Commit | Line | Data |
---|---|---|
db66e574 | 1 | /* |
ab5be9fa MJ |
2 | * Copyright (C) 2017 Julien Desfossez <jdesfossez@efficios.com> |
3 | * Copyright (C) 2018 Jérémie Galarneau <jeremie.galarneau@efficios.com> | |
db66e574 | 4 | * |
ab5be9fa | 5 | * SPDX-License-Identifier: GPL-2.0-only |
db66e574 | 6 | * |
db66e574 JD |
7 | */ |
8 | ||
9 | #define _LGPL_SOURCE | |
10 | #include <lttng/trigger/trigger.h> | |
11 | #include <common/error.h> | |
12 | #include <common/config/session-config.h> | |
13 | #include <common/defaults.h> | |
14 | #include <common/utils.h> | |
15 | #include <common/futex.h> | |
16 | #include <common/align.h> | |
17 | #include <common/time.h> | |
18 | #include <common/hashtable/utils.h> | |
19 | #include <sys/eventfd.h> | |
20 | #include <sys/stat.h> | |
21 | #include <time.h> | |
22 | #include <signal.h> | |
23 | #include <inttypes.h> | |
24 | ||
25 | #include <common/kernel-ctl/kernel-ctl.h> | |
26 | #include <lttng/notification/channel-internal.h> | |
5c408ad8 | 27 | #include <lttng/rotate-internal.h> |
db66e574 JD |
28 | |
29 | #include "rotation-thread.h" | |
30 | #include "lttng-sessiond.h" | |
31 | #include "health-sessiond.h" | |
32 | #include "rotate.h" | |
33 | #include "cmd.h" | |
34 | #include "session.h" | |
8e319828 | 35 | #include "timer.h" |
17dd1232 | 36 | #include "notification-thread-commands.h" |
64d9b072 JG |
37 | #include "utils.h" |
38 | #include "thread.h" | |
db66e574 JD |
39 | |
40 | #include <urcu.h> | |
41 | #include <urcu/list.h> | |
db66e574 | 42 | |
90936dcf JD |
43 | struct lttng_notification_channel *rotate_notification_channel = NULL; |
44 | ||
92816cc3 | 45 | struct rotation_thread { |
db66e574 JD |
46 | struct lttng_poll_event events; |
47 | }; | |
48 | ||
92816cc3 JG |
49 | struct rotation_thread_job { |
50 | enum rotation_thread_job_type type; | |
c7031a2c | 51 | struct ltt_session *session; |
92816cc3 JG |
52 | /* List member in struct rotation_thread_timer_queue. */ |
53 | struct cds_list_head head; | |
54 | }; | |
55 | ||
56 | /* | |
57 | * The timer thread enqueues jobs and wakes up the rotation thread. | |
58 | * When the rotation thread wakes up, it empties the queue. | |
59 | */ | |
60 | struct rotation_thread_timer_queue { | |
61 | struct lttng_pipe *event_pipe; | |
62 | struct cds_list_head list; | |
63 | pthread_mutex_t lock; | |
64 | }; | |
65 | ||
66 | struct rotation_thread_handle { | |
92816cc3 JG |
67 | struct rotation_thread_timer_queue *rotation_timer_queue; |
68 | /* Access to the notification thread cmd_queue */ | |
69 | struct notification_thread_handle *notification_thread_handle; | |
64d9b072 JG |
70 | /* Thread-specific quit pipe. */ |
71 | struct lttng_pipe *quit_pipe; | |
92816cc3 JG |
72 | }; |
73 | ||
db66e574 | 74 | static |
92816cc3 | 75 | const char *get_job_type_str(enum rotation_thread_job_type job_type) |
db66e574 | 76 | { |
92816cc3 JG |
77 | switch (job_type) { |
78 | case ROTATION_THREAD_JOB_TYPE_CHECK_PENDING_ROTATION: | |
79 | return "CHECK_PENDING_ROTATION"; | |
80 | case ROTATION_THREAD_JOB_TYPE_SCHEDULED_ROTATION: | |
81 | return "SCHEDULED_ROTATION"; | |
82 | default: | |
83 | abort(); | |
84 | } | |
db66e574 JD |
85 | } |
86 | ||
92816cc3 | 87 | struct rotation_thread_timer_queue *rotation_thread_timer_queue_create(void) |
db66e574 | 88 | { |
92816cc3 | 89 | struct rotation_thread_timer_queue *queue = NULL; |
db66e574 | 90 | |
92816cc3 JG |
91 | queue = zmalloc(sizeof(*queue)); |
92 | if (!queue) { | |
93 | PERROR("Failed to allocate timer rotate queue"); | |
94 | goto end; | |
95 | } | |
db66e574 | 96 | |
92816cc3 JG |
97 | queue->event_pipe = lttng_pipe_open(FD_CLOEXEC | O_NONBLOCK); |
98 | CDS_INIT_LIST_HEAD(&queue->list); | |
99 | pthread_mutex_init(&queue->lock, NULL); | |
100 | end: | |
101 | return queue; | |
db66e574 JD |
102 | } |
103 | ||
92816cc3 JG |
104 | void rotation_thread_timer_queue_destroy( |
105 | struct rotation_thread_timer_queue *queue) | |
db66e574 | 106 | { |
92816cc3 JG |
107 | if (!queue) { |
108 | return; | |
db66e574 JD |
109 | } |
110 | ||
92816cc3 JG |
111 | lttng_pipe_destroy(queue->event_pipe); |
112 | ||
113 | pthread_mutex_lock(&queue->lock); | |
64d9b072 | 114 | assert(cds_list_empty(&queue->list)); |
92816cc3 JG |
