diff --git a/daemons/based/pacemaker-based.c b/daemons/based/pacemaker-based.c index 7d3c884c1c..ab00e4fedc 100644 --- a/daemons/based/pacemaker-based.c +++ b/daemons/based/pacemaker-based.c @@ -1,433 +1,435 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #define SUMMARY "daemon for managing the configuration of a Pacemaker cluster" extern int init_remote_listener(int port, gboolean encrypted); gboolean cib_shutdown_flag = FALSE; int cib_status = pcmk_ok; pcmk_cluster_t *crm_cluster = NULL; GMainLoop *mainloop = NULL; gchar *cib_root = NULL; static gboolean preserve_status = FALSE; gboolean cib_writes_enabled = TRUE; gboolean stand_alone = FALSE; int remote_fd = 0; int remote_tls_fd = 0; GHashTable *config_hash = NULL; static void cib_init(void); void cib_shutdown(int nsig); static bool startCib(const char *filename); extern int write_cib_contents(gpointer p); static crm_exit_t exit_code = CRM_EX_OK; static void cib_enable_writes(int nsig) { crm_info("(Re)enabling disk writes"); cib_writes_enabled = TRUE; } /*! * \internal * \brief Set up options, users, and groups for stand-alone mode * * \param[out] error GLib error object * * \return Standard Pacemaker return code */ static int setup_stand_alone(GError **error) { int rc = 0; struct passwd *pwentry = NULL; preserve_status = TRUE; cib_writes_enabled = FALSE; errno = 0; pwentry = getpwnam(CRM_DAEMON_USER); if (pwentry == NULL) { exit_code = CRM_EX_FATAL; if (errno != 0) { g_set_error(error, PCMK__EXITC_ERROR, exit_code, "Error getting password DB entry for %s: %s", CRM_DAEMON_USER, strerror(errno)); return errno; } g_set_error(error, PCMK__EXITC_ERROR, exit_code, "Password DB entry for '%s' not found", CRM_DAEMON_USER); return ENXIO; } rc = setgid(pwentry->pw_gid); if (rc < 0) { exit_code = CRM_EX_FATAL; g_set_error(error, PCMK__EXITC_ERROR, exit_code, "Could not set group to %d: %s", pwentry->pw_gid, strerror(errno)); return errno; } rc = initgroups(CRM_DAEMON_USER, pwentry->pw_gid); if (rc < 0) { exit_code = CRM_EX_FATAL; g_set_error(error, PCMK__EXITC_ERROR, exit_code, "Could not setup groups for user %d: %s", pwentry->pw_uid, strerror(errno)); return errno; } rc = setuid(pwentry->pw_uid); if (rc < 0) { exit_code = CRM_EX_FATAL; g_set_error(error, PCMK__EXITC_ERROR, exit_code, "Could not set user to %d: %s", pwentry->pw_uid, strerror(errno)); return errno; } return pcmk_rc_ok; } /* @COMPAT Deprecated since 2.1.8. Use pcmk_list_cluster_options() or * crm_attribute --list-options=cluster instead of querying daemon metadata. + * + * NOTE: pcs (as of at least 0.11.8) uses this */ static int based_metadata(pcmk__output_t *out) { return pcmk__daemon_metadata(out, "pacemaker-based", "Cluster Information Base manager options", "Cluster options used by Pacemaker's Cluster " "Information Base manager", pcmk__opt_based); } static GOptionEntry entries[] = { { "stand-alone", 's', G_OPTION_FLAG_NONE, G_OPTION_ARG_NONE, &stand_alone, "(Advanced use only) Run in stand-alone mode", NULL }, { "disk-writes", 'w', G_OPTION_FLAG_NONE, G_OPTION_ARG_NONE, &cib_writes_enabled, "(Advanced use only) Enable disk writes (enabled by default unless in " "stand-alone mode)", NULL }, { "cib-root", 'r', G_OPTION_FLAG_NONE, G_OPTION_ARG_FILENAME, &cib_root, "(Advanced use only) Directory where the CIB XML file should be located " "(default: " CRM_CONFIG_DIR ")", NULL }, { NULL } }; static pcmk__supported_format_t formats[] = { PCMK__SUPPORTED_FORMAT_NONE, PCMK__SUPPORTED_FORMAT_TEXT, PCMK__SUPPORTED_FORMAT_XML, { NULL, NULL, NULL } }; static GOptionContext * build_arg_context(pcmk__common_args_t *args, GOptionGroup **group) { GOptionContext *context = NULL; context = pcmk__build_arg_context(args, "text (default), xml", group, NULL); pcmk__add_main_args(context, entries); return context; } int main(int argc, char **argv) { int rc = pcmk_rc_ok; crm_ipc_t *old_instance = NULL; pcmk__output_t *out = NULL; GError *error = NULL; GOptionGroup *output_group = NULL; pcmk__common_args_t *args = pcmk__new_common_args(SUMMARY); gchar **processed_args = pcmk__cmdline_preproc(argv, "r"); GOptionContext *context = build_arg_context(args, &output_group); crm_log_preinit(NULL, argc, argv); pcmk__register_formats(output_group, formats); if (!g_option_context_parse_strv(context, &processed_args, &error)) { exit_code = CRM_EX_USAGE; goto done; } rc = pcmk__output_new(&out, args->output_ty, args->output_dest, argv); if (rc != pcmk_rc_ok) { exit_code = CRM_EX_ERROR; g_set_error(&error, PCMK__EXITC_ERROR, exit_code, "Error creating output format %s: %s", args->output_ty, pcmk_rc_str(rc)); goto done; } if (args->version) { out->version(out, false); goto done; } mainloop_add_signal(SIGTERM, cib_shutdown); mainloop_add_signal(SIGPIPE, cib_enable_writes); cib_writer = mainloop_add_trigger(G_PRIORITY_LOW, write_cib_contents, NULL); if ((g_strv_length(processed_args) >= 2) && pcmk__str_eq(processed_args[1], "metadata", pcmk__str_none)) { rc = based_metadata(out); if (rc != pcmk_rc_ok) { exit_code = CRM_EX_FATAL; g_set_error(&error, PCMK__EXITC_ERROR, exit_code, "Unable to display metadata: %s", pcmk_rc_str(rc)); } goto done; } pcmk__cli_init_logging("pacemaker-based", args->verbosity); crm_log_init(NULL, LOG_INFO, TRUE, FALSE, argc, argv, FALSE); crm_notice("Starting Pacemaker CIB manager"); old_instance = crm_ipc_new(PCMK__SERVER_BASED_RO, 0); if (old_instance == NULL) { /* crm_ipc_new() will have already logged an error message with * crm_err() */ exit_code = CRM_EX_FATAL; goto done; } if (pcmk__connect_generic_ipc(old_instance) == pcmk_rc_ok) { /* IPC end-point already up */ crm_ipc_close(old_instance); crm_ipc_destroy(old_instance); crm_err("pacemaker-based is already active, aborting startup"); goto done; } else { /* not up or not authentic, we'll proceed either way */ crm_ipc_destroy(old_instance); old_instance = NULL; } if (stand_alone) { rc = setup_stand_alone(&error); if (rc != pcmk_rc_ok) { goto done; } } if (cib_root == NULL) { cib_root = g_strdup(CRM_CONFIG_DIR); } else { crm_notice("Using custom config location: %s", cib_root); } if (!pcmk__daemon_can_write(cib_root, NULL)) { exit_code = CRM_EX_FATAL; crm_err("Terminating due to bad permissions on %s", cib_root); g_set_error(&error, PCMK__EXITC_ERROR, exit_code, "Bad permissions on %s (see logs for details)", cib_root); goto done; } pcmk__cluster_init_node_caches(); // Read initial CIB, connect to cluster, and start IPC servers cib_init(); // Run the main loop mainloop = g_main_loop_new(NULL, FALSE); crm_notice("Pacemaker CIB manager successfully started and accepting connections"); g_main_loop_run(mainloop); /* If main loop returned, clean up and exit. We disconnect in case * terminate_cib() was called with fast=-1. */ pcmk_cluster_disconnect(crm_cluster); pcmk__stop_based_ipc(ipcs_ro, ipcs_rw, ipcs_shm); done: g_strfreev(processed_args); pcmk__free_arg_context(context); pcmk__cluster_destroy_node_caches(); if (config_hash != NULL) { g_hash_table_destroy(config_hash); } pcmk__client_cleanup(); pcmk_cluster_free(crm_cluster); g_free(cib_root); pcmk__output_and_clear_error(&error, out); if (out != NULL) { out->finish(out, exit_code, true, NULL); pcmk__output_free(out); } pcmk__unregister_formats(); crm_exit(exit_code); } #if SUPPORT_COROSYNC static void cib_cs_dispatch(cpg_handle_t handle, const struct cpg_name *groupName, uint32_t nodeid, uint32_t pid, void *msg, size_t msg_len) { xmlNode *xml = NULL; const char *from = NULL; char *data = pcmk__cpg_message_data(handle, nodeid, pid, msg, &from); if(data == NULL) { return; } xml = pcmk__xml_parse(data); if (xml == NULL) { crm_err("Invalid XML: '%.120s'", data); free(data); return; } crm_xml_add(xml, PCMK__XA_SRC, from); cib_peer_callback(xml, NULL); pcmk__xml_free(xml); free(data); } static void cib_cs_destroy(gpointer user_data) { if (cib_shutdown_flag) { crm_info("Corosync disconnection complete"); } else { crm_crit("Lost connection to cluster layer, shutting down"); terminate_cib(__func__, CRM_EX_DISCONNECT); } } #endif static void cib_peer_update_callback(enum pcmk__node_update type, pcmk__node_status_t *node, const void *data) { switch (type) { case pcmk__node_update_name: case pcmk__node_update_state: if (cib_shutdown_flag && (pcmk__cluster_num_active_nodes() < 2) && (pcmk__ipc_client_count() == 0)) { crm_info("No more peers"); terminate_cib(__func__, -1); } break; default: break; } } static void cib_init(void) { crm_cluster = pcmk_cluster_new(); #if SUPPORT_COROSYNC if (pcmk_get_cluster_layer() == pcmk_cluster_layer_corosync) { pcmk_cluster_set_destroy_fn(crm_cluster, cib_cs_destroy); pcmk_cpg_set_deliver_fn(crm_cluster, cib_cs_dispatch); pcmk_cpg_set_confchg_fn(crm_cluster, pcmk__cpg_confchg_cb); } #endif // SUPPORT_COROSYNC config_hash = pcmk__strkey_table(free, free); if (startCib("cib.xml") == FALSE) { crm_crit("Cannot start CIB... terminating"); crm_exit(CRM_EX_NOINPUT); } if (!stand_alone) { pcmk__cluster_set_status_callback(&cib_peer_update_callback); if (pcmk_cluster_connect(crm_cluster) != pcmk_rc_ok) { crm_crit("Cannot sign in to the cluster... terminating"); crm_exit(CRM_EX_FATAL); } } pcmk__serve_based_ipc(&ipcs_ro, &ipcs_rw, &ipcs_shm, &ipc_ro_callbacks, &ipc_rw_callbacks); if (stand_alone) { based_is_primary = true; } } static bool startCib(const char *filename) { gboolean active = FALSE; xmlNode *cib = readCibXmlFile(cib_root, filename, !preserve_status); if (activateCibXml(cib, TRUE, "start") == 0) { int port = 0; active = TRUE; cib_read_config(config_hash, cib); pcmk__scan_port(crm_element_value(cib, PCMK_XA_REMOTE_TLS_PORT), &port); if (port >= 0) { remote_tls_fd = init_remote_listener(port, TRUE); } pcmk__scan_port(crm_element_value(cib, PCMK_XA_REMOTE_CLEAR_PORT), &port); if (port >= 0) { remote_fd = init_remote_listener(port, FALSE); } } return active; } diff --git a/daemons/controld/pacemaker-controld.c b/daemons/controld/pacemaker-controld.c index 913518bc9f..b1e5938c07 100644 --- a/daemons/controld/pacemaker-controld.c +++ b/daemons/controld/pacemaker-controld.c @@ -1,223 +1,225 @@ /* * 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 #define SUMMARY "daemon for coordinating a Pacemaker cluster's response " \ "to events" _Noreturn void crmd_init(void); extern void init_dotfile(void); controld_globals_t controld_globals = { // Automatic initialization to 0, false, or NULL is fine for most members .fsa_state = S_STARTING, .fsa_actions = A_NOTHING, }; static pcmk__supported_format_t formats[] = { PCMK__SUPPORTED_FORMAT_NONE, PCMK__SUPPORTED_FORMAT_TEXT, PCMK__SUPPORTED_FORMAT_XML, { NULL, NULL, NULL } }; /* @COMPAT Deprecated since 2.1.8. Use pcmk_list_cluster_options() or * crm_attribute --list-options=cluster instead of querying daemon metadata. + * + * NOTE: pcs (as of at least 0.11.8) uses this */ static int controld_metadata(pcmk__output_t *out) { return pcmk__daemon_metadata(out, "pacemaker-controld", "Pacemaker controller options", "Cluster options used by Pacemaker's " "controller", pcmk__opt_controld); } static GOptionContext * build_arg_context(pcmk__common_args_t *args, GOptionGroup **group) { return pcmk__build_arg_context(args, "text (default), xml", group, NULL); } int main(int argc, char **argv) { int rc = pcmk_rc_ok; crm_exit_t exit_code = CRM_EX_OK; bool initialize = true; crm_ipc_t *old_instance = NULL; pcmk__output_t *out = NULL; GError *error = NULL; GOptionGroup *output_group = NULL; pcmk__common_args_t *args = pcmk__new_common_args(SUMMARY); gchar **processed_args = pcmk__cmdline_preproc(argv, NULL); GOptionContext *context = build_arg_context(args, &output_group); crm_log_preinit(NULL, argc, argv); pcmk__register_formats(output_group, formats); if (!g_option_context_parse_strv(context, &processed_args, &error)) { exit_code = CRM_EX_USAGE; goto done; } rc = pcmk__output_new(&out, args->output_ty, args->output_dest, argv); if (rc != pcmk_rc_ok) { exit_code = CRM_EX_ERROR; g_set_error(&error, PCMK__EXITC_ERROR, exit_code, "Error creating output format %s: %s", args->output_ty, pcmk_rc_str(rc)); goto done; } if (args->version) { out->version(out, false); initialize = false; goto done; } if ((g_strv_length(processed_args) >= 2) && pcmk__str_eq(processed_args[1], "metadata", pcmk__str_none)) { initialize = false; rc = controld_metadata(out); if (rc != pcmk_rc_ok) { exit_code = CRM_EX_FATAL; g_set_error(&error, PCMK__EXITC_ERROR, exit_code, "Unable to display metadata: %s", pcmk_rc_str(rc)); } goto done; } pcmk__cli_init_logging("pacemaker-controld", args->verbosity); crm_log_init(NULL, LOG_INFO, TRUE, FALSE, argc, argv, FALSE); crm_notice("Starting Pacemaker controller"); old_instance = crm_ipc_new(CRM_SYSTEM_CRMD, 0); if (old_instance == NULL) { /* crm_ipc_new will have already printed an error message with crm_err. */ exit_code = CRM_EX_FATAL; goto done; } if (pcmk__connect_generic_ipc(old_instance) == pcmk_rc_ok) { /* IPC end-point already up */ crm_ipc_close(old_instance); crm_ipc_destroy(old_instance); crm_err("pacemaker-controld is