Commit | Line | Data |
---|---|---|
db66e574 JD |
1 | /* |
2 | * Copyright (C) 2017 - Julien Desfossez <jdesfossez@efficios.com> | |
92816cc3 | 3 | * Copyright (C) 2018 - Jérémie Galarneau <jeremie.galarneau@efficios.com> |
db66e574 JD |
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> | |
5c408ad8 | 37 | #include <lttng/rotate-internal.h> |
db66e574 JD |
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" | |
8e319828 | 45 | #include "timer.h" |
17dd1232 | 46 | #include "notification-thread-commands.h" |
db66e574 JD |
47 | |
48 | #include <urcu.h> | |
49 | #include <urcu/list.h> | |
db66e574 | 50 | |
90936dcf JD |
51 | struct lttng_notification_channel *rotate_notification_channel = NULL; |
52 | ||
92816cc3 | 53 | struct rotation_thread { |
db66e574 JD |
54 | struct lttng_poll_event events; |
55 | }; | |
56 | ||
92816cc3 JG |
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 | int quit_pipe; | |
76 | struct rotation_thread_timer_queue *rotation_timer_queue; | |
77 | /* Access to the notification thread cmd_queue */ | |
78 | struct notification_thread_handle *notification_thread_handle; | |
79 | sem_t *notification_thread_ready; | |
80 | }; | |
81 | ||
db66e574 | 82 | static |
92816cc3 | 83 | const char *get_job_type_str(enum rotation_thread_job_type job_type) |
db66e574 | 84 | { |
92816cc3 JG |
85 | switch (job_type) { |
86 | case ROTATION_THREAD_JOB_TYPE_CHECK_PENDING_ROTATION: | |
87 | return "CHECK_PENDING_ROTATION"; | |
88 | case ROTATION_THREAD_JOB_TYPE_SCHEDULED_ROTATION: | |
89 | return "SCHEDULED_ROTATION"; | |
90 | default: | |
91 | abort(); | |
92 | } | |
db66e574 JD |
93 | } |
94 | ||
92816cc3 | 95 | struct rotation_thread_timer_queue *rotation_thread_timer_queue_create(void) |
db66e574 | 96 | { |
92816cc3 | 97 | struct rotation_thread_timer_queue *queue = NULL; |
db66e574 | 98 | |
92816cc3 JG |
99 | queue = zmalloc(sizeof(*queue)); |
100 | if (!queue) { | |
101 | PERROR("Failed to allocate timer rotate queue"); | |
102 | goto end; | |
103 | } | |
db66e574 | 104 | |
92816cc3 JG |
105 | queue->event_pipe = lttng_pipe_open(FD_CLOEXEC | O_NONBLOCK); |
106 | CDS_INIT_LIST_HEAD(&queue->list); | |
107 | pthread_mutex_init(&queue->lock, NULL); | |
108 | end: | |
109 | return queue; | |
db66e574 JD |
110 | } |
111 | ||
92816cc3 | 112 | void log_job_destruction(const struct rotation_thread_job *job) |
db66e574 | 113 | { |
92816cc3 JG |
114 | enum lttng_error_level log_level; |
115 | const char *job_type_str = get_job_type_str(job->type); | |
116 | ||
117 | switch (job->type) { | |
118 | case ROTATION_THREAD_JOB_TYPE_SCHEDULED_ROTATION: | |
119 | /* | |
120 | * Not a problem, the scheduled rotation is racing with the teardown | |
121 | * of the daemon. In this case, the rotation will not happen, which | |
122 | * is not a problem (or at least, not important enough to delay | |
123 | * the shutdown of the session daemon). | |
124 | */ | |
125 | log_level = PRINT_DBG; | |
126 | break; | |
127 | case ROTATION_THREAD_JOB_TYPE_CHECK_PENDING_ROTATION: | |
128 | /* This is not expected to happen; warn the user. */ | |
129 | log_level = PRINT_WARN; | |
130 | break; | |
131 | default: | |
132 | abort(); | |
db66e574 JD |
133 | } |
134 | ||
92816cc3 JG |
135 | LOG(log_level, "Rotation thread timer queue still contains job of type %s targeting session %" PRIu64 " on destruction", |
136 | job_type_str, job->session_id); | |
db66e574 JD |
137 | } |
138 | ||
92816cc3 JG |
139 | void rotation_thread_timer_queue_destroy( |
140 | struct rotation_thread_timer_queue *queue) | |
db66e574 | 141 | { |
92816cc3 | 142 | struct rotation_thread_job *job, *tmp_job; |
db66e574 | 143 | |
92816cc3 JG |
144 | if (!queue) { |
145 | return; | |
db66e574 JD |
146 | } |
147 | ||
92816cc3 JG |
148 | lttng_pipe_destroy(queue->event_pipe); |
149 | ||
150 | pthread_mutex_lock(&queue->lock); | |
151 | /* Empty wait queue. */ | |
152 | cds_list_for_each_entry_safe(job, tmp_job, &queue->list, head) { | |
153 | log_job_destruction(job); | |
154 | cds_list_del(&job->head); | |
155 | free(job); | |
db66e574 | 156 | } |
92816cc3 JG |
157 | pthread_mutex_unlock(&queue->lock); |
158 | pthread_mutex_destroy(&queue->lock); | |
159 | free(queue); | |
160 | } | |
db66e574 | 161 | |
92816cc3 JG |
162 | /* |
163 | * Destroy the thread data previously created by the init function. | |
164 | */ | |
165 | void rotation_thread_handle_destroy( | |
166 | struct rotation_thread_handle *handle) | |