115 | pthread_mutex_unlock(&queue->lock); |
116 | pthread_mutex_destroy(&queue->lock); | |
117 | free(queue); | |
118 | } | |
db66e574 | 119 | |
92816cc3 JG |
120 | /* |
121 | * Destroy the thread data previously created by the init function. | |
122 | */ | |
123 | void rotation_thread_handle_destroy( | |
124 | struct rotation_thread_handle *handle) | |
125 | { | |
64d9b072 | 126 | lttng_pipe_destroy(handle->quit_pipe); |
db66e574 JD |
127 | free(handle); |
128 | } | |
129 | ||
130 | struct rotation_thread_handle *rotation_thread_handle_create( | |
90936dcf | 131 | struct rotation_thread_timer_queue *rotation_timer_queue, |
c8a9de5a | 132 | struct notification_thread_handle *notification_thread_handle) |
db66e574 JD |
133 | { |
134 | struct rotation_thread_handle *handle; | |
135 | ||
136 | handle = zmalloc(sizeof(*handle)); | |
137 | if (!handle) { | |
138 | goto end; | |
139 | } | |
140 | ||
92816cc3 JG |
141 | handle->rotation_timer_queue = rotation_timer_queue; |
142 | handle->notification_thread_handle = notification_thread_handle; | |
64d9b072 JG |
143 | handle->quit_pipe = lttng_pipe_open(FD_CLOEXEC); |
144 | if (!handle->quit_pipe) { | |
145 | goto error; | |
146 | } | |
92816cc3 JG |
147 | |
148 | end: | |
149 | return handle; | |
64d9b072 JG |
150 | error: |
151 | rotation_thread_handle_destroy(handle); | |
152 | return NULL; | |
92816cc3 JG |
153 | } |
154 | ||
155 | /* | |
156 | * Called with the rotation_thread_timer_queue lock held. | |
157 | * Return true if the same timer job already exists in the queue, false if not. | |
158 | */ | |
159 | static | |
160 | bool timer_job_exists(const struct rotation_thread_timer_queue *queue, | |
c7031a2c JG |
161 | enum rotation_thread_job_type job_type, |
162 | struct ltt_session *session) | |
92816cc3 JG |
163 | { |
164 | bool exists = false; | |
165 | struct rotation_thread_job *job; | |
166 | ||
167 | cds_list_for_each_entry(job, &queue->list, head) { | |
c7031a2c | 168 | if (job->session == session && job->type == job_type) { |
92816cc3 JG |
169 | exists = true; |
170 | goto end; | |
db66e574 | 171 | } |
db66e574 | 172 | } |
92816cc3 JG |
173 | end: |
174 | return exists; | |
175 | } | |
176 | ||
177 | void rotation_thread_enqueue_job(struct rotation_thread_timer_queue *queue, | |
c7031a2c JG |
178 | enum rotation_thread_job_type job_type, |
179 | struct ltt_session *session) | |
92816cc3 JG |
180 | { |
181 | int ret; | |
be2956e7 | 182 | const char dummy = '!'; |
92816cc3 JG |
183 | struct rotation_thread_job *job = NULL; |
184 | const char *job_type_str = get_job_type_str(job_type); | |
185 | ||
186 | pthread_mutex_lock(&queue->lock); | |
c7031a2c | 187 | if (timer_job_exists(queue, job_type, session)) { |
92816cc3 JG |
188 | /* |
189 | * This timer job is already pending, we don't need to add | |
190 | * it. | |
191 | */ | |
192 | goto end; | |
db66e574 | 193 | } |
92816cc3 JG |
194 | |
195 | job = zmalloc(sizeof(struct rotation_thread_job)); | |
196 | if (!job) { | |
c7031a2c JG |
197 | PERROR("Failed to allocate rotation thread job of type \"%s\" for session \"%s\"", |
198 | job_type_str, session->name); | |
92816cc3 JG |
199 | goto end; |
200 | } | |
c7031a2c JG |
201 | /* No reason for this to fail as the caller must hold a reference. */ |
202 | (void) session_get(session); | |
203 | ||
204 | job->session = session; | |
92816cc3 | 205 | job->type = job_type; |
92816cc3 JG |
206 | cds_list_add_tail(&job->head, &queue->list); |
207 | ||
be2956e7 JG |
208 | ret = lttng_write(lttng_pipe_get_writefd(queue->event_pipe), &dummy, |
209 | sizeof(dummy)); | |
92816cc3 JG |
210 | if (ret < 0) { |
211 | /* | |
212 | * We do not want to block in the timer handler, the job has | |
213 | * been enqueued in the list, the wakeup pipe is probably full, | |
214 | * the job will be processed when the rotation_thread catches | |
215 | * up. | |
216 | */ | |
217 | if (errno == EAGAIN || errno == EWOULDBLOCK) { | |
218 | /* | |
219 | * Not an error, but would be surprising and indicate | |
220 | * that the rotation thread can't keep up with the | |
221 | * current load. | |
222 | */ | |
223 | DBG("Wake-up pipe of rotation thread job queue is full"); | |
224 | goto end; | |
db66e574 | 225 | } |
c7031a2c JG |
