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 { | |
92816cc3 JG |
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 | ||
db66e574 | 81 | static |
92816cc3 | 82 | const char *get_job_type_str(enum rotation_thread_job_type job_type) |
db66e574 | 83 | { |
92816cc3 JG |
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 | } | |
db66e574 JD |
92 | } |
93 | ||
92816cc3 | 94 | struct rotation_thread_timer_queue *rotation_thread_timer_queue_create(void) |
db66e574 | 95 | { |
92816cc3 | 96 | struct rotation_thread_timer_queue *queue = NULL; |
db66e574 | 97 | |
92816cc3 JG |
98 | queue = zmalloc(sizeof(*queue)); |
99 | if (!queue) { | |
100 | PERROR("Failed to allocate timer rotate queue"); | |
101 | goto end; | |
102 | } | |
db66e574 | 103 | |
92816cc3 JG |
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; | |
db66e574 JD |
109 | } |
110 | ||
92816cc3 | 111 | void log_job_destruction(const struct rotation_thread_job *job) |
db66e574 | 112 | { |
92816cc3 JG |
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(); | |
db66e574 JD |
132 | } |
133 | ||
92816cc3 JG |
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); | |
db66e574 JD |
136 | } |
137 | ||
92816cc3 JG |
138 | void rotation_thread_timer_queue_destroy( |
139 | struct rotation_thread_timer_queue *queue) | |
db66e574 | 140 | { |
92816cc3 | 141 | struct rotation_thread_job *job, *tmp_job; |
db66e574 | 142 | |
92816cc3 JG |
143 | if (!queue) { |
144 | return; | |
db66e574 JD |
145 | } |
146 | ||
92816cc3 JG |
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); | |
db66e574 | 155 | } |
92816cc3 JG |
156 | pthread_mutex_unlock(&queue->lock); |
157 | pthread_mutex_destroy(&queue->lock); | |
158 | free(queue); | |
159 | } | |
db66e574 | 160 | |
92816cc3 JG |
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 | { | |
db66e574 JD |
167 | free(handle); |
168 | } | |
169 | ||
170 | struct rotation_thread_handle *rotation_thread_handle_create( | |
90936dcf JD |
171 | struct rotation_thread_timer_queue *rotation_timer_queue, |
172 | struct notification_thread_handle *notification_thread_handle, | |
173 | sem_t *notification_thread_ready) | |
db66e574 JD |
174 | { |
175 | struct rotation_thread_handle *handle; | |
176 | ||
177 | handle = zmalloc(sizeof(*handle)); | |
178 | if (!handle) { | |
179 | goto end; | |
180 | } | |
181 | ||
92816cc3 JG |
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; | |
db66e574 | 205 | } |
db66e574 | 206 | } |
92816cc3 JG |
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; | |
db66e574 | 226 | } |
92816cc3 JG |
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; | |
db66e574 | 255 | } |
92816cc3 JG |
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; | |
db66e574 | 259 | } |
db66e574 JD |
260 | |
261 | end: | |
92816cc3 | 262 | pthread_mutex_unlock(&queue->lock); |
db66e574 JD |
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 | /* | |
92816cc3 JG |
272 | * Create pollset with size 2: |
273 | * - quit pipe, | |
274 | * - rotation thread timer queue pipe, | |
db66e574 | 275 | */ |
a7333da7 JG |
276 | ret = sessiond_set_thread_pollset(poll_set, 2); |
277 | if (ret) { | |
db66e574 JD |
278 | goto error; |
279 | } | |
d086f507 JD |
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 | } | |
db66e574 | 287 | |
db66e574 JD |
288 | return ret; |
289 | error: | |
