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" |
64d9b072 JG |
47 | #include "utils.h" |
48 | #include "thread.h" | |
db66e574 JD |
49 | |
50 | #include <urcu.h> | |
51 | #include <urcu/list.h> | |
db66e574 | 52 | |
90936dcf JD |
53 | struct lttng_notification_channel *rotate_notification_channel = NULL; |
54 | ||
92816cc3 | 55 | struct rotation_thread { |
db66e574 JD |
56 | struct lttng_poll_event events; |
57 | }; | |
58 | ||
92816cc3 JG |
59 | struct rotation_thread_job { |
60 | enum rotation_thread_job_type type; | |
c7031a2c | 61 | struct ltt_session *session; |
92816cc3 JG |
62 | /* List member in struct rotation_thread_timer_queue. */ |
63 | struct cds_list_head head; | |
64 | }; | |
65 | ||
66 | /* | |
67 | * The timer thread enqueues jobs and wakes up the rotation thread. | |
68 | * When the rotation thread wakes up, it empties the queue. | |
69 | */ | |
70 | struct rotation_thread_timer_queue { | |
71 | struct lttng_pipe *event_pipe; | |
72 | struct cds_list_head list; | |
73 | pthread_mutex_t lock; | |
74 | }; | |
75 | ||
76 | struct rotation_thread_handle { | |
92816cc3 JG |
77 | struct rotation_thread_timer_queue *rotation_timer_queue; |
78 | /* Access to the notification thread cmd_queue */ | |
79 | struct notification_thread_handle *notification_thread_handle; | |
64d9b072 JG |
80 | /* Thread-specific quit pipe. */ |
81 | struct lttng_pipe *quit_pipe; | |
92816cc3 JG |
82 | }; |
83 | ||
db66e574 | 84 | static |
92816cc3 | 85 | const char *get_job_type_str(enum rotation_thread_job_type job_type) |
db66e574 | 86 | { |
92816cc3 JG |
87 | switch (job_type) { |
88 | case ROTATION_THREAD_JOB_TYPE_CHECK_PENDING_ROTATION: | |
89 | return "CHECK_PENDING_ROTATION"; | |
90 | case ROTATION_THREAD_JOB_TYPE_SCHEDULED_ROTATION: | |
91 | return "SCHEDULED_ROTATION"; | |
92 | default: | |
93 | abort(); | |
94 | } | |
db66e574 JD |
95 | } |
96 | ||
92816cc3 | 97 | struct rotation_thread_timer_queue *rotation_thread_timer_queue_create(void) |
db66e574 | 98 | { |
92816cc3 | 99 | struct rotation_thread_timer_queue *queue = NULL; |
db66e574 | 100 | |
92816cc3 JG |
101 | queue = zmalloc(sizeof(*queue)); |
102 | if (!queue) { | |
103 | PERROR("Failed to allocate timer rotate queue"); | |
104 | goto end; | |
105 | } | |
db66e574 | 106 | |
92816cc3 JG |
107 | queue->event_pipe = lttng_pipe_open(FD_CLOEXEC | O_NONBLOCK); |
108 | CDS_INIT_LIST_HEAD(&queue->list); | |
109 | pthread_mutex_init(&queue->lock, NULL); | |
110 | end: | |
111 | return queue; | |
db66e574 JD |
112 | } |
113 | ||
92816cc3 | 114 | void log_job_destruction(const struct rotation_thread_job *job) |
db66e574 | 115 | { |
92816cc3 JG |
116 | enum lttng_error_level log_level; |
117 | const char *job_type_str = get_job_type_str(job->type); | |
118 | ||
119 | switch (job->type) { | |
120 | case ROTATION_THREAD_JOB_TYPE_SCHEDULED_ROTATION: | |
121 | /* | |
122 | * Not a problem, the scheduled rotation is racing with the teardown | |
123 | * of the daemon. In this case, the rotation will not happen, which | |
124 | * is not a problem (or at least, not important enough to delay | |
125 | * the shutdown of the session daemon). | |
126 | */ | |
127 | log_level = PRINT_DBG; | |
128 | break; | |
129 | case ROTATION_THREAD_JOB_TYPE_CHECK_PENDING_ROTATION: | |
130 | /* This is not expected to happen; warn the user. */ | |
131 | log_level = PRINT_WARN; | |
132 | break; | |
133 | default: | |
134 | abort(); | |
db66e574 JD |
135 | } |
136 | ||
c7031a2c JG |
137 | LOG(log_level, "Rotation thread timer queue still contains job of type %s targeting session \"%s\" on destruction", |
138 | job_type_str, job->session->name); | |
db66e574 JD |
139 | } |
140 | ||
92816cc3 JG |
141 | void rotation_thread_timer_queue_destroy( |
142 | struct rotation_thread_timer_queue *queue) | |
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); | |
64d9b072 | 151 | assert(cds_list_empty(&queue->list)); |
92816cc3 JG |
152 | pthread_mutex_unlock(&queue->lock); |
153 | pthread_mutex_destroy(&queue->lock); | |
154 | free(queue); | |
155 | } | |
db66e574 | 156 | |
92816cc3 JG |
157 | /* |
158 | * Destroy the thread data previously created by the init function. | |
159 | */ | |
160 | void rotation_thread_handle_destroy( | |
161 | struct rotation_thread_handle *handle) | |
162 | { | |
64d9b072 | 163 | lttng_pipe_destroy(handle->quit_pipe); |
