rotate pending working on the relay
[deliverable/lttng-tools.git] / src / common / ust-consumer / ust-consumer.c
1 /*
2 * Copyright (C) 2011 - Julien Desfossez <julien.desfossez@polymtl.ca>
3 * Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
4 * Copyright (C) 2017 - Jérémie Galarneau <jeremie.galarneau@efficios.com>
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License, version 2 only,
8 * as published by the Free Software Foundation.
9 *
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
14 *
15 * You should have received a copy of the GNU General Public License along
16 * with this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
18 */
19
20 #define _LGPL_SOURCE
21 #include <assert.h>
22 #include <lttng/ust-ctl.h>
23 #include <poll.h>
24 #include <pthread.h>
25 #include <stdlib.h>
26 #include <string.h>
27 #include <sys/mman.h>
28 #include <sys/socket.h>
29 #include <sys/stat.h>
30 #include <sys/types.h>
31 #include <inttypes.h>
32 #include <unistd.h>
33 #include <urcu/list.h>
34 #include <signal.h>
35
36 #include <bin/lttng-consumerd/health-consumerd.h>
37 #include <common/common.h>
38 #include <common/sessiond-comm/sessiond-comm.h>
39 #include <common/relayd/relayd.h>
40 #include <common/compat/fcntl.h>
41 #include <common/compat/endian.h>
42 #include <common/consumer/consumer-metadata-cache.h>
43 #include <common/consumer/consumer-stream.h>
44 #include <common/consumer/consumer-timer.h>
45 #include <common/utils.h>
46 #include <common/index/index.h>
47
48 #include "ust-consumer.h"
49
50 #define INT_MAX_STR_LEN 12 /* includes \0 */
51
52 extern struct lttng_consumer_global_data consumer_data;
53 extern int consumer_poll_timeout;
54
55 /*
56 * Free channel object and all streams associated with it. This MUST be used
57 * only and only if the channel has _NEVER_ been added to the global channel
58 * hash table.
59 */
60 static void destroy_channel(struct lttng_consumer_channel *channel)
61 {
62 struct lttng_consumer_stream *stream, *stmp;
63
64 assert(channel);
65
66 DBG("UST consumer cleaning stream list");
67
68 cds_list_for_each_entry_safe(stream, stmp, &channel->streams.head,
69 send_node) {
70
71 health_code_update();
72
73 cds_list_del(&stream->send_node);
74 ustctl_destroy_stream(stream->ustream);
75 free(stream);
76 }
77
78 /*
79 * If a channel is available meaning that was created before the streams
80 * were, delete it.
81 */
82 if (channel->uchan) {
83 lttng_ustconsumer_del_channel(channel);
84 lttng_ustconsumer_free_channel(channel);
85 }
86 free(channel);
87 }
88
89 /*
90 * Add channel to internal consumer state.
91 *
92 * Returns 0 on success or else a negative value.
93 */
94 static int add_channel(struct lttng_consumer_channel *channel,
95 struct lttng_consumer_local_data *ctx)
96 {
97 int ret = 0;
98
99 assert(channel);
100 assert(ctx);
101
102 if (ctx->on_recv_channel != NULL) {
103 ret = ctx->on_recv_channel(channel);
104 if (ret == 0) {
105 ret = consumer_add_channel(channel, ctx);
106 } else if (ret < 0) {
107 /* Most likely an ENOMEM. */
108 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
109 goto error;
110 }
111 } else {
112 ret = consumer_add_channel(channel, ctx);
113 }
114
115 DBG("UST consumer channel added (key: %" PRIu64 ")", channel->key);
116
117 error:
118 return ret;
119 }
120
121 /*
122 * Allocate and return a consumer channel object.
123 */
124 static struct lttng_consumer_channel *allocate_channel(uint64_t session_id,
125 const char *pathname, const char *name, uid_t uid, gid_t gid,
126 uint64_t relayd_id, uint64_t key, enum lttng_event_output output,
127 uint64_t tracefile_size, uint64_t tracefile_count,
128 uint64_t session_id_per_pid, unsigned int monitor,
129 unsigned int live_timer_interval,
130 const char *root_shm_path, const char *shm_path)
131 {
132 assert(pathname);
133 assert(name);
134
135 return consumer_allocate_channel(key, session_id, pathname, name, uid,
136 gid, relayd_id, output, tracefile_size,
137 tracefile_count, session_id_per_pid, monitor,
138 live_timer_interval, root_shm_path, shm_path);
139 }
140
141 /*
142 * Allocate and return a consumer stream object. If _alloc_ret is not NULL, the
143 * error value if applicable is set in it else it is kept untouched.
144 *
145 * Return NULL on error else the newly allocated stream object.
146 */
147 static struct lttng_consumer_stream *allocate_stream(int cpu, int key,
148 struct lttng_consumer_channel *channel,
149 struct lttng_consumer_local_data *ctx, int *_alloc_ret)
150 {
151 int alloc_ret;
152 struct lttng_consumer_stream *stream = NULL;
153
154 assert(channel);
155 assert(ctx);
156
157 stream = consumer_allocate_stream(channel->key,
158 key,
159 LTTNG_CONSUMER_ACTIVE_STREAM,
160 channel->name,
161 channel->uid,
162 channel->gid,
163 channel->relayd_id,
164 channel->session_id,
165 cpu,
166 &alloc_ret,
167 channel->type,
168 channel->monitor);
169 if (stream == NULL) {
170 switch (alloc_ret) {
171 case -ENOENT:
172 /*
173 * We could not find the channel. Can happen if cpu hotplug
174 * happens while tearing down.
175 */
176 DBG3("Could not find channel");
177 break;
178 case -ENOMEM:
179 case -EINVAL:
180 default:
181 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
182 break;
183 }
184 goto error;
185 }
186 consumer_stream_copy_ro_channel_values(stream, channel);
187 stream->chan = channel;
188
189 error:
190 if (_alloc_ret) {
191 *_alloc_ret = alloc_ret;
192 }
193 return stream;
194 }
195
196 /*
197 * Send the given stream pointer to the corresponding thread.
198 *
199 * Returns 0 on success else a negative value.
200 */
201 static int send_stream_to_thread(struct lttng_consumer_stream *stream,
202 struct lttng_consumer_local_data *ctx)
203 {
204 int ret;
205 struct lttng_pipe *stream_pipe;
206
207 /* Get the right pipe where the stream will be sent. */
208 if (stream->metadata_flag) {
209 ret = consumer_add_metadata_stream(stream);
210 if (ret) {
211 ERR("Consumer add metadata stream %" PRIu64 " failed.",
212 stream->key);
213 goto error;
214 }
215 stream_pipe = ctx->consumer_metadata_pipe;
216 } else {
217 ret = consumer_add_data_stream(stream);
218 if (ret) {
219 ERR("Consumer add stream %" PRIu64 " failed.",
220 stream->key);
221 goto error;
222 }
223 stream_pipe = ctx->consumer_data_pipe;
224 }
225
226 /*
227 * From this point on, the stream's ownership has been moved away from
228 * the channel and becomes globally visible.
229 */
230 stream->globally_visible = 1;
231
232 ret = lttng_pipe_write(stream_pipe, &stream, sizeof(stream));
233 if (ret < 0) {
234 ERR("Consumer write %s stream to pipe %d",
235 stream->metadata_flag ? "metadata" : "data",
236 lttng_pipe_get_writefd(stream_pipe));
237 if (stream->metadata_flag) {
238 consumer_del_stream_for_metadata(stream);
239 } else {
240 consumer_del_stream_for_data(stream);
241 }
242 }
243 error:
244 return ret;
245 }
246
247 static
248 int get_stream_shm_path(char *stream_shm_path, const char *shm_path, int cpu)
249 {
250 char cpu_nr[INT_MAX_STR_LEN]; /* int max len */
251 int ret;
252
253 strncpy(stream_shm_path, shm_path, PATH_MAX);
254 stream_shm_path[PATH_MAX - 1] = '\0';
255 ret = snprintf(cpu_nr, INT_MAX_STR_LEN, "%i", cpu);
256 if (ret < 0) {
257 PERROR("snprintf");
258 goto end;
259 }
260 strncat(stream_shm_path, cpu_nr,
261 PATH_MAX - strlen(stream_shm_path) - 1);
262 ret = 0;
263 end:
264 return ret;
265 }
266
267 /*
268 * Create streams for the given channel using liblttng-ust-ctl.
269 *
270 * Return 0 on success else a negative value.
271 */
272 static int create_ust_streams(struct lttng_consumer_channel *channel,
273 struct lttng_consumer_local_data *ctx)
274 {
275 int ret, cpu = 0;
276 struct ustctl_consumer_stream *ustream;
277 struct lttng_consumer_stream *stream;
278
279 assert(channel);
280 assert(ctx);
281
282 /*
283 * While a stream is available from ustctl. When NULL is returned, we've
284 * reached the end of the possible stream for the channel.
285 */
286 while ((ustream = ustctl_create_stream(channel->uchan, cpu))) {
287 int wait_fd;
288 int ust_metadata_pipe[2];
289
290 health_code_update();
291
292 if (channel->type == CONSUMER_CHANNEL_TYPE_METADATA && channel->monitor) {
293 ret = utils_create_pipe_cloexec_nonblock(ust_metadata_pipe);
294 if (ret < 0) {
295 ERR("Create ust metadata poll pipe");
296 goto error;
297 }
298 wait_fd = ust_metadata_pipe[0];
299 } else {
300 wait_fd = ustctl_stream_get_wait_fd(ustream);
301 }
302
303 /* Allocate consumer stream object. */
304 stream = allocate_stream(cpu, wait_fd, channel, ctx, &ret);
305 if (!stream) {
306 goto error_alloc;
307 }
308 stream->ustream = ustream;
309 /*
310 * Store it so we can save multiple function calls afterwards since
311 * this value is used heavily in the stream threads. This is UST
312 * specific so this is why it's done after allocation.
313 */
314 stream->wait_fd = wait_fd;
315
316 /*
317 * Increment channel refcount since the channel reference has now been
318 * assigned in the allocation process above.
319 */
320 if (stream->chan->monitor) {
321 uatomic_inc(&stream->chan->refcount);
322 }
323
324 /*
325 * Order is important this is why a list is used. On error, the caller
326 * should clean this list.
327 */
328 cds_list_add_tail(&stream->send_node, &channel->streams.head);
329
330 ret = ustctl_get_max_subbuf_size(stream->ustream,
331 &stream->max_sb_size);
332 if (ret < 0) {
333 ERR("ustctl_get_max_subbuf_size failed for stream %s",
334 stream->name);
335 goto error;
336 }
337
338 /* Do actions once stream has been received. */
339 if (ctx->on_recv_stream) {
340 ret = ctx->on_recv_stream(stream);
341 if (ret < 0) {
342 goto error;
343 }
344 }
345
346 DBG("UST consumer add stream %s (key: %" PRIu64 ") with relayd id %" PRIu64,
347 stream->name, stream->key, stream->relayd_stream_id);
348
349 /* Set next CPU stream. */
350 channel->streams.count = ++cpu;
351
352 /* Keep stream reference when creating metadata. */
353 if (channel->type == CONSUMER_CHANNEL_TYPE_METADATA) {
354 channel->metadata_stream = stream;
355 if (channel->monitor) {
356 /* Set metadata poll pipe if we created one */
357 memcpy(stream->ust_metadata_poll_pipe,
358 ust_metadata_pipe,
359 sizeof(ust_metadata_pipe));
360 }
361 }
362 }
363
364 return 0;
365
366 error:
367 error_alloc:
368 return ret;
369 }
370
371 /*
372 * create_posix_shm is never called concurrently within a process.
373 */
374 static
375 int create_posix_shm(void)
376 {
377 char tmp_name[NAME_MAX];
378 int shmfd, ret;
379
380 ret = snprintf(tmp_name, NAME_MAX, "/ust-shm-consumer-%d", getpid());
381 if (ret < 0) {
382 PERROR("snprintf");
383 return -1;
384 }
385 /*
386 * Allocate shm, and immediately unlink its shm oject, keeping
387 * only the file descriptor as a reference to the object.
388 * We specifically do _not_ use the / at the beginning of the
389 * pathname so that some OS implementations can keep it local to
390 * the process (POSIX leaves this implementation-defined).
391 */
392 shmfd = shm_open(tmp_name, O_CREAT | O_EXCL | O_RDWR, 0700);
393 if (shmfd < 0) {
394 PERROR("shm_open");
395 goto error_shm_open;
396 }
397 ret = shm_unlink(tmp_name);
398 if (ret < 0 && errno != ENOENT) {
399 PERROR("shm_unlink");
400 goto error_shm_release;
401 }
402 return shmfd;
403
404 error_shm_release:
405 ret = close(shmfd);
406 if (ret) {
407 PERROR("close");
408 }
409 error_shm_open:
410 return -1;
411 }
412
413 static int open_ust_stream_fd(struct lttng_consumer_channel *channel,
414 struct ustctl_consumer_channel_attr *attr,
415 int cpu)
416 {
417 char shm_path[PATH_MAX];
418 int ret;
419
420 if (!channel->shm_path[0]) {
421 return create_posix_shm();
422 }
423 ret = get_stream_shm_path(shm_path, channel->shm_path, cpu);
424 if (ret) {
425 goto error_shm_path;
426 }
427 return run_as_open(shm_path,
428 O_RDWR | O_CREAT | O_EXCL, S_IRUSR | S_IWUSR,
429 channel->uid, channel->gid);
430
431 error_shm_path:
432 return -1;
433 }
434
435 /*
436 * Create an UST channel with the given attributes and send it to the session
437 * daemon using the ust ctl API.
