Fix: keep ust/kernel session items around for destroy notifier
[lttng-tools.git] / src / bin / lttng-sessiond / rotation-thread.c
1 /*
2 * Copyright (C) 2017 - Julien Desfossez <jdesfossez@efficios.com>
3 * Copyright (C) 2018 - Jérémie Galarneau <jeremie.galarneau@efficios.com>
4 *
5 * This program is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License, version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
12 * more details.
13 *
14 * You should have received a copy of the GNU General Public License along with
15 * this program; if not, write to the Free Software Foundation, Inc., 51
16 * Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
17 */
18
19 #define _LGPL_SOURCE
20 #include <lttng/trigger/trigger.h>
21 #include <common/error.h>
22 #include <common/config/session-config.h>
23 #include <common/defaults.h>
24 #include <common/utils.h>
25 #include <common/futex.h>
26 #include <common/align.h>
27 #include <common/time.h>
28 #include <common/hashtable/utils.h>
29 #include <sys/eventfd.h>
30 #include <sys/stat.h>
31 #include <time.h>
32 #include <signal.h>
33 #include <inttypes.h>
34
35 #include <common/kernel-ctl/kernel-ctl.h>
36 #include <lttng/notification/channel-internal.h>
37 #include <lttng/rotate-internal.h>
38
39 #include "rotation-thread.h"
40 #include "lttng-sessiond.h"
41 #include "health-sessiond.h"
42 #include "rotate.h"
43 #include "cmd.h"
44 #include "session.h"
45 #include "timer.h"
46 #include "notification-thread-commands.h"
47 #include "utils.h"
48 #include "thread.h"
49
50 #include <urcu.h>
51 #include <urcu/list.h>
52
53 struct lttng_notification_channel *rotate_notification_channel = NULL;
54
55 struct rotation_thread {
56 struct lttng_poll_event events;
57 };
58
59 struct rotation_thread_job {
60 enum rotation_thread_job_type type;
61 struct ltt_session *session;
62 /* List member in struct rotation_thread_timer_queue. */
63 struct cds_list_head head;
64 };
65
66 /*
67 * The timer thread enqueues jobs and wakes up the rotation thread.
68 * When the rotation thread wakes up, it empties the queue.
69 */
70 struct rotation_thread_timer_queue {
71 struct lttng_pipe *event_pipe;
72 struct cds_list_head list;
73 pthread_mutex_t lock;
74 };
75
76 struct rotation_thread_handle {
77 struct rotation_thread_timer_queue *rotation_timer_queue;
78 /* Access to the notification thread cmd_queue */
79 struct notification_thread_handle *notification_thread_handle;
80 /* Thread-specific quit pipe. */
81 struct lttng_pipe *quit_pipe;
82 };
83
84 static
85 const char *get_job_type_str(enum rotation_thread_job_type job_type)
86 {
87 switch (job_type) {
88 case ROTATION_THREAD_JOB_TYPE_CHECK_PENDING_ROTATION:
89 return "CHECK_PENDING_ROTATION";
90 case ROTATION_THREAD_JOB_TYPE_SCHEDULED_ROTATION:
91 return "SCHEDULED_ROTATION";
92 default:
93 abort();
94 }
95 }
96
97 struct rotation_thread_timer_queue *rotation_thread_timer_queue_create(void)
98 {
99 struct rotation_thread_timer_queue *queue = NULL;
100
101 queue = zmalloc(sizeof(*queue));
102 if (!queue) {
103 PERROR("Failed to allocate timer rotate queue");
104 goto end;
105 }
106
107 queue->event_pipe = lttng_pipe_open(FD_CLOEXEC | O_NONBLOCK);
108 CDS_INIT_LIST_HEAD(&queue->list);
109 pthread_mutex_init(&queue->lock, NULL);
110 end:
111 return queue;
112 }
113
114 void log_job_destruction(const struct rotation_thread_job *job)
115 {
116 enum lttng_error_level log_level;
117 const char *job_type_str = get_job_type_str(job->type);
118
119 switch (job->type) {
120 case ROTATION_THREAD_JOB_TYPE_SCHEDULED_ROTATION:
121 /*
122 * Not a problem, the scheduled rotation is racing with the teardown
123 * of the daemon. In this case, the rotation will not happen, which
124 * is not a problem (or at least, not important enough to delay
125 * the shutdown of the session daemon).
