This source file includes following definitions.
- lrm_connection_destroy
- make_stop_id
- copy_instance_keys
- copy_meta_keys
- history_remove_recurring_op
- history_free_recurring_ops
- history_free
- update_history_cache
- send_task_ok_ack
- op_node_name
- lrm_op_callback
- try_local_executor_connect
- do_lrm_control
- lrm_state_verify_stopped
- is_rsc_active
- build_active_RAs
- controld_query_executor_state
- controld_rc2event
- controld_trigger_delete_refresh
- notify_deleted
- lrm_remove_deleted_rsc
- lrm_remove_deleted_op
- delete_rsc_entry
- last_failed_matches_op
- lrm_clear_last_failure
- cancel_op
- cancel_action_by_key
- cancel_op_key
- get_lrm_resource
- delete_resource
- get_fake_call_id
- fake_op_status
- force_reprobe
- synthesize_lrmd_failure
- lrm_op_target
- fail_lrm_resource
- handle_reprobe_op
- do_lrm_cancel
- do_lrm_delete
- new_metadata_cb_data
- free_metadata_cb_data
- metadata_complete
- do_lrm_invoke
- construct_op
- controld_ack_event_directly
- verify_stopped
- stop_recurring_action_by_rsc
- stop_recurring_actions
- should_cancel_recurring
- should_nack_action
- do_lrm_rsc_op
- unescape_newlines
- did_lrm_rsc_op_fail
- log_executor_event
- process_lrm_event
1
2
3
4
5
6
7
8
9
10 #include <crm_internal.h>
11
12 #include <regex.h>
13 #include <sys/param.h>
14 #include <sys/types.h>
15 #include <sys/wait.h>
16
17 #include <crm/crm.h>
18 #include <crm/lrmd.h>
19 #include <crm/services.h>
20 #include <crm/common/xml.h>
21 #include <crm/lrmd_internal.h>
22
23 #include <pacemaker-internal.h>
24 #include <pacemaker-controld.h>
25
26 #define START_DELAY_THRESHOLD 5 * 60 * 1000
27 #define MAX_LRM_REG_FAILS 30
28
29 struct delete_event_s {
30 int rc;
31 const char *rsc;
32 lrm_state_t *lrm_state;
33 };
34
35 static gboolean is_rsc_active(lrm_state_t * lrm_state, const char *rsc_id);
36 static gboolean build_active_RAs(lrm_state_t * lrm_state, xmlNode * rsc_list);
37 static gboolean stop_recurring_actions(gpointer key, gpointer value, gpointer user_data);
38
39 static lrmd_event_data_t *construct_op(const lrm_state_t *lrm_state,
40 const xmlNode *rsc_op,
41 const char *rsc_id,
42 const char *operation);
43 static void do_lrm_rsc_op(lrm_state_t *lrm_state, lrmd_rsc_info_t *rsc,
44 xmlNode *msg, struct ra_metadata_s *md);
45
46 static gboolean lrm_state_verify_stopped(lrm_state_t * lrm_state, enum crmd_fsa_state cur_state,
47 int log_level);
48
49 static void
50 lrm_connection_destroy(void)
51 {
52 if (pcmk_is_set(controld_globals.fsa_input_register, R_LRM_CONNECTED)) {
53 crm_crit("Lost connection to local executor");
54 register_fsa_input(C_FSA_INTERNAL, I_ERROR, NULL);
55 controld_clear_fsa_input_flags(R_LRM_CONNECTED);
56 }
57 }
58
59 static char *
60 make_stop_id(const char *rsc, int call_id)
61 {
62 return crm_strdup_printf("%s:%d", rsc, call_id);
63 }
64
65 static void
66 copy_instance_keys(gpointer key, gpointer value, gpointer user_data)
67 {
68 if (strstr(key, CRM_META "_") == NULL) {
69 pcmk__insert_dup(user_data, (const char *) key, (const char *) value);
70 }
71 }
72
73 static void
74 copy_meta_keys(gpointer key, gpointer value, gpointer user_data)
75 {
76 if (strstr(key, CRM_META "_") != NULL) {
77 pcmk__insert_dup(user_data, (const char *) key, (const char *) value);
78 }
79 }
80
81
82
83
84
85
86
87
88
89
90 static gboolean
91 history_remove_recurring_op(rsc_history_t *history, const lrmd_event_data_t *op)
92 {
93 GList *iter;
94
95 for (iter = history->recurring_op_list; iter != NULL; iter = iter->next) {
96 lrmd_event_data_t *existing = iter->data;
97
98 if ((op->interval_ms == existing->interval_ms)
99 && pcmk__str_eq(op->rsc_id, existing->rsc_id, pcmk__str_none)
100 && pcmk__str_eq(op->op_type, existing->op_type, pcmk__str_casei)) {
101
102 history->recurring_op_list = g_list_delete_link(history->recurring_op_list, iter);
103 lrmd_free_event(existing);
104 return TRUE;
105 }
106 }
107 return FALSE;
108 }
109
110
111
112
113
114
115
116 static void
117 history_free_recurring_ops(rsc_history_t *history)
118 {
119 GList *iter;
120
121 for (iter = history->recurring_op_list; iter != NULL; iter = iter->next) {
122 lrmd_free_event(iter->data);
123 }
124 g_list_free(history->recurring_op_list);
125 history->recurring_op_list = NULL;
126 }
127
128
129
130
131
132
133
134 void
135 history_free(gpointer data)
136 {
137 rsc_history_t *history = (rsc_history_t*)data;
138
139 if (history->stop_params) {
140 g_hash_table_destroy(history->stop_params);
141 }
142
143
144 free(history->rsc.type);
145 free(history->rsc.standard);
146 free(history->rsc.provider);
147
148 lrmd_free_event(history->failed);
149 lrmd_free_event(history->last);
150 free(history->id);
151 history_free_recurring_ops(history);
152 free(history);
153 }
154
155 static void
156 update_history_cache(lrm_state_t * lrm_state, lrmd_rsc_info_t * rsc, lrmd_event_data_t * op)
157 {
158 int target_rc = 0;
159 rsc_history_t *entry = NULL;
160
161 if (op->rsc_deleted) {
162 crm_debug("Purged history for '%s' after %s", op->rsc_id, op->op_type);
163 controld_delete_resource_history(op->rsc_id, lrm_state->node_name,
164 NULL, crmd_cib_smart_opt());
165 return;
166 }
167
168 if (pcmk__str_eq(op->op_type, PCMK_ACTION_NOTIFY, pcmk__str_casei)) {
169 return;
170 }
171
172 crm_debug("Updating history for '%s' with %s op", op->rsc_id, op->op_type);
173
174 entry = g_hash_table_lookup(lrm_state->resource_history, op->rsc_id);
175 if (entry == NULL && rsc) {
176 entry = pcmk__assert_alloc(1, sizeof(rsc_history_t));
177 entry->id = pcmk__str_copy(op->rsc_id);
178 g_hash_table_insert(lrm_state->resource_history, entry->id, entry);
179
180 entry->rsc.id = entry->id;
181 entry->rsc.type = pcmk__str_copy(rsc->type);
182 entry->rsc.standard = pcmk__str_copy(rsc->standard);
183 entry->rsc.provider = pcmk__str_copy(rsc->provider);
184
185 } else if (entry == NULL) {
186 crm_info("Resource %s no longer exists, not updating cache", op->rsc_id);
187 return;
188 }
189
190 entry->last_callid = op->call_id;
191 target_rc = rsc_op_expected_rc(op);
192 if (op->op_status == PCMK_EXEC_CANCELLED) {
193 if (op->interval_ms > 0) {
194 crm_trace("Removing cancelled recurring op: " PCMK__OP_FMT,
195 op->rsc_id, op->op_type, op->interval_ms);
196 history_remove_recurring_op(entry, op);
197 return;
198 } else {
199 crm_trace("Skipping " PCMK__OP_FMT " rc=%d, status=%d",
200 op->rsc_id, op->op_type, op->interval_ms, op->rc,
201 op->op_status);
202 }
203
204 } else if (did_rsc_op_fail(op, target_rc)) {
205
206
207
208 if (entry->failed) {
209 lrmd_free_event(entry->failed);
210 }
211 entry->failed = lrmd_copy_event(op);
212
213 } else if (op->interval_ms == 0) {
214 if (entry->last) {
215 lrmd_free_event(entry->last);
216 }
217 entry->last = lrmd_copy_event(op);
218
219 if (op->params && pcmk__strcase_any_of(op->op_type, PCMK_ACTION_START,
220 PCMK_ACTION_RELOAD,
221 PCMK_ACTION_RELOAD_AGENT,
222 PCMK_ACTION_MONITOR, NULL)) {
223 if (entry->stop_params) {
224 g_hash_table_destroy(entry->stop_params);
225 }
226 entry->stop_params = pcmk__strkey_table(free, free);
227
228 g_hash_table_foreach(op->params, copy_instance_keys, entry->stop_params);
229 }
230 }
231
232 if (op->interval_ms > 0) {
233
234 history_remove_recurring_op(entry, op);
235
236 crm_trace("Adding recurring op: " PCMK__OP_FMT,
237 op->rsc_id, op->op_type, op->interval_ms);
238 entry->recurring_op_list = g_list_prepend(entry->recurring_op_list, lrmd_copy_event(op));
239
240 } else if ((entry->recurring_op_list != NULL)
241 && !pcmk__str_eq(op->op_type, PCMK_ACTION_MONITOR,
242 pcmk__str_casei)) {
243 crm_trace("Dropping %d recurring ops because of: " PCMK__OP_FMT,
244 g_list_length(entry->recurring_op_list), op->rsc_id,
245 op->op_type, op->interval_ms);
246 history_free_recurring_ops(entry);
247 }
248 }
249
250
251
252
253
254
255
256
257
258
259
260
261
262 static void
263 send_task_ok_ack(const lrm_state_t *lrm_state, const ha_msg_input_t *input,
264 const char *rsc_id, const lrmd_rsc_info_t *rsc,
265 const char *task, const char *ack_host, const char *ack_sys)
266 {
267 lrmd_event_data_t *op = construct_op(lrm_state, input->xml, rsc_id, task);
268
269 lrmd__set_result(op, PCMK_OCF_OK, PCMK_EXEC_DONE, NULL);
270 controld_ack_event_directly(ack_host, ack_sys, rsc, op, rsc_id);
271 lrmd_free_event(op);
272 }
273
274 static inline const char *