167 | { | |
db66e574 JD |
168 | free(handle); |
169 | } | |
170 | ||
171 | struct rotation_thread_handle *rotation_thread_handle_create( | |
92816cc3 | 172 | int quit_pipe, |
90936dcf JD |
173 | struct rotation_thread_timer_queue *rotation_timer_queue, |
174 | struct notification_thread_handle *notification_thread_handle, | |
175 | sem_t *notification_thread_ready) | |
db66e574 JD |
176 | { |
177 | struct rotation_thread_handle *handle; | |
178 | ||
179 | handle = zmalloc(sizeof(*handle)); | |
180 | if (!handle) { | |
181 | goto end; | |
182 | } | |
183 | ||
92816cc3 JG |
184 | handle->quit_pipe = quit_pipe; |
185 | handle->rotation_timer_queue = rotation_timer_queue; | |
186 | handle->notification_thread_handle = notification_thread_handle; | |
187 | handle->notification_thread_ready = notification_thread_ready; | |
188 | ||
189 | end: | |
190 | return handle; | |
191 | } | |
192 | ||
193 | /* | |
194 | * Called with the rotation_thread_timer_queue lock held. | |
195 | * Return true if the same timer job already exists in the queue, false if not. | |
196 | */ | |
197 | static | |
198 | bool timer_job_exists(const struct rotation_thread_timer_queue *queue, | |
199 | enum rotation_thread_job_type job_type, uint64_t session_id) | |
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_id == session_id && job->type == job_type) { | |
206 | exists = true; | |
207 | goto end; | |
db66e574 | 208 | } |
db66e574 | 209 | } |
92816cc3 JG |
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, uint64_t session_id) | |
216 | { | |
217 | int ret; | |
218 | const char * const dummy = "!"; | |
219 | struct rotation_thread_job *job = NULL; | |
220 | const char *job_type_str = get_job_type_str(job_type); | |
221 | ||
222 | pthread_mutex_lock(&queue->lock); | |
223 | if (timer_job_exists(queue, session_id, job_type)) { | |
224 | /* | |
225 | * This timer job is already pending, we don't need to add | |
226 | * it. | |
227 | */ | |
228 | goto end; | |
db66e574 | 229 | } |
92816cc3 JG |
230 | |
231 | job = zmalloc(sizeof(struct rotation_thread_job)); | |
232 | if (!job) { | |
233 | PERROR("Failed to allocate rotation thread job of type \"%s\" for session id %" PRIu64, | |
234 | job_type_str, session_id); | |
235 | goto end; | |
236 | } | |
237 | job->type = job_type; | |
238 | job->session_id = session_id; | |
239 | cds_list_add_tail(&job->head, &queue->list); | |
240 | ||
241 | ret = lttng_write(lttng_pipe_get_writefd(queue->event_pipe), dummy, | |
242 | 1); | |
243 | if (ret < 0) { | |
244 | /* | |
245 | * We do not want to block in the timer handler, the job has | |
246 | * been enqueued in the list, the wakeup pipe is probably full, | |
247 | * the job will be processed when the rotation_thread catches | |
248 | * up. | |
249 | */ | |
250 | if (errno == EAGAIN || errno == EWOULDBLOCK) { | |
251 | /* | |
252 | * Not an error, but would be surprising and indicate | |
253 | * that the rotation thread can't keep up with the | |
254 | * current load. | |
255 | */ | |
256 | DBG("Wake-up pipe of rotation thread job queue is full"); | |
257 | goto end; | |
db66e574 | 258 | } |
92816cc3 JG |
259 | PERROR("Failed to wake-up the rotation thread after pushing a job of type \"%s\" for session id %" PRIu64, |
260 | job_type_str, session_id); | |
261 | goto end; | |
db66e574 | 262 | } |
db66e574 JD |
263 | |
264 | end: | |
92816cc3 | 265 | pthread_mutex_unlock(&queue->lock); |
db66e574 JD |
266 | } |
267 | ||
268 | static | |
269 | int init_poll_set(struct lttng_poll_event *poll_set, | |
270 | struct rotation_thread_handle *handle) | |
271 | { | |
272 | int ret; | |
273 | ||
274 | /* | |
92816cc3 JG |
275 | * Create pollset with size 2: |
276 | * - quit pipe, | |
277 | * - rotation thread timer queue pipe, | |
db66e574 | 278 | */ |
92816cc3 | 279 | ret = lttng_poll_create(poll_set, 2, LTTNG_CLOEXEC); |
db66e574 JD |
280 | if (ret < 0) { |
281 | goto end; | |
282 | } | |
283 | ||
92816cc3 | 284 | ret = lttng_poll_add(poll_set, handle->quit_pipe, |
db66e574 JD |
285 | LPOLLIN | LPOLLERR); |
286 | if (ret < 0) { | |
92816cc3 | 287 | ERR("[rotation-thread] Failed to add quit_pipe fd to pollset"); |
db66e574 JD |
288 | goto error; |
289 | } | |
d086f507 JD |
290 | ret = lttng_poll_add(poll_set, |
291 | lttng_pipe_get_readfd(handle->rotation_timer_queue->event_pipe), | |
292 | LPOLLIN | LPOLLERR); | |
293 | if (ret < 0) { | |