226 | PERROR("Failed to wake-up the rotation thread after pushing a job of type \"%s\" for session \"%s\"", |
227 | job_type_str, session->name); | |
92816cc3 | 228 | goto end; |
db66e574 | 229 | } |
db66e574 JD |
230 | |
231 | end: | |
92816cc3 | 232 | pthread_mutex_unlock(&queue->lock); |
db66e574 JD |
233 | } |
234 | ||
235 | static | |
236 | int init_poll_set(struct lttng_poll_event *poll_set, | |
237 | struct rotation_thread_handle *handle) | |
238 | { | |
239 | int ret; | |
240 | ||
241 | /* | |
64d9b072 JG |
242 | * Create pollset with size 3: |
243 | * - rotation thread quit pipe, | |
92816cc3 | 244 | * - rotation thread timer queue pipe, |
64d9b072 | 245 | * - notification channel sock, |
db66e574 | 246 | */ |
64d9b072 JG |
247 | ret = lttng_poll_create(poll_set, 5, LTTNG_CLOEXEC); |
248 | if (ret < 0) { | |
db66e574 JD |
249 | goto error; |
250 | } | |
64d9b072 JG |
251 | |
252 | ret = lttng_poll_add(poll_set, | |
253 | lttng_pipe_get_readfd(handle->quit_pipe), | |
254 | LPOLLIN | LPOLLERR); | |
255 | if (ret < 0) { | |
256 | ERR("[rotation-thread] Failed to add quit pipe read fd to poll set"); | |
257 | goto error; | |
258 | } | |
259 | ||
d086f507 JD |
260 | ret = lttng_poll_add(poll_set, |
261 | lttng_pipe_get_readfd(handle->rotation_timer_queue->event_pipe), | |
262 | LPOLLIN | LPOLLERR); | |
263 | if (ret < 0) { | |
64d9b072 | 264 | ERR("[rotation-thread] Failed to add rotate_pending fd to poll set"); |
d086f507 JD |
265 | goto error; |
266 | } | |
db66e574 | 267 | |
db66e574 JD |
268 | return ret; |
269 | error: | |
270 | lttng_poll_clean(poll_set); | |
271 | return ret; | |
272 | } | |
273 | ||
274 | static | |
92816cc3 | 275 | void fini_thread_state(struct rotation_thread *state) |
db66e574 JD |
276 | { |
277 | lttng_poll_clean(&state->events); | |
90936dcf JD |
278 | if (rotate_notification_channel) { |
279 | lttng_notification_channel_destroy(rotate_notification_channel); | |
280 | } | |
db66e574 JD |
281 | } |
282 | ||
283 | static | |
284 | int init_thread_state(struct rotation_thread_handle *handle, | |
92816cc3 | 285 | struct rotation_thread *state) |
db66e574 JD |
286 | { |
287 | int ret; | |
288 | ||
289 | memset(state, 0, sizeof(*state)); | |
290 | lttng_poll_init(&state->events); | |
291 | ||
292 | ret = init_poll_set(&state->events, handle); | |
293 | if (ret) { | |
294 | ERR("[rotation-thread] Failed to initialize rotation thread poll set"); | |
295 | goto end; | |
296 | } | |
297 | ||
90936dcf JD |
298 | rotate_notification_channel = lttng_notification_channel_create( |
299 | lttng_session_daemon_notification_endpoint); | |
300 | if (!rotate_notification_channel) { | |
301 | ERR("[rotation-thread] Could not create notification channel"); | |
302 | ret = -1; | |
303 | goto end; | |
304 | } | |
305 | ret = lttng_poll_add(&state->events, rotate_notification_channel->socket, | |
306 | LPOLLIN | LPOLLERR); | |
307 | if (ret < 0) { | |
308 | ERR("[rotation-thread] Failed to add notification fd to pollset"); | |
309 | goto end; | |
310 | } | |
311 | ||
db66e574 JD |
312 | end: |
313 | return ret; | |
314 | } | |
315 | ||
316 | static | |
d2956687 JG |
317 | void check_session_rotation_pending_on_consumers(struct ltt_session *session, |
318 | bool *_rotation_completed) | |
92816cc3 | 319 | { |
db582e11 | 320 | int ret = 0; |
92816cc3 JG |
321 | struct consumer_socket *socket; |
322 | struct cds_lfht_iter iter; | |
d2956687 JG |
323 | enum consumer_trace_chunk_exists_status exists_status; |
324 | uint64_t relayd_id; | |
325 | bool chunk_exists_on_peer = false; | |
326 | enum lttng_trace_chunk_status chunk_status; | |
327 | ||
328 | assert(session->chunk_being_archived); | |
92816cc3 JG |
329 | |
330 | /* | |
331 | * Check for a local pending rotation on all consumers (32-bit | |
332 | * user space, 64-bit user space, and kernel). | |
333 | */ | |
92816cc3 JG |
334 | rcu_read_lock(); |
335 | if (!session->ust_session) { | |
336 | goto skip_ust; | |
337 | } | |
338 | cds_lfht_for_each_entry(session->ust_session->consumer->socks->ht, | |
339 | &iter, socket, node.node) { | |
d2956687 JG |
340 | relayd_id = session->ust_session->consumer->type == CONSUMER_DST_LOCAL ? |
341 | -1ULL : | |
342 | session->ust_session->consumer->net_seq_index; | |