db66e574 JD |
164 | free(handle); |
165 | } | |
166 | ||
167 | struct rotation_thread_handle *rotation_thread_handle_create( | |
90936dcf | 168 | struct rotation_thread_timer_queue *rotation_timer_queue, |
c8a9de5a | 169 | struct notification_thread_handle *notification_thread_handle) |
db66e574 JD |
170 | { |
171 | struct rotation_thread_handle *handle; | |
172 | ||
173 | handle = zmalloc(sizeof(*handle)); | |
174 | if (!handle) { | |
175 | goto end; | |
176 | } | |
177 | ||
92816cc3 JG |
178 | handle->rotation_timer_queue = rotation_timer_queue; |
179 | handle->notification_thread_handle = notification_thread_handle; | |
64d9b072 JG |
180 | handle->quit_pipe = lttng_pipe_open(FD_CLOEXEC); |
181 | if (!handle->quit_pipe) { | |
182 | goto error; | |
183 | } | |
92816cc3 JG |
184 | |
185 | end: | |
186 | return handle; | |
64d9b072 JG |
187 | error: |
188 | rotation_thread_handle_destroy(handle); | |
189 | return NULL; | |
92816cc3 JG |
190 | } |
191 | ||
192 | /* | |
193 | * Called with the rotation_thread_timer_queue lock held. | |
194 | * Return true if the same timer job already exists in the queue, false if not. | |
195 | */ | |
196 | static | |
197 | bool timer_job_exists(const struct rotation_thread_timer_queue *queue, | |
c7031a2c JG |
198 | enum rotation_thread_job_type job_type, |
199 | struct ltt_session *session) | |
92816cc3 JG |
200 | { |
201 | bool exists = false; | |
202 | struct rotation_thread_job *job; | |
203 | ||
204 | cds_list_for_each_entry(job, &queue->list, head) { | |
c7031a2c | 205 | if (job->session == session && job->type == job_type) { |
92816cc3 JG |
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, | |
c7031a2c JG |
215 | enum rotation_thread_job_type job_type, |
216 | struct ltt_session *session) | |
92816cc3 JG |
217 | { |
218 | int ret; | |
be2956e7 | 219 | const char dummy = '!'; |
92816cc3 JG |
220 | struct rotation_thread_job *job = NULL; |
221 | const char *job_type_str = get_job_type_str(job_type); | |
222 | ||
223 | pthread_mutex_lock(&queue->lock); | |
c7031a2c | 224 | if (timer_job_exists(queue, job_type, session)) { |
92816cc3 JG |
225 | /* |
226 | * This timer job is already pending, we don't need to add | |
227 | * it. | |
228 | */ | |
229 | goto end; | |
db66e574 | 230 | } |
92816cc3 JG |
231 | |
232 | job = zmalloc(sizeof(struct rotation_thread_job)); | |
233 | if (!job) { | |
c7031a2c JG |
234 | PERROR("Failed to allocate rotation thread job of type \"%s\" for session \"%s\"", |
235 | job_type_str, session->name); | |
92816cc3 JG |
236 | goto end; |
237 | } | |
c7031a2c JG |
238 | /* No reason for this to fail as the caller must hold a reference. */ |
239 | (void) session_get(session); | |
240 | ||
241 | job->session = session; | |
92816cc3 | 242 | job->type = job_type; |
92816cc3 JG |
243 | cds_list_add_tail(&job->head, &queue->list); |
244 | ||
be2956e7 JG |
245 | ret = lttng_write(lttng_pipe_get_writefd(queue->event_pipe), &dummy, |
246 | sizeof(dummy)); | |
92816cc3 JG |
247 | if (ret < 0) { |
248 | /* | |
249 | * We do not want to block in the timer handler, the job has | |
250 | * been enqueued in the list, the wakeup pipe is probably full, | |
251 | * the job will be processed when the rotation_thread catches | |
252 | * up. | |
253 | */ | |
254 | if (errno == EAGAIN || errno == EWOULDBLOCK) { | |
255 | /* | |
256 | * Not an error, but would be surprising and indicate | |
257 | * that the rotation thread can't keep up with the | |
258 | * current load. | |
259 | */ | |
260 | DBG("Wake-up pipe of rotation thread job queue is full"); | |
261 | goto end; | |
db66e574 | 262 | } |
c7031a2c JG |
263 | PERROR("Failed to wake-up the rotation thread after pushing a job of type \"%s\" for session \"%s\"", |
264 | job_type_str, session->name); | |
92816cc3 | 265 | goto end; |
db66e574 | 266 | } |
db66e574 JD |
267 | |
268 | end: | |
92816cc3 | 269 | pthread_mutex_unlock(&queue->lock); |
db66e574 JD |
270 | } |
271 | ||
272 | static | |
273 | int init_poll_set(struct lttng_poll_event *poll_set, | |
274 | struct rotation_thread_handle *handle) | |
275 | { | |
276 | int ret; | |
277 | ||
278 | /* | |
64d9b072 JG |
279 | * Create pollset with size 3: |
280 | * - rotation thread quit pipe, | |
92816cc3 | 281 | * - rotation thread timer queue pipe, |
