluster information * * \param[in,out] api Controller connection * * \return Standard Pacemaker return code * \note Event callback will get a reply of type pcmk_controld_reply_nodes. */ int pcmk_controld_api_list_nodes(pcmk_ipc_api_t *api) { xmlNode *request; int rc = EINVAL; request = create_controller_request(api, PCMK__CONTROLD_CMD_NODES, NULL, NULL); if (request != NULL) { rc = send_controller_request(api, request, true); pcmk__xml_free(request); } return rc; } // \return Standard Pacemaker return code static int controller_resource_op(pcmk_ipc_api_t *api, const char *op, const char *target_node, const char *router_node, bool cib_only, const char *rsc_id, const char *rsc_long_id, const char *standard, const char *provider, const char *type) { int rc = pcmk_rc_ok; char *key; xmlNode *request, *msg_data, *xml_rsc, *params; if (api == NULL) { return EINVAL; } if (router_node == NULL) { router_node = target_node; } msg_data = pcmk__xe_create(NULL, PCMK__XE_RSC_OP); /* The controller logs the transition key from resource op requests, so we * need to have *something* for it. * @TODO don't use "crm-resource" */ key = pcmk__transition_key(0, getpid(), 0, "xxxxxxxx-xrsc-opxx-xcrm-resourcexxxx"); crm_xml_add(msg_data, PCMK__XA_TRANSITION_KEY, key); free(key); crm_xml_add(msg_data, PCMK__META_ON_NODE, target_node); if (!pcmk__str_eq(router_node, target_node, pcmk__str_casei)) { crm_xml_add(msg_data, PCMK__XA_ROUTER_NODE, router_node); } if (cib_only) { // Indicate that only the CIB needs to be cleaned crm_xml_add(msg_data, PCMK__XA_MODE, PCMK__VALUE_CIB); } xml_rsc = pcmk__xe_create(msg_data, PCMK_XE_PRIMITIVE); crm_xml_add(xml_rsc, PCMK_XA_ID, rsc_id); crm_xml_add(xml_rsc, PCMK__XA_LONG_ID, rsc_long_id); crm_xml_add(xml_rsc, PCMK_XA_CLASS, standard); crm_xml_add(xml_rsc, PCMK_XA_PROVIDER, provider); crm_xml_add(xml_rsc, PCMK_XA_TYPE, type); params = pcmk__xe_create(msg_data, PCMK__XE_ATTRIBUTES); crm_xml_add(params, PCMK_XA_CRM_FEATURE_SET, CRM_FEATURE_SET); // The controller parses the timeout from the request key = crm_meta_name(PCMK_META_TIMEOUT); crm_xml_add(params, key, "60000"); /* 1 minute */ //@TODO pass as arg free(key); request = create_controller_request(api, op, router_node, msg_data); rc = send_controller_request(api, request, true); pcmk__xml_free(msg_data); pcmk__xml_free(request); return rc; } /*! * \brief Ask the controller to fail a resource * * \param[in,out] api Controller connection * \param[in] target_node Name of node resource is on * \param[in] router_node Router node for target * \param[in] rsc_id ID of resource to fail * \param[in] rsc_long_id Long ID of resource (if any) * \param[in] standard Standard of resource * \param[in] provider Provider of resource (if any) * \param[in] type Type of resource to fail * * \return Standard Pacemaker return code * \note Event callback will get a reply of type pcmk_controld_reply_resource. */ int pcmk_controld_api_fail(pcmk_ipc_api_t *api, const char *target_node, const char *router_node, const char *rsc_id, const char *rsc_long_id, const char *standard, const char *provider, const char *type) { crm_debug("Sending %s IPC request to fail %s (a.k.a. %s) on %s via %s", pcmk_ipc_name(api, true), pcmk__s(rsc_id, "unknown resource"), pcmk__s(rsc_long_id, "no other names"), pcmk__s(target_node, "unspecified node"), pcmk__s(router_node, "unspecified node")); return controller_resource_op(api, CRM_OP_LRM_FAIL, target_node, router_node, false, rsc_id, rsc_long_id, standard, provider, type); } /*! * \brief Ask the controller to refresh a resource * * \param[in,out] api Controller connection * \param[in] target_node Name of node resource is on * \param[in] router_node Router node for target * \param[in] rsc_id ID of resource to refresh * \param[in] rsc_long_id Long ID of resource (if any) * \param[in] standard Standard of resource * \param[in] provider Provider of resource (if any) * \param[in] type Type of resource * \param[in] cib_only If true, clean resource from CIB only * * \return Standard Pacemaker return code * \note Event callback will get a reply of type pcmk_controld_reply_resource. */ int pcmk_controld_api_refresh(pcmk_ipc_api_t *api, const char *target_node, const char *router_node, const char *rsc_id, const char *rsc_long_id, const char *standard, const char *provider, const char *type, bool cib_only) { crm_debug("Sending %s IPC request to refresh %s (a.k.a. %s) on %s via %s", pcmk_ipc_name(api, true), pcmk__s(rsc_id, "unknown resource"), pcmk__s(rsc_long_id, "no other names"), pcmk__s(target_node, "unspecified node"), pcmk__s(router_node, "unspecified node")); return controller_resource_op(api, CRM_OP_LRM_DELETE, target_node, router_node, cib_only, rsc_id, rsc_long_id, standard, provider, type); } /*! * \brief Get the number of IPC replies currently expected from the controller * * \param[in] api Controller IPC API connection * * \return Number of replies expected */ unsigned int pcmk_controld_api_replies_expected(const pcmk_ipc_api_t *api) { struct controld_api_private_s *private = api->api_data; return private->replies_expected; } /*! * \internal * \brief Create XML for a controller IPC "hello" message */ static xmlNode * create_hello_message(const char *uuid, const char *client_name, const char *major_version, const char *minor_version) { xmlNode *hello_node = NULL; xmlNode *hello = NULL; char *sender_system = NULL; if (pcmk__str_empty(uuid) || pcmk__str_empty(client_name) || pcmk__str_empty(major_version) || pcmk__str_empty(minor_version)) { crm_err("Could not create IPC hello message from %s (UUID %s): " "missing information", client_name? client_name : "unknown client", uuid? uuid : "unknown"); return NULL; } hello_node = pcmk__xe_create(NULL, PCMK__XE_OPTIONS); crm_xml_add(hello_node, PCMK__XA_MAJOR_VERSION, major_version); crm_xml_add(hello_node, PCMK__XA_MINOR_VERSION, minor_version); crm_xml_add(hello_node, PCMK__XA_CLIENT_NAME, client_name); // @TODO Nothing uses this. Drop, or keep for debugging? crm_xml_add(hello_node, PCMK__XA_CLIENT_UUID, uuid); sender_system = crm_strdup_printf("%s_%s", uuid, client_name); hello = pcmk__new_request(pcmk_ipc_controld, sender_system, NULL, NULL, CRM_OP_HELLO, hello_node); free(sender_system); pcmk__xml_free(hello_node); if (hello == NULL) { crm_err("Could not create IPC hello message from %s (UUID %s): " "Request creation failed", client_name, uuid); return NULL; } crm_trace("Created hello message from %s (UUID %s)", client_name, uuid); return hello; } diff --git a/lib/common/xml_element.c b/lib/common/xml_element.c index f68c005ba3..eca3f5a859 100644 --- a/lib/common/xml_element.c +++ b/lib/common/xml_element.c @@ -1,1794 +1,1827 @@ /* * Copyright 2004-2025 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #include #include // va_start(), etc. #include // uint32_t #include // NULL, etc. #include // free(), etc. #include // strchr(), etc. #include // time_t, etc. #include // xmlNode, etc. #include // xmlValidateNameValue() #include // xmlChar #include #include // crm_xml_add(), etc. #include // pcmk_rc_ok, etc. #include #include "crmcommon_private.h" /*! * \internal * \brief Find first XML child element matching given criteria * * \param[in] parent XML element to search (can be \c NULL) * \param[in] node_name If not \c NULL, only match children of this type * \param[in] attr_n If not \c NULL, only match children with an attribute * of this name. * \param[in] attr_v If \p attr_n and this are not NULL, only match children * with an attribute named \p attr_n and this value * * \return Matching XML child element, or \c NULL if none found */ xmlNode * pcmk__xe_first_child(const xmlNode *parent, const char *node_name, const char *attr_n, const char *attr_v) { xmlNode *child = NULL; CRM_CHECK((attr_v == NULL) || (attr_n != NULL), return NULL); if (parent == NULL) { return NULL; } child = parent->children; while ((child != NULL) && (child->type != XML_ELEMENT_NODE)) { child = child->next; } for (; child != NULL; child = pcmk__xe_next(child, NULL)) { const char *value = NULL; if ((node_name != NULL) && !pcmk__xe_is(child, node_name)) { // Node name mismatch continue; } if (attr_n == NULL) { // No attribute match needed return child; } value = pcmk__xe_get(child, attr_n); if ((attr_v == NULL) && (value != NULL)) { // attr_v == NULL: Attribute attr_n must be set (to any value) return child; } if ((attr_v != NULL) && (pcmk__str_eq(value, attr_v, pcmk__str_none))) { // attr_v != NULL: Attribute attr_n must be set to value attr_v return child; } } if (attr_n == NULL) { crm_trace("%s XML has no child element of %s type", (const char *) parent->name, pcmk__s(node_name, "any")); } else { crm_trace("%s XML has no child element of %s type with %s='%s'", (const char *) parent->name, pcmk__s(node_name, "any"), attr_n, attr_v); } return NULL; } /*! * \internal * \brief Return next sibling element of an XML element * * \param[in] xml XML element to check * \param[in] element_name If not NULL, get next sibling with this element name * * \return Next desired sibling of \p xml (or NULL if none) */ xmlNode * pcmk__xe_next(const xmlNode *xml, const char *element_name) { for (xmlNode *next = (xml == NULL)? NULL : xml->next; next != NULL; next = next->next) { if ((next->type == XML_ELEMENT_NODE) && ((element_name == NULL) || pcmk__xe_is(next, element_name))) { return next; } } return NULL; } /*! * \internal * \brief Parse an integer score from an XML attribute * * \param[in] xml XML element with attribute to parse * \param[in] name Name of attribute to parse * \param[out] score Where to store parsed score (can be NULL to * just validate) * \param[in] default_score What to return if the attribute value is not * present or invalid * * \return Standard Pacemaker return code */ int pcmk__xe_get_score(const xmlNode *xml, const char *name, int *score, int default_score) { const char *value = NULL; CRM_CHECK((xml != NULL) && (name != NULL), return EINVAL); value = pcmk__xe_get(xml, name); return pcmk_parse_score(value, score, default_score); } /*! * \internal * \brief Set an XML attribute, expanding \c ++ and \c += where appropriate * * If \p target already has an attribute named \p name set to an integer value * and \p value is an addition assignment expression on \p name, then expand * \p value to an integer and set attribute \p name to the expanded value in * \p target. * * Otherwise, set attribute \p name on \p target using the literal \p value. * * The original attribute value in \p target and the number in an assignment * expression in \p value are parsed and added as scores (that is, their values * are capped at \c INFINITY and \c -INFINITY). For more details, refer to * \c pcmk_parse_score(). * * For example, suppose \p target has an attribute named \c "X" with value * \c "5", and that \p name is \c "X". * * If \p value is \c "X++", the new value of \c "X" in \p target is \c "6". * * If \p value is \c "X+=3", the new value of \c "X" in \p target is \c "8". * * If \p value is \c "val", the new value of \c "X" in \p target is \c "val". * * If \p value is \c "Y++", the new value of \c "X" in \p target is \c "Y++". * * \param[in,out] target XML node whose attribute to set * \param[in] name Name of the attribute to set * \param[in] value New value of attribute to set (if NULL, initial value * will be left unchanged) * * \return Standard Pacemaker return code (specifically, \c EINVAL on invalid * argument, or \c pcmk_rc_ok otherwise) */ int pcmk__xe_set_score(xmlNode *target, const char *name, const char *value) { const char *old_value = NULL; CRM_CHECK((target != NULL) && (name != NULL), return EINVAL); if (value == NULL) { // @TODO Maybe instead delete the attribute or set it to 0 return pcmk_rc_ok; } old_value = pcmk__xe_get(target, name); // If no previous value, skip to default case and set the value unexpanded. if (old_value != NULL) { const char *n = name; const char *v = value; // Stop at first character that differs between name and value for (; (*n == *v) && (*n != '\0'); n++, v++); // If value begins with name followed by a "++" or "+=" if ((*n == '\0') && (*v++ == '+') && ((*v == '+') || (*v == '='))) { int add = 1; int old_value_i = 0; int rc = pcmk_rc_ok; // If we're expanding ourselves, no previous value was set; use 0 if (old_value != value) { rc = pcmk_parse_score(old_value, &old_value_i, 0); if (rc != pcmk_rc_ok) { // @TODO This is inconsistent with old_value==NULL crm_trace("Using 0 before incrementing %s because '%s' " "is not a score", name, old_value); } } /* value="X++": new value of X is old_value + 1 * value="X+=Y": new value of X is old_value + Y (for some number Y) */ if (*v != '+') { rc = pcmk_parse_score(++v, &add, 0); if (rc != pcmk_rc_ok) { // @TODO We should probably skip expansion instead crm_trace("Not incrementing %s because '%s' does not have " "a valid increment", name, value); } } crm_xml_add_int(target, name, pcmk__add_scores(old_value_i, add)); return pcmk_rc_ok; } } // Default case: set the attribute unexpanded (with value treated literally) if (old_value != value) { crm_xml_add(target, name, value); } return pcmk_rc_ok; } /*! * \internal * \brief Copy XML attributes from a source element to a target element * * This is similar to \c xmlCopyPropList() except that attributes are marked * as dirty for change tracking purposes. * * \param[in,out] target XML element to receive copied attributes from \p src * \param[in] src XML element whose attributes to copy to \p target * \param[in] flags Group of enum pcmk__xa_flags * * \return Standard Pacemaker return code */ int pcmk__xe_copy_attrs(xmlNode *target, const xmlNode *src, uint32_t flags) { CRM_CHECK((src != NULL) && (target != NULL), return EINVAL); for (xmlAttr *attr = pcmk__xe_first_attr(src); attr != NULL; attr = attr->next) { const char *name = (const char *) attr->name; const char *value = pcmk__xml_attr_value(attr); if (pcmk_is_set(flags, pcmk__xaf_no_overwrite) && (pcmk__xe_get(target, name) != NULL)) { continue; } if (pcmk_is_set(flags, pcmk__xaf_score_update)) { pcmk__xe_set_score(target, name, value); } else { crm_xml_add(target, name, value); } } return pcmk_rc_ok; } /*! * \internal * \brief Compare two XML attributes by name * * \param[in] a First XML attribute to compare * \param[in] b Second XML attribute to compare * * \retval negative \c a->name is \c NULL or comes before \c b->name * lexicographically * \retval 0 \c a->name and \c b->name are equal * \retval positive \c b->name is \c NULL or comes before \c a->name * lexicographically */ static gint compare_xml_attr(gconstpointer a, gconstpointer b) { const xmlAttr *attr_a = a; const xmlAttr *attr_b = b; return pcmk__strcmp((const char *) attr_a->name, (const char *) attr_b->name, pcmk__str_none); } /*! * \internal * \brief Sort an XML element's attributes by name * * This does not consider ACLs and does not mark the attributes as deleted or * dirty. Upon return, all attributes still exist and are set to the same values * as before the call. The only thing that may change is the order of the * attribute list. * * \param[in,out] xml XML element whose attributes to sort */ void pcmk__xe_sort_attrs(xmlNode *xml) { GSList *attr_list = NULL; for (xmlAttr *iter = pcmk__xe_first_attr(xml); iter != NULL; iter = iter->next) { attr_list = g_slist_prepend(attr_list, iter); } attr_list = g_slist_sort(attr_list, compare_xml_attr); for (GSList *iter = attr_list; iter != NULL; iter = iter->next) { xmlNode *attr = iter->data; xmlUnlinkNode(attr); xmlAddChild(xml, attr); } g_slist_free(attr_list); } /*! * \internal * \brief Remove a named attribute from an XML element * * \param[in,out] element XML element to remove an attribute from * \param[in] name Name of attribute to remove */ void pcmk__xe_remove_attr(xmlNode *element, const char *name) { if (name != NULL) { pcmk__xa_remove(xmlHasProp(element, (const xmlChar *) name), false); } } /*! * \internal * \brief Remove a named attribute from an XML element * * This is a wrapper for \c pcmk__xe_remove_attr() for use with * \c pcmk__xml_tree_foreach(). * * \param[in,out] xml XML element to remove an attribute from * \param[in] user_data Name of attribute to remove * * \return \c true (to continue traversing the tree) * * \note This is compatible with \c pcmk__xml_tree_foreach(). */ bool pcmk__xe_remove_attr_cb(xmlNode *xml, void *user_data) { const char *name = user_data; pcmk__xe_remove_attr(xml, name); return true; } /*! * \internal * \brief Remove an XML element's attributes that match some criteria * * \param[in,out] element XML element to modify * \param[in] force If \c true, remove matching attributes immediately, * ignoring ACLs and change tracking * \param[in] match If not NULL, only remove attributes for which * this function returns true * \param[in,out] user_data Data to pass to \p match */ void pcmk__xe_remove_matching_attrs(xmlNode *element, bool force, bool (*match)(xmlAttrPtr, void *), void *user_data) { xmlAttrPtr next = NULL; for (xmlAttrPtr a = pcmk__xe_first_attr(element); a != NULL; a = next) { next = a->next; // Grab now because attribute might get removed if ((match == NULL) || match(a, user_data)) { if (pcmk__xa_remove(a, force) != pcmk_rc_ok) { return; } } } } /*! * \internal * \brief Create a new XML element under a given parent * * \param[in,out] parent XML element that will be the new element's parent * (\c NULL to create a new XML document with the new * node as root) * \param[in] name Name of new element * * \return Newly created XML element (guaranteed not to be \c NULL) */ xmlNode * pcmk__xe_create(xmlNode *parent, const char *name) { xmlNode *node = NULL; pcmk__assert(!pcmk__str_empty(name)); if (parent == NULL) { xmlDoc *doc = pcmk__xml_new_doc(); node = xmlNewDocRawNode(doc, NULL, (const xmlChar *) name, NULL); pcmk__mem_assert(node); xmlDocSetRootElement(doc, node); } else { node = xmlNewChild(parent, NULL, (const xmlChar *) name, NULL); pcmk__mem_assert(node); } pcmk__xml_new_private_data(node); return node; } /*! * \internal * \brief Set a formatted string as an XML node's content * * \param[in,out] node Node whose content to set * \param[in] format printf(3)-style format string * \param[in] ... Arguments for \p format * * \note This function escapes special characters. \c xmlNodeSetContent() does * not. */ G_GNUC_PRINTF(2, 3) void pcmk__xe_set_content(xmlNode *node, const char *format, ...) { if (node != NULL) { const char *content = NULL; char *buf = NULL; /* xmlNodeSetContent() frees node->children and replaces it with new * text. If this function is called for a node that already has a non- * text child, it's a bug. */ CRM_CHECK((node->children == NULL) || (node->children->type == XML_TEXT_NODE), return); if (strchr(format, '%') == NULL) { // Nothing to format