Fix: don't send error to sessiond on orderly shutdown
[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 * Search for a relayd object related to the stream. If found, send the stream
214 * to the relayd.
215 *
216 * On success, returns 0 else a negative value.
217 */
218 static int send_stream_to_relayd(struct lttng_consumer_stream *stream)
219 {
220 int ret = 0;
221 struct consumer_relayd_sock_pair *relayd;
222
223 assert(stream);
224
225 relayd = consumer_find_relayd(stream->net_seq_idx);
226 if (relayd != NULL) {
227 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
228 /* Add stream on the relayd */
229 ret = relayd_add_stream(&relayd->control_sock, stream->name,
230 stream->chan->pathname, &stream->relayd_stream_id,
231 stream->chan->tracefile_size,
232 stream->chan->tracefile_count);
233 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
234 if (ret < 0) {
235 goto error;
236 }
237 } else if (stream->net_seq_idx != (uint64_t) -1ULL) {
238 ERR("Network sequence index %" PRIu64 " unknown. Not adding stream.",
239 stream->net_seq_idx);
240 ret = -1;
241 goto error;
242 }
243
244 error:
245 return ret;
246 }
247
248 /*
249 * Create streams for the given channel using liblttng-ust-ctl.
250 *
251 * Return 0 on success else a negative value.
252 */
253 static int create_ust_streams(struct lttng_consumer_channel *channel,
254 struct lttng_consumer_local_data *ctx)
255 {
256 int ret, cpu = 0;
257 struct ustctl_consumer_stream *ustream;
258 struct lttng_consumer_stream *stream;
259
260 assert(channel);
261 assert(ctx);
262
263 /*
264 * While a stream is available from ustctl. When NULL is returned, we've
265 * reached the end of the possible stream for the channel.
266 */
267 while ((ustream = ustctl_create_stream(channel->uchan, cpu))) {
268 int wait_fd;
269
270 wait_fd = ustctl_stream_get_wait_fd(ustream);
271
272 /* Allocate consumer stream object. */
273 stream = allocate_stream(cpu, wait_fd, channel, ctx, &ret);
274 if (!stream) {
275 goto error_alloc;
276 }
277 stream->ustream = ustream;
278 /*
279 * Store it so we can save multiple function calls afterwards since
280 * this value is used heavily in the stream threads. This is UST
281 * specific so this is why it's done after allocation.
282 */
283 stream->wait_fd = wait_fd;
284
285 /*
286 * Increment channel refcount since the channel reference has now been
287 * assigned in the allocation process above.
288 */
289 if (stream->chan->monitor) {
290 uatomic_inc(&stream->chan->refcount);
291 }
292
293 /*
294 * Order is important this is why a list is used. On error, the caller
295 * should clean this list.
296 */
297 cds_list_add_tail(&stream->send_node, &channel->streams.head);
298
299 ret = ustctl_get_max_subbuf_size(stream->ustream,
300 &stream->max_sb_size);
301 if (ret < 0) {
302 ERR("ustctl_get_max_subbuf_size failed for stream %s",
303 stream->name);
304 goto error;
305 }
306
307 /* Do actions once stream has been received. */
308 if (ctx->on_recv_stream) {
309 ret = ctx->on_recv_stream(stream);
310 if (ret < 0) {
311 goto error;
312 }
313 }
314
315 DBG("UST consumer add stream %s (key: %" PRIu64 ") with relayd id %" PRIu64,
316 stream->name, stream->key, stream->relayd_stream_id);
317
318 /* Set next CPU stream. */
319 channel->streams.count = ++cpu;
320
321 /* Keep stream reference when creating metadata. */
322 if (channel->type == CONSUMER_CHANNEL_TYPE_METADATA) {
323 channel->metadata_stream = stream;
324 }
325 }
326
327 return 0;
328
329 error:
330 error_alloc:
331 return ret;
332 }
333
334 /*
335 * Create an UST channel with the given attributes and send it to the session
336 * daemon using the ust ctl API.
337 *
338 * Return 0 on success or else a negative value.
339 */
340 static int create_ust_channel(struct ustctl_consumer_channel_attr *attr,
341 struct ustctl_consumer_channel **chanp)
342 {
343 int ret;
344 struct ustctl_consumer_channel *channel;
345
346 assert(attr);
347 assert(chanp);
348
349 DBG3("Creating channel to ustctl with attr: [overwrite: %d, "
350 "subbuf_size: %" PRIu64 ", num_subbuf: %" PRIu64 ", "
351 "switch_timer_interval: %u, read_timer_interval: %u, "
352 "output: %d, type: %d", attr->overwrite, attr->subbuf_size,
353 attr->num_subbuf, attr->switch_timer_interval,
354 attr->read_timer_interval, attr->output, attr->type);
355
356 channel = ustctl_create_channel(attr);
357 if (!channel) {
358 ret = -1;
359 goto error_create;
360 }
361
362 *chanp = channel;
363
364 return 0;
365
366 error_create:
367 return ret;
368 }
369
370 /*
371 * Send a single given stream to the session daemon using the sock.
372 *
373 * Return 0 on success else a negative value.
374 */
375 static int send_sessiond_stream(int sock, struct lttng_consumer_stream *stream)
376 {
377 int ret;
378
379 assert(stream);
380 assert(sock >= 0);
381
382 DBG2("UST consumer sending stream %" PRIu64 " to sessiond", stream->key);
383
384 /* Send stream to session daemon. */
385 ret = ustctl_send_stream_to_sessiond(sock, stream->ustream);
386 if (ret < 0) {
387 goto error;
388 }
389
390 error:
391 return ret;
392 }
393
394 /*
395 * Send channel to sessiond.
396 *
397 * Return 0 on success or else a negative value.
398 */
399 static int send_sessiond_channel(int sock,
400 struct lttng_consumer_channel *channel,
401 struct lttng_consumer_local_data *ctx, int *relayd_error)
402 {
403 int ret, ret_code = LTTNG_OK;
404 struct lttng_consumer_stream *stream;
405
406 assert(channel);
407 assert(ctx);
408 assert(sock >= 0);
409
410 DBG("UST consumer sending channel %s to sessiond", channel->name);
411
412 cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
413 /* Try to send the stream to the relayd if one is available. */
414 ret = send_stream_to_relayd(stream);
415 if (ret < 0) {
416 /*
417 * Flag that the relayd was the problem here probably due to a
418 * communicaton error on the socket.
419 */
420 if (relayd_error) {
421 *relayd_error = 1;
422 }
423 ret_code = LTTNG_ERR_RELAYD_CONNECT_FAIL;
424 }
425 }
426
427 /* Inform sessiond that we are about to send channel and streams. */
428 ret = consumer_send_status_msg(sock, ret_code);
429 if (ret < 0 || ret_code != LTTNG_OK) {
430 /*
431 * Either the session daemon is not responding or the relayd died so we
432 * stop now.
