This source file includes following definitions.
- do_ha_control
- do_shutdown
- do_shutdown_req
- crmd_fast_exit
- crmd_exit
- do_exit
- sigpipe_ignore
- do_startup
- accept_controller_client
- dispatch_controller_ipc
- crmd_ipc_closed
- crmd_ipc_destroy
- do_stop
- do_started
- do_recover
- crmd_metadata
- verify_crmd_options
- crmd_pref
- config_query_callback
- crm_read_options
- do_read_config
- crm_shutdown
1
2
3
4
5
6
7
8
9
10 #include <crm_internal.h>
11
12 #include <sys/param.h>
13 #include <sys/types.h>
14 #include <sys/stat.h>
15
16 #include <crm/crm.h>
17 #include <crm/msg_xml.h>
18 #include <crm/pengine/rules.h>
19 #include <crm/cluster/internal.h>
20 #include <crm/cluster/election_internal.h>
21 #include <crm/common/ipc_internal.h>
22
23 #include <pacemaker-controld.h>
24
25 qb_ipcs_service_t *ipcs = NULL;
26
27 #if SUPPORT_COROSYNC
28 extern gboolean crm_connect_corosync(crm_cluster_t * cluster);
29 #endif
30
31 void crm_shutdown(int nsig);
32 gboolean crm_read_options(gpointer user_data);
33
34 gboolean fsa_has_quorum = FALSE;
35 crm_trigger_t *fsa_source = NULL;
36 crm_trigger_t *config_read = NULL;
37 bool no_quorum_suicide_escalation = FALSE;
38 bool controld_shutdown_lock_enabled = false;
39
40
41 void
42 do_ha_control(long long action,
43 enum crmd_fsa_cause cause,
44 enum crmd_fsa_state cur_state,
45 enum crmd_fsa_input current_input, fsa_data_t * msg_data)
46 {
47 gboolean registered = FALSE;
48 static crm_cluster_t *cluster = NULL;
49
50 if (cluster == NULL) {
51 cluster = calloc(1, sizeof(crm_cluster_t));
52 }
53
54 if (action & A_HA_DISCONNECT) {
55 crm_cluster_disconnect(cluster);
56 crm_info("Disconnected from the cluster");
57
58 controld_set_fsa_input_flags(R_HA_DISCONNECTED);
59 }
60
61 if (action & A_HA_CONNECT) {
62 crm_set_status_callback(&peer_update_callback);
63 crm_set_autoreap(FALSE);
64
65 if (is_corosync_cluster()) {
66 #if SUPPORT_COROSYNC
67 registered = crm_connect_corosync(cluster);
68 #endif
69 }
70
71 if (registered == TRUE) {
72 controld_election_init(cluster->uname);
73 fsa_our_uname = cluster->uname;
74 fsa_our_uuid = cluster->uuid;
75 if(cluster->uuid == NULL) {
76 crm_err("Could not obtain local uuid");
77 registered = FALSE;
78 }
79 }
80
81 if (registered == FALSE) {
82 controld_set_fsa_input_flags(R_HA_DISCONNECTED);
83 register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL);
84 return;
85 }
86
87 populate_cib_nodes(node_update_none, __func__);
88 controld_clear_fsa_input_flags(R_HA_DISCONNECTED);
89 crm_info("Connected to the cluster");
90 }
91
92 if (action & ~(A_HA_CONNECT | A_HA_DISCONNECT)) {
93 crm_err("Unexpected action %s in %s", fsa_action2string(action),
94 __func__);
95 }
96 }
97
98
99 void
100 do_shutdown(long long action,
101 enum crmd_fsa_cause cause,
102 enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data)
103 {
104
105 controld_set_fsa_input_flags(R_SHUTDOWN);
106 controld_disconnect_fencer(FALSE);
107 }
108
109
110 void
111 do_shutdown_req(long long action,
112 enum crmd_fsa_cause cause,
113 enum crmd_fsa_state cur_state,
114 enum crmd_fsa_input current_input, fsa_data_t * msg_data)
115 {
116 xmlNode *msg = NULL;
117
118 controld_set_fsa_input_flags(R_SHUTDOWN);
