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> |
b0563fed | 28 | #include <lttng/location-internal.h> |
c19ea82d | 29 | #include <lttng/condition/condition-internal.h> |
69d5d2ae | 30 | #include <lttng/notification/notification-internal.h> |
db66e574 JD |
31 | |
32 | #include "rotation-thread.h" | |
33 | #include "lttng-sessiond.h" | |
34 | #include "health-sessiond.h" | |
35 | #include "rotate.h" | |
36 | #include "cmd.h" | |
37 | #include "session.h" | |
8e319828 | 38 | #include "timer.h" |
17dd1232 | 39 | #include "notification-thread-commands.h" |
64d9b072 JG |
40 | #include "utils.h" |
41 | #include "thread.h" | |
db66e574 JD |
42 | |
43 | #include <urcu.h> | |
44 | #include <urcu/list.h> | |
db66e574 | 45 | |
90936dcf JD |
46 | struct lttng_notification_channel *rotate_notification_channel = NULL; |
47 | ||
92816cc3 | 48 | struct rotation_thread { |
db66e574 JD |
49 | struct lttng_poll_event events; |
50 | }; | |
51 | ||
92816cc3 JG |
52 | struct rotation_thread_job { |
53 | enum rotation_thread_job_type type; | |
c7031a2c | 54 | struct ltt_session *session; |
92816cc3 JG |
55 | /* List member in struct rotation_thread_timer_queue. */ |
56 | struct cds_list_head head; | |
57 | }; | |
58 | ||
59 | /* | |
60 | * The timer thread enqueues jobs and wakes up the rotation thread. | |
61 | * When the rotation thread wakes up, it empties the queue. | |
62 | */ | |
63 | struct rotation_thread_timer_queue { | |
64 | struct lttng_pipe *event_pipe; | |
65 | struct cds_list_head list; | |
66 | pthread_mutex_t lock; | |
67 | }; | |
68 | ||
69 | struct rotation_thread_handle { | |
92816cc3 JG |
70 | struct rotation_thread_timer_queue *rotation_timer_queue; |
71 | /* Access to the notification thread cmd_queue */ | |
72 | struct notification_thread_handle *notification_thread_handle; | |
64d9b072 JG |
73 | /* Thread-specific quit pipe. */ |
74 | struct lttng_pipe *quit_pipe; | |
92816cc3 JG |
75 | }; |
76 | ||
db66e574 | 77 | static |
92816cc3 | 78 | const char *get_job_type_str(enum rotation_thread_job_type job_type) |
db66e574 | 79 | { |
92816cc3 JG |
80 | switch (job_type) { |
81 | case ROTATION_THREAD_JOB_TYPE_CHECK_PENDING_ROTATION: | |
82 | return "CHECK_PENDING_ROTATION"; | |
83 | case ROTATION_THREAD_JOB_TYPE_SCHEDULED_ROTATION: | |
84 | return "SCHEDULED_ROTATION"; | |
85 | default: | |
86 | abort(); | |
87 | } | |
db66e574 JD |
88 | } |
89 | ||
92816cc3 | 90 | struct rotation_thread_timer_queue *rotation_thread_timer_queue_create(void) |
db66e574 | 91 | { |
92816cc3 | 92 | struct rotation_thread_timer_queue *queue = NULL; |
db66e574 | 93 | |
92816cc3 JG |
94 | queue = zmalloc(sizeof(*queue)); |
95 | if (!queue) { | |
96 | PERROR("Failed to allocate timer rotate queue"); | |
97 | goto end; | |
98 | } | |
db66e574 | 99 | |
92816cc3 JG |
100 | queue->event_pipe = lttng_pipe_open(FD_CLOEXEC | O_NONBLOCK); |
101 | CDS_INIT_LIST_HEAD(&queue->list); | |
102 | pthread_mutex_init(&queue->lock, NULL); | |
103 | end: | |
104 | return queue; | |
db66e574 JD |
105 | } |
106 | ||
92816cc3 JG |
107 | void rotation_thread_timer_queue_destroy( |
108 | struct rotation_thread_timer_queue *queue) | |
db66e574 | 109 | { |
92816cc3 JG |
110 | if (!queue) { |
111 | return; | |
db66e574 JD |
112 | } |
113 | ||
92816cc3 JG |
114 | lttng_pipe_destroy(queue->event_pipe); |
115 | ||
116 | pthread_mutex_lock(&queue->lock); | |
64d9b072 | 117 | assert(cds_list_empty(&queue->list)); |
92816cc3 JG |
118 | pthread_mutex_unlock(&queue->lock); |
119 | pthread_mutex_destroy(&queue->lock); | |
120 | free(queue); | |
121 | } | |
db66e574 | 122 | |
92816cc3 JG |
123 | /* |
124 | * Destroy the thread data previously created by the init function. | |
125 | */ | |
126 | void rotation_thread_handle_destroy( | |
127 | struct rotation_thread_handle *handle) | |
128 | { | |
64d9b072 | 129 | lttng_pipe_destroy(handle->quit_pipe); |
db66e574 JD |
130 | free(handle); |
131 | } | |
132 | ||
133 | struct rotation_thread_handle *rotation_thread_handle_create( | |
90936dcf | 134 | struct rotation_thread_timer_queue *rotation_timer_queue, |
c8a9de5a | 135 | struct notification_thread_handle *notification_thread_handle) |
