Fix: data pending race
[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 ret = consumer_add_metadata_stream(stream);
198 if (ret) {
199 ERR("Consumer add metadata stream %" PRIu64 " failed.",
200 stream->key);
201 goto error;
202 }
203 stream_pipe = ctx->consumer_metadata_pipe;
204 } else {
205 ret = consumer_add_data_stream(stream);
206 if (ret) {
207 ERR("Consumer add stream %" PRIu64 " failed.",
208 stream->key);
209 goto error;
210 }
211 stream_pipe = ctx->consumer_data_pipe;
212 }
213
214 /*
215 * From this point on, the stream's ownership has been moved away from
216 * the channel and becomes globally visible.
217 */
218 stream->globally_visible = 1;
219
220 ret = lttng_pipe_write(stream_pipe, &stream, sizeof(stream));
221 if (ret < 0) {
222 ERR("Consumer write %s stream to pipe %d",
223 stream->metadata_flag ? "metadata" : "data",
224 lttng_pipe_get_writefd(stream_pipe));
225 if (stream->metadata_flag) {
226 consumer_del_stream_for_metadata(stream);
227 } else {
228 consumer_del_stream_for_data(stream);
229 }
230 }
231 error:
232 return ret;
233 }
234
235 /*
236 * Create streams for the given channel using liblttng-ust-ctl.
237 *
238 * Return 0 on success else a negative value.
239 */
240 static int create_ust_streams(struct lttng_consumer_channel *channel,
241 struct lttng_consumer_local_data *ctx)
242 {
243 int ret, cpu = 0;
244 struct ustctl_consumer_stream *ustream;
245 struct lttng_consumer_stream *stream;
246
247 assert(channel);
248 assert(ctx);
249
250 /*
251 * While a stream is available from ustctl. When NULL is returned, we've
252 * reached the end of the possible stream for the channel.
253 */
254 while ((ustream = ustctl_create_stream(channel->uchan, cpu))) {
255 int wait_fd;
256 int ust_metadata_pipe[2];
257
258 if (channel->type == CONSUMER_CHANNEL_TYPE_METADATA && channel->monitor) {
259 ret = utils_create_pipe_cloexec_nonblock(ust_metadata_pipe);
260 if (ret < 0) {
261 ERR("Create ust metadata poll pipe");
262 goto error;
263 }
264 wait_fd = ust_metadata_pipe[0];
265 } else {
266 wait_fd = ustctl_stream_get_wait_fd(ustream);
267 }
268
269 /* Allocate consumer stream object. */
270 stream = allocate_stream(cpu, wait_fd, channel, ctx, &ret);
271 if (!stream) {
272 goto error_alloc;
273 }
274 stream->ustream = ustream;
275 /*
276 * Store it so we can save multiple function calls afterwards since
277 * this value is used heavily in the stream threads. This is UST
278 * specific so this is why it's done after allocation.
279 */
280 stream->wait_fd = wait_fd;
281
282 /*
283 * Increment channel refcount since the channel reference has now been
284 * assigned in the allocation process above.
285 */
286 if (stream->chan->monitor) {
287 uatomic_inc(&stream->chan->refcount);
288 }
289
290 /*
291 * Order is important this is why a list is used. On error, the caller
292 * should clean this list.
293 */
294 cds_list_add_tail(&stream->send_node, &channel->streams.head);
295
296 ret = ustctl_get_max_subbuf_size(stream->ustream,
297 &stream->max_sb_size);
298 if (ret < 0) {
299 ERR("ustctl_get_max_subbuf_size failed for stream %s",
300 stream->name);
301 goto error;
302 }
303
304 /* Do actions once stream has been received. */
305 if (ctx->on_recv_stream) {
306 ret = ctx->on_recv_stream(stream);
307 if (ret < 0) {
308 goto error;
309 }
310 }
311
312 DBG("UST consumer add stream %s (key: %" PRIu64 ") with relayd id %" PRIu64,
313 stream->name, stream->key, stream->relayd_stream_id);
314
315 /* Set next CPU stream. */
316 channel->streams.count = ++cpu;
317
318 /* Keep stream reference when creating metadata. */
319 if (channel->type == CONSUMER_CHANNEL_TYPE_METADATA) {
320 channel->metadata_stream = stream;
321 stream->ust_metadata_poll_pipe[0] = ust_metadata_pipe[0];
322 stream->ust_metadata_poll_pipe[1] = ust_metadata_pipe[1];
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 DBG("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 if (channel->relayd_id != (uint64_t) -1ULL) {
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 = consumer_send_relayd_stream(stream, stream->chan->pathname);
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
428 /* Inform sessiond that we are about to send channel and streams. */
429 ret = consumer_send_status_msg(sock, ret_code);
430 if (ret < 0 || ret_code != LTTNG_OK) {
431 /*
432 * Either the session daemon is not responding or the relayd died so we
433 * stop now.
434 */
435 goto error;
436 }
437
438 /* Send channel to sessiond. */
439 ret = ustctl_send_channel_to_sessiond(sock, channel->uchan);
440 if (ret < 0) {
441 goto error;
442 }
443
444 ret = ustctl_channel_close_wakeup_fd(channel->uchan);
445 if (ret < 0) {
446 goto error;
447 }
448
449 /* The channel was sent successfully to the sessiond at this point. */
450 cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
451 /* Send stream to session daemon. */
452 ret = send_sessiond_stream(sock, stream);
453 if (ret < 0) {
454 goto error;
455 }
456 }
457
458 /* Tell sessiond there is no more stream. */
459 ret = ustctl_send_stream_to_sessiond(sock, NULL);
460 if (ret < 0) {
461 goto error;
462 }
463
464 DBG("UST consumer NULL stream sent to sessiond");
465
466 return 0;
467
468 error:
469 if (ret_code != LTTNG_OK) {
470 ret = -1;
471 }
472 return ret;
473 }
474
475 /*
476 * Creates a channel and streams and add the channel it to the channel internal
477 * state. The created stream must ONLY be sent once the GET_CHANNEL command is
478 * received.
479 *
480 * Return 0 on success or else, a negative value is returned and the channel
481 * MUST be destroyed by consumer_del_channel().
482 */
483 static int ask_channel(struct lttng_consumer_local_data *ctx, int sock,
484 struct lttng_consumer_channel *channel,
485 struct ustctl_consumer_channel_attr *attr)
486 {
487 int ret;
488
489 assert(ctx);
490 assert(channel);
491 assert(attr);
492
493 /*
494 * This value is still used by the kernel consumer since for the kernel,
495 * the stream ownership is not IN the consumer so we need to have the
496 * number of left stream that needs to be initialized so we can know when
497 * to delete the channel (see consumer.c).
498 *
499 * As for the user space tracer now, the consumer creates and sends the
500 * stream to the session daemon which only sends them to the application
501 * once every stream of a channel is received making this value useless
502 * because we they will be added to the poll thread before the application
503 * receives them. This ensures that a stream can not hang up during
504 * initilization of a channel.