64d9b072 | 282 | * - notification channel sock, |
db66e574 | 283 | */ |
64d9b072 JG |
284 | ret = lttng_poll_create(poll_set, 5, LTTNG_CLOEXEC); |
285 | if (ret < 0) { | |
db66e574 JD |
286 | goto error; |
287 | } | |
64d9b072 JG |
288 | |
289 | ret = lttng_poll_add(poll_set, | |
290 | lttng_pipe_get_readfd(handle->quit_pipe), | |
291 | LPOLLIN | LPOLLERR); | |
292 | if (ret < 0) { | |
293 | ERR("[rotation-thread] Failed to add quit pipe read fd to poll set"); | |
294 | goto error; | |
295 | } | |
296 | ||
d086f507 JD |
297 | ret = lttng_poll_add(poll_set, |
298 | lttng_pipe_get_readfd(handle->rotation_timer_queue->event_pipe), | |
299 | LPOLLIN | LPOLLERR); | |
300 | if (ret < 0) { | |
64d9b072 | 301 | ERR("[rotation-thread] Failed to add rotate_pending fd to poll set"); |
d086f507 JD |
302 | goto error; |
303 | } | |
db66e574 | 304 | |
db66e574 JD |
305 | return ret; |
306 | error: | |
307 | lttng_poll_clean(poll_set); | |
308 | return ret; | |
309 | } | |
310 | ||
311 | static | |
92816cc3 | 312 | void fini_thread_state(struct rotation_thread *state) |
db66e574 JD |
313 | { |
314 | lttng_poll_clean(&state->events); | |
90936dcf JD |
315 | if (rotate_notification_channel) { |
316 | lttng_notification_channel_destroy(rotate_notification_channel); | |
317 | } | |
db66e574 JD |
318 | } |
319 | ||
320 | static | |
321 | int init_thread_state(struct rotation_thread_handle *handle, | |
92816cc3 | 322 | struct rotation_thread *state) |
db66e574 JD |
323 | { |
324 | int ret; | |
325 | ||
326 | memset(state, 0, sizeof(*state)); | |
327 | lttng_poll_init(&state->events); | |
328 | ||
329 | ret = init_poll_set(&state->events, handle); | |
330 | if (ret) { | |
331 | ERR("[rotation-thread] Failed to initialize rotation thread poll set"); | |
332 | goto end; | |
333 | } | |
334 | ||
90936dcf JD |
335 | rotate_notification_channel = lttng_notification_channel_create( |
336 | lttng_session_daemon_notification_endpoint); | |
337 | if (!rotate_notification_channel) { | |
338 | ERR("[rotation-thread] Could not create notification channel"); | |
339 | ret = -1; | |
340 | goto end; | |
341 | } | |
342 | ret = lttng_poll_add(&state->events, rotate_notification_channel->socket, | |
343 | LPOLLIN | LPOLLERR); | |
344 | if (ret < 0) { | |
345 | ERR("[rotation-thread] Failed to add notification fd to pollset"); | |
346 | goto end; | |
347 | } | |
348 | ||
db66e574 JD |
349 | end: |
350 | return ret; | |
351 | } | |
352 | ||
353 | static | |
d2956687 JG |
354 | void check_session_rotation_pending_on_consumers(struct ltt_session *session, |
355 | bool *_rotation_completed) | |
92816cc3 | 356 | { |
db582e11 | 357 | int ret = 0; |
92816cc3 JG |
358 | struct consumer_socket *socket; |
359 | struct cds_lfht_iter iter; | |
d2956687 JG |
360 | enum consumer_trace_chunk_exists_status exists_status; |
361 | uint64_t relayd_id; | |
362 | bool chunk_exists_on_peer = false; | |
363 | enum lttng_trace_chunk_status chunk_status; | |
364 | ||
365 | assert(session->chunk_being_archived); | |
92816cc3 JG |
366 | |
367 | /* | |
368 | * Check for a local pending rotation on all consumers (32-bit | |
369 | * user space, 64-bit user space, and kernel). | |
370 | */ | |
92816cc3 JG |
371 | rcu_read_lock(); |
372 | if (!session->ust_session) { | |
373 | goto skip_ust; | |
374 | } | |
375 | cds_lfht_for_each_entry(session->ust_session->consumer->socks->ht, | |
376 | &iter, socket, node.node) { | |
d2956687 JG |
377 | relayd_id = session->ust_session->consumer->type == CONSUMER_DST_LOCAL ? |
378 | -1ULL : | |
379 | session->ust_session->consumer->net_seq_index; | |
380 | ||
381 | pthread_mutex_lock(socket->lock); | |
382 | ret = consumer_trace_chunk_exists(socket, | |
383 | relayd_id, | |
384 | session->id, session->chunk_being_archived, | |
385 | &exists_status); | |
386 | if (ret) { | |
387 | pthread_mutex_unlock(socket->lock); | |
388 | ERR("Error occured while checking rotation status on consumer daemon"); | |
92816cc3 | 389 | goto end; |
db66e574 | 390 | } |
d2956687 JG |
391 | |
392 | if (exists_status != CONSUMER_TRACE_CHUNK_EXISTS_STATUS_UNKNOWN_CHUNK) { | |
393 | pthread_mutex_unlock(socket->lock); | |
394 | chunk_exists_on_peer = true; | |
395 | goto end; | |
396 | } | |
397 | pthread_mutex_unlock(socket->lock); | |
398 | } | |
db66e574 | 399 | |
92816cc3 JG |
400 | skip_ust: |
