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 JG |
114 | void rotation_thread_timer_queue_destroy( |
115 | struct rotation_thread_timer_queue *queue) | |
db66e574 | 116 | { |
92816cc3 JG |
117 | if (!queue) { |
118 | return; | |
db66e574 JD |
119 | } |
120 | ||
92816cc3 JG |
121 | lttng_pipe_destroy(queue->event_pipe); |
122 | ||
123 | pthread_mutex_lock(&queue->lock); | |
64d9b072 | 124 | assert(cds_list_empty(&queue->list)); |
92816cc3 JG |
125 | pthread_mutex_unlock(&queue->lock); |
126 | pthread_mutex_destroy(&queue->lock); | |
127 | free(queue); | |
128 | } | |
db66e574 | 129 | |
92816cc3 JG |
130 | /* |
131 | * Destroy the thread data previously created by the init function. | |
132 | */ | |
133 | void rotation_thread_handle_destroy( | |
134 | struct rotation_thread_handle *handle) | |
135 | { | |
64d9b072 | 136 | lttng_pipe_destroy(handle->quit_pipe); |
db66e574 JD |
137 | free(handle); |
138 | } | |
139 | ||
140 | struct rotation_thread_handle *rotation_thread_handle_create( | |
90936dcf | 141 | struct rotation_thread_timer_queue *rotation_timer_queue, |
c8a9de5a | 142 | struct notification_thread_handle *notification_thread_handle) |
db66e574 JD |
143 | { |
144 | struct rotation_thread_handle *handle; | |
145 | ||
146 | handle = zmalloc(sizeof(*handle)); | |
147 | if (!handle) { | |
148 | goto end; | |
149 | } | |
150 | ||
92816cc3 JG |
151 | handle->rotation_timer_queue = rotation_timer_queue; |
152 | handle->notification_thread_handle = notification_thread_handle; | |
64d9b072 JG |
153 | handle->quit_pipe = lttng_pipe_open(FD_CLOEXEC); |
154 | if (!handle->quit_pipe) { | |
155 | goto error; | |
156 | } | |
92816cc3 JG |
157 | |
158 | end: | |
159 | return handle; | |
64d9b072 JG |
160 | error: |
161 | rotation_thread_handle_destroy(handle); | |
162 | return NULL; | |
92816cc3 JG |
163 | } |
164 | ||
165 | /* | |
166 | * Called with the rotation_thread_timer_queue lock held. | |
167 | * Return true if the same timer job already exists in the queue, false if not. | |
168 | */ | |
169 | static | |
170 | bool timer_job_exists(const struct rotation_thread_timer_queue *queue, | |
c7031a2c JG |
171 | enum rotation_thread_job_type job_type, |
172 | struct ltt_session *session) | |
92816cc3 JG |
173 | { |
174 | bool exists = false; | |
175 | struct rotation_thread_job *job; | |
176 | ||
177 | cds_list_for_each_entry(job, &queue->list, head) { | |
c7031a2c | 178 | if (job->session == session && job->type == job_type) { |
92816cc3 JG |
179 | exists = true; |
180 | goto end; | |
db66e574 | 181 | } |
db66e574 | 182 | } |
92816cc3 JG |
183 | end: |
184 | return exists; | |
185 | } | |
186 | ||
187 | void rotation_thread_enqueue_job(struct rotation_thread_timer_queue *queue, | |
c7031a2c JG |
188 | enum rotation_thread_job_type job_type, |
189 | struct ltt_session *session) | |
92816cc3 JG |
190 | { |
191 | int ret; | |
be2956e7 | 192 | const char dummy = '!'; |
92816cc3 JG |
193 | struct rotation_thread_job *job = NULL; |
194 | const char *job_type_str = get_job_type_str(job_type); | |
195 | ||
196 | pthread_mutex_lock(&queue->lock); | |
c7031a2c | 197 | if (timer_job_exists(queue, job_type, session)) { |
92816cc3 JG |
198 | /* |
199 | * This timer job is already pending, we don't need to add | |
200 | * it. | |
201 | */ | |
202 | goto end; | |
db66e574 | 203 | } |
92816cc3 JG |
204 | |
205 | job = zmalloc(sizeof(struct rotation_thread_job)); | |
206 | if (!job) { | |
c7031a2c JG |
207 | PERROR("Failed to allocate rotation thread job of type \"%s\" for session \"%s\"", |
208 | job_type_str, session->name); | |
92816cc3 JG |
209 | goto end; |
210 | } | |
c7031a2c JG |
211 | /* No reason for this to fail as the caller must hold a reference. */ |
212 | (void) session_get(session); | |