126 */
127 log_level = PRINT_DBG;
128 break;
129 case ROTATION_THREAD_JOB_TYPE_CHECK_PENDING_ROTATION:
130 /* This is not expected to happen; warn the user. */
131 log_level = PRINT_WARN;
132 break;
133 default:
134 abort();
135 }
136
137 LOG(log_level, "Rotation thread timer queue still contains job of type %s targeting session \"%s\" on destruction",
138 job_type_str, job->session->name);
139 }
140
141 void rotation_thread_timer_queue_destroy(
142 struct rotation_thread_timer_queue *queue)
143 {
144 if (!queue) {
145 return;
146 }
147
148 lttng_pipe_destroy(queue->event_pipe);
149
150 pthread_mutex_lock(&queue->lock);
151 assert(cds_list_empty(&queue->list));
152 pthread_mutex_unlock(&queue->lock);
153 pthread_mutex_destroy(&queue->lock);
154 free(queue);
155 }
156
157 /*
158 * Destroy the thread data previously created by the init function.
159 */
160 void rotation_thread_handle_destroy(
161 struct rotation_thread_handle *handle)
162 {
163 lttng_pipe_destroy(handle->quit_pipe);
164 free(handle);
165 }
166
167 struct rotation_thread_handle *rotation_thread_handle_create(
168 struct rotation_thread_timer_queue *rotation_timer_queue,
169 struct notification_thread_handle *notification_thread_handle)
170 {
171 struct rotation_thread_handle *handle;
172
173 handle = zmalloc(sizeof(*handle));
174 if (!handle) {
175 goto end;
176 }
177
178 handle->rotation_timer_queue = rotation_timer_queue;
179 handle->notification_thread_handle = notification_thread_handle;
180 handle->quit_pipe = lttng_pipe_open(FD_CLOEXEC);
181 if (!handle->quit_pipe) {
182 goto error;
183 }
184
185 end:
186 return handle;
187 error:
188 rotation_thread_handle_destroy(handle);
189 return NULL;
190 }
191
192 /*
193 * Called with the rotation_thread_timer_queue lock held.
194 * Return true if the same timer job already exists in the queue, false if not.
195 */
196 static
197 bool timer_job_exists(const struct rotation_thread_timer_queue *queue,
198 enum rotation_thread_job_type job_type,
199 struct ltt_session *session)
200 {
201 bool exists = false;
202 struct rotation_thread_job *job;
203
204 cds_list_for_each_entry(job, &queue->list, head) {
205 if (job->session == session && job->type == job_type) {
206 exists = true;
207 goto end;
208 }
209 }
210 end:
211 return exists;
212 }
213
214 void rotation_thread_enqueue_job(struct rotation_thread_timer_queue *queue,
215 enum rotation_thread_job_type job_type,
216 struct ltt_session *session)
217 {
218 int ret;
219 const char dummy = '!';
220 struct rotation_thread_job *job = NULL;
221 const char *job_type_str = get_job_type_str(job_type);
222
223 pthread_mutex_lock(&queue->lock);
224 if (timer_job_exists(queue, job_type, session)) {
225 /*
226 * This timer job is already pending, we don't need to add
227 * it.
228 */
229 goto end;
230 }
231
232 job = zmalloc(sizeof(struct rotation_thread_job));
233 if (!job) {
234 PERROR("Failed to allocate rotation thread job of type \"%s\" for session \"%s\"",
235 job_type_str, session->name);
236 goto end;
237 }
238 /* No reason for this to fail as the caller must hold a reference. */
239 (void) session_get(session);
240
241 job->session = session;
242 job->type = job_type;
243 cds_list_add_tail(&job->head, &queue->list);
244
245 ret = lttng_write(lttng_pipe_get_writefd(queue->event_pipe), &dummy,
246 sizeof(dummy));
247 if (ret < 0) {
248 /*
249 * We do not want to block in the timer handler, the job has
250 * been enqueued in the list, the wakeup pipe is probably full,
251 * the job will be processed when the rotation_thread catches
252 * up.
