Add UST snapshot support
[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 if (stream == NULL) {
157 switch (alloc_ret) {
158 case -ENOENT:
159 /*
160 * We could not find the channel. Can happen if cpu hotplug
161 * happens while tearing down.
162 */
163 DBG3("Could not find channel");
164 break;
165 case -ENOMEM:
166 case -EINVAL:
167 default:
168 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
169 break;
170 }
171 goto error;
172 }
173
174 stream->chan = channel;
175
176 error:
177 if (_alloc_ret) {
178 *_alloc_ret = alloc_ret;
179 }
180 return stream;
181 }
182
183 /*
184 * Send the given stream pointer to the corresponding thread.
185 *
186 * Returns 0 on success else a negative value.
187 */
188 static int send_stream_to_thread(struct lttng_consumer_stream *stream,
189 struct lttng_consumer_local_data *ctx)
190 {
191 int ret;
192 struct lttng_pipe *stream_pipe;
193
194 /* Get the right pipe where the stream will be sent. */
195 if (stream->metadata_flag) {
196 stream_pipe = ctx->consumer_metadata_pipe;
197 } else {
198 stream_pipe = ctx->consumer_data_pipe;
199 }
200
201 ret = lttng_pipe_write(stream_pipe, &stream, sizeof(stream));
202 if (ret < 0) {
203 ERR("Consumer write %s stream to pipe %d",
204 stream->metadata_flag ? "metadata" : "data",
205 lttng_pipe_get_writefd(stream_pipe));
206 }
207
208 return ret;
209 }
210
211 /*
212 * Search for a relayd object related to the stream. If found, send the stream
213 * to the relayd.
214 *
215 * On success, returns 0 else a negative value.
216 */
217 static int send_stream_to_relayd(struct lttng_consumer_stream *stream)
218 {
219 int ret = 0;
220 struct consumer_relayd_sock_pair *relayd;
221
222 assert(stream);
223
224 relayd = consumer_find_relayd(stream->net_seq_idx);
225 if (relayd != NULL) {
226 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
227 /* Add stream on the relayd */
228 ret = relayd_add_stream(&relayd->control_sock, stream->name,
229 stream->chan->pathname, &stream->relayd_stream_id,
230 stream->chan->tracefile_size,
231 stream->chan->tracefile_count);
232 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
233 if (ret < 0) {
234 goto error;
235 }
236 } else if (stream->net_seq_idx != (uint64_t) -1ULL) {
237 ERR("Network sequence index %" PRIu64 " unknown. Not adding stream.",
238 stream->net_seq_idx);
239 ret = -1;
240 goto error;
241 }
242
243 error:
244 return ret;
245 }
246
247 /*
248 * Create streams for the given channel using liblttng-ust-ctl.
249 *
250 * Return 0 on success else a negative value.
251 */
252 static int create_ust_streams(struct lttng_consumer_channel *channel,
253 struct lttng_consumer_local_data *ctx)
254 {
255 int ret, cpu = 0;
256 struct ustctl_consumer_stream *ustream;
257 struct lttng_consumer_stream *stream;
258
259 assert(channel);
260 assert(ctx);
261
262 /*
263 * While a stream is available from ustctl. When NULL is returned, we've
264 * reached the end of the possible stream for the channel.
265 */
266 while ((ustream = ustctl_create_stream(channel->uchan, cpu))) {
267 int wait_fd;
268
269 wait_fd = ustctl_stream_get_wait_fd(ustream);
270
271 /* Allocate consumer stream object. */
272 stream = allocate_stream(cpu, wait_fd, channel, ctx, &ret);
273 if (!stream) {
274 goto error_alloc;
275 }
276 stream->ustream = ustream;
277 /*
278 * Store it so we can save multiple function calls afterwards since
279 * this value is used heavily in the stream threads. This is UST
280 * specific so this is why it's done after allocation.
281 */
282 stream->wait_fd = wait_fd;
283
284 /*
285 * Increment channel refcount since the channel reference has now been
286 * assigned in the allocation process above.
287 */
288 if (stream->chan->monitor) {
289 uatomic_inc(&stream->chan->refcount);
290 }
291
292 /*
293 * Order is important this is why a list is used. On error, the caller
294 * should clean this list.
295 */
296 cds_list_add_tail(&stream->send_node, &channel->streams.head);
297
298 ret = ustctl_get_max_subbuf_size(stream->ustream,
299 &stream->max_sb_size);
300 if (ret < 0) {
301 ERR("ustctl_get_max_subbuf_size failed for stream %s",
302 stream->name);
303 goto error;
304 }
305
306 /* Do actions once stream has been received. */
307 if (ctx->on_recv_stream) {
308 ret = ctx->on_recv_stream(stream);
309 if (ret < 0) {
310 goto error;
311 }
312 }
313
314 DBG("UST consumer add stream %s (key: %" PRIu64 ") with relayd id %" PRIu64,
315 stream->name, stream->key, stream->relayd_stream_id);
316
317 /* Set next CPU stream. */
318 channel->streams.count = ++cpu;
319
320 /* Keep stream reference when creating metadata. */
321 if (channel->type == CONSUMER_CHANNEL_TYPE_METADATA) {
322 channel->metadata_stream = stream;
323 }
324 }
325
326 return 0;
327
328 error:
329 error_alloc:
330 return ret;
331 }
332
333 /*
334 * Create an UST channel with the given attributes and send it to the session
335 * daemon using the ust ctl API.
336 *
337 * Return 0 on success or else a negative value.
338 */
339 static int create_ust_channel(struct ustctl_consumer_channel_attr *attr,
340 struct ustctl_consumer_channel **chanp)
341 {
342 int ret;
343 struct ustctl_consumer_channel *channel;
344
345 assert(attr);
346 assert(chanp);
347
348 DBG3("Creating channel to ustctl with attr: [overwrite: %d, "
349 "subbuf_size: %" PRIu64 ", num_subbuf: %" PRIu64 ", "
350 "switch_timer_interval: %u, read_timer_interval: %u, "
351 "output: %d, type: %d", attr->overwrite, attr->subbuf_size,
352 attr->num_subbuf, attr->switch_timer_interval,
353 attr->read_timer_interval, attr->output, attr->type);
354
355 channel = ustctl_create_channel(attr);
356 if (!channel) {
357 ret = -1;
358 goto error_create;
359 }
360
361 *chanp = channel;
362
363 return 0;
364
365 error_create:
366 return ret;
367 }
368
369 /*
370 * Send a single given stream to the session daemon using the sock.
371 *
372 * Return 0 on success else a negative value.
373 */
374 static int send_sessiond_stream(int sock, struct lttng_consumer_stream *stream)
375 {
376 int ret;
377
378 assert(stream);
379 assert(sock >= 0);
380
381 DBG2("UST consumer sending stream %" PRIu64 " to sessiond", stream->key);
382
383 /* Send stream to session daemon. */
384 ret = ustctl_send_stream_to_sessiond(sock, stream->ustream);
385 if (ret < 0) {
386 goto error;
387 }
388
389 error:
390 return ret;
391 }
392
393 /*
394 * Send channel to sessiond.
395 *
396 * Return 0 on success or else a negative value.
