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> | |
c9e313bc SM |
11 | #include <common/error.hpp> |
12 | #include <common/config/session-config.hpp> | |
13 | #include <common/defaults.hpp> | |
14 | #include <common/utils.hpp> | |
15 | #include <common/futex.hpp> | |
16 | #include <common/align.hpp> | |
17 | #include <common/time.hpp> | |
18 | #include <common/hashtable/utils.hpp> | |
db66e574 JD |
19 | #include <sys/stat.h> |
20 | #include <time.h> | |
21 | #include <signal.h> | |
22 | #include <inttypes.h> | |
23 | ||
c9e313bc SM |
24 | #include <common/kernel-ctl/kernel-ctl.hpp> |
25 | #include <lttng/notification/channel-internal.hpp> | |
26 | #include <lttng/rotate-internal.hpp> | |
27 | #include <lttng/location-internal.hpp> | |
28 | #include <lttng/condition/condition-internal.hpp> | |
c08136a3 | 29 | #include <lttng/notification/notification-internal.hpp> |
db66e574 | 30 | |
c9e313bc SM |
31 | #include "rotation-thread.hpp" |
32 | #include "lttng-sessiond.hpp" | |
33 | #include "health-sessiond.hpp" | |
34 | #include "rotate.hpp" | |
35 | #include "cmd.hpp" | |
36 | #include "session.hpp" | |
37 | #include "timer.hpp" | |
38 | #include "notification-thread-commands.hpp" | |
39 | #include "utils.hpp" | |
40 | #include "thread.hpp" | |
db66e574 JD |
41 | |
42 | #include <urcu.h> | |
43 | #include <urcu/list.h> | |
db66e574 | 44 | |
90936dcf JD |
45 | struct lttng_notification_channel *rotate_notification_channel = NULL; |
46 | ||
92816cc3 | 47 | struct rotation_thread { |
db66e574 JD |
48 | struct lttng_poll_event events; |
49 | }; | |
50 | ||
92816cc3 JG |
51 | /* |
52 | * The timer thread enqueues jobs and wakes up the rotation thread. | |
53 | * When the rotation thread wakes up, it empties the queue. | |
54 | */ | |
55 | struct rotation_thread_timer_queue { | |
56 | struct lttng_pipe *event_pipe; | |
57 | struct cds_list_head list; | |
58 | pthread_mutex_t lock; | |
59 | }; | |
60 | ||
61 | struct rotation_thread_handle { | |
92816cc3 JG |
62 | struct rotation_thread_timer_queue *rotation_timer_queue; |
63 | /* Access to the notification thread cmd_queue */ | |
64 | struct notification_thread_handle *notification_thread_handle; | |
64d9b072 JG |
65 | /* Thread-specific quit pipe. */ |
66 | struct lttng_pipe *quit_pipe; | |
92816cc3 JG |
67 | }; |
68 | ||
f1494934 JG |
69 | namespace { |
70 | struct rotation_thread_job { | |
71 | enum rotation_thread_job_type type; | |
72 | struct ltt_session *session; | |
73 | /* List member in struct rotation_thread_timer_queue. */ | |
74 | struct cds_list_head head; | |
75 | }; | |
76 | } /* namespace */ | |
77 | ||
db66e574 | 78 | static |
92816cc3 | 79 | const char *get_job_type_str(enum rotation_thread_job_type job_type) |
db66e574 | 80 | { |
92816cc3 JG |
81 | switch (job_type) { |
82 | case ROTATION_THREAD_JOB_TYPE_CHECK_PENDING_ROTATION: | |
83 | return "CHECK_PENDING_ROTATION"; | |
84 | case ROTATION_THREAD_JOB_TYPE_SCHEDULED_ROTATION: | |
85 | return "SCHEDULED_ROTATION"; | |
86 | default: | |
87 | abort(); | |
88 | } | |
db66e574 JD |
89 | } |
90 | ||
92816cc3 | 91 | struct rotation_thread_timer_queue *rotation_thread_timer_queue_create(void) |
db66e574 | 92 | { |
92816cc3 | 93 | struct rotation_thread_timer_queue *queue = NULL; |
db66e574 | 94 | |
64803277 | 95 | queue = zmalloc<rotation_thread_timer_queue>(); |
92816cc3 JG |
96 | if (!queue) { |
97 | PERROR("Failed to allocate timer rotate queue"); | |
98 | goto end; | |
99 | } | |
db66e574 | 100 | |
92816cc3 JG |
101 | queue->event_pipe = lttng_pipe_open(FD_CLOEXEC | O_NONBLOCK); |
102 | CDS_INIT_LIST_HEAD(&queue->list); | |
103 | pthread_mutex_init(&queue->lock, NULL); | |
104 | end: | |
105 | return queue; | |
db66e574 JD |
106 | } |
107 | ||
92816cc3 JG |
108 | void rotation_thread_timer_queue_destroy( |
109 | struct rotation_thread_timer_queue *queue) | |
db66e574 | 110 | { |
92816cc3 JG |
111 | if (!queue) { |
112 | return; | |
db66e574 JD |
113 | } |
114 | ||
92816cc3 JG |
115 | lttng_pipe_destroy(queue->event_pipe); |
116 | ||
117 | pthread_mutex_lock(&queue->lock); | |
a0377dfe | 118 | LTTNG_ASSERT(cds_list_empty(&queue->list)); |
92816cc3 JG |
119 | pthread_mutex_unlock(&queue->lock); |
120 | pthread_mutex_destroy(&queue->lock); | |
121 | free(queue); | |
122 | } | |
db66e574 | 123 | |
92816cc3 JG |
124 | /* |
