Fix deadlock: don't take channel lock in timer
[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 <common/common.h>
36 #include <common/sessiond-comm/sessiond-comm.h>
37 #include <common/relayd/relayd.h>
38 #include <common/compat/fcntl.h>
39 #include <common/consumer-metadata-cache.h>
40 #include <common/consumer-stream.h>
41 #include <common/consumer-timer.h>
42 #include <common/utils.h>
43
44 #include "ust-consumer.h"
45
46 extern struct lttng_consumer_global_data consumer_data;
47 extern int consumer_poll_timeout;
48 extern volatile int consumer_quit;
49
50 /*
51 * Free channel object and all streams associated with it. This MUST be used
52 * only and only if the channel has _NEVER_ been added to the global channel
53 * hash table.
54 */
55 static void destroy_channel(struct lttng_consumer_channel *channel)
56 {
57 struct lttng_consumer_stream *stream, *stmp;
58
59 assert(channel);
60
61 DBG("UST consumer cleaning stream list");
62
63 cds_list_for_each_entry_safe(stream, stmp, &channel->streams.head,
64 send_node) {
65 cds_list_del(&stream->send_node);
66 ustctl_destroy_stream(stream->ustream);
67 free(stream);
68 }
69
70 /*
71 * If a channel is available meaning that was created before the streams
72 * were, delete it.
73 */
74 if (channel->uchan) {
75 lttng_ustconsumer_del_channel(channel);
76 }
77 free(channel);
78 }
79
80 /*
81 * Add channel to internal consumer state.
82 *
83 * Returns 0 on success or else a negative value.
84 */
85 static int add_channel(struct lttng_consumer_channel *channel,
86 struct lttng_consumer_local_data *ctx)
87 {
88 int ret = 0;
89
90 assert(channel);
91 assert(ctx);
92
93 if (ctx->on_recv_channel != NULL) {
94 ret = ctx->on_recv_channel(channel);
95 if (ret == 0) {
96 ret = consumer_add_channel(channel, ctx);
97 } else if (ret < 0) {
98 /* Most likely an ENOMEM. */
99 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
100 goto error;
101 }
102 } else {
103 ret = consumer_add_channel(channel, ctx);
104 }
105
106 DBG("UST consumer channel added (key: %" PRIu64 ")", channel->key);
107
108 error:
109 return ret;
110 }
111
112 /*
113 * Allocate and return a consumer channel object.
114 */
115 static struct lttng_consumer_channel *allocate_channel(uint64_t session_id,
116 const char *pathname, const char *name, uid_t uid, gid_t gid,
117 uint64_t relayd_id, uint64_t key, enum lttng_event_output output,
118 uint64_t tracefile_size, uint64_t tracefile_count,
119 uint64_t session_id_per_pid, unsigned int monitor)
120 {
121 assert(pathname);
122 assert(name);
123
124 return consumer_allocate_channel(key, session_id, pathname, name, uid,
125 gid, relayd_id, output, tracefile_size,
126 tracefile_count, session_id_per_pid, monitor);
127 }
128
129 /*
130 * Allocate and return a consumer stream object. If _alloc_ret is not NULL, the
131 * error value if applicable is set in it else it is kept untouched.
132 *
133 * Return NULL on error else the newly allocated stream object.
134 */
135 static struct lttng_consumer_stream *allocate_stream(int cpu, int key,
136 struct lttng_consumer_channel *channel,
137 struct lttng_consumer_local_data *ctx, int *_alloc_ret)
138 {
139 int alloc_ret;
140 struct lttng_consumer_stream *stream = NULL;
141
142 assert(channel);
143 assert(ctx);
144
145 stream = consumer_allocate_stream(channel->key,
146 key,
147 LTTNG_CONSUMER_ACTIVE_STREAM,
148 channel->name,
149 channel->uid,
150 channel->gid,
151 channel->relayd_id,
152 channel->session_id,
153 cpu,
154 &alloc_ret,
155 channel->type,
156 channel->monitor);
157 if (stream == NULL) {
158 switch (alloc_ret) {
159 case -ENOENT:
160 /*
161 * We could not find the channel. Can happen if cpu hotplug
162 * happens while tearing down.
163 */
164 DBG3("Could not find channel");
165 break;
166 case -ENOMEM:
167 case -EINVAL:
168 default:
169 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
170 break;
171 }
172 goto error;
173 }
174
175 stream->chan = channel;
176
177 error:
178 if (_alloc_ret) {
179 *_alloc_ret = alloc_ret;
180 }
181 return stream;
182 }
183
184 /*
185 * Send the given stream pointer to the corresponding thread.
186 *
187 * Returns 0 on success else a negative value.
188 */
189 static int send_stream_to_thread(struct lttng_consumer_stream *stream,
190 struct lttng_consumer_local_data *ctx)
191 {
192 int ret;
193 struct lttng_pipe *stream_pipe;
194
195 /* Get the right pipe where the stream will be sent. */
196 if (stream->metadata_flag) {
197 stream_pipe = ctx->consumer_metadata_pipe;
198 } else {
199 stream_pipe = ctx->consumer_data_pipe;
200 }
201
202 ret = lttng_pipe_write(stream_pipe, &stream, sizeof(stream));
203 if (ret < 0) {
204 ERR("Consumer write %s stream to pipe %d",
205 stream->metadata_flag ? "metadata" : "data",
206 lttng_pipe_get_writefd(stream_pipe));
207 }
208
209 return ret;
210 }
211
212 /*
213 * Create streams for the given channel using liblttng-ust-ctl.
214 *
215 * Return 0 on success else a negative value.
216 */
217 static int create_ust_streams(struct lttng_consumer_channel *channel,
218 struct lttng_consumer_local_data *ctx)
219 {
220 int ret, cpu = 0;
221 struct ustctl_consumer_stream *ustream;
222 struct lttng_consumer_stream *stream;
223
224 assert(channel);
225 assert(ctx);
226
227 /*
228 * While a stream is available from ustctl. When NULL is returned, we've
229 * reached the end of the possible stream for the channel.
230 */
231 while ((ustream = ustctl_create_stream(channel->uchan, cpu))) {
232 int wait_fd;
233 int ust_metadata_pipe[2];
234
235 if (channel->type == CONSUMER_CHANNEL_TYPE_METADATA && channel->monitor) {
236 ret = utils_create_pipe_cloexec_nonblock(ust_metadata_pipe);
237 if (ret < 0) {
238 ERR("Create ust metadata poll pipe");
239 goto error;
240 }
241 wait_fd = ust_metadata_pipe[0];
242 } else {
243 wait_fd = ustctl_stream_get_wait_fd(ustream);
244 }
245
246 /* Allocate consumer stream object. */
247 stream = allocate_stream(cpu, wait_fd, channel, ctx, &ret);
248 if (!stream) {
249 goto error_alloc;
250 }
251 stream->ustream = ustream;
252 /*
253 * Store it so we can save multiple function calls afterwards since
254 * this value is used heavily in the stream threads. This is UST
255 * specific so this is why it's done after allocation.
256 */
257 stream->wait_fd = wait_fd;
258
259 /*
260 * Increment channel refcount since the channel reference has now been
261 * assigned in the allocation process above.
262 */
263 if (stream->chan->monitor) {
264 uatomic_inc(&stream->chan->refcount);
265 }
266
267 /*
268 * Order is important this is why a list is used. On error, the caller
269 * should clean this list.