294 | ERR("[rotation-thread] Failed to add rotate_pending fd to pollset"); | |
295 | goto error; | |
296 | } | |
db66e574 JD |
297 | |
298 | end: | |
299 | return ret; | |
300 | error: | |
301 | lttng_poll_clean(poll_set); | |
302 | return ret; | |
303 | } | |
304 | ||
305 | static | |
92816cc3 | 306 | void fini_thread_state(struct rotation_thread *state) |
db66e574 JD |
307 | { |
308 | lttng_poll_clean(&state->events); | |
90936dcf JD |
309 | if (rotate_notification_channel) { |
310 | lttng_notification_channel_destroy(rotate_notification_channel); | |
311 | } | |
db66e574 JD |
312 | } |
313 | ||
314 | static | |
315 | int init_thread_state(struct rotation_thread_handle *handle, | |
92816cc3 | 316 | struct rotation_thread *state) |
db66e574 JD |
317 | { |
318 | int ret; | |
319 | ||
320 | memset(state, 0, sizeof(*state)); | |
321 | lttng_poll_init(&state->events); | |
322 | ||
323 | ret = init_poll_set(&state->events, handle); | |
324 | if (ret) { | |
325 | ERR("[rotation-thread] Failed to initialize rotation thread poll set"); | |
326 | goto end; | |
327 | } | |
328 | ||
90936dcf JD |
329 | /* |
330 | * We wait until the notification thread is ready to create the | |
331 | * notification channel and add it to the poll_set. | |
332 | */ | |
333 | sem_wait(handle->notification_thread_ready); | |
334 | rotate_notification_channel = lttng_notification_channel_create( | |
335 | lttng_session_daemon_notification_endpoint); | |
336 | if (!rotate_notification_channel) { | |
337 | ERR("[rotation-thread] Could not create notification channel"); | |
338 | ret = -1; | |
339 | goto end; | |
340 | } | |
341 | ret = lttng_poll_add(&state->events, rotate_notification_channel->socket, | |
342 | LPOLLIN | LPOLLERR); | |
343 | if (ret < 0) { | |
344 | ERR("[rotation-thread] Failed to add notification fd to pollset"); | |
345 | goto end; | |
346 | } | |
347 | ||
db66e574 JD |
348 | end: |
349 | return ret; | |
350 | } | |
351 | ||
352 | static | |
92816cc3 JG |
353 | int check_session_rotation_pending_local_on_consumer( |
354 | const struct ltt_session *session, | |
355 | struct consumer_socket *socket, bool *rotation_completed) | |
db66e574 | 356 | { |
92816cc3 JG |
357 | int ret; |
358 | ||
359 | pthread_mutex_lock(socket->lock); | |
360 | DBG("[rotation-thread] Checking for locally pending rotation on the %s consumer for session %s", | |
361 | lttng_consumer_type_str(socket->type), | |
362 | session->name); | |
363 | ret = consumer_check_rotation_pending_local(socket, | |
364 | session->id, | |
365 | session->current_archive_id - 1); | |
366 | pthread_mutex_unlock(socket->lock); | |
367 | ||
368 | if (ret == 0) { | |
369 | /* Rotation was completed on this consumer. */ | |
370 | DBG("[rotation-thread] Local rotation of trace archive %" PRIu64 " of session \"%s\" was completed on the %s consumer", | |
371 | session->current_archive_id - 1, | |
372 | session->name, | |
373 | lttng_consumer_type_str(socket->type)); | |
374 | *rotation_completed = true; | |
375 | } else if (ret == 1) { | |
376 | /* Rotation pending on this consumer. */ | |
377 | DBG("[rotation-thread] Local rotation of trace archive %" PRIu64 " of session \"%s\" is pending on the %s consumer", | |
378 | session->current_archive_id - 1, | |
379 | session->name, | |
380 | lttng_consumer_type_str(socket->type)); | |
381 | *rotation_completed = false; | |
382 | ret = 0; | |
db66e574 | 383 | } else { |
92816cc3 JG |
384 | /* Not a fatal error. */ |
385 | ERR("[rotation-thread] Encountered an error when checking if local rotation of trace archive %" PRIu64 " of session \"%s\" is pending on the %s consumer", | |
386 | session->current_archive_id - 1, | |
387 | session->name, | |
388 | lttng_consumer_type_str(socket->type)); | |
389 | *rotation_completed = false; | |
db66e574 | 390 | } |
92816cc3 JG |
391 | return ret; |
392 | } | |
db66e574 | 393 | |
92816cc3 JG |
394 | static |
395 | int check_session_rotation_pending_local(struct ltt_session *session) | |
396 | { | |
db582e11 | 397 | int ret = 0; |
92816cc3 JG |
398 | struct consumer_socket *socket; |
399 | struct cds_lfht_iter iter; | |
400 | bool rotation_completed = true; | |
401 | ||
402 | /* | |
403 | * Check for a local pending rotation on all consumers (32-bit | |
404 | * user space, 64-bit user space, and kernel). | |
405 | */ | |
406 | DBG("[rotation-thread] Checking for pending local rotation on session \"%s\", trace archive %" PRIu64, | |