438 *
439 * Return 0 on success or else a negative value.
440 */
441 static int create_ust_channel(struct lttng_consumer_channel *channel,
442 struct ustctl_consumer_channel_attr *attr,
443 struct ustctl_consumer_channel **ust_chanp)
444 {
445 int ret, nr_stream_fds, i, j;
446 int *stream_fds;
447 struct ustctl_consumer_channel *ust_channel;
448
449 assert(channel);
450 assert(attr);
451 assert(ust_chanp);
452
453 DBG3("Creating channel to ustctl with attr: [overwrite: %d, "
454 "subbuf_size: %" PRIu64 ", num_subbuf: %" PRIu64 ", "
455 "switch_timer_interval: %u, read_timer_interval: %u, "
456 "output: %d, type: %d", attr->overwrite, attr->subbuf_size,
457 attr->num_subbuf, attr->switch_timer_interval,
458 attr->read_timer_interval, attr->output, attr->type);
459
460 if (channel->type == CONSUMER_CHANNEL_TYPE_METADATA)
461 nr_stream_fds = 1;
462 else
463 nr_stream_fds = ustctl_get_nr_stream_per_channel();
464 stream_fds = zmalloc(nr_stream_fds * sizeof(*stream_fds));
465 if (!stream_fds) {
466 ret = -1;
467 goto error_alloc;
468 }
469 for (i = 0; i < nr_stream_fds; i++) {
470 stream_fds[i] = open_ust_stream_fd(channel, attr, i);
471 if (stream_fds[i] < 0) {
472 ret = -1;
473 goto error_open;
474 }
475 }
476 ust_channel = ustctl_create_channel(attr, stream_fds, nr_stream_fds);
477 if (!ust_channel) {
478 ret = -1;
479 goto error_create;
480 }
481 channel->nr_stream_fds = nr_stream_fds;
482 channel->stream_fds = stream_fds;
483 *ust_chanp = ust_channel;
484
485 return 0;
486
487 error_create:
488 error_open:
489 for (j = i - 1; j >= 0; j--) {
490 int closeret;
491
492 closeret = close(stream_fds[j]);
493 if (closeret) {
494 PERROR("close");
495 }
496 if (channel->shm_path[0]) {
497 char shm_path[PATH_MAX];
498
499 closeret = get_stream_shm_path(shm_path,
500 channel->shm_path, j);
501 if (closeret) {
502 ERR("Cannot get stream shm path");
503 }
504 closeret = run_as_unlink(shm_path,
505 channel->uid, channel->gid);
506 if (closeret) {
507 PERROR("unlink %s", shm_path);
508 }
509 }
510 }
511 /* Try to rmdir all directories under shm_path root. */
512 if (channel->root_shm_path[0]) {
513 (void) run_as_rmdir_recursive(channel->root_shm_path,
514 channel->uid, channel->gid);
515 }
516 free(stream_fds);
517 error_alloc:
518 return ret;
519 }
520
521 /*
522 * Send a single given stream to the session daemon using the sock.
523 *
524 * Return 0 on success else a negative value.
525 */
526 static int send_sessiond_stream(int sock, struct lttng_consumer_stream *stream)
527 {
528 int ret;
529
530 assert(stream);
531 assert(sock >= 0);
532
533 DBG("UST consumer sending stream %" PRIu64 " to sessiond", stream->key);
534
535 /* Send stream to session daemon. */
536 ret = ustctl_send_stream_to_sessiond(sock, stream->ustream);
537 if (ret < 0) {
538 goto error;
539 }
540
541 error:
542 return ret;
543 }
544
545 /*
546 * Send channel to sessiond.
547 *
548 * Return 0 on success or else a negative value.
549 */
550 static int send_sessiond_channel(int sock,
551 struct lttng_consumer_channel *channel,
552 struct lttng_consumer_local_data *ctx, int *relayd_error)
553 {
554 int ret, ret_code = LTTCOMM_CONSUMERD_SUCCESS;
555 struct lttng_consumer_stream *stream;
556 uint64_t net_seq_idx = -1ULL;
557
558 assert(channel);
559 assert(ctx);
560 assert(sock >= 0);
561
562 DBG("UST consumer sending channel %s to sessiond", channel->name);
563
564 if (channel->relayd_id != (uint64_t) -1ULL) {
565 cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
566
567 health_code_update();
568
569 /* Try to send the stream to the relayd if one is available. */
570 ret = consumer_send_relayd_stream(stream, stream->chan->pathname,
571 LTTNG_DOMAIN_UST);
572 if (ret < 0) {
573 /*
574 * Flag that the relayd was the problem here probably due to a
575 * communicaton error on the socket.
576 */
577 if (relayd_error) {
578 *relayd_error = 1;
579 }
580 ret_code = LTTCOMM_CONSUMERD_RELAYD_FAIL;
581 }
582 if (net_seq_idx == -1ULL) {
583 net_seq_idx = stream->net_seq_idx;
584 }
585 }
586 }
587
588 /* Inform sessiond that we are about to send channel and streams. */
589 ret = consumer_send_status_msg(sock, ret_code);
590 if (ret < 0 || ret_code != LTTCOMM_CONSUMERD_SUCCESS) {
591 /*
592 * Either the session daemon is not responding or the relayd died so we
593 * stop now.
594 */
595 goto error;
596 }
597
598 /* Send channel to sessiond. */
599 ret = ustctl_send_channel_to_sessiond(sock, channel->uchan);
600 if (ret < 0) {
601 goto error;
602 }
603
604 ret = ustctl_channel_close_wakeup_fd(channel->uchan);
605 if (ret < 0) {
606 goto error;
607 }
608
609 /* The channel was sent successfully to the sessiond at this point. */
610 cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
611
612 health_code_update();
613
614 /* Send stream to session daemon. */
615 ret = send_sessiond_stream(sock, stream);
616 if (ret < 0) {
617 goto error;
618 }
619 }
620
621 /* Tell sessiond there is no more stream. */
622 ret = ustctl_send_stream_to_sessiond(sock, NULL);
623 if (ret < 0) {
624 goto error;
625 }
626
627 DBG("UST consumer NULL stream sent to sessiond");
628
629 return 0;
630
631 error:
632 if (ret_code != LTTCOMM_CONSUMERD_SUCCESS) {
633 ret = -1;
634 }
635 return ret;
636 }
637
638 /*
639 * Creates a channel and streams and add the channel it to the channel internal
640 * state. The created stream must ONLY be sent once the GET_CHANNEL command is
641 * received.
642 *
643 * Return 0 on success or else, a negative value is returned and the channel
644 * MUST be destroyed by consumer_del_channel().
645 */
646 static int ask_channel(struct lttng_consumer_local_data *ctx, int sock,
647 struct lttng_consumer_channel *channel,
648 struct ustctl_consumer_channel_attr *attr)
649 {
650 int ret;
651
652 assert(ctx);
653 assert(channel);
654 assert(attr);
655
656 /*
657 * This value is still used by the kernel consumer since for the kernel,
658 * the stream ownership is not IN the consumer so we need to have the
659 * number of left stream that needs to be initialized so we can know when
660 * to delete the channel (see consumer.c).
661 *
662 * As for the user space tracer now, the consumer creates and sends the
663 * stream to the session daemon which only sends them to the application
664 * once every stream of a channel is received making this value useless
665 * because we they will be added to the poll thread before the application
666 * receives them. This ensures that a stream can not hang up during
667 * initilization of a channel.
668 */
669 channel->nb_init_stream_left = 0;
670
671 /* The reply msg status is handled in the following call. */
672 ret = create_ust_channel(channel, attr, &channel->uchan);
673 if (ret < 0) {
674 goto end;
675 }
676
677 channel->wait_fd = ustctl_channel_get_wait_fd(channel->uchan);
678
679 /*
680 * For the snapshots (no monitor), we create the metadata streams
681 * on demand, not during the channel creation.
682 */
683 if (channel->type == CONSUMER_CHANNEL_TYPE_METADATA && !channel->monitor) {
684 ret = 0;
685 goto end;
686 }
687
688 /* Open all streams for this channel. */
689 ret = create_ust_streams(channel, ctx);
690 if (ret < 0) {
691 goto end;
692 }
693
694 end:
695 return ret;
696 }
697
698 /*
699 * Send all stream of a channel to the right thread handling it.
700 *
701 * On error, return a negative value else 0 on success.
702 */
703 static int send_streams_to_thread(struct lttng_consumer_channel *channel,
704 struct lttng_consumer_local_data *ctx)
705 {
706 int ret = 0;
707 struct lttng_consumer_stream *stream, *stmp;
708
709 assert(channel);
710 assert(ctx);
711
712 /* Send streams to the corresponding thread. */
713 cds_list_for_each_entry_safe(stream, stmp, &channel->streams.head,
714 send_node) {
715
716 health_code_update();
717
718 /* Sending the stream to the thread. */
719 ret = send_stream_to_thread(stream, ctx);
720 if (ret < 0) {
721 /*
722 * If we are unable to send the stream to the thread, there is
723 * a big problem so just stop everything.
724 */
725 /* Remove node from the channel stream list. */
726 cds_list_del(&stream->send_node);
727 goto error;
728 }
729
730 /* Remove node from the channel stream list. */
731 cds_list_del(&stream->send_node);
732
733 }
734
735 error:
736 return ret;
737 }
738
739 /*
740 * Flush channel's streams using the given key to retrieve the channel.
741 *
742 * Return 0 on success else an LTTng error code.
743 */
744 static int flush_channel(uint64_t chan_key)
745 {
746 int ret = 0;
747 struct lttng_consumer_channel *channel;
748 struct lttng_consumer_stream *stream;
749 struct lttng_ht *ht;
750 struct lttng_ht_iter iter;
751
752 DBG("UST consumer flush channel key %" PRIu64, chan_key);
753
754 rcu_read_lock();
755 channel = consumer_find_channel(chan_key);
756 if (!channel) {
757 ERR("UST consumer flush channel %" PRIu64 " not found", chan_key);
758 ret = LTTNG_ERR_UST_CHAN_NOT_FOUND;
759 goto error;
760 }
761
762 ht = consumer_data.stream_per_chan_id_ht;
763
764 /* For each stream of the channel id, flush it. */
765 cds_lfht_for_each_entry_duplicate(ht->ht,
766 ht->hash_fct(&channel->key, lttng_ht_seed), ht->match_fct,
767 &channel->key, &iter.iter, stream, node_channel_id.node) {
768
769 health_code_update();
770
771 pthread_mutex_lock(&stream->lock);
772 if (!stream->quiescent) {
773 ustctl_flush_buffer(stream->ustream, 0);
774 stream->quiescent = true;
775 }
776 pthread_mutex_unlock(&stream->lock);
777 }
778 error:
779 rcu_read_unlock();
780 return ret;
781 }
782
783 /*
784 * Clear quiescent state from channel's streams using the given key to
785 * retrieve the channel.
786 *
787 * Return 0 on success else an LTTng error code.
788 */
789 static int clear_quiescent_channel(uint64_t chan_key)
790 {
791 int ret = 0;
792 struct lttng_consumer_channel *channel;
793 struct lttng_consumer_stream *stream;
794 struct lttng_ht *ht;
795 struct lttng_ht_iter iter;
796
797 DBG("UST consumer clear quiescent channel key %" PRIu64, chan_key);
798
799 rcu_read_lock();
800 channel = consumer_find_channel(chan_key);
801 if (!channel) {
802 ERR("UST consumer clear quiescent channel %" PRIu64 " not found", chan_key);
803 ret = LTTNG_ERR_UST_CHAN_NOT_FOUND;
804 goto error;
805 }
806
807 ht = consumer_data.stream_per_chan_id_ht;
808
809 /* For each stream of the channel id, clear quiescent state. */
810 cds_lfht_for_each_entry_duplicate(ht->ht,
811 ht->hash_fct(&channel->key, lttng_ht_seed), ht->match_fct,
812 &channel->key, &iter.iter, stream, node_channel_id.node) {
813
814 health_code_update();
815
816 pthread_mutex_lock(&stream->lock);
817 stream->quiescent = false;
818 pthread_mutex_unlock(&stream->lock);
819 }
820 error:
821 rcu_read_unlock();
822 return ret;
823 }
824
825 /*
826 * Close metadata stream wakeup_fd using the given key to retrieve the channel.
827 * RCU read side lock MUST be acquired before calling this function.
828 *
829 * Return 0 on success else an LTTng error code.
830 */
831 static int close_metadata(uint64_t chan_key)
832 {
833 int ret = 0;
834 struct lttng_consumer_channel *channel;
835
836 DBG("UST consumer close metadata key %" PRIu64, chan_key);
837
838 channel = consumer_find_channel(chan_key);
839 if (!channel) {
840 /*
841 * This is possible if the metadata thread has issue a delete because
842 * the endpoint point of the stream hung up. There is no way the
843 * session daemon can know about it thus use a DBG instead of an actual
844 * error.