397 */
398 static int send_sessiond_channel(int sock,
399 struct lttng_consumer_channel *channel,
400 struct lttng_consumer_local_data *ctx, int *relayd_error)
401 {
402 int ret, ret_code = LTTNG_OK;
403 struct lttng_consumer_stream *stream;
404
405 assert(channel);
406 assert(ctx);
407 assert(sock >= 0);
408
409 DBG("UST consumer sending channel %s to sessiond", channel->name);
410
411 cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
412 /* Try to send the stream to the relayd if one is available. */
413 ret = send_stream_to_relayd(stream);
414 if (ret < 0) {
415 /*
416 * Flag that the relayd was the problem here probably due to a
417 * communicaton error on the socket.
418 */
419 if (relayd_error) {
420 *relayd_error = 1;
421 }
422 ret_code = LTTNG_ERR_RELAYD_CONNECT_FAIL;
423 }
424 }
425
426 /* Inform sessiond that we are about to send channel and streams. */
427 ret = consumer_send_status_msg(sock, ret_code);
428 if (ret < 0 || ret_code != LTTNG_OK) {
429 /*
430 * Either the session daemon is not responding or the relayd died so we
431 * stop now.
432 */
433 goto error;
434 }
435
436 /* Send channel to sessiond. */
437 ret = ustctl_send_channel_to_sessiond(sock, channel->uchan);
438 if (ret < 0) {
439 goto error;
440 }
441
442 ret = ustctl_channel_close_wakeup_fd(channel->uchan);
443 if (ret < 0) {
444 goto error;
445 }
446
447 /* The channel was sent successfully to the sessiond at this point. */
448 cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
449 /* Send stream to session daemon. */
450 ret = send_sessiond_stream(sock, stream);
451 if (ret < 0) {
452 goto error;
453 }
454 }
455
456 /* Tell sessiond there is no more stream. */
457 ret = ustctl_send_stream_to_sessiond(sock, NULL);
458 if (ret < 0) {
459 goto error;
460 }
461
462 DBG("UST consumer NULL stream sent to sessiond");
463
464 return 0;
465
466 error:
467 if (ret_code != LTTNG_OK) {
468 ret = -1;
469 }
470 return ret;
471 }
472
473 /*
474 * Creates a channel and streams and add the channel it to the channel internal
475 * state. The created stream must ONLY be sent once the GET_CHANNEL command is
476 * received.
477 *
478 * Return 0 on success or else, a negative value is returned and the channel
479 * MUST be destroyed by consumer_del_channel().
480 */
481 static int ask_channel(struct lttng_consumer_local_data *ctx, int sock,
482 struct lttng_consumer_channel *channel,
483 struct ustctl_consumer_channel_attr *attr)
484 {
485 int ret;
486
487 assert(ctx);
488 assert(channel);
489 assert(attr);
490
491 /*
492 * This value is still used by the kernel consumer since for the kernel,
493 * the stream ownership is not IN the consumer so we need to have the
494 * number of left stream that needs to be initialized so we can know when
495 * to delete the channel (see consumer.c).
496 *
497 * As for the user space tracer now, the consumer creates and sends the
498 * stream to the session daemon which only sends them to the application
499 * once every stream of a channel is received making this value useless
500 * because we they will be added to the poll thread before the application
501 * receives them. This ensures that a stream can not hang up during
502 * initilization of a channel.
503 */
504 channel->nb_init_stream_left = 0;
505
506 /* The reply msg status is handled in the following call. */
507 ret = create_ust_channel(attr, &channel->uchan);
508 if (ret < 0) {
509 goto end;
510 }
511
512 channel->wait_fd = ustctl_channel_get_wait_fd(channel->uchan);
513
514 /*
515 * For the snapshots (no monitor), we create the metadata streams
516 * on demand, not during the channel creation.
517 */
518 if (channel->type == CONSUMER_CHANNEL_TYPE_METADATA && !channel->monitor) {
519 ret = 0;
520 goto end;
521 }
522
523 /* Open all streams for this channel. */
524 ret = create_ust_streams(channel, ctx);
525 if (ret < 0) {
526 goto end;
527 }
528
529 end:
530 return ret;
531 }
532
533 /*
534 * Send all stream of a channel to the right thread handling it.
535 *
536 * On error, return a negative value else 0 on success.
537 */
538 static int send_streams_to_thread(struct lttng_consumer_channel *channel,
539 struct lttng_consumer_local_data *ctx)
540 {
541 int ret = 0;
542 struct lttng_consumer_stream *stream, *stmp;
543
544 assert(channel);
545 assert(ctx);
546
547 /* Send streams to the corresponding thread. */
548 cds_list_for_each_entry_safe(stream, stmp, &channel->streams.head,
549 send_node) {
550 /* Sending the stream to the thread. */
551 ret = send_stream_to_thread(stream, ctx);
552 if (ret < 0) {
553 /*
554 * If we are unable to send the stream to the thread, there is
555 * a big problem so just stop everything.
556 */
557 goto error;
558 }
559
560 /* Remove node from the channel stream list. */
561 cds_list_del(&stream->send_node);
562 }
563
564 error:
565 return ret;
566 }
567
568 /*
569 * Write metadata to the given channel using ustctl to convert the string to
570 * the ringbuffer.
571 * Called only from consumer_metadata_cache_write.
572 * The metadata cache lock MUST be acquired to write in the cache.
573 *
574 * Return 0 on success else a negative value.
575 */
576 int lttng_ustconsumer_push_metadata(struct lttng_consumer_channel *metadata,
577 const char *metadata_str, uint64_t target_offset, uint64_t len)
578 {
579 int ret;
580
581 assert(metadata);
582 assert(metadata_str);
583
584 DBG("UST consumer writing metadata to channel %s", metadata->name);
585
586 if (!metadata->metadata_stream) {
587 ret = 0;
588 goto error;
589 }
590
591 assert(target_offset <= metadata->metadata_cache->max_offset);
592 ret = ustctl_write_metadata_to_channel(metadata->uchan,
593 metadata_str + target_offset, len);
594 if (ret < 0) {
595 ERR("ustctl write metadata fail with ret %d, len %" PRIu64, ret, len);
596 goto error;
597 }
598
599 ustctl_flush_buffer(metadata->metadata_stream->ustream, 1);
600
601 error:
602 return ret;
603 }
604
605 /*
606 * Flush channel's streams using the given key to retrieve the channel.
607 *
608 * Return 0 on success else an LTTng error code.
609 */
610 static int flush_channel(uint64_t chan_key)
611 {
612 int ret = 0;
613 struct lttng_consumer_channel *channel;
614 struct lttng_consumer_stream *stream;
615 struct lttng_ht *ht;
616 struct lttng_ht_iter iter;
617
618 DBG("UST consumer flush channel key %" PRIu64, chan_key);
619
620 rcu_read_lock();
621 channel = consumer_find_channel(chan_key);
622 if (!channel) {
623 ERR("UST consumer flush channel %" PRIu64 " not found", chan_key);
624 ret = LTTNG_ERR_UST_CHAN_NOT_FOUND;
625 goto error;
626 }
627
628 ht = consumer_data.stream_per_chan_id_ht;
629
630 /* For each stream of the channel id, flush it. */
631 cds_lfht_for_each_entry_duplicate(ht->ht,
632 ht->hash_fct(&channel->key, lttng_ht_seed), ht->match_fct,
633 &channel->key, &iter.iter, stream, node_channel_id.node) {
634 ustctl_flush_buffer(stream->ustream, 1);
635 }
636 error:
637 rcu_read_unlock();
638 return ret;
639 }
640
641 /*
642 * Close metadata stream wakeup_fd using the given key to retrieve the channel.
643 * RCU read side lock MUST be acquired before calling this function.