290 | lttng_poll_clean(poll_set); | |
291 | return ret; | |
292 | } | |
293 | ||
294 | static | |
92816cc3 | 295 | void fini_thread_state(struct rotation_thread *state) |
db66e574 JD |
296 | { |
297 | lttng_poll_clean(&state->events); | |
90936dcf JD |
298 | if (rotate_notification_channel) { |
299 | lttng_notification_channel_destroy(rotate_notification_channel); | |
300 | } | |
db66e574 JD |
301 | } |
302 | ||
303 | static | |
304 | int init_thread_state(struct rotation_thread_handle *handle, | |
92816cc3 | 305 | struct rotation_thread *state) |
db66e574 JD |
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 | ||
90936dcf JD |
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 | ||
db66e574 JD |
337 | end: |
338 | return ret; | |
339 | } | |
340 | ||
341 | static | |
92816cc3 JG |
342 | int check_session_rotation_pending_local_on_consumer( |
343 | const struct ltt_session *session, | |
344 | struct consumer_socket *socket, bool *rotation_completed) | |
db66e574 | 345 | { |
92816cc3 JG |
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; | |
db66e574 | 372 | } else { |
92816cc3 JG |
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; | |
db66e574 | 379 | } |
92816cc3 JG |
380 | return ret; |
381 | } | |
db66e574 | 382 | |
92816cc3 JG |
383 | static |
384 | int check_session_rotation_pending_local(struct ltt_session *session) | |
385 | { | |
db582e11 | 386 | int ret = 0; |
92816cc3 JG |
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; | |
db66e574 | 408 | } |
db66e574 JD |
409 | } |
410 | ||
92816cc3 JG |
411 | skip_ust: |
412 | if (!session->kernel_session) { | |
413 | goto skip_kernel; | |
db66e574 | 414 | } |
92816cc3 JG |
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(); | |
db66e574 | 426 | |
92816cc3 JG |
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; | |
db66e574 | 432 | } |
92816cc3 | 433 | if (ret) { |
2961f09e JG |
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 | } | |
db66e574 | 440 | } |
92816cc3 JG |
441 | return 0; |
442 | } | |
db66e574 | 443 | |
92816cc3 JG |
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; | |
db66e574 | 452 | |
92816cc3 JG |
453 | /* |
454 | * Check for a pending rotation on any consumer as we only use | |
455 | * it as a "tunnel" to the relayd. | |
456 | */ | |
17dd1232 | 457 | |
92816cc3 JG |
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; | |
db66e574 | 467 | } |
92816cc3 | 468 | assert(cds_lfht_iter_get_node(&iter)); |
db66e574 | 469 | |
92816cc3 JG |
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); | |
2961f09e JG |
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 | } | |
92816cc3 JG |
505 | rotation_completed = false; |
506 | } | |
db66e574 | 507 | |
db66e574 | 508 | rcu_read_unlock(); |
92816cc3 JG |
509 | |
510 | if (rotation_completed) { | |
cb70f0d5 | 511 | DBG("[rotation-thread] Rotation of trace archive %" PRIu64 " of session \"%s\" is complete on the relay", |
92816cc3 JG |
512 | session->current_archive_id - 1, |
513 | session->name); | |
514 | session->rotation_pending_relay = false; | |
515 | } | |
516 | return 0; | |
db66e574 JD |
517 | } |
518 | ||
d88744a4 | 519 | /* |
92816cc3 | 520 | * Check if the last rotation was completed, called with session lock held. |
d88744a4 JD |
521 | */ |
522 | static | |
92816cc3 JG |
523 | int check_session_rotation_pending(struct ltt_session *session, |