845 */
846 DBG("UST consumer close metadata %" PRIu64 " not found", chan_key);
847 ret = LTTNG_ERR_UST_CHAN_NOT_FOUND;
848 goto error;
849 }
850
851 pthread_mutex_lock(&consumer_data.lock);
852 pthread_mutex_lock(&channel->lock);
853
854 if (cds_lfht_is_node_deleted(&channel->node.node)) {
855 goto error_unlock;
856 }
857
858 lttng_ustconsumer_close_metadata(channel);
859
860 error_unlock:
861 pthread_mutex_unlock(&channel->lock);
862 pthread_mutex_unlock(&consumer_data.lock);
863 error:
864 return ret;
865 }
866
867 /*
868 * RCU read side lock MUST be acquired before calling this function.
869 *
870 * Return 0 on success else an LTTng error code.
871 */
872 static int setup_metadata(struct lttng_consumer_local_data *ctx, uint64_t key)
873 {
874 int ret;
875 struct lttng_consumer_channel *metadata;
876
877 DBG("UST consumer setup metadata key %" PRIu64, key);
878
879 metadata = consumer_find_channel(key);
880 if (!metadata) {
881 ERR("UST consumer push metadata %" PRIu64 " not found", key);
882 ret = LTTNG_ERR_UST_CHAN_NOT_FOUND;
883 goto end;
884 }
885
886 /*
887 * In no monitor mode, the metadata channel has no stream(s) so skip the
888 * ownership transfer to the metadata thread.
889 */
890 if (!metadata->monitor) {
891 DBG("Metadata channel in no monitor");
892 ret = 0;
893 goto end;
894 }
895
896 /*
897 * Send metadata stream to relayd if one available. Availability is
898 * known if the stream is still in the list of the channel.
899 */
900 if (cds_list_empty(&metadata->streams.head)) {
901 ERR("Metadata channel key %" PRIu64 ", no stream available.", key);
902 ret = LTTCOMM_CONSUMERD_ERROR_METADATA;
903 goto error_no_stream;
904 }
905
906 /* Send metadata stream to relayd if needed. */
907 if (metadata->metadata_stream->net_seq_idx != (uint64_t) -1ULL) {
908 ret = consumer_send_relayd_stream(metadata->metadata_stream,
909 metadata->pathname, LTTNG_DOMAIN_UST);
910 if (ret < 0) {
911 ret = LTTCOMM_CONSUMERD_ERROR_METADATA;
912 goto error;
913 }
914 ret = consumer_send_relayd_streams_sent(
915 metadata->metadata_stream->net_seq_idx);
916 if (ret < 0) {
917 ret = LTTCOMM_CONSUMERD_RELAYD_FAIL;
918 goto error;
919 }
920 }
921
922 ret = send_streams_to_thread(metadata, ctx);
923 if (ret < 0) {
924 /*
925 * If we are unable to send the stream to the thread, there is
926 * a big problem so just stop everything.
927 */
928 ret = LTTCOMM_CONSUMERD_FATAL;
929 goto error;
930 }
931 /* List MUST be empty after or else it could be reused. */
932 assert(cds_list_empty(&metadata->streams.head));
933
934 ret = 0;
935 goto end;
936
937 error:
938 /*
939 * Delete metadata channel on error. At this point, the metadata stream can
940 * NOT be monitored by the metadata thread thus having the guarantee that
941 * the stream is still in the local stream list of the channel. This call
942 * will make sure to clean that list.
943 */
944 consumer_stream_destroy(metadata->metadata_stream, NULL);
945 cds_list_del(&metadata->metadata_stream->send_node);
946 metadata->metadata_stream = NULL;
947 error_no_stream:
948 end:
949 return ret;
950 }
951
952 /*
953 * Snapshot the whole metadata.
954 *
955 * Returns 0 on success, < 0 on error
956 */
957 static int snapshot_metadata(uint64_t key, char *path, uint64_t relayd_id,
958 struct lttng_consumer_local_data *ctx)
959 {
960 int ret = 0;
961 struct lttng_consumer_channel *metadata_channel;
962 struct lttng_consumer_stream *metadata_stream;
963
964 assert(path);
965 assert(ctx);
966
967 DBG("UST consumer snapshot metadata with key %" PRIu64 " at path %s",
968 key, path);
969
970 rcu_read_lock();
971
972 metadata_channel = consumer_find_channel(key);
973 if (!metadata_channel) {
974 ERR("UST snapshot metadata channel not found for key %" PRIu64,
975 key);
976 ret = -1;
977 goto error;
978 }
979 assert(!metadata_channel->monitor);
980
981 health_code_update();
982
983 /*
984 * Ask the sessiond if we have new metadata waiting and update the
985 * consumer metadata cache.
986 */
987 ret = lttng_ustconsumer_request_metadata(ctx, metadata_channel, 0, 1);
988 if (ret < 0) {
989 goto error;
990 }
991
992 health_code_update();
993
994 /*
995 * The metadata stream is NOT created in no monitor mode when the channel
996 * is created on a sessiond ask channel command.
997 */
998 ret = create_ust_streams(metadata_channel, ctx);
999 if (ret < 0) {
1000 goto error;
1001 }
1002
1003 metadata_stream = metadata_channel->metadata_stream;
1004 assert(metadata_stream);
1005
1006 if (relayd_id != (uint64_t) -1ULL) {
1007 metadata_stream->net_seq_idx = relayd_id;
1008 ret = consumer_send_relayd_stream(metadata_stream, path,
1009 LTTNG_DOMAIN_UST);
1010 if (ret < 0) {
1011 goto error_stream;
1012 }
1013 } else {
1014 ret = utils_create_stream_file(path, metadata_stream->name,
1015 metadata_stream->chan->tracefile_size,
1016 metadata_stream->tracefile_count_current,
1017 metadata_stream->uid, metadata_stream->gid, NULL);
1018 if (ret < 0) {
1019 goto error_stream;
1020 }
1021 metadata_stream->out_fd = ret;
1022 metadata_stream->tracefile_size_current = 0;
1023 }
1024
1025 do {
1026 health_code_update();
1027
1028 ret = lttng_consumer_read_subbuffer(metadata_stream, ctx);
1029 if (ret < 0) {
1030 goto error_stream;
1031 }
1032 } while (ret > 0);
1033
1034 error_stream:
1035 /*
1036 * Clean up the stream completly because the next snapshot will use a new
1037 * metadata stream.
1038 */
1039 consumer_stream_destroy(metadata_stream, NULL);
1040 cds_list_del(&metadata_stream->send_node);
1041 metadata_channel->metadata_stream = NULL;
1042
1043 error:
1044 rcu_read_unlock();
1045 return ret;
1046 }
1047
1048 /*
1049 * Take a snapshot of all the stream of a channel.
1050 *
1051 * Returns 0 on success, < 0 on error
1052 */
1053 static int snapshot_channel(uint64_t key, char *path, uint64_t relayd_id,
1054 uint64_t nb_packets_per_stream, struct lttng_consumer_local_data *ctx)
1055 {
1056 int ret;
1057 unsigned use_relayd = 0;
1058 unsigned long consumed_pos, produced_pos;
1059 struct lttng_consumer_channel *channel;
1060 struct lttng_consumer_stream *stream;
1061
1062 assert(path);
1063 assert(ctx);
1064
1065 rcu_read_lock();
1066
1067 if (relayd_id != (uint64_t) -1ULL) {
1068 use_relayd = 1;
1069 }
1070
1071 channel = consumer_find_channel(key);
1072 if (!channel) {
1073 ERR("UST snapshot channel not found for key %" PRIu64, key);
1074 ret = -1;
1075 goto error;
1076 }
1077 assert(!channel->monitor);
1078 DBG("UST consumer snapshot channel %" PRIu64, key);
1079
1080 cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
1081 health_code_update();
1082
1083 /* Lock stream because we are about to change its state. */
1084 pthread_mutex_lock(&stream->lock);
1085 stream->net_seq_idx = relayd_id;
1086
1087 if (use_relayd) {
1088 ret = consumer_send_relayd_stream(stream, path,
1089 LTTNG_DOMAIN_UST);
1090 if (ret < 0) {
1091 goto error_unlock;
1092 }
1093 } else {
1094 ret = utils_create_stream_file(path, stream->name,
1095 stream->chan->tracefile_size,
1096 stream->tracefile_count_current,
1097 stream->uid, stream->gid, NULL);
1098 if (ret < 0) {
1099 goto error_unlock;
1100 }
1101 stream->out_fd = ret;
1102 stream->tracefile_size_current = 0;
1103
1104 DBG("UST consumer snapshot stream %s/%s (%" PRIu64 ")", path,
1105 stream->name, stream->key);
1106 }
1107
1108 /*
1109 * If tracing is active, we want to perform a "full" buffer flush.
1110 * Else, if quiescent, it has already been done by the prior stop.
1111 */
1112 if (!stream->quiescent) {
1113 ustctl_flush_buffer(stream->ustream, 0);
1114 }
1115
1116 ret = lttng_ustconsumer_take_snapshot(stream);
1117 if (ret < 0) {
1118 ERR("Taking UST snapshot");
1119 goto error_unlock;
1120 }
1121
1122 ret = lttng_ustconsumer_get_produced_snapshot(stream, &produced_pos);
1123 if (ret < 0) {
1124 ERR("Produced UST snapshot position");
1125 goto error_unlock;
1126 }
1127
1128 ret = lttng_ustconsumer_get_consumed_snapshot(stream, &consumed_pos);
1129 if (ret < 0) {
1130 ERR("Consumerd UST snapshot position");
1131 goto error_unlock;
1132 }
1133
1134 /*
1135 * The original value is sent back if max stream size is larger than
1136 * the possible size of the snapshot. Also, we assume that the session
1137 * daemon should never send a maximum stream size that is lower than
1138 * subbuffer size.
1139 */
1140 consumed_pos = consumer_get_consume_start_pos(consumed_pos,
1141 produced_pos, nb_packets_per_stream,
1142 stream->max_sb_size);
1143
1144 while (consumed_pos < produced_pos) {
1145 ssize_t read_len;
1146 unsigned long len, padded_len;
1147
1148 health_code_update();
1149
1150 DBG("UST consumer taking snapshot at pos %lu", consumed_pos);
1151
1152 ret = ustctl_get_subbuf(stream->ustream, &consumed_pos);
1153 if (ret < 0) {
1154 if (ret != -EAGAIN) {
1155 PERROR("ustctl_get_subbuf snapshot");
1156 goto error_close_stream;
1157 }
1158 DBG("UST consumer get subbuf failed. Skipping it.");
1159 consumed_pos += stream->max_sb_size;
1160 stream->chan->lost_packets++;
1161 continue;
1162 }
1163
1164 ret = ustctl_get_subbuf_size(stream->ustream, &len);
1165 if (ret < 0) {
1166 ERR("Snapshot ustctl_get_subbuf_size");
1167 goto error_put_subbuf;
1168 }
1169
1170 ret = ustctl_get_padded_subbuf_size(stream->ustream, &padded_len);
1171 if (ret < 0) {
1172 ERR("Snapshot ustctl_get_padded_subbuf_size");
1173 goto error_put_subbuf;
1174 }
1175
1176 read_len = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, len,
1177 padded_len - len, NULL);
1178 if (use_relayd) {
1179 if (read_len != len) {
1180 ret = -EPERM;
1181 goto error_put_subbuf;
1182 }
1183 } else {
1184 if (read_len != padded_len) {
1185 ret = -EPERM;
1186 goto error_put_subbuf;
1187 }
1188 }
1189
1190 ret = ustctl_put_subbuf(stream->ustream);
1191 if (ret < 0) {
1192 ERR("Snapshot ustctl_put_subbuf");
1193 goto error_close_stream;
1194 }
1195 consumed_pos += stream->max_sb_size;
1196 }
1197
1198 /* Simply close the stream so we can use it on the next snapshot. */
1199 consumer_stream_close(stream);
1200 pthread_mutex_unlock(&stream->lock);
1201 }
1202
1203 rcu_read_unlock();
1204 return 0;
1205
1206 error_put_subbuf:
1207 if (ustctl_put_subbuf(stream->ustream) < 0) {
1208 ERR("Snapshot ustctl_put_subbuf");
1209 }
1210 error_close_stream:
1211 consumer_stream_close(stream);
1212 error_unlock:
1213 pthread_mutex_unlock(&stream->lock);
1214 error:
1215 rcu_read_unlock();
1216 return ret;
1217 }
1218
1219 /*
1220 * Receive the metadata updates from the sessiond. Supports receiving
1221 * overlapping metadata, but is needs to always belong to a contiguous
1222 * range starting from 0.
1223 * Be careful about the locks held when calling this function: it needs
1224 * the metadata cache flush to concurrently progress in order to
1225 * complete.