db66e574 JD |
136 | { |
137 | struct rotation_thread_handle *handle; | |
138 | ||
139 | handle = zmalloc(sizeof(*handle)); | |
140 | if (!handle) { | |
141 | goto end; | |
142 | } | |
143 | ||
92816cc3 JG |
144 | handle->rotation_timer_queue = rotation_timer_queue; |
145 | handle->notification_thread_handle = notification_thread_handle; | |
64d9b072 JG |
146 | handle->quit_pipe = lttng_pipe_open(FD_CLOEXEC); |
147 | if (!handle->quit_pipe) { | |
148 | goto error; | |
149 | } | |
92816cc3 JG |
150 | |
151 | end: | |
152 | return handle; | |
64d9b072 JG |
153 | error: |
154 | rotation_thread_handle_destroy(handle); | |
155 | return NULL; | |
92816cc3 JG |
156 | } |
157 | ||
158 | /* | |
159 | * Called with the rotation_thread_timer_queue lock held. | |
160 | * Return true if the same timer job already exists in the queue, false if not. | |
161 | */ | |
162 | static | |
163 | bool timer_job_exists(const struct rotation_thread_timer_queue *queue, | |
c7031a2c JG |
164 | enum rotation_thread_job_type job_type, |
165 | struct ltt_session *session) | |
92816cc3 JG |
166 | { |
167 | bool exists = false; | |
168 | struct rotation_thread_job *job; | |
169 | ||
170 | cds_list_for_each_entry(job, &queue->list, head) { | |
c7031a2c | 171 | if (job->session == session && job->type == job_type) { |
92816cc3 JG |
172 | exists = true; |
173 | goto end; | |
db66e574 | 174 | } |
db66e574 | 175 | } |
92816cc3 JG |
176 | end: |
177 | return exists; | |
178 | } | |
179 | ||
180 | void rotation_thread_enqueue_job(struct rotation_thread_timer_queue *queue, | |
c7031a2c JG |
181 | enum rotation_thread_job_type job_type, |
182 | struct ltt_session *session) | |
92816cc3 JG |
183 | { |
184 | int ret; | |
be2956e7 | 185 | const char dummy = '!'; |
92816cc3 JG |
186 | struct rotation_thread_job *job = NULL; |
187 | const char *job_type_str = get_job_type_str(job_type); | |
188 | ||
189 | pthread_mutex_lock(&queue->lock); | |
c7031a2c | 190 | if (timer_job_exists(queue, job_type, session)) { |
92816cc3 JG |
191 | /* |
192 | * This timer job is already pending, we don't need to add | |
193 | * it. | |
194 | */ | |
195 | goto end; | |
db66e574 | 196 | } |
92816cc3 JG |
197 | |
198 | job = zmalloc(sizeof(struct rotation_thread_job)); | |
199 | if (!job) { | |
c7031a2c JG |
200 | PERROR("Failed to allocate rotation thread job of type \"%s\" for session \"%s\"", |
201 | job_type_str, session->name); | |
92816cc3 JG |
202 | goto end; |
203 | } | |
c7031a2c JG |
204 | /* No reason for this to fail as the caller must hold a reference. */ |
205 | (void) session_get(session); | |
206 | ||
207 | job->session = session; | |
92816cc3 | 208 | job->type = job_type; |
92816cc3 JG |
209 | cds_list_add_tail(&job->head, &queue->list); |
210 | ||
be2956e7 JG |
211 | ret = lttng_write(lttng_pipe_get_writefd(queue->event_pipe), &dummy, |
212 | sizeof(dummy)); | |
92816cc3 JG |
213 | if (ret < 0) { |
214 | /* | |
215 | * We do not want to block in the timer handler, the job has | |
216 | * been enqueued in the list, the wakeup pipe is probably full, | |
217 | * the job will be processed when the rotation_thread catches | |
218 | * up. | |
219 | */ | |
220 | if (errno == EAGAIN || errno == EWOULDBLOCK) { | |
221 | /* | |
222 | * Not an error, but would be surprising and indicate | |
223 | * that the rotation thread can't keep up with the | |
224 | * current load. | |
225 | */ | |
226 | DBG("Wake-up pipe of rotation thread job queue is full"); | |
227 | goto end; | |
db66e574 | 228 | } |
c7031a2c JG |
229 | PERROR("Failed to wake-up the rotation thread after pushing a job of type \"%s\" for session \"%s\"", |
230 | job_type_str, session->name); | |
92816cc3 | 231 | goto end; |
db66e574 | 232 | } |
db66e574 JD |
233 | |
234 | end: | |
92816cc3 | 235 | pthread_mutex_unlock(&queue->lock); |
db66e574 JD |
236 | } |
237 | ||
238 | static | |
239 | int init_poll_set(struct lttng_poll_event *poll_set, | |
240 | struct rotation_thread_handle *handle) | |
241 | { | |
242 | int ret; | |
243 | ||
244 | /* | |
64d9b072 JG |
245 | * Create pollset with size 3: |
246 | * - rotation thread quit pipe, | |
