Fix: free ust channel object after grace period in consumer
[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 *
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License, version 2 only,
7 * as published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License along
15 * with this program; if not, write to the Free Software Foundation, Inc.,
16 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
17 */
18
19 #define _GNU_SOURCE
20 #include <assert.h>
21 #include <lttng/ust-ctl.h>
22 #include <poll.h>
23 #include <pthread.h>
24 #include <stdlib.h>
25 #include <string.h>
26 #include <sys/mman.h>
27 #include <sys/socket.h>
28 #include <sys/stat.h>
29 #include <sys/types.h>
30 #include <inttypes.h>
31 #include <unistd.h>
32 #include <urcu/list.h>
33 #include <signal.h>
34
35 #include <bin/lttng-consumerd/health-consumerd.h>
36 #include <common/common.h>
37 #include <common/sessiond-comm/sessiond-comm.h>
38 #include <common/relayd/relayd.h>
39 #include <common/compat/fcntl.h>
40 #include <common/compat/endian.h>
41 #include <common/consumer-metadata-cache.h>
42 #include <common/consumer-stream.h>
43 #include <common/consumer-timer.h>
44 #include <common/utils.h>
45 #include <common/index/index.h>
46
47 #include "ust-consumer.h"
48
49 extern struct lttng_consumer_global_data consumer_data;
50 extern int consumer_poll_timeout;
51 extern volatile int consumer_quit;
52
53 /*
54 * Free channel object and all streams associated with it. This MUST be used
55 * only and only if the channel has _NEVER_ been added to the global channel
56 * hash table.
57 */
58 static void destroy_channel(struct lttng_consumer_channel *channel)
59 {
60 struct lttng_consumer_stream *stream, *stmp;
61
62 assert(channel);
63
64 DBG("UST consumer cleaning stream list");
65
66 cds_list_for_each_entry_safe(stream, stmp, &channel->streams.head,
67 send_node) {
68
69 health_code_update();
70
71 cds_list_del(&stream->send_node);
72 ustctl_destroy_stream(stream->ustream);
73 free(stream);
74 }
75
76 /*
77 * If a channel is available meaning that was created before the streams
78 * were, delete it.
79 */
80 if (channel->uchan) {
81 lttng_ustconsumer_del_channel(channel);
82 lttng_ustconsumer_free_channel(channel);
83 }
84 free(channel);
85 }
86
87 /*
88 * Add channel to internal consumer state.
89 *
90 * Returns 0 on success or else a negative value.
91 */
92 static int add_channel(struct lttng_consumer_channel *channel,
93 struct lttng_consumer_local_data *ctx)
94 {
95 int ret = 0;
96
97 assert(channel);
98 assert(ctx);
99
100 if (ctx->on_recv_channel != NULL) {
101 ret = ctx->on_recv_channel(channel);
102 if (ret == 0) {
103 ret = consumer_add_channel(channel, ctx);
104 } else if (ret < 0) {
105 /* Most likely an ENOMEM. */
106 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
107 goto error;
108 }
109 } else {
110 ret = consumer_add_channel(channel, ctx);
111 }
112
113 DBG("UST consumer channel added (key: %" PRIu64 ")", channel->key);
114
115 error:
116 return ret;
117 }
118
119 /*
120 * Allocate and return a consumer channel object.
121 */
122 static struct lttng_consumer_channel *allocate_channel(uint64_t session_id,
123 const char *pathname, const char *name, uid_t uid, gid_t gid,
124 uint64_t relayd_id, uint64_t key, enum lttng_event_output output,
125 uint64_t tracefile_size, uint64_t tracefile_count,
126 uint64_t session_id_per_pid, unsigned int monitor,
127 unsigned int live_timer_interval)
128 {
129 assert(pathname);
130 assert(name);
131
132 return consumer_allocate_channel(key, session_id, pathname, name, uid,
133 gid, relayd_id, output, tracefile_size,
134 tracefile_count, session_id_per_pid, monitor, live_timer_interval);
135 }
136
137 /*
138 * Allocate and return a consumer stream object. If _alloc_ret is not NULL, the
139 * error value if applicable is set in it else it is kept untouched.
140 *
141 * Return NULL on error else the newly allocated stream object.
142 */
143 static struct lttng_consumer_stream *allocate_stream(int cpu, int key,
144 struct lttng_consumer_channel *channel,
145 struct lttng_consumer_local_data *ctx, int *_alloc_ret)
146 {
147 int alloc_ret;
148 struct lttng_consumer_stream *stream = NULL;
149
150 assert(channel);
151 assert(ctx);
152
153 stream = consumer_allocate_stream(channel->key,
154 key,
155 LTTNG_CONSUMER_ACTIVE_STREAM,
156 channel->name,
157 channel->uid,
158 channel->gid,
159 channel->relayd_id,
160 channel->session_id,
161 cpu,
162 &alloc_ret,
163 channel->type,
164 channel->monitor);
165 if (stream == NULL) {
166 switch (alloc_ret) {
167 case -ENOENT:
168 /*
169 * We could not find the channel. Can happen if cpu hotplug
170 * happens while tearing down.
171 */
172 DBG3("Could not find channel");
173 break;
174 case -ENOMEM:
175 case -EINVAL:
176 default:
177 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
178 break;
179 }
180 goto error;
181 }
182
183 stream->chan = channel;
184
185 error:
186 if (_alloc_ret) {
187 *_alloc_ret = alloc_ret;
188 }
189 return stream;
190 }
191
192 /*
193 * Send the given stream pointer to the corresponding thread.
194 *
195 * Returns 0 on success else a negative value.
196 */
197 static int send_stream_to_thread(struct lttng_consumer_stream *stream,
198 struct lttng_consumer_local_data *ctx)
199 {
200 int ret;
201 struct lttng_pipe *stream_pipe;
202
203 /* Get the right pipe where the stream will be sent. */
204 if (stream->metadata_flag) {
205 ret = consumer_add_metadata_stream(stream);
206 if (ret) {
207 ERR("Consumer add metadata stream %" PRIu64 " failed.",
208 stream->key);
209 goto error;
210 }
211 stream_pipe = ctx->consumer_metadata_pipe;
212 } else {
213 ret = consumer_add_data_stream(stream);
214 if (ret) {
215 ERR("Consumer add stream %" PRIu64 " failed.",
216 stream->key);
217 goto error;
218 }
219 stream_pipe = ctx->consumer_data_pipe;
220 }
221
222 /*
223 * From this point on, the stream's ownership has been moved away from
224 * the channel and becomes globally visible.
225 */
226 stream->globally_visible = 1;
227
228 ret = lttng_pipe_write(stream_pipe, &stream, sizeof(stream));
229 if (ret < 0) {
230 ERR("Consumer write %s stream to pipe %d",
231 stream->metadata_flag ? "metadata" : "data",
232 lttng_pipe_get_writefd(stream_pipe));
233 if (stream->metadata_flag) {
234 consumer_del_stream_for_metadata(stream);
235 } else {
236 consumer_del_stream_for_data(stream);
237 }
238 }
239 error:
240 return ret;
241 }
242
243 /*
244 * Create streams for the given channel using liblttng-ust-ctl.
245 *
246 * Return 0 on success else a negative value.
247 */
248 static int create_ust_streams(struct lttng_consumer_channel *channel,
249 struct lttng_consumer_local_data *ctx)
250 {
251 int ret, cpu = 0;
252 struct ustctl_consumer_stream *ustream;
253 struct lttng_consumer_stream *stream;
254
255 assert(channel);
256 assert(ctx);
257
258 /*
259 * While a stream is available from ustctl. When NULL is returned, we've
260 * reached the end of the possible stream for the channel.
261 */
262 while ((ustream = ustctl_create_stream(channel->uchan, cpu))) {
263 int wait_fd;
264 int ust_metadata_pipe[2];
265
266 health_code_update();
267
268 if (channel->type == CONSUMER_CHANNEL_TYPE_METADATA && channel->monitor) {
269 ret = utils_create_pipe_cloexec_nonblock(ust_metadata_pipe);
270 if (ret < 0) {
271 ERR("Create ust metadata poll pipe");
272 goto error;
273 }
274 wait_fd = ust_metadata_pipe[0];
275 } else {
276 wait_fd = ustctl_stream_get_wait_fd(ustream);
277 }
278
279 /* Allocate consumer stream object. */
280 stream = allocate_stream(cpu, wait_fd, channel, ctx, &ret);
281 if (!stream) {
282 goto error_alloc;
283 }
284 stream->ustream = ustream;
285 /*
286 * Store it so we can save multiple function calls afterwards since
287 * this value is used heavily in the stream threads. This is UST
288 * specific so this is why it's done after allocation.
289 */
290 stream->wait_fd = wait_fd;
291
292 /*
293 * Increment channel refcount since the channel reference has now been
294 * assigned in the allocation process above.
295 */
296 if (stream->chan->monitor) {
297 uatomic_inc(&stream->chan->refcount);
298 }
299
300 /*
301 * Order is important this is why a list is used. On error, the caller
302 * should clean this list.
303 */
304 cds_list_add_tail(&stream->send_node, &channel->streams.head);
305
306 ret = ustctl_get_max_subbuf_size(stream->ustream,
307 &stream->max_sb_size);
308 if (ret < 0) {
309 ERR("ustctl_get_max_subbuf_size failed for stream %s",
310 stream->name);
311 goto error;
312 }
313
314 /* Do actions once stream has been received. */
315 if (ctx->on_recv_stream) {
316 ret = ctx->on_recv_stream(stream);
317 if (ret < 0) {
318 goto error;
319 }
320 }
321
322 DBG("UST consumer add stream %s (key: %" PRIu64 ") with relayd id %" PRIu64,
323 stream->name, stream->key, stream->relayd_stream_id);
324
325 /* Set next CPU stream. */
326 channel->streams.count = ++cpu;
327
328 /* Keep stream reference when creating metadata. */
329 if (channel->type == CONSUMER_CHANNEL_TYPE_METADATA) {
330 channel->metadata_stream = stream;
331 if (channel->monitor) {
332 /* Set metadata poll pipe if we created one */
333 memcpy(stream->ust_metadata_poll_pipe,
334 ust_metadata_pipe,
335 sizeof(ust_metadata_pipe));
336 }
337 }
338 }
339
340 return 0;
341
342 error:
343 error_alloc:
344 return ret;
345 }
346
347 /*
348 * Create an UST channel with the given attributes and send it to the session
349 * daemon using the ust ctl API.