275 op_node_name(lrmd_event_data_t *op)
276 {
277 return pcmk__s(op->remote_nodename,
278 controld_globals.cluster->priv->node_name);
279 }
280
281 void
282 lrm_op_callback(lrmd_event_data_t * op)
283 {
284 CRM_CHECK(op != NULL, return);
285 switch (op->type) {
286 case lrmd_event_disconnect:
287 if (op->remote_nodename == NULL) {
288
289
290
291 lrm_connection_destroy();
292 }
293 break;
294
295 case lrmd_event_exec_complete:
296 {
297 lrm_state_t *lrm_state =
298 controld_get_executor_state(op_node_name(op), false);
299
300 pcmk__assert(lrm_state != NULL);
301 process_lrm_event(lrm_state, op, NULL, NULL);
302 }
303 break;
304
305 default:
306 break;
307 }
308 }
309
310 static void
311 try_local_executor_connect(long long action, fsa_data_t *msg_data,
312 lrm_state_t *lrm_state)
313 {
314 int rc = pcmk_rc_ok;
315
316 crm_debug("Connecting to the local executor");
317
318
319 rc = controld_connect_local_executor(lrm_state);
320 if (rc == pcmk_rc_ok) {
321 controld_set_fsa_input_flags(R_LRM_CONNECTED);
322 crm_info("Connection to the local executor established");
323 return;
324 }
325
326
327 if (lrm_state->num_lrm_register_fails < MAX_LRM_REG_FAILS) {
328 crm_warn("Failed to connect to the local executor %d time%s "
329 "(%d max): %s", lrm_state->num_lrm_register_fails,
330 pcmk__plural_s(lrm_state->num_lrm_register_fails),
331 MAX_LRM_REG_FAILS, pcmk_rc_str(rc));
332 controld_start_wait_timer();
333 crmd_fsa_stall(FALSE);
334 return;
335 }
336
337
338 crm_err("Failed to connect to the executor the max allowed "
339 "%d time%s: %s", lrm_state->num_lrm_register_fails,
340 pcmk__plural_s(lrm_state->num_lrm_register_fails),
341 pcmk_rc_str(rc));
342 register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL);
343 }
344
345
346 void
347 do_lrm_control(long long action,
348 enum crmd_fsa_cause cause,
349 enum crmd_fsa_state cur_state,
350 enum crmd_fsa_input current_input, fsa_data_t * msg_data)
351 {
352
353
354
355
356
357 lrm_state_t *lrm_state = NULL;
358
359 if (controld_globals.cluster->priv->node_name == NULL) {
360 return;
361 }
362 lrm_state = controld_get_executor_state(NULL, true);
363 if (lrm_state == NULL) {
364 register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL);
365 return;
366 }
367
368 if (action & A_LRM_DISCONNECT) {
369 if (lrm_state_verify_stopped(lrm_state, cur_state, LOG_INFO) == FALSE) {
370 if (action == A_LRM_DISCONNECT) {
371 crmd_fsa_stall(FALSE);
372 return;
373 }
374 }
375
376 controld_clear_fsa_input_flags(R_LRM_CONNECTED);
377 lrm_state_disconnect(lrm_state);
378 lrm_state_reset_tables(lrm_state, FALSE);
379 }
380
381 if (action & A_LRM_CONNECT) {
382 try_local_executor_connect(action, msg_data, lrm_state);
383 }
384
385 if (action & ~(A_LRM_CONNECT | A_LRM_DISCONNECT)) {
386 crm_err("Unexpected action %s in %s", fsa_action2string(action),
387 __func__);
388 }
389 }
390
391 static gboolean
392 lrm_state_verify_stopped(lrm_state_t * lrm_state, enum crmd_fsa_state cur_state, int log_level)
393 {
394 int counter = 0;
395 gboolean rc = TRUE;
396 const char *when = "lrm disconnect";
397
398 GHashTableIter gIter;
399 const char *key = NULL;
400 rsc_history_t *entry = NULL;
401 active_op_t *pending = NULL;
402
403 crm_debug("Checking for active resources before exit");
404
405 if (cur_state == S_TERMINATE) {
406 log_level = LOG_ERR;
407 when = "shutdown";
408
409 } else if (pcmk_is_set(controld_globals.fsa_input_register, R_SHUTDOWN)) {
410 when = "shutdown... waiting";
411 }
412
413 if ((lrm_state->active_ops != NULL) && lrm_state_is_connected(lrm_state)) {
414 guint removed = g_hash_table_foreach_remove(lrm_state->active_ops,
415 stop_recurring_actions,
416 lrm_state);
417 guint nremaining = g_hash_table_size(lrm_state->active_ops);
418
419 if (removed || nremaining) {
420 crm_notice("Stopped %u recurring operation%s at %s (%u remaining)",
421 removed, pcmk__plural_s(removed), when, nremaining);
422 }
423 }
424
425 if (lrm_state->active_ops != NULL) {
426 g_hash_table_iter_init(&gIter, lrm_state->active_ops);
427 while (g_hash_table_iter_next(&gIter, NULL, (void **)&pending)) {
428
429 if (pending->interval_ms == 0) {
430 counter++;
431 }
432 }
433 }
434
435 if (counter > 0) {
436 do_crm_log(log_level, "%d pending executor operation%s at %s",
437 counter, pcmk__plural_s(counter), when);
438
439 if ((cur_state == S_TERMINATE)
440 || !pcmk_is_set(controld_globals.fsa_input_register,
441 R_SENT_RSC_STOP)) {
442 g_hash_table_iter_init(&gIter, lrm_state->active_ops);
443 while (g_hash_table_iter_next(&gIter, (gpointer*)&key, (gpointer*)&pending)) {
444 do_crm_log(log_level, "Pending action: %s (%s)", key, pending->op_key);
445 }
446
447 } else {
448 rc = FALSE;
449 }
450 return rc;
451 }
452
453 if (lrm_state->resource_history == NULL) {
454 return rc;
455 }
456
457 if (pcmk_is_set(controld_globals.fsa_input_register, R_SHUTDOWN)) {
458
459 when = "shutdown";
460 }
461
462 counter = 0;
463 g_hash_table_iter_init(&gIter, lrm_state->resource_history);
464 while (g_hash_table_iter_next(&gIter, NULL, (gpointer*)&entry)) {
465 if (is_rsc_active(lrm_state, entry->id) == FALSE) {
466 continue;
467 }
468
469 counter++;
470 if (log_level == LOG_ERR) {
471 crm_info("Found %s active at %s", entry->id, when);
472 } else {
473 crm_trace("Found %s active at %s", entry->id, when);
474 }
475 if (lrm_state->active_ops != NULL) {
476 GHashTableIter hIter;
477
478 g_hash_table_iter_init(&hIter, lrm_state->active_ops);
479 while (g_hash_table_iter_next(&hIter, (gpointer*)&key, (gpointer*)&pending)) {
480 if (pcmk__str_eq(entry->id, pending->rsc_id, pcmk__str_none)) {
481 crm_notice("%sction %s (%s) incomplete at %s",
482 pending->interval_ms == 0 ? "A" : "Recurring a",
483 key, pending->op_key, when);
484 }
485 }
486 }
487 }
488
489 if (counter) {
490 crm_err("%d resource%s active at %s",
491 counter, (counter == 1)? " was" : "s were", when);
492 }
493
494 return rc;
495 }
496
497 static gboolean
498 is_rsc_active(lrm_state_t * lrm_state, const char *rsc_id)
499 {
500 rsc_history_t *entry = NULL;
501
502 entry = g_hash_table_lookup(lrm_state->resource_history, rsc_id);
503 if (entry == NULL || entry->last == NULL) {
504 return FALSE;
505 }
506
507 crm_trace("Processing %s: %s.%d=%d", rsc_id, entry->last->op_type,
508 entry->last->interval_ms, entry->last->rc);
509 if ((entry->last->rc == PCMK_OCF_OK)
510 && pcmk__str_eq(entry->last->op_type, PCMK_ACTION_STOP,
511 pcmk__str_casei)) {
512 return FALSE;
513
514 } else if (entry->last->rc == PCMK_OCF_OK
515 && pcmk__str_eq(entry->last->op_type, PCMK_ACTION_MIGRATE_TO,
516 pcmk__str_casei)) {
517
518 return FALSE;
519
520 } else if (entry->last->rc == PCMK_OCF_NOT_RUNNING) {
521 return FALSE;
522
523 } else if ((entry->last->interval_ms == 0)
524 && (entry->last->rc == PCMK_OCF_NOT_CONFIGURED)) {
525
526 return FALSE;
527 }
528
529 return TRUE;
530 }
531
532 static gboolean
533 build_active_RAs(lrm_state_t * lrm_state, xmlNode * rsc_list)
534 {
535 GHashTableIter iter;
536 rsc_history_t *entry = NULL;
537
538 g_hash_table_iter_init(&iter, lrm_state->resource_history);
539 while (g_hash_table_iter_next(&iter, NULL, (void **)&entry)) {
540
541 GList *gIter = NULL;
542 xmlNode *xml_rsc = pcmk__xe_create(rsc_list, PCMK__XE_LRM_RESOURCE);
543
544 crm_xml_add(xml_rsc, PCMK_XA_ID, entry->id);
545 crm_xml_add(xml_rsc, PCMK_XA_TYPE, entry->rsc.type);
546 crm_xml_add(xml_rsc, PCMK_XA_CLASS, entry->rsc.standard);
547 crm_xml_add(xml_rsc, PCMK_XA_PROVIDER, entry->rsc.provider);
548
549 if (entry->last && entry->last->params) {
550 static const char *name = CRM_META "_" PCMK__META_CONTAINER;
551 const char *container = g_hash_table_lookup(entry->last->params,
552 name);
553
554 if (container) {
555 crm_trace("Resource %s is a part of container resource %s", entry->id, container);
556 crm_xml_add(xml_rsc, PCMK__META_CONTAINER, container);
557 }
558 }
559 controld_add_resource_history_xml(xml_rsc, &(entry->rsc), entry->failed,
560 lrm_state->node_name);
561 controld_add_resource_history_xml(xml_rsc, &(entry->rsc), entry->last,
562 lrm_state->node_name);
563 for (gIter = entry->recurring_op_list; gIter != NULL; gIter = gIter->next) {
564 controld_add_resource_history_xml(xml_rsc, &(entry->rsc), gIter->data,
565 lrm_state->node_name);
566 }
567 }
568
569 return FALSE;
570 }
571
572 xmlNode *
573 controld_query_executor_state(void)
