f78e8ea25adf53bed9ad917e87338deaac328c95
[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
48 #include <urcu.h>
49 #include <urcu/list.h>
50
51 struct lttng_notification_channel *rotate_notification_channel = NULL;
52
53 struct rotation_thread {
54 struct lttng_poll_event events;
55 };
56
57 struct rotation_thread_job {
58 enum rotation_thread_job_type type;
59 uint64_t session_id;
60 /* List member in struct rotation_thread_timer_queue. */
61 struct cds_list_head head;
62 };
63
64 /*
65 * The timer thread enqueues jobs and wakes up the rotation thread.
66 * When the rotation thread wakes up, it empties the queue.
67 */
68 struct rotation_thread_timer_queue {
69 struct lttng_pipe *event_pipe;
70 struct cds_list_head list;
71 pthread_mutex_t lock;
72 };
73
74 struct rotation_thread_handle {
75 struct rotation_thread_timer_queue *rotation_timer_queue;
76 /* Access to the notification thread cmd_queue */
77 struct notification_thread_handle *notification_thread_handle;
78 sem_t *notification_thread_ready;
79 };
80
81 static
82 const char *get_job_type_str(enum rotation_thread_job_type job_type)
83 {
84 switch (job_type) {
85 case ROTATION_THREAD_JOB_TYPE_CHECK_PENDING_ROTATION:
86 return "CHECK_PENDING_ROTATION";
87 case ROTATION_THREAD_JOB_TYPE_SCHEDULED_ROTATION:
88 return "SCHEDULED_ROTATION";
89 default:
90 abort();
91 }
92 }
93
94 struct rotation_thread_timer_queue *rotation_thread_timer_queue_create(void)
95 {
96 struct rotation_thread_timer_queue *queue = NULL;
97
98 queue = zmalloc(sizeof(*queue));
99 if (!queue) {
100 PERROR("Failed to allocate timer rotate queue");
101 goto end;
102 }
103
104 queue->event_pipe = lttng_pipe_open(FD_CLOEXEC | O_NONBLOCK);
105 CDS_INIT_LIST_HEAD(&queue->list);
106 pthread_mutex_init(&queue->lock, NULL);
107 end:
108 return queue;
109 }
110
111 void log_job_destruction(const struct rotation_thread_job *job)
112 {
113 enum lttng_error_level log_level;
114 const char *job_type_str = get_job_type_str(job->type);
115
116 switch (job->type) {
117 case ROTATION_THREAD_JOB_TYPE_SCHEDULED_ROTATION:
118 /*
119 * Not a problem, the scheduled rotation is racing with the teardown
120 * of the daemon. In this case, the rotation will not happen, which
121 * is not a problem (or at least, not important enough to delay
122 * the shutdown of the session daemon).
123 */
124 log_level = PRINT_DBG;
125 break;
126 case ROTATION_THREAD_JOB_TYPE_CHECK_PENDING_ROTATION:
127 /* This is not expected to happen; warn the user. */
128 log_level = PRINT_WARN;
129 break;
130 default:
131 abort();
132 }
133
134 LOG(log_level, "Rotation thread timer queue still contains job of type %s targeting session %" PRIu64 " on destruction",
135 job_type_str, job->session_id);
136 }
137
138 void rotation_thread_timer_queue_destroy(
139 struct rotation_thread_timer_queue *queue)
140 {
141 struct rotation_thread_job *job, *tmp_job;
142
143 if (!queue) {
144 return;
145 }
146
147 lttng_pipe_destroy(queue->event_pipe);
148
149 pthread_mutex_lock(&queue->lock);
150 /* Empty wait queue. */
151 cds_list_for_each_entry_safe(job, tmp_job, &queue->list, head) {
152 log_job_destruction(job);
153 cds_list_del(&job->head);
154 free(job);
155 }
156 pthread_mutex_unlock(&queue->lock);
157 pthread_mutex_destroy(&queue->lock);
158 free(queue);
159 }
160
161 /*
162 * Destroy the thread data previously created by the init function.