270 */
271 cds_list_add_tail(&stream->send_node, &channel->streams.head);
272
273 ret = ustctl_get_max_subbuf_size(stream->ustream,
274 &stream->max_sb_size);
275 if (ret < 0) {
276 ERR("ustctl_get_max_subbuf_size failed for stream %s",
277 stream->name);
278 goto error;
279 }
280
281 /* Do actions once stream has been received. */
282 if (ctx->on_recv_stream) {
283 ret = ctx->on_recv_stream(stream);
284 if (ret < 0) {
285 goto error;
286 }
287 }
288
289 DBG("UST consumer add stream %s (key: %" PRIu64 ") with relayd id %" PRIu64,
290 stream->name, stream->key, stream->relayd_stream_id);
291
292 /* Set next CPU stream. */
293 channel->streams.count = ++cpu;
294
295 /* Keep stream reference when creating metadata. */
296 if (channel->type == CONSUMER_CHANNEL_TYPE_METADATA) {
297 channel->metadata_stream = stream;
298 stream->ust_metadata_poll_pipe[0] = ust_metadata_pipe[0];
299 stream->ust_metadata_poll_pipe[1] = ust_metadata_pipe[1];
300 }
301 }
302
303 return 0;
304
305 error:
306 error_alloc:
307 return ret;
308 }
309
310 /*
311 * Create an UST channel with the given attributes and send it to the session
312 * daemon using the ust ctl API.
313 *
314 * Return 0 on success or else a negative value.
315 */
316 static int create_ust_channel(struct ustctl_consumer_channel_attr *attr,
317 struct ustctl_consumer_channel **chanp)
318 {
319 int ret;
320 struct ustctl_consumer_channel *channel;
321
322 assert(attr);
323 assert(chanp);
324
325 DBG3("Creating channel to ustctl with attr: [overwrite: %d, "
326 "subbuf_size: %" PRIu64 ", num_subbuf: %" PRIu64 ", "
327 "switch_timer_interval: %u, read_timer_interval: %u, "
328 "output: %d, type: %d", attr->overwrite, attr->subbuf_size,
329 attr->num_subbuf, attr->switch_timer_interval,
330 attr->read_timer_interval, attr->output, attr->type);
331
332 channel = ustctl_create_channel(attr);
333 if (!channel) {
334 ret = -1;
335 goto error_create;
336 }
337
338 *chanp = channel;
339
340 return 0;
341
342 error_create:
343 return ret;
344 }
345
346 /*
347 * Send a single given stream to the session daemon using the sock.
348 *
349 * Return 0 on success else a negative value.
350 */
351 static int send_sessiond_stream(int sock, struct lttng_consumer_stream *stream)
352 {
353 int ret;
354
355 assert(stream);
356 assert(sock >= 0);
357
358 DBG("UST consumer sending stream %" PRIu64 " to sessiond", stream->key);
359
360 /* Send stream to session daemon. */
361 ret = ustctl_send_stream_to_sessiond(sock, stream->ustream);
362 if (ret < 0) {
363 goto error;
364 }
365
366 error:
367 return ret;
368 }
369
370 /*
371 * Send channel to sessiond.
372 *
373 * Return 0 on success or else a negative value.
374 */
375 static int send_sessiond_channel(int sock,
376 struct lttng_consumer_channel *channel,
377 struct lttng_consumer_local_data *ctx, int *relayd_error)
378 {
379 int ret, ret_code = LTTNG_OK;
380 struct lttng_consumer_stream *stream;
381
382 assert(channel);
383 assert(ctx);
384 assert(sock >= 0);
385
386 DBG("UST consumer sending channel %s to sessiond", channel->name);
387
388 if (channel->relayd_id != (uint64_t) -1ULL) {
389 cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
390 /* Try to send the stream to the relayd if one is available. */
391 ret = consumer_send_relayd_stream(stream, stream->chan->pathname);
392 if (ret < 0) {
393 /*
394 * Flag that the relayd was the problem here probably due to a
395 * communicaton error on the socket.
396 */
397 if (relayd_error) {
398 *relayd_error = 1;
399 }
400 ret_code = LTTNG_ERR_RELAYD_CONNECT_FAIL;
401 }
402 }
403 }
404
405 /* Inform sessiond that we are about to send channel and streams. */
406 ret = consumer_send_status_msg(sock, ret_code);
407 if (ret < 0 || ret_code != LTTNG_OK) {
408 /*
409 * Either the session daemon is not responding or the relayd died so we
410 * stop now.
411 */
412 goto error;
413 }
414
415 /* Send channel to sessiond. */
416 ret = ustctl_send_channel_to_sessiond(sock, channel->uchan);
417 if (ret < 0) {
418 goto error;
419 }
420
421 ret = ustctl_channel_close_wakeup_fd(channel->uchan);
422 if (ret < 0) {
423 goto error;
424 }
425
426 /* The channel was sent successfully to the sessiond at this point. */
427 cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
428 /* Send stream to session daemon. */
429 ret = send_sessiond_stream(sock, stream);
430 if (ret < 0) {
431 goto error;
432 }
433 }
434
435 /* Tell sessiond there is no more stream. */
436 ret = ustctl_send_stream_to_sessiond(sock, NULL);
437 if (ret < 0) {
438 goto error;
439 }
440
441 DBG("UST consumer NULL stream sent to sessiond");
442
443 return 0;
444
445 error:
446 if (ret_code != LTTNG_OK) {
447 ret = -1;
448 }
449 return ret;
450 }
451
452 /*
453 * Creates a channel and streams and add the channel it to the channel internal
454 * state. The created stream must ONLY be sent once the GET_CHANNEL command is
455 * received.
456 *
457 * Return 0 on success or else, a negative value is returned and the channel
458 * MUST be destroyed by consumer_del_channel().
459 */
460 static int ask_channel(struct lttng_consumer_local_data *ctx, int sock,
461 struct lttng_consumer_channel *channel,
462 struct ustctl_consumer_channel_attr *attr)
463 {
464 int ret;
465
466 assert(ctx);
467 assert(channel);
468 assert(attr);
469
470 /*
471 * This value is still used by the kernel consumer since for the kernel,
472 * the stream ownership is not IN the consumer so we need to have the
473 * number of left stream that needs to be initialized so we can know when
474 * to delete the channel (see consumer.c).
475 *
476 * As for the user space tracer now, the consumer creates and sends the
477 * stream to the session daemon which only sends them to the application
478 * once every stream of a channel is received making this value useless
479 * because we they will be added to the poll thread before the application
480 * receives them. This ensures that a stream can not hang up during
481 * initilization of a channel.
482 */
483 channel->nb_init_stream_left = 0;
484
485 /* The reply msg status is handled in the following call. */
486 ret = create_ust_channel(attr, &channel->uchan);
487 if (ret < 0) {
488 goto end;
489 }
490
491 channel->wait_fd = ustctl_channel_get_wait_fd(channel->uchan);
492
493 /*
494 * For the snapshots (no monitor), we create the metadata streams
495 * on demand, not during the channel creation.
496 */
497 if (channel->type == CONSUMER_CHANNEL_TYPE_METADATA && !channel->monitor) {
498 ret = 0;
499 goto end;
500 }
501
502 /* Open all streams for this channel. */
503 ret = create_ust_streams(channel, ctx);
504 if (ret < 0) {
505 goto end;
506 }
507
508 end:
509 return ret;
510 }
511
512 /*
513 * Send all stream of a channel to the right thread handling it.
514 *
515 * On error, return a negative value else 0 on success.
516 */
517 static int send_streams_to_thread(struct lttng_consumer_channel *channel,
518 struct lttng_consumer_local_data *ctx)
519 {
520 int ret = 0;
521 struct lttng_consumer_stream *stream, *stmp;
522
523 assert(channel);
524 assert(ctx);
525
526 /* Send streams to the corresponding thread. */
527 cds_list_for_each_entry_safe(stream, stmp, &channel->streams.head,
528 send_node) {
529 /* Sending the stream to the thread. */
530 ret = send_stream_to_thread(stream, ctx);
531 if (ret < 0) {
532 /*
533 * If we are unable to send the stream to the thread, there is
534 * a big problem so just stop everything.
535 */
536 goto error;
537 }
538
539 /* Remove node from the channel stream list. */
540 cds_list_del(&stream->send_node);
541
542 /*
543 * From this point on, the stream's ownership has been moved away from
544 * the channel and becomes globally visible.
545 */
546 stream->globally_visible = 1;
547 }
548
549 error:
550 return ret;
551 }
552
553 /*
554 * Flush channel's streams using the given key to retrieve the channel.
555 *
556 * Return 0 on success else an LTTng error code.