350 *
351 * Return 0 on success or else a negative value.
352 */
353 static int create_ust_channel(struct ustctl_consumer_channel_attr *attr,
354 struct ustctl_consumer_channel **chanp)
355 {
356 int ret;
357 struct ustctl_consumer_channel *channel;
358
359 assert(attr);
360 assert(chanp);
361
362 DBG3("Creating channel to ustctl with attr: [overwrite: %d, "
363 "subbuf_size: %" PRIu64 ", num_subbuf: %" PRIu64 ", "
364 "switch_timer_interval: %u, read_timer_interval: %u, "
365 "output: %d, type: %d", attr->overwrite, attr->subbuf_size,
366 attr->num_subbuf, attr->switch_timer_interval,
367 attr->read_timer_interval, attr->output, attr->type);
368
369 channel = ustctl_create_channel(attr);
370 if (!channel) {
371 ret = -1;
372 goto error_create;
373 }
374
375 *chanp = channel;
376
377 return 0;
378
379 error_create:
380 return ret;
381 }
382
383 /*
384 * Send a single given stream to the session daemon using the sock.
385 *
386 * Return 0 on success else a negative value.
387 */
388 static int send_sessiond_stream(int sock, struct lttng_consumer_stream *stream)
389 {
390 int ret;
391
392 assert(stream);
393 assert(sock >= 0);
394
395 DBG("UST consumer sending stream %" PRIu64 " to sessiond", stream->key);
396
397 /* Send stream to session daemon. */
398 ret = ustctl_send_stream_to_sessiond(sock, stream->ustream);
399 if (ret < 0) {
400 goto error;
401 }
402
403 error:
404 return ret;
405 }
406
407 /*
408 * Send channel to sessiond.
409 *
410 * Return 0 on success or else a negative value.
411 */
412 static int send_sessiond_channel(int sock,
413 struct lttng_consumer_channel *channel,
414 struct lttng_consumer_local_data *ctx, int *relayd_error)
415 {
416 int ret, ret_code = LTTCOMM_CONSUMERD_SUCCESS;
417 struct lttng_consumer_stream *stream;
418 uint64_t net_seq_idx = -1ULL;
419
420 assert(channel);
421 assert(ctx);
422 assert(sock >= 0);
423
424 DBG("UST consumer sending channel %s to sessiond", channel->name);
425
426 if (channel->relayd_id != (uint64_t) -1ULL) {
427 cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
428
429 health_code_update();
430
431 /* Try to send the stream to the relayd if one is available. */
432 ret = consumer_send_relayd_stream(stream, stream->chan->pathname);
433 if (ret < 0) {
434 /*
435 * Flag that the relayd was the problem here probably due to a
436 * communicaton error on the socket.
437 */
438 if (relayd_error) {
439 *relayd_error = 1;
440 }
441 ret_code = LTTNG_ERR_RELAYD_CONNECT_FAIL;
442 }
443 if (net_seq_idx == -1ULL) {
444 net_seq_idx = stream->net_seq_idx;
445 }
446 }
447 }
448
449 /* Inform sessiond that we are about to send channel and streams. */
450 ret = consumer_send_status_msg(sock, ret_code);
451 if (ret < 0 || ret_code != LTTCOMM_CONSUMERD_SUCCESS) {
452 /*
453 * Either the session daemon is not responding or the relayd died so we
454 * stop now.
455 */
456 goto error;
457 }
458
459 /* Send channel to sessiond. */
460 ret = ustctl_send_channel_to_sessiond(sock, channel->uchan);
461 if (ret < 0) {
462 goto error;
463 }
464
465 ret = ustctl_channel_close_wakeup_fd(channel->uchan);
466 if (ret < 0) {
467 goto error;
468 }
469
470 /* The channel was sent successfully to the sessiond at this point. */
471 cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
472
473 health_code_update();
474
475 /* Send stream to session daemon. */
476 ret = send_sessiond_stream(sock, stream);
477 if (ret < 0) {
478 goto error;
479 }
480 }
481
482 /* Tell sessiond there is no more stream. */
483 ret = ustctl_send_stream_to_sessiond(sock, NULL);
484 if (ret < 0) {
485 goto error;
486 }
487
488 DBG("UST consumer NULL stream sent to sessiond");
489
490 return 0;
491
492 error:
493 if (ret_code != LTTCOMM_CONSUMERD_SUCCESS) {
494 ret = -1;
495 }
496 return ret;
497 }
498
499 /*
500 * Creates a channel and streams and add the channel it to the channel internal
501 * state. The created stream must ONLY be sent once the GET_CHANNEL command is
502 * received.
503 *
504 * Return 0 on success or else, a negative value is returned and the channel
505 * MUST be destroyed by consumer_del_channel().
506 */
507 static int ask_channel(struct lttng_consumer_local_data *ctx, int sock,
508 struct lttng_consumer_channel *channel,
509 struct ustctl_consumer_channel_attr *attr)
510 {
511 int ret;
512
513 assert(ctx);
514 assert(channel);
515 assert(attr);
516
517 /*
518 * This value is still used by the kernel consumer since for the kernel,
519 * the stream ownership is not IN the consumer so we need to have the
520 * number of left stream that needs to be initialized so we can know when
521 * to delete the channel (see consumer.c).
522 *
523 * As for the user space tracer now, the consumer creates and sends the
524 * stream to the session daemon which only sends them to the application
525 * once every stream of a channel is received making this value useless
526 * because we they will be added to the poll thread before the application
527 * receives them. This ensures that a stream can not hang up during
528 * initilization of a channel.
529 */
530 channel->nb_init_stream_left = 0;
531
532 /* The reply msg status is handled in the following call. */
533 ret = create_ust_channel(attr, &channel->uchan);
534 if (ret < 0) {
535 goto end;
536 }
537
538 channel->wait_fd = ustctl_channel_get_wait_fd(channel->uchan);
539
540 /*
541 * For the snapshots (no monitor), we create the metadata streams
542 * on demand, not during the channel creation.
543 */
544 if (channel->type == CONSUMER_CHANNEL_TYPE_METADATA && !channel->monitor) {
545 ret = 0;
546 goto end;
547 }
548
549 /* Open all streams for this channel. */
550 ret = create_ust_streams(channel, ctx);
551 if (ret < 0) {
552 goto end;
553 }
554
555 end:
556 return ret;
557 }
558
559 /*
560 * Send all stream of a channel to the right thread handling it.
561 *
562 * On error, return a negative value else 0 on success.
563 */
564 static int send_streams_to_thread(struct lttng_consumer_channel *channel,
565 struct lttng_consumer_local_data *ctx)
566 {
567 int ret = 0;
568 struct lttng_consumer_stream *stream, *stmp;
569
570 assert(channel);
571 assert(ctx);
572
573 /* Send streams to the corresponding thread. */
574 cds_list_for_each_entry_safe(stream, stmp, &channel->streams.head,
575 send_node) {
576
577 health_code_update();
578
579 /* Sending the stream to the thread. */
580 ret = send_stream_to_thread(stream, ctx);
581 if (ret < 0) {
582 /*
583 * If we are unable to send the stream to the thread, there is
584 * a big problem so just stop everything.
585 */
586 /* Remove node from the channel stream list. */
587 cds_list_del(&stream->send_node);
588 goto error;
589 }
590
591 /* Remove node from the channel stream list. */
592 cds_list_del(&stream->send_node);
593
594 }
595
596 error:
597 return ret;
598 }
599
600 /*
601 * Flush channel's streams using the given key to retrieve the channel.
602 *
603 * Return 0 on success else an LTTng error code.
604 */
605 static int flush_channel(uint64_t chan_key)
606 {
607 int ret = 0;
608 struct lttng_consumer_channel *channel;
609 struct lttng_consumer_stream *stream;
610 struct lttng_ht *ht;
611 struct lttng_ht_iter iter;
612
613 DBG("UST consumer flush channel key %" PRIu64, chan_key);
614
615 rcu_read_lock();
616 channel = consumer_find_channel(chan_key);
617 if (!channel) {
618 ERR("UST consumer flush channel %" PRIu64 " not found", chan_key);
619 ret = LTTNG_ERR_UST_CHAN_NOT_FOUND;
620 goto error;
621 }
622
623 ht = consumer_data.stream_per_chan_id_ht;
624
625 /* For each stream of the channel id, flush it. */
626 cds_lfht_for_each_entry_duplicate(ht->ht,
627 ht->hash_fct(&channel->key, lttng_ht_seed), ht->match_fct,
628 &channel->key, &iter.iter, stream, node_channel_id.node) {
629
630 health_code_update();
631
632 ustctl_flush_buffer(stream->ustream, 1);
633 }
634 error:
635 rcu_read_unlock();
636 return ret;
637 }
638
639 /*
640 * Close metadata stream wakeup_fd using the given key to retrieve the channel.
641 * RCU read side lock MUST be acquired before calling this function.
642 *
643 * Return 0 on success else an LTTng error code.
644 */
645 static int close_metadata(uint64_t chan_key)
646 {
647 int ret = 0;
648 struct lttng_consumer_channel *channel;
649
650 DBG("UST consumer close metadata key %" PRIu64, chan_key);
651
652 channel = consumer_find_channel(chan_key);
653 if (!channel) {
654 /*
655 * This is possible if the metadata thread has issue a delete because
656 * the endpoint point of the stream hung up. There is no way the
657 * session daemon can know about it thus use a DBG instead of an actual
658 * error.
659 */
660 DBG("UST consumer close metadata %" PRIu64 " not found", chan_key);
661 ret = LTTNG_ERR_UST_CHAN_NOT_FOUND;
662 goto error;
663 }
664
665 pthread_mutex_lock(&consumer_data.lock);
666 pthread_mutex_lock(&channel->lock);
667
668 if (cds_lfht_is_node_deleted(&channel->node.node)) {
669 goto error_unlock;
670 }
671
672 lttng_ustconsumer_close_metadata(channel);
673
674 error_unlock:
675 pthread_mutex_unlock(&channel->lock);
676 pthread_mutex_unlock(&consumer_data.lock);
677 error:
678 return ret;
679 }
680
681 /*
682 * RCU read side lock MUST be acquired before calling this function.