433 */
434 goto error;
435 }
436
437 /* Send channel to sessiond. */
438 ret = ustctl_send_channel_to_sessiond(sock, channel->uchan);
439 if (ret < 0) {
440 goto error;
441 }
442
443 ret = ustctl_channel_close_wakeup_fd(channel->uchan);
444 if (ret < 0) {
445 goto error;
446 }
447
448 /* The channel was sent successfully to the sessiond at this point. */
449 cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
450 /* Send stream to session daemon. */
451 ret = send_sessiond_stream(sock, stream);
452 if (ret < 0) {
453 goto error;
454 }
455 }
456
457 /* Tell sessiond there is no more stream. */
458 ret = ustctl_send_stream_to_sessiond(sock, NULL);
459 if (ret < 0) {
460 goto error;
461 }
462
463 DBG("UST consumer NULL stream sent to sessiond");
464
465 return 0;
466
467 error:
468 if (ret_code != LTTNG_OK) {
469 ret = -1;
470 }
471 return ret;
472 }
473
474 /*
475 * Creates a channel and streams and add the channel it to the channel internal
476 * state. The created stream must ONLY be sent once the GET_CHANNEL command is
477 * received.
478 *
479 * Return 0 on success or else, a negative value is returned and the channel
480 * MUST be destroyed by consumer_del_channel().
481 */
482 static int ask_channel(struct lttng_consumer_local_data *ctx, int sock,
483 struct lttng_consumer_channel *channel,
484 struct ustctl_consumer_channel_attr *attr)
485 {
486 int ret;
487
488 assert(ctx);
489 assert(channel);
490 assert(attr);
491
492 /*
493 * This value is still used by the kernel consumer since for the kernel,
494 * the stream ownership is not IN the consumer so we need to have the
495 * number of left stream that needs to be initialized so we can know when
496 * to delete the channel (see consumer.c).
497 *
498 * As for the user space tracer now, the consumer creates and sends the
499 * stream to the session daemon which only sends them to the application
500 * once every stream of a channel is received making this value useless
501 * because we they will be added to the poll thread before the application
502 * receives them. This ensures that a stream can not hang up during
503 * initilization of a channel.
504 */
505 channel->nb_init_stream_left = 0;
506
507 /* The reply msg status is handled in the following call. */
508 ret = create_ust_channel(attr, &channel->uchan);
509 if (ret < 0) {
510 goto end;
511 }
512
513 channel->wait_fd = ustctl_channel_get_wait_fd(channel->uchan);
514
515 /*
516 * For the snapshots (no monitor), we create the metadata streams
517 * on demand, not during the channel creation.
518 */
519 if (channel->type == CONSUMER_CHANNEL_TYPE_METADATA && !channel->monitor) {
520 ret = 0;
521 goto end;
522 }
523
524 /* Open all streams for this channel. */
525 ret = create_ust_streams(channel, ctx);
526 if (ret < 0) {
527 goto end;
528 }
529
530 end:
531 return ret;
532 }
533
534 /*
535 * Send all stream of a channel to the right thread handling it.
536 *
537 * On error, return a negative value else 0 on success.
538 */
539 static int send_streams_to_thread(struct lttng_consumer_channel *channel,
540 struct lttng_consumer_local_data *ctx)
541 {
542 int ret = 0;
543 struct lttng_consumer_stream *stream, *stmp;
544
545 assert(channel);
546 assert(ctx);
547
548 /* Send streams to the corresponding thread. */
549 cds_list_for_each_entry_safe(stream, stmp, &channel->streams.head,
550 send_node) {
551 /* Sending the stream to the thread. */
552 ret = send_stream_to_thread(stream, ctx);
553 if (ret < 0) {
554 /*
555 * If we are unable to send the stream to the thread, there is
556 * a big problem so just stop everything.
557 */
558 goto error;
559 }
560
561 /* Remove node from the channel stream list. */
562 cds_list_del(&stream->send_node);
563
564 /*
565 * From this point on, the stream's ownership has been moved away from
566 * the channel and becomes globally visible.
567 */
568 stream->globally_visible = 1;
569 }
570
571 error:
572 return ret;
573 }
574
575 /*
576 * Write metadata to the given channel using ustctl to convert the string to
577 * the ringbuffer.
578 * Called only from consumer_metadata_cache_write.
579 * The metadata cache lock MUST be acquired to write in the cache.
580 *
581 * Return 0 on success else a negative value.
582 */
583 int lttng_ustconsumer_push_metadata(struct lttng_consumer_channel *metadata,
584 const char *metadata_str, uint64_t target_offset, uint64_t len)
585 {
586 int ret;
587
588 assert(metadata);
589 assert(metadata_str);
590
591 DBG("UST consumer writing metadata to channel %s", metadata->name);
592
593 if (!metadata->metadata_stream) {
594 ret = 0;
595 goto error;
596 }
597
598 assert(target_offset <= metadata->metadata_cache->max_offset);
599 ret = ustctl_write_metadata_to_channel(metadata->uchan,
600 metadata_str + target_offset, len);
601 if (ret < 0) {
602 ERR("ustctl write metadata fail with ret %d, len %" PRIu64, ret, len);
603 goto error;
604 }
605
606 ustctl_flush_buffer(metadata->metadata_stream->ustream, 1);
607
608 error:
609 return ret;
610 }
611
612 /*
613 * Flush channel's streams using the given key to retrieve the channel.
614 *
615 * Return 0 on success else an LTTng error code.
616 */
617 static int flush_channel(uint64_t chan_key)
618 {
619 int ret = 0;
620 struct lttng_consumer_channel *channel;
621 struct lttng_consumer_stream *stream;
622 struct lttng_ht *ht;
623 struct lttng_ht_iter iter;
624
625 DBG("UST consumer flush channel key %" PRIu64, chan_key);
626
627 rcu_read_lock();
628 channel = consumer_find_channel(chan_key);
629 if (!channel) {
630 ERR("UST consumer flush channel %" PRIu64 " not found", chan_key);
631 ret = LTTNG_ERR_UST_CHAN_NOT_FOUND;
632 goto error;
633 }
634
635 ht = consumer_data.stream_per_chan_id_ht;
636
637 /* For each stream of the channel id, flush it. */
638 cds_lfht_for_each_entry_duplicate(ht->ht,
639 ht->hash_fct(&channel->key, lttng_ht_seed), ht->match_fct,
640 &channel->key, &iter.iter, stream, node_channel_id.node) {
641 ustctl_flush_buffer(stream->ustream, 1);
642 }
643 error:
644 rcu_read_unlock();
645 return ret;
646 }
647
648 /*
649 * Close metadata stream wakeup_fd using the given key to retrieve the channel.
650 * RCU read side lock MUST be acquired before calling this function.
651 *
652 * Return 0 on success else an LTTng error code.
653 */
654 static int close_metadata(uint64_t chan_key)
655 {
656 int ret = 0;
657 struct lttng_consumer_channel *channel;
658
659 DBG("UST consumer close metadata key %" PRIu64, chan_key);
660
661 channel = consumer_find_channel(chan_key);
662 if (!channel) {
663 /*
664 * This is possible if the metadata thread has issue a delete because
665 * the endpoint point of the stream hung up. There is no way the
666 * session daemon can know about it thus use a DBG instead of an actual
667 * error.