163 */
164 void rotation_thread_handle_destroy(
165 struct rotation_thread_handle *handle)
166 {
167 free(handle);
168 }
169
170 struct rotation_thread_handle *rotation_thread_handle_create(
171 struct rotation_thread_timer_queue *rotation_timer_queue,
172 struct notification_thread_handle *notification_thread_handle,
173 sem_t *notification_thread_ready)
174 {
175 struct rotation_thread_handle *handle;
176
177 handle = zmalloc(sizeof(*handle));
178 if (!handle) {
179 goto end;
180 }
181
182 handle->rotation_timer_queue = rotation_timer_queue;
183 handle->notification_thread_handle = notification_thread_handle;
184 handle->notification_thread_ready = notification_thread_ready;
185
186 end:
187 return handle;
188 }
189
190 /*
191 * Called with the rotation_thread_timer_queue lock held.
192 * Return true if the same timer job already exists in the queue, false if not.
193 */
194 static
195 bool timer_job_exists(const struct rotation_thread_timer_queue *queue,
196 enum rotation_thread_job_type job_type, uint64_t session_id)
197 {
198 bool exists = false;
199 struct rotation_thread_job *job;
200
201 cds_list_for_each_entry(job, &queue->list, head) {
202 if (job->session_id == session_id && job->type == job_type) {
203 exists = true;
204 goto end;
205 }
206 }
207 end:
208 return exists;
209 }
210
211 void rotation_thread_enqueue_job(struct rotation_thread_timer_queue *queue,
212 enum rotation_thread_job_type job_type, uint64_t session_id)
213 {
214 int ret;
215 const char * const dummy = "!";
216 struct rotation_thread_job *job = NULL;
217 const char *job_type_str = get_job_type_str(job_type);
218
219 pthread_mutex_lock(&queue->lock);
220 if (timer_job_exists(queue, session_id, job_type)) {
221 /*
222 * This timer job is already pending, we don't need to add
223 * it.
224 */
225 goto end;
226 }
227
228 job = zmalloc(sizeof(struct rotation_thread_job));
229 if (!job) {
230 PERROR("Failed to allocate rotation thread job of type \"%s\" for session id %" PRIu64,
231 job_type_str, session_id);
232 goto end;
233 }
234 job->type = job_type;
235 job->session_id = session_id;
236 cds_list_add_tail(&job->head, &queue->list);
237
238 ret = lttng_write(lttng_pipe_get_writefd(queue->event_pipe), dummy,
239 1);
240 if (ret < 0) {
241 /*
242 * We do not want to block in the timer handler, the job has
243 * been enqueued in the list, the wakeup pipe is probably full,
244 * the job will be processed when the rotation_thread catches
245 * up.
246 */
247 if (errno == EAGAIN || errno == EWOULDBLOCK) {
248 /*
249 * Not an error, but would be surprising and indicate
250 * that the rotation thread can't keep up with the
251 * current load.