644 *
645 * Return 0 on success else an LTTng error code.
646 */
647 static int close_metadata(uint64_t chan_key)
648 {
649 int ret = 0;
650 struct lttng_consumer_channel *channel;
651
652 DBG("UST consumer close metadata key %" PRIu64, chan_key);
653
654 channel = consumer_find_channel(chan_key);
655 if (!channel) {
656 /*
657 * This is possible if the metadata thread has issue a delete because
658 * the endpoint point of the stream hung up. There is no way the
659 * session daemon can know about it thus use a DBG instead of an actual
660 * error.
661 */
662 DBG("UST consumer close metadata %" PRIu64 " not found", chan_key);
663 ret = LTTNG_ERR_UST_CHAN_NOT_FOUND;
664 goto error;
665 }
666
667 pthread_mutex_lock(&consumer_data.lock);
668
669 if (cds_lfht_is_node_deleted(&channel->node.node)) {
670 goto error_unlock;
671 }
672
673 if (channel->switch_timer_enabled == 1) {
674 DBG("Deleting timer on metadata channel");
675 consumer_timer_switch_stop(channel);
676 }
677
678 if (channel->metadata_stream) {
679 ret = ustctl_stream_close_wakeup_fd(channel->metadata_stream->ustream);
680 if (ret < 0) {
681 ERR("UST consumer unable to close fd of metadata (ret: %d)", ret);
682 ret = LTTCOMM_CONSUMERD_ERROR_METADATA;
683 goto error_unlock;
684 }
685 }
686
687 error_unlock:
688 pthread_mutex_unlock(&consumer_data.lock);
689 error:
690 return ret;
691 }
692
693 /*
694 * RCU read side lock MUST be acquired before calling this function.
695 *
696 * Return 0 on success else an LTTng error code.
697 */
698 static int setup_metadata(struct lttng_consumer_local_data *ctx, uint64_t key)
699 {
700 int ret;
701 struct lttng_consumer_channel *metadata;
702
703 DBG("UST consumer setup metadata key %" PRIu64, key);
704
705 metadata = consumer_find_channel(key);
706 if (!metadata) {
707 ERR("UST consumer push metadata %" PRIu64 " not found", key);
708 ret = LTTNG_ERR_UST_CHAN_NOT_FOUND;
709 goto end;
710 }
711
712 /*
713 * In no monitor mode, the metadata channel has no stream(s) so skip the
714 * ownership transfer to the metadata thread.
715 */
716 if (!metadata->monitor) {
717 DBG("Metadata channel in no monitor");
718 ret = 0;
719 goto end;
720 }
721
722 /*
723 * Send metadata stream to relayd if one available. Availability is
724 * known if the stream is still in the list of the channel.
725 */
726 if (cds_list_empty(&metadata->streams.head)) {
727 ERR("Metadata channel key %" PRIu64 ", no stream available.", key);
728 ret = LTTCOMM_CONSUMERD_ERROR_METADATA;
729 goto error;
730 }
731
732 /* Send metadata stream to relayd if needed. */
733 ret = send_stream_to_relayd(metadata->metadata_stream);
734 if (ret < 0) {
735 ret = LTTCOMM_CONSUMERD_ERROR_METADATA;
736 goto error;
737 }
738
739 ret = send_streams_to_thread(metadata, ctx);
740 if (ret < 0) {
741 /*
742 * If we are unable to send the stream to the thread, there is
743 * a big problem so just stop everything.
744 */
745 ret = LTTCOMM_CONSUMERD_FATAL;
746 goto error;
747 }
748 /* List MUST be empty after or else it could be reused. */
749 assert(cds_list_empty(&metadata->streams.head));
750
751 ret = 0;
752 goto end;
753
754 error:
755 /*
756 * Delete metadata channel on error. At this point, the metadata stream can
757 * NOT be monitored by the metadata thread thus having the guarantee that
758 * the stream is still in the local stream list of the channel. This call
759 * will make sure to clean that list.
760 */
761 consumer_del_channel(metadata);
762 end:
763 return ret;
764 }
765
766 /*
767 * Snapshot the whole metadata.
768 *
769 * Returns 0 on success, < 0 on error
770 */
771 static int snapshot_metadata(uint64_t key, char *path, uint64_t relayd_id,
772 struct lttng_consumer_local_data *ctx)
773 {
774 int ret = 0;
775 ssize_t write_len;
776 uint64_t total_len = 0;
777 struct lttng_consumer_channel *metadata_channel;
778 struct lttng_consumer_stream *metadata_stream;
779
780 assert(path);
781 assert(ctx);
782
783 DBG("UST consumer snapshot metadata with key %" PRIu64 " at path %s",
784 key, path);
785
786 rcu_read_lock();
787
788 metadata_channel = consumer_find_channel(key);
789 if (!metadata_channel) {
790 ERR("UST snapshot metadata channel not found for key %lu", key);
791 ret = -1;
792 goto error;
793 }
794 assert(!metadata_channel->monitor);
795
796 /*
797 * Ask the sessiond if we have new metadata waiting and update the
798 * consumer metadata cache.
799 */
800 ret = lttng_ustconsumer_request_metadata(ctx, metadata_channel);
801 if (ret < 0) {
802 goto error;
803 }
804
805 /*
806 * The metadata stream is NOT created in no monitor mode when the channel
807 * is created on a sessiond ask channel command.
808 */
809 ret = create_ust_streams(metadata_channel, ctx);
810 if (ret < 0) {
811 goto error;
812 }
813
814 metadata_stream = metadata_channel->metadata_stream;
815 assert(metadata_stream);
816
817 if (relayd_id != (uint64_t) -1ULL) {
818 metadata_stream->net_seq_idx = relayd_id;
819 ret = consumer_send_relayd_stream(metadata_stream, path);
820 if (ret < 0) {
821 goto error_stream;
822 }
823 } else {
824 ret = utils_create_stream_file(path, metadata_stream->name,
825 metadata_stream->chan->tracefile_size,
826 metadata_stream->tracefile_count_current,
827 metadata_stream->uid, metadata_stream->gid);
828 if (ret < 0) {
829 goto error_stream;
830 }
831 metadata_stream->out_fd = ret;
832 metadata_stream->tracefile_size_current = 0;
833 }
834
835 pthread_mutex_lock(&metadata_channel->metadata_cache->lock);
836 while (total_len < metadata_channel->metadata_cache->total_bytes_written) {
837 /*
838 * Write at most one packet of metadata into the channel
839 * to avoid blocking here.
840 */
841 write_len = ustctl_write_one_packet_to_channel(metadata_channel->uchan,
842 metadata_channel->metadata_cache->data,
843 metadata_channel->metadata_cache->total_bytes_written);
844 if (write_len < 0) {
845 ERR("UST consumer snapshot writing metadata packet");
846 ret = -1;
847 goto error_unlock;
848 }
849 total_len += write_len;
850
851 DBG("Written %" PRIu64 " bytes to metadata (left: %" PRIu64 ")",
852 write_len,
853 metadata_channel->metadata_cache->total_bytes_written - write_len);
854 ustctl_flush_buffer(metadata_stream->ustream, 1);
855 ret = lttng_consumer_read_subbuffer(metadata_stream, ctx);
856 if (ret < 0) {
857 goto error_unlock;
858 }
859 }
860
861 error_unlock:
862 pthread_mutex_unlock(&metadata_channel->metadata_cache->lock);
863
864 error_stream:
865 /*
866 * Clean up the stream completly because the next snapshot will use a new
867 * metadata stream.