683 *
684 * Return 0 on success else an LTTng error code.
685 */
686 static int setup_metadata(struct lttng_consumer_local_data *ctx, uint64_t key)
687 {
688 int ret;
689 struct lttng_consumer_channel *metadata;
690
691 DBG("UST consumer setup metadata key %" PRIu64, key);
692
693 metadata = consumer_find_channel(key);
694 if (!metadata) {
695 ERR("UST consumer push metadata %" PRIu64 " not found", key);
696 ret = LTTNG_ERR_UST_CHAN_NOT_FOUND;
697 goto end;
698 }
699
700 /*
701 * In no monitor mode, the metadata channel has no stream(s) so skip the
702 * ownership transfer to the metadata thread.
703 */
704 if (!metadata->monitor) {
705 DBG("Metadata channel in no monitor");
706 ret = 0;
707 goto end;
708 }
709
710 /*
711 * Send metadata stream to relayd if one available. Availability is
712 * known if the stream is still in the list of the channel.
713 */
714 if (cds_list_empty(&metadata->streams.head)) {
715 ERR("Metadata channel key %" PRIu64 ", no stream available.", key);
716 ret = LTTCOMM_CONSUMERD_ERROR_METADATA;
717 goto error_no_stream;
718 }
719
720 /* Send metadata stream to relayd if needed. */
721 if (metadata->metadata_stream->net_seq_idx != (uint64_t) -1ULL) {
722 ret = consumer_send_relayd_stream(metadata->metadata_stream,
723 metadata->pathname);
724 if (ret < 0) {
725 ret = LTTCOMM_CONSUMERD_ERROR_METADATA;
726 goto error;
727 }
728 ret = consumer_send_relayd_streams_sent(
729 metadata->metadata_stream->net_seq_idx);
730 if (ret < 0) {
731 ret = LTTCOMM_CONSUMERD_RELAYD_FAIL;
732 goto error;
733 }
734 }
735
736 ret = send_streams_to_thread(metadata, ctx);
737 if (ret < 0) {
738 /*
739 * If we are unable to send the stream to the thread, there is
740 * a big problem so just stop everything.
741 */
742 ret = LTTCOMM_CONSUMERD_FATAL;
743 goto error;
744 }
745 /* List MUST be empty after or else it could be reused. */
746 assert(cds_list_empty(&metadata->streams.head));
747
748 ret = 0;
749 goto end;
750
751 error:
752 /*
753 * Delete metadata channel on error. At this point, the metadata stream can
754 * NOT be monitored by the metadata thread thus having the guarantee that
755 * the stream is still in the local stream list of the channel. This call
756 * will make sure to clean that list.
757 */
758 consumer_stream_destroy(metadata->metadata_stream, NULL);
759 cds_list_del(&metadata->metadata_stream->send_node);
760 metadata->metadata_stream = NULL;
761 error_no_stream:
762 end:
763 return ret;
764 }
765
766 /*
767 * Snapshot the whole metadata.
768 *
769 * Returns 0 on success, < 0 on error
770 */
771 static int snapshot_metadata(uint64_t key, char *path, uint64_t relayd_id,
772 struct lttng_consumer_local_data *ctx)
773 {
774 int ret = 0;
775 struct lttng_consumer_channel *metadata_channel;
776 struct lttng_consumer_stream *metadata_stream;
777
778 assert(path);
779 assert(ctx);
780
781 DBG("UST consumer snapshot metadata with key %" PRIu64 " at path %s",
782 key, path);
783
784 rcu_read_lock();
785
786 metadata_channel = consumer_find_channel(key);
787 if (!metadata_channel) {
788 ERR("UST snapshot metadata channel not found for key %" PRIu64,
789 key);
790 ret = -1;
791 goto error;
792 }
793 assert(!metadata_channel->monitor);
794
795 health_code_update();
796
797 /*
798 * Ask the sessiond if we have new metadata waiting and update the
799 * consumer metadata cache.
800 */
801 ret = lttng_ustconsumer_request_metadata(ctx, metadata_channel, 0, 1);
802 if (ret < 0) {
803 goto error;
804 }
805
806 health_code_update();
807
808 /*
809 * The metadata stream is NOT created in no monitor mode when the channel
810 * is created on a sessiond ask channel command.
811 */
812 ret = create_ust_streams(metadata_channel, ctx);
813 if (ret < 0) {
814 goto error;
815 }
816
817 metadata_stream = metadata_channel->metadata_stream;
818 assert(metadata_stream);
819
820 if (relayd_id != (uint64_t) -1ULL) {
821 metadata_stream->net_seq_idx = relayd_id;
822 ret = consumer_send_relayd_stream(metadata_stream, path);
823 if (ret < 0) {
824 goto error_stream;
825 }
826 } else {
827 ret = utils_create_stream_file(path, metadata_stream->name,
828 metadata_stream->chan->tracefile_size,
829 metadata_stream->tracefile_count_current,
830 metadata_stream->uid, metadata_stream->gid, NULL);
831 if (ret < 0) {
832 goto error_stream;
833 }
834 metadata_stream->out_fd = ret;
835 metadata_stream->tracefile_size_current = 0;
836 }
837
838 do {
839 health_code_update();
840
841 ret = lttng_consumer_read_subbuffer(metadata_stream, ctx);
842 if (ret < 0) {
843 goto error_stream;
844 }
845 } while (ret > 0);
846
847 error_stream:
848 /*
849 * Clean up the stream completly because the next snapshot will use a new
850 * metadata stream.
851 */
852 consumer_stream_destroy(metadata_stream, NULL);
853 cds_list_del(&metadata_stream->send_node);
854 metadata_channel->metadata_stream = NULL;
855
856 error:
857 rcu_read_unlock();
858 return ret;
859 }
860
861 /*
862 * Take a snapshot of all the stream of a channel.
863 *
864 * Returns 0 on success, < 0 on error
865 */
866 static int snapshot_channel(uint64_t key, char *path, uint64_t relayd_id,
867 uint64_t nb_packets_per_stream, struct lttng_consumer_local_data *ctx)
868 {
869 int ret;
870 unsigned use_relayd = 0;
871 unsigned long consumed_pos, produced_pos;
872 struct lttng_consumer_channel *channel;
873 struct lttng_consumer_stream *stream;
874
875 assert(path);
876 assert(ctx);
877
878 rcu_read_lock();
879
880 if (relayd_id != (uint64_t) -1ULL) {
881 use_relayd = 1;
882 }
883
884 channel = consumer_find_channel(key);
885 if (!channel) {
886 ERR("UST snapshot channel not found for key %" PRIu64, key);
887 ret = -1;
888 goto error;
889 }
890 assert(!channel->monitor);
891 DBG("UST consumer snapshot channel %" PRIu64, key);
892
893 cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
894
895 health_code_update();
896
897 /* Lock stream because we are about to change its state. */
898 pthread_mutex_lock(&stream->lock);
899 stream->net_seq_idx = relayd_id;
900
901 if (use_relayd) {
902 ret = consumer_send_relayd_stream(stream, path);
903 if (ret < 0) {
904 goto error_unlock;
905 }
906 } else {
907 ret = utils_create_stream_file(path, stream->name,
908 stream->chan->tracefile_size,
909 stream->tracefile_count_current,
910 stream->uid, stream->gid, NULL);
911 if (ret < 0) {
912 goto error_unlock;
913 }
914 stream->out_fd = ret;
915 stream->tracefile_size_current = 0;
916
917 DBG("UST consumer snapshot stream %s/%s (%" PRIu64 ")", path,
918 stream->name, stream->key);
919 }
920 if (relayd_id != -1ULL) {
921 ret = consumer_send_relayd_streams_sent(relayd_id);
922 if (ret < 0) {
923 goto error_unlock;
924 }
925 }
926
927 ustctl_flush_buffer(stream->ustream, 1);
928
929 ret = lttng_ustconsumer_take_snapshot(stream);
930 if (ret < 0) {
931 ERR("Taking UST snapshot");
932 goto error_unlock;
933 }
934
935 ret = lttng_ustconsumer_get_produced_snapshot(stream, &produced_pos);
936 if (ret < 0) {
937 ERR("Produced UST snapshot position");
938 goto error_unlock;
939 }
940
941 ret = lttng_ustconsumer_get_consumed_snapshot(stream, &consumed_pos);
942 if (ret < 0) {
943 ERR("Consumerd UST snapshot position");
944 goto error_unlock;
945 }
946
947 /*
948 * The original value is sent back if max stream size is larger than
949 * the possible size of the snapshot. Also, we assume that the session
950 * daemon should never send a maximum stream size that is lower than
951 * subbuffer size.