92816cc3 | 247 | * - rotation thread timer queue pipe, |
64d9b072 | 248 | * - notification channel sock, |
db66e574 | 249 | */ |
64d9b072 JG |
250 | ret = lttng_poll_create(poll_set, 5, LTTNG_CLOEXEC); |
251 | if (ret < 0) { | |
db66e574 JD |
252 | goto error; |
253 | } | |
64d9b072 JG |
254 | |
255 | ret = lttng_poll_add(poll_set, | |
256 | lttng_pipe_get_readfd(handle->quit_pipe), | |
257 | LPOLLIN | LPOLLERR); | |
258 | if (ret < 0) { | |
259 | ERR("[rotation-thread] Failed to add quit pipe read fd to poll set"); | |
260 | goto error; | |
261 | } | |
262 | ||
d086f507 JD |
263 | ret = lttng_poll_add(poll_set, |
264 | lttng_pipe_get_readfd(handle->rotation_timer_queue->event_pipe), | |
265 | LPOLLIN | LPOLLERR); | |
266 | if (ret < 0) { | |
64d9b072 | 267 | ERR("[rotation-thread] Failed to add rotate_pending fd to poll set"); |
d086f507 JD |
268 | goto error; |
269 | } | |
db66e574 | 270 | |
db66e574 JD |
271 | return ret; |
272 | error: | |
273 | lttng_poll_clean(poll_set); | |
274 | return ret; | |
275 | } | |
276 | ||
277 | static | |
92816cc3 | 278 | void fini_thread_state(struct rotation_thread *state) |
db66e574 JD |
279 | { |
280 | lttng_poll_clean(&state->events); | |
90936dcf JD |
281 | if (rotate_notification_channel) { |
282 | lttng_notification_channel_destroy(rotate_notification_channel); | |
283 | } | |
db66e574 JD |
284 | } |
285 | ||
286 | static | |
287 | int init_thread_state(struct rotation_thread_handle *handle, | |
92816cc3 | 288 | struct rotation_thread *state) |
db66e574 JD |
289 | { |
290 | int ret; | |
291 | ||
292 | memset(state, 0, sizeof(*state)); | |
293 | lttng_poll_init(&state->events); | |
294 | ||
295 | ret = init_poll_set(&state->events, handle); | |
296 | if (ret) { | |
297 | ERR("[rotation-thread] Failed to initialize rotation thread poll set"); | |
298 | goto end; | |
299 | } | |
300 | ||
90936dcf JD |
301 | rotate_notification_channel = lttng_notification_channel_create( |
302 | lttng_session_daemon_notification_endpoint); | |
303 | if (!rotate_notification_channel) { | |
304 | ERR("[rotation-thread] Could not create notification channel"); | |
305 | ret = -1; | |
306 | goto end; | |
307 | } | |
308 | ret = lttng_poll_add(&state->events, rotate_notification_channel->socket, | |
309 | LPOLLIN | LPOLLERR); | |
310 | if (ret < 0) { | |
311 | ERR("[rotation-thread] Failed to add notification fd to pollset"); | |
312 | goto end; | |
313 | } | |
314 | ||
db66e574 JD |
315 | end: |
316 | return ret; | |
317 | } | |
318 | ||
319 | static | |
d2956687 JG |
320 | void check_session_rotation_pending_on_consumers(struct ltt_session *session, |
321 | bool *_rotation_completed) | |
92816cc3 | 322 | { |
db582e11 | 323 | int ret = 0; |
92816cc3 JG |
324 | struct consumer_socket *socket; |
325 | struct cds_lfht_iter iter; | |
d2956687 JG |
326 | enum consumer_trace_chunk_exists_status exists_status; |
327 | uint64_t relayd_id; | |
328 | bool chunk_exists_on_peer = false; | |
329 | enum lttng_trace_chunk_status chunk_status; | |
330 | ||
331 | assert(session->chunk_being_archived); | |
92816cc3 JG |
332 | |
333 | /* | |
334 | * Check for a local pending rotation on all consumers (32-bit | |
335 | * user space, 64-bit user space, and kernel). | |
336 | */ | |
92816cc3 JG |
337 | rcu_read_lock(); |
338 | if (!session->ust_session) { | |
339 | goto skip_ust; | |
340 | } | |
341 | cds_lfht_for_each_entry(session->ust_session->consumer->socks->ht, | |
342 | &iter, socket, node.node) { | |
d2956687 JG |
343 | relayd_id = session->ust_session->consumer->type == CONSUMER_DST_LOCAL ? |
344 | -1ULL : | |
345 | session->ust_session->consumer->net_seq_index; | |
346 | ||
347 | pthread_mutex_lock(socket->lock); | |
348 | ret = consumer_trace_chunk_exists(socket, | |
349 | relayd_id, | |
350 | session->id, session->chunk_being_archived, | |
351 | &exists_status); | |
352 | if (ret) { | |
353 | pthread_mutex_unlock(socket->lock); | |
83ed9e90 | 354 | ERR("Error occurred while checking rotation status on consumer daemon"); |
92816cc3 | 355 | goto end; |
db66e574 | 356 | } |
d2956687 JG |
357 | |
358 | if (exists_status != CONSUMER_TRACE_CHUNK_EXISTS_STATUS_UNKNOWN_CHUNK) { | |