868 */
869 cds_list_del(&metadata_stream->send_node);
870 consumer_stream_destroy(metadata_stream, NULL);
871 metadata_channel->metadata_stream = NULL;
872
873 error:
874 rcu_read_unlock();
875 return ret;
876 }
877
878 /*
879 * Take a snapshot of all the stream of a channel.
880 *
881 * Returns 0 on success, < 0 on error
882 */
883 static int snapshot_channel(uint64_t key, char *path, uint64_t relayd_id,
884 struct lttng_consumer_local_data *ctx)
885 {
886 int ret;
887 unsigned use_relayd = 0;
888 unsigned long consumed_pos, produced_pos;
889 struct lttng_consumer_channel *channel;
890 struct lttng_consumer_stream *stream;
891
892 assert(path);
893 assert(ctx);
894
895 rcu_read_lock();
896
897 if (relayd_id != (uint64_t) -1ULL) {
898 use_relayd = 1;
899 }
900
901 channel = consumer_find_channel(key);
902 if (!channel) {
903 ERR("UST snapshot channel not found for key %lu", key);
904 ret = -1;
905 goto error;
906 }
907 assert(!channel->monitor);
908 DBG("UST consumer snapshot channel %lu", key);
909
910 cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
911 /* Lock stream because we are about to change its state. */
912 pthread_mutex_lock(&stream->lock);
913 stream->net_seq_idx = relayd_id;
914
915 if (use_relayd) {
916 ret = consumer_send_relayd_stream(stream, path);
917 if (ret < 0) {
918 goto error_unlock;
919 }
920 } else {
921 ret = utils_create_stream_file(path, stream->name,
922 stream->chan->tracefile_size,
923 stream->tracefile_count_current,
924 stream->uid, stream->gid);
925 if (ret < 0) {
926 goto error_unlock;
927 }
928 stream->out_fd = ret;
929 stream->tracefile_size_current = 0;
930
931 DBG("UST consumer snapshot stream %s/%s (%" PRIu64 ")", path,
932 stream->name, stream->key);
933 }
934
935 ustctl_flush_buffer(stream->ustream, 1);
936
937 ret = lttng_ustconsumer_take_snapshot(stream);
938 if (ret < 0) {
939 ERR("Taking UST snapshot");
940 goto error_unlock;
941 }
942
943 ret = lttng_ustconsumer_get_produced_snapshot(stream, &produced_pos);
944 if (ret < 0) {
945 ERR("Produced UST snapshot position");
946 goto error_unlock;
947 }
948
949 ret = lttng_ustconsumer_get_consumed_snapshot(stream, &consumed_pos);
950 if (ret < 0) {
951 ERR("Consumerd UST snapshot position");
952 goto error_unlock;
953 }
954
955 while (consumed_pos < produced_pos) {
956 ssize_t read_len;
957 unsigned long len, padded_len;
958
959 DBG("UST consumer taking snapshot at pos %lu", consumed_pos);
960
961 ret = ustctl_get_subbuf(stream->ustream, &consumed_pos);
962 if (ret < 0) {
963 if (ret != -EAGAIN) {
964 PERROR("ustctl_get_subbuf snapshot");
965 goto error_close_stream;
966 }
967 DBG("UST consumer get subbuf failed. Skipping it.");
968 consumed_pos += stream->max_sb_size;
969 continue;
970 }
971
972 ret = ustctl_get_subbuf_size(stream->ustream, &len);
973 if (ret < 0) {
974 ERR("Snapshot ustctl_get_subbuf_size");
975 goto error_put_subbuf;
976 }
977
978 ret = ustctl_get_padded_subbuf_size(stream->ustream, &padded_len);
979 if (ret < 0) {
980 ERR("Snapshot ustctl_get_padded_subbuf_size");
981 goto error_put_subbuf;
982 }
983
984 read_len = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, len,
985 padded_len - len);
986 if (use_relayd) {
987 if (read_len != len) {
988 ret = -1;
989 goto error_put_subbuf;
990 }
991 } else {
992 if (read_len != padded_len) {
993 ret = -1;
994 goto error_put_subbuf;
995 }
996 }
997
998 ret = ustctl_put_subbuf(stream->ustream);
999 if (ret < 0) {
1000 ERR("Snapshot ustctl_put_subbuf");
1001 goto error_close_stream;
1002 }
1003 consumed_pos += stream->max_sb_size;
1004 }
1005
1006 /* Simply close the stream so we can use it on the next snapshot. */
1007 consumer_stream_close(stream);
1008 pthread_mutex_unlock(&stream->lock);
1009 }
1010
1011 rcu_read_unlock();
1012 return 0;
1013
1014 error_put_subbuf:
1015 if (ustctl_put_subbuf(stream->ustream) < 0) {
1016 ERR("Snapshot ustctl_put_subbuf");
1017 }
1018 error_close_stream:
1019 consumer_stream_close(stream);
1020 error_unlock:
1021 pthread_mutex_unlock(&stream->lock);
1022 error:
1023 rcu_read_unlock();
1024 return ret;
1025 }
1026
1027 /*
1028 * Receive the metadata updates from the sessiond.
1029 */
1030 int lttng_ustconsumer_recv_metadata(int sock, uint64_t key, uint64_t offset,
1031 uint64_t len, struct lttng_consumer_channel *channel)
1032 {
1033 int ret, ret_code = LTTNG_OK;
1034 char *metadata_str;
1035
1036 DBG("UST consumer push metadata key %" PRIu64 " of len %" PRIu64, key, len);
1037
1038 metadata_str = zmalloc(len * sizeof(char));
1039 if (!metadata_str) {
1040 PERROR("zmalloc metadata string");
1041 ret_code = LTTCOMM_CONSUMERD_ENOMEM;
1042 goto end;
1043 }
1044
1045 /* Receive metadata string. */
1046 ret = lttcomm_recv_unix_sock(sock, metadata_str, len);
1047 if (ret < 0) {
1048 /* Session daemon is dead so return gracefully. */
1049 ret_code = ret;
1050 goto end_free;
1051 }
1052
1053 /*
1054 * XXX: The consumer data lock is acquired before calling metadata cache
1055 * write which calls push metadata that MUST be protected by the consumer
1056 * lock in order to be able to check the validity of the metadata stream of
1057 * the channel.
1058 *
1059 * Note that this will be subject to change to better fine grained locking
1060 * and ultimately try to get rid of this global consumer data lock.
1061 */
1062 pthread_mutex_lock(&consumer_data.lock);
1063
1064 pthread_mutex_lock(&channel->metadata_cache->lock);
1065 ret = consumer_metadata_cache_write(channel, offset, len, metadata_str);
1066 if (ret < 0) {
1067 /* Unable to handle metadata. Notify session daemon. */
1068 ret_code = LTTCOMM_CONSUMERD_ERROR_METADATA;
1069 /*
1070 * Skip metadata flush on write error since the offset and len might
1071 * not have been updated which could create an infinite loop below when
1072 * waiting for the metadata cache to be flushed.
1073 */
1074 pthread_mutex_unlock(&channel->metadata_cache->lock);
1075 pthread_mutex_unlock(&consumer_data.lock);
1076 goto end_free;
1077 }
1078 pthread_mutex_unlock(&channel->metadata_cache->lock);
1079 pthread_mutex_unlock(&consumer_data.lock);
1080
1081 while (consumer_metadata_cache_flushed(channel, offset + len)) {
1082 DBG("Waiting for metadata to be flushed");
1083 usleep(DEFAULT_METADATA_AVAILABILITY_WAIT_TIME);
1084 }
1085
1086 end_free:
1087 free(metadata_str);
1088 end:
1089 return ret_code;
1090 }
1091
1092 /*
1093 * Receive command from session daemon and process it.