524 | struct notification_thread_handle *notification_thread_handle) | |
d88744a4 JD |
525 | { |
526 | int ret; | |
92816cc3 JG |
527 | struct lttng_trace_archive_location *location; |
528 | time_t now; | |
d88744a4 | 529 | |
92816cc3 JG |
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 | } | |
17dd1232 | 539 | |
d88744a4 | 540 | /* |
92816cc3 JG |
541 | * No need to check for a pending rotation on the relay |
542 | * since the rotation is not even completed locally yet. | |
d88744a4 | 543 | */ |
92816cc3 JG |
544 | if (session->rotation_pending_local) { |
545 | goto end; | |
17dd1232 | 546 | } |
92816cc3 JG |
547 | } |
548 | ||
549 | if (session->rotation_pending_relay) { | |
550 | /* Updates session->rotation_pending_relay as needed. */ | |
551 | ret = check_session_rotation_pending_relay(session); | |
d88744a4 | 552 | if (ret) { |
92816cc3 JG |
553 | goto end; |
554 | } | |
555 | ||
556 | if (session->rotation_pending_relay) { | |
d88744a4 JD |
557 | goto end; |
558 | } | |
559 | } | |
560 | ||
92816cc3 JG |
561 | DBG("[rotation-thread] Rotation of trace archive %" PRIu64 " completed for " |
562 | "session %s", session->current_archive_id - 1, | |
563 | session->name); | |
d88744a4 | 564 | |
92816cc3 JG |
565 | /* Rename the completed trace archive's location. */ |
566 | now = time(NULL); | |
567 | if (now == (time_t) -1) { | |
2961f09e JG |
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 | } | |
92816cc3 JG |
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 | ||
9402c5b9 JG |
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 | ||
92816cc3 | 643 | ret = 0; |
d88744a4 | 644 | end: |
92816cc3 JG |
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 | ||
d88744a4 JD |
657 | return ret; |
658 | } | |
659 | ||
ed1e52a3 | 660 | /* Call with the session and session_list locks held. */ |
259c2674 | 661 | static |
92816cc3 | 662 | int launch_session_rotation(struct ltt_session *session) |
259c2674 JD |
663 | { |
664 | int ret; | |
92816cc3 | 665 | struct lttng_rotate_session_return rotation_return; |
259c2674 | 666 | |
92816cc3 JG |
667 | DBG("[rotation-thread] Launching scheduled time-based rotation on session \"%s\"", |
668 | session->name); | |
259c2674 | 669 | |
92816cc3 JG |
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)); | |
259c2674 | 678 | } |
92816cc3 JG |
679 | return 0; |
680 | } | |
259c2674 | 681 | |
92816cc3 JG |
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; | |
259c2674 | 687 | |
92816cc3 JG |
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(); | |
259c2674 | 698 | } |
259c2674 JD |
699 | return ret; |
700 | } | |
701 | ||
d88744a4 | 702 | static |
92816cc3 JG |
703 | int handle_job_queue(struct rotation_thread_handle *handle, |
704 | struct rotation_thread *state, | |
d88744a4 JD |
705 | struct rotation_thread_timer_queue *queue) |
706 | { | |
707 | int ret = 0; | |
708 | int fd = lttng_pipe_get_readfd(queue->event_pipe); | |
92816cc3 | 709 | char buf; |
d88744a4 | 710 | |
92816cc3 | 711 | ret = lttng_read(fd, &buf, 1); |
d88744a4 JD |
712 | if (ret != 1) { |
713 | ERR("[rotation-thread] Failed to read from wakeup pipe (fd = %i)", fd); | |
714 | ret = -1; | |
715 | goto end; | |
716 | } | |
717 | ||
718 | for (;;) { | |
e32d7f27 | 719 | struct ltt_session *session; |
92816cc3 | 720 | struct rotation_thread_job *job; |