213 | ||
214 | job->session = session; | |
92816cc3 | 215 | job->type = job_type; |
92816cc3 JG |
216 | cds_list_add_tail(&job->head, &queue->list); |
217 | ||
be2956e7 JG |
218 | ret = lttng_write(lttng_pipe_get_writefd(queue->event_pipe), &dummy, |
219 | sizeof(dummy)); | |
92816cc3 JG |
220 | if (ret < 0) { |
221 | /* | |
222 | * We do not want to block in the timer handler, the job has | |
223 | * been enqueued in the list, the wakeup pipe is probably full, | |
224 | * the job will be processed when the rotation_thread catches | |
225 | * up. | |
226 | */ | |
227 | if (errno == EAGAIN || errno == EWOULDBLOCK) { | |
228 | /* | |
229 | * Not an error, but would be surprising and indicate | |
230 | * that the rotation thread can't keep up with the | |
231 | * current load. | |
232 | */ | |
233 | DBG("Wake-up pipe of rotation thread job queue is full"); | |
234 | goto end; | |
db66e574 | 235 | } |
c7031a2c JG |
236 | PERROR("Failed to wake-up the rotation thread after pushing a job of type \"%s\" for session \"%s\"", |
237 | job_type_str, session->name); | |
92816cc3 | 238 | goto end; |
db66e574 | 239 | } |
db66e574 JD |
240 | |
241 | end: | |
92816cc3 | 242 | pthread_mutex_unlock(&queue->lock); |
db66e574 JD |
243 | } |
244 | ||
245 | static | |
246 | int init_poll_set(struct lttng_poll_event *poll_set, | |
247 | struct rotation_thread_handle *handle) | |
248 | { | |
249 | int ret; | |
250 | ||
251 | /* | |
64d9b072 JG |
252 | * Create pollset with size 3: |
253 | * - rotation thread quit pipe, | |
92816cc3 | 254 | * - rotation thread timer queue pipe, |
64d9b072 | 255 | * - notification channel sock, |
db66e574 | 256 | */ |
64d9b072 JG |
257 | ret = lttng_poll_create(poll_set, 5, LTTNG_CLOEXEC); |
258 | if (ret < 0) { | |
db66e574 JD |
259 | goto error; |
260 | } | |
64d9b072 JG |
261 | |
262 | ret = lttng_poll_add(poll_set, | |
263 | lttng_pipe_get_readfd(handle->quit_pipe), | |
264 | LPOLLIN | LPOLLERR); | |
265 | if (ret < 0) { | |
266 | ERR("[rotation-thread] Failed to add quit pipe read fd to poll set"); | |
267 | goto error; | |
268 | } | |
269 | ||
d086f507 JD |
270 | ret = lttng_poll_add(poll_set, |
271 | lttng_pipe_get_readfd(handle->rotation_timer_queue->event_pipe), | |
272 | LPOLLIN | LPOLLERR); | |
273 | if (ret < 0) { | |
64d9b072 | 274 | ERR("[rotation-thread] Failed to add rotate_pending fd to poll set"); |
d086f507 JD |
275 | goto error; |
276 | } | |
db66e574 | 277 | |
db66e574 JD |
278 | return ret; |
279 | error: | |
280 | lttng_poll_clean(poll_set); | |
281 | return ret; | |
282 | } | |
283 | ||
284 | static | |
92816cc3 | 285 | void fini_thread_state(struct rotation_thread *state) |
db66e574 JD |
286 | { |
287 | lttng_poll_clean(&state->events); | |
90936dcf JD |
288 | if (rotate_notification_channel) { |
289 | lttng_notification_channel_destroy(rotate_notification_channel); | |
290 | } | |
db66e574 JD |
291 | } |
292 | ||
293 | static | |
294 | int init_thread_state(struct rotation_thread_handle *handle, | |
92816cc3 | 295 | struct rotation_thread *state) |
db66e574 JD |
296 | { |
297 | int ret; | |
298 | ||
299 | memset(state, 0, sizeof(*state)); | |
300 | lttng_poll_init(&state->events); | |
301 | ||
302 | ret = init_poll_set(&state->events, handle); | |
303 | if (ret) { | |
304 | ERR("[rotation-thread] Failed to initialize rotation thread poll set"); | |
305 | goto end; | |
306 | } | |
307 | ||
90936dcf JD |
308 | rotate_notification_channel = lttng_notification_channel_create( |
309 | lttng_session_daemon_notification_endpoint); | |
310 | if (!rotate_notification_channel) { | |
311 | ERR("[rotation-thread] Could not create notification channel"); | |
312 | ret = -1; | |
313 | goto end; | |
314 | } | |