557 */
558 static int flush_channel(uint64_t chan_key)
559 {
560 int ret = 0;
561 struct lttng_consumer_channel *channel;
562 struct lttng_consumer_stream *stream;
563 struct lttng_ht *ht;
564 struct lttng_ht_iter iter;
565
566 DBG("UST consumer flush channel key %" PRIu64, chan_key);
567
568 rcu_read_lock();
569 channel = consumer_find_channel(chan_key);
570 if (!channel) {
571 ERR("UST consumer flush channel %" PRIu64 " not found", chan_key);
572 ret = LTTNG_ERR_UST_CHAN_NOT_FOUND;
573 goto error;
574 }
575
576 ht = consumer_data.stream_per_chan_id_ht;
577
578 /* For each stream of the channel id, flush it. */
579 cds_lfht_for_each_entry_duplicate(ht->ht,
580 ht->hash_fct(&channel->key, lttng_ht_seed), ht->match_fct,
581 &channel->key, &iter.iter, stream, node_channel_id.node) {
582 ustctl_flush_buffer(stream->ustream, 1);
583 }
584 error:
585 rcu_read_unlock();
586 return ret;
587 }
588
589 /*
590 * Close metadata stream wakeup_fd using the given key to retrieve the channel.
591 * RCU read side lock MUST be acquired before calling this function.
592 *
593 * Return 0 on success else an LTTng error code.
594 */
595 static int close_metadata(uint64_t chan_key)
596 {
597 int ret = 0;
598 struct lttng_consumer_channel *channel;
599
600 DBG("UST consumer close metadata key %" PRIu64, chan_key);
601
602 channel = consumer_find_channel(chan_key);
603 if (!channel) {
604 /*
605 * This is possible if the metadata thread has issue a delete because
606 * the endpoint point of the stream hung up. There is no way the
607 * session daemon can know about it thus use a DBG instead of an actual
608 * error.
609 */
610 DBG("UST consumer close metadata %" PRIu64 " not found", chan_key);
611 ret = LTTNG_ERR_UST_CHAN_NOT_FOUND;
612 goto error;
613 }
614
615 pthread_mutex_lock(&consumer_data.lock);
616 pthread_mutex_lock(&channel->lock);
617
618 if (cds_lfht_is_node_deleted(&channel->node.node)) {
619 goto error_unlock;
620 }
621
622 if (channel->switch_timer_enabled == 1) {
623 DBG("Deleting timer on metadata channel");
624 consumer_timer_switch_stop(channel);
625 }
626
627 if (channel->metadata_stream) {
628 ret = ustctl_stream_close_wakeup_fd(channel->metadata_stream->ustream);
629 if (ret < 0) {
630 ERR("UST consumer unable to close fd of metadata (ret: %d)", ret);
631 ret = LTTCOMM_CONSUMERD_ERROR_METADATA;
632 goto error_unlock;
633 }
634 if (channel->monitor) {
635 /* close the read-side in consumer_del_metadata_stream */
636 ret = close(channel->metadata_stream->ust_metadata_poll_pipe[1]);
637 if (ret < 0) {
638 PERROR("Close UST metadata write-side poll pipe");
639 }
640 }
641 }
642
643 error_unlock:
644 pthread_mutex_unlock(&channel->lock);
645 pthread_mutex_unlock(&consumer_data.lock);
646 error:
647 return ret;
648 }
649
650 /*
651 * RCU read side lock MUST be acquired before calling this function.
652 *
653 * Return 0 on success else an LTTng error code.
654 */
655 static int setup_metadata(struct lttng_consumer_local_data *ctx, uint64_t key)
656 {
657 int ret;
658 struct lttng_consumer_channel *metadata;
659
660 DBG("UST consumer setup metadata key %" PRIu64, key);
661
662 metadata = consumer_find_channel(key);
663 if (!metadata) {
664 ERR("UST consumer push metadata %" PRIu64 " not found", key);
665 ret = LTTNG_ERR_UST_CHAN_NOT_FOUND;
666 goto end;
667 }
668
669 /*
670 * In no monitor mode, the metadata channel has no stream(s) so skip the
671 * ownership transfer to the metadata thread.
672 */
673 if (!metadata->monitor) {
674 DBG("Metadata channel in no monitor");
675 ret = 0;
676 goto end;
677 }
678
679 /*
680 * Send metadata stream to relayd if one available. Availability is
681 * known if the stream is still in the list of the channel.
682 */
683 if (cds_list_empty(&metadata->streams.head)) {
684 ERR("Metadata channel key %" PRIu64 ", no stream available.", key);
685 ret = LTTCOMM_CONSUMERD_ERROR_METADATA;
686 goto error_no_stream;
687 }
688
689 /* Send metadata stream to relayd if needed. */
690 if (metadata->metadata_stream->net_seq_idx != (uint64_t) -1ULL) {
691 ret = consumer_send_relayd_stream(metadata->metadata_stream,
692 metadata->pathname);
693 if (ret < 0) {
694 ret = LTTCOMM_CONSUMERD_ERROR_METADATA;
695 goto error;
696 }
697 }
698
699 ret = send_streams_to_thread(metadata, ctx);
700 if (ret < 0) {
701 /*
702 * If we are unable to send the stream to the thread, there is
703 * a big problem so just stop everything.
704 */
705 ret = LTTCOMM_CONSUMERD_FATAL;
706 goto error;
707 }
708 /* List MUST be empty after or else it could be reused. */
709 assert(cds_list_empty(&metadata->streams.head));
710
711 ret = 0;
712 goto end;
713
714 error:
715 /*
716 * Delete metadata channel on error. At this point, the metadata stream can
717 * NOT be monitored by the metadata thread thus having the guarantee that
718 * the stream is still in the local stream list of the channel. This call
719 * will make sure to clean that list.
720 */
721 cds_list_del(&metadata->metadata_stream->send_node);
722 consumer_stream_destroy(metadata->metadata_stream, NULL);
723 error_no_stream:
724 end:
725 return ret;
726 }
727
728 /*
729 * Snapshot the whole metadata.
730 *
731 * Returns 0 on success, < 0 on error
732 */
733 static int snapshot_metadata(uint64_t key, char *path, uint64_t relayd_id,
734 struct lttng_consumer_local_data *ctx)
735 {
736 int ret = 0;
737 struct lttng_consumer_channel *metadata_channel;
738 struct lttng_consumer_stream *metadata_stream;
739
740 assert(path);
741 assert(ctx);
742
743 DBG("UST consumer snapshot metadata with key %" PRIu64 " at path %s",
744 key, path);
745
746 rcu_read_lock();
747
748 metadata_channel = consumer_find_channel(key);
749 if (!metadata_channel) {
750 ERR("UST snapshot metadata channel not found for key %lu", key);
751 ret = -1;
752 goto error;
753 }
754 assert(!metadata_channel->monitor);
755
756 /*
757 * Ask the sessiond if we have new metadata waiting and update the
758 * consumer metadata cache.
759 */
760 ret = lttng_ustconsumer_request_metadata(ctx, metadata_channel, 0);
761 if (ret < 0) {
762 goto error;
763 }
764
765 /*
766 * The metadata stream is NOT created in no monitor mode when the channel
767 * is created on a sessiond ask channel command.
768 */
769 ret = create_ust_streams(metadata_channel, ctx);
770 if (ret < 0) {
771 goto error;
772 }
773
774 metadata_stream = metadata_channel->metadata_stream;
775 assert(metadata_stream);
776
777 if (relayd_id != (uint64_t) -1ULL) {
778 metadata_stream->net_seq_idx = relayd_id;
779 ret = consumer_send_relayd_stream(metadata_stream, path);
780 if (ret < 0) {
781 goto error_stream;
782 }
783 } else {
784 ret = utils_create_stream_file(path, metadata_stream->name,
785 metadata_stream->chan->tracefile_size,
786 metadata_stream->tracefile_count_current,
787 metadata_stream->uid, metadata_stream->gid);
788 if (ret < 0) {
789 goto error_stream;
790 }
791 metadata_stream->out_fd = ret;
792 metadata_stream->tracefile_size_current = 0;
793 }
794
795 pthread_mutex_lock(&metadata_channel->metadata_cache->lock);
796
797 do {
798 ret = lttng_consumer_read_subbuffer(metadata_stream, ctx);
799 if (ret < 0) {
800 goto error_unlock;
801 }
802 } while (ret > 0);
803
804 error_unlock:
805 pthread_mutex_unlock(&metadata_channel->metadata_cache->lock);
806
807 error_stream:
808 /*
809 * Clean up the stream completly because the next snapshot will use a new
810 * metadata stream.