343 | ||
344 | pthread_mutex_lock(socket->lock); | |
345 | ret = consumer_trace_chunk_exists(socket, | |
346 | relayd_id, | |
347 | session->id, session->chunk_being_archived, | |
348 | &exists_status); | |
349 | if (ret) { | |
350 | pthread_mutex_unlock(socket->lock); | |
83ed9e90 | 351 | ERR("Error occurred while checking rotation status on consumer daemon"); |
92816cc3 | 352 | goto end; |
db66e574 | 353 | } |
d2956687 JG |
354 | |
355 | if (exists_status != CONSUMER_TRACE_CHUNK_EXISTS_STATUS_UNKNOWN_CHUNK) { | |
356 | pthread_mutex_unlock(socket->lock); | |
357 | chunk_exists_on_peer = true; | |
358 | goto end; | |
359 | } | |
360 | pthread_mutex_unlock(socket->lock); | |
361 | } | |
db66e574 | 362 | |
92816cc3 JG |
363 | skip_ust: |
364 | if (!session->kernel_session) { | |
365 | goto skip_kernel; | |
db66e574 | 366 | } |
92816cc3 JG |
367 | cds_lfht_for_each_entry(session->kernel_session->consumer->socks->ht, |
368 | &iter, socket, node.node) { | |
d2956687 JG |
369 | pthread_mutex_lock(socket->lock); |
370 | relayd_id = session->kernel_session->consumer->type == CONSUMER_DST_LOCAL ? | |
371 | -1ULL : | |
372 | session->kernel_session->consumer->net_seq_index; | |
373 | ||
374 | ret = consumer_trace_chunk_exists(socket, | |
375 | relayd_id, | |
376 | session->id, session->chunk_being_archived, | |
377 | &exists_status); | |
378 | if (ret) { | |
379 | pthread_mutex_unlock(socket->lock); | |
83ed9e90 | 380 | ERR("Error occurred while checking rotation status on consumer daemon"); |
92816cc3 JG |
381 | goto end; |
382 | } | |
d2956687 JG |
383 | |
384 | if (exists_status != CONSUMER_TRACE_CHUNK_EXISTS_STATUS_UNKNOWN_CHUNK) { | |
385 | pthread_mutex_unlock(socket->lock); | |
386 | chunk_exists_on_peer = true; | |
387 | goto end; | |
388 | } | |
389 | pthread_mutex_unlock(socket->lock); | |
92816cc3 JG |
390 | } |
391 | skip_kernel: | |
392 | end: | |
393 | rcu_read_unlock(); | |
db66e574 | 394 | |
d2956687 JG |
395 | if (!chunk_exists_on_peer) { |
396 | uint64_t chunk_being_archived_id; | |
397 | ||
398 | chunk_status = lttng_trace_chunk_get_id( | |
399 | session->chunk_being_archived, | |
400 | &chunk_being_archived_id); | |
401 | assert(chunk_status == LTTNG_TRACE_CHUNK_STATUS_OK); | |
402 | DBG("[rotation-thread] Rotation of trace archive %" PRIu64 " of session \"%s\" is complete on all consumers", | |
403 | chunk_being_archived_id, | |
92816cc3 | 404 | session->name); |
db66e574 | 405 | } |
d2956687 | 406 | *_rotation_completed = !chunk_exists_on_peer; |
92816cc3 | 407 | if (ret) { |
2961f09e JG |
408 | ret = session_reset_rotation_state(session, |
409 | LTTNG_ROTATION_STATE_ERROR); | |
410 | if (ret) { | |
411 | ERR("Failed to reset rotation state of session \"%s\"", | |
412 | session->name); | |
413 | } | |
db66e574 | 414 | } |
db66e574 JD |
415 | } |
416 | ||
d88744a4 | 417 | /* |
92816cc3 | 418 | * Check if the last rotation was completed, called with session lock held. |
d2956687 JG |
419 | * Should only return non-zero in the event of a fatal error. Doing so will |
420 | * shutdown the thread. | |
d88744a4 JD |
421 | */ |
422 | static | |
92816cc3 JG |
423 | int check_session_rotation_pending(struct ltt_session *session, |
424 | struct notification_thread_handle *notification_thread_handle) | |
d88744a4 JD |
425 | { |
426 | int ret; | |
92816cc3 | 427 | struct lttng_trace_archive_location *location; |
d2956687 JG |
428 | enum lttng_trace_chunk_status chunk_status; |
429 | bool rotation_completed = false; | |
430 | const char *archived_chunk_name; | |
431 | uint64_t chunk_being_archived_id; | |
432 | ||
dc1d5967 FD |
433 | if (!session->chunk_being_archived) { |
434 | ret = 0; | |
435 | goto end; | |
436 | } | |
437 | ||
d2956687 JG |
438 | chunk_status = lttng_trace_chunk_get_id(session->chunk_being_archived, |
439 | &chunk_being_archived_id); | |
440 | assert(chunk_status == LTTNG_TRACE_CHUNK_STATUS_OK); | |
d88744a4 | 441 | |
92816cc3 | 442 | DBG("[rotation-thread] Checking for pending rotation on session \"%s\", trace archive %" PRIu64, |
d2956687 JG |
443 | session->name, chunk_being_archived_id); |
444 | ||
faf1bdcf JG |
445 | /* |
446 | * The rotation-pending check timer of a session is launched in | |
447 | * one-shot mode. If the rotation is incomplete, the rotation | |