already active, aborting startup"); initialize = false; goto done; } else { /* not up or not authentic, we'll proceed either way */ crm_ipc_destroy(old_instance); old_instance = NULL; } if (pcmk__daemon_can_write(PE_STATE_DIR, NULL) == FALSE) { exit_code = CRM_EX_FATAL; crm_err("Terminating due to bad permissions on " PE_STATE_DIR); g_set_error(&error, PCMK__EXITC_ERROR, exit_code, "Bad permissions on " PE_STATE_DIR " (see logs for details)"); goto done; } else if (pcmk__daemon_can_write(CRM_CONFIG_DIR, NULL) == FALSE) { exit_code = CRM_EX_FATAL; crm_err("Terminating due to bad permissions on " CRM_CONFIG_DIR); g_set_error(&error, PCMK__EXITC_ERROR, exit_code, "Bad permissions on " CRM_CONFIG_DIR " (see logs for details)"); goto done; } if (pcmk__log_output_new(&(controld_globals.logger_out)) != pcmk_rc_ok) { exit_code = CRM_EX_FATAL; goto done; } pcmk__output_set_log_level(controld_globals.logger_out, LOG_TRACE); 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(); if ((exit_code == CRM_EX_OK) && initialize) { // Does not return crmd_init(); } crm_exit(exit_code); } void crmd_init(void) { crm_exit_t exit_code = CRM_EX_OK; enum crmd_fsa_state state; init_dotfile(); register_fsa_input(C_STARTUP, I_STARTUP, NULL); pcmk__cluster_init_node_caches(); state = s_crmd_fsa(C_STARTUP); if (state == S_PENDING || state == S_STARTING) { /* Create the mainloop and run it... */ crm_trace("Starting %s's mainloop", crm_system_name); controld_globals.mainloop = g_main_loop_new(NULL, FALSE); g_main_loop_run(controld_globals.mainloop); if (pcmk_is_set(controld_globals.fsa_input_register, R_STAYDOWN)) { crm_info("Inhibiting automated respawn"); exit_code = CRM_EX_FATAL; } } else { crm_err("Startup of %s failed. Current state: %s", crm_system_name, fsa_state2string(state)); exit_code = CRM_EX_ERROR; } crm_info("%s[%lu] exiting with status %d (%s)", crm_system_name, (unsigned long) getpid(), exit_code, crm_exit_str(exit_code)); crmd_fast_exit(exit_code); } diff --git a/daemons/fenced/pacemaker-fenced.c b/daemons/fenced/pacemaker-fenced.c index 5c0b9b24c5..94b1428e8e 100644 --- a/daemons/fenced/pacemaker-fenced.c +++ b/daemons/fenced/pacemaker-fenced.c @@ -1,674 +1,676 @@ /* * Copyright 2009-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include #include #include // PRIu32, PRIx32 #include #include #include #include #include #include #include #include #include #include #include #include #define SUMMARY "daemon for executing fencing devices in a Pacemaker cluster" long long stonith_watchdog_timeout_ms = 0; GList *stonith_watchdog_targets = NULL; static GMainLoop *mainloop = NULL; gboolean stand_alone = FALSE; gboolean stonith_shutdown_flag = FALSE; static qb_ipcs_service_t *ipcs = NULL; static pcmk__output_t *out = NULL; pcmk__supported_format_t formats[] = { PCMK__SUPPORTED_FORMAT_NONE, PCMK__SUPPORTED_FORMAT_TEXT, PCMK__SUPPORTED_FORMAT_XML, { NULL, NULL, NULL } }; static struct { bool no_cib_connect; gchar **log_files; } options; crm_exit_t exit_code = CRM_EX_OK; static void stonith_cleanup(void); static int32_t st_ipc_accept(qb_ipcs_connection_t * c, uid_t uid, gid_t gid) { if (stonith_shutdown_flag) { crm_info("Ignoring new client [%d] during shutdown", pcmk__client_pid(c)); return -ECONNREFUSED; } if (pcmk__new_client(c, uid, gid) == NULL) { return -ENOMEM; } return 0; } /* Exit code means? */ static int32_t st_ipc_dispatch(qb_ipcs_connection_t * qbc, void *data, size_t size) { uint32_t id = 0; uint32_t flags = 0; int call_options = 0; xmlNode *request = NULL; pcmk__client_t *c = pcmk__find_client(qbc); const char *op = NULL; if (c == NULL) { crm_info("Invalid client: %p", qbc); return 0; } request = pcmk__client_data2xml(c, data, &id, &flags); if (request == NULL) { pcmk__ipc_send_ack(c, id, flags, PCMK__XE_NACK, NULL, CRM_EX_PROTOCOL); return 0; } op = crm_element_value(request, PCMK__XA_CRM_TASK); if(pcmk__str_eq(op, CRM_OP_RM_NODE_CACHE, pcmk__str_casei)) { crm_xml_add(request, PCMK__XA_T, PCMK__VALUE_STONITH_NG); crm_xml_add(request, PCMK__XA_ST_OP, op); crm_xml_add(request, PCMK__XA_ST_CLIENTID, c->id); crm_xml_add(request, PCMK__XA_ST_CLIENTNAME, pcmk__client_name(c)); crm_xml_add(request, PCMK__XA_ST_CLIENTNODE, fenced_get_local_node()); pcmk__cluster_send_message(NULL, pcmk_ipc_fenced, request); pcmk__xml_free(request); return 0; } if (c->name == NULL) { const char *value = crm_element_value(request, PCMK__XA_ST_CLIENTNAME); c->name = crm_strdup_printf("%s.%u", pcmk__s(value, "unknown"), c->pid); } crm_element_value_int(request, PCMK__XA_ST_CALLOPT, &call_options); crm_trace("Flags %#08" PRIx32 "/%#08x for command %" PRIu32 " from client %s", flags, call_options, id, pcmk__client_name(c)); if (pcmk_is_set(call_options, st_opt_sync_call)) { CRM_ASSERT(flags & crm_ipc_client_response); CRM_LOG_ASSERT(c->request_id == 0); /* This means the client has two synchronous events in-flight */ c->request_id = id; /* Reply only to the last one */ } crm_xml_add(request, PCMK__XA_ST_CLIENTID, c->id); crm_xml_add(request, PCMK__XA_ST_CLIENTNAME, pcmk__client_name(c)); crm_xml_add(request, PCMK__XA_ST_CLIENTNODE, fenced_get_local_node()); crm_log_xml_trace(request, "ipc-received"); stonith_command(c, id, flags, request, NULL); pcmk__xml_free(request); return 0; } /* Error code means? */ static int32_t st_ipc_closed(qb_ipcs_connection_t * c) { pcmk__client_t *client = pcmk__find_client(c); if (client == NULL) { return 0; } crm_trace("Connection %p closed", c); pcmk__free_client(client); /* 0 means: yes, go ahead and destroy the connection */ return 0; } static void st_ipc_destroy(qb_ipcs_connection_t * c) { crm_trace("Connection %p destroyed", c); st_ipc_closed(c); } static void stonith_peer_callback(xmlNode * msg, void *private_data) { const char *remote_peer = crm_element_value(msg, PCMK__XA_SRC); const char *op = crm_element_value(msg, PCMK__XA_ST_OP); if (pcmk__str_eq(op, STONITH_OP_POKE, pcmk__str_none)) { return; } crm_log_xml_trace(msg, "Peer[inbound]"); stonith_command(NULL, 0, 0, msg, remote_peer); } #if SUPPORT_COROSYNC static void stonith_peer_ais_callback(cpg_handle_t handle, const struct cpg_name *groupName, uint32_t nodeid, uint32_t pid, void *msg, size_t msg_len) { xmlNode *xml = NULL; const char *from = NULL; char *data = pcmk__cpg_message_data(handle, nodeid, pid, msg, &from); if(data == NULL) { return; } xml = pcmk__xml_parse(data); if (xml == NULL) { crm_err("Invalid XML: '%.120s'", data); free(data); return; } crm_xml_add(xml, PCMK__XA_SRC, from); stonith_peer_callback(xml, NULL); pcmk__xml_free(xml); free(data); } static void stonith_peer_cs_destroy(gpointer user_data) { crm_crit("Lost connection to cluster layer, shutting down"); stonith_shutdown(0); } #endif void do_local_reply(const xmlNode *notify_src, pcmk__client_t *client, int call_options) { /* send callback to originating child */ int local_rc = pcmk_rc_ok; int rid = 0; uint32_t ipc_flags = crm_ipc_server_event; if (pcmk_is_set(call_options, st_opt_sync_call)) { CRM_LOG_ASSERT(client->request_id); rid = client->request_id; client->request_id = 0; ipc_flags = crm_ipc_flags_none; } local_rc = pcmk__ipc_send_xml(client, rid, notify_src, ipc_flags); if (local_rc == pcmk_rc_ok) { crm_trace("Sent response %d to client %s", rid, pcmk__client_name(client)); } else { crm_warn("%synchronous reply to client %s failed: %s", (pcmk_is_set(call_options, st_opt_sync_call)? "S" : "As"), pcmk__client_name(client), pcmk_rc_str(local_rc)); } } uint64_t get_stonith_flag(const char *name) { if (pcmk__str_eq(name, PCMK__VALUE_ST_NOTIFY_FENCE, pcmk__str_none)) { return st_callback_notify_fence; } else if (pcmk__str_eq(name, STONITH_OP_DEVICE_ADD, pcmk__str_casei)) { return st_callback_device_add; } else if (pcmk__str_eq(name, STONITH_OP_DEVICE_DEL, pcmk__str_casei)) { return st_callback_device_del; } else if (pcmk__str_eq(name, PCMK__VALUE_ST_NOTIFY_HISTORY, pcmk__str_none)) { return st_callback_notify_history; } else if (pcmk__str_eq(name, PCMK__VALUE_ST_NOTIFY_HISTORY_SYNCED, pcmk__str_none)) { return st_callback_notify_history_synced; } return st_callback_unknown; } static void stonith_notify_client(gpointer key, gpointer value, gpointer user_data) { const xmlNode *update_msg = user_data; pcmk__client_t *client = value; const char *type = NULL; CRM_CHECK(client != NULL, return); CRM_CHECK(update_msg != NULL, return); type = crm_element_value(update_msg, PCMK__XA_SUBT); CRM_CHECK(type != NULL, crm_log_xml_err(update_msg, "notify"); return); if (client->ipcs == NULL) { crm_trace("Skipping client with NULL channel"); return; } if (pcmk_is_set(client->flags, get_stonith_flag(type))) { int rc = pcmk__ipc_send_xml(client, 0, update_msg, crm_ipc_server_event); if (rc != pcmk_rc_ok) { crm_warn("%s notification of client %s failed: %s " QB_XS " id=%.8s rc=%d", type, pcmk__client_name(client), pcmk_rc_str(rc), client->id, rc); } else { crm_trace("Sent %s notification to client %s", type, pcmk__client_name(client)); } } } void do_stonith_async_timeout_update(const char *client_id, const char *call_id, int timeout) { pcmk__client_t *client = NULL; xmlNode *notify_data = NULL; if (!timeout || !call_id || !client_id) { return; } client = pcmk__find_client_by_id(client_id); if (!client) { return; } notify_data = pcmk__xe_create(NULL, PCMK__XE_ST_ASYNC_TIMEOUT_VALUE); crm_xml_add(notify_data, PCMK__XA_T, PCMK__VALUE_ST_ASYNC_TIMEOUT_VALUE); crm_xml_add(notify_data, PCMK__XA_ST_CALLID, call_id); crm_xml_add_int(notify_data, PCMK__XA_ST_TIMEOUT, timeout); crm_trace("timeout update is %d for client %s and call id %s", timeout, client_id, call_id); if (client) { pcmk__ipc_send_xml(client, 0, notify_data, crm_ipc_server_event); } pcmk__xml_free(notify_data); } /*! * \internal * \brief Notify relevant IPC clients of a fencing operation result * * \param[in] type Notification type * \param[in] result Result of fencing operation (assume success if NULL) * \param[in] data If not NULL, add to notification as call data */ void fenced_send_notification(const char *type, const pcmk__action_result_t *result, xmlNode *data) { /* TODO: Standardize the contents of data */ xmlNode *update_msg = pcmk__xe_create(NULL, PCMK__XE_NOTIFY); CRM_LOG_ASSERT(type != NULL); crm_xml_add(update_msg, PCMK__XA_T, PCMK__VALUE_ST_NOTIFY); crm_xml_add(update_msg, PCMK__XA_SUBT, type); crm_xml_add(update_msg, PCMK__XA_ST_OP, type); stonith__xe_set_result(update_msg, result); if (data != NULL) { xmlNode *wrapper = pcmk__xe_create(update_msg, PCMK__XE_ST_CALLDATA); pcmk__xml_copy(wrapper, data); } crm_trace("Notifying clients"); pcmk__foreach_ipc_client(stonith_notify_client, update_msg); pcmk__xml_free(update_msg); crm_trace("Notify complete"); } /*! * \internal * \brief Send notifications for a configuration change to subscribed clients * * \param[in] op Notification type (\c STONITH_OP_DEVICE_ADD, * \c STONITH_OP_DEVICE_DEL, \c STONITH_OP_LEVEL_ADD, or * \c STONITH_OP_LEVEL_DEL) * \param[in] result Operation result * \param[in] desc Description of what changed (either device ID or string * representation of level * ([])) */ void fenced_send_config_notification(const char *op, const pcmk__action_result_t *result, const char *desc) { xmlNode *notify_data = pcmk__xe_create(NULL, op); crm_xml_add(notify_data, PCMK__XA_ST_DEVICE_ID, desc); fenced_send_notification(op, result, notify_data); pcmk__xml_free(notify_data); } /*! * \internal * \brief Check whether a node does watchdog-fencing * * \param[in] node Name of node to check * * \return TRUE if node found in stonith_watchdog_targets * or stonith_watchdog_targets is empty indicating * all nodes are doing watchdog-fencing */ gboolean node_does_watchdog_fencing(const char *node) { return ((stonith_watchdog_targets == NULL) || pcmk__str_in_list(node, stonith_watchdog_targets, pcmk__str_casei)); } void stonith_shutdown(int nsig) { crm_info("Terminating with %d clients", pcmk__ipc_client_count()); stonith_shutdown_flag = TRUE; if (mainloop != NULL && g_main_loop_is_running(mainloop)) { g_main_loop_quit(mainloop); } } static void stonith_cleanup(void) { fenced_cib_cleanup(); if (ipcs) { qb_ipcs_destroy(ipcs); } pcmk__cluster_destroy_node_caches(); pcmk__client_cleanup(); free_stonith_remote_op_list(); free_topology_list(); free_device_list(); free_metadata_cache(); fenced_unregister_handlers(); } static gboolean stand_alone_cpg_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error) { stand_alone = FALSE; options.no_cib_connect = true; return TRUE; } struct qb_ipcs_service_handlers ipc_callbacks = { .connection_accept = st_ipc_accept, .connection_created = NULL, .msg_process = st_ipc_dispatch, .connection_closed = st_ipc_closed, .connection_destroyed = st_ipc_destroy }; /*! * \internal * \brief Callback for peer status changes * * \param[in] type What changed * \param[in] node What peer had the change * \param[in] data Previous value of what changed */ static void st_peer_update_callback(enum pcmk__node_update type, pcmk__node_status_t *node, const void *data) { if ((type != pcmk__node_update_processes) && !pcmk_is_set(node->flags, pcmk__node_status_remote)) { /* * This is a hack until we can send to a nodeid and/or we fix node name lookups * These messages are ignored in stonith_peer_callback() */ xmlNode *query = pcmk__xe_create(NULL, PCMK__XE_STONITH_COMMAND); crm_xml_add(query, PCMK__XA_T, PCMK__VALUE_STONITH_NG); crm_xml_add(query, PCMK__XA_ST_OP, STONITH_OP_POKE); crm_debug("Broadcasting our uname because of node %" PRIu32, node->cluster_layer_id); pcmk__cluster_send_message(NULL, pcmk_ipc_fenced, query); pcmk__xml_free(query); } } /* @COMPAT Deprecated since 2.1.8. Use pcmk_list_fence_attrs() or * crm_resource --list-options=fencing instead of querying daemon metadata. + * + * NOTE: pcs (as of at least 0.11.8) uses this */ static int fencer_metadata(void) { const char *name = "pacemaker-fenced"; const char *desc_short = N_("Instance attributes available for all " "\"stonith\"-class resources"); const char *desc_long = N_("Instance attributes available for all " "\"stonith\"-class resources and used by " "Pacemaker's fence daemon"); return pcmk__daemon_metadata(out, name, desc_short, desc_long, pcmk__opt_fencing); } static GOptionEntry entries[] = { { "stand-alone", 's', G_OPTION_FLAG_NONE, G_OPTION_ARG_NONE, &stand_alone, N_("Deprecated (will be removed in a future release)"), NULL }, { "stand-alone-w-cpg", 'c', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, stand_alone_cpg_cb, N_("Intended for use in regression testing only"), NULL }, { "logfile", 'l', G_OPTION_FLAG_NONE, G_OPTION_ARG_FILENAME_ARRAY, &options.log_files, N_("Send logs to the additional named logfile"), NULL }, { NULL } }; static GOptionContext * build_arg_context(pcmk__common_args_t *args, GOptionGroup **group) { GOptionContext *context = NULL; context = pcmk__build_arg_context(args, "text (default), xml", group, NULL); pcmk__add_main_args(context, entries); return context; } int main(int argc, char **argv) { int rc = pcmk_rc_ok; pcmk_cluster_t *cluster = NULL; crm_ipc_t *old_instance = NULL; GError *error = NULL; GOptionGroup *output_group = NULL; pcmk__common_args_t *args = pcmk__new_common_args(SUMMARY); gchar **processed_args = pcmk__cmdline_preproc(argv, "l"); GOptionContext *context = build_arg_context(args, &output_group); crm_log_preinit(NULL, argc, argv); pcmk__register_formats(output_group, formats); if (!g_option_context_parse_strv(context, &processed_args, &error)) { exit_code = CRM_EX_USAGE; goto done; } rc = pcmk__output_new(&out, args->output_ty, args->output_dest, argv); if (rc != pcmk_rc_ok) { exit_code = CRM_EX_ERROR; g_set_error(&error, PCMK__EXITC_ERROR, exit_code, "Error creating output format %s: %s", args->output_ty, pcmk_rc_str(rc)); goto done; } if (args->version) { out->version(out, false); goto done; } if ((g_strv_length(processed_args) >= 2) && pcmk__str_eq(processed_args[1], "metadata", pcmk__str_none)) { rc = fencer_metadata(); if (rc != pcmk_rc_ok) { exit_code = CRM_EX_FATAL; g_set_error(&error, PCMK__EXITC_ERROR, exit_code, "Unable to display metadata: %s", pcmk_rc_str(rc)); } goto done; } // Open additional log files pcmk__add_logfiles(options.log_files, out); crm_log_init(NULL, LOG_INFO + args->verbosity, TRUE, (args->verbosity > 0), argc, argv, FALSE); crm_notice("Starting Pacemaker fencer"); old_instance = crm_ipc_new("stonith-ng", 0); if (old_instance == NULL) { /* crm_ipc_new() will have already logged an error message with * crm_err() */ exit_code = CRM_EX_FATAL; goto done; } if (pcmk__connect_generic_ipc(old_instance) == pcmk_rc_ok) { // IPC endpoint already up crm_ipc_close(old_instance); crm_ipc_destroy(old_instance); crm_err("pacemaker-fenced is already active, aborting startup"); goto done; } else { // Not up or not authentic, we'll proceed either way crm_ipc_destroy(old_instance); old_instance = NULL; } mainloop_add_signal(SIGTERM, stonith_shutdown); pcmk__cluster_init_node_caches(); rc = fenced_scheduler_init(); if (rc != pcmk_rc_ok) { exit_code = CRM_EX_FATAL; g_set_error(&error, PCMK__EXITC_ERROR, exit_code, "Error initializing scheduler data: %s", pcmk_rc_str(rc)); goto done; } cluster = pcmk_cluster_new(); if (!stand_alone) { #if SUPPORT_COROSYNC if (pcmk_get_cluster_layer() == pcmk_cluster_layer_corosync) { pcmk_cluster_set_destroy_fn(cluster, stonith_peer_cs_destroy); pcmk_cpg_set_deliver_fn(cluster, stonith_peer_ais_callback); pcmk_cpg_set_confchg_fn(cluster, pcmk__cpg_confchg_cb); } #endif // SUPPORT_COROSYNC pcmk__cluster_set_status_callback(&st_peer_update_callback); if (pcmk_cluster_connect(cluster) != pcmk_rc_ok) { exit_code = CRM_EX_FATAL; crm_crit("Cannot sign in to the cluster... terminating"); goto done; } fenced_set_local_node(cluster->priv->node_name); if (!options.no_cib_connect) { setup_cib(); } } else { fenced_set_local_node("localhost"); crm_warn("Stand-alone mode is deprecated and will be removed " "in a future release"); } init_device_list(); init_topology_list(); pcmk__serve_fenced_ipc(&ipcs, &ipc_callbacks); // Create the mainloop and run it... mainloop = g_main_loop_new(NULL, FALSE); crm_notice("Pacemaker fencer successfully started and accepting connections"); g_main_loop_run(mainloop); done: g_strfreev(processed_args); pcmk__free_arg_context(context); g_strfreev(options.log_files); stonith_cleanup(); pcmk_cluster_free(cluster); fenced_scheduler_cleanup(); pcmk__output_and_clear_error(&error, out); if (out != NULL) { out->finish(out, exit_code, true, NULL); pcmk__output_free(out); } pcmk__unregister_formats(); crm_exit(exit_code); } diff --git a/daemons/schedulerd/pacemaker-schedulerd.c b/daemons/schedulerd/pacemaker-schedulerd.c index d1b93625f4..3dda91cefc 100644 --- a/daemons/schedulerd/pacemaker-schedulerd.c +++ b/daemons/schedulerd/pacemaker-schedulerd.c @@ -1,199 +1,201 @@ /* * 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 "pacemaker-schedulerd.h" #define SUMMARY "pacemaker-schedulerd - daemon for calculating a Pacemaker cluster's response to events" struct { gchar **remainder; } options; pcmk__output_t *logger_out = NULL; static pcmk__output_t *out = NULL; static GMainLoop *mainloop = NULL; static qb_ipcs_service_t *ipcs = NULL; static crm_exit_t exit_code = CRM_EX_OK; pcmk__supported_format_t formats[] = { PCMK__SUPPORTED_FORMAT_NONE, PCMK__SUPPORTED_FORMAT_TEXT, PCMK__SUPPORTED_FORMAT_XML, { NULL, NULL, NULL } }; void pengine_shutdown(int nsig); /* @COMPAT Deprecated since 2.1.8. Use pcmk_list_cluster_options() or * crm_attribute --list-options=cluster instead of querying daemon metadata. + * + * NOTE: pcs (as of at least 0.11.8) uses this */ static int scheduler_metadata(pcmk__output_t *out) { return pcmk__daemon_metadata(out, "pacemaker-schedulerd", "Pacemaker scheduler options", "Cluster options used by Pacemaker's " "scheduler", pcmk__opt_schedulerd); } static GOptionContext * build_arg_context(pcmk__common_args_t *args, GOptionGroup **group) { GOptionContext *context = NULL; GOptionEntry extra_prog_entries[] = { { G_OPTION_REMAINING, 0, G_OPTION_FLAG_NONE, G_OPTION_ARG_STRING_ARRAY, &options.remainder, NULL, NULL }, { NULL } }; context = pcmk__build_arg_context(args, "text (default), xml", group, NULL); pcmk__add_main_args(context, extra_prog_entries); return context; } int main(int argc, char **argv) { GError *error = NULL; 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, NULL); GOptionContext *context = build_arg_context(args, &output_group); crm_log_preinit(NULL, argc, argv); mainloop_add_signal(SIGTERM, pengine_shutdown); pcmk__register_formats(output_group, formats); if (!g_option_context_parse_strv(context, &processed_args, &error)) { exit_code = CRM_EX_USAGE; goto done; } rc = pcmk__output_new(&out, args->output_ty, args->output_dest, argv); if ((rc != pcmk_rc_ok) || (out == NULL)) { exit_code = CRM_EX_FATAL; g_set_error(&error, PCMK__EXITC_ERROR, exit_code, "Error creating output format %s: %s", args->output_ty, pcmk_rc_str(rc)); goto done; } pe__register_messages(out); pcmk__register_lib_messages(out); if (options.remainder) { if (g_strv_length(options.remainder) == 1 && pcmk__str_eq("metadata", options.remainder[0], pcmk__str_casei)) { rc = scheduler_metadata(out); if (rc != pcmk_rc_ok) { exit_code = CRM_EX_FATAL; g_set_error(&error, PCMK__EXITC_ERROR, exit_code, "Unable to display metadata: %s", pcmk_rc_str(rc)); } } else { exit_code = CRM_EX_USAGE; g_set_error(&error, PCMK__EXITC_ERROR, exit_code, "Unsupported extra command line parameters"); } goto done; } if (args->version) { out->version(out, false); goto done; } pcmk__cli_init_logging("pacemaker-schedulerd", args->verbosity); crm_log_init(NULL, LOG_INFO, TRUE, FALSE, argc, argv, FALSE); crm_notice("Starting Pacemaker scheduler"); if (pcmk__daemon_can_write(PE_STATE_DIR, NULL) == FALSE) { crm_err("Terminating due to bad permissions on " PE_STATE_DIR); exit_code = CRM_EX_FATAL; g_set_error(&error, PCMK__EXITC_ERROR, exit_code, "ERROR: Bad permissions on %s (see logs for details)", PE_STATE_DIR); goto done; } ipcs = pcmk__serve_schedulerd_ipc(&ipc_callbacks); if (ipcs == NULL) { g_set_error(&error, PCMK__EXITC_ERROR, exit_code, "Failed to create pacemaker-schedulerd server: exiting and inhibiting respawn"); exit_code = CRM_EX_FATAL; goto done; } if (pcmk__log_output_new(&logger_out) != pcmk_rc_ok) { exit_code = CRM_EX_FATAL; goto done; } pe__register_messages(logger_out); pcmk__register_lib_messages(logger_out); pcmk__output_set_log_level(logger_out, LOG_TRACE); /* Create the mainloop and run it... */ mainloop = g_main_loop_new(NULL, FALSE); crm_notice("Pacemaker scheduler successfully started and accepting connections"); g_main_loop_run(mainloop); done: g_strfreev(options.remainder); g_strfreev(processed_args); pcmk__free_arg_context(context); pcmk__output_and_clear_error(&error, out); pengine_shutdown(0); } void pengine_shutdown(int nsig) { if (ipcs != NULL) { crm_trace("Closing IPC server"); mainloop_del_ipc_server(ipcs); ipcs = NULL; } if (logger_out != NULL) { logger_out->finish(logger_out, exit_code, true, NULL); pcmk__output_free(logger_out); logger_out = NULL; } if (out != NULL) { out->finish(out, exit_code, true, NULL); pcmk__output_free(out); out = NULL; } pcmk__unregister_formats(); crm_exit(exit_code); } diff --git a/include/crm/cluster/internal.h b/include/crm/cluster/internal.h index f708376c6e..bb882f2252 100644 --- a/include/crm/cluster/internal.h +++ b/include/crm/cluster/internal.h @@ -1,322 +1,319 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #ifndef PCMK__CRM_CLUSTER_INTERNAL__H #define PCMK__CRM_CLUSTER_INTERNAL__H #include #include // uint32_t, uint64_t #include // gboolean #include // xmlNode #include // enum crm_ipc_server #include #if SUPPORT_COROSYNC #include // cpg_name, cpg_handle_t #endif #ifdef __cplusplus extern "C" { #endif +// @TODO Replace this with a pcmk__node_status_flags value enum crm_proc_flag { - /* @COMPAT When pcmk__node_status_t:processes is made internal, we can merge - * this into node flags or turn it into a boolean. Until then, in theory - * something could depend on these particular numeric values. - */ crm_proc_none = 0x00000001, // Cluster layers crm_proc_cpg = 0x04000000, }; /*! * \internal * \enum pcmk__node_status_flags * \brief Boolean flags for a \c pcmk__node_status_t object * * Some flags may not be related to status specifically. However, we keep these * separate from enum pcmk__node_flags because they're used with * different object types. */ enum pcmk__node_status_flags { /*! * Node is a Pacemaker Remote node and should not be considered for cluster * membership */ pcmk__node_status_remote = (UINT32_C(1) << 0), //! Node's cache entry is dirty pcmk__node_status_dirty = (UINT32_C(1) << 1), }; // Used with node cache search functions enum pcmk__node_search_flags { //! Does not affect search pcmk__node_search_none = 0, //! Search for cluster nodes from membership cache pcmk__node_search_cluster_member = (1 << 0), //! Search for remote nodes pcmk__node_search_remote = (1 << 1), //! Search for cluster member nodes and remote nodes pcmk__node_search_any = pcmk__node_search_cluster_member |pcmk__node_search_remote, //! Search for cluster nodes from CIB (as of last cache refresh) pcmk__node_search_cluster_cib = (1 << 2), }; /*! * \internal * \enum pcmk__node_update * \brief Type of update to a \c pcmk__node_status_t object */ enum pcmk__node_update { pcmk__node_update_name, //!< Node name updated pcmk__node_update_state, //!< Node connection state updated pcmk__node_update_processes, //!< Node process group membership updated }; //! Implementation of pcmk__cluster_private_t struct pcmk__cluster_private { // @TODO Drop and replace with per-daemon cluster-layer ID global variables? uint32_t node_id; //!< Local node ID at cluster layer // @TODO Drop and replace with per-daemon node name global variables? char *node_name; //!< Local node name at cluster layer #if SUPPORT_COROSYNC /* @TODO Make these members a separate struct and use void *cluster_data * here instead, to abstract the cluster layer further. */ struct cpg_name group; //!