125 | * Destroy the thread data previously created by the init function. | |
126 | */ | |
127 | void rotation_thread_handle_destroy( | |
128 | struct rotation_thread_handle *handle) | |
129 | { | |
64d9b072 | 130 | lttng_pipe_destroy(handle->quit_pipe); |
db66e574 JD |
131 | free(handle); |
132 | } | |
133 | ||
134 | struct rotation_thread_handle *rotation_thread_handle_create( | |
90936dcf | 135 | struct rotation_thread_timer_queue *rotation_timer_queue, |
c8a9de5a | 136 | struct notification_thread_handle *notification_thread_handle) |
db66e574 JD |
137 | { |
138 | struct rotation_thread_handle *handle; | |
139 | ||
64803277 | 140 | handle = zmalloc<rotation_thread_handle>(); |
db66e574 JD |
141 | if (!handle) { |
142 | goto end; | |
143 | } | |
144 | ||
92816cc3 JG |
145 | handle->rotation_timer_queue = rotation_timer_queue; |
146 | handle->notification_thread_handle = notification_thread_handle; | |
64d9b072 JG |
147 | handle->quit_pipe = lttng_pipe_open(FD_CLOEXEC); |
148 | if (!handle->quit_pipe) { | |
149 | goto error; | |
150 | } | |
92816cc3 JG |
151 | |
152 | end: | |
153 | return handle; | |
64d9b072 JG |
154 | error: |
155 | rotation_thread_handle_destroy(handle); | |
156 | return NULL; | |
92816cc3 JG |
157 | } |
158 | ||
159 | /* | |
160 | * Called with the rotation_thread_timer_queue lock held. | |
161 | * Return true if the same timer job already exists in the queue, false if not. | |
162 | */ | |
163 | static | |
164 | bool timer_job_exists(const struct rotation_thread_timer_queue *queue, | |
c7031a2c JG |
165 | enum rotation_thread_job_type job_type, |
166 | struct ltt_session *session) | |
92816cc3 JG |
167 | { |
168 | bool exists = false; | |
169 | struct rotation_thread_job *job; | |
170 | ||
171 | cds_list_for_each_entry(job, &queue->list, head) { | |
c7031a2c | 172 | if (job->session == session && job->type == job_type) { |
92816cc3 JG |
173 | exists = true; |
174 | goto end; | |
db66e574 | 175 | } |
db66e574 | 176 | } |
92816cc3 JG |
177 | end: |
178 | return exists; | |
179 | } | |
180 | ||
181 | void rotation_thread_enqueue_job(struct rotation_thread_timer_queue *queue, | |
c7031a2c JG |
182 | enum rotation_thread_job_type job_type, |
183 | struct ltt_session *session) | |
92816cc3 JG |
184 | { |
185 | int ret; | |
be2956e7 | 186 | const char dummy = '!'; |
92816cc3 JG |
187 | struct rotation_thread_job *job = NULL; |
188 | const char *job_type_str = get_job_type_str(job_type); | |
189 | ||
190 | pthread_mutex_lock(&queue->lock); | |
c7031a2c | 191 | if (timer_job_exists(queue, job_type, session)) { |
92816cc3 JG |
192 | /* |
193 | * This timer job is already pending, we don't need to add | |
194 | * it. | |
195 | */ | |
196 | goto end; | |
db66e574 | 197 | } |
92816cc3 | 198 | |
64803277 | 199 | job = zmalloc<rotation_thread_job>(); |
92816cc3 | 200 | if (!job) { |
c7031a2c JG |
201 | PERROR("Failed to allocate rotation thread job of type \"%s\" for session \"%s\"", |
202 | job_type_str, session->name); | |
92816cc3 JG |
203 | goto end; |
204 | } | |
c7031a2c JG |
205 | /* No reason for this to fail as the caller must hold a reference. */ |
206 | (void) session_get(session); | |
207 | ||
208 | job->session = session; | |
92816cc3 | 209 | job->type = job_type; |
92816cc3 JG |
210 | cds_list_add_tail(&job->head, &queue->list); |
211 | ||
be2956e7 JG |
212 | ret = lttng_write(lttng_pipe_get_writefd(queue->event_pipe), &dummy, |
213 | sizeof(dummy)); | |
92816cc3 JG |
214 | if (ret < 0) { |
215 | /* | |
216 | * We do not want to block in the timer handler, the job has | |
217 | * been enqueued in the list, the wakeup pipe is probably full, | |
218 | * the job will be processed when the rotation_thread catches | |
219 | * up. | |
220 | */ | |
942003e5 MJ |
221 | DIAGNOSTIC_PUSH |
222 | DIAGNOSTIC_IGNORE_LOGICAL_OP | |
92816cc3 | 223 | if (errno == EAGAIN || errno == EWOULDBLOCK) { |
942003e5 | 224 | DIAGNOSTIC_POP |
92816cc3 JG |
225 | /* |
226 | * Not an error, but would be surprising and indicate | |
227 | * that the rotation thread can't keep up with the | |
228 | * current load. | |
229 | */ | |
230 | DBG("Wake-up pipe of rotation thread job queue is full"); | |
231 | goto end; | |
db66e574 | 232 | } |
c7031a2c JG |
233 | PERROR("Failed to wake-up the rotation thread after pushing a job of type \"%s\" for session \"%s\"", |
234 | job_type_str, session->name); | |