407 | session->name, session->current_archive_id - 1); | |
408 | ||
409 | rcu_read_lock(); | |
410 | if (!session->ust_session) { | |
411 | goto skip_ust; | |
412 | } | |
413 | cds_lfht_for_each_entry(session->ust_session->consumer->socks->ht, | |
414 | &iter, socket, node.node) { | |
415 | ret = check_session_rotation_pending_local_on_consumer(session, | |
416 | socket, &rotation_completed); | |
417 | if (ret || !rotation_completed) { | |
418 | goto end; | |
db66e574 | 419 | } |
db66e574 JD |
420 | } |
421 | ||
92816cc3 JG |
422 | skip_ust: |
423 | if (!session->kernel_session) { | |
424 | goto skip_kernel; | |
db66e574 | 425 | } |
92816cc3 JG |
426 | cds_lfht_for_each_entry(session->kernel_session->consumer->socks->ht, |
427 | &iter, socket, node.node) { | |
428 | ret = check_session_rotation_pending_local_on_consumer(session, | |
429 | socket, &rotation_completed); | |
430 | if (ret || !rotation_completed) { | |
431 | goto end; | |
432 | } | |
433 | } | |
434 | skip_kernel: | |
435 | end: | |
436 | rcu_read_unlock(); | |
db66e574 | 437 | |
92816cc3 JG |
438 | if (rotation_completed) { |
439 | DBG("[rotation-thread] Local rotation of trace archive %" PRIu64 " of session \"%s\" is complete on all consumers", | |
440 | session->current_archive_id - 1, | |
441 | session->name); | |
442 | session->rotation_pending_local = false; | |
db66e574 | 443 | } |
92816cc3 JG |
444 | if (ret) { |
445 | session->rotation_state = LTTNG_ROTATION_STATE_ERROR; | |
db66e574 | 446 | } |
92816cc3 JG |
447 | return 0; |
448 | } | |
db66e574 | 449 | |
92816cc3 JG |
450 | static |
451 | int check_session_rotation_pending_relay(struct ltt_session *session) | |
452 | { | |
453 | int ret; | |
454 | struct consumer_socket *socket; | |
455 | struct cds_lfht_iter iter; | |
456 | bool rotation_completed = true; | |
457 | const struct consumer_output *output; | |
db66e574 | 458 | |
92816cc3 JG |
459 | /* |
460 | * Check for a pending rotation on any consumer as we only use | |
461 | * it as a "tunnel" to the relayd. | |
462 | */ | |
17dd1232 | 463 | |
92816cc3 JG |
464 | rcu_read_lock(); |
465 | if (session->ust_session) { | |
466 | cds_lfht_first(session->ust_session->consumer->socks->ht, | |
467 | &iter); | |
468 | output = session->ust_session->consumer; | |
469 | } else { | |
470 | cds_lfht_first(session->kernel_session->consumer->socks->ht, | |
471 | &iter); | |
472 | output = session->kernel_session->consumer; | |
db66e574 | 473 | } |
92816cc3 | 474 | assert(cds_lfht_iter_get_node(&iter)); |
db66e574 | 475 | |
92816cc3 JG |
476 | socket = caa_container_of(cds_lfht_iter_get_node(&iter), |
477 | typeof(*socket), node.node); | |
478 | ||
479 | pthread_mutex_lock(socket->lock); | |
480 | DBG("[rotation-thread] Checking for pending relay rotation on session \"%s\", trace archive %" PRIu64 " through the %s consumer", | |
481 | session->name, session->current_archive_id - 1, | |
482 | lttng_consumer_type_str(socket->type)); | |
483 | ret = consumer_check_rotation_pending_relay(socket, | |
484 | output, | |
485 | session->id, | |
486 | session->current_archive_id - 1); | |
487 | pthread_mutex_unlock(socket->lock); | |
488 | ||
489 | if (ret == 0) { | |
490 | /* Rotation was completed on the relay. */ | |
491 | DBG("[rotation-thread] Relay rotation of trace archive %" PRIu64 " of session \"%s\" was completed", | |
492 | session->current_archive_id - 1, | |
493 | session->name); | |
494 | } else if (ret == 1) { | |
495 | /* Rotation pending on relay. */ | |
496 | DBG("[rotation-thread] Relay rotation of trace archive %" PRIu64 " of session \"%s\" is pending", | |
497 | session->current_archive_id - 1, | |
498 | session->name); | |
499 | rotation_completed = false; | |
500 | } else { | |
501 | /* Not a fatal error. */ | |
502 | ERR("[rotation-thread] Encountered an error when checking if rotation of trace archive %" PRIu64 " of session \"%s\" is pending on the relay", | |
503 | session->current_archive_id - 1, | |
504 | session->name); | |
505 | session->rotation_state = LTTNG_ROTATION_STATE_ERROR; | |
506 | rotation_completed = false; | |
507 | } | |
db66e574 | 508 | |
db66e574 | 509 | rcu_read_unlock(); |
92816cc3 JG |
510 | |
511 | if (rotation_completed) { | |
512 | DBG("[rotation-thread] Totation of trace archive %" PRIu64 " of session \"%s\" is complete on the relay", | |
513 | session->current_archive_id - 1, | |
514 | session->name); | |
515 | session->rotation_pending_relay = false; | |
