1 /*
2 * Copyright 2010-2023 the Pacemaker project contributors
3 *
4 * The version control history for this file may have further details.
5 *
6 * This source code is licensed under the GNU Lesser General Public License
7 * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY.
8 */
9
10 #include <crm_internal.h>
11
12 #ifndef _GNU_SOURCE
13 # define _GNU_SOURCE
14 #endif
15
16 #include <sys/types.h>
17 #include <sys/stat.h>
18 #include <stdio.h>
19 #include <errno.h>
20 #include <unistd.h>
21 #include <dirent.h>
22 #include <fcntl.h>
23
24 #include <crm/crm.h>
25 #include <crm/common/mainloop.h>
26 #include <crm/services.h>
27 #include <crm/services_internal.h>
28 #include <crm/stonith-ng.h>
29 #include <crm/msg_xml.h>
30 #include "services_private.h"
31 #include "services_ocf.h"
32 #include "services_lsb.h"
33
34 #if SUPPORT_UPSTART
35 # include <upstart.h>
36 #endif
37
38 #if SUPPORT_SYSTEMD
39 # include <systemd.h>
40 #endif
41
42 #if SUPPORT_NAGIOS
43 # include <services_nagios.h>
44 #endif
45
46 /* TODO: Develop a rollover strategy */
47
48 static int operations = 0;
49 static GHashTable *recurring_actions = NULL;
50
51 /* ops waiting to run async because of conflicting active
52 * pending ops */
53 static GList *blocked_ops = NULL;
54
55 /* ops currently active (in-flight) */
56 static GList *inflight_ops = NULL;
57
58 static void handle_blocked_ops(void);
59
60 /*!
61 * \brief Find first service class that can provide a specified agent
62 *
63 * \param[in] agent Name of agent to search for
64 *
65 * \return Service class if found, NULL otherwise
66 *
67 * \note The priority is LSB, then systemd, then upstart. It would be preferable
68 * to put systemd first, but LSB merely requires a file existence check,
69 * while systemd requires contacting D-Bus.
70 */
71 const char *
72 resources_find_service_class(const char *agent)
73 {
74 if (services__lsb_agent_exists(agent)) {
75 return PCMK_RESOURCE_CLASS_LSB;
76 }
77
78 #if SUPPORT_SYSTEMD
79 if (systemd_unit_exists(agent)) {
80 return PCMK_RESOURCE_CLASS_SYSTEMD;
81 }
82 #endif
83
84 #if SUPPORT_UPSTART
85 if (upstart_job_exists(agent)) {
86 return PCMK_RESOURCE_CLASS_UPSTART;
87 }
88 #endif
89 return NULL;
90 }
91
92 static inline void
93 init_recurring_actions(void)
94 {
95 if (recurring_actions == NULL) {
96 recurring_actions = pcmk__strkey_table(NULL, NULL);
97 }
98 }
99
100 /*!
101 * \internal
102 * \brief Check whether op is in-flight systemd or upstart op
103 *
104 * \param[in] op Operation to check
105 *
106 * \return TRUE if op is in-flight systemd or upstart op
107 */
108 static inline gboolean
109 inflight_systemd_or_upstart(const svc_action_t *op)
110 {
111 return pcmk__strcase_any_of(op->standard, PCMK_RESOURCE_CLASS_SYSTEMD,
112 PCMK_RESOURCE_CLASS_UPSTART, NULL) &&
113 g_list_find(inflight_ops, op) != NULL;
114 }
115
116 /*!
117 * \internal
118 * \brief Expand "service" alias to an actual resource class
119 *
120 * \param[in] rsc Resource name (for logging only)
121 * \param[in] standard Resource class as configured
122 * \param[in] agent Agent name to look for
123 *
124 * \return Newly allocated string with actual resource class
125 *
126 * \note The caller is responsible for calling free() on the result.
127 */
128 static char *
129 expand_resource_class(const char *rsc, const char *standard, const char *agent)
130 {
131 char *expanded_class = NULL;
132
133 if (strcasecmp(standard, PCMK_RESOURCE_CLASS_SERVICE) == 0) {
134 const char *found_class = resources_find_service_class(agent);
135
136 if (found_class) {
137 crm_debug("Found %s agent %s for %s", found_class, agent, rsc);
138 expanded_class = strdup(found_class);
139 } else {
140 crm_info("Assuming resource class lsb for agent %s for %s",
141 agent, rsc);
142 expanded_class = strdup(PCMK_RESOURCE_CLASS_LSB);
143 }
144 } else {
145 expanded_class = strdup(standard);
146 }
147 CRM_ASSERT(expanded_class);
148 return expanded_class;
149 }
150
151 /*!