252 */
253 DBG("Wake-up pipe of rotation thread job queue is full");
254 goto end;
255 }
256 PERROR("Failed to wake-up the rotation thread after pushing a job of type \"%s\" for session id %" PRIu64,
257 job_type_str, session_id);
258 goto end;
259 }
260
261 end:
262 pthread_mutex_unlock(&queue->lock);
263 }
264
265 static
266 int init_poll_set(struct lttng_poll_event *poll_set,
267 struct rotation_thread_handle *handle)
268 {
269 int ret;
270
271 /*
272 * Create pollset with size 2:
273 * - quit pipe,
274 * - rotation thread timer queue pipe,
275 */
276 ret = sessiond_set_thread_pollset(poll_set, 2);
277 if (ret) {
278 goto error;
279 }
280 ret = lttng_poll_add(poll_set,
281 lttng_pipe_get_readfd(handle->rotation_timer_queue->event_pipe),
282 LPOLLIN | LPOLLERR);
283 if (ret < 0) {
284 ERR("[rotation-thread] Failed to add rotate_pending fd to pollset");
285 goto error;
286 }
287
288 return ret;
289 error:
290 lttng_poll_clean(poll_set);
291 return ret;
292 }
293
294 static
295 void fini_thread_state(struct rotation_thread *state)
296 {
297 lttng_poll_clean(&state->events);
298 if (rotate_notification_channel) {
299 lttng_notification_channel_destroy(rotate_notification_channel);
300 }
301 }
302
303 static
304 int init_thread_state(struct rotation_thread_handle *handle,
305 struct rotation_thread *state)
306 {
307 int ret;
308
309 memset(state, 0, sizeof(*state));
310 lttng_poll_init(&state->events);
311
312 ret = init_poll_set(&state->events, handle);
313 if (ret) {
314 ERR("[rotation-thread] Failed to initialize rotation thread poll set");
315 goto end;
316 }
317
318 /*
319 * We wait until the notification thread is ready to create the
320 * notification channel and add it to the poll_set.
321 */
322 sem_wait(handle->notification_thread_ready);
323 rotate_notification_channel = lttng_notification_channel_create(
324 lttng_session_daemon_notification_endpoint);
325 if (!rotate_notification_channel) {
326 ERR("[rotation-thread] Could not create notification channel");
327 ret = -1;
328 goto end;
329 }
330 ret = lttng_poll_add(&state->events, rotate_notification_channel->socket,
331 LPOLLIN | LPOLLERR);
332 if (ret < 0) {
333 ERR("[rotation-thread] Failed to add notification fd to pollset");
334 goto end;
335 }
336
337 end:
338 return ret;
339 }
340
341 static
342 int check_session_rotation_pending_local_on_consumer(
343 const struct ltt_session *session,
344 struct consumer_socket *socket, bool *rotation_completed)
345 {
346 int ret;
347
348 pthread_mutex_lock(socket->lock);
349 DBG("[rotation-thread] Checking for locally pending rotation on the %s consumer for session %s",
350 lttng_consumer_type_str(socket->type),
351 session->name);
352 ret = consumer_check_rotation_pending_local(socket,
353 session->id,
354 session->current_archive_id - 1);
355 pthread_mutex_unlock(socket->lock);
356
357 if (ret == 0) {
358 /* Rotation was completed on this consumer. */
359 DBG("[rotation-thread] Local rotation of trace archive %" PRIu64 " of session \"%s\" was completed on the %s consumer",
360 session->current_archive_id - 1,
361 session->name,
362 lttng_consumer_type_str(socket->type));
363 *rotation_completed = true;
364 } else if (ret == 1) {
365 /* Rotation pending on this consumer. */
366 DBG("[rotation-thread] Local rotation of trace archive %" PRIu64 " of session \"%s\" is pending on the %s consumer",
367 session->current_archive_id - 1,
368 session->name,
369 lttng_consumer_type_str(socket->type));
370 *rotation_completed = false;
371 ret = 0;
372 } else {
373 /* Not a fatal error. */
374 ERR("[rotation-thread] Encountered an error when checking if local rotation of trace archive %" PRIu64 " of session \"%s\" is pending on the %s consumer",
375 session->current_archive_id - 1,
376 session->name,
377 lttng_consumer_type_str(socket->type));
378 *rotation_completed = false;
379 }
380 return ret;
381 }
382
383 static
384 int check_session_rotation_pending_local(struct ltt_session *session)
385 {
386 int ret = 0;
387 struct consumer_socket *socket;
388 struct cds_lfht_iter iter;
389 bool rotation_completed = true;
390
391 /*
392 * Check for a local pending rotation on all consumers (32-bit
393 * user space, 64-bit user space, and kernel).