952 */
953 consumed_pos = consumer_get_consume_start_pos(consumed_pos,
954 produced_pos, nb_packets_per_stream,
955 stream->max_sb_size);
956
957 while (consumed_pos < produced_pos) {
958 ssize_t read_len;
959 unsigned long len, padded_len;
960
961 health_code_update();
962
963 DBG("UST consumer taking snapshot at pos %lu", consumed_pos);
964
965 ret = ustctl_get_subbuf(stream->ustream, &consumed_pos);
966 if (ret < 0) {
967 if (ret != -EAGAIN) {
968 PERROR("ustctl_get_subbuf snapshot");
969 goto error_close_stream;
970 }
971 DBG("UST consumer get subbuf failed. Skipping it.");
972 consumed_pos += stream->max_sb_size;
973 continue;
974 }
975
976 ret = ustctl_get_subbuf_size(stream->ustream, &len);
977 if (ret < 0) {
978 ERR("Snapshot ustctl_get_subbuf_size");
979 goto error_put_subbuf;
980 }
981
982 ret = ustctl_get_padded_subbuf_size(stream->ustream, &padded_len);
983 if (ret < 0) {
984 ERR("Snapshot ustctl_get_padded_subbuf_size");
985 goto error_put_subbuf;
986 }
987
988 read_len = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, len,
989 padded_len - len, NULL);
990 if (use_relayd) {
991 if (read_len != len) {
992 ret = -EPERM;
993 goto error_put_subbuf;
994 }
995 } else {
996 if (read_len != padded_len) {
997 ret = -EPERM;
998 goto error_put_subbuf;
999 }
1000 }
1001
1002 ret = ustctl_put_subbuf(stream->ustream);
1003 if (ret < 0) {
1004 ERR("Snapshot ustctl_put_subbuf");
1005 goto error_close_stream;
1006 }
1007 consumed_pos += stream->max_sb_size;
1008 }
1009
1010 /* Simply close the stream so we can use it on the next snapshot. */
1011 consumer_stream_close(stream);
1012 pthread_mutex_unlock(&stream->lock);
1013 }
1014
1015 rcu_read_unlock();
1016 return 0;
1017
1018 error_put_subbuf:
1019 if (ustctl_put_subbuf(stream->ustream) < 0) {
1020 ERR("Snapshot ustctl_put_subbuf");
1021 }
1022 error_close_stream:
1023 consumer_stream_close(stream);
1024 error_unlock:
1025 pthread_mutex_unlock(&stream->lock);
1026 error:
1027 rcu_read_unlock();
1028 return ret;
1029 }
1030
1031 /*
1032 * Receive the metadata updates from the sessiond. Supports receiving
1033 * overlapping metadata, but is needs to always belong to a contiguous
1034 * range starting from 0.
1035 * Be careful about the locks held when calling this function: it needs
1036 * the metadata cache flush to concurrently progress in order to
1037 * complete.
1038 */
1039 int lttng_ustconsumer_recv_metadata(int sock, uint64_t key, uint64_t offset,
1040 uint64_t len, struct lttng_consumer_channel *channel,
1041 int timer, int wait)
1042 {
1043 int ret, ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1044 char *metadata_str;
1045
1046 DBG("UST consumer push metadata key %" PRIu64 " of len %" PRIu64, key, len);
1047
1048 metadata_str = zmalloc(len * sizeof(char));
1049 if (!metadata_str) {
1050 PERROR("zmalloc metadata string");
1051 ret_code = LTTCOMM_CONSUMERD_ENOMEM;
1052 goto end;
1053 }
1054
1055 health_code_update();
1056
1057 /* Receive metadata string. */
1058 ret = lttcomm_recv_unix_sock(sock, metadata_str, len);
1059 if (ret < 0) {
1060 /* Session daemon is dead so return gracefully. */
1061 ret_code = ret;
1062 goto end_free;
1063 }
1064
1065 health_code_update();
1066
1067 pthread_mutex_lock(&channel->metadata_cache->lock);
1068 ret = consumer_metadata_cache_write(channel, offset, len, metadata_str);
1069 if (ret < 0) {
1070 /* Unable to handle metadata. Notify session daemon. */
1071 ret_code = LTTCOMM_CONSUMERD_ERROR_METADATA;
1072 /*
1073 * Skip metadata flush on write error since the offset and len might
1074 * not have been updated which could create an infinite loop below when
1075 * waiting for the metadata cache to be flushed.
1076 */
1077 pthread_mutex_unlock(&channel->metadata_cache->lock);
1078 goto end_free;
1079 }
1080 pthread_mutex_unlock(&channel->metadata_cache->lock);
1081
1082 if (!wait) {
1083 goto end_free;
1084 }
1085 while (consumer_metadata_cache_flushed(channel, offset + len, timer)) {
1086 DBG("Waiting for metadata to be flushed");
1087
1088 health_code_update();
1089
1090 usleep(DEFAULT_METADATA_AVAILABILITY_WAIT_TIME);
1091 }
1092
1093 end_free:
1094 free(metadata_str);
1095 end:
1096 return ret_code;
1097 }
1098
1099 /*
1100 * Receive command from session daemon and process it.
1101 *
1102 * Return 1 on success else a negative value or 0.
1103 */
1104 int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
1105 int sock, struct pollfd *consumer_sockpoll)
1106 {
1107 ssize_t ret;
1108 enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1109 struct lttcomm_consumer_msg msg;
1110 struct lttng_consumer_channel *channel = NULL;
1111
1112 health_code_update();
1113
1114 ret = lttcomm_recv_unix_sock(sock, &msg, sizeof(msg));
1115 if (ret != sizeof(msg)) {
1116 DBG("Consumer received unexpected message size %zd (expects %zu)",
1117 ret, sizeof(msg));
1118 /*
1119 * The ret value might 0 meaning an orderly shutdown but this is ok
1120 * since the caller handles this.
1121 */
1122 if (ret > 0) {
1123 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD);
1124 ret = -1;
1125 }
1126 return ret;
1127 }
1128
1129 health_code_update();
1130
1131 /* deprecated */
1132 assert(msg.cmd_type != LTTNG_CONSUMER_STOP);
1133
1134 health_code_update();
1135
1136 /* relayd needs RCU read-side lock */
1137 rcu_read_lock();
1138
1139 switch (msg.cmd_type) {
1140 case LTTNG_CONSUMER_ADD_RELAYD_SOCKET:
1141 {
1142 /* Session daemon status message are handled in the following call. */
1143 ret = consumer_add_relayd_socket(msg.u.relayd_sock.net_index,
1144 msg.u.relayd_sock.type, ctx, sock, consumer_sockpoll,
1145 &msg.u.relayd_sock.sock, msg.u.relayd_sock.session_id,
1146 msg.u.relayd_sock.relayd_session_id);
1147 goto end_nosignal;
1148 }
1149 case LTTNG_CONSUMER_DESTROY_RELAYD:
1150 {
1151 uint64_t index = msg.u.destroy_relayd.net_seq_idx;
1152 struct consumer_relayd_sock_pair *relayd;
1153
1154 DBG("UST consumer destroying relayd %" PRIu64, index);
1155
1156 /* Get relayd reference if exists. */
1157 relayd = consumer_find_relayd(index);
1158 if (relayd == NULL) {
1159 DBG("Unable to find relayd %" PRIu64, index);
1160 ret_code = LTTCOMM_CONSUMERD_RELAYD_FAIL;
1161 }
1162
1163 /*
1164 * Each relayd socket pair has a refcount of stream attached to it
1165 * which tells if the relayd is still active or not depending on the
1166 * refcount value.
1167 *
1168 * This will set the destroy flag of the relayd object and destroy it
1169 * if the refcount reaches zero when called.
1170 *
1171 * The destroy can happen either here or when a stream fd hangs up.
1172 */
1173 if (relayd) {
1174 consumer_flag_relayd_for_destroy(relayd);
1175 }
1176
1177 goto end_msg_sessiond;
1178 }
1179 case LTTNG_CONSUMER_UPDATE_STREAM:
1180 {
1181 rcu_read_unlock();
1182 return -ENOSYS;
1183 }
1184 case LTTNG_CONSUMER_DATA_PENDING:
1185 {
1186 int ret, is_data_pending;
1187 uint64_t id = msg.u.data_pending.session_id;
1188
1189 DBG("UST consumer data pending command for id %" PRIu64, id);
1190
1191 is_data_pending = consumer_data_pending(id);
1192
1193 /* Send back returned value to session daemon */
1194 ret = lttcomm_send_unix_sock(sock, &is_data_pending,
1195 sizeof(is_data_pending));
1196 if (ret < 0) {
1197 DBG("Error when sending the data pending ret code: %d", ret);
1198 goto error_fatal;
1199 }
1200
1201 /*
1202 * No need to send back a status message since the data pending
1203 * returned value is the response.
1204 */
1205 break;
1206 }
1207 case LTTNG_CONSUMER_ASK_CHANNEL_CREATION:
1208 {
1209 int ret;
1210 struct ustctl_consumer_channel_attr attr;
1211
1212 /* Create a plain object and reserve a channel key. */
1213 channel = allocate_channel(msg.u.ask_channel.session_id,
1214 msg.u.ask_channel.pathname, msg.u.ask_channel.name,
1215 msg.u.ask_channel.uid, msg.u.ask_channel.gid,
1216 msg.u.ask_channel.relayd_id, msg.u.ask_channel.key,
1217 (enum lttng_event_output) msg.u.ask_channel.output,
1218 msg.u.ask_channel.tracefile_size,
1219 msg.u.ask_channel.tracefile_count,
1220 msg.u.ask_channel.session_id_per_pid,
1221 msg.u.ask_channel.monitor,
1222 msg.u.ask_channel.live_timer_interval);
1223 if (!channel) {
1224 goto end_channel_error;
1225 }
1226
1227 /*
1228 * Assign UST application UID to the channel. This value is ignored for
1229 * per PID buffers. This is specific to UST thus setting this after the
1230 * allocation.
1231 */
1232 channel->ust_app_uid = msg.u.ask_channel.ust_app_uid;
1233
1234 /* Build channel attributes from received message. */
1235 attr.subbuf_size = msg.u.ask_channel.subbuf_size;
1236 attr.num_subbuf = msg.u.ask_channel.num_subbuf;
1237 attr.overwrite = msg.u.ask_channel.overwrite;
1238 attr.switch_timer_interval = msg.u.ask_channel.switch_timer_interval;
1239 attr.read_timer_interval = msg.u.ask_channel.read_timer_interval;
1240 attr.chan_id = msg.u.ask_channel.chan_id;
1241 memcpy(attr.uuid, msg.u.ask_channel.uuid, sizeof(attr.uuid));
1242
1243 /* Match channel buffer type to the UST abi. */
1244 switch (msg.u.ask_channel.output) {
1245 case LTTNG_EVENT_MMAP:
1246 default:
1247 attr.output = LTTNG_UST_MMAP;
1248 break;
1249 }
1250
1251 /* Translate and save channel type. */
1252 switch (msg.u.ask_channel.type) {
1253 case LTTNG_UST_CHAN_PER_CPU:
1254 channel->type = CONSUMER_CHANNEL_TYPE_DATA;
1255 attr.type = LTTNG_UST_CHAN_PER_CPU;
1256 /*
1257 * Set refcount to 1 for owner. Below, we will
1258 * pass ownership to the
1259 * consumer_thread_channel_poll() thread.