152 * \internal
153 * \brief Create a simple svc_action_t instance
154 *
155 * \return Newly allocated instance (or NULL if not enough memory)
156 */
157 static svc_action_t *
158 new_action(void)
159 {
160 svc_action_t *op = calloc(1, sizeof(svc_action_t));
161
162 if (op == NULL) {
163 return NULL;
164 }
165
166 op->opaque = calloc(1, sizeof(svc_action_private_t));
167 if (op->opaque == NULL) {
168 free(op);
169 return NULL;
170 }
171
172 // Initialize result
173 services__set_result(op, PCMK_OCF_UNKNOWN, PCMK_EXEC_UNKNOWN, NULL);
174 return op;
175 }
176
177 static bool
178 required_argument_missing(uint32_t ra_caps, const char *name,
179 const char *standard, const char *provider,
180 const char *agent, const char *action)
181 {
182 if (pcmk__str_empty(name)) {
183 crm_info("Cannot create operation without resource name (bug?)");
184 return true;
185 }
186
187 if (pcmk__str_empty(standard)) {
188 crm_info("Cannot create operation for %s without resource class (bug?)",
189 name);
190 return true;
191 }
192
193 if (pcmk_is_set(ra_caps, pcmk_ra_cap_provider)
194 && pcmk__str_empty(provider)) {
195 crm_info("Cannot create operation for %s resource %s "
196 "without provider (bug?)", standard, name);
197 return true;
198 }
199
200 if (pcmk__str_empty(agent)) {
201 crm_info("Cannot create operation for %s without agent name (bug?)",
202 name);
203 return true;
204 }
205
206 if (pcmk__str_empty(action)) {
207 crm_info("Cannot create operation for %s without action name (bug?)",
208 name);
209 return true;
210 }
211 return false;
212 }
213
214 // \return Standard Pacemaker return code (pcmk_rc_ok or ENOMEM)
215 static int
216 copy_action_arguments(svc_action_t *op, uint32_t ra_caps, const char *name,
217 const char *standard, const char *provider,
218 const char *agent, const char *action)
219 {
220 op->rsc = strdup(name);
221 if (op->rsc == NULL) {
222 return ENOMEM;
223 }
224
225 op->agent = strdup(agent);
226 if (op->agent == NULL) {
227 return ENOMEM;
228 }
229
230 op->standard = expand_resource_class(name, standard, agent);
231 if (op->standard == NULL) {
232 return ENOMEM;
233 }
234
235 if (pcmk_is_set(ra_caps, pcmk_ra_cap_status)
236 && pcmk__str_eq(action, PCMK_ACTION_MONITOR, pcmk__str_casei)) {
237 action = PCMK_ACTION_STATUS;
238 }
239 op->action = strdup(action);
240 if (op->action == NULL) {
241 return ENOMEM;
242 }
243
244 if (pcmk_is_set(ra_caps, pcmk_ra_cap_provider)) {
245 op->provider = strdup(provider);
246 if (op->provider == NULL) {
247 return ENOMEM;
248 }
249 }
250 return pcmk_rc_ok;
251 }
252
253 svc_action_t *
254 services__create_resource_action(const char *name, const char *standard,
255 const char *provider, const char *agent,
256 const char *action, guint interval_ms, int timeout,
257 GHashTable *params, enum svc_action_flags flags)
258 {
259 svc_action_t *op = NULL;
260 uint32_t ra_caps = pcmk_get_ra_caps(standard);
261 int rc = pcmk_rc_ok;
262
263 op = new_action();
264 if (op == NULL) {
265 crm_crit("Cannot prepare action: %s", strerror(ENOMEM));
266 if (params != NULL) {
267 g_hash_table_destroy(params);
268 }
269 return NULL;
270 }
271
272 op->interval_ms = interval_ms;
273 op->timeout = timeout;
274 op->flags = flags;
275 op->sequence = ++operations;
276
277 // Take ownership of params
278 if (pcmk_is_set(ra_caps, pcmk_ra_cap_params)) {
279 op->params = params;
280 } else if (params != NULL) {
281 g_hash_table_destroy(params);
282 params = NULL;
283 }
284
285 if (required_argument_missing(ra_caps, name, standard, provider, agent,
286 action)) {
287 services__set_result(op, services__generic_error(op),
288 PCMK_EXEC_ERROR_FATAL,
289 "Required agent or action information missing");
290 return op;
291 }
292
293 op->id = pcmk__op_key(name, action, interval_ms);
294
295 if (copy_action_arguments(op, ra_caps, name, standard, provider, agent,
296 action) != pcmk_rc_ok) {
297 crm_crit("Cannot prepare %s action for %s: %s",
298 action, name, strerror(ENOMEM));
299 services__handle_exec_error(op, ENOMEM);
300 return op;
301 }
302
303 if (strcasecmp(op->standard, PCMK_RESOURCE_CLASS_OCF) == 0) {
304 rc = services__ocf_prepare(op);
305
306 } else if (strcasecmp(op->standard, PCMK_RESOURCE_CLASS_LSB) == 0) {
307 rc = services__lsb_prepare(op);
308
309 #if SUPPORT_SYSTEMD
310 } else if (strcasecmp(op->standard, PCMK_RESOURCE_CLASS_SYSTEMD) == 0) {
311 rc = services__systemd_prepare(op);
312 #endif
313 #if SUPPORT_UPSTART
314 } else if (strcasecmp(op->standard, PCMK_RESOURCE_CLASS_UPSTART) == 0) {
315 rc = services__upstart_prepare(op);
316 #endif
317 #if SUPPORT_NAGIOS
318 } else if (strcasecmp(op->standard, PCMK_RESOURCE_CLASS_NAGIOS) == 0) {
319 rc = services__nagios_prepare(op);
320 #endif
321 } else {
322 crm_info("Unknown resource standard: %s", op->standard);
323 rc = ENOENT;
324 }
325
326 if (rc != pcmk_rc_ok) {
327 crm_info("Cannot prepare %s operation for %s: %s",
328 action, name, strerror(rc));
329 services__handle_exec_error(op, rc);
330 }
331 return op;
332 }
333
334 svc_action_t *
335 resources_action_create(const char *name, const char *standard,
336 const char *provider, const char *agent,
337 const char *action, guint interval_ms, int timeout,
338 GHashTable *params, enum svc_action_flags flags)
339 {
340 svc_action_t *op = services__create_resource_action(name, standard,
341 provider, agent, action, interval_ms, timeout,
342 params, flags);
343 if (op == NULL || op->rc != 0) {
344 services_action_free(op);
345 return NULL;
346 } else {
347 // Preserve public API backward compatibility
348 op->rc = PCMK_OCF_OK;
349 op->status = PCMK_EXEC_DONE;
350
351 return op;
352 }
353 }
354
355 svc_action_t *
356 services_action_create_generic(const char *exec, const char *args[])
357 {
358 svc_action_t *op = new_action();
359
360 CRM_ASSERT(op != NULL);
361
362 op->opaque->exec = strdup(exec);
363 op->opaque->args[0] = strdup(exec);
364 if ((op->opaque->exec == NULL) || (op->opaque->args[0] == NULL)) {
365 crm_crit("Cannot prepare action for '%s': %s", exec, strerror(ENOMEM));
366 services__set_result(op, PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_ERROR,
367 strerror(ENOMEM));
368 return op;
369 }
370
371 if (args == NULL) {
372 return op;
373 }
374
375 for (int cur_arg = 1; args[cur_arg - 1] != NULL; cur_arg++) {
376
377 if (cur_arg == PCMK__NELEM(op->opaque->args)) {
378 crm_info("Cannot prepare action for '%s': Too many arguments",
379 exec);
380 services__set_result(op, PCMK_OCF_UNKNOWN_ERROR,
381 PCMK_EXEC_ERROR_HARD, "Too many arguments");
382 break;
383 }
384
385 op->opaque->args[cur_arg] = strdup(args[cur_arg - 1]);
386 if (op->opaque->args[cur_arg] == NULL) {
387 crm_crit("Cannot prepare action for '%s': %s",
388 exec, strerror(ENOMEM));
389 services__set_result(op, PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_ERROR,
390 strerror(ENOMEM));
391 break;
392 }
393 }
394
395 return op;
396 }
397
398 /*!