668 */
669 DBG("UST consumer close metadata %" PRIu64 " not found", chan_key);
670 ret = LTTNG_ERR_UST_CHAN_NOT_FOUND;
671 goto error;
672 }
673
674 pthread_mutex_lock(&consumer_data.lock);
675
676 if (cds_lfht_is_node_deleted(&channel->node.node)) {
677 goto error_unlock;
678 }
679
680 if (channel->switch_timer_enabled == 1) {
681 DBG("Deleting timer on metadata channel");
682 consumer_timer_switch_stop(channel);
683 }
684
685 if (channel->metadata_stream) {
686 ret = ustctl_stream_close_wakeup_fd(channel->metadata_stream->ustream);
687 if (ret < 0) {
688 ERR("UST consumer unable to close fd of metadata (ret: %d)", ret);
689 ret = LTTCOMM_CONSUMERD_ERROR_METADATA;
690 goto error_unlock;
691 }
692 }
693
694 error_unlock:
695 pthread_mutex_unlock(&consumer_data.lock);
696 error:
697 return ret;
698 }
699
700 /*
701 * RCU read side lock MUST be acquired before calling this function.
702 *
703 * Return 0 on success else an LTTng error code.
704 */
705 static int setup_metadata(struct lttng_consumer_local_data *ctx, uint64_t key)
706 {
707 int ret;
708 struct lttng_consumer_channel *metadata;
709
710 DBG("UST consumer setup metadata key %" PRIu64, key);
711
712 metadata = consumer_find_channel(key);
713 if (!metadata) {
714 ERR("UST consumer push metadata %" PRIu64 " not found", key);
715 ret = LTTNG_ERR_UST_CHAN_NOT_FOUND;
716 goto end;
717 }
718
719 /*
720 * In no monitor mode, the metadata channel has no stream(s) so skip the
721 * ownership transfer to the metadata thread.
722 */
723 if (!metadata->monitor) {
724 DBG("Metadata channel in no monitor");
725 ret = 0;
726 goto end;
727 }
728
729 /*
730 * Send metadata stream to relayd if one available. Availability is
731 * known if the stream is still in the list of the channel.
732 */
733 if (cds_list_empty(&metadata->streams.head)) {
734 ERR("Metadata channel key %" PRIu64 ", no stream available.", key);
735 ret = LTTCOMM_CONSUMERD_ERROR_METADATA;
736 goto error_no_stream;
737 }
738
739 /* Send metadata stream to relayd if needed. */
740 ret = send_stream_to_relayd(metadata->metadata_stream);
741 if (ret < 0) {
742 ret = LTTCOMM_CONSUMERD_ERROR_METADATA;
743 goto error;
744 }
745
746 ret = send_streams_to_thread(metadata, ctx);
747 if (ret < 0) {
748 /*
749 * If we are unable to send the stream to the thread, there is
750 * a big problem so just stop everything.
751 */
752 ret = LTTCOMM_CONSUMERD_FATAL;
753 goto error;
754 }
755 /* List MUST be empty after or else it could be reused. */
756 assert(cds_list_empty(&metadata->streams.head));
757
758 ret = 0;
759 goto end;
760
761 error:
762 /*
763 * Delete metadata channel on error. At this point, the metadata stream can
764 * NOT be monitored by the metadata thread thus having the guarantee that
765 * the stream is still in the local stream list of the channel. This call
766 * will make sure to clean that list.
767 */
768 cds_list_del(&metadata->metadata_stream->send_node);
769 consumer_stream_destroy(metadata->metadata_stream, NULL);
770 error_no_stream:
771 end:
772 return ret;
773 }
774
775 /*
776 * Snapshot the whole metadata.
777 *
778 * Returns 0 on success, < 0 on error
779 */
780 static int snapshot_metadata(uint64_t key, char *path, uint64_t relayd_id,
781 struct lttng_consumer_local_data *ctx)
782 {
783 int ret = 0;
784 ssize_t write_len;
785 uint64_t total_len = 0;
786 struct lttng_consumer_channel *metadata_channel;
787 struct lttng_consumer_stream *metadata_stream;
788
789 assert(path);
790 assert(ctx);
791
792 DBG("UST consumer snapshot metadata with key %" PRIu64 " at path %s",
793 key, path);
794
795 rcu_read_lock();
796
797 metadata_channel = consumer_find_channel(key);
798 if (!metadata_channel) {
799 ERR("UST snapshot metadata channel not found for key %lu", key);
800 ret = -1;
801 goto error;
802 }
803 assert(!metadata_channel->monitor);
804
805 /*
806 * Ask the sessiond if we have new metadata waiting and update the
807 * consumer metadata cache.
808 */
809 ret = lttng_ustconsumer_request_metadata(ctx, metadata_channel);
810 if (ret < 0) {
811 goto error;
812 }
813
814 /*
815 * The metadata stream is NOT created in no monitor mode when the channel
816 * is created on a sessiond ask channel command.
817 */
818 ret = create_ust_streams(metadata_channel, ctx);
819 if (ret < 0) {
820 goto error;
821 }
822
823 metadata_stream = metadata_channel->metadata_stream;
824 assert(metadata_stream);
825
826 if (relayd_id != (uint64_t) -1ULL) {
827 metadata_stream->net_seq_idx = relayd_id;
828 ret = consumer_send_relayd_stream(metadata_stream, path);
829 if (ret < 0) {
830 goto error_stream;
831 }
832 } else {
833 ret = utils_create_stream_file(path, metadata_stream->name,
834 metadata_stream->chan->tracefile_size,
835 metadata_stream->tracefile_count_current,
836 metadata_stream->uid, metadata_stream->gid);
837 if (ret < 0) {
838 goto error_stream;
839 }
840 metadata_stream->out_fd = ret;
841 metadata_stream->tracefile_size_current = 0;
842 }
843
844 pthread_mutex_lock(&metadata_channel->metadata_cache->lock);
845 while (total_len < metadata_channel->metadata_cache->total_bytes_written) {
846 /*
847 * Write at most one packet of metadata into the channel
848 * to avoid blocking here.
849 */
850 write_len = ustctl_write_one_packet_to_channel(metadata_channel->uchan,
851 metadata_channel->metadata_cache->data,
852 metadata_channel->metadata_cache->total_bytes_written);
853 if (write_len < 0) {
854 ERR("UST consumer snapshot writing metadata packet");
855 ret = -1;
856 goto error_unlock;
857 }
858 total_len += write_len;
859
860 DBG("Written %" PRIu64 " bytes to metadata (left: %" PRIu64 ")",
861 write_len,
862 metadata_channel->metadata_cache->total_bytes_written - write_len);
863 ustctl_flush_buffer(metadata_stream->ustream, 1);
864 ret = lttng_consumer_read_subbuffer(metadata_stream, ctx);
865 if (ret < 0) {
866 goto error_unlock;
867 }
868 }
869
870 error_unlock:
871 pthread_mutex_unlock(&metadata_channel->metadata_cache->lock);
872
873 error_stream:
874 /*
875 * Clean up the stream completly because the next snapshot will use a new
876 * metadata stream.
877 */
878 cds_list_del(&metadata_stream->send_node);
879 consumer_stream_destroy(metadata_stream, NULL);
880 metadata_channel->metadata_stream = NULL;
881
882 error:
883 rcu_read_unlock();
884 return ret;
885 }
886
887 /*
888 * Take a snapshot of all the stream of a channel.