315 | ret = lttng_poll_add(&state->events, rotate_notification_channel->socket, | |
316 | LPOLLIN | LPOLLERR); | |
317 | if (ret < 0) { | |
318 | ERR("[rotation-thread] Failed to add notification fd to pollset"); | |
319 | goto end; | |
320 | } | |
321 | ||
db66e574 JD |
322 | end: |
323 | return ret; | |
324 | } | |
325 | ||
326 | static | |
d2956687 JG |
327 | void check_session_rotation_pending_on_consumers(struct ltt_session *session, |
328 | bool *_rotation_completed) | |
92816cc3 | 329 | { |
db582e11 | 330 | int ret = 0; |
92816cc3 JG |
331 | struct consumer_socket *socket; |
332 | struct cds_lfht_iter iter; | |
d2956687 JG |
333 | enum consumer_trace_chunk_exists_status exists_status; |
334 | uint64_t relayd_id; | |
335 | bool chunk_exists_on_peer = false; | |
336 | enum lttng_trace_chunk_status chunk_status; | |
337 | ||
338 | assert(session->chunk_being_archived); | |
92816cc3 JG |
339 | |
340 | /* | |
341 | * Check for a local pending rotation on all consumers (32-bit | |
342 | * user space, 64-bit user space, and kernel). | |
343 | */ | |
92816cc3 JG |
344 | rcu_read_lock(); |
345 | if (!session->ust_session) { | |
346 | goto skip_ust; | |
347 | } | |
348 | cds_lfht_for_each_entry(session->ust_session->consumer->socks->ht, | |
349 | &iter, socket, node.node) { | |
d2956687 JG |
350 | relayd_id = session->ust_session->consumer->type == CONSUMER_DST_LOCAL ? |
351 | -1ULL : | |
352 | session->ust_session->consumer->net_seq_index; | |
353 | ||
354 | pthread_mutex_lock(socket->lock); | |
355 | ret = consumer_trace_chunk_exists(socket, | |
356 | relayd_id, | |
357 | session->id, session->chunk_being_archived, | |
358 | &exists_status); | |
359 | if (ret) { | |
360 | pthread_mutex_unlock(socket->lock); | |
83ed9e90 | 361 | ERR("Error occurred while checking rotation status on consumer daemon"); |
92816cc3 | 362 | goto end; |
db66e574 | 363 | } |
d2956687 JG |
364 | |
365 | if (exists_status != CONSUMER_TRACE_CHUNK_EXISTS_STATUS_UNKNOWN_CHUNK) { | |
366 | pthread_mutex_unlock(socket->lock); | |
367 | chunk_exists_on_peer = true; | |
368 | goto end; | |
369 | } | |
370 | pthread_mutex_unlock(socket->lock); | |
371 | } | |
db66e574 | 372 | |
92816cc3 JG |
373 | skip_ust: |
374 | if (!session->kernel_session) { | |
375 | goto skip_kernel; | |
db66e574 | 376 | } |
92816cc3 JG |
377 | cds_lfht_for_each_entry(session->kernel_session->consumer->socks->ht, |
378 | &iter, socket, node.node) { | |
d2956687 JG |
379 | pthread_mutex_lock(socket->lock); |
380 | relayd_id = session->kernel_session->consumer->type == CONSUMER_DST_LOCAL ? | |
381 | -1ULL : | |
382 | session->kernel_session->consumer->net_seq_index; | |
383 | ||
384 | ret = consumer_trace_chunk_exists(socket, | |
385 | relayd_id, | |
386 | session->id, session->chunk_being_archived, | |
387 | &exists_status); | |
388 | if (ret) { | |
389 | pthread_mutex_unlock(socket->lock); | |
83ed9e90 | 390 | ERR("Error occurred while checking rotation status on consumer daemon"); |
92816cc3 JG |
391 | goto end; |
392 | } | |
d2956687 JG |
393 | |
394 | if (exists_status != CONSUMER_TRACE_CHUNK_EXISTS_STATUS_UNKNOWN_CHUNK) { | |
395 | pthread_mutex_unlock(socket->lock); | |
396 | chunk_exists_on_peer = true; | |
397 | goto end; | |
398 | } | |
399 | pthread_mutex_unlock(socket->lock); | |
92816cc3 JG |
400 | } |
401 | skip_kernel: | |
402 | end: | |
403 | rcu_read_unlock(); | |
db66e574 | 404 | |
d2956687 JG |
405 | if (!chunk_exists_on_peer) { |
406 | uint64_t chunk_being_archived_id; | |
407 | ||
408 | chunk_status = lttng_trace_chunk_get_id( | |
409 | session->chunk_being_archived, | |
410 | &chunk_being_archived_id); | |
411 | assert(chunk_status == LTTNG_TRACE_CHUNK_STATUS_OK); | |
412 | DBG("[rotation-thread] Rotation of trace archive %" PRIu64 " of session \"%s\" is complete on all consumers", | |