401 | if (!session->kernel_session) { | |
402 | goto skip_kernel; | |
db66e574 | 403 | } |
92816cc3 JG |
404 | cds_lfht_for_each_entry(session->kernel_session->consumer->socks->ht, |
405 | &iter, socket, node.node) { | |
d2956687 JG |
406 | pthread_mutex_lock(socket->lock); |
407 | relayd_id = session->kernel_session->consumer->type == CONSUMER_DST_LOCAL ? | |
408 | -1ULL : | |
409 | session->kernel_session->consumer->net_seq_index; | |
410 | ||
411 | ret = consumer_trace_chunk_exists(socket, | |
412 | relayd_id, | |
413 | session->id, session->chunk_being_archived, | |
414 | &exists_status); | |
415 | if (ret) { | |
416 | pthread_mutex_unlock(socket->lock); | |
417 | ERR("Error occured while checking rotation status on consumer daemon"); | |
92816cc3 JG |
418 | goto end; |
419 | } | |
d2956687 JG |
420 | |
421 | if (exists_status != CONSUMER_TRACE_CHUNK_EXISTS_STATUS_UNKNOWN_CHUNK) { | |
422 | pthread_mutex_unlock(socket->lock); | |
423 | chunk_exists_on_peer = true; | |
424 | goto end; | |
425 | } | |
426 | pthread_mutex_unlock(socket->lock); | |
92816cc3 JG |
427 | } |
428 | skip_kernel: | |
429 | end: | |
430 | rcu_read_unlock(); | |
db66e574 | 431 | |
d2956687 JG |
432 | if (!chunk_exists_on_peer) { |
433 | uint64_t chunk_being_archived_id; | |
434 | ||
435 | chunk_status = lttng_trace_chunk_get_id( | |
436 | session->chunk_being_archived, | |
437 | &chunk_being_archived_id); | |
438 | assert(chunk_status == LTTNG_TRACE_CHUNK_STATUS_OK); | |
439 | DBG("[rotation-thread] Rotation of trace archive %" PRIu64 " of session \"%s\" is complete on all consumers", | |
440 | chunk_being_archived_id, | |
92816cc3 | 441 | session->name); |
db66e574 | 442 | } |
d2956687 | 443 | *_rotation_completed = !chunk_exists_on_peer; |
92816cc3 | 444 | if (ret) { |
2961f09e JG |
445 | ret = session_reset_rotation_state(session, |
446 | LTTNG_ROTATION_STATE_ERROR); | |
447 | if (ret) { | |
448 | ERR("Failed to reset rotation state of session \"%s\"", | |
449 | session->name); | |
450 | } | |
db66e574 | 451 | } |
db66e574 JD |
452 | } |
453 | ||
d88744a4 | 454 | /* |
92816cc3 | 455 | * Check if the last rotation was completed, called with session lock held. |
d2956687 JG |
456 | * Should only return non-zero in the event of a fatal error. Doing so will |
457 | * shutdown the thread. | |
d88744a4 JD |
458 | */ |
459 | static | |
92816cc3 JG |
460 | int check_session_rotation_pending(struct ltt_session *session, |
461 | struct notification_thread_handle *notification_thread_handle) | |
d88744a4 JD |
462 | { |
463 | int ret; | |
92816cc3 | 464 | struct lttng_trace_archive_location *location; |
d2956687 JG |
465 | enum lttng_trace_chunk_status chunk_status; |
466 | bool rotation_completed = false; | |
467 | const char *archived_chunk_name; | |
468 | uint64_t chunk_being_archived_id; | |
469 | ||
dc1d5967 FD |
470 | if (!session->chunk_being_archived) { |
471 | ret = 0; | |
472 | goto end; | |
473 | } | |
474 | ||
d2956687 JG |
475 | chunk_status = lttng_trace_chunk_get_id(session->chunk_being_archived, |
476 | &chunk_being_archived_id); | |
477 | assert(chunk_status == LTTNG_TRACE_CHUNK_STATUS_OK); | |
d88744a4 | 478 | |
92816cc3 | 479 | DBG("[rotation-thread] Checking for pending rotation on session \"%s\", trace archive %" PRIu64, |
d2956687 JG |
480 | session->name, chunk_being_archived_id); |
481 | ||
faf1bdcf JG |
482 | /* |
483 | * The rotation-pending check timer of a session is launched in | |
484 | * one-shot mode. If the rotation is incomplete, the rotation | |
485 | * thread will re-enable the pending-check timer. | |
486 | * | |
487 | * The timer thread can't stop the timer itself since it is involved | |
488 | * in the check for the timer's quiescence. | |
489 | */ | |
490 | ret = timer_session_rotation_pending_check_stop(session); | |
491 | if (ret) { | |
6ae1bf46 | 492 | goto check_ongoing_rotation; |
faf1bdcf JG |
493 | } |
494 | ||
d2956687 JG |
495 | check_session_rotation_pending_on_consumers(session, |
496 | &rotation_completed); | |
d2956687 JG |
497 | if (!rotation_completed || |
498 | session->rotation_state == LTTNG_ROTATION_STATE_ERROR) { | |
6ae1bf46 | 499 | goto check_ongoing_rotation; |
92816cc3 JG |
500 | } |
501 | ||
92816cc3 JG |
502 | /* |
503 | * Now we can clear the "ONGOING" state in the session. New | |
504 | * rotations can start now. | |