1260 */
1261 channel->refcount = 1;
1262 break;
1263 case LTTNG_UST_CHAN_METADATA:
1264 channel->type = CONSUMER_CHANNEL_TYPE_METADATA;
1265 attr.type = LTTNG_UST_CHAN_METADATA;
1266 break;
1267 default:
1268 assert(0);
1269 goto error_fatal;
1270 };
1271
1272 health_code_update();
1273
1274 ret = ask_channel(ctx, sock, channel, &attr);
1275 if (ret < 0) {
1276 goto end_channel_error;
1277 }
1278
1279 if (msg.u.ask_channel.type == LTTNG_UST_CHAN_METADATA) {
1280 ret = consumer_metadata_cache_allocate(channel);
1281 if (ret < 0) {
1282 ERR("Allocating metadata cache");
1283 goto end_channel_error;
1284 }
1285 consumer_timer_switch_start(channel, attr.switch_timer_interval);
1286 attr.switch_timer_interval = 0;
1287 } else {
1288 consumer_timer_live_start(channel,
1289 msg.u.ask_channel.live_timer_interval);
1290 }
1291
1292 health_code_update();
1293
1294 /*
1295 * Add the channel to the internal state AFTER all streams were created
1296 * and successfully sent to session daemon. This way, all streams must
1297 * be ready before this channel is visible to the threads.
1298 * If add_channel succeeds, ownership of the channel is
1299 * passed to consumer_thread_channel_poll().
1300 */
1301 ret = add_channel(channel, ctx);
1302 if (ret < 0) {
1303 if (msg.u.ask_channel.type == LTTNG_UST_CHAN_METADATA) {
1304 if (channel->switch_timer_enabled == 1) {
1305 consumer_timer_switch_stop(channel);
1306 }
1307 consumer_metadata_cache_destroy(channel);
1308 }
1309 if (channel->live_timer_enabled == 1) {
1310 consumer_timer_live_stop(channel);
1311 }
1312 goto end_channel_error;
1313 }
1314
1315 health_code_update();
1316
1317 /*
1318 * Channel and streams are now created. Inform the session daemon that
1319 * everything went well and should wait to receive the channel and
1320 * streams with ustctl API.
1321 */
1322 ret = consumer_send_status_channel(sock, channel);
1323 if (ret < 0) {
1324 /*
1325 * There is probably a problem on the socket.
1326 */
1327 goto error_fatal;
1328 }
1329
1330 break;
1331 }
1332 case LTTNG_CONSUMER_GET_CHANNEL:
1333 {
1334 int ret, relayd_err = 0;
1335 uint64_t key = msg.u.get_channel.key;
1336 struct lttng_consumer_channel *channel;
1337
1338 channel = consumer_find_channel(key);
1339 if (!channel) {
1340 ERR("UST consumer get channel key %" PRIu64 " not found", key);
1341 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
1342 goto end_msg_sessiond;
1343 }
1344
1345 health_code_update();
1346
1347 /* Send everything to sessiond. */
1348 ret = send_sessiond_channel(sock, channel, ctx, &relayd_err);
1349 if (ret < 0) {
1350 if (relayd_err) {
1351 /*
1352 * We were unable to send to the relayd the stream so avoid
1353 * sending back a fatal error to the thread since this is OK
1354 * and the consumer can continue its work. The above call
1355 * has sent the error status message to the sessiond.
1356 */
1357 goto end_nosignal;
1358 }
1359 /*
1360 * The communicaton was broken hence there is a bad state between
1361 * the consumer and sessiond so stop everything.
1362 */
1363 goto error_fatal;
1364 }
1365
1366 health_code_update();
1367
1368 /*
1369 * In no monitor mode, the streams ownership is kept inside the channel
1370 * so don't send them to the data thread.
1371 */
1372 if (!channel->monitor) {
1373 goto end_msg_sessiond;
1374 }
1375
1376 ret = send_streams_to_thread(channel, ctx);
1377 if (ret < 0) {
1378 /*
1379 * If we are unable to send the stream to the thread, there is
1380 * a big problem so just stop everything.
1381 */
1382 goto error_fatal;
1383 }
1384 /* List MUST be empty after or else it could be reused. */
1385 assert(cds_list_empty(&channel->streams.head));
1386 goto end_msg_sessiond;
1387 }
1388 case LTTNG_CONSUMER_DESTROY_CHANNEL:
1389 {
1390 uint64_t key = msg.u.destroy_channel.key;
1391
1392 /*
1393 * Only called if streams have not been sent to stream
1394 * manager thread. However, channel has been sent to
1395 * channel manager thread.
1396 */
1397 notify_thread_del_channel(ctx, key);
1398 goto end_msg_sessiond;
1399 }
1400 case LTTNG_CONSUMER_CLOSE_METADATA:
1401 {
1402 int ret;
1403
1404 ret = close_metadata(msg.u.close_metadata.key);
1405 if (ret != 0) {
1406 ret_code = ret;
1407 }
1408
1409 goto end_msg_sessiond;
1410 }
1411 case LTTNG_CONSUMER_FLUSH_CHANNEL:
1412 {
1413 int ret;
1414
1415 ret = flush_channel(msg.u.flush_channel.key);
1416 if (ret != 0) {
1417 ret_code = ret;
1418 }
1419
1420 goto end_msg_sessiond;
1421 }
1422 case LTTNG_CONSUMER_PUSH_METADATA:
1423 {
1424 int ret;
1425 uint64_t len = msg.u.push_metadata.len;
1426 uint64_t key = msg.u.push_metadata.key;
1427 uint64_t offset = msg.u.push_metadata.target_offset;
1428 struct lttng_consumer_channel *channel;
1429
1430 DBG("UST consumer push metadata key %" PRIu64 " of len %" PRIu64, key,
1431 len);
1432
1433 channel = consumer_find_channel(key);
1434 if (!channel) {
1435 /*
1436 * This is possible if the metadata creation on the consumer side
1437 * is in flight vis-a-vis a concurrent push metadata from the
1438 * session daemon. Simply return that the channel failed and the
1439 * session daemon will handle that message correctly considering
1440 * that this race is acceptable thus the DBG() statement here.
1441 */
1442 DBG("UST consumer push metadata %" PRIu64 " not found", key);
1443 ret_code = LTTCOMM_CONSUMERD_CHANNEL_FAIL;
1444 goto end_msg_sessiond;
1445 }
1446
1447 health_code_update();
1448
1449 if (!len) {
1450 /*
1451 * There is nothing to receive. We have simply
1452 * checked whether the channel can be found.
1453 */
1454 ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1455 goto end_msg_sessiond;
1456 }
1457
1458 /* Tell session daemon we are ready to receive the metadata. */
1459 ret = consumer_send_status_msg(sock, LTTCOMM_CONSUMERD_SUCCESS);
1460 if (ret < 0) {
1461 /* Somehow, the session daemon is not responding anymore. */
1462 goto error_fatal;
1463 }
1464
1465 health_code_update();
1466
1467 /* Wait for more data. */
1468 health_poll_entry();
1469 ret = lttng_consumer_poll_socket(consumer_sockpoll);
1470 health_poll_exit();
1471 if (ret) {
1472 goto error_fatal;
1473 }
1474
1475 health_code_update();
1476
1477 ret = lttng_ustconsumer_recv_metadata(sock, key, offset,
1478 len, channel, 0, 1);
1479 if (ret < 0) {
1480 /* error receiving from sessiond */
1481 goto error_fatal;
1482 } else {
1483 ret_code = ret;
1484 goto end_msg_sessiond;
1485 }
1486 }
1487 case LTTNG_CONSUMER_SETUP_METADATA:
1488 {
1489 int ret;
1490
1491 ret = setup_metadata(ctx, msg.u.setup_metadata.key);
1492 if (ret) {
1493 ret_code = ret;
1494 }
1495 goto end_msg_sessiond;
1496 }
1497 case LTTNG_CONSUMER_SNAPSHOT_CHANNEL:
1498 {
1499 if (msg.u.snapshot_channel.metadata) {
1500 ret = snapshot_metadata(msg.u.snapshot_channel.key,
1501 msg.u.snapshot_channel.pathname,
1502 msg.u.snapshot_channel.relayd_id,
1503 ctx);
1504 if (ret < 0) {
1505 ERR("Snapshot metadata failed");
1506 ret_code = LTTCOMM_CONSUMERD_ERROR_METADATA;
1507 }
1508 } else {
1509 ret = snapshot_channel(msg.u.snapshot_channel.key,
1510 msg.u.snapshot_channel.pathname,
1511 msg.u.snapshot_channel.relayd_id,
1512 msg.u.snapshot_channel.nb_packets_per_stream,
1513 ctx);
1514 if (ret < 0) {
1515 ERR("Snapshot channel failed");
1516 ret_code = LTTCOMM_CONSUMERD_CHANNEL_FAIL;
1517 }
1518 }
1519
1520 health_code_update();
1521 ret = consumer_send_status_msg(sock, ret_code);
1522 if (ret < 0) {
1523 /* Somehow, the session daemon is not responding anymore. */
1524 goto end_nosignal;
1525 }
1526 health_code_update();
1527 break;
1528 }
1529 default:
1530 break;
1531 }
1532
1533 end_nosignal:
1534 rcu_read_unlock();
1535
1536 health_code_update();
1537
1538 /*
1539 * Return 1 to indicate success since the 0 value can be a socket
1540 * shutdown during the recv() or send() call.
1541 */
1542 return 1;
1543
1544 end_msg_sessiond:
1545 /*
1546 * The returned value here is not useful since either way we'll return 1 to
1547 * the caller because the session daemon socket management is done
1548 * elsewhere. Returning a negative code or 0 will shutdown the consumer.
1549 */
1550 ret = consumer_send_status_msg(sock, ret_code);
1551 if (ret < 0) {
1552 goto error_fatal;
1553 }
1554 rcu_read_unlock();
1555
1556 health_code_update();
1557
1558 return 1;
1559 end_channel_error:
1560 if (channel) {
1561 /*
1562 * Free channel here since no one has a reference to it. We don't
1563 * free after that because a stream can store this pointer.