413 | chunk_being_archived_id, | |
92816cc3 | 414 | session->name); |
db66e574 | 415 | } |
d2956687 | 416 | *_rotation_completed = !chunk_exists_on_peer; |
92816cc3 | 417 | if (ret) { |
2961f09e JG |
418 | ret = session_reset_rotation_state(session, |
419 | LTTNG_ROTATION_STATE_ERROR); | |
420 | if (ret) { | |
421 | ERR("Failed to reset rotation state of session \"%s\"", | |
422 | session->name); | |
423 | } | |
db66e574 | 424 | } |
db66e574 JD |
425 | } |
426 | ||
d88744a4 | 427 | /* |
92816cc3 | 428 | * Check if the last rotation was completed, called with session lock held. |
d2956687 JG |
429 | * Should only return non-zero in the event of a fatal error. Doing so will |
430 | * shutdown the thread. | |
d88744a4 JD |
431 | */ |
432 | static | |
92816cc3 JG |
433 | int check_session_rotation_pending(struct ltt_session *session, |
434 | struct notification_thread_handle *notification_thread_handle) | |
d88744a4 JD |
435 | { |
436 | int ret; | |
92816cc3 | 437 | struct lttng_trace_archive_location *location; |
d2956687 JG |
438 | enum lttng_trace_chunk_status chunk_status; |
439 | bool rotation_completed = false; | |
440 | const char *archived_chunk_name; | |
441 | uint64_t chunk_being_archived_id; | |
442 | ||
dc1d5967 FD |
443 | if (!session->chunk_being_archived) { |
444 | ret = 0; | |
445 | goto end; | |
446 | } | |
447 | ||
d2956687 JG |
448 | chunk_status = lttng_trace_chunk_get_id(session->chunk_being_archived, |
449 | &chunk_being_archived_id); | |
450 | assert(chunk_status == LTTNG_TRACE_CHUNK_STATUS_OK); | |
d88744a4 | 451 | |
92816cc3 | 452 | DBG("[rotation-thread] Checking for pending rotation on session \"%s\", trace archive %" PRIu64, |
d2956687 JG |
453 | session->name, chunk_being_archived_id); |
454 | ||
faf1bdcf JG |
455 | /* |
456 | * The rotation-pending check timer of a session is launched in | |
457 | * one-shot mode. If the rotation is incomplete, the rotation | |
458 | * thread will re-enable the pending-check timer. | |
459 | * | |
460 | * The timer thread can't stop the timer itself since it is involved | |
461 | * in the check for the timer's quiescence. | |
462 | */ | |
463 | ret = timer_session_rotation_pending_check_stop(session); | |
464 | if (ret) { | |
6ae1bf46 | 465 | goto check_ongoing_rotation; |
faf1bdcf JG |
466 | } |
467 | ||
d2956687 JG |
468 | check_session_rotation_pending_on_consumers(session, |
469 | &rotation_completed); | |
d2956687 JG |
470 | if (!rotation_completed || |
471 | session->rotation_state == LTTNG_ROTATION_STATE_ERROR) { | |
6ae1bf46 | 472 | goto check_ongoing_rotation; |
92816cc3 JG |
473 | } |
474 | ||
92816cc3 JG |
475 | /* |
476 | * Now we can clear the "ONGOING" state in the session. New | |
477 | * rotations can start now. | |
478 | */ | |
d2956687 JG |
479 | chunk_status = lttng_trace_chunk_get_name(session->chunk_being_archived, |
480 | &archived_chunk_name, NULL); | |
481 | assert(chunk_status == LTTNG_TRACE_CHUNK_STATUS_OK); | |
482 | free(session->last_archived_chunk_name); | |
483 | session->last_archived_chunk_name = strdup(archived_chunk_name); | |
484 | if (!session->last_archived_chunk_name) { | |
485 | PERROR("Failed to duplicate archived chunk name"); | |
486 | } | |
487 | session_reset_rotation_state(session, LTTNG_ROTATION_STATE_COMPLETED); | |
92816cc3 | 488 | |
7fdbed1c JG |
489 | if (!session->quiet_rotation) { |
490 | location = session_get_trace_archive_location(session); | |
491 | /* Ownership of location is transferred. */ | |
492 | ret = notification_thread_command_session_rotation_completed( | |
493 | notification_thread_handle, | |
494 | session->name, | |
495 | session->uid, | |
496 | session->gid, | |
497 | session->last_archived_chunk_id.value, | |