92816cc3 | 235 | goto end; |
db66e574 | 236 | } |
db66e574 JD |
237 | |
238 | end: | |
92816cc3 | 239 | pthread_mutex_unlock(&queue->lock); |
db66e574 JD |
240 | } |
241 | ||
242 | static | |
243 | int init_poll_set(struct lttng_poll_event *poll_set, | |
244 | struct rotation_thread_handle *handle) | |
245 | { | |
246 | int ret; | |
247 | ||
248 | /* | |
64d9b072 JG |
249 | * Create pollset with size 3: |
250 | * - rotation thread quit pipe, | |
92816cc3 | 251 | * - rotation thread timer queue pipe, |
64d9b072 | 252 | * - notification channel sock, |
db66e574 | 253 | */ |
64d9b072 JG |
254 | ret = lttng_poll_create(poll_set, 5, LTTNG_CLOEXEC); |
255 | if (ret < 0) { | |
db66e574 JD |
256 | goto error; |
257 | } | |
64d9b072 JG |
258 | |
259 | ret = lttng_poll_add(poll_set, | |
260 | lttng_pipe_get_readfd(handle->quit_pipe), | |
261 | LPOLLIN | LPOLLERR); | |
262 | if (ret < 0) { | |
bd0514a5 | 263 | ERR("Failed to add quit pipe read fd to poll set"); |
64d9b072 JG |
264 | goto error; |
265 | } | |
266 | ||
d086f507 JD |
267 | ret = lttng_poll_add(poll_set, |
268 | lttng_pipe_get_readfd(handle->rotation_timer_queue->event_pipe), | |
269 | LPOLLIN | LPOLLERR); | |
270 | if (ret < 0) { | |
bd0514a5 | 271 | ERR("Failed to add rotate_pending fd to poll set"); |
d086f507 JD |
272 | goto error; |
273 | } | |
db66e574 | 274 | |
db66e574 JD |
275 | return ret; |
276 | error: | |
277 | lttng_poll_clean(poll_set); | |
278 | return ret; | |
279 | } | |
280 | ||
281 | static | |
92816cc3 | 282 | void fini_thread_state(struct rotation_thread *state) |
db66e574 JD |
283 | { |
284 | lttng_poll_clean(&state->events); | |
90936dcf JD |
285 | if (rotate_notification_channel) { |
286 | lttng_notification_channel_destroy(rotate_notification_channel); | |
287 | } | |
db66e574 JD |
288 | } |
289 | ||
290 | static | |
291 | int init_thread_state(struct rotation_thread_handle *handle, | |
92816cc3 | 292 | struct rotation_thread *state) |
db66e574 JD |
293 | { |
294 | int ret; | |
295 | ||
296 | memset(state, 0, sizeof(*state)); | |
297 | lttng_poll_init(&state->events); | |
298 | ||
299 | ret = init_poll_set(&state->events, handle); | |
300 | if (ret) { | |
bd0514a5 | 301 | ERR("Failed to initialize rotation thread poll set"); |
db66e574 JD |
302 | goto end; |
303 | } | |
304 | ||
90936dcf JD |
305 | rotate_notification_channel = lttng_notification_channel_create( |
306 | lttng_session_daemon_notification_endpoint); | |
307 | if (!rotate_notification_channel) { | |
bd0514a5 | 308 | ERR("Could not create notification channel"); |
90936dcf JD |
309 | ret = -1; |
310 | goto end; | |
311 | } | |
312 | ret = lttng_poll_add(&state->events, rotate_notification_channel->socket, | |
313 | LPOLLIN | LPOLLERR); | |
314 | if (ret < 0) { | |
bd0514a5 | 315 | ERR("Failed to add notification fd to pollset"); |
90936dcf JD |
316 | goto end; |
317 | } | |
318 | ||
db66e574 JD |
319 | end: |
320 | return ret; | |
321 | } | |
322 | ||
323 | static | |
d2956687 JG |
324 | void check_session_rotation_pending_on_consumers(struct ltt_session *session, |
325 | bool *_rotation_completed) | |
92816cc3 | 326 | { |
db582e11 | 327 | int ret = 0; |
92816cc3 JG |
328 | struct consumer_socket *socket; |
329 | struct cds_lfht_iter iter; | |
d2956687 JG |
330 | enum consumer_trace_chunk_exists_status exists_status; |
331 | uint64_t relayd_id; | |
332 | bool chunk_exists_on_peer = false; | |
333 | enum lttng_trace_chunk_status chunk_status; | |
334 | ||
a0377dfe | 335 | LTTNG_ASSERT(session->chunk_being_archived); |
92816cc3 JG |
336 | |
337 | /* | |
338 | * Check for a local pending rotation on all consumers (32-bit | |
339 | * user space, 64-bit user space, and kernel). | |
340 | */ | |
92816cc3 JG |
341 | rcu_read_lock(); |
342 | if (!session->ust_session) { | |
343 | goto skip_ust; | |
344 | } | |
345 | cds_lfht_for_each_entry(session->ust_session->consumer->socks->ht, | |
346 | &iter, socket, node.node) { | |
d2956687 JG |
347 | relayd_id = session->ust_session->consumer->type == CONSUMER_DST_LOCAL ? |
348 | -1ULL : | |
349 | session->ust_session->consumer->net_seq_index; | |
350 | ||
351 | pthread_mutex_lock(socket->lock); | |
352 | ret = consumer_trace_chunk_exists(socket, | |
353 | relayd_id, | |
354 | session->id, session->chunk_being_archived, | |
355 | &exists_status); | |
356 | if (ret) { | |