< Corosync CPG name cpg_handle_t cpg_handle; //!< Corosync CPG handle #endif // SUPPORT_COROSYNC }; //! Node status data (may be a cluster node or a Pacemaker Remote node) typedef struct pcmk__node_status { //! Node name as known to cluster layer, or Pacemaker Remote node name char *name; /* @COMPAT This is less than ideal since the value is not a valid XML ID * (for Corosync, it's the string equivalent of the node's numeric node ID, * but XML IDs can't start with a number) and the three elements should have * different IDs. * * Ideally, we would use something like node-NODEID, node_state-NODEID, and * transient_attributes-NODEID as the element IDs. Unfortunately changing it * would be impractical due to backward compatibility; older nodes in a * rolling upgrade will always write and expect the value in the old format. */ /*! * Value of the PCMK_XA_ID XML attribute to use with the node's * PCMK_XE_NODE, PCMK_XE_NODE_STATE, and PCMK_XE_TRANSIENT_ATTRIBUTES * XML elements in the CIB */ char *xml_id; char *state; // @TODO change to enum //! Group of enum pcmk__node_status_flags uint32_t flags; /*! * Most recent cluster membership in which node was seen (0 for Pacemaker * Remote nodes) */ uint64_t membership_id; uint32_t processes; // @TODO most not needed, merge into flags /* @TODO When we can break public API compatibility, we can make the rest of * these members separate structs and use void *cluster_data and * void *user_data here instead, to abstract the cluster layer further. */ //! Arbitrary data (must be freeable by \c free()) void *user_data; char *expected; time_t peer_lost; char *conn_host; time_t when_member; // Since when node has been a cluster member time_t when_online; // Since when peer has been online in CPG /* @TODO The following are currently needed only by the Corosync stack. * Eventually consider moving them to a cluster-layer-specific data object. */ uint32_t cluster_layer_id; //!< Cluster-layer numeric node ID time_t when_lost; //!< When CPG membership was last lost } pcmk__node_status_t; /*! * \internal * \brief Return the process bit corresponding to the current cluster stack * * \return Process flag if detectable, otherwise 0 */ static inline uint32_t crm_get_cluster_proc(void) { switch (pcmk_get_cluster_layer()) { case pcmk_cluster_layer_corosync: return crm_proc_cpg; default: break; } return crm_proc_none; } /*! * \internal * \brief Get log-friendly string description of a Corosync return code * * \param[in] error Corosync return code * * \return Log-friendly string description corresponding to \p error */ static inline const char * pcmk__cs_err_str(int error) { # if SUPPORT_COROSYNC switch (error) { case CS_OK: return "OK"; case CS_ERR_LIBRARY: return "Library error"; case CS_ERR_VERSION: return "Version error"; case CS_ERR_INIT: return "Initialization error"; case CS_ERR_TIMEOUT: return "Timeout"; case CS_ERR_TRY_AGAIN: return "Try again"; case CS_ERR_INVALID_PARAM: return "Invalid parameter"; case CS_ERR_NO_MEMORY: return "No memory"; case CS_ERR_BAD_HANDLE: return "Bad handle"; case CS_ERR_BUSY: return "Busy"; case CS_ERR_ACCESS: return "Access error"; case CS_ERR_NOT_EXIST: return "Doesn't exist"; case CS_ERR_NAME_TOO_LONG: return "Name too long"; case CS_ERR_EXIST: return "Exists"; case CS_ERR_NO_SPACE: return "No space"; case CS_ERR_INTERRUPT: return "Interrupt"; case CS_ERR_NAME_NOT_FOUND: return "Name not found"; case CS_ERR_NO_RESOURCES: return "No resources"; case CS_ERR_NOT_SUPPORTED: return "Not supported"; case CS_ERR_BAD_OPERATION: return "Bad operation"; case CS_ERR_FAILED_OPERATION: return "Failed operation"; case CS_ERR_MESSAGE_ERROR: return "Message error"; case CS_ERR_QUEUE_FULL: return "Queue full"; case CS_ERR_QUEUE_NOT_AVAILABLE: return "Queue not available"; case CS_ERR_BAD_FLAGS: return "Bad flags"; case CS_ERR_TOO_BIG: return "Too big"; case CS_ERR_NO_SECTIONS: return "No sections"; } # endif return "Corosync error"; } # if SUPPORT_COROSYNC #if 0 /* This is the new way to do it, but we still support all Corosync 2 versions, * and this isn't always available. A better alternative here would be to check * for support in the configure script and enable this conditionally. */ #define pcmk__init_cmap(handle) cmap_initialize_map((handle), CMAP_MAP_ICMAP) #else #define pcmk__init_cmap(handle) cmap_initialize(handle) #endif char *pcmk__corosync_cluster_name(void); bool pcmk__corosync_add_nodes(xmlNode *xml_parent); void pcmk__cpg_confchg_cb(cpg_handle_t handle, const struct cpg_name *group_name, const struct cpg_address *member_list, size_t member_list_entries, const struct cpg_address *left_list, size_t left_list_entries, const struct cpg_address *joined_list, size_t joined_list_entries); char *pcmk__cpg_message_data(cpg_handle_t handle, uint32_t sender_id, uint32_t pid, void *content, const char **from); # endif const char *pcmk__cluster_node_uuid(pcmk__node_status_t *node); char *pcmk__cluster_node_name(uint32_t nodeid); const char *pcmk__cluster_local_node_name(void); const char *pcmk__node_name_from_uuid(const char *uuid); pcmk__node_status_t *crm_update_peer_proc(const char *source, pcmk__node_status_t *peer, uint32_t flag, const char *status); pcmk__node_status_t *pcmk__update_peer_state(const char *source, pcmk__node_status_t *node, const char *state, uint64_t membership); void pcmk__update_peer_expected(const char *source, pcmk__node_status_t *node, const char *expected); void pcmk__reap_unseen_nodes(uint64_t ring_id); void pcmk__corosync_quorum_connect(gboolean (*dispatch)(unsigned long long, gboolean), void (*destroy) (gpointer)); bool pcmk__cluster_send_message(const pcmk__node_status_t *node, enum pcmk_ipc_server service, const xmlNode *data); // Membership extern GHashTable *pcmk__peer_cache; extern GHashTable *pcmk__remote_peer_cache; bool pcmk__cluster_has_quorum(void); void pcmk__cluster_init_node_caches(void); void pcmk__cluster_destroy_node_caches(void); void pcmk__cluster_set_autoreap(bool enable); void pcmk__cluster_set_status_callback(void (*dispatch)(enum pcmk__node_update, pcmk__node_status_t *, const void *)); bool pcmk__cluster_is_node_active(const pcmk__node_status_t *node); unsigned int pcmk__cluster_num_active_nodes(void); unsigned int pcmk__cluster_num_remote_nodes(void); pcmk__node_status_t *pcmk__cluster_lookup_remote_node(const char *node_name); void pcmk__cluster_forget_cluster_node(uint32_t id, const char *node_name); void pcmk__cluster_forget_remote_node(const char *node_name); pcmk__node_status_t *pcmk__search_node_caches(unsigned int id, const char *uname, uint32_t flags); void pcmk__purge_node_from_cache(const char *node_name, uint32_t node_id); void pcmk__refresh_node_caches_from_cib(xmlNode *cib); pcmk__node_status_t *pcmk__get_node(unsigned int id, const char *uname, const char *uuid, uint32_t flags); #ifdef __cplusplus } #endif #endif // PCMK__CRM_CLUSTER_INTERNAL__H diff --git a/tools/crm_mon.c b/tools/crm_mon.c index 06689db0c0..117934f80e 100644 --- a/tools/crm_mon.c +++ b/tools/crm_mon.c @@ -1,2195 +1,2188 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include // pcmk__ends_with_ext() #include #include #include #include #include #include #include #include #include #include #include #include #include #include // stonith__* #include "crm_mon.h" #define SUMMARY "Provides a summary of cluster's current state.\n\n" \ "Outputs varying levels of detail in a number of different formats." /* * Definitions indicating which items to print */ static uint32_t show; static uint32_t show_opts = pcmk_show_pending; /* * Definitions indicating how to output */ static mon_output_format_t output_format = mon_output_unset; /* other globals */ static GIOChannel *io_channel = NULL; static GMainLoop *mainloop = NULL; static guint reconnect_timer = 0; static mainloop_timer_t *refresh_timer = NULL; static enum pcmk_pacemakerd_state pcmkd_state = pcmk_pacemakerd_state_invalid; static cib_t *cib = NULL; static stonith_t *st = NULL; static xmlNode *current_cib = NULL; static GError *error = NULL; static pcmk__common_args_t *args = NULL; static pcmk__output_t *out = NULL; static GOptionContext *context = NULL; static gchar **processed_args = NULL; static time_t last_refresh = 0; volatile crm_trigger_t *refresh_trigger = NULL; static pcmk_scheduler_t *scheduler = NULL; static enum pcmk__fence_history fence_history = pcmk__fence_history_none; int interactive_fence_level = 0; static pcmk__supported_format_t formats[] = { #if CURSES_ENABLED CRM_MON_SUPPORTED_FORMAT_CURSES, #endif PCMK__SUPPORTED_FORMAT_HTML, PCMK__SUPPORTED_FORMAT_NONE, PCMK__SUPPORTED_FORMAT_TEXT, PCMK__SUPPORTED_FORMAT_XML, { NULL, NULL, NULL } }; PCMK__OUTPUT_ARGS("crm-mon-disconnected", "const char *", "enum pcmk_pacemakerd_state") static int crm_mon_disconnected_default(pcmk__output_t *out, va_list args) { return pcmk_rc_no_output; } PCMK__OUTPUT_ARGS("crm-mon-disconnected", "const char *", "enum pcmk_pacemakerd_state") static int crm_mon_disconnected_html(pcmk__output_t *out, va_list args) { const char *desc = va_arg(args, const char *); enum pcmk_pacemakerd_state state = (enum pcmk_pacemakerd_state) va_arg(args, int); if (out->dest != stdout) { out->reset(out); } pcmk__output_create_xml_text_node(out, PCMK__XE_SPAN, "Not connected to CIB"); if (desc != NULL) { pcmk__output_create_xml_text_node(out, PCMK__XE_SPAN, ": "); pcmk__output_create_xml_text_node(out, PCMK__XE_SPAN, desc); } if (state != pcmk_pacemakerd_state_invalid) { const char *state_s = pcmk__pcmkd_state_enum2friendly(state); pcmk__output_create_xml_text_node(out, PCMK__XE_SPAN, " ("); pcmk__output_create_xml_text_node(out, PCMK__XE_SPAN, state_s); pcmk__output_create_xml_text_node(out, PCMK__XE_SPAN, ")"); } out->finish(out, CRM_EX_DISCONNECT, true, NULL); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("crm-mon-disconnected", "const char *", "enum pcmk_pacemakerd_state") static int crm_mon_disconnected_text(pcmk__output_t *out, va_list args) { const char *desc = va_arg(args, const char *); enum pcmk_pacemakerd_state state = (enum pcmk_pacemakerd_state) va_arg(args, int); int rc = pcmk_rc_ok; if (out->dest != stdout) { out->reset(out); } if (state != pcmk_pacemakerd_state_invalid) { rc = out->info(out, "Not connected to CIB%s%s (%s)", (desc != NULL)? ": " : "", pcmk__s(desc, ""), pcmk__pcmkd_state_enum2friendly(state)); } else { rc = out->info(out, "Not connected to CIB%s%s", (desc != NULL)? ": " : "", pcmk__s(desc, "")); } out->finish(out, CRM_EX_DISCONNECT, true, NULL); return rc; } PCMK__OUTPUT_ARGS("crm-mon-disconnected", "const char *", "enum pcmk_pacemakerd_state") static int crm_mon_disconnected_xml(pcmk__output_t *out, va_list args) { const char *desc = va_arg(args, const char *); enum pcmk_pacemakerd_state state = (enum pcmk_pacemakerd_state) va_arg(args, int); const char *state_s = NULL; if (out->dest != stdout) { out->reset(out); } if (state != pcmk_pacemakerd_state_invalid) { state_s = pcmk_pacemakerd_api_daemon_state_enum2text(state); } pcmk__output_create_xml_node(out, PCMK_XE_CRM_MON_DISCONNECTED, PCMK_XA_DESCRIPTION, desc, PCMK_XA_PACEMAKERD_STATE, state_s, NULL); out->finish(out, CRM_EX_DISCONNECT, true, NULL); return pcmk_rc_ok; } static pcmk__message_entry_t fmt_functions[] = { { "crm-mon-disconnected", "default", crm_mon_disconnected_default }, { "crm-mon-disconnected", "html", crm_mon_disconnected_html }, { "crm-mon-disconnected", "text", crm_mon_disconnected_text }, { "crm-mon-disconnected", "xml", crm_mon_disconnected_xml }, { NULL, NULL, NULL }, }; #define RECONNECT_MSECS 5000 struct { guint reconnect_ms; enum mon_exec_mode exec_mode; gboolean fence_connect; gboolean print_pending; gboolean show_bans; gboolean watch_fencing; char *pid_file; char *external_agent; char *external_recipient; char *neg_location_prefix; char *only_node; char *only_rsc; GSList *user_includes_excludes; GSList *includes_excludes; } options = { .reconnect_ms = RECONNECT_MSECS, .exec_mode = mon_exec_unset, .fence_connect = TRUE, }; static crm_exit_t clean_up(crm_exit_t exit_code); static void crm_diff_update(const char *event, xmlNode * msg); static void clean_up_on_connection_failure(int rc); static int mon_refresh_display(gpointer user_data); static int setup_cib_connection(void); static int setup_fencer_connection(void); static int setup_api_connections(void); static void mon_st_callback_event(stonith_t * st, stonith_event_t * e); static void mon_st_callback_display(stonith_t * st, stonith_event_t * e); static void refresh_after_event(gboolean data_updated, gboolean enforce); static uint32_t all_includes(mon_output_format_t fmt) { if ((fmt == mon_output_plain) || (fmt == mon_output_console)) { return ~pcmk_section_options; } else { return pcmk_section_all; } } static uint32_t default_includes(mon_output_format_t fmt) { switch (fmt) { case mon_output_plain: case mon_output_console: case mon_output_html: return