1094 *
1095 * Return 1 on success else a negative value or 0.
1096 */
1097 int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
1098 int sock, struct pollfd *consumer_sockpoll)
1099 {
1100 ssize_t ret;
1101 enum lttng_error_code ret_code = LTTNG_OK;
1102 struct lttcomm_consumer_msg msg;
1103 struct lttng_consumer_channel *channel = NULL;
1104
1105 ret = lttcomm_recv_unix_sock(sock, &msg, sizeof(msg));
1106 if (ret != sizeof(msg)) {
1107 DBG("Consumer received unexpected message size %zd (expects %zu)",
1108 ret, sizeof(msg));
1109 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD);
1110 /*
1111 * The ret value might 0 meaning an orderly shutdown but this is ok
1112 * since the caller handles this.
1113 */
1114 if (ret > 0) {
1115 ret = -1;
1116 }
1117 return ret;
1118 }
1119 if (msg.cmd_type == LTTNG_CONSUMER_STOP) {
1120 /*
1121 * Notify the session daemon that the command is completed.
1122 *
1123 * On transport layer error, the function call will print an error
1124 * message so handling the returned code is a bit useless since we
1125 * return an error code anyway.
1126 */
1127 (void) consumer_send_status_msg(sock, ret_code);
1128 return -ENOENT;
1129 }
1130
1131 /* relayd needs RCU read-side lock */
1132 rcu_read_lock();
1133
1134 switch (msg.cmd_type) {
1135 case LTTNG_CONSUMER_ADD_RELAYD_SOCKET:
1136 {
1137 /* Session daemon status message are handled in the following call. */
1138 ret = consumer_add_relayd_socket(msg.u.relayd_sock.net_index,
1139 msg.u.relayd_sock.type, ctx, sock, consumer_sockpoll,
1140 &msg.u.relayd_sock.sock, msg.u.relayd_sock.session_id);
1141 goto end_nosignal;
1142 }
1143 case LTTNG_CONSUMER_DESTROY_RELAYD:
1144 {
1145 uint64_t index = msg.u.destroy_relayd.net_seq_idx;
1146 struct consumer_relayd_sock_pair *relayd;
1147
1148 DBG("UST consumer destroying relayd %" PRIu64, index);
1149
1150 /* Get relayd reference if exists. */
1151 relayd = consumer_find_relayd(index);
1152 if (relayd == NULL) {
1153 DBG("Unable to find relayd %" PRIu64, index);
1154 ret_code = LTTNG_ERR_NO_CONSUMER;
1155 }
1156
1157 /*
1158 * Each relayd socket pair has a refcount of stream attached to it
1159 * which tells if the relayd is still active or not depending on the
1160 * refcount value.
1161 *
1162 * This will set the destroy flag of the relayd object and destroy it
1163 * if the refcount reaches zero when called.
1164 *
1165 * The destroy can happen either here or when a stream fd hangs up.
1166 */
1167 if (relayd) {
1168 consumer_flag_relayd_for_destroy(relayd);
1169 }
1170
1171 goto end_msg_sessiond;
1172 }
1173 case LTTNG_CONSUMER_UPDATE_STREAM:
1174 {
1175 rcu_read_unlock();
1176 return -ENOSYS;
1177 }
1178 case LTTNG_CONSUMER_DATA_PENDING:
1179 {
1180 int ret, is_data_pending;
1181 uint64_t id = msg.u.data_pending.session_id;
1182
1183 DBG("UST consumer data pending command for id %" PRIu64, id);
1184
1185 is_data_pending = consumer_data_pending(id);
1186
1187 /* Send back returned value to session daemon */
1188 ret = lttcomm_send_unix_sock(sock, &is_data_pending,
1189 sizeof(is_data_pending));
1190 if (ret < 0) {
1191 DBG("Error when sending the data pending ret code: %d", ret);
1192 goto error_fatal;
1193 }
1194
1195 /*
1196 * No need to send back a status message since the data pending
1197 * returned value is the response.
1198 */
1199 break;
1200 }
1201 case LTTNG_CONSUMER_ASK_CHANNEL_CREATION:
1202 {
1203 int ret;
1204 struct ustctl_consumer_channel_attr attr;
1205
1206 /* Create a plain object and reserve a channel key. */
1207 channel = allocate_channel(msg.u.ask_channel.session_id,
1208 msg.u.ask_channel.pathname, msg.u.ask_channel.name,
1209 msg.u.ask_channel.uid, msg.u.ask_channel.gid,
1210 msg.u.ask_channel.relayd_id, msg.u.ask_channel.key,
1211 (enum lttng_event_output) msg.u.ask_channel.output,
1212 msg.u.ask_channel.tracefile_size,
1213 msg.u.ask_channel.tracefile_count,
1214 msg.u.ask_channel.session_id_per_pid,
1215 msg.u.ask_channel.monitor);
1216 if (!channel) {
1217 goto end_channel_error;
1218 }
1219
1220 /* Build channel attributes from received message. */
1221 attr.subbuf_size = msg.u.ask_channel.subbuf_size;
1222 attr.num_subbuf = msg.u.ask_channel.num_subbuf;
1223 attr.overwrite = msg.u.ask_channel.overwrite;
1224 attr.switch_timer_interval = msg.u.ask_channel.switch_timer_interval;
1225 attr.read_timer_interval = msg.u.ask_channel.read_timer_interval;
1226 attr.chan_id = msg.u.ask_channel.chan_id;
1227 attr.output = msg.u.ask_channel.output;
1228 memcpy(attr.uuid, msg.u.ask_channel.uuid, sizeof(attr.uuid));
1229
1230 /* Translate and save channel type. */
1231 switch (msg.u.ask_channel.type) {
1232 case LTTNG_UST_CHAN_PER_CPU:
1233 channel->type = CONSUMER_CHANNEL_TYPE_DATA;
1234 attr.type = LTTNG_UST_CHAN_PER_CPU;
1235 /*
1236 * Set refcount to 1 for owner. Below, we will
1237 * pass ownership to the
1238 * consumer_thread_channel_poll() thread.
1239 */
1240 channel->refcount = 1;
1241 break;
1242 case LTTNG_UST_CHAN_METADATA:
1243 channel->type = CONSUMER_CHANNEL_TYPE_METADATA;
1244 attr.type = LTTNG_UST_CHAN_METADATA;
1245 break;
1246 default:
1247 assert(0);
1248 goto error_fatal;
1249 };
1250
1251 ret = ask_channel(ctx, sock, channel, &attr);
1252 if (ret < 0) {
1253 goto end_channel_error;
1254 }
1255
1256 if (msg.u.ask_channel.type == LTTNG_UST_CHAN_METADATA) {
1257 ret = consumer_metadata_cache_allocate(channel);
1258 if (ret < 0) {
1259 ERR("Allocating metadata cache");
1260 goto end_channel_error;
1261 }
1262 consumer_timer_switch_start(channel, attr.switch_timer_interval);
1263 attr.switch_timer_interval = 0;
1264 }
1265
1266 /*
1267 * Add the channel to the internal state AFTER all streams were created
1268 * and successfully sent to session daemon. This way, all streams must
1269 * be ready before this channel is visible to the threads.