1564 */
1565 destroy_channel(channel);
1566 }
1567 /* We have to send a status channel message indicating an error. */
1568 ret = consumer_send_status_channel(sock, NULL);
1569 if (ret < 0) {
1570 /* Stop everything if session daemon can not be notified. */
1571 goto error_fatal;
1572 }
1573 rcu_read_unlock();
1574
1575 health_code_update();
1576
1577 return 1;
1578 error_fatal:
1579 rcu_read_unlock();
1580 /* This will issue a consumer stop. */
1581 return -1;
1582 }
1583
1584 /*
1585 * Wrapper over the mmap() read offset from ust-ctl library. Since this can be
1586 * compiled out, we isolate it in this library.
1587 */
1588 int lttng_ustctl_get_mmap_read_offset(struct lttng_consumer_stream *stream,
1589 unsigned long *off)
1590 {
1591 assert(stream);
1592 assert(stream->ustream);
1593
1594 return ustctl_get_mmap_read_offset(stream->ustream, off);
1595 }
1596
1597 /*
1598 * Wrapper over the mmap() read offset from ust-ctl library. Since this can be
1599 * compiled out, we isolate it in this library.
1600 */
1601 void *lttng_ustctl_get_mmap_base(struct lttng_consumer_stream *stream)
1602 {
1603 assert(stream);
1604 assert(stream->ustream);
1605
1606 return ustctl_get_mmap_base(stream->ustream);
1607 }
1608
1609 /*
1610 * Take a snapshot for a specific fd
1611 *
1612 * Returns 0 on success, < 0 on error
1613 */
1614 int lttng_ustconsumer_take_snapshot(struct lttng_consumer_stream *stream)
1615 {
1616 assert(stream);
1617 assert(stream->ustream);
1618
1619 return ustctl_snapshot(stream->ustream);
1620 }
1621
1622 /*
1623 * Get the produced position
1624 *
1625 * Returns 0 on success, < 0 on error
1626 */
1627 int lttng_ustconsumer_get_produced_snapshot(
1628 struct lttng_consumer_stream *stream, unsigned long *pos)
1629 {
1630 assert(stream);
1631 assert(stream->ustream);
1632 assert(pos);
1633
1634 return ustctl_snapshot_get_produced(stream->ustream, pos);
1635 }
1636
1637 /*
1638 * Get the consumed position
1639 *
1640 * Returns 0 on success, < 0 on error
1641 */
1642 int lttng_ustconsumer_get_consumed_snapshot(
1643 struct lttng_consumer_stream *stream, unsigned long *pos)
1644 {
1645 assert(stream);
1646 assert(stream->ustream);
1647 assert(pos);
1648
1649 return ustctl_snapshot_get_consumed(stream->ustream, pos);
1650 }
1651
1652 void lttng_ustconsumer_flush_buffer(struct lttng_consumer_stream *stream,
1653 int producer)
1654 {
1655 assert(stream);
1656 assert(stream->ustream);
1657
1658 ustctl_flush_buffer(stream->ustream, producer);
1659 }
1660
1661 int lttng_ustconsumer_get_current_timestamp(
1662 struct lttng_consumer_stream *stream, uint64_t *ts)
1663 {
1664 assert(stream);
1665 assert(stream->ustream);
1666 assert(ts);
1667
1668 return ustctl_get_current_timestamp(stream->ustream, ts);
1669 }
1670
1671 /*
1672 * Called when the stream signal the consumer that it has hang up.
1673 */
1674 void lttng_ustconsumer_on_stream_hangup(struct lttng_consumer_stream *stream)
1675 {
1676 assert(stream);
1677 assert(stream->ustream);
1678
1679 ustctl_flush_buffer(stream->ustream, 0);
1680 stream->hangup_flush_done = 1;
1681 }
1682
1683 void lttng_ustconsumer_del_channel(struct lttng_consumer_channel *chan)
1684 {
1685 assert(chan);
1686 assert(chan->uchan);
1687
1688 if (chan->switch_timer_enabled == 1) {
1689 consumer_timer_switch_stop(chan);
1690 }
1691 }
1692
1693 void lttng_ustconsumer_free_channel(struct lttng_consumer_channel *chan)
1694 {
1695 assert(chan);
1696 assert(chan->uchan);
1697
1698 consumer_metadata_cache_destroy(chan);
1699 ustctl_destroy_channel(chan->uchan);
1700 }
1701
1702 void lttng_ustconsumer_del_stream(struct lttng_consumer_stream *stream)
1703 {
1704 assert(stream);
1705 assert(stream->ustream);
1706
1707 if (stream->chan->switch_timer_enabled == 1) {
1708 consumer_timer_switch_stop(stream->chan);
1709 }
1710 ustctl_destroy_stream(stream->ustream);
1711 }
1712
1713 int lttng_ustconsumer_get_wakeup_fd(struct lttng_consumer_stream *stream)
1714 {
1715 assert(stream);
1716 assert(stream->ustream);
1717
1718 return ustctl_stream_get_wakeup_fd(stream->ustream);
1719 }
1720
1721 int lttng_ustconsumer_close_wakeup_fd(struct lttng_consumer_stream *stream)
1722 {
1723 assert(stream);
1724 assert(stream->ustream);
1725
1726 return ustctl_stream_close_wakeup_fd(stream->ustream);
1727 }
1728
1729 /*
1730 * Populate index values of a UST stream. Values are set in big endian order.
1731 *
1732 * Return 0 on success or else a negative value.
1733 */
1734 static int get_index_values(struct ctf_packet_index *index,
1735 struct ustctl_consumer_stream *ustream)
1736 {
1737 int ret;
1738
1739 ret = ustctl_get_timestamp_begin(ustream, &index->timestamp_begin);
1740 if (ret < 0) {
1741 PERROR("ustctl_get_timestamp_begin");
1742 goto error;
1743 }
1744 index->timestamp_begin = htobe64(index->timestamp_begin);
1745
1746 ret = ustctl_get_timestamp_end(ustream, &index->timestamp_end);
1747 if (ret < 0) {
1748 PERROR("ustctl_get_timestamp_end");
1749 goto error;
1750 }
1751 index->timestamp_end = htobe64(index->timestamp_end);
1752
1753 ret = ustctl_get_events_discarded(ustream, &index->events_discarded);
1754 if (ret < 0) {
1755 PERROR("ustctl_get_events_discarded");
1756 goto error;
1757 }
1758 index->events_discarded = htobe64(index->events_discarded);
1759
1760 ret = ustctl_get_content_size(ustream, &index->content_size);
1761 if (ret < 0) {
1762 PERROR("ustctl_get_content_size");
1763 goto error;
1764 }
1765 index->content_size = htobe64(index->content_size);
1766
1767 ret = ustctl_get_packet_size(ustream, &index->packet_size);
1768 if (ret < 0) {
1769 PERROR("ustctl_get_packet_size");
1770 goto error;
1771 }
1772 index->packet_size = htobe64(index->packet_size);
1773
1774 ret = ustctl_get_stream_id(ustream, &index->stream_id);
1775 if (ret < 0) {
1776 PERROR("ustctl_get_stream_id");
1777 goto error;
1778 }
1779 index->stream_id = htobe64(index->stream_id);
1780
1781 error:
1782 return ret;
1783 }
1784
1785 /*
1786 * Write up to one packet from the metadata cache to the channel.
1787 *
1788 * Returns the number of bytes pushed in the cache, or a negative value
1789 * on error.
1790 */
1791 static
1792 int commit_one_metadata_packet(struct lttng_consumer_stream *stream)
1793 {
1794 ssize_t write_len;
1795 int ret;
1796
1797 pthread_mutex_lock(&stream->chan->metadata_cache->lock);
1798 if (stream->chan->metadata_cache->max_offset
1799 == stream->ust_metadata_pushed) {
1800 ret = 0;
1801 goto end;
1802 }
1803
1804 write_len = ustctl_write_one_packet_to_channel(stream->chan->uchan,
1805 &stream->chan->metadata_cache->data[stream->ust_metadata_pushed],
1806 stream->chan->metadata_cache->max_offset
1807 - stream->ust_metadata_pushed);
1808 assert(write_len != 0);
1809 if (write_len < 0) {
1810 ERR("Writing one metadata packet");
1811 ret = -1;
1812 goto end;
1813 }
1814 stream->ust_metadata_pushed += write_len;
1815
1816 assert(stream->chan->metadata_cache->max_offset >=
1817 stream->ust_metadata_pushed);
1818 ret = write_len;
1819
1820 end:
1821 pthread_mutex_unlock(&stream->chan->metadata_cache->lock);
1822 return ret;
1823 }
1824
1825
1826 /*
1827 * Sync metadata meaning request them to the session daemon and snapshot to the
1828 * metadata thread can consumer them.
1829 *
1830 * Metadata stream lock is held here, but we need to release it when
1831 * interacting with sessiond, else we cause a deadlock with live
1832 * awaiting on metadata to be pushed out.
1833 *
1834 * Return 0 if new metadatda is available, EAGAIN if the metadata stream
1835 * is empty or a negative value on error.
1836 */
1837 int lttng_ustconsumer_sync_metadata(struct lttng_consumer_local_data *ctx,
1838 struct lttng_consumer_stream *metadata)
1839 {
1840 int ret;
1841 int retry = 0;
1842
1843 assert(ctx);
1844 assert(metadata);
1845
1846 pthread_mutex_unlock(&metadata->lock);
1847 /*
1848 * Request metadata from the sessiond, but don't wait for the flush
1849 * because we locked the metadata thread.
1850 */
1851 ret = lttng_ustconsumer_request_metadata(ctx, metadata->chan, 0, 0);
1852 if (ret < 0) {
1853 goto end;
1854 }
1855 pthread_mutex_lock(&metadata->lock);
1856
1857 ret = commit_one_metadata_packet(metadata);
1858 if (ret <= 0) {
1859 goto end;
1860 } else if (ret > 0) {
1861 retry = 1;
1862 }
1863
1864 ustctl_flush_buffer(metadata->ustream, 1);
1865 ret = ustctl_snapshot(metadata->ustream);
1866 if (ret < 0) {
1867 if (errno != EAGAIN) {
1868 ERR("Sync metadata, taking UST snapshot");
1869 goto end;
1870 }
1871 DBG("No new metadata when syncing them.");
1872 /* No new metadata, exit. */
1873 ret = ENODATA;
1874 goto end;
1875 }
1876
1877 /*
1878 * After this flush, we still need to extract metadata.