1226 */
1227 int lttng_ustconsumer_recv_metadata(int sock, uint64_t key, uint64_t offset,
1228 uint64_t len, uint64_t version,
1229 struct lttng_consumer_channel *channel, int timer, int wait)
1230 {
1231 int ret, ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1232 char *metadata_str;
1233
1234 DBG("UST consumer push metadata key %" PRIu64 " of len %" PRIu64, key, len);
1235
1236 metadata_str = zmalloc(len * sizeof(char));
1237 if (!metadata_str) {
1238 PERROR("zmalloc metadata string");
1239 ret_code = LTTCOMM_CONSUMERD_ENOMEM;
1240 goto end;
1241 }
1242
1243 health_code_update();
1244
1245 /* Receive metadata string. */
1246 ret = lttcomm_recv_unix_sock(sock, metadata_str, len);
1247 if (ret < 0) {
1248 /* Session daemon is dead so return gracefully. */
1249 ret_code = ret;
1250 goto end_free;
1251 }
1252
1253 health_code_update();
1254
1255 pthread_mutex_lock(&channel->metadata_cache->lock);
1256 ret = consumer_metadata_cache_write(channel, offset, len, version,
1257 metadata_str);
1258 if (ret < 0) {
1259 /* Unable to handle metadata. Notify session daemon. */
1260 ret_code = LTTCOMM_CONSUMERD_ERROR_METADATA;
1261 /*
1262 * Skip metadata flush on write error since the offset and len might
1263 * not have been updated which could create an infinite loop below when
1264 * waiting for the metadata cache to be flushed.
1265 */
1266 pthread_mutex_unlock(&channel->metadata_cache->lock);
1267 goto end_free;
1268 }
1269 pthread_mutex_unlock(&channel->metadata_cache->lock);
1270
1271 if (!wait) {
1272 goto end_free;
1273 }
1274 while (consumer_metadata_cache_flushed(channel, offset + len, timer)) {
1275 DBG("Waiting for metadata to be flushed");
1276
1277 health_code_update();
1278
1279 usleep(DEFAULT_METADATA_AVAILABILITY_WAIT_TIME);
1280 }
1281
1282 end_free:
1283 free(metadata_str);
1284 end:
1285 return ret_code;
1286 }
1287
1288 /*
1289 * Receive command from session daemon and process it.
1290 *
1291 * Return 1 on success else a negative value or 0.
1292 */
1293 int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
1294 int sock, struct pollfd *consumer_sockpoll)
1295 {
1296 ssize_t ret;
1297 enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1298 struct lttcomm_consumer_msg msg;
1299 struct lttng_consumer_channel *channel = NULL;
1300
1301 health_code_update();
1302
1303 ret = lttcomm_recv_unix_sock(sock, &msg, sizeof(msg));
1304 if (ret != sizeof(msg)) {
1305 DBG("Consumer received unexpected message size %zd (expects %zu)",
1306 ret, sizeof(msg));
1307 /*
1308 * The ret value might 0 meaning an orderly shutdown but this is ok
1309 * since the caller handles this.
1310 */
1311 if (ret > 0) {
1312 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD);
1313 ret = -1;
1314 }
1315 return ret;
1316 }
1317
1318 health_code_update();
1319
1320 /* deprecated */
1321 assert(msg.cmd_type != LTTNG_CONSUMER_STOP);
1322
1323 health_code_update();
1324
1325 /* relayd needs RCU read-side lock */
1326 rcu_read_lock();
1327
1328 switch (msg.cmd_type) {
1329 case LTTNG_CONSUMER_ADD_RELAYD_SOCKET:
1330 {
1331 /* Session daemon status message are handled in the following call. */
1332 consumer_add_relayd_socket(msg.u.relayd_sock.net_index,
1333 msg.u.relayd_sock.type, ctx, sock, consumer_sockpoll,
1334 &msg.u.relayd_sock.sock, msg.u.relayd_sock.session_id,
1335 msg.u.relayd_sock.relayd_session_id);
1336 goto end_nosignal;
1337 }
1338 case LTTNG_CONSUMER_DESTROY_RELAYD:
1339 {
1340 uint64_t index = msg.u.destroy_relayd.net_seq_idx;
1341 struct consumer_relayd_sock_pair *relayd;
1342
1343 DBG("UST consumer destroying relayd %" PRIu64, index);
1344
1345 /* Get relayd reference if exists. */
1346 relayd = consumer_find_relayd(index);
1347 if (relayd == NULL) {
1348 DBG("Unable to find relayd %" PRIu64, index);
1349 ret_code = LTTCOMM_CONSUMERD_RELAYD_FAIL;
1350 }
1351
1352 /*
1353 * Each relayd socket pair has a refcount of stream attached to it
1354 * which tells if the relayd is still active or not depending on the
1355 * refcount value.
1356 *
1357 * This will set the destroy flag of the relayd object and destroy it
1358 * if the refcount reaches zero when called.
1359 *
1360 * The destroy can happen either here or when a stream fd hangs up.
1361 */
1362 if (relayd) {
1363 consumer_flag_relayd_for_destroy(relayd);
1364 }
1365
1366 goto end_msg_sessiond;
1367 }
1368 case LTTNG_CONSUMER_UPDATE_STREAM:
1369 {
1370 rcu_read_unlock();
1371 return -ENOSYS;
1372 }
1373 case LTTNG_CONSUMER_DATA_PENDING:
1374 {
1375 int ret, is_data_pending;
1376 uint64_t id = msg.u.data_pending.session_id;
1377
1378 DBG("UST consumer data pending command for id %" PRIu64, id);
1379
1380 is_data_pending = consumer_data_pending(id);
1381
1382 /* Send back returned value to session daemon */
1383 ret = lttcomm_send_unix_sock(sock, &is_data_pending,
1384 sizeof(is_data_pending));
1385 if (ret < 0) {
1386 DBG("Error when sending the data pending ret code: %d", ret);
1387 goto error_fatal;
1388 }
1389
1390 /*
1391 * No need to send back a status message since the data pending
1392 * returned value is the response.
1393 */
1394 break;
1395 }
1396 case LTTNG_CONSUMER_ASK_CHANNEL_CREATION:
1397 {
1398 int ret;
1399 struct ustctl_consumer_channel_attr attr;
1400
1401 /* Create a plain object and reserve a channel key. */
1402 channel = allocate_channel(msg.u.ask_channel.session_id,
1403 msg.u.ask_channel.pathname, msg.u.ask_channel.name,
1404 msg.u.ask_channel.uid, msg.u.ask_channel.gid,
1405 msg.u.ask_channel.relayd_id, msg.u.ask_channel.key,
1406 (enum lttng_event_output) msg.u.ask_channel.output,
1407 msg.u.ask_channel.tracefile_size,
1408 msg.u.ask_channel.tracefile_count,
1409 msg.u.ask_channel.session_id_per_pid,
1410 msg.u.ask_channel.monitor,
1411 msg.u.ask_channel.live_timer_interval,
1412 msg.u.ask_channel.root_shm_path,
1413 msg.u.ask_channel.shm_path);
1414 if (!channel) {
1415 goto end_channel_error;
1416 }
1417
1418 /*
1419 * Assign UST application UID to the channel. This value is ignored for
1420 * per PID buffers. This is specific to UST thus setting this after the
1421 * allocation.
1422 */
1423 channel->ust_app_uid = msg.u.ask_channel.ust_app_uid;
1424
1425 /* Build channel attributes from received message. */
1426 attr.subbuf_size = msg.u.ask_channel.subbuf_size;
1427 attr.num_subbuf = msg.u.ask_channel.num_subbuf;
1428 attr.overwrite = msg.u.ask_channel.overwrite;
1429 attr.switch_timer_interval = msg.u.ask_channel.switch_timer_interval;
1430 attr.read_timer_interval = msg.u.ask_channel.read_timer_interval;
1431 attr.chan_id = msg.u.ask_channel.chan_id;
1432 memcpy(attr.uuid, msg.u.ask_channel.uuid, sizeof(attr.uuid));
1433 attr.blocking_timeout= msg.u.ask_channel.blocking_timeout;
1434
1435 /* Match channel buffer type to the UST abi. */
1436 switch (msg.u.ask_channel.output) {
1437 case LTTNG_EVENT_MMAP:
1438 default:
1439 attr.output = LTTNG_UST_MMAP;
1440 break;
1441 }
1442
1443 /* Translate and save channel type. */
1444 switch (msg.u.ask_channel.type) {
1445 case LTTNG_UST_CHAN_PER_CPU:
1446 channel->type = CONSUMER_CHANNEL_TYPE_DATA;
1447 attr.type = LTTNG_UST_CHAN_PER_CPU;
1448 /*
1449 * Set refcount to 1 for owner. Below, we will
1450 * pass ownership to the
1451 * consumer_thread_channel_poll() thread.
1452 */
1453 channel->refcount = 1;
1454 break;
1455 case LTTNG_UST_CHAN_METADATA:
1456 channel->type = CONSUMER_CHANNEL_TYPE_METADATA;
1457 attr.type = LTTNG_UST_CHAN_METADATA;
1458 break;
1459 default:
1460 assert(0);
1461 goto error_fatal;
1462 };
1463
1464 health_code_update();
1465
1466 ret = ask_channel(ctx, sock, channel, &attr);
1467 if (ret < 0) {
1468 goto end_channel_error;
1469 }
1470
1471 if (msg.u.ask_channel.type == LTTNG_UST_CHAN_METADATA) {
1472 ret = consumer_metadata_cache_allocate(channel);
1473 if (ret < 0) {
1474 ERR("Allocating metadata cache");
1475 goto end_channel_error;
1476 }
1477 consumer_timer_switch_start(channel, attr.switch_timer_interval);
1478 attr.switch_timer_interval = 0;
1479 } else {
1480 int monitor_start_ret;
1481
1482 consumer_timer_live_start(channel,
1483 msg.u.ask_channel.live_timer_interval);
1484 monitor_start_ret = consumer_timer_monitor_start(
1485 channel,
1486 msg.u.ask_channel.monitor_timer_interval);
1487 if (monitor_start_ret < 0) {
1488 ERR("Starting channel monitoring timer failed");
1489 goto end_channel_error;
1490 }
1491 }
1492
1493 health_code_update();
1494
1495 /*
1496 * Add the channel to the internal state AFTER all streams were created
1497 * and successfully sent to session daemon. This way, all streams must
1498 * be ready before this channel is visible to the threads.
1499 * If add_channel succeeds, ownership of the channel is
1500 * passed to consumer_thread_channel_poll().
1501 */
1502 ret = add_channel(channel, ctx);
1503 if (ret < 0) {
1504 if (msg.u.ask_channel.type == LTTNG_UST_CHAN_METADATA) {
1505 if (channel->switch_timer_enabled == 1) {
1506 consumer_timer_switch_stop(channel);
1507 }
1508 consumer_metadata_cache_destroy(channel);
1509 }
1510 if (channel->live_timer_enabled == 1) {
1511 consumer_timer_live_stop(channel);
1512 }
1513 if (channel->monitor_timer_enabled == 1) {
1514 consumer_timer_monitor_stop(channel);
1515 }
1516 goto end_channel_error;
1517 }
1518
1519 health_code_update();
1520
1521 /*
1522 * Channel and streams are now created. Inform the session daemon that
1523 * everything went well and should wait to receive the channel and
1524 * streams with ustctl API.
1525 */
1526 ret = consumer_send_status_channel(sock, channel);
1527 if (ret < 0) {
1528 /*
1529 * There is probably a problem on the socket.
1530 */
1531 goto error_fatal;
1532 }
1533
1534 break;
1535 }
1536 case LTTNG_CONSUMER_GET_CHANNEL:
1537 {
1538 int ret, relayd_err = 0;
1539 uint64_t key = msg.u.get_channel.key;
1540 struct lttng_consumer_channel *channel;
1541
1542 channel = consumer_find_channel(key);
1543 if (!channel) {
1544 ERR("UST consumer get channel key %" PRIu64 " not found", key);
1545 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
1546 goto end_msg_sessiond;
1547 }
1548
1549 health_code_update();
1550
1551 /* Send everything to sessiond. */
1552 ret = send_sessiond_channel(sock, channel, ctx, &relayd_err);
1553 if (ret < 0) {
1554 if (relayd_err) {
1555 /*
1556 * We were unable to send to the relayd the stream so avoid
1557 * sending back a fatal error to the thread since this is OK
1558 * and the consumer can continue its work. The above call
1559 * has sent the error status message to the sessiond.
1560 */
1561 goto end_nosignal;
1562 }
1563 /*
1564 * The communicaton was broken hence there is a bad state between
1565 * the consumer and sessiond so stop everything.
1566 */
1567 goto error_fatal;
1568 }
1569
1570 health_code_update();
1571
1572 /*
1573 * In no monitor mode, the streams ownership is kept inside the channel
1574 * so don't send them to the data thread.
1575 */
1576 if (!channel->monitor) {
1577 goto end_msg_sessiond;
1578 }
1579
1580 ret = send_streams_to_thread(channel, ctx);
1581 if (ret < 0) {
1582 /*
1583 * If we are unable to send the stream to the thread, there is
1584 * a big problem so just stop everything.
1585 */
1586 goto error_fatal;
1587 }
1588 /* List MUST be empty after or else it could be reused. */
1589 assert(cds_list_empty(&channel->streams.head));
1590 goto end_msg_sessiond;
1591 }
1592 case LTTNG_CONSUMER_DESTROY_CHANNEL:
1593 {
1594 uint64_t key = msg.u.destroy_channel.key;
1595
1596 /*
1597 * Only called if streams have not been sent to stream
1598 * manager thread. However, channel has been sent to
1599 * channel manager thread.