119
120 crm_info("Sending shutdown request to all peers (DC is %s)",
121 (fsa_our_dc? fsa_our_dc : "not set"));
122 msg = create_request(CRM_OP_SHUTDOWN_REQ, NULL, NULL, CRM_SYSTEM_CRMD, CRM_SYSTEM_CRMD, NULL);
123
124 if (send_cluster_message(NULL, crm_msg_crmd, msg, TRUE) == FALSE) {
125 register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL);
126 }
127 free_xml(msg);
128 }
129
130 extern char *max_generation_from;
131 extern xmlNode *max_generation_xml;
132 extern GHashTable *resource_history;
133 extern GHashTable *voted;
134
135 void
136 crmd_fast_exit(crm_exit_t exit_code)
137 {
138 if (pcmk_is_set(fsa_input_register, R_STAYDOWN)) {
139 crm_warn("Inhibiting respawn "CRM_XS" remapping exit code %d to %d",
140 exit_code, CRM_EX_FATAL);
141 exit_code = CRM_EX_FATAL;
142
143 } else if ((exit_code == CRM_EX_OK)
144 && pcmk_is_set(fsa_input_register, R_IN_RECOVERY)) {
145 crm_err("Could not recover from internal error");
146 exit_code = CRM_EX_ERROR;
147 }
148 crm_exit(exit_code);
149 }
150
151 crm_exit_t
152 crmd_exit(crm_exit_t exit_code)
153 {
154 GList *gIter = NULL;
155 GMainLoop *mloop = crmd_mainloop;
156
157 static bool in_progress = FALSE;
158
159 if (in_progress && (exit_code == CRM_EX_OK)) {
160 crm_debug("Exit is already in progress");
161 return exit_code;
162
163 } else if(in_progress) {
164 crm_notice("Error during shutdown process, exiting now with status %d (%s)",
165 exit_code, crm_exit_str(exit_code));
166 crm_write_blackbox(SIGTRAP, NULL);
167 crmd_fast_exit(exit_code);
168 }
169
170 in_progress = TRUE;
171 crm_trace("Preparing to exit with status %d (%s)",
172 exit_code, crm_exit_str(exit_code));
173
174
175 controld_set_fsa_input_flags(R_HA_DISCONNECTED);
176
177
178
179 if(ipcs) {
180 crm_trace("Closing IPC server");
181 mainloop_del_ipc_server(ipcs);
182 ipcs = NULL;
183 }
184
185 controld_close_attrd_ipc();
186 pe_subsystem_free();
187 controld_disconnect_fencer(TRUE);
188
189 if ((exit_code == CRM_EX_OK) && (crmd_mainloop == NULL)) {
190 crm_debug("No mainloop detected");
191 exit_code = CRM_EX_ERROR;
192 }
193
194
195
196
197
198
199
200 if (exit_code != CRM_EX_OK) {
201 crm_notice("Forcing immediate exit with status %d (%s)",
202 exit_code, crm_exit_str(exit_code));
203 crm_write_blackbox(SIGTRAP, NULL);
204 crmd_fast_exit(exit_code);
205 }
206
207
208
209 for (gIter = fsa_message_queue; gIter != NULL; gIter = gIter->next) {
210 fsa_data_t *fsa_data = gIter->data;
211
212 crm_info("Dropping %s: [ state=%s cause=%s origin=%s ]",
213 fsa_input2string(fsa_data->fsa_input),
214 fsa_state2string(fsa_state),
215 fsa_cause2string(fsa_data->fsa_cause), fsa_data->origin);
216 delete_fsa_input(fsa_data);
217 }
218
219 controld_clear_fsa_input_flags(R_MEMBERSHIP);
220 g_list_free(fsa_message_queue); fsa_message_queue = NULL;
221
222 metadata_cache_fini();
223 controld_election_fini();
224
225
226
227
228
229 controld_disconnect_cib_manager();
230
231 verify_stopped(fsa_state, LOG_WARNING);
232 controld_clear_fsa_input_flags(R_LRM_CONNECTED);
233 lrm_state_destroy_all();
234
235
236 mainloop_destroy_trigger(fsa_source); fsa_source = NULL;
237
238 mainloop_destroy_trigger(config_read); config_read = NULL;
