Port: Add Solaris paths compat
[lttng-tools.git] / src / common / consumer-timer.c
1 /*
2 * Copyright (C) 2012 - Julien Desfossez <julien.desfossez@efficios.com>
3 * David Goulet <dgoulet@efficios.com>
4 *
5 * This program is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License, version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
12 * more details.
13 *
14 * You should have received a copy of the GNU General Public License along with
15 * this program; if not, write to the Free Software Foundation, Inc., 51
16 * Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
17 */
18
19 #define _LGPL_SOURCE
20 #include <assert.h>
21 #include <inttypes.h>
22 #include <signal.h>
23
24 #include <bin/lttng-consumerd/health-consumerd.h>
25 #include <common/common.h>
26 #include <common/compat/endian.h>
27 #include <common/kernel-ctl/kernel-ctl.h>
28 #include <common/kernel-consumer/kernel-consumer.h>
29 #include <common/consumer-stream.h>
30
31 #include "consumer-timer.h"
32 #include "consumer-testpoint.h"
33 #include "ust-consumer/ust-consumer.h"
34
35 static struct timer_signal_data timer_signal = {
36 .tid = 0,
37 .setup_done = 0,
38 .qs_done = 0,
39 .lock = PTHREAD_MUTEX_INITIALIZER,
40 };
41
42 /*
43 * Set custom signal mask to current thread.
44 */
45 static void setmask(sigset_t *mask)
46 {
47 int ret;
48
49 ret = sigemptyset(mask);
50 if (ret) {
51 PERROR("sigemptyset");
52 }
53 ret = sigaddset(mask, LTTNG_CONSUMER_SIG_SWITCH);
54 if (ret) {
55 PERROR("sigaddset switch");
56 }
57 ret = sigaddset(mask, LTTNG_CONSUMER_SIG_TEARDOWN);
58 if (ret) {
59 PERROR("sigaddset teardown");
60 }
61 ret = sigaddset(mask, LTTNG_CONSUMER_SIG_LIVE);
62 if (ret) {
63 PERROR("sigaddset live");
64 }
65 }
66
67 /*
68 * Execute action on a timer switch.
69 *
70 * Beware: metadata_switch_timer() should *never* take a mutex also held
71 * while consumer_timer_switch_stop() is called. It would result in
72 * deadlocks.
73 */
74 static void metadata_switch_timer(struct lttng_consumer_local_data *ctx,
75 int sig, siginfo_t *si, void *uc)
76 {
77 int ret;
78 struct lttng_consumer_channel *channel;
79
80 channel = si->si_value.sival_ptr;
81 assert(channel);
82
83 if (channel->switch_timer_error) {
84 return;
85 }
86
87 DBG("Switch timer for channel %" PRIu64, channel->key);
88 switch (ctx->type) {
89 case LTTNG_CONSUMER32_UST:
90 case LTTNG_CONSUMER64_UST:
91 /*
92 * Locks taken by lttng_ustconsumer_request_metadata():
93 * - metadata_socket_lock
94 * - Calling lttng_ustconsumer_recv_metadata():
95 * - channel->metadata_cache->lock
96 * - Calling consumer_metadata_cache_flushed():
97 * - channel->timer_lock
98 * - channel->metadata_cache->lock
99 *
100 * Ensure that neither consumer_data.lock nor
101 * channel->lock are taken within this function, since
102 * they are held while consumer_timer_switch_stop() is
103 * called.