448 | * thread will re-enable the pending-check timer. | |
449 | * | |
450 | * The timer thread can't stop the timer itself since it is involved | |
451 | * in the check for the timer's quiescence. | |
452 | */ | |
453 | ret = timer_session_rotation_pending_check_stop(session); | |
454 | if (ret) { | |
6ae1bf46 | 455 | goto check_ongoing_rotation; |
faf1bdcf JG |
456 | } |
457 | ||
d2956687 JG |
458 | check_session_rotation_pending_on_consumers(session, |
459 | &rotation_completed); | |
d2956687 JG |
460 | if (!rotation_completed || |
461 | session->rotation_state == LTTNG_ROTATION_STATE_ERROR) { | |
6ae1bf46 | 462 | goto check_ongoing_rotation; |
92816cc3 JG |
463 | } |
464 | ||
92816cc3 JG |
465 | /* |
466 | * Now we can clear the "ONGOING" state in the session. New | |
467 | * rotations can start now. | |
468 | */ | |
d2956687 JG |
469 | chunk_status = lttng_trace_chunk_get_name(session->chunk_being_archived, |
470 | &archived_chunk_name, NULL); | |
471 | assert(chunk_status == LTTNG_TRACE_CHUNK_STATUS_OK); | |
472 | free(session->last_archived_chunk_name); | |
473 | session->last_archived_chunk_name = strdup(archived_chunk_name); | |
474 | if (!session->last_archived_chunk_name) { | |
475 | PERROR("Failed to duplicate archived chunk name"); | |
476 | } | |
477 | session_reset_rotation_state(session, LTTNG_ROTATION_STATE_COMPLETED); | |
92816cc3 | 478 | |
7fdbed1c JG |
479 | if (!session->quiet_rotation) { |
480 | location = session_get_trace_archive_location(session); | |
481 | /* Ownership of location is transferred. */ | |
482 | ret = notification_thread_command_session_rotation_completed( | |
483 | notification_thread_handle, | |
484 | session->name, | |
485 | session->uid, | |
486 | session->gid, | |
487 | session->last_archived_chunk_id.value, | |
488 | location); | |
489 | if (ret != LTTNG_OK) { | |
490 | ERR("[rotation-thread] Failed to notify notification thread of completed rotation for session %s", | |
491 | session->name); | |
492 | } | |
92816cc3 JG |
493 | } |
494 | ||
495 | ret = 0; | |
6ae1bf46 | 496 | check_ongoing_rotation: |
92816cc3 | 497 | if (session->rotation_state == LTTNG_ROTATION_STATE_ONGOING) { |
d2956687 JG |
498 | uint64_t chunk_being_archived_id; |
499 | ||
500 | chunk_status = lttng_trace_chunk_get_id( | |
501 | session->chunk_being_archived, | |
502 | &chunk_being_archived_id); | |
503 | assert(chunk_status == LTTNG_TRACE_CHUNK_STATUS_OK); | |
504 | ||
92816cc3 | 505 | DBG("[rotation-thread] Rotation of trace archive %" PRIu64 " is still pending for session %s", |
d2956687 | 506 | chunk_being_archived_id, session->name); |
92816cc3 JG |
507 | ret = timer_session_rotation_pending_check_start(session, |
508 | DEFAULT_ROTATE_PENDING_TIMER); | |
509 | if (ret) { | |
d2956687 | 510 | ERR("Failed to re-enable rotation pending timer"); |
92816cc3 JG |
511 | ret = -1; |
512 | goto end; | |
513 | } | |
514 | } | |
515 | ||
6ae1bf46 | 516 | end: |
d88744a4 JD |
517 | return ret; |
518 | } | |
519 | ||
ed1e52a3 | 520 | /* Call with the session and session_list locks held. */ |
259c2674 | 521 | static |
92816cc3 | 522 | int launch_session_rotation(struct ltt_session *session) |
259c2674 JD |
523 | { |
524 | int ret; | |
92816cc3 | 525 | struct lttng_rotate_session_return rotation_return; |
259c2674 | 526 | |
92816cc3 JG |
527 | DBG("[rotation-thread] Launching scheduled time-based rotation on session \"%s\"", |
528 | session->name); | |
259c2674 | 529 | |
343defc2 MD |
530 | ret = cmd_rotate_session(session, &rotation_return, false, |
531 | LTTNG_TRACE_CHUNK_COMMAND_TYPE_MOVE_TO_COMPLETED); | |
92816cc3 JG |
532 | if (ret == LTTNG_OK) { |
533 | DBG("[rotation-thread] Scheduled time-based rotation successfully launched on session \"%s\"", | |
534 | session->name); | |
535 | } else { | |
536 | /* Don't consider errors as fatal. */ | |
537 | DBG("[rotation-thread] Scheduled time-based rotation aborted for session %s: %s", | |
538 | session->name, lttng_strerror(ret)); | |
259c2674 | 539 | } |
92816cc3 JG |
540 | return 0; |
541 | } | |
259c2674 | 542 | |
92816cc3 JG |
543 | static |
544 | int run_job(struct rotation_thread_job *job, struct ltt_session *session, | |