239 mainloop_destroy_trigger(transition_trigger); transition_trigger = NULL;
240
241 pcmk__client_cleanup();
242 crm_peer_destroy();
243
244 controld_free_fsa_timers();
245 te_cleanup_stonith_history_sync(NULL, TRUE);
246 controld_free_sched_timer();
247
248 free(fsa_our_dc_version); fsa_our_dc_version = NULL;
249 free(fsa_our_uname); fsa_our_uname = NULL;
250 free(fsa_our_uuid); fsa_our_uuid = NULL;
251 free(fsa_our_dc); fsa_our_dc = NULL;
252
253 free(fsa_cluster_name); fsa_cluster_name = NULL;
254
255 free(te_uuid); te_uuid = NULL;
256 free(failed_stop_offset); failed_stop_offset = NULL;
257 free(failed_start_offset); failed_start_offset = NULL;
258
259 free(max_generation_from); max_generation_from = NULL;
260 free_xml(max_generation_xml); max_generation_xml = NULL;
261
262 mainloop_destroy_signal(SIGPIPE);
263 mainloop_destroy_signal(SIGUSR1);
264 mainloop_destroy_signal(SIGTERM);
265 mainloop_destroy_signal(SIGTRAP);
266
267
268 if (mloop) {
269 GMainContext *ctx = g_main_loop_get_context(crmd_mainloop);
270
271
272 crmd_mainloop = NULL;
273
274
275 mainloop_destroy_signal(SIGCHLD);
276
277 crm_trace("Draining mainloop %d %d", g_main_loop_is_running(mloop), g_main_context_pending(ctx));
278
279 {
280 int lpc = 0;
281
282 while((g_main_context_pending(ctx) && lpc < 10)) {
283 lpc++;
284 crm_trace("Iteration %d", lpc);
285 g_main_context_dispatch(ctx);
286 }
287 }
288
289 crm_trace("Closing mainloop %d %d", g_main_loop_is_running(mloop), g_main_context_pending(ctx));
290 g_main_loop_quit(mloop);
291
292
293 g_main_loop_unref(mloop);
294 } else {
295 mainloop_destroy_signal(SIGCHLD);
296 }
297
298 cib_delete(fsa_cib_conn);
299 fsa_cib_conn = NULL;
300
301 throttle_fini();
302
303
304 crm_trace("Done preparing for exit with status %d (%s)",
305 exit_code, crm_exit_str(exit_code));
306 return exit_code;
307 }
308
309
310 void
311 do_exit(long long action,
312 enum crmd_fsa_cause cause,
313 enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data)
314 {
315 crm_exit_t exit_code = CRM_EX_OK;
316 int log_level = LOG_INFO;
317 const char *exit_type = "gracefully";
318
319 if (action & A_EXIT_1) {
320 log_level = LOG_ERR;
321 exit_type = "forcefully";
322 exit_code = CRM_EX_ERROR;
323 }
324
325 verify_stopped(cur_state, LOG_ERR);
326 do_crm_log(log_level, "Performing %s - %s exiting the controller",
327 fsa_action2string(action), exit_type);
328
329 crm_info("[%s] stopped (%d)", crm_system_name, exit_code);
330 crmd_exit(exit_code);
331 }
332
333 static void sigpipe_ignore(int nsig) { return; }
334
335
336 void
337 do_startup(long long action,
338 enum crmd_fsa_cause cause,
339 enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data)
340 {
341 crm_debug("Registering Signal Handlers");
342 mainloop_add_signal(SIGTERM, crm_shutdown);
343 mainloop_add_signal(SIGPIPE, sigpipe_ignore);
344
345 fsa_source = mainloop_add_trigger(G_PRIORITY_HIGH, crm_fsa_trigger, NULL);
346 config_read = mainloop_add_trigger(G_PRIORITY_HIGH, crm_read_options, NULL);
347 transition_trigger = mainloop_add_trigger(G_PRIORITY_LOW, te_graph_trigger, NULL);
348
349 crm_debug("Creating CIB manager and executor objects");
350 fsa_cib_conn = cib_new();