104 */
105 ret = lttng_ustconsumer_request_metadata(ctx, channel, 1, 1);
106 if (ret < 0) {
107 channel->switch_timer_error = 1;
108 }
109 break;
110 case LTTNG_CONSUMER_KERNEL:
111 case LTTNG_CONSUMER_UNKNOWN:
112 assert(0);
113 break;
114 }
115 }
116
117 static int send_empty_index(struct lttng_consumer_stream *stream, uint64_t ts,
118 uint64_t stream_id)
119 {
120 int ret;
121 struct ctf_packet_index index;
122
123 memset(&index, 0, sizeof(index));
124 index.stream_id = htobe64(stream_id);
125 index.timestamp_end = htobe64(ts);
126 ret = consumer_stream_write_index(stream, &index);
127 if (ret < 0) {
128 goto error;
129 }
130
131 error:
132 return ret;
133 }
134
135 int consumer_flush_kernel_index(struct lttng_consumer_stream *stream)
136 {
137 uint64_t ts, stream_id;
138 int ret;
139
140 ret = kernctl_get_current_timestamp(stream->wait_fd, &ts);
141 if (ret < 0) {
142 ERR("Failed to get the current timestamp");
143 goto end;
144 }
145 ret = kernctl_buffer_flush(stream->wait_fd);
146 if (ret < 0) {
147 ERR("Failed to flush kernel stream");
148 goto end;
149 }
150 ret = kernctl_snapshot(stream->wait_fd);
151 if (ret < 0) {
152 if (errno != EAGAIN && errno != ENODATA) {
153 PERROR("live timer kernel snapshot");
154 ret = -1;
155 goto end;
156 }
157 ret = kernctl_get_stream_id(stream->wait_fd, &stream_id);
158 if (ret < 0) {
159 PERROR("kernctl_get_stream_id");
160 goto end;
161 }
162 DBG("Stream %" PRIu64 " empty, sending beacon", stream->key);
163 ret = send_empty_index(stream, ts, stream_id);
164 if (ret < 0) {
165 goto end;
166 }
167 }
168 ret = 0;
169 end:
170 return ret;
171 }
172
173 static int check_kernel_stream(struct lttng_consumer_stream *stream)
174 {
175 int ret;
176
177 /*
178 * While holding the stream mutex, try to take a snapshot, if it
179 * succeeds, it means that data is ready to be sent, just let the data
180 * thread handle that. Otherwise, if the snapshot returns EAGAIN, it
181 * means that there is no data to read after the flush, so we can
182 * safely send the empty index.
183 *
184 * Doing a trylock and checking if waiting on metadata if
185 * trylock fails. Bail out of the stream is indeed waiting for
186 * metadata to be pushed. Busy wait on trylock otherwise.
187 */
188 for (;;) {
189 ret = pthread_mutex_trylock(&stream->lock);
190 switch (ret) {
191 case 0:
192 break; /* We have the lock. */
193 case EBUSY:
194 pthread_mutex_lock(&stream->metadata_timer_lock);
195 if (stream->waiting_on_metadata) {
196 ret = 0;
197 stream->missed_metadata_flush = true;
198 pthread_mutex_unlock(&stream->metadata_timer_lock);
199 goto end; /* Bail out. */
200 }
201 pthread_mutex_unlock(&stream->metadata_timer_lock);
202 /* Try again. */
203 caa_cpu_relax();
204 continue;
205 default:
206 ERR("Unexpected pthread_mutex_trylock error %d", ret);
207 ret = -1;
208 goto end;
209 }
210 break;
211 }
212 ret = consumer_flush_kernel_index(stream);
213 pthread_mutex_unlock(&stream->lock);
214 end:
215 return ret;
216 }
217
218 int consumer_flush_ust_index(struct lttng_consumer_stream *stream)
219 {
220 uint64_t ts, stream_id;
221 int ret;
222
223 ret = cds_lfht_is_node_deleted(&stream->node.node);
224 if (ret) {
225 goto end;
226 }
227
228 ret = lttng_ustconsumer_get_current_timestamp(stream, &ts);
229 if (ret < 0) {
230 ERR("Failed to get the current timestamp");
231 goto end;
232 }
233 lttng_ustconsumer_flush_buffer(stream, 1);
234 ret = lttng_ustconsumer_take_snapshot(stream);
235 if (ret < 0) {
236 if (ret != -EAGAIN) {
237 ERR("Taking UST snapshot");
238 ret = -1;
239 goto end;
240 }
241 ret = lttng_ustconsumer_get_stream_id(stream, &stream_id);
242 if (ret < 0) {
243 PERROR("ustctl_get_stream_id");
244 goto end;
245 }
246 DBG("Stream %" PRIu64 " empty, sending beacon", stream->key);
247 ret = send_empty_index(stream, ts, stream_id);
248 if (ret < 0) {
249 goto end;
250 }
251 }
252 ret = 0;
253 end:
254 return ret;
255 }
256
257 static int check_ust_stream(struct lttng_consumer_stream *stream)
258 {
259 int ret;
260
261 assert(stream);
262 assert(stream->ustream);
263 /*
264 * While holding the stream mutex, try to take a snapshot, if it
265 * succeeds, it means that data is ready to be sent, just let the data
266 * thread handle that. Otherwise, if the snapshot returns EAGAIN, it
267 * means that there is no data to read after the flush, so we can
268 * safely send the empty index.