574 {
575
576 xmlNode *xml_state = NULL;
577 xmlNode *xml_data = NULL;
578 xmlNode *rsc_list = NULL;
579 pcmk__node_status_t *peer = NULL;
580 lrm_state_t *lrm_state = controld_get_executor_state(NULL, false);
581
582 if (!lrm_state) {
583 crm_err("Could not get executor state for local node");
584 return NULL;
585 }
586
587 peer = pcmk__get_node(0, lrm_state->node_name, NULL, pcmk__node_search_any);
588 CRM_CHECK(peer != NULL, return NULL);
589
590 xml_state = create_node_state_update(peer,
591 controld_node_update_cluster
592 |controld_node_update_peer,
593 NULL, __func__);
594 if (xml_state == NULL) {
595 return NULL;
596 }
597
598 xml_data = pcmk__xe_create(xml_state, PCMK__XE_LRM);
599 crm_xml_add(xml_data, PCMK_XA_ID, peer->xml_id);
600 rsc_list = pcmk__xe_create(xml_data, PCMK__XE_LRM_RESOURCES);
601
602
603 build_active_RAs(lrm_state, rsc_list);
604
605 crm_log_xml_trace(xml_state, "Current executor state");
606
607 return xml_state;
608 }
609
610
611
612
613
614
615
616
617 void
618 controld_rc2event(lrmd_event_data_t *event, int rc)
619 {
620
621
622
623 switch (rc) {
624 case pcmk_rc_ok:
625 lrmd__set_result(event, PCMK_OCF_OK, PCMK_EXEC_DONE, NULL);
626 break;
627 case EACCES:
628 lrmd__set_result(event, PCMK_OCF_INSUFFICIENT_PRIV,
629 PCMK_EXEC_ERROR, NULL);
630 break;
631 default:
632 lrmd__set_result(event, PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_ERROR,
633 NULL);
634 break;
635 }
636 }
637
638
639
640
641
642
643
644
645
646
647
648
649 void
650 controld_trigger_delete_refresh(const char *from_sys, const char *rsc_id)
651 {
652 if (!pcmk__str_eq(from_sys, CRM_SYSTEM_TENGINE, pcmk__str_casei)) {
653 char *now_s = crm_strdup_printf("%lld", (long long) time(NULL));
654
655 crm_debug("Triggering a refresh after %s cleaned %s", from_sys, rsc_id);
656 cib__update_node_attr(controld_globals.logger_out,
657 controld_globals.cib_conn, cib_none,
658 PCMK_XE_CRM_CONFIG, NULL, NULL, NULL, NULL,
659 "last-lrm-refresh", now_s, NULL, NULL);
660 free(now_s);
661 }
662 }
663
664 static void
665 notify_deleted(lrm_state_t * lrm_state, ha_msg_input_t * input, const char *rsc_id, int rc)
666 {
667 lrmd_event_data_t *op = NULL;
668 const char *from_sys = crm_element_value(input->msg, PCMK__XA_CRM_SYS_FROM);
669 const char *from_host = crm_element_value(input->msg, PCMK__XA_SRC);
670
671 crm_info("Notifying %s on %s that %s was%s deleted",
672 from_sys, (from_host? from_host : "localhost"), rsc_id,
673 ((rc == pcmk_ok)? "" : " not"));
674 op = construct_op(lrm_state, input->xml, rsc_id, PCMK_ACTION_DELETE);
675 controld_rc2event(op, pcmk_legacy2rc(rc));
676 controld_ack_event_directly(from_host, from_sys, NULL, op, rsc_id);
677 lrmd_free_event(op);
678 controld_trigger_delete_refresh(from_sys, rsc_id);
679 }
680
681 static gboolean
682 lrm_remove_deleted_rsc(gpointer key, gpointer value, gpointer user_data)
683 {
684 struct delete_event_s *event = user_data;
685 struct pending_deletion_op_s *op = value;
686
687 if (pcmk__str_eq(event->rsc, op->rsc, pcmk__str_none)) {
688 notify_deleted(event->lrm_state, op->input, event->rsc, event->rc);
689 return TRUE;
690 }
691 return FALSE;
692 }
693
694 static gboolean
695 lrm_remove_deleted_op(gpointer key, gpointer value, gpointer user_data)
696 {
697 const char *rsc = user_data;
698 active_op_t *pending = value;
699
700 if (pcmk__str_eq(rsc, pending->rsc_id, pcmk__str_none)) {
701 crm_info("Removing op %s:%d for deleted resource %s",
702 pending->op_key, pending->call_id, rsc);
703 return TRUE;
704 }
705 return FALSE;
706 }
707
708 static void
709 delete_rsc_entry(lrm_state_t *lrm_state, ha_msg_input_t *input,
710 const char *rsc_id, GHashTableIter *rsc_iter, int rc,
711 const char *user_name, bool from_cib)
712 {
713 struct delete_event_s event;
714
715 CRM_CHECK(rsc_id != NULL, return);
716
717 if (rc == pcmk_ok) {
718 char *rsc_id_copy = pcmk__str_copy(rsc_id);
719
720 if (rsc_iter) {
721 g_hash_table_iter_remove(rsc_iter);
722 } else {
723 g_hash_table_remove(lrm_state->resource_history, rsc_id_copy);
724 }
725
726 if (from_cib) {
727 controld_delete_resource_history(rsc_id_copy, lrm_state->node_name,
728 user_name, crmd_cib_smart_opt());
729 }
730 g_hash_table_foreach_remove(lrm_state->active_ops,
731 lrm_remove_deleted_op, rsc_id_copy);
732 free(rsc_id_copy);
733 }
734
735 if (input) {
736 notify_deleted(lrm_state, input, rsc_id, rc);
737 }
738
739 event.rc = rc;
740 event.rsc = rsc_id;
741 event.lrm_state = lrm_state;
742 g_hash_table_foreach_remove(lrm_state->deletion_ops, lrm_remove_deleted_rsc, &event);
743 }
744
745 static inline gboolean
746 last_failed_matches_op(rsc_history_t *entry, const char *op, guint interval_ms)
747 {
748 if (entry == NULL) {
749 return FALSE;
750 }
751 if (op == NULL) {
752 return TRUE;
753 }
754 return (pcmk__str_eq(op, entry->failed->op_type, pcmk__str_casei)
755 && (interval_ms == entry->failed->interval_ms));
756 }
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771 void
772 lrm_clear_last_failure(const char *rsc_id, const char *node_name,
773 const char *operation, guint interval_ms)
774 {
775 lrm_state_t *lrm_state = controld_get_executor_state(node_name, false);
776
777 if (lrm_state == NULL) {
778 return;
779 }
780 if (lrm_state->resource_history != NULL) {
781 rsc_history_t *entry = g_hash_table_lookup(lrm_state->resource_history,
782 rsc_id);
783
784 if (last_failed_matches_op(entry, operation, interval_ms)) {
785 lrmd_free_event(entry->failed);
786 entry->failed = NULL;
787 }
788 }
789 }
790
791
792 static gboolean
793 cancel_op(lrm_state_t * lrm_state, const char *rsc_id, const char *key, int op, gboolean remove)
794 {
795 int rc = pcmk_ok;
796 char *local_key = NULL;
797 active_op_t *pending = NULL;
798
799 CRM_CHECK(op != 0, return FALSE);
800 CRM_CHECK(rsc_id != NULL, return FALSE);
801 if (key == NULL) {
802 local_key = make_stop_id(rsc_id, op);
803 key = local_key;
804 }
805 pending = g_hash_table_lookup(lrm_state->active_ops, key);
806
807 if (pending) {
808 if (remove && !pcmk_is_set(pending->flags, active_op_remove)) {
809 controld_set_active_op_flags(pending, active_op_remove);
810 crm_debug("Scheduling %s for removal", key);
811 }
812
813 if (pcmk_is_set(pending->flags, active_op_cancelled)) {
814 crm_debug("Operation %s already cancelled", key);
815 free(local_key);
816 return FALSE;
817 }
818 controld_set_active_op_flags(pending, active_op_cancelled);
819
820 } else {
821 crm_info("No pending op found for %s", key);
822 free(local_key);
823 return FALSE;
824 }
825
826 crm_debug("Cancelling op %d for %s (%s)", op, rsc_id, key);
827 rc = lrm_state_cancel(lrm_state, pending->rsc_id, pending->op_type,
828 pending->interval_ms);
829 if (rc == pcmk_ok) {
830 crm_debug("Op %d for %s (%s): cancelled", op, rsc_id, key);
831 free(local_key);
832 return TRUE;
833 }
834
835 crm_debug("Op %d for %s (%s): Nothing to cancel", op, rsc_id, key);
836
837
838
839
840
841
842
843
844
845 free(local_key);
846 return FALSE;
847 }
848
849 struct cancel_data {
850 gboolean done;
851 gboolean remove;
852 const char *key;
853 lrmd_rsc_info_t *rsc;
854 lrm_state_t *lrm_state;
855 };
856
857 static gboolean
858 cancel_action_by_key(gpointer key, gpointer value, gpointer user_data)
859 {
860 gboolean remove = FALSE;
861 struct cancel_data *data = user_data;
862 active_op_t *op = value;
863
864 if (pcmk__str_eq(op->op_key, data->key, pcmk__str_none)) {
865 data->done = TRUE;
866 remove = !cancel_op(data->lrm_state, data->rsc->id, key, op->call_id, data->remove);
867 }
868 return remove;
869 }
870
871 static gboolean
872 cancel_op_key(lrm_state_t * lrm_state, lrmd_rsc_info_t * rsc, const char *key, gboolean remove)
873 {
874 guint removed = 0;
875 struct cancel_data data;
876
877 CRM_CHECK(rsc != NULL, return FALSE);
878 CRM_CHECK(key != NULL, return FALSE);
879
880 data.key = key;
881 data.rsc = rsc;
882 data.done = FALSE;
883 data.remove = remove;
884 data.lrm_state = lrm_state;
885
886 removed = g_hash_table_foreach_remove(lrm_state->active_ops,
887 cancel_action_by_key, &data);
888 crm_trace("Removed %u op cache entries, new size: %u",
889 removed, g_hash_table_size(lrm_state->active_ops));
890 return data.done;
891 }
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910 static int
911 get_lrm_resource(lrm_state_t *lrm_state, const xmlNode *rsc_xml,