content = format; } else { va_list ap; va_start(ap, format); if (pcmk__str_eq(format, "%s", pcmk__str_none)) { // No need to make a copy content = va_arg(ap, const char *); } else { pcmk__assert(vasprintf(&buf, format, ap) >= 0); content = buf; } va_end(ap); } xmlNodeSetContent(node, (const xmlChar *) content); free(buf); } } /*! * \internal * \brief Set a formatted string as an XML element's ID * * If the formatted string would not be a valid ID, it's first sanitized by * \c pcmk__xml_sanitize_id(). * * \param[in,out] node Node whose ID to set * \param[in] format printf(3)-style format string * \param[in] ... Arguments for \p format */ G_GNUC_PRINTF(2, 3) void pcmk__xe_set_id(xmlNode *node, const char *format, ...) { char *id = NULL; va_list ap; pcmk__assert(!pcmk__str_empty(format)); if (node == NULL) { return; } va_start(ap, format); pcmk__assert(vasprintf(&id, format, ap) >= 0); va_end(ap); if (!xmlValidateNameValue((const xmlChar *) id)) { pcmk__xml_sanitize_id(id); } crm_xml_add(node, PCMK_XA_ID, id); free(id); } /*! * \internal * \brief Add a "last written" attribute to an XML element, set to current time * * \param[in,out] xe XML element to add attribute to * * \return Value that was set, or NULL on error */ const char * pcmk__xe_add_last_written(xmlNode *xe) { char *now_s = pcmk__epoch2str(NULL, 0); const char *result = NULL; result = crm_xml_add(xe, PCMK_XA_CIB_LAST_WRITTEN, pcmk__s(now_s, "Could not determine current time")); free(now_s); return result; } /*! * \internal * \brief Merge one XML tree into another * * Here, "merge" means: * 1. Copy attribute values from \p update to the target, overwriting in case of * conflict. * 2. Descend through \p update and the target in parallel. At each level, for * each child of \p update, look for a matching child of the target. * a. For each child, if a match is found, go to step 1, recursively merging * the child of \p update into the child of the target. * b. Otherwise, copy the child of \p update as a child of the target. * * A match is defined as the first child of the same type within the target, * with: * * the \c PCMK_XA_ID attribute matching, if set in \p update; otherwise, * * the \c PCMK_XA_ID_REF attribute matching, if set in \p update * * This function does not delete any elements or attributes from the target. It * may add elements or overwrite attributes, as described above. * * \param[in,out] parent If \p target is NULL and this is not, add or update * child of this XML node that matches \p update * \param[in,out] target If not NULL, update this XML * \param[in] update Make the desired XML match this (must not be \c NULL) * \param[in] flags Group of enum pcmk__xa_flags * * \note At least one of \p parent and \p target must be non-NULL. * \note This function is recursive. For the top-level call, \p parent is * \c NULL and \p target is not \c NULL. For recursive calls, \p target is * \c NULL and \p parent is not \c NULL. */ static void update_xe(xmlNode *parent, xmlNode *target, xmlNode *update, uint32_t flags) { // @TODO Try to refactor further, possibly using pcmk__xml_tree_foreach() const char *update_name = NULL; const char *update_id_attr = NULL; const char *update_id_val = NULL; char *trace_s = NULL; crm_log_xml_trace(update, "update"); crm_log_xml_trace(target, "target"); CRM_CHECK(update != NULL, goto done); if (update->type == XML_COMMENT_NODE) { pcmk__xc_update(parent, target, update); goto done; } update_name = (const char *) update->name; CRM_CHECK(update_name != NULL, goto done); CRM_CHECK((target != NULL) || (parent != NULL), goto done); update_id_val = pcmk__xe_id(update); if (update_id_val != NULL) { update_id_attr = PCMK_XA_ID; } else { update_id_val = pcmk__xe_get(update, PCMK_XA_ID_REF); if (update_id_val != NULL) { update_id_attr = PCMK_XA_ID_REF; } } pcmk__if_tracing( { if (update_id_attr != NULL) { trace_s = crm_strdup_printf("<%s %s=%s/>", update_name, update_id_attr, update_id_val); } else { trace_s = crm_strdup_printf("<%s/>", update_name); } }, {} ); if (target == NULL) { // Recursive call target = pcmk__xe_first_child(parent, update_name, update_id_attr, update_id_val); } if (target == NULL) { // Recursive call with no existing matching child target = pcmk__xe_create(parent, update_name); crm_trace("Added %s", pcmk__s(trace_s, update_name)); } else { // Either recursive call with match, or top-level call crm_trace("Found node %s to update", pcmk__s(trace_s, update_name)); } CRM_CHECK(pcmk__xe_is(target, (const char *) update->name), return); pcmk__xe_copy_attrs(target, update, flags); for (xmlNode *child = pcmk__xml_first_child(update); child != NULL; child = pcmk__xml_next(child)) { crm_trace("Updating child of %s", pcmk__s(trace_s, update_name)); update_xe(target, NULL, child, flags); } crm_trace("Finished with %s", pcmk__s(trace_s, update_name)); done: free(trace_s); } /*! * \internal * \brief Delete an XML subtree if it matches a search element * * A match is defined as follows: * * \p xml and \p user_data are both element nodes of the same type. * * If \p user_data has attributes set, \p xml has those attributes set to the * same values. (\p xml may have additional attributes set to arbitrary * values.) * * \param[in,out] xml XML subtree to delete upon match * \param[in] user_data Search element * * \return \c true to continue traversing the tree, or \c false to stop (because * \p xml was deleted) * * \note This is compatible with \c pcmk__xml_tree_foreach(). */ static bool delete_xe_if_matching(xmlNode *xml, void *user_data) { xmlNode *search = user_data; if (!pcmk__xe_is(search, (const char *) xml->name)) { // No match: either not both elements, or different element types return true; } for (const xmlAttr *attr = pcmk__xe_first_attr(search); attr != NULL; attr = attr->next) { const char *search_val = pcmk__xml_attr_value(attr); const char *xml_val = pcmk__xe_get(xml, (const char *) attr->name); if (!pcmk__str_eq(search_val, xml_val, pcmk__str_casei)) { // No match: an attr in xml doesn't match the attr in search return true; } } crm_log_xml_trace(xml, "delete-match"); crm_log_xml_trace(search, "delete-search"); pcmk__xml_free(xml); // Found a match and deleted it; stop traversing tree return false; } /*! * \internal * \brief Search an XML tree depth-first and delete the first matching element * * This function does not attempt to match the tree root (\p xml). * * A match with a node \c node is defined as follows: * * \c node and \p search are both element nodes of the same type. * * If \p search has attributes set, \c node has those attributes set to the * same values. (\c node may have additional attributes set to arbitrary * values.) * * \param[in,out] xml XML subtree to search * \param[in] search Element to match against * * \return Standard Pacemaker return code (specifically, \c pcmk_rc_ok on * successful deletion and an error code otherwise) */ int pcmk__xe_delete_match(xmlNode *xml, xmlNode *search) { // See @COMPAT comment in pcmk__xe_replace_match() CRM_CHECK((xml != NULL) && (search != NULL), return EINVAL); for (xml = pcmk__xe_first_child(xml, NULL, NULL, NULL); xml != NULL; xml = pcmk__xe_next(xml, NULL)) { if (!pcmk__xml_tree_foreach(xml, delete_xe_if_matching, search)) { // Found and deleted an element return pcmk_rc_ok; } } // No match found in this subtree return ENXIO; } /*! * \internal * \brief Replace one XML node with a copy of another XML node * * This function handles change tracking and applies ACLs. * * \param[in,out] old XML node to replace * \param[in] new XML node to copy as replacement for \p old * * \note This frees \p old. */ static void replace_node(xmlNode *old, xmlNode *new) { // Pass old for its doc; it won't remain the parent of new new = pcmk__xml_copy(old, new); old = xmlReplaceNode(old, new); // old == NULL means memory allocation error pcmk__assert(old != NULL); // May be unnecessary but avoids slight changes to some test outputs pcmk__xml_tree_foreach(new, pcmk__xml_reset_node_flags, NULL); if (pcmk__xml_doc_all_flags_set(new->doc, pcmk__xf_tracking)) { // Replaced sections may have included relevant ACLs pcmk__apply_acl(new); } pcmk__xml_mark_changes(old, new); pcmk__xml_free_node(old); } /*! * \internal * \brief Replace one XML subtree with a copy of another if the two match * * A match is defined as follows: * * \p xml and \p user_data are both element nodes of the same type. * * If \p user_data has the \c PCMK_XA_ID attribute set, then \p xml has * \c PCMK_XA_ID set to the same value. * * \param[in,out] xml XML subtree to replace with \p user_data upon match * \param[in] user_data XML to replace \p xml with a copy of upon match * * \return \c true to continue traversing the tree, or \c false to stop (because * \p xml was replaced by \p user_data) * * \note This is compatible with \c pcmk__xml_tree_foreach(). */ static bool replace_xe_if_matching(xmlNode *xml, void *user_data) { xmlNode *replace = user_data; const char *xml_id = NULL; const char *replace_id = NULL; xml_id = pcmk__xe_id(xml); replace_id = pcmk__xe_id(replace); if (!pcmk__xe_is(replace, (const char *) xml->name)) { // No match: either not both elements, or different element types return true; } if ((replace_id != NULL) && !pcmk__str_eq(replace_id, xml_id, pcmk__str_none)) { // No match: ID was provided in replace and doesn't match xml's ID return