253 */
254 if (errno == EAGAIN || errno == EWOULDBLOCK) {
255 /*
256 * Not an error, but would be surprising and indicate
257 * that the rotation thread can't keep up with the
258 * current load.
259 */
260 DBG("Wake-up pipe of rotation thread job queue is full");
261 goto end;
262 }
263 PERROR("Failed to wake-up the rotation thread after pushing a job of type \"%s\" for session \"%s\"",
264 job_type_str, session->name);
265 goto end;
266 }
267
268 end:
269 pthread_mutex_unlock(&queue->lock);
270 }
271
272 static
273 int init_poll_set(struct lttng_poll_event *poll_set,
274 struct rotation_thread_handle *handle)
275 {
276 int ret;
277
278 /*
279 * Create pollset with size 3:
280 * - rotation thread quit pipe,
281 * - rotation thread timer queue pipe,
282 * - notification channel sock,
283 */
284 ret = lttng_poll_create(poll_set, 5, LTTNG_CLOEXEC);
285 if (ret < 0) {
286 goto error;
287 }
288
289 ret = lttng_poll_add(poll_set,
290 lttng_pipe_get_readfd(handle->quit_pipe),
291 LPOLLIN | LPOLLERR);
292 if (ret < 0) {
293 ERR("[rotation-thread] Failed to add quit pipe read fd to poll set");
294 goto error;
295 }
296
297 ret = lttng_poll_add(poll_set,
298 lttng_pipe_get_readfd(handle->rotation_timer_queue->event_pipe),
299 LPOLLIN | LPOLLERR);
300 if (ret < 0) {
301 ERR("[rotation-thread] Failed to add rotate_pending fd to poll set");
302 goto error;
303 }
304
305 return ret;
306 error:
307 lttng_poll_clean(poll_set);
308 return ret;
309 }
310
311 static
312 void fini_thread_state(struct rotation_thread *state)
313 {
314 lttng_poll_clean(&state->events);
315 if (rotate_notification_channel) {
316 lttng_notification_channel_destroy(rotate_notification_channel);
317 }
318 }
319
320 static
321 int init_thread_state(struct rotation_thread_handle *handle,
322 struct rotation_thread *state)
323 {
324 int ret;
325
326 memset(state, 0, sizeof(*state));
327 lttng_poll_init(&state->events);
328
329 ret = init_poll_set(&state->events, handle);
330 if (ret) {
331 ERR("[rotation-thread] Failed to initialize rotation thread poll set");
332 goto end;
333 }
334
335 rotate_notification_channel = lttng_notification_channel_create(
336 lttng_session_daemon_notification_endpoint);
337 if (!rotate_notification_channel) {
338 ERR("[rotation-thread] Could not create notification channel");
339 ret = -1;
340 goto end;
341 }
342 ret = lttng_poll_add(&state->events, rotate_notification_channel->socket,
343 LPOLLIN | LPOLLERR);
344 if (ret < 0) {
345 ERR("[rotation-thread] Failed to add notification fd to pollset");
346 goto end;
347 }
348
349 end:
350 return ret;
351 }
352
353 static
354 void check_session_rotation_pending_on_consumers(struct ltt_session *session,
355 bool *_rotation_completed)
356 {
357 int ret = 0;
358 struct consumer_socket *socket;
359 struct cds_lfht_iter iter;
360 enum consumer_trace_chunk_exists_status exists_status;
361 uint64_t relayd_id;
362 bool chunk_exists_on_peer = false;
363 enum lttng_trace_chunk_status chunk_status;
364
365 assert(session->chunk_being_archived);
366
367 /*
368 * Check for a local pending rotation on all consumers (32-bit
369 * user space, 64-bit user space, and kernel).