545 | struct notification_thread_handle *notification_thread_handle) | |
546 | { | |
547 | int ret; | |
259c2674 | 548 | |
92816cc3 JG |
549 | switch (job->type) { |
550 | case ROTATION_THREAD_JOB_TYPE_SCHEDULED_ROTATION: | |
551 | ret = launch_session_rotation(session); | |
552 | break; | |
553 | case ROTATION_THREAD_JOB_TYPE_CHECK_PENDING_ROTATION: | |
554 | ret = check_session_rotation_pending(session, | |
555 | notification_thread_handle); | |
556 | break; | |
557 | default: | |
558 | abort(); | |
259c2674 | 559 | } |
259c2674 JD |
560 | return ret; |
561 | } | |
562 | ||
d88744a4 | 563 | static |
92816cc3 JG |
564 | int handle_job_queue(struct rotation_thread_handle *handle, |
565 | struct rotation_thread *state, | |
d88744a4 JD |
566 | struct rotation_thread_timer_queue *queue) |
567 | { | |
568 | int ret = 0; | |
d88744a4 JD |
569 | |
570 | for (;;) { | |
e32d7f27 | 571 | struct ltt_session *session; |
92816cc3 | 572 | struct rotation_thread_job *job; |
d88744a4 | 573 | |
92816cc3 | 574 | /* Take the queue lock only to pop an element from the list. */ |
d88744a4 JD |
575 | pthread_mutex_lock(&queue->lock); |
576 | if (cds_list_empty(&queue->list)) { | |
577 | pthread_mutex_unlock(&queue->lock); | |
578 | break; | |
579 | } | |
92816cc3 JG |
580 | job = cds_list_first_entry(&queue->list, |
581 | typeof(*job), head); | |
582 | cds_list_del(&job->head); | |
d88744a4 JD |
583 | pthread_mutex_unlock(&queue->lock); |
584 | ||
d88744a4 | 585 | session_lock_list(); |
c7031a2c | 586 | session = job->session; |
d88744a4 | 587 | if (!session) { |
c7031a2c JG |
588 | DBG("[rotation-thread] Session \"%s\" not found", |
589 | session->name); | |
d88744a4 | 590 | /* |
92816cc3 JG |
591 | * This is a non-fatal error, and we cannot report it to |
592 | * the user (timer), so just print the error and | |
593 | * continue the processing. | |
594 | * | |
595 | * While the timer thread will purge pending signals for | |
596 | * a session on the session's destruction, it is | |
597 | * possible for a job targeting that session to have | |
598 | * already been queued before it was destroyed. | |
d88744a4 | 599 | */ |
92816cc3 | 600 | free(job); |
e32d7f27 | 601 | session_put(session); |
5b8139c6 | 602 | session_unlock_list(); |
d88744a4 JD |
603 | continue; |
604 | } | |
605 | ||
d88744a4 | 606 | session_lock(session); |
92816cc3 | 607 | ret = run_job(job, session, handle->notification_thread_handle); |
d88744a4 | 608 | session_unlock(session); |
faf1bdcf | 609 | /* Release reference held by the job. */ |
e32d7f27 | 610 | session_put(session); |
ed1e52a3 | 611 | session_unlock_list(); |
92816cc3 | 612 | free(job); |
d88744a4 | 613 | if (ret) { |
d88744a4 JD |
614 | goto end; |
615 | } | |
616 | } | |
617 | ||
618 | ret = 0; | |
619 | ||
620 | end: | |
621 | return ret; | |
622 | } | |
623 | ||
92816cc3 JG |
624 | static |
625 | int handle_condition(const struct lttng_condition *condition, | |
90936dcf JD |
626 | const struct lttng_evaluation *evaluation, |
627 | struct notification_thread_handle *notification_thread_handle) | |
628 | { | |
629 | int ret = 0; | |
630 | const char *condition_session_name = NULL; | |
631 | enum lttng_condition_type condition_type; | |
632 | enum lttng_condition_status condition_status; | |
633 | enum lttng_evaluation_status evaluation_status; | |
634 | uint64_t consumed; | |
635 | struct ltt_session *session; | |
636 | ||
637 | condition_type = lttng_condition_get_type(condition); | |
638 | ||
639 | if (condition_type != LTTNG_CONDITION_TYPE_SESSION_CONSUMED_SIZE) { | |
640 | ret = -1; | |
641 | ERR("[rotation-thread] Condition type and session usage type are not the same"); | |
642 | goto end; | |
643 | } | |
644 | ||
645 | /* Fetch info to test */ | |
646 | condition_status = lttng_condition_session_consumed_size_get_session_name( | |
647 | condition, &condition_session_name); | |
648 | if (condition_status != LTTNG_CONDITION_STATUS_OK) { | |
649 | ERR("[rotation-thread] Session name could not be fetched"); | |
650 | ret = -1; | |
651 | goto end; | |
652 | } | |
653 | evaluation_status = lttng_evaluation_session_consumed_size_get_consumed_size(evaluation, | |
654 | &consumed); | |
655 | if (evaluation_status != LTTNG_EVALUATION_STATUS_OK) { | |
