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 | |
28ab034a JG |
10 | #include "cmd.hpp" |
11 | #include "health-sessiond.hpp" | |
12 | #include "lttng-sessiond.hpp" | |
13 | #include "notification-thread-commands.hpp" | |
14 | #include "rotate.hpp" | |
15 | #include "rotation-thread.hpp" | |
16 | #include "session.hpp" | |
17 | #include "thread.hpp" | |
18 | #include "timer.hpp" | |
19 | #include "utils.hpp" | |
20 | ||
21 | #include <common/align.hpp> | |
c9e313bc SM |
22 | #include <common/config/session-config.hpp> |
23 | #include <common/defaults.hpp> | |
28ab034a | 24 | #include <common/error.hpp> |
c9e313bc | 25 | #include <common/futex.hpp> |
c9e313bc | 26 | #include <common/hashtable/utils.hpp> |
c9e313bc | 27 | #include <common/kernel-ctl/kernel-ctl.hpp> |
28ab034a JG |
28 | #include <common/time.hpp> |
29 | #include <common/utils.hpp> | |
30 | ||
c9e313bc | 31 | #include <lttng/condition/condition-internal.hpp> |
28ab034a JG |
32 | #include <lttng/location-internal.hpp> |
33 | #include <lttng/notification/channel-internal.hpp> | |
c08136a3 | 34 | #include <lttng/notification/notification-internal.hpp> |
28ab034a JG |
35 | #include <lttng/rotate-internal.hpp> |
36 | #include <lttng/trigger/trigger.h> | |
db66e574 | 37 | |
28ab034a JG |
38 | #include <inttypes.h> |
39 | #include <signal.h> | |
40 | #include <sys/stat.h> | |
41 | #include <time.h> | |
db66e574 JD |
42 | #include <urcu.h> |
43 | #include <urcu/list.h> | |
db66e574 | 44 | |
cd9adb8b | 45 | struct lttng_notification_channel *rotate_notification_channel = nullptr; |
90936dcf | 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 | ||
28ab034a | 78 | static 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 | ||
cd9adb8b | 90 | struct rotation_thread_timer_queue *rotation_thread_timer_queue_create() |
db66e574 | 91 | { |
cd9adb8b | 92 | struct rotation_thread_timer_queue *queue = nullptr; |
db66e574 | 93 | |
64803277 | 94 | queue = zmalloc<rotation_thread_timer_queue>(); |
92816cc3 JG |
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); | |
cd9adb8b | 102 | pthread_mutex_init(&queue->lock, nullptr); |
92816cc3 JG |
103 | end: |
104 | return queue; | |
db66e574 JD |
105 | } |
106 | ||
28ab034a | 107 | void rotation_thread_timer_queue_destroy(struct rotation_thread_timer_queue *queue) |
db66e574 | 108 | { |
92816cc3 JG |
109 | if (!queue) { |
110 | return; | |
db66e574 JD |
111 | } |
112 | ||
92816cc3 JG |
113 | lttng_pipe_destroy(queue->event_pipe); |
114 | ||
115 | pthread_mutex_lock(&queue->lock); | |
a0377dfe | 116 | LTTNG_ASSERT(cds_list_empty(&queue->list)); |
92816cc3 JG |
117 | pthread_mutex_unlock(&queue->lock); |
118 | pthread_mutex_destroy(&queue->lock); | |
119 | free(queue); | |
120 | } | |
db66e574 | 121 | |
92816cc3 JG |
122 | /* |
123 | * Destroy the thread data previously created by the init function. | |
124 | */ | |
28ab034a | 125 | void rotation_thread_handle_destroy(struct rotation_thread_handle *handle) |
92816cc3 | 126 | { |
64d9b072 | 127 | lttng_pipe_destroy(handle->quit_pipe); |
db66e574 JD |
128 | free(handle); |
129 | } | |
130 | ||
28ab034a JG |
131 | struct rotation_thread_handle * |
132 | rotation_thread_handle_create(struct rotation_thread_timer_queue *rotation_timer_queue, | |
133 | struct notification_thread_handle *notification_thread_handle) | |
db66e574 JD |
134 | { |
135 | struct rotation_thread_handle *handle; | |
136 | ||
64803277 | 137 | handle = zmalloc<rotation_thread_handle>(); |
db66e574 JD |
138 | if (!handle) { |
139 | goto end; | |
140 | } | |
141 | ||
92816cc3 JG |
142 | handle->rotation_timer_queue = rotation_timer_queue; |
143 | handle->notification_thread_handle = notification_thread_handle; | |
64d9b072 JG |
144 | handle->quit_pipe = lttng_pipe_open(FD_CLOEXEC); |
145 | if (!handle->quit_pipe) { | |
146 | goto error; | |
147 | } | |
92816cc3 JG |
148 | |
149 | end: | |
150 | return handle; | |
64d9b072 JG |
151 | error: |
152 | rotation_thread_handle_destroy(handle); | |
cd9adb8b | 153 | return nullptr; |
92816cc3 JG |
154 | } |
155 | ||