1600 */
1601 notify_thread_del_channel(ctx, key);
1602 goto end_msg_sessiond;
1603 }
1604 case LTTNG_CONSUMER_CLOSE_METADATA:
1605 {
1606 int ret;
1607
1608 ret = close_metadata(msg.u.close_metadata.key);
1609 if (ret != 0) {
1610 ret_code = ret;
1611 }
1612
1613 goto end_msg_sessiond;
1614 }
1615 case LTTNG_CONSUMER_FLUSH_CHANNEL:
1616 {
1617 int ret;
1618
1619 ret = flush_channel(msg.u.flush_channel.key);
1620 if (ret != 0) {
1621 ret_code = ret;
1622 }
1623
1624 goto end_msg_sessiond;
1625 }
1626 case LTTNG_CONSUMER_CLEAR_QUIESCENT_CHANNEL:
1627 {
1628 int ret;
1629
1630 ret = clear_quiescent_channel(
1631 msg.u.clear_quiescent_channel.key);
1632 if (ret != 0) {
1633 ret_code = ret;
1634 }
1635
1636 goto end_msg_sessiond;
1637 }
1638 case LTTNG_CONSUMER_PUSH_METADATA:
1639 {
1640 int ret;
1641 uint64_t len = msg.u.push_metadata.len;
1642 uint64_t key = msg.u.push_metadata.key;
1643 uint64_t offset = msg.u.push_metadata.target_offset;
1644 uint64_t version = msg.u.push_metadata.version;
1645 struct lttng_consumer_channel *channel;
1646
1647 DBG("UST consumer push metadata key %" PRIu64 " of len %" PRIu64, key,
1648 len);
1649
1650 channel = consumer_find_channel(key);
1651 if (!channel) {
1652 /*
1653 * This is possible if the metadata creation on the consumer side
1654 * is in flight vis-a-vis a concurrent push metadata from the
1655 * session daemon. Simply return that the channel failed and the
1656 * session daemon will handle that message correctly considering
1657 * that this race is acceptable thus the DBG() statement here.
1658 */
1659 DBG("UST consumer push metadata %" PRIu64 " not found", key);
1660 ret_code = LTTCOMM_CONSUMERD_CHANNEL_FAIL;
1661 goto end_msg_sessiond;
1662 }
1663
1664 health_code_update();
1665
1666 if (!len) {
1667 /*
1668 * There is nothing to receive. We have simply
1669 * checked whether the channel can be found.
1670 */
1671 ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1672 goto end_msg_sessiond;
1673 }
1674
1675 /* Tell session daemon we are ready to receive the metadata. */
1676 ret = consumer_send_status_msg(sock, LTTCOMM_CONSUMERD_SUCCESS);
1677 if (ret < 0) {
1678 /* Somehow, the session daemon is not responding anymore. */
1679 goto error_fatal;
1680 }
1681
1682 health_code_update();
1683
1684 /* Wait for more data. */
1685 health_poll_entry();
1686 ret = lttng_consumer_poll_socket(consumer_sockpoll);
1687 health_poll_exit();
1688 if (ret) {
1689 goto error_fatal;
1690 }
1691
1692 health_code_update();
1693
1694 ret = lttng_ustconsumer_recv_metadata(sock, key, offset,
1695 len, version, channel, 0, 1);
1696 if (ret < 0) {
1697 /* error receiving from sessiond */
1698 goto error_fatal;
1699 } else {
1700 ret_code = ret;
1701 goto end_msg_sessiond;
1702 }
1703 }
1704 case LTTNG_CONSUMER_SETUP_METADATA:
1705 {
1706 int ret;
1707
1708 ret = setup_metadata(ctx, msg.u.setup_metadata.key);
1709 if (ret) {
1710 ret_code = ret;
1711 }
1712 goto end_msg_sessiond;
1713 }
1714 case LTTNG_CONSUMER_SNAPSHOT_CHANNEL:
1715 {
1716 if (msg.u.snapshot_channel.metadata) {
1717 ret = snapshot_metadata(msg.u.snapshot_channel.key,
1718 msg.u.snapshot_channel.pathname,
1719 msg.u.snapshot_channel.relayd_id,
1720 ctx);
1721 if (ret < 0) {
1722 ERR("Snapshot metadata failed");
1723 ret_code = LTTCOMM_CONSUMERD_ERROR_METADATA;
1724 }
1725 } else {
1726 ret = snapshot_channel(msg.u.snapshot_channel.key,
1727 msg.u.snapshot_channel.pathname,
1728 msg.u.snapshot_channel.relayd_id,
1729 msg.u.snapshot_channel.nb_packets_per_stream,
1730 ctx);
1731 if (ret < 0) {
1732 ERR("Snapshot channel failed");
1733 ret_code = LTTCOMM_CONSUMERD_CHANNEL_FAIL;
1734 }
1735 }
1736
1737 health_code_update();
1738 ret = consumer_send_status_msg(sock, ret_code);
1739 if (ret < 0) {
1740 /* Somehow, the session daemon is not responding anymore. */
1741 goto end_nosignal;
1742 }
1743 health_code_update();
1744 break;
1745 }
1746 case LTTNG_CONSUMER_DISCARDED_EVENTS:
1747 {
1748 int ret = 0;
1749 uint64_t discarded_events;
1750 struct lttng_ht_iter iter;
1751 struct lttng_ht *ht;
1752 struct lttng_consumer_stream *stream;
1753 uint64_t id = msg.u.discarded_events.session_id;
1754 uint64_t key = msg.u.discarded_events.channel_key;
1755
1756 DBG("UST consumer discarded events command for session id %"
1757 PRIu64, id);
1758 rcu_read_lock();
1759 pthread_mutex_lock(&consumer_data.lock);
1760
1761 ht = consumer_data.stream_list_ht;
1762
1763 /*
1764 * We only need a reference to the channel, but they are not
1765 * directly indexed, so we just use the first matching stream
1766 * to extract the information we need, we default to 0 if not
1767 * found (no events are dropped if the channel is not yet in
1768 * use).
1769 */
1770 discarded_events = 0;
1771 cds_lfht_for_each_entry_duplicate(ht->ht,
1772 ht->hash_fct(&id, lttng_ht_seed),
1773 ht->match_fct, &id,
1774 &iter.iter, stream, node_session_id.node) {
1775 if (stream->chan->key == key) {
1776 discarded_events = stream->chan->discarded_events;
1777 break;
1778 }
1779 }
1780 pthread_mutex_unlock(&consumer_data.lock);
1781 rcu_read_unlock();
1782
1783 DBG("UST consumer discarded events command for session id %"
1784 PRIu64 ", channel key %" PRIu64, id, key);
1785
1786 health_code_update();
1787
1788 /* Send back returned value to session daemon */
1789 ret = lttcomm_send_unix_sock(sock, &discarded_events, sizeof(discarded_events));
1790 if (ret < 0) {
1791 PERROR("send discarded events");
1792 goto error_fatal;
1793 }
1794
1795 break;
1796 }
1797 case LTTNG_CONSUMER_LOST_PACKETS:
1798 {
1799 int ret;
1800 uint64_t lost_packets;
1801 struct lttng_ht_iter iter;
1802 struct lttng_ht *ht;
1803 struct lttng_consumer_stream *stream;
1804 uint64_t id = msg.u.lost_packets.session_id;
1805 uint64_t key = msg.u.lost_packets.channel_key;
1806
1807 DBG("UST consumer lost packets command for session id %"
1808 PRIu64, id);
1809 rcu_read_lock();
1810 pthread_mutex_lock(&consumer_data.lock);
1811
1812 ht = consumer_data.stream_list_ht;
1813
1814 /*
1815 * We only need a reference to the channel, but they are not
1816 * directly indexed, so we just use the first matching stream
1817 * to extract the information we need, we default to 0 if not
1818 * found (no packets lost if the channel is not yet in use).
1819 */
1820 lost_packets = 0;
1821 cds_lfht_for_each_entry_duplicate(ht->ht,
1822 ht->hash_fct(&id, lttng_ht_seed),
1823 ht->match_fct, &id,
1824 &iter.iter, stream, node_session_id.node) {
1825 if (stream->chan->key == key) {
1826 lost_packets = stream->chan->lost_packets;
1827 break;
1828 }
1829 }
1830 pthread_mutex_unlock(&consumer_data.lock);
1831 rcu_read_unlock();
1832
1833 DBG("UST consumer lost packets command for session id %"
1834 PRIu64 ", channel key %" PRIu64, id, key);
1835
1836 health_code_update();
1837
1838 /* Send back returned value to session daemon */
1839 ret = lttcomm_send_unix_sock(sock, &lost_packets,
1840 sizeof(lost_packets));
1841 if (ret < 0) {
1842 PERROR("send lost packets");
1843 goto error_fatal;
1844 }
1845
1846 break;
1847 }
1848 case LTTNG_CONSUMER_SET_CHANNEL_MONITOR_PIPE:
1849 {
1850 int channel_monitor_pipe;
1851
1852 ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1853 /* Successfully received the command's type. */
1854 ret = consumer_send_status_msg(sock, ret_code);
1855 if (ret < 0) {
1856 goto error_fatal;
1857 }
1858
1859 ret = lttcomm_recv_fds_unix_sock(sock, &channel_monitor_pipe,
1860 1);
1861 if (ret != sizeof(channel_monitor_pipe)) {
1862 ERR("Failed to receive channel monitor pipe");
1863 goto error_fatal;
1864 }
1865
1866 DBG("Received channel monitor pipe (%d)", channel_monitor_pipe);
1867 ret = consumer_timer_thread_set_channel_monitor_pipe(
1868 channel_monitor_pipe);
1869 if (!ret) {
1870 int flags;
1871
1872 ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1873 /* Set the pipe as non-blocking. */
1874 ret = fcntl(channel_monitor_pipe, F_GETFL, 0);
1875 if (ret == -1) {
1876 PERROR("fcntl get flags of the channel monitoring pipe");
1877 goto error_fatal;
1878 }
1879 flags = ret;
1880
1881 ret = fcntl(channel_monitor_pipe, F_SETFL,
1882 flags | O_NONBLOCK);
1883 if (ret == -1) {
1884 PERROR("fcntl set O_NONBLOCK flag of the channel monitoring pipe");
1885 goto error_fatal;
1886 }
1887 DBG("Channel monitor pipe set as non-blocking");
1888 } else {
1889 ret_code = LTTCOMM_CONSUMERD_ALREADY_SET;
1890 }
1891 goto end_msg_sessiond;
1892 }
1893 case LTTNG_CONSUMER_SET_CHANNEL_ROTATE_PIPE:
1894 {
1895 int channel_rotate_pipe;
1896 int flags;
1897
1898 ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1899 /* Successfully received the command's type. */
1900 ret = consumer_send_status_msg(sock, ret_code);
1901 if (ret < 0) {
1902 goto error_fatal;
1903 }
1904
1905 ret = lttcomm_recv_fds_unix_sock(sock, &channel_rotate_pipe,
1906 1);
1907 if (ret != sizeof(channel_rotate_pipe)) {
1908 ERR("Failed to receive channel rotate pipe");
1909 goto error_fatal;
1910 }
1911
1912 DBG("Received channel rotate pipe (%d)", channel_rotate_pipe);
1913 ctx->channel_rotate_pipe = channel_rotate_pipe;
1914 /* Set the pipe as non-blocking. */
1915 ret = fcntl(channel_rotate_pipe, F_GETFL, 0);
1916 if (ret == -1) {
1917 PERROR("fcntl get flags of the channel rotate pipe");
1918 goto error_fatal;
1919 }
1920 flags = ret;
1921
1922 ret = fcntl(channel_rotate_pipe, F_SETFL,
1923 flags | O_NONBLOCK);
1924 if (ret == -1) {
1925 PERROR("fcntl set O_NONBLOCK flag of the channel rotate pipe");
1926 goto error_fatal;
1927 }
1928 DBG("Channel rotate pipe set as non-blocking");
1929 ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1930 ret = consumer_send_status_msg(sock, ret_code);
1931 if (ret < 0) {
1932 goto error_fatal;
1933 }
1934 break;
1935 }
1936 case LTTNG_CONSUMER_ROTATE_CHANNEL:
1937 {
1938 ret = lttng_consumer_rotate_channel(msg.u.rotate_channel.key,
1939 msg.u.rotate_channel.pathname,
1940 msg.u.rotate_channel.relayd_id,
1941 msg.u.rotate_channel.metadata,
1942 msg.u.rotate_channel.new_chunk_id,
1943 ctx);
1944 if (ret < 0) {
1945 ERR("Rotate channel failed");
1946 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
1947 }
1948
1949 health_code_update();
1950
1951 ret = consumer_send_status_msg(sock, ret_code);
1952 if (ret < 0) {
1953 /* Somehow, the session daemon is not responding anymore. */
1954 goto end_nosignal;
1955 }
1956
1957 /*
1958 * Rotate the streams that are ready right now.
1959 * FIXME: this is a second consecutive iteration over the
1960 * streams in a channel, there is probably a better way to
1961 * handle this, but it needs to be after the
1962 * consumer_send_status_msg() call.