498 | location); | |
499 | if (ret != LTTNG_OK) { | |
500 | ERR("[rotation-thread] Failed to notify notification thread of completed rotation for session %s", | |
501 | session->name); | |
502 | } | |
92816cc3 JG |
503 | } |
504 | ||
505 | ret = 0; | |
6ae1bf46 | 506 | check_ongoing_rotation: |
92816cc3 | 507 | if (session->rotation_state == LTTNG_ROTATION_STATE_ONGOING) { |
d2956687 JG |
508 | uint64_t chunk_being_archived_id; |
509 | ||
510 | chunk_status = lttng_trace_chunk_get_id( | |
511 | session->chunk_being_archived, | |
512 | &chunk_being_archived_id); | |
513 | assert(chunk_status == LTTNG_TRACE_CHUNK_STATUS_OK); | |
514 | ||
92816cc3 | 515 | DBG("[rotation-thread] Rotation of trace archive %" PRIu64 " is still pending for session %s", |
d2956687 | 516 | chunk_being_archived_id, session->name); |
92816cc3 JG |
517 | ret = timer_session_rotation_pending_check_start(session, |
518 | DEFAULT_ROTATE_PENDING_TIMER); | |
519 | if (ret) { | |
d2956687 | 520 | ERR("Failed to re-enable rotation pending timer"); |
92816cc3 JG |
521 | ret = -1; |
522 | goto end; | |
523 | } | |
524 | } | |
525 | ||
6ae1bf46 | 526 | end: |
d88744a4 JD |
527 | return ret; |
528 | } | |
529 | ||
ed1e52a3 | 530 | /* Call with the session and session_list locks held. */ |
259c2674 | 531 | static |
92816cc3 | 532 | int launch_session_rotation(struct ltt_session *session) |
259c2674 JD |
533 | { |
534 | int ret; | |
92816cc3 | 535 | struct lttng_rotate_session_return rotation_return; |
259c2674 | 536 | |
92816cc3 JG |
537 | DBG("[rotation-thread] Launching scheduled time-based rotation on session \"%s\"", |
538 | session->name); | |
259c2674 | 539 | |
7fdbed1c | 540 | ret = cmd_rotate_session(session, &rotation_return, false); |
92816cc3 JG |
541 | if (ret == LTTNG_OK) { |
542 | DBG("[rotation-thread] Scheduled time-based rotation successfully launched on session \"%s\"", | |
543 | session->name); | |
544 | } else { | |
545 | /* Don't consider errors as fatal. */ | |
546 | DBG("[rotation-thread] Scheduled time-based rotation aborted for session %s: %s", | |
547 | session->name, lttng_strerror(ret)); | |
259c2674 | 548 | } |
92816cc3 JG |
549 | return 0; |
550 | } | |
259c2674 | 551 | |
92816cc3 JG |
552 | static |
553 | int run_job(struct rotation_thread_job *job, struct ltt_session *session, | |
554 | struct notification_thread_handle *notification_thread_handle) | |
555 | { | |
556 | int ret; | |
259c2674 | 557 | |
92816cc3 JG |
558 | switch (job->type) { |
559 | case ROTATION_THREAD_JOB_TYPE_SCHEDULED_ROTATION: | |
560 | ret = launch_session_rotation(session); | |
561 | break; | |
562 | case ROTATION_THREAD_JOB_TYPE_CHECK_PENDING_ROTATION: | |
563 | ret = check_session_rotation_pending(session, | |
564 | notification_thread_handle); | |
565 | break; | |
566 | default: | |
567 | abort(); | |
259c2674 | 568 | } |
259c2674 JD |
569 | return ret; |
570 | } | |
571 | ||
d88744a4 | 572 | static |
92816cc3 JG |
573 | int handle_job_queue(struct rotation_thread_handle *handle, |
574 | struct rotation_thread *state, | |
d88744a4 JD |
575 | struct rotation_thread_timer_queue *queue) |
576 | { | |
577 | int ret = 0; | |
d88744a4 JD |
578 | |
579 | for (;;) { | |
e32d7f27 | 580 | struct ltt_session *session; |
92816cc3 | 581 | struct rotation_thread_job *job; |
d88744a4 | 582 | |
92816cc3 | 583 | /* Take the queue lock only to pop an element from the list. */ |
d88744a4 JD |
584 | pthread_mutex_lock(&queue->lock); |
585 | if (cds_list_empty(&queue->list)) { | |
586 | pthread_mutex_unlock(&queue->lock); | |
587 | break; | |
588 | } | |
92816cc3 JG |
589 | job = cds_list_first_entry(&queue->list, |
590 | typeof(*job), head); | |
591 | cds_list_del(&job->head); | |
d88744a4 JD |
592 | pthread_mutex_unlock(&queue->lock); |