359 | pthread_mutex_unlock(socket->lock); | |
360 | chunk_exists_on_peer = true; | |
361 | goto end; | |
362 | } | |
363 | pthread_mutex_unlock(socket->lock); | |
364 | } | |
db66e574 | 365 | |
92816cc3 JG |
366 | skip_ust: |
367 | if (!session->kernel_session) { | |
368 | goto skip_kernel; | |
db66e574 | 369 | } |
92816cc3 JG |
370 | cds_lfht_for_each_entry(session->kernel_session->consumer->socks->ht, |
371 | &iter, socket, node.node) { | |
d2956687 JG |
372 | pthread_mutex_lock(socket->lock); |
373 | relayd_id = session->kernel_session->consumer->type == CONSUMER_DST_LOCAL ? | |
374 | -1ULL : | |
375 | session->kernel_session->consumer->net_seq_index; | |
376 | ||
377 | ret = consumer_trace_chunk_exists(socket, | |
378 | relayd_id, | |
379 | session->id, session->chunk_being_archived, | |
380 | &exists_status); | |
381 | if (ret) { | |
382 | pthread_mutex_unlock(socket->lock); | |
83ed9e90 | 383 | ERR("Error occurred while checking rotation status on consumer daemon"); |
92816cc3 JG |
384 | goto end; |
385 | } | |
d2956687 JG |
386 | |
387 | if (exists_status != CONSUMER_TRACE_CHUNK_EXISTS_STATUS_UNKNOWN_CHUNK) { | |
388 | pthread_mutex_unlock(socket->lock); | |
389 | chunk_exists_on_peer = true; | |
390 | goto end; | |
391 | } | |
392 | pthread_mutex_unlock(socket->lock); | |
92816cc3 JG |
393 | } |
394 | skip_kernel: | |
395 | end: | |
396 | rcu_read_unlock(); | |
db66e574 | 397 | |
d2956687 JG |
398 | if (!chunk_exists_on_peer) { |
399 | uint64_t chunk_being_archived_id; | |
400 | ||
401 | chunk_status = lttng_trace_chunk_get_id( | |
402 | session->chunk_being_archived, | |
403 | &chunk_being_archived_id); | |
404 | assert(chunk_status == LTTNG_TRACE_CHUNK_STATUS_OK); | |
405 | DBG("[rotation-thread] Rotation of trace archive %" PRIu64 " of session \"%s\" is complete on all consumers", | |
406 | chunk_being_archived_id, | |
92816cc3 | 407 | session->name); |
db66e574 | 408 | } |
d2956687 | 409 | *_rotation_completed = !chunk_exists_on_peer; |
92816cc3 | 410 | if (ret) { |
2961f09e JG |
411 | ret = session_reset_rotation_state(session, |
412 | LTTNG_ROTATION_STATE_ERROR); | |
413 | if (ret) { | |
414 | ERR("Failed to reset rotation state of session \"%s\"", | |
415 | session->name); | |
416 | } | |
db66e574 | 417 | } |
db66e574 JD |
418 | } |
419 | ||
d88744a4 | 420 | /* |
92816cc3 | 421 | * Check if the last rotation was completed, called with session lock held. |
d2956687 JG |
422 | * Should only return non-zero in the event of a fatal error. Doing so will |
423 | * shutdown the thread. | |
d88744a4 JD |
424 | */ |
425 | static | |
92816cc3 JG |
426 | int check_session_rotation_pending(struct ltt_session *session, |
427 | struct notification_thread_handle *notification_thread_handle) | |
d88744a4 JD |
428 | { |
429 | int ret; | |
92816cc3 | 430 | struct lttng_trace_archive_location *location; |
d2956687 JG |
431 | enum lttng_trace_chunk_status chunk_status; |
432 | bool rotation_completed = false; | |
433 | const char *archived_chunk_name; | |
434 | uint64_t chunk_being_archived_id; | |
435 | ||
dc1d5967 FD |
436 | if (!session->chunk_being_archived) { |
437 | ret = 0; | |
438 | goto end; | |
439 | } | |
440 | ||
d2956687 JG |
441 | chunk_status = lttng_trace_chunk_get_id(session->chunk_being_archived, |
442 | &chunk_being_archived_id); | |
443 | assert(chunk_status == LTTNG_TRACE_CHUNK_STATUS_OK); | |
d88744a4 | 444 | |
92816cc3 | 445 | DBG("[rotation-thread] Checking for pending rotation on session \"%s\", trace archive %" PRIu64, |
d2956687 JG |
446 | session->name, chunk_being_archived_id); |
447 | ||
faf1bdcf JG |
448 | /* |
449 | * The rotation-pending check timer of a session is launched in | |
450 | * one-shot mode. If the rotation is incomplete, the rotation | |
451 | * thread will re-enable the pending-check timer. | |
452 | * | |
453 | * The timer thread can't stop the timer itself since it is involved | |
454 | * in the check for the timer's quiescence. | |
455 | */ | |
456 | ret = timer_session_rotation_pending_check_stop(session); | |
457 | if (ret) { | |
6ae1bf46 | 458 | goto check_ongoing_rotation; |
faf1bdcf JG |
459 | } |
460 | ||