pcmk_section_summary |pcmk_section_nodes |pcmk_section_resources |pcmk_section_failures; case mon_output_xml: return all_includes(fmt); default: return 0; } } struct { const char *name; uint32_t bit; } sections[] = { { "attributes", pcmk_section_attributes }, { "bans", pcmk_section_bans }, { "counts", pcmk_section_counts }, { "dc", pcmk_section_dc }, { "failcounts", pcmk_section_failcounts }, { "failures", pcmk_section_failures }, { PCMK_VALUE_FENCING, pcmk_section_fencing_all }, { "fencing-failed", pcmk_section_fence_failed }, { "fencing-pending", pcmk_section_fence_pending }, { "fencing-succeeded", pcmk_section_fence_worked }, { "maint-mode", pcmk_section_maint_mode }, { "nodes", pcmk_section_nodes }, { "operations", pcmk_section_operations }, { "options", pcmk_section_options }, { "resources", pcmk_section_resources }, { "stack", pcmk_section_stack }, { "summary", pcmk_section_summary }, { "tickets", pcmk_section_tickets }, { "times", pcmk_section_times }, { NULL } }; static uint32_t find_section_bit(const char *name) { for (int i = 0; sections[i].name != NULL; i++) { if (pcmk__str_eq(sections[i].name, name, pcmk__str_casei)) { return sections[i].bit; } } return 0; } static gboolean apply_exclude(const gchar *excludes, GError **error) { char **parts = NULL; gboolean result = TRUE; parts = g_strsplit(excludes, ",", 0); for (char **s = parts; *s != NULL; s++) { uint32_t bit = find_section_bit(*s); if (pcmk__str_eq(*s, "all", pcmk__str_none)) { show = 0; } else if (pcmk__str_eq(*s, PCMK_VALUE_NONE, pcmk__str_none)) { show = all_includes(output_format); } else if (bit != 0) { show &= ~bit; } else { g_set_error(error, PCMK__EXITC_ERROR, CRM_EX_USAGE, "--exclude options: all, attributes, bans, counts, dc, " "failcounts, failures, fencing, fencing-failed, " "fencing-pending, fencing-succeeded, maint-mode, nodes, " PCMK_VALUE_NONE ", operations, options, resources, " "stack, summary, tickets, times"); result = FALSE; break; } } g_strfreev(parts); return result; } static gboolean apply_include(const gchar *includes, GError **error) { char **parts = NULL; gboolean result = TRUE; parts = g_strsplit(includes, ",", 0); for (char **s = parts; *s != NULL; s++) { uint32_t bit = find_section_bit(*s); if (pcmk__str_eq(*s, "all", pcmk__str_none)) { show = all_includes(output_format); } else if (pcmk__starts_with(*s, "bans")) { show |= pcmk_section_bans; if (options.neg_location_prefix != NULL) { free(options.neg_location_prefix); options.neg_location_prefix = NULL; } if (strlen(*s) > 4 && (*s)[4] == ':') { options.neg_location_prefix = strdup(*s+5); } } else if (pcmk__str_any_of(*s, PCMK_VALUE_DEFAULT, "defaults", NULL)) { show |= default_includes(output_format); } else if (pcmk__str_eq(*s, PCMK_VALUE_NONE, pcmk__str_none)) { show = 0; } else if (bit != 0) { show |= bit; } else { g_set_error(error, PCMK__EXITC_ERROR, CRM_EX_USAGE, "--include options: all, attributes, bans[:PREFIX], counts, dc, " PCMK_VALUE_DEFAULT ", failcounts, failures, fencing, " "fencing-failed, fencing-pending, fencing-succeeded, " "maint-mode, nodes, " PCMK_VALUE_NONE ", operations, " "options, resources, stack, summary, tickets, times"); result = FALSE; break; } } g_strfreev(parts); return result; } static gboolean apply_include_exclude(GSList *lst, GError **error) { gboolean rc = TRUE; GSList *node = lst; while (node != NULL) { char *s = node->data; if (pcmk__starts_with(s, "--include=")) { rc = apply_include(s+10, error); } else if (pcmk__starts_with(s, "-I=")) { rc = apply_include(s+3, error); } else if (pcmk__starts_with(s, "--exclude=")) { rc = apply_exclude(s+10, error); } else if (pcmk__starts_with(s, "-U=")) { rc = apply_exclude(s+3, error); } if (rc != TRUE) { break; } node = node->next; } return rc; } static gboolean user_include_exclude_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **err) { char *s = crm_strdup_printf("%s=%s", option_name, optarg); options.user_includes_excludes = g_slist_append(options.user_includes_excludes, s); return TRUE; } static gboolean include_exclude_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **err) { char *s = crm_strdup_printf("%s=%s", option_name, optarg); options.includes_excludes = g_slist_append(options.includes_excludes, s); return TRUE; } static gboolean as_xml_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **err) { pcmk__str_update(&args->output_ty, "xml"); output_format = mon_output_legacy_xml; return TRUE; } static gboolean fence_history_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **err) { if (optarg == NULL) { interactive_fence_level = 2; } else { pcmk__scan_min_int(optarg, &interactive_fence_level, 0); } switch (interactive_fence_level) { case 3: options.fence_connect = TRUE; fence_history = pcmk__fence_history_full; return include_exclude_cb("--include", PCMK_VALUE_FENCING, data, err); case 2: options.fence_connect = TRUE; fence_history = pcmk__fence_history_full; return include_exclude_cb("--include", PCMK_VALUE_FENCING, data, err); case 1: options.fence_connect = TRUE; fence_history = pcmk__fence_history_full; return include_exclude_cb("--include", "fencing-failed,fencing-pending", data, err); case 0: options.fence_connect = FALSE; fence_history = pcmk__fence_history_none; return include_exclude_cb("--exclude", PCMK_VALUE_FENCING, data, err); default: g_set_error(err, PCMK__EXITC_ERROR, CRM_EX_INVALID_PARAM, "Fence history must be 0-3"); return FALSE; } } static gboolean group_by_node_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **err) { show_opts |= pcmk_show_rscs_by_node; return TRUE; } static gboolean hide_headers_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **err) { return user_include_exclude_cb("--exclude", "summary", data, err); } static gboolean inactive_resources_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **err) { show_opts |= pcmk_show_inactive_rscs; return TRUE; } static gboolean print_brief_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **err) { show_opts |= pcmk_show_brief; return TRUE; } static gboolean print_detail_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **err) { show_opts |= pcmk_show_details; return TRUE; } static gboolean print_description_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **err) { show_opts |= pcmk_show_description; return TRUE; } static gboolean print_timing_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **err) { show_opts |= pcmk_show_timing; return user_include_exclude_cb("--include", "operations", data, err); } static gboolean reconnect_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **err) { int rc = crm_get_msec(optarg); if (rc == -1) { g_set_error(err, PCMK__EXITC_ERROR, CRM_EX_INVALID_PARAM, "Invalid value for -i: %s", optarg); return FALSE; } else { pcmk_parse_interval_spec(optarg, &options.reconnect_ms); if (options.exec_mode != mon_exec_daemonized) { // Reconnect interval applies to daemonized too, so don't override options.exec_mode = mon_exec_update; } } return TRUE; } /*! * \internal * \brief Enable one-shot mode * * \param[in] option_name Name of option being parsed (ignored) * \param[in] optarg Value to be parsed (ignored) * \param[in] data User data (ignored) * \param[out] err Where to store error (ignored) */ static gboolean one_shot_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **err) { options.exec_mode = mon_exec_one_shot; return TRUE; } /*! * \internal * \brief Enable daemonized mode * * \param[in] option_name Name of option being parsed (ignored) * \param[in] optarg Value to be parsed (ignored) * \param[in] data User data (ignored) * \param[out] err Where to store error (ignored) */ static gboolean daemonize_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **err) { options.exec_mode = mon_exec_daemonized; return TRUE; } static gboolean show_attributes_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **err) { return user_include_exclude_cb("--include", "attributes", data, err); } static gboolean show_bans_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **err) { if (optarg != NULL) { char *s = crm_strdup_printf("bans:%s", optarg); gboolean rc = user_include_exclude_cb("--include", s, data, err); free(s); return rc; } else { return user_include_exclude_cb("--include", "bans", data, err); } } static gboolean show_failcounts_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **err) { return user_include_exclude_cb("--include", "failcounts", data, err); } static gboolean show_operations_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **err) { return user_include_exclude_cb("--include", "failcounts,operations", data, err); } static gboolean show_tickets_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **err) { return user_include_exclude_cb("--include", "tickets", data, err); } static gboolean use_cib_file_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **err) { setenv("CIB_file", optarg, 1); options.exec_mode = mon_exec_one_shot; return TRUE; } #define INDENT " " /* *INDENT-OFF* */ static GOptionEntry addl_entries[] = { { "interval", 'i', 0, G_OPTION_ARG_CALLBACK, reconnect_cb, "Update frequency (default is 5 seconds)", "TIMESPEC" }, { "one-shot", '1', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, one_shot_cb, "Display the cluster status once and exit", NULL }, { "daemonize", 'd', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, daemonize_cb, "Run in the background as a daemon.\n" INDENT "Requires at least one of --output-to and --external-agent.", NULL }, { "pid-file", 'p', 0, G_OPTION_ARG_FILENAME, &options.pid_file, "(Advanced) Daemon pid file location", "FILE" }, { "external-agent", 'E', 0, G_OPTION_ARG_FILENAME, &options.external_agent, "A program to run when resource operations take place", "FILE" }, { "external-recipient", 'e', 0, G_OPTION_ARG_STRING, &options.external_recipient, "A recipient for your program (assuming you want the program to send something to someone).", "RCPT" }, { "watch-fencing", 'W', 0, G_OPTION_ARG_NONE, &options.watch_fencing, "Listen for fencing events. For use with --external-agent.", NULL }, { "xml-file", 'x', G_OPTION_FLAG_HIDDEN, G_OPTION_ARG_CALLBACK, use_cib_file_cb, NULL, NULL }, { NULL } }; static GOptionEntry display_entries[] = { { "include", 'I', 0, G_OPTION_ARG_CALLBACK, user_include_exclude_cb, "A list of sections to include in the output.\n" INDENT "See `Output Control` help for more information.", "SECTION(s)" }, { "exclude", 'U', 0, G_OPTION_ARG_CALLBACK, user_include_exclude_cb, "A list of sections to exclude from the output.\n" INDENT "See `Output Control` help for more information.", "SECTION(s)" }, { "node", 0, 0, G_OPTION_ARG_STRING, &options.only_node, "When displaying information about nodes, show only what's related to the given\n" INDENT "node, or to all nodes tagged with the given tag", "NODE" }, { "resource", 0, 0, G_OPTION_ARG_STRING, &options.only_rsc, "When displaying information about resources, show only what's related to the given\n" INDENT "resource, or to all resources tagged with the given tag", "RSC" }, { "group-by-node", 'n', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, group_by_node_cb, "Group resources by node", NULL }, { "inactive", 'r', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, inactive_resources_cb, "Display inactive resources", NULL }, { "failcounts", 'f', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, show_failcounts_cb, "Display resource fail counts", NULL }, { "operations", 'o', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, show_operations_cb, "Display resource operation history", NULL }, { "timing-details", 't', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, print_timing_cb, "Display resource operation history with timing details", NULL }, { "tickets", 'c', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, show_tickets_cb, "Display cluster tickets", NULL }, { "fence-history", 'm', G_OPTION_FLAG_OPTIONAL_ARG, G_OPTION_ARG_CALLBACK, fence_history_cb, "Show fence history:\n" INDENT "0=off, 1=failures and pending (default without option),\n" INDENT "2=add successes (default without value for option),\n" INDENT "3=show full history without reduction to most recent of each flavor", "LEVEL" }, { "neg-locations", 'L', G_OPTION_FLAG_OPTIONAL_ARG, G_OPTION_ARG_CALLBACK, show_bans_cb, "Display negative location constraints [optionally filtered by id prefix]", NULL }, { "show-node-attributes", 'A', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, show_attributes_cb, "Display node attributes", NULL }, { "hide-headers", 'D', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, hide_headers_cb, "Hide all headers", NULL }, { "show-detail", 'R', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, print_detail_cb, "Show more details (node IDs, individual clone instances)", NULL }, { "show-description", 0, G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, print_description_cb, "Show resource descriptions", NULL }, { "brief", 'b', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, print_brief_cb, "Brief output", NULL }, { "pending", 'j', G_OPTION_FLAG_HIDDEN, G_OPTION_ARG_NONE, &options.print_pending, "Display pending state if '" PCMK_META_RECORD_PENDING "' is enabled", NULL }, { NULL } }; static GOptionEntry deprecated_entries[] = { - /* @COMPAT At least the following resource agents use --as-xml at time of - * writing: - * * galera - * * pgsql - * * rabbitmq-cluster - * * redis - * - * Removing this option must wait until those are updated and we no longer - * support building on any platforms that ship the older agents in stock - * packages. + /* @COMPAT resource-agents <4.15.0 uses --as-xml, so removing this option + * must wait until we no longer support building on any platforms that ship + * the older agents. */ { "as-xml", 'X', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, as_xml_cb, "Write cluster status as XML to stdout. This will enable one-shot mode.