505 */
506 channel->nb_init_stream_left = 0;
507
508 /* The reply msg status is handled in the following call. */
509 ret = create_ust_channel(attr, &channel->uchan);
510 if (ret < 0) {
511 goto end;
512 }
513
514 channel->wait_fd = ustctl_channel_get_wait_fd(channel->uchan);
515
516 /*
517 * For the snapshots (no monitor), we create the metadata streams
518 * on demand, not during the channel creation.
519 */
520 if (channel->type == CONSUMER_CHANNEL_TYPE_METADATA && !channel->monitor) {
521 ret = 0;
522 goto end;
523 }
524
525 /* Open all streams for this channel. */
526 ret = create_ust_streams(channel, ctx);
527 if (ret < 0) {
528 goto end;
529 }
530
531 end:
532 return ret;
533 }
534
535 /*
536 * Send all stream of a channel to the right thread handling it.
537 *
538 * On error, return a negative value else 0 on success.
539 */
540 static int send_streams_to_thread(struct lttng_consumer_channel *channel,
541 struct lttng_consumer_local_data *ctx)
542 {
543 int ret = 0;
544 struct lttng_consumer_stream *stream, *stmp;
545
546 assert(channel);
547 assert(ctx);
548
549 /* Send streams to the corresponding thread. */
550 cds_list_for_each_entry_safe(stream, stmp, &channel->streams.head,
551 send_node) {
552 /* Sending the stream to the thread. */
553 ret = send_stream_to_thread(stream, ctx);
554 if (ret < 0) {
555 /*
556 * If we are unable to send the stream to the thread, there is
557 * a big problem so just stop everything.
558 */
559 /* Remove node from the channel stream list. */
560 cds_list_del(&stream->send_node);
561 goto error;
562 }
563
564 /* Remove node from the channel stream list. */
565 cds_list_del(&stream->send_node);
566
567 }
568
569 error:
570 return ret;
571 }
572
573 /*
574 * Flush channel's streams using the given key to retrieve the channel.
575 *
576 * Return 0 on success else an LTTng error code.
577 */
578 static int flush_channel(uint64_t chan_key)
579 {
580 int ret = 0;
581 struct lttng_consumer_channel *channel;
582 struct lttng_consumer_stream *stream;
583 struct lttng_ht *ht;
584 struct lttng_ht_iter iter;
585
586 DBG("UST consumer flush channel key %" PRIu64, chan_key);
587
588 rcu_read_lock();
589 channel = consumer_find_channel(chan_key);
590 if (!channel) {
591 ERR("UST consumer flush channel %" PRIu64 " not found", chan_key);
592 ret = LTTNG_ERR_UST_CHAN_NOT_FOUND;
593 goto error;
594 }
595
596 ht = consumer_data.stream_per_chan_id_ht;
597
598 /* For each stream of the channel id, flush it. */
599 cds_lfht_for_each_entry_duplicate(ht->ht,
600 ht->hash_fct(&channel->key, lttng_ht_seed), ht->match_fct,
601 &channel->key, &iter.iter, stream, node_channel_id.node) {
602 ustctl_flush_buffer(stream->ustream, 1);
603 }
604 error:
605 rcu_read_unlock();
606 return ret;
607 }
608
609 /*
610 * Close metadata stream wakeup_fd using the given key to retrieve the channel.
611 * RCU read side lock MUST be acquired before calling this function.
612 *
613 * Return 0 on success else an LTTng error code.
614 */
615 static int close_metadata(uint64_t chan_key)
616 {
617 int ret = 0;
618 struct lttng_consumer_channel *channel;
619
620 DBG("UST consumer close metadata key %" PRIu64, chan_key);
621
622 channel = consumer_find_channel(chan_key);
623 if (!channel) {
624 /*
625 * This is possible if the metadata thread has issue a delete because
626 * the endpoint point of the stream hung up. There is no way the
627 * session daemon can know about it thus use a DBG instead of an actual
628 * error.
629 */
630 DBG("UST consumer close metadata %" PRIu64 " not found", chan_key);
631 ret = LTTNG_ERR_UST_CHAN_NOT_FOUND;
632 goto error;
633 }
634
635 pthread_mutex_lock(&consumer_data.lock);
636 pthread_mutex_lock(&channel->lock);
637
638 if (cds_lfht_is_node_deleted(&channel->node.node)) {
639 goto error_unlock;
640 }
641
642 if (channel->switch_timer_enabled == 1) {
643 DBG("Deleting timer on metadata channel");
644 consumer_timer_switch_stop(channel);
645 }
646
647 if (channel->metadata_stream) {
648 ret = ustctl_stream_close_wakeup_fd(channel->metadata_stream->ustream);
649 if (ret < 0) {
650 ERR("UST consumer unable to close fd of metadata (ret: %d)", ret);
651 ret = LTTCOMM_CONSUMERD_ERROR_METADATA;
652 goto error_unlock;
653 }
654 if (channel->monitor) {
655 /* close the read-side in consumer_del_metadata_stream */
656 ret = close(channel->metadata_stream->ust_metadata_poll_pipe[1]);
657 if (ret < 0) {
658 PERROR("Close UST metadata write-side poll pipe");
659 }
660 }
661 }
662
663 error_unlock:
664 pthread_mutex_unlock(&channel->lock);
665 pthread_mutex_unlock(&consumer_data.lock);
666 error:
667 return ret;
668 }
669
670 /*
671 * RCU read side lock MUST be acquired before calling this function.
672 *
673 * Return 0 on success else an LTTng error code.
674 */
675 static int setup_metadata(struct lttng_consumer_local_data *ctx, uint64_t key)
676 {
677 int ret;
678 struct lttng_consumer_channel *metadata;
679
680 DBG("UST consumer setup metadata key %" PRIu64, key);
681
682 metadata = consumer_find_channel(key);
683 if (!metadata) {
684 ERR("UST consumer push metadata %" PRIu64 " not found", key);
685 ret = LTTNG_ERR_UST_CHAN_NOT_FOUND;
686 goto end;
687 }
688
689 /*
690 * In no monitor mode, the metadata channel has no stream(s) so skip the
691 * ownership transfer to the metadata thread.
692 */
693 if (!metadata->monitor) {
694 DBG("Metadata channel in no monitor");
695 ret = 0;
696 goto end;
697 }
698
699 /*
700 * Send metadata stream to relayd if one available. Availability is
701 * known if the stream is still in the list of the channel.
702 */
703 if (cds_list_empty(&metadata->streams.head)) {
704 ERR("Metadata channel key %" PRIu64 ", no stream available.", key);
705 ret = LTTCOMM_CONSUMERD_ERROR_METADATA;
706 goto error_no_stream;
707 }
708
709 /* Send metadata stream to relayd if needed. */
710 if (metadata->metadata_stream->net_seq_idx != (uint64_t) -1ULL) {
711 ret = consumer_send_relayd_stream(metadata->metadata_stream,
712 metadata->pathname);
713 if (ret < 0) {
714 ret = LTTCOMM_CONSUMERD_ERROR_METADATA;
715 goto error;
716 }
717 }
718
719 ret = send_streams_to_thread(metadata, ctx);
720 if (ret < 0) {
721 /*
722 * If we are unable to send the stream to the thread, there is
723 * a big problem so just stop everything.