d2956687 JG |
461 | check_session_rotation_pending_on_consumers(session, |
462 | &rotation_completed); | |
d2956687 JG |
463 | if (!rotation_completed || |
464 | session->rotation_state == LTTNG_ROTATION_STATE_ERROR) { | |
6ae1bf46 | 465 | goto check_ongoing_rotation; |
92816cc3 JG |
466 | } |
467 | ||
92816cc3 JG |
468 | /* |
469 | * Now we can clear the "ONGOING" state in the session. New | |
470 | * rotations can start now. | |
471 | */ | |
d2956687 JG |
472 | chunk_status = lttng_trace_chunk_get_name(session->chunk_being_archived, |
473 | &archived_chunk_name, NULL); | |
474 | assert(chunk_status == LTTNG_TRACE_CHUNK_STATUS_OK); | |
475 | free(session->last_archived_chunk_name); | |
476 | session->last_archived_chunk_name = strdup(archived_chunk_name); | |
477 | if (!session->last_archived_chunk_name) { | |
478 | PERROR("Failed to duplicate archived chunk name"); | |
479 | } | |
480 | session_reset_rotation_state(session, LTTNG_ROTATION_STATE_COMPLETED); | |
92816cc3 | 481 | |
7fdbed1c JG |
482 | if (!session->quiet_rotation) { |
483 | location = session_get_trace_archive_location(session); | |
7fdbed1c JG |
484 | ret = notification_thread_command_session_rotation_completed( |
485 | notification_thread_handle, | |
486 | session->name, | |
487 | session->uid, | |
488 | session->gid, | |
489 | session->last_archived_chunk_id.value, | |
490 | location); | |
b0563fed | 491 | lttng_trace_archive_location_put(location); |
7fdbed1c JG |
492 | if (ret != LTTNG_OK) { |
493 | ERR("[rotation-thread] Failed to notify notification thread of completed rotation for session %s", | |
494 | session->name); | |
495 | } | |
92816cc3 JG |
496 | } |
497 | ||
498 | ret = 0; | |
6ae1bf46 | 499 | check_ongoing_rotation: |
92816cc3 | 500 | if (session->rotation_state == LTTNG_ROTATION_STATE_ONGOING) { |
d2956687 JG |
501 | uint64_t chunk_being_archived_id; |
502 | ||
503 | chunk_status = lttng_trace_chunk_get_id( | |
504 | session->chunk_being_archived, | |
505 | &chunk_being_archived_id); | |
506 | assert(chunk_status == LTTNG_TRACE_CHUNK_STATUS_OK); | |
507 | ||
92816cc3 | 508 | DBG("[rotation-thread] Rotation of trace archive %" PRIu64 " is still pending for session %s", |
d2956687 | 509 | chunk_being_archived_id, session->name); |
92816cc3 JG |
510 | ret = timer_session_rotation_pending_check_start(session, |
511 | DEFAULT_ROTATE_PENDING_TIMER); | |
512 | if (ret) { | |
d2956687 | 513 | ERR("Failed to re-enable rotation pending timer"); |
92816cc3 JG |
514 | ret = -1; |
515 | goto end; | |
516 | } | |
517 | } | |
518 | ||
6ae1bf46 | 519 | end: |
d88744a4 JD |
520 | return ret; |
521 | } | |
522 | ||
ed1e52a3 | 523 | /* Call with the session and session_list locks held. */ |
259c2674 | 524 | static |
92816cc3 | 525 | int launch_session_rotation(struct ltt_session *session) |
259c2674 JD |
526 | { |
527 | int ret; | |
92816cc3 | 528 | struct lttng_rotate_session_return rotation_return; |
259c2674 | 529 | |
92816cc3 JG |
530 | DBG("[rotation-thread] Launching scheduled time-based rotation on session \"%s\"", |
531 | session->name); | |
259c2674 | 532 | |
343defc2 MD |
533 | ret = cmd_rotate_session(session, &rotation_return, false, |
534 | LTTNG_TRACE_CHUNK_COMMAND_TYPE_MOVE_TO_COMPLETED); | |
92816cc3 JG |
535 | if (ret == LTTNG_OK) { |
536 | DBG("[rotation-thread] Scheduled time-based rotation successfully launched on session \"%s\"", | |
537 | session->name); | |
538 | } else { | |
539 | /* Don't consider errors as fatal. */ | |
540 | DBG("[rotation-thread] Scheduled time-based rotation aborted for session %s: %s", | |
541 | session->name, lttng_strerror(ret)); | |
259c2674 | 542 | } |
92816cc3 JG |
543 | return 0; |
544 | } | |
259c2674 | 545 | |
92816cc3 JG |
546 | static |
547 | int run_job(struct rotation_thread_job *job, struct ltt_session *session, | |
548 | struct notification_thread_handle *notification_thread_handle) | |
549 | { | |
550 | int ret; | |
259c2674 | 551 | |
92816cc3 JG |
552 | switch (job->type) { |
553 | case ROTATION_THREAD_JOB_TYPE_SCHEDULED_ROTATION: | |
554 | ret = launch_session_rotation(session); | |
555 | break; | |
556 | case ROTATION_THREAD_JOB_TYPE_CHECK_PENDING_ROTATION: | |