d88744a4 | 721 | |
92816cc3 | 722 | /* Take the queue lock only to pop an element from the list. */ |
d88744a4 JD |
723 | pthread_mutex_lock(&queue->lock); |
724 | if (cds_list_empty(&queue->list)) { | |
725 | pthread_mutex_unlock(&queue->lock); | |
726 | break; | |
727 | } | |
92816cc3 JG |
728 | job = cds_list_first_entry(&queue->list, |
729 | typeof(*job), head); | |
730 | cds_list_del(&job->head); | |
d88744a4 JD |
731 | pthread_mutex_unlock(&queue->lock); |
732 | ||
d88744a4 | 733 | session_lock_list(); |
92816cc3 | 734 | session = session_find_by_id(job->session_id); |
d88744a4 JD |
735 | if (!session) { |
736 | DBG("[rotation-thread] Session %" PRIu64 " not found", | |
92816cc3 | 737 | job->session_id); |
d88744a4 | 738 | /* |
92816cc3 JG |
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. | |
d88744a4 JD |
747 | */ |
748 | session_unlock_list(); | |
92816cc3 | 749 | free(job); |
e32d7f27 | 750 | session_put(session); |
d88744a4 JD |
751 | continue; |
752 | } | |
753 | ||
d88744a4 | 754 | session_lock(session); |
92816cc3 | 755 | ret = run_job(job, session, handle->notification_thread_handle); |
d88744a4 | 756 | session_unlock(session); |
e32d7f27 | 757 | session_put(session); |
ed1e52a3 | 758 | session_unlock_list(); |
92816cc3 | 759 | free(job); |
d88744a4 | 760 | if (ret) { |
d88744a4 JD |
761 | goto end; |
762 | } | |
763 | } | |
764 | ||
765 | ret = 0; | |
766 | ||
767 | end: | |
768 | return ret; | |
769 | } | |
770 | ||
92816cc3 JG |
771 | static |
772 | int handle_condition(const struct lttng_condition *condition, | |
90936dcf JD |
773 | const struct lttng_evaluation *evaluation, |
774 | struct notification_thread_handle *notification_thread_handle) | |
775 | { | |
776 | int ret = 0; | |
777 | const char *condition_session_name = NULL; | |
778 | enum lttng_condition_type condition_type; | |
779 | enum lttng_condition_status condition_status; | |
780 | enum lttng_evaluation_status evaluation_status; | |
781 | uint64_t consumed; | |
782 | struct ltt_session *session; | |
783 | ||
784 | condition_type = lttng_condition_get_type(condition); | |
785 | ||
786 | if (condition_type != LTTNG_CONDITION_TYPE_SESSION_CONSUMED_SIZE) { | |
787 | ret = -1; | |
788 | ERR("[rotation-thread] Condition type and session usage type are not the same"); | |
789 | goto end; | |
790 | } | |
791 | ||
792 | /* Fetch info to test */ | |
793 | condition_status = lttng_condition_session_consumed_size_get_session_name( | |
794 | condition, &condition_session_name); | |
795 | if (condition_status != LTTNG_CONDITION_STATUS_OK) { | |
796 | ERR("[rotation-thread] Session name could not be fetched"); | |
797 | ret = -1; | |
798 | goto end; | |
799 | } | |
800 | evaluation_status = lttng_evaluation_session_consumed_size_get_consumed_size(evaluation, | |
801 | &consumed); | |
802 | if (evaluation_status != LTTNG_EVALUATION_STATUS_OK) { | |
803 | ERR("[rotation-thread] Failed to get evaluation"); | |
804 | ret = -1; | |
805 | goto end; | |
806 | } | |
807 | ||
808 | session_lock_list(); | |
809 | session = session_find_by_name(condition_session_name); | |
810 | if (!session) { | |
811 | ret = -1; | |
812 | session_unlock_list(); | |
813 | ERR("[rotation-thread] Session \"%s\" not found", | |
814 | condition_session_name); | |
815 | goto end; | |
816 | } | |
817 | session_lock(session); | |