156 | /* | |
157 | * Called with the rotation_thread_timer_queue lock held. | |
158 | * Return true if the same timer job already exists in the queue, false if not. | |
159 | */ | |
28ab034a JG |
160 | static bool timer_job_exists(const struct rotation_thread_timer_queue *queue, |
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 | ||
28ab034a | 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, | |
28ab034a 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 = '!'; |
cd9adb8b | 183 | struct rotation_thread_job *job = nullptr; |
92816cc3 JG |
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 | 194 | |
64803277 | 195 | job = zmalloc<rotation_thread_job>(); |
92816cc3 | 196 | if (!job) { |
c7031a2c | 197 | PERROR("Failed to allocate rotation thread job of type \"%s\" for session \"%s\"", |
28ab034a JG |
198 | job_type_str, |
199 | session->name); | |
92816cc3 JG |
200 | goto end; |
201 | } | |
c7031a2c JG |
202 | /* No reason for this to fail as the caller must hold a reference. */ |
203 | (void) session_get(session); | |
204 | ||
205 | job->session = session; | |
92816cc3 | 206 | job->type = job_type; |
92816cc3 JG |
207 | cds_list_add_tail(&job->head, &queue->list); |
208 | ||
28ab034a | 209 | ret = lttng_write(lttng_pipe_get_writefd(queue->event_pipe), &dummy, 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 | */ | |
942003e5 MJ |
217 | DIAGNOSTIC_PUSH |
218 | DIAGNOSTIC_IGNORE_LOGICAL_OP | |
92816cc3 | 219 | if (errno == EAGAIN || errno == EWOULDBLOCK) { |
28ab034a | 220 | DIAGNOSTIC_POP |
92816cc3 JG |
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 | 229 | PERROR("Failed to wake-up the rotation thread after pushing a job of type \"%s\" for session \"%s\"", |
28ab034a JG |
230 | job_type_str, |
231 | session->name); | |
92816cc3 | 232 | goto end; |
db66e574 | 233 | } |
db66e574 JD |
234 | |
235 | end: | |
92816cc3 | 236 | pthread_mutex_unlock(&queue->lock); |
db66e574 JD |
237 | } |
238 | ||
28ab034a | 239 | static int init_poll_set(struct lttng_poll_event *poll_set, struct rotation_thread_handle *handle) |
db66e574 JD |
240 | { |
241 | int ret; | |
242 | ||
243 | /* | |
64d9b072 JG |
244 | * Create pollset with size 3: |
245 | * - rotation thread quit pipe, | |
92816cc3 | 246 | * - rotation thread timer queue pipe, |
64d9b072 | 247 | * - notification channel sock, |
db66e574 | 248 | */ |
64d9b072 JG |
249 | ret = lttng_poll_create(poll_set, 5, LTTNG_CLOEXEC); |
250 | if (ret < 0) { | |
db66e574 JD |
251 | goto error; |
252 | } | |
64d9b072 | 253 | |
28ab034a | 254 | ret = lttng_poll_add(poll_set, lttng_pipe_get_readfd(handle->quit_pipe), LPOLLIN); |
64d9b072 | 255 | if (ret < 0) { |
bd0514a5 | 256 | ERR("Failed to add quit pipe read fd to poll set"); |
64d9b072 JG |
257 | goto error; |
258 | } | |
259 | ||
28ab034a JG |
260 | ret = lttng_poll_add( |
261 | poll_set, lttng_pipe_get_readfd(handle->rotation_timer_queue->event_pipe), LPOLLIN); | |
d086f507 | 262 | if (ret < 0) { |
bd0514a5 | 263 | ERR("Failed to add rotate_pending fd to poll set"); |
d086f507 JD |
264 | goto error; |
265 | } | |
db66e574 | 266 | |
db66e574 JD |
267 | return ret; |
268 | error: | |
269 | lttng_poll_clean(poll_set); | |
270 | return ret; | |
271 | } | |
272 | ||
28ab034a | 273 | static void fini_thread_state(struct rotation_thread *state) |
db66e574 JD |
274 | { |
275 | lttng_poll_clean(&state->events); | |
90936dcf JD |
276 | if (rotate_notification_channel) { |
277 | lttng_notification_channel_destroy(rotate_notification_channel); | |
278 | } | |
db66e574 JD |
279 | } |
280 | ||
28ab034a | 281 | static int init_thread_state(struct rotation_thread_handle *handle, struct rotation_thread *state) |
db66e574 JD |
282 | { |
283 | int ret; | |
284 | ||
285 | memset(state, 0, sizeof(*state)); | |
286 | lttng_poll_init(&state->events); | |
287 | ||
288 | ret = init_poll_set(&state->events, handle); | |
289 | if (ret) { | |
bd0514a5 | 290 | ERR("Failed to initialize rotation thread poll set"); |
db66e574 JD |
291 | goto end; |
292 | } | |
293 | ||
28ab034a JG |
294 | rotate_notification_channel = |