811 */
812 cds_list_del(&metadata_stream->send_node);
813 consumer_stream_destroy(metadata_stream, NULL);
814 metadata_channel->metadata_stream = NULL;
815
816 error:
817 rcu_read_unlock();
818 return ret;
819 }
820
821 /*
822 * Take a snapshot of all the stream of a channel.
823 *
824 * Returns 0 on success, < 0 on error
825 */
826 static int snapshot_channel(uint64_t key, char *path, uint64_t relayd_id,
827 uint64_t max_stream_size, struct lttng_consumer_local_data *ctx)
828 {
829 int ret;
830 unsigned use_relayd = 0;
831 unsigned long consumed_pos, produced_pos;
832 struct lttng_consumer_channel *channel;
833 struct lttng_consumer_stream *stream;
834
835 assert(path);
836 assert(ctx);
837
838 rcu_read_lock();
839
840 if (relayd_id != (uint64_t) -1ULL) {
841 use_relayd = 1;
842 }
843
844 channel = consumer_find_channel(key);
845 if (!channel) {
846 ERR("UST snapshot channel not found for key %lu", key);
847 ret = -1;
848 goto error;
849 }
850 assert(!channel->monitor);
851 DBG("UST consumer snapshot channel %lu", key);
852
853 cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
854 /* Lock stream because we are about to change its state. */
855 pthread_mutex_lock(&stream->lock);
856 stream->net_seq_idx = relayd_id;
857
858 if (use_relayd) {
859 ret = consumer_send_relayd_stream(stream, path);
860 if (ret < 0) {
861 goto error_unlock;
862 }
863 } else {
864 ret = utils_create_stream_file(path, stream->name,
865 stream->chan->tracefile_size,
866 stream->tracefile_count_current,
867 stream->uid, stream->gid);
868 if (ret < 0) {
869 goto error_unlock;
870 }
871 stream->out_fd = ret;
872 stream->tracefile_size_current = 0;
873
874 DBG("UST consumer snapshot stream %s/%s (%" PRIu64 ")", path,
875 stream->name, stream->key);
876 }
877
878 ustctl_flush_buffer(stream->ustream, 1);
879
880 ret = lttng_ustconsumer_take_snapshot(stream);
881 if (ret < 0) {
882 ERR("Taking UST snapshot");
883 goto error_unlock;
884 }
885
886 ret = lttng_ustconsumer_get_produced_snapshot(stream, &produced_pos);
887 if (ret < 0) {
888 ERR("Produced UST snapshot position");
889 goto error_unlock;
890 }
891
892 ret = lttng_ustconsumer_get_consumed_snapshot(stream, &consumed_pos);
893 if (ret < 0) {
894 ERR("Consumerd UST snapshot position");
895 goto error_unlock;
896 }
897
898 /*
899 * The original value is sent back if max stream size is larger than
900 * the possible size of the snapshot. Also, we asume that the session
901 * daemon should never send a maximum stream size that is lower than
902 * subbuffer size.
903 */
904 consumed_pos = consumer_get_consumed_maxsize(consumed_pos,
905 produced_pos, max_stream_size);
906
907 while (consumed_pos < produced_pos) {
908 ssize_t read_len;
909 unsigned long len, padded_len;
910
911 DBG("UST consumer taking snapshot at pos %lu", consumed_pos);
912
913 ret = ustctl_get_subbuf(stream->ustream, &consumed_pos);
914 if (ret < 0) {
915 if (ret != -EAGAIN) {
916 PERROR("ustctl_get_subbuf snapshot");
917 goto error_close_stream;
918 }
919 DBG("UST consumer get subbuf failed. Skipping it.");
920 consumed_pos += stream->max_sb_size;
921 continue;
922 }
923
924 ret = ustctl_get_subbuf_size(stream->ustream, &len);
925 if (ret < 0) {
926 ERR("Snapshot ustctl_get_subbuf_size");
927 goto error_put_subbuf;
928 }
929
930 ret = ustctl_get_padded_subbuf_size(stream->ustream, &padded_len);
931 if (ret < 0) {
932 ERR("Snapshot ustctl_get_padded_subbuf_size");
933 goto error_put_subbuf;
934 }
935
936 read_len = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, len,
937 padded_len - len);
938 if (use_relayd) {
939 if (read_len != len) {
940 ret = -1;
941 goto error_put_subbuf;
942 }
943 } else {
944 if (read_len != padded_len) {
945 ret = -1;
946 goto error_put_subbuf;
947 }
948 }
949
950 ret = ustctl_put_subbuf(stream->ustream);
951 if (ret < 0) {
952 ERR("Snapshot ustctl_put_subbuf");
953 goto error_close_stream;
954 }
955 consumed_pos += stream->max_sb_size;
956 }
957
958 /* Simply close the stream so we can use it on the next snapshot. */
959 consumer_stream_close(stream);
960 pthread_mutex_unlock(&stream->lock);
961 }
962
963 rcu_read_unlock();
964 return 0;
965
966 error_put_subbuf:
967 if (ustctl_put_subbuf(stream->ustream) < 0) {
968 ERR("Snapshot ustctl_put_subbuf");
969 }
970 error_close_stream:
971 consumer_stream_close(stream);
972 error_unlock:
973 pthread_mutex_unlock(&stream->lock);
974 error:
975 rcu_read_unlock();
976 return ret;
977 }
978
979 /*
980 * Receive the metadata updates from the sessiond.
981 */
982 int lttng_ustconsumer_recv_metadata(int sock, uint64_t key, uint64_t offset,
983 uint64_t len, struct lttng_consumer_channel *channel,
984 int timer)
985 {
986 int ret, ret_code = LTTNG_OK;
987 char *metadata_str;
988
989 DBG("UST consumer push metadata key %" PRIu64 " of len %" PRIu64, key, len);
990
991 metadata_str = zmalloc(len * sizeof(char));
992 if (!metadata_str) {
993 PERROR("zmalloc metadata string");
994 ret_code = LTTCOMM_CONSUMERD_ENOMEM;
995 goto end;
996 }
997
998 /* Receive metadata string. */
999 ret = lttcomm_recv_unix_sock(sock, metadata_str, len);
1000 if (ret < 0) {
1001 /* Session daemon is dead so return gracefully. */
1002 ret_code = ret;
1003 goto end_free;
1004 }
1005
1006 pthread_mutex_lock(&channel->metadata_cache->lock);
1007 ret = consumer_metadata_cache_write(channel, offset, len, metadata_str);
1008 if (ret < 0) {
1009 /* Unable to handle metadata. Notify session daemon. */
1010 ret_code = LTTCOMM_CONSUMERD_ERROR_METADATA;
1011 /*
1012 * Skip metadata flush on write error since the offset and len might
1013 * not have been updated which could create an infinite loop below when
1014 * waiting for the metadata cache to be flushed.
1015 */
1016 pthread_mutex_unlock(&channel->metadata_cache->lock);
1017 goto end_free;
1018 }
1019 pthread_mutex_unlock(&channel->metadata_cache->lock);
1020
1021 while (consumer_metadata_cache_flushed(channel, offset + len, timer)) {
1022 DBG("Waiting for metadata to be flushed");
1023 usleep(DEFAULT_METADATA_AVAILABILITY_WAIT_TIME);
1024 }
1025
1026 end_free:
1027 free(metadata_str);
1028 end:
1029 return ret_code;
1030 }
1031
1032 /*
1033 * Receive command from session daemon and process it.
1034 *
1035 * Return 1 on success else a negative value or 0.