724 */
725 ret = LTTCOMM_CONSUMERD_FATAL;
726 goto error;
727 }
728 /* List MUST be empty after or else it could be reused. */
729 assert(cds_list_empty(&metadata->streams.head));
730
731 ret = 0;
732 goto end;
733
734 error:
735 /*
736 * Delete metadata channel on error. At this point, the metadata stream can
737 * NOT be monitored by the metadata thread thus having the guarantee that
738 * the stream is still in the local stream list of the channel. This call
739 * will make sure to clean that list.
740 */
741 cds_list_del(&metadata->metadata_stream->send_node);
742 consumer_stream_destroy(metadata->metadata_stream, NULL);
743 error_no_stream:
744 end:
745 return ret;
746 }
747
748 /*
749 * Snapshot the whole metadata.
750 *
751 * Returns 0 on success, < 0 on error
752 */
753 static int snapshot_metadata(uint64_t key, char *path, uint64_t relayd_id,
754 struct lttng_consumer_local_data *ctx)
755 {
756 int ret = 0;
757 struct lttng_consumer_channel *metadata_channel;
758 struct lttng_consumer_stream *metadata_stream;
759
760 assert(path);
761 assert(ctx);
762
763 DBG("UST consumer snapshot metadata with key %" PRIu64 " at path %s",
764 key, path);
765
766 rcu_read_lock();
767
768 metadata_channel = consumer_find_channel(key);
769 if (!metadata_channel) {
770 ERR("UST snapshot metadata channel not found for key %lu", key);
771 ret = -1;
772 goto error;
773 }
774 assert(!metadata_channel->monitor);
775
776 /*
777 * Ask the sessiond if we have new metadata waiting and update the
778 * consumer metadata cache.
779 */
780 ret = lttng_ustconsumer_request_metadata(ctx, metadata_channel, 0);
781 if (ret < 0) {
782 goto error;
783 }
784
785 /*
786 * The metadata stream is NOT created in no monitor mode when the channel
787 * is created on a sessiond ask channel command.
788 */
789 ret = create_ust_streams(metadata_channel, ctx);
790 if (ret < 0) {
791 goto error;
792 }
793
794 metadata_stream = metadata_channel->metadata_stream;
795 assert(metadata_stream);
796
797 if (relayd_id != (uint64_t) -1ULL) {
798 metadata_stream->net_seq_idx = relayd_id;
799 ret = consumer_send_relayd_stream(metadata_stream, path);
800 if (ret < 0) {
801 goto error_stream;
802 }
803 } else {
804 ret = utils_create_stream_file(path, metadata_stream->name,
805 metadata_stream->chan->tracefile_size,
806 metadata_stream->tracefile_count_current,
807 metadata_stream->uid, metadata_stream->gid);
808 if (ret < 0) {
809 goto error_stream;
810 }
811 metadata_stream->out_fd = ret;
812 metadata_stream->tracefile_size_current = 0;
813 }
814
815 pthread_mutex_lock(&metadata_channel->metadata_cache->lock);
816
817 do {
818 ret = lttng_consumer_read_subbuffer(metadata_stream, ctx);
819 if (ret < 0) {
820 goto error_unlock;
821 }
822 } while (ret > 0);
823
824 error_unlock:
825 pthread_mutex_unlock(&metadata_channel->metadata_cache->lock);
826
827 error_stream:
828 /*
829 * Clean up the stream completly because the next snapshot will use a new
830 * metadata stream.
831 */
832 cds_list_del(&metadata_stream->send_node);
833 consumer_stream_destroy(metadata_stream, NULL);
834 metadata_channel->metadata_stream = NULL;
835
836 error:
837 rcu_read_unlock();
838 return ret;
839 }
840
841 /*
842 * Take a snapshot of all the stream of a channel.
843 *
844 * Returns 0 on success, < 0 on error
845 */
846 static int snapshot_channel(uint64_t key, char *path, uint64_t relayd_id,
847 uint64_t max_stream_size, struct lttng_consumer_local_data *ctx)
848 {
849 int ret;
850 unsigned use_relayd = 0;
851 unsigned long consumed_pos, produced_pos;
852 struct lttng_consumer_channel *channel;
853 struct lttng_consumer_stream *stream;
854
855 assert(path);
856 assert(ctx);
857
858 rcu_read_lock();
859
860 if (relayd_id != (uint64_t) -1ULL) {
861 use_relayd = 1;
862 }
863
864 channel = consumer_find_channel(key);
865 if (!channel) {
866 ERR("UST snapshot channel not found for key %lu", key);
867 ret = -1;
868 goto error;
869 }
870 assert(!channel->monitor);
871 DBG("UST consumer snapshot channel %lu", key);
872
873 cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
874 /* Lock stream because we are about to change its state. */
875 pthread_mutex_lock(&stream->lock);
876 stream->net_seq_idx = relayd_id;
877
878 if (use_relayd) {
879 ret = consumer_send_relayd_stream(stream, path);
880 if (ret < 0) {
881 goto error_unlock;
882 }
883 } else {
884 ret = utils_create_stream_file(path, stream->name,
885 stream->chan->tracefile_size,
886 stream->tracefile_count_current,
887 stream->uid, stream->gid);
888 if (ret < 0) {
889 goto error_unlock;
890 }
891 stream->out_fd = ret;
892 stream->tracefile_size_current = 0;
893
894 DBG("UST consumer snapshot stream %s/%s (%" PRIu64 ")", path,
895 stream->name, stream->key);
896 }
897
898 ustctl_flush_buffer(stream->ustream, 1);
899
900 ret = lttng_ustconsumer_take_snapshot(stream);
901 if (ret < 0) {
902 ERR("Taking UST snapshot");
903 goto error_unlock;
904 }
905
906 ret = lttng_ustconsumer_get_produced_snapshot(stream, &produced_pos);
907 if (ret < 0) {
908 ERR("Produced UST snapshot position");
909 goto error_unlock;
910 }
911
912 ret = lttng_ustconsumer_get_consumed_snapshot(stream, &consumed_pos);
913 if (ret < 0) {
914 ERR("Consumerd UST snapshot position");
915 goto error_unlock;
916 }
917
918 /*
919 * The original value is sent back if max stream size is larger than
920 * the possible size of the snapshot. Also, we asume that the session
921 * daemon should never send a maximum stream size that is lower than
922 * subbuffer size.