295 | lttng_notification_channel_create(lttng_session_daemon_notification_endpoint); | |
90936dcf | 296 | if (!rotate_notification_channel) { |
bd0514a5 | 297 | ERR("Could not create notification channel"); |
90936dcf JD |
298 | ret = -1; |
299 | goto end; | |
300 | } | |
28ab034a | 301 | ret = lttng_poll_add(&state->events, rotate_notification_channel->socket, LPOLLIN); |
90936dcf | 302 | if (ret < 0) { |
bd0514a5 | 303 | ERR("Failed to add notification fd to pollset"); |
90936dcf JD |
304 | goto end; |
305 | } | |
306 | ||
db66e574 JD |
307 | end: |
308 | return ret; | |
309 | } | |
310 | ||
28ab034a JG |
311 | static void check_session_rotation_pending_on_consumers(struct ltt_session *session, |
312 | bool *_rotation_completed) | |
92816cc3 | 313 | { |
db582e11 | 314 | int ret = 0; |
92816cc3 JG |
315 | struct consumer_socket *socket; |
316 | struct cds_lfht_iter iter; | |
d2956687 JG |
317 | enum consumer_trace_chunk_exists_status exists_status; |
318 | uint64_t relayd_id; | |
319 | bool chunk_exists_on_peer = false; | |
320 | enum lttng_trace_chunk_status chunk_status; | |
321 | ||
a0377dfe | 322 | LTTNG_ASSERT(session->chunk_being_archived); |
92816cc3 JG |
323 | |
324 | /* | |
325 | * Check for a local pending rotation on all consumers (32-bit | |
326 | * user space, 64-bit user space, and kernel). | |
327 | */ | |
92816cc3 JG |
328 | rcu_read_lock(); |
329 | if (!session->ust_session) { | |
330 | goto skip_ust; | |
331 | } | |
28ab034a JG |
332 | cds_lfht_for_each_entry ( |
333 | session->ust_session->consumer->socks->ht, &iter, socket, node.node) { | |
d2956687 | 334 | relayd_id = session->ust_session->consumer->type == CONSUMER_DST_LOCAL ? |
28ab034a JG |
335 | -1ULL : |
336 | session->ust_session->consumer->net_seq_index; | |
d2956687 JG |
337 | |
338 | pthread_mutex_lock(socket->lock); | |
339 | ret = consumer_trace_chunk_exists(socket, | |
28ab034a JG |
340 | relayd_id, |
341 | session->id, | |
342 | session->chunk_being_archived, | |
343 | &exists_status); | |
d2956687 JG |
344 | if (ret) { |
345 | pthread_mutex_unlock(socket->lock); | |
83ed9e90 | 346 | ERR("Error occurred while checking rotation status on consumer daemon"); |
92816cc3 | 347 | goto end; |
db66e574 | 348 | } |
d2956687 | 349 | |
16100d7a | 350 | if (exists_status != CONSUMER_TRACE_CHUNK_EXISTS_STATUS_UNKNOWN_CHUNK) { |
d2956687 JG |
351 | pthread_mutex_unlock(socket->lock); |
352 | chunk_exists_on_peer = true; | |
353 | goto end; | |
16100d7a | 354 | } |
d2956687 | 355 | pthread_mutex_unlock(socket->lock); |
16100d7a | 356 | } |
db66e574 | 357 | |
92816cc3 JG |
358 | skip_ust: |
359 | if (!session->kernel_session) { | |
360 | goto skip_kernel; | |
db66e574 | 361 | } |
28ab034a JG |
362 | cds_lfht_for_each_entry ( |
363 | session->kernel_session->consumer->socks->ht, &iter, socket, node.node) { | |
d2956687 JG |
364 | pthread_mutex_lock(socket->lock); |
365 | relayd_id = session->kernel_session->consumer->type == CONSUMER_DST_LOCAL ? | |
28ab034a JG |
366 | -1ULL : |
367 | session->kernel_session->consumer->net_seq_index; | |
d2956687 JG |
368 | |
369 | ret = consumer_trace_chunk_exists(socket, | |
28ab034a JG |
370 | relayd_id, |
371 | session->id, | |
372 | session->chunk_being_archived, | |
373 | &exists_status); | |
d2956687 JG |
374 | if (ret) { |
375 | pthread_mutex_unlock(socket->lock); | |
83ed9e90 | 376 | ERR("Error occurred while checking rotation status on consumer daemon"); |
92816cc3 JG |
377 | goto end; |
378 | } | |
d2956687 | 379 | |
16100d7a | 380 | if (exists_status != CONSUMER_TRACE_CHUNK_EXISTS_STATUS_UNKNOWN_CHUNK) { |
d2956687 JG |
381 | pthread_mutex_unlock(socket->lock); |
382 | chunk_exists_on_peer = true; | |
383 | goto end; | |
16100d7a | 384 | } |
d2956687 | 385 | pthread_mutex_unlock(socket->lock); |
92816cc3 JG |
386 | } |
387 | skip_kernel: | |
388 | end: | |
389 | rcu_read_unlock(); | |
db66e574 | 390 | |
d2956687 JG |
391 | if (!chunk_exists_on_peer) { |
392 | uint64_t chunk_being_archived_id; | |
393 | ||
28ab034a JG |