1879 */
1880 if (retry) {
1881 ret = EAGAIN;
1882 }
1883
1884 end:
1885 return ret;
1886 }
1887
1888 /*
1889 * Return 0 on success else a negative value.
1890 */
1891 static int notify_if_more_data(struct lttng_consumer_stream *stream,
1892 struct lttng_consumer_local_data *ctx)
1893 {
1894 int ret;
1895 struct ustctl_consumer_stream *ustream;
1896
1897 assert(stream);
1898 assert(ctx);
1899
1900 ustream = stream->ustream;
1901
1902 /*
1903 * First, we are going to check if there is a new subbuffer available
1904 * before reading the stream wait_fd.
1905 */
1906 /* Get the next subbuffer */
1907 ret = ustctl_get_next_subbuf(ustream);
1908 if (ret) {
1909 /* No more data found, flag the stream. */
1910 stream->has_data = 0;
1911 ret = 0;
1912 goto end;
1913 }
1914
1915 ret = ustctl_put_subbuf(ustream);
1916 assert(!ret);
1917
1918 /* This stream still has data. Flag it and wake up the data thread. */
1919 stream->has_data = 1;
1920
1921 if (stream->monitor && !stream->hangup_flush_done && !ctx->has_wakeup) {
1922 ssize_t writelen;
1923
1924 writelen = lttng_pipe_write(ctx->consumer_wakeup_pipe, "!", 1);
1925 if (writelen < 0 && errno != EAGAIN && errno != EWOULDBLOCK) {
1926 ret = writelen;
1927 goto end;
1928 }
1929
1930 /* The wake up pipe has been notified. */
1931 ctx->has_wakeup = 1;
1932 }
1933 ret = 0;
1934
1935 end:
1936 return ret;
1937 }
1938
1939 /*
1940 * Read subbuffer from the given stream.
1941 *
1942 * Stream lock MUST be acquired.
1943 *
1944 * Return 0 on success else a negative value.
1945 */
1946 int lttng_ustconsumer_read_subbuffer(struct lttng_consumer_stream *stream,
1947 struct lttng_consumer_local_data *ctx)
1948 {
1949 unsigned long len, subbuf_size, padding;
1950 int err, write_index = 1;
1951 long ret = 0;
1952 struct ustctl_consumer_stream *ustream;
1953 struct ctf_packet_index index;
1954
1955 assert(stream);
1956 assert(stream->ustream);
1957 assert(ctx);
1958
1959 DBG("In UST read_subbuffer (wait_fd: %d, name: %s)", stream->wait_fd,
1960 stream->name);
1961
1962 /* Ease our life for what's next. */
1963 ustream = stream->ustream;
1964
1965 /*
1966 * We can consume the 1 byte written into the wait_fd by UST. Don't trigger
1967 * error if we cannot read this one byte (read returns 0), or if the error
1968 * is EAGAIN or EWOULDBLOCK.
1969 *
1970 * This is only done when the stream is monitored by a thread, before the
1971 * flush is done after a hangup and if the stream is not flagged with data
1972 * since there might be nothing to consume in the wait fd but still have
1973 * data available flagged by the consumer wake up pipe.
1974 */
1975 if (stream->monitor && !stream->hangup_flush_done && !stream->has_data) {
1976 char dummy;
1977 ssize_t readlen;
1978
1979 readlen = lttng_read(stream->wait_fd, &dummy, 1);
1980 if (readlen < 0 && errno != EAGAIN && errno != EWOULDBLOCK) {
1981 ret = readlen;
1982 goto end;
1983 }
1984 }
1985
1986 retry:
1987 /* Get the next subbuffer */
1988 err = ustctl_get_next_subbuf(ustream);
1989 if (err != 0) {
1990 /*
1991 * Populate metadata info if the existing info has
1992 * already been read.
1993 */
1994 if (stream->metadata_flag) {
1995 ret = commit_one_metadata_packet(stream);
1996 if (ret <= 0) {
1997 goto end;
1998 }
1999 ustctl_flush_buffer(stream->ustream, 1);
2000 goto retry;
2001 }
2002
2003 ret = err; /* ustctl_get_next_subbuf returns negative, caller expect positive. */
2004 /*
2005 * This is a debug message even for single-threaded consumer,
2006 * because poll() have more relaxed criterions than get subbuf,
2007 * so get_subbuf may fail for short race windows where poll()
2008 * would issue wakeups.
2009 */
2010 DBG("Reserving sub buffer failed (everything is normal, "
2011 "it is due to concurrency) [ret: %d]", err);
2012 goto end;
2013 }
2014 assert(stream->chan->output == CONSUMER_CHANNEL_MMAP);
2015
2016 if (!stream->metadata_flag) {
2017 index.offset = htobe64(stream->out_fd_offset);
2018 ret = get_index_values(&index, ustream);
2019 if (ret < 0) {
2020 goto end;
2021 }
2022 } else {
2023 write_index = 0;
2024 }
2025
2026 /* Get the full padded subbuffer size */
2027 err = ustctl_get_padded_subbuf_size(ustream, &len);
2028 assert(err == 0);
2029
2030 /* Get subbuffer data size (without padding) */
2031 err = ustctl_get_subbuf_size(ustream, &subbuf_size);
2032 assert(err == 0);
2033
2034 /* Make sure we don't get a subbuffer size bigger than the padded */
2035 assert(len >= subbuf_size);
2036
2037 padding = len - subbuf_size;
2038 /* write the subbuffer to the tracefile */
2039 ret = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, subbuf_size, padding, &index);
2040 /*
2041 * The mmap operation should write subbuf_size amount of data when network
2042 * streaming or the full padding (len) size when we are _not_ streaming.
2043 */
2044 if ((ret != subbuf_size && stream->net_seq_idx != (uint64_t) -1ULL) ||
2045 (ret != len && stream->net_seq_idx == (uint64_t) -1ULL)) {
2046 /*
2047 * Display the error but continue processing to try to release the
2048 * subbuffer. This is a DBG statement since any unexpected kill or
2049 * signal, the application gets unregistered, relayd gets closed or
2050 * anything that affects the buffer lifetime will trigger this error.
2051 * So, for the sake of the user, don't print this error since it can
2052 * happen and it is OK with the code flow.
2053 */
2054 DBG("Error writing to tracefile "
2055 "(ret: %ld != len: %lu != subbuf_size: %lu)",
2056 ret, len, subbuf_size);
2057 write_index = 0;
2058 }
2059 err = ustctl_put_next_subbuf(ustream);
2060 assert(err == 0);
2061
2062 /*
2063 * This will consumer the byte on the wait_fd if and only if there is not
2064 * next subbuffer to be acquired.
2065 */
2066 if (!stream->metadata_flag) {
2067 ret = notify_if_more_data(stream, ctx);
2068 if (ret < 0) {
2069 goto end;
2070 }
2071 }
2072
2073 /* Write index if needed. */
2074 if (!write_index) {
2075 goto end;
2076 }
2077
2078 if (stream->chan->live_timer_interval && !stream->metadata_flag) {
2079 /*
2080 * In live, block until all the metadata is sent.
2081 */
2082 pthread_mutex_lock(&stream->metadata_timer_lock);
2083 assert(!stream->missed_metadata_flush);
2084 stream->waiting_on_metadata = true;
2085 pthread_mutex_unlock(&stream->metadata_timer_lock);
2086
2087 err = consumer_stream_sync_metadata(ctx, stream->session_id);
2088
2089 pthread_mutex_lock(&stream->metadata_timer_lock);
2090 stream->waiting_on_metadata = false;
2091 if (stream->missed_metadata_flush) {
2092 stream->missed_metadata_flush = false;
2093 pthread_mutex_unlock(&stream->metadata_timer_lock);
2094 (void) consumer_flush_ust_index(stream);
2095 } else {
2096 pthread_mutex_unlock(&stream->metadata_timer_lock);
2097 }
2098
2099 if (err < 0) {
2100 goto end;
2101 }
2102 }
2103
2104 assert(!stream->metadata_flag);
2105 err = consumer_stream_write_index(stream, &index);
2106 if (err < 0) {
2107 goto end;
2108 }
2109
2110 end:
2111 return ret;
2112 }
2113
2114 /*
2115 * Called when a stream is created.
2116 *
2117 * Return 0 on success or else a negative value.
2118 */
2119 int lttng_ustconsumer_on_recv_stream(struct lttng_consumer_stream *stream)
2120 {
2121 int ret;
2122
2123 assert(stream);
2124
2125 /* Don't create anything if this is set for streaming. */
2126 if (stream->net_seq_idx == (uint64_t) -1ULL && stream->chan->monitor) {
2127 ret = utils_create_stream_file(stream->chan->pathname, stream->name,
2128 stream->chan->tracefile_size, stream->tracefile_count_current,
2129 stream->uid, stream->gid, NULL);
2130 if (ret < 0) {
2131 goto error;
2132 }
2133 stream->out_fd = ret;
2134 stream->tracefile_size_current = 0;
2135
2136 if (!stream->metadata_flag) {
2137 ret = index_create_file(stream->chan->pathname,
2138 stream->name, stream->uid, stream->gid,
2139 stream->chan->tracefile_size,
2140 stream->tracefile_count_current);
2141 if (ret < 0) {
2142 goto error;
2143 }
2144 stream->index_fd = ret;
2145 }
2146 }
2147 ret = 0;
2148
2149 error:
2150 return ret;
2151 }
2152
2153 /*
2154 * Check if data is still being extracted from the buffers for a specific
2155 * stream. Consumer data lock MUST be acquired before calling this function
2156 * and the stream lock.
2157 *
2158 * Return 1 if the traced data are still getting read else 0 meaning that the
2159 * data is available for trace viewer reading.