889 *
890 * Returns 0 on success, < 0 on error
891 */
892 static int snapshot_channel(uint64_t key, char *path, uint64_t relayd_id,
893 struct lttng_consumer_local_data *ctx)
894 {
895 int ret;
896 unsigned use_relayd = 0;
897 unsigned long consumed_pos, produced_pos;
898 struct lttng_consumer_channel *channel;
899 struct lttng_consumer_stream *stream;
900
901 assert(path);
902 assert(ctx);
903
904 rcu_read_lock();
905
906 if (relayd_id != (uint64_t) -1ULL) {
907 use_relayd = 1;
908 }
909
910 channel = consumer_find_channel(key);
911 if (!channel) {
912 ERR("UST snapshot channel not found for key %lu", key);
913 ret = -1;
914 goto error;
915 }
916 assert(!channel->monitor);
917 DBG("UST consumer snapshot channel %lu", key);
918
919 cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
920 /* Lock stream because we are about to change its state. */
921 pthread_mutex_lock(&stream->lock);
922 stream->net_seq_idx = relayd_id;
923
924 if (use_relayd) {
925 ret = consumer_send_relayd_stream(stream, path);
926 if (ret < 0) {
927 goto error_unlock;
928 }
929 } else {
930 ret = utils_create_stream_file(path, stream->name,
931 stream->chan->tracefile_size,
932 stream->tracefile_count_current,
933 stream->uid, stream->gid);
934 if (ret < 0) {
935 goto error_unlock;
936 }
937 stream->out_fd = ret;
938 stream->tracefile_size_current = 0;
939
940 DBG("UST consumer snapshot stream %s/%s (%" PRIu64 ")", path,
941 stream->name, stream->key);
942 }
943
944 ustctl_flush_buffer(stream->ustream, 1);
945
946 ret = lttng_ustconsumer_take_snapshot(stream);
947 if (ret < 0) {
948 ERR("Taking UST snapshot");
949 goto error_unlock;
950 }
951
952 ret = lttng_ustconsumer_get_produced_snapshot(stream, &produced_pos);
953 if (ret < 0) {
954 ERR("Produced UST snapshot position");
955 goto error_unlock;
956 }
957
958 ret = lttng_ustconsumer_get_consumed_snapshot(stream, &consumed_pos);
959 if (ret < 0) {
960 ERR("Consumerd UST snapshot position");
961 goto error_unlock;
962 }
963
964 while (consumed_pos < produced_pos) {
965 ssize_t read_len;
966 unsigned long len, padded_len;
967
968 DBG("UST consumer taking snapshot at pos %lu", consumed_pos);
969
970 ret = ustctl_get_subbuf(stream->ustream, &consumed_pos);
971 if (ret < 0) {
972 if (ret != -EAGAIN) {
973 PERROR("ustctl_get_subbuf snapshot");
974 goto error_close_stream;
975 }
976 DBG("UST consumer get subbuf failed. Skipping it.");
977 consumed_pos += stream->max_sb_size;
978 continue;
979 }
980
981 ret = ustctl_get_subbuf_size(stream->ustream, &len);
982 if (ret < 0) {
983 ERR("Snapshot ustctl_get_subbuf_size");
984 goto error_put_subbuf;
985 }
986
987 ret = ustctl_get_padded_subbuf_size(stream->ustream, &padded_len);
988 if (ret < 0) {
989 ERR("Snapshot ustctl_get_padded_subbuf_size");
990 goto error_put_subbuf;
991 }
992
993 read_len = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, len,
994 padded_len - len);
995 if (use_relayd) {
996 if (read_len != len) {
997 ret = -1;
998 goto error_put_subbuf;
999 }
1000 } else {
1001 if (read_len != padded_len) {
1002 ret = -1;
1003 goto error_put_subbuf;
1004 }
1005 }
1006
1007 ret = ustctl_put_subbuf(stream->ustream);
1008 if (ret < 0) {
1009 ERR("Snapshot ustctl_put_subbuf");
1010 goto error_close_stream;
1011 }
1012 consumed_pos += stream->max_sb_size;
1013 }
1014
1015 /* Simply close the stream so we can use it on the next snapshot. */
1016 consumer_stream_close(stream);
1017 pthread_mutex_unlock(&stream->lock);
1018 }
1019
1020 rcu_read_unlock();
1021 return 0;
1022
1023 error_put_subbuf:
1024 if (ustctl_put_subbuf(stream->ustream) < 0) {
1025 ERR("Snapshot ustctl_put_subbuf");
1026 }
1027 error_close_stream:
1028 consumer_stream_close(stream);
1029 error_unlock:
1030 pthread_mutex_unlock(&stream->lock);
1031 error:
1032 rcu_read_unlock();
1033 return ret;
1034 }
1035
1036 /*
1037 * Receive the metadata updates from the sessiond.
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 {
1042 int ret, ret_code = LTTNG_OK;
1043 char *metadata_str;
1044
1045 DBG("UST consumer push metadata key %" PRIu64 " of len %" PRIu64, key, len);
1046
1047 metadata_str = zmalloc(len * sizeof(char));
1048 if (!metadata_str) {
1049 PERROR("zmalloc metadata string");
1050 ret_code = LTTCOMM_CONSUMERD_ENOMEM;
1051 goto end;
1052 }
1053
1054 /* Receive metadata string. */
1055 ret = lttcomm_recv_unix_sock(sock, metadata_str, len);
1056 if (ret < 0) {
1057 /* Session daemon is dead so return gracefully. */
1058 ret_code = ret;
1059 goto end_free;
1060 }
1061
1062 /*
1063 * XXX: The consumer data lock is acquired before calling metadata cache
1064 * write which calls push metadata that MUST be protected by the consumer
1065 * lock in order to be able to check the validity of the metadata stream of
1066 * the channel.
1067 *
1068 * Note that this will be subject to change to better fine grained locking
1069 * and ultimately try to get rid of this global consumer data lock.
1070 */
1071 pthread_mutex_lock(&consumer_data.lock);
1072
1073 pthread_mutex_lock(&channel->metadata_cache->lock);
1074 ret = consumer_metadata_cache_write(channel, offset, len, metadata_str);
1075 if (ret < 0) {
1076 /* Unable to handle metadata. Notify session daemon. */
1077 ret_code = LTTCOMM_CONSUMERD_ERROR_METADATA;
1078 /*
1079 * Skip metadata flush on write error since the offset and len might
1080 * not have been updated which could create an infinite loop below when
1081 * waiting for the metadata cache to be flushed.
1082 */
1083 pthread_mutex_unlock(&channel->metadata_cache->lock);
1084 pthread_mutex_unlock(&consumer_data.lock);
1085 goto end_free;
1086 }
1087 pthread_mutex_unlock(&channel->metadata_cache->lock);
1088 pthread_mutex_unlock(&consumer_data.lock);
1089
1090 while (consumer_metadata_cache_flushed(channel, offset + len)) {
1091 DBG("Waiting for metadata to be flushed");
1092 usleep(DEFAULT_METADATA_AVAILABILITY_WAIT_TIME);
1093 }
1094
1095 end_free:
1096 free(metadata_str);
1097 end:
1098 return ret_code;
1099 }
1100
1101 /*
1102 * Receive command from session daemon and process it.
1103 *
1104 * Return 1 on success else a negative value or 0.