370 */
371 rcu_read_lock();
372 if (!session->ust_session) {
373 goto skip_ust;
374 }
375 cds_lfht_for_each_entry(session->ust_session->consumer->socks->ht,
376 &iter, socket, node.node) {
377 relayd_id = session->ust_session->consumer->type == CONSUMER_DST_LOCAL ?
378 -1ULL :
379 session->ust_session->consumer->net_seq_index;
380
381 pthread_mutex_lock(socket->lock);
382 ret = consumer_trace_chunk_exists(socket,
383 relayd_id,
384 session->id, session->chunk_being_archived,
385 &exists_status);
386 if (ret) {
387 pthread_mutex_unlock(socket->lock);
388 ERR("Error occured while checking rotation status on consumer daemon");
389 goto end;
390 }
391
392 if (exists_status != CONSUMER_TRACE_CHUNK_EXISTS_STATUS_UNKNOWN_CHUNK) {
393 pthread_mutex_unlock(socket->lock);
394 chunk_exists_on_peer = true;
395 goto end;
396 }
397 pthread_mutex_unlock(socket->lock);
398 }
399
400 skip_ust:
401 if (!session->kernel_session) {
402 goto skip_kernel;
403 }
404 cds_lfht_for_each_entry(session->kernel_session->consumer->socks->ht,
405 &iter, socket, node.node) {
406 pthread_mutex_lock(socket->lock);
407 relayd_id = session->kernel_session->consumer->type == CONSUMER_DST_LOCAL ?
408 -1ULL :
409 session->kernel_session->consumer->net_seq_index;
410
411 ret = consumer_trace_chunk_exists(socket,
412 relayd_id,
413 session->id, session->chunk_being_archived,
414 &exists_status);
415 if (ret) {
416 pthread_mutex_unlock(socket->lock);
417 ERR("Error occured while checking rotation status on consumer daemon");
418 goto end;
419 }
420
421 if (exists_status != CONSUMER_TRACE_CHUNK_EXISTS_STATUS_UNKNOWN_CHUNK) {
422 pthread_mutex_unlock(socket->lock);
423 chunk_exists_on_peer = true;
424 goto end;
425 }
426 pthread_mutex_unlock(socket->lock);
427 }
428 skip_kernel:
429 end:
430 rcu_read_unlock();
431
432 if (!chunk_exists_on_peer) {
433 uint64_t chunk_being_archived_id;
434
435 chunk_status = lttng_trace_chunk_get_id(
436 session->chunk_being_archived,
437 &chunk_being_archived_id);
438 assert(chunk_status == LTTNG_TRACE_CHUNK_STATUS_OK);
439 DBG("[rotation-thread] Rotation of trace archive %" PRIu64 " of session \"%s\" is complete on all consumers",
440 chunk_being_archived_id,
441 session->name);
442 }
443 *_rotation_completed = !chunk_exists_on_peer;
444 if (ret) {
445 ret = session_reset_rotation_state(session,
446 LTTNG_ROTATION_STATE_ERROR);
447 if (ret) {
448 ERR("Failed to reset rotation state of session \"%s\"",
449 session->name);
450 }
451 }
452 }
453
454 /*
455 * Check if the last rotation was completed, called with session lock held.
456 * Should only return non-zero in the event of a fatal error. Doing so will
457 * shutdown the thread.