1036 */
1037 int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
1038 int sock, struct pollfd *consumer_sockpoll)
1039 {
1040 ssize_t ret;
1041 enum lttng_error_code ret_code = LTTNG_OK;
1042 struct lttcomm_consumer_msg msg;
1043 struct lttng_consumer_channel *channel = NULL;
1044
1045 ret = lttcomm_recv_unix_sock(sock, &msg, sizeof(msg));
1046 if (ret != sizeof(msg)) {
1047 DBG("Consumer received unexpected message size %zd (expects %zu)",
1048 ret, sizeof(msg));
1049 /*
1050 * The ret value might 0 meaning an orderly shutdown but this is ok
1051 * since the caller handles this.
1052 */
1053 if (ret > 0) {
1054 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD);
1055 ret = -1;
1056 }
1057 return ret;
1058 }
1059 if (msg.cmd_type == LTTNG_CONSUMER_STOP) {
1060 /*
1061 * Notify the session daemon that the command is completed.
1062 *
1063 * On transport layer error, the function call will print an error
1064 * message so handling the returned code is a bit useless since we
1065 * return an error code anyway.
1066 */
1067 (void) consumer_send_status_msg(sock, ret_code);
1068 return -ENOENT;
1069 }
1070
1071 /* relayd needs RCU read-side lock */
1072 rcu_read_lock();
1073
1074 switch (msg.cmd_type) {
1075 case LTTNG_CONSUMER_ADD_RELAYD_SOCKET:
1076 {
1077 /* Session daemon status message are handled in the following call. */
1078 ret = consumer_add_relayd_socket(msg.u.relayd_sock.net_index,
1079 msg.u.relayd_sock.type, ctx, sock, consumer_sockpoll,
1080 &msg.u.relayd_sock.sock, msg.u.relayd_sock.session_id);
1081 goto end_nosignal;
1082 }
1083 case LTTNG_CONSUMER_DESTROY_RELAYD:
1084 {
1085 uint64_t index = msg.u.destroy_relayd.net_seq_idx;
1086 struct consumer_relayd_sock_pair *relayd;
1087
1088 DBG("UST consumer destroying relayd %" PRIu64, index);
1089
1090 /* Get relayd reference if exists. */
1091 relayd = consumer_find_relayd(index);
1092 if (relayd == NULL) {
1093 DBG("Unable to find relayd %" PRIu64, index);
1094 ret_code = LTTNG_ERR_NO_CONSUMER;
1095 }
1096
1097 /*
1098 * Each relayd socket pair has a refcount of stream attached to it
1099 * which tells if the relayd is still active or not depending on the
1100 * refcount value.
1101 *
1102 * This will set the destroy flag of the relayd object and destroy it
1103 * if the refcount reaches zero when called.
1104 *
1105 * The destroy can happen either here or when a stream fd hangs up.
1106 */
1107 if (relayd) {
1108 consumer_flag_relayd_for_destroy(relayd);
1109 }
1110
1111 goto end_msg_sessiond;
1112 }
1113 case LTTNG_CONSUMER_UPDATE_STREAM:
1114 {
1115 rcu_read_unlock();
1116 return -ENOSYS;
1117 }
1118 case LTTNG_CONSUMER_DATA_PENDING:
1119 {
1120 int ret, is_data_pending;
1121 uint64_t id = msg.u.data_pending.session_id;
1122
1123 DBG("UST consumer data pending command for id %" PRIu64, id);
1124
1125 is_data_pending = consumer_data_pending(id);
1126
1127 /* Send back returned value to session daemon */
1128 ret = lttcomm_send_unix_sock(sock, &is_data_pending,
1129 sizeof(is_data_pending));
1130 if (ret < 0) {
1131 DBG("Error when sending the data pending ret code: %d", ret);
1132 goto error_fatal;
1133 }
1134
1135 /*
1136 * No need to send back a status message since the data pending
1137 * returned value is the response.
1138 */
1139 break;
1140 }
1141 case LTTNG_CONSUMER_ASK_CHANNEL_CREATION:
1142 {
1143 int ret;
1144 struct ustctl_consumer_channel_attr attr;
1145
1146 /* Create a plain object and reserve a channel key. */
1147 channel = allocate_channel(msg.u.ask_channel.session_id,
1148 msg.u.ask_channel.pathname, msg.u.ask_channel.name,
1149 msg.u.ask_channel.uid, msg.u.ask_channel.gid,
1150 msg.u.ask_channel.relayd_id, msg.u.ask_channel.key,
1151 (enum lttng_event_output) msg.u.ask_channel.output,
1152 msg.u.ask_channel.tracefile_size,
1153 msg.u.ask_channel.tracefile_count,
1154 msg.u.ask_channel.session_id_per_pid,
1155 msg.u.ask_channel.monitor);
1156 if (!channel) {
1157 goto end_channel_error;
1158 }
1159
1160 /* Build channel attributes from received message. */
1161 attr.subbuf_size = msg.u.ask_channel.subbuf_size;
1162 attr.num_subbuf = msg.u.ask_channel.num_subbuf;
1163 attr.overwrite = msg.u.ask_channel.overwrite;
1164 attr.switch_timer_interval = msg.u.ask_channel.switch_timer_interval;
1165 attr.read_timer_interval = msg.u.ask_channel.read_timer_interval;
1166 attr.chan_id = msg.u.ask_channel.chan_id;
1167 attr.output = msg.u.ask_channel.output;
1168 memcpy(attr.uuid, msg.u.ask_channel.uuid, sizeof(attr.uuid));
1169
1170 /* Translate and save channel type. */
1171 switch (msg.u.ask_channel.type) {
1172 case LTTNG_UST_CHAN_PER_CPU:
1173 channel->type = CONSUMER_CHANNEL_TYPE_DATA;
1174 attr.type = LTTNG_UST_CHAN_PER_CPU;
1175 /*
1176 * Set refcount to 1 for owner. Below, we will
1177 * pass ownership to the
1178 * consumer_thread_channel_poll() thread.
1179 */
1180 channel->refcount = 1;
1181 break;
1182 case LTTNG_UST_CHAN_METADATA:
1183 channel->type = CONSUMER_CHANNEL_TYPE_METADATA;
1184 attr.type = LTTNG_UST_CHAN_METADATA;
1185 break;
1186 default:
1187 assert(0);
1188 goto error_fatal;
1189 };
1190
1191 ret = ask_channel(ctx, sock, channel, &attr);
1192 if (ret < 0) {
1193 goto end_channel_error;
1194 }
1195
1196 if (msg.u.ask_channel.type == LTTNG_UST_CHAN_METADATA) {
1197 ret = consumer_metadata_cache_allocate(channel);
1198 if (ret < 0) {
1199 ERR("Allocating metadata cache");
1200 goto end_channel_error;
1201 }
1202 consumer_timer_switch_start(channel, attr.switch_timer_interval);
1203 attr.switch_timer_interval = 0;
1204 }
1205
1206 /*
1207 * Add the channel to the internal state AFTER all streams were created
1208 * and successfully sent to session daemon. This way, all streams must
1209 * be ready before this channel is visible to the threads.
1210 * If add_channel succeeds, ownership of the channel is
1211 * passed to consumer_thread_channel_poll().
1212 */
1213 ret = add_channel(channel, ctx);
1214 if (ret < 0) {
1215 if (msg.u.ask_channel.type == LTTNG_UST_CHAN_METADATA) {
1216 if (channel->switch_timer_enabled == 1) {
1217 consumer_timer_switch_stop(channel);
1218 }
1219 consumer_metadata_cache_destroy(channel);
1220 }
1221 goto end_channel_error;
1222 }
1223
1224 /*
1225 * Channel and streams are now created. Inform the session daemon that
1226 * everything went well and should wait to receive the channel and
1227 * streams with ustctl API.
1228 */
1229 ret = consumer_send_status_channel(sock, channel);
1230 if (ret < 0) {
1231 /*
1232 * There is probably a problem on the socket.