394 | chunk_status = lttng_trace_chunk_get_id(session->chunk_being_archived, |
395 | &chunk_being_archived_id); | |
a0377dfe | 396 | LTTNG_ASSERT(chunk_status == LTTNG_TRACE_CHUNK_STATUS_OK); |
28ab034a JG |
397 | DBG("Rotation of trace archive %" PRIu64 |
398 | " of session \"%s\" is complete on all consumers", | |
399 | chunk_being_archived_id, | |
400 | session->name); | |
db66e574 | 401 | } |
d2956687 | 402 | *_rotation_completed = !chunk_exists_on_peer; |
92816cc3 | 403 | if (ret) { |
28ab034a | 404 | ret = session_reset_rotation_state(session, LTTNG_ROTATION_STATE_ERROR); |
2961f09e | 405 | if (ret) { |
28ab034a | 406 | ERR("Failed to reset rotation state of session \"%s\"", session->name); |
2961f09e | 407 | } |
db66e574 | 408 | } |
db66e574 JD |
409 | } |
410 | ||
d88744a4 | 411 | /* |
92816cc3 | 412 | * Check if the last rotation was completed, called with session lock held. |
d2956687 JG |
413 | * Should only return non-zero in the event of a fatal error. Doing so will |
414 | * shutdown the thread. | |
d88744a4 | 415 | */ |
28ab034a JG |
416 | static int |
417 | check_session_rotation_pending(struct ltt_session *session, | |
418 | struct notification_thread_handle *notification_thread_handle) | |
d88744a4 JD |
419 | { |
420 | int ret; | |
92816cc3 | 421 | struct lttng_trace_archive_location *location; |
d2956687 JG |
422 | enum lttng_trace_chunk_status chunk_status; |
423 | bool rotation_completed = false; | |
424 | const char *archived_chunk_name; | |
425 | uint64_t chunk_being_archived_id; | |
426 | ||
dc1d5967 FD |
427 | if (!session->chunk_being_archived) { |
428 | ret = 0; | |
429 | goto end; | |
430 | } | |
431 | ||
28ab034a JG |
432 | chunk_status = |
433 | lttng_trace_chunk_get_id(session->chunk_being_archived, &chunk_being_archived_id); | |
a0377dfe | 434 | LTTNG_ASSERT(chunk_status == LTTNG_TRACE_CHUNK_STATUS_OK); |
d88744a4 | 435 | |
bd0514a5 | 436 | DBG("Checking for pending rotation on session \"%s\", trace archive %" PRIu64, |
28ab034a JG |
437 | session->name, |
438 | chunk_being_archived_id); | |
d2956687 | 439 | |
faf1bdcf JG |
440 | /* |
441 | * The rotation-pending check timer of a session is launched in | |
442 | * one-shot mode. If the rotation is incomplete, the rotation | |
443 | * thread will re-enable the pending-check timer. | |
444 | * | |
445 | * The timer thread can't stop the timer itself since it is involved | |
446 | * in the check for the timer's quiescence. | |
447 | */ | |
448 | ret = timer_session_rotation_pending_check_stop(session); | |
449 | if (ret) { | |
6ae1bf46 | 450 | goto check_ongoing_rotation; |
faf1bdcf JG |
451 | } |
452 | ||
28ab034a JG |
453 | check_session_rotation_pending_on_consumers(session, &rotation_completed); |
454 | if (!rotation_completed || session->rotation_state == LTTNG_ROTATION_STATE_ERROR) { | |
6ae1bf46 | 455 | goto check_ongoing_rotation; |
92816cc3 JG |
456 | } |
457 | ||
92816cc3 JG |
458 | /* |
459 | * Now we can clear the "ONGOING" state in the session. New | |
460 | * rotations can start now. | |
461 | */ | |
28ab034a | 462 | chunk_status = lttng_trace_chunk_get_name( |
cd9adb8b | 463 | session->chunk_being_archived, &archived_chunk_name, nullptr); |
a0377dfe | 464 | LTTNG_ASSERT(chunk_status == LTTNG_TRACE_CHUNK_STATUS_OK); |
d2956687 JG |
465 | free(session->last_archived_chunk_name); |
466 | session->last_archived_chunk_name = strdup(archived_chunk_name); | |
467 | if (!session->last_archived_chunk_name) { | |
468 | PERROR("Failed to duplicate archived chunk name"); | |
469 | } | |
470 | session_reset_rotation_state(session, LTTNG_ROTATION_STATE_COMPLETED); | |
92816cc3 | 471 | |
7fdbed1c JG |
472 | if (!session->quiet_rotation) { |
473 | location = session_get_trace_archive_location(session); | |
7fdbed1c | 474 | ret = notification_thread_command_session_rotation_completed( |
28ab034a JG |
475 | notification_thread_handle, |
476 | session->id, | |
477 | session->last_archived_chunk_id.value, | |
478 | location); | |
d3740619 | 479 | lttng_trace_archive_location_put(location); |
7fdbed1c | 480 | if (ret != LTTNG_OK) { |