357 | pthread_mutex_unlock(socket->lock); | |
83ed9e90 | 358 | ERR("Error occurred while checking rotation status on consumer daemon"); |
92816cc3 | 359 | goto end; |
db66e574 | 360 | } |
d2956687 | 361 | |
16100d7a | 362 | if (exists_status != CONSUMER_TRACE_CHUNK_EXISTS_STATUS_UNKNOWN_CHUNK) { |
d2956687 JG |
363 | pthread_mutex_unlock(socket->lock); |
364 | chunk_exists_on_peer = true; | |
365 | goto end; | |
16100d7a | 366 | } |
d2956687 | 367 | pthread_mutex_unlock(socket->lock); |
16100d7a | 368 | } |
db66e574 | 369 | |
92816cc3 JG |
370 | skip_ust: |
371 | if (!session->kernel_session) { | |
372 | goto skip_kernel; | |
db66e574 | 373 | } |
92816cc3 JG |
374 | cds_lfht_for_each_entry(session->kernel_session->consumer->socks->ht, |
375 | &iter, socket, node.node) { | |
d2956687 JG |
376 | pthread_mutex_lock(socket->lock); |
377 | relayd_id = session->kernel_session->consumer->type == CONSUMER_DST_LOCAL ? | |
378 | -1ULL : | |
379 | session->kernel_session->consumer->net_seq_index; | |
380 | ||
381 | ret = consumer_trace_chunk_exists(socket, | |
382 | relayd_id, | |
383 | session->id, session->chunk_being_archived, | |
384 | &exists_status); | |
385 | if (ret) { | |
386 | pthread_mutex_unlock(socket->lock); | |
83ed9e90 | 387 | ERR("Error occurred while checking rotation status on consumer daemon"); |
92816cc3 JG |
388 | goto end; |
389 | } | |
d2956687 | 390 | |
16100d7a | 391 | if (exists_status != CONSUMER_TRACE_CHUNK_EXISTS_STATUS_UNKNOWN_CHUNK) { |
d2956687 JG |
392 | pthread_mutex_unlock(socket->lock); |
393 | chunk_exists_on_peer = true; | |
394 | goto end; | |
16100d7a | 395 | } |
d2956687 | 396 | pthread_mutex_unlock(socket->lock); |
92816cc3 JG |
397 | } |
398 | skip_kernel: | |
399 | end: | |
400 | rcu_read_unlock(); | |
db66e574 | 401 | |
d2956687 JG |
402 | if (!chunk_exists_on_peer) { |
403 | uint64_t chunk_being_archived_id; | |
404 | ||
405 | chunk_status = lttng_trace_chunk_get_id( | |
406 | session->chunk_being_archived, | |
407 | &chunk_being_archived_id); | |
a0377dfe | 408 | LTTNG_ASSERT(chunk_status == LTTNG_TRACE_CHUNK_STATUS_OK); |
bd0514a5 | 409 | DBG("Rotation of trace archive %" PRIu64 " of session \"%s\" is complete on all consumers", |
d2956687 | 410 | chunk_being_archived_id, |
92816cc3 | 411 | session->name); |
db66e574 | 412 | } |
d2956687 | 413 | *_rotation_completed = !chunk_exists_on_peer; |
92816cc3 | 414 | if (ret) { |
2961f09e JG |
415 | ret = session_reset_rotation_state(session, |
416 | LTTNG_ROTATION_STATE_ERROR); | |
417 | if (ret) { | |
418 | ERR("Failed to reset rotation state of session \"%s\"", | |
419 | session->name); | |
420 | } | |
db66e574 | 421 | } |
db66e574 JD |
422 | } |
423 | ||
d88744a4 | 424 | /* |
92816cc3 | 425 | * Check if the last rotation was completed, called with session lock held. |
d2956687 JG |
426 | * Should only return non-zero in the event of a fatal error. Doing so will |
427 | * shutdown the thread. | |
d88744a4 JD |
428 | */ |
429 | static | |
92816cc3 JG |
430 | int check_session_rotation_pending(struct ltt_session *session, |
431 | struct notification_thread_handle *notification_thread_handle) | |
d88744a4 JD |
432 | { |
433 | int ret; | |
92816cc3 | 434 | struct lttng_trace_archive_location *location; |
d2956687 JG |
435 | enum lttng_trace_chunk_status chunk_status; |
436 | bool rotation_completed = false; | |
437 | const char *archived_chunk_name; | |
438 | uint64_t chunk_being_archived_id; | |
439 | ||
dc1d5967 FD |
440 | if (!session->chunk_being_archived) { |
441 | ret = 0; | |
442 | goto end; | |
443 | } | |
444 | ||
d2956687 JG |
445 | chunk_status = lttng_trace_chunk_get_id(session->chunk_being_archived, |
446 | &chunk_being_archived_id); | |
a0377dfe | 447 | LTTNG_ASSERT(chunk_status == LTTNG_TRACE_CHUNK_STATUS_OK); |
d88744a4 | 448 | |
bd0514a5 | 449 | DBG("Checking for pending rotation on session \"%s\", trace archive %" PRIu64, |
d2956687 JG |
450 | session->name, chunk_being_archived_id); |
451 | ||
faf1bdcf JG |
452 | /* |
453 | * The rotation-pending check timer of a session is launched in | |
454 | * one-shot mode. If the rotation is incomplete, the rotation | |
455 | * thread will re-enable the pending-check timer. | |
456 | * | |
457 | * The timer thread can't stop the timer itself since it is involved | |
458 | * in the check for the timer's quiescence. | |
459 | */ | |