912 gboolean do_create, lrmd_rsc_info_t **rsc_info)
913 {
914 const char *id = pcmk__xe_id(rsc_xml);
915
916 CRM_CHECK(lrm_state && rsc_xml && rsc_info, return -EINVAL);
917 CRM_CHECK(id, return -EINVAL);
918
919 if (lrm_state_is_connected(lrm_state) == FALSE) {
920 return -ENOTCONN;
921 }
922
923 crm_trace("Retrieving resource information for %s from the executor", id);
924 *rsc_info = lrm_state_get_rsc_info(lrm_state, id, 0);
925
926
927 if (!*rsc_info) {
928 const char *long_id = crm_element_value(rsc_xml, PCMK__XA_LONG_ID);
929
930 if (long_id) {
931 *rsc_info = lrm_state_get_rsc_info(lrm_state, long_id, 0);
932 }
933 }
934
935 if ((*rsc_info == NULL) && do_create) {
936 const char *class = crm_element_value(rsc_xml, PCMK_XA_CLASS);
937 const char *provider = crm_element_value(rsc_xml, PCMK_XA_PROVIDER);
938 const char *type = crm_element_value(rsc_xml, PCMK_XA_TYPE);
939 int rc;
940
941 crm_trace("Registering resource %s with the executor", id);
942 rc = lrm_state_register_rsc(lrm_state, id, class, provider, type,
943 lrmd_opt_drop_recurring);
944 if (rc != pcmk_ok) {
945 fsa_data_t *msg_data = NULL;
946
947 crm_err("Could not register resource %s with the executor on %s: %s "
948 QB_XS " rc=%d",
949 id, lrm_state->node_name, pcmk_strerror(rc), rc);
950
951
952
953
954
955 if (lrm_state_is_local(lrm_state) == TRUE) {
956 register_fsa_error(C_FSA_INTERNAL, I_FAIL, NULL);
957 }
958 return rc;
959 }
960
961 *rsc_info = lrm_state_get_rsc_info(lrm_state, id, 0);
962 }
963 return *rsc_info? pcmk_ok : -ENODEV;
964 }
965
966 static void
967 delete_resource(lrm_state_t *lrm_state, const char *id, lrmd_rsc_info_t *rsc,
968 GHashTableIter *iter, const char *sys, const char *user,
969 ha_msg_input_t *request, bool unregister, bool from_cib)
970 {
971 int rc = pcmk_ok;
972
973 crm_info("Removing resource %s from executor for %s%s%s",
974 id, sys, (user? " as " : ""), (user? user : ""));
975
976 if (rsc && unregister) {
977 rc = lrm_state_unregister_rsc(lrm_state, id, 0);
978 }
979
980 if (rc == pcmk_ok) {
981 crm_trace("Resource %s deleted from executor", id);
982 } else if (rc == -EINPROGRESS) {
983 crm_info("Deletion of resource '%s' from executor is pending", id);
984 if (request) {
985 struct pending_deletion_op_s *op = NULL;
986 char *ref = crm_element_value_copy(request->msg, PCMK_XA_REFERENCE);
987
988 op = pcmk__assert_alloc(1, sizeof(struct pending_deletion_op_s));
989 op->rsc = pcmk__str_copy(rsc->id);
990 op->input = copy_ha_msg_input(request);
991 g_hash_table_insert(lrm_state->deletion_ops, ref, op);
992 }
993 return;
994 } else {
995 crm_warn("Could not delete '%s' from executor for %s%s%s: %s "
996 QB_XS " rc=%d", id, sys, (user? " as " : ""),
997 (user? user : ""), pcmk_strerror(rc), rc);
998 }
999
1000 delete_rsc_entry(lrm_state, request, id, iter, rc, user, from_cib);
1001 }
1002
1003 static int
1004 get_fake_call_id(lrm_state_t *lrm_state, const char *rsc_id)
1005 {
1006 int call_id = 999999999;
1007 rsc_history_t *entry = NULL;
1008
1009 if(lrm_state) {
1010 entry = g_hash_table_lookup(lrm_state->resource_history, rsc_id);
1011 }
1012
1013
1014
1015
1016 if (entry) {
1017 call_id = entry->last_callid + 1;
1018 }
1019
1020 if (call_id < 0) {
1021 call_id = 1;
1022 }
1023 return call_id;
1024 }
1025
1026 static void
1027 fake_op_status(lrm_state_t *lrm_state, lrmd_event_data_t *op, int op_status,
1028 enum ocf_exitcode op_exitcode, const char *exit_reason)
1029 {
1030 op->call_id = get_fake_call_id(lrm_state, op->rsc_id);
1031 op->t_run = time(NULL);
1032 op->t_rcchange = op->t_run;
1033 lrmd__set_result(op, op_exitcode, op_status, exit_reason);
1034 }
1035
1036 static void
1037 force_reprobe(lrm_state_t *lrm_state, const char *from_sys,
1038 const char *from_host, const char *user_name,
1039 gboolean is_remote_node, bool reprobe_all_nodes)
1040 {
1041 GHashTableIter gIter;
1042 rsc_history_t *entry = NULL;
1043
1044 crm_info("Clearing resource history on node %s", lrm_state->node_name);
1045 g_hash_table_iter_init(&gIter, lrm_state->resource_history);
1046 while (g_hash_table_iter_next(&gIter, NULL, (void **)&entry)) {
1047
1048
1049
1050 bool unregister = true;
1051
1052 if (is_remote_lrmd_ra(NULL, NULL, entry->id)) {
1053 unregister = false;
1054
1055 if (reprobe_all_nodes) {
1056 lrm_state_t *remote_lrm_state =
1057 controld_get_executor_state(entry->id, false);
1058
1059 if (remote_lrm_state != NULL) {
1060
1061
1062
1063 force_reprobe(remote_lrm_state, from_sys, from_host,
1064 user_name, TRUE, reprobe_all_nodes);
1065 }
1066 }
1067 }
1068
1069
1070
1071
1072 delete_resource(lrm_state, entry->id, &entry->rsc, &gIter, from_sys,
1073 user_name, NULL, unregister, false);
1074 }
1075
1076
1077 controld_delete_node_state(lrm_state->node_name, controld_section_lrm,
1078 cib_none);
1079 }
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095 static void
1096 synthesize_lrmd_failure(lrm_state_t *lrm_state, const xmlNode *action,
1097 int op_status, enum ocf_exitcode rc,
1098 const char *exit_reason)
1099 {
1100 lrmd_event_data_t *op = NULL;
1101 const char *operation = crm_element_value(action, PCMK_XA_OPERATION);
1102 const char *target_node = crm_element_value(action, PCMK__META_ON_NODE);
1103 xmlNode *xml_rsc = pcmk__xe_first_child(action, PCMK_XE_PRIMITIVE, NULL,
1104 NULL);
1105
1106 if ((xml_rsc == NULL) || (pcmk__xe_id(xml_rsc) == NULL)) {
1107
1108 crm_info("Can't fake %s failure (%d) on %s without resource configuration",
1109 crm_element_value(action, PCMK__XA_OPERATION_KEY), rc,
1110 target_node);
1111 return;
1112
1113 } else if(operation == NULL) {
1114
1115 crm_info("Can't fake %s failure (%d) on %s without operation",
1116 pcmk__xe_id(xml_rsc), rc, target_node);
1117 return;
1118 }
1119
1120 op = construct_op(lrm_state, action, pcmk__xe_id(xml_rsc), operation);
1121
1122 if (pcmk__str_eq(operation, PCMK_ACTION_NOTIFY, pcmk__str_casei)) {
1123
1124 fake_op_status(lrm_state, op, PCMK_EXEC_DONE, PCMK_OCF_OK, NULL);
1125 } else {
1126 fake_op_status(lrm_state, op, op_status, rc, exit_reason);
1127 }
1128
1129 crm_info("Faking " PCMK__OP_FMT " result (%d) on %s",
1130 op->rsc_id, op->op_type, op->interval_ms, op->rc, target_node);
1131
1132
1133 process_lrm_event(lrm_state, op, NULL, action);
1134 lrmd_free_event(op);
1135 }
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146 static const char *
1147 lrm_op_target(const xmlNode *xml)
1148 {
1149 const char *target = NULL;
1150
1151 if (xml) {
1152 target = crm_element_value(xml, PCMK__META_ON_NODE);
1153 }
1154 if (target == NULL) {
1155 target = controld_globals.cluster->priv->node_name;
1156 }
1157 return target;
1158 }
1159
1160 static void
1161 fail_lrm_resource(xmlNode *xml, lrm_state_t *lrm_state, const char *user_name,
1162 const char *from_host, const char *from_sys)
1163 {
1164 lrmd_event_data_t *op = NULL;
1165 lrmd_rsc_info_t *rsc = NULL;
1166 xmlNode *xml_rsc = pcmk__xe_first_child(xml, PCMK_XE_PRIMITIVE, NULL, NULL);
1167
1168 CRM_CHECK(xml_rsc != NULL, return);
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178 op = construct_op(lrm_state, xml, pcmk__xe_id(xml_rsc), "asyncmon");
1179
1180 free((char*) op->user_data);
1181 op->user_data = NULL;
1182 op->interval_ms = 0;
1183
1184 if (user_name && !pcmk__is_privileged(user_name)) {
1185 crm_err("%s does not have permission to fail %s",
1186 user_name, pcmk__xe_id(xml_rsc));
1187 fake_op_status(lrm_state, op, PCMK_EXEC_ERROR,
1188 PCMK_OCF_INSUFFICIENT_PRIV,
1189 "Unprivileged user cannot fail resources");
1190 controld_ack_event_directly(from_host, from_sys, NULL, op,
1191 pcmk__xe_id(xml_rsc));
1192 lrmd_free_event(op);
1193 return;
1194 }
1195
1196
1197 if (get_lrm_resource(lrm_state, xml_rsc, TRUE, &rsc) == pcmk_ok) {
1198 crm_info("Failing resource %s...", rsc->id);
1199 fake_op_status(lrm_state, op, PCMK_EXEC_DONE, PCMK_OCF_UNKNOWN_ERROR,
1200 "Simulated failure");
1201 process_lrm_event(lrm_state, op, NULL, xml);
1202 op->rc = PCMK_OCF_OK;
1203 lrmd_free_rsc_info(rsc);
1204
1205 } else {
1206 crm_info("Cannot find/create resource in order to fail it...");
1207 crm_log_xml_warn(xml, "bad input");
1208 fake_op_status(lrm_state, op, PCMK_EXEC_ERROR, PCMK_OCF_UNKNOWN_ERROR,
1209 "Cannot fail unknown resource");
1210 }
1211
1212 controld_ack_event_directly(from_host, from_sys, NULL, op,
1213 pcmk__xe_id(xml_rsc));
1214 lrmd_free_event(op);