bd0514a5 | 481 | ERR("Failed to notify notification thread of completed rotation for session %s", |
28ab034a | 482 | session->name); |
7fdbed1c | 483 | } |
92816cc3 JG |
484 | } |
485 | ||
486 | ret = 0; | |
6ae1bf46 | 487 | check_ongoing_rotation: |
92816cc3 | 488 | if (session->rotation_state == LTTNG_ROTATION_STATE_ONGOING) { |
28ab034a JG |
489 | chunk_status = lttng_trace_chunk_get_id(session->chunk_being_archived, |
490 | &chunk_being_archived_id); | |
a0377dfe | 491 | LTTNG_ASSERT(chunk_status == LTTNG_TRACE_CHUNK_STATUS_OK); |
d2956687 | 492 | |
bd0514a5 | 493 | DBG("Rotation of trace archive %" PRIu64 " is still pending for session %s", |
28ab034a JG |
494 | chunk_being_archived_id, |
495 | session->name); | |
92816cc3 | 496 | ret = timer_session_rotation_pending_check_start(session, |
28ab034a | 497 | DEFAULT_ROTATE_PENDING_TIMER); |
92816cc3 | 498 | if (ret) { |
d2956687 | 499 | ERR("Failed to re-enable rotation pending timer"); |
92816cc3 JG |
500 | ret = -1; |
501 | goto end; | |
502 | } | |
503 | } | |
504 | ||
6ae1bf46 | 505 | end: |
d88744a4 JD |
506 | return ret; |
507 | } | |
508 | ||
ed1e52a3 | 509 | /* Call with the session and session_list locks held. */ |
28ab034a | 510 | static int launch_session_rotation(struct ltt_session *session) |
259c2674 JD |
511 | { |
512 | int ret; | |
92816cc3 | 513 | struct lttng_rotate_session_return rotation_return; |
259c2674 | 514 | |
28ab034a | 515 | DBG("Launching scheduled time-based rotation on session \"%s\"", session->name); |
259c2674 | 516 | |
28ab034a JG |
517 | ret = cmd_rotate_session( |
518 | session, &rotation_return, false, LTTNG_TRACE_CHUNK_COMMAND_TYPE_MOVE_TO_COMPLETED); | |
92816cc3 | 519 | if (ret == LTTNG_OK) { |
bd0514a5 | 520 | DBG("Scheduled time-based rotation successfully launched on session \"%s\"", |
28ab034a | 521 | session->name); |
92816cc3 JG |
522 | } else { |
523 | /* Don't consider errors as fatal. */ | |
bd0514a5 | 524 | DBG("Scheduled time-based rotation aborted for session %s: %s", |
28ab034a JG |
525 | session->name, |
526 | lttng_strerror(ret)); | |
259c2674 | 527 | } |
92816cc3 JG |
528 | return 0; |
529 | } | |
259c2674 | 530 | |
28ab034a JG |
531 | static int run_job(struct rotation_thread_job *job, |
532 | struct ltt_session *session, | |
533 | struct notification_thread_handle *notification_thread_handle) | |
92816cc3 JG |
534 | { |
535 | int ret; | |
259c2674 | 536 | |
92816cc3 JG |
537 | switch (job->type) { |
538 | case ROTATION_THREAD_JOB_TYPE_SCHEDULED_ROTATION: | |
16100d7a | 539 | ret = launch_session_rotation(session); |
92816cc3 JG |
540 | break; |
541 | case ROTATION_THREAD_JOB_TYPE_CHECK_PENDING_ROTATION: | |
28ab034a | 542 | ret = check_session_rotation_pending(session, notification_thread_handle); |
92816cc3 JG |
543 | break; |
544 | default: | |
545 | abort(); | |
259c2674 | 546 | } |
259c2674 JD |
547 | return ret; |
548 | } | |
549 | ||
28ab034a JG |
550 | static int handle_job_queue(struct rotation_thread_handle *handle, |
551 | struct rotation_thread *state __attribute__((unused)), | |
552 | struct rotation_thread_timer_queue *queue) | |
d88744a4 JD |
553 | { |
554 | int ret = 0; | |
d88744a4 JD |
555 | |
556 | for (;;) { | |
e32d7f27 | 557 | struct ltt_session *session; |
92816cc3 | 558 | struct rotation_thread_job *job; |
d88744a4 | 559 | |
92816cc3 | 560 | /* Take the queue lock only to pop an element from the list. */ |
d88744a4 JD |
561 | pthread_mutex_lock(&queue->lock); |
562 | if (cds_list_empty(&queue->list)) { | |
563 | pthread_mutex_unlock(&queue->lock); | |
564 | break; | |
565 | } | |
28ab034a | 566 | job = cds_list_first_entry(&queue->list, typeof(*job), head); |
92816cc3 | 567 | cds_list_del(&job->head); |
d88744a4 JD |
568 | pthread_mutex_unlock(&queue->lock); |
569 | ||
d88744a4 | 570 | session_lock_list(); |
c7031a2c | 571 | session = job->session; |
d88744a4 | 572 | if (!session) { |
28ab034a | 573 | DBG("Session \"%s\" not found", session->name != NULL ? session->name : ""); |
d88744a4 | 574 | /* |
92816cc3 JG |