818 | session_unlock_list(); | |
819 | ||
820 | ret = unsubscribe_session_consumed_size_rotation(session, | |
821 | notification_thread_handle); | |
822 | if (ret) { | |
490b3229 | 823 | goto end_unlock; |
90936dcf JD |
824 | } |
825 | ||
826 | ret = cmd_rotate_session(session, NULL); | |
827 | if (ret == -LTTNG_ERR_ROTATION_PENDING) { | |
828 | DBG("Rotate already pending, subscribe to the next threshold value"); | |
90936dcf JD |
829 | } else if (ret != LTTNG_OK) { |
830 | ERR("[rotation-thread] Failed to rotate on size notification with error: %s", | |
831 | lttng_strerror(ret)); | |
832 | ret = -1; | |
833 | goto end_unlock; | |
834 | } | |
835 | ret = subscribe_session_consumed_size_rotation(session, | |
836 | consumed + session->rotate_size, | |
837 | notification_thread_handle); | |
838 | if (ret) { | |
839 | ERR("[rotation-thread] Failed to subscribe to session consumed size condition"); | |
840 | goto end_unlock; | |
841 | } | |
842 | ret = 0; | |
843 | ||
844 | end_unlock: | |
845 | session_unlock(session); | |
e32d7f27 | 846 | session_put(session); |
90936dcf JD |
847 | end: |
848 | return ret; | |
849 | } | |
850 | ||
851 | static | |
92816cc3 | 852 | int handle_notification_channel(int fd, |
90936dcf | 853 | struct rotation_thread_handle *handle, |
92816cc3 | 854 | struct rotation_thread *state) |
90936dcf JD |
855 | { |
856 | int ret; | |
d73ee93f JG |
857 | bool notification_pending; |
858 | struct lttng_notification *notification = NULL; | |
90936dcf JD |
859 | enum lttng_notification_channel_status status; |
860 | const struct lttng_evaluation *notification_evaluation; | |
861 | const struct lttng_condition *notification_condition; | |
862 | ||
d73ee93f JG |
863 | status = lttng_notification_channel_has_pending_notification( |
864 | rotate_notification_channel, ¬ification_pending); | |
865 | if (status != LTTNG_NOTIFICATION_CHANNEL_STATUS_OK) { | |
a9577b76 | 866 | ERR("[rotation-thread ]Error occurred while checking for pending notification"); |
d73ee93f JG |
867 | ret = -1; |
868 | goto end; | |
869 | } | |
870 | ||
871 | if (!notification_pending) { | |
872 | ret = 0; | |
873 | goto end; | |
874 | } | |
875 | ||
90936dcf JD |
876 | /* Receive the next notification. */ |
877 | status = lttng_notification_channel_get_next_notification( | |
878 | rotate_notification_channel, | |
879 | ¬ification); | |
880 | ||
881 | switch (status) { | |
882 | case LTTNG_NOTIFICATION_CHANNEL_STATUS_OK: | |
883 | break; | |
884 | case LTTNG_NOTIFICATION_CHANNEL_STATUS_NOTIFICATIONS_DROPPED: | |
885 | /* Not an error, we will wait for the next one */ | |
886 | ret = 0; | |
887 | goto end;; | |
888 | case LTTNG_NOTIFICATION_CHANNEL_STATUS_CLOSED: | |
889 | ERR("Notification channel was closed"); | |
890 | ret = -1; | |
891 | goto end; | |
892 | default: | |
893 | /* Unhandled conditions / errors. */ | |
894 | ERR("Unknown notification channel status"); | |
895 | ret = -1; | |
896 | goto end; | |
897 | } | |
898 | ||
899 | notification_condition = lttng_notification_get_condition(notification); | |
900 | notification_evaluation = lttng_notification_get_evaluation(notification); | |
901 | ||
902 | ret = handle_condition(notification_condition, notification_evaluation, | |
903 | handle->notification_thread_handle); | |
904 | ||
905 | end: | |
906 | lttng_notification_destroy(notification); | |