394 */
395 DBG("[rotation-thread] Checking for pending local rotation on session \"%s\", trace archive %" PRIu64,
396 session->name, session->current_archive_id - 1);
397
398 rcu_read_lock();
399 if (!session->ust_session) {
400 goto skip_ust;
401 }
402 cds_lfht_for_each_entry(session->ust_session->consumer->socks->ht,
403 &iter, socket, node.node) {
404 ret = check_session_rotation_pending_local_on_consumer(session,
405 socket, &rotation_completed);
406 if (ret || !rotation_completed) {
407 goto end;
408 }
409 }
410
411 skip_ust:
412 if (!session->kernel_session) {
413 goto skip_kernel;
414 }
415 cds_lfht_for_each_entry(session->kernel_session->consumer->socks->ht,
416 &iter, socket, node.node) {
417 ret = check_session_rotation_pending_local_on_consumer(session,
418 socket, &rotation_completed);
419 if (ret || !rotation_completed) {
420 goto end;
421 }
422 }
423 skip_kernel:
424 end:
425 rcu_read_unlock();
426
427 if (rotation_completed) {
428 DBG("[rotation-thread] Local rotation of trace archive %" PRIu64 " of session \"%s\" is complete on all consumers",
429 session->current_archive_id - 1,
430 session->name);
431 session->rotation_pending_local = false;
432 }
433 if (ret) {
434 ret = session_reset_rotation_state(session,
435 LTTNG_ROTATION_STATE_ERROR);
436 if (ret) {
437 ERR("Failed to reset rotation state of session \"%s\"",
438 session->name);
439 }
440 }
441 return 0;
442 }
443
444 static
445 int check_session_rotation_pending_relay(struct ltt_session *session)
446 {
447 int ret;
448 struct consumer_socket *socket;
449 struct cds_lfht_iter iter;
450 bool rotation_completed = true;
451 const struct consumer_output *output;
452
453 /*
454 * Check for a pending rotation on any consumer as we only use
455 * it as a "tunnel" to the relayd.
456 */
457
458 rcu_read_lock();
459 if (session->ust_session) {
460 cds_lfht_first(session->ust_session->consumer->socks->ht,
461 &iter);
462 output = session->ust_session->consumer;
463 } else {
464 cds_lfht_first(session->kernel_session->consumer->socks->ht,
465 &iter);
466 output = session->kernel_session->consumer;
467 }
468 assert(cds_lfht_iter_get_node(&iter));
469
470 socket = caa_container_of(cds_lfht_iter_get_node(&iter),
471 typeof(*socket), node.node);
472
473 pthread_mutex_lock(socket->lock);
474 DBG("[rotation-thread] Checking for pending relay rotation on session \"%s\", trace archive %" PRIu64 " through the %s consumer",
475 session->name, session->current_archive_id - 1,
476 lttng_consumer_type_str(socket->type));
477 ret = consumer_check_rotation_pending_relay(socket,
478 output,
479 session->id,
480 session->current_archive_id - 1);
481 pthread_mutex_unlock(socket->lock);
482
483 if (ret == 0) {
484 /* Rotation was completed on the relay. */
485 DBG("[rotation-thread] Relay rotation of trace archive %" PRIu64 " of session \"%s\" was completed",
486 session->current_archive_id - 1,
487 session->name);
488 } else if (ret == 1) {
489 /* Rotation pending on relay. */
490 DBG("[rotation-thread] Relay rotation of trace archive %" PRIu64 " of session \"%s\" is pending",
491 session->current_archive_id - 1,
492 session->name);
493 rotation_completed = false;
494 } else {
495 /* Not a fatal error. */
496 ERR("[rotation-thread] Encountered an error when checking if rotation of trace archive %" PRIu64 " of session \"%s\" is pending on the relay",
497 session->current_archive_id - 1,
498 session->name);
499 ret = session_reset_rotation_state(session,
500 LTTNG_ROTATION_STATE_ERROR);
501 if (ret) {
502 ERR("Failed to reset rotation state of session \"%s\"",
503 session->name);
504 }
505 rotation_completed = false;
506 }
507
508 rcu_read_unlock();
509
510 if (rotation_completed) {
511 DBG("[rotation-thread] Rotation of trace archive %" PRIu64 " of session \"%s\" is complete on the relay",
512 session->current_archive_id - 1,
513 session->name);
514 session->rotation_pending_relay = false;
515 }
516 return 0;
517 }
518
519 /*
520 * Check if the last rotation was completed, called with session lock held.