923 */
924 consumed_pos = consumer_get_consumed_maxsize(consumed_pos,
925 produced_pos, max_stream_size);
926
927 while (consumed_pos < produced_pos) {
928 ssize_t read_len;
929 unsigned long len, padded_len;
930
931 DBG("UST consumer taking snapshot at pos %lu", consumed_pos);
932
933 ret = ustctl_get_subbuf(stream->ustream, &consumed_pos);
934 if (ret < 0) {
935 if (ret != -EAGAIN) {
936 PERROR("ustctl_get_subbuf snapshot");
937 goto error_close_stream;
938 }
939 DBG("UST consumer get subbuf failed. Skipping it.");
940 consumed_pos += stream->max_sb_size;
941 continue;
942 }
943
944 ret = ustctl_get_subbuf_size(stream->ustream, &len);
945 if (ret < 0) {
946 ERR("Snapshot ustctl_get_subbuf_size");
947 goto error_put_subbuf;
948 }
949
950 ret = ustctl_get_padded_subbuf_size(stream->ustream, &padded_len);
951 if (ret < 0) {
952 ERR("Snapshot ustctl_get_padded_subbuf_size");
953 goto error_put_subbuf;
954 }
955
956 read_len = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, len,
957 padded_len - len);
958 if (use_relayd) {
959 if (read_len != len) {
960 ret = -1;
961 goto error_put_subbuf;
962 }
963 } else {
964 if (read_len != padded_len) {
965 ret = -1;
966 goto error_put_subbuf;
967 }
968 }
969
970 ret = ustctl_put_subbuf(stream->ustream);
971 if (ret < 0) {
972 ERR("Snapshot ustctl_put_subbuf");
973 goto error_close_stream;
974 }
975 consumed_pos += stream->max_sb_size;
976 }
977
978 /* Simply close the stream so we can use it on the next snapshot. */
979 consumer_stream_close(stream);
980 pthread_mutex_unlock(&stream->lock);
981 }
982
983 rcu_read_unlock();
984 return 0;
985
986 error_put_subbuf:
987 if (ustctl_put_subbuf(stream->ustream) < 0) {
988 ERR("Snapshot ustctl_put_subbuf");
989 }
990 error_close_stream:
991 consumer_stream_close(stream);
992 error_unlock:
993 pthread_mutex_unlock(&stream->lock);
994 error:
995 rcu_read_unlock();
996 return ret;
997 }
998
999 /*
1000 * Receive the metadata updates from the sessiond.
1001 */
1002 int lttng_ustconsumer_recv_metadata(int sock, uint64_t key, uint64_t offset,
1003 uint64_t len, struct lttng_consumer_channel *channel,
1004 int timer)
1005 {
1006 int ret, ret_code = LTTNG_OK;
1007 char *metadata_str;
1008
1009 DBG("UST consumer push metadata key %" PRIu64 " of len %" PRIu64, key, len);
1010
1011 metadata_str = zmalloc(len * sizeof(char));
1012 if (!metadata_str) {
1013 PERROR("zmalloc metadata string");
1014 ret_code = LTTCOMM_CONSUMERD_ENOMEM;
1015 goto end;
1016 }
1017
1018 /* Receive metadata string. */
1019 ret = lttcomm_recv_unix_sock(sock, metadata_str, len);
1020 if (ret < 0) {
1021 /* Session daemon is dead so return gracefully. */
1022 ret_code = ret;
1023 goto end_free;
1024 }
1025
1026 pthread_mutex_lock(&channel->metadata_cache->lock);
1027 ret = consumer_metadata_cache_write(channel, offset, len, metadata_str);
1028 if (ret < 0) {
1029 /* Unable to handle metadata. Notify session daemon. */
1030 ret_code = LTTCOMM_CONSUMERD_ERROR_METADATA;
1031 /*
1032 * Skip metadata flush on write error since the offset and len might
1033 * not have been updated which could create an infinite loop below when
1034 * waiting for the metadata cache to be flushed.
1035 */
1036 pthread_mutex_unlock(&channel->metadata_cache->lock);
1037 goto end_free;
1038 }
1039 pthread_mutex_unlock(&channel->metadata_cache->lock);
1040
1041 while (consumer_metadata_cache_flushed(channel, offset + len, timer)) {
1042 DBG("Waiting for metadata to be flushed");
1043 usleep(DEFAULT_METADATA_AVAILABILITY_WAIT_TIME);
1044 }
1045
1046 end_free:
1047 free(metadata_str);
1048 end:
1049 return ret_code;
1050 }
1051
1052 /*
1053 * Receive command from session daemon and process it.
1054 *
1055 * Return 1 on success else a negative value or 0.
1056 */
1057 int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
1058 int sock, struct pollfd *consumer_sockpoll)
1059 {
1060 ssize_t ret;
1061 enum lttng_error_code ret_code = LTTNG_OK;
1062 struct lttcomm_consumer_msg msg;
1063 struct lttng_consumer_channel *channel = NULL;
1064
1065 ret = lttcomm_recv_unix_sock(sock, &msg, sizeof(msg));
1066 if (ret != sizeof(msg)) {
1067 DBG("Consumer received unexpected message size %zd (expects %zu)",
1068 ret, sizeof(msg));
1069 /*
1070 * The ret value might 0 meaning an orderly shutdown but this is ok
1071 * since the caller handles this.
1072 */
1073 if (ret > 0) {
1074 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD);
1075 ret = -1;
1076 }
1077 return ret;
1078 }
1079 if (msg.cmd_type == LTTNG_CONSUMER_STOP) {
1080 /*
1081 * Notify the session daemon that the command is completed.
1082 *
1083 * On transport layer error, the function call will print an error
1084 * message so handling the returned code is a bit useless since we
1085 * return an error code anyway.
1086 */
1087 (void) consumer_send_status_msg(sock, ret_code);
1088 return -ENOENT;
1089 }
1090
1091 /* relayd needs RCU read-side lock */
1092 rcu_read_lock();
1093
1094 switch (msg.cmd_type) {
1095 case LTTNG_CONSUMER_ADD_RELAYD_SOCKET:
1096 {
1097 /* Session daemon status message are handled in the following call. */
1098 ret = consumer_add_relayd_socket(msg.u.relayd_sock.net_index,
1099 msg.u.relayd_sock.type, ctx, sock, consumer_sockpoll,
1100 &msg.u.relayd_sock.sock, msg.u.relayd_sock.session_id);
1101 goto end_nosignal;
1102 }
1103 case LTTNG_CONSUMER_DESTROY_RELAYD:
1104 {
1105 uint64_t index = msg.u.destroy_relayd.net_seq_idx;
1106 struct consumer_relayd_sock_pair *relayd;
1107
1108 DBG("UST consumer destroying relayd %" PRIu64, index);
1109
1110 /* Get relayd reference if exists. */
1111 relayd = consumer_find_relayd(index);
1112 if (relayd == NULL) {
1113 DBG("Unable to find relayd %" PRIu64, index);
1114 ret_code = LTTNG_ERR_NO_CONSUMER;
1115 }
1116
1117 /*
1118 * Each relayd socket pair has a refcount of stream attached to it
1119 * which tells if the relayd is still active or not depending on the
1120 * refcount value.
1121 *
1122 * This will set the destroy flag of the relayd object and destroy it
1123 * if the refcount reaches zero when called.
1124 *
1125 * The destroy can happen either here or when a stream fd hangs up.
1126 */
1127 if (relayd) {
1128 consumer_flag_relayd_for_destroy(relayd);
1129 }
1130
1131 goto end_msg_sessiond;
1132 }
1133 case LTTNG_CONSUMER_UPDATE_STREAM:
1134 {
1135 rcu_read_unlock();
1136 return -ENOSYS;
1137 }
1138 case LTTNG_CONSUMER_DATA_PENDING:
1139 {
1140 int ret, is_data_pending;
1141 uint64_t id = msg.u.data_pending.session_id;
1142
1143 DBG("UST consumer data pending command for id %" PRIu64, id);
1144
1145 is_data_pending = consumer_data_pending(id);
1146
1147 /* Send back returned value to session daemon */
1148 ret = lttcomm_send_unix_sock(sock, &is_data_pending,
1149 sizeof(is_data_pending));
1150 if (ret < 0) {
1151 DBG("Error when sending the data pending ret code: %d", ret);
1152 goto error_fatal;
1153 }
1154
1155 /*
1156 * No need to send back a status message since the data pending
1157 * returned value is the response.