656 | ERR("[rotation-thread] Failed to get evaluation"); | |
657 | ret = -1; | |
658 | goto end; | |
659 | } | |
660 | ||
661 | session_lock_list(); | |
662 | session = session_find_by_name(condition_session_name); | |
663 | if (!session) { | |
664 | ret = -1; | |
665 | session_unlock_list(); | |
666 | ERR("[rotation-thread] Session \"%s\" not found", | |
667 | condition_session_name); | |
668 | goto end; | |
669 | } | |
670 | session_lock(session); | |
90936dcf JD |
671 | |
672 | ret = unsubscribe_session_consumed_size_rotation(session, | |
673 | notification_thread_handle); | |
674 | if (ret) { | |
490b3229 | 675 | goto end_unlock; |
90936dcf JD |
676 | } |
677 | ||
343defc2 MD |
678 | ret = cmd_rotate_session(session, NULL, false, |
679 | LTTNG_TRACE_CHUNK_COMMAND_TYPE_MOVE_TO_COMPLETED); | |
90936dcf JD |
680 | if (ret == -LTTNG_ERR_ROTATION_PENDING) { |
681 | DBG("Rotate already pending, subscribe to the next threshold value"); | |
90936dcf JD |
682 | } else if (ret != LTTNG_OK) { |
683 | ERR("[rotation-thread] Failed to rotate on size notification with error: %s", | |
684 | lttng_strerror(ret)); | |
685 | ret = -1; | |
686 | goto end_unlock; | |
687 | } | |
688 | ret = subscribe_session_consumed_size_rotation(session, | |
689 | consumed + session->rotate_size, | |
690 | notification_thread_handle); | |
691 | if (ret) { | |
692 | ERR("[rotation-thread] Failed to subscribe to session consumed size condition"); | |
693 | goto end_unlock; | |
694 | } | |
695 | ret = 0; | |
696 | ||
697 | end_unlock: | |
698 | session_unlock(session); | |
e32d7f27 | 699 | session_put(session); |
5b8139c6 | 700 | session_unlock_list(); |
90936dcf JD |
701 | end: |
702 | return ret; | |
703 | } | |
704 | ||
705 | static | |
92816cc3 | 706 | int handle_notification_channel(int fd, |
90936dcf | 707 | struct rotation_thread_handle *handle, |
92816cc3 | 708 | struct rotation_thread *state) |
90936dcf JD |
709 | { |
710 | int ret; | |
d73ee93f JG |
711 | bool notification_pending; |
712 | struct lttng_notification *notification = NULL; | |
90936dcf JD |
713 | enum lttng_notification_channel_status status; |
714 | const struct lttng_evaluation *notification_evaluation; | |
715 | const struct lttng_condition *notification_condition; | |
716 | ||
d73ee93f JG |
717 | status = lttng_notification_channel_has_pending_notification( |
718 | rotate_notification_channel, ¬ification_pending); | |
719 | if (status != LTTNG_NOTIFICATION_CHANNEL_STATUS_OK) { | |
a9577b76 | 720 | ERR("[rotation-thread ]Error occurred while checking for pending notification"); |
d73ee93f JG |
721 | ret = -1; |
722 | goto end; | |
723 | } | |
724 | ||
725 | if (!notification_pending) { | |
726 | ret = 0; | |
727 | goto end; | |
728 | } | |
729 | ||
90936dcf JD |
730 | /* Receive the next notification. */ |
731 | status = lttng_notification_channel_get_next_notification( | |
732 | rotate_notification_channel, | |
733 | ¬ification); | |
734 | ||
735 | switch (status) { | |
736 | case LTTNG_NOTIFICATION_CHANNEL_STATUS_OK: | |
737 | break; | |
738 | case LTTNG_NOTIFICATION_CHANNEL_STATUS_NOTIFICATIONS_DROPPED: | |
739 | /* Not an error, we will wait for the next one */ | |
740 | ret = 0; | |
741 | goto end;; | |
742 | case LTTNG_NOTIFICATION_CHANNEL_STATUS_CLOSED: | |
743 | ERR("Notification channel was closed"); | |
744 | ret = -1; | |
745 | goto end; | |
746 | default: | |
747 | /* Unhandled conditions / errors. */ | |
748 | ERR("Unknown notification channel status"); | |
749 | ret = -1; | |
750 | goto end; | |
751 | } | |
752 | ||
753 | notification_condition = lttng_notification_get_condition(notification); | |
754 | notification_evaluation = lttng_notification_get_evaluation(notification); | |
755 | ||
756 | ret = handle_condition(notification_condition, notification_evaluation, | |
757 | handle->notification_thread_handle); | |
758 | ||
759 | end: | |
760 | lttng_notification_destroy(notification); | |
90936dcf JD |
761 | return ret; |
762 | } | |
763 | ||
2c0c9bbc | 764 | static |
db66e574 JD |
765 | void *thread_rotation(void *data) |
766 | { | |
767 | int ret; | |
768 | struct rotation_thread_handle *handle = data; | |
92816cc3 | 769 | struct rotation_thread thread; |
87380d40 | 770 | int queue_pipe_fd; |
db66e574 JD |
771 | |