399 * \brief Create an alert agent action
400 *
401 * \param[in] id Alert ID
402 * \param[in] exec Path to alert agent executable
403 * \param[in] timeout Action timeout
404 * \param[in] params Parameters to use with action
405 * \param[in] sequence Action sequence number
406 * \param[in] cb_data Data to pass to callback function
407 *
408 * \return New action on success, NULL on error
409 * \note It is the caller's responsibility to free cb_data.
410 * The caller should not free params explicitly.
411 */
412 svc_action_t *
413 services_alert_create(const char *id, const char *exec, int timeout,
414 GHashTable *params, int sequence, void *cb_data)
415 {
416 svc_action_t *action = services_action_create_generic(exec, NULL);
417
418 action->id = strdup(id);
419 action->standard = strdup(PCMK_RESOURCE_CLASS_ALERT);
420 CRM_ASSERT((action->id != NULL) && (action->standard != NULL));
421
422 action->timeout = timeout;
423 action->params = params;
424 action->sequence = sequence;
425 action->cb_data = cb_data;
426 return action;
427 }
428
429 /*!
430 * \brief Set the user and group that an action will execute as
431 *
432 * \param[in,out] op Action to modify
433 * \param[in] user Name of user to execute action as
434 * \param[in] group Name of group to execute action as
435 *
436 * \return pcmk_ok on success, -errno otherwise
437 *
438 * \note This will have no effect unless the process executing the action runs
439 * as root, and the action is not a systemd or upstart action.
440 * We could implement this for systemd by adding User= and Group= to
441 * [Service] in the override file, but that seems more likely to cause
442 * problems than be useful.
443 */
444 int
445 services_action_user(svc_action_t *op, const char *user)
446 {
447 CRM_CHECK((op != NULL) && (user != NULL), return -EINVAL);
448 return crm_user_lookup(user, &(op->opaque->uid), &(op->opaque->gid));
449 }
450
451 /*!
452 * \brief Execute an alert agent action
453 *
454 * \param[in,out] action Action to execute
455 * \param[in] cb Function to call when action completes
456 *
457 * \return TRUE if the library will free action, FALSE otherwise
458 *
459 * \note If this function returns FALSE, it is the caller's responsibility to
460 * free the action with services_action_free(). However, unless someone
461 * intentionally creates a recurring alert action, this will never return
462 * FALSE.
463 */
464 gboolean
465 services_alert_async(svc_action_t *action, void (*cb)(svc_action_t *op))
466 {
467 action->synchronous = false;
468 action->opaque->callback = cb;
469 return services__execute_file(action) == pcmk_rc_ok;
470 }
471
472 #if HAVE_DBUS
473 /*!
474 * \internal
475 * \brief Update operation's pending DBus call, unreferencing old one if needed
476 *
477 * \param[in,out] op Operation to modify
478 * \param[in] pending Pending call to set
479 */
480 void
481 services_set_op_pending(svc_action_t *op, DBusPendingCall *pending)
482 {
483 if (op->opaque->pending && (op->opaque->pending != pending)) {
484 if (pending) {
485 crm_info("Lost pending %s DBus call (%p)", op->id, op->opaque->pending);
486 } else {
487 crm_trace("Done with pending %s DBus call (%p)", op->id, op->opaque->pending);
488 }
489 dbus_pending_call_unref(op->opaque->pending);
490 }
491 op->opaque->pending = pending;
492 if (pending) {
493 crm_trace("Updated pending %s DBus call (%p)", op->id, pending);
494 } else {
495 crm_trace("Cleared pending %s DBus call", op->id);
496 }
497 }
498 #endif
499
500 void
501 services_action_cleanup(svc_action_t * op)
502 {
503 if ((op == NULL) || (op->opaque == NULL)) {
504 return;
505 }
506
507 #if HAVE_DBUS
508 if(op->opaque->timerid != 0) {
509 crm_trace("Removing timer for call %s to %s", op->action, op->rsc);
510 g_source_remove(op->opaque->timerid);
511 op->opaque->timerid = 0;
512 }
513
514 if(op->opaque->pending) {
515 if (dbus_pending_call_get_completed(op->opaque->pending)) {
516 // This should never be the case
517 crm_warn("Result of %s op %s was unhandled",
518 op->standard, op->id);
519 } else {
520 crm_debug("Will ignore any result of canceled %s op %s",
521 op->standard, op->id);
522 }
523 dbus_pending_call_cancel(op->opaque->pending);
524 services_set_op_pending(op, NULL);
525 }
526 #endif
527
528 if (op->opaque->stderr_gsource) {
529 mainloop_del_fd(op->opaque->stderr_gsource);
530 op->opaque->stderr_gsource = NULL;
531 }
532
533 if (op->opaque->stdout_gsource) {
534 mainloop_del_fd(op->opaque->stdout_gsource);
535 op->opaque->stdout_gsource = NULL;
536 }
537 }
538
539 /*!