269 *
270 * Doing a trylock and checking if waiting on metadata if
271 * trylock fails. Bail out of the stream is indeed waiting for
272 * metadata to be pushed. Busy wait on trylock otherwise.
273 */
274 for (;;) {
275 ret = pthread_mutex_trylock(&stream->lock);
276 switch (ret) {
277 case 0:
278 break; /* We have the lock. */
279 case EBUSY:
280 pthread_mutex_lock(&stream->metadata_timer_lock);
281 if (stream->waiting_on_metadata) {
282 ret = 0;
283 stream->missed_metadata_flush = true;
284 pthread_mutex_unlock(&stream->metadata_timer_lock);
285 goto end; /* Bail out. */
286 }
287 pthread_mutex_unlock(&stream->metadata_timer_lock);
288 /* Try again. */
289 caa_cpu_relax();
290 continue;
291 default:
292 ERR("Unexpected pthread_mutex_trylock error %d", ret);
293 ret = -1;
294 goto end;
295 }
296 break;
297 }
298 ret = consumer_flush_ust_index(stream);
299 pthread_mutex_unlock(&stream->lock);
300 end:
301 return ret;
302 }
303
304 /*
305 * Execute action on a live timer
306 */
307 static void live_timer(struct lttng_consumer_local_data *ctx,
308 int sig, siginfo_t *si, void *uc)
309 {
310 int ret;
311 struct lttng_consumer_channel *channel;
312 struct lttng_consumer_stream *stream;
313 struct lttng_ht *ht;
314 struct lttng_ht_iter iter;
315
316 channel = si->si_value.sival_ptr;
317 assert(channel);
318
319 if (channel->switch_timer_error) {
320 goto error;
321 }
322 ht = consumer_data.stream_per_chan_id_ht;
323
324 DBG("Live timer for channel %" PRIu64, channel->key);
325
326 rcu_read_lock();
327 switch (ctx->type) {
328 case LTTNG_CONSUMER32_UST:
329 case LTTNG_CONSUMER64_UST:
330 cds_lfht_for_each_entry_duplicate(ht->ht,
331 ht->hash_fct(&channel->key, lttng_ht_seed),
332 ht->match_fct, &channel->key, &iter.iter,
333 stream, node_channel_id.node) {
334 ret = check_ust_stream(stream);
335 if (ret < 0) {
336 goto error_unlock;
337 }
338 }
339 break;
340 case LTTNG_CONSUMER_KERNEL:
341 cds_lfht_for_each_entry_duplicate(ht->ht,
342 ht->hash_fct(&channel->key, lttng_ht_seed),
343 ht->match_fct, &channel->key, &iter.iter,
344 stream, node_channel_id.node) {
345 ret = check_kernel_stream(stream);
346 if (ret < 0) {
347 goto error_unlock;
348 }
349 }
350 break;
351 case LTTNG_CONSUMER_UNKNOWN:
352 assert(0);
353 break;
354 }
355
356 error_unlock:
357 rcu_read_unlock();
358
359 error:
360 return;
361 }
362
363 static
364 void consumer_timer_signal_thread_qs(unsigned int signr)
365 {
366 sigset_t pending_set;
367 int ret;
368
369 /*
370 * We need to be the only thread interacting with the thread
371 * that manages signals for teardown synchronization.