521 */
522 static
523 int check_session_rotation_pending(struct ltt_session *session,
524 struct notification_thread_handle *notification_thread_handle)
525 {
526 int ret;
527 struct lttng_trace_archive_location *location;
528 time_t now;
529
530 DBG("[rotation-thread] Checking for pending rotation on session \"%s\", trace archive %" PRIu64,
531 session->name, session->current_archive_id - 1);
532
533 if (session->rotation_pending_local) {
534 /* Updates session->rotation_pending_local as needed. */
535 ret = check_session_rotation_pending_local(session);
536 if (ret) {
537 goto end;
538 }
539
540 /*
541 * No need to check for a pending rotation on the relay
542 * since the rotation is not even completed locally yet.
543 */
544 if (session->rotation_pending_local) {
545 goto end;
546 }
547 }
548
549 if (session->rotation_pending_relay) {
550 /* Updates session->rotation_pending_relay as needed. */
551 ret = check_session_rotation_pending_relay(session);
552 if (ret) {
553 goto end;
554 }
555
556 if (session->rotation_pending_relay) {
557 goto end;
558 }
559 }
560
561 DBG("[rotation-thread] Rotation of trace archive %" PRIu64 " completed for "
562 "session %s", session->current_archive_id - 1,
563 session->name);
564
565 /* Rename the completed trace archive's location. */
566 now = time(NULL);
567 if (now == (time_t) -1) {
568 ret = session_reset_rotation_state(session,
569 LTTNG_ROTATION_STATE_ERROR);
570 if (ret) {
571 ERR("Failed to reset rotation state of session \"%s\"",
572 session->name);
573 }
574 ret = LTTNG_ERR_UNK;
575 goto end;
576 }
577
578 ret = rename_completed_chunk(session, now);
579 if (ret < 0) {
580 ERR("Failed to rename completed rotation chunk");
581 goto end;
582 }
583 session->last_chunk_start_ts = session->current_chunk_start_ts;
584
585 /*
586 * Now we can clear the "ONGOING" state in the session. New
587 * rotations can start now.
588 */
589 session->rotation_state = LTTNG_ROTATION_STATE_COMPLETED;
590
591 /* Ownership of location is transferred. */
592 location = session_get_trace_archive_location(session);
593 ret = notification_thread_command_session_rotation_completed(
594 notification_thread_handle,
595 session->name,
596 session->uid,
597 session->gid,
598 session->current_archive_id,
599 location);
600 if (ret != LTTNG_OK) {
601 ERR("[rotation-thread] Failed to notify notification thread of completed rotation for session %s",
602 session->name);
603 }
604
605 if (!session->active) {
606 /*
607 * A stop command was issued during the rotation, it is
608 * up to the rotation completion check to perform the
609 * renaming of the last chunk that was produced.