458 */
459 static
460 int check_session_rotation_pending(struct ltt_session *session,
461 struct notification_thread_handle *notification_thread_handle)
462 {
463 int ret;
464 struct lttng_trace_archive_location *location;
465 enum lttng_trace_chunk_status chunk_status;
466 bool rotation_completed = false;
467 const char *archived_chunk_name;
468 uint64_t chunk_being_archived_id;
469
470 chunk_status = lttng_trace_chunk_get_id(session->chunk_being_archived,
471 &chunk_being_archived_id);
472 assert(chunk_status == LTTNG_TRACE_CHUNK_STATUS_OK);
473
474 DBG("[rotation-thread] Checking for pending rotation on session \"%s\", trace archive %" PRIu64,
475 session->name, chunk_being_archived_id);
476
477 if (!session->chunk_being_archived) {
478 ret = 0;
479 goto end;
480 }
481
482 /*
483 * The rotation-pending check timer of a session is launched in
484 * one-shot mode. If the rotation is incomplete, the rotation
485 * thread will re-enable the pending-check timer.
486 *
487 * The timer thread can't stop the timer itself since it is involved
488 * in the check for the timer's quiescence.
489 */
490 ret = timer_session_rotation_pending_check_stop(session);
491 if (ret) {
492 goto end;
493 }
494
495 check_session_rotation_pending_on_consumers(session,
496 &rotation_completed);
497 if (!rotation_completed ||
498 session->rotation_state == LTTNG_ROTATION_STATE_ERROR) {
499 goto end;
500 }
501
502 /*
503 * Now we can clear the "ONGOING" state in the session. New
504 * rotations can start now.
505 */
506 chunk_status = lttng_trace_chunk_get_name(session->chunk_being_archived,
507 &archived_chunk_name, NULL);
508 assert(chunk_status == LTTNG_TRACE_CHUNK_STATUS_OK);
509 free(session->last_archived_chunk_name);
510 session->last_archived_chunk_name = strdup(archived_chunk_name);
511 if (!session->last_archived_chunk_name) {
512 PERROR("Failed to duplicate archived chunk name");
513 }
514 session_reset_rotation_state(session, LTTNG_ROTATION_STATE_COMPLETED);
515
516 if (!session->quiet_rotation) {
517 location = session_get_trace_archive_location(session);
518 /* Ownership of location is transferred. */
519 ret = notification_thread_command_session_rotation_completed(
520 notification_thread_handle,
521 session->name,
522 session->uid,
523 session->gid,
524 session->last_archived_chunk_id.value,
525 location);
526 if (ret != LTTNG_OK) {
527 ERR("[rotation-thread] Failed to notify notification thread of completed rotation for session %s",
528 session->name);
529 }
530 }
531
532 if (!session->active && !session->quiet_rotation) {
533 /*
534 * A stop command was issued during the rotation, it is
535 * up to the rotation completion check to perform the
536 * renaming of the last chunk that was produced.
537 */
538 ret = notification_thread_command_session_rotation_ongoing(
539 notification_thread_handle,
540 session->name,
541 session->uid,
542 session->gid,
543 session->most_recent_chunk_id.value);
544 if (ret != LTTNG_OK) {
545 ERR("[rotation-thread] Failed to notify notification thread of completed rotation for session %s",
546 session->name);
547 }
548
549 /* Ownership of location is transferred. */
550 location = session_get_trace_archive_location(session);
551 ret = notification_thread_command_session_rotation_completed(
552 notification_thread_handle,
553 session->name,
554 session->uid,
555 session->gid,
556 session->most_recent_chunk_id.value,
557 location);
558 if (ret != LTTNG_OK) {
559 ERR("[rotation-thread] Failed to notify notification thread of completed rotation for session %s",
560 session->name);
561 }
562 }
563
564 ret = 0;
565 end:
566 if (session->rotation_state == LTTNG_ROTATION_STATE_ONGOING) {
567 uint64_t chunk_being_archived_id;
568
569 chunk_status = lttng_trace_chunk_get_id(
570 session->chunk_being_archived,
571 &chunk_being_archived_id);
572 assert(chunk_status == LTTNG_TRACE_CHUNK_STATUS_OK);
573
574 DBG("[rotation-thread] Rotation of trace archive %" PRIu64 " is still pending for session %s",
575 chunk_being_archived_id, session->name);
576 ret = timer_session_rotation_pending_check_start(session,
577 DEFAULT_ROTATE_PENDING_TIMER);