351
352 lrm_state_init_local();
353 if (controld_init_fsa_timers() == FALSE) {
354 register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL);
355 }
356 }
357
358
359 static int32_t
360 accept_controller_client(qb_ipcs_connection_t *c, uid_t uid, gid_t gid)
361 {
362 crm_trace("Accepting new IPC client connection");
363 if (pcmk__new_client(c, uid, gid) == NULL) {
364 return -EIO;
365 }
366 return 0;
367 }
368
369
370 static int32_t
371 dispatch_controller_ipc(qb_ipcs_connection_t * c, void *data, size_t size)
372 {
373 uint32_t id = 0;
374 uint32_t flags = 0;
375 pcmk__client_t *client = pcmk__find_client(c);
376
377 xmlNode *msg = pcmk__client_data2xml(client, data, &id, &flags);
378
379 if (msg == NULL) {
380 pcmk__ipc_send_ack(client, id, flags, "ack", CRM_EX_PROTOCOL);
381 return 0;
382 }
383 pcmk__ipc_send_ack(client, id, flags, "ack", CRM_EX_INDETERMINATE);
384
385 CRM_ASSERT(client->user != NULL);
386 pcmk__update_acl_user(msg, F_CRM_USER, client->user);
387
388 crm_xml_add(msg, F_CRM_SYS_FROM, client->id);
389 if (controld_authorize_ipc_message(msg, client, NULL)) {
390 crm_trace("Processing IPC message from client %s",
391 pcmk__client_name(client));
392 route_message(C_IPC_MESSAGE, msg);
393 }
394
395 trigger_fsa();
396 free_xml(msg);
397 return 0;
398 }
399
400 static int32_t
401 crmd_ipc_closed(qb_ipcs_connection_t * c)
402 {
403 pcmk__client_t *client = pcmk__find_client(c);
404
405 if (client) {
406 crm_trace("Disconnecting %sregistered client %s (%p/%p)",
407 (client->userdata? "" : "un"), pcmk__client_name(client),
408 c, client);
409 free(client->userdata);
410 pcmk__free_client(client);
411 trigger_fsa();
412 }
413 return 0;
414 }
415
416 static void
417 crmd_ipc_destroy(qb_ipcs_connection_t * c)
418 {
419 crm_trace("Connection %p", c);
420 crmd_ipc_closed(c);
421 }
422
423
424 void
425 do_stop(long long action,
426 enum crmd_fsa_cause cause,
427 enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data)
428 {
429 crm_trace("Closing IPC server");
430 mainloop_del_ipc_server(ipcs); ipcs = NULL;
431 register_fsa_input(C_FSA_INTERNAL, I_TERMINATE, NULL);
432 }
433
434
435 void
436 do_started(long long action,
437 enum crmd_fsa_cause cause,
438 enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data)
439 {
440 static struct qb_ipcs_service_handlers crmd_callbacks = {
441 .connection_accept = accept_controller_client,
442 .connection_created = NULL,
443 .msg_process = dispatch_controller_ipc,
444 .connection_closed = crmd_ipc_closed,
445 .connection_destroyed = crmd_ipc_destroy
446 };
447
448 if (cur_state != S_STARTING) {
449 crm_err("Start cancelled... %s", fsa_state2string(cur_state));
450 return;
451
452 } else if (!pcmk_is_set(fsa_input_register, R_MEMBERSHIP)) {
453 crm_info("Delaying start, no membership data (%.16llx)", R_MEMBERSHIP);
454
455 crmd_fsa_stall(TRUE);
456 return;
457
458 } else if (!pcmk_is_set(fsa_input_register, R_LRM_CONNECTED)) {
459 crm_info("Delaying start, not connected to executor (%.16llx)", R_LRM_CONNECTED);
460
461 crmd_fsa_stall(TRUE);
462 return;
463
464 } else if (!pcmk_is_set(fsa_input_register, R_CIB_CONNECTED)) {
465 crm_info("Delaying start, CIB not connected (%.16llx)", R_CIB_CONNECTED);