505 | */ | |
d2956687 JG |
506 | chunk_status = lttng_trace_chunk_get_name(session->chunk_being_archived, |
507 | &archived_chunk_name, NULL); | |
508 | assert(chunk_status == LTTNG_TRACE_CHUNK_STATUS_OK); | |
509 | free(session->last_archived_chunk_name); | |
510 | session->last_archived_chunk_name = strdup(archived_chunk_name); | |
511 | if (!session->last_archived_chunk_name) { | |
512 | PERROR("Failed to duplicate archived chunk name"); | |
513 | } | |
514 | session_reset_rotation_state(session, LTTNG_ROTATION_STATE_COMPLETED); | |
92816cc3 | 515 | |
7fdbed1c JG |
516 | if (!session->quiet_rotation) { |
517 | location = session_get_trace_archive_location(session); | |
518 | /* Ownership of location is transferred. */ | |
519 | ret = notification_thread_command_session_rotation_completed( | |
520 | notification_thread_handle, | |
521 | session->name, | |
522 | session->uid, | |
523 | session->gid, | |
524 | session->last_archived_chunk_id.value, | |
525 | location); | |
526 | if (ret != LTTNG_OK) { | |
527 | ERR("[rotation-thread] Failed to notify notification thread of completed rotation for session %s", | |
528 | session->name); | |
529 | } | |
92816cc3 JG |
530 | } |
531 | ||
7fdbed1c | 532 | if (!session->active && !session->quiet_rotation) { |
9402c5b9 JG |
533 | /* |
534 | * A stop command was issued during the rotation, it is | |
535 | * up to the rotation completion check to perform the | |
536 | * renaming of the last chunk that was produced. | |
537 | */ | |
538 | ret = notification_thread_command_session_rotation_ongoing( | |
539 | notification_thread_handle, | |
540 | session->name, | |
541 | session->uid, | |
542 | session->gid, | |
d2956687 | 543 | session->most_recent_chunk_id.value); |
9402c5b9 JG |
544 | if (ret != LTTNG_OK) { |
545 | ERR("[rotation-thread] Failed to notify notification thread of completed rotation for session %s", | |
546 | session->name); | |
547 | } | |
548 | ||
9402c5b9 JG |
549 | /* Ownership of location is transferred. */ |
550 | location = session_get_trace_archive_location(session); | |
551 | ret = notification_thread_command_session_rotation_completed( | |
552 | notification_thread_handle, | |
553 | session->name, | |
554 | session->uid, | |
555 | session->gid, | |
d2956687 | 556 | session->most_recent_chunk_id.value, |
9402c5b9 JG |
557 | location); |
558 | if (ret != LTTNG_OK) { | |
559 | ERR("[rotation-thread] Failed to notify notification thread of completed rotation for session %s", | |
560 | session->name); | |
561 | } | |
562 | } | |
563 | ||
92816cc3 | 564 | ret = 0; |
6ae1bf46 | 565 | check_ongoing_rotation: |
92816cc3 | 566 | if (session->rotation_state == LTTNG_ROTATION_STATE_ONGOING) { |
d2956687 JG |
567 | uint64_t chunk_being_archived_id; |
568 | ||
569 | chunk_status = lttng_trace_chunk_get_id( | |
570 | session->chunk_being_archived, | |
571 | &chunk_being_archived_id); | |
572 | assert(chunk_status == LTTNG_TRACE_CHUNK_STATUS_OK); | |
573 | ||
92816cc3 | 574 | DBG("[rotation-thread] Rotation of trace archive %" PRIu64 " is still pending for session %s", |
d2956687 | 575 | chunk_being_archived_id, session->name); |
92816cc3 JG |
576 | ret = timer_session_rotation_pending_check_start(session, |
577 | DEFAULT_ROTATE_PENDING_TIMER); | |
578 | if (ret) { | |
d2956687 | 579 | ERR("Failed to re-enable rotation pending timer"); |
92816cc3 JG |
580 | ret = -1; |
581 | goto end; | |
582 | } | |
583 | } | |
584 | ||
6ae1bf46 | 585 | end: |
d88744a4 JD |
586 | return ret; |
587 | } | |
588 | ||
ed1e52a3 | 589 | /* Call with the session and session_list locks held. */ |
259c2674 | 590 | static |
92816cc3 | 591 | int launch_session_rotation(struct ltt_session *session) |
259c2674 JD |
592 | { |
593 | int ret; | |
92816cc3 | 594 | struct lttng_rotate_session_return rotation_return; |
259c2674 | 595 | |
92816cc3 JG |
596 | DBG("[rotation-thread] Launching scheduled time-based rotation on session \"%s\"", |
597 | session->name); | |
259c2674 | 598 | |
7fdbed1c | 599 | ret = cmd_rotate_session(session, &rotation_return, false); |
92816cc3 JG |
600 | if (ret == LTTNG_OK) { |
601 | DBG("[rotation-thread] Scheduled time-based rotation successfully launched on session \"%s\"", | |
602 | session->name); | |
603 | } else { | |
604 | /* Don't consider errors as fatal. */ | |