2160 */
2161 int lttng_ustconsumer_data_pending(struct lttng_consumer_stream *stream)
2162 {
2163 int ret;
2164
2165 assert(stream);
2166 assert(stream->ustream);
2167
2168 DBG("UST consumer checking data pending");
2169
2170 if (stream->endpoint_status != CONSUMER_ENDPOINT_ACTIVE) {
2171 ret = 0;
2172 goto end;
2173 }
2174
2175 if (stream->chan->type == CONSUMER_CHANNEL_TYPE_METADATA) {
2176 uint64_t contiguous, pushed;
2177
2178 /* Ease our life a bit. */
2179 contiguous = stream->chan->metadata_cache->max_offset;
2180 pushed = stream->ust_metadata_pushed;
2181
2182 /*
2183 * We can simply check whether all contiguously available data
2184 * has been pushed to the ring buffer, since the push operation
2185 * is performed within get_next_subbuf(), and because both
2186 * get_next_subbuf() and put_next_subbuf() are issued atomically
2187 * thanks to the stream lock within
2188 * lttng_ustconsumer_read_subbuffer(). This basically means that
2189 * whetnever ust_metadata_pushed is incremented, the associated
2190 * metadata has been consumed from the metadata stream.
2191 */
2192 DBG("UST consumer metadata pending check: contiguous %" PRIu64 " vs pushed %" PRIu64,
2193 contiguous, pushed);
2194 assert(((int64_t) (contiguous - pushed)) >= 0);
2195 if ((contiguous != pushed) ||
2196 (((int64_t) contiguous - pushed) > 0 || contiguous == 0)) {
2197 ret = 1; /* Data is pending */
2198 goto end;
2199 }
2200 } else {
2201 ret = ustctl_get_next_subbuf(stream->ustream);
2202 if (ret == 0) {
2203 /*
2204 * There is still data so let's put back this
2205 * subbuffer.
2206 */
2207 ret = ustctl_put_subbuf(stream->ustream);
2208 assert(ret == 0);
2209 ret = 1; /* Data is pending */
2210 goto end;
2211 }
2212 }
2213
2214 /* Data is NOT pending so ready to be read. */
2215 ret = 0;
2216
2217 end:
2218 return ret;
2219 }
2220
2221 /*
2222 * Stop a given metadata channel timer if enabled and close the wait fd which
2223 * is the poll pipe of the metadata stream.
2224 *
2225 * This MUST be called with the metadata channel acquired.
2226 */
2227 void lttng_ustconsumer_close_metadata(struct lttng_consumer_channel *metadata)
2228 {
2229 int ret;
2230
2231 assert(metadata);
2232 assert(metadata->type == CONSUMER_CHANNEL_TYPE_METADATA);
2233
2234 DBG("Closing metadata channel key %" PRIu64, metadata->key);
2235
2236 if (metadata->switch_timer_enabled == 1) {
2237 consumer_timer_switch_stop(metadata);
2238 }
2239
2240 if (!metadata->metadata_stream) {
2241 goto end;
2242 }
2243
2244 /*
2245 * Closing write side so the thread monitoring the stream wakes up if any
2246 * and clean the metadata stream.
2247 */
2248 if (metadata->metadata_stream->ust_metadata_poll_pipe[1] >= 0) {
2249 ret = close(metadata->metadata_stream->ust_metadata_poll_pipe[1]);
2250 if (ret < 0) {
2251 PERROR("closing metadata pipe write side");
2252 }
2253 metadata->metadata_stream->ust_metadata_poll_pipe[1] = -1;
2254 }
2255
2256 end:
2257 return;
2258 }
2259
2260 /*
2261 * Close every metadata stream wait fd of the metadata hash table. This
2262 * function MUST be used very carefully so not to run into a race between the
2263 * metadata thread handling streams and this function closing their wait fd.
2264 *
2265 * For UST, this is used when the session daemon hangs up. Its the metadata
2266 * producer so calling this is safe because we are assured that no state change
2267 * can occur in the metadata thread for the streams in the hash table.
2268 */
2269 void lttng_ustconsumer_close_all_metadata(struct lttng_ht *metadata_ht)
2270 {
2271 struct lttng_ht_iter iter;
2272 struct lttng_consumer_stream *stream;
2273
2274 assert(metadata_ht);
2275 assert(metadata_ht->ht);
2276
2277 DBG("UST consumer closing all metadata streams");
2278
2279 rcu_read_lock();
2280 cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream,
2281 node.node) {
2282
2283 health_code_update();
2284
2285 pthread_mutex_lock(&stream->chan->lock);
2286 lttng_ustconsumer_close_metadata(stream->chan);
2287 pthread_mutex_unlock(&stream->chan->lock);
2288
2289 }
2290 rcu_read_unlock();
2291 }
2292
2293 void lttng_ustconsumer_close_stream_wakeup(struct lttng_consumer_stream *stream)
2294 {
2295 int ret;
2296
2297 ret = ustctl_stream_close_wakeup_fd(stream->ustream);
2298 if (ret < 0) {
2299 ERR("Unable to close wakeup fd");
2300 }
2301 }
2302
2303 /*
2304 * Please refer to consumer-timer.c before adding any lock within this
2305 * function or any of its callees. Timers have a very strict locking
2306 * semantic with respect to teardown. Failure to respect this semantic
2307 * introduces deadlocks.
2308 *
2309 * DON'T hold the metadata lock when calling this function, else this
2310 * can cause deadlock involving consumer awaiting for metadata to be
2311 * pushed out due to concurrent interaction with the session daemon.
2312 */
2313 int lttng_ustconsumer_request_metadata(struct lttng_consumer_local_data *ctx,
2314 struct lttng_consumer_channel *channel, int timer, int wait)
2315 {
2316 struct lttcomm_metadata_request_msg request;
2317 struct lttcomm_consumer_msg msg;
2318 enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS;
2319 uint64_t len, key, offset;
2320 int ret;
2321
2322 assert(channel);
2323 assert(channel->metadata_cache);
2324
2325 memset(&request, 0, sizeof(request));
2326
2327 /* send the metadata request to sessiond */
2328 switch (consumer_data.type) {
2329 case LTTNG_CONSUMER64_UST:
2330 request.bits_per_long = 64;
2331 break;
2332 case LTTNG_CONSUMER32_UST:
2333 request.bits_per_long = 32;
2334 break;
2335 default:
2336 request.bits_per_long = 0;
2337 break;
2338 }
2339
2340 request.session_id = channel->session_id;
2341 request.session_id_per_pid = channel->session_id_per_pid;
2342 /*
2343 * Request the application UID here so the metadata of that application can
2344 * be sent back. The channel UID corresponds to the user UID of the session
2345 * used for the rights on the stream file(s).
2346 */
2347 request.uid = channel->ust_app_uid;
2348 request.key = channel->key;
2349
2350 DBG("Sending metadata request to sessiond, session id %" PRIu64
2351 ", per-pid %" PRIu64 ", app UID %u and channek key %" PRIu64,
2352 request.session_id, request.session_id_per_pid, request.uid,
2353 request.key);
2354
2355 pthread_mutex_lock(&ctx->metadata_socket_lock);
2356
2357 health_code_update();
2358
2359 ret = lttcomm_send_unix_sock(ctx->consumer_metadata_socket, &request,
2360 sizeof(request));
2361 if (ret < 0) {
2362 ERR("Asking metadata to sessiond");
2363 goto end;
2364 }
2365
2366 health_code_update();
2367
2368 /* Receive the metadata from sessiond */
2369 ret = lttcomm_recv_unix_sock(ctx->consumer_metadata_socket, &msg,
2370 sizeof(msg));
2371 if (ret != sizeof(msg)) {
2372 DBG("Consumer received unexpected message size %d (expects %zu)",
2373 ret, sizeof(msg));
2374 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD);
2375 /*
2376 * The ret value might 0 meaning an orderly shutdown but this is ok
2377 * since the caller handles this.
2378 */
2379 goto end;
2380 }
2381
2382 health_code_update();
2383
2384 if (msg.cmd_type == LTTNG_ERR_UND) {
2385 /* No registry found */
2386 (void) consumer_send_status_msg(ctx->consumer_metadata_socket,
2387 ret_code);
2388 ret = 0;
2389 goto end;
2390 } else if (msg.cmd_type != LTTNG_CONSUMER_PUSH_METADATA) {
2391 ERR("Unexpected cmd_type received %d", msg.cmd_type);
2392 ret = -1;
2393 goto end;
2394 }
2395
2396 len = msg.u.push_metadata.len;
2397 key = msg.u.push_metadata.key;
2398 offset = msg.u.push_metadata.target_offset;
2399
2400 assert(key == channel->key);
2401 if (len == 0) {
2402 DBG("No new metadata to receive for key %" PRIu64, key);
2403 }
2404
2405 health_code_update();
2406
2407 /* Tell session daemon we are ready to receive the metadata. */
2408 ret = consumer_send_status_msg(ctx->consumer_metadata_socket,
2409 LTTCOMM_CONSUMERD_SUCCESS);
2410 if (ret < 0 || len == 0) {
2411 /*
2412 * Somehow, the session daemon is not responding anymore or there is
2413 * nothing to receive.
2414 */
2415 goto end;
2416 }
2417
2418 health_code_update();
2419
2420 ret = lttng_ustconsumer_recv_metadata(ctx->consumer_metadata_socket,
2421 key, offset, len, channel, timer, wait);
2422 if (ret >= 0) {
2423 /*
2424 * Only send the status msg if the sessiond is alive meaning a positive
2425 * ret code.
2426 */
2427 (void) consumer_send_status_msg(ctx->consumer_metadata_socket, ret);
2428 }
2429 ret = 0;
2430
2431 end:
2432 health_code_update();
2433
2434 pthread_mutex_unlock(&ctx->metadata_socket_lock);
2435 return ret;
2436 }
2437
2438 /*
2439 * Return the ustctl call for the get stream id.
2440 */
2441 int lttng_ustconsumer_get_stream_id(struct lttng_consumer_stream *stream,
2442 uint64_t *stream_id)
2443 {
2444 assert(stream);
2445 assert(stream_id);
2446
2447 return ustctl_get_stream_id(stream->ustream, stream_id);
2448 }
This page took 0.083594 seconds and 4 git commands to generate.