466
467 crmd_fsa_stall(TRUE);
468 return;
469
470 } else if (!pcmk_is_set(fsa_input_register, R_READ_CONFIG)) {
471 crm_info("Delaying start, Config not read (%.16llx)", R_READ_CONFIG);
472
473 crmd_fsa_stall(TRUE);
474 return;
475
476 } else if (!pcmk_is_set(fsa_input_register, R_PEER_DATA)) {
477
478 crm_info("Delaying start, No peer data (%.16llx)", R_PEER_DATA);
479 crmd_fsa_stall(TRUE);
480 return;
481 }
482
483 crm_debug("Init server comms");
484 ipcs = pcmk__serve_controld_ipc(&crmd_callbacks);
485 if (ipcs == NULL) {
486 crm_err("Failed to create IPC server: shutting down and inhibiting respawn");
487 register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL);
488 } else {
489 crm_notice("Pacemaker controller successfully started and accepting connections");
490 }
491 controld_trigger_fencer_connect();
492
493 controld_clear_fsa_input_flags(R_STARTING);
494 register_fsa_input(msg_data->fsa_cause, I_PENDING, NULL);
495 }
496
497
498 void
499 do_recover(long long action,
500 enum crmd_fsa_cause cause,
501 enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data)
502 {
503 controld_set_fsa_input_flags(R_IN_RECOVERY);
504 crm_warn("Fast-tracking shutdown in response to errors");
505
506 register_fsa_input(C_FSA_INTERNAL, I_TERMINATE, NULL);
507 }
508
509 static pcmk__cluster_option_t crmd_opts[] = {
510
511
512
513
514
515 {
516 "dc-version", NULL, "string", NULL, "none", NULL,
517 "Pacemaker version on cluster node elected Designated Controller (DC)",
518 "Includes a hash which identifies the exact changeset the code was "
519 "built from. Used for diagnostic purposes."
520 },
521 {
522 "cluster-infrastructure", NULL, "string", NULL, "corosync", NULL,
523 "The messaging stack on which Pacemaker is currently running",
524 "Used for informational and diagnostic purposes."
525 },
526 {
527 "cluster-name", NULL, "string", NULL, NULL, NULL,
528 "An arbitrary name for the cluster",
529 "This optional value is mostly for users' convenience as desired "
530 "in administration, but may also be used in Pacemaker "
531 "configuration rules via the #cluster-name node attribute, and "
532 "by higher-level tools and resource agents."
533 },
534 {
535 XML_CONFIG_ATTR_DC_DEADTIME, NULL, "time",
536 NULL, "20s", pcmk__valid_interval_spec,
537 "How long to wait for a response from other nodes during start-up",
538 "The optimal value will depend on the speed and load of your network "
539 "and the type of switches used."
540 },
541 {
542 XML_CONFIG_ATTR_RECHECK, NULL, "time",
543 "Zero disables polling, while positive values are an interval in seconds"
544 "(unless other units are specified, for example \"5min\")",
545 "15min", pcmk__valid_interval_spec,
546 "Polling interval to recheck cluster state and evaluate rules "
547 "with date specifications",
548 "Pacemaker is primarily event-driven, and looks ahead to know when to "
549 "recheck cluster state for failure timeouts and most time-based "
550 "rules. However, it will also recheck the cluster after this "
551 "amount of inactivity, to evaluate rules with date specifications "
552 "and serve as a fail-safe for certain types of scheduler bugs."