\n" INDENT "Use --output-as=xml instead.", NULL }, { NULL } }; /* *INDENT-ON* */ /* Reconnect to the CIB and fencing agent after reconnect_ms has passed. This sounds * like it would be more broadly useful, but only ever happens after a disconnect via * mon_cib_connection_destroy. */ static gboolean reconnect_after_timeout(gpointer data) { #if CURSES_ENABLED if (output_format == mon_output_console) { clear(); refresh(); } #endif out->transient(out, "Reconnecting..."); if (setup_api_connections() == pcmk_rc_ok) { // Trigger redrawing the screen (needs reconnect_timer == 0) reconnect_timer = 0; refresh_after_event(FALSE, TRUE); return G_SOURCE_REMOVE; } out->message(out, "crm-mon-disconnected", "Latest connection attempt failed", pcmkd_state); reconnect_timer = g_timeout_add(options.reconnect_ms, reconnect_after_timeout, NULL); return G_SOURCE_REMOVE; } /* Called from various places when we are disconnected from the CIB or from the * fencing agent. If the CIB connection is still valid, this function will also * attempt to sign off and reconnect. */ static void mon_cib_connection_destroy(gpointer user_data) { const char *msg = "Connection to the cluster lost"; pcmkd_state = pcmk_pacemakerd_state_invalid; /* No crm-mon-disconnected message for console; a working implementation * is not currently worth the effort */ out->transient(out, "%s", msg); out->message(out, "crm-mon-disconnected", msg, pcmkd_state); if (refresh_timer != NULL) { /* we'll trigger a refresh after reconnect */ mainloop_timer_stop(refresh_timer); } if (reconnect_timer) { /* we'll trigger a new reconnect-timeout at the end */ g_source_remove(reconnect_timer); reconnect_timer = 0; } /* the client API won't properly reconnect notifications if they are still * in the table - so remove them */ if (st != NULL) { if (st->state != stonith_disconnected) { st->cmds->disconnect(st); } st->cmds->remove_notification(st, NULL); } if (cib) { cib->cmds->signoff(cib); reconnect_timer = g_timeout_add(options.reconnect_ms, reconnect_after_timeout, NULL); } } /* Signal handler installed into the mainloop for normal program shutdown */ static void mon_shutdown(int nsig) { clean_up(CRM_EX_OK); } #if CURSES_ENABLED static volatile sighandler_t ncurses_winch_handler; /* Signal handler installed the regular way (not into the main loop) for when * the screen is resized. Commonly, this happens when running in an xterm and * the user changes its size. */ static void mon_winresize(int nsig) { static int not_done; int lines = 0, cols = 0; if (!not_done++) { if (ncurses_winch_handler) /* the original ncurses WINCH signal handler does the * magic of retrieving the new window size; * otherwise, we'd have to use ioctl or tgetent */ (*ncurses_winch_handler) (SIGWINCH); getmaxyx(stdscr, lines, cols); resizeterm(lines, cols); /* Alert the mainloop code we'd like the refresh_trigger to run next * time the mainloop gets around to checking. */ mainloop_set_trigger((crm_trigger_t *) refresh_trigger); } not_done--; } #endif static int setup_fencer_connection(void) { int rc = pcmk_ok; if (options.fence_connect && st == NULL) { st = stonith_api_new(); } if (!options.fence_connect || st == NULL || st->state != stonith_disconnected) { return rc; } rc = st->cmds->connect(st, crm_system_name, NULL); if (rc == pcmk_ok) { crm_trace("Setting up stonith callbacks"); if (options.watch_fencing) { st->cmds->register_notification(st, PCMK__VALUE_ST_NOTIFY_DISCONNECT, mon_st_callback_event); st->cmds->register_notification(st, PCMK__VALUE_ST_NOTIFY_FENCE, mon_st_callback_event); } else { st->cmds->register_notification(st, PCMK__VALUE_ST_NOTIFY_DISCONNECT, mon_st_callback_display); st->cmds->register_notification(st, PCMK__VALUE_ST_NOTIFY_HISTORY, mon_st_callback_display); } } else { stonith_api_delete(st); st = NULL; } return rc; } static int setup_cib_connection(void) { int rc = pcmk_rc_ok; CRM_CHECK(cib != NULL, return EINVAL); if (cib->state != cib_disconnected) { // Already connected with notifications registered for CIB updates return rc; } rc = cib__signon_query(out, &cib, ¤t_cib); if (rc == pcmk_rc_ok) { rc = pcmk_legacy2rc(cib->cmds->set_connection_dnotify(cib, mon_cib_connection_destroy)); if (rc == EPROTONOSUPPORT) { out->err(out, "CIB client does not support connection loss " "notifications; crm_mon will be unable to reconnect after " "connection loss"); rc = pcmk_rc_ok; } if (rc == pcmk_rc_ok) { cib->cmds->del_notify_callback(cib, PCMK__VALUE_CIB_DIFF_NOTIFY, crm_diff_update); rc = cib->cmds->add_notify_callback(cib, PCMK__VALUE_CIB_DIFF_NOTIFY, crm_diff_update); rc = pcmk_legacy2rc(rc); } if (rc != pcmk_rc_ok) { if (rc == EPROTONOSUPPORT) { out->err(out, "CIB client does not support CIB diff " "notifications"); } else { out->err(out, "CIB diff notification setup failed"); } out->err(out, "Cannot monitor CIB changes; exiting"); cib__clean_up_connection(&cib); stonith_api_delete(st); st = NULL; } } return rc; } /* This is used to set up the fencing options after the interactive UI has been stared. * fence_history_cb can't be used because it builds up a list of includes/excludes that * then have to be processed with apply_include_exclude and that could affect other * things. */ static void set_fencing_options(int level) { switch (level) { case 3: options.fence_connect = TRUE; fence_history = pcmk__fence_history_full; show |= pcmk_section_fencing_all; break; case 2: options.fence_connect = TRUE; fence_history = pcmk__fence_history_full; show |= pcmk_section_fencing_all; break; case 1: options.fence_connect = TRUE; fence_history = pcmk__fence_history_full; show |= pcmk_section_fence_failed | pcmk_section_fence_pending; break; default: interactive_fence_level = 0; options.fence_connect = FALSE; fence_history = pcmk__fence_history_none; show &= ~pcmk_section_fencing_all; break; } } static int setup_api_connections(void) { int rc = pcmk_rc_ok; CRM_CHECK(cib != NULL, return EINVAL); if (cib->state != cib_disconnected) { return rc; } if (cib->variant == cib_native) { rc = pcmk__pacemakerd_status(out, crm_system_name, options.reconnect_ms / 2, false, &pcmkd_state); if (rc != pcmk_rc_ok) { return rc; } switch (pcmkd_state) { case pcmk_pacemakerd_state_running: case pcmk_pacemakerd_state_remote: case pcmk_pacemakerd_state_shutting_down: /* Fencer and CIB may still be available while shutting down or * running on a Pacemaker Remote node */ break; default: // Fencer and CIB are definitely unavailable return ENOTCONN; } setup_fencer_connection(); } rc = setup_cib_connection(); return rc; } #if CURSES_ENABLED static const char * get_option_desc(char c) { const char *desc = "No help available"; for (GOptionEntry *entry = display_entries; entry != NULL; entry++) { if (entry->short_name == c) { desc = entry->description; break; } } return desc; } #define print_option_help(out, option, condition) \ curses_formatted_printf(out, "%c %c: \t%s\n", ((condition)? '*': ' '), option, get_option_desc(option)); /* This function is called from the main loop when there is something to be read * on stdin, like an interactive user's keystroke. All it does is read the keystroke, * set flags (or show the page showing which keystrokes are valid), and redraw the * screen. It does not do anything with connections to the CIB or fencing agent * agent what would happen in mon_refresh_display. */ static gboolean detect_user_input(GIOChannel *channel, GIOCondition condition, gpointer user_data) { int c; gboolean config_mode = FALSE; gboolean rc = G_SOURCE_CONTINUE; /* If the attached pty device (pseudo-terminal) has been closed/deleted, * the condition (G_IO_IN | G_IO_ERR | G_IO_HUP) occurs. * Exit with an error, otherwise the process would persist in the * background and significantly raise the CPU usage. */ if ((condition & G_IO_ERR) && (condition & G_IO_HUP)) { rc = G_SOURCE_REMOVE; clean_up(CRM_EX_IOERR); } /* The connection/fd has been closed. Refresh the screen and remove this * event source hence ignore stdin. */ if (condition & (G_IO_HUP | G_IO_NVAL)) { rc = G_SOURCE_REMOVE; } if ((condition & G_IO_IN) == 0) { return rc; } while (1) { /* Get user input */ c = getchar(); switch (c) { case 'm': interactive_fence_level++; if (interactive_fence_level > 3) { interactive_fence_level = 0; } set_fencing_options(interactive_fence_level); break; case 'c': show ^= pcmk_section_tickets; break; case 'f': show ^= pcmk_section_failcounts; break; case 'n': show_opts ^= pcmk_show_rscs_by_node; break; case 'o': show ^= pcmk_section_operations; if (!pcmk_is_set(show, pcmk_section_operations)) { show_opts &= ~pcmk_show_timing; } break; case 'r': show_opts ^= pcmk_show_inactive_rscs; break; case 'R': show_opts ^= pcmk_show_details; break; case 't': show_opts ^= pcmk_show_timing; if (pcmk_is_set(show_opts, pcmk_show_timing)) { show |= pcmk_section_operations; } break; case 'A': show ^= pcmk_section_attributes; break; case 'L': show ^= pcmk_section_bans; break; case 'D': /* If any header is shown, clear them all, otherwise set them all */ if (pcmk_any_flags_set(show, pcmk_section_summary)) { show &= ~pcmk_section_summary; } else { show |= pcmk_section_summary; } /* Regardless, we don't show options in console mode. */ show &= ~pcmk_section_options; break; case 'b': show_opts ^= pcmk_show_brief; break; case 'j': show_opts ^= pcmk_show_pending; break; case '?': config_mode = TRUE; break; default: /* All other keys just redraw the screen. */ goto refresh; } if (!config_mode) goto refresh; clear(); refresh(); curses_formatted_printf(out, "%s", "Display option change mode\n"); print_option_help(out, 'c', pcmk_is_set(show, pcmk_section_tickets)); print_option_help(out, 'f', pcmk_is_set(show, pcmk_section_failcounts)); print_option_help(out, 'n', pcmk_is_set(show_opts, pcmk_show_rscs_by_node)); print_option_help(out, 'o', pcmk_is_set(show, pcmk_section_operations)); print_option_help(out, 'r', pcmk_is_set(show_opts, pcmk_show_inactive_rscs)); print_option_help(out, 't', pcmk_is_set(show_opts, pcmk_show_timing)); print_option_help(out, 'A', pcmk_is_set(show, pcmk_section_attributes)); print_option_help(out, 'L', pcmk_is_set(show, pcmk_section_bans)); print_option_help(out, 'D', !pcmk_is_set(show, pcmk_section_summary)); print_option_help(out, 'R', pcmk_any_flags_set(show_opts, pcmk_show_details)); print_option_help(out, 'b', pcmk_is_set(show_opts, pcmk_show_brief)); print_option_help(out, 'j', pcmk_is_set(show_opts, pcmk_show_pending)); curses_formatted_printf(out, "%d m: \t%s\n", interactive_fence_level, get_option_desc('m')); curses_formatted_printf(out, "%s", "\nToggle fields via field letter, type any other key to return\n"); } refresh: refresh_after_event(FALSE, TRUE); return rc; } #endif // CURSES_ENABLED // Basically crm_signal_handler(SIGCHLD, SIG_IGN) plus the SA_NOCLDWAIT flag static void avoid_zombies(void) { struct sigaction sa; memset(&sa, 0, sizeof(struct sigaction)); if (sigemptyset(&sa.sa_mask) < 0) { crm_warn("Cannot avoid zombies: %s", pcmk_rc_str(errno)); return; } sa.sa_handler = SIG_IGN; sa.sa_flags = SA_RESTART|SA_NOCLDWAIT; if (sigaction(SIGCHLD, &sa, NULL) < 0) { crm_warn("Cannot avoid zombies: %s", pcmk_rc_str(errno)); } } 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 } }; #if CURSES_ENABLED const char *fmts = "console (default), html, text, xml, none"; #else const char *fmts = "text (default), html, xml, none"; #endif // CURSES_ENABLED const char *desc = NULL; desc = "Notes:\n\n" "Time Specification:\n\n" "The TIMESPEC in any command line option can be specified in many\n" "different formats. It can be an integer number of seconds, a\n" "number plus units (us/usec/ms/msec/s/sec/m/min/h/hr), or an ISO\n" "8601 period specification.\n\n" "Output Control:\n\n" "By default, a particular set of sections are written to the\n" "output destination. The default varies based on the output\n" "format: XML includes all sections by default, while other output\n" "formats include less. This set can be modified with the --include\n" "and --exclude command line options. Each option may be passed\n" "multiple times, and each can specify a comma-separated list of\n" "sections. The options are applied to the default set, in order\n" "from left to right as they are passed on the command line. For a\n" "list of valid sections, pass --include=list or --exclude=list.