1270 * If add_channel succeeds, ownership of the channel is
1271 * passed to consumer_thread_channel_poll().
1272 */
1273 ret = add_channel(channel, ctx);
1274 if (ret < 0) {
1275 if (msg.u.ask_channel.type == LTTNG_UST_CHAN_METADATA) {
1276 if (channel->switch_timer_enabled == 1) {
1277 consumer_timer_switch_stop(channel);
1278 }
1279 consumer_metadata_cache_destroy(channel);
1280 }
1281 goto end_channel_error;
1282 }
1283
1284 /*
1285 * Channel and streams are now created. Inform the session daemon that
1286 * everything went well and should wait to receive the channel and
1287 * streams with ustctl API.
1288 */
1289 ret = consumer_send_status_channel(sock, channel);
1290 if (ret < 0) {
1291 /*
1292 * There is probably a problem on the socket.
1293 */
1294 goto error_fatal;
1295 }
1296
1297 break;
1298 }
1299 case LTTNG_CONSUMER_GET_CHANNEL:
1300 {
1301 int ret, relayd_err = 0;
1302 uint64_t key = msg.u.get_channel.key;
1303 struct lttng_consumer_channel *channel;
1304
1305 channel = consumer_find_channel(key);
1306 if (!channel) {
1307 ERR("UST consumer get channel key %" PRIu64 " not found", key);
1308 ret_code = LTTNG_ERR_UST_CHAN_NOT_FOUND;
1309 goto end_msg_sessiond;
1310 }
1311
1312 /* Send everything to sessiond. */
1313 ret = send_sessiond_channel(sock, channel, ctx, &relayd_err);
1314 if (ret < 0) {
1315 if (relayd_err) {
1316 /*
1317 * We were unable to send to the relayd the stream so avoid
1318 * sending back a fatal error to the thread since this is OK
1319 * and the consumer can continue its work. The above call
1320 * has sent the error status message to the sessiond.
1321 */
1322 goto end_nosignal;
1323 }
1324 /*
1325 * The communicaton was broken hence there is a bad state between
1326 * the consumer and sessiond so stop everything.
1327 */
1328 goto error_fatal;
1329 }
1330
1331 /*
1332 * In no monitor mode, the streams ownership is kept inside the channel
1333 * so don't send them to the data thread.
1334 */
1335 if (!channel->monitor) {
1336 goto end_msg_sessiond;
1337 }
1338
1339 ret = send_streams_to_thread(channel, ctx);
1340 if (ret < 0) {
1341 /*
1342 * If we are unable to send the stream to the thread, there is
1343 * a big problem so just stop everything.
1344 */
1345 goto error_fatal;
1346 }
1347 /* List MUST be empty after or else it could be reused. */
1348 assert(cds_list_empty(&channel->streams.head));
1349 goto end_msg_sessiond;
1350 }
1351 case LTTNG_CONSUMER_DESTROY_CHANNEL:
1352 {
1353 uint64_t key = msg.u.destroy_channel.key;
1354
1355 /*
1356 * Only called if streams have not been sent to stream
1357 * manager thread. However, channel has been sent to
1358 * channel manager thread.
1359 */
1360 notify_thread_del_channel(ctx, key);
1361 goto end_msg_sessiond;
1362 }
1363 case LTTNG_CONSUMER_CLOSE_METADATA:
1364 {
1365 int ret;
1366
1367 ret = close_metadata(msg.u.close_metadata.key);
1368 if (ret != 0) {
1369 ret_code = ret;
1370 }
1371
1372 goto end_msg_sessiond;
1373 }
1374 case LTTNG_CONSUMER_FLUSH_CHANNEL:
1375 {
1376 int ret;
1377
1378 ret = flush_channel(msg.u.flush_channel.key);
1379 if (ret != 0) {
1380 ret_code = ret;
1381 }
1382
1383 goto end_msg_sessiond;
1384 }
1385 case LTTNG_CONSUMER_PUSH_METADATA:
1386 {
1387 int ret;
1388 uint64_t len = msg.u.push_metadata.len;
1389 uint64_t key = msg.u.push_metadata.key;
1390 uint64_t offset = msg.u.push_metadata.target_offset;
1391 struct lttng_consumer_channel *channel;
1392
1393 DBG("UST consumer push metadata key %" PRIu64 " of len %" PRIu64, key,
1394 len);
1395
1396 channel = consumer_find_channel(key);
1397 if (!channel) {
1398 ERR("UST consumer push metadata %" PRIu64 " not found", key);
1399 ret_code = LTTNG_ERR_UST_CHAN_NOT_FOUND;
1400 goto end_msg_sessiond;
1401 }
1402
1403 /* Tell session daemon we are ready to receive the metadata. */
1404 ret = consumer_send_status_msg(sock, LTTNG_OK);
1405 if (ret < 0) {
1406 /* Somehow, the session daemon is not responding anymore. */
1407 goto error_fatal;
1408 }
1409
1410 /* Wait for more data. */
1411 if (lttng_consumer_poll_socket(consumer_sockpoll) < 0) {
1412 goto error_fatal;
1413 }
1414
1415 ret = lttng_ustconsumer_recv_metadata(sock, key, offset,
1416 len, channel);
1417 if (ret < 0) {
1418 /* error receiving from sessiond */
1419 goto error_fatal;
1420 } else {
1421 ret_code = ret;
1422 goto end_msg_sessiond;
1423 }
1424 }
1425 case LTTNG_CONSUMER_SETUP_METADATA:
1426 {
1427 int ret;
1428
1429 ret = setup_metadata(ctx, msg.u.setup_metadata.key);
1430 if (ret) {
1431 ret_code = ret;
1432 }
1433 goto end_msg_sessiond;
1434 }
1435 case LTTNG_CONSUMER_SNAPSHOT_CHANNEL:
1436 {
1437 if (msg.u.snapshot_channel.metadata) {
1438 ret = snapshot_metadata(msg.u.snapshot_channel.key,
1439 msg.u.snapshot_channel.pathname,
1440 msg.u.snapshot_channel.relayd_id,
1441 ctx);
1442 if (ret < 0) {
1443 ERR("Snapshot metadata failed");
1444 ret_code = LTTNG_ERR_UST_META_FAIL;
1445 }
1446 } else {
1447 ret = snapshot_channel(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 channel failed");
1453 ret_code = LTTNG_ERR_UST_CHAN_FAIL;
1454 }
1455 }
1456
1457 ret = consumer_send_status_msg(sock, ret_code);
1458 if (ret < 0) {
1459 /* Somehow, the session daemon is not responding anymore. */
1460 goto end_nosignal;
1461 }
1462 break;
1463 }
1464 default:
1465 break;
1466 }
1467
1468 end_nosignal:
1469 rcu_read_unlock();
1470
1471 /*
1472 * Return 1 to indicate success since the 0 value can be a socket
1473 * shutdown during the recv() or send() call.
1474 */
1475 return 1;
1476
1477 end_msg_sessiond:
1478 /*
1479 * The returned value here is not useful since either way we'll return 1 to
1480 * the caller because the session daemon socket management is done
1481 * elsewhere. Returning a negative code or 0 will shutdown the consumer.
1482 */
1483 ret = consumer_send_status_msg(sock, ret_code);
1484 if (ret < 0) {
1485 goto error_fatal;
1486 }
1487 rcu_read_unlock();
1488 return 1;
1489 end_channel_error:
1490 if (channel) {
1491 /*
1492 * Free channel here since no one has a reference to it. We don't
1493 * free after that because a stream can store this pointer.