460 | ret = timer_session_rotation_pending_check_stop(session); | |
461 | if (ret) { | |
6ae1bf46 | 462 | goto check_ongoing_rotation; |
faf1bdcf JG |
463 | } |
464 | ||
d2956687 JG |
465 | check_session_rotation_pending_on_consumers(session, |
466 | &rotation_completed); | |
d2956687 JG |
467 | if (!rotation_completed || |
468 | session->rotation_state == LTTNG_ROTATION_STATE_ERROR) { | |
6ae1bf46 | 469 | goto check_ongoing_rotation; |
92816cc3 JG |
470 | } |
471 | ||
92816cc3 JG |
472 | /* |
473 | * Now we can clear the "ONGOING" state in the session. New | |
474 | * rotations can start now. | |
475 | */ | |
d2956687 JG |
476 | chunk_status = lttng_trace_chunk_get_name(session->chunk_being_archived, |
477 | &archived_chunk_name, NULL); | |
a0377dfe | 478 | LTTNG_ASSERT(chunk_status == LTTNG_TRACE_CHUNK_STATUS_OK); |
d2956687 JG |
479 | free(session->last_archived_chunk_name); |
480 | session->last_archived_chunk_name = strdup(archived_chunk_name); | |
481 | if (!session->last_archived_chunk_name) { | |
482 | PERROR("Failed to duplicate archived chunk name"); | |
483 | } | |
484 | session_reset_rotation_state(session, LTTNG_ROTATION_STATE_COMPLETED); | |
92816cc3 | 485 | |
7fdbed1c JG |
486 | if (!session->quiet_rotation) { |
487 | location = session_get_trace_archive_location(session); | |
7fdbed1c JG |
488 | ret = notification_thread_command_session_rotation_completed( |
489 | notification_thread_handle, | |
490 | session->name, | |
491 | session->uid, | |
492 | session->gid, | |
493 | session->last_archived_chunk_id.value, | |
494 | location); | |
d3740619 | 495 | lttng_trace_archive_location_put(location); |
7fdbed1c | 496 | if (ret != LTTNG_OK) { |
bd0514a5 | 497 | ERR("Failed to notify notification thread of completed rotation for session %s", |
7fdbed1c JG |
498 | session->name); |
499 | } | |
92816cc3 JG |
500 | } |
501 | ||
502 | ret = 0; | |
6ae1bf46 | 503 | check_ongoing_rotation: |
92816cc3 | 504 | if (session->rotation_state == LTTNG_ROTATION_STATE_ONGOING) { |
d2956687 JG |
505 | chunk_status = lttng_trace_chunk_get_id( |
506 | session->chunk_being_archived, | |
507 | &chunk_being_archived_id); | |
a0377dfe | 508 | LTTNG_ASSERT(chunk_status == LTTNG_TRACE_CHUNK_STATUS_OK); |
d2956687 | 509 | |
bd0514a5 | 510 | DBG("Rotation of trace archive %" PRIu64 " is still pending for session %s", |
d2956687 | 511 | chunk_being_archived_id, session->name); |
92816cc3 JG |
512 | ret = timer_session_rotation_pending_check_start(session, |
513 | DEFAULT_ROTATE_PENDING_TIMER); | |
514 | if (ret) { | |
d2956687 | 515 | ERR("Failed to re-enable rotation pending timer"); |
92816cc3 JG |
516 | ret = -1; |
517 | goto end; | |
518 | } | |
519 | } | |
520 | ||
6ae1bf46 | 521 | end: |
d88744a4 JD |
522 | return ret; |
523 | } | |
524 | ||
ed1e52a3 | 525 | /* Call with the session and session_list locks held. */ |
259c2674 | 526 | static |
92816cc3 | 527 | int launch_session_rotation(struct ltt_session *session) |
259c2674 JD |
528 | { |
529 | int ret; | |
92816cc3 | 530 | struct lttng_rotate_session_return rotation_return; |
259c2674 | 531 | |
bd0514a5 | 532 | DBG("Launching scheduled time-based rotation on session \"%s\"", |
92816cc3 | 533 | session->name); |
259c2674 | 534 | |
343defc2 MD |
535 | ret = cmd_rotate_session(session, &rotation_return, false, |
536 | LTTNG_TRACE_CHUNK_COMMAND_TYPE_MOVE_TO_COMPLETED); | |
92816cc3 | 537 | if (ret == LTTNG_OK) { |
bd0514a5 | 538 | DBG("Scheduled time-based rotation successfully launched on session \"%s\"", |
92816cc3 JG |
539 | session->name); |
540 | } else { | |
541 | /* Don't consider errors as fatal. */ | |
bd0514a5 | 542 | DBG("Scheduled time-based rotation aborted for session %s: %s", |
92816cc3 | 543 | session->name, lttng_strerror(ret)); |
259c2674 | 544 | } |
92816cc3 JG |
545 | return 0; |
546 | } | |
259c2674 | 547 | |
92816cc3 JG |
548 | static |
549 | int run_job(struct rotation_thread_job *job, struct ltt_session *session, | |
550 | struct notification_thread_handle *notification_thread_handle) | |
551 | { | |
552 | int ret; | |
259c2674 | 553 | |
92816cc3 JG |
554 | switch (job->type) { |
555 | case ROTATION_THREAD_JOB_TYPE_SCHEDULED_ROTATION: | |
16100d7a | 556 | ret = launch_session_rotation(session); |
92816cc3 JG |
557 | break; |
558 | case ROTATION_THREAD_JOB_TYPE_CHECK_PENDING_ROTATION: | |