1105 */
1106 int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
1107 int sock, struct pollfd *consumer_sockpoll)
1108 {
1109 ssize_t ret;
1110 enum lttng_error_code ret_code = LTTNG_OK;
1111 struct lttcomm_consumer_msg msg;
1112 struct lttng_consumer_channel *channel = NULL;
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 if (msg.cmd_type == LTTNG_CONSUMER_STOP) {
1129 /*
1130 * Notify the session daemon that the command is completed.
1131 *
1132 * On transport layer error, the function call will print an error
1133 * message so handling the returned code is a bit useless since we
1134 * return an error code anyway.
1135 */
1136 (void) consumer_send_status_msg(sock, ret_code);
1137 return -ENOENT;
1138 }
1139
1140 /* relayd needs RCU read-side lock */
1141 rcu_read_lock();
1142
1143 switch (msg.cmd_type) {
1144 case LTTNG_CONSUMER_ADD_RELAYD_SOCKET:
1145 {
1146 /* Session daemon status message are handled in the following call. */
1147 ret = consumer_add_relayd_socket(msg.u.relayd_sock.net_index,
1148 msg.u.relayd_sock.type, ctx, sock, consumer_sockpoll,
1149 &msg.u.relayd_sock.sock, msg.u.relayd_sock.session_id);
1150 goto end_nosignal;
1151 }
1152 case LTTNG_CONSUMER_DESTROY_RELAYD:
1153 {
1154 uint64_t index = msg.u.destroy_relayd.net_seq_idx;
1155 struct consumer_relayd_sock_pair *relayd;
1156
1157 DBG("UST consumer destroying relayd %" PRIu64, index);
1158
1159 /* Get relayd reference if exists. */
1160 relayd = consumer_find_relayd(index);
1161 if (relayd == NULL) {
1162 DBG("Unable to find relayd %" PRIu64, index);
1163 ret_code = LTTNG_ERR_NO_CONSUMER;
1164 }
1165
1166 /*
1167 * Each relayd socket pair has a refcount of stream attached to it
1168 * which tells if the relayd is still active or not depending on the
1169 * refcount value.
1170 *
1171 * This will set the destroy flag of the relayd object and destroy it
1172 * if the refcount reaches zero when called.
1173 *
1174 * The destroy can happen either here or when a stream fd hangs up.
1175 */
1176 if (relayd) {
1177 consumer_flag_relayd_for_destroy(relayd);
1178 }
1179
1180 goto end_msg_sessiond;
1181 }
1182 case LTTNG_CONSUMER_UPDATE_STREAM:
1183 {
1184 rcu_read_unlock();
1185 return -ENOSYS;
1186 }
1187 case LTTNG_CONSUMER_DATA_PENDING:
1188 {
1189 int ret, is_data_pending;
1190 uint64_t id = msg.u.data_pending.session_id;
1191
1192 DBG("UST consumer data pending command for id %" PRIu64, id);
1193
1194 is_data_pending = consumer_data_pending(id);
1195
1196 /* Send back returned value to session daemon */
1197 ret = lttcomm_send_unix_sock(sock, &is_data_pending,
1198 sizeof(is_data_pending));
1199 if (ret < 0) {
1200 DBG("Error when sending the data pending ret code: %d", ret);
1201 goto error_fatal;
1202 }
1203
1204 /*
1205 * No need to send back a status message since the data pending
1206 * returned value is the response.
1207 */
1208 break;
1209 }
1210 case LTTNG_CONSUMER_ASK_CHANNEL_CREATION:
1211 {
1212 int ret;
1213 struct ustctl_consumer_channel_attr attr;
1214
1215 /* Create a plain object and reserve a channel key. */
1216 channel = allocate_channel(msg.u.ask_channel.session_id,
1217 msg.u.ask_channel.pathname, msg.u.ask_channel.name,
1218 msg.u.ask_channel.uid, msg.u.ask_channel.gid,
1219 msg.u.ask_channel.relayd_id, msg.u.ask_channel.key,
1220 (enum lttng_event_output) msg.u.ask_channel.output,
1221 msg.u.ask_channel.tracefile_size,
1222 msg.u.ask_channel.tracefile_count,
1223 msg.u.ask_channel.session_id_per_pid,
1224 msg.u.ask_channel.monitor);
1225 if (!channel) {
1226 goto end_channel_error;
1227 }
1228
1229 /* Build channel attributes from received message. */
1230 attr.subbuf_size = msg.u.ask_channel.subbuf_size;
1231 attr.num_subbuf = msg.u.ask_channel.num_subbuf;
1232 attr.overwrite = msg.u.ask_channel.overwrite;
1233 attr.switch_timer_interval = msg.u.ask_channel.switch_timer_interval;
1234 attr.read_timer_interval = msg.u.ask_channel.read_timer_interval;
1235 attr.chan_id = msg.u.ask_channel.chan_id;
1236 attr.output = msg.u.ask_channel.output;
1237 memcpy(attr.uuid, msg.u.ask_channel.uuid, sizeof(attr.uuid));
1238
1239 /* Translate and save channel type. */
1240 switch (msg.u.ask_channel.type) {
1241 case LTTNG_UST_CHAN_PER_CPU:
1242 channel->type = CONSUMER_CHANNEL_TYPE_DATA;
1243 attr.type = LTTNG_UST_CHAN_PER_CPU;
1244 /*
1245 * Set refcount to 1 for owner. Below, we will
1246 * pass ownership to the
1247 * consumer_thread_channel_poll() thread.
1248 */
1249 channel->refcount = 1;
1250 break;
1251 case LTTNG_UST_CHAN_METADATA:
1252 channel->type = CONSUMER_CHANNEL_TYPE_METADATA;
1253 attr.type = LTTNG_UST_CHAN_METADATA;
1254 break;
1255 default:
1256 assert(0);
1257 goto error_fatal;
1258 };
1259
1260 ret = ask_channel(ctx, sock, channel, &attr);
1261 if (ret < 0) {
1262 goto end_channel_error;
1263 }
1264
1265 if (msg.u.ask_channel.type == LTTNG_UST_CHAN_METADATA) {
1266 ret = consumer_metadata_cache_allocate(channel);
1267 if (ret < 0) {
1268 ERR("Allocating metadata cache");
1269 goto end_channel_error;
1270 }
1271 consumer_timer_switch_start(channel, attr.switch_timer_interval);
1272 attr.switch_timer_interval = 0;
1273 }
1274
1275 /*
1276 * Add the channel to the internal state AFTER all streams were created
1277 * and successfully sent to session daemon. This way, all streams must
1278 * be ready before this channel is visible to the threads.
1279 * If add_channel succeeds, ownership of the channel is
1280 * passed to consumer_thread_channel_poll().
1281 */
1282 ret = add_channel(channel, ctx);
1283 if (ret < 0) {
1284 if (msg.u.ask_channel.type == LTTNG_UST_CHAN_METADATA) {
1285 if (channel->switch_timer_enabled == 1) {
1286 consumer_timer_switch_stop(channel);
1287 }
1288 consumer_metadata_cache_destroy(channel);
1289 }
1290 goto end_channel_error;
1291 }
1292
1293 /*
1294 * Channel and streams are now created. Inform the session daemon that
1295 * everything went well and should wait to receive the channel and
1296 * streams with ustctl API.