1215 }
1216
1217 static void
1218 handle_reprobe_op(lrm_state_t *lrm_state, xmlNode *msg, const char *from_sys,
1219 const char *from_host, const char *user_name,
1220 gboolean is_remote_node, bool reprobe_all_nodes)
1221 {
1222 crm_notice("Forcing the status of all resources to be redetected");
1223 force_reprobe(lrm_state, from_sys, from_host, user_name, is_remote_node,
1224 reprobe_all_nodes);
1225
1226 if (!pcmk__strcase_any_of(from_sys, CRM_SYSTEM_PENGINE, CRM_SYSTEM_TENGINE, NULL)) {
1227 xmlNode *reply = pcmk__new_reply(msg, NULL);
1228
1229 crm_debug("ACK'ing re-probe from %s (%s)", from_sys, from_host);
1230
1231 if (relay_message(reply, TRUE) == FALSE) {
1232 crm_log_xml_err(reply, "Unable to route reply");
1233 }
1234 pcmk__xml_free(reply);
1235 }
1236 }
1237
1238 static bool do_lrm_cancel(ha_msg_input_t *input, lrm_state_t *lrm_state,
1239 lrmd_rsc_info_t *rsc, const char *from_host, const char *from_sys)
1240 {
1241 char *op_key = NULL;
1242 char *meta_key = NULL;
1243 int call = 0;
1244 const char *call_id = NULL;
1245 const char *op_task = NULL;
1246 guint interval_ms = 0;
1247 gboolean in_progress = FALSE;
1248 xmlNode *params = pcmk__xe_first_child(input->xml, PCMK__XE_ATTRIBUTES,
1249 NULL, NULL);
1250
1251 CRM_CHECK(params != NULL, return FALSE);
1252
1253 meta_key = crm_meta_name(PCMK_XA_OPERATION);
1254 op_task = crm_element_value(params, meta_key);
1255 free(meta_key);
1256 CRM_CHECK(op_task != NULL, return FALSE);
1257
1258 meta_key = crm_meta_name(PCMK_META_INTERVAL);
1259 if (crm_element_value_ms(params, meta_key, &interval_ms) != pcmk_ok) {
1260 free(meta_key);
1261 return FALSE;
1262 }
1263 free(meta_key);
1264
1265 op_key = pcmk__op_key(rsc->id, op_task, interval_ms);
1266
1267 meta_key = crm_meta_name(PCMK__XA_CALL_ID);
1268 call_id = crm_element_value(params, meta_key);
1269 free(meta_key);
1270
1271 crm_debug("Scheduler requested op %s (call=%s) be cancelled",
1272 op_key, (call_id? call_id : "NA"));
1273 pcmk__scan_min_int(call_id, &call, 0);
1274 if (call == 0) {
1275
1276 in_progress = cancel_op_key(lrm_state, rsc, op_key, TRUE);
1277
1278 } else {
1279
1280 in_progress = cancel_op(lrm_state, rsc->id, NULL, call, TRUE);
1281 }
1282
1283
1284 if (!in_progress || is_remote_lrmd_ra(NULL, NULL, rsc->id)) {
1285 char *op_id = make_stop_id(rsc->id, call);
1286
1287 if (is_remote_lrmd_ra(NULL, NULL, rsc->id) == FALSE) {
1288 crm_info("Nothing known about operation %d for %s", call, op_key);
1289 }
1290 controld_delete_action_history_by_key(rsc->id, lrm_state->node_name,
1291 op_key, call);
1292 send_task_ok_ack(lrm_state, input, rsc->id, rsc, op_task,
1293 from_host, from_sys);
1294
1295
1296 if (lrm_state->active_ops != NULL) {
1297 g_hash_table_remove(lrm_state->active_ops, op_id);
1298 }
1299 free(op_id);
1300 }
1301
1302 free(op_key);
1303 return TRUE;
1304 }
1305
1306 static void
1307 do_lrm_delete(ha_msg_input_t *input, lrm_state_t *lrm_state,
1308 lrmd_rsc_info_t *rsc, const char *from_sys, const char *from_host,
1309 bool crm_rsc_delete, const char *user_name)
1310 {
1311 bool unregister = true;
1312 int cib_rc = controld_delete_resource_history(rsc->id, lrm_state->node_name,
1313 user_name,
1314 cib_dryrun|cib_sync_call);
1315
1316 if (cib_rc != pcmk_rc_ok) {
1317 lrmd_event_data_t *op = NULL;
1318
1319 op = construct_op(lrm_state, input->xml, rsc->id, PCMK_ACTION_DELETE);
1320
1321
1322
1323
1324 lrmd__set_result(op, pcmk_rc2ocf(cib_rc), PCMK_EXEC_ERROR, NULL);
1325 controld_ack_event_directly(from_host, from_sys, NULL, op, rsc->id);
1326 lrmd_free_event(op);
1327 return;
1328 }
1329
1330 if (crm_rsc_delete && is_remote_lrmd_ra(NULL, NULL, rsc->id)) {
1331 unregister = false;
1332 }
1333
1334 delete_resource(lrm_state, rsc->id, rsc, NULL, from_sys,
1335 user_name, input, unregister, true);
1336 }
1337
1338
1339 struct metadata_cb_data {
1340 lrmd_rsc_info_t *rsc;
1341 xmlNode *input_xml;
1342 };
1343
1344 static struct metadata_cb_data *
1345 new_metadata_cb_data(lrmd_rsc_info_t *rsc, xmlNode *input_xml)
1346 {
1347 struct metadata_cb_data *data = NULL;
1348
1349 data = pcmk__assert_alloc(1, sizeof(struct metadata_cb_data));
1350 data->input_xml = pcmk__xml_copy(NULL, input_xml);
1351 data->rsc = lrmd_copy_rsc_info(rsc);
1352 return data;
1353 }
1354
1355 static void
1356 free_metadata_cb_data(struct metadata_cb_data *data)
1357 {
1358 lrmd_free_rsc_info(data->rsc);
1359 pcmk__xml_free(data->input_xml);
1360 free(data);
1361 }
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371 static void
1372 metadata_complete(int pid, const pcmk__action_result_t *result, void *user_data)
1373 {
1374 struct metadata_cb_data *data = (struct metadata_cb_data *) user_data;
1375
1376 struct ra_metadata_s *md = NULL;
1377 lrm_state_t *lrm_state =
1378 controld_get_executor_state(lrm_op_target(data->input_xml), false);
1379
1380 if ((lrm_state != NULL) && pcmk__result_ok(result)) {
1381 md = controld_cache_metadata(lrm_state->metadata_cache, data->rsc,
1382 result->action_stdout);
1383 }
1384 if (!pcmk_is_set(controld_globals.fsa_input_register, R_HA_DISCONNECTED)) {
1385 do_lrm_rsc_op(lrm_state, data->rsc, data->input_xml, md);
1386 }
1387 free_metadata_cb_data(data);
1388 }
1389
1390
1391 void
1392 do_lrm_invoke(long long action,
1393 enum crmd_fsa_cause cause,
1394 enum crmd_fsa_state cur_state,
1395 enum crmd_fsa_input current_input, fsa_data_t * msg_data)
1396 {
1397 lrm_state_t *lrm_state = NULL;
1398 const char *crm_op = NULL;
1399 const char *from_sys = NULL;
1400 const char *from_host = NULL;
1401 const char *operation = NULL;
1402 ha_msg_input_t *input = fsa_typed_data(fsa_dt_ha_msg);
1403 const char *user_name = NULL;
1404 const char *target_node = lrm_op_target(input->xml);
1405 gboolean is_remote_node = FALSE;
1406 bool crm_rsc_delete = FALSE;
1407
1408
1409 is_remote_node = !controld_is_local_node(target_node);
1410
1411 lrm_state = controld_get_executor_state(target_node, false);
1412 if ((lrm_state == NULL) && is_remote_node) {
1413 crm_err("Failing action because local node has never had connection to remote node %s",
1414 target_node);
1415 synthesize_lrmd_failure(NULL, input->xml, PCMK_EXEC_NOT_CONNECTED,
1416 PCMK_OCF_UNKNOWN_ERROR,
1417 "Local node has no connection to remote");
1418 return;
1419 }
1420 pcmk__assert(lrm_state != NULL);
1421
1422 user_name = pcmk__update_acl_user(input->msg, PCMK__XA_CRM_USER, NULL);
1423 crm_op = crm_element_value(input->msg, PCMK__XA_CRM_TASK);
1424 from_sys = crm_element_value(input->msg, PCMK__XA_CRM_SYS_FROM);
1425 if (!pcmk__str_eq(from_sys, CRM_SYSTEM_TENGINE, pcmk__str_none)) {
1426 from_host = crm_element_value(input->msg, PCMK__XA_SRC);
1427 }
1428
1429 if (pcmk__str_eq(crm_op, PCMK_ACTION_LRM_DELETE, pcmk__str_none)) {
1430 if (!pcmk__str_eq(from_sys, CRM_SYSTEM_TENGINE, pcmk__str_none)) {
1431 crm_rsc_delete = TRUE;
1432 }
1433 operation = PCMK_ACTION_DELETE;
1434
1435 } else if (input->xml != NULL) {
1436 operation = crm_element_value(input->xml, PCMK_XA_OPERATION);
1437 }
1438
1439 CRM_CHECK(!pcmk__str_empty(crm_op) || !pcmk__str_empty(operation), return);
1440
1441 crm_trace("'%s' execution request from %s as %s user",
1442 pcmk__s(crm_op, operation),
1443 pcmk__s(from_sys, "unknown subsystem"),
1444 pcmk__s(user_name, "current"));
1445
1446 if (pcmk__str_eq(crm_op, CRM_OP_LRM_FAIL, pcmk__str_none)) {
1447 fail_lrm_resource(input->xml, lrm_state, user_name, from_host,
1448 from_sys);
1449
1450 } else if (pcmk__str_eq(crm_op, CRM_OP_REPROBE, pcmk__str_none)
1451 || pcmk__str_eq(operation, CRM_OP_REPROBE, pcmk__str_none)) {
1452 const char *raw_target = NULL;
1453
1454 if (input->xml != NULL) {
1455
1456 raw_target = crm_element_value(input->xml, PCMK__META_ON_NODE);
1457 }
1458 handle_reprobe_op(lrm_state, input->msg, from_sys, from_host, user_name,
1459 is_remote_node, (raw_target == NULL));
1460
1461 } else if (operation != NULL) {
1462 lrmd_rsc_info_t *rsc = NULL;
1463 xmlNode *xml_rsc = pcmk__xe_first_child(input->xml, PCMK_XE_PRIMITIVE,
1464 NULL, NULL);
1465 gboolean create_rsc = !pcmk__str_eq(operation, PCMK_ACTION_DELETE,
1466 pcmk__str_none);
1467 int rc;
1468
1469
1470 CRM_CHECK((xml_rsc != NULL) && (pcmk__xe_id(xml_rsc) != NULL), return);
1471
1472 rc = get_lrm_resource(lrm_state, xml_rsc, create_rsc, &rsc);
1473 if (rc == -ENOTCONN) {