516 | } | |
517 | return 0; | |
db66e574 JD |
518 | } |
519 | ||
d88744a4 | 520 | /* |
92816cc3 | 521 | * Check if the last rotation was completed, called with session lock held. |
d88744a4 JD |
522 | */ |
523 | static | |
92816cc3 JG |
524 | int check_session_rotation_pending(struct ltt_session *session, |
525 | struct notification_thread_handle *notification_thread_handle) | |
d88744a4 JD |
526 | { |
527 | int ret; | |
92816cc3 JG |
528 | struct lttng_trace_archive_location *location; |
529 | time_t now; | |
d88744a4 | 530 | |
92816cc3 JG |
531 | DBG("[rotation-thread] Checking for pending rotation on session \"%s\", trace archive %" PRIu64, |
532 | session->name, session->current_archive_id - 1); | |
533 | ||
534 | if (session->rotation_pending_local) { | |
535 | /* Updates session->rotation_pending_local as needed. */ | |
536 | ret = check_session_rotation_pending_local(session); | |
537 | if (ret) { | |
538 | goto end; | |
539 | } | |
17dd1232 | 540 | |
d88744a4 | 541 | /* |
92816cc3 JG |
542 | * No need to check for a pending rotation on the relay |
543 | * since the rotation is not even completed locally yet. | |
d88744a4 | 544 | */ |
92816cc3 JG |
545 | if (session->rotation_pending_local) { |
546 | goto end; | |
17dd1232 | 547 | } |
92816cc3 JG |
548 | } |
549 | ||
550 | if (session->rotation_pending_relay) { | |
551 | /* Updates session->rotation_pending_relay as needed. */ | |
552 | ret = check_session_rotation_pending_relay(session); | |
d88744a4 | 553 | if (ret) { |
92816cc3 JG |
554 | goto end; |
555 | } | |
556 | ||
557 | if (session->rotation_pending_relay) { | |
d88744a4 JD |
558 | goto end; |
559 | } | |
560 | } | |
561 | ||
92816cc3 JG |
562 | DBG("[rotation-thread] Rotation of trace archive %" PRIu64 " completed for " |
563 | "session %s", session->current_archive_id - 1, | |
564 | session->name); | |
d88744a4 | 565 | |
92816cc3 JG |
566 | /* Rename the completed trace archive's location. */ |
567 | now = time(NULL); | |
568 | if (now == (time_t) -1) { | |
569 | session->rotation_state = LTTNG_ROTATION_STATE_ERROR; | |
570 | ret = LTTNG_ERR_UNK; | |
571 | goto end; | |
572 | } | |
573 | ||
574 | ret = rename_completed_chunk(session, now); | |
575 | if (ret < 0) { | |
576 | ERR("Failed to rename completed rotation chunk"); | |
577 | goto end; | |
578 | } | |
579 | session->last_chunk_start_ts = session->current_chunk_start_ts; | |
580 | ||
581 | /* | |
582 | * Now we can clear the "ONGOING" state in the session. New | |
583 | * rotations can start now. | |
584 | */ | |
585 | session->rotation_state = LTTNG_ROTATION_STATE_COMPLETED; | |
586 | ||
587 | /* Ownership of location is transferred. */ | |
588 | location = session_get_trace_archive_location(session); | |
589 | ret = notification_thread_command_session_rotation_completed( | |
590 | notification_thread_handle, | |
591 | session->name, | |
592 | session->uid, | |
593 | session->gid, | |
594 | session->current_archive_id, | |
595 | location); | |
596 | if (ret != LTTNG_OK) { | |
597 | ERR("[rotation-thread] Failed to notify notification thread of completed rotation for session %s", | |
598 | session->name); | |
599 | } | |
600 | ||
9402c5b9 JG |
601 | if (!session->active) { |
602 | /* | |
603 | * A stop command was issued during the rotation, it is | |
604 | * up to the rotation completion check to perform the | |
605 | * renaming of the last chunk that was produced. | |
606 | */ | |
607 | ret = notification_thread_command_session_rotation_ongoing( | |
608 | notification_thread_handle, | |
609 | session->name, | |
610 | session->uid, | |
611 | session->gid, | |
612 | session->current_archive_id); | |
613 | if (ret != LTTNG_OK) { | |
614 | ERR("[rotation-thread] Failed to notify notification thread of completed rotation for session %s", | |
615 | session->name); | |
616 | } | |
617 | ||
618 | ret = rename_active_chunk(session); | |
619 | if (ret < 0) { | |
620 | ERR("[rotation-thread] Failed to rename active rotation chunk"); | |
621 | goto end; | |
622 | } | |
623 | ||
624 | /* Ownership of location is transferred. */ | |
625 | location = session_get_trace_archive_location(session); | |
626 | ret = notification_thread_command_session_rotation_completed( | |
627 | notification_thread_handle, | |
628 | session->name, | |
629 | session->uid, | |
630 | session->gid, | |
631 | session->current_archive_id, | |
632 | location); | |
633 | if (ret != LTTNG_OK) { | |