1494 */
1495 destroy_channel(channel);
1496 }
1497 /* We have to send a status channel message indicating an error. */
1498 ret = consumer_send_status_channel(sock, NULL);
1499 if (ret < 0) {
1500 /* Stop everything if session daemon can not be notified. */
1501 goto error_fatal;
1502 }
1503 rcu_read_unlock();
1504 return 1;
1505 error_fatal:
1506 rcu_read_unlock();
1507 /* This will issue a consumer stop. */
1508 return -1;
1509 }
1510
1511 /*
1512 * Wrapper over the mmap() read offset from ust-ctl library. Since this can be
1513 * compiled out, we isolate it in this library.
1514 */
1515 int lttng_ustctl_get_mmap_read_offset(struct lttng_consumer_stream *stream,
1516 unsigned long *off)
1517 {
1518 assert(stream);
1519 assert(stream->ustream);
1520
1521 return ustctl_get_mmap_read_offset(stream->ustream, off);
1522 }
1523
1524 /*
1525 * Wrapper over the mmap() read offset from ust-ctl library. Since this can be
1526 * compiled out, we isolate it in this library.
1527 */
1528 void *lttng_ustctl_get_mmap_base(struct lttng_consumer_stream *stream)
1529 {
1530 assert(stream);
1531 assert(stream->ustream);
1532
1533 return ustctl_get_mmap_base(stream->ustream);
1534 }
1535
1536 /*
1537 * Take a snapshot for a specific fd
1538 *
1539 * Returns 0 on success, < 0 on error
1540 */
1541 int lttng_ustconsumer_take_snapshot(struct lttng_consumer_stream *stream)
1542 {
1543 assert(stream);
1544 assert(stream->ustream);
1545
1546 return ustctl_snapshot(stream->ustream);
1547 }
1548
1549 /*
1550 * Get the produced position
1551 *
1552 * Returns 0 on success, < 0 on error
1553 */
1554 int lttng_ustconsumer_get_produced_snapshot(
1555 struct lttng_consumer_stream *stream, unsigned long *pos)
1556 {
1557 assert(stream);
1558 assert(stream->ustream);
1559 assert(pos);
1560
1561 return ustctl_snapshot_get_produced(stream->ustream, pos);
1562 }
1563
1564 /*
1565 * Get the consumed position
1566 *
1567 * Returns 0 on success, < 0 on error
1568 */
1569 int lttng_ustconsumer_get_consumed_snapshot(
1570 struct lttng_consumer_stream *stream, unsigned long *pos)
1571 {
1572 assert(stream);
1573 assert(stream->ustream);
1574 assert(pos);
1575
1576 return ustctl_snapshot_get_consumed(stream->ustream, pos);
1577 }
1578
1579 /*
1580 * Called when the stream signal the consumer that it has hang up.
1581 */
1582 void lttng_ustconsumer_on_stream_hangup(struct lttng_consumer_stream *stream)
1583 {
1584 assert(stream);
1585 assert(stream->ustream);
1586
1587 ustctl_flush_buffer(stream->ustream, 0);
1588 stream->hangup_flush_done = 1;
1589 }
1590
1591 void lttng_ustconsumer_del_channel(struct lttng_consumer_channel *chan)
1592 {
1593 assert(chan);
1594 assert(chan->uchan);
1595
1596 if (chan->switch_timer_enabled == 1) {
1597 consumer_timer_switch_stop(chan);
1598 }
1599 consumer_metadata_cache_destroy(chan);
1600 ustctl_destroy_channel(chan->uchan);
1601 }
1602
1603 void lttng_ustconsumer_del_stream(struct lttng_consumer_stream *stream)
1604 {
1605 assert(stream);
1606 assert(stream->ustream);
1607
1608 if (stream->chan->switch_timer_enabled == 1) {
1609 consumer_timer_switch_stop(stream->chan);
1610 }
1611 ustctl_destroy_stream(stream->ustream);
1612 }
1613
1614 int lttng_ustconsumer_read_subbuffer(struct lttng_consumer_stream *stream,
1615 struct lttng_consumer_local_data *ctx)
1616 {
1617 unsigned long len, subbuf_size, padding;
1618 int err;
1619 long ret = 0;
1620 char dummy;
1621 struct ustctl_consumer_stream *ustream;
1622
1623 assert(stream);
1624 assert(stream->ustream);
1625 assert(ctx);
1626
1627 DBG2("In UST read_subbuffer (wait_fd: %d, name: %s)", stream->wait_fd,
1628 stream->name);
1629
1630 /* Ease our life for what's next. */
1631 ustream = stream->ustream;
1632
1633 /* We can consume the 1 byte written into the wait_fd by UST */
1634 if (!stream->hangup_flush_done) {
1635 ssize_t readlen;
1636
1637 do {
1638 readlen = read(stream->wait_fd, &dummy, 1);
1639 } while (readlen == -1 && errno == EINTR);
1640 if (readlen == -1) {
1641 ret = readlen;
1642 goto end;
1643 }
1644 }
1645
1646 /* Get the next subbuffer */
1647 err = ustctl_get_next_subbuf(ustream);
1648 if (err != 0) {
1649 ret = err; /* ustctl_get_next_subbuf returns negative, caller expect positive. */
1650 /*
1651 * This is a debug message even for single-threaded consumer,
1652 * because poll() have more relaxed criterions than get subbuf,
1653 * so get_subbuf may fail for short race windows where poll()
1654 * would issue wakeups.
1655 */
1656 DBG("Reserving sub buffer failed (everything is normal, "
1657 "it is due to concurrency) [ret: %d]", err);
1658 goto end;
1659 }
1660 assert(stream->chan->output == CONSUMER_CHANNEL_MMAP);
1661 /* Get the full padded subbuffer size */
1662 err = ustctl_get_padded_subbuf_size(ustream, &len);
1663 assert(err == 0);
1664
1665 /* Get subbuffer data size (without padding) */
1666 err = ustctl_get_subbuf_size(ustream, &subbuf_size);
1667 assert(err == 0);
1668
1669 /* Make sure we don't get a subbuffer size bigger than the padded */
1670 assert(len >= subbuf_size);
1671
1672 padding = len - subbuf_size;
1673 /* write the subbuffer to the tracefile */
1674 ret = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, subbuf_size, padding);
1675 /*
1676 * The mmap operation should write subbuf_size amount of data when network
1677 * streaming or the full padding (len) size when we are _not_ streaming.
1678 */
1679 if ((ret != subbuf_size && stream->net_seq_idx != (uint64_t) -1ULL) ||
1680 (ret != len && stream->net_seq_idx == (uint64_t) -1ULL)) {
1681 /*
1682 * Display the error but continue processing to try to release the
1683 * subbuffer. This is a DBG statement since any unexpected kill or
1684 * signal, the application gets unregistered, relayd gets closed or
1685 * anything that affects the buffer lifetime will trigger this error.
1686 * So, for the sake of the user, don't print this error since it can
1687 * happen and it is OK with the code flow.
1688 */
1689 DBG("Error writing to tracefile "
1690 "(ret: %ld != len: %lu != subbuf_size: %lu)",
1691 ret, len, subbuf_size);
1692 }
1693 err = ustctl_put_next_subbuf(ustream);
1694 assert(err == 0);
1695
1696 end:
1697 return ret;
1698 }
1699
1700 /*
1701 * Called when a stream is created.