578 if (ret) {
579 ERR("Failed to re-enable rotation pending timer");
580 ret = -1;
581 goto end;
582 }
583 }
584
585 return ret;
586 }
587
588 /* Call with the session and session_list locks held. */
589 static
590 int launch_session_rotation(struct ltt_session *session)
591 {
592 int ret;
593 struct lttng_rotate_session_return rotation_return;
594
595 DBG("[rotation-thread] Launching scheduled time-based rotation on session \"%s\"",
596 session->name);
597
598 ret = cmd_rotate_session(session, &rotation_return, false);
599 if (ret == LTTNG_OK) {
600 DBG("[rotation-thread] Scheduled time-based rotation successfully launched on session \"%s\"",
601 session->name);
602 } else {
603 /* Don't consider errors as fatal. */
604 DBG("[rotation-thread] Scheduled time-based rotation aborted for session %s: %s",
605 session->name, lttng_strerror(ret));
606 }
607 return 0;
608 }
609
610 static
611 int run_job(struct rotation_thread_job *job, struct ltt_session *session,
612 struct notification_thread_handle *notification_thread_handle)
613 {
614 int ret;
615
616 switch (job->type) {
617 case ROTATION_THREAD_JOB_TYPE_SCHEDULED_ROTATION:
618 ret = launch_session_rotation(session);
619 break;
620 case ROTATION_THREAD_JOB_TYPE_CHECK_PENDING_ROTATION:
621 ret = check_session_rotation_pending(session,
622 notification_thread_handle);
623 break;
624 default:
625 abort();
626 }
627 return ret;
628 }
629
630 static
631 int handle_job_queue(struct rotation_thread_handle *handle,
632 struct rotation_thread *state,
633 struct rotation_thread_timer_queue *queue)
634 {
635 int ret = 0;
636
637 for (;;) {
638 struct ltt_session *session;
639 struct rotation_thread_job *job;
640
641 /* Take the queue lock only to pop an element from the list. */
642 pthread_mutex_lock(&queue->lock);
643 if (cds_list_empty(&queue->list)) {
644 pthread_mutex_unlock(&queue->lock);
645 break;
646 }
647 job = cds_list_first_entry(&queue->list,
648 typeof(*job), head);
649 cds_list_del(&job->head);
650 pthread_mutex_unlock(&queue->lock);
651
652 session_lock_list();
653 session = job->session;
654 if (!session) {
655 DBG("[rotation-thread] Session \"%s\" not found",
656 session->name);
657 /*
658 * This is a non-fatal error, and we cannot report it to
659 * the user (timer), so just print the error and
660 * continue the processing.
661 *
662 * While the timer thread will purge pending signals for
663 * a session on the session's destruction, it is
664 * possible for a job targeting that session to have
665 * already been queued before it was destroyed.
666 */
667 free(job);
668 session_put(session);
669 session_unlock_list();
670 continue;
671 }
672
673 session_lock(session);
674 ret = run_job(job, session, handle->notification_thread_handle);
675 session_unlock(session);
676 /* Release reference held by the job. */
677 session_put(session);
678 session_unlock_list();
679 free(job);
680 if (ret) {
681 goto end;
682 }
683 }
684
685 ret = 0;
686
687 end:
688 return ret;
689 }
690
691 static
692 int handle_condition(const struct lttng_condition *condition,
693 const struct lttng_evaluation *evaluation,
694 struct notification_thread_handle *notification_thread_handle)
695 {
696 int ret = 0;
697 const char *condition_session_name = NULL;
698 enum lttng_condition_type condition_type;
699 enum lttng_condition_status condition_status;
700 enum lttng_evaluation_status evaluation_status;
701 uint64_t consumed;
702 struct ltt_session *session;
703
704 condition_type = lttng_condition_get_type(condition);
705
706 if (condition_type != LTTNG_CONDITION_TYPE_SESSION_CONSUMED_SIZE) {
707 ret = -1;
708 ERR("[rotation-thread] Condition type and session usage type are not the same");
709 goto end;
710 }
711
712 /* Fetch info to test */
713 condition_status = lttng_condition_session_consumed_size_get_session_name(
714 condition, &condition_session_name);
715 if (condition_status != LTTNG_CONDITION_STATUS_OK) {
716 ERR("[rotation-thread] Session name could not be fetched");
717 ret = -1;
718 goto end;
719 }
720 evaluation_status = lttng_evaluation_session_consumed_size_get_consumed_size(evaluation,