540 * \internal
541 * \brief Map an actual resource action result to a standard OCF result
542 *
543 * \param[in] standard Agent standard (must not be "service")
544 * \param[in] action Action that result is for
545 * \param[in] exit_status Actual agent exit status
546 *
547 * \return Standard OCF result
548 */
549 enum ocf_exitcode
550 services_result2ocf(const char *standard, const char *action, int exit_status)
551 {
552 if (pcmk__str_eq(standard, PCMK_RESOURCE_CLASS_OCF, pcmk__str_casei)) {
553 return services__ocf2ocf(exit_status);
554
555 #if SUPPORT_SYSTEMD
556 } else if (pcmk__str_eq(standard, PCMK_RESOURCE_CLASS_SYSTEMD,
557 pcmk__str_casei)) {
558 return services__systemd2ocf(exit_status);
559 #endif
560
561 #if SUPPORT_UPSTART
562 } else if (pcmk__str_eq(standard, PCMK_RESOURCE_CLASS_UPSTART,
563 pcmk__str_casei)) {
564 return services__upstart2ocf(exit_status);
565 #endif
566
567 #if SUPPORT_NAGIOS
568 } else if (pcmk__str_eq(standard, PCMK_RESOURCE_CLASS_NAGIOS,
569 pcmk__str_casei)) {
570 return services__nagios2ocf(exit_status);
571 #endif
572
573 } else if (pcmk__str_eq(standard, PCMK_RESOURCE_CLASS_LSB,
574 pcmk__str_casei)) {
575 return services__lsb2ocf(action, exit_status);
576
577 } else {
578 crm_warn("Treating result from unknown standard '%s' as OCF",
579 ((standard == NULL)? "unspecified" : standard));
580 return services__ocf2ocf(exit_status);
581 }
582 }
583
584 void
585 services_action_free(svc_action_t * op)
586 {
587 unsigned int i;
588
589 if (op == NULL) {
590 return;
591 }
592
593 /* The operation should be removed from all tracking lists by this point.
594 * If it's not, we have a bug somewhere, so bail. That may lead to a
595 * memory leak, but it's better than a use-after-free segmentation fault.
596 */
597 CRM_CHECK(g_list_find(inflight_ops, op) == NULL, return);
598 CRM_CHECK(g_list_find(blocked_ops, op) == NULL, return);
599 CRM_CHECK((recurring_actions == NULL)
600 || (g_hash_table_lookup(recurring_actions, op->id) == NULL),
601 return);
602
603 services_action_cleanup(op);
604
605 if (op->opaque->repeat_timer) {
606 g_source_remove(op->opaque->repeat_timer);
607 op->opaque->repeat_timer = 0;
608 }
609
610 free(op->id);
611 free(op->opaque->exec);
612
613 for (i = 0; i < PCMK__NELEM(op->opaque->args); i++) {
614 free(op->opaque->args[i]);
615 }
616
617 free(op->opaque->exit_reason);
618 free(op->opaque);
619 free(op->rsc);
620 free(op->action);
621
622 free(op->standard);
623 free(op->agent);
624 free(op->provider);
625
626 free(op->stdout_data);
627 free(op->stderr_data);
628
629 if (op->params) {
630 g_hash_table_destroy(op->params);
631 op->params = NULL;
632 }
633
634 free(op);
635 }
636
637 gboolean
638 cancel_recurring_action(svc_action_t * op)
639 {
640 crm_info("Cancelling %s operation %s", op->standard, op->id);
641
642 if (recurring_actions) {
643 g_hash_table_remove(recurring_actions, op->id);
644 }
645
646 if (op->opaque->repeat_timer) {
647 g_source_remove(op->opaque->repeat_timer);
648 op->opaque->repeat_timer = 0;
649 }
650
651 return TRUE;
652 }
653
654 /*!
655 * \brief Cancel a recurring action
656 *
657 * \param[in] name Name of resource that operation is for
658 * \param[in] action Name of operation to cancel
659 * \param[in] interval_ms Interval of operation to cancel
660 *
661 * \return TRUE if action was successfully cancelled, FALSE otherwise
662 */
663 gboolean
664 services_action_cancel(const char *name, const char *action, guint interval_ms)
665 {
666 gboolean cancelled = FALSE;
667 char *id = pcmk__op_key(name, action, interval_ms);
668 svc_action_t *op = NULL;
669
670 /* We can only cancel a recurring action */
671 init_recurring_actions();
672 op = g_hash_table_lookup(recurring_actions, id);
673 if (op == NULL) {
674 goto done;
675 }
676
677 // Tell services__finalize_async_op() not to reschedule the operation
678 op->cancel = TRUE;
679
680 /* Stop tracking it as a recurring operation, and stop its repeat timer */
681 cancel_recurring_action(op);
682
683 /* If the op has a PID, it's an in-flight child process, so kill it.
684 *
685 * Whether the kill succeeds or fails, the main loop will send the op to
686 * async_action_complete() (and thus services__finalize_async_op()) when the
687 * process goes away.
688 */
689 if (op->pid != 0) {
690 crm_info("Terminating in-flight op %s[%d] early because it was cancelled",
691 id, op->pid);
692 cancelled = mainloop_child_kill(op->pid);
693 if (cancelled == FALSE) {
694 crm_err("Termination of %s[%d] failed", id, op->pid);
695 }
696 goto done;
697 }
698
699 #if HAVE_DBUS
700 // In-flight systemd and upstart ops don't have a pid
701 if (inflight_systemd_or_upstart(op)) {
702 inflight_ops = g_list_remove(inflight_ops, op);
703
704 /* This will cause any result that comes in later to be discarded, so we
705 * don't call the callback and free the operation twice.
706 */
707 services_action_cleanup(op);
708 }
709 #endif
710
711 /* The rest of this is essentially equivalent to
712 * services__finalize_async_op(), minus the handle_blocked_ops() call.
713 */
714
715 // Report operation as cancelled
716 services__set_cancelled(op);
717 if (op->opaque->callback) {
718 op->opaque->callback(op);
719 }
720
721 blocked_ops = g_list_remove(blocked_ops, op);
722 services_action_free(op);
723 cancelled = TRUE;
724 // @TODO Initiate handle_blocked_ops() asynchronously
725
726 done:
727 free(id);
728 return cancelled;
729 }
730
731 gboolean
732 services_action_kick(const char *name, const char *action, guint interval_ms)
733 {
734 svc_action_t * op = NULL;
735 char *id = pcmk__op_key(name, action, interval_ms);
736
737 init_recurring_actions();
738 op = g_hash_table_lookup(recurring_actions, id);
739 free(id);
740
741 if (op == NULL) {
742 return FALSE;
743 }
744
745
746 if (op->pid || inflight_systemd_or_upstart(op)) {
747 return TRUE;
748 } else {
749 if (op->opaque->repeat_timer) {
750 g_source_remove(op->opaque->repeat_timer);
751 op->opaque->repeat_timer = 0;
752 }
753 recurring_action_timer(op);
754 return TRUE;
755 }
756
757 }
758
759 /*!