1474 synthesize_lrmd_failure(lrm_state, input->xml,
1475 PCMK_EXEC_NOT_CONNECTED,
1476 PCMK_OCF_UNKNOWN_ERROR,
1477 "Not connected to remote executor");
1478 return;
1479
1480 } else if ((rc < 0) && !create_rsc) {
1481
1482
1483
1484 crm_debug("Not registering resource '%s' for a %s event "
1485 QB_XS " get-rc=%d (%s) transition-key=%s",
1486 pcmk__xe_id(xml_rsc), operation,
1487 rc, pcmk_strerror(rc), pcmk__xe_id(input->xml));
1488 delete_rsc_entry(lrm_state, input, pcmk__xe_id(xml_rsc), NULL,
1489 pcmk_ok, user_name, true);
1490 return;
1491
1492 } else if (rc == -EINVAL) {
1493
1494 crm_err("Invalid resource definition for %s", pcmk__xe_id(xml_rsc));
1495 crm_log_xml_warn(input->msg, "invalid resource");
1496 synthesize_lrmd_failure(lrm_state, input->xml, PCMK_EXEC_ERROR,
1497 PCMK_OCF_NOT_CONFIGURED,
1498 "Invalid resource definition");
1499 return;
1500
1501 } else if (rc < 0) {
1502
1503 crm_err("Could not register resource '%s' with executor: %s "
1504 QB_XS " rc=%d",
1505 pcmk__xe_id(xml_rsc), pcmk_strerror(rc), rc);
1506 crm_log_xml_warn(input->msg, "failed registration");
1507 synthesize_lrmd_failure(lrm_state, input->xml, PCMK_EXEC_ERROR,
1508 PCMK_OCF_INVALID_PARAM,
1509 "Could not register resource with executor");
1510 return;
1511 }
1512
1513 if (pcmk__str_eq(operation, PCMK_ACTION_CANCEL, pcmk__str_none)) {
1514 if (!do_lrm_cancel(input, lrm_state, rsc, from_host, from_sys)) {
1515 crm_log_xml_warn(input->xml, "Bad command");
1516 }
1517
1518 } else if (pcmk__str_eq(operation, PCMK_ACTION_DELETE,
1519 pcmk__str_none)) {
1520 do_lrm_delete(input, lrm_state, rsc, from_sys, from_host,
1521 crm_rsc_delete, user_name);
1522
1523 } else {
1524 struct ra_metadata_s *md = NULL;
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534 if (strcmp(operation, PCMK_ACTION_START) != 0) {
1535 md = controld_get_rsc_metadata(lrm_state, rsc,
1536 controld_metadata_from_cache);
1537 }
1538
1539 if ((md == NULL) && crm_op_needs_metadata(rsc->standard,
1540 operation)) {
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552 struct metadata_cb_data *data = NULL;
1553
1554 data = new_metadata_cb_data(rsc, input->xml);
1555 crm_info("Retrieving metadata for %s (%s%s%s:%s) asynchronously",
1556 rsc->id, rsc->standard,
1557 ((rsc->provider == NULL)? "" : ":"),
1558 ((rsc->provider == NULL)? "" : rsc->provider),
1559 rsc->type);
1560 (void) lrmd__metadata_async(rsc, metadata_complete,
1561 (void *) data);
1562 } else {
1563 do_lrm_rsc_op(lrm_state, rsc, input->xml, md);
1564 }
1565 }
1566
1567 lrmd_free_rsc_info(rsc);
1568
1569 } else {
1570 crm_err("Invalid execution request: unknown command '%s' (bug?)",
1571 crm_op);
1572 register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL);
1573 }
1574 }
1575
1576 static lrmd_event_data_t *
1577 construct_op(const lrm_state_t *lrm_state, const xmlNode *rsc_op,
1578 const char *rsc_id, const char *operation)
1579 {
1580 lrmd_event_data_t *op = NULL;
1581 const char *op_delay = NULL;
1582 const char *op_timeout = NULL;
1583 GHashTable *params = NULL;
1584
1585 xmlNode *primitive = NULL;
1586 const char *class = NULL;
1587
1588 const char *transition = NULL;
1589
1590 pcmk__assert((rsc_id != NULL) && (operation != NULL));
1591
1592 op = lrmd_new_event(rsc_id, operation, 0);
1593 op->type = lrmd_event_exec_complete;
1594 op->timeout = 0;
1595 op->start_delay = 0;
1596 lrmd__set_result(op, PCMK_OCF_UNKNOWN, PCMK_EXEC_PENDING, NULL);
1597
1598 if (rsc_op == NULL) {
1599 CRM_LOG_ASSERT(pcmk__str_eq(operation, PCMK_ACTION_STOP,
1600 pcmk__str_casei));
1601 op->user_data = NULL;
1602
1603
1604
1605
1606
1607 op->params = pcmk__strkey_table(free, free);
1608
1609 pcmk__insert_dup(op->params, PCMK_XA_CRM_FEATURE_SET, CRM_FEATURE_SET);
1610
1611 crm_trace("Constructed %s op for %s", operation, rsc_id);
1612 return op;
1613 }
1614
1615 params = xml2list(rsc_op);
1616 g_hash_table_remove(params, CRM_META "_" PCMK__META_OP_TARGET_RC);
1617
1618 op_delay = crm_meta_value(params, PCMK_META_START_DELAY);
1619 pcmk__scan_min_int(op_delay, &op->start_delay, 0);
1620
1621 op_timeout = crm_meta_value(params, PCMK_META_TIMEOUT);
1622 pcmk__scan_min_int(op_timeout, &op->timeout, 0);
1623
1624 if (pcmk__guint_from_hash(params, CRM_META "_" PCMK_META_INTERVAL, 0,
1625 &(op->interval_ms)) != pcmk_rc_ok) {
1626 op->interval_ms = 0;
1627 }
1628
1629
1630
1631 primitive = pcmk__xe_first_child(rsc_op, PCMK_XE_PRIMITIVE, NULL, NULL);
1632 class = crm_element_value(primitive, PCMK_XA_CLASS);
1633
1634 if (pcmk_is_set(pcmk_get_ra_caps(class), pcmk_ra_cap_fence_params)
1635 && pcmk__str_eq(operation, PCMK_ACTION_MONITOR, pcmk__str_casei)
1636 && (op->interval_ms > 0)) {
1637
1638 op_timeout = g_hash_table_lookup(params, "pcmk_monitor_timeout");
1639 if (op_timeout != NULL) {
1640 long long timeout_ms = crm_get_msec(op_timeout);
1641
1642 op->timeout = (int) QB_MIN(timeout_ms, INT_MAX);
1643 }
1644 }
1645
1646 if (!pcmk__str_eq(operation, PCMK_ACTION_STOP, pcmk__str_casei)) {
1647 op->params = params;
1648
1649 } else {
1650 rsc_history_t *entry = NULL;
1651
1652 if (lrm_state) {
1653 entry = g_hash_table_lookup(lrm_state->resource_history, rsc_id);
1654 }
1655
1656
1657
1658 if (!entry || !entry->stop_params) {
1659 op->params = params;
1660 } else {
1661
1662
1663 op->params = pcmk__strkey_table(free, free);
1664
1665 g_hash_table_foreach(params, copy_meta_keys, op->params);
1666 g_hash_table_foreach(entry->stop_params, copy_instance_keys, op->params);
1667 g_hash_table_destroy(params);
1668 params = NULL;
1669 }
1670 }
1671
1672
1673 if (op->timeout <= 0) {
1674 op->timeout = op->interval_ms;
1675 }
1676 if (op->start_delay < 0) {
1677 op->start_delay = 0;
1678 }
1679
1680 transition = crm_element_value(rsc_op, PCMK__XA_TRANSITION_KEY);
1681 CRM_CHECK(transition != NULL, return op);
1682
1683 op->user_data = pcmk__str_copy(transition);
1684
1685 if (op->interval_ms != 0) {
1686 if (pcmk__strcase_any_of(operation, PCMK_ACTION_START, PCMK_ACTION_STOP,
1687 NULL)) {
1688 crm_err("Start and Stop actions cannot have an interval: %u",
1689 op->interval_ms);
1690 op->interval_ms = 0;
1691 }
1692 }
1693
1694 crm_trace("Constructed %s op for %s: interval=%u",
1695 operation, rsc_id, op->interval_ms);
1696
1697 return op;
1698 }
1699
1700
1701
1702
1703
1704
1705
1706
1707
1708
1709
1710
1711
1712
1713 void
1714 controld_ack_event_directly(const char *to_host, const char *to_sys,
1715 const lrmd_rsc_info_t *rsc, lrmd_event_data_t *op,
1716 const char *rsc_id)
1717 {
1718 xmlNode *reply = NULL;
1719 xmlNode *update, *iter;
1720 pcmk__node_status_t *peer = NULL;
1721
1722 CRM_CHECK(op != NULL, return);
1723 if (op->rsc_id == NULL) {
1724
1725 pcmk__assert(rsc_id != NULL);
1726 op->rsc_id = pcmk__str_copy(rsc_id);
1727 }
1728 if (to_sys == NULL) {
1729 to_sys = CRM_SYSTEM_TENGINE;
1730 }
1731
1732 peer = controld_get_local_node_status();
1733 update = create_node_state_update(peer, controld_node_update_none, NULL,
1734 __func__);
1735
1736 iter = pcmk__xe_create(update, PCMK__XE_LRM);
1737 crm_xml_add(iter, PCMK_XA_ID, controld_globals.our_uuid);
1738 iter = pcmk__xe_create(iter, PCMK__XE_LRM_RESOURCES);
1739 iter = pcmk__xe_create(iter, PCMK__XE_LRM_RESOURCE);
1740
1741 crm_xml_add(iter, PCMK_XA_ID, op->rsc_id);
1742
1743 controld_add_resource_history_xml(iter, rsc, op,
1744 controld_globals.cluster->priv->node_name);
1745
1746
1747
1748
1749
1750
1751
1752 reply = pcmk__new_message(pcmk_ipc_controld, "direct-ack", CRM_SYSTEM_LRMD,
1753 to_host, to_sys, CRM_OP_INVOKE_LRM, update);
1754
1755 crm_log_xml_trace(update, "[direct ACK]");
1756
1757 crm_debug("ACK'ing resource op " PCMK__OP_FMT " from %s: %s",
1758 op->rsc_id, op->op_type, op->interval_ms, op->user_data,
1759 crm_element_value(reply, PCMK_XA_REFERENCE));
1760
1761 if (relay_message(reply, TRUE) == FALSE) {
1762 crm_log_xml_err(reply, "Unable to route reply");
1763 }
1764
1765 pcmk__xml_free(update);
1766 pcmk__xml_free(reply);
1767 }
1768
1769 gboolean
1770 verify_stopped(enum crmd_fsa_state cur_state, int log_level)
1771 {
1772 gboolean res = TRUE;
1773 GList *lrm_state_list = lrm_state_get_list();
1774 GList *state_entry;
1775
1776 for (state_entry = lrm_state_list; state_entry != NULL; state_entry = state_entry->next) {
1777 lrm_state_t *lrm_state = state_entry->data;
1778
1779 if (!lrm_state_verify_stopped(lrm_state, cur_state, log_level)) {