372 */
373 pthread_mutex_lock(&timer_signal.lock);
374
375 /* Ensure we don't have any signal queued for this channel. */
376 for (;;) {
377 ret = sigemptyset(&pending_set);
378 if (ret == -1) {
379 PERROR("sigemptyset");
380 }
381 ret = sigpending(&pending_set);
382 if (ret == -1) {
383 PERROR("sigpending");
384 }
385 if (!sigismember(&pending_set, LTTNG_CONSUMER_SIG_SWITCH)) {
386 break;
387 }
388 caa_cpu_relax();
389 }
390
391 /*
392 * From this point, no new signal handler will be fired that would try to
393 * access "chan". However, we still need to wait for any currently
394 * executing handler to complete.
395 */
396 cmm_smp_mb();
397 CMM_STORE_SHARED(timer_signal.qs_done, 0);
398 cmm_smp_mb();
399
400 /*
401 * Kill with LTTNG_CONSUMER_SIG_TEARDOWN, so signal management thread wakes
402 * up.
403 */
404 kill(getpid(), LTTNG_CONSUMER_SIG_TEARDOWN);
405
406 while (!CMM_LOAD_SHARED(timer_signal.qs_done)) {
407 caa_cpu_relax();
408 }
409 cmm_smp_mb();
410
411 pthread_mutex_unlock(&timer_signal.lock);
412 }
413
414 /*
415 * Set the timer for periodical metadata flush.
416 */
417 void consumer_timer_switch_start(struct lttng_consumer_channel *channel,
418 unsigned int switch_timer_interval)
419 {
420 int ret;
421 struct sigevent sev;
422 struct itimerspec its;
423
424 assert(channel);
425 assert(channel->key);
426
427 if (switch_timer_interval == 0) {
428 return;
429 }
430
431 sev.sigev_notify = SIGEV_SIGNAL;
432 sev.sigev_signo = LTTNG_CONSUMER_SIG_SWITCH;
433 sev.sigev_value.sival_ptr = channel;
434 ret = timer_create(CLOCKID, &sev, &channel->switch_timer);
435 if (ret == -1) {
436 PERROR("timer_create");
437 }
438 channel->switch_timer_enabled = 1;
439
440 its.it_value.tv_sec = switch_timer_interval / 1000000;
441 its.it_value.tv_nsec = switch_timer_interval % 1000000;
442 its.it_interval.tv_sec = its.it_value.tv_sec;
443 its.it_interval.tv_nsec = its.it_value.tv_nsec;
444
445 ret = timer_settime(channel->switch_timer, 0, &its, NULL);
446 if (ret == -1) {
447 PERROR("timer_settime");
448 }
449 }
450
451 /*
452 * Stop and delete timer.
453 */
454 void consumer_timer_switch_stop(struct lttng_consumer_channel *channel)
455 {
456 int ret;
457
458 assert(channel);
459
460 ret = timer_delete(channel->switch_timer);
461 if (ret == -1) {
462 PERROR("timer_delete");
463 }
464
465 consumer_timer_signal_thread_qs(LTTNG_CONSUMER_SIG_SWITCH);
466
467 channel->switch_timer = 0;
468 channel->switch_timer_enabled = 0;
469 }
470
471 /*
472 * Set the timer for the live mode.