760 * \internal
761 * \brief Add a new recurring operation, checking for duplicates
762 *
763 * \param[in,out] op Operation to add
764 *
765 * \return TRUE if duplicate found (and reschedule), FALSE otherwise
766 */
767 static gboolean
768 handle_duplicate_recurring(svc_action_t *op)
769 {
770 svc_action_t * dup = NULL;
771
772 /* check for duplicates */
773 dup = g_hash_table_lookup(recurring_actions, op->id);
774
775 if (dup && (dup != op)) {
776 /* update user data */
777 if (op->opaque->callback) {
778 dup->opaque->callback = op->opaque->callback;
779 dup->cb_data = op->cb_data;
780 op->cb_data = NULL;
781 }
782 /* immediately execute the next interval */
783 if (dup->pid != 0) {
784 if (op->opaque->repeat_timer) {
785 g_source_remove(op->opaque->repeat_timer);
786 op->opaque->repeat_timer = 0;
787 }
788 recurring_action_timer(dup);
789 }
790 /* free the duplicate */
791 services_action_free(op);
792 return TRUE;
793 }
794
795 return FALSE;
796 }
797
798 /*!
799 * \internal
800 * \brief Execute an action appropriately according to its standard
801 *
802 * \param[in,out] op Action to execute
803 *
804 * \return Standard Pacemaker return code
805 * \retval EBUSY Recurring operation could not be initiated
806 * \retval pcmk_rc_error Synchronous action failed
807 * \retval pcmk_rc_ok Synchronous action succeeded, or asynchronous action
808 * should not be freed (because it's pending or because
809 * it failed to execute and was already freed)
810 *
811 * \note If the return value for an asynchronous action is not pcmk_rc_ok, the
812 * caller is responsible for freeing the action.
813 */
814 static int
815 execute_action(svc_action_t *op)
816 {
817 #if SUPPORT_UPSTART
818 if (pcmk__str_eq(op->standard, PCMK_RESOURCE_CLASS_UPSTART,
819 pcmk__str_casei)) {
820 return services__execute_upstart(op);
821 }
822 #endif
823
824 #if SUPPORT_SYSTEMD
825 if (pcmk__str_eq(op->standard, PCMK_RESOURCE_CLASS_SYSTEMD,
826 pcmk__str_casei)) {
827 return services__execute_systemd(op);
828 }
829 #endif
830
831 return services__execute_file(op);
832 }
833
834 void
835 services_add_inflight_op(svc_action_t * op)
836 {
837 if (op == NULL) {
838 return;
839 }
840
841 CRM_ASSERT(op->synchronous == FALSE);
842
843 /* keep track of ops that are in-flight to avoid collisions in the same namespace */
844 if (op->rsc) {
845 inflight_ops = g_list_append(inflight_ops, op);
846 }
847 }
848
849 /*!
850 * \internal
851 * \brief Stop tracking an operation that completed
852 *
853 * \param[in] op Operation to stop tracking
854 */
855 void
856 services_untrack_op(const svc_action_t *op)
857 {
858 /* Op is no longer in-flight or blocked */
859 inflight_ops = g_list_remove(inflight_ops, op);
860 blocked_ops = g_list_remove(blocked_ops, op);
861
862 /* Op is no longer blocking other ops, so check if any need to run */
863 handle_blocked_ops();
864 }
865
866 gboolean
867 services_action_async_fork_notify(svc_action_t * op,
868 void (*action_callback) (svc_action_t *),
869 void (*action_fork_callback) (svc_action_t *))
870 {
871 CRM_CHECK(op != NULL, return TRUE);
872
873 op->synchronous = false;
874 if (action_callback != NULL) {
875 op->opaque->callback = action_callback;
876 }
877 if (action_fork_callback != NULL) {
878 op->opaque->fork_callback = action_fork_callback;
879 }
880
881 if (op->interval_ms > 0) {
882 init_recurring_actions();
883 if (handle_duplicate_recurring(op)) {
884 /* entry rescheduled, dup freed */
885 /* exit early */
886 return TRUE;
887 }
888 g_hash_table_replace(recurring_actions, op->id, op);
889 }
890
891 if (!pcmk_is_set(op->flags, SVC_ACTION_NON_BLOCKED)
892 && op->rsc && is_op_blocked(op->rsc)) {
893 blocked_ops = g_list_append(blocked_ops, op);
894 return TRUE;
895 }
896
897 return execute_action(op) == pcmk_rc_ok;
898 }
899
900 gboolean
901 services_action_async(svc_action_t * op,
902 void (*action_callback) (svc_action_t *))
903 {
904 return services_action_async_fork_notify(op, action_callback, NULL);
905 }
906
907 static gboolean processing_blocked_ops = FALSE;
908
909 gboolean
910 is_op_blocked(const char *rsc)
911 {
912 GList *gIter = NULL;
913 svc_action_t *op = NULL;
914
915 for (gIter = inflight_ops; gIter != NULL; gIter = gIter->next) {
916 op = gIter->data;
917 if (pcmk__str_eq(op->rsc, rsc, pcmk__str_casei)) {
918 return TRUE;
919 }
920 }
921
922 return FALSE;
923 }
924
925 static void
926 handle_blocked_ops(void)
927 {
928 GList *executed_ops = NULL;
929 GList *gIter = NULL;
930 svc_action_t *op = NULL;
931
932 if (processing_blocked_ops) {
933 /* avoid nested calling of this function */
934 return;
935 }
936
937 processing_blocked_ops = TRUE;
938
939 /* n^2 operation here, but blocked ops are incredibly rare. this list
940 * will be empty 99% of the time. */
941 for (gIter = blocked_ops; gIter != NULL; gIter = gIter->next) {
942 op = gIter->data;
943 if (is_op_blocked(op->rsc)) {
944 continue;
945 }
946 executed_ops = g_list_append(executed_ops, op);
947 if (execute_action(op) != pcmk_rc_ok) {
948 /* this can cause this function to be called recursively
949 * which is why we have processing_blocked_ops static variable */
950 services__finalize_async_op(op);
951 }
952 }
953
954 for (gIter = executed_ops; gIter != NULL; gIter = gIter->next) {
955 op = gIter->data;
956 blocked_ops = g_list_remove(blocked_ops, op);
957 }
958 g_list_free(executed_ops);
959
960 processing_blocked_ops = FALSE;
961 }
962
963 /*!