1963 */
1964 ret = lttng_consumer_rotate_ready_streams(
1965 msg.u.rotate_channel.key, ctx);
1966 if (ret < 0) {
1967 ERR("Rotate channel failed");
1968 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
1969 }
1970 break;
1971 }
1972 case LTTNG_CONSUMER_ROTATE_RENAME:
1973 {
1974 DBG("Consumer rename session %" PRIu64 " after rotation",
1975 msg.u.rotate_rename.session_id);
1976 ret = lttng_consumer_rotate_rename(msg.u.rotate_rename.current_path,
1977 msg.u.rotate_rename.new_path,
1978 msg.u.rotate_rename.uid,
1979 msg.u.rotate_rename.gid,
1980 msg.u.rotate_rename.relayd_id);
1981 if (ret < 0) {
1982 ERR("Rotate rename failed");
1983 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
1984 }
1985
1986 health_code_update();
1987
1988 ret = consumer_send_status_msg(sock, ret_code);
1989 if (ret < 0) {
1990 /* Somehow, the session daemon is not responding anymore. */
1991 goto end_nosignal;
1992 }
1993
1994 }
1995 case LTTNG_CONSUMER_ROTATE_PENDING_RELAY:
1996 {
1997 DBG("Consumer rotate pending on relay for session %" PRIu64,
1998 msg.u.rotate_pending_relay.session_id);
1999 ret = lttng_consumer_rotate_pending_relay(
2000 msg.u.rotate_pending_relay.session_id,
2001 msg.u.rotate_pending_relay.relayd_id,
2002 msg.u.rotate_pending_relay.chunk_id);
2003 if (ret < 0) {
2004 ERR("Rotate pending relay failed");
2005 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
2006 }
2007
2008 health_code_update();
2009 ret_code = ret;
2010
2011 ret = consumer_send_status_msg(sock, ret_code);
2012 if (ret < 0) {
2013 /* Somehow, the session daemon is not responding anymore. */
2014 goto end_nosignal;
2015 }
2016
2017 }
2018 default:
2019 break;
2020 }
2021
2022 end_nosignal:
2023 rcu_read_unlock();
2024
2025 health_code_update();
2026
2027 /*
2028 * Return 1 to indicate success since the 0 value can be a socket
2029 * shutdown during the recv() or send() call.
2030 */
2031 return 1;
2032
2033 end_msg_sessiond:
2034 /*
2035 * The returned value here is not useful since either way we'll return 1 to
2036 * the caller because the session daemon socket management is done
2037 * elsewhere. Returning a negative code or 0 will shutdown the consumer.
2038 */
2039 ret = consumer_send_status_msg(sock, ret_code);
2040 if (ret < 0) {
2041 goto error_fatal;
2042 }
2043 rcu_read_unlock();
2044
2045 health_code_update();
2046
2047 return 1;
2048 end_channel_error:
2049 if (channel) {
2050 /*
2051 * Free channel here since no one has a reference to it. We don't
2052 * free after that because a stream can store this pointer.
2053 */
2054 destroy_channel(channel);
2055 }
2056 /* We have to send a status channel message indicating an error. */
2057 ret = consumer_send_status_channel(sock, NULL);
2058 if (ret < 0) {
2059 /* Stop everything if session daemon can not be notified. */
2060 goto error_fatal;
2061 }
2062 rcu_read_unlock();
2063
2064 health_code_update();
2065
2066 return 1;
2067 error_fatal:
2068 rcu_read_unlock();
2069 /* This will issue a consumer stop. */
2070 return -1;
2071 }
2072
2073 /*
2074 * Wrapper over the mmap() read offset from ust-ctl library. Since this can be
2075 * compiled out, we isolate it in this library.
2076 */
2077 int lttng_ustctl_get_mmap_read_offset(struct lttng_consumer_stream *stream,
2078 unsigned long *off)
2079 {
2080 assert(stream);
2081 assert(stream->ustream);
2082
2083 return ustctl_get_mmap_read_offset(stream->ustream, off);
2084 }
2085
2086 /*
2087 * Wrapper over the mmap() read offset from ust-ctl library. Since this can be
2088 * compiled out, we isolate it in this library.
2089 */
2090 void *lttng_ustctl_get_mmap_base(struct lttng_consumer_stream *stream)
2091 {
2092 assert(stream);
2093 assert(stream->ustream);
2094
2095 return ustctl_get_mmap_base(stream->ustream);
2096 }
2097
2098 void lttng_ustctl_flush_buffer(struct lttng_consumer_stream *stream,
2099 int producer_active)
2100 {
2101 assert(stream);
2102 assert(stream->ustream);
2103
2104 ustctl_flush_buffer(stream->ustream, producer_active);
2105 }
2106
2107 /*
2108 * Take a snapshot for a specific stream.
2109 *
2110 * Returns 0 on success, < 0 on error
2111 */
2112 int lttng_ustconsumer_take_snapshot(struct lttng_consumer_stream *stream)
2113 {
2114 assert(stream);
2115 assert(stream->ustream);
2116
2117 return ustctl_snapshot(stream->ustream);
2118 }
2119
2120 /*
2121 * Sample consumed and produced positions for a specific stream.
2122 *
2123 * Returns 0 on success, < 0 on error.
2124 */
2125 int lttng_ustconsumer_sample_snapshot_positions(
2126 struct lttng_consumer_stream *stream)
2127 {
2128 assert(stream);
2129 assert(stream->ustream);
2130
2131 return ustctl_snapshot_sample_positions(stream->ustream);
2132 }
2133
2134 /*
2135 * Get the produced position
2136 *
2137 * Returns 0 on success, < 0 on error
2138 */
2139 int lttng_ustconsumer_get_produced_snapshot(
2140 struct lttng_consumer_stream *stream, unsigned long *pos)
2141 {
2142 assert(stream);
2143 assert(stream->ustream);
2144 assert(pos);
2145
2146 return ustctl_snapshot_get_produced(stream->ustream, pos);
2147 }
2148
2149 /*
2150 * Get the consumed position
2151 *
2152 * Returns 0 on success, < 0 on error
2153 */
2154 int lttng_ustconsumer_get_consumed_snapshot(
2155 struct lttng_consumer_stream *stream, unsigned long *pos)
2156 {
2157 assert(stream);
2158 assert(stream->ustream);
2159 assert(pos);
2160
2161 return ustctl_snapshot_get_consumed(stream->ustream, pos);
2162 }
2163
2164 void lttng_ustconsumer_flush_buffer(struct lttng_consumer_stream *stream,
2165 int producer)
2166 {
2167 assert(stream);
2168 assert(stream->ustream);
2169
2170 ustctl_flush_buffer(stream->ustream, producer);
2171 }
2172
2173 int lttng_ustconsumer_get_current_timestamp(
2174 struct lttng_consumer_stream *stream, uint64_t *ts)
2175 {
2176 assert(stream);
2177 assert(stream->ustream);
2178 assert(ts);
2179
2180 return ustctl_get_current_timestamp(stream->ustream, ts);
2181 }
2182
2183 int lttng_ustconsumer_get_sequence_number(
2184 struct lttng_consumer_stream *stream, uint64_t *seq)
2185 {
2186 assert(stream);
2187 assert(stream->ustream);
2188 assert(seq);
2189
2190 return ustctl_get_sequence_number(stream->ustream, seq);
2191 }
2192
2193 /*
2194 * Called when the stream signals the consumer that it has hung up.
2195 */
2196 void lttng_ustconsumer_on_stream_hangup(struct lttng_consumer_stream *stream)
2197 {
2198 assert(stream);
2199 assert(stream->ustream);
2200
2201 pthread_mutex_lock(&stream->lock);
2202 if (!stream->quiescent) {
2203 ustctl_flush_buffer(stream->ustream, 0);
2204 stream->quiescent = true;
2205 }
2206 pthread_mutex_unlock(&stream->lock);
2207 stream->hangup_flush_done = 1;
2208 }
2209
2210 void lttng_ustconsumer_del_channel(struct lttng_consumer_channel *chan)
2211 {
2212 int i;
2213
2214 assert(chan);
2215 assert(chan->uchan);
2216
2217 if (chan->switch_timer_enabled == 1) {
2218 consumer_timer_switch_stop(chan);
2219 }
2220 for (i = 0; i < chan->nr_stream_fds; i++) {
2221 int ret;
2222
2223 ret = close(chan->stream_fds[i]);
2224 if (ret) {
2225 PERROR("close");
2226 }
2227 if (chan->shm_path[0]) {
2228 char shm_path[PATH_MAX];
2229
2230 ret = get_stream_shm_path(shm_path, chan->shm_path, i);
2231 if (ret) {
2232 ERR("Cannot get stream shm path");
2233 }
2234 ret = run_as_unlink(shm_path, chan->uid, chan->gid);
2235 if (ret) {
2236 PERROR("unlink %s", shm_path);
2237 }
2238 }
2239 }
2240 }
2241
2242 void lttng_ustconsumer_free_channel(struct lttng_consumer_channel *chan)
2243 {
2244 assert(chan);
2245 assert(chan->uchan);
2246
2247 consumer_metadata_cache_destroy(chan);
2248 ustctl_destroy_channel(chan->uchan);
2249 /* Try to rmdir all directories under shm_path root. */
2250 if (chan->root_shm_path[0]) {
2251 (void) run_as_rmdir_recursive(chan->root_shm_path,
2252 chan->uid, chan->gid);
2253 }
2254 free(chan->stream_fds);
2255 }
2256
2257 void lttng_ustconsumer_del_stream(struct lttng_consumer_stream *stream)
2258 {
2259 assert(stream);
2260 assert(stream->ustream);
2261
2262 if (stream->chan->switch_timer_enabled == 1) {
2263 consumer_timer_switch_stop(stream->chan);
2264 }
2265 ustctl_destroy_stream(stream->ustream);
2266 }
2267
2268 int lttng_ustconsumer_get_wakeup_fd(struct lttng_consumer_stream *stream)
2269 {
2270 assert(stream);
2271 assert(stream->ustream);
2272
2273 return ustctl_stream_get_wakeup_fd(stream->ustream);
2274 }
2275
2276 int lttng_ustconsumer_close_wakeup_fd(struct lttng_consumer_stream *stream)
2277 {
2278 assert(stream);
2279 assert(stream->ustream);
2280
2281 return ustctl_stream_close_wakeup_fd(stream->ustream);
2282 }
2283
2284 /*
2285 * Populate index values of a UST stream. Values are set in big endian order.
2286 *
2287 * Return 0 on success or else a negative value.
2288 */
2289 static int get_index_values(struct ctf_packet_index *index,
2290 struct ustctl_consumer_stream *ustream)
2291 {
2292 int ret;
2293
2294 ret = ustctl_get_timestamp_begin(ustream, &index->timestamp_begin);
2295 if (ret < 0) {
2296 PERROR("ustctl_get_timestamp_begin");
2297 goto error;
2298 }
2299 index->timestamp_begin = htobe64(index->timestamp_begin);
2300
2301 ret = ustctl_get_timestamp_end(ustream, &index->timestamp_end);
2302 if (ret < 0) {
2303 PERROR("ustctl_get_timestamp_end");
2304 goto error;
2305 }
2306 index->timestamp_end = htobe64(index->timestamp_end);
2307
2308 ret = ustctl_get_events_discarded(ustream, &index->events_discarded);
2309 if (ret < 0) {
2310 PERROR("ustctl_get_events_discarded");
2311 goto error;
2312 }
2313 index->events_discarded = htobe64(index->events_discarded);
2314
2315 ret = ustctl_get_content_size(ustream, &index->content_size);
2316 if (ret < 0) {
2317 PERROR("ustctl_get_content_size");
2318 goto error;
2319 }
2320 index->content_size = htobe64(index->content_size);
2321
2322 ret = ustctl_get_packet_size(ustream, &index->packet_size);
2323 if (ret < 0) {
2324 PERROR("ustctl_get_packet_size");
2325 goto error;
2326 }
2327 index->packet_size = htobe64(index->packet_size);
2328
2329 ret = ustctl_get_stream_id(ustream, &index->stream_id);
2330 if (ret < 0) {
2331 PERROR("ustctl_get_stream_id");
2332 goto error;
2333 }
2334 index->stream_id = htobe64(index->stream_id);
2335
2336 ret = ustctl_get_instance_id(ustream, &index->stream_instance_id);
2337 if (ret < 0) {
2338 PERROR("ustctl_get_instance_id");
2339 goto error;
2340 }
2341 index->stream_instance_id = htobe64(index->stream_instance_id);
2342
2343 ret = ustctl_get_sequence_number(ustream, &index->packet_seq_num);
2344 if (ret < 0) {
2345 PERROR("ustctl_get_sequence_number");
2346 goto error;
2347 }
2348 index->packet_seq_num = htobe64(index->packet_seq_num);
2349
2350 error:
2351 return ret;
2352 }
2353
2354 static
2355 void metadata_stream_reset_cache(struct lttng_consumer_stream *stream,
2356 struct consumer_metadata_cache *cache)
2357 {
2358 DBG("Metadata stream update to version %" PRIu64,
2359 cache->version);
2360 stream->ust_metadata_pushed = 0;
2361 stream->metadata_version = cache->version;
2362 stream->reset_metadata_flag = 1;
2363 }
2364
2365 /*
2366 * Check if the version of the metadata stream and metadata cache match.
2367 * If the cache got updated, reset the metadata stream.
2368 * The stream lock and metadata cache lock MUST be held.
2369 * Return 0 on success, a negative value on error.
2370 */
2371 static
2372 int metadata_stream_check_version(struct lttng_consumer_stream *stream)
2373 {
2374 int ret = 0;
2375 struct consumer_metadata_cache *cache = stream->chan->metadata_cache;
2376
2377 if (cache->version == stream->metadata_version) {
2378 goto end;
2379 }
2380 metadata_stream_reset_cache(stream, cache);
2381
2382 end:
2383 return ret;
2384 }
2385
2386 /*
2387 * Write up to one packet from the metadata cache to the channel.