1233 */
1234 goto error_fatal;
1235 }
1236
1237 break;
1238 }
1239 case LTTNG_CONSUMER_GET_CHANNEL:
1240 {
1241 int ret, relayd_err = 0;
1242 uint64_t key = msg.u.get_channel.key;
1243 struct lttng_consumer_channel *channel;
1244
1245 channel = consumer_find_channel(key);
1246 if (!channel) {
1247 ERR("UST consumer get channel key %" PRIu64 " not found", key);
1248 ret_code = LTTNG_ERR_UST_CHAN_NOT_FOUND;
1249 goto end_msg_sessiond;
1250 }
1251
1252 /* Send everything to sessiond. */
1253 ret = send_sessiond_channel(sock, channel, ctx, &relayd_err);
1254 if (ret < 0) {
1255 if (relayd_err) {
1256 /*
1257 * We were unable to send to the relayd the stream so avoid
1258 * sending back a fatal error to the thread since this is OK
1259 * and the consumer can continue its work. The above call
1260 * has sent the error status message to the sessiond.
1261 */
1262 goto end_nosignal;
1263 }
1264 /*
1265 * The communicaton was broken hence there is a bad state between
1266 * the consumer and sessiond so stop everything.
1267 */
1268 goto error_fatal;
1269 }
1270
1271 /*
1272 * In no monitor mode, the streams ownership is kept inside the channel
1273 * so don't send them to the data thread.
1274 */
1275 if (!channel->monitor) {
1276 goto end_msg_sessiond;
1277 }
1278
1279 ret = send_streams_to_thread(channel, ctx);
1280 if (ret < 0) {
1281 /*
1282 * If we are unable to send the stream to the thread, there is
1283 * a big problem so just stop everything.
1284 */
1285 goto error_fatal;
1286 }
1287 /* List MUST be empty after or else it could be reused. */
1288 assert(cds_list_empty(&channel->streams.head));
1289 goto end_msg_sessiond;
1290 }
1291 case LTTNG_CONSUMER_DESTROY_CHANNEL:
1292 {
1293 uint64_t key = msg.u.destroy_channel.key;
1294
1295 /*
1296 * Only called if streams have not been sent to stream
1297 * manager thread. However, channel has been sent to
1298 * channel manager thread.
1299 */
1300 notify_thread_del_channel(ctx, key);
1301 goto end_msg_sessiond;
1302 }
1303 case LTTNG_CONSUMER_CLOSE_METADATA:
1304 {
1305 int ret;
1306
1307 ret = close_metadata(msg.u.close_metadata.key);
1308 if (ret != 0) {
1309 ret_code = ret;
1310 }
1311
1312 goto end_msg_sessiond;
1313 }
1314 case LTTNG_CONSUMER_FLUSH_CHANNEL:
1315 {
1316 int ret;
1317
1318 ret = flush_channel(msg.u.flush_channel.key);
1319 if (ret != 0) {
1320 ret_code = ret;
1321 }
1322
1323 goto end_msg_sessiond;
1324 }
1325 case LTTNG_CONSUMER_PUSH_METADATA:
1326 {
1327 int ret;
1328 uint64_t len = msg.u.push_metadata.len;
1329 uint64_t key = msg.u.push_metadata.key;
1330 uint64_t offset = msg.u.push_metadata.target_offset;
1331 struct lttng_consumer_channel *channel;
1332
1333 DBG("UST consumer push metadata key %" PRIu64 " of len %" PRIu64, key,
1334 len);
1335
1336 channel = consumer_find_channel(key);
1337 if (!channel) {
1338 ERR("UST consumer push metadata %" PRIu64 " not found", key);
1339 ret_code = LTTNG_ERR_UST_CHAN_NOT_FOUND;
1340 goto end_msg_sessiond;
1341 }
1342
1343 /* Tell session daemon we are ready to receive the metadata. */
1344 ret = consumer_send_status_msg(sock, LTTNG_OK);
1345 if (ret < 0) {
1346 /* Somehow, the session daemon is not responding anymore. */
1347 goto error_fatal;
1348 }
1349
1350 /* Wait for more data. */
1351 if (lttng_consumer_poll_socket(consumer_sockpoll) < 0) {
1352 goto error_fatal;
1353 }
1354
1355 ret = lttng_ustconsumer_recv_metadata(sock, key, offset,
1356 len, channel, 0);
1357 if (ret < 0) {
1358 /* error receiving from sessiond */
1359 goto error_fatal;
1360 } else {
1361 ret_code = ret;
1362 goto end_msg_sessiond;
1363 }
1364 }
1365 case LTTNG_CONSUMER_SETUP_METADATA:
1366 {
1367 int ret;
1368
1369 ret = setup_metadata(ctx, msg.u.setup_metadata.key);
1370 if (ret) {
1371 ret_code = ret;
1372 }
1373 goto end_msg_sessiond;
1374 }
1375 case LTTNG_CONSUMER_SNAPSHOT_CHANNEL:
1376 {
1377 if (msg.u.snapshot_channel.metadata) {
1378 ret = snapshot_metadata(msg.u.snapshot_channel.key,
1379 msg.u.snapshot_channel.pathname,
1380 msg.u.snapshot_channel.relayd_id,
1381 ctx);
1382 if (ret < 0) {
1383 ERR("Snapshot metadata failed");
1384 ret_code = LTTNG_ERR_UST_META_FAIL;
1385 }
1386 } else {
1387 ret = snapshot_channel(msg.u.snapshot_channel.key,
1388 msg.u.snapshot_channel.pathname,
1389 msg.u.snapshot_channel.relayd_id,
1390 msg.u.snapshot_channel.max_stream_size,
1391 ctx);
1392 if (ret < 0) {
1393 ERR("Snapshot channel failed");
1394 ret_code = LTTNG_ERR_UST_CHAN_FAIL;
1395 }
1396 }
1397
1398 ret = consumer_send_status_msg(sock, ret_code);
1399 if (ret < 0) {
1400 /* Somehow, the session daemon is not responding anymore. */
1401 goto end_nosignal;
1402 }
1403 break;
1404 }
1405 default:
1406 break;
1407 }
1408
1409 end_nosignal:
1410 rcu_read_unlock();
1411
1412 /*
1413 * Return 1 to indicate success since the 0 value can be a socket
1414 * shutdown during the recv() or send() call.
1415 */
1416 return 1;
1417
1418 end_msg_sessiond:
1419 /*
1420 * The returned value here is not useful since either way we'll return 1 to
1421 * the caller because the session daemon socket management is done
1422 * elsewhere. Returning a negative code or 0 will shutdown the consumer.
1423 */
1424 ret = consumer_send_status_msg(sock, ret_code);
1425 if (ret < 0) {
1426 goto error_fatal;
1427 }
1428 rcu_read_unlock();
1429 return 1;
1430 end_channel_error:
1431 if (channel) {
1432 /*
1433 * Free channel here since no one has a reference to it. We don't
1434 * free after that because a stream can store this pointer.
1435 */
1436 destroy_channel(channel);
1437 }
1438 /* We have to send a status channel message indicating an error. */
1439 ret = consumer_send_status_channel(sock, NULL);
1440 if (ret < 0) {
1441 /* Stop everything if session daemon can not be notified. */
1442 goto error_fatal;
1443 }
1444 rcu_read_unlock();
1445 return 1;
1446 error_fatal:
1447 rcu_read_unlock();
1448 /* This will issue a consumer stop. */
1449 return -1;
1450 }
1451
1452 /*
1453 * Wrapper over the mmap() read offset from ust-ctl library. Since this can be
1454 * compiled out, we isolate it in this library.
1455 */
1456 int lttng_ustctl_get_mmap_read_offset(struct lttng_consumer_stream *stream,
1457 unsigned long *off)
1458 {
1459 assert(stream);
1460 assert(stream->ustream);
1461
1462 return ustctl_get_mmap_read_offset(stream->ustream, off);
1463 }
1464
1465 /*
1466 * Wrapper over the mmap() read offset from ust-ctl library. Since this can be
1467 * compiled out, we isolate it in this library.