1297 */
1298 ret = consumer_send_status_channel(sock, channel);
1299 if (ret < 0) {
1300 /*
1301 * There is probably a problem on the socket.
1302 */
1303 goto error_fatal;
1304 }
1305
1306 break;
1307 }
1308 case LTTNG_CONSUMER_GET_CHANNEL:
1309 {
1310 int ret, relayd_err = 0;
1311 uint64_t key = msg.u.get_channel.key;
1312 struct lttng_consumer_channel *channel;
1313
1314 channel = consumer_find_channel(key);
1315 if (!channel) {
1316 ERR("UST consumer get channel key %" PRIu64 " not found", key);
1317 ret_code = LTTNG_ERR_UST_CHAN_NOT_FOUND;
1318 goto end_msg_sessiond;
1319 }
1320
1321 /* Send everything to sessiond. */
1322 ret = send_sessiond_channel(sock, channel, ctx, &relayd_err);
1323 if (ret < 0) {
1324 if (relayd_err) {
1325 /*
1326 * We were unable to send to the relayd the stream so avoid
1327 * sending back a fatal error to the thread since this is OK
1328 * and the consumer can continue its work. The above call
1329 * has sent the error status message to the sessiond.
1330 */
1331 goto end_nosignal;
1332 }
1333 /*
1334 * The communicaton was broken hence there is a bad state between
1335 * the consumer and sessiond so stop everything.
1336 */
1337 goto error_fatal;
1338 }
1339
1340 /*
1341 * In no monitor mode, the streams ownership is kept inside the channel
1342 * so don't send them to the data thread.
1343 */
1344 if (!channel->monitor) {
1345 goto end_msg_sessiond;
1346 }
1347
1348 ret = send_streams_to_thread(channel, ctx);
1349 if (ret < 0) {
1350 /*
1351 * If we are unable to send the stream to the thread, there is
1352 * a big problem so just stop everything.
1353 */
1354 goto error_fatal;
1355 }
1356 /* List MUST be empty after or else it could be reused. */
1357 assert(cds_list_empty(&channel->streams.head));
1358 goto end_msg_sessiond;
1359 }
1360 case LTTNG_CONSUMER_DESTROY_CHANNEL:
1361 {
1362 uint64_t key = msg.u.destroy_channel.key;
1363
1364 /*
1365 * Only called if streams have not been sent to stream
1366 * manager thread. However, channel has been sent to
1367 * channel manager thread.
1368 */
1369 notify_thread_del_channel(ctx, key);
1370 goto end_msg_sessiond;
1371 }
1372 case LTTNG_CONSUMER_CLOSE_METADATA:
1373 {
1374 int ret;
1375
1376 ret = close_metadata(msg.u.close_metadata.key);
1377 if (ret != 0) {
1378 ret_code = ret;
1379 }
1380
1381 goto end_msg_sessiond;
1382 }
1383 case LTTNG_CONSUMER_FLUSH_CHANNEL:
1384 {
1385 int ret;
1386
1387 ret = flush_channel(msg.u.flush_channel.key);
1388 if (ret != 0) {
1389 ret_code = ret;
1390 }
1391
1392 goto end_msg_sessiond;
1393 }
1394 case LTTNG_CONSUMER_PUSH_METADATA:
1395 {
1396 int ret;
1397 uint64_t len = msg.u.push_metadata.len;
1398 uint64_t key = msg.u.push_metadata.key;
1399 uint64_t offset = msg.u.push_metadata.target_offset;
1400 struct lttng_consumer_channel *channel;
1401
1402 DBG("UST consumer push metadata key %" PRIu64 " of len %" PRIu64, key,
1403 len);
1404
1405 channel = consumer_find_channel(key);
1406 if (!channel) {
1407 ERR("UST consumer push metadata %" PRIu64 " not found", key);
1408 ret_code = LTTNG_ERR_UST_CHAN_NOT_FOUND;
1409 goto end_msg_sessiond;
1410 }
1411
1412 /* Tell session daemon we are ready to receive the metadata. */
1413 ret = consumer_send_status_msg(sock, LTTNG_OK);
1414 if (ret < 0) {
1415 /* Somehow, the session daemon is not responding anymore. */
1416 goto error_fatal;
1417 }
1418
1419 /* Wait for more data. */
1420 if (lttng_consumer_poll_socket(consumer_sockpoll) < 0) {
1421 goto error_fatal;
1422 }
1423
1424 ret = lttng_ustconsumer_recv_metadata(sock, key, offset,
1425 len, channel);
1426 if (ret < 0) {
1427 /* error receiving from sessiond */
1428 goto error_fatal;
1429 } else {
1430 ret_code = ret;
1431 goto end_msg_sessiond;
1432 }
1433 }
1434 case LTTNG_CONSUMER_SETUP_METADATA:
1435 {
1436 int ret;
1437
1438 ret = setup_metadata(ctx, msg.u.setup_metadata.key);
1439 if (ret) {
1440 ret_code = ret;
1441 }
1442 goto end_msg_sessiond;
1443 }
1444 case LTTNG_CONSUMER_SNAPSHOT_CHANNEL:
1445 {
1446 if (msg.u.snapshot_channel.metadata) {
1447 ret = snapshot_metadata(msg.u.snapshot_channel.key,
1448 msg.u.snapshot_channel.pathname,
1449 msg.u.snapshot_channel.relayd_id,
1450 ctx);
1451 if (ret < 0) {
1452 ERR("Snapshot metadata failed");
1453 ret_code = LTTNG_ERR_UST_META_FAIL;
1454 }
1455 } else {
1456 ret = snapshot_channel(msg.u.snapshot_channel.key,
1457 msg.u.snapshot_channel.pathname,
1458 msg.u.snapshot_channel.relayd_id,
1459 ctx);
1460 if (ret < 0) {
1461 ERR("Snapshot channel failed");
1462 ret_code = LTTNG_ERR_UST_CHAN_FAIL;
1463 }
1464 }
1465
1466 ret = consumer_send_status_msg(sock, ret_code);
1467 if (ret < 0) {
1468 /* Somehow, the session daemon is not responding anymore. */
1469 goto end_nosignal;
1470 }
1471 break;
1472 }
1473 default:
1474 break;
1475 }
1476
1477 end_nosignal:
1478 rcu_read_unlock();
1479
1480 /*
1481 * Return 1 to indicate success since the 0 value can be a socket
1482 * shutdown during the recv() or send() call.
1483 */
1484 return 1;
1485
1486 end_msg_sessiond:
1487 /*
1488 * The returned value here is not useful since either way we'll return 1 to
1489 * the caller because the session daemon socket management is done
1490 * elsewhere. Returning a negative code or 0 will shutdown the consumer.
1491 */
1492 ret = consumer_send_status_msg(sock, ret_code);
1493 if (ret < 0) {
1494 goto error_fatal;
1495 }
1496 rcu_read_unlock();
1497 return 1;
1498 end_channel_error:
1499 if (channel) {
1500 /*
1501 * Free channel here since no one has a reference to it. We don't
1502 * free after that because a stream can store this pointer.