1158 */
1159 break;
1160 }
1161 case LTTNG_CONSUMER_ASK_CHANNEL_CREATION:
1162 {
1163 int ret;
1164 struct ustctl_consumer_channel_attr attr;
1165
1166 /* Create a plain object and reserve a channel key. */
1167 channel = allocate_channel(msg.u.ask_channel.session_id,
1168 msg.u.ask_channel.pathname, msg.u.ask_channel.name,
1169 msg.u.ask_channel.uid, msg.u.ask_channel.gid,
1170 msg.u.ask_channel.relayd_id, msg.u.ask_channel.key,
1171 (enum lttng_event_output) msg.u.ask_channel.output,
1172 msg.u.ask_channel.tracefile_size,
1173 msg.u.ask_channel.tracefile_count,
1174 msg.u.ask_channel.session_id_per_pid,
1175 msg.u.ask_channel.monitor);
1176 if (!channel) {
1177 goto end_channel_error;
1178 }
1179
1180 /* Build channel attributes from received message. */
1181 attr.subbuf_size = msg.u.ask_channel.subbuf_size;
1182 attr.num_subbuf = msg.u.ask_channel.num_subbuf;
1183 attr.overwrite = msg.u.ask_channel.overwrite;
1184 attr.switch_timer_interval = msg.u.ask_channel.switch_timer_interval;
1185 attr.read_timer_interval = msg.u.ask_channel.read_timer_interval;
1186 attr.chan_id = msg.u.ask_channel.chan_id;
1187 attr.output = msg.u.ask_channel.output;
1188 memcpy(attr.uuid, msg.u.ask_channel.uuid, sizeof(attr.uuid));
1189
1190 /* Translate and save channel type. */
1191 switch (msg.u.ask_channel.type) {
1192 case LTTNG_UST_CHAN_PER_CPU:
1193 channel->type = CONSUMER_CHANNEL_TYPE_DATA;
1194 attr.type = LTTNG_UST_CHAN_PER_CPU;
1195 /*
1196 * Set refcount to 1 for owner. Below, we will
1197 * pass ownership to the
1198 * consumer_thread_channel_poll() thread.
1199 */
1200 channel->refcount = 1;
1201 break;
1202 case LTTNG_UST_CHAN_METADATA:
1203 channel->type = CONSUMER_CHANNEL_TYPE_METADATA;
1204 attr.type = LTTNG_UST_CHAN_METADATA;
1205 break;
1206 default:
1207 assert(0);
1208 goto error_fatal;
1209 };
1210
1211 ret = ask_channel(ctx, sock, channel, &attr);
1212 if (ret < 0) {
1213 goto end_channel_error;
1214 }
1215
1216 if (msg.u.ask_channel.type == LTTNG_UST_CHAN_METADATA) {
1217 ret = consumer_metadata_cache_allocate(channel);
1218 if (ret < 0) {
1219 ERR("Allocating metadata cache");
1220 goto end_channel_error;
1221 }
1222 consumer_timer_switch_start(channel, attr.switch_timer_interval);
1223 attr.switch_timer_interval = 0;
1224 }
1225
1226 /*
1227 * Add the channel to the internal state AFTER all streams were created
1228 * and successfully sent to session daemon. This way, all streams must
1229 * be ready before this channel is visible to the threads.
1230 * If add_channel succeeds, ownership of the channel is
1231 * passed to consumer_thread_channel_poll().
1232 */
1233 ret = add_channel(channel, ctx);
1234 if (ret < 0) {
1235 if (msg.u.ask_channel.type == LTTNG_UST_CHAN_METADATA) {
1236 if (channel->switch_timer_enabled == 1) {
1237 consumer_timer_switch_stop(channel);
1238 }
1239 consumer_metadata_cache_destroy(channel);
1240 }
1241 goto end_channel_error;
1242 }
1243
1244 /*
1245 * Channel and streams are now created. Inform the session daemon that
1246 * everything went well and should wait to receive the channel and
1247 * streams with ustctl API.
1248 */
1249 ret = consumer_send_status_channel(sock, channel);
1250 if (ret < 0) {
1251 /*
1252 * There is probably a problem on the socket.
1253 */
1254 goto error_fatal;
1255 }
1256
1257 break;
1258 }
1259 case LTTNG_CONSUMER_GET_CHANNEL:
1260 {
1261 int ret, relayd_err = 0;
1262 uint64_t key = msg.u.get_channel.key;
1263 struct lttng_consumer_channel *channel;
1264
1265 channel = consumer_find_channel(key);
1266 if (!channel) {
1267 ERR("UST consumer get channel key %" PRIu64 " not found", key);
1268 ret_code = LTTNG_ERR_UST_CHAN_NOT_FOUND;
1269 goto end_msg_sessiond;
1270 }
1271
1272 /* Send everything to sessiond. */
1273 ret = send_sessiond_channel(sock, channel, ctx, &relayd_err);
1274 if (ret < 0) {
1275 if (relayd_err) {
1276 /*
1277 * We were unable to send to the relayd the stream so avoid
1278 * sending back a fatal error to the thread since this is OK
1279 * and the consumer can continue its work. The above call
1280 * has sent the error status message to the sessiond.
1281 */
1282 goto end_nosignal;
1283 }
1284 /*
1285 * The communicaton was broken hence there is a bad state between
1286 * the consumer and sessiond so stop everything.
1287 */
1288 goto error_fatal;
1289 }
1290
1291 /*
1292 * In no monitor mode, the streams ownership is kept inside the channel
1293 * so don't send them to the data thread.
1294 */
1295 if (!channel->monitor) {
1296 goto end_msg_sessiond;
1297 }
1298
1299 ret = send_streams_to_thread(channel, ctx);
1300 if (ret < 0) {
1301 /*
1302 * If we are unable to send the stream to the thread, there is
1303 * a big problem so just stop everything.
1304 */
1305 goto error_fatal;
1306 }
1307 /* List MUST be empty after or else it could be reused. */
1308 assert(cds_list_empty(&channel->streams.head));
1309 goto end_msg_sessiond;
1310 }
1311 case LTTNG_CONSUMER_DESTROY_CHANNEL:
1312 {
1313 uint64_t key = msg.u.destroy_channel.key;
1314
1315 /*
1316 * Only called if streams have not been sent to stream
1317 * manager thread. However, channel has been sent to
1318 * channel manager thread.