90936dcf JD |
907 | return ret; |
908 | } | |
909 | ||
db66e574 JD |
910 | void *thread_rotation(void *data) |
911 | { | |
912 | int ret; | |
913 | struct rotation_thread_handle *handle = data; | |
92816cc3 | 914 | struct rotation_thread thread; |
db66e574 JD |
915 | |
916 | DBG("[rotation-thread] Started rotation thread"); | |
917 | ||
918 | if (!handle) { | |
919 | ERR("[rotation-thread] Invalid thread context provided"); | |
920 | goto end; | |
921 | } | |
922 | ||
03732c32 JG |
923 | rcu_register_thread(); |
924 | rcu_thread_online(); | |
925 | ||
db66e574 JD |
926 | health_register(health_sessiond, HEALTH_SESSIOND_TYPE_ROTATION); |
927 | health_code_update(); | |
928 | ||
92816cc3 | 929 | ret = init_thread_state(handle, &thread); |
db66e574 | 930 | if (ret) { |
f5f8e5cd | 931 | goto error; |
db66e574 JD |
932 | } |
933 | ||
934 | /* Ready to handle client connections. */ | |
935 | sessiond_notify_ready(); | |
936 | ||
937 | while (true) { | |
938 | int fd_count, i; | |
939 | ||
940 | health_poll_entry(); | |
941 | DBG("[rotation-thread] Entering poll wait"); | |
92816cc3 | 942 | ret = lttng_poll_wait(&thread.events, -1); |
db66e574 JD |
943 | DBG("[rotation-thread] Poll wait returned (%i)", ret); |
944 | health_poll_exit(); | |
945 | if (ret < 0) { | |
946 | /* | |
947 | * Restart interrupted system call. | |
948 | */ | |
949 | if (errno == EINTR) { | |
950 | continue; | |
951 | } | |
952 | ERR("[rotation-thread] Error encountered during lttng_poll_wait (%i)", ret); | |
953 | goto error; | |
954 | } | |
955 | ||
956 | fd_count = ret; | |
957 | for (i = 0; i < fd_count; i++) { | |
92816cc3 JG |
958 | int fd = LTTNG_POLL_GETFD(&thread.events, i); |
959 | uint32_t revents = LTTNG_POLL_GETEV(&thread.events, i); | |
db66e574 JD |
960 | |
961 | DBG("[rotation-thread] Handling fd (%i) activity (%u)", | |
962 | fd, revents); | |
963 | ||
92816cc3 JG |
964 | if (revents & LPOLLERR) { |
965 | ERR("[rotation-thread] Polling returned an error on fd %i", fd); | |
966 | goto error; | |
967 | } | |
968 | ||
a7333da7 | 969 | if (sessiond_check_thread_quit_pipe(fd, revents)) { |
db66e574 | 970 | DBG("[rotation-thread] Quit pipe activity"); |
92816cc3 | 971 | /* TODO flush the queue. */ |
db66e574 | 972 | goto exit; |
d88744a4 | 973 | } else if (fd == lttng_pipe_get_readfd(handle->rotation_timer_queue->event_pipe)) { |
92816cc3 JG |
974 | ret = handle_job_queue(handle, &thread, |
975 | handle->rotation_timer_queue); | |
d88744a4 JD |
976 | if (ret) { |
977 | ERR("[rotation-thread] Failed to handle rotation timer pipe event"); | |
978 | goto error; | |
979 | } | |
90936dcf | 980 | } else if (fd == rotate_notification_channel->socket) { |
92816cc3 JG |
981 | ret = handle_notification_channel(fd, handle, |
982 | &thread); | |
90936dcf | 983 | if (ret) { |
a9577b76 | 984 | ERR("[rotation-thread] Error occurred while handling activity on notification channel socket"); |
90936dcf JD |
985 | goto error; |
986 | } | |
db66e574 JD |
987 | } |
988 | } | |
989 | } | |
990 | exit: | |
991 | error: | |
992 | DBG("[rotation-thread] Exit"); | |
92816cc3 | 993 | fini_thread_state(&thread); |
db66e574 | 994 | health_unregister(health_sessiond); |
03732c32 JG |
995 | rcu_thread_offline(); |
996 | rcu_unregister_thread(); | |
db66e574 JD |
997 | end: |
998 | return NULL; | |
999 | } |