1468 */
1469 void *lttng_ustctl_get_mmap_base(struct lttng_consumer_stream *stream)
1470 {
1471 assert(stream);
1472 assert(stream->ustream);
1473
1474 return ustctl_get_mmap_base(stream->ustream);
1475 }
1476
1477 /*
1478 * Take a snapshot for a specific fd
1479 *
1480 * Returns 0 on success, < 0 on error
1481 */
1482 int lttng_ustconsumer_take_snapshot(struct lttng_consumer_stream *stream)
1483 {
1484 assert(stream);
1485 assert(stream->ustream);
1486
1487 return ustctl_snapshot(stream->ustream);
1488 }
1489
1490 /*
1491 * Get the produced position
1492 *
1493 * Returns 0 on success, < 0 on error
1494 */
1495 int lttng_ustconsumer_get_produced_snapshot(
1496 struct lttng_consumer_stream *stream, unsigned long *pos)
1497 {
1498 assert(stream);
1499 assert(stream->ustream);
1500 assert(pos);
1501
1502 return ustctl_snapshot_get_produced(stream->ustream, pos);
1503 }
1504
1505 /*
1506 * Get the consumed position
1507 *
1508 * Returns 0 on success, < 0 on error
1509 */
1510 int lttng_ustconsumer_get_consumed_snapshot(
1511 struct lttng_consumer_stream *stream, unsigned long *pos)
1512 {
1513 assert(stream);
1514 assert(stream->ustream);
1515 assert(pos);
1516
1517 return ustctl_snapshot_get_consumed(stream->ustream, pos);
1518 }
1519
1520 /*
1521 * Called when the stream signal the consumer that it has hang up.
1522 */
1523 void lttng_ustconsumer_on_stream_hangup(struct lttng_consumer_stream *stream)
1524 {
1525 assert(stream);
1526 assert(stream->ustream);
1527
1528 ustctl_flush_buffer(stream->ustream, 0);
1529 stream->hangup_flush_done = 1;
1530 }
1531
1532 void lttng_ustconsumer_del_channel(struct lttng_consumer_channel *chan)
1533 {
1534 assert(chan);
1535 assert(chan->uchan);
1536
1537 if (chan->switch_timer_enabled == 1) {
1538 consumer_timer_switch_stop(chan);
1539 }
1540 consumer_metadata_cache_destroy(chan);
1541 ustctl_destroy_channel(chan->uchan);
1542 }
1543
1544 void lttng_ustconsumer_del_stream(struct lttng_consumer_stream *stream)
1545 {
1546 assert(stream);
1547 assert(stream->ustream);
1548
1549 if (stream->chan->switch_timer_enabled == 1) {
1550 consumer_timer_switch_stop(stream->chan);
1551 }
1552 ustctl_destroy_stream(stream->ustream);
1553 }
1554
1555 int lttng_ustconsumer_read_subbuffer(struct lttng_consumer_stream *stream,
1556 struct lttng_consumer_local_data *ctx)
1557 {
1558 unsigned long len, subbuf_size, padding;
1559 int err;
1560 long ret = 0;
1561 char dummy;
1562 struct ustctl_consumer_stream *ustream;
1563
1564 assert(stream);
1565 assert(stream->ustream);
1566 assert(ctx);
1567
1568 DBG("In UST read_subbuffer (wait_fd: %d, name: %s)", stream->wait_fd,
1569 stream->name);
1570
1571 /* Ease our life for what's next. */
1572 ustream = stream->ustream;
1573
1574 /* We can consume the 1 byte written into the wait_fd by UST */
1575 if (stream->monitor && !stream->hangup_flush_done) {
1576 ssize_t readlen;
1577
1578 do {
1579 readlen = read(stream->wait_fd, &dummy, 1);
1580 } while (readlen == -1 && errno == EINTR);
1581 if (readlen == -1 && errno != EAGAIN && errno != EWOULDBLOCK) {
1582 ret = readlen;
1583 goto end;
1584 }
1585 }
1586
1587 retry:
1588 /* Get the next subbuffer */
1589 err = ustctl_get_next_subbuf(ustream);
1590 if (err != 0) {
1591 /*
1592 * Populate metadata info if the existing info has
1593 * already been read.
1594 */
1595 if (stream->metadata_flag) {
1596 ssize_t write_len;
1597
1598 if (stream->chan->metadata_cache->contiguous
1599 == stream->ust_metadata_pushed) {
1600 ret = 0;
1601 goto end;
1602 }
1603
1604 write_len = ustctl_write_one_packet_to_channel(stream->chan->uchan,
1605 &stream->chan->metadata_cache->data[stream->ust_metadata_pushed],
1606 stream->chan->metadata_cache->contiguous
1607 - stream->ust_metadata_pushed);
1608 assert(write_len != 0);
1609 if (write_len < 0) {
1610 ERR("Writing one metadata packet");
1611 ret = -1;
1612 goto end;
1613 }
1614 stream->ust_metadata_pushed += write_len;
1615 ustctl_flush_buffer(stream->ustream, 1);
1616 goto retry;
1617 }
1618
1619 ret = err; /* ustctl_get_next_subbuf returns negative, caller expect positive. */
1620 /*
1621 * This is a debug message even for single-threaded consumer,
1622 * because poll() have more relaxed criterions than get subbuf,
1623 * so get_subbuf may fail for short race windows where poll()
1624 * would issue wakeups.
1625 */
1626 DBG("Reserving sub buffer failed (everything is normal, "
1627 "it is due to concurrency) [ret: %d]", err);
1628 goto end;
1629 }
1630 assert(stream->chan->output == CONSUMER_CHANNEL_MMAP);
1631 /* Get the full padded subbuffer size */
1632 err = ustctl_get_padded_subbuf_size(ustream, &len);
1633 assert(err == 0);
1634
1635 /* Get subbuffer data size (without padding) */
1636 err = ustctl_get_subbuf_size(ustream, &subbuf_size);
1637 assert(err == 0);
1638
1639 /* Make sure we don't get a subbuffer size bigger than the padded */
1640 assert(len >= subbuf_size);
1641
1642 padding = len - subbuf_size;
1643 /* write the subbuffer to the tracefile */
1644 ret = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, subbuf_size, padding);
1645 /*
1646 * The mmap operation should write subbuf_size amount of data when network
1647 * streaming or the full padding (len) size when we are _not_ streaming.
1648 */
1649 if ((ret != subbuf_size && stream->net_seq_idx != (uint64_t) -1ULL) ||
1650 (ret != len && stream->net_seq_idx == (uint64_t) -1ULL)) {
1651 /*
1652 * Display the error but continue processing to try to release the
1653 * subbuffer. This is a DBG statement since any unexpected kill or
1654 * signal, the application gets unregistered, relayd gets closed or
1655 * anything that affects the buffer lifetime will trigger this error.
1656 * So, for the sake of the user, don't print this error since it can
1657 * happen and it is OK with the code flow.
1658 */
1659 DBG("Error writing to tracefile "
1660 "(ret: %ld != len: %lu != subbuf_size: %lu)",
1661 ret, len, subbuf_size);
1662 }
1663 err = ustctl_put_next_subbuf(ustream);
1664 assert(err == 0);
1665
1666 end:
1667 return ret;
1668 }
1669
1670 /*
1671 * Called when a stream is created.
1672 *
1673 * Return 0 on success or else a negative value.
1674 */
1675 int lttng_ustconsumer_on_recv_stream(struct lttng_consumer_stream *stream)
1676 {
1677 int ret;
1678
1679 assert(stream);
1680
1681 /* Don't create anything if this is set for streaming. */
1682 if (stream->net_seq_idx == (uint64_t) -1ULL && stream->chan->monitor) {
1683 ret = utils_create_stream_file(stream->chan->pathname, stream->name,
1684 stream->chan->tracefile_size, stream->tracefile_count_current,
1685 stream->uid, stream->gid);
1686 if (ret < 0) {
1687 goto error;
1688 }
1689 stream->out_fd = ret;
1690 stream->tracefile_size_current = 0;
1691 }
1692 ret = 0;
1693
1694 error:
1695 return ret;
1696 }
1697
1698 /*
1699 * Check if data is still being extracted from the buffers for a specific
1700 * stream. Consumer data lock MUST be acquired before calling this function
1701 * and the stream lock.