557 | ret = check_session_rotation_pending(session, | |
558 | notification_thread_handle); | |
559 | break; | |
560 | default: | |
561 | abort(); | |
259c2674 | 562 | } |
259c2674 JD |
563 | return ret; |
564 | } | |
565 | ||
d88744a4 | 566 | static |
92816cc3 JG |
567 | int handle_job_queue(struct rotation_thread_handle *handle, |
568 | struct rotation_thread *state, | |
d88744a4 JD |
569 | struct rotation_thread_timer_queue *queue) |
570 | { | |
571 | int ret = 0; | |
d88744a4 JD |
572 | |
573 | for (;;) { | |
e32d7f27 | 574 | struct ltt_session *session; |
92816cc3 | 575 | struct rotation_thread_job *job; |
d88744a4 | 576 | |
92816cc3 | 577 | /* Take the queue lock only to pop an element from the list. */ |
d88744a4 JD |
578 | pthread_mutex_lock(&queue->lock); |
579 | if (cds_list_empty(&queue->list)) { | |
580 | pthread_mutex_unlock(&queue->lock); | |
581 | break; | |
582 | } | |
92816cc3 JG |
583 | job = cds_list_first_entry(&queue->list, |
584 | typeof(*job), head); | |
585 | cds_list_del(&job->head); | |
d88744a4 JD |
586 | pthread_mutex_unlock(&queue->lock); |
587 | ||
d88744a4 | 588 | session_lock_list(); |
c7031a2c | 589 | session = job->session; |
d88744a4 | 590 | if (!session) { |
12ee1056 | 591 | DBG("[rotation-thread] Session not found"); |
d88744a4 | 592 | /* |
92816cc3 JG |
593 | * This is a non-fatal error, and we cannot report it to |
594 | * the user (timer), so just print the error and | |
595 | * continue the processing. | |
596 | * | |
597 | * While the timer thread will purge pending signals for | |
598 | * a session on the session's destruction, it is | |
599 | * possible for a job targeting that session to have | |
600 | * already been queued before it was destroyed. | |
d88744a4 | 601 | */ |
92816cc3 | 602 | free(job); |
5b8139c6 | 603 | session_unlock_list(); |
d88744a4 JD |
604 | continue; |
605 | } | |
606 | ||
d88744a4 | 607 | session_lock(session); |
92816cc3 | 608 | ret = run_job(job, session, handle->notification_thread_handle); |
d88744a4 | 609 | session_unlock(session); |
faf1bdcf | 610 | /* Release reference held by the job. */ |
e32d7f27 | 611 | session_put(session); |
ed1e52a3 | 612 | session_unlock_list(); |
92816cc3 | 613 | free(job); |
d88744a4 | 614 | if (ret) { |
d88744a4 JD |
615 | goto end; |
616 | } | |
617 | } | |
618 | ||
619 | ret = 0; | |
620 | ||
621 | end: | |
622 | return ret; | |
623 | } | |
624 | ||
92816cc3 | 625 | static |
69d5d2ae | 626 | int handle_condition(const struct lttng_notification *notification, |
90936dcf JD |
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; | |
69d5d2ae JG |
636 | const struct lttng_condition *condition = |
637 | lttng_notification_get_const_condition(notification); | |
638 | const struct lttng_evaluation *evaluation = | |
639 | lttng_notification_get_const_evaluation(notification); | |
90936dcf JD |
640 | |
641 | condition_type = lttng_condition_get_type(condition); | |
642 | ||
643 | if (condition_type != LTTNG_CONDITION_TYPE_SESSION_CONSUMED_SIZE) { | |
644 | ret = -1; | |
645 | ERR("[rotation-thread] Condition type and session usage type are not the same"); | |
646 | goto end; | |
647 | } | |
648 | ||
649 | /* Fetch info to test */ | |
650 | condition_status = lttng_condition_session_consumed_size_get_session_name( | |
651 | condition, &condition_session_name); | |
652 | if (condition_status != LTTNG_CONDITION_STATUS_OK) { | |
653 | ERR("[rotation-thread] Session name could not be fetched"); | |
654 | ret = -1; | |
655 | goto end; | |
656 | } | |
657 | evaluation_status = lttng_evaluation_session_consumed_size_get_consumed_size(evaluation, | |
658 | &consumed); | |
659 | if (evaluation_status != LTTNG_EVALUATION_STATUS_OK) { | |
660 | ERR("[rotation-thread] Failed to get evaluation"); | |
661 | ret = -1; | |
662 | goto end; | |
663 | } | |
664 | ||
665 | session_lock_list(); | |
666 | session = session_find_by_name(condition_session_name); | |
667 | if (!session) { | |
c19ea82d | 668 | DBG("[rotation-thread] Failed to find session while handling notification: session name = `%s`", |
90936dcf | 669 | condition_session_name); |
c19ea82d JG |
670 | /* |