610 */
611 ret = notification_thread_command_session_rotation_ongoing(
612 notification_thread_handle,
613 session->name,
614 session->uid,
615 session->gid,
616 session->current_archive_id);
617 if (ret != LTTNG_OK) {
618 ERR("[rotation-thread] Failed to notify notification thread of completed rotation for session %s",
619 session->name);
620 }
621
622 ret = rename_active_chunk(session);
623 if (ret < 0) {
624 ERR("[rotation-thread] Failed to rename active rotation chunk");
625 goto end;
626 }
627
628 /* Ownership of location is transferred. */
629 location = session_get_trace_archive_location(session);
630 ret = notification_thread_command_session_rotation_completed(
631 notification_thread_handle,
632 session->name,
633 session->uid,
634 session->gid,
635 session->current_archive_id,
636 location);
637 if (ret != LTTNG_OK) {
638 ERR("[rotation-thread] Failed to notify notification thread of completed rotation for session %s",
639 session->name);
640 }
641 }
642
643 ret = 0;
644 end:
645 if (session->rotation_state == LTTNG_ROTATION_STATE_ONGOING) {
646 DBG("[rotation-thread] Rotation of trace archive %" PRIu64 " is still pending for session %s",
647 session->current_archive_id - 1, session->name);
648 ret = timer_session_rotation_pending_check_start(session,
649 DEFAULT_ROTATE_PENDING_TIMER);
650 if (ret) {
651 ERR("Re-enabling rotate pending timer");
652 ret = -1;
653 goto end;
654 }
655 }
656
657 return ret;
658 }
659
660 /* Call with the session and session_list locks held. */
661 static
662 int launch_session_rotation(struct ltt_session *session)
663 {
664 int ret;
665 struct lttng_rotate_session_return rotation_return;
666
667 DBG("[rotation-thread] Launching scheduled time-based rotation on session \"%s\"",
668 session->name);
669
670 ret = cmd_rotate_session(session, &rotation_return);
671 if (ret == LTTNG_OK) {
672 DBG("[rotation-thread] Scheduled time-based rotation successfully launched on session \"%s\"",
673 session->name);
674 } else {
675 /* Don't consider errors as fatal. */
676 DBG("[rotation-thread] Scheduled time-based rotation aborted for session %s: %s",
677 session->name, lttng_strerror(ret));
678 }
679 return 0;
680 }
681
682 static
683 int run_job(struct rotation_thread_job *job, struct ltt_session *session,
684 struct notification_thread_handle *notification_thread_handle)
685 {
686 int ret;
687
688 switch (job->type) {
689 case ROTATION_THREAD_JOB_TYPE_SCHEDULED_ROTATION:
690 ret = launch_session_rotation(session);
691 break;
692 case ROTATION_THREAD_JOB_TYPE_CHECK_PENDING_ROTATION:
693 ret = check_session_rotation_pending(session,
694 notification_thread_handle);
695 break;
696 default:
697 abort();
698 }
699 return ret;
700 }
701
702 static
703 int handle_job_queue(struct rotation_thread_handle *handle,
704 struct rotation_thread *state,
705 struct rotation_thread_timer_queue *queue)
706 {
707 int ret = 0;
708 int fd = lttng_pipe_get_readfd(queue->event_pipe);
709 struct ltt_session *session;
710 char buf;
711
712 ret = lttng_read(fd, &buf, 1);
713 if (ret != 1) {
714 ERR("[rotation-thread] Failed to read from wakeup pipe (fd = %i)", fd);
715 ret = -1;
716 goto end;
717 }
718
719 for (;;) {
720 struct rotation_thread_job *job;
721
722 /* Take the queue lock only to pop an element from the list. */
723 pthread_mutex_lock(&queue->lock);
724 if (cds_list_empty(&queue->list)) {
725 pthread_mutex_unlock(&queue->lock);
726 break;
727 }
728 job = cds_list_first_entry(&queue->list,
729 typeof(*job), head);
730 cds_list_del(&job->head);
731 pthread_mutex_unlock(&queue->lock);
732
733 session_lock_list();
734 session = session_find_by_id(job->session_id);
735 if (!session) {
736 DBG("[rotation-thread] Session %" PRIu64 " not found",
737 job->session_id);
738 /*
739 * This is a non-fatal error, and we cannot report it to
740 * the user (timer), so just print the error and
741 * continue the processing.
742 *
743 * While the timer thread will purge pending signals for
744 * a session on the session's destruction, it is
745 * possible for a job targeting that session to have
746 * already been queued before it was destroyed.