\n\n" "Interactive Use:\n\n" #if CURSES_ENABLED "When run interactively, crm_mon can be told to hide and show\n" "various sections of output. To see a help screen explaining the\n" "options, press '?'. Any key stroke aside from those listed will\n" "cause the screen to refresh.\n\n" #else "The local installation of Pacemaker was built without support for\n" "interactive (console) mode. A curses library must be available at\n" "build time to support interactive mode.\n\n" #endif // CURSES_ENABLED "Examples:\n\n" #if CURSES_ENABLED "Display the cluster status on the console with updates as they\n" "occur:\n\n" "\tcrm_mon\n\n" #endif // CURSES_ENABLED "Display the cluster status once and exit:\n\n" "\tcrm_mon -1\n\n" "Display the cluster status, group resources by node, and include\n" "inactive resources in the list:\n\n" "\tcrm_mon --group-by-node --inactive\n\n" "Start crm_mon as a background daemon and have it write the\n" "cluster status to an HTML file:\n\n" "\tcrm_mon --daemonize --output-as html " "--output-to /path/to/docroot/filename.html\n\n" "Display the cluster status as XML:\n\n" "\tcrm_mon --output-as xml\n\n"; context = pcmk__build_arg_context(args, fmts, group, NULL); pcmk__add_main_args(context, extra_prog_entries); g_option_context_set_description(context, desc); pcmk__add_arg_group(context, "display", "Display Options:", "Show display options", display_entries); pcmk__add_arg_group(context, "additional", "Additional Options:", "Show additional options", addl_entries); pcmk__add_arg_group(context, "deprecated", "Deprecated Options:", "Show deprecated options", deprecated_entries); return context; } /*! * \internal * \brief Set output format based on \c --output-as arguments and mode arguments * * When the deprecated \c --as-xml argument is parsed, a callback function sets * \c output_format. Otherwise, this function does the same based on the current * \c --output-as arguments and the \c --one-shot and \c --daemonize arguments. * * \param[in,out] args Command line arguments */ static void reconcile_output_format(pcmk__common_args_t *args) { if (output_format != mon_output_unset) { /* The deprecated --as-xml argument was used, and we're finished. Note * that this means the deprecated argument takes precedence. */ return; } if (pcmk__str_eq(args->output_ty, PCMK_VALUE_NONE, pcmk__str_none)) { output_format = mon_output_none; } else if (pcmk__str_eq(args->output_ty, "html", pcmk__str_none)) { output_format = mon_output_html; umask(S_IWGRP | S_IWOTH); // World-readable HTML } else if (pcmk__str_eq(args->output_ty, "xml", pcmk__str_none)) { output_format = mon_output_xml; #if CURSES_ENABLED } else if (pcmk__str_eq(args->output_ty, "console", pcmk__str_null_matches)) { /* Console is the default format if no conflicting options are given. * * Use text output instead if one of the following conditions is met: * * We've requested daemonized or one-shot mode (console output is * incompatible with modes other than mon_exec_update) * * We requested the version, which is effectively one-shot * * We specified a non-stdout output destination (console mode is * compatible only with stdout) */ if ((options.exec_mode == mon_exec_daemonized) || (options.exec_mode == mon_exec_one_shot) || args->version || !pcmk__str_eq(args->output_dest, "-", pcmk__str_null_matches)) { pcmk__str_update(&args->output_ty, "text"); output_format = mon_output_plain; } else { pcmk__str_update(&args->output_ty, "console"); output_format = mon_output_console; crm_enable_stderr(FALSE); } #endif // CURSES_ENABLED } else if (pcmk__str_eq(args->output_ty, "text", pcmk__str_null_matches)) { /* Text output was explicitly requested, or it's the default because * curses is not enabled */ pcmk__str_update(&args->output_ty, "text"); output_format = mon_output_plain; } // Otherwise, invalid format. Let pcmk__output_new() throw an error. } /*! * \internal * \brief Set execution mode to the output format's default if appropriate * * \param[in,out] args Command line arguments */ static void set_default_exec_mode(const pcmk__common_args_t *args) { if (output_format == mon_output_console) { /* Update is the only valid mode for console, but set here instead of * reconcile_output_format() for isolation and consistency */ options.exec_mode = mon_exec_update; } else if (options.exec_mode == mon_exec_unset) { // Default to one-shot mode for all other formats options.exec_mode = mon_exec_one_shot; } else if ((options.exec_mode == mon_exec_update) && pcmk__str_eq(args->output_dest, "-", pcmk__str_null_matches)) { // If not using console format, update mode cannot be used with stdout options.exec_mode = mon_exec_one_shot; } } static void clean_up_on_connection_failure(int rc) { if (rc == ENOTCONN) { if (pcmkd_state == pcmk_pacemakerd_state_remote) { g_set_error(&error, PCMK__EXITC_ERROR, CRM_EX_ERROR, "Error: remote-node not connected to cluster"); } else { g_set_error(&error, PCMK__EXITC_ERROR, CRM_EX_ERROR, "Error: cluster is not available on this node"); } } else { g_set_error(&error, PCMK__EXITC_ERROR, CRM_EX_ERROR, "Connection to cluster failed: %s", pcmk_rc_str(rc)); } clean_up(pcmk_rc2exitc(rc)); } static void one_shot(void) { int rc = pcmk__status(out, cib, fence_history, show, show_opts, options.only_node, options.only_rsc, options.neg_location_prefix, 0); if (rc == pcmk_rc_ok) { clean_up(pcmk_rc2exitc(rc)); } else { clean_up_on_connection_failure(rc); } } static void exit_on_invalid_cib(void) { if (cib != NULL) { return; } // Shouldn't really be possible g_set_error(&error, PCMK__EXITC_ERROR, CRM_EX_ERROR, "Invalid CIB source"); clean_up(CRM_EX_ERROR); } int main(int argc, char **argv) { int rc = pcmk_rc_ok; GOptionGroup *output_group = NULL; args = pcmk__new_common_args(SUMMARY); context = build_arg_context(args, &output_group); pcmk__register_formats(output_group, formats); options.pid_file = strdup("/tmp/ClusterMon.pid"); pcmk__cli_init_logging("crm_mon", 0); // Avoid needing to wait for subprocesses forked for -E/--external-agent avoid_zombies(); processed_args = pcmk__cmdline_preproc(argv, "eimpxEILU"); fence_history_cb("--fence-history", "1", NULL, NULL); /* Set an HTML title regardless of what format we will eventually use. * Doing this here means the user can give their own title on the command * line. */ if (!pcmk__force_args(context, &error, "%s --html-title \"Cluster Status\"", g_get_prgname())) { return clean_up(CRM_EX_USAGE); } if (!g_option_context_parse_strv(context, &processed_args, &error)) { return clean_up(CRM_EX_USAGE); } for (int i = 0; i < args->verbosity; i++) { crm_bump_log_level(argc, argv); } if (!args->version) { if (args->quiet) { include_exclude_cb("--exclude", "times", NULL, NULL); } if (options.watch_fencing) { fence_history_cb("--fence-history", "0", NULL, NULL); options.fence_connect = TRUE; } /* create the cib-object early to be able to do further * decisions based on the cib-source */ cib = cib_new(); exit_on_invalid_cib(); switch (cib->variant) { case cib_native: // Everything (fencer, CIB, pcmkd status) should be available break; case cib_file: // Live fence history is not meaningful fence_history_cb("--fence-history", "0", NULL, NULL); /* Notifications are unsupported; nothing to monitor * @COMPAT: Let setup_cib_connection() handle this by exiting? */ options.exec_mode = mon_exec_one_shot; break; case cib_remote: // We won't receive any fencing updates fence_history_cb("--fence-history", "0", NULL, NULL); break; default: /* something is odd */ exit_on_invalid_cib(); break; } if ((options.exec_mode == mon_exec_daemonized) && !options.external_agent && pcmk__str_eq(args->output_dest, "-", pcmk__str_null_matches)) { g_set_error(&error, PCMK__EXITC_ERROR, CRM_EX_USAGE, "--daemonize requires at least one of --output-to " "(with value not set to '-') and --external-agent"); return clean_up(CRM_EX_USAGE); } } reconcile_output_format(args); set_default_exec_mode(args); rc = pcmk__output_new(&out, args->output_ty, args->output_dest, argv); if (rc != pcmk_rc_ok) { g_set_error(&error, PCMK__EXITC_ERROR, CRM_EX_ERROR, "Error creating output format %s: %s", args->output_ty, pcmk_rc_str(rc)); return clean_up(CRM_EX_ERROR); } if (output_format == mon_output_legacy_xml) { output_format = mon_output_xml; pcmk__output_set_legacy_xml(out); } /* output_format MUST NOT BE CHANGED AFTER THIS POINT. */ /* If we had a valid format for pcmk__output_new(), output_format should be * set by now. */ CRM_ASSERT(output_format != mon_output_unset); if (output_format == mon_output_plain) { pcmk__output_text_set_fancy(out, true); } if (options.exec_mode == mon_exec_daemonized) { if (!options.external_agent && (output_format == mon_output_none)) { g_set_error(&error, PCMK__EXITC_ERROR, CRM_EX_USAGE, "--daemonize requires --external-agent if used with " "--output-as=none"); return clean_up(CRM_EX_USAGE); } crm_enable_stderr(FALSE); cib_delete(cib); cib = NULL; pcmk__daemonize(crm_system_name, options.pid_file); cib = cib_new(); exit_on_invalid_cib(); } show = default_includes(output_format); /* Apply --include/--exclude flags we used internally. There's no error reporting * here because this would be a programming error. */ apply_include_exclude(options.includes_excludes, &error); /* And now apply any --include/--exclude flags the user gave on the command line. * These are done in a separate pass from the internal ones because we want to * make sure whatever the user specifies overrides whatever we do. */ if (!apply_include_exclude(options.user_includes_excludes, &error)) { return clean_up(CRM_EX_USAGE); } /* Sync up the initial value of interactive_fence_level with whatever was set with * --include/--exclude= options. */ if (pcmk_all_flags_set(show, pcmk_section_fencing_all)) { interactive_fence_level = 3; } else if (pcmk_is_set(show, pcmk_section_fence_worked)) { interactive_fence_level = 2; } else if (pcmk_any_flags_set(show, pcmk_section_fence_failed | pcmk_section_fence_pending)) { interactive_fence_level = 1; } else { interactive_fence_level = 0; } pcmk__register_lib_messages(out); crm_mon_register_messages(out); pe__register_messages(out); stonith__register_messages(out); // Messages internal to this file, nothing curses-specific pcmk__register_messages(out, fmt_functions); if (args->version) { out->version(out, false); return clean_up(CRM_EX_OK); } if (output_format == mon_output_xml) { show_opts |= pcmk_show_inactive_rscs | pcmk_show_timing; } if ((output_format == mon_output_html) && (out->dest != stdout)) { char *content = pcmk__itoa(options.reconnect_ms / 1000); pcmk__html_add_header(PCMK__XE_META, PCMK__XA_HTTP_EQUIV, PCMK__VALUE_REFRESH, PCMK__XA_CONTENT, content, NULL); free(content); } crm_info("Starting %s", crm_system_name); cib__set_output(cib, out); if (options.exec_mode == mon_exec_one_shot) { one_shot(); } scheduler = pe_new_working_set(); pcmk__mem_assert(scheduler); scheduler->priv->out = out; if ((cib->variant == cib_native) && pcmk_is_set(show, pcmk_section_times)) { // Currently used only in the times section pcmk__query_node_name(out, 0, &(scheduler->priv->local_node_name), 0); } out->message(out, "crm-mon-disconnected", "Waiting for initial connection", pcmkd_state); do { out->transient(out, "Connecting to cluster..."); rc = setup_api_connections(); if (rc != pcmk_rc_ok) { if ((rc == ENOTCONN) || (rc == ECONNREFUSED)) { out->transient(out, "Connection failed. Retrying in %ums...", options.reconnect_ms); } // Give some time to view all output even if we won't retry pcmk__sleep_ms(options.reconnect_ms); #if CURSES_ENABLED if (output_format == mon_output_console) { clear(); refresh(); } #endif } } while ((rc == ENOTCONN) || (rc == ECONNREFUSED)); if (rc != pcmk_rc_ok) { clean_up_on_connection_failure(rc); } set_fencing_options(interactive_fence_level); mon_refresh_display(NULL); mainloop = g_main_loop_new(NULL, FALSE); mainloop_add_signal(SIGTERM, mon_shutdown); mainloop_add_signal(SIGINT, mon_shutdown); #if CURSES_ENABLED if (output_format == mon_output_console) { ncurses_winch_handler = crm_signal_handler(SIGWINCH, mon_winresize); if (ncurses_winch_handler == SIG_DFL || ncurses_winch_handler == SIG_IGN || ncurses_winch_handler == SIG_ERR) ncurses_winch_handler = NULL; io_channel = g_io_channel_unix_new(STDIN_FILENO); g_io_add_watch(io_channel, (G_IO_IN | G_IO_ERR | G_IO_HUP | G_IO_NVAL), detect_user_input, NULL); } #endif /* When refresh_trigger->trigger is set to TRUE, call mon_refresh_display. In * this file, that is anywhere mainloop_set_trigger is called. */ refresh_trigger = mainloop_add_trigger(G_PRIORITY_LOW, mon_refresh_display, NULL); g_main_loop_run(mainloop); g_main_loop_unref(mainloop); crm_info("Exiting %s", crm_system_name); return clean_up(CRM_EX_OK); } static