559 | ret = check_session_rotation_pending(session, | |
560 | notification_thread_handle); | |
561 | break; | |
562 | default: | |
563 | abort(); | |
259c2674 | 564 | } |
259c2674 JD |
565 | return ret; |
566 | } | |
567 | ||
d88744a4 | 568 | static |
92816cc3 | 569 | int handle_job_queue(struct rotation_thread_handle *handle, |
f46376a1 | 570 | struct rotation_thread *state __attribute__((unused)), |
d88744a4 JD |
571 | struct rotation_thread_timer_queue *queue) |
572 | { | |
573 | int ret = 0; | |
d88744a4 JD |
574 | |
575 | for (;;) { | |
e32d7f27 | 576 | struct ltt_session *session; |
92816cc3 | 577 | struct rotation_thread_job *job; |
d88744a4 | 578 | |
92816cc3 | 579 | /* Take the queue lock only to pop an element from the list. */ |
d88744a4 JD |
580 | pthread_mutex_lock(&queue->lock); |
581 | if (cds_list_empty(&queue->list)) { | |
582 | pthread_mutex_unlock(&queue->lock); | |
583 | break; | |
584 | } | |
92816cc3 JG |
585 | job = cds_list_first_entry(&queue->list, |
586 | typeof(*job), head); | |
587 | cds_list_del(&job->head); | |
d88744a4 JD |
588 | pthread_mutex_unlock(&queue->lock); |
589 | ||
d88744a4 | 590 | session_lock_list(); |
c7031a2c | 591 | session = job->session; |
d88744a4 | 592 | if (!session) { |
bd0514a5 | 593 | DBG("Session \"%s\" not found", |
739b8e17 | 594 | session->name != NULL ? session->name : ""); |
d88744a4 | 595 | /* |
92816cc3 JG |
596 | * This is a non-fatal error, and we cannot report it to |
597 | * the user (timer), so just print the error and | |
598 | * continue the processing. | |
599 | * | |
600 | * While the timer thread will purge pending signals for | |
601 | * a session on the session's destruction, it is | |
602 | * possible for a job targeting that session to have | |
603 | * already been queued before it was destroyed. | |
d88744a4 | 604 | */ |
92816cc3 | 605 | free(job); |
e32d7f27 | 606 | session_put(session); |
5b8139c6 | 607 | session_unlock_list(); |
d88744a4 JD |
608 | continue; |
609 | } | |
610 | ||
d88744a4 | 611 | session_lock(session); |
16100d7a | 612 | ret = run_job(job, session, handle->notification_thread_handle); |
d88744a4 | 613 | session_unlock(session); |
faf1bdcf | 614 | /* Release reference held by the job. */ |
e32d7f27 | 615 | session_put(session); |
ed1e52a3 | 616 | session_unlock_list(); |
92816cc3 | 617 | free(job); |
d88744a4 | 618 | if (ret) { |
d88744a4 JD |
619 | goto end; |
620 | } | |
621 | } | |
622 | ||
623 | ret = 0; | |
624 | ||
625 | end: | |
626 | return ret; | |
627 | } | |
628 | ||
92816cc3 | 629 | static |
c08136a3 | 630 | int handle_condition(const struct lttng_notification *notification, |
90936dcf JD |
631 | struct notification_thread_handle *notification_thread_handle) |
632 | { | |
633 | int ret = 0; | |
634 | const char *condition_session_name = NULL; | |
635 | enum lttng_condition_type condition_type; | |
636 | enum lttng_condition_status condition_status; | |
637 | enum lttng_evaluation_status evaluation_status; | |
638 | uint64_t consumed; | |
639 | struct ltt_session *session; | |
c08136a3 JG |
640 | const struct lttng_condition *condition = |
641 | lttng_notification_get_const_condition(notification); | |
642 | const struct lttng_evaluation *evaluation = | |
643 | lttng_notification_get_const_evaluation(notification); | |
90936dcf JD |
644 | |
645 | condition_type = lttng_condition_get_type(condition); | |
646 | ||
647 | if (condition_type != LTTNG_CONDITION_TYPE_SESSION_CONSUMED_SIZE) { | |
648 | ret = -1; | |
bd0514a5 | 649 | ERR("Condition type and session usage type are not the same"); |
90936dcf JD |
650 | goto end; |
651 | } | |
652 | ||
653 | /* Fetch info to test */ | |
654 | condition_status = lttng_condition_session_consumed_size_get_session_name( | |
655 | condition, &condition_session_name); | |
656 | if (condition_status != LTTNG_CONDITION_STATUS_OK) { | |
bd0514a5 | 657 | ERR("Session name could not be fetched"); |
90936dcf JD |
658 | ret = -1; |
659 | goto end; | |
660 | } | |
661 | evaluation_status = lttng_evaluation_session_consumed_size_get_consumed_size(evaluation, | |
662 | &consumed); | |
663 | if (evaluation_status != LTTNG_EVALUATION_STATUS_OK) { | |
bd0514a5 | 664 | ERR("Failed to get evaluation"); |
90936dcf JD |
665 | ret = -1; |
666 | goto end; | |
667 | } | |
668 | ||
669 | session_lock_list(); | |
670 | session = session_find_by_name(condition_session_name); | |