634 | ERR("[rotation-thread] Failed to notify notification thread of completed rotation for session %s", | |
635 | session->name); | |
636 | } | |
637 | } | |
638 | ||
92816cc3 | 639 | ret = 0; |
d88744a4 | 640 | end: |
92816cc3 JG |
641 | if (session->rotation_state == LTTNG_ROTATION_STATE_ONGOING) { |
642 | DBG("[rotation-thread] Rotation of trace archive %" PRIu64 " is still pending for session %s", | |
643 | session->current_archive_id - 1, session->name); | |
644 | ret = timer_session_rotation_pending_check_start(session, | |
645 | DEFAULT_ROTATE_PENDING_TIMER); | |
646 | if (ret) { | |
647 | ERR("Re-enabling rotate pending timer"); | |
648 | ret = -1; | |
649 | goto end; | |
650 | } | |
651 | } | |
652 | ||
d88744a4 JD |
653 | return ret; |
654 | } | |
655 | ||
92816cc3 | 656 | /* Call with the session lock held. */ |
259c2674 | 657 | static |
92816cc3 | 658 | int launch_session_rotation(struct ltt_session *session) |
259c2674 JD |
659 | { |
660 | int ret; | |
92816cc3 | 661 | struct lttng_rotate_session_return rotation_return; |
259c2674 | 662 | |
92816cc3 JG |
663 | DBG("[rotation-thread] Launching scheduled time-based rotation on session \"%s\"", |
664 | session->name); | |
259c2674 | 665 | |
92816cc3 JG |
666 | ret = cmd_rotate_session(session, &rotation_return); |
667 | if (ret == LTTNG_OK) { | |
668 | DBG("[rotation-thread] Scheduled time-based rotation successfully launched on session \"%s\"", | |
669 | session->name); | |
670 | } else { | |
671 | /* Don't consider errors as fatal. */ | |
672 | DBG("[rotation-thread] Scheduled time-based rotation aborted for session %s: %s", | |
673 | session->name, lttng_strerror(ret)); | |
259c2674 | 674 | } |
92816cc3 JG |
675 | return 0; |
676 | } | |
259c2674 | 677 | |
92816cc3 JG |
678 | static |
679 | int run_job(struct rotation_thread_job *job, struct ltt_session *session, | |
680 | struct notification_thread_handle *notification_thread_handle) | |
681 | { | |
682 | int ret; | |
259c2674 | 683 | |
92816cc3 JG |
684 | switch (job->type) { |
685 | case ROTATION_THREAD_JOB_TYPE_SCHEDULED_ROTATION: | |
686 | ret = launch_session_rotation(session); | |
687 | break; | |
688 | case ROTATION_THREAD_JOB_TYPE_CHECK_PENDING_ROTATION: | |
689 | ret = check_session_rotation_pending(session, | |
690 | notification_thread_handle); | |
691 | break; | |
692 | default: | |
693 | abort(); | |
259c2674 | 694 | } |
259c2674 JD |
695 | return ret; |
696 | } | |
697 | ||
d88744a4 | 698 | static |
92816cc3 JG |
699 | int handle_job_queue(struct rotation_thread_handle *handle, |
700 | struct rotation_thread *state, | |
d88744a4 JD |
701 | struct rotation_thread_timer_queue *queue) |
702 | { | |
703 | int ret = 0; | |
704 | int fd = lttng_pipe_get_readfd(queue->event_pipe); | |
705 | struct ltt_session *session; | |
92816cc3 | 706 | char buf; |
d88744a4 | 707 | |
92816cc3 | 708 | ret = lttng_read(fd, &buf, 1); |
d88744a4 JD |
709 | if (ret != 1) { |
710 | ERR("[rotation-thread] Failed to read from wakeup pipe (fd = %i)", fd); | |
711 | ret = -1; | |
712 | goto end; | |
713 | } | |
714 | ||
715 | for (;;) { | |
92816cc3 | 716 | struct rotation_thread_job *job; |
d88744a4 | 717 | |
92816cc3 | 718 | /* Take the queue lock only to pop an element from the list. */ |
d88744a4 JD |
719 | pthread_mutex_lock(&queue->lock); |
720 | if (cds_list_empty(&queue->list)) { | |
721 | pthread_mutex_unlock(&queue->lock); | |
722 | break; | |
723 | } | |
92816cc3 JG |
724 | job = cds_list_first_entry(&queue->list, |
725 | typeof(*job), head); | |
726 | cds_list_del(&job->head); | |
d88744a4 JD |
727 | pthread_mutex_unlock(&queue->lock); |
728 | ||
d88744a4 | 729 | session_lock_list(); |
92816cc3 | 730 | session = session_find_by_id(job->session_id); |
d88744a4 JD |
731 | if (!session) { |
732 | DBG("[rotation-thread] Session %" PRIu64 " not found", | |
92816cc3 | 733 | job->session_id); |
d88744a4 | 734 | /* |
92816cc3 JG |
735 | * This is a non-fatal error, and we cannot report it to |
736 | * the user (timer), so just print the error and | |
737 | * continue the processing. | |
738 | * | |
739 | * While the timer thread will purge pending signals for | |
740 | * a session on the session's destruction, it is | |
741 | * possible for a job targeting that session to have | |
742 | * already been queued before it was destroyed. | |
d88744a4 JD |
743 | */ |
744 | session_unlock_list(); | |
92816cc3 | 745 | free(job); |
d88744a4 JD |