605 | DBG("[rotation-thread] Scheduled time-based rotation aborted for session %s: %s", | |
606 | session->name, lttng_strerror(ret)); | |
259c2674 | 607 | } |
92816cc3 JG |
608 | return 0; |
609 | } | |
259c2674 | 610 | |
92816cc3 JG |
611 | static |
612 | int run_job(struct rotation_thread_job *job, struct ltt_session *session, | |
613 | struct notification_thread_handle *notification_thread_handle) | |
614 | { | |
615 | int ret; | |
259c2674 | 616 | |
92816cc3 JG |
617 | switch (job->type) { |
618 | case ROTATION_THREAD_JOB_TYPE_SCHEDULED_ROTATION: | |
619 | ret = launch_session_rotation(session); | |
620 | break; | |
621 | case ROTATION_THREAD_JOB_TYPE_CHECK_PENDING_ROTATION: | |
622 | ret = check_session_rotation_pending(session, | |
623 | notification_thread_handle); | |
624 | break; | |
625 | default: | |
626 | abort(); | |
259c2674 | 627 | } |
259c2674 JD |
628 | return ret; |
629 | } | |
630 | ||
d88744a4 | 631 | static |
92816cc3 JG |
632 | int handle_job_queue(struct rotation_thread_handle *handle, |
633 | struct rotation_thread *state, | |
d88744a4 JD |
634 | struct rotation_thread_timer_queue *queue) |
635 | { | |
636 | int ret = 0; | |
d88744a4 JD |
637 | |
638 | for (;;) { | |
e32d7f27 | 639 | struct ltt_session *session; |
92816cc3 | 640 | struct rotation_thread_job *job; |
d88744a4 | 641 | |
92816cc3 | 642 | /* Take the queue lock only to pop an element from the list. */ |
d88744a4 JD |
643 | pthread_mutex_lock(&queue->lock); |
644 | if (cds_list_empty(&queue->list)) { | |
645 | pthread_mutex_unlock(&queue->lock); | |
646 | break; | |
647 | } | |
92816cc3 JG |
648 | job = cds_list_first_entry(&queue->list, |
649 | typeof(*job), head); | |
650 | cds_list_del(&job->head); | |
d88744a4 JD |
651 | pthread_mutex_unlock(&queue->lock); |
652 | ||
d88744a4 | 653 | session_lock_list(); |
c7031a2c | 654 | session = job->session; |
d88744a4 | 655 | if (!session) { |
c7031a2c JG |
656 | DBG("[rotation-thread] Session \"%s\" not found", |
657 | session->name); | |
d88744a4 | 658 | /* |
92816cc3 JG |
659 | * This is a non-fatal error, and we cannot report it to |
660 | * the user (timer), so just print the error and | |
661 | * continue the processing. | |
662 | * | |
663 | * While the timer thread will purge pending signals for | |
664 | * a session on the session's destruction, it is | |
665 | * possible for a job targeting that session to have | |
666 | * already been queued before it was destroyed. | |
d88744a4 | 667 | */ |
92816cc3 | 668 | free(job); |
e32d7f27 | 669 | session_put(session); |
5b8139c6 | 670 | session_unlock_list(); |
d88744a4 JD |
671 | continue; |
672 | } | |
673 | ||
d88744a4 | 674 | session_lock(session); |
92816cc3 | 675 | ret = run_job(job, session, handle->notification_thread_handle); |
d88744a4 | 676 | session_unlock(session); |
faf1bdcf | 677 | /* Release reference held by the job. */ |
e32d7f27 | 678 | session_put(session); |
ed1e52a3 | 679 | session_unlock_list(); |
92816cc3 | 680 | free(job); |
d88744a4 | 681 | if (ret) { |
d88744a4 JD |
682 | goto end; |
683 | } | |
684 | } | |
685 | ||
686 | ret = 0; | |
687 | ||
688 | end: | |
689 | return ret; | |
690 | } | |
691 | ||
92816cc3 JG |
692 | static |
693 | int handle_condition(const struct lttng_condition *condition, | |
90936dcf JD |
694 | const struct lttng_evaluation *evaluation, |
695 | struct notification_thread_handle *notification_thread_handle) | |
696 | { | |
697 | int ret = 0; | |
698 | const char *condition_session_name = NULL; | |
699 | enum lttng_condition_type condition_type; | |
700 | enum lttng_condition_status condition_status; | |
701 | enum lttng_evaluation_status evaluation_status; | |
702 | uint64_t consumed; | |
703 | struct ltt_session *session; | |
704 | ||
705 | condition_type = lttng_condition_get_type(condition); | |
706 | ||
707 | if (condition_type != LTTNG_CONDITION_TYPE_SESSION_CONSUMED_SIZE) { | |
708 | ret = -1; | |
709 | ERR("[rotation-thread] Condition type and session usage type are not the same"); | |
710 | goto end; | |
711 | } | |
712 | ||
713 | /* Fetch info to test */ | |
714 | condition_status = lttng_condition_session_consumed_size_get_session_name( | |
715 | condition, &condition_session_name); | |
716 | if (condition_status != LTTNG_CONDITION_STATUS_OK) { | |
717 | ERR("[rotation-thread] Session name could not be fetched"); | |