575 | * This is a non-fatal error, and we cannot report it to |
576 | * the user (timer), so just print the error and | |
577 | * continue the processing. | |
578 | * | |
579 | * While the timer thread will purge pending signals for | |
580 | * a session on the session's destruction, it is | |
581 | * possible for a job targeting that session to have | |
582 | * already been queued before it was destroyed. | |
d88744a4 | 583 | */ |
92816cc3 | 584 | free(job); |
e32d7f27 | 585 | session_put(session); |
5b8139c6 | 586 | session_unlock_list(); |
d88744a4 JD |
587 | continue; |
588 | } | |
589 | ||
d88744a4 | 590 | session_lock(session); |
16100d7a | 591 | ret = run_job(job, session, handle->notification_thread_handle); |
d88744a4 | 592 | session_unlock(session); |
faf1bdcf | 593 | /* Release reference held by the job. */ |
e32d7f27 | 594 | session_put(session); |
ed1e52a3 | 595 | session_unlock_list(); |
92816cc3 | 596 | free(job); |
d88744a4 | 597 | if (ret) { |
d88744a4 JD |
598 | goto end; |
599 | } | |
600 | } | |
601 | ||
602 | ret = 0; | |
603 | ||
604 | end: | |
605 | return ret; | |
606 | } | |
607 | ||
28ab034a JG |
608 | static int handle_condition(const struct lttng_notification *notification, |
609 | struct notification_thread_handle *notification_thread_handle) | |
90936dcf JD |
610 | { |
611 | int ret = 0; | |
cd9adb8b | 612 | const char *condition_session_name = nullptr; |
90936dcf JD |
613 | enum lttng_condition_type condition_type; |
614 | enum lttng_condition_status condition_status; | |
615 | enum lttng_evaluation_status evaluation_status; | |
616 | uint64_t consumed; | |
617 | struct ltt_session *session; | |
c08136a3 | 618 | const struct lttng_condition *condition = |
28ab034a | 619 | lttng_notification_get_const_condition(notification); |
c08136a3 | 620 | const struct lttng_evaluation *evaluation = |
28ab034a | 621 | lttng_notification_get_const_evaluation(notification); |
90936dcf JD |
622 | |
623 | condition_type = lttng_condition_get_type(condition); | |
624 | ||
625 | if (condition_type != LTTNG_CONDITION_TYPE_SESSION_CONSUMED_SIZE) { | |
626 | ret = -1; | |
bd0514a5 | 627 | ERR("Condition type and session usage type are not the same"); |
90936dcf JD |
628 | goto end; |
629 | } | |
630 | ||
631 | /* Fetch info to test */ | |
632 | condition_status = lttng_condition_session_consumed_size_get_session_name( | |
28ab034a | 633 | condition, &condition_session_name); |
90936dcf | 634 | if (condition_status != LTTNG_CONDITION_STATUS_OK) { |
bd0514a5 | 635 | ERR("Session name could not be fetched"); |
90936dcf JD |
636 | ret = -1; |
637 | goto end; | |
638 | } | |
28ab034a JG |
639 | evaluation_status = |
640 | lttng_evaluation_session_consumed_size_get_consumed_size(evaluation, &consumed); | |
90936dcf | 641 | if (evaluation_status != LTTNG_EVALUATION_STATUS_OK) { |
bd0514a5 | 642 | ERR("Failed to get evaluation"); |
90936dcf JD |
643 | ret = -1; |
644 | goto end; | |
645 | } | |
646 | ||
647 | session_lock_list(); | |
648 | session = session_find_by_name(condition_session_name); | |
649 | if (!session) { | |
eb2827a4 | 650 | DBG("Failed to find session while handling notification: notification type = %s, session name = `%s`", |
28ab034a JG |
651 | lttng_condition_type_str(condition_type), |
652 | condition_session_name); | |
eb2827a4 JG |
653 | /* |
654 | * Not a fatal error: a session can be destroyed before we get | |
655 | * the chance to handle the notification. | |
656 | */ | |
657 | ret = 0; | |
658 | session_unlock_list(); | |
90936dcf JD |
659 | goto end; |
660 | } | |
661 | session_lock(session); | |
90936dcf | 662 | |
c08136a3 | 663 | if (!lttng_trigger_is_equal(session->rotate_trigger, |
28ab034a | 664 | lttng_notification_get_const_trigger(notification))) { |
c08136a3 JG |
665 | /* Notification does not originate from our rotation trigger. */ |
666 | ret = 0; | |
667 | goto end_unlock; | |
668 | } | |
669 | ||
28ab034a | 670 | ret = unsubscribe_session_consumed_size_rotation(session, notification_thread_handle); |
90936dcf | 671 | if (ret) { |
490b3229 | 672 | goto end_unlock; |