964 * \internal
965 * \brief Execute a meta-data action appropriately to standard
966 *
967 * \param[in,out] op Meta-data action to execute
968 *
969 * \return Standard Pacemaker return code
970 */
971 static int
972 execute_metadata_action(svc_action_t *op)
973 {
974 const char *class = op->standard;
975
|
(1) Event path: |
Condition "op->agent == NULL", taking false branch. |
976 if (op->agent == NULL) {
977 crm_info("Meta-data requested without specifying agent");
978 services__set_result(op, services__generic_error(op),
979 PCMK_EXEC_ERROR_FATAL, "Agent not specified");
980 return EINVAL;
981 }
982
|
(2) Event path: |
Condition "class == NULL", taking false branch. |
983 if (class == NULL) {
984 crm_info("Meta-data requested for agent %s without specifying class",
985 op->agent);
986 services__set_result(op, services__generic_error(op),
987 PCMK_EXEC_ERROR_FATAL,
988 "Agent standard not specified");
989 return EINVAL;
990 }
991
|
(3) Event path: |
Condition "!strcmp(class, "service")", taking true branch. |
992 if (!strcmp(class, PCMK_RESOURCE_CLASS_SERVICE)) {
993 class = resources_find_service_class(op->agent);
994 }
|
(4) Event path: |
Condition "class == NULL", taking false branch. |
995 if (class == NULL) {
996 crm_info("Meta-data requested for %s, but could not determine class",
997 op->agent);
998 services__set_result(op, services__generic_error(op),
999 PCMK_EXEC_ERROR_HARD,
1000 "Agent standard could not be determined");
1001 return EINVAL;
1002 }
1003
|
(5) Event path: |
Condition "pcmk__str_eq(class, "lsb", pcmk__str_casei)", taking false branch. |
1004 if (pcmk__str_eq(class, PCMK_RESOURCE_CLASS_LSB, pcmk__str_casei)) {
1005 return pcmk_legacy2rc(services__get_lsb_metadata(op->agent,
1006 &op->stdout_data));
1007 }
1008
1009 #if SUPPORT_NAGIOS
|
(6) Event path: |
Condition "pcmk__str_eq(class, "nagios", pcmk__str_casei)", taking true branch. |
1010 if (pcmk__str_eq(class, PCMK_RESOURCE_CLASS_NAGIOS, pcmk__str_casei)) {
|
(7) Event string_null_source: |
Function "services__get_nagios_metadata" does not terminate string "op->stdout_data". [details] |
1011 return pcmk_legacy2rc(services__get_nagios_metadata(op->agent,
1012 &op->stdout_data));
1013 }
1014 #endif
1015
1016 return execute_action(op);
1017 }
1018
1019 gboolean
1020 services_action_sync(svc_action_t * op)
1021 {
1022 gboolean rc = TRUE;
1023
|
(1) Event path: |
Condition "op == NULL", taking false branch. |
1024 if (op == NULL) {
1025 crm_trace("No operation to execute");
1026 return FALSE;
1027 }
1028
1029 op->synchronous = true;
1030
|
(2) Event path: |
Condition "pcmk__str_eq(op->action, "meta-data", pcmk__str_casei)", taking true branch. |
1031 if (pcmk__str_eq(op->action, PCMK_ACTION_META_DATA, pcmk__str_casei)) {
1032 /* Synchronous meta-data operations are handled specially. Since most
1033 * resource classes don't provide any meta-data, it has to be
1034 * synthesized from available information about the agent.
1035 *
1036 * services_action_async() doesn't treat meta-data actions specially, so
1037 * it will result in an error for classes that don't support the action.
1038 */
|
(3) Event string_null_source: |
Function "execute_metadata_action" does not terminate string "op->stdout_data". [details] |
|
(4) Event path: |
Condition "execute_metadata_action(op) == pcmk_rc_ok", taking true branch. |
1039 rc = (execute_metadata_action(op) == pcmk_rc_ok);
|
(5) Event path: |
Falling through to end of if statement. |
1040 } else {
1041 rc = (execute_action(op) == pcmk_rc_ok);
1042 }
|
(6) Event path: |
Switch case default. |
|
(7) Event path: |
Condition "trace_cs == NULL", taking true branch. |
|
(8) Event path: |
Condition "crm_is_callsite_active(trace_cs, _level, 0)", taking false branch. |
|
(9) Event path: |
Breaking from switch. |
1043 crm_trace(" > " PCMK__OP_FMT ": %s = %d",
1044 op->rsc, op->action, op->interval_ms, op->opaque->exec, op->rc);
|
(10) Event path: |
Condition "op->stdout_data", taking true branch. |
1045 if (op->stdout_data) {
|
(11) Event path: |
Switch case default. |
|
(12) Event path: |
Condition "trace_cs == NULL", taking true branch. |
|
(13) Event path: |
Condition "crm_is_callsite_active(trace_cs, _level, 0)", taking false branch. |
|
(14) Event path: |
Breaking from switch. |
1046 crm_trace(" > stdout: %s", op->stdout_data);
1047 }
|
(15) Event path: |
Condition "op->stderr_data", taking true branch. |
1048 if (op->stderr_data) {
|
(16) Event path: |
Switch case default. |
|
(17) Event path: |
Condition "trace_cs == NULL", taking true branch. |
|
(18) Event path: |
Condition "crm_is_callsite_active(trace_cs, _level, 0)", taking false branch. |
|
(19) Event path: |
Breaking from switch. |
1049 crm_trace(" > stderr: %s", op->stderr_data);
1050 }
1051 return rc;
1052 }
1053
1054 GList *
1055 get_directory_list(const char *root, gboolean files, gboolean executable)
1056 {
1057 return services_os_get_directory_list(root, files, executable);
1058 }
1059
1060 GList *
1061 resources_list_standards(void)
1062 {
1063 GList *standards = NULL;
1064
1065 standards = g_list_append(standards, strdup(PCMK_RESOURCE_CLASS_OCF));