1780
1781 res = FALSE;
1782 }
1783 }
1784
1785 controld_set_fsa_input_flags(R_SENT_RSC_STOP);
1786 g_list_free(lrm_state_list); lrm_state_list = NULL;
1787 return res;
1788 }
1789
1790 struct stop_recurring_action_s {
1791 lrmd_rsc_info_t *rsc;
1792 lrm_state_t *lrm_state;
1793 };
1794
1795 static gboolean
1796 stop_recurring_action_by_rsc(gpointer key, gpointer value, gpointer user_data)
1797 {
1798 gboolean remove = FALSE;
1799 struct stop_recurring_action_s *event = user_data;
1800 active_op_t *op = value;
1801
1802 if ((op->interval_ms != 0)
1803 && pcmk__str_eq(op->rsc_id, event->rsc->id, pcmk__str_none)) {
1804
1805 crm_debug("Cancelling op %d for %s (%s)", op->call_id, op->rsc_id, (char*)key);
1806 remove = !cancel_op(event->lrm_state, event->rsc->id, key, op->call_id, FALSE);
1807 }
1808
1809 return remove;
1810 }
1811
1812 static gboolean
1813 stop_recurring_actions(gpointer key, gpointer value, gpointer user_data)
1814 {
1815 gboolean remove = FALSE;
1816 lrm_state_t *lrm_state = user_data;
1817 active_op_t *op = value;
1818
1819 if (op->interval_ms != 0) {
1820 crm_info("Cancelling op %d for %s (%s)", op->call_id, op->rsc_id,
1821 (const char *) key);
1822 remove = !cancel_op(lrm_state, op->rsc_id, key, op->call_id, FALSE);
1823 }
1824
1825 return remove;
1826 }
1827
1828
1829
1830
1831
1832
1833
1834
1835
1836
1837
1838 static bool
1839 should_cancel_recurring(const char *rsc_id, const char *action, guint interval_ms)
1840 {
1841 if (is_remote_lrmd_ra(NULL, NULL, rsc_id) && (interval_ms == 0)
1842 && (strcmp(action, PCMK_ACTION_MIGRATE_TO) == 0)) {
1843
1844
1845
1846
1847 return false;
1848 }
1849
1850
1851 return (interval_ms == 0)
1852 && !pcmk__str_any_of(action, PCMK_ACTION_MONITOR,
1853 PCMK_ACTION_NOTIFY, NULL);
1854 }
1855
1856
1857
1858
1859
1860
1861
1862
1863
1864
1865 static const char *
1866 should_nack_action(const char *action)
1867 {
1868 if (pcmk_is_set(controld_globals.fsa_input_register, R_SHUTDOWN)
1869 && pcmk__str_eq(action, PCMK_ACTION_START, pcmk__str_none)) {
1870
1871 register_fsa_input(C_SHUTDOWN, I_SHUTDOWN, NULL);
1872 return "Not attempting start due to shutdown in progress";
1873 }
1874
1875 switch (controld_globals.fsa_state) {
1876 case S_NOT_DC:
1877 case S_POLICY_ENGINE:
1878 case S_TRANSITION_ENGINE:
1879 break;
1880 default:
1881 if (!pcmk__str_eq(action, PCMK_ACTION_STOP, pcmk__str_none)) {
1882 return "Controller cannot attempt actions at this time";
1883 }
1884 break;
1885 }
1886 return NULL;
1887 }
1888
1889 static void
1890 do_lrm_rsc_op(lrm_state_t *lrm_state, lrmd_rsc_info_t *rsc, xmlNode *msg,
1891 struct ra_metadata_s *md)
1892 {
1893 int rc;
1894 int call_id = 0;
1895 char *op_id = NULL;
1896 lrmd_event_data_t *op = NULL;
1897 fsa_data_t *msg_data = NULL;
1898 const char *transition = NULL;
1899 const char *operation = NULL;
1900 const char *nack_reason = NULL;
1901
1902 CRM_CHECK((rsc != NULL) && (msg != NULL), return);
1903
1904 operation = crm_element_value(msg, PCMK_XA_OPERATION);
1905 CRM_CHECK(!pcmk__str_empty(operation), return);
1906
1907 transition = crm_element_value(msg, PCMK__XA_TRANSITION_KEY);
1908 if (pcmk__str_empty(transition)) {
1909 crm_log_xml_err(msg, "Missing transition number");
1910 }
1911
1912 if (lrm_state == NULL) {
1913
1914 crm_err("Cannot execute %s of %s: No executor connection "
1915 QB_XS " transition_key=%s",
1916 operation, rsc->id, pcmk__s(transition, ""));
1917 synthesize_lrmd_failure(NULL, msg, PCMK_EXEC_INVALID,
1918 PCMK_OCF_UNKNOWN_ERROR,
1919 "No executor connection");
1920 return;
1921 }
1922
1923 if (pcmk__str_any_of(operation, PCMK_ACTION_RELOAD,
1924 PCMK_ACTION_RELOAD_AGENT, NULL)) {
1925
1926
1927
1928
1929
1930 if ((md != NULL)
1931 && pcmk_is_set(md->ra_flags, ra_supports_legacy_reload)) {
1932 operation = PCMK_ACTION_RELOAD;
1933 } else {
1934 operation = PCMK_ACTION_RELOAD_AGENT;
1935 }
1936 }
1937
1938 op = construct_op(lrm_state, msg, rsc->id, operation);
1939 CRM_CHECK(op != NULL, return);
1940
1941 if (should_cancel_recurring(rsc->id, operation, op->interval_ms)) {
1942 guint removed = 0;
1943 struct stop_recurring_action_s data;
1944
1945 data.rsc = rsc;
1946 data.lrm_state = lrm_state;
1947 removed = g_hash_table_foreach_remove(lrm_state->active_ops,
1948 stop_recurring_action_by_rsc,
1949 &data);
1950
1951 if (removed) {
1952 crm_debug("Stopped %u recurring operation%s in preparation for "
1953 PCMK__OP_FMT, removed, pcmk__plural_s(removed),
1954 rsc->id, operation, op->interval_ms);
1955 }
1956 }
1957
1958 nack_reason = should_nack_action(operation);
1959 if (nack_reason != NULL) {
1960 crm_notice("Not requesting local execution of %s operation for %s on %s"
1961 " in state %s: %s",
1962 pcmk__readable_action(op->op_type, op->interval_ms), rsc->id,
1963 lrm_state->node_name,
1964 fsa_state2string(controld_globals.fsa_state), nack_reason);
1965
1966 lrmd__set_result(op, PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_INVALID,
1967 nack_reason);
1968 controld_ack_event_directly(NULL, NULL, rsc, op, rsc->id);
1969 lrmd_free_event(op);
1970 free(op_id);
1971 return;
1972 }
1973
1974 crm_notice("Requesting local execution of %s operation for %s on %s "
1975 QB_XS " transition %s",
1976 pcmk__readable_action(op->op_type, op->interval_ms), rsc->id,
1977 lrm_state->node_name, pcmk__s(transition, ""));
1978
1979 controld_record_pending_op(lrm_state->node_name, rsc, op);
1980
1981 op_id = pcmk__op_key(rsc->id, op->op_type, op->interval_ms);
1982
1983 if (op->interval_ms > 0) {
1984
1985 cancel_op_key(lrm_state, rsc, op_id, FALSE);
1986 }
1987
1988 rc = controld_execute_resource_agent(lrm_state, rsc->id, op->op_type,
1989 op->user_data, op->interval_ms,
1990 op->timeout, op->start_delay,
1991 op->params, &call_id);
1992 if (rc == pcmk_rc_ok) {
1993
1994
1995
1996 char *call_id_s = make_stop_id(rsc->id, call_id);
1997 active_op_t *pending = NULL;
1998
1999 pending = pcmk__assert_alloc(1, sizeof(active_op_t));
2000 crm_trace("Recording pending op: %d - %s %s", call_id, op_id, call_id_s);
2001
2002 pending->call_id = call_id;
2003 pending->interval_ms = op->interval_ms;
2004 pending->op_type = pcmk__str_copy(operation);
2005 pending->op_key = pcmk__str_copy(op_id);
2006 pending->rsc_id = pcmk__str_copy(rsc->id);
2007 pending->start_time = time(NULL);
2008 pending->user_data = pcmk__str_copy(op->user_data);
2009 if (crm_element_value_epoch(msg, PCMK_OPT_SHUTDOWN_LOCK,
2010 &(pending->lock_time)) != pcmk_ok) {
2011 pending->lock_time = 0;
2012 }
2013 g_hash_table_replace(lrm_state->active_ops, call_id_s, pending);
2014
2015 if ((op->interval_ms > 0)
2016 && (op->start_delay > START_DELAY_THRESHOLD)) {
2017 int target_rc = PCMK_OCF_OK;
2018
2019 crm_info("Faking confirmation of %s: execution postponed for over 5 minutes", op_id);
2020 decode_transition_key(op->user_data, NULL, NULL, NULL, &target_rc);
2021 lrmd__set_result(op, target_rc, PCMK_EXEC_DONE, NULL);
2022 controld_ack_event_directly(NULL, NULL, rsc, op, rsc->id);
2023 }
2024
2025 pending->params = op->params;
2026 op->params = NULL;
2027
2028 } else if (lrm_state_is_local(lrm_state)) {
2029 crm_err("Could not initiate %s action for resource %s locally: %s "
2030 QB_XS " rc=%d", operation, rsc->id, pcmk_rc_str(rc), rc);
2031 fake_op_status(lrm_state, op, PCMK_EXEC_NOT_CONNECTED,
2032 PCMK_OCF_UNKNOWN_ERROR, pcmk_rc_str(rc));
2033 process_lrm_event(lrm_state, op, NULL, NULL);
2034 register_fsa_error(C_FSA_INTERNAL, I_FAIL, NULL);
2035
2036 } else {
2037 crm_err("Could not initiate %s action for resource %s remotely on %s: "
2038 "%s " QB_XS " rc=%d",
2039 operation, rsc->id, lrm_state->node_name, pcmk_rc_str(rc), rc);
2040 fake_op_status(lrm_state, op, PCMK_EXEC_NOT_CONNECTED,
2041 PCMK_OCF_UNKNOWN_ERROR, pcmk_rc_str(rc));
2042 process_lrm_event(lrm_state, op, NULL, NULL);
2043 }
2044
2045 free(op_id);
2046 lrmd_free_event(op);
2047 }
2048
2049 static char *
2050 unescape_newlines(const char *string)
2051 {
2052 char *pch = NULL;
2053 char *ret = NULL;
2054 static const char *escaped_newline = "\\n";
2055
2056 if (!string) {
2057 return NULL;
2058 }
2059
2060 ret = pcmk__str_copy(string);
2061 pch = strstr(ret, escaped_newline);
2062 while (pch != NULL) {
2063
2064
2065
2066 pch[0] = '\n';
2067 pch[1] = ' ';
2068 pch = strstr(pch, escaped_newline);
2069 }
2070
2071 return ret;