671 | * Not a fatal error: a session can be destroyed before we get | |
672 | * the chance to handle the notification. | |
673 | */ | |
674 | ret = 0; | |
675 | session_unlock_list(); | |
90936dcf JD |
676 | goto end; |
677 | } | |
678 | session_lock(session); | |
90936dcf | 679 | |
69d5d2ae JG |
680 | if (!lttng_trigger_is_equal(session->rotate_trigger, |
681 | lttng_notification_get_const_trigger(notification))) { | |
682 | /* Notification does not originate from our rotation trigger. */ | |
683 | ret = 0; | |
684 | goto end_unlock; | |
685 | } | |
686 | ||
90936dcf JD |
687 | ret = unsubscribe_session_consumed_size_rotation(session, |
688 | notification_thread_handle); | |
689 | if (ret) { | |
490b3229 | 690 | goto end_unlock; |
90936dcf JD |
691 | } |
692 | ||
343defc2 MD |
693 | ret = cmd_rotate_session(session, NULL, false, |
694 | LTTNG_TRACE_CHUNK_COMMAND_TYPE_MOVE_TO_COMPLETED); | |
90936dcf JD |
695 | if (ret == -LTTNG_ERR_ROTATION_PENDING) { |
696 | DBG("Rotate already pending, subscribe to the next threshold value"); | |
90936dcf JD |
697 | } else if (ret != LTTNG_OK) { |
698 | ERR("[rotation-thread] Failed to rotate on size notification with error: %s", | |
699 | lttng_strerror(ret)); | |
700 | ret = -1; | |
701 | goto end_unlock; | |
702 | } | |
703 | ret = subscribe_session_consumed_size_rotation(session, | |
704 | consumed + session->rotate_size, | |
705 | notification_thread_handle); | |
706 | if (ret) { | |
707 | ERR("[rotation-thread] Failed to subscribe to session consumed size condition"); | |
708 | goto end_unlock; | |
709 | } | |
710 | ret = 0; | |
711 | ||
712 | end_unlock: | |
713 | session_unlock(session); | |
e32d7f27 | 714 | session_put(session); |
5b8139c6 | 715 | session_unlock_list(); |
90936dcf JD |
716 | end: |
717 | return ret; | |
718 | } | |
719 | ||
720 | static | |
92816cc3 | 721 | int handle_notification_channel(int fd, |
90936dcf | 722 | struct rotation_thread_handle *handle, |
92816cc3 | 723 | struct rotation_thread *state) |
90936dcf JD |
724 | { |
725 | int ret; | |
d73ee93f JG |
726 | bool notification_pending; |
727 | struct lttng_notification *notification = NULL; | |
90936dcf | 728 | enum lttng_notification_channel_status status; |
90936dcf | 729 | |
d73ee93f JG |
730 | status = lttng_notification_channel_has_pending_notification( |
731 | rotate_notification_channel, ¬ification_pending); | |
732 | if (status != LTTNG_NOTIFICATION_CHANNEL_STATUS_OK) { | |
a9577b76 | 733 | ERR("[rotation-thread ]Error occurred while checking for pending notification"); |
d73ee93f JG |
734 | ret = -1; |
735 | goto end; | |
736 | } | |
737 | ||
738 | if (!notification_pending) { | |
739 | ret = 0; | |
740 | goto end; | |
741 | } | |
742 | ||
90936dcf JD |
743 | /* Receive the next notification. */ |
744 | status = lttng_notification_channel_get_next_notification( | |
745 | rotate_notification_channel, | |
746 | ¬ification); | |
747 | ||
748 | switch (status) { | |
749 | case LTTNG_NOTIFICATION_CHANNEL_STATUS_OK: | |
750 | break; | |
751 | case LTTNG_NOTIFICATION_CHANNEL_STATUS_NOTIFICATIONS_DROPPED: | |
752 | /* Not an error, we will wait for the next one */ | |
753 | ret = 0; | |
754 | goto end;; | |
755 | case LTTNG_NOTIFICATION_CHANNEL_STATUS_CLOSED: | |
756 | ERR("Notification channel was closed"); | |
757 | ret = -1; | |
758 | goto end; | |
759 | default: | |
760 | /* Unhandled conditions / errors. */ | |
761 | ERR("Unknown notification channel status"); | |
762 | ret = -1; | |
763 | goto end; | |
764 | } | |
765 | ||
69d5d2ae | 766 | ret = handle_condition(notification, |
90936dcf JD |
767 | handle->notification_thread_handle); |
768 | ||
769 | end: | |
770 | lttng_notification_destroy(notification); | |
90936dcf JD |
771 | return ret; |
772 | } | |
773 | ||
2c0c9bbc | 774 | static |
db66e574 JD |
775 | void *thread_rotation(void *data) |
776 | { | |
777 | int ret; | |
778 | struct rotation_thread_handle *handle = data; | |
92816cc3 | 779 | struct rotation_thread thread; |
87380d40 | 780 | int queue_pipe_fd; |
db66e574 JD |
781 | |
782 | DBG("[rotation-thread] Started rotation thread"); | |
f620cc28 JG |
783 | rcu_register_thread(); |