553 },
554 {
555 "load-threshold", NULL, "percentage", NULL,
556 "80%", pcmk__valid_utilization,
557 "Maximum amount of system load that should be used by cluster nodes",
558 "The cluster will slow down its recovery process when the amount of "
559 "system resources used (currently CPU) approaches this limit",
560 },
561 {
562 "node-action-limit", NULL, "integer", NULL,
563 "0", pcmk__valid_number,
564 "Maximum number of jobs that can be scheduled per node "
565 "(defaults to 2x cores)"
566 },
567 { XML_CONFIG_ATTR_FENCE_REACTION, NULL, "string", NULL, "stop", NULL,
568 "How a cluster node should react if notified of its own fencing",
569 "A cluster node may receive notification of its own fencing if fencing "
570 "is misconfigured, or if fabric fencing is in use that doesn't cut "
571 "cluster communication. Allowed values are \"stop\" to attempt to "
572 "immediately stop Pacemaker and stay stopped, or \"panic\" to attempt "
573 "to immediately reboot the local node, falling back to stop on failure."
574 },
575 {
576 XML_CONFIG_ATTR_ELECTION_FAIL, NULL, "time", NULL,
577 "2min", pcmk__valid_interval_spec,
578 "*** Advanced Use Only ***",
579 "Declare an election failed if it is not decided within this much "
580 "time. If you need to adjust this value, it probably indicates "
581 "the presence of a bug."
582 },
583 {
584 XML_CONFIG_ATTR_FORCE_QUIT, NULL, "time", NULL,
585 "20min", pcmk__valid_interval_spec,
586 "*** Advanced Use Only ***",
587 "Exit immediately if shutdown does not complete within this much "
588 "time. If you need to adjust this value, it probably indicates "
589 "the presence of a bug."
590 },
591 {
592 "join-integration-timeout", "crmd-integration-timeout", "time", NULL,
593 "3min", pcmk__valid_interval_spec,
594 "*** Advanced Use Only ***",
595 "If you need to adjust this value, it probably indicates "
596 "the presence of a bug."
597 },
598 {
599 "join-finalization-timeout", "crmd-finalization-timeout", "time", NULL,
600 "30min", pcmk__valid_interval_spec,
601 "*** Advanced Use Only ***",
602 "If you need to adjust this value, it probably indicates "
603 "the presence of a bug."
604 },
605 {
606 "transition-delay", "crmd-transition-delay", "time", NULL,
607 "0s", pcmk__valid_interval_spec,
608 "*** Advanced Use Only *** Enabling this option will slow down "
609 "cluster recovery under all conditions",
610 "Delay cluster recovery for this much time to allow for additional "
611 "events to occur. Useful if your configuration is sensitive to "
612 "the order in which ping updates arrive."
613 },
614 {
615 "stonith-watchdog-timeout", NULL, "time", NULL,
616 "0", controld_verify_stonith_watchdog_timeout,
617 "How long to wait before we can assume nodes are safely down "
618 "when watchdog-based self-fencing via SBD is in use",
619 "If nonzero, along with `have-watchdog=true` automatically set by the "
620 "cluster, when fencing is required, watchdog-based self-fencing "
621 "will be performed via SBD without requiring a fencing resource "
622 "explicitly configured. "
623 "If `stonith-watchdog-timeout` is set to a positive value, unseen "
624 "nodes are assumed to self-fence within this much time. +WARNING:+ "
625 "It must be ensured that this value is larger than the "
626 "`SBD_WATCHDOG_TIMEOUT` environment variable on all nodes. "
627 "Pacemaker verifies the settings individually on all nodes and "
628 "prevents startup or shuts down if configured wrongly on the fly. "
629 "It's strongly recommended that `SBD_WATCHDOG_TIMEOUT` is set to "
630 "the same value on all nodes. "
631 "If `stonith-watchdog-timeout` is set to a negative value, and "
632 "`SBD_WATCHDOG_TIMEOUT` is set, twice that value will be used. "
633 "+WARNING:+ In this case, it's essential (currently not verified by "
634 "Pacemaker) that `SBD_WATCHDOG_TIMEOUT` is set to the same value on "
635 "all nodes."