772 | DBG("[rotation-thread] Started rotation thread"); | |
f620cc28 JG |
773 | rcu_register_thread(); |
774 | rcu_thread_online(); | |
775 | health_register(health_sessiond, HEALTH_SESSIOND_TYPE_ROTATION); | |
776 | health_code_update(); | |
db66e574 JD |
777 | |
778 | if (!handle) { | |
779 | ERR("[rotation-thread] Invalid thread context provided"); | |
780 | goto end; | |
781 | } | |
782 | ||
87380d40 JR |
783 | queue_pipe_fd = lttng_pipe_get_readfd( |
784 | handle->rotation_timer_queue->event_pipe); | |
785 | ||
db66e574 | 786 | |
92816cc3 | 787 | ret = init_thread_state(handle, &thread); |
db66e574 | 788 | if (ret) { |
f5f8e5cd | 789 | goto error; |
db66e574 JD |
790 | } |
791 | ||
db66e574 JD |
792 | while (true) { |
793 | int fd_count, i; | |
794 | ||
795 | health_poll_entry(); | |
796 | DBG("[rotation-thread] Entering poll wait"); | |
92816cc3 | 797 | ret = lttng_poll_wait(&thread.events, -1); |
db66e574 JD |
798 | DBG("[rotation-thread] Poll wait returned (%i)", ret); |
799 | health_poll_exit(); | |
800 | if (ret < 0) { | |
801 | /* | |
802 | * Restart interrupted system call. | |
803 | */ | |
804 | if (errno == EINTR) { | |
805 | continue; | |
806 | } | |
807 | ERR("[rotation-thread] Error encountered during lttng_poll_wait (%i)", ret); | |
808 | goto error; | |
809 | } | |
810 | ||
811 | fd_count = ret; | |
812 | for (i = 0; i < fd_count; i++) { | |
92816cc3 JG |
813 | int fd = LTTNG_POLL_GETFD(&thread.events, i); |
814 | uint32_t revents = LTTNG_POLL_GETEV(&thread.events, i); | |
db66e574 JD |
815 | |
816 | DBG("[rotation-thread] Handling fd (%i) activity (%u)", | |
817 | fd, revents); | |
818 | ||
92816cc3 JG |
819 | if (revents & LPOLLERR) { |
820 | ERR("[rotation-thread] Polling returned an error on fd %i", fd); | |
821 | goto error; | |
822 | } | |
823 | ||
85e17b27 JG |
824 | if (fd == rotate_notification_channel->socket) { |
825 | ret = handle_notification_channel(fd, handle, | |
826 | &thread); | |
827 | if (ret) { | |
828 | ERR("[rotation-thread] Error occurred while handling activity on notification channel socket"); | |
829 | goto error; | |
830 | } | |
831 | } else { | |
832 | /* Job queue or quit pipe activity. */ | |
85e17b27 JG |
833 | |
834 | /* | |
835 | * The job queue is serviced if there is | |
836 | * activity on the quit pipe to ensure it is | |
837 | * flushed and all references held in the queue | |
838 | * are released. | |
839 | */ | |
92816cc3 JG |
840 | ret = handle_job_queue(handle, &thread, |
841 | handle->rotation_timer_queue); | |
d88744a4 JD |
842 | if (ret) { |
843 | ERR("[rotation-thread] Failed to handle rotation timer pipe event"); | |
844 | goto error; | |
845 | } | |
85e17b27 | 846 | |
64d9b072 JG |
847 | if (fd == queue_pipe_fd) { |
848 | char buf; | |
849 | ||
850 | ret = lttng_read(fd, &buf, 1); | |
851 | if (ret != 1) { | |
852 | ERR("[rotation-thread] Failed to read from wakeup pipe (fd = %i)", fd); | |
64d9b072 JG |
853 | goto error; |
854 | } | |
855 | } else { | |
85e17b27 JG |
856 | DBG("[rotation-thread] Quit pipe activity"); |
857 | goto exit; | |
90936dcf | 858 | } |
db66e574 JD |
859 | } |
860 | } | |
861 | } | |
862 | exit: | |
863 | error: | |
864 | DBG("[rotation-thread] Exit"); | |
92816cc3 | 865 | fini_thread_state(&thread); |
f620cc28 | 866 | end: |
db66e574 | 867 | health_unregister(health_sessiond); |
03732c32 JG |
868 | rcu_thread_offline(); |
869 | rcu_unregister_thread(); | |
db66e574 JD |
870 | return NULL; |
871 | } | |
64d9b072 JG |
872 | |
873 | static | |
874 | bool shutdown_rotation_thread(void *thread_data) | |
875 | { | |
876 | struct rotation_thread_handle *handle = thread_data; | |
877 | const int write_fd = lttng_pipe_get_writefd(handle->quit_pipe); | |
878 | ||
879 | return notify_thread_pipe(write_fd) == 1; | |
880 | } | |
881 | ||
882 | bool launch_rotation_thread(struct rotation_thread_handle *handle) | |
883 | { | |
884 | struct lttng_thread *thread; | |
885 | ||
886 | thread = lttng_thread_create("Rotation", | |
887 | thread_rotation, | |
888 | shutdown_rotation_thread, | |
889 | NULL, | |
890 | handle); | |
891 | if (!thread) { | |
892 | goto error; | |
893 | } | |
894 | lttng_thread_put(thread); | |
895 | return true; | |
896 | error: | |
897 | return false; | |
898 | } |