747 */
748 session_unlock_list();
749 free(job);
750 continue;
751 }
752
753 session_lock(session);
754 ret = run_job(job, session, handle->notification_thread_handle);
755 session_unlock(session);
756 session_unlock_list();
757 free(job);
758 if (ret) {
759 goto end;
760 }
761 }
762
763 ret = 0;
764
765 end:
766 return ret;
767 }
768
769 static
770 int handle_condition(const struct lttng_condition *condition,
771 const struct lttng_evaluation *evaluation,
772 struct notification_thread_handle *notification_thread_handle)
773 {
774 int ret = 0;
775 const char *condition_session_name = NULL;
776 enum lttng_condition_type condition_type;
777 enum lttng_condition_status condition_status;
778 enum lttng_evaluation_status evaluation_status;
779 uint64_t consumed;
780 struct ltt_session *session;
781
782 condition_type = lttng_condition_get_type(condition);
783
784 if (condition_type != LTTNG_CONDITION_TYPE_SESSION_CONSUMED_SIZE) {
785 ret = -1;
786 ERR("[rotation-thread] Condition type and session usage type are not the same");
787 goto end;
788 }
789
790 /* Fetch info to test */
791 condition_status = lttng_condition_session_consumed_size_get_session_name(
792 condition, &condition_session_name);
793 if (condition_status != LTTNG_CONDITION_STATUS_OK) {
794 ERR("[rotation-thread] Session name could not be fetched");
795 ret = -1;
796 goto end;
797 }
798 evaluation_status = lttng_evaluation_session_consumed_size_get_consumed_size(evaluation,
799 &consumed);
800 if (evaluation_status != LTTNG_EVALUATION_STATUS_OK) {
801 ERR("[rotation-thread] Failed to get evaluation");
802 ret = -1;
803 goto end;
804 }
805
806 session_lock_list();
807 session = session_find_by_name(condition_session_name);
808 if (!session) {
809 ret = -1;
810 session_unlock_list();
811 ERR("[rotation-thread] Session \"%s\" not found",
812 condition_session_name);
813 goto end;
814 }
815 session_lock(session);
816 session_unlock_list();
817
818 ret = unsubscribe_session_consumed_size_rotation(session,
819 notification_thread_handle);
820 if (ret) {
821 goto end_unlock;
822 }
823
824 ret = cmd_rotate_session(session, NULL);
825 if (ret == -LTTNG_ERR_ROTATION_PENDING) {
826 DBG("Rotate already pending, subscribe to the next threshold value");
827 } else if (ret != LTTNG_OK) {
828 ERR("[rotation-thread] Failed to rotate on size notification with error: %s",
829 lttng_strerror(ret));
830 ret = -1;
831 goto end_unlock;
832 }
833 ret = subscribe_session_consumed_size_rotation(session,
834 consumed + session->rotate_size,
835 notification_thread_handle);
836 if (ret) {
837 ERR("[rotation-thread] Failed to subscribe to session consumed size condition");
838 goto end_unlock;
839 }
840 ret = 0;
841
842 end_unlock:
843 session_unlock(session);
844 end:
845 return ret;
846 }
847
848 static
849 int handle_notification_channel(int fd,
850 struct rotation_thread_handle *handle,
851 struct rotation_thread *state)
852 {
853 int ret;
854 bool notification_pending;
855 struct lttng_notification *notification = NULL;
856 enum lttng_notification_channel_status status;
857 const struct lttng_evaluation *notification_evaluation;
858 const struct lttng_condition *notification_condition;
859
860 status = lttng_notification_channel_has_pending_notification(
861 rotate_notification_channel, &notification_pending);
862 if (status != LTTNG_NOTIFICATION_CHANNEL_STATUS_OK) {
863 ERR("[rotation-thread ]Error occurred while checking for pending notification");
864 ret = -1;
865 goto end;
866 }
867
868 if (!notification_pending) {