636 },
637 {
638 "stonith-max-attempts", NULL, "integer", NULL,
639 "10", pcmk__valid_positive_number,
640 "How many times fencing can fail before it will no longer be "
641 "immediately re-attempted on a target"
642 },
643
644
645 {
646 "no-quorum-policy", NULL, "select", "stop, freeze, ignore, demote, suicide",
647 "stop", pcmk__valid_quorum, NULL, NULL
648 },
649 {
650 XML_CONFIG_ATTR_SHUTDOWN_LOCK, NULL, "boolean", NULL,
651 "false", pcmk__valid_boolean, NULL, NULL
652 },
653 };
654
655 void
656 crmd_metadata(void)
657 {
658 pcmk__print_option_metadata("pacemaker-controld",
659 "Pacemaker controller options",
660 "Cluster options used by Pacemaker's "
661 "controller (formerly called crmd)",
662 crmd_opts, PCMK__NELEM(crmd_opts));
663 }
664
665 static void
666 verify_crmd_options(GHashTable * options)
667 {
668 pcmk__validate_cluster_options(options, crmd_opts, PCMK__NELEM(crmd_opts));
669 }
670
671 static const char *
672 crmd_pref(GHashTable * options, const char *name)
673 {
674 return pcmk__cluster_option(options, crmd_opts, PCMK__NELEM(crmd_opts),
675 name);
676 }
677
678 static void
679 config_query_callback(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data)
680 {
681 const char *value = NULL;
682 GHashTable *config_hash = NULL;
683 crm_time_t *now = crm_time_new(NULL);
684 xmlNode *crmconfig = NULL;
685 xmlNode *alerts = NULL;
686
687 if (rc != pcmk_ok) {
688 fsa_data_t *msg_data = NULL;
689
690 crm_err("Local CIB query resulted in an error: %s", pcmk_strerror(rc));
691 register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL);
692
693 if (rc == -EACCES || rc == -pcmk_err_schema_validation) {
694 crm_err("The cluster is mis-configured - shutting down and staying down");
695 controld_set_fsa_input_flags(R_STAYDOWN);
696 }
697 goto bail;
698 }
699
700 crmconfig = output;
701 if ((crmconfig) &&
702 (crm_element_name(crmconfig)) &&
703 (strcmp(crm_element_name(crmconfig), XML_CIB_TAG_CRMCONFIG) != 0)) {
704 crmconfig = first_named_child(crmconfig, XML_CIB_TAG_CRMCONFIG);
705 }
706 if (!crmconfig) {
707 fsa_data_t *msg_data = NULL;
708
709 crm_err("Local CIB query for " XML_CIB_TAG_CRMCONFIG " section failed");
710 register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL);
711 goto bail;
712 }
713
714 crm_debug("Call %d : Parsing CIB options", call_id);
715 config_hash = pcmk__strkey_table(free, free);
716 pe_unpack_nvpairs(crmconfig, crmconfig, XML_CIB_TAG_PROPSET, NULL,
717 config_hash, CIB_OPTIONS_FIRST, FALSE, now, NULL);
718
719 verify_crmd_options(config_hash);
720
721 value = crmd_pref(config_hash, XML_CONFIG_ATTR_DC_DEADTIME);
722 election_trigger->period_ms = crm_parse_interval_spec(value);
723
724 value = crmd_pref(config_hash, "node-action-limit");
725 throttle_update_job_max(value);
726
727 value = crmd_pref(config_hash, "load-threshold");
728 if(value) {
729 throttle_set_load_target(strtof(value, NULL) / 100.0);
730 }
731
732 value = crmd_pref(config_hash, "no-quorum-policy");
733 if (pcmk__str_eq(value, "suicide", pcmk__str_casei) && pcmk__locate_sbd()) {
734 no_quorum_suicide_escalation = TRUE;
735 }
736
737 set_fence_reaction(crmd_pref(config_hash, XML_CONFIG_ATTR_FENCE_REACTION));
738