1503 */
1504 destroy_channel(channel);
1505 }
1506 /* We have to send a status channel message indicating an error. */
1507 ret = consumer_send_status_channel(sock, NULL);
1508 if (ret < 0) {
1509 /* Stop everything if session daemon can not be notified. */
1510 goto error_fatal;
1511 }
1512 rcu_read_unlock();
1513 return 1;
1514 error_fatal:
1515 rcu_read_unlock();
1516 /* This will issue a consumer stop. */
1517 return -1;
1518 }
1519
1520 /*
1521 * Wrapper over the mmap() read offset from ust-ctl library. Since this can be
1522 * compiled out, we isolate it in this library.
1523 */
1524 int lttng_ustctl_get_mmap_read_offset(struct lttng_consumer_stream *stream,
1525 unsigned long *off)
1526 {
1527 assert(stream);
1528 assert(stream->ustream);
1529
1530 return ustctl_get_mmap_read_offset(stream->ustream, off);
1531 }
1532
1533 /*
1534 * Wrapper over the mmap() read offset from ust-ctl library. Since this can be
1535 * compiled out, we isolate it in this library.
1536 */
1537 void *lttng_ustctl_get_mmap_base(struct lttng_consumer_stream *stream)
1538 {
1539 assert(stream);
1540 assert(stream->ustream);
1541
1542 return ustctl_get_mmap_base(stream->ustream);
1543 }
1544
1545 /*
1546 * Take a snapshot for a specific fd
1547 *
1548 * Returns 0 on success, < 0 on error
1549 */
1550 int lttng_ustconsumer_take_snapshot(struct lttng_consumer_stream *stream)
1551 {
1552 assert(stream);
1553 assert(stream->ustream);
1554
1555 return ustctl_snapshot(stream->ustream);
1556 }
1557
1558 /*
1559 * Get the produced position
1560 *
1561 * Returns 0 on success, < 0 on error
1562 */
1563 int lttng_ustconsumer_get_produced_snapshot(
1564 struct lttng_consumer_stream *stream, unsigned long *pos)
1565 {
1566 assert(stream);
1567 assert(stream->ustream);
1568 assert(pos);
1569
1570 return ustctl_snapshot_get_produced(stream->ustream, pos);
1571 }
1572
1573 /*
1574 * Get the consumed position
1575 *
1576 * Returns 0 on success, < 0 on error
1577 */
1578 int lttng_ustconsumer_get_consumed_snapshot(
1579 struct lttng_consumer_stream *stream, unsigned long *pos)
1580 {
1581 assert(stream);
1582 assert(stream->ustream);
1583 assert(pos);
1584
1585 return ustctl_snapshot_get_consumed(stream->ustream, pos);
1586 }
1587
1588 /*
1589 * Called when the stream signal the consumer that it has hang up.
1590 */
1591 void lttng_ustconsumer_on_stream_hangup(struct lttng_consumer_stream *stream)
1592 {
1593 assert(stream);
1594 assert(stream->ustream);
1595
1596 ustctl_flush_buffer(stream->ustream, 0);
1597 stream->hangup_flush_done = 1;
1598 }
1599
1600 void lttng_ustconsumer_del_channel(struct lttng_consumer_channel *chan)
1601 {
1602 assert(chan);
1603 assert(chan->uchan);
1604
1605 if (chan->switch_timer_enabled == 1) {
1606 consumer_timer_switch_stop(chan);
1607 }
1608 consumer_metadata_cache_destroy(chan);
1609 ustctl_destroy_channel(chan->uchan);
1610 }
1611
1612 void lttng_ustconsumer_del_stream(struct lttng_consumer_stream *stream)
1613 {
1614 assert(stream);
1615 assert(stream->ustream);
1616
1617 if (stream->chan->switch_timer_enabled == 1) {
1618 consumer_timer_switch_stop(stream->chan);
1619 }
1620 ustctl_destroy_stream(stream->ustream);
1621 }
1622
1623 int lttng_ustconsumer_read_subbuffer(struct lttng_consumer_stream *stream,
1624 struct lttng_consumer_local_data *ctx)
1625 {
1626 unsigned long len, subbuf_size, padding;
1627 int err;
1628 long ret = 0;
1629 char dummy;
1630 struct ustctl_consumer_stream *ustream;
1631
1632 assert(stream);
1633 assert(stream->ustream);
1634 assert(ctx);
1635
1636 DBG2("In UST read_subbuffer (wait_fd: %d, name: %s)", stream->wait_fd,
1637 stream->name);
1638
1639 /* Ease our life for what's next. */
1640 ustream = stream->ustream;
1641
1642 /* We can consume the 1 byte written into the wait_fd by UST */
1643 if (!stream->hangup_flush_done) {
1644 ssize_t readlen;
1645
1646 do {
1647 readlen = read(stream->wait_fd, &dummy, 1);
1648 } while (readlen == -1 && errno == EINTR);
1649 if (readlen == -1) {
1650 ret = readlen;
1651 goto end;
1652 }
1653 }
1654
1655 /* Get the next subbuffer */
1656 err = ustctl_get_next_subbuf(ustream);
1657 if (err != 0) {
1658 ret = err; /* ustctl_get_next_subbuf returns negative, caller expect positive. */
1659 /*
1660 * This is a debug message even for single-threaded consumer,
1661 * because poll() have more relaxed criterions than get subbuf,
1662 * so get_subbuf may fail for short race windows where poll()
1663 * would issue wakeups.
1664 */
1665 DBG("Reserving sub buffer failed (everything is normal, "
1666 "it is due to concurrency) [ret: %d]", err);
1667 goto end;
1668 }
1669 assert(stream->chan->output == CONSUMER_CHANNEL_MMAP);
1670 /* Get the full padded subbuffer size */
1671 err = ustctl_get_padded_subbuf_size(ustream, &len);
1672 assert(err == 0);
1673
1674 /* Get subbuffer data size (without padding) */
1675 err = ustctl_get_subbuf_size(ustream, &subbuf_size);
1676 assert(err == 0);
1677
1678 /* Make sure we don't get a subbuffer size bigger than the padded */
1679 assert(len >= subbuf_size);
1680
1681 padding = len - subbuf_size;
1682 /* write the subbuffer to the tracefile */
1683 ret = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, subbuf_size, padding);
1684 /*
1685 * The mmap operation should write subbuf_size amount of data when network
1686 * streaming or the full padding (len) size when we are _not_ streaming.
1687 */
1688 if ((ret != subbuf_size && stream->net_seq_idx != (uint64_t) -1ULL) ||
1689 (ret != len && stream->net_seq_idx == (uint64_t) -1ULL)) {
1690 /*
1691 * Display the error but continue processing to try to release the
1692 * subbuffer. This is a DBG statement since any unexpected kill or
1693 * signal, the application gets unregistered, relayd gets closed or
1694 * anything that affects the buffer lifetime will trigger this error.
1695 * So, for the sake of the user, don't print this error since it can
1696 * happen and it is OK with the code flow.
1697 */
1698 DBG("Error writing to tracefile "
1699 "(ret: %ld != len: %lu != subbuf_size: %lu)",
1700 ret, len, subbuf_size);
1701 }
1702 err = ustctl_put_next_subbuf(ustream);
1703 assert(err == 0);
1704
1705 end:
1706 return ret;
1707 }
1708
1709 /*
1710 * Called when a stream is created.
1711 *
1712 * Return 0 on success or else a negative value.