869 ret = 0;
870 goto end;
871 }
872
873 /* Receive the next notification. */
874 status = lttng_notification_channel_get_next_notification(
875 rotate_notification_channel,
876 &notification);
877
878 switch (status) {
879 case LTTNG_NOTIFICATION_CHANNEL_STATUS_OK:
880 break;
881 case LTTNG_NOTIFICATION_CHANNEL_STATUS_NOTIFICATIONS_DROPPED:
882 /* Not an error, we will wait for the next one */
883 ret = 0;
884 goto end;;
885 case LTTNG_NOTIFICATION_CHANNEL_STATUS_CLOSED:
886 ERR("Notification channel was closed");
887 ret = -1;
888 goto end;
889 default:
890 /* Unhandled conditions / errors. */
891 ERR("Unknown notification channel status");
892 ret = -1;
893 goto end;
894 }
895
896 notification_condition = lttng_notification_get_condition(notification);
897 notification_evaluation = lttng_notification_get_evaluation(notification);
898
899 ret = handle_condition(notification_condition, notification_evaluation,
900 handle->notification_thread_handle);
901
902 end:
903 lttng_notification_destroy(notification);
904 return ret;
905 }
906
907 void *thread_rotation(void *data)
908 {
909 int ret;
910 struct rotation_thread_handle *handle = data;
911 struct rotation_thread thread;
912
913 DBG("[rotation-thread] Started rotation thread");
914
915 if (!handle) {
916 ERR("[rotation-thread] Invalid thread context provided");
917 goto end;
918 }
919
920 rcu_register_thread();
921 rcu_thread_online();
922
923 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_ROTATION);
924 health_code_update();
925
926 ret = init_thread_state(handle, &thread);
927 if (ret) {
928 goto error;
929 }
930
931 /* Ready to handle client connections. */
932 sessiond_notify_ready();
933
934 while (true) {
935 int fd_count, i;
936
937 health_poll_entry();
938 DBG("[rotation-thread] Entering poll wait");
939 ret = lttng_poll_wait(&thread.events, -1);
940 DBG("[rotation-thread] Poll wait returned (%i)", ret);
941 health_poll_exit();
942 if (ret < 0) {
943 /*
944 * Restart interrupted system call.
945 */
946 if (errno == EINTR) {
947 continue;
948 }
949 ERR("[rotation-thread] Error encountered during lttng_poll_wait (%i)", ret);
950 goto error;
951 }
952
953 fd_count = ret;
954 for (i = 0; i < fd_count; i++) {
955 int fd = LTTNG_POLL_GETFD(&thread.events, i);
956 uint32_t revents = LTTNG_POLL_GETEV(&thread.events, i);
957
958 DBG("[rotation-thread] Handling fd (%i) activity (%u)",
959 fd, revents);
960
961 if (revents & LPOLLERR) {
962 ERR("[rotation-thread] Polling returned an error on fd %i", fd);
963 goto error;
964 }
965
966 if (sessiond_check_thread_quit_pipe(fd, revents)) {
967 DBG("[rotation-thread] Quit pipe activity");
968 /* TODO flush the queue. */
969 goto exit;
970 } else if (fd == lttng_pipe_get_readfd(handle->rotation_timer_queue->event_pipe)) {
971 ret = handle_job_queue(handle, &thread,
972 handle->rotation_timer_queue);
973 if (ret) {
974 ERR("[rotation-thread] Failed to handle rotation timer pipe event");
975 goto error;
976 }
977 } else if (fd == rotate_notification_channel->socket) {
978 ret = handle_notification_channel(fd, handle,
979 &thread);
980 if (ret) {
981 ERR("[rotation-thread] Error occurred while handling activity on notification channel socket");
982 goto error;
983 }
984 }
985 }
986 }
987 exit:
988 error:
989 DBG("[rotation-thread] Exit");
990 fini_thread_state(&thread);
991 health_unregister(health_sessiond);
992 rcu_thread_offline();
993 rcu_unregister_thread();
994 end:
995 return NULL;
996 }
This page took 0.051455 seconds and 4 git commands to generate.