593 | ||
d88744a4 | 594 | session_lock_list(); |
c7031a2c | 595 | session = job->session; |
d88744a4 | 596 | if (!session) { |
c7031a2c JG |
597 | DBG("[rotation-thread] Session \"%s\" not found", |
598 | session->name); | |
d88744a4 | 599 | /* |
92816cc3 JG |
600 | * This is a non-fatal error, and we cannot report it to |
601 | * the user (timer), so just print the error and | |
602 | * continue the processing. | |
603 | * | |
604 | * While the timer thread will purge pending signals for | |
605 | * a session on the session's destruction, it is | |
606 | * possible for a job targeting that session to have | |
607 | * already been queued before it was destroyed. | |
d88744a4 | 608 | */ |
92816cc3 | 609 | free(job); |
e32d7f27 | 610 | session_put(session); |
5b8139c6 | 611 | session_unlock_list(); |
d88744a4 JD |
612 | continue; |
613 | } | |
614 | ||
d88744a4 | 615 | session_lock(session); |
92816cc3 | 616 | ret = run_job(job, session, handle->notification_thread_handle); |
d88744a4 | 617 | session_unlock(session); |
faf1bdcf | 618 | /* Release reference held by the job. */ |
e32d7f27 | 619 | session_put(session); |
ed1e52a3 | 620 | session_unlock_list(); |
92816cc3 | 621 | free(job); |
d88744a4 | 622 | if (ret) { |
d88744a4 JD |
623 | goto end; |
624 | } | |
625 | } | |
626 | ||
627 | ret = 0; | |
628 | ||
629 | end: | |
630 | return ret; | |
631 | } | |
632 | ||
92816cc3 JG |
633 | static |
634 | int handle_condition(const struct lttng_condition *condition, | |
90936dcf JD |
635 | const struct lttng_evaluation *evaluation, |
636 | struct notification_thread_handle *notification_thread_handle) | |
637 | { | |
638 | int ret = 0; | |
639 | const char *condition_session_name = NULL; | |
640 | enum lttng_condition_type condition_type; | |
641 | enum lttng_condition_status condition_status; | |
642 | enum lttng_evaluation_status evaluation_status; | |
643 | uint64_t consumed; | |
644 | struct ltt_session *session; | |
645 | ||
646 | condition_type = lttng_condition_get_type(condition); | |
647 | ||
648 | if (condition_type != LTTNG_CONDITION_TYPE_SESSION_CONSUMED_SIZE) { | |
649 | ret = -1; | |
650 | ERR("[rotation-thread] Condition type and session usage type are not the same"); | |
651 | goto end; | |
652 | } | |
653 | ||
654 | /* Fetch info to test */ | |
655 | condition_status = lttng_condition_session_consumed_size_get_session_name( | |
656 | condition, &condition_session_name); | |
657 | if (condition_status != LTTNG_CONDITION_STATUS_OK) { | |
658 | ERR("[rotation-thread] Session name could not be fetched"); | |
659 | ret = -1; | |
660 | goto end; | |
661 | } | |
662 | evaluation_status = lttng_evaluation_session_consumed_size_get_consumed_size(evaluation, | |
663 | &consumed); | |
664 | if (evaluation_status != LTTNG_EVALUATION_STATUS_OK) { | |
665 | ERR("[rotation-thread] Failed to get evaluation"); | |
666 | ret = -1; | |
667 | goto end; | |
668 | } | |
669 | ||
670 | session_lock_list(); | |
671 | session = session_find_by_name(condition_session_name); | |
672 | if (!session) { | |
673 | ret = -1; | |
674 | session_unlock_list(); | |
675 | ERR("[rotation-thread] Session \"%s\" not found", | |
676 | condition_session_name); | |
677 | goto end; | |
678 | } | |
679 | session_lock(session); | |
90936dcf JD |
680 | |
681 | ret = unsubscribe_session_consumed_size_rotation(session, | |
682 | notification_thread_handle); | |
683 | if (ret) { | |
490b3229 | 684 | goto end_unlock; |
90936dcf JD |
685 | } |
686 | ||
7fdbed1c | 687 | ret = cmd_rotate_session(session, NULL, false); |
90936dcf JD |
688 | if (ret == -LTTNG_ERR_ROTATION_PENDING) { |
689 | DBG("Rotate already pending, subscribe to the next threshold value"); | |
90936dcf JD |
690 | } else if (ret != LTTNG_OK) { |