671 | if (!session) { | |
eb2827a4 JG |
672 | DBG("Failed to find session while handling notification: notification type = %s, session name = `%s`", |
673 | lttng_condition_type_str(condition_type), | |
90936dcf | 674 | condition_session_name); |
eb2827a4 JG |
675 | /* |
676 | * Not a fatal error: a session can be destroyed before we get | |
677 | * the chance to handle the notification. | |
678 | */ | |
679 | ret = 0; | |
680 | session_unlock_list(); | |
90936dcf JD |
681 | goto end; |
682 | } | |
683 | session_lock(session); | |
90936dcf | 684 | |
c08136a3 JG |
685 | if (!lttng_trigger_is_equal(session->rotate_trigger, |
686 | lttng_notification_get_const_trigger(notification))) { | |
687 | /* Notification does not originate from our rotation trigger. */ | |
688 | ret = 0; | |
689 | goto end_unlock; | |
690 | } | |
691 | ||
90936dcf JD |
692 | ret = unsubscribe_session_consumed_size_rotation(session, |
693 | notification_thread_handle); | |
694 | if (ret) { | |
490b3229 | 695 | goto end_unlock; |
90936dcf JD |
696 | } |
697 | ||
343defc2 MD |
698 | ret = cmd_rotate_session(session, NULL, false, |
699 | LTTNG_TRACE_CHUNK_COMMAND_TYPE_MOVE_TO_COMPLETED); | |
90936dcf JD |
700 | if (ret == -LTTNG_ERR_ROTATION_PENDING) { |
701 | DBG("Rotate already pending, subscribe to the next threshold value"); | |
90936dcf | 702 | } else if (ret != LTTNG_OK) { |
bd0514a5 | 703 | ERR("Failed to rotate on size notification with error: %s", |
90936dcf JD |
704 | lttng_strerror(ret)); |
705 | ret = -1; | |
706 | goto end_unlock; | |
707 | } | |
708 | ret = subscribe_session_consumed_size_rotation(session, | |
709 | consumed + session->rotate_size, | |
710 | notification_thread_handle); | |
711 | if (ret) { | |
bd0514a5 | 712 | ERR("Failed to subscribe to session consumed size condition"); |
90936dcf JD |
713 | goto end_unlock; |
714 | } | |
715 | ret = 0; | |
716 | ||
717 | end_unlock: | |
718 | session_unlock(session); | |
e32d7f27 | 719 | session_put(session); |
5b8139c6 | 720 | session_unlock_list(); |
90936dcf JD |
721 | end: |
722 | return ret; | |
723 | } | |
724 | ||
725 | static | |
f46376a1 | 726 | int handle_notification_channel(int fd __attribute__((unused)), |
90936dcf | 727 | struct rotation_thread_handle *handle, |
f46376a1 | 728 | struct rotation_thread *state __attribute__((unused))) |
90936dcf JD |
729 | { |
730 | int ret; | |
d73ee93f JG |
731 | bool notification_pending; |
732 | struct lttng_notification *notification = NULL; | |
90936dcf | 733 | enum lttng_notification_channel_status status; |
90936dcf | 734 | |
d73ee93f JG |
735 | status = lttng_notification_channel_has_pending_notification( |
736 | rotate_notification_channel, ¬ification_pending); | |
737 | if (status != LTTNG_NOTIFICATION_CHANNEL_STATUS_OK) { | |
bd0514a5 | 738 | ERR("Error occurred while checking for pending notification"); |
d73ee93f JG |
739 | ret = -1; |
740 | goto end; | |
741 | } | |
742 | ||
743 | if (!notification_pending) { | |
744 | ret = 0; | |
745 | goto end; | |
746 | } | |
747 | ||
90936dcf JD |
748 | /* Receive the next notification. */ |
749 | status = lttng_notification_channel_get_next_notification( | |
750 | rotate_notification_channel, | |
751 | ¬ification); | |
752 | ||
753 | switch (status) { | |
754 | case LTTNG_NOTIFICATION_CHANNEL_STATUS_OK: | |
755 | break; | |
756 | case LTTNG_NOTIFICATION_CHANNEL_STATUS_NOTIFICATIONS_DROPPED: | |
757 | /* Not an error, we will wait for the next one */ | |
758 | ret = 0; | |
759 | goto end;; | |
760 | case LTTNG_NOTIFICATION_CHANNEL_STATUS_CLOSED: | |
761 | ERR("Notification channel was closed"); | |
762 | ret = -1; | |
763 | goto end; | |
764 | default: | |
765 | /* Unhandled conditions / errors. */ | |
766 | ERR("Unknown notification channel status"); | |
767 | ret = -1; | |
768 | goto end; | |
769 | } | |
770 | ||
c08136a3 | 771 | ret = handle_condition(notification, |
90936dcf JD |
772 | handle->notification_thread_handle); |
773 | ||
774 | end: | |
775 | lttng_notification_destroy(notification); | |
90936dcf JD |
776 | return ret; |
777 | } | |
778 | ||
2c0c9bbc | 779 | static |
db66e574 JD |
780 | void *thread_rotation(void *data) |
781 | { | |
782 | int ret; | |
7966af57 | 783 | struct rotation_thread_handle *handle = (rotation_thread_handle *) data; |
92816cc3 | 784 | struct rotation_thread thread; |