718 | ret = -1; | |
719 | goto end; | |
720 | } | |
721 | evaluation_status = lttng_evaluation_session_consumed_size_get_consumed_size(evaluation, | |
722 | &consumed); | |
723 | if (evaluation_status != LTTNG_EVALUATION_STATUS_OK) { | |
724 | ERR("[rotation-thread] Failed to get evaluation"); | |
725 | ret = -1; | |
726 | goto end; | |
727 | } | |
728 | ||
729 | session_lock_list(); | |
730 | session = session_find_by_name(condition_session_name); | |
731 | if (!session) { | |
732 | ret = -1; | |
733 | session_unlock_list(); | |
734 | ERR("[rotation-thread] Session \"%s\" not found", | |
735 | condition_session_name); | |
736 | goto end; | |
737 | } | |
738 | session_lock(session); | |
90936dcf JD |
739 | |
740 | ret = unsubscribe_session_consumed_size_rotation(session, | |
741 | notification_thread_handle); | |
742 | if (ret) { | |
490b3229 | 743 | goto end_unlock; |
90936dcf JD |
744 | } |
745 | ||
7fdbed1c | 746 | ret = cmd_rotate_session(session, NULL, false); |
90936dcf JD |
747 | if (ret == -LTTNG_ERR_ROTATION_PENDING) { |
748 | DBG("Rotate already pending, subscribe to the next threshold value"); | |
90936dcf JD |
749 | } else if (ret != LTTNG_OK) { |
750 | ERR("[rotation-thread] Failed to rotate on size notification with error: %s", | |
751 | lttng_strerror(ret)); | |
752 | ret = -1; | |
753 | goto end_unlock; | |
754 | } | |
755 | ret = subscribe_session_consumed_size_rotation(session, | |
756 | consumed + session->rotate_size, | |
757 | notification_thread_handle); | |
758 | if (ret) { | |
759 | ERR("[rotation-thread] Failed to subscribe to session consumed size condition"); | |
760 | goto end_unlock; | |
761 | } | |
762 | ret = 0; | |
763 | ||
764 | end_unlock: | |
765 | session_unlock(session); | |
e32d7f27 | 766 | session_put(session); |
5b8139c6 | 767 | session_unlock_list(); |
90936dcf JD |
768 | end: |
769 | return ret; | |
770 | } | |
771 | ||
772 | static | |
92816cc3 | 773 | int handle_notification_channel(int fd, |
90936dcf | 774 | struct rotation_thread_handle *handle, |
92816cc3 | 775 | struct rotation_thread *state) |
90936dcf JD |
776 | { |
777 | int ret; | |
d73ee93f JG |
778 | bool notification_pending; |
779 | struct lttng_notification *notification = NULL; | |
90936dcf JD |
780 | enum lttng_notification_channel_status status; |
781 | const struct lttng_evaluation *notification_evaluation; | |
782 | const struct lttng_condition *notification_condition; | |
783 | ||
d73ee93f JG |
784 | status = lttng_notification_channel_has_pending_notification( |
785 | rotate_notification_channel, ¬ification_pending); | |
786 | if (status != LTTNG_NOTIFICATION_CHANNEL_STATUS_OK) { | |
a9577b76 | 787 | ERR("[rotation-thread ]Error occurred while checking for pending notification"); |
d73ee93f JG |
788 | ret = -1; |
789 | goto end; | |
790 | } | |
791 | ||
792 | if (!notification_pending) { | |
793 | ret = 0; | |
794 | goto end; | |
795 | } | |
796 | ||
90936dcf JD |
797 | /* Receive the next notification. */ |
798 | status = lttng_notification_channel_get_next_notification( | |
799 | rotate_notification_channel, | |
800 | ¬ification); | |
801 | ||
802 | switch (status) { | |
803 | case LTTNG_NOTIFICATION_CHANNEL_STATUS_OK: | |
804 | break; | |
805 | case LTTNG_NOTIFICATION_CHANNEL_STATUS_NOTIFICATIONS_DROPPED: | |
806 | /* Not an error, we will wait for the next one */ | |
807 | ret = 0; | |
808 | goto end;; | |
809 | case LTTNG_NOTIFICATION_CHANNEL_STATUS_CLOSED: | |
810 | ERR("Notification channel was closed"); | |
811 | ret = -1; | |
812 | goto end; | |
813 | default: | |
814 | /* Unhandled conditions / errors. */ | |
815 | ERR("Unknown notification channel status"); | |
816 | ret = -1; | |
817 | goto end; | |
818 | } | |
819 | ||
820 | notification_condition = lttng_notification_get_condition(notification); | |
821 | notification_evaluation = lttng_notification_get_evaluation(notification); | |
822 | ||
823 | ret = handle_condition(notification_condition, notification_evaluation, | |
824 | handle->notification_thread_handle); | |
825 | ||
826 | end: | |
827 | lttng_notification_destroy(notification); | |
90936dcf JD |
828 | return ret; |
829 | } | |
830 | ||
db66e574 JD |
831 | void *thread_rotation(void *data) |
832 | { | |
833 | int ret; | |
834 | struct rotation_thread_handle *handle = data; | |