2072 }
2073
2074 static bool
2075 did_lrm_rsc_op_fail(lrm_state_t *lrm_state, const char * rsc_id,
2076 const char * op_type, guint interval_ms)
2077 {
2078 rsc_history_t *entry = NULL;
2079
2080 CRM_CHECK(lrm_state != NULL, return FALSE);
2081 CRM_CHECK(rsc_id != NULL, return FALSE);
2082 CRM_CHECK(op_type != NULL, return FALSE);
2083
2084 entry = g_hash_table_lookup(lrm_state->resource_history, rsc_id);
2085 if (entry == NULL || entry->failed == NULL) {
2086 return FALSE;
2087 }
2088
2089 if (pcmk__str_eq(entry->failed->rsc_id, rsc_id, pcmk__str_none)
2090 && pcmk__str_eq(entry->failed->op_type, op_type, pcmk__str_casei)
2091 && entry->failed->interval_ms == interval_ms) {
2092 return TRUE;
2093 }
2094
2095 return FALSE;
2096 }
2097
2098
2099
2100
2101
2102
2103
2104
2105
2106
2107 static void
2108 log_executor_event(const lrmd_event_data_t *op, const char *op_key,
2109 const char *node_name, gboolean confirmed)
2110 {
2111 int log_level = LOG_ERR;
2112 GString *str = g_string_sized_new(100);
2113
2114 pcmk__g_strcat(str,
2115 "Result of ",
2116 pcmk__readable_action(op->op_type, op->interval_ms),
2117 " operation for ", op->rsc_id, NULL);
2118
2119 if (node_name != NULL) {
2120 pcmk__g_strcat(str, " on ", node_name, NULL);
2121 }
2122
2123 switch (op->op_status) {
2124 case PCMK_EXEC_DONE:
2125 log_level = LOG_NOTICE;
2126 pcmk__g_strcat(str, ": ", crm_exit_str((crm_exit_t) op->rc), NULL);
2127 break;
2128
2129 case PCMK_EXEC_TIMEOUT:
2130 pcmk__g_strcat(str,
2131 ": ", pcmk_exec_status_str(op->op_status), " after ",
2132 pcmk__readable_interval(op->timeout), NULL);
2133 break;
2134
2135 case PCMK_EXEC_CANCELLED:
2136 log_level = LOG_INFO;
2137 pcmk__g_strcat(str, ": ", pcmk_exec_status_str(op->op_status),
2138 NULL);
2139 break;
2140
2141 default:
2142 pcmk__g_strcat(str, ": ", pcmk_exec_status_str(op->op_status),
2143 NULL);
2144 break;
2145 }
2146
2147 if ((op->exit_reason != NULL)
2148 && ((op->op_status != PCMK_EXEC_DONE) || (op->rc != PCMK_OCF_OK))) {
2149
2150 pcmk__g_strcat(str, " (", op->exit_reason, ")", NULL);
2151 }
2152
2153 g_string_append(str, " " QB_XS);
2154 g_string_append_printf(str, " graph action %sconfirmed; call=%d key=%s",
2155 (confirmed? "" : "un"), op->call_id, op_key);
2156 if (op->op_status == PCMK_EXEC_DONE) {
2157 g_string_append_printf(str, " rc=%d", op->rc);
2158 }
2159
2160 do_crm_log(log_level, "%s", str->str);
2161 g_string_free(str, TRUE);
2162
2163
2164
2165
2166 if ((op->output != NULL) && (op->rc != PCMK_OCF_OK)) {
2167 char *prefix = crm_strdup_printf(PCMK__OP_FMT "@%s output",
2168 op->rsc_id, op->op_type,
2169 op->interval_ms, node_name);
2170
2171 crm_log_output(LOG_NOTICE, prefix, op->output);
2172 free(prefix);
2173 }
2174 }
2175
2176 void
2177 process_lrm_event(lrm_state_t *lrm_state, lrmd_event_data_t *op,
2178 active_op_t *pending, const xmlNode *action_xml)
2179 {
2180 char *op_id = NULL;
2181 char *op_key = NULL;
2182
2183 gboolean remove = FALSE;
2184 gboolean removed = FALSE;
2185 bool need_direct_ack = FALSE;
2186 lrmd_rsc_info_t *rsc = NULL;
2187 const char *node_name = NULL;
2188
2189 CRM_CHECK(op != NULL, return);
2190 CRM_CHECK(op->rsc_id != NULL, return);
2191
2192
2193 if (compare_version(controld_globals.dc_version, "3.2.0") < 0) {
2194 switch (op->op_status) {
2195 case PCMK_EXEC_NOT_CONNECTED:
2196 lrmd__set_result(op, PCMK_OCF_CONNECTION_DIED,
2197 PCMK_EXEC_ERROR, op->exit_reason);
2198 break;
2199 case PCMK_EXEC_INVALID:
2200 lrmd__set_result(op, CRM_DIRECT_NACK_RC, PCMK_EXEC_ERROR,
2201 op->exit_reason);
2202 break;
2203 default:
2204 break;
2205 }
2206 }
2207
2208 op_id = make_stop_id(op->rsc_id, op->call_id);
2209 op_key = pcmk__op_key(op->rsc_id, op->op_type, op->interval_ms);
2210
2211
2212 if (lrm_state) {
2213 rsc = lrm_state_get_rsc_info(lrm_state, op->rsc_id, 0);
2214 }
2215 if ((rsc == NULL) && action_xml) {
2216 xmlNode *xml = pcmk__xe_first_child(action_xml, PCMK_XE_PRIMITIVE, NULL,
2217 NULL);
2218
2219 const char *standard = crm_element_value(xml, PCMK_XA_CLASS);
2220 const char *provider = crm_element_value(xml, PCMK_XA_PROVIDER);
2221 const char *type = crm_element_value(xml, PCMK_XA_TYPE);
2222
2223 if (standard && type) {
2224 crm_info("%s agent information not cached, using %s%s%s:%s from action XML",
2225 op->rsc_id, standard,
2226 (provider? ":" : ""), (provider? provider : ""), type);
2227 rsc = lrmd_new_rsc_info(op->rsc_id, standard, provider, type);
2228 } else {
2229 crm_err("Can't process %s result because %s agent information not cached or in XML",
2230 op_key, op->rsc_id);
2231 }
2232 }
2233
2234
2235 if (lrm_state) {
2236 node_name = lrm_state->node_name;
2237 } else if (action_xml) {
2238 node_name = crm_element_value(action_xml, PCMK__META_ON_NODE);
2239 }
2240
2241 if(pending == NULL) {
2242 remove = TRUE;
2243 if (lrm_state) {
2244 pending = g_hash_table_lookup(lrm_state->active_ops, op_id);
2245 }
2246 }
2247
2248 if (op->op_status == PCMK_EXEC_ERROR) {
2249 switch(op->rc) {
2250 case PCMK_OCF_NOT_RUNNING:
2251 case PCMK_OCF_RUNNING_PROMOTED:
2252 case PCMK_OCF_DEGRADED:
2253 case PCMK_OCF_DEGRADED_PROMOTED:
2254
2255 op->op_status = PCMK_EXEC_DONE;
2256 break;
2257 default:
2258
2259 break;
2260 }
2261 }
2262
2263 if (op->op_status != PCMK_EXEC_CANCELLED) {
2264
2265
2266
2267
2268 need_direct_ack = TRUE;
2269
2270 if (controld_action_is_recordable(op->op_type)) {
2271 if (node_name && rsc) {
2272
2273 time_t lock_time = (pending == NULL)? 0 : pending->lock_time;
2274
2275 controld_update_resource_history(node_name, rsc, op, lock_time);
2276 need_direct_ack = FALSE;
2277
2278 } else if (op->rsc_deleted) {
2279
2280
2281
2282
2283 crm_notice("Not recording %s result in CIB because "
2284 "resource information was removed since it was initiated",
2285 op_key);
2286 } else {
2287
2288
2289
2290
2291 crm_err("Unable to record %s result in CIB: %s", op_key,
2292 (node_name? "No resource information" : "No node name"));
2293 }
2294 }
2295
2296 } else if (op->interval_ms == 0) {
2297
2298
2299
2300
2301 need_direct_ack = TRUE;
2302
2303 } else if (pending == NULL) {
2304
2305
2306
2307
2308 } else if (op->user_data == NULL) {
2309
2310
2311
2312 crm_err("Recurring operation %s was cancelled without transition information",
2313 op_key);
2314
2315 } else if (pcmk_is_set(pending->flags, active_op_remove)) {
2316
2317
2318
2319 if (lrm_state) {
2320 controld_delete_action_history(op);
2321 }
2322
2323
2324
2325
2326
2327
2328
2329 if (did_lrm_rsc_op_fail(lrm_state, pending->rsc_id,
2330 pending->op_type, pending->interval_ms)) {
2331 need_direct_ack = TRUE;
2332 }
2333
2334 } else if (op->rsc_deleted) {
2335
2336
2337
2338
2339 crm_debug("Recurring op %s was cancelled due to resource deletion",
2340 op_key);
2341 need_direct_ack = TRUE;
2342
2343 } else {
2344
2345
2346
2347
2348 }
2349
2350 if (need_direct_ack) {
2351 controld_ack_event_directly(NULL, NULL, NULL, op, op->rsc_id);
2352 }
2353
2354 if(remove == FALSE) {
2355
2356 removed = TRUE;
2357
2358 } else if (lrm_state && ((op->interval_ms == 0)
2359 || (op->op_status == PCMK_EXEC_CANCELLED))) {
2360
2361 gboolean found = g_hash_table_remove(lrm_state->active_ops, op_id);
2362
2363 if (op->interval_ms != 0) {
2364 removed = TRUE;
2365 } else if (found) {
2366 removed = TRUE;
2367 crm_trace("Op %s (call=%d, stop-id=%s, remaining=%u): Confirmed",
2368 op_key, op->call_id, op_id,
2369 g_hash_table_size(lrm_state->active_ops));
2370 }
2371 }
2372
2373 log_executor_event(op, op_key, node_name, removed);
2374
2375 if (lrm_state) {
2376 if (!pcmk__str_eq(op->op_type, PCMK_ACTION_META_DATA,
2377 pcmk__str_casei)) {
2378 crmd_alert_resource_op(lrm_state->node_name, op);
2379 } else if (rsc && (op->rc == PCMK_OCF_OK)) {
2380 char *metadata = unescape_newlines(op->output);
2381
2382 controld_cache_metadata(lrm_state->metadata_cache, rsc, metadata);
2383 free(metadata);
2384 }
2385 }
2386
2387 if (op->rsc_deleted) {
2388 crm_info("Deletion of resource '%s' complete after %s", op->rsc_id, op_key);
2389 if (lrm_state) {
2390 delete_rsc_entry(lrm_state, NULL, op->rsc_id, NULL, pcmk_ok, NULL,
2391 true);
2392 }
2393 }
2394
2395
2396
2397
2398 controld_trigger_fsa();
2399 if (lrm_state && rsc) {
2400 update_history_cache(lrm_state, rsc, op);
2401 }
2402
2403 lrmd_free_rsc_info(rsc);
2404 free(op_key);
2405 free(op_id);
2406 }