87380d40 | 785 | int queue_pipe_fd; |
db66e574 | 786 | |
bd0514a5 | 787 | DBG("Started rotation thread"); |
f620cc28 JG |
788 | rcu_register_thread(); |
789 | rcu_thread_online(); | |
412d7227 | 790 | health_register(the_health_sessiond, HEALTH_SESSIOND_TYPE_ROTATION); |
f620cc28 | 791 | health_code_update(); |
db66e574 JD |
792 | |
793 | if (!handle) { | |
bd0514a5 | 794 | ERR("Invalid thread context provided"); |
db66e574 JD |
795 | goto end; |
796 | } | |
797 | ||
87380d40 JR |
798 | queue_pipe_fd = lttng_pipe_get_readfd( |
799 | handle->rotation_timer_queue->event_pipe); | |
800 | ||
db66e574 | 801 | |
92816cc3 | 802 | ret = init_thread_state(handle, &thread); |
db66e574 | 803 | if (ret) { |
f5f8e5cd | 804 | goto error; |
db66e574 JD |
805 | } |
806 | ||
db66e574 JD |
807 | while (true) { |
808 | int fd_count, i; | |
809 | ||
810 | health_poll_entry(); | |
bd0514a5 | 811 | DBG("Entering poll wait"); |
92816cc3 | 812 | ret = lttng_poll_wait(&thread.events, -1); |
bd0514a5 | 813 | DBG("Poll wait returned (%i)", ret); |
db66e574 JD |
814 | health_poll_exit(); |
815 | if (ret < 0) { | |
816 | /* | |
817 | * Restart interrupted system call. | |
818 | */ | |
819 | if (errno == EINTR) { | |
820 | continue; | |
821 | } | |
bd0514a5 | 822 | ERR("Error encountered during lttng_poll_wait (%i)", ret); |
db66e574 JD |
823 | goto error; |
824 | } | |
825 | ||
826 | fd_count = ret; | |
827 | for (i = 0; i < fd_count; i++) { | |
92816cc3 JG |
828 | int fd = LTTNG_POLL_GETFD(&thread.events, i); |
829 | uint32_t revents = LTTNG_POLL_GETEV(&thread.events, i); | |
db66e574 | 830 | |
bd0514a5 | 831 | DBG("Handling fd (%i) activity (%u)", |
db66e574 JD |
832 | fd, revents); |
833 | ||
92816cc3 | 834 | if (revents & LPOLLERR) { |
bd0514a5 | 835 | ERR("Polling returned an error on fd %i", fd); |
92816cc3 JG |
836 | goto error; |
837 | } | |
838 | ||
85e17b27 JG |
839 | if (fd == rotate_notification_channel->socket) { |
840 | ret = handle_notification_channel(fd, handle, | |
841 | &thread); | |
842 | if (ret) { | |
bd0514a5 | 843 | ERR("Error occurred while handling activity on notification channel socket"); |
85e17b27 JG |
844 | goto error; |
845 | } | |
846 | } else { | |
847 | /* Job queue or quit pipe activity. */ | |
85e17b27 JG |
848 | |
849 | /* | |
850 | * The job queue is serviced if there is | |
851 | * activity on the quit pipe to ensure it is | |
852 | * flushed and all references held in the queue | |
853 | * are released. | |
854 | */ | |
92816cc3 JG |
855 | ret = handle_job_queue(handle, &thread, |
856 | handle->rotation_timer_queue); | |
d88744a4 | 857 | if (ret) { |
bd0514a5 | 858 | ERR("Failed to handle rotation timer pipe event"); |
d88744a4 JD |
859 | goto error; |
860 | } | |
85e17b27 | 861 | |
64d9b072 JG |
862 | if (fd == queue_pipe_fd) { |
863 | char buf; | |
864 | ||
865 | ret = lttng_read(fd, &buf, 1); | |
866 | if (ret != 1) { | |
bd0514a5 | 867 | ERR("Failed to read from wakeup pipe (fd = %i)", fd); |
64d9b072 JG |
868 | goto error; |
869 | } | |
870 | } else { | |
bd0514a5 | 871 | DBG("Quit pipe activity"); |
85e17b27 | 872 | goto exit; |
90936dcf | 873 | } |
db66e574 JD |
874 | } |
875 | } | |
876 | } | |
877 | exit: | |
878 | error: | |
bd0514a5 | 879 | DBG("Thread exit"); |
92816cc3 | 880 | fini_thread_state(&thread); |
f620cc28 | 881 | end: |
412d7227 | 882 | health_unregister(the_health_sessiond); |
03732c32 JG |
883 | rcu_thread_offline(); |
884 | rcu_unregister_thread(); | |
db66e574 JD |
885 | return NULL; |
886 | } | |
64d9b072 JG |
887 | |
888 | static | |
889 | bool shutdown_rotation_thread(void *thread_data) | |
890 | { | |
7966af57 | 891 | struct rotation_thread_handle *handle = (rotation_thread_handle *) thread_data; |
64d9b072 JG |
892 | const int write_fd = lttng_pipe_get_writefd(handle->quit_pipe); |
893 | ||
894 | return notify_thread_pipe(write_fd) == 1; | |
895 | } | |
896 | ||
897 | bool launch_rotation_thread(struct rotation_thread_handle *handle) | |
898 | { | |
899 | struct lttng_thread *thread; | |
900 | ||
901 | thread = lttng_thread_create("Rotation", | |
902 | thread_rotation, | |
903 | shutdown_rotation_thread, | |
904 | NULL, | |
905 | handle); | |
906 | if (!thread) { | |
907 | goto error; | |
908 | } | |
909 | lttng_thread_put(thread); | |
910 | return true; | |
911 | error: | |
912 | return false; | |
913 | } |