90936dcf JD |
673 | } |
674 | ||
2545db87 | 675 | ret = cmd_rotate_session( |
cd9adb8b | 676 | session, nullptr, false, LTTNG_TRACE_CHUNK_COMMAND_TYPE_MOVE_TO_COMPLETED); |
2545db87 JG |
677 | switch (ret) { |
678 | case LTTNG_OK: | |
679 | break; | |
680 | case -LTTNG_ERR_ROTATION_PENDING: | |
90936dcf | 681 | DBG("Rotate already pending, subscribe to the next threshold value"); |
2545db87 JG |
682 | break; |
683 | case -LTTNG_ERR_ROTATION_MULTIPLE_AFTER_STOP: | |
684 | DBG("Rotation already happened since last stop, subscribe to the next threshold value"); | |
685 | break; | |
686 | case -LTTNG_ERR_ROTATION_AFTER_STOP_CLEAR: | |
687 | DBG("Rotation already happened since last stop and clear, subscribe to the next threshold value"); | |
688 | break; | |
689 | default: | |
690 | ERR("Failed to rotate on size notification with error: %s", lttng_strerror(ret)); | |
90936dcf JD |
691 | ret = -1; |
692 | goto end_unlock; | |
693 | } | |
2545db87 JG |
694 | |
695 | ret = subscribe_session_consumed_size_rotation( | |
28ab034a | 696 | session, consumed + session->rotate_size, notification_thread_handle); |
90936dcf | 697 | if (ret) { |
bd0514a5 | 698 | ERR("Failed to subscribe to session consumed size condition"); |
90936dcf JD |
699 | goto end_unlock; |
700 | } | |
701 | ret = 0; | |
702 | ||
703 | end_unlock: | |
704 | session_unlock(session); | |
e32d7f27 | 705 | session_put(session); |
5b8139c6 | 706 | session_unlock_list(); |
90936dcf JD |
707 | end: |
708 | return ret; | |
709 | } | |
710 | ||
28ab034a JG |
711 | static int handle_notification_channel(int fd __attribute__((unused)), |
712 | struct rotation_thread_handle *handle, | |
713 | struct rotation_thread *state __attribute__((unused))) | |
90936dcf JD |
714 | { |
715 | int ret; | |
d73ee93f | 716 | bool notification_pending; |
cd9adb8b | 717 | struct lttng_notification *notification = nullptr; |
90936dcf | 718 | enum lttng_notification_channel_status status; |
90936dcf | 719 | |
28ab034a JG |
720 | status = lttng_notification_channel_has_pending_notification(rotate_notification_channel, |
721 | ¬ification_pending); | |
d73ee93f | 722 | if (status != LTTNG_NOTIFICATION_CHANNEL_STATUS_OK) { |
bd0514a5 | 723 | ERR("Error occurred while checking for pending notification"); |
d73ee93f JG |
724 | ret = -1; |
725 | goto end; | |
726 | } | |
727 | ||
728 | if (!notification_pending) { | |
729 | ret = 0; | |
730 | goto end; | |
731 | } | |
732 | ||
90936dcf | 733 | /* Receive the next notification. */ |
28ab034a JG |
734 | status = lttng_notification_channel_get_next_notification(rotate_notification_channel, |
735 | ¬ification); | |
90936dcf JD |
736 | |
737 | switch (status) { | |
738 | case LTTNG_NOTIFICATION_CHANNEL_STATUS_OK: | |
739 | break; | |
740 | case LTTNG_NOTIFICATION_CHANNEL_STATUS_NOTIFICATIONS_DROPPED: | |
741 | /* Not an error, we will wait for the next one */ | |
742 | ret = 0; | |
28ab034a JG |
743 | goto end; |
744 | ; | |
90936dcf JD |
745 | case LTTNG_NOTIFICATION_CHANNEL_STATUS_CLOSED: |
746 | ERR("Notification channel was closed"); | |
747 | ret = -1; | |
748 | goto end; | |
749 | default: | |
750 | /* Unhandled conditions / errors. */ | |
751 | ERR("Unknown notification channel status"); | |
752 | ret = -1; | |
753 | goto end; | |
754 | } | |
755 | ||
28ab034a | 756 | ret = handle_condition(notification, handle->notification_thread_handle); |
90936dcf JD |
757 | |
758 | end: | |
759 | lttng_notification_destroy(notification); | |
90936dcf JD |
760 | return ret; |
761 | } | |
762 | ||
28ab034a | 763 | static void *thread_rotation(void *data) |
db66e574 JD |
764 | { |
765 | int ret; | |
7966af57 | 766 | struct rotation_thread_handle *handle = (rotation_thread_handle *) data; |
92816cc3 | 767 | struct rotation_thread thread; |
87380d40 | 768 | int queue_pipe_fd; |
db66e574 | 769 | |
bd0514a5 | 770 | DBG("Started rotation thread"); |
f620cc28 JG |
771 | rcu_register_thread(); |
772 | rcu_thread_online(); | |
412d7227 | 773 | health_register(the_health_sessiond, HEALTH_SESSIOND_TYPE_ROTATION); |