1319 */
1320 notify_thread_del_channel(ctx, key);
1321 goto end_msg_sessiond;
1322 }
1323 case LTTNG_CONSUMER_CLOSE_METADATA:
1324 {
1325 int ret;
1326
1327 ret = close_metadata(msg.u.close_metadata.key);
1328 if (ret != 0) {
1329 ret_code = ret;
1330 }
1331
1332 goto end_msg_sessiond;
1333 }
1334 case LTTNG_CONSUMER_FLUSH_CHANNEL:
1335 {
1336 int ret;
1337
1338 ret = flush_channel(msg.u.flush_channel.key);
1339 if (ret != 0) {
1340 ret_code = ret;
1341 }
1342
1343 goto end_msg_sessiond;
1344 }
1345 case LTTNG_CONSUMER_PUSH_METADATA:
1346 {
1347 int ret;
1348 uint64_t len = msg.u.push_metadata.len;
1349 uint64_t key = msg.u.push_metadata.key;
1350 uint64_t offset = msg.u.push_metadata.target_offset;
1351 struct lttng_consumer_channel *channel;
1352
1353 DBG("UST consumer push metadata key %" PRIu64 " of len %" PRIu64, key,
1354 len);
1355
1356 channel = consumer_find_channel(key);
1357 if (!channel) {
1358 ERR("UST consumer push metadata %" PRIu64 " not found", key);
1359 ret_code = LTTNG_ERR_UST_CHAN_NOT_FOUND;
1360 goto end_msg_sessiond;
1361 }
1362
1363 /* Tell session daemon we are ready to receive the metadata. */
1364 ret = consumer_send_status_msg(sock, LTTNG_OK);
1365 if (ret < 0) {
1366 /* Somehow, the session daemon is not responding anymore. */
1367 goto error_fatal;
1368 }
1369
1370 /* Wait for more data. */
1371 if (lttng_consumer_poll_socket(consumer_sockpoll) < 0) {
1372 goto error_fatal;
1373 }
1374
1375 ret = lttng_ustconsumer_recv_metadata(sock, key, offset,
1376 len, channel, 0);
1377 if (ret < 0) {
1378 /* error receiving from sessiond */
1379 goto error_fatal;
1380 } else {
1381 ret_code = ret;
1382 goto end_msg_sessiond;
1383 }
1384 }
1385 case LTTNG_CONSUMER_SETUP_METADATA:
1386 {
1387 int ret;
1388
1389 ret = setup_metadata(ctx, msg.u.setup_metadata.key);
1390 if (ret) {
1391 ret_code = ret;
1392 }
1393 goto end_msg_sessiond;
1394 }
1395 case LTTNG_CONSUMER_SNAPSHOT_CHANNEL:
1396 {
1397 if (msg.u.snapshot_channel.metadata) {
1398 ret = snapshot_metadata(msg.u.snapshot_channel.key,
1399 msg.u.snapshot_channel.pathname,
1400 msg.u.snapshot_channel.relayd_id,
1401 ctx);
1402 if (ret < 0) {
1403 ERR("Snapshot metadata failed");
1404 ret_code = LTTNG_ERR_UST_META_FAIL;
1405 }
1406 } else {
1407 ret = snapshot_channel(msg.u.snapshot_channel.key,
1408 msg.u.snapshot_channel.pathname,
1409 msg.u.snapshot_channel.relayd_id,
1410 msg.u.snapshot_channel.max_stream_size,
1411 ctx);
1412 if (ret < 0) {
1413 ERR("Snapshot channel failed");
1414 ret_code = LTTNG_ERR_UST_CHAN_FAIL;
1415 }
1416 }
1417
1418 ret = consumer_send_status_msg(sock, ret_code);
1419 if (ret < 0) {
1420 /* Somehow, the session daemon is not responding anymore. */
1421 goto end_nosignal;
1422 }
1423 break;
1424 }
1425 default:
1426 break;
1427 }
1428
1429 end_nosignal:
1430 rcu_read_unlock();
1431
1432 /*
1433 * Return 1 to indicate success since the 0 value can be a socket
1434 * shutdown during the recv() or send() call.
1435 */
1436 return 1;
1437
1438 end_msg_sessiond:
1439 /*
1440 * The returned value here is not useful since either way we'll return 1 to
1441 * the caller because the session daemon socket management is done
1442 * elsewhere. Returning a negative code or 0 will shutdown the consumer.
1443 */
1444 ret = consumer_send_status_msg(sock, ret_code);
1445 if (ret < 0) {
1446 goto error_fatal;
1447 }
1448 rcu_read_unlock();
1449 return 1;
1450 end_channel_error:
1451 if (channel) {
1452 /*
1453 * Free channel here since no one has a reference to it. We don't
1454 * free after that because a stream can store this pointer.
1455 */
1456 destroy_channel(channel);
1457 }
1458 /* We have to send a status channel message indicating an error. */
1459 ret = consumer_send_status_channel(sock, NULL);
1460 if (ret < 0) {
1461 /* Stop everything if session daemon can not be notified. */
1462 goto error_fatal;
1463 }
1464 rcu_read_unlock();
1465 return 1;
1466 error_fatal:
1467 rcu_read_unlock();
1468 /* This will issue a consumer stop. */
1469 return -1;
1470 }
1471
1472 /*
1473 * Wrapper over the mmap() read offset from ust-ctl library. Since this can be
1474 * compiled out, we isolate it in this library.
1475 */
1476 int lttng_ustctl_get_mmap_read_offset(struct lttng_consumer_stream *stream,
1477 unsigned long *off)
1478 {
1479 assert(stream);
1480 assert(stream->ustream);
1481
1482 return ustctl_get_mmap_read_offset(stream->ustream, off);
1483 }
1484
1485 /*
1486 * Wrapper over the mmap() read offset from ust-ctl library. Since this can be
1487 * compiled out, we isolate it in this library.
1488 */
1489 void *lttng_ustctl_get_mmap_base(struct lttng_consumer_stream *stream)
1490 {
1491 assert(stream);
1492 assert(stream->ustream);
1493
1494 return ustctl_get_mmap_base(stream->ustream);
1495 }
1496
1497 /*
1498 * Take a snapshot for a specific fd
1499 *
1500 * Returns 0 on success, < 0 on error
1501 */
1502 int lttng_ustconsumer_take_snapshot(struct lttng_consumer_stream *stream)
1503 {
1504 assert(stream);
1505 assert(stream->ustream);
1506
1507 return ustctl_snapshot(stream->ustream);
1508 }
1509
1510 /*
1511 * Get the produced position
1512 *
1513 * Returns 0 on success, < 0 on error
1514 */
1515 int lttng_ustconsumer_get_produced_snapshot(
1516 struct lttng_consumer_stream *stream, unsigned long *pos)
1517 {
1518 assert(stream);
1519 assert(stream->ustream);
1520 assert(pos);
1521
1522 return ustctl_snapshot_get_produced(stream->ustream, pos);
1523 }
1524
1525 /*
1526 * Get the consumed position
1527 *
1528 * Returns 0 on success, < 0 on error
1529 */
1530 int lttng_ustconsumer_get_consumed_snapshot(
1531 struct lttng_consumer_stream *stream, unsigned long *pos)
1532 {
1533 assert(stream);
1534 assert(stream->ustream);
1535 assert(pos);
1536
1537 return ustctl_snapshot_get_consumed(stream->ustream, pos);
1538 }
1539
1540 /*
1541 * Called when the stream signal the consumer that it has hang up.