784 | rcu_thread_online(); | |
785 | health_register(health_sessiond, HEALTH_SESSIOND_TYPE_ROTATION); | |
786 | health_code_update(); | |
db66e574 JD |
787 | |
788 | if (!handle) { | |
789 | ERR("[rotation-thread] Invalid thread context provided"); | |
790 | goto end; | |
791 | } | |
792 | ||
87380d40 JR |
793 | queue_pipe_fd = lttng_pipe_get_readfd( |
794 | handle->rotation_timer_queue->event_pipe); | |
795 | ||
db66e574 | 796 | |
92816cc3 | 797 | ret = init_thread_state(handle, &thread); |
db66e574 | 798 | if (ret) { |
f5f8e5cd | 799 | goto error; |
db66e574 JD |
800 | } |
801 | ||
db66e574 JD |
802 | while (true) { |
803 | int fd_count, i; | |
804 | ||
805 | health_poll_entry(); | |
806 | DBG("[rotation-thread] Entering poll wait"); | |
92816cc3 | 807 | ret = lttng_poll_wait(&thread.events, -1); |
db66e574 JD |
808 | DBG("[rotation-thread] Poll wait returned (%i)", ret); |
809 | health_poll_exit(); | |
810 | if (ret < 0) { | |
811 | /* | |
812 | * Restart interrupted system call. | |
813 | */ | |
814 | if (errno == EINTR) { | |
815 | continue; | |
816 | } | |
817 | ERR("[rotation-thread] Error encountered during lttng_poll_wait (%i)", ret); | |
818 | goto error; | |
819 | } | |
820 | ||
821 | fd_count = ret; | |
822 | for (i = 0; i < fd_count; i++) { | |
92816cc3 JG |
823 | int fd = LTTNG_POLL_GETFD(&thread.events, i); |
824 | uint32_t revents = LTTNG_POLL_GETEV(&thread.events, i); | |
db66e574 JD |
825 | |
826 | DBG("[rotation-thread] Handling fd (%i) activity (%u)", | |
827 | fd, revents); | |
828 | ||
92816cc3 JG |
829 | if (revents & LPOLLERR) { |
830 | ERR("[rotation-thread] Polling returned an error on fd %i", fd); | |
831 | goto error; | |
832 | } | |
833 | ||
85e17b27 JG |
834 | if (fd == rotate_notification_channel->socket) { |
835 | ret = handle_notification_channel(fd, handle, | |
836 | &thread); | |
837 | if (ret) { | |
838 | ERR("[rotation-thread] Error occurred while handling activity on notification channel socket"); | |
839 | goto error; | |
840 | } | |
841 | } else { | |
842 | /* Job queue or quit pipe activity. */ | |
85e17b27 JG |
843 | |
844 | /* | |
845 | * The job queue is serviced if there is | |
846 | * activity on the quit pipe to ensure it is | |
847 | * flushed and all references held in the queue | |
848 | * are released. | |
849 | */ | |
92816cc3 JG |
850 | ret = handle_job_queue(handle, &thread, |
851 | handle->rotation_timer_queue); | |
d88744a4 JD |
852 | if (ret) { |
853 | ERR("[rotation-thread] Failed to handle rotation timer pipe event"); | |
854 | goto error; | |
855 | } | |
85e17b27 | 856 | |
64d9b072 JG |
857 | if (fd == queue_pipe_fd) { |
858 | char buf; | |
859 | ||
860 | ret = lttng_read(fd, &buf, 1); | |
861 | if (ret != 1) { | |
862 | ERR("[rotation-thread] Failed to read from wakeup pipe (fd = %i)", fd); | |
64d9b072 JG |
863 | goto error; |
864 | } | |
865 | } else { | |
85e17b27 JG |
866 | DBG("[rotation-thread] Quit pipe activity"); |
867 | goto exit; | |
90936dcf | 868 | } |
db66e574 JD |
869 | } |
870 | } | |
871 | } | |
872 | exit: | |
873 | error: | |
874 | DBG("[rotation-thread] Exit"); | |
92816cc3 | 875 | fini_thread_state(&thread); |
f620cc28 | 876 | end: |
db66e574 | 877 | health_unregister(health_sessiond); |
03732c32 JG |
878 | rcu_thread_offline(); |
879 | rcu_unregister_thread(); | |
db66e574 JD |
880 | return NULL; |
881 | } | |
64d9b072 JG |
882 | |
883 | static | |
884 | bool shutdown_rotation_thread(void *thread_data) | |
885 | { | |
886 | struct rotation_thread_handle *handle = thread_data; | |
887 | const int write_fd = lttng_pipe_get_writefd(handle->quit_pipe); | |
888 | ||
889 | return notify_thread_pipe(write_fd) == 1; | |
890 | } | |
891 | ||
892 | bool launch_rotation_thread(struct rotation_thread_handle *handle) | |
893 | { | |
894 | struct lttng_thread *thread; | |
895 | ||
896 | thread = lttng_thread_create("Rotation", | |
897 | thread_rotation, | |
898 | shutdown_rotation_thread, | |
899 | NULL, | |
900 | handle); | |
901 | if (!thread) { | |
902 | goto error; | |
903 | } | |
904 | lttng_thread_put(thread); | |
905 | return true; | |
906 | error: | |
907 | return false; | |
908 | } |