739 value = crmd_pref(config_hash,"stonith-max-attempts");
740 update_stonith_max_attempts(value);
741
742 value = crmd_pref(config_hash, XML_CONFIG_ATTR_FORCE_QUIT);
743 shutdown_escalation_timer->period_ms = crm_parse_interval_spec(value);
744 crm_debug("Shutdown escalation occurs if DC has not responded to request in %ums",
745 shutdown_escalation_timer->period_ms);
746
747 value = crmd_pref(config_hash, XML_CONFIG_ATTR_ELECTION_FAIL);
748 controld_set_election_period(value);
749
750 value = crmd_pref(config_hash, XML_CONFIG_ATTR_RECHECK);
751 recheck_interval_ms = crm_parse_interval_spec(value);
752 crm_debug("Re-run scheduler after %dms of inactivity", recheck_interval_ms);
753
754 value = crmd_pref(config_hash, "transition-delay");
755 transition_timer->period_ms = crm_parse_interval_spec(value);
756
757 value = crmd_pref(config_hash, "join-integration-timeout");
758 integration_timer->period_ms = crm_parse_interval_spec(value);
759
760 value = crmd_pref(config_hash, "join-finalization-timeout");
761 finalization_timer->period_ms = crm_parse_interval_spec(value);
762
763 value = crmd_pref(config_hash, XML_CONFIG_ATTR_SHUTDOWN_LOCK);
764 controld_shutdown_lock_enabled = crm_is_true(value);
765
766 free(fsa_cluster_name);
767 fsa_cluster_name = NULL;
768
769 value = g_hash_table_lookup(config_hash, "cluster-name");
770 if (value) {
771 fsa_cluster_name = strdup(value);
772 }
773
774 alerts = first_named_child(output, XML_CIB_TAG_ALERTS);
775 crmd_unpack_alerts(alerts);
776
777 controld_set_fsa_input_flags(R_READ_CONFIG);
778 crm_trace("Triggering FSA: %s", __func__);
779 mainloop_set_trigger(fsa_source);
780
781 g_hash_table_destroy(config_hash);
782 bail:
783 crm_time_free(now);
784 }
785
786 gboolean
787 crm_read_options(gpointer user_data)
788 {
789 int call_id =
790 fsa_cib_conn->cmds->query(fsa_cib_conn,
791 "//" XML_CIB_TAG_CRMCONFIG " | //" XML_CIB_TAG_ALERTS,
792 NULL, cib_xpath | cib_scope_local);
793
794 fsa_register_cib_callback(call_id, FALSE, NULL, config_query_callback);
795 crm_trace("Querying the CIB... call %d", call_id);
796 return TRUE;
797 }
798
799
800 void
801 do_read_config(long long action,
802 enum crmd_fsa_cause cause,
803 enum crmd_fsa_state cur_state,
804 enum crmd_fsa_input current_input, fsa_data_t * msg_data)
805 {
806 throttle_init();
807 mainloop_set_trigger(config_read);
808 }
809
810 void
811 crm_shutdown(int nsig)
812 {
813 if ((crmd_mainloop == NULL) || !g_main_loop_is_running(crmd_mainloop)) {
814 crmd_exit(CRM_EX_OK);
815 return;
816 }
817
818 if (pcmk_is_set(fsa_input_register, R_SHUTDOWN)) {
819 crm_err("Escalating shutdown");
820 register_fsa_input_before(C_SHUTDOWN, I_ERROR, NULL);
821 return;
822 }
823
824 controld_set_fsa_input_flags(R_SHUTDOWN);
825 register_fsa_input(C_SHUTDOWN, I_SHUTDOWN, NULL);
826
827 if (shutdown_escalation_timer->period_ms == 0) {
828 const char *value = crmd_pref(NULL, XML_CONFIG_ATTR_FORCE_QUIT);
829
830 shutdown_escalation_timer->period_ms = crm_parse_interval_spec(value);
831 }
832
833 crm_notice("Initiating controller shutdown sequence " CRM_XS
834 " limit=%ums", shutdown_escalation_timer->period_ms);
835 controld_start_timer(shutdown_escalation_timer);
836 }