473 */
474 void consumer_timer_live_start(struct lttng_consumer_channel *channel,
475 int live_timer_interval)
476 {
477 int ret;
478 struct sigevent sev;
479 struct itimerspec its;
480
481 assert(channel);
482 assert(channel->key);
483
484 if (live_timer_interval <= 0) {
485 return;
486 }
487
488 sev.sigev_notify = SIGEV_SIGNAL;
489 sev.sigev_signo = LTTNG_CONSUMER_SIG_LIVE;
490 sev.sigev_value.sival_ptr = channel;
491 ret = timer_create(CLOCKID, &sev, &channel->live_timer);
492 if (ret == -1) {
493 PERROR("timer_create");
494 }
495 channel->live_timer_enabled = 1;
496
497 its.it_value.tv_sec = live_timer_interval / 1000000;
498 its.it_value.tv_nsec = live_timer_interval % 1000000;
499 its.it_interval.tv_sec = its.it_value.tv_sec;
500 its.it_interval.tv_nsec = its.it_value.tv_nsec;
501
502 ret = timer_settime(channel->live_timer, 0, &its, NULL);
503 if (ret == -1) {
504 PERROR("timer_settime");
505 }
506 }
507
508 /*
509 * Stop and delete timer.
510 */
511 void consumer_timer_live_stop(struct lttng_consumer_channel *channel)
512 {
513 int ret;
514
515 assert(channel);
516
517 ret = timer_delete(channel->live_timer);
518 if (ret == -1) {
519 PERROR("timer_delete");
520 }
521
522 consumer_timer_signal_thread_qs(LTTNG_CONSUMER_SIG_LIVE);
523
524 channel->live_timer = 0;
525 channel->live_timer_enabled = 0;
526 }
527
528 /*
529 * Block the RT signals for the entire process. It must be called from the
530 * consumer main before creating the threads
531 */
532 int consumer_signal_init(void)
533 {
534 int ret;
535 sigset_t mask;
536
537 /* Block signal for entire process, so only our thread processes it. */
538 setmask(&mask);
539 ret = pthread_sigmask(SIG_BLOCK, &mask, NULL);
540 if (ret) {
541 errno = ret;
542 PERROR("pthread_sigmask");
543 return -1;
544 }
545 return 0;
546 }
547
548 /*
549 * This thread is the sighandler for signals LTTNG_CONSUMER_SIG_SWITCH,
550 * LTTNG_CONSUMER_SIG_TEARDOWN and LTTNG_CONSUMER_SIG_LIVE.
551 */
552 void *consumer_timer_thread(void *data)
553 {
554 int signr;
555 sigset_t mask;
556 siginfo_t info;
557 struct lttng_consumer_local_data *ctx = data;
558
559 rcu_register_thread();
560
561 health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_METADATA_TIMER);
562
563 if (testpoint(consumerd_thread_metadata_timer)) {
564 goto error_testpoint;
565 }
566
567 health_code_update();
568
569 /* Only self thread will receive signal mask. */
570 setmask(&mask);
571 CMM_STORE_SHARED(timer_signal.tid, pthread_self());
572
573 while (1) {
574 health_code_update();
575
576 health_poll_entry();
577 signr = sigwaitinfo(&mask, &info);
578 health_poll_exit();
579 if (signr == -1) {
580 if (errno != EINTR) {
581 PERROR("sigwaitinfo");
582 }
583 continue;
584 } else if (signr == LTTNG_CONSUMER_SIG_SWITCH) {
585 metadata_switch_timer(ctx, info.si_signo, &info, NULL);
586 } else if (signr == LTTNG_CONSUMER_SIG_TEARDOWN) {
587 cmm_smp_mb();
588 CMM_STORE_SHARED(timer_signal.qs_done, 1);
589 cmm_smp_mb();
590 DBG("Signal timer metadata thread teardown");
591 } else if (signr == LTTNG_CONSUMER_SIG_LIVE) {
592 live_timer(ctx, info.si_signo, &info, NULL);
593 } else {
594 ERR("Unexpected signal %d\n", info.si_signo);
595 }
596 }
597
598 error_testpoint:
599 /* Only reached in testpoint error */
600 health_error();
601 health_unregister(health_consumerd);
602
603 rcu_unregister_thread();
604
605 /* Never return */
606 return NULL;
607 }
This page took 0.041051 seconds and 4 git commands to generate.