92816cc3 | 835 | struct rotation_thread thread; |
87380d40 | 836 | int queue_pipe_fd; |
db66e574 JD |
837 | |
838 | DBG("[rotation-thread] Started rotation thread"); | |
839 | ||
840 | if (!handle) { | |
841 | ERR("[rotation-thread] Invalid thread context provided"); | |
842 | goto end; | |
843 | } | |
844 | ||
87380d40 JR |
845 | queue_pipe_fd = lttng_pipe_get_readfd( |
846 | handle->rotation_timer_queue->event_pipe); | |
847 | ||
03732c32 JG |
848 | rcu_register_thread(); |
849 | rcu_thread_online(); | |
850 | ||
db66e574 JD |
851 | health_register(health_sessiond, HEALTH_SESSIOND_TYPE_ROTATION); |
852 | health_code_update(); | |
853 | ||
92816cc3 | 854 | ret = init_thread_state(handle, &thread); |
db66e574 | 855 | if (ret) { |
f5f8e5cd | 856 | goto error; |
db66e574 JD |
857 | } |
858 | ||
db66e574 JD |
859 | while (true) { |
860 | int fd_count, i; | |
861 | ||
862 | health_poll_entry(); | |
863 | DBG("[rotation-thread] Entering poll wait"); | |
92816cc3 | 864 | ret = lttng_poll_wait(&thread.events, -1); |
db66e574 JD |
865 | DBG("[rotation-thread] Poll wait returned (%i)", ret); |
866 | health_poll_exit(); | |
867 | if (ret < 0) { | |
868 | /* | |
869 | * Restart interrupted system call. | |
870 | */ | |
871 | if (errno == EINTR) { | |
872 | continue; | |
873 | } | |
874 | ERR("[rotation-thread] Error encountered during lttng_poll_wait (%i)", ret); | |
875 | goto error; | |
876 | } | |
877 | ||
878 | fd_count = ret; | |
879 | for (i = 0; i < fd_count; i++) { | |
92816cc3 JG |
880 | int fd = LTTNG_POLL_GETFD(&thread.events, i); |
881 | uint32_t revents = LTTNG_POLL_GETEV(&thread.events, i); | |
db66e574 JD |
882 | |
883 | DBG("[rotation-thread] Handling fd (%i) activity (%u)", | |
884 | fd, revents); | |
885 | ||
92816cc3 JG |
886 | if (revents & LPOLLERR) { |
887 | ERR("[rotation-thread] Polling returned an error on fd %i", fd); | |
888 | goto error; | |
889 | } | |
890 | ||
85e17b27 JG |
891 | if (fd == rotate_notification_channel->socket) { |
892 | ret = handle_notification_channel(fd, handle, | |
893 | &thread); | |
894 | if (ret) { | |
895 | ERR("[rotation-thread] Error occurred while handling activity on notification channel socket"); | |
896 | goto error; | |
897 | } | |
898 | } else { | |
899 | /* Job queue or quit pipe activity. */ | |
85e17b27 JG |
900 | |
901 | /* | |
902 | * The job queue is serviced if there is | |
903 | * activity on the quit pipe to ensure it is | |
904 | * flushed and all references held in the queue | |
905 | * are released. | |
906 | */ | |
92816cc3 JG |
907 | ret = handle_job_queue(handle, &thread, |
908 | handle->rotation_timer_queue); | |
d88744a4 JD |
909 | if (ret) { |
910 | ERR("[rotation-thread] Failed to handle rotation timer pipe event"); | |
911 | goto error; | |
912 | } | |
85e17b27 | 913 | |
64d9b072 JG |
914 | if (fd == queue_pipe_fd) { |
915 | char buf; | |
916 | ||
917 | ret = lttng_read(fd, &buf, 1); | |
918 | if (ret != 1) { | |
919 | ERR("[rotation-thread] Failed to read from wakeup pipe (fd = %i)", fd); | |
64d9b072 JG |
920 | goto error; |
921 | } | |
922 | } else { | |
85e17b27 JG |
923 | DBG("[rotation-thread] Quit pipe activity"); |
924 | goto exit; | |
90936dcf | 925 | } |
db66e574 JD |
926 | } |
927 | } | |
928 | } | |
929 | exit: | |
930 | error: | |
931 | DBG("[rotation-thread] Exit"); | |
92816cc3 | 932 | fini_thread_state(&thread); |
db66e574 | 933 | health_unregister(health_sessiond); |
03732c32 JG |
934 | rcu_thread_offline(); |
935 | rcu_unregister_thread(); | |
db66e574 JD |
936 | end: |
937 | return NULL; | |
938 | } | |
64d9b072 JG |
939 | |
940 | static | |
941 | bool shutdown_rotation_thread(void *thread_data) | |
942 | { | |
943 | struct rotation_thread_handle *handle = thread_data; | |
944 | const int write_fd = lttng_pipe_get_writefd(handle->quit_pipe); | |
945 | ||
946 | return notify_thread_pipe(write_fd) == 1; | |
947 | } | |
948 | ||
949 | bool launch_rotation_thread(struct rotation_thread_handle *handle) | |
950 | { | |
951 | struct lttng_thread *thread; | |
952 | ||
953 | thread = lttng_thread_create("Rotation", | |
954 | thread_rotation, | |
955 | shutdown_rotation_thread, | |
956 | NULL, | |
957 | handle); | |
958 | if (!thread) { | |
959 | goto error; | |
960 | } | |
961 | lttng_thread_put(thread); | |
962 | return true; | |
963 | error: | |
964 | return false; | |
965 | } |