746 | continue; |
747 | } | |
748 | ||
d88744a4 JD |
749 | session_lock(session); |
750 | session_unlock_list(); | |
751 | ||
92816cc3 | 752 | ret = run_job(job, session, handle->notification_thread_handle); |
d88744a4 | 753 | session_unlock(session); |
92816cc3 | 754 | free(job); |
d88744a4 | 755 | if (ret) { |
d88744a4 JD |
756 | goto end; |
757 | } | |
758 | } | |
759 | ||
760 | ret = 0; | |
761 | ||
762 | end: | |
763 | return ret; | |
764 | } | |
765 | ||
92816cc3 JG |
766 | static |
767 | int handle_condition(const struct lttng_condition *condition, | |
90936dcf JD |
768 | const struct lttng_evaluation *evaluation, |
769 | struct notification_thread_handle *notification_thread_handle) | |
770 | { | |
771 | int ret = 0; | |
772 | const char *condition_session_name = NULL; | |
773 | enum lttng_condition_type condition_type; | |
774 | enum lttng_condition_status condition_status; | |
775 | enum lttng_evaluation_status evaluation_status; | |
776 | uint64_t consumed; | |
777 | struct ltt_session *session; | |
778 | ||
779 | condition_type = lttng_condition_get_type(condition); | |
780 | ||
781 | if (condition_type != LTTNG_CONDITION_TYPE_SESSION_CONSUMED_SIZE) { | |
782 | ret = -1; | |
783 | ERR("[rotation-thread] Condition type and session usage type are not the same"); | |
784 | goto end; | |
785 | } | |
786 | ||
787 | /* Fetch info to test */ | |
788 | condition_status = lttng_condition_session_consumed_size_get_session_name( | |
789 | condition, &condition_session_name); | |
790 | if (condition_status != LTTNG_CONDITION_STATUS_OK) { | |
791 | ERR("[rotation-thread] Session name could not be fetched"); | |
792 | ret = -1; | |
793 | goto end; | |
794 | } | |
795 | evaluation_status = lttng_evaluation_session_consumed_size_get_consumed_size(evaluation, | |
796 | &consumed); | |
797 | if (evaluation_status != LTTNG_EVALUATION_STATUS_OK) { | |
798 | ERR("[rotation-thread] Failed to get evaluation"); | |
799 | ret = -1; | |
800 | goto end; | |
801 | } | |
802 | ||
803 | session_lock_list(); | |
804 | session = session_find_by_name(condition_session_name); | |
805 | if (!session) { | |
806 | ret = -1; | |
807 | session_unlock_list(); | |
808 | ERR("[rotation-thread] Session \"%s\" not found", | |
809 | condition_session_name); | |
810 | goto end; | |
811 | } | |
812 | session_lock(session); | |
813 | session_unlock_list(); | |
814 | ||
815 | ret = unsubscribe_session_consumed_size_rotation(session, | |
816 | notification_thread_handle); | |
817 | if (ret) { | |
818 | goto end; | |
819 | } | |
820 | ||
821 | ret = cmd_rotate_session(session, NULL); | |
822 | if (ret == -LTTNG_ERR_ROTATION_PENDING) { | |
823 | DBG("Rotate already pending, subscribe to the next threshold value"); | |
90936dcf JD |
824 | } else if (ret != LTTNG_OK) { |
825 | ERR("[rotation-thread] Failed to rotate on size notification with error: %s", | |
826 | lttng_strerror(ret)); | |
827 | ret = -1; | |
828 | goto end_unlock; | |
829 | } | |
830 | ret = subscribe_session_consumed_size_rotation(session, | |
831 | consumed + session->rotate_size, | |
832 | notification_thread_handle); | |
833 | if (ret) { | |
834 | ERR("[rotation-thread] Failed to subscribe to session consumed size condition"); | |
835 | goto end_unlock; | |
836 | } | |
837 | ret = 0; | |
838 | ||
839 | end_unlock: | |
840 | session_unlock(session); | |
841 | end: | |
842 | return ret; | |
843 | } | |
844 | ||
845 | static | |
92816cc3 | 846 | int handle_notification_channel(int fd, |
90936dcf | 847 | struct rotation_thread_handle *handle, |
92816cc3 | 848 | struct rotation_thread *state) |
90936dcf JD |
849 | { |
850 | int ret; | |
d73ee93f JG |
851 | bool notification_pending; |
852 | struct lttng_notification *notification = NULL; | |
90936dcf JD |
853 | enum lttng_notification_channel_status status; |
854 | const struct lttng_evaluation *notification_evaluation; | |
855 | const struct lttng_condition *notification_condition; | |
856 | ||
d73ee93f JG |
857 | status = lttng_notification_channel_has_pending_notification( |
858 | rotate_notification_channel, ¬ification_pending); | |
859 | if (status != LTTNG_NOTIFICATION_CHANNEL_STATUS_OK) { | |
860 | ERR("[rotation-thread ]Error occured while checking for pending notification"); | |
861 | ret = -1; | |
862 | goto end; | |
863 | } | |
864 | ||
865 | if (!notification_pending) { | |
866 | ret = 0; | |
867 | goto end; | |
868 | } | |
869 | ||
90936dcf JD |