721 &consumed);
722 if (evaluation_status != LTTNG_EVALUATION_STATUS_OK) {
723 ERR("[rotation-thread] Failed to get evaluation");
724 ret = -1;
725 goto end;
726 }
727
728 session_lock_list();
729 session = session_find_by_name(condition_session_name);
730 if (!session) {
731 ret = -1;
732 session_unlock_list();
733 ERR("[rotation-thread] Session \"%s\" not found",
734 condition_session_name);
735 goto end;
736 }
737 session_lock(session);
738
739 ret = unsubscribe_session_consumed_size_rotation(session,
740 notification_thread_handle);
741 if (ret) {
742 goto end_unlock;
743 }
744
745 ret = cmd_rotate_session(session, NULL, false);
746 if (ret == -LTTNG_ERR_ROTATION_PENDING) {
747 DBG("Rotate already pending, subscribe to the next threshold value");
748 } else if (ret != LTTNG_OK) {
749 ERR("[rotation-thread] Failed to rotate on size notification with error: %s",
750 lttng_strerror(ret));
751 ret = -1;
752 goto end_unlock;
753 }
754 ret = subscribe_session_consumed_size_rotation(session,
755 consumed + session->rotate_size,
756 notification_thread_handle);
757 if (ret) {
758 ERR("[rotation-thread] Failed to subscribe to session consumed size condition");
759 goto end_unlock;
760 }
761 ret = 0;
762
763 end_unlock:
764 session_unlock(session);
765 session_put(session);
766 session_unlock_list();
767 end:
768 return ret;
769 }
770
771 static
772 int handle_notification_channel(int fd,
773 struct rotation_thread_handle *handle,
774 struct rotation_thread *state)
775 {
776 int ret;
777 bool notification_pending;
778 struct lttng_notification *notification = NULL;
779 enum lttng_notification_channel_status status;
780 const struct lttng_evaluation *notification_evaluation;
781 const struct lttng_condition *notification_condition;
782
783 status = lttng_notification_channel_has_pending_notification(
784 rotate_notification_channel, &notification_pending);
785 if (status != LTTNG_NOTIFICATION_CHANNEL_STATUS_OK) {
786 ERR("[rotation-thread ]Error occurred while checking for pending notification");
787 ret = -1;
788 goto end;
789 }
790
791 if (!notification_pending) {
792 ret = 0;
793 goto end;
794 }
795
796 /* Receive the next notification. */
797 status = lttng_notification_channel_get_next_notification(
798 rotate_notification_channel,
799 &notification);
800
801 switch (status) {
802 case LTTNG_NOTIFICATION_CHANNEL_STATUS_OK:
803 break;
804 case LTTNG_NOTIFICATION_CHANNEL_STATUS_NOTIFICATIONS_DROPPED:
805 /* Not an error, we will wait for the next one */
806 ret = 0;
807 goto end;;
808 case LTTNG_NOTIFICATION_CHANNEL_STATUS_CLOSED:
809 ERR("Notification channel was closed");
810 ret = -1;
811 goto end;
812 default:
813 /* Unhandled conditions / errors. */
814 ERR("Unknown notification channel status");
815 ret = -1;
816 goto end;
817 }
818
819 notification_condition = lttng_notification_get_condition(notification);
820 notification_evaluation = lttng_notification_get_evaluation(notification);
821
822 ret = handle_condition(notification_condition, notification_evaluation,
823 handle->notification_thread_handle);
824
825 end:
826 lttng_notification_destroy(notification);
827 return ret;
828 }
829
830 void *thread_rotation(void *data)
831 {
832 int ret;
833 struct rotation_thread_handle *handle = data;
834 struct rotation_thread thread;
835 const int queue_pipe_fd = lttng_pipe_get_readfd(
836 handle->rotation_timer_queue->event_pipe);
837
838 DBG("[rotation-thread] Started rotation thread");
839
840 if (!handle) {
841 ERR("[rotation-thread] Invalid thread context provided");
842 goto end;
843 }
844
845 rcu_register_thread();
846 rcu_thread_online();
847
848 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_ROTATION);
849 health_code_update();
850
851 ret = init_thread_state(handle, &thread);
852 if (ret) {
853 goto error;
854 }
855
856 while (true) {
857 int fd_count, i;
858
859 health_poll_entry();
860 DBG("[rotation-thread] Entering poll wait");
861 ret = lttng_poll_wait(&thread.events, -1);
862 DBG("[rotation-thread] Poll wait returned (%i)", ret);
863 health_poll_exit();
864 if (ret < 0) {
865 /*
866 * Restart interrupted system call.