true; } crm_log_xml_trace(xml, "replace-match"); crm_log_xml_trace(replace, "replace-with"); replace_node(xml, replace); // Found a match and replaced it; stop traversing tree return false; } /*! * \internal * \brief Search an XML tree depth-first and replace the first matching element * * This function does not attempt to match the tree root (\p xml). * * A match with a node \c node is defined as follows: * * \c node and \p replace are both element nodes of the same type. * * If \p replace has the \c PCMK_XA_ID attribute set, then \c node has * \c PCMK_XA_ID set to the same value. * * \param[in,out] xml XML tree to search * \param[in] replace XML to replace a matching element with a copy of * * \return Standard Pacemaker return code (specifically, \c pcmk_rc_ok on * successful replacement and an error code otherwise) */ int pcmk__xe_replace_match(xmlNode *xml, xmlNode *replace) { /* @COMPAT Some of this behavior (like not matching the tree root, which is * allowed by pcmk__xe_update_match()) is questionable for general use but * required for backward compatibility by cib_process_replace() and * cib_process_delete(). Behavior can change at a major version release if * desired. */ CRM_CHECK((xml != NULL) && (replace != NULL), return EINVAL); for (xml = pcmk__xe_first_child(xml, NULL, NULL, NULL); xml != NULL; xml = pcmk__xe_next(xml, NULL)) { if (!pcmk__xml_tree_foreach(xml, replace_xe_if_matching, replace)) { // Found and replaced an element return pcmk_rc_ok; } } // No match found in this subtree return ENXIO; } //! User data for \c update_xe_if_matching() struct update_data { xmlNode *update; //!< Update source uint32_t flags; //!< Group of enum pcmk__xa_flags }; /*! * \internal * \brief Update one XML subtree with another if the two match * * "Update" means to merge a source subtree into a target subtree (see * \c update_xe()). * * A match is defined as follows: * * \p xml and \p user_data->update are both element nodes of the same type. * * \p xml and \p user_data->update have the same \c PCMK_XA_ID attribute * value, or \c PCMK_XA_ID is unset in both * * \param[in,out] xml XML subtree to update with \p user_data->update * upon match * \param[in] user_data struct update_data object * * \return \c true to continue traversing the tree, or \c false to stop (because * \p xml was updated by \p user_data->update) * * \note This is compatible with \c pcmk__xml_tree_foreach(). */ static bool update_xe_if_matching(xmlNode *xml, void *user_data) { struct update_data *data = user_data; xmlNode *update = data->update; if (!pcmk__xe_is(update, (const char *) xml->name)) { // No match: either not both elements, or different element types return true; } if (!pcmk__str_eq(pcmk__xe_id(xml), pcmk__xe_id(update), pcmk__str_none)) { // No match: ID mismatch return true; } crm_log_xml_trace(xml, "update-match"); crm_log_xml_trace(update, "update-with"); update_xe(NULL, xml, update, data->flags); // Found a match and replaced it; stop traversing tree return false; } /*! * \internal * \brief Search an XML tree depth-first and update the first matching element * * "Update" means to merge a source subtree into a target subtree (see * \c update_xe()). * * A match with a node \c node is defined as follows: * * \c node and \p update are both element nodes of the same type. * * \c node and \p update have the same \c PCMK_XA_ID attribute value, or * \c PCMK_XA_ID is unset in both * * \param[in,out] xml XML tree to search * \param[in] update XML to update a matching element with * \param[in] flags Group of enum pcmk__xa_flags * * \return Standard Pacemaker return code (specifically, \c pcmk_rc_ok on * successful update and an error code otherwise) */ int pcmk__xe_update_match(xmlNode *xml, xmlNode *update, uint32_t flags) { /* @COMPAT In pcmk__xe_delete_match() and pcmk__xe_replace_match(), we * compare IDs only if the equivalent of the update argument has an ID. * Here, we're stricter: we consider it a mismatch if only one element has * an ID attribute, or if both elements have IDs but they don't match. * * Perhaps we should align the behavior at a major version release. */ struct update_data data = { .update = update, .flags = flags, }; CRM_CHECK((xml != NULL) && (update != NULL), return EINVAL); if (!pcmk__xml_tree_foreach(xml, update_xe_if_matching, &data)) { // Found and updated an element return pcmk_rc_ok; } // No match found in this subtree return ENXIO; } void pcmk__xe_set_propv(xmlNodePtr node, va_list pairs) { while (true) { const char *name, *value; name = va_arg(pairs, const char *); if (name == NULL) { return; } value = va_arg(pairs, const char *); if (value != NULL) { crm_xml_add(node, name, value); } } } void pcmk__xe_set_props(xmlNodePtr node, ...) { va_list pairs; va_start(pairs, node); pcmk__xe_set_propv(node, pairs); va_end(pairs); } int pcmk__xe_foreach_child(xmlNode *xml, const char *child_element_name, int (*handler)(xmlNode *xml, void *userdata), void *userdata) { xmlNode *children = (xml? xml->children : NULL); pcmk__assert(handler != NULL); for (xmlNode *node = children; node != NULL; node = node->next) { if ((node->type == XML_ELEMENT_NODE) && ((child_element_name == NULL) || pcmk__xe_is(node, child_element_name))) { int rc = handler(node, userdata); if (rc != pcmk_rc_ok) { return rc; } } } return pcmk_rc_ok; } // XML attribute handling /*! * \brief Create an XML attribute with specified name and value * * \param[in,out] node XML node to modify * \param[in] name Attribute name to set * \param[in] value Attribute value to set * * \return New value on success, \c NULL otherwise * \note This does nothing if node, name, or value are \c NULL or empty. */ const char * crm_xml_add(xmlNode *node, const char *name, const char *value) { // @TODO Replace with internal function that returns the new attribute bool dirty = FALSE; xmlAttr *attr = NULL; CRM_CHECK(node != NULL, return NULL); CRM_CHECK(name != NULL, return NULL); if (value == NULL) { return NULL; } if (pcmk__xml_doc_all_flags_set(node->doc, pcmk__xf_tracking)) { const char *old = pcmk__xe_get(node, name); if (old == NULL || value == NULL || strcmp(old, value) != 0) { dirty = TRUE; } } if (dirty && (pcmk__check_acl(node, name, pcmk__xf_acl_create) == FALSE)) { crm_trace("Cannot add %s=%s to %s", name, value, node->name); return NULL; } attr = xmlSetProp(node, (const xmlChar *) name, (const xmlChar *) value); /* If the attribute already exists, this does nothing. Attribute values * don't get private data. */ pcmk__xml_new_private_data((xmlNode *) attr); if (dirty) { pcmk__mark_xml_attr_dirty(attr); } CRM_CHECK(attr && attr->children && attr->children->content, return NULL); return (char *)attr->children->content; } /*! * \brief Create an XML attribute with specified name and integer value * * This is like \c crm_xml_add() but taking an integer value. * * \param[in,out] node XML node to modify * \param[in] name Attribute name to set * \param[in] value Attribute value to set * * \return New value as string on success, \c NULL otherwise * \note This does nothing if node or name are \c NULL or empty. */ const char * crm_xml_add_int(xmlNode *node, const char *name, int value) { char *number = pcmk__itoa(value); const char *added = crm_xml_add(node, name, number); free(number); return added; } // Maximum size of null-terminated string representation of 64-bit integer // -9223372036854775808 #define LLSTRSIZE 21 /*! * \brief Create an XML attribute with specified name and long long int value * * This is like \c crm_xml_add() but taking a long long int value. It is a * useful equivalent for defined types like time_t, etc. * * \param[in,out] xml XML node to modify * \param[in] name Attribute name to set * \param[in] value Attribute value to set * * \return New value as string on success, \c NULL otherwise * \note This does nothing if xml or name are \c NULL or empty. * This does not support greater than 64-bit values. */ const char * crm_xml_add_ll(xmlNode *xml, const char *name, long long value) { char s[LLSTRSIZE] = { '\0', }; if (snprintf(s, LLSTRSIZE, "%lld", (long long) value) == LLSTRSIZE) { return NULL; } return crm_xml_add(xml, name, s); } /*! * \internal * \brief Retrieve the value of an XML attribute * * \param[in] xml XML element whose attribute to get * \param[in] attr_name Attribute name * * \return Value of specified attribute (may be \c NULL) */ const char * pcmk__xe_get(const xmlNode *xml, const char *attr_name) { xmlAttr *attr = NULL; CRM_CHECK((xml != NULL) && (attr_name != NULL), return NULL); attr = xmlHasProp(xml, (const xmlChar *) attr_name); if ((attr == NULL) || (attr->children == NULL)) { return NULL; } return (const char *) attr->children->content; } /*! * \internal * \brief Retrieve a flag group from an XML attribute value * * This is like \c pcmk__xe_get() but returns the value as a \c uint32_t. * * \param[in] xml XML node to check * \param[in] name Attribute name to check (must not be NULL) * \param[out] dest Where to store flags (may be NULL to just * validate type) * \param[in] default_value What to use for missing or invalid value * * \return Standard Pacemaker return code */ int pcmk__xe_get_flags(const xmlNode *xml, const char *name, uint32_t *dest, uint32_t default_value) { const char *value = NULL; long long value_ll = 0LL; int rc = pcmk_rc_ok; if (dest != NULL) { *dest = default_value; } if (name == NULL) { return EINVAL; } if (xml == NULL) { return pcmk_rc_ok; } value = pcmk__xe_get(xml, name); if (value == NULL) { return pcmk_rc_ok; } rc = pcmk__scan_ll(value, &value_ll, default_value); if ((value_ll < 0) || (value_ll > UINT32_MAX)) { value_ll = default_value; if (rc == pcmk_rc_ok) { rc = pcmk_rc_bad_input; } } if (dest != NULL) { *dest = (uint32_t) value_ll; } return rc; } /*! * \internal * \brief Retrieve a \c guint value from an XML attribute * * This is like \c pcmk__xe_get() but returns the value as a \c guint. * * \param[in] xml XML element whose attribute to get * \param[in] attr Attribute name * \param[out] dest Where to store attribute value (unchanged on error) * * \return Standard Pacemaker return code */ int pcmk__xe_get_guint(const xmlNode *xml, const char *attr, guint *dest) { long long value_ll = 0; int rc = pcmk_rc_ok; CRM_CHECK((xml != NULL) && (attr != NULL) && (dest != NULL), return EINVAL); rc = pcmk__xe_get_ll(xml, attr, &value_ll); if (rc != pcmk_rc_ok) { return rc; } if ((value_ll < 0) || (value_ll > G_MAXUINT)) { return ERANGE; } *dest = (guint) value_ll; return pcmk_rc_ok; } /*! * \internal * \brief Set an XML attribute using a \c guint value * * This is like \c crm_xml_add() but takes a \c guint. * * \param[in,out] xml XML node to modify * \param[in] attr Attribute name * \param[in] value Attribute value to set */ void pcmk__xe_set_guint(xmlNode *xml, const char *attr, guint value) { char *value_s = NULL; CRM_CHECK((xml != NULL) && (attr != NULL), return); value_s = crm_strdup_printf("%u", value); crm_xml_add(xml, attr, value_s); free(value_s); } /*! * \internal * \brief Retrieve an \c int value from an XML attribute * * This is like \c pcmk__xe_get() but returns the value as an \c int. * * \param[in] xml XML element whose attribute to get * \param[in] attr Attribute name * \param[out] dest Where to store element value (unchanged on error) * * \return Standard Pacemaker return code */ int pcmk__xe_get_int(const xmlNode *xml, const char *attr, int *dest) { long long value_ll = 0; int rc = pcmk_rc_ok; CRM_CHECK((xml != NULL) && (attr != NULL) && (dest != NULL), return EINVAL); rc = pcmk__xe_get_ll(xml, attr, &value_ll); if (rc != pcmk_rc_ok) { return rc; } if ((value_ll < INT_MIN) || (value_ll > INT_MAX)) { return ERANGE; } *dest = (int) value_ll; return pcmk_rc_ok; } /*! * \internal * \brief Retrieve a long long value from an XML attribute * * This is like \c pcmk__xe_get() but returns the value as a long long. * * \param[in] xml XML element whose attribute to get * \param[in] attr Attribute name * \param[out] dest Where to store element value (unchanged on error) * * \return Standard Pacemaker return code */ int pcmk__xe_get_ll(const xmlNode *xml, const char *attr, long long *dest) { const char *value = NULL; long long value_ll = 0; int rc = pcmk_rc_ok; CRM_CHECK((xml != NULL) && (attr != NULL) && (dest != NULL), return EINVAL); value = pcmk__xe_get(xml, attr); if (value == NULL) { return ENXIO; } rc = pcmk__scan_ll(value, &value_ll, PCMK__PARSE_INT_DEFAULT); if (rc != pcmk_rc_ok) { return rc; } *dest = value_ll; return pcmk_rc_ok; } +/*! + * \internal + * \brief Set an XML attribute using a long long value + * + * This is like \c crm_xml_add() but takes a long long. + * + * \param[in,out] xml XML node to modify + * \param[in] attr Attribute name + * \param[in] value Attribute value to set + * + * \return Standard Pacemaker return code + */ +int +pcmk__xe_set_ll(xmlNode *xml, const char *attr, long long value) +{ + char *value_s = NULL; + int rc = pcmk_rc_ok; + + CRM_CHECK((xml != NULL) && (attr != NULL), return EINVAL); + + value_s = crm_strdup_printf("%lld", value); + + errno = 0; + if (crm_xml_add(xml, attr, value_s) == NULL) { + // ACL denial + rc = (errno != 0)? errno : EACCES; + } + + free(value_s); + return rc; +} + /*! * \internal * \brief Retrieve a \c time_t value from an XML attribute * * This is like \c pcmk__xe_get() but returns the value as a \c time_t. * * \param[in] xml XML element whose attribute to get * \param[in] attr Attribute name * \param[out] dest Where to store attribute value (unchanged on error) * * \return Standard Pacemaker return code */ int pcmk__xe_get_time(const xmlNode *xml, const char *attr, time_t *dest) { long long value_ll = 0; int rc = pcmk_rc_ok; CRM_CHECK((xml != NULL) && (attr != NULL) && (dest != NULL), return EINVAL); rc = pcmk__xe_get_ll(xml, attr, &value_ll); if (rc != pcmk_rc_ok) { return rc; } /* We don't do any bounds checking, since there are no constants provided * for the bounds of time_t, and calculating them isn't worth the effort. If * there are XML values beyond the native sizes, there will likely be worse * problems anyway. */ *dest = (time_t) value_ll; return pcmk_rc_ok; } /*! * \internal * \brief Set an XML attribute using a \c time_t value * * This is like \c crm_xml_add() but takes a \c time_t. * * \param[in,out] xml XML element whose attribute to set * \param[in] attr Attribute name * \param[in] value Attribute value to set (in seconds) */ void pcmk__xe_set_time(xmlNode *xml, const char *attr, time_t value) { // Could be inline, but keep it underneath pcmk__xe_get_time() CRM_CHECK((xml != NULL) && (attr != NULL), return); - crm_xml_add_ll(xml, attr, (long long) value); + pcmk__xe_set_ll(xml, attr, (long long) value); } /*! * \internal * \brief Retrieve the values of XML second/microsecond attributes as time * * This is like \c pcmk__xe_get() but returns the value as a * struct timeval. * * \param[in] xml XML element whose attributes to get * \param[in] sec_attr Name of XML attribute for seconds * \param[in] usec_attr Name of XML attribute for microseconds * \param[out] dest Where to store result (unchanged on error) * * \return Standard Pacemaker return code */ int pcmk__xe_get_timeval(const xmlNode *xml, const char *sec_attr, const char *usec_attr, struct timeval *dest) { long long value_ll = 0; struct timeval result = { 0, 0 }; int rc = pcmk_rc_ok; // Could allow one of sec_attr and usec_attr to be NULL in the future CRM_CHECK((xml != NULL) && (sec_attr != NULL) && (usec_attr != NULL) && (dest != NULL), return EINVAL); // No bounds checking; see comment in pcmk__xe_get_time() // Parse seconds rc = pcmk__xe_get_time(xml, sec_attr, &(result.tv_sec)); if (rc != pcmk_rc_ok) { return rc; } // Parse microseconds rc = pcmk__xe_get_ll(xml, usec_attr, &value_ll); if (rc != pcmk_rc_ok) { return rc; } result.tv_usec = (suseconds_t) value_ll; *dest = result; return pcmk_rc_ok; } /*! * \internal * \brief Set XML attribute values for seconds and microseconds * * This is like \c crm_xml_add() but takes a struct timeval *. * * \param[in,out] xml XML element whose attributes to set * \param[in] sec_attr Name of XML attribute for seconds * \param[in] usec_attr Name of XML attribute for microseconds * \param[in] value Attribute values to set * * \note This does nothing if \p value is \c NULL. */ void pcmk__xe_set_timeval(xmlNode *xml, const char *sec_attr, const char *usec_attr, const struct timeval *value) { CRM_CHECK((xml != NULL) && (sec_attr != NULL) && (usec_attr != NULL), return); if (value == NULL) { return; } - if (crm_xml_add_ll(xml, sec_attr, (long long) value->tv_sec) == NULL) { + if (pcmk__xe_set_ll(xml, sec_attr, + (long long) value->tv_sec) != pcmk_rc_ok) { return; } /* Seconds were added successfully. Ignore any errors adding microseconds. * * It would be nice to make this atomic: revert the seconds attribute if * adding the microseconds attribute fails. That's somewhat complicated due * to change tracking: the chain of parents is already marked dirty, etc. In * practice, microseconds should succeed if seconds succeeded, unless it's * due to memory allocation failure. Nothing checks the return values of * these setter functions at time of writing, anyway. */ - crm_xml_add_ll(xml, usec_attr, (long long) value->tv_usec); + pcmk__xe_set_ll(xml, usec_attr, (long long) value->tv_usec); } /*! * \internal * \brief Get a date/time object from an XML attribute value * * \param[in] xml XML with attribute to parse (from CIB) * \param[in] attr Name of attribute to parse * \param[out] t Where to create date/time object * (\p *t must be NULL initially) * * \return Standard Pacemaker return code * \note The caller is responsible for freeing \p *t using crm_time_free(). */ int pcmk__xe_get_datetime(const xmlNode *xml, const char *attr, crm_time_t **t) { const char *value = NULL; if ((t == NULL) || (*t != NULL) || (xml == NULL) || (attr == NULL)) { return EINVAL; } value = pcmk__xe_get(xml, attr); if (value != NULL) { *t = crm_time_new(value); if (*t == NULL) { return pcmk_rc_unpack_error; } } return pcmk_rc_ok; } /*! * \internal * \brief Add a boolean attribute to an XML node. * * \param[in,out] node XML node to add attributes to * \param[in] name XML attribute to create * \param[in] value Value to give to the attribute */ void pcmk__xe_set_bool_attr(xmlNodePtr node, const char *name, bool value) { crm_xml_add(node, name, pcmk__btoa(value)); } /*! * \internal * \brief Extract a boolean attribute's value from an XML element, with * error checking * * \param[in] node XML node to get attribute from * \param[in] name XML attribute to get * \param[out] value Destination for the value of the attribute * * \return EINVAL if \p name or \p value are NULL, ENODATA if \p node is * NULL or the attribute does not exist, pcmk_rc_unknown_format * if