int send_custom_trap(const char *node, const char *rsc, const char *task, int target_rc, int rc, int status, const char *desc) { pid_t pid; /*setenv needs chars, these are ints */ char *rc_s = pcmk__itoa(rc); char *status_s = pcmk__itoa(status); char *target_rc_s = pcmk__itoa(target_rc); crm_debug("Sending external notification to '%s' via '%s'", options.external_recipient, options.external_agent); if(rsc) { setenv("CRM_notify_rsc", rsc, 1); } if (options.external_recipient) { setenv("CRM_notify_recipient", options.external_recipient, 1); } setenv("CRM_notify_node", node, 1); setenv("CRM_notify_task", task, 1); setenv("CRM_notify_desc", desc, 1); setenv("CRM_notify_rc", rc_s, 1); setenv("CRM_notify_target_rc", target_rc_s, 1); setenv("CRM_notify_status", status_s, 1); pid = fork(); if (pid == -1) { out->err(out, "notification fork() failed: %s", strerror(errno)); } if (pid == 0) { /* crm_debug("notification: I am the child. Executing the nofitication program."); */ execl(options.external_agent, options.external_agent, NULL); crm_exit(CRM_EX_ERROR); } crm_trace("Finished running custom notification program '%s'.", options.external_agent); free(target_rc_s); free(status_s); free(rc_s); return 0; } static int handle_rsc_op(xmlNode *xml, void *userdata) { const char *node_id = (const char *) userdata; int rc = -1; int status = -1; int target_rc = -1; gboolean notify = TRUE; char *rsc = NULL; char *task = NULL; const char *desc = NULL; const char *magic = NULL; const char *id = NULL; const char *node = NULL; xmlNode *n = xml; xmlNode * rsc_op = xml; if(strcmp((const char*)xml->name, PCMK__XE_LRM_RSC_OP) != 0) { pcmk__xe_foreach_child(xml, NULL, handle_rsc_op, (void *) node_id); return pcmk_rc_ok; } id = pcmk__xe_history_key(rsc_op); magic = crm_element_value(rsc_op, PCMK__XA_TRANSITION_MAGIC); if (magic == NULL) { /* non-change */ return pcmk_rc_ok; } if (!decode_transition_magic(magic, NULL, NULL, NULL, &status, &rc, &target_rc)) { crm_err("Invalid event %s detected for %s", magic, id); return pcmk_rc_ok; } if (parse_op_key(id, &rsc, &task, NULL) == FALSE) { crm_err("Invalid event detected for %s", id); goto bail; } node = crm_element_value(rsc_op, PCMK__META_ON_NODE); while ((n != NULL) && !pcmk__xe_is(n, PCMK__XE_NODE_STATE)) { n = n->parent; } if(node == NULL && n) { node = crm_element_value(n, PCMK_XA_UNAME); } if (node == NULL && n) { node = pcmk__xe_id(n); } if (node == NULL) { node = node_id; } if (node == NULL) { crm_err("No node detected for event %s (%s)", magic, id); goto bail; } /* look up where we expected it to be? */ desc = pcmk_rc_str(pcmk_rc_ok); if ((status == PCMK_EXEC_DONE) && (target_rc == rc)) { crm_notice("%s of %s on %s completed: %s", task, rsc, node, desc); if (rc == PCMK_OCF_NOT_RUNNING) { notify = FALSE; } } else if (status == PCMK_EXEC_DONE) { desc = services_ocf_exitcode_str(rc); crm_warn("%s of %s on %s failed: %s", task, rsc, node, desc); } else { desc = pcmk_exec_status_str(status); crm_warn("%s of %s on %s failed: %s", task, rsc, node, desc); } if (notify && options.external_agent) { send_custom_trap(node, rsc, task, target_rc, rc, status, desc); } bail: free(rsc); free(task); return pcmk_rc_ok; } /* This function is just a wrapper around mainloop_set_trigger so that it can be * called from a mainloop directly. It's simply another way of ensuring the screen * gets redrawn. */ static gboolean mon_trigger_refresh(gpointer user_data) { mainloop_set_trigger((crm_trigger_t *) refresh_trigger); return FALSE; } static int handle_op_for_node(xmlNode *xml, void *userdata) { const char *node = crm_element_value(xml, PCMK_XA_UNAME); if (node == NULL) { node = pcmk__xe_id(xml); } handle_rsc_op(xml, (void *) node); return pcmk_rc_ok; } static int crm_diff_update_element(xmlNode *change, void *userdata) { const char *name = NULL; const char *op = crm_element_value(change, PCMK_XA_OPERATION); const char *xpath = crm_element_value(change, PCMK_XA_PATH); xmlNode *match = NULL; const char *node = NULL; if (op == NULL) { return pcmk_rc_ok; } else if (strcmp(op, PCMK_VALUE_CREATE) == 0) { match = change->children; } else if (pcmk__str_any_of(op, PCMK_VALUE_MOVE, PCMK_VALUE_DELETE, NULL)) { return pcmk_rc_ok; } else if (strcmp(op, PCMK_VALUE_MODIFY) == 0) { match = pcmk__xe_first_child(change, PCMK_XE_CHANGE_RESULT, NULL, NULL); if(match) { match = match->children; } } if(match) { name = (const char *)match->name; } crm_trace("Handling %s operation for %s %p, %s", op, xpath, match, name); if(xpath == NULL) { /* Version field, ignore */ } else if(name == NULL) { crm_debug("No result for %s operation to %s", op, xpath); CRM_ASSERT(pcmk__str_any_of(op, PCMK_VALUE_MOVE, PCMK_VALUE_DELETE, NULL)); } else if (strcmp(name, PCMK_XE_CIB) == 0) { pcmk__xe_foreach_child(pcmk__xe_first_child(match, PCMK_XE_STATUS, NULL, NULL), NULL, handle_op_for_node, NULL); } else if (strcmp(name, PCMK_XE_STATUS) == 0) { pcmk__xe_foreach_child(match, NULL, handle_op_for_node, NULL); } else if (strcmp(name, PCMK__XE_NODE_STATE) == 0) { node = crm_element_value(match, PCMK_XA_UNAME); if (node == NULL) { node = pcmk__xe_id(match); } handle_rsc_op(match, (void *) node); } else if (strcmp(name, PCMK__XE_LRM) == 0) { node = pcmk__xe_id(match); handle_rsc_op(match, (void *) node); } else if (strcmp(name, PCMK__XE_LRM_RESOURCES) == 0) { char *local_node = pcmk__xpath_node_id(xpath, PCMK__XE_LRM); handle_rsc_op(match, local_node); free(local_node); } else if (strcmp(name, PCMK__XE_LRM_RESOURCE) == 0) { char *local_node = pcmk__xpath_node_id(xpath, PCMK__XE_LRM); handle_rsc_op(match, local_node); free(local_node); } else if (strcmp(name, PCMK__XE_LRM_RSC_OP) == 0) { char *local_node = pcmk__xpath_node_id(xpath, PCMK__XE_LRM); handle_rsc_op(match, local_node); free(local_node); } else { crm_trace("Ignoring %s operation for %s %p, %s", op, xpath, match, name); } return pcmk_rc_ok; } static void crm_diff_update(const char *event, xmlNode * msg) { int rc = -1; static bool stale = FALSE; gboolean cib_updated = FALSE; xmlNode *wrapper = pcmk__xe_first_child(msg, PCMK__XE_CIB_UPDATE_RESULT, NULL, NULL); xmlNode *diff = pcmk__xe_first_child(wrapper, NULL, NULL, NULL); out->progress(out, false); if (current_cib != NULL) { rc = xml_apply_patchset(current_cib, diff, TRUE); switch (rc) { case -pcmk_err_diff_resync: case -pcmk_err_diff_failed: crm_notice("[%s] Patch aborted: %s (%d)", event, pcmk_strerror(rc), rc); pcmk__xml_free(current_cib); current_cib = NULL; break; case pcmk_ok: cib_updated = TRUE; break; default: crm_notice("[%s] ABORTED: %s (%d)", event, pcmk_strerror(rc), rc); pcmk__xml_free(current_cib); current_cib = NULL; } } if (current_cib == NULL) { crm_trace("Re-requesting the full cib"); cib->cmds->query(cib, NULL, ¤t_cib, cib_sync_call); } if (options.external_agent) { int format = 0; crm_element_value_int(diff, PCMK_XA_FORMAT, &format); if (format == 2) { xmlNode *wrapper = pcmk__xe_first_child(msg, PCMK__XE_CIB_UPDATE_RESULT, NULL, NULL); xmlNode *diff = pcmk__xe_first_child(wrapper, NULL, NULL, NULL); pcmk__xe_foreach_child(diff, NULL, crm_diff_update_element, NULL); } else { crm_err("Unknown patch format: %d", format); } } if (current_cib == NULL) { if(!stale) { out->info(out, "--- Stale data ---"); } stale = TRUE; return; } stale = FALSE; refresh_after_event(cib_updated, FALSE); } static int mon_refresh_display(gpointer user_data) { int rc = pcmk_rc_ok; last_refresh = time(NULL); if (output_format == mon_output_none) { return G_SOURCE_REMOVE; } if (fence_history == pcmk__fence_history_full && !pcmk_all_flags_set(show, pcmk_section_fencing_all) && output_format != mon_output_xml) { fence_history = pcmk__fence_history_reduced; } // Get an up-to-date pacemakerd status for the cluster summary if (cib->variant == cib_native) { pcmk__pacemakerd_status(out, crm_system_name, options.reconnect_ms / 2, false, &pcmkd_state); } if (out->dest != stdout) { out->reset(out); } rc = pcmk__output_cluster_status(scheduler, st, cib, current_cib, pcmkd_state, fence_history, show, show_opts, options.only_node,options.only_rsc, options.neg_location_prefix); if (rc == pcmk_rc_schema_validation) { clean_up(CRM_EX_CONFIG); return G_SOURCE_REMOVE; } if (out->dest != stdout) { out->finish(out, CRM_EX_OK, true, NULL); } return G_SOURCE_CONTINUE; } /* This function is called for fencing events (see setup_fencer_connection() for * which ones) when --watch-fencing is used on the command line */ static void mon_st_callback_event(stonith_t * st, stonith_event_t * e) { if (st->state == stonith_disconnected) { /* disconnect cib as well and have everything reconnect */ mon_cib_connection_destroy(NULL); } else if (options.external_agent) { char *desc = stonith__event_description(e); send_custom_trap(e->target, NULL, e->operation, pcmk_ok, e->result, 0, desc); free(desc); } } /* Cause the screen to be redrawn (via mainloop_set_trigger) when various conditions are met: * * - If the last update occurred more than reconnect_ms ago (defaults to 5s, but * can be changed via the -i command line option), or * - After every 10 CIB updates, or * - If it's been 2s since the last update * * This function sounds like it would be more broadly useful, but it is only called when a * fencing event is received or a CIB diff occurrs. */ static void refresh_after_event(gboolean data_updated, gboolean enforce) { static int updates = 0; time_t now = time(NULL); if (data_updated) { updates++; } if(refresh_timer == NULL) { refresh_timer = mainloop_timer_add("refresh", 2000, FALSE, mon_trigger_refresh, NULL); } if (reconnect_timer > 0) { /* we will receive a refresh request after successful reconnect */ mainloop_timer_stop(refresh_timer); return; } /* as we're not handling initial failure of fencer-connection as * fatal give it a retry here * not getting here if cib-reconnection is already on the way */ setup_fencer_connection(); if (enforce || ((now - last_refresh) > (options.reconnect_ms / 1000)) || updates >= 10) { mainloop_set_trigger((crm_trigger_t *) refresh_trigger); mainloop_timer_stop(refresh_timer); updates = 0; } else { mainloop_timer_start(refresh_timer); } } /* This function is called for fencing events (see setup_fencer_connection() for * which ones) when --watch-fencing is NOT used on the command line */ static void mon_st_callback_display(stonith_t * st, stonith_event_t * e) { if (st->state == stonith_disconnected) { /* disconnect cib as well and have everything reconnect */ mon_cib_connection_destroy(NULL); } else { out->progress(out, false); refresh_after_event(TRUE, FALSE); } } /* * De-init ncurses, disconnect from the CIB manager, disconnect fencing, * deallocate memory and show usage-message if requested. * * We don't actually return, but nominally returning crm_exit_t allows a usage * like "return clean_up(exit_code);" which helps static analysis understand the * code flow. */ static crm_exit_t clean_up(crm_exit_t exit_code) { /* Quitting crm_mon is much more complicated than it ought to be. */ /* (1) Close connections, free things, etc. */ if (io_channel != NULL) { g_io_channel_shutdown(io_channel, TRUE, NULL); } cib__clean_up_connection(&cib); stonith_api_delete(st); free(options.neg_location_prefix); free(options.only_node); free(options.only_rsc); free(options.pid_file); g_slist_free_full(options.includes_excludes, free); g_strfreev(processed_args); pe_free_working_set(scheduler); /* (2) If this is abnormal termination and we're in curses mode, shut down * curses first. Any messages displayed to the screen before curses is shut * down will be lost because doing the shut down will also restore the * screen to whatever it looked like before crm_mon was started. */ if (((error != NULL) || (exit_code == CRM_EX_USAGE)) && (output_format == mon_output_console) && (out != NULL)) { out->finish(out, exit_code, false, NULL); pcmk__output_free(out); out = NULL; } /* (3) If this is a command line usage related failure, print the usage * message. */ if (exit_code == CRM_EX_USAGE && (output_format == mon_output_console || output_format == mon_output_plain)) { char *help = g_option_context_get_help(context, TRUE, NULL); fprintf(stderr, "%s", help); g_free(help); } pcmk__free_arg_context(context); /* (4) If this is any kind of error, print the error out and exit. Make * sure to handle situations both before and after formatted output is * set up. We want errors to appear formatted if at all possible. */ if (error != NULL) { if (out != NULL) { out->err(out, "%s: %s", g_get_prgname(), error->message); out->finish(out, exit_code, true, NULL); pcmk__output_free(out); } else { fprintf(stderr, "%s: %s\n", g_get_prgname(), error->message); } g_clear_error(&error); crm_exit(exit_code); } /* (5) Print formatted output to the screen if we made it far enough in * crm_mon to be able to do so. */ if (out != NULL) { if (options.exec_mode != mon_exec_daemonized) { out->finish(out, exit_code, true, NULL); } pcmk__output_free(out); pcmk__unregister_formats(); } crm_exit(exit_code); }