f620cc28 | 774 | health_code_update(); |
db66e574 JD |
775 | |
776 | if (!handle) { | |
bd0514a5 | 777 | ERR("Invalid thread context provided"); |
db66e574 JD |
778 | goto end; |
779 | } | |
780 | ||
28ab034a | 781 | queue_pipe_fd = lttng_pipe_get_readfd(handle->rotation_timer_queue->event_pipe); |
db66e574 | 782 | |
92816cc3 | 783 | ret = init_thread_state(handle, &thread); |
db66e574 | 784 | if (ret) { |
f5f8e5cd | 785 | goto error; |
db66e574 JD |
786 | } |
787 | ||
db66e574 JD |
788 | while (true) { |
789 | int fd_count, i; | |
790 | ||
791 | health_poll_entry(); | |
bd0514a5 | 792 | DBG("Entering poll wait"); |
92816cc3 | 793 | ret = lttng_poll_wait(&thread.events, -1); |
bd0514a5 | 794 | DBG("Poll wait returned (%i)", ret); |
db66e574 JD |
795 | health_poll_exit(); |
796 | if (ret < 0) { | |
797 | /* | |
798 | * Restart interrupted system call. | |
799 | */ | |
800 | if (errno == EINTR) { | |
801 | continue; | |
802 | } | |
bd0514a5 | 803 | ERR("Error encountered during lttng_poll_wait (%i)", ret); |
db66e574 JD |
804 | goto error; |
805 | } | |
806 | ||
807 | fd_count = ret; | |
808 | for (i = 0; i < fd_count; i++) { | |
92816cc3 JG |
809 | int fd = LTTNG_POLL_GETFD(&thread.events, i); |
810 | uint32_t revents = LTTNG_POLL_GETEV(&thread.events, i); | |
db66e574 | 811 | |
28ab034a | 812 | DBG("Handling fd (%i) activity (%u)", fd, revents); |
db66e574 | 813 | |
92816cc3 | 814 | if (revents & LPOLLERR) { |
bd0514a5 | 815 | ERR("Polling returned an error on fd %i", fd); |
92816cc3 JG |
816 | goto error; |
817 | } | |
818 | ||
85e17b27 | 819 | if (fd == rotate_notification_channel->socket) { |
28ab034a | 820 | ret = handle_notification_channel(fd, handle, &thread); |
85e17b27 | 821 | if (ret) { |
bd0514a5 | 822 | ERR("Error occurred while handling activity on notification channel socket"); |
85e17b27 JG |
823 | goto error; |
824 | } | |
825 | } else { | |
826 | /* Job queue or quit pipe activity. */ | |
85e17b27 JG |
827 | |
828 | /* | |
829 | * The job queue is serviced if there is | |
830 | * activity on the quit pipe to ensure it is | |
831 | * flushed and all references held in the queue | |
832 | * are released. | |
833 | */ | |
28ab034a JG |
834 | ret = handle_job_queue( |
835 | handle, &thread, handle->rotation_timer_queue); | |
d88744a4 | 836 | if (ret) { |
bd0514a5 | 837 | ERR("Failed to handle rotation timer pipe event"); |
d88744a4 JD |
838 | goto error; |
839 | } | |
85e17b27 | 840 | |
64d9b072 JG |
841 | if (fd == queue_pipe_fd) { |
842 | char buf; | |
843 | ||
844 | ret = lttng_read(fd, &buf, 1); | |
845 | if (ret != 1) { | |
28ab034a JG |
846 | ERR("Failed to read from wakeup pipe (fd = %i)", |
847 | fd); | |
64d9b072 JG |
848 | goto error; |
849 | } | |
850 | } else { | |
bd0514a5 | 851 | DBG("Quit pipe activity"); |
85e17b27 | 852 | goto exit; |
90936dcf | 853 | } |
db66e574 JD |
854 | } |
855 | } | |
856 | } | |
857 | exit: | |
858 | error: | |
bd0514a5 | 859 | DBG("Thread exit"); |
92816cc3 | 860 | fini_thread_state(&thread); |
f620cc28 | 861 | end: |
412d7227 | 862 | health_unregister(the_health_sessiond); |
03732c32 JG |
863 | rcu_thread_offline(); |
864 | rcu_unregister_thread(); | |
cd9adb8b | 865 | return nullptr; |
db66e574 | 866 | } |
64d9b072 | 867 | |
28ab034a | 868 | static bool shutdown_rotation_thread(void *thread_data) |
64d9b072 | 869 | { |
7966af57 | 870 | struct rotation_thread_handle *handle = (rotation_thread_handle *) thread_data; |
64d9b072 JG |
871 | const int write_fd = lttng_pipe_get_writefd(handle->quit_pipe); |
872 | ||
873 | return notify_thread_pipe(write_fd) == 1; | |
874 | } | |
875 | ||
876 | bool launch_rotation_thread(struct rotation_thread_handle *handle) | |
877 | { | |
878 | struct lttng_thread *thread; | |
879 | ||
28ab034a | 880 | thread = lttng_thread_create( |
cd9adb8b | 881 | "Rotation", thread_rotation, shutdown_rotation_thread, nullptr, handle); |
64d9b072 JG |
882 | if (!thread) { |
883 | goto error; | |
884 | } | |
885 | lttng_thread_put(thread); | |
886 | return true; | |
887 | error: | |
888 | return false; | |
889 | } |