2388 *
2389 * Returns the number of bytes pushed in the cache, or a negative value
2390 * on error.
2391 */
2392 static
2393 int commit_one_metadata_packet(struct lttng_consumer_stream *stream)
2394 {
2395 ssize_t write_len;
2396 int ret;
2397
2398 pthread_mutex_lock(&stream->chan->metadata_cache->lock);
2399 ret = metadata_stream_check_version(stream);
2400 if (ret < 0) {
2401 goto end;
2402 }
2403 if (stream->chan->metadata_cache->max_offset
2404 == stream->ust_metadata_pushed) {
2405 ret = 0;
2406 goto end;
2407 }
2408
2409 write_len = ustctl_write_one_packet_to_channel(stream->chan->uchan,
2410 &stream->chan->metadata_cache->data[stream->ust_metadata_pushed],
2411 stream->chan->metadata_cache->max_offset
2412 - stream->ust_metadata_pushed);
2413 assert(write_len != 0);
2414 if (write_len < 0) {
2415 ERR("Writing one metadata packet");
2416 ret = -1;
2417 goto end;
2418 }
2419 stream->ust_metadata_pushed += write_len;
2420
2421 assert(stream->chan->metadata_cache->max_offset >=
2422 stream->ust_metadata_pushed);
2423 ret = write_len;
2424
2425 end:
2426 pthread_mutex_unlock(&stream->chan->metadata_cache->lock);
2427 return ret;
2428 }
2429
2430
2431 /*
2432 * Sync metadata meaning request them to the session daemon and snapshot to the
2433 * metadata thread can consumer them.
2434 *
2435 * Metadata stream lock is held here, but we need to release it when
2436 * interacting with sessiond, else we cause a deadlock with live
2437 * awaiting on metadata to be pushed out.
2438 *
2439 * Return 0 if new metadatda is available, EAGAIN if the metadata stream
2440 * is empty or a negative value on error.
2441 */
2442 int lttng_ustconsumer_sync_metadata(struct lttng_consumer_local_data *ctx,
2443 struct lttng_consumer_stream *metadata)
2444 {
2445 int ret;
2446 int retry = 0;
2447
2448 assert(ctx);
2449 assert(metadata);
2450
2451 pthread_mutex_unlock(&metadata->lock);
2452 /*
2453 * Request metadata from the sessiond, but don't wait for the flush
2454 * because we locked the metadata thread.
2455 */
2456 ret = lttng_ustconsumer_request_metadata(ctx, metadata->chan, 0, 0);
2457 pthread_mutex_lock(&metadata->lock);
2458 if (ret < 0) {
2459 goto end;
2460 }
2461
2462 ret = commit_one_metadata_packet(metadata);
2463 if (ret <= 0) {
2464 goto end;
2465 } else if (ret > 0) {
2466 retry = 1;
2467 }
2468
2469 ustctl_flush_buffer(metadata->ustream, 1);
2470 ret = ustctl_snapshot(metadata->ustream);
2471 if (ret < 0) {
2472 if (errno != EAGAIN) {
2473 ERR("Sync metadata, taking UST snapshot");
2474 goto end;
2475 }
2476 DBG("No new metadata when syncing them.");
2477 /* No new metadata, exit. */
2478 ret = ENODATA;
2479 goto end;
2480 }
2481
2482 /*
2483 * After this flush, we still need to extract metadata.
2484 */
2485 if (retry) {
2486 ret = EAGAIN;
2487 }
2488
2489 end:
2490 return ret;
2491 }
2492
2493 /*
2494 * Return 0 on success else a negative value.
2495 */
2496 static int notify_if_more_data(struct lttng_consumer_stream *stream,
2497 struct lttng_consumer_local_data *ctx)
2498 {
2499 int ret;
2500 struct ustctl_consumer_stream *ustream;
2501
2502 assert(stream);
2503 assert(ctx);
2504
2505 ustream = stream->ustream;
2506
2507 /*
2508 * First, we are going to check if there is a new subbuffer available
2509 * before reading the stream wait_fd.
2510 */
2511 /* Get the next subbuffer */
2512 ret = ustctl_get_next_subbuf(ustream);
2513 if (ret) {
2514 /* No more data found, flag the stream. */
2515 stream->has_data = 0;
2516 ret = 0;
2517 goto end;
2518 }
2519
2520 ret = ustctl_put_subbuf(ustream);
2521 assert(!ret);
2522
2523 /* This stream still has data. Flag it and wake up the data thread. */
2524 stream->has_data = 1;
2525
2526 if (stream->monitor && !stream->hangup_flush_done && !ctx->has_wakeup) {
2527 ssize_t writelen;
2528
2529 writelen = lttng_pipe_write(ctx->consumer_wakeup_pipe, "!", 1);
2530 if (writelen < 0 && errno != EAGAIN && errno != EWOULDBLOCK) {
2531 ret = writelen;
2532 goto end;
2533 }
2534
2535 /* The wake up pipe has been notified. */
2536 ctx->has_wakeup = 1;
2537 }
2538 ret = 0;
2539
2540 end:
2541 return ret;
2542 }
2543
2544 static
2545 int update_stream_stats(struct lttng_consumer_stream *stream)
2546 {
2547 int ret;
2548 uint64_t seq, discarded;
2549
2550 ret = ustctl_get_sequence_number(stream->ustream, &seq);
2551 if (ret < 0) {
2552 PERROR("ustctl_get_sequence_number");
2553 goto end;
2554 }
2555 /*
2556 * Start the sequence when we extract the first packet in case we don't
2557 * start at 0 (for example if a consumer is not connected to the
2558 * session immediately after the beginning).
2559 */
2560 if (stream->last_sequence_number == -1ULL) {
2561 stream->last_sequence_number = seq;
2562 } else if (seq > stream->last_sequence_number) {
2563 stream->chan->lost_packets += seq -
2564 stream->last_sequence_number - 1;
2565 } else {
2566 /* seq <= last_sequence_number */
2567 ERR("Sequence number inconsistent : prev = %" PRIu64
2568 ", current = %" PRIu64,
2569 stream->last_sequence_number, seq);
2570 ret = -1;
2571 goto end;
2572 }
2573 stream->last_sequence_number = seq;
2574
2575 ret = ustctl_get_events_discarded(stream->ustream, &discarded);
2576 if (ret < 0) {
2577 PERROR("kernctl_get_events_discarded");
2578 goto end;
2579 }
2580 if (discarded < stream->last_discarded_events) {
2581 /*
2582 * Overflow has occurred. We assume only one wrap-around
2583 * has occurred.
2584 */
2585 stream->chan->discarded_events +=
2586 (1ULL << (CAA_BITS_PER_LONG - 1)) -
2587 stream->last_discarded_events + discarded;
2588 } else {
2589 stream->chan->discarded_events += discarded -
2590 stream->last_discarded_events;
2591 }
2592 stream->last_discarded_events = discarded;
2593 ret = 0;
2594
2595 end:
2596 return ret;
2597 }
2598
2599 /*
2600 * Read subbuffer from the given stream.
2601 *
2602 * Stream lock MUST be acquired.
2603 *
2604 * Return 0 on success else a negative value.
2605 */
2606 int lttng_ustconsumer_read_subbuffer(struct lttng_consumer_stream *stream,
2607 struct lttng_consumer_local_data *ctx)
2608 {
2609 unsigned long len, subbuf_size, padding;
2610 int err, write_index = 1, rotation_ret, rotate_ready;
2611 long ret = 0;
2612 struct ustctl_consumer_stream *ustream;
2613 struct ctf_packet_index index;
2614
2615 assert(stream);
2616 assert(stream->ustream);
2617 assert(ctx);
2618
2619 DBG("In UST read_subbuffer (wait_fd: %d, name: %s)", stream->wait_fd,
2620 stream->name);
2621
2622 /* Ease our life for what's next. */
2623 ustream = stream->ustream;
2624
2625 /*
2626 * We can consume the 1 byte written into the wait_fd by UST. Don't trigger
2627 * error if we cannot read this one byte (read returns 0), or if the error
2628 * is EAGAIN or EWOULDBLOCK.
2629 *
2630 * This is only done when the stream is monitored by a thread, before the
2631 * flush is done after a hangup and if the stream is not flagged with data
2632 * since there might be nothing to consume in the wait fd but still have
2633 * data available flagged by the consumer wake up pipe.
2634 */
2635 if (stream->monitor && !stream->hangup_flush_done && !stream->has_data) {
2636 char dummy;
2637 ssize_t readlen;
2638
2639 readlen = lttng_read(stream->wait_fd, &dummy, 1);
2640 if (readlen < 0 && errno != EAGAIN && errno != EWOULDBLOCK) {
2641 ret = readlen;
2642 goto error;
2643 }
2644 }
2645
2646 retry:
2647 /* Get the next subbuffer */
2648 err = ustctl_get_next_subbuf(ustream);
2649 if (err != 0) {
2650 /*
2651 * Populate metadata info if the existing info has
2652 * already been read.
2653 */
2654 if (stream->metadata_flag) {
2655 ret = commit_one_metadata_packet(stream);
2656 if (ret <= 0) {
2657 goto error;
2658 }
2659 ustctl_flush_buffer(stream->ustream, 1);
2660 goto retry;
2661 }
2662
2663 ret = err; /* ustctl_get_next_subbuf returns negative, caller expect positive. */
2664 /*
2665 * This is a debug message even for single-threaded consumer,
2666 * because poll() have more relaxed criterions than get subbuf,
2667 * so get_subbuf may fail for short race windows where poll()
2668 * would issue wakeups.
2669 */
2670 DBG("Reserving sub buffer failed (everything is normal, "
2671 "it is due to concurrency) [ret: %d]", err);
2672 goto error;
2673 }
2674 assert(stream->chan->output == CONSUMER_CHANNEL_MMAP);
2675
2676 if (!stream->metadata_flag) {
2677 index.offset = htobe64(stream->out_fd_offset);
2678 ret = get_index_values(&index, ustream);
2679 if (ret < 0) {
2680 err = ustctl_put_subbuf(ustream);
2681 assert(err == 0);
2682 goto error;
2683 }
2684
2685 /* Update the stream's sequence and discarded events count. */
2686 ret = update_stream_stats(stream);
2687 if (ret < 0) {
2688 PERROR("kernctl_get_events_discarded");
2689 err = ustctl_put_subbuf(ustream);
2690 assert(err == 0);
2691 goto error;
2692 }
2693 } else {
2694 write_index = 0;
2695 }
2696
2697 /* Get the full padded subbuffer size */
2698 err = ustctl_get_padded_subbuf_size(ustream, &len);
2699 assert(err == 0);
2700
2701 /* Get subbuffer data size (without padding) */
2702 err = ustctl_get_subbuf_size(ustream, &subbuf_size);
2703 assert(err == 0);
2704
2705 /* Make sure we don't get a subbuffer size bigger than the padded */
2706 assert(len >= subbuf_size);
2707
2708 padding = len - subbuf_size;
2709
2710 rotate_ready = lttng_consumer_stream_is_rotate_ready(stream, len);
2711 if (rotate_ready < 0) {
2712 ERR("Failed to check if stream is ready for rotation");
2713 ret = -1;
2714 err = ustctl_put_subbuf(ustream);
2715 assert(err == 0);
2716 goto error;
2717 }
2718 stream->rotate_ready = rotate_ready;
2719
2720 /* write the subbuffer to the tracefile */
2721 ret = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, subbuf_size, padding, &index);
2722 /*
2723 * The mmap operation should write subbuf_size amount of data when network
2724 * streaming or the full padding (len) size when we are _not_ streaming.
2725 */
2726 if ((ret != subbuf_size && stream->net_seq_idx != (uint64_t) -1ULL) ||
2727 (ret != len && stream->net_seq_idx == (uint64_t) -1ULL)) {
2728 /*
2729 * Display the error but continue processing to try to release the
2730 * subbuffer. This is a DBG statement since any unexpected kill or
2731 * signal, the application gets unregistered, relayd gets closed or
2732 * anything that affects the buffer lifetime will trigger this error.
2733 * So, for the sake of the user, don't print this error since it can
2734 * happen and it is OK with the code flow.
2735 */
2736 DBG("Error writing to tracefile "
2737 "(ret: %ld != len: %lu != subbuf_size: %lu)",
2738 ret, len, subbuf_size);
2739 write_index = 0;
2740 }
2741 err = ustctl_put_next_subbuf(ustream);
2742 assert(err == 0);
2743
2744 /*
2745 * This will consumer the byte on the wait_fd if and only if there is not
2746 * next subbuffer to be acquired.
2747 */
2748 if (!stream->metadata_flag) {
2749 ret = notify_if_more_data(stream, ctx);
2750 if (ret < 0) {
2751 goto error;
2752 }
2753 }
2754
2755 /* Write index if needed. */
2756 if (!write_index) {
2757 goto rotate;
2758 }
2759
2760 if (stream->chan->live_timer_interval && !stream->metadata_flag) {
2761 /*
2762 * In live, block until all the metadata is sent.