1066 standards = g_list_append(standards, strdup(PCMK_RESOURCE_CLASS_LSB));
1067 standards = g_list_append(standards, strdup(PCMK_RESOURCE_CLASS_SERVICE));
1068
1069 #if SUPPORT_SYSTEMD
1070 {
1071 GList *agents = systemd_unit_listall();
1072
1073 if (agents != NULL) {
1074 standards = g_list_append(standards,
1075 strdup(PCMK_RESOURCE_CLASS_SYSTEMD));
1076 g_list_free_full(agents, free);
1077 }
1078 }
1079 #endif
1080
1081 #if SUPPORT_UPSTART
1082 {
1083 GList *agents = upstart_job_listall();
1084
1085 if (agents != NULL) {
1086 standards = g_list_append(standards,
1087 strdup(PCMK_RESOURCE_CLASS_UPSTART));
1088 g_list_free_full(agents, free);
1089 }
1090 }
1091 #endif
1092
1093 #if SUPPORT_NAGIOS
1094 {
1095 GList *agents = services__list_nagios_agents();
1096
1097 if (agents != NULL) {
1098 standards = g_list_append(standards,
1099 strdup(PCMK_RESOURCE_CLASS_NAGIOS));
1100 g_list_free_full(agents, free);
1101 }
1102 }
1103 #endif
1104
1105 return standards;
1106 }
1107
1108 GList *
1109 resources_list_providers(const char *standard)
1110 {
1111 if (pcmk_is_set(pcmk_get_ra_caps(standard), pcmk_ra_cap_provider)) {
1112 return resources_os_list_ocf_providers();
1113 }
1114
1115 return NULL;
1116 }
1117
1118 GList *
1119 resources_list_agents(const char *standard, const char *provider)
1120 {
1121 if ((standard == NULL)
1122 || (strcasecmp(standard, PCMK_RESOURCE_CLASS_SERVICE) == 0)) {
1123
1124 GList *tmp1;
1125 GList *tmp2;
1126 GList *result = services__list_lsb_agents();
1127
1128 if (standard == NULL) {
1129 tmp1 = result;
1130 tmp2 = resources_os_list_ocf_agents(NULL);
1131 if (tmp2) {
1132 result = g_list_concat(tmp1, tmp2);
1133 }
1134 }
1135 #if SUPPORT_SYSTEMD
1136 tmp1 = result;
1137 tmp2 = systemd_unit_listall();
1138 if (tmp2) {
1139 result = g_list_concat(tmp1, tmp2);
1140 }
1141 #endif
1142
1143 #if SUPPORT_UPSTART
1144 tmp1 = result;
1145 tmp2 = upstart_job_listall();
1146 if (tmp2) {
1147 result = g_list_concat(tmp1, tmp2);
1148 }
1149 #endif
1150
1151 return result;
1152
1153 } else if (strcasecmp(standard, PCMK_RESOURCE_CLASS_OCF) == 0) {
1154 return resources_os_list_ocf_agents(provider);
1155 } else if (strcasecmp(standard, PCMK_RESOURCE_CLASS_LSB) == 0) {
1156 return services__list_lsb_agents();
1157 #if SUPPORT_SYSTEMD
1158 } else if (strcasecmp(standard, PCMK_RESOURCE_CLASS_SYSTEMD) == 0) {
1159 return systemd_unit_listall();
1160 #endif
1161 #if SUPPORT_UPSTART
1162 } else if (strcasecmp(standard, PCMK_RESOURCE_CLASS_UPSTART) == 0) {
1163 return upstart_job_listall();
1164 #endif
1165 #if SUPPORT_NAGIOS
1166 } else if (strcasecmp(standard, PCMK_RESOURCE_CLASS_NAGIOS) == 0) {
1167 return services__list_nagios_agents();
1168 #endif
1169 }
1170
1171 return NULL;
1172 }
1173
1174 gboolean
1175 resources_agent_exists(const char *standard, const char *provider, const char *agent)
1176 {
1177 GList *standards = NULL;
1178 GList *providers = NULL;
1179 GList *iter = NULL;
1180 gboolean rc = FALSE;
1181 gboolean has_providers = FALSE;
1182
1183 standards = resources_list_standards();
1184 for (iter = standards; iter != NULL; iter = iter->next) {
1185 if (pcmk__str_eq(iter->data, standard, pcmk__str_none)) {
1186 rc = TRUE;
1187 break;
1188 }
1189 }
1190
1191 if (rc == FALSE) {
1192 goto done;
1193 }
1194
1195 rc = FALSE;
1196
1197 has_providers = pcmk_is_set(pcmk_get_ra_caps(standard), pcmk_ra_cap_provider);
1198 if (has_providers == TRUE && provider != NULL) {
1199 providers = resources_list_providers(standard);
1200 for (iter = providers; iter != NULL; iter = iter->next) {
1201 if (pcmk__str_eq(iter->data, provider, pcmk__str_none)) {
1202 rc = TRUE;
1203 break;
1204 }
1205 }
1206 } else if (has_providers == FALSE && provider == NULL) {
1207 rc = TRUE;
1208 }
1209
1210 if (rc == FALSE) {
1211 goto done;
1212 }
1213
1214 if (pcmk__str_eq(standard, PCMK_RESOURCE_CLASS_SERVICE, pcmk__str_casei)) {
1215 if (services__lsb_agent_exists(agent)) {
1216 rc = TRUE;
1217 #if SUPPORT_SYSTEMD
1218 } else if (systemd_unit_exists(agent)) {
1219 rc = TRUE;
1220 #endif
1221
1222 #if SUPPORT_UPSTART
1223 } else if (upstart_job_exists(agent)) {
1224 rc = TRUE;
1225 #endif
1226 } else {
1227 rc = FALSE;
1228 }
1229
1230 } else if (pcmk__str_eq(standard, PCMK_RESOURCE_CLASS_OCF, pcmk__str_casei)) {
1231 rc = services__ocf_agent_exists(provider, agent);
1232
1233 } else if (pcmk__str_eq(standard, PCMK_RESOURCE_CLASS_LSB, pcmk__str_casei)) {
1234 rc = services__lsb_agent_exists(agent);
1235
1236 #if SUPPORT_SYSTEMD
1237 } else if (pcmk__str_eq(standard, PCMK_RESOURCE_CLASS_SYSTEMD, pcmk__str_casei)) {
1238 rc = systemd_unit_exists(agent);
1239 #endif
1240
1241 #if SUPPORT_UPSTART
1242 } else if (pcmk__str_eq(standard, PCMK_RESOURCE_CLASS_UPSTART, pcmk__str_casei)) {
1243 rc = upstart_job_exists(agent);
1244 #endif
1245
1246 #if SUPPORT_NAGIOS
1247 } else if (pcmk__str_eq(standard, PCMK_RESOURCE_CLASS_NAGIOS, pcmk__str_casei)) {
1248 rc = services__nagios_agent_exists(agent);
1249 #endif
1250
1251 } else {
1252 rc = FALSE;
1253 }
1254
1255 done:
1256 g_list_free(standards);
1257 g_list_free(providers);
1258 return rc;
1259 }
1260
1261 /*!