691 | ERR("[rotation-thread] Failed to rotate on size notification with error: %s", | |
692 | lttng_strerror(ret)); | |
693 | ret = -1; | |
694 | goto end_unlock; | |
695 | } | |
696 | ret = subscribe_session_consumed_size_rotation(session, | |
697 | consumed + session->rotate_size, | |
698 | notification_thread_handle); | |
699 | if (ret) { | |
700 | ERR("[rotation-thread] Failed to subscribe to session consumed size condition"); | |
701 | goto end_unlock; | |
702 | } | |
703 | ret = 0; | |
704 | ||
705 | end_unlock: | |
706 | session_unlock(session); | |
e32d7f27 | 707 | session_put(session); |
5b8139c6 | 708 | session_unlock_list(); |
90936dcf JD |
709 | end: |
710 | return ret; | |
711 | } | |
712 | ||
713 | static | |
92816cc3 | 714 | int handle_notification_channel(int fd, |
90936dcf | 715 | struct rotation_thread_handle *handle, |
92816cc3 | 716 | struct rotation_thread *state) |
90936dcf JD |
717 | { |
718 | int ret; | |
d73ee93f JG |
719 | bool notification_pending; |
720 | struct lttng_notification *notification = NULL; | |
90936dcf JD |
721 | enum lttng_notification_channel_status status; |
722 | const struct lttng_evaluation *notification_evaluation; | |
723 | const struct lttng_condition *notification_condition; | |
724 | ||
d73ee93f JG |
725 | status = lttng_notification_channel_has_pending_notification( |
726 | rotate_notification_channel, ¬ification_pending); | |
727 | if (status != LTTNG_NOTIFICATION_CHANNEL_STATUS_OK) { | |
a9577b76 | 728 | ERR("[rotation-thread ]Error occurred while checking for pending notification"); |
d73ee93f JG |
729 | ret = -1; |
730 | goto end; | |
731 | } | |
732 | ||
733 | if (!notification_pending) { | |
734 | ret = 0; | |
735 | goto end; | |
736 | } | |
737 | ||
90936dcf JD |
738 | /* Receive the next notification. */ |
739 | status = lttng_notification_channel_get_next_notification( | |
740 | rotate_notification_channel, | |
741 | ¬ification); | |
742 | ||
743 | switch (status) { | |
744 | case LTTNG_NOTIFICATION_CHANNEL_STATUS_OK: | |
745 | break; | |
746 | case LTTNG_NOTIFICATION_CHANNEL_STATUS_NOTIFICATIONS_DROPPED: | |
747 | /* Not an error, we will wait for the next one */ | |
748 | ret = 0; | |
749 | goto end;; | |
750 | case LTTNG_NOTIFICATION_CHANNEL_STATUS_CLOSED: | |
751 | ERR("Notification channel was closed"); | |
752 | ret = -1; | |
753 | goto end; | |
754 | default: | |
755 | /* Unhandled conditions / errors. */ | |
756 | ERR("Unknown notification channel status"); | |
757 | ret = -1; | |
758 | goto end; | |
759 | } | |
760 | ||
761 | notification_condition = lttng_notification_get_condition(notification); | |
762 | notification_evaluation = lttng_notification_get_evaluation(notification); | |
763 | ||
764 | ret = handle_condition(notification_condition, notification_evaluation, | |
765 | handle->notification_thread_handle); | |
766 | ||
767 | end: | |
768 | lttng_notification_destroy(notification); | |
90936dcf JD |
769 | return ret; |
770 | } | |
771 | ||
2c0c9bbc | 772 | static |
db66e574 JD |
773 | void *thread_rotation(void *data) |
774 | { | |
775 | int ret; | |
776 | struct rotation_thread_handle *handle = data; | |
92816cc3 | 777 | struct rotation_thread thread; |
87380d40 | 778 | int queue_pipe_fd; |
db66e574 JD |
779 | |
780 | DBG("[rotation-thread] Started rotation thread"); | |
f620cc28 JG |
781 | rcu_register_thread(); |
782 | rcu_thread_online(); | |
783 | health_register(health_sessiond, HEALTH_SESSIOND_TYPE_ROTATION); | |
784 | health_code_update(); | |
db66e574 JD |
785 | |
786 | if (!handle) { | |
787 | ERR("[rotation-thread] Invalid thread context provided"); | |
788 | goto end; | |
789 | } | |
790 | ||
87380d40 JR |
791 | queue_pipe_fd = lttng_pipe_get_readfd( |
792 | handle->rotation_timer_queue->event_pipe); | |