the attribute is not a boolean, and pcmk_rc_ok otherwise. * * \note \p value only has any meaning if the return value is pcmk_rc_ok. */ int pcmk__xe_get_bool_attr(const xmlNode *node, const char *name, bool *value) { const char *xml_value = NULL; int ret, rc; if (node == NULL) { return ENODATA; } else if (name == NULL || value == NULL) { return EINVAL; } xml_value = pcmk__xe_get(node, name); if (xml_value == NULL) { return ENODATA; } rc = crm_str_to_boolean(xml_value, &ret); if (rc == 1) { *value = ret; return pcmk_rc_ok; } else { return pcmk_rc_bad_input; } } /*! * \internal * \brief Extract a boolean attribute's value from an XML element * * \param[in] node XML node to get attribute from * \param[in] name XML attribute to get * * \return True if the given \p name is an attribute on \p node and has * the value \c PCMK_VALUE_TRUE, False in all other cases */ bool pcmk__xe_attr_is_true(const xmlNode *node, const char *name) { bool value = false; int rc; rc = pcmk__xe_get_bool_attr(node, name, &value); return rc == pcmk_rc_ok && value == true; } // Deprecated functions kept only for backward API compatibility // LCOV_EXCL_START #include // gboolean, GSList #include // pcmk_xml_attrs2nvpairs(), etc. #include // crm_xml_sanitize_id() #include xmlNode * expand_idref(xmlNode *input, xmlNode *top) { return pcmk__xe_resolve_idref(input, top); } void crm_xml_set_id(xmlNode *xml, const char *format, ...) { va_list ap; int len = 0; char *id = NULL; /* equivalent to crm_strdup_printf() */ va_start(ap, format); len = vasprintf(&id, format, ap); va_end(ap); pcmk__assert(len > 0); crm_xml_sanitize_id(id); crm_xml_add(xml, PCMK_XA_ID, id); free(id); } xmlNode * sorted_xml(xmlNode *input, xmlNode *parent, gboolean recursive) { xmlNode *child = NULL; GSList *nvpairs = NULL; xmlNode *result = NULL; CRM_CHECK(input != NULL, return NULL); result = pcmk__xe_create(parent, (const char *) input->name); nvpairs = pcmk_xml_attrs2nvpairs(input); nvpairs = pcmk_sort_nvpairs(nvpairs); pcmk_nvpairs2xml_attrs(nvpairs, result); pcmk_free_nvpairs(nvpairs); for (child = pcmk__xe_first_child(input, NULL, NULL, NULL); child != NULL; child = pcmk__xe_next(child, NULL)) { if (recursive) { sorted_xml(child, result, recursive); } else { pcmk__xml_copy(result, child); } } return result; } const char * crm_element_value(const xmlNode *data, const char *name) { xmlAttr *attr = NULL; if (data == NULL) { crm_err("Couldn't find %s in NULL", name ? name : ""); CRM_LOG_ASSERT(data != NULL); return NULL; } else if (name == NULL) { crm_err("Couldn't find NULL in %s", data->name); return NULL; } attr = xmlHasProp(data, (const xmlChar *) name); if (!attr || !attr->children) { return NULL; } return (const char *) attr->children->content; } const char * crm_copy_xml_element(const xmlNode *obj1, xmlNode *obj2, const char *element) { const char *value = crm_element_value(obj1, element); crm_xml_add(obj2, element, value); return value; } int crm_element_value_ll(const xmlNode *data, const char *name, long long *dest) { const char *value = NULL; CRM_CHECK(dest != NULL, return -1); value = crm_element_value(data, name); if (value != NULL) { int rc = pcmk__scan_ll(value, dest, PCMK__PARSE_INT_DEFAULT); if (rc == pcmk_rc_ok) { return 0; } crm_warn("Using default for %s " "because '%s' is not a valid integer: %s", name, value, pcmk_rc_str(rc)); } return -1; } int crm_element_value_timeval(const xmlNode *xml, const char *name_sec, const char *name_usec, struct timeval *dest) { long long value_i = 0; CRM_CHECK(dest != NULL, return -EINVAL); dest->tv_sec = 0; dest->tv_usec = 0; if (xml == NULL) { return pcmk_ok; } // No bounds checking; see comment in pcmk__xe_get_time() // Parse seconds errno = 0; if (crm_element_value_ll(xml, name_sec, &value_i) < 0) { return -errno; } dest->tv_sec = (time_t) value_i; // Parse microseconds if (crm_element_value_ll(xml, name_usec, &value_i) < 0) { return -errno; } dest->tv_usec = (suseconds_t) value_i; return pcmk_ok; } int crm_element_value_epoch(const xmlNode *xml, const char *name, time_t *dest) { long long value_ll = 0; if (crm_element_value_ll(xml, name, &value_ll) < 0) { return -1; } // No bounds checking; see comment in pcmk__xe_get_time() *dest = (time_t) value_ll; return pcmk_ok; } int crm_element_value_ms(const xmlNode *data, const char *name, guint *dest) { const char *value = NULL; long long value_ll; int rc = pcmk_rc_ok; CRM_CHECK(dest != NULL, return -1); *dest = 0; value = crm_element_value(data, name); rc = pcmk__scan_ll(value, &value_ll, 0LL); if (rc != pcmk_rc_ok) { crm_warn("Using default for %s " "because '%s' is not valid milliseconds: %s", name, value, pcmk_rc_str(rc)); return -1; } if ((value_ll < 0) || (value_ll > G_MAXUINT)) { crm_warn("Using default for %s because '%s' is out of range", name, value); return -1; } *dest = (guint) value_ll; return pcmk_ok; } int crm_element_value_int(const xmlNode *data, const char *name, int *dest) { const char *value = NULL; CRM_CHECK(dest != NULL, return -1); value = crm_element_value(data, name); if (value) { long long value_ll; int rc = pcmk__scan_ll(value, &value_ll, 0LL); *dest = PCMK__PARSE_INT_DEFAULT; if (rc != pcmk_rc_ok) { crm_warn("Using default for %s " "because '%s' is not a valid integer: %s", name, value, pcmk_rc_str(rc)); } else if ((value_ll < INT_MIN) || (value_ll > INT_MAX)) { crm_warn("Using default for %s because '%s' is out of range", name, value); } else { *dest = (int) value_ll; return 0; } } return -1; } char * crm_element_value_copy(const xmlNode *data, const char *name) { CRM_CHECK((data != NULL) && (name != NULL), return NULL); return pcmk__str_copy(pcmk__xe_get(data, name)); } const char * crm_xml_add_timeval(xmlNode *xml, const char *name_sec, const char *name_usec, const struct timeval *value) { const char *added = NULL; if (xml && name_sec && value) { added = crm_xml_add_ll(xml, name_sec, (long long) value->tv_sec); if (added && name_usec) { // Any error is ignored (we successfully added seconds) crm_xml_add_ll(xml, name_usec, (long long) value->tv_usec); } } return added; } const char * crm_xml_add_ms(xmlNode *node, const char *name, guint ms) { char *number = crm_strdup_printf("%u", ms); const char *added = crm_xml_add(node, name, number); free(number); return added; } // LCOV_EXCL_STOP // End deprecated API diff --git a/tools/crm_node.c b/tools/crm_node.c index 81fc56b808..f5a2148f87 100644 --- a/tools/crm_node.c +++ b/tools/crm_node.c @@ -1,875 +1,875 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #define SUMMARY "crm_node - Tool for displaying low-level node information" struct { gboolean corosync; gboolean dangerous_cmd; gboolean force_flag; char command; int nodeid; char *target_uname; } options = { .command = '\0', .force_flag = FALSE }; gboolean command_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error); gboolean name_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error); gboolean remove_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error); static GError *error = NULL; static GMainLoop *mainloop = NULL; static crm_exit_t exit_code = CRM_EX_OK; static pcmk__output_t *out = NULL; #define INDENT " " static GOptionEntry command_entries[] = { { "cluster-id", 'i', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, command_cb, "Display this node's cluster id", NULL }, { "list", 'l', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, command_cb, "Display all known members (past and present) of this cluster", NULL }, { "name", 'n', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, command_cb, "Display the name used by the cluster for this node", NULL }, { "partition", 'p', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, command_cb, "Display the members of this partition", NULL }, { "quorum", 'q', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, command_cb, "Display a 1 if our partition has quorum, 0 if not", NULL }, { "name-for-id", 'N', 0, G_OPTION_ARG_CALLBACK, name_cb, "Display the name used by the cluster for the node with the specified ID", "ID" }, { "remove", 'R', 0, G_OPTION_ARG_CALLBACK, remove_cb, "(Advanced) Remove the (stopped) node with the specified name from Pacemaker's\n" INDENT "configuration and caches (the node must already have been removed from\n" INDENT "the underlying cluster stack configuration", "NAME" }, { NULL } }; static GOptionEntry addl_entries[] = { { "force", 'f', 0, G_OPTION_ARG_NONE, &options.force_flag, NULL, NULL }, #if SUPPORT_COROSYNC /* Unused and deprecated */ { "corosync", 'C', G_OPTION_FLAG_HIDDEN, G_OPTION_ARG_NONE, &options.corosync, NULL, NULL }, #endif // @TODO add timeout option for when IPC replies are needed { NULL } }; static pcmk__supported_format_t formats[] = { PCMK__SUPPORTED_FORMAT_NONE, PCMK__SUPPORTED_FORMAT_TEXT, PCMK__SUPPORTED_FORMAT_XML, { NULL, NULL, NULL } }; gboolean command_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error) { if (pcmk__str_eq("-i", option_name, pcmk__str_casei) || pcmk__str_eq("--cluster-id", option_name, pcmk__str_casei)) { options.command = 'i'; } else if (pcmk__str_eq("-l", option_name, pcmk__str_casei) || pcmk__str_eq("--list", option_name, pcmk__str_casei)) { options.command = 'l'; } else if (pcmk__str_eq("-n", option_name, pcmk__str_casei) || pcmk__str_eq("--name", option_name, pcmk__str_casei)) { options.command = 'n'; } else if (pcmk__str_eq("-p", option_name, pcmk__str_casei) || pcmk__str_eq("--partition", option_name, pcmk__str_casei)) { options.command = 'p'; } else if (pcmk__str_eq("-q", option_name, pcmk__str_casei) || pcmk__str_eq("--quorum", option_name, pcmk__str_casei)) { options.command = 'q'; } else { g_set_error(error, PCMK__EXITC_ERROR, CRM_EX_INVALID_PARAM, "Unknown param passed to command_cb: %s", option_name); return FALSE; } return TRUE; } gboolean name_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error) { options.command = 'N'; pcmk__scan_min_int(optarg, &(options.nodeid), 0); return TRUE; } gboolean remove_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error) { if (optarg == NULL) { g_set_error(error, PCMK__EXITC_ERROR, CRM_EX_INVALID_PARAM, "-R option requires an argument"); return FALSE; } options.command = 'R'; options.dangerous_cmd = TRUE; pcmk__str_update(&options.target_uname, optarg); return TRUE; } PCMK__OUTPUT_ARGS("node-id", "uint32_t") static int node_id_default(pcmk__output_t *out, va_list args) { uint32_t node_id = va_arg(args, uint32_t); out->info(out, "%" PRIu32, node_id); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("node-id", "uint32_t") static int node_id_xml(pcmk__output_t *out, va_list args) { uint32_t node_id = va_arg(args, uint32_t); char *id_s = crm_strdup_printf("%" PRIu32, node_id); pcmk__output_create_xml_node(out, PCMK_XE_NODE_INFO, PCMK_XA_NODEID, id_s, NULL); free(id_s); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("simple-node-list", "GList *") static int simple_node_list_default(pcmk__output_t *out, va_list args) { GList *nodes = va_arg(args, GList *); for (GList *node_iter = nodes; node_iter != NULL; node_iter = node_iter->next) { pcmk_controld_api_node_t *node = node_iter->data; out->info(out, "%" PRIu32 " %s %s", node->id, pcmk__s(node->uname, ""), pcmk__s(node->state, "")); } return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("simple-node-list", "GList *") static int simple_node_list_xml(pcmk__output_t *out, va_list args) { GList *nodes = va_arg(args, GList *); out->begin_list(out, NULL, NULL, PCMK_XE_NODES); for (GList *node_iter = nodes; node_iter != NULL; node_iter = node_iter->next) { pcmk_controld_api_node_t *node = node_iter->data; char *id_s = crm_strdup_printf("%" PRIu32, node->id); pcmk__output_create_xml_node(out, PCMK_XE_NODE, PCMK_XA_ID, id_s, PCMK_XA_NAME, node->uname, PCMK_XA_STATE, node->state, NULL); free(id_s); } out->end_list(out); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("node-name", "uint32_t", "const char *") static int node_name_default(pcmk__output_t *out, va_list args) { uint32_t node_id G_GNUC_UNUSED = va_arg(args, uint32_t); const char *node_name = va_arg(args, const char *); out->info(out, "%s", node_name); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("node-name", "uint32_t", "const char *") static int node_name_xml(pcmk__output_t *out, va_list args) { uint32_t node_id = va_arg(args, uint32_t); const char *node_name = va_arg(args, const char *); char *id_s = crm_strdup_printf("%" PRIu32, node_id); pcmk__output_create_xml_node(out, PCMK_XE_NODE_INFO, PCMK_XA_NODEID, id_s, PCMK_XA_UNAME, node_name, NULL); free(id_s); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("partition-list", "GList *") static int partition_list_default(pcmk__output_t *out, va_list args) { GList *nodes = va_arg(args, GList *); GString *buffer = NULL; for (GList *node_iter = nodes; node_iter != NULL; node_iter = node_iter->next) { pcmk_controld_api_node_t *node = node_iter->data; if (pcmk__str_eq(node->state, "member", pcmk__str_none)) { pcmk__add_separated_word(&buffer, 128, pcmk__s(node->uname, ""), " "); } } if (buffer != NULL) { out->info(out, "%s", buffer->str); g_string_free(buffer, TRUE); return pcmk_rc_ok; } return pcmk_rc_no_output; } PCMK__OUTPUT_ARGS("partition-list", "GList *") static int partition_list_xml(pcmk__output_t *out, va_list args) { GList *nodes = va_arg(args, GList *); out->begin_list(out, NULL, NULL, PCMK_XE_NODES); for (GList *node_iter = nodes; node_iter != NULL; node_iter = node_iter->next) { pcmk_controld_api_node_t *node = node_iter->data; if (pcmk__str_eq(node->state, "member", pcmk__str_none)) { char *id_s = crm_strdup_printf("%" PRIu32, node->id); pcmk__output_create_xml_node(out, PCMK_XE_NODE, PCMK_XA_ID, id_s, PCMK_XA_NAME, node->uname, PCMK_XA_STATE, node->state, NULL); free(id_s); } } out->end_list(out); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("quorum", "bool") static int quorum_default(pcmk__output_t *out, va_list args) { bool have_quorum = va_arg(args, int); out->info(out, "%d", have_quorum); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("quorum", "bool") static int quorum_xml(pcmk__output_t *out, va_list args) { bool have_quorum = va_arg(args, int); pcmk__output_create_xml_node(out, PCMK_XE_CLUSTER_INFO, PCMK_XA_QUORUM, pcmk__btoa(have_quorum), NULL); return pcmk_rc_ok; } static pcmk__message_entry_t fmt_functions[] = { { "node-id", "default", node_id_default }, { "node-id", "xml", node_id_xml }, { "node-name", "default", node_name_default }, { "node-name", "xml", node_name_xml }, { "partition-list", "default", partition_list_default }, { "partition-list", "xml", partition_list_xml }, { "quorum", "default", quorum_default }, { "quorum", "xml", quorum_xml }, { "simple-node-list", "default", simple_node_list_default }, { "simple-node-list", "xml", simple_node_list_xml }, { NULL, NULL, NULL } }; static gint sort_node(gconstpointer a, gconstpointer b) { const pcmk_controld_api_node_t *node_a = a; const pcmk_controld_api_node_t *node_b = b; return pcmk__numeric_strcasecmp((node_a->uname? node_a->uname : ""), (node_b->uname? node_b->uname : "")); } static void controller_event_cb(pcmk_ipc_api_t *controld_api, enum pcmk_ipc_event event_type, crm_exit_t status, void *event_data, void *user_data) { pcmk_controld_api_reply_t *reply = event_data; switch (event_type) { case pcmk_ipc_event_disconnect: if (exit_code == CRM_EX_DISCONNECT) { // Unexpected g_set_error(&error, PCMK__EXITC_ERROR, exit_code, "Lost connection to controller"); } goto done; break; case pcmk_ipc_event_reply: break; default: return; } if (status != CRM_EX_OK) { exit_code = status; g_set_error(&error, PCMK__EXITC_ERROR, status, "Bad reply from controller: %s", crm_exit_str(status)); goto done; } if (reply->reply_type != pcmk_controld_reply_nodes) { g_set_error(&error, PCMK__EXITC_ERROR, CRM_EX_INDETERMINATE, "Unknown reply type %d from controller", reply->reply_type); goto done; } reply->data.nodes = g_list_sort(reply->data.nodes, sort_node); if (options.command == 'p') { out->message(out, "partition-list", reply->data.nodes); } else if (options.command == 'l') { out->message(out, "simple-node-list", reply->data.nodes); } // Success exit_code = CRM_EX_OK; done: pcmk_disconnect_ipc(controld_api); pcmk_quit_main_loop(mainloop, 10); } static void run_controller_mainloop(void) { pcmk_ipc_api_t *controld_api = NULL; int rc; // Set disconnect exit code to handle unexpected disconnects exit_code = CRM_EX_DISCONNECT; // Create controller IPC object rc = pcmk_new_ipc_api(&controld_api, pcmk_ipc_controld); if (rc != pcmk_rc_ok) { g_set_error(&error, PCMK__RC_ERROR, rc, "Could not connect to controller: %s", pcmk_rc_str(rc)); return; } pcmk_register_ipc_callback(controld_api, controller_event_cb, NULL); // Connect to controller rc = pcmk__connect_ipc(controld_api, pcmk_ipc_dispatch_main, 5); if (rc != pcmk_rc_ok) { exit_code = pcmk_rc2exitc(rc); g_set_error(&error, PCMK__EXITC_ERROR, exit_code, "Could not connect to %s: %s", pcmk_ipc_name(controld_api, true), pcmk_rc_str(rc)); return; } rc = pcmk_controld_api_list_nodes(controld_api); if (rc != pcmk_rc_ok) { pcmk_disconnect_ipc(controld_api); exit_code = pcmk_rc2exitc(rc); g_set_error(&error, PCMK__EXITC_ERROR, exit_code, "Could not ping controller: %s", pcmk_rc_str(rc)); return; } // Run main loop to get controller reply via controller_event_cb() mainloop = g_main_loop_new(NULL, FALSE); g_main_loop_run(mainloop); g_main_loop_unref(mainloop); mainloop = NULL; pcmk_free_ipc_api(controld_api); } static void print_node_id(void) { uint32_t nodeid = 0; int rc = pcmk__query_node_info(out, &nodeid, NULL, NULL, NULL, NULL, NULL, false, 0); if (rc != pcmk_rc_ok) { /* pcmk__query_node_info already sets an error message on the output object, * so there's no need to call g_set_error here. That would just create a * duplicate error message in the output. */ exit_code = pcmk_rc2exitc(rc); return; } rc = out->message(out, "node-id", nodeid); if (rc != pcmk_rc_ok) { g_set_error(&error, PCMK__RC_ERROR, rc, "Could not print node ID: %s", pcmk_rc_str(rc)); } exit_code = pcmk_rc2exitc(rc); } static void print_node_name(uint32_t nodeid) { int rc = pcmk_rc_ok; char *node_name = NULL; if (nodeid == 0) { // Check environment first (i.e. when called by resource agent) const char *name = getenv("OCF_RESKEY_" CRM_META "_" PCMK__META_ON_NODE); if (name != NULL) { rc = out->message(out, "node-name", 0UL, name); goto done; } } // Otherwise ask the controller /* pcmk__query_node_name already sets an error message on the output object, * so there's no need to call g_set_error here. That would just create a * duplicate error message in the output. */ rc = pcmk__query_node_name(out, nodeid, &node_name, 0); if (rc != pcmk_rc_ok) { exit_code = pcmk_rc2exitc(rc); return; } rc = out->message(out, "node-name", 0UL, node_name); done: if (node_name != NULL) { free(node_name); } if (rc != pcmk_rc_ok) { g_set_error(&error, PCMK__RC_ERROR, rc, "Could not print node name: %s", pcmk_rc_str(rc)); } exit_code = pcmk_rc2exitc(rc); } static void print_quorum(void) { bool quorum; int rc = pcmk__query_node_info(out, NULL, NULL, NULL, NULL, &quorum, NULL, false, 0); if (rc != pcmk_rc_ok) { /* pcmk__query_node_info already sets an error message on the output object, * so there's no need to call g_set_error here. That would just create a * duplicate error message in the output. */ exit_code = pcmk_rc2exitc(rc); return; } rc = out->message(out, "quorum", quorum); if (rc != pcmk_rc_ok) { g_set_error(&error, PCMK__RC_ERROR, rc, "Could not print quorum status: %s", pcmk_rc_str(rc)); } exit_code = pcmk_rc2exitc(rc); } /*! * \internal * \brief Extend a transaction by removing a node from a CIB section * * \param[in,out] cib Active CIB connection * \param[in] element CIB element containing node name and/or ID * \param[in] section CIB section that \p element is in * \param[in] node_name Name of node to purge (NULL to leave unspecified) * \param[in] node_id Node ID of node to purge (0 to leave unspecified) * * \note At least one of node_name and node_id must be specified. * \return Standard Pacemaker return code */ static int remove_from_section(cib_t *cib, const char *element, const char *section, const char *node_name, long node_id) { int rc = pcmk_rc_ok; xmlNode *xml = pcmk__xe_create(NULL, element); crm_xml_add(xml, PCMK_XA_UNAME, node_name); if (node_id > 0) { - crm_xml_add_ll(xml, PCMK_XA_ID, node_id); + pcmk__xe_set_ll(xml, PCMK_XA_ID, (long long) node_id); } rc = cib->cmds->remove(cib, section, xml, cib_transaction); pcmk__xml_free(xml); return (rc >= 0)? pcmk_rc_ok : pcmk_legacy2rc(rc); } /*! * \internal * \brief Purge a node from CIB * * \param[in] node_name Name of node to purge (or NULL to leave unspecified) * \param[in] node_id Node ID of node to purge (or 0 to leave unspecified) * * \note At least one of node_name and node_id must be specified. * \return Standard Pacemaker return code */ static int purge_node_from_cib(const char *node_name, long node_id) { int rc = pcmk_rc_ok; int commit_rc = pcmk_rc_ok; cib_t *cib = NULL; // Connect to CIB and start a transaction cib = cib_new(); if (cib == NULL) { return ENOTCONN; } rc = cib__signon_attempts(cib, cib_command, 5); if (rc == pcmk_ok) { rc = cib->cmds->init_transaction(cib); } if (rc != pcmk_ok) { rc = pcmk_legacy2rc(rc); cib__clean_up_connection(&cib); return rc; } // Remove from configuration and status rc = remove_from_section(cib, PCMK_XE_NODE, PCMK_XE_NODES, node_name, node_id); if (rc == pcmk_rc_ok) { rc = remove_from_section(cib, PCMK__XE_NODE_STATE, PCMK_XE_STATUS, node_name, node_id); } // Commit the transaction commit_rc = cib->cmds->end_transaction(cib, (rc == pcmk_rc_ok), cib_sync_call); cib__clean_up_connection(&cib); if ((rc == pcmk_rc_ok) && (commit_rc == pcmk_ok)) { crm_debug("Purged node %s (%ld) from CIB", pcmk__s(node_name, "by ID"), node_id); } return rc; } /*! * \internal * \brief Purge a node from a single server's peer cache * * \param[in] server IPC server to send request to * \param[in] node_name Name of node to purge (or NULL to leave unspecified) * \param[in] node_id Node ID of node to purge (or 0 to leave unspecified) * * \note At least one of node_name and node_id must be specified. * \return Standard Pacemaker return code */ static int purge_node_from(enum pcmk_ipc_server server, const char *node_name, long node_id) { pcmk_ipc_api_t *api = NULL; int rc; rc = pcmk_new_ipc_api(&api, server); if (rc != pcmk_rc_ok) { goto done; } rc = pcmk__connect_ipc(api, pcmk_ipc_dispatch_sync, 5); if (rc != pcmk_rc_ok) { goto done; } rc = pcmk_ipc_purge_node(api, node_name, node_id); done: if (rc != pcmk_rc_ok) { // Debug message already logged on success g_set_error(&error, PCMK__RC_ERROR, rc, "Could not purge node %s from %s: %s", pcmk__s(node_name, "by ID"), pcmk_ipc_name(api, true), pcmk_rc_str(rc)); } pcmk_free_ipc_api(api); return rc; } /*! * \internal * \brief Purge a node from the fencer's peer cache * * \param[in] node_name Name of node to purge (or NULL to leave unspecified) * \param[in] node_id Node ID of node to purge (or 0 to leave unspecified) * * \note At least one of node_name and node_id must be specified. * \return Standard Pacemaker return code */ static int purge_node_from_fencer(const char *node_name, long node_id) { int rc = pcmk_rc_ok; crm_ipc_t *conn = NULL; xmlNode *cmd = NULL; conn = crm_ipc_new("stonith-ng", 0); if (conn == NULL) { rc = ENOTCONN; exit_code = pcmk_rc2exitc(rc); g_set_error(&error, PCMK__EXITC_ERROR, exit_code, "Could not connect to fencer to purge node %s", pcmk__s(node_name, "by ID")); return rc; } rc = pcmk__connect_generic_ipc(conn); if (rc != pcmk_rc_ok) { exit_code = pcmk_rc2exitc(rc); g_set_error(&error, PCMK__EXITC_ERROR, exit_code, "Could not connect to fencer to purge node %s: %s", pcmk__s(node_name, "by ID"), pcmk_rc_str(rc)); crm_ipc_destroy(conn); return rc; } cmd = pcmk__new_request(pcmk_ipc_fenced, crm_system_name, NULL, PCMK__VALUE_STONITH_NG, CRM_OP_RM_NODE_CACHE, NULL); if (node_id > 0) { - crm_xml_add_ll(cmd, PCMK_XA_ID, node_id); + pcmk__xe_set_ll(cmd, PCMK_XA_ID, (long long) node_id); } crm_xml_add(cmd, PCMK_XA_UNAME, node_name); rc = crm_ipc_send(conn, cmd, 0, 0, NULL); if (rc >= 0) { rc = pcmk_rc_ok; crm_debug("Purged node %s (%ld) from fencer", pcmk__s(node_name, "by ID"), node_id); } else { rc = pcmk_legacy2rc(rc); fprintf(stderr, "Could not purge node %s from fencer: %s\n", pcmk__s(node_name, "by ID"), pcmk_rc_str(rc)); } pcmk__xml_free(cmd); crm_ipc_close(conn); crm_ipc_destroy(conn); return rc; } static void remove_node(const char *target_uname) { int rc = pcmk_rc_ok; long nodeid = 0; const char *node_name = NULL; char *endptr = NULL; const enum pcmk_ipc_server servers[] = { pcmk_ipc_controld, pcmk_ipc_attrd, }; // Check whether node was specified by name or numeric ID errno = 0; nodeid = strtol(target_uname, &endptr, 10); if ((errno != 0) || (endptr == target_uname) || (*endptr != '\0') || (nodeid <= 0)) { // It's not a positive integer, so assume it's a node name nodeid = 0; node_name = target_uname; } for (int i = 0; i < PCMK__NELEM(servers); ++i) { rc = purge_node_from(servers[i], node_name, nodeid); if (rc != pcmk_rc_ok) { exit_code = pcmk_rc2exitc(rc); return; } } // The fencer hasn't been converted to pcmk_ipc_api_t yet rc = purge_node_from_fencer(node_name, nodeid); if (rc != pcmk_rc_ok) { exit_code = pcmk_rc2exitc(rc); return; } // Lastly, purge the node from the CIB itself rc = purge_node_from_cib(node_name, nodeid); exit_code = pcmk_rc2exitc(rc); } static GOptionContext * build_arg_context(pcmk__common_args_t *args, GOptionGroup **group) { GOptionContext *context = NULL; GOptionEntry extra_prog_entries[] = { { "quiet", 'Q', 0, G_OPTION_ARG_NONE, &(args->quiet), "Be less descriptive in output.", NULL }, { NULL } }; context = pcmk__build_arg_context(args, "text (default), xml", group, NULL); /* Add the -q option, which cannot be part of the globally supported options * because some tools use that flag for something else. */ pcmk__add_main_args(context, extra_prog_entries); pcmk__add_arg_group(context, "commands", "Commands:", "Show command help", command_entries); pcmk__add_arg_group(context, "additional", "Additional Options:", "Show additional options", addl_entries); return context; } int main(int argc, char **argv) { int rc = pcmk_rc_ok; GOptionGroup *output_group = NULL; pcmk__common_args_t *args = pcmk__new_common_args(SUMMARY); gchar **processed_args = pcmk__cmdline_preproc(argv, "NR"); GOptionContext *context = build_arg_context(args, &output_group); pcmk__register_formats(output_group, formats); if (!g_option_context_parse_strv(context, &processed_args, &error)) { exit_code = CRM_EX_USAGE; goto done; } pcmk__cli_init_logging("crm_node", args->verbosity); rc = pcmk__output_new(&out, args->output_ty, args->output_dest, argv); if (rc != pcmk_rc_ok) { exit_code = pcmk_rc2exitc(rc); g_set_error(&error, PCMK__EXITC_ERROR, exit_code, "Error creating output format %s: %s", args->output_ty, pcmk_rc_str(rc)); goto done; } if (args->version) { out->version(out, false); goto done; } if (options.command == 0) { char *help = g_option_context_get_help(context, TRUE, NULL); out->err(out, "%s", help); g_free(help); exit_code = CRM_EX_USAGE; goto done; } if (options.dangerous_cmd && options.force_flag == FALSE) { exit_code = CRM_EX_USAGE; g_set_error(&error, PCMK__EXITC_ERROR, exit_code, "The supplied command is considered dangerous." " To prevent accidental destruction of the cluster," " the --force flag is required in order to proceed."); goto done; } pcmk__register_lib_messages(out); pcmk__register_messages(out, fmt_functions); switch (options.command) { case 'i': print_node_id(); break; case 'n': print_node_name(0); break; case 'q': print_quorum(); break; case 'N': print_node_name(options.nodeid); break; case 'R': remove_node(options.target_uname); break; case 'l': case 'p': run_controller_mainloop(); break; default: break; } done: g_strfreev(processed_args); pcmk__free_arg_context(context); pcmk__output_and_clear_error(&error, out); if (out != NULL) { out->finish(out, exit_code, true, NULL); pcmk__output_free(out); } pcmk__unregister_formats(); return crm_exit(exit_code); }