1262 * \internal
1263 * \brief Set the result of an action
1264 *
1265 * \param[out] action Where to set action result
1266 * \param[in] agent_status Exit status to set
1267 * \param[in] exec_status Execution status to set
1268 * \param[in] reason Human-friendly description of event to set
1269 */
1270 void
1271 services__set_result(svc_action_t *action, int agent_status,
1272 enum pcmk_exec_status exec_status, const char *reason)
1273 {
1274 if (action == NULL) {
1275 return;
1276 }
1277
1278 action->rc = agent_status;
1279 action->status = exec_status;
1280
1281 if (!pcmk__str_eq(action->opaque->exit_reason, reason,
1282 pcmk__str_none)) {
1283 free(action->opaque->exit_reason);
1284 action->opaque->exit_reason = (reason == NULL)? NULL : strdup(reason);
1285 }
1286 }
1287
1288 /*!
1289 * \internal
1290 * \brief Set the result of an action, with a formatted exit reason
1291 *
1292 * \param[out] action Where to set action result
1293 * \param[in] agent_status Exit status to set
1294 * \param[in] exec_status Execution status to set
1295 * \param[in] format printf-style format for a human-friendly
1296 * description of reason for result
1297 * \param[in] ... arguments for \p format
1298 */
1299 void
1300 services__format_result(svc_action_t *action, int agent_status,
1301 enum pcmk_exec_status exec_status,
1302 const char *format, ...)
1303 {
1304 va_list ap;
1305 int len = 0;
1306 char *reason = NULL;
1307
1308 if (action == NULL) {
1309 return;
1310 }
1311
1312 action->rc = agent_status;
1313 action->status = exec_status;
1314
1315 if (format != NULL) {
1316 va_start(ap, format);
1317 len = vasprintf(&reason, format, ap);
1318 CRM_ASSERT(len > 0);
1319 va_end(ap);
1320 }
1321 free(action->opaque->exit_reason);
1322 action->opaque->exit_reason = reason;
1323 }
1324
1325 /*!
1326 * \internal
1327 * \brief Set the result of an action to cancelled
1328 *
1329 * \param[out] action Where to set action result
1330 *
1331 * \note This sets execution status but leaves the exit status unchanged
1332 */
1333 void
1334 services__set_cancelled(svc_action_t *action)
1335 {
1336 if (action != NULL) {
1337 action->status = PCMK_EXEC_CANCELLED;
1338 free(action->opaque->exit_reason);
1339 action->opaque->exit_reason = NULL;
1340 }
1341 }
1342
1343 /*!
1344 * \internal
1345 * \brief Get a readable description of what an action is for
1346 *
1347 * \param[in] action Action to check
1348 *
1349 * \return Readable name for the kind of \p action
1350 */
1351 const char *
1352 services__action_kind(const svc_action_t *action)
1353 {
1354 if ((action == NULL) || (action->standard == NULL)) {
1355 return "Process";
1356 } else if (pcmk__str_eq(action->standard, PCMK_RESOURCE_CLASS_STONITH,
1357 pcmk__str_none)) {
1358 return "Fence agent";
1359 } else if (pcmk__str_eq(action->standard, PCMK_RESOURCE_CLASS_ALERT,
1360 pcmk__str_none)) {
1361 return "Alert agent";
1362 } else {
1363 return "Resource agent";
1364 }
1365 }
1366
1367 /*!
1368 * \internal
1369 * \brief Get the exit reason of an action
1370 *
1371 * \param[in] action Action to check
1372 *
1373 * \return Action's exit reason (or NULL if none)
1374 */
1375 const char *
1376 services__exit_reason(const svc_action_t *action)
1377 {
1378 return action->opaque->exit_reason;
1379 }
1380
1381 /*!
1382 * \internal
1383 * \brief Steal stdout from an action
1384 *
1385 * \param[in,out] action Action whose stdout is desired
1386 *
1387 * \return Action's stdout (which may be NULL)
1388 * \note Upon return, \p action will no longer track the output, so it is the
1389 * caller's responsibility to free the return value.
1390 */
1391 char *
1392 services__grab_stdout(svc_action_t *action)
1393 {
1394 char *output = action->stdout_data;
1395
1396 action->stdout_data = NULL;
1397 return output;
1398 }
1399
1400 /*!
1401 * \internal
1402 * \brief Steal stderr from an action
1403 *
1404 * \param[in,out] action Action whose stderr is desired
1405 *
1406 * \return Action's stderr (which may be NULL)
1407 * \note Upon return, \p action will no longer track the output, so it is the
1408 * caller's responsibility to free the return value.
1409 */
1410 char *
1411 services__grab_stderr(svc_action_t *action)
1412 {
1413 char *output = action->stderr_data;
1414
1415 action->stderr_data = NULL;
1416 return output;
1417 }
1418