1542 */
1543 void lttng_ustconsumer_on_stream_hangup(struct lttng_consumer_stream *stream)
1544 {
1545 assert(stream);
1546 assert(stream->ustream);
1547
1548 ustctl_flush_buffer(stream->ustream, 0);
1549 stream->hangup_flush_done = 1;
1550 }
1551
1552 void lttng_ustconsumer_del_channel(struct lttng_consumer_channel *chan)
1553 {
1554 assert(chan);
1555 assert(chan->uchan);
1556
1557 if (chan->switch_timer_enabled == 1) {
1558 consumer_timer_switch_stop(chan);
1559 }
1560 consumer_metadata_cache_destroy(chan);
1561 ustctl_destroy_channel(chan->uchan);
1562 }
1563
1564 void lttng_ustconsumer_del_stream(struct lttng_consumer_stream *stream)
1565 {
1566 assert(stream);
1567 assert(stream->ustream);
1568
1569 if (stream->chan->switch_timer_enabled == 1) {
1570 consumer_timer_switch_stop(stream->chan);
1571 }
1572 ustctl_destroy_stream(stream->ustream);
1573 }
1574
1575 int lttng_ustconsumer_read_subbuffer(struct lttng_consumer_stream *stream,
1576 struct lttng_consumer_local_data *ctx)
1577 {
1578 unsigned long len, subbuf_size, padding;
1579 int err;
1580 long ret = 0;
1581 char dummy;
1582 struct ustctl_consumer_stream *ustream;
1583
1584 assert(stream);
1585 assert(stream->ustream);
1586 assert(ctx);
1587
1588 DBG("In UST read_subbuffer (wait_fd: %d, name: %s)", stream->wait_fd,
1589 stream->name);
1590
1591 /* Ease our life for what's next. */
1592 ustream = stream->ustream;
1593
1594 /* We can consume the 1 byte written into the wait_fd by UST */
1595 if (stream->monitor && !stream->hangup_flush_done) {
1596 ssize_t readlen;
1597
1598 do {
1599 readlen = read(stream->wait_fd, &dummy, 1);
1600 } while (readlen == -1 && errno == EINTR);
1601 if (readlen == -1 && errno != EAGAIN && errno != EWOULDBLOCK) {
1602 ret = readlen;
1603 goto end;
1604 }
1605 }
1606
1607 retry:
1608 /* Get the next subbuffer */
1609 err = ustctl_get_next_subbuf(ustream);
1610 if (err != 0) {
1611 /*
1612 * Populate metadata info if the existing info has
1613 * already been read.
1614 */
1615 if (stream->metadata_flag) {
1616 ssize_t write_len;
1617
1618 if (stream->chan->metadata_cache->contiguous
1619 == stream->ust_metadata_pushed) {
1620 ret = 0;
1621 goto end;
1622 }
1623
1624 write_len = ustctl_write_one_packet_to_channel(stream->chan->uchan,
1625 &stream->chan->metadata_cache->data[stream->ust_metadata_pushed],
1626 stream->chan->metadata_cache->contiguous
1627 - stream->ust_metadata_pushed);
1628 assert(write_len != 0);
1629 if (write_len < 0) {
1630 ERR("Writing one metadata packet");
1631 ret = -1;
1632 goto end;
1633 }
1634 stream->ust_metadata_pushed += write_len;
1635 ustctl_flush_buffer(stream->ustream, 1);
1636 goto retry;
1637 }
1638
1639 ret = err; /* ustctl_get_next_subbuf returns negative, caller expect positive. */
1640 /*
1641 * This is a debug message even for single-threaded consumer,
1642 * because poll() have more relaxed criterions than get subbuf,
1643 * so get_subbuf may fail for short race windows where poll()
1644 * would issue wakeups.
1645 */
1646 DBG("Reserving sub buffer failed (everything is normal, "
1647 "it is due to concurrency) [ret: %d]", err);
1648 goto end;
1649 }
1650 assert(stream->chan->output == CONSUMER_CHANNEL_MMAP);
1651 /* Get the full padded subbuffer size */
1652 err = ustctl_get_padded_subbuf_size(ustream, &len);
1653 assert(err == 0);
1654
1655 /* Get subbuffer data size (without padding) */
1656 err = ustctl_get_subbuf_size(ustream, &subbuf_size);
1657 assert(err == 0);
1658
1659 /* Make sure we don't get a subbuffer size bigger than the padded */
1660 assert(len >= subbuf_size);
1661
1662 padding = len - subbuf_size;
1663 /* write the subbuffer to the tracefile */
1664 ret = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, subbuf_size, padding);
1665 /*
1666 * The mmap operation should write subbuf_size amount of data when network
1667 * streaming or the full padding (len) size when we are _not_ streaming.
1668 */
1669 if ((ret != subbuf_size && stream->net_seq_idx != (uint64_t) -1ULL) ||
1670 (ret != len && stream->net_seq_idx == (uint64_t) -1ULL)) {
1671 /*
1672 * Display the error but continue processing to try to release the
1673 * subbuffer. This is a DBG statement since any unexpected kill or
1674 * signal, the application gets unregistered, relayd gets closed or
1675 * anything that affects the buffer lifetime will trigger this error.
1676 * So, for the sake of the user, don't print this error since it can
1677 * happen and it is OK with the code flow.
1678 */
1679 DBG("Error writing to tracefile "
1680 "(ret: %ld != len: %lu != subbuf_size: %lu)",
1681 ret, len, subbuf_size);
1682 }
1683 err = ustctl_put_next_subbuf(ustream);
1684 assert(err == 0);
1685
1686 end:
1687 return ret;
1688 }
1689
1690 /*
1691 * Called when a stream is created.
1692 *
1693 * Return 0 on success or else a negative value.
1694 */
1695 int lttng_ustconsumer_on_recv_stream(struct lttng_consumer_stream *stream)
1696 {
1697 int ret;
1698
1699 assert(stream);
1700
1701 /* Don't create anything if this is set for streaming. */
1702 if (stream->net_seq_idx == (uint64_t) -1ULL && stream->chan->monitor) {
1703 ret = utils_create_stream_file(stream->chan->pathname, stream->name,
1704 stream->chan->tracefile_size, stream->tracefile_count_current,
1705 stream->uid, stream->gid);
1706 if (ret < 0) {
1707 goto error;
1708 }
1709 stream->out_fd = ret;
1710 stream->tracefile_size_current = 0;
1711 }
1712 ret = 0;
1713
1714 error:
1715 return ret;
1716 }
1717
1718 /*
1719 * Check if data is still being extracted from the buffers for a specific
1720 * stream. Consumer data lock MUST be acquired before calling this function
1721 * and the stream lock.
1722 *
1723 * Return 1 if the traced data are still getting read else 0 meaning that the
1724 * data is available for trace viewer reading.
1725 */
1726 int lttng_ustconsumer_data_pending(struct lttng_consumer_stream *stream)
1727 {
1728 int ret;
1729
1730 assert(stream);
1731 assert(stream->ustream);
1732
1733 DBG("UST consumer checking data pending");
1734
1735 if (stream->endpoint_status != CONSUMER_ENDPOINT_ACTIVE) {
1736 ret = 0;
1737 goto end;
1738 }
1739
1740 if (stream->chan->type == CONSUMER_CHANNEL_TYPE_METADATA) {
1741 /*
1742 * We can simply check whether all contiguously available data
1743 * has been pushed to the ring buffer, since the push operation
1744 * is performed within get_next_subbuf(), and because both
1745 * get_next_subbuf() and put_next_subbuf() are issued atomically
1746 * thanks to the stream lock within
1747 * lttng_ustconsumer_read_subbuffer(). This basically means that
1748 * whetnever ust_metadata_pushed is incremented, the associated
1749 * metadata has been consumed from the metadata stream.
1750 */
1751 DBG("UST consumer metadata pending check: contiguous %" PRIu64 " vs pushed %" PRIu64,
1752 stream->chan->metadata_cache->contiguous,
1753 stream->ust_metadata_pushed);
1754 if (stream->chan->metadata_cache->contiguous
1755 != stream->ust_metadata_pushed) {
1756 ret = 1; /* Data is pending */
1757 goto end;
1758 }
1759 } else {
1760 ret = ustctl_get_next_subbuf(stream->ustream);
1761 if (ret == 0) {
1762 /*
1763 * There is still data so let's put back this
1764 * subbuffer.
1765 */
1766 ret = ustctl_put_subbuf(stream->ustream);
1767 assert(ret == 0);
1768 ret = 1; /* Data is pending */
1769 goto end;
1770 }
1771 }
1772
1773 /* Data is NOT pending so ready to be read. */
1774 ret = 0;
1775
1776 end:
1777 return ret;
1778 }
1779
1780 /*
1781 * Close every metadata stream wait fd of the metadata hash table. This
1782 * function MUST be used very carefully so not to run into a race between the
1783 * metadata thread handling streams and this function closing their wait fd.
1784 *
1785 * For UST, this is used when the session daemon hangs up. Its the metadata
1786 * producer so calling this is safe because we are assured that no state change
1787 * can occur in the metadata thread for the streams in the hash table.
1788 */
1789 void lttng_ustconsumer_close_metadata(struct lttng_ht *metadata_ht)
1790 {
1791 int ret;
1792 struct lttng_ht_iter iter;
1793 struct lttng_consumer_stream *stream;
1794
1795 assert(metadata_ht);
1796 assert(metadata_ht->ht);
1797
1798 DBG("UST consumer closing all metadata streams");
1799
1800 rcu_read_lock();
1801 cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream,
1802 node.node) {
1803 int fd = stream->wait_fd;
1804
1805 /*
1806 * Whatever happens here we have to continue to try to close every
1807 * streams. Let's report at least the error on failure.
1808 */
1809 ret = ustctl_stream_close_wakeup_fd(stream->ustream);
1810 if (ret) {
1811 ERR("Unable to close metadata stream fd %d ret %d", fd, ret);
1812 }
1813 DBG("Metadata wait fd %d closed", fd);
1814 }
1815 rcu_read_unlock();
1816 }
1817
1818 void lttng_ustconsumer_close_stream_wakeup(struct lttng_consumer_stream *stream)
1819 {
1820 int ret;
1821
1822 ret = ustctl_stream_close_wakeup_fd(stream->ustream);
1823 if (ret < 0) {
1824 ERR("Unable to close wakeup fd");
1825 }
1826 }
1827
1828 /*
1829 * Please refer to consumer-timer.c before adding any lock within this
1830 * function or any of its callees. Timers have a very strict locking
1831 * semantic with respect to teardown. Failure to respect this semantic
1832 * introduces deadlocks.
1833 */
1834 int lttng_ustconsumer_request_metadata(struct lttng_consumer_local_data *ctx,
1835 struct lttng_consumer_channel *channel, int timer)
1836 {
1837 struct lttcomm_metadata_request_msg request;
1838 struct lttcomm_consumer_msg msg;
1839 enum lttng_error_code ret_code = LTTNG_OK;
1840 uint64_t len, key, offset;
1841 int ret;
1842
1843 assert(channel);
1844 assert(channel->metadata_cache);
1845
1846 /* send the metadata request to sessiond */
1847 switch (consumer_data.type) {
1848 case LTTNG_CONSUMER64_UST:
1849 request.bits_per_long = 64;
1850 break;
1851 case LTTNG_CONSUMER32_UST:
1852 request.bits_per_long = 32;
1853 break;
1854 default:
1855 request.bits_per_long = 0;
1856 break;
1857 }
1858
1859 request.session_id = channel->session_id;
1860 request.session_id_per_pid = channel->session_id_per_pid;
1861 request.uid = channel->uid;
1862 request.key = channel->key;
1863 DBG("Sending metadata request to sessiond, session id %" PRIu64
1864 ", per-pid %" PRIu64,
1865 channel->session_id,
1866 channel->session_id_per_pid);
1867
1868 pthread_mutex_lock(&ctx->metadata_socket_lock);
1869 ret = lttcomm_send_unix_sock(ctx->consumer_metadata_socket, &request,
1870 sizeof(request));
1871 if (ret < 0) {
1872 ERR("Asking metadata to sessiond");
1873 goto end;
1874 }
1875
1876 /* Receive the metadata from sessiond */
1877 ret = lttcomm_recv_unix_sock(ctx->consumer_metadata_socket, &msg,
1878 sizeof(msg));
1879 if (ret != sizeof(msg)) {
1880 DBG("Consumer received unexpected message size %d (expects %zu)",
1881 ret, sizeof(msg));
1882 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD);
1883 /*
1884 * The ret value might 0 meaning an orderly shutdown but this is ok
1885 * since the caller handles this.
1886 */
1887 goto end;
1888 }
1889
1890 if (msg.cmd_type == LTTNG_ERR_UND) {
1891 /* No registry found */
1892 (void) consumer_send_status_msg(ctx->consumer_metadata_socket,
1893 ret_code);
1894 ret = 0;
1895 goto end;
1896 } else if (msg.cmd_type != LTTNG_CONSUMER_PUSH_METADATA) {
1897 ERR("Unexpected cmd_type received %d", msg.cmd_type);
1898 ret = -1;
1899 goto end;
1900 }
1901
1902 len = msg.u.push_metadata.len;
1903 key = msg.u.push_metadata.key;
1904 offset = msg.u.push_metadata.target_offset;
1905
1906 assert(key == channel->key);
1907 if (len == 0) {
1908 DBG("No new metadata to receive for key %" PRIu64, key);
1909 }
1910
1911 /* Tell session daemon we are ready to receive the metadata. */
1912 ret = consumer_send_status_msg(ctx->consumer_metadata_socket,
1913 LTTNG_OK);
1914 if (ret < 0 || len == 0) {
1915 /*
1916 * Somehow, the session daemon is not responding anymore or there is
1917 * nothing to receive.
1918 */
1919 goto end;
1920 }
1921
1922 ret_code = lttng_ustconsumer_recv_metadata(ctx->consumer_metadata_socket,
1923 key, offset, len, channel, timer);
1924 if (ret_code >= 0) {
1925 /*
1926 * Only send the status msg if the sessiond is alive meaning a positive
1927 * ret code.
1928 */
1929 (void) consumer_send_status_msg(ctx->consumer_metadata_socket, ret_code);
1930 }
1931 ret = 0;
1932
1933 end:
1934 pthread_mutex_unlock(&ctx->metadata_socket_lock);
1935 return ret;
1936 }
This page took 0.103348 seconds and 6 git commands to generate.