2763 */
2764 pthread_mutex_lock(&stream->metadata_timer_lock);
2765 assert(!stream->missed_metadata_flush);
2766 stream->waiting_on_metadata = true;
2767 pthread_mutex_unlock(&stream->metadata_timer_lock);
2768
2769 err = consumer_stream_sync_metadata(ctx, stream->session_id);
2770
2771 pthread_mutex_lock(&stream->metadata_timer_lock);
2772 stream->waiting_on_metadata = false;
2773 if (stream->missed_metadata_flush) {
2774 stream->missed_metadata_flush = false;
2775 pthread_mutex_unlock(&stream->metadata_timer_lock);
2776 (void) consumer_flush_ust_index(stream);
2777 } else {
2778 pthread_mutex_unlock(&stream->metadata_timer_lock);
2779 }
2780
2781 if (err < 0) {
2782 goto error;
2783 }
2784 }
2785
2786 assert(!stream->metadata_flag);
2787 err = consumer_stream_write_index(stream, &index);
2788 if (err < 0) {
2789 goto error;
2790 }
2791
2792 rotate:
2793 if (stream->rotate_ready) {
2794 rotation_ret = lttng_consumer_rotate_stream(ctx, stream);
2795 if (rotation_ret < 0) {
2796 ret = -1;
2797 ERR("Stream rotation error");
2798 goto error;
2799 }
2800 }
2801 error:
2802 return ret;
2803 }
2804
2805 /*
2806 * Called when a stream is created.
2807 *
2808 * Return 0 on success or else a negative value.
2809 */
2810 int lttng_ustconsumer_on_recv_stream(struct lttng_consumer_stream *stream)
2811 {
2812 int ret;
2813
2814 assert(stream);
2815
2816 /* Don't create anything if this is set for streaming. */
2817 if (stream->net_seq_idx == (uint64_t) -1ULL && stream->chan->monitor) {
2818 ret = utils_create_stream_file(stream->chan->pathname, stream->name,
2819 stream->chan->tracefile_size, stream->tracefile_count_current,
2820 stream->uid, stream->gid, NULL);
2821 if (ret < 0) {
2822 goto error;
2823 }
2824 stream->out_fd = ret;
2825 stream->tracefile_size_current = 0;
2826
2827 if (!stream->metadata_flag) {
2828 struct lttng_index_file *index_file;
2829
2830 index_file = lttng_index_file_create(stream->chan->pathname,
2831 stream->name, stream->uid, stream->gid,
2832 stream->chan->tracefile_size,
2833 stream->tracefile_count_current,
2834 CTF_INDEX_MAJOR, CTF_INDEX_MINOR);
2835 if (!index_file) {
2836 goto error;
2837 }
2838 assert(!stream->index_file);
2839 stream->index_file = index_file;
2840 }
2841 }
2842 ret = 0;
2843
2844 error:
2845 return ret;
2846 }
2847
2848 /*
2849 * Check if data is still being extracted from the buffers for a specific
2850 * stream. Consumer data lock MUST be acquired before calling this function
2851 * and the stream lock.
2852 *
2853 * Return 1 if the traced data are still getting read else 0 meaning that the
2854 * data is available for trace viewer reading.
2855 */
2856 int lttng_ustconsumer_data_pending(struct lttng_consumer_stream *stream)
2857 {
2858 int ret;
2859
2860 assert(stream);
2861 assert(stream->ustream);
2862
2863 DBG("UST consumer checking data pending");
2864
2865 if (stream->endpoint_status != CONSUMER_ENDPOINT_ACTIVE) {
2866 ret = 0;
2867 goto end;
2868 }
2869
2870 if (stream->chan->type == CONSUMER_CHANNEL_TYPE_METADATA) {
2871 uint64_t contiguous, pushed;
2872
2873 /* Ease our life a bit. */
2874 contiguous = stream->chan->metadata_cache->max_offset;
2875 pushed = stream->ust_metadata_pushed;
2876
2877 /*
2878 * We can simply check whether all contiguously available data
2879 * has been pushed to the ring buffer, since the push operation
2880 * is performed within get_next_subbuf(), and because both
2881 * get_next_subbuf() and put_next_subbuf() are issued atomically
2882 * thanks to the stream lock within
2883 * lttng_ustconsumer_read_subbuffer(). This basically means that
2884 * whetnever ust_metadata_pushed is incremented, the associated
2885 * metadata has been consumed from the metadata stream.
2886 */
2887 DBG("UST consumer metadata pending check: contiguous %" PRIu64 " vs pushed %" PRIu64,
2888 contiguous, pushed);
2889 assert(((int64_t) (contiguous - pushed)) >= 0);
2890 if ((contiguous != pushed) ||
2891 (((int64_t) contiguous - pushed) > 0 || contiguous == 0)) {
2892 ret = 1; /* Data is pending */
2893 goto end;
2894 }
2895 } else {
2896 ret = ustctl_get_next_subbuf(stream->ustream);
2897 if (ret == 0) {
2898 /*
2899 * There is still data so let's put back this
2900 * subbuffer.
2901 */
2902 ret = ustctl_put_subbuf(stream->ustream);
2903 assert(ret == 0);
2904 ret = 1; /* Data is pending */
2905 goto end;
2906 }
2907 }
2908
2909 /* Data is NOT pending so ready to be read. */
2910 ret = 0;
2911
2912 end:
2913 return ret;
2914 }
2915
2916 /*
2917 * Stop a given metadata channel timer if enabled and close the wait fd which
2918 * is the poll pipe of the metadata stream.
2919 *
2920 * This MUST be called with the metadata channel acquired.
2921 */
2922 void lttng_ustconsumer_close_metadata(struct lttng_consumer_channel *metadata)
2923 {
2924 int ret;
2925
2926 assert(metadata);
2927 assert(metadata->type == CONSUMER_CHANNEL_TYPE_METADATA);
2928
2929 DBG("Closing metadata channel key %" PRIu64, metadata->key);
2930
2931 if (metadata->switch_timer_enabled == 1) {
2932 consumer_timer_switch_stop(metadata);
2933 }
2934
2935 if (!metadata->metadata_stream) {
2936 goto end;
2937 }
2938
2939 /*
2940 * Closing write side so the thread monitoring the stream wakes up if any
2941 * and clean the metadata stream.
2942 */
2943 if (metadata->metadata_stream->ust_metadata_poll_pipe[1] >= 0) {
2944 ret = close(metadata->metadata_stream->ust_metadata_poll_pipe[1]);
2945 if (ret < 0) {
2946 PERROR("closing metadata pipe write side");
2947 }
2948 metadata->metadata_stream->ust_metadata_poll_pipe[1] = -1;
2949 }
2950
2951 end:
2952 return;
2953 }
2954
2955 /*
2956 * Close every metadata stream wait fd of the metadata hash table. This
2957 * function MUST be used very carefully so not to run into a race between the
2958 * metadata thread handling streams and this function closing their wait fd.
2959 *
2960 * For UST, this is used when the session daemon hangs up. Its the metadata
2961 * producer so calling this is safe because we are assured that no state change
2962 * can occur in the metadata thread for the streams in the hash table.
2963 */
2964 void lttng_ustconsumer_close_all_metadata(struct lttng_ht *metadata_ht)
2965 {
2966 struct lttng_ht_iter iter;
2967 struct lttng_consumer_stream *stream;
2968
2969 assert(metadata_ht);
2970 assert(metadata_ht->ht);
2971
2972 DBG("UST consumer closing all metadata streams");
2973
2974 rcu_read_lock();
2975 cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream,
2976 node.node) {
2977
2978 health_code_update();
2979
2980 pthread_mutex_lock(&stream->chan->lock);
2981 lttng_ustconsumer_close_metadata(stream->chan);
2982 pthread_mutex_unlock(&stream->chan->lock);
2983
2984 }
2985 rcu_read_unlock();
2986 }
2987
2988 void lttng_ustconsumer_close_stream_wakeup(struct lttng_consumer_stream *stream)
2989 {
2990 int ret;
2991
2992 ret = ustctl_stream_close_wakeup_fd(stream->ustream);
2993 if (ret < 0) {
2994 ERR("Unable to close wakeup fd");
2995 }
2996 }
2997
2998 /*
2999 * Please refer to consumer-timer.c before adding any lock within this
3000 * function or any of its callees. Timers have a very strict locking
3001 * semantic with respect to teardown. Failure to respect this semantic
3002 * introduces deadlocks.
3003 *
3004 * DON'T hold the metadata lock when calling this function, else this
3005 * can cause deadlock involving consumer awaiting for metadata to be
3006 * pushed out due to concurrent interaction with the session daemon.
3007 */
3008 int lttng_ustconsumer_request_metadata(struct lttng_consumer_local_data *ctx,
3009 struct lttng_consumer_channel *channel, int timer, int wait)
3010 {
3011 struct lttcomm_metadata_request_msg request;
3012 struct lttcomm_consumer_msg msg;
3013 enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS;
3014 uint64_t len, key, offset, version;
3015 int ret;
3016
3017 assert(channel);
3018 assert(channel->metadata_cache);
3019
3020 memset(&request, 0, sizeof(request));
3021
3022 /* send the metadata request to sessiond */
3023 switch (consumer_data.type) {
3024 case LTTNG_CONSUMER64_UST:
3025 request.bits_per_long = 64;
3026 break;
3027 case LTTNG_CONSUMER32_UST:
3028 request.bits_per_long = 32;
3029 break;
3030 default:
3031 request.bits_per_long = 0;
3032 break;
3033 }
3034
3035 request.session_id = channel->session_id;
3036 request.session_id_per_pid = channel->session_id_per_pid;
3037 /*
3038 * Request the application UID here so the metadata of that application can
3039 * be sent back. The channel UID corresponds to the user UID of the session
3040 * used for the rights on the stream file(s).
3041 */
3042 request.uid = channel->ust_app_uid;
3043 request.key = channel->key;
3044
3045 DBG("Sending metadata request to sessiond, session id %" PRIu64
3046 ", per-pid %" PRIu64 ", app UID %u and channek key %" PRIu64,
3047 request.session_id, request.session_id_per_pid, request.uid,
3048 request.key);
3049
3050 pthread_mutex_lock(&ctx->metadata_socket_lock);
3051
3052 health_code_update();
3053
3054 ret = lttcomm_send_unix_sock(ctx->consumer_metadata_socket, &request,
3055 sizeof(request));
3056 if (ret < 0) {
3057 ERR("Asking metadata to sessiond");
3058 goto end;
3059 }
3060
3061 health_code_update();
3062
3063 /* Receive the metadata from sessiond */
3064 ret = lttcomm_recv_unix_sock(ctx->consumer_metadata_socket, &msg,
3065 sizeof(msg));
3066 if (ret != sizeof(msg)) {
3067 DBG("Consumer received unexpected message size %d (expects %zu)",
3068 ret, sizeof(msg));
3069 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD);
3070 /*
3071 * The ret value might 0 meaning an orderly shutdown but this is ok
3072 * since the caller handles this.
3073 */
3074 goto end;
3075 }
3076
3077 health_code_update();
3078
3079 if (msg.cmd_type == LTTNG_ERR_UND) {
3080 /* No registry found */
3081 (void) consumer_send_status_msg(ctx->consumer_metadata_socket,
3082 ret_code);
3083 ret = 0;
3084 goto end;
3085 } else if (msg.cmd_type != LTTNG_CONSUMER_PUSH_METADATA) {
3086 ERR("Unexpected cmd_type received %d", msg.cmd_type);
3087 ret = -1;
3088 goto end;
3089 }
3090
3091 len = msg.u.push_metadata.len;
3092 key = msg.u.push_metadata.key;
3093 offset = msg.u.push_metadata.target_offset;
3094 version = msg.u.push_metadata.version;
3095
3096 assert(key == channel->key);
3097 if (len == 0) {
3098 DBG("No new metadata to receive for key %" PRIu64, key);
3099 }
3100
3101 health_code_update();
3102
3103 /* Tell session daemon we are ready to receive the metadata. */
3104 ret = consumer_send_status_msg(ctx->consumer_metadata_socket,
3105 LTTCOMM_CONSUMERD_SUCCESS);
3106 if (ret < 0 || len == 0) {
3107 /*
3108 * Somehow, the session daemon is not responding anymore or there is
3109 * nothing to receive.
3110 */
3111 goto end;
3112 }
3113
3114 health_code_update();
3115
3116 ret = lttng_ustconsumer_recv_metadata(ctx->consumer_metadata_socket,
3117 key, offset, len, version, channel, timer, wait);
3118 if (ret >= 0) {
3119 /*
3120 * Only send the status msg if the sessiond is alive meaning a positive
3121 * ret code.
3122 */
3123 (void) consumer_send_status_msg(ctx->consumer_metadata_socket, ret);
3124 }
3125 ret = 0;
3126
3127 end:
3128 health_code_update();
3129
3130 pthread_mutex_unlock(&ctx->metadata_socket_lock);
3131 return ret;
3132 }
3133
3134 /*
3135 * Return the ustctl call for the get stream id.
3136 */
3137 int lttng_ustconsumer_get_stream_id(struct lttng_consumer_stream *stream,
3138 uint64_t *stream_id)
3139 {
3140 assert(stream);
3141 assert(stream_id);
3142
3143 return ustctl_get_stream_id(stream->ustream, stream_id);
3144 }
This page took 0.0979179999999999 seconds and 5 git commands to generate.