1702 *
1703 * Return 1 if the traced data are still getting read else 0 meaning that the
1704 * data is available for trace viewer reading.
1705 */
1706 int lttng_ustconsumer_data_pending(struct lttng_consumer_stream *stream)
1707 {
1708 int ret;
1709
1710 assert(stream);
1711 assert(stream->ustream);
1712
1713 DBG("UST consumer checking data pending");
1714
1715 if (stream->endpoint_status != CONSUMER_ENDPOINT_ACTIVE) {
1716 ret = 0;
1717 goto end;
1718 }
1719
1720 if (stream->chan->type == CONSUMER_CHANNEL_TYPE_METADATA) {
1721 /*
1722 * We can simply check whether all contiguously available data
1723 * has been pushed to the ring buffer, since the push operation
1724 * is performed within get_next_subbuf(), and because both
1725 * get_next_subbuf() and put_next_subbuf() are issued atomically
1726 * thanks to the stream lock within
1727 * lttng_ustconsumer_read_subbuffer(). This basically means that
1728 * whetnever ust_metadata_pushed is incremented, the associated
1729 * metadata has been consumed from the metadata stream.
1730 */
1731 DBG("UST consumer metadata pending check: contiguous %" PRIu64 " vs pushed %" PRIu64,
1732 stream->chan->metadata_cache->contiguous,
1733 stream->ust_metadata_pushed);
1734 if (stream->chan->metadata_cache->contiguous
1735 != stream->ust_metadata_pushed) {
1736 ret = 1; /* Data is pending */
1737 goto end;
1738 }
1739 } else {
1740 ret = ustctl_get_next_subbuf(stream->ustream);
1741 if (ret == 0) {
1742 /*
1743 * There is still data so let's put back this
1744 * subbuffer.
1745 */
1746 ret = ustctl_put_subbuf(stream->ustream);
1747 assert(ret == 0);
1748 ret = 1; /* Data is pending */
1749 goto end;
1750 }
1751 }
1752
1753 /* Data is NOT pending so ready to be read. */
1754 ret = 0;
1755
1756 end:
1757 return ret;
1758 }
1759
1760 /*
1761 * Close every metadata stream wait fd of the metadata hash table. This
1762 * function MUST be used very carefully so not to run into a race between the
1763 * metadata thread handling streams and this function closing their wait fd.
1764 *
1765 * For UST, this is used when the session daemon hangs up. Its the metadata
1766 * producer so calling this is safe because we are assured that no state change
1767 * can occur in the metadata thread for the streams in the hash table.
1768 */
1769 void lttng_ustconsumer_close_metadata(struct lttng_ht *metadata_ht)
1770 {
1771 int ret;
1772 struct lttng_ht_iter iter;
1773 struct lttng_consumer_stream *stream;
1774
1775 assert(metadata_ht);
1776 assert(metadata_ht->ht);
1777
1778 DBG("UST consumer closing all metadata streams");
1779
1780 rcu_read_lock();
1781 cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream,
1782 node.node) {
1783 int fd = stream->wait_fd;
1784
1785 /*
1786 * Whatever happens here we have to continue to try to close every
1787 * streams. Let's report at least the error on failure.
1788 */
1789 ret = ustctl_stream_close_wakeup_fd(stream->ustream);
1790 if (ret) {
1791 ERR("Unable to close metadata stream fd %d ret %d", fd, ret);
1792 }
1793 DBG("Metadata wait fd %d closed", fd);
1794 }
1795 rcu_read_unlock();
1796 }
1797
1798 void lttng_ustconsumer_close_stream_wakeup(struct lttng_consumer_stream *stream)
1799 {
1800 int ret;
1801
1802 ret = ustctl_stream_close_wakeup_fd(stream->ustream);
1803 if (ret < 0) {
1804 ERR("Unable to close wakeup fd");
1805 }
1806 }
1807
1808 /*
1809 * Please refer to consumer-timer.c before adding any lock within this
1810 * function or any of its callees. Timers have a very strict locking
1811 * semantic with respect to teardown. Failure to respect this semantic
1812 * introduces deadlocks.
1813 */
1814 int lttng_ustconsumer_request_metadata(struct lttng_consumer_local_data *ctx,
1815 struct lttng_consumer_channel *channel, int timer)
1816 {
1817 struct lttcomm_metadata_request_msg request;
1818 struct lttcomm_consumer_msg msg;
1819 enum lttng_error_code ret_code = LTTNG_OK;
1820 uint64_t len, key, offset;
1821 int ret;
1822
1823 assert(channel);
1824 assert(channel->metadata_cache);
1825
1826 /* send the metadata request to sessiond */
1827 switch (consumer_data.type) {
1828 case LTTNG_CONSUMER64_UST:
1829 request.bits_per_long = 64;
1830 break;
1831 case LTTNG_CONSUMER32_UST:
1832 request.bits_per_long = 32;
1833 break;
1834 default:
1835 request.bits_per_long = 0;
1836 break;
1837 }
1838
1839 request.session_id = channel->session_id;
1840 request.session_id_per_pid = channel->session_id_per_pid;
1841 request.uid = channel->uid;
1842 request.key = channel->key;
1843 DBG("Sending metadata request to sessiond, session id %" PRIu64
1844 ", per-pid %" PRIu64,
1845 channel->session_id,
1846 channel->session_id_per_pid);
1847
1848 pthread_mutex_lock(&ctx->metadata_socket_lock);
1849 ret = lttcomm_send_unix_sock(ctx->consumer_metadata_socket, &request,
1850 sizeof(request));
1851 if (ret < 0) {
1852 ERR("Asking metadata to sessiond");
1853 goto end;
1854 }
1855
1856 /* Receive the metadata from sessiond */
1857 ret = lttcomm_recv_unix_sock(ctx->consumer_metadata_socket, &msg,
1858 sizeof(msg));
1859 if (ret != sizeof(msg)) {
1860 DBG("Consumer received unexpected message size %d (expects %zu)",
1861 ret, sizeof(msg));
1862 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD);
1863 /*
1864 * The ret value might 0 meaning an orderly shutdown but this is ok
1865 * since the caller handles this.
1866 */
1867 goto end;
1868 }
1869
1870 if (msg.cmd_type == LTTNG_ERR_UND) {
1871 /* No registry found */
1872 (void) consumer_send_status_msg(ctx->consumer_metadata_socket,
1873 ret_code);
1874 ret = 0;
1875 goto end;
1876 } else if (msg.cmd_type != LTTNG_CONSUMER_PUSH_METADATA) {
1877 ERR("Unexpected cmd_type received %d", msg.cmd_type);
1878 ret = -1;
1879 goto end;
1880 }
1881
1882 len = msg.u.push_metadata.len;
1883 key = msg.u.push_metadata.key;
1884 offset = msg.u.push_metadata.target_offset;
1885
1886 assert(key == channel->key);
1887 if (len == 0) {
1888 DBG("No new metadata to receive for key %" PRIu64, key);
1889 }
1890
1891 /* Tell session daemon we are ready to receive the metadata. */
1892 ret = consumer_send_status_msg(ctx->consumer_metadata_socket,
1893 LTTNG_OK);
1894 if (ret < 0 || len == 0) {
1895 /*
1896 * Somehow, the session daemon is not responding anymore or there is
1897 * nothing to receive.
1898 */
1899 goto end;
1900 }
1901
1902 ret_code = lttng_ustconsumer_recv_metadata(ctx->consumer_metadata_socket,
1903 key, offset, len, channel, timer);
1904 if (ret_code >= 0) {
1905 /*
1906 * Only send the status msg if the sessiond is alive meaning a positive
1907 * ret code.
1908 */
1909 (void) consumer_send_status_msg(ctx->consumer_metadata_socket, ret_code);
1910 }
1911 ret = 0;
1912
1913 end:
1914 pthread_mutex_unlock(&ctx->metadata_socket_lock);
1915 return ret;
1916 }
This page took 0.102822 seconds and 6 git commands to generate.