793 | ||
db66e574 | 794 | |
92816cc3 | 795 | ret = init_thread_state(handle, &thread); |
db66e574 | 796 | if (ret) { |
f5f8e5cd | 797 | goto error; |
db66e574 JD |
798 | } |
799 | ||
db66e574 JD |
800 | while (true) { |
801 | int fd_count, i; | |
802 | ||
803 | health_poll_entry(); | |
804 | DBG("[rotation-thread] Entering poll wait"); | |
92816cc3 | 805 | ret = lttng_poll_wait(&thread.events, -1); |
db66e574 JD |
806 | DBG("[rotation-thread] Poll wait returned (%i)", ret); |
807 | health_poll_exit(); | |
808 | if (ret < 0) { | |
809 | /* | |
810 | * Restart interrupted system call. | |
811 | */ | |
812 | if (errno == EINTR) { | |
813 | continue; | |
814 | } | |
815 | ERR("[rotation-thread] Error encountered during lttng_poll_wait (%i)", ret); | |
816 | goto error; | |
817 | } | |
818 | ||
819 | fd_count = ret; | |
820 | for (i = 0; i < fd_count; i++) { | |
92816cc3 JG |
821 | int fd = LTTNG_POLL_GETFD(&thread.events, i); |
822 | uint32_t revents = LTTNG_POLL_GETEV(&thread.events, i); | |
db66e574 JD |
823 | |
824 | DBG("[rotation-thread] Handling fd (%i) activity (%u)", | |
825 | fd, revents); | |
826 | ||
92816cc3 JG |
827 | if (revents & LPOLLERR) { |
828 | ERR("[rotation-thread] Polling returned an error on fd %i", fd); | |
829 | goto error; | |
830 | } | |
831 | ||
85e17b27 JG |
832 | if (fd == rotate_notification_channel->socket) { |
833 | ret = handle_notification_channel(fd, handle, | |
834 | &thread); | |
835 | if (ret) { | |
836 | ERR("[rotation-thread] Error occurred while handling activity on notification channel socket"); | |
837 | goto error; | |
838 | } | |
839 | } else { | |
840 | /* Job queue or quit pipe activity. */ | |
85e17b27 JG |
841 | |
842 | /* | |
843 | * The job queue is serviced if there is | |
844 | * activity on the quit pipe to ensure it is | |
845 | * flushed and all references held in the queue | |
846 | * are released. | |
847 | */ | |
92816cc3 JG |
848 | ret = handle_job_queue(handle, &thread, |
849 | handle->rotation_timer_queue); | |
d88744a4 JD |
850 | if (ret) { |
851 | ERR("[rotation-thread] Failed to handle rotation timer pipe event"); | |
852 | goto error; | |
853 | } | |
85e17b27 | 854 | |
64d9b072 JG |
855 | if (fd == queue_pipe_fd) { |
856 | char buf; | |
857 | ||
858 | ret = lttng_read(fd, &buf, 1); | |
859 | if (ret != 1) { | |
860 | ERR("[rotation-thread] Failed to read from wakeup pipe (fd = %i)", fd); | |
64d9b072 JG |
861 | goto error; |
862 | } | |
863 | } else { | |
85e17b27 JG |
864 | DBG("[rotation-thread] Quit pipe activity"); |
865 | goto exit; | |
90936dcf | 866 | } |
db66e574 JD |
867 | } |
868 | } | |
869 | } | |
870 | exit: | |
871 | error: | |
872 | DBG("[rotation-thread] Exit"); | |
92816cc3 | 873 | fini_thread_state(&thread); |
f620cc28 | 874 | end: |
db66e574 | 875 | health_unregister(health_sessiond); |
03732c32 JG |
876 | rcu_thread_offline(); |
877 | rcu_unregister_thread(); | |
db66e574 JD |
878 | return NULL; |
879 | } | |
64d9b072 JG |
880 | |
881 | static | |
882 | bool shutdown_rotation_thread(void *thread_data) | |
883 | { | |
884 | struct rotation_thread_handle *handle = thread_data; | |
885 | const int write_fd = lttng_pipe_get_writefd(handle->quit_pipe); | |
886 | ||
887 | return notify_thread_pipe(write_fd) == 1; | |
888 | } | |
889 | ||
890 | bool launch_rotation_thread(struct rotation_thread_handle *handle) | |
891 | { | |
892 | struct lttng_thread *thread; | |
893 | ||
894 | thread = lttng_thread_create("Rotation", | |
895 | thread_rotation, | |
896 | shutdown_rotation_thread, | |
897 | NULL, | |
898 | handle); | |
899 | if (!thread) { | |
900 | goto error; | |
901 | } | |
902 | lttng_thread_put(thread); | |
903 | return true; | |
904 | error: | |
905 | return false; | |
906 | } |