867 */
868 if (errno == EINTR) {
869 continue;
870 }
871 ERR("[rotation-thread] Error encountered during lttng_poll_wait (%i)", ret);
872 goto error;
873 }
874
875 fd_count = ret;
876 for (i = 0; i < fd_count; i++) {
877 int fd = LTTNG_POLL_GETFD(&thread.events, i);
878 uint32_t revents = LTTNG_POLL_GETEV(&thread.events, i);
879
880 DBG("[rotation-thread] Handling fd (%i) activity (%u)",
881 fd, revents);
882
883 if (revents & LPOLLERR) {
884 ERR("[rotation-thread] Polling returned an error on fd %i", fd);
885 goto error;
886 }
887
888 if (fd == rotate_notification_channel->socket) {
889 ret = handle_notification_channel(fd, handle,
890 &thread);
891 if (ret) {
892 ERR("[rotation-thread] Error occurred while handling activity on notification channel socket");
893 goto error;
894 }
895 } else {
896 /* Job queue or quit pipe activity. */
897
898 /*
899 * The job queue is serviced if there is
900 * activity on the quit pipe to ensure it is
901 * flushed and all references held in the queue
902 * are released.
903 */
904 ret = handle_job_queue(handle, &thread,
905 handle->rotation_timer_queue);
906 if (ret) {
907 ERR("[rotation-thread] Failed to handle rotation timer pipe event");
908 goto error;
909 }
910
911 if (fd == queue_pipe_fd) {
912 char buf;
913
914 ret = lttng_read(fd, &buf, 1);
915 if (ret != 1) {
916 ERR("[rotation-thread] Failed to read from wakeup pipe (fd = %i)", fd);
917 ret = -1;
918 goto error;
919 }
920 } else {
921 DBG("[rotation-thread] Quit pipe activity");
922 goto exit;
923 }
924 }
925 }
926 }
927 exit:
928 error:
929 DBG("[rotation-thread] Exit");
930 fini_thread_state(&thread);
931 health_unregister(health_sessiond);
932 rcu_thread_offline();
933 rcu_unregister_thread();
934 end:
935 return NULL;
936 }
937
938 static
939 bool shutdown_rotation_thread(void *thread_data)
940 {
941 struct rotation_thread_handle *handle = thread_data;
942 const int write_fd = lttng_pipe_get_writefd(handle->quit_pipe);
943
944 return notify_thread_pipe(write_fd) == 1;
945 }
946
947 bool launch_rotation_thread(struct rotation_thread_handle *handle)
948 {
949 struct lttng_thread *thread;
950
951 thread = lttng_thread_create("Rotation",
952 thread_rotation,
953 shutdown_rotation_thread,
954 NULL,
955 handle);
956 if (!thread) {
957 goto error;
958 }
959 lttng_thread_put(thread);
960 return true;
961 error:
962 return false;
963 }
This page took 0.05504 seconds and 4 git commands to generate.