870 | /* Receive the next notification. */ |
871 | status = lttng_notification_channel_get_next_notification( | |
872 | rotate_notification_channel, | |
873 | ¬ification); | |
874 | ||
875 | switch (status) { | |
876 | case LTTNG_NOTIFICATION_CHANNEL_STATUS_OK: | |
877 | break; | |
878 | case LTTNG_NOTIFICATION_CHANNEL_STATUS_NOTIFICATIONS_DROPPED: | |
879 | /* Not an error, we will wait for the next one */ | |
880 | ret = 0; | |
881 | goto end;; | |
882 | case LTTNG_NOTIFICATION_CHANNEL_STATUS_CLOSED: | |
883 | ERR("Notification channel was closed"); | |
884 | ret = -1; | |
885 | goto end; | |
886 | default: | |
887 | /* Unhandled conditions / errors. */ | |
888 | ERR("Unknown notification channel status"); | |
889 | ret = -1; | |
890 | goto end; | |
891 | } | |
892 | ||
893 | notification_condition = lttng_notification_get_condition(notification); | |
894 | notification_evaluation = lttng_notification_get_evaluation(notification); | |
895 | ||
896 | ret = handle_condition(notification_condition, notification_evaluation, | |
897 | handle->notification_thread_handle); | |
898 | ||
899 | end: | |
900 | lttng_notification_destroy(notification); | |
90936dcf JD |
901 | return ret; |
902 | } | |
903 | ||
db66e574 JD |
904 | void *thread_rotation(void *data) |
905 | { | |
906 | int ret; | |
907 | struct rotation_thread_handle *handle = data; | |
92816cc3 | 908 | struct rotation_thread thread; |
db66e574 JD |
909 | |
910 | DBG("[rotation-thread] Started rotation thread"); | |
911 | ||
912 | if (!handle) { | |
913 | ERR("[rotation-thread] Invalid thread context provided"); | |
914 | goto end; | |
915 | } | |
916 | ||
03732c32 JG |
917 | rcu_register_thread(); |
918 | rcu_thread_online(); | |
919 | ||
db66e574 JD |
920 | health_register(health_sessiond, HEALTH_SESSIOND_TYPE_ROTATION); |
921 | health_code_update(); | |
922 | ||
92816cc3 | 923 | ret = init_thread_state(handle, &thread); |
db66e574 | 924 | if (ret) { |
f5f8e5cd | 925 | goto error; |
db66e574 JD |
926 | } |
927 | ||
928 | /* Ready to handle client connections. */ | |
929 | sessiond_notify_ready(); | |
930 | ||
931 | while (true) { | |
932 | int fd_count, i; | |
933 | ||
934 | health_poll_entry(); | |
935 | DBG("[rotation-thread] Entering poll wait"); | |
92816cc3 | 936 | ret = lttng_poll_wait(&thread.events, -1); |
db66e574 JD |
937 | DBG("[rotation-thread] Poll wait returned (%i)", ret); |
938 | health_poll_exit(); | |
939 | if (ret < 0) { | |
940 | /* | |
941 | * Restart interrupted system call. | |
942 | */ | |
943 | if (errno == EINTR) { | |
944 | continue; | |
945 | } | |
946 | ERR("[rotation-thread] Error encountered during lttng_poll_wait (%i)", ret); | |
947 | goto error; | |
948 | } | |
949 | ||
950 | fd_count = ret; | |
951 | for (i = 0; i < fd_count; i++) { | |
92816cc3 JG |
952 | int fd = LTTNG_POLL_GETFD(&thread.events, i); |
953 | uint32_t revents = LTTNG_POLL_GETEV(&thread.events, i); | |
db66e574 JD |
954 | |
955 | DBG("[rotation-thread] Handling fd (%i) activity (%u)", | |
956 | fd, revents); | |
957 | ||
92816cc3 JG |
958 | if (revents & LPOLLERR) { |
959 | ERR("[rotation-thread] Polling returned an error on fd %i", fd); | |
960 | goto error; | |
961 | } | |
962 | ||
963 | if (fd == handle->quit_pipe) { | |
db66e574 | 964 | DBG("[rotation-thread] Quit pipe activity"); |
92816cc3 | 965 | /* TODO flush the queue. */ |
db66e574 | 966 | goto exit; |
d88744a4 | 967 | } else if (fd == lttng_pipe_get_readfd(handle->rotation_timer_queue->event_pipe)) { |
92816cc3 JG |
968 | ret = handle_job_queue(handle, &thread, |
969 | handle->rotation_timer_queue); | |
d88744a4 JD |
970 | if (ret) { |
971 | ERR("[rotation-thread] Failed to handle rotation timer pipe event"); | |
972 | goto error; | |
973 | } | |
90936dcf | 974 | } else if (fd == rotate_notification_channel->socket) { |
92816cc3 JG |
975 | ret = handle_notification_channel(fd, handle, |
976 | &thread); | |
90936dcf JD |
977 | if (ret) { |
978 | ERR("[rotation-thread] Error occured while handling activity on notification channel socket"); | |
979 | goto error; | |
980 | } | |
db66e574 JD |
981 | } |
982 | } | |
983 | } | |
984 | exit: | |
985 | error: | |
986 | DBG("[rotation-thread] Exit"); | |
92816cc3 | 987 | fini_thread_state(&thread); |
db66e574 | 988 | health_unregister(health_sessiond); |
03732c32 JG |
989 | rcu_thread_offline(); |
990 | rcu_unregister_thread(); | |
db66e574 JD |
991 | end: |
992 | return NULL; | |
993 | } |