1713 */
1714 int lttng_ustconsumer_on_recv_stream(struct lttng_consumer_stream *stream)
1715 {
1716 int ret;
1717
1718 assert(stream);
1719
1720 /* Don't create anything if this is set for streaming. */
1721 if (stream->net_seq_idx == (uint64_t) -1ULL && stream->chan->monitor) {
1722 ret = utils_create_stream_file(stream->chan->pathname, stream->name,
1723 stream->chan->tracefile_size, stream->tracefile_count_current,
1724 stream->uid, stream->gid);
1725 if (ret < 0) {
1726 goto error;
1727 }
1728 stream->out_fd = ret;
1729 stream->tracefile_size_current = 0;
1730 }
1731 ret = 0;
1732
1733 error:
1734 return ret;
1735 }
1736
1737 /*
1738 * Check if data is still being extracted from the buffers for a specific
1739 * stream. Consumer data lock MUST be acquired before calling this function
1740 * and the stream lock.
1741 *
1742 * Return 1 if the traced data are still getting read else 0 meaning that the
1743 * data is available for trace viewer reading.
1744 */
1745 int lttng_ustconsumer_data_pending(struct lttng_consumer_stream *stream)
1746 {
1747 int ret;
1748
1749 assert(stream);
1750 assert(stream->ustream);
1751
1752 DBG("UST consumer checking data pending");
1753
1754 ret = ustctl_get_next_subbuf(stream->ustream);
1755 if (ret == 0) {
1756 /* There is still data so let's put back this subbuffer. */
1757 ret = ustctl_put_subbuf(stream->ustream);
1758 assert(ret == 0);
1759 ret = 1; /* Data is pending */
1760 goto end;
1761 }
1762
1763 /* Data is NOT pending so ready to be read. */
1764 ret = 0;
1765
1766 end:
1767 return ret;
1768 }
1769
1770 /*
1771 * Close every metadata stream wait fd of the metadata hash table. This
1772 * function MUST be used very carefully so not to run into a race between the
1773 * metadata thread handling streams and this function closing their wait fd.
1774 *
1775 * For UST, this is used when the session daemon hangs up. Its the metadata
1776 * producer so calling this is safe because we are assured that no state change
1777 * can occur in the metadata thread for the streams in the hash table.
1778 */
1779 void lttng_ustconsumer_close_metadata(struct lttng_ht *metadata_ht)
1780 {
1781 int ret;
1782 struct lttng_ht_iter iter;
1783 struct lttng_consumer_stream *stream;
1784
1785 assert(metadata_ht);
1786 assert(metadata_ht->ht);
1787
1788 DBG("UST consumer closing all metadata streams");
1789
1790 rcu_read_lock();
1791 cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream,
1792 node.node) {
1793 int fd = stream->wait_fd;
1794
1795 /*
1796 * Whatever happens here we have to continue to try to close every
1797 * streams. Let's report at least the error on failure.
1798 */
1799 ret = ustctl_stream_close_wakeup_fd(stream->ustream);
1800 if (ret) {
1801 ERR("Unable to close metadata stream fd %d ret %d", fd, ret);
1802 }
1803 DBG("Metadata wait fd %d closed", fd);
1804 }
1805 rcu_read_unlock();
1806 }
1807
1808 void lttng_ustconsumer_close_stream_wakeup(struct lttng_consumer_stream *stream)
1809 {
1810 int ret;
1811
1812 ret = ustctl_stream_close_wakeup_fd(stream->ustream);
1813 if (ret < 0) {
1814 ERR("Unable to close wakeup fd");
1815 }
1816 }
1817
1818 int lttng_ustconsumer_request_metadata(struct lttng_consumer_local_data *ctx,
1819 struct lttng_consumer_channel *channel)
1820 {
1821 struct lttcomm_metadata_request_msg request;
1822 struct lttcomm_consumer_msg msg;
1823 enum lttng_error_code ret_code = LTTNG_OK;
1824 uint64_t len, key, offset;
1825 int ret;
1826
1827 assert(channel);
1828 assert(channel->metadata_cache);
1829
1830 /* send the metadata request to sessiond */
1831 switch (consumer_data.type) {
1832 case LTTNG_CONSUMER64_UST:
1833 request.bits_per_long = 64;
1834 break;
1835 case LTTNG_CONSUMER32_UST:
1836 request.bits_per_long = 32;
1837 break;
1838 default:
1839 request.bits_per_long = 0;
1840 break;
1841 }
1842
1843 request.session_id = channel->session_id;
1844 request.session_id_per_pid = channel->session_id_per_pid;
1845 request.uid = channel->uid;
1846 request.key = channel->key;
1847 DBG("Sending metadata request to sessiond, session id %" PRIu64
1848 ", per-pid %" PRIu64,
1849 channel->session_id,
1850 channel->session_id_per_pid);
1851
1852 ret = lttcomm_send_unix_sock(ctx->consumer_metadata_socket, &request,
1853 sizeof(request));
1854 if (ret < 0) {
1855 ERR("Asking metadata to sessiond");
1856 goto end;
1857 }
1858
1859 /* Receive the metadata from sessiond */
1860 ret = lttcomm_recv_unix_sock(ctx->consumer_metadata_socket, &msg,
1861 sizeof(msg));
1862 if (ret != sizeof(msg)) {
1863 DBG("Consumer received unexpected message size %d (expects %zu)",
1864 ret, sizeof(msg));
1865 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD);
1866 /*
1867 * The ret value might 0 meaning an orderly shutdown but this is ok
1868 * since the caller handles this.
1869 */
1870 goto end;
1871 }
1872
1873 if (msg.cmd_type == LTTNG_ERR_UND) {
1874 /* No registry found */
1875 (void) consumer_send_status_msg(ctx->consumer_metadata_socket,
1876 ret_code);
1877 ret = 0;
1878 goto end;
1879 } else if (msg.cmd_type != LTTNG_CONSUMER_PUSH_METADATA) {
1880 ERR("Unexpected cmd_type received %d", msg.cmd_type);
1881 ret = -1;
1882 goto end;
1883 }
1884
1885 len = msg.u.push_metadata.len;
1886 key = msg.u.push_metadata.key;
1887 offset = msg.u.push_metadata.target_offset;
1888
1889 assert(key == channel->key);
1890 if (len == 0) {
1891 DBG("No new metadata to receive for key %" PRIu64, key);
1892 }
1893
1894 /* Tell session daemon we are ready to receive the metadata. */
1895 ret = consumer_send_status_msg(ctx->consumer_metadata_socket,
1896 LTTNG_OK);
1897 if (ret < 0 || len == 0) {
1898 /*
1899 * Somehow, the session daemon is not responding anymore or there is
1900 * nothing to receive.
1901 */
1902 goto end;
1903 }
1904
1905 ret_code = lttng_ustconsumer_recv_metadata(ctx->consumer_metadata_socket,
1906 key, offset, len, channel);
1907 if (ret_code >= 0) {
1908 /*
1909 * Only send the status msg if the sessiond is alive meaning a positive
1910 * ret code.
1911 */
1912 (void) consumer_send_status_msg(ctx->consumer_metadata_socket, ret_code);
1913 }
1914 ret = 0;
1915
1916 end:
1917 return ret;
1918 }
This page took 0.10443 seconds and 5 git commands to generate.