b2c454deb54302362806dd80180f9b235f29e33a
[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
37 #include <bin/lttng-consumerd/health-consumerd.h>
38 #include <common/common.h>
39 #include <common/sessiond-comm/sessiond-comm.h>
40 #include <common/relayd/relayd.h>
41 #include <common/compat/fcntl.h>
42 #include <common/compat/endian.h>
43 #include <common/consumer/consumer-metadata-cache.h>
44 #include <common/consumer/consumer-stream.h>
45 #include <common/consumer/consumer-timer.h>
46 #include <common/utils.h>
47 #include <common/index/index.h>
48
49 #include "ust-consumer.h"
50
51 #define INT_MAX_STR_LEN 12 /* includes \0 */
52
53 extern struct lttng_consumer_global_data consumer_data;
54 extern int consumer_poll_timeout;
55
56 /*
57 * Free channel object and all streams associated with it. This MUST be used
58 * only and only if the channel has _NEVER_ been added to the global channel
59 * hash table.
60 */
61 static void destroy_channel(struct lttng_consumer_channel *channel)
62 {
63 struct lttng_consumer_stream *stream, *stmp;
64
65 assert(channel);
66
67 DBG("UST consumer cleaning stream list");
68
69 cds_list_for_each_entry_safe(stream, stmp, &channel->streams.head,
70 send_node) {
71
72 health_code_update();
73
74 cds_list_del(&stream->send_node);
75 ustctl_destroy_stream(stream->ustream);
76 lttng_trace_chunk_put(stream->trace_chunk);
77 free(stream);
78 }
79
80 /*
81 * If a channel is available meaning that was created before the streams
82 * were, delete it.
83 */
84 if (channel->uchan) {
85 lttng_ustconsumer_del_channel(channel);
86 lttng_ustconsumer_free_channel(channel);
87 }
88 free(channel);
89 }
90
91 /*
92 * Add channel to internal consumer state.
93 *
94 * Returns 0 on success or else a negative value.
95 */
96 static int add_channel(struct lttng_consumer_channel *channel,
97 struct lttng_consumer_local_data *ctx)
98 {
99 int ret = 0;
100
101 assert(channel);
102 assert(ctx);
103
104 if (ctx->on_recv_channel != NULL) {
105 ret = ctx->on_recv_channel(channel);
106 if (ret == 0) {
107 ret = consumer_add_channel(channel, ctx);
108 } else if (ret < 0) {
109 /* Most likely an ENOMEM. */
110 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
111 goto error;
112 }
113 } else {
114 ret = consumer_add_channel(channel, ctx);
115 }
116
117 DBG("UST consumer channel added (key: %" PRIu64 ")", channel->key);
118
119 error:
120 return ret;
121 }
122
123 /*
124 * Allocate and return a consumer channel object.
125 */
126 static struct lttng_consumer_channel *allocate_channel(uint64_t session_id,
127 const uint64_t *chunk_id, const char *pathname, const char *name,
128 uint64_t relayd_id, uint64_t key, enum lