Fix: ust-consumer: metadata cache lock not taken when sampling max offset
[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 * Copyright (C) 2017 - Jérémie Galarneau <jeremie.galarneau@efficios.com>
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License, version 2 only,
8 * as published by the Free Software Foundation.
9 *
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
14 *
15 * You should have received a copy of the GNU General Public License along
16 * with this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
18 */
19
20 #define _LGPL_SOURCE
21 #include <assert.h>
22 #include <lttng/ust-ctl.h>
23 #include <poll.h>
24 #include <pthread.h>
25 #include <stdlib.h>
26 #include <string.h>
27 #include <sys/mman.h>
28 #include <sys/socket.h>
29 #include <sys/stat.h>
30 #include <sys/types.h>
31 #include <inttypes.h>
32 #include <unistd.h>
33 #include <urcu/list.h>
34 #include <signal.h>
35 #include <stdbool.h>
36 #include <stdint.h>
37
38 #include <bin/lttng-consumerd/health-consumerd.h>
39 #include <common/common.h>
40 #include <common/sessiond-comm/sessiond-comm.h>
41 #include <common/relayd/relayd.h>
42 #include <common/compat/fcntl.h>
43 #include <common/compat/endian.h>
44 #include <common/consumer/consumer-metadata-cache.h>
45 #include <common/consumer/consumer-stream.h>
46 #include <common/consumer/consumer-timer.h>
47 #include <common/utils.h>
48 #include <common/index/index.h>
49 #include <common/consumer/consumer.h>
50 #include <common/optional.h>
51
52 #include "ust-consumer.h"
53
54 #define INT_MAX_STR_LEN 12 /* includes \0 */
55
56 extern struct lttng_consumer_global_data consumer_data;
57 extern int consumer_poll_timeout;
58
59 /*
60 * Free channel object and all streams associated with it. This MUST be used
61 * only and only if the channel has _NEVER_ been added to the global channel
62 * hash table.
63 */
64 static void destroy_channel(struct lttng_consumer_channel *channel)
65 {
66 struct lttng_consumer_stream *stream, *stmp;
67
68 assert(channel);
69
70 DBG("UST consumer cleaning stream list");
71
72 cds_list_for_each_entry_safe(stream, stmp, &channel->streams.head,
73 send_node) {
74
75 health_code_update();
76
77 cds_list_del(&stream->send_node);
78 ustctl_destroy_stream(stream->ustream);
79 lttng_trace_chunk_put(stream->trace_chunk);
80 free(stream);
81 }
82
83 /*
84 * If a channel is available meaning that was created before the streams
85 * were, delete it.
86 */
87 if (channel->uchan) {
88 lttng_ustconsumer_del_channel(channel);
89 lttng_ustconsumer_free_channel(channel);
90 }
91
92 if (channel->trace_chunk) {
93 lttng_trace_chunk_put(channel->trace_chunk);
94 }
95
96 free(channel);
97 }
98
99 /*
100 * Add channel to internal consumer state.
101 *
102 * Returns 0 on success or else a negative value.
103 */
104 static int add_channel(struct lttng_consumer_channel *channel,
105 struct lttng_consumer_local_data *ctx)
106 {
107 int ret = 0;
108
109 assert(channel);
110 assert(ctx);
111
112 if (ctx->on_recv_channel != NULL) {
113 ret = ctx->on_recv_channel(channel);
114 if (ret == 0) {
115 ret = consumer_add_channel(channel, ctx);
116 } else if (ret < 0) {
117 /* Most likely an ENOMEM. */
118 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
119 goto error;
120 }
121 } else {
122 ret = consumer_add_channel(channel, ctx);
123 }
124
125 DBG("UST consumer channel added (key: %" PRIu64 ")", channel->key);
126
127 error:
128 return ret;
129 }
130
131 /*
132 * Allocate and return a consumer stream object. If _alloc_ret is not NULL, the
133 * error value if applicable is set in it else it is kept untouched.
134 *
135 * Return NULL on error else the newly allocated stream object.
136 */
137 static struct lttng_consumer_stream *allocate_stream(int cpu, int key,
138 struct lttng_consumer_channel *channel,
139 struct lttng_consumer_local_data *ctx, int *_alloc_ret)
140 {
141 int alloc_ret;
142 struct lttng_consumer_stream *stream = NULL;
143
144 assert(channel);
145 assert(ctx);
146
147 stream = consumer_stream_create(
148 channel,
149 channel->key,
150 key,
151 channel->name,
152 channel->relayd_id,
153 channel->session_id,
154 channel->trace_chunk,
155 cpu,
156 &alloc_ret,
157 channel->type,
158 channel->monitor);
159 if (stream == NULL) {
160 switch (alloc_ret) {
161 case -ENOENT:
162 /*
163 * We could not find the channel. Can happen if cpu hotplug
164 * happens while tearing down.
165 */
166 DBG3("Could not find channel");
167 break;
168 case -ENOMEM:
169 case -EINVAL:
170 default:
171 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
172 break;
173 }
174 goto error;
175 }
176
177 consumer_stream_update_channel_attributes(stream, channel);
178
179 error:
180 if (_alloc_ret) {
181 *_alloc_ret = alloc_ret;
182 }
183 return stream;
184 }
185
186 /*
187 * Send the given stream pointer to the corresponding thread.
188 *
189 * Returns 0 on success else a negative value.
190 */
191 static int send_stream_to_thread(struct lttng_consumer_stream *stream,
192 struct lttng_consumer_local_data *ctx)
193 {
194 int ret;
195 struct lttng_pipe *stream_pipe;
196
197 /* Get the right pipe where the stream will be sent. */
198 if (stream->metadata_flag) {
199 consumer_add_metadata_stream(stream);
200 stream_pipe = ctx->consumer_metadata_pipe;
201 } else {
202 consumer_add_data_stream(stream);
203 stream_pipe = ctx->consumer_data_pipe;
204 }
205
206 /*
207 * From this point on, the stream's ownership has been moved away from
208 * the channel and it becomes globally visible. Hence, remove it from
209 * the local stream list to prevent the stream from being both local and
210 * global.
211 */
212 stream->globally_visible = 1;
213 cds_list_del(&stream->send_node);
214
215 ret = lttng_pipe_write(stream_pipe, &stream, sizeof(stream));
216 if (ret < 0) {
217 ERR("Consumer write %s stream to pipe %d",
218 stream->metadata_flag ? "metadata" : "data",
219 lttng_pipe_get_writefd(stream_pipe));
220 if (stream->metadata_flag) {
221 consumer_del_stream_for_metadata(stream);
222 } else {
223 consumer_del_stream_for_data(stream);
224 }
225 goto error;
226 }
227
228 error:
229 return ret;
230 }
231
232 static
233 int get_stream_shm_path(char *stream_shm_path, const char *shm_path, int cpu)
234 {
235 char cpu_nr[INT_MAX_STR_LEN]; /* int max len */
236 int ret;
237
238 strncpy(stream_shm_path, shm_path, PATH_MAX);
239 stream_shm_path[PATH_MAX - 1] = '\0';
240 ret = snprintf(cpu_nr, INT_MAX_STR_LEN, "%i", cpu);
241 if (ret < 0) {
242 PERROR("snprintf");
243 goto end;
244 }
245 strncat(stream_shm_path, cpu_nr,
246 PATH_MAX - strlen(stream_shm_path) - 1);
247 ret = 0;
248 end:
249 return ret;
250 }
251
252 /*
253 * Create streams for the given channel using liblttng-ust-ctl.
254 * The channel lock must be acquired by the caller.
255 *
256 * Return 0 on success else a negative value.
257 */
258 static int create_ust_streams(struct lttng_consumer_channel *channel,
259 struct lttng_consumer_local_data *ctx)
260 {
261 int ret, cpu = 0;
262 struct ustctl_consumer_stream *ustream;
263 struct lttng_consumer_stream *stream;
264 pthread_mutex_t *current_stream_lock = NULL;
265
266 assert(channel);
267 assert(ctx);
268
269 /*
270 * While a stream is available from ustctl. When NULL is returned, we've
271 * reached the end of the possible stream for the channel.
272 */
273 while ((ustream = ustctl_create_stream(channel->uchan, cpu))) {
274 int wait_fd;
275 int ust_metadata_pipe[2];
276
277 health_code_update();
278
279 if (channel->type == CONSUMER_CHANNEL_TYPE_METADATA && channel->monitor) {
280 ret = utils_create_pipe_cloexec_nonblock(ust_metadata_pipe);
281 if (ret < 0) {
282 ERR("Create ust metadata poll pipe");
283 goto error;
284 }
285 wait_fd = ust_metadata_pipe[0];
286 } else {
287 wait_fd = ustctl_stream_get_wait_fd(ustream);
288 }
289
290 /* Allocate consumer stream object. */
291 stream = allocate_stream(cpu, wait_fd, channel, ctx, &ret);
292 if (!stream) {
293 goto error_alloc;
294 }
295 stream->ustream = ustream;
296 /*
297 * Store it so we can save multiple function calls afterwards since
298 * this value is used heavily in the stream threads. This is UST
299 * specific so this is why it's done after allocation.
300 */
301 stream->wait_fd = wait_fd;
302
303 /*
304 * Increment channel refcount since the channel reference has now been
305 * assigned in the allocation process above.
306 */
307 if (stream->chan->monitor) {
308 uatomic_inc(&stream->chan->refcount);
309 }
310
311 pthread_mutex_lock(&stream->lock);
312 current_stream_lock = &stream->lock;
313 /*
314 * Order is important this is why a list is used. On error, the caller
315 * should clean this list.
316 */
317 cds_list_add_tail(&stream->send_node, &channel->streams.head);
318
319 ret = ustctl_get_max_subbuf_size(stream->ustream,
320 &stream->max_sb_size);
321 if (ret < 0) {
322 ERR("ustctl_get_max_subbuf_size failed for stream %s",
323 stream->name);
324 goto error;
325 }
326
327 /* Do actions once stream has been received. */
328 if (ctx->on_recv_stream) {
329 ret = ctx->on_recv_stream(stream);
330 if (ret < 0) {
331 goto error;
332 }
333 }
334
335 DBG("UST consumer add stream %s (key: %" PRIu64 ") with relayd id %" PRIu64,
336 stream->name, stream->key, stream->relayd_stream_id);
337
338 /* Set next CPU stream. */
339 channel->streams.count = ++cpu;
340
341 /* Keep stream reference when creating metadata. */
342 if (channel->type == CONSUMER_CHANNEL_TYPE_METADATA) {
343 channel->metadata_stream = stream;
344 if (channel->monitor) {
345 /* Set metadata poll pipe if we created one */
346 memcpy(stream->ust_metadata_poll_pipe,
347 ust_metadata_pipe,
348 sizeof(ust_metadata_pipe));
349 }
350 }
351 pthread_mutex_unlock(&stream->lock);
352 current_stream_lock = NULL;
353 }
354
355 return 0;
356
357 error:
358 error_alloc:
359 if (current_stream_lock) {
360 pthread_mutex_unlock(current_stream_lock);
361 }
362 return ret;
363 }
364
365 /*
366 * create_posix_shm is never called concurrently within a process.
367 */
368 static
369 int create_posix_shm(void)
370 {
371 char tmp_name[NAME_MAX];
372 int shmfd, ret;
373
374 ret = snprintf(tmp_name, NAME_MAX, "/ust-shm-consumer-%d", getpid());
375 if (ret < 0) {
376 PERROR("snprintf");
377 return -1;
378 }
379 /*
380 * Allocate shm, and immediately unlink its shm oject, keeping
381 * only the file descriptor as a reference to the object.
382 * We specifically do _not_ use the / at the beginning of the
383 * pathname so that some OS implementations can keep it local to
384 * the process (POSIX leaves this implementation-defined).
385 */
386 shmfd = shm_open(tmp_name, O_CREAT | O_EXCL | O_RDWR, 0700);
387 if (shmfd < 0) {
388 PERROR("shm_open");
389 goto error_shm_open;
390 }
391 ret = shm_unlink(tmp_name);
392 if (ret < 0 && errno != ENOENT) {
393 PERROR("shm_unlink");
394 goto error_shm_release;
395 }
396 return shmfd;
397
398 error_shm_release:
399 ret = close(shmfd);
400 if (ret) {
401 PERROR("close");
402 }
403 error_shm_open:
404 return -1;
405 }
406
407 static int open_ust_stream_fd(struct lttng_consumer_channel *channel, int cpu,
408 const struct lttng_credentials *session_credentials)
409 {
410 char shm_path[PATH_MAX];
411 int ret;
412
413 if (!channel->shm_path[0]) {
414 return create_posix_shm();
415 }
416 ret = get_stream_shm_path(shm_path, channel->shm_path, cpu);
417 if (ret) {
418 goto error_shm_path;
419 }
420 return run_as_open(shm_path,
421 O_RDWR | O_CREAT | O_EXCL, S_IRUSR | S_IWUSR,
422 session_credentials->uid, session_credentials->gid);
423
424 error_shm_path:
425 return -1;
426 }
427
428 /*
429 * Create an UST channel with the given attributes and send it to the session
430 * daemon using the ust ctl API.
431 *
432 * Return 0 on success or else a negative value.
433 */
434 static int create_ust_channel(struct lttng_consumer_channel *channel,
435 struct ustctl_consumer_channel_attr *attr,
436 struct ustctl_consumer_channel **ust_chanp)
437 {
438 int ret, nr_stream_fds, i, j;
439 int *stream_fds;
440 struct ustctl_consumer_channel *ust_channel;
441
442 assert(channel);
443 assert(attr);
444 assert(ust_chanp);
445 assert(channel->buffer_credentials.is_set);
446
447 DBG3("Creating channel to ustctl with attr: [overwrite: %d, "
448 "subbuf_size: %" PRIu64 ", num_subbuf: %" PRIu64 ", "
449 "switch_timer_interval: %u, read_timer_interval: %u, "
450 "output: %d, type: %d", attr->overwrite, attr->subbuf_size,
451 attr->num_subbuf, attr->switch_timer_interval,
452 attr->read_timer_interval, attr->output, attr->type);
453
454 if (channel->type == CONSUMER_CHANNEL_TYPE_METADATA)
455 nr_stream_fds = 1;
456 else
457 nr_stream_fds = ustctl_get_nr_stream_per_channel();
458 stream_fds = zmalloc(nr_stream_fds * sizeof(*stream_fds));
459 if (!stream_fds) {
460 ret = -1;
461 goto error_alloc;
462 }
463 for (i = 0; i < nr_stream_fds; i++) {
464 stream_fds[i] = open_ust_stream_fd(channel, i,
465 &channel->buffer_credentials.value);
466 if (stream_fds[i] < 0) {
467 ret = -1;
468 goto error_open;
469 }
470 }
471 ust_channel = ustctl_create_channel(attr, stream_fds, nr_stream_fds);
472 if (!ust_channel) {
473 ret = -1;
474 goto error_create;
475 }
476 channel->nr_stream_fds = nr_stream_fds;
477 channel->stream_fds = stream_fds;
478 *ust_chanp = ust_channel;
479
480 return 0;
481
482 error_create:
483 error_open:
484 for (j = i - 1; j >= 0; j--) {
485 int closeret;
486
487 closeret = close(stream_fds[j]);
488 if (closeret) {
489 PERROR("close");
490 }
491 if (channel->shm_path[0]) {
492 char shm_path[PATH_MAX];
493
494 closeret = get_stream_shm_path(shm_path,
495 channel->shm_path, j);
496 if (closeret) {
497 ERR("Cannot get stream shm path");
498 }
499 closeret = run_as_unlink(shm_path,
500 channel->buffer_credentials.value.uid,
501 channel->buffer_credentials.value.gid);
502 if (closeret) {
503 PERROR("unlink %s", shm_path);
504 }
505 }
506 }
507 /* Try to rmdir all directories under shm_path root. */
508 if (channel->root_shm_path[0]) {
509 (void) run_as_rmdir_recursive(channel->root_shm_path,
510 channel->buffer_credentials.value.uid,
511 channel->buffer_credentials.value.gid,
512 LTTNG_DIRECTORY_HANDLE_SKIP_NON_EMPTY_FLAG);
513 }
514 free(stream_fds);
515 error_alloc:
516 return ret;
517 }
518
519 /*
520 * Send a single given stream to the session daemon using the sock.
521 *
522 * Return 0 on success else a negative value.
523 */
524 static int send_sessiond_stream(int sock, struct lttng_consumer_stream *stream)
525 {
526 int ret;
527
528 assert(stream);
529 assert(sock >= 0);
530
531 DBG("UST consumer sending stream %" PRIu64 " to sessiond", stream->key);
532
533 /* Send stream to session daemon. */
534 ret = ustctl_send_stream_to_sessiond(sock, stream->ustream);
535 if (ret < 0) {
536 goto error;
537 }
538
539 error:
540 return ret;
541 }
542
543 /*
544 * Send channel to sessiond and relayd if applicable.
545 *
546 * Return 0 on success or else a negative value.
547 */
548 static int send_channel_to_sessiond_and_relayd(int sock,
549 struct lttng_consumer_channel *channel,
550 struct lttng_consumer_local_data *ctx, int *relayd_error)
551 {
552 int ret, ret_code = LTTCOMM_CONSUMERD_SUCCESS;
553 struct lttng_consumer_stream *stream;
554 uint64_t net_seq_idx = -1ULL;
555
556 assert(channel);
557 assert(ctx);
558 assert(sock >= 0);
559
560 DBG("UST consumer sending channel %s to sessiond", channel->name);
561
562 if (channel->relayd_id != (uint64_t) -1ULL) {
563 cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
564
565 health_code_update();
566
567 /* Try to send the stream to the relayd if one is available. */
568 DBG("Sending stream %" PRIu64 " of channel \"%s\" to relayd",
569 stream->key, channel->name);
570 ret = consumer_send_relayd_stream(stream, stream->chan->pathname);
571 if (ret < 0) {
572 /*
573 * Flag that the relayd was the problem here probably due to a
574 * communicaton error on the socket.
575 */
576 if (relayd_error) {
577 *relayd_error = 1;
578 }
579 ret_code = LTTCOMM_CONSUMERD_RELAYD_FAIL;
580 }
581 if (net_seq_idx == -1ULL) {
582 net_seq_idx = stream->net_seq_idx;
583 }
584 }
585 }
586
587 /* Inform sessiond that we are about to send channel and streams. */
588 ret = consumer_send_status_msg(sock, ret_code);
589 if (ret < 0 || ret_code != LTTCOMM_CONSUMERD_SUCCESS) {
590 /*
591 * Either the session daemon is not responding or the relayd died so we
592 * stop now.
593 */
594 goto error;
595 }
596
597 /* Send channel to sessiond. */
598 ret = ustctl_send_channel_to_sessiond(sock, channel->uchan);
599 if (ret < 0) {
600 goto error;
601 }
602
603 ret = ustctl_channel_close_wakeup_fd(channel->uchan);
604 if (ret < 0) {
605 goto error;
606 }
607
608 /* The channel was sent successfully to the sessiond at this point. */
609 cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
610
611 health_code_update();
612
613 /* Send stream to session daemon. */
614 ret = send_sessiond_stream(sock, stream);
615 if (ret < 0) {
616 goto error;
617 }
618 }
619
620 /* Tell sessiond there is no more stream. */
621 ret = ustctl_send_stream_to_sessiond(sock, NULL);
622 if (ret < 0) {
623 goto error;
624 }
625
626 DBG("UST consumer NULL stream sent to sessiond");
627
628 return 0;
629
630 error:
631 if (ret_code != LTTCOMM_CONSUMERD_SUCCESS) {
632 ret = -1;
633 }
634 return ret;
635 }
636
637 /*
638 * Creates a channel and streams and add the channel it to the channel internal
639 * state. The created stream must ONLY be sent once the GET_CHANNEL command is
640 * received.
641 *
642 * Return 0 on success or else, a negative value is returned and the channel
643 * MUST be destroyed by consumer_del_channel().
644 */
645 static int ask_channel(struct lttng_consumer_local_data *ctx,
646 struct lttng_consumer_channel *channel,
647 struct ustctl_consumer_channel_attr *attr)
648 {
649 int ret;
650
651 assert(ctx);
652 assert(channel);
653 assert(attr);
654
655 /*
656 * This value is still used by the kernel consumer since for the kernel,
657 * the stream ownership is not IN the consumer so we need to have the
658 * number of left stream that needs to be initialized so we can know when
659 * to delete the channel (see consumer.c).
660 *
661 * As for the user space tracer now, the consumer creates and sends the
662 * stream to the session daemon which only sends them to the application
663 * once every stream of a channel is received making this value useless
664 * because we they will be added to the poll thread before the application
665 * receives them. This ensures that a stream can not hang up during
666 * initilization of a channel.
667 */
668 channel->nb_init_stream_left = 0;
669
670 /* The reply msg status is handled in the following call. */
671 ret = create_ust_channel(channel, attr, &channel->uchan);
672 if (ret < 0) {
673 goto end;
674 }
675
676 channel->wait_fd = ustctl_channel_get_wait_fd(channel->uchan);
677
678 /*
679 * For the snapshots (no monitor), we create the metadata streams
680 * on demand, not during the channel creation.
681 */
682 if (channel->type == CONSUMER_CHANNEL_TYPE_METADATA && !channel->monitor) {
683 ret = 0;
684 goto end;
685 }
686
687 /* Open all streams for this channel. */
688 pthread_mutex_lock(&channel->lock);
689 ret = create_ust_streams(channel, ctx);
690 pthread_mutex_unlock(&channel->lock);
691 if (ret < 0) {
692 goto end;
693 }
694
695 end:
696 return ret;
697 }
698
699 /*
700 * Send all stream of a channel to the right thread handling it.
701 *
702 * On error, return a negative value else 0 on success.
703 */
704 static int send_streams_to_thread(struct lttng_consumer_channel *channel,
705 struct lttng_consumer_local_data *ctx)
706 {
707 int ret = 0;
708 struct lttng_consumer_stream *stream, *stmp;
709
710 assert(channel);
711 assert(ctx);
712
713 /* Send streams to the corresponding thread. */
714 cds_list_for_each_entry_safe(stream, stmp, &channel->streams.head,
715 send_node) {
716
717 health_code_update();
718
719 /* Sending the stream to the thread. */
720 ret = send_stream_to_thread(stream, ctx);
721 if (ret < 0) {
722 /*
723 * If we are unable to send the stream to the thread, there is
724 * a big problem so just stop everything.
725 */
726 goto error;
727 }
728 }
729
730 error:
731 return ret;
732 }
733
734 /*
735 * Flush channel's streams using the given key to retrieve the channel.
736 *
737 * Return 0 on success else an LTTng error code.
738 */
739 static int flush_channel(uint64_t chan_key)
740 {
741 int ret = 0;
742 struct lttng_consumer_channel *channel;
743 struct lttng_consumer_stream *stream;
744 struct lttng_ht *ht;
745 struct lttng_ht_iter iter;
746
747 DBG("UST consumer flush channel key %" PRIu64, chan_key);
748
749 rcu_read_lock();
750 channel = consumer_find_channel(chan_key);
751 if (!channel) {
752 ERR("UST consumer flush channel %" PRIu64 " not found", chan_key);
753 ret = LTTNG_ERR_UST_CHAN_NOT_FOUND;
754 goto error;
755 }
756
757 ht = consumer_data.stream_per_chan_id_ht;
758
759 /* For each stream of the channel id, flush it. */
760 cds_lfht_for_each_entry_duplicate(ht->ht,
761 ht->hash_fct(&channel->key, lttng_ht_seed), ht->match_fct,
762 &channel->key, &iter.iter, stream, node_channel_id.node) {
763
764 health_code_update();
765
766 pthread_mutex_lock(&stream->lock);
767
768 /*
769 * Protect against concurrent teardown of a stream.
770 */
771 if (cds_lfht_is_node_deleted(&stream->node.node)) {
772 goto next;
773 }
774
775 if (!stream->quiescent) {
776 ustctl_flush_buffer(stream->ustream, 0);
777 stream->quiescent = true;
778 }
779 next:
780 pthread_mutex_unlock(&stream->lock);
781 }
782 error:
783 rcu_read_unlock();
784 return ret;
785 }
786
787 /*
788 * Clear quiescent state from channel's streams using the given key to
789 * retrieve the channel.
790 *
791 * Return 0 on success else an LTTng error code.
792 */
793 static int clear_quiescent_channel(uint64_t chan_key)
794 {
795 int ret = 0;
796 struct lttng_consumer_channel *channel;
797 struct lttng_consumer_stream *stream;
798 struct lttng_ht *ht;
799 struct lttng_ht_iter iter;
800
801 DBG("UST consumer clear quiescent channel key %" PRIu64, chan_key);
802
803 rcu_read_lock();
804 channel = consumer_find_channel(chan_key);
805 if (!channel) {
806 ERR("UST consumer clear quiescent channel %" PRIu64 " not found", chan_key);
807 ret = LTTNG_ERR_UST_CHAN_NOT_FOUND;
808 goto error;
809 }
810
811 ht = consumer_data.stream_per_chan_id_ht;
812
813 /* For each stream of the channel id, clear quiescent state. */
814 cds_lfht_for_each_entry_duplicate(ht->ht,
815 ht->hash_fct(&channel->key, lttng_ht_seed), ht->match_fct,
816 &channel->key, &iter.iter, stream, node_channel_id.node) {
817
818 health_code_update();
819
820 pthread_mutex_lock(&stream->lock);
821 stream->quiescent = false;
822 pthread_mutex_unlock(&stream->lock);
823 }
824 error:
825 rcu_read_unlock();
826 return ret;
827 }
828
829 /*
830 * Close metadata stream wakeup_fd using the given key to retrieve the channel.
831 *
832 * Return 0 on success else an LTTng error code.
833 */
834 static int close_metadata(uint64_t chan_key)
835 {
836 int ret = 0;
837 struct lttng_consumer_channel *channel;
838 unsigned int channel_monitor;
839
840 DBG("UST consumer close metadata key %" PRIu64, chan_key);
841
842 channel = consumer_find_channel(chan_key);
843 if (!channel) {
844 /*
845 * This is possible if the metadata thread has issue a delete because
846 * the endpoint point of the stream hung up. There is no way the
847 * session daemon can know about it thus use a DBG instead of an actual
848 * error.
849 */
850 DBG("UST consumer close metadata %" PRIu64 " not found", chan_key);
851 ret = LTTNG_ERR_UST_CHAN_NOT_FOUND;
852 goto error;
853 }
854
855 pthread_mutex_lock(&consumer_data.lock);
856 pthread_mutex_lock(&channel->lock);
857 channel_monitor = channel->monitor;
858 if (cds_lfht_is_node_deleted(&channel->node.node)) {
859 goto error_unlock;
860 }
861
862 lttng_ustconsumer_close_metadata(channel);
863 pthread_mutex_unlock(&channel->lock);
864 pthread_mutex_unlock(&consumer_data.lock);
865
866 /*
867 * The ownership of a metadata channel depends on the type of
868 * session to which it belongs. In effect, the monitor flag is checked
869 * to determine if this metadata channel is in "snapshot" mode or not.
870 *
871 * In the non-snapshot case, the metadata channel is created along with
872 * a single stream which will remain present until the metadata channel
873 * is destroyed (on the destruction of its session). In this case, the
874 * metadata stream in "monitored" by the metadata poll thread and holds
875 * the ownership of its channel.
876 *
877 * Closing the metadata will cause the metadata stream's "metadata poll
878 * pipe" to be closed. Closing this pipe will wake-up the metadata poll
879 * thread which will teardown the metadata stream which, in return,
880 * deletes the metadata channel.
881 *
882 * In the snapshot case, the metadata stream is created and destroyed
883 * on every snapshot record. Since the channel doesn't have an owner
884 * other than the session daemon, it is safe to destroy it immediately
885 * on reception of the CLOSE_METADATA command.
886 */
887 if (!channel_monitor) {
888 /*
889 * The channel and consumer_data locks must be
890 * released before this call since consumer_del_channel
891 * re-acquires the channel and consumer_data locks to teardown
892 * the channel and queue its reclamation by the "call_rcu"
893 * worker thread.
894 */
895 consumer_del_channel(channel);
896 }
897
898 return ret;
899 error_unlock:
900 pthread_mutex_unlock(&channel->lock);
901 pthread_mutex_unlock(&consumer_data.lock);
902 error:
903 return ret;
904 }
905
906 /*
907 * RCU read side lock MUST be acquired before calling this function.
908 *
909 * Return 0 on success else an LTTng error code.
910 */
911 static int setup_metadata(struct lttng_consumer_local_data *ctx, uint64_t key)
912 {
913 int ret;
914 struct lttng_consumer_channel *metadata;
915
916 DBG("UST consumer setup metadata key %" PRIu64, key);
917
918 metadata = consumer_find_channel(key);
919 if (!metadata) {
920 ERR("UST consumer push metadata %" PRIu64 " not found", key);
921 ret = LTTNG_ERR_UST_CHAN_NOT_FOUND;
922 goto end;
923 }
924
925 /*
926 * In no monitor mode, the metadata channel has no stream(s) so skip the
927 * ownership transfer to the metadata thread.
928 */
929 if (!metadata->monitor) {
930 DBG("Metadata channel in no monitor");
931 ret = 0;
932 goto end;
933 }
934
935 /*
936 * Send metadata stream to relayd if one available. Availability is
937 * known if the stream is still in the list of the channel.
938 */
939 if (cds_list_empty(&metadata->streams.head)) {
940 ERR("Metadata channel key %" PRIu64 ", no stream available.", key);
941 ret = LTTCOMM_CONSUMERD_ERROR_METADATA;
942 goto error_no_stream;
943 }
944
945 /* Send metadata stream to relayd if needed. */
946 if (metadata->metadata_stream->net_seq_idx != (uint64_t) -1ULL) {
947 ret = consumer_send_relayd_stream(metadata->metadata_stream,
948 metadata->pathname);
949 if (ret < 0) {
950 ret = LTTCOMM_CONSUMERD_ERROR_METADATA;
951 goto error;
952 }
953 ret = consumer_send_relayd_streams_sent(
954 metadata->metadata_stream->net_seq_idx);
955 if (ret < 0) {
956 ret = LTTCOMM_CONSUMERD_RELAYD_FAIL;
957 goto error;
958 }
959 }
960
961 /*
962 * Ownership of metadata stream is passed along. Freeing is handled by
963 * the callee.
964 */
965 ret = send_streams_to_thread(metadata, ctx);
966 if (ret < 0) {
967 /*
968 * If we are unable to send the stream to the thread, there is
969 * a big problem so just stop everything.
970 */
971 ret = LTTCOMM_CONSUMERD_FATAL;
972 goto send_streams_error;
973 }
974 /* List MUST be empty after or else it could be reused. */
975 assert(cds_list_empty(&metadata->streams.head));
976
977 ret = 0;
978 goto end;
979
980 error:
981 /*
982 * Delete metadata channel on error. At this point, the metadata stream can
983 * NOT be monitored by the metadata thread thus having the guarantee that
984 * the stream is still in the local stream list of the channel. This call
985 * will make sure to clean that list.
986 */
987 consumer_stream_destroy(metadata->metadata_stream, NULL);
988 cds_list_del(&metadata->metadata_stream->send_node);
989 metadata->metadata_stream = NULL;
990 send_streams_error:
991 error_no_stream:
992 end:
993 return ret;
994 }
995
996 /*
997 * Snapshot the whole metadata.
998 * RCU read-side lock must be held by the caller.
999 *
1000 * Returns 0 on success, < 0 on error
1001 */
1002 static int snapshot_metadata(struct lttng_consumer_channel *metadata_channel,
1003 uint64_t key, char *path, uint64_t relayd_id,
1004 struct lttng_consumer_local_data *ctx)
1005 {
1006 int ret = 0;
1007 struct lttng_consumer_stream *metadata_stream;
1008
1009 assert(path);
1010 assert(ctx);
1011
1012 DBG("UST consumer snapshot metadata with key %" PRIu64 " at path %s",
1013 key, path);
1014
1015 rcu_read_lock();
1016
1017 assert(!metadata_channel->monitor);
1018
1019 health_code_update();
1020
1021 /*
1022 * Ask the sessiond if we have new metadata waiting and update the
1023 * consumer metadata cache.
1024 */
1025 ret = lttng_ustconsumer_request_metadata(ctx, metadata_channel, 0, 1);
1026 if (ret < 0) {
1027 goto error;
1028 }
1029
1030 health_code_update();
1031
1032 /*
1033 * The metadata stream is NOT created in no monitor mode when the channel
1034 * is created on a sessiond ask channel command.
1035 */
1036 ret = create_ust_streams(metadata_channel, ctx);
1037 if (ret < 0) {
1038 goto error;
1039 }
1040
1041 metadata_stream = metadata_channel->metadata_stream;
1042 assert(metadata_stream);
1043
1044 pthread_mutex_lock(&metadata_stream->lock);
1045 if (relayd_id != (uint64_t) -1ULL) {
1046 metadata_stream->net_seq_idx = relayd_id;
1047 ret = consumer_send_relayd_stream(metadata_stream, path);
1048 } else {
1049 ret = consumer_stream_create_output_files(metadata_stream,
1050 false);
1051 }
1052 pthread_mutex_unlock(&metadata_stream->lock);
1053 if (ret < 0) {
1054 goto error_stream;
1055 }
1056
1057 do {
1058 health_code_update();
1059
1060 ret = lttng_consumer_read_subbuffer(metadata_stream, ctx, true);
1061 if (ret < 0) {
1062 goto error_stream;
1063 }
1064 } while (ret > 0);
1065
1066 error_stream:
1067 /*
1068 * Clean up the stream completly because the next snapshot will use a new
1069 * metadata stream.
1070 */
1071 consumer_stream_destroy(metadata_stream, NULL);
1072 cds_list_del(&metadata_stream->send_node);
1073 metadata_channel->metadata_stream = NULL;
1074
1075 error:
1076 rcu_read_unlock();
1077 return ret;
1078 }
1079
1080 static
1081 int get_current_subbuf_addr(struct lttng_consumer_stream *stream,
1082 const char **addr)
1083 {
1084 int ret;
1085 unsigned long mmap_offset;
1086 const char *mmap_base;
1087
1088 mmap_base = ustctl_get_mmap_base(stream->ustream);
1089 if (!mmap_base) {
1090 ERR("Failed to get mmap base for stream `%s`",
1091 stream->name);
1092 ret = -EPERM;
1093 goto error;
1094 }
1095
1096 ret = ustctl_get_mmap_read_offset(stream->ustream, &mmap_offset);
1097 if (ret != 0) {
1098 ERR("Failed to get mmap offset for stream `%s`", stream->name);
1099 ret = -EINVAL;
1100 goto error;
1101 }
1102
1103 *addr = mmap_base + mmap_offset;
1104 error:
1105 return ret;
1106
1107 }
1108
1109 /*
1110 * Take a snapshot of all the stream of a channel.
1111 * RCU read-side lock and the channel lock must be held by the caller.
1112 *
1113 * Returns 0 on success, < 0 on error
1114 */
1115 static int snapshot_channel(struct lttng_consumer_channel *channel,
1116 uint64_t key, char *path, uint64_t relayd_id,
1117 uint64_t nb_packets_per_stream,
1118 struct lttng_consumer_local_data *ctx)
1119 {
1120 int ret;
1121 unsigned use_relayd = 0;
1122 unsigned long consumed_pos, produced_pos;
1123 struct lttng_consumer_stream *stream;
1124
1125 assert(path);
1126 assert(ctx);
1127
1128 rcu_read_lock();
1129
1130 if (relayd_id != (uint64_t) -1ULL) {
1131 use_relayd = 1;
1132 }
1133
1134 assert(!channel->monitor);
1135 DBG("UST consumer snapshot channel %" PRIu64, key);
1136
1137 cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
1138 health_code_update();
1139
1140 /* Lock stream because we are about to change its state. */
1141 pthread_mutex_lock(&stream->lock);
1142 assert(channel->trace_chunk);
1143 if (!lttng_trace_chunk_get(channel->trace_chunk)) {
1144 /*
1145 * Can't happen barring an internal error as the channel
1146 * holds a reference to the trace chunk.
1147 */
1148 ERR("Failed to acquire reference to channel's trace chunk");
1149 ret = -1;
1150 goto error_unlock;
1151 }
1152 assert(!stream->trace_chunk);
1153 stream->trace_chunk = channel->trace_chunk;
1154
1155 stream->net_seq_idx = relayd_id;
1156
1157 if (use_relayd) {
1158 ret = consumer_send_relayd_stream(stream, path);
1159 if (ret < 0) {
1160 goto error_unlock;
1161 }
1162 } else {
1163 ret = consumer_stream_create_output_files(stream,
1164 false);
1165 if (ret < 0) {
1166 goto error_unlock;
1167 }
1168 DBG("UST consumer snapshot stream (%" PRIu64 ")",
1169 stream->key);
1170 }
1171
1172 /*
1173 * If tracing is active, we want to perform a "full" buffer flush.
1174 * Else, if quiescent, it has already been done by the prior stop.
1175 */
1176 if (!stream->quiescent) {
1177 ustctl_flush_buffer(stream->ustream, 0);
1178 }
1179
1180 ret = lttng_ustconsumer_take_snapshot(stream);
1181 if (ret < 0) {
1182 ERR("Taking UST snapshot");
1183 goto error_unlock;
1184 }
1185
1186 ret = lttng_ustconsumer_get_produced_snapshot(stream, &produced_pos);
1187 if (ret < 0) {
1188 ERR("Produced UST snapshot position");
1189 goto error_unlock;
1190 }
1191
1192 ret = lttng_ustconsumer_get_consumed_snapshot(stream, &consumed_pos);
1193 if (ret < 0) {
1194 ERR("Consumerd UST snapshot position");
1195 goto error_unlock;
1196 }
1197
1198 /*
1199 * The original value is sent back if max stream size is larger than
1200 * the possible size of the snapshot. Also, we assume that the session
1201 * daemon should never send a maximum stream size that is lower than
1202 * subbuffer size.
1203 */
1204 consumed_pos = consumer_get_consume_start_pos(consumed_pos,
1205 produced_pos, nb_packets_per_stream,
1206 stream->max_sb_size);
1207
1208 while ((long) (consumed_pos - produced_pos) < 0) {
1209 ssize_t read_len;
1210 unsigned long len, padded_len;
1211 const char *subbuf_addr;
1212 struct lttng_buffer_view subbuf_view;
1213
1214 health_code_update();
1215
1216 DBG("UST consumer taking snapshot at pos %lu", consumed_pos);
1217
1218 ret = ustctl_get_subbuf(stream->ustream, &consumed_pos);
1219 if (ret < 0) {
1220 if (ret != -EAGAIN) {
1221 PERROR("ustctl_get_subbuf snapshot");
1222 goto error_close_stream;
1223 }
1224 DBG("UST consumer get subbuf failed. Skipping it.");
1225 consumed_pos += stream->max_sb_size;
1226 stream->chan->lost_packets++;
1227 continue;
1228 }
1229
1230 ret = ustctl_get_subbuf_size(stream->ustream, &len);
1231 if (ret < 0) {
1232 ERR("Snapshot ustctl_get_subbuf_size");
1233 goto error_put_subbuf;
1234 }
1235
1236 ret = ustctl_get_padded_subbuf_size(stream->ustream, &padded_len);
1237 if (ret < 0) {
1238 ERR("Snapshot ustctl_get_padded_subbuf_size");
1239 goto error_put_subbuf;
1240 }
1241
1242 ret = get_current_subbuf_addr(stream, &subbuf_addr);
1243 if (ret) {
1244 goto error_put_subbuf;
1245 }
1246
1247 subbuf_view = lttng_buffer_view_init(
1248 subbuf_addr, 0, padded_len);
1249 read_len = lttng_consumer_on_read_subbuffer_mmap(
1250 stream, &subbuf_view, padded_len - len);
1251 if (use_relayd) {
1252 if (read_len != len) {
1253 ret = -EPERM;
1254 goto error_put_subbuf;
1255 }
1256 } else {
1257 if (read_len != padded_len) {
1258 ret = -EPERM;
1259 goto error_put_subbuf;
1260 }
1261 }
1262
1263 ret = ustctl_put_subbuf(stream->ustream);
1264 if (ret < 0) {
1265 ERR("Snapshot ustctl_put_subbuf");
1266 goto error_close_stream;
1267 }
1268 consumed_pos += stream->max_sb_size;
1269 }
1270
1271 /* Simply close the stream so we can use it on the next snapshot. */
1272 consumer_stream_close(stream);
1273 pthread_mutex_unlock(&stream->lock);
1274 }
1275
1276 rcu_read_unlock();
1277 return 0;
1278
1279 error_put_subbuf:
1280 if (ustctl_put_subbuf(stream->ustream) < 0) {
1281 ERR("Snapshot ustctl_put_subbuf");
1282 }
1283 error_close_stream:
1284 consumer_stream_close(stream);
1285 error_unlock:
1286 pthread_mutex_unlock(&stream->lock);
1287 rcu_read_unlock();
1288 return ret;
1289 }
1290
1291 /*
1292 * Receive the metadata updates from the sessiond. Supports receiving
1293 * overlapping metadata, but is needs to always belong to a contiguous
1294 * range starting from 0.
1295 * Be careful about the locks held when calling this function: it needs
1296 * the metadata cache flush to concurrently progress in order to
1297 * complete.
1298 */
1299 int lttng_ustconsumer_recv_metadata(int sock, uint64_t key, uint64_t offset,
1300 uint64_t len, uint64_t version,
1301 struct lttng_consumer_channel *channel, int timer, int wait)
1302 {
1303 int ret, ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1304 char *metadata_str;
1305
1306 DBG("UST consumer push metadata key %" PRIu64 " of len %" PRIu64, key, len);
1307
1308 metadata_str = zmalloc(len * sizeof(char));
1309 if (!metadata_str) {
1310 PERROR("zmalloc metadata string");
1311 ret_code = LTTCOMM_CONSUMERD_ENOMEM;
1312 goto end;
1313 }
1314
1315 health_code_update();
1316
1317 /* Receive metadata string. */
1318 ret = lttcomm_recv_unix_sock(sock, metadata_str, len);
1319 if (ret < 0) {
1320 /* Session daemon is dead so return gracefully. */
1321 ret_code = ret;
1322 goto end_free;
1323 }
1324
1325 health_code_update();
1326
1327 pthread_mutex_lock(&channel->metadata_cache->lock);
1328 ret = consumer_metadata_cache_write(channel, offset, len, version,
1329 metadata_str);
1330 if (ret < 0) {
1331 /* Unable to handle metadata. Notify session daemon. */
1332 ret_code = LTTCOMM_CONSUMERD_ERROR_METADATA;
1333 /*
1334 * Skip metadata flush on write error since the offset and len might
1335 * not have been updated which could create an infinite loop below when
1336 * waiting for the metadata cache to be flushed.
1337 */
1338 pthread_mutex_unlock(&channel->metadata_cache->lock);
1339 goto end_free;
1340 }
1341 pthread_mutex_unlock(&channel->metadata_cache->lock);
1342
1343 if (!wait) {
1344 goto end_free;
1345 }
1346 while (consumer_metadata_cache_flushed(channel, offset + len, timer)) {
1347 DBG("Waiting for metadata to be flushed");
1348
1349 health_code_update();
1350
1351 usleep(DEFAULT_METADATA_AVAILABILITY_WAIT_TIME);
1352 }
1353
1354 end_free:
1355 free(metadata_str);
1356 end:
1357 return ret_code;
1358 }
1359
1360 /*
1361 * Receive command from session daemon and process it.
1362 *
1363 * Return 1 on success else a negative value or 0.
1364 */
1365 int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
1366 int sock, struct pollfd *consumer_sockpoll)
1367 {
1368 ssize_t ret;
1369 enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1370 struct lttcomm_consumer_msg msg;
1371 struct lttng_consumer_channel *channel = NULL;
1372
1373 health_code_update();
1374
1375 ret = lttcomm_recv_unix_sock(sock, &msg, sizeof(msg));
1376 if (ret != sizeof(msg)) {
1377 DBG("Consumer received unexpected message size %zd (expects %zu)",
1378 ret, sizeof(msg));
1379 /*
1380 * The ret value might 0 meaning an orderly shutdown but this is ok
1381 * since the caller handles this.
1382 */
1383 if (ret > 0) {
1384 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD);
1385 ret = -1;
1386 }
1387 return ret;
1388 }
1389
1390 health_code_update();
1391
1392 /* deprecated */
1393 assert(msg.cmd_type != LTTNG_CONSUMER_STOP);
1394
1395 health_code_update();
1396
1397 /* relayd needs RCU read-side lock */
1398 rcu_read_lock();
1399
1400 switch (msg.cmd_type) {
1401 case LTTNG_CONSUMER_ADD_RELAYD_SOCKET:
1402 {
1403 /* Session daemon status message are handled in the following call. */
1404 consumer_add_relayd_socket(msg.u.relayd_sock.net_index,
1405 msg.u.relayd_sock.type, ctx, sock, consumer_sockpoll,
1406 &msg.u.relayd_sock.sock, msg.u.relayd_sock.session_id,
1407 msg.u.relayd_sock.relayd_session_id);
1408 goto end_nosignal;
1409 }
1410 case LTTNG_CONSUMER_DESTROY_RELAYD:
1411 {
1412 uint64_t index = msg.u.destroy_relayd.net_seq_idx;
1413 struct consumer_relayd_sock_pair *relayd;
1414
1415 DBG("UST consumer destroying relayd %" PRIu64, index);
1416
1417 /* Get relayd reference if exists. */
1418 relayd = consumer_find_relayd(index);
1419 if (relayd == NULL) {
1420 DBG("Unable to find relayd %" PRIu64, index);
1421 ret_code = LTTCOMM_CONSUMERD_RELAYD_FAIL;
1422 }
1423
1424 /*
1425 * Each relayd socket pair has a refcount of stream attached to it
1426 * which tells if the relayd is still active or not depending on the
1427 * refcount value.
1428 *
1429 * This will set the destroy flag of the relayd object and destroy it
1430 * if the refcount reaches zero when called.
1431 *
1432 * The destroy can happen either here or when a stream fd hangs up.
1433 */
1434 if (relayd) {
1435 consumer_flag_relayd_for_destroy(relayd);
1436 }
1437
1438 goto end_msg_sessiond;
1439 }
1440 case LTTNG_CONSUMER_UPDATE_STREAM:
1441 {
1442 rcu_read_unlock();
1443 return -ENOSYS;
1444 }
1445 case LTTNG_CONSUMER_DATA_PENDING:
1446 {
1447 int ret, is_data_pending;
1448 uint64_t id = msg.u.data_pending.session_id;
1449
1450 DBG("UST consumer data pending command for id %" PRIu64, id);
1451
1452 is_data_pending = consumer_data_pending(id);
1453
1454 /* Send back returned value to session daemon */
1455 ret = lttcomm_send_unix_sock(sock, &is_data_pending,
1456 sizeof(is_data_pending));
1457 if (ret < 0) {
1458 DBG("Error when sending the data pending ret code: %d", ret);
1459 goto error_fatal;
1460 }
1461
1462 /*
1463 * No need to send back a status message since the data pending
1464 * returned value is the response.
1465 */
1466 break;
1467 }
1468 case LTTNG_CONSUMER_ASK_CHANNEL_CREATION:
1469 {
1470 int ret;
1471 struct ustctl_consumer_channel_attr attr;
1472 const uint64_t chunk_id = msg.u.ask_channel.chunk_id.value;
1473 const struct lttng_credentials buffer_credentials = {
1474 .uid = msg.u.ask_channel.buffer_credentials.uid,
1475 .gid = msg.u.ask_channel.buffer_credentials.gid,
1476 };
1477
1478 /* Create a plain object and reserve a channel key. */
1479 channel = consumer_allocate_channel(
1480 msg.u.ask_channel.key,
1481 msg.u.ask_channel.session_id,
1482 msg.u.ask_channel.chunk_id.is_set ?
1483 &chunk_id : NULL,
1484 msg.u.ask_channel.pathname,
1485 msg.u.ask_channel.name,
1486 msg.u.ask_channel.relayd_id,
1487 (enum lttng_event_output) msg.u.ask_channel.output,
1488 msg.u.ask_channel.tracefile_size,
1489 msg.u.ask_channel.tracefile_count,
1490 msg.u.ask_channel.session_id_per_pid,
1491 msg.u.ask_channel.monitor,
1492 msg.u.ask_channel.live_timer_interval,
1493 msg.u.ask_channel.is_live,
1494 msg.u.ask_channel.root_shm_path,
1495 msg.u.ask_channel.shm_path);
1496 if (!channel) {
1497 goto end_channel_error;
1498 }
1499
1500 LTTNG_OPTIONAL_SET(&channel->buffer_credentials,
1501 buffer_credentials);
1502
1503 /*
1504 * Assign UST application UID to the channel. This value is ignored for
1505 * per PID buffers. This is specific to UST thus setting this after the
1506 * allocation.
1507 */
1508 channel->ust_app_uid = msg.u.ask_channel.ust_app_uid;
1509
1510 /* Build channel attributes from received message. */
1511 attr.subbuf_size = msg.u.ask_channel.subbuf_size;
1512 attr.num_subbuf = msg.u.ask_channel.num_subbuf;
1513 attr.overwrite = msg.u.ask_channel.overwrite;
1514 attr.switch_timer_interval = msg.u.ask_channel.switch_timer_interval;
1515 attr.read_timer_interval = msg.u.ask_channel.read_timer_interval;
1516 attr.chan_id = msg.u.ask_channel.chan_id;
1517 memcpy(attr.uuid, msg.u.ask_channel.uuid, sizeof(attr.uuid));
1518 attr.blocking_timeout= msg.u.ask_channel.blocking_timeout;
1519
1520 /* Match channel buffer type to the UST abi. */
1521 switch (msg.u.ask_channel.output) {
1522 case LTTNG_EVENT_MMAP:
1523 default:
1524 attr.output = LTTNG_UST_MMAP;
1525 break;
1526 }
1527
1528 /* Translate and save channel type. */
1529 switch (msg.u.ask_channel.type) {
1530 case LTTNG_UST_CHAN_PER_CPU:
1531 channel->type = CONSUMER_CHANNEL_TYPE_DATA;
1532 attr.type = LTTNG_UST_CHAN_PER_CPU;
1533 /*
1534 * Set refcount to 1 for owner. Below, we will
1535 * pass ownership to the
1536 * consumer_thread_channel_poll() thread.
1537 */
1538 channel->refcount = 1;
1539 break;
1540 case LTTNG_UST_CHAN_METADATA:
1541 channel->type = CONSUMER_CHANNEL_TYPE_METADATA;
1542 attr.type = LTTNG_UST_CHAN_METADATA;
1543 break;
1544 default:
1545 assert(0);
1546 goto error_fatal;
1547 };
1548
1549 health_code_update();
1550
1551 ret = ask_channel(ctx, channel, &attr);
1552 if (ret < 0) {
1553 goto end_channel_error;
1554 }
1555
1556 if (msg.u.ask_channel.type == LTTNG_UST_CHAN_METADATA) {
1557 ret = consumer_metadata_cache_allocate(channel);
1558 if (ret < 0) {
1559 ERR("Allocating metadata cache");
1560 goto end_channel_error;
1561 }
1562 consumer_timer_switch_start(channel, attr.switch_timer_interval);
1563 attr.switch_timer_interval = 0;
1564 } else {
1565 int monitor_start_ret;
1566
1567 consumer_timer_live_start(channel,
1568 msg.u.ask_channel.live_timer_interval);
1569 monitor_start_ret = consumer_timer_monitor_start(
1570 channel,
1571 msg.u.ask_channel.monitor_timer_interval);
1572 if (monitor_start_ret < 0) {
1573 ERR("Starting channel monitoring timer failed");
1574 goto end_channel_error;
1575 }
1576 }
1577
1578 health_code_update();
1579
1580 /*
1581 * Add the channel to the internal state AFTER all streams were created
1582 * and successfully sent to session daemon. This way, all streams must
1583 * be ready before this channel is visible to the threads.
1584 * If add_channel succeeds, ownership of the channel is
1585 * passed to consumer_thread_channel_poll().
1586 */
1587 ret = add_channel(channel, ctx);
1588 if (ret < 0) {
1589 if (msg.u.ask_channel.type == LTTNG_UST_CHAN_METADATA) {
1590 if (channel->switch_timer_enabled == 1) {
1591 consumer_timer_switch_stop(channel);
1592 }
1593 consumer_metadata_cache_destroy(channel);
1594 }
1595 if (channel->live_timer_enabled == 1) {
1596 consumer_timer_live_stop(channel);
1597 }
1598 if (channel->monitor_timer_enabled == 1) {
1599 consumer_timer_monitor_stop(channel);
1600 }
1601 goto end_channel_error;
1602 }
1603
1604 health_code_update();
1605
1606 /*
1607 * Channel and streams are now created. Inform the session daemon that
1608 * everything went well and should wait to receive the channel and
1609 * streams with ustctl API.
1610 */
1611 ret = consumer_send_status_channel(sock, channel);
1612 if (ret < 0) {
1613 /*
1614 * There is probably a problem on the socket.
1615 */
1616 goto error_fatal;
1617 }
1618
1619 break;
1620 }
1621 case LTTNG_CONSUMER_GET_CHANNEL:
1622 {
1623 int ret, relayd_err = 0;
1624 uint64_t key = msg.u.get_channel.key;
1625 struct lttng_consumer_channel *channel;
1626
1627 channel = consumer_find_channel(key);
1628 if (!channel) {
1629 ERR("UST consumer get channel key %" PRIu64 " not found", key);
1630 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
1631 goto end_get_channel;
1632 }
1633
1634 health_code_update();
1635
1636 /* Send the channel to sessiond (and relayd, if applicable). */
1637 ret = send_channel_to_sessiond_and_relayd(sock, channel, ctx,
1638 &relayd_err);
1639 if (ret < 0) {
1640 if (relayd_err) {
1641 /*
1642 * We were unable to send to the relayd the stream so avoid
1643 * sending back a fatal error to the thread since this is OK
1644 * and the consumer can continue its work. The above call
1645 * has sent the error status message to the sessiond.
1646 */
1647 goto end_get_channel_nosignal;
1648 }
1649 /*
1650 * The communicaton was broken hence there is a bad state between
1651 * the consumer and sessiond so stop everything.
1652 */
1653 goto error_get_channel_fatal;
1654 }
1655
1656 health_code_update();
1657
1658 /*
1659 * In no monitor mode, the streams ownership is kept inside the channel
1660 * so don't send them to the data thread.
1661 */
1662 if (!channel->monitor) {
1663 goto end_get_channel;
1664 }
1665
1666 ret = send_streams_to_thread(channel, ctx);
1667 if (ret < 0) {
1668 /*
1669 * If we are unable to send the stream to the thread, there is
1670 * a big problem so just stop everything.
1671 */
1672 goto error_get_channel_fatal;
1673 }
1674 /* List MUST be empty after or else it could be reused. */
1675 assert(cds_list_empty(&channel->streams.head));
1676 end_get_channel:
1677 goto end_msg_sessiond;
1678 error_get_channel_fatal:
1679 goto error_fatal;
1680 end_get_channel_nosignal:
1681 goto end_nosignal;
1682 }
1683 case LTTNG_CONSUMER_DESTROY_CHANNEL:
1684 {
1685 uint64_t key = msg.u.destroy_channel.key;
1686
1687 /*
1688 * Only called if streams have not been sent to stream
1689 * manager thread. However, channel has been sent to
1690 * channel manager thread.
1691 */
1692 notify_thread_del_channel(ctx, key);
1693 goto end_msg_sessiond;
1694 }
1695 case LTTNG_CONSUMER_CLOSE_METADATA:
1696 {
1697 int ret;
1698
1699 ret = close_metadata(msg.u.close_metadata.key);
1700 if (ret != 0) {
1701 ret_code = ret;
1702 }
1703
1704 goto end_msg_sessiond;
1705 }
1706 case LTTNG_CONSUMER_FLUSH_CHANNEL:
1707 {
1708 int ret;
1709
1710 ret = flush_channel(msg.u.flush_channel.key);
1711 if (ret != 0) {
1712 ret_code = ret;
1713 }
1714
1715 goto end_msg_sessiond;
1716 }
1717 case LTTNG_CONSUMER_CLEAR_QUIESCENT_CHANNEL:
1718 {
1719 int ret;
1720
1721 ret = clear_quiescent_channel(
1722 msg.u.clear_quiescent_channel.key);
1723 if (ret != 0) {
1724 ret_code = ret;
1725 }
1726
1727 goto end_msg_sessiond;
1728 }
1729 case LTTNG_CONSUMER_PUSH_METADATA:
1730 {
1731 int ret;
1732 uint64_t len = msg.u.push_metadata.len;
1733 uint64_t key = msg.u.push_metadata.key;
1734 uint64_t offset = msg.u.push_metadata.target_offset;
1735 uint64_t version = msg.u.push_metadata.version;
1736 struct lttng_consumer_channel *channel;
1737
1738 DBG("UST consumer push metadata key %" PRIu64 " of len %" PRIu64, key,
1739 len);
1740
1741 channel = consumer_find_channel(key);
1742 if (!channel) {
1743 /*
1744 * This is possible if the metadata creation on the consumer side
1745 * is in flight vis-a-vis a concurrent push metadata from the
1746 * session daemon. Simply return that the channel failed and the
1747 * session daemon will handle that message correctly considering
1748 * that this race is acceptable thus the DBG() statement here.
1749 */
1750 DBG("UST consumer push metadata %" PRIu64 " not found", key);
1751 ret_code = LTTCOMM_CONSUMERD_CHANNEL_FAIL;
1752 goto end_push_metadata_msg_sessiond;
1753 }
1754
1755 health_code_update();
1756
1757 if (!len) {
1758 /*
1759 * There is nothing to receive. We have simply
1760 * checked whether the channel can be found.
1761 */
1762 ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1763 goto end_push_metadata_msg_sessiond;
1764 }
1765
1766 /* Tell session daemon we are ready to receive the metadata. */
1767 ret = consumer_send_status_msg(sock, LTTCOMM_CONSUMERD_SUCCESS);
1768 if (ret < 0) {
1769 /* Somehow, the session daemon is not responding anymore. */
1770 goto error_push_metadata_fatal;
1771 }
1772
1773 health_code_update();
1774
1775 /* Wait for more data. */
1776 health_poll_entry();
1777 ret = lttng_consumer_poll_socket(consumer_sockpoll);
1778 health_poll_exit();
1779 if (ret) {
1780 goto error_push_metadata_fatal;
1781 }
1782
1783 health_code_update();
1784
1785 ret = lttng_ustconsumer_recv_metadata(sock, key, offset,
1786 len, version, channel, 0, 1);
1787 if (ret < 0) {
1788 /* error receiving from sessiond */
1789 goto error_push_metadata_fatal;
1790 } else {
1791 ret_code = ret;
1792 goto end_push_metadata_msg_sessiond;
1793 }
1794 end_push_metadata_msg_sessiond:
1795 goto end_msg_sessiond;
1796 error_push_metadata_fatal:
1797 goto error_fatal;
1798 }
1799 case LTTNG_CONSUMER_SETUP_METADATA:
1800 {
1801 int ret;
1802
1803 ret = setup_metadata(ctx, msg.u.setup_metadata.key);
1804 if (ret) {
1805 ret_code = ret;
1806 }
1807 goto end_msg_sessiond;
1808 }
1809 case LTTNG_CONSUMER_SNAPSHOT_CHANNEL:
1810 {
1811 struct lttng_consumer_channel *channel;
1812 uint64_t key = msg.u.snapshot_channel.key;
1813
1814 channel = consumer_find_channel(key);
1815 if (!channel) {
1816 DBG("UST snapshot channel not found for key %" PRIu64, key);
1817 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
1818 } else {
1819 if (msg.u.snapshot_channel.metadata) {
1820 ret = snapshot_metadata(channel, key,
1821 msg.u.snapshot_channel.pathname,
1822 msg.u.snapshot_channel.relayd_id,
1823 ctx);
1824 if (ret < 0) {
1825 ERR("Snapshot metadata failed");
1826 ret_code = LTTCOMM_CONSUMERD_SNAPSHOT_FAILED;
1827 }
1828 } else {
1829 ret = snapshot_channel(channel, key,
1830 msg.u.snapshot_channel.pathname,
1831 msg.u.snapshot_channel.relayd_id,
1832 msg.u.snapshot_channel.nb_packets_per_stream,
1833 ctx);
1834 if (ret < 0) {
1835 ERR("Snapshot channel failed");
1836 ret_code = LTTCOMM_CONSUMERD_SNAPSHOT_FAILED;
1837 }
1838 }
1839 }
1840 health_code_update();
1841 ret = consumer_send_status_msg(sock, ret_code);
1842 if (ret < 0) {
1843 /* Somehow, the session daemon is not responding anymore. */
1844 goto end_nosignal;
1845 }
1846 health_code_update();
1847 break;
1848 }
1849 case LTTNG_CONSUMER_DISCARDED_EVENTS:
1850 {
1851 int ret = 0;
1852 uint64_t discarded_events;
1853 struct lttng_ht_iter iter;
1854 struct lttng_ht *ht;
1855 struct lttng_consumer_stream *stream;
1856 uint64_t id = msg.u.discarded_events.session_id;
1857 uint64_t key = msg.u.discarded_events.channel_key;
1858
1859 DBG("UST consumer discarded events command for session id %"
1860 PRIu64, id);
1861 rcu_read_lock();
1862 pthread_mutex_lock(&consumer_data.lock);
1863
1864 ht = consumer_data.stream_list_ht;
1865
1866 /*
1867 * We only need a reference to the channel, but they are not
1868 * directly indexed, so we just use the first matching stream
1869 * to extract the information we need, we default to 0 if not
1870 * found (no events are dropped if the channel is not yet in
1871 * use).
1872 */
1873 discarded_events = 0;
1874 cds_lfht_for_each_entry_duplicate(ht->ht,
1875 ht->hash_fct(&id, lttng_ht_seed),
1876 ht->match_fct, &id,
1877 &iter.iter, stream, node_session_id.node) {
1878 if (stream->chan->key == key) {
1879 discarded_events = stream->chan->discarded_events;
1880 break;
1881 }
1882 }
1883 pthread_mutex_unlock(&consumer_data.lock);
1884 rcu_read_unlock();
1885
1886 DBG("UST consumer discarded events command for session id %"
1887 PRIu64 ", channel key %" PRIu64, id, key);
1888
1889 health_code_update();
1890
1891 /* Send back returned value to session daemon */
1892 ret = lttcomm_send_unix_sock(sock, &discarded_events, sizeof(discarded_events));
1893 if (ret < 0) {
1894 PERROR("send discarded events");
1895 goto error_fatal;
1896 }
1897
1898 break;
1899 }
1900 case LTTNG_CONSUMER_LOST_PACKETS:
1901 {
1902 int ret;
1903 uint64_t lost_packets;
1904 struct lttng_ht_iter iter;
1905 struct lttng_ht *ht;
1906 struct lttng_consumer_stream *stream;
1907 uint64_t id = msg.u.lost_packets.session_id;
1908 uint64_t key = msg.u.lost_packets.channel_key;
1909
1910 DBG("UST consumer lost packets command for session id %"
1911 PRIu64, id);
1912 rcu_read_lock();
1913 pthread_mutex_lock(&consumer_data.lock);
1914
1915 ht = consumer_data.stream_list_ht;
1916
1917 /*
1918 * We only need a reference to the channel, but they are not
1919 * directly indexed, so we just use the first matching stream
1920 * to extract the information we need, we default to 0 if not
1921 * found (no packets lost if the channel is not yet in use).
1922 */
1923 lost_packets = 0;
1924 cds_lfht_for_each_entry_duplicate(ht->ht,
1925 ht->hash_fct(&id, lttng_ht_seed),
1926 ht->match_fct, &id,
1927 &iter.iter, stream, node_session_id.node) {
1928 if (stream->chan->key == key) {
1929 lost_packets = stream->chan->lost_packets;
1930 break;
1931 }
1932 }
1933 pthread_mutex_unlock(&consumer_data.lock);
1934 rcu_read_unlock();
1935
1936 DBG("UST consumer lost packets command for session id %"
1937 PRIu64 ", channel key %" PRIu64, id, key);
1938
1939 health_code_update();
1940
1941 /* Send back returned value to session daemon */
1942 ret = lttcomm_send_unix_sock(sock, &lost_packets,
1943 sizeof(lost_packets));
1944 if (ret < 0) {
1945 PERROR("send lost packets");
1946 goto error_fatal;
1947 }
1948
1949 break;
1950 }
1951 case LTTNG_CONSUMER_SET_CHANNEL_MONITOR_PIPE:
1952 {
1953 int channel_monitor_pipe;
1954
1955 ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1956 /* Successfully received the command's type. */
1957 ret = consumer_send_status_msg(sock, ret_code);
1958 if (ret < 0) {
1959 goto error_fatal;
1960 }
1961
1962 ret = lttcomm_recv_fds_unix_sock(sock, &channel_monitor_pipe,
1963 1);
1964 if (ret != sizeof(channel_monitor_pipe)) {
1965 ERR("Failed to receive channel monitor pipe");
1966 goto error_fatal;
1967 }
1968
1969 DBG("Received channel monitor pipe (%d)", channel_monitor_pipe);
1970 ret = consumer_timer_thread_set_channel_monitor_pipe(
1971 channel_monitor_pipe);
1972 if (!ret) {
1973 int flags;
1974
1975 ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1976 /* Set the pipe as non-blocking. */
1977 ret = fcntl(channel_monitor_pipe, F_GETFL, 0);
1978 if (ret == -1) {
1979 PERROR("fcntl get flags of the channel monitoring pipe");
1980 goto error_fatal;
1981 }
1982 flags = ret;
1983
1984 ret = fcntl(channel_monitor_pipe, F_SETFL,
1985 flags | O_NONBLOCK);
1986 if (ret == -1) {
1987 PERROR("fcntl set O_NONBLOCK flag of the channel monitoring pipe");
1988 goto error_fatal;
1989 }
1990 DBG("Channel monitor pipe set as non-blocking");
1991 } else {
1992 ret_code = LTTCOMM_CONSUMERD_ALREADY_SET;
1993 }
1994 goto end_msg_sessiond;
1995 }
1996 case LTTNG_CONSUMER_ROTATE_CHANNEL:
1997 {
1998 struct lttng_consumer_channel *channel;
1999 uint64_t key = msg.u.rotate_channel.key;
2000
2001 channel = consumer_find_channel(key);
2002 if (!channel) {
2003 DBG("Channel %" PRIu64 " not found", key);
2004 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
2005 } else {
2006 /*
2007 * Sample the rotate position of all the streams in
2008 * this channel.
2009 */
2010 ret = lttng_consumer_rotate_channel(channel, key,
2011 msg.u.rotate_channel.relayd_id,
2012 msg.u.rotate_channel.metadata,
2013 ctx);
2014 if (ret < 0) {
2015 ERR("Rotate channel failed");
2016 ret_code = LTTCOMM_CONSUMERD_ROTATION_FAIL;
2017 }
2018
2019 health_code_update();
2020 }
2021 ret = consumer_send_status_msg(sock, ret_code);
2022 if (ret < 0) {
2023 /* Somehow, the session daemon is not responding anymore. */
2024 goto end_rotate_channel_nosignal;
2025 }
2026
2027 /*
2028 * Rotate the streams that are ready right now.
2029 * FIXME: this is a second consecutive iteration over the
2030 * streams in a channel, there is probably a better way to
2031 * handle this, but it needs to be after the
2032 * consumer_send_status_msg() call.
2033 */
2034 if (channel) {
2035 ret = lttng_consumer_rotate_ready_streams(
2036 channel, key, ctx);
2037 if (ret < 0) {
2038 ERR("Rotate channel failed");
2039 }
2040 }
2041 break;
2042 end_rotate_channel_nosignal:
2043 goto end_nosignal;
2044 }
2045 case LTTNG_CONSUMER_INIT:
2046 {
2047 ret_code = lttng_consumer_init_command(ctx,
2048 msg.u.init.sessiond_uuid);
2049 health_code_update();
2050 ret = consumer_send_status_msg(sock, ret_code);
2051 if (ret < 0) {
2052 /* Somehow, the session daemon is not responding anymore. */
2053 goto end_nosignal;
2054 }
2055 break;
2056 }
2057 case LTTNG_CONSUMER_CREATE_TRACE_CHUNK:
2058 {
2059 const struct lttng_credentials credentials = {
2060 .uid = msg.u.create_trace_chunk.credentials.value.uid,
2061 .gid = msg.u.create_trace_chunk.credentials.value.gid,
2062 };
2063 const bool is_local_trace =
2064 !msg.u.create_trace_chunk.relayd_id.is_set;
2065 const uint64_t relayd_id =
2066 msg.u.create_trace_chunk.relayd_id.value;
2067 const char *chunk_override_name =
2068 *msg.u.create_trace_chunk.override_name ?
2069 msg.u.create_trace_chunk.override_name :
2070 NULL;
2071 LTTNG_OPTIONAL(struct lttng_directory_handle) chunk_directory_handle =
2072 LTTNG_OPTIONAL_INIT;
2073
2074 /*
2075 * The session daemon will only provide a chunk directory file
2076 * descriptor for local traces.
2077 */
2078 if (is_local_trace) {
2079 int chunk_dirfd;
2080
2081 /* Acnowledge the reception of the command. */
2082 ret = consumer_send_status_msg(sock,
2083 LTTCOMM_CONSUMERD_SUCCESS);
2084 if (ret < 0) {
2085 /* Somehow, the session daemon is not responding anymore. */
2086 goto end_nosignal;
2087 }
2088
2089 ret = lttcomm_recv_fds_unix_sock(sock, &chunk_dirfd, 1);
2090 if (ret != sizeof(chunk_dirfd)) {
2091 ERR("Failed to receive trace chunk directory file descriptor");
2092 goto error_fatal;
2093 }
2094
2095 DBG("Received trace chunk directory fd (%d)",
2096 chunk_dirfd);
2097 ret = lttng_directory_handle_init_from_dirfd(
2098 &chunk_directory_handle.value,
2099 chunk_dirfd);
2100 if (ret) {
2101 ERR("Failed to initialize chunk directory handle from directory file descriptor");
2102 if (close(chunk_dirfd)) {
2103 PERROR("Failed to close chunk directory file descriptor");
2104 }
2105 goto error_fatal;
2106 }
2107 chunk_directory_handle.is_set = true;
2108 }
2109
2110 ret_code = lttng_consumer_create_trace_chunk(
2111 !is_local_trace ? &relayd_id : NULL,
2112 msg.u.create_trace_chunk.session_id,
2113 msg.u.create_trace_chunk.chunk_id,
2114 (time_t) msg.u.create_trace_chunk
2115 .creation_timestamp,
2116 chunk_override_name,
2117 msg.u.create_trace_chunk.credentials.is_set ?
2118 &credentials :
2119 NULL,
2120 chunk_directory_handle.is_set ?
2121 &chunk_directory_handle.value :
2122 NULL);
2123
2124 if (chunk_directory_handle.is_set) {
2125 lttng_directory_handle_fini(
2126 &chunk_directory_handle.value);
2127 }
2128 goto end_msg_sessiond;
2129 }
2130 case LTTNG_CONSUMER_CLOSE_TRACE_CHUNK:
2131 {
2132 enum lttng_trace_chunk_command_type close_command =
2133 msg.u.close_trace_chunk.close_command.value;
2134 const uint64_t relayd_id =
2135 msg.u.close_trace_chunk.relayd_id.value;
2136 struct lttcomm_consumer_close_trace_chunk_reply reply;
2137 char closed_trace_chunk_path[LTTNG_PATH_MAX];
2138 int ret;
2139
2140 ret_code = lttng_consumer_close_trace_chunk(
2141 msg.u.close_trace_chunk.relayd_id.is_set ?
2142 &relayd_id :
2143 NULL,
2144 msg.u.close_trace_chunk.session_id,
2145 msg.u.close_trace_chunk.chunk_id,
2146 (time_t) msg.u.close_trace_chunk.close_timestamp,
2147 msg.u.close_trace_chunk.close_command.is_set ?
2148 &close_command :
2149 NULL, closed_trace_chunk_path);
2150 reply.ret_code = ret_code;
2151 reply.path_length = strlen(closed_trace_chunk_path) + 1;
2152 ret = lttcomm_send_unix_sock(sock, &reply, sizeof(reply));
2153 if (ret != sizeof(reply)) {
2154 goto error_fatal;
2155 }
2156 ret = lttcomm_send_unix_sock(sock, closed_trace_chunk_path,
2157 reply.path_length);
2158 if (ret != reply.path_length) {
2159 goto error_fatal;
2160 }
2161 goto end_nosignal;
2162 }
2163 case LTTNG_CONSUMER_TRACE_CHUNK_EXISTS:
2164 {
2165 const uint64_t relayd_id =
2166 msg.u.trace_chunk_exists.relayd_id.value;
2167
2168 ret_code = lttng_consumer_trace_chunk_exists(
2169 msg.u.trace_chunk_exists.relayd_id.is_set ?
2170 &relayd_id : NULL,
2171 msg.u.trace_chunk_exists.session_id,
2172 msg.u.trace_chunk_exists.chunk_id);
2173 goto end_msg_sessiond;
2174 }
2175 default:
2176 break;
2177 }
2178
2179 end_nosignal:
2180 /*
2181 * Return 1 to indicate success since the 0 value can be a socket
2182 * shutdown during the recv() or send() call.
2183 */
2184 ret = 1;
2185 goto end;
2186
2187 end_msg_sessiond:
2188 /*
2189 * The returned value here is not useful since either way we'll return 1 to
2190 * the caller because the session daemon socket management is done
2191 * elsewhere. Returning a negative code or 0 will shutdown the consumer.
2192 */
2193 ret = consumer_send_status_msg(sock, ret_code);
2194 if (ret < 0) {
2195 goto error_fatal;
2196 }
2197 ret = 1;
2198 goto end;
2199
2200 end_channel_error:
2201 if (channel) {
2202 /*
2203 * Free channel here since no one has a reference to it. We don't
2204 * free after that because a stream can store this pointer.
2205 */
2206 destroy_channel(channel);
2207 }
2208 /* We have to send a status channel message indicating an error. */
2209 ret = consumer_send_status_channel(sock, NULL);
2210 if (ret < 0) {
2211 /* Stop everything if session daemon can not be notified. */
2212 goto error_fatal;
2213 }
2214 ret = 1;
2215 goto end;
2216
2217 error_fatal:
2218 /* This will issue a consumer stop. */
2219 ret = -1;
2220 goto end;
2221
2222 end:
2223 rcu_read_unlock();
2224 health_code_update();
2225 return ret;
2226 }
2227
2228 void lttng_ustctl_flush_buffer(struct lttng_consumer_stream *stream,
2229 int producer_active)
2230 {
2231 assert(stream);
2232 assert(stream->ustream);
2233
2234 ustctl_flush_buffer(stream->ustream, producer_active);
2235 }
2236
2237 /*
2238 * Take a snapshot for a specific stream.
2239 *
2240 * Returns 0 on success, < 0 on error
2241 */
2242 int lttng_ustconsumer_take_snapshot(struct lttng_consumer_stream *stream)
2243 {
2244 assert(stream);
2245 assert(stream->ustream);
2246
2247 return ustctl_snapshot(stream->ustream);
2248 }
2249
2250 /*
2251 * Sample consumed and produced positions for a specific stream.
2252 *
2253 * Returns 0 on success, < 0 on error.
2254 */
2255 int lttng_ustconsumer_sample_snapshot_positions(
2256 struct lttng_consumer_stream *stream)
2257 {
2258 assert(stream);
2259 assert(stream->ustream);
2260
2261 return ustctl_snapshot_sample_positions(stream->ustream);
2262 }
2263
2264 /*
2265 * Get the produced position
2266 *
2267 * Returns 0 on success, < 0 on error
2268 */
2269 int lttng_ustconsumer_get_produced_snapshot(
2270 struct lttng_consumer_stream *stream, unsigned long *pos)
2271 {
2272 assert(stream);
2273 assert(stream->ustream);
2274 assert(pos);
2275
2276 return ustctl_snapshot_get_produced(stream->ustream, pos);
2277 }
2278
2279 /*
2280 * Get the consumed position
2281 *
2282 * Returns 0 on success, < 0 on error
2283 */
2284 int lttng_ustconsumer_get_consumed_snapshot(
2285 struct lttng_consumer_stream *stream, unsigned long *pos)
2286 {
2287 assert(stream);
2288 assert(stream->ustream);
2289 assert(pos);
2290
2291 return ustctl_snapshot_get_consumed(stream->ustream, pos);
2292 }
2293
2294 void lttng_ustconsumer_flush_buffer(struct lttng_consumer_stream *stream,
2295 int producer)
2296 {
2297 assert(stream);
2298 assert(stream->ustream);
2299
2300 ustctl_flush_buffer(stream->ustream, producer);
2301 }
2302
2303 int lttng_ustconsumer_get_current_timestamp(
2304 struct lttng_consumer_stream *stream, uint64_t *ts)
2305 {
2306 assert(stream);
2307 assert(stream->ustream);
2308 assert(ts);
2309
2310 return ustctl_get_current_timestamp(stream->ustream, ts);
2311 }
2312
2313 int lttng_ustconsumer_get_sequence_number(
2314 struct lttng_consumer_stream *stream, uint64_t *seq)
2315 {
2316 assert(stream);
2317 assert(stream->ustream);
2318 assert(seq);
2319
2320 return ustctl_get_sequence_number(stream->ustream, seq);
2321 }
2322
2323 /*
2324 * Called when the stream signals the consumer that it has hung up.
2325 */
2326 void lttng_ustconsumer_on_stream_hangup(struct lttng_consumer_stream *stream)
2327 {
2328 assert(stream);
2329 assert(stream->ustream);
2330
2331 pthread_mutex_lock(&stream->lock);
2332 if (!stream->quiescent) {
2333 ustctl_flush_buffer(stream->ustream, 0);
2334 stream->quiescent = true;
2335 }
2336 pthread_mutex_unlock(&stream->lock);
2337 stream->hangup_flush_done = 1;
2338 }
2339
2340 void lttng_ustconsumer_del_channel(struct lttng_consumer_channel *chan)
2341 {
2342 int i;
2343
2344 assert(chan);
2345 assert(chan->uchan);
2346 assert(chan->buffer_credentials.is_set);
2347
2348 if (chan->switch_timer_enabled == 1) {
2349 consumer_timer_switch_stop(chan);
2350 }
2351 for (i = 0; i < chan->nr_stream_fds; i++) {
2352 int ret;
2353
2354 ret = close(chan->stream_fds[i]);
2355 if (ret) {
2356 PERROR("close");
2357 }
2358 if (chan->shm_path[0]) {
2359 char shm_path[PATH_MAX];
2360
2361 ret = get_stream_shm_path(shm_path, chan->shm_path, i);
2362 if (ret) {
2363 ERR("Cannot get stream shm path");
2364 }
2365 ret = run_as_unlink(shm_path,
2366 chan->buffer_credentials.value.uid,
2367 chan->buffer_credentials.value.gid);
2368 if (ret) {
2369 PERROR("unlink %s", shm_path);
2370 }
2371 }
2372 }
2373 }
2374
2375 void lttng_ustconsumer_free_channel(struct lttng_consumer_channel *chan)
2376 {
2377 assert(chan);
2378 assert(chan->uchan);
2379 assert(chan->buffer_credentials.is_set);
2380
2381 consumer_metadata_cache_destroy(chan);
2382 ustctl_destroy_channel(chan->uchan);
2383 /* Try to rmdir all directories under shm_path root. */
2384 if (chan->root_shm_path[0]) {
2385 (void) run_as_rmdir_recursive(chan->root_shm_path,
2386 chan->buffer_credentials.value.uid,
2387 chan->buffer_credentials.value.gid,
2388 LTTNG_DIRECTORY_HANDLE_SKIP_NON_EMPTY_FLAG);
2389 }
2390 free(chan->stream_fds);
2391 }
2392
2393 void lttng_ustconsumer_del_stream(struct lttng_consumer_stream *stream)
2394 {
2395 assert(stream);
2396 assert(stream->ustream);
2397
2398 if (stream->chan->switch_timer_enabled == 1) {
2399 consumer_timer_switch_stop(stream->chan);
2400 }
2401 ustctl_destroy_stream(stream->ustream);
2402 }
2403
2404 int lttng_ustconsumer_get_wakeup_fd(struct lttng_consumer_stream *stream)
2405 {
2406 assert(stream);
2407 assert(stream->ustream);
2408
2409 return ustctl_stream_get_wakeup_fd(stream->ustream);
2410 }
2411
2412 int lttng_ustconsumer_close_wakeup_fd(struct lttng_consumer_stream *stream)
2413 {
2414 assert(stream);
2415 assert(stream->ustream);
2416
2417 return ustctl_stream_close_wakeup_fd(stream->ustream);
2418 }
2419
2420 static
2421 void metadata_stream_reset_cache_consumed_position(
2422 struct lttng_consumer_stream *stream)
2423 {
2424 DBG("Reset metadata cache of session %" PRIu64,
2425 stream->chan->session_id);
2426 stream->ust_metadata_pushed = 0;
2427 }
2428
2429 /*
2430 * Write up to one packet from the metadata cache to the channel.
2431 *
2432 * Returns the number of bytes pushed from the cache into the ring buffer, or a
2433 * negative value on error.
2434 */
2435 static
2436 int commit_one_metadata_packet(struct lttng_consumer_stream *stream)
2437 {
2438 ssize_t write_len;
2439 int ret;
2440
2441 pthread_mutex_lock(&stream->chan->metadata_cache->lock);
2442 if (stream->chan->metadata_cache->max_offset ==
2443 stream->ust_metadata_pushed) {
2444 /*
2445 * In the context of a user space metadata channel, a
2446 * change in version can be detected in two ways:
2447 * 1) During the pre-consume of the `read_subbuffer` loop,
2448 * 2) When populating the metadata ring buffer (i.e. here).
2449 *
2450 * This function is invoked when there is no metadata
2451 * available in the ring-buffer. If all data was consumed
2452 * up to the size of the metadata cache, there is no metadata
2453 * to insert in the ring-buffer.
2454 *
2455 * However, the metadata version could still have changed (a
2456 * regeneration without any new data will yield the same cache
2457 * size).
2458 *
2459 * The cache's version is checked for a version change and the
2460 * consumed position is reset if one occurred.
2461 *
2462 * This check is only necessary for the user space domain as
2463 * it has to manage the cache explicitly. If this reset was not
2464 * performed, no metadata would be consumed (and no reset would
2465 * occur as part of the pre-consume) until the metadata size
2466 * exceeded the cache size.
2467 */
2468 if (stream->metadata_version !=
2469 stream->chan->metadata_cache->version) {
2470 metadata_stream_reset_cache_consumed_position(stream);
2471 consumer_stream_metadata_set_version(stream,
2472 stream->chan->metadata_cache->version);
2473 } else {
2474 ret = 0;
2475 goto end;
2476 }
2477 }
2478
2479 write_len = ustctl_write_one_packet_to_channel(stream->chan->uchan,
2480 &stream->chan->metadata_cache->data[stream->ust_metadata_pushed],
2481 stream->chan->metadata_cache->max_offset
2482 - stream->ust_metadata_pushed);
2483 assert(write_len != 0);
2484 if (write_len < 0) {
2485 ERR("Writing one metadata packet");
2486 ret = write_len;
2487 goto end;
2488 }
2489 stream->ust_metadata_pushed += write_len;
2490
2491 assert(stream->chan->metadata_cache->max_offset >=
2492 stream->ust_metadata_pushed);
2493 ret = write_len;
2494
2495 /*
2496 * Switch packet (but don't open the next one) on every commit of
2497 * a metadata packet. Since the subbuffer is fully filled (with padding,
2498 * if needed), the stream is "quiescent" after this commit.
2499 */
2500 ustctl_flush_buffer(stream->ustream, 1);
2501 stream->quiescent = true;
2502 end:
2503 pthread_mutex_unlock(&stream->chan->metadata_cache->lock);
2504 return ret;
2505 }
2506
2507
2508 /*
2509 * Sync metadata meaning request them to the session daemon and snapshot to the
2510 * metadata thread can consumer them.
2511 *
2512 * Metadata stream lock is held here, but we need to release it when
2513 * interacting with sessiond, else we cause a deadlock with live
2514 * awaiting on metadata to be pushed out.
2515 *
2516 * The RCU read side lock must be held by the caller.
2517 */
2518 enum sync_metadata_status lttng_ustconsumer_sync_metadata(
2519 struct lttng_consumer_local_data *ctx,
2520 struct lttng_consumer_stream *metadata_stream)
2521 {
2522 int ret;
2523 enum sync_metadata_status status;
2524 struct lttng_consumer_channel *metadata_channel;
2525
2526 assert(ctx);
2527 assert(metadata_stream);
2528
2529 metadata_channel = metadata_stream->chan;
2530 pthread_mutex_unlock(&metadata_stream->lock);
2531 /*
2532 * Request metadata from the sessiond, but don't wait for the flush
2533 * because we locked the metadata thread.
2534 */
2535 ret = lttng_ustconsumer_request_metadata(ctx, metadata_channel, 0, 0);
2536 pthread_mutex_lock(&metadata_stream->lock);
2537 if (ret < 0) {
2538 status = SYNC_METADATA_STATUS_ERROR;
2539 goto end;
2540 }
2541
2542 /*
2543 * The metadata stream and channel can be deleted while the
2544 * metadata stream lock was released. The streamed is checked
2545 * for deletion before we use it further.
2546 *
2547 * Note that it is safe to access a logically-deleted stream since its
2548 * existence is still guaranteed by the RCU read side lock. However,
2549 * it should no longer be used. The close/deletion of the metadata
2550 * channel and stream already guarantees that all metadata has been
2551 * consumed. Therefore, there is nothing left to do in this function.
2552 */
2553 if (consumer_stream_is_deleted(metadata_stream)) {
2554 DBG("Metadata stream %" PRIu64 " was deleted during the metadata synchronization",
2555 metadata_stream->key);
2556 status = SYNC_METADATA_STATUS_NO_DATA;
2557 goto end;
2558 }
2559
2560 ret = commit_one_metadata_packet(metadata_stream);
2561 if (ret < 0) {
2562 status = SYNC_METADATA_STATUS_ERROR;
2563 goto end;
2564 } else if (ret > 0) {
2565 status = SYNC_METADATA_STATUS_NEW_DATA;
2566 } else /* ret == 0 */ {
2567 status = SYNC_METADATA_STATUS_NO_DATA;
2568 goto end;
2569 }
2570
2571 ret = ustctl_snapshot(metadata_stream->ustream);
2572 if (ret < 0) {
2573 ERR("Failed to take a snapshot of the metadata ring-buffer positions, ret = %d", ret);
2574 status = SYNC_METADATA_STATUS_ERROR;
2575 goto end;
2576 }
2577
2578 end:
2579 return status;
2580 }
2581
2582 /*
2583 * Return 0 on success else a negative value.
2584 */
2585 static int notify_if_more_data(struct lttng_consumer_stream *stream,
2586 struct lttng_consumer_local_data *ctx)
2587 {
2588 int ret;
2589 struct ustctl_consumer_stream *ustream;
2590
2591 assert(stream);
2592 assert(ctx);
2593
2594 ustream = stream->ustream;
2595
2596 /*
2597 * First, we are going to check if there is a new subbuffer available
2598 * before reading the stream wait_fd.
2599 */
2600 /* Get the next subbuffer */
2601 ret = ustctl_get_next_subbuf(ustream);
2602 if (ret) {
2603 /* No more data found, flag the stream. */
2604 stream->has_data = 0;
2605 ret = 0;
2606 goto end;
2607 }
2608
2609 ret = ustctl_put_subbuf(ustream);
2610 assert(!ret);
2611
2612 /* This stream still has data. Flag it and wake up the data thread. */
2613 stream->has_data = 1;
2614
2615 if (stream->monitor && !stream->hangup_flush_done && !ctx->has_wakeup) {
2616 ssize_t writelen;
2617
2618 writelen = lttng_pipe_write(ctx->consumer_wakeup_pipe, "!", 1);
2619 if (writelen < 0 && errno != EAGAIN && errno != EWOULDBLOCK) {
2620 ret = writelen;
2621 goto end;
2622 }
2623
2624 /* The wake up pipe has been notified. */
2625 ctx->has_wakeup = 1;
2626 }
2627 ret = 0;
2628
2629 end:
2630 return ret;
2631 }
2632
2633 static int consumer_stream_ust_on_wake_up(struct lttng_consumer_stream *stream)
2634 {
2635 int ret = 0;
2636
2637 /*
2638 * We can consume the 1 byte written into the wait_fd by
2639 * UST. Don't trigger error if we cannot read this one byte
2640 * (read returns 0), or if the error is EAGAIN or EWOULDBLOCK.
2641 *
2642 * This is only done when the stream is monitored by a thread,
2643 * before the flush is done after a hangup and if the stream
2644 * is not flagged with data since there might be nothing to
2645 * consume in the wait fd but still have data available
2646 * flagged by the consumer wake up pipe.
2647 */
2648 if (stream->monitor && !stream->hangup_flush_done && !stream->has_data) {
2649 char dummy;
2650 ssize_t readlen;
2651
2652 readlen = lttng_read(stream->wait_fd, &dummy, 1);
2653 if (readlen < 0 && errno != EAGAIN && errno != EWOULDBLOCK) {
2654 ret = readlen;
2655 }
2656 }
2657
2658 return ret;
2659 }
2660
2661 static int extract_common_subbuffer_info(struct lttng_consumer_stream *stream,
2662 struct stream_subbuffer *subbuf)
2663 {
2664 int ret;
2665
2666 ret = ustctl_get_subbuf_size(
2667 stream->ustream, &subbuf->info.data.subbuf_size);
2668 if (ret) {
2669 goto end;
2670 }
2671
2672 ret = ustctl_get_padded_subbuf_size(
2673 stream->ustream, &subbuf->info.data.padded_subbuf_size);
2674 if (ret) {
2675 goto end;
2676 }
2677
2678 end:
2679 return ret;
2680 }
2681
2682 static int extract_metadata_subbuffer_info(struct lttng_consumer_stream *stream,
2683 struct stream_subbuffer *subbuf)
2684 {
2685 int ret;
2686
2687 ret = extract_common_subbuffer_info(stream, subbuf);
2688 if (ret) {
2689 goto end;
2690 }
2691
2692 subbuf->info.metadata.version = stream->metadata_version;
2693
2694 end:
2695 return ret;
2696 }
2697
2698 static int extract_data_subbuffer_info(struct lttng_consumer_stream *stream,
2699 struct stream_subbuffer *subbuf)
2700 {
2701 int ret;
2702
2703 ret = extract_common_subbuffer_info(stream, subbuf);
2704 if (ret) {
2705 goto end;
2706 }
2707
2708 ret = ustctl_get_packet_size(
2709 stream->ustream, &subbuf->info.data.packet_size);
2710 if (ret < 0) {
2711 PERROR("Failed to get sub-buffer packet size");
2712 goto end;
2713 }
2714
2715 ret = ustctl_get_content_size(
2716 stream->ustream, &subbuf->info.data.content_size);
2717 if (ret < 0) {
2718 PERROR("Failed to get sub-buffer content size");
2719 goto end;
2720 }
2721
2722 ret = ustctl_get_timestamp_begin(
2723 stream->ustream, &subbuf->info.data.timestamp_begin);
2724 if (ret < 0) {
2725 PERROR("Failed to get sub-buffer begin timestamp");
2726 goto end;
2727 }
2728
2729 ret = ustctl_get_timestamp_end(
2730 stream->ustream, &subbuf->info.data.timestamp_end);
2731 if (ret < 0) {
2732 PERROR("Failed to get sub-buffer end timestamp");
2733 goto end;
2734 }
2735
2736 ret = ustctl_get_events_discarded(
2737 stream->ustream, &subbuf->info.data.events_discarded);
2738 if (ret) {
2739 PERROR("Failed to get sub-buffer events discarded count");
2740 goto end;
2741 }
2742
2743 ret = ustctl_get_sequence_number(stream->ustream,
2744 &subbuf->info.data.sequence_number.value);
2745 if (ret) {
2746 /* May not be supported by older LTTng-modules. */
2747 if (ret != -ENOTTY) {
2748 PERROR("Failed to get sub-buffer sequence number");
2749 goto end;
2750 }
2751 } else {
2752 subbuf->info.data.sequence_number.is_set = true;
2753 }
2754
2755 ret = ustctl_get_stream_id(
2756 stream->ustream, &subbuf->info.data.stream_id);
2757 if (ret < 0) {
2758 PERROR("Failed to get stream id");
2759 goto end;
2760 }
2761
2762 ret = ustctl_get_instance_id(stream->ustream,
2763 &subbuf->info.data.stream_instance_id.value);
2764 if (ret) {
2765 /* May not be supported by older LTTng-modules. */
2766 if (ret != -ENOTTY) {
2767 PERROR("Failed to get stream instance id");
2768 goto end;
2769 }
2770 } else {
2771 subbuf->info.data.stream_instance_id.is_set = true;
2772 }
2773 end:
2774 return ret;
2775 }
2776
2777 static int get_next_subbuffer_common(struct lttng_consumer_stream *stream,
2778 struct stream_subbuffer *subbuffer)
2779 {
2780 int ret;
2781 const char *addr;
2782
2783 ret = stream->read_subbuffer_ops.extract_subbuffer_info(
2784 stream, subbuffer);
2785 if (ret) {
2786 goto end;
2787 }
2788
2789 ret = get_current_subbuf_addr(stream, &addr);
2790 if (ret) {
2791 goto end;
2792 }
2793
2794 subbuffer->buffer.buffer = lttng_buffer_view_init(
2795 addr, 0, subbuffer->info.data.padded_subbuf_size);
2796 assert(subbuffer->buffer.buffer.data != NULL);
2797 end:
2798 return ret;
2799 }
2800
2801 static int get_next_subbuffer(struct lttng_consumer_stream *stream,
2802 struct stream_subbuffer *subbuffer)
2803 {
2804 int ret;
2805
2806 ret = ustctl_get_next_subbuf(stream->ustream);
2807 if (ret) {
2808 goto end;
2809 }
2810
2811 ret = get_next_subbuffer_common(stream, subbuffer);
2812 if (ret) {
2813 goto end;
2814 }
2815 end:
2816 return ret;
2817 }
2818
2819 static int get_next_subbuffer_metadata(struct lttng_consumer_stream *stream,
2820 struct stream_subbuffer *subbuffer)
2821 {
2822 int ret;
2823 bool cache_empty;
2824 bool got_subbuffer;
2825 bool coherent;
2826 bool buffer_empty;
2827 unsigned long consumed_pos, produced_pos;
2828
2829 do {
2830 ret = ustctl_get_next_subbuf(stream->ustream);
2831 if (ret == 0) {
2832 got_subbuffer = true;
2833 } else {
2834 got_subbuffer = false;
2835 if (ret != -EAGAIN) {
2836 /* Fatal error. */
2837 goto end;
2838 }
2839 }
2840
2841 /*
2842 * Determine if the cache is empty and ensure that a sub-buffer
2843 * is made available if the cache is not empty.
2844 */
2845 if (!got_subbuffer) {
2846 ret = commit_one_metadata_packet(stream);
2847 if (ret < 0 && ret != -ENOBUFS) {
2848 goto end;
2849 } else if (ret == 0) {
2850 /* Not an error, the cache is empty. */
2851 cache_empty = true;
2852 ret = -ENODATA;
2853 goto end;
2854 } else {
2855 cache_empty = false;
2856 }
2857 } else {
2858 pthread_mutex_lock(&stream->chan->metadata_cache->lock);
2859 cache_empty = stream->chan->metadata_cache->max_offset ==
2860 stream->ust_metadata_pushed;
2861 pthread_mutex_unlock(&stream->chan->metadata_cache->lock);
2862 }
2863 } while (!got_subbuffer);
2864
2865 /* Populate sub-buffer infos and view. */
2866 ret = get_next_subbuffer_common(stream, subbuffer);
2867 if (ret) {
2868 goto end;
2869 }
2870
2871 ret = lttng_ustconsumer_sample_snapshot_positions(stream);
2872 if (ret < 0) {
2873 /*
2874 * -EAGAIN is not expected since we got a sub-buffer and haven't
2875 * pushed the consumption position yet (on put_next).
2876 */
2877 PERROR("Failed to take a snapshot of metadata buffer positions");
2878 goto end;
2879 }
2880
2881 ret = lttng_ustconsumer_get_consumed_snapshot(stream, &consumed_pos);
2882 if (ret) {
2883 PERROR("Failed to get metadata consumed position");
2884 goto end;
2885 }
2886
2887 ret = lttng_ustconsumer_get_produced_snapshot(stream, &produced_pos);
2888 if (ret) {
2889 PERROR("Failed to get metadata produced position");
2890 goto end;
2891 }
2892
2893 /* Last sub-buffer of the ring buffer ? */
2894 buffer_empty = (consumed_pos + stream->max_sb_size) == produced_pos;
2895
2896 /*
2897 * The sessiond registry lock ensures that coherent units of metadata
2898 * are pushed to the consumer daemon at once. Hence, if a sub-buffer is
2899 * acquired, the cache is empty, and it is the only available sub-buffer
2900 * available, it is safe to assume that it is "coherent".
2901 */
2902 coherent = got_subbuffer && cache_empty && buffer_empty;
2903
2904 LTTNG_OPTIONAL_SET(&subbuffer->info.metadata.coherent, coherent);
2905 end:
2906 return ret;
2907 }
2908
2909 static int put_next_subbuffer(struct lttng_consumer_stream *stream,
2910 struct stream_subbuffer *subbuffer)
2911 {
2912 const int ret = ustctl_put_next_subbuf(stream->ustream);
2913
2914 assert(ret == 0);
2915 return ret;
2916 }
2917
2918 static int signal_metadata(struct lttng_consumer_stream *stream,
2919 struct lttng_consumer_local_data *ctx)
2920 {
2921 return pthread_cond_broadcast(&stream->metadata_rdv) ? -errno : 0;
2922 }
2923
2924 static int lttng_ustconsumer_set_stream_ops(
2925 struct lttng_consumer_stream *stream)
2926 {
2927 int ret = 0;
2928
2929 stream->read_subbuffer_ops.on_wake_up = consumer_stream_ust_on_wake_up;
2930 if (stream->metadata_flag) {
2931 stream->read_subbuffer_ops.get_next_subbuffer =
2932 get_next_subbuffer_metadata;
2933 stream->read_subbuffer_ops.extract_subbuffer_info =
2934 extract_metadata_subbuffer_info;
2935 stream->read_subbuffer_ops.reset_metadata =
2936 metadata_stream_reset_cache_consumed_position;
2937 if (stream->chan->is_live) {
2938 stream->read_subbuffer_ops.on_sleep = signal_metadata;
2939 ret = consumer_stream_enable_metadata_bucketization(
2940 stream);
2941 if (ret) {
2942 goto end;
2943 }
2944 }
2945 } else {
2946 stream->read_subbuffer_ops.get_next_subbuffer =
2947 get_next_subbuffer;
2948 stream->read_subbuffer_ops.extract_subbuffer_info =
2949 extract_data_subbuffer_info;
2950 stream->read_subbuffer_ops.on_sleep = notify_if_more_data;
2951 if (stream->chan->is_live) {
2952 stream->read_subbuffer_ops.send_live_beacon =
2953 consumer_flush_ust_index;
2954 }
2955 }
2956
2957 stream->read_subbuffer_ops.put_next_subbuffer = put_next_subbuffer;
2958 end:
2959 return ret;
2960 }
2961
2962 /*
2963 * Called when a stream is created.
2964 *
2965 * Return 0 on success or else a negative value.
2966 */
2967 int lttng_ustconsumer_on_recv_stream(struct lttng_consumer_stream *stream)
2968 {
2969 int ret;
2970
2971 assert(stream);
2972
2973 /*
2974 * Don't create anything if this is set for streaming or if there is
2975 * no current trace chunk on the parent channel.
2976 */
2977 if (stream->net_seq_idx == (uint64_t) -1ULL && stream->chan->monitor &&
2978 stream->chan->trace_chunk) {
2979 ret = consumer_stream_create_output_files(stream, true);
2980 if (ret) {
2981 goto error;
2982 }
2983 }
2984
2985 lttng_ustconsumer_set_stream_ops(stream);
2986 ret = 0;
2987
2988 error:
2989 return ret;
2990 }
2991
2992 /*
2993 * Check if data is still being extracted from the buffers for a specific
2994 * stream. Consumer data lock MUST be acquired before calling this function
2995 * and the stream lock.
2996 *
2997 * Return 1 if the traced data are still getting read else 0 meaning that the
2998 * data is available for trace viewer reading.
2999 */
3000 int lttng_ustconsumer_data_pending(struct lttng_consumer_stream *stream)
3001 {
3002 int ret;
3003
3004 assert(stream);
3005 assert(stream->ustream);
3006
3007 DBG("UST consumer checking data pending");
3008
3009 if (stream->endpoint_status != CONSUMER_ENDPOINT_ACTIVE) {
3010 ret = 0;
3011 goto end;
3012 }
3013
3014 if (stream->chan->type == CONSUMER_CHANNEL_TYPE_METADATA) {
3015 uint64_t contiguous, pushed;
3016
3017 /* Ease our life a bit. */
3018 pthread_mutex_lock(&stream->chan->metadata_cache->lock);
3019 contiguous = stream->chan->metadata_cache->max_offset;
3020 pthread_mutex_unlock(&stream->chan->metadata_cache->lock);
3021 pushed = stream->ust_metadata_pushed;
3022
3023 /*
3024 * We can simply check whether all contiguously available data
3025 * has been pushed to the ring buffer, since the push operation
3026 * is performed within get_next_subbuf(), and because both
3027 * get_next_subbuf() and put_next_subbuf() are issued atomically
3028 * thanks to the stream lock within
3029 * lttng_ustconsumer_read_subbuffer(). This basically means that
3030 * whetnever ust_metadata_pushed is incremented, the associated
3031 * metadata has been consumed from the metadata stream.
3032 */
3033 DBG("UST consumer metadata pending check: contiguous %" PRIu64 " vs pushed %" PRIu64,
3034 contiguous, pushed);
3035 assert(((int64_t) (contiguous - pushed)) >= 0);
3036 if ((contiguous != pushed) ||
3037 (((int64_t) contiguous - pushed) > 0 || contiguous == 0)) {
3038 ret = 1; /* Data is pending */
3039 goto end;
3040 }
3041 } else {
3042 ret = ustctl_get_next_subbuf(stream->ustream);
3043 if (ret == 0) {
3044 /*
3045 * There is still data so let's put back this
3046 * subbuffer.
3047 */
3048 ret = ustctl_put_subbuf(stream->ustream);
3049 assert(ret == 0);
3050 ret = 1; /* Data is pending */
3051 goto end;
3052 }
3053 }
3054
3055 /* Data is NOT pending so ready to be read. */
3056 ret = 0;
3057
3058 end:
3059 return ret;
3060 }
3061
3062 /*
3063 * Stop a given metadata channel timer if enabled and close the wait fd which
3064 * is the poll pipe of the metadata stream.
3065 *
3066 * This MUST be called with the metadata channel lock acquired.
3067 */
3068 void lttng_ustconsumer_close_metadata(struct lttng_consumer_channel *metadata)
3069 {
3070 int ret;
3071
3072 assert(metadata);
3073 assert(metadata->type == CONSUMER_CHANNEL_TYPE_METADATA);
3074
3075 DBG("Closing metadata channel key %" PRIu64, metadata->key);
3076
3077 if (metadata->switch_timer_enabled == 1) {
3078 consumer_timer_switch_stop(metadata);
3079 }
3080
3081 if (!metadata->metadata_stream) {
3082 goto end;
3083 }
3084
3085 /*
3086 * Closing write side so the thread monitoring the stream wakes up if any
3087 * and clean the metadata stream.
3088 */
3089 if (metadata->metadata_stream->ust_metadata_poll_pipe[1] >= 0) {
3090 ret = close(metadata->metadata_stream->ust_metadata_poll_pipe[1]);
3091 if (ret < 0) {
3092 PERROR("closing metadata pipe write side");
3093 }
3094 metadata->metadata_stream->ust_metadata_poll_pipe[1] = -1;
3095 }
3096
3097 end:
3098 return;
3099 }
3100
3101 /*
3102 * Close every metadata stream wait fd of the metadata hash table. This
3103 * function MUST be used very carefully so not to run into a race between the
3104 * metadata thread handling streams and this function closing their wait fd.
3105 *
3106 * For UST, this is used when the session daemon hangs up. Its the metadata
3107 * producer so calling this is safe because we are assured that no state change
3108 * can occur in the metadata thread for the streams in the hash table.
3109 */
3110 void lttng_ustconsumer_close_all_metadata(struct lttng_ht *metadata_ht)
3111 {
3112 struct lttng_ht_iter iter;
3113 struct lttng_consumer_stream *stream;
3114
3115 assert(metadata_ht);
3116 assert(metadata_ht->ht);
3117
3118 DBG("UST consumer closing all metadata streams");
3119
3120 rcu_read_lock();
3121 cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream,
3122 node.node) {
3123
3124 health_code_update();
3125
3126 pthread_mutex_lock(&stream->chan->lock);
3127 lttng_ustconsumer_close_metadata(stream->chan);
3128 pthread_mutex_unlock(&stream->chan->lock);
3129
3130 }
3131 rcu_read_unlock();
3132 }
3133
3134 void lttng_ustconsumer_close_stream_wakeup(struct lttng_consumer_stream *stream)
3135 {
3136 int ret;
3137
3138 ret = ustctl_stream_close_wakeup_fd(stream->ustream);
3139 if (ret < 0) {
3140 ERR("Unable to close wakeup fd");
3141 }
3142 }
3143
3144 /*
3145 * Please refer to consumer-timer.c before adding any lock within this
3146 * function or any of its callees. Timers have a very strict locking
3147 * semantic with respect to teardown. Failure to respect this semantic
3148 * introduces deadlocks.
3149 *
3150 * DON'T hold the metadata lock when calling this function, else this
3151 * can cause deadlock involving consumer awaiting for metadata to be
3152 * pushed out due to concurrent interaction with the session daemon.
3153 */
3154 int lttng_ustconsumer_request_metadata(struct lttng_consumer_local_data *ctx,
3155 struct lttng_consumer_channel *channel, int timer, int wait)
3156 {
3157 struct lttcomm_metadata_request_msg request;
3158 struct lttcomm_consumer_msg msg;
3159 enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS;
3160 uint64_t len, key, offset, version;
3161 int ret;
3162
3163 assert(channel);
3164 assert(channel->metadata_cache);
3165
3166 memset(&request, 0, sizeof(request));
3167
3168 /* send the metadata request to sessiond */
3169 switch (consumer_data.type) {
3170 case LTTNG_CONSUMER64_UST:
3171 request.bits_per_long = 64;
3172 break;
3173 case LTTNG_CONSUMER32_UST:
3174 request.bits_per_long = 32;
3175 break;
3176 default:
3177 request.bits_per_long = 0;
3178 break;
3179 }
3180
3181 request.session_id = channel->session_id;
3182 request.session_id_per_pid = channel->session_id_per_pid;
3183 /*
3184 * Request the application UID here so the metadata of that application can
3185 * be sent back. The channel UID corresponds to the user UID of the session
3186 * used for the rights on the stream file(s).
3187 */
3188 request.uid = channel->ust_app_uid;
3189 request.key = channel->key;
3190
3191 DBG("Sending metadata request to sessiond, session id %" PRIu64
3192 ", per-pid %" PRIu64 ", app UID %u and channel key %" PRIu64,
3193 request.session_id, request.session_id_per_pid, request.uid,
3194 request.key);
3195
3196 pthread_mutex_lock(&ctx->metadata_socket_lock);
3197
3198 health_code_update();
3199
3200 ret = lttcomm_send_unix_sock(ctx->consumer_metadata_socket, &request,
3201 sizeof(request));
3202 if (ret < 0) {
3203 ERR("Asking metadata to sessiond");
3204 goto end;
3205 }
3206
3207 health_code_update();
3208
3209 /* Receive the metadata from sessiond */
3210 ret = lttcomm_recv_unix_sock(ctx->consumer_metadata_socket, &msg,
3211 sizeof(msg));
3212 if (ret != sizeof(msg)) {
3213 DBG("Consumer received unexpected message size %d (expects %zu)",
3214 ret, sizeof(msg));
3215 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD);
3216 /*
3217 * The ret value might 0 meaning an orderly shutdown but this is ok
3218 * since the caller handles this.
3219 */
3220 goto end;
3221 }
3222
3223 health_code_update();
3224
3225 if (msg.cmd_type == LTTNG_ERR_UND) {
3226 /* No registry found */
3227 (void) consumer_send_status_msg(ctx->consumer_metadata_socket,
3228 ret_code);
3229 ret = 0;
3230 goto end;
3231 } else if (msg.cmd_type != LTTNG_CONSUMER_PUSH_METADATA) {
3232 ERR("Unexpected cmd_type received %d", msg.cmd_type);
3233 ret = -1;
3234 goto end;
3235 }
3236
3237 len = msg.u.push_metadata.len;
3238 key = msg.u.push_metadata.key;
3239 offset = msg.u.push_metadata.target_offset;
3240 version = msg.u.push_metadata.version;
3241
3242 assert(key == channel->key);
3243 if (len == 0) {
3244 DBG("No new metadata to receive for key %" PRIu64, key);
3245 }
3246
3247 health_code_update();
3248
3249 /* Tell session daemon we are ready to receive the metadata. */
3250 ret = consumer_send_status_msg(ctx->consumer_metadata_socket,
3251 LTTCOMM_CONSUMERD_SUCCESS);
3252 if (ret < 0 || len == 0) {
3253 /*
3254 * Somehow, the session daemon is not responding anymore or there is
3255 * nothing to receive.
3256 */
3257 goto end;
3258 }
3259
3260 health_code_update();
3261
3262 ret = lttng_ustconsumer_recv_metadata(ctx->consumer_metadata_socket,
3263 key, offset, len, version, channel, timer, wait);
3264 if (ret >= 0) {
3265 /*
3266 * Only send the status msg if the sessiond is alive meaning a positive
3267 * ret code.
3268 */
3269 (void) consumer_send_status_msg(ctx->consumer_metadata_socket, ret);
3270 }
3271 ret = 0;
3272
3273 end:
3274 health_code_update();
3275
3276 pthread_mutex_unlock(&ctx->metadata_socket_lock);
3277 return ret;
3278 }
3279
3280 /*
3281 * Return the ustctl call for the get stream id.
3282 */
3283 int lttng_ustconsumer_get_stream_id(struct lttng_consumer_stream *stream,
3284 uint64_t *stream_id)
3285 {
3286 assert(stream);
3287 assert(stream_id);
3288
3289 return ustctl_get_stream_id(stream->ustream, stream_id);
3290 }
This page took 0.200991 seconds and 4 git commands to generate.