1702 *
1703 * Return 0 on success or else a negative value.
1704 */
1705 int lttng_ustconsumer_on_recv_stream(struct lttng_consumer_stream *stream)
1706 {
1707 int ret;
1708
1709 assert(stream);
1710
1711 /* Don't create anything if this is set for streaming. */
1712 if (stream->net_seq_idx == (uint64_t) -1ULL && stream->chan->monitor) {
1713 ret = utils_create_stream_file(stream->chan->pathname, stream->name,
1714 stream->chan->tracefile_size, stream->tracefile_count_current,
1715 stream->uid, stream->gid);
1716 if (ret < 0) {
1717 goto error;
1718 }
1719 stream->out_fd = ret;
1720 stream->tracefile_size_current = 0;
1721 }
1722 ret = 0;
1723
1724 error:
1725 return ret;
1726 }
1727
1728 /*
1729 * Check if data is still being extracted from the buffers for a specific
1730 * stream. Consumer data lock MUST be acquired before calling this function
1731 * and the stream lock.
1732 *
1733 * Return 1 if the traced data are still getting read else 0 meaning that the
1734 * data is available for trace viewer reading.
1735 */
1736 int lttng_ustconsumer_data_pending(struct lttng_consumer_stream *stream)
1737 {
1738 int ret;
1739
1740 assert(stream);
1741 assert(stream->ustream);
1742
1743 DBG("UST consumer checking data pending");
1744
1745 ret = ustctl_get_next_subbuf(stream->ustream);
1746 if (ret == 0) {
1747 /* There is still data so let's put back this subbuffer. */
1748 ret = ustctl_put_subbuf(stream->ustream);
1749 assert(ret == 0);
1750 ret = 1; /* Data is pending */
1751 goto end;
1752 }
1753
1754 /* Data is NOT pending so ready to be read. */
1755 ret = 0;
1756
1757 end:
1758 return ret;
1759 }
1760
1761 /*
1762 * Close every metadata stream wait fd of the metadata hash table. This
1763 * function MUST be used very carefully so not to run into a race between the
1764 * metadata thread handling streams and this function closing their wait fd.
1765 *
1766 * For UST, this is used when the session daemon hangs up. Its the metadata
1767 * producer so calling this is safe because we are assured that no state change
1768 * can occur in the metadata thread for the streams in the hash table.
1769 */
1770 void lttng_ustconsumer_close_metadata(struct lttng_ht *metadata_ht)
1771 {
1772 int ret;
1773 struct lttng_ht_iter iter;
1774 struct lttng_consumer_stream *stream;
1775
1776 assert(metadata_ht);
1777 assert(metadata_ht->ht);
1778
1779 DBG("UST consumer closing all metadata streams");
1780
1781 rcu_read_lock();
1782 cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream,
1783 node.node) {
1784 int fd = stream->wait_fd;
1785
1786 /*
1787 * Whatever happens here we have to continue to try to close every
1788 * streams. Let's report at least the error on failure.
1789 */
1790 ret = ustctl_stream_close_wakeup_fd(stream->ustream);
1791 if (ret) {
1792 ERR("Unable to close metadata stream fd %d ret %d", fd, ret);
1793 }
1794 DBG("Metadata wait fd %d closed", fd);
1795 }
1796 rcu_read_unlock();
1797 }
1798
1799 void lttng_ustconsumer_close_stream_wakeup(struct lttng_consumer_stream *stream)
1800 {
1801 int ret;
1802
1803 ret = ustctl_stream_close_wakeup_fd(stream->ustream);
1804 if (ret < 0) {
1805 ERR("Unable to close wakeup fd");
1806 }
1807 }
1808
1809 int lttng_ustconsumer_request_metadata(struct lttng_consumer_local_data *ctx,
1810 struct lttng_consumer_channel *channel)
1811 {
1812 struct lttcomm_metadata_request_msg request;
1813 struct lttcomm_consumer_msg msg;
1814 enum lttng_error_code ret_code = LTTNG_OK;
1815 uint64_t len, key, offset;
1816 int ret;
1817
1818 assert(channel);
1819 assert(channel->metadata_cache);
1820
1821 /* send the metadata request to sessiond */
1822 switch (consumer_data.type) {
1823 case LTTNG_CONSUMER64_UST:
1824 request.bits_per_long = 64;
1825 break;
1826 case LTTNG_CONSUMER32_UST:
1827 request.bits_per_long = 32;
1828 break;
1829 default:
1830 request.bits_per_long = 0;
1831 break;
1832 }
1833
1834 request.session_id = channel->session_id;
1835 request.session_id_per_pid = channel->session_id_per_pid;
1836 request.uid = channel->uid;
1837 request.key = channel->key;
1838 DBG("Sending metadata request to sessiond, session id %" PRIu64
1839 ", per-pid %" PRIu64,
1840 channel->session_id,
1841 channel->session_id_per_pid);
1842
1843 ret = lttcomm_send_unix_sock(ctx->consumer_metadata_socket, &request,
1844 sizeof(request));
1845 if (ret < 0) {
1846 ERR("Asking metadata to sessiond");
1847 goto end;
1848 }
1849
1850 /* Receive the metadata from sessiond */
1851 ret = lttcomm_recv_unix_sock(ctx->consumer_metadata_socket, &msg,
1852 sizeof(msg));
1853 if (ret != sizeof(msg)) {
1854 DBG("Consumer received unexpected message size %d (expects %zu)",
1855 ret, sizeof(msg));
1856 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD);
1857 /*
1858 * The ret value might 0 meaning an orderly shutdown but this is ok
1859 * since the caller handles this.
1860 */
1861 goto end;
1862 }
1863
1864 if (msg.cmd_type == LTTNG_ERR_UND) {
1865 /* No registry found */
1866 (void) consumer_send_status_msg(ctx->consumer_metadata_socket,
1867 ret_code);
1868 ret = 0;
1869 goto end;
1870 } else if (msg.cmd_type != LTTNG_CONSUMER_PUSH_METADATA) {
1871 ERR("Unexpected cmd_type received %d", msg.cmd_type);
1872 ret = -1;
1873 goto end;
1874 }
1875
1876 len = msg.u.push_metadata.len;
1877 key = msg.u.push_metadata.key;
1878 offset = msg.u.push_metadata.target_offset;
1879
1880 assert(key == channel->key);
1881 if (len == 0) {
1882 DBG("No new metadata to receive for key %" PRIu64, key);
1883 }
1884
1885 /* Tell session daemon we are ready to receive the metadata. */
1886 ret = consumer_send_status_msg(ctx->consumer_metadata_socket,
1887 LTTNG_OK);
1888 if (ret < 0 || len == 0) {
1889 /*
1890 * Somehow, the session daemon is not responding anymore or there is
1891 * nothing to receive.
1892 */
1893 goto end;
1894 }
1895
1896 ret_code = lttng_ustconsumer_recv_metadata(ctx->consumer_metadata_socket,
1897 key, offset, len, channel);
1898 if (ret_code >= 0) {
1899 /*
1900 * Only send the status msg if the sessiond is alive meaning a positive
1901 * ret code.
1902 */
1903 (void) consumer_send_status_msg(ctx->consumer_metadata_socket, ret_code);
1904 }
1905 ret = 0;
1906
1907 end:
1908 return ret;
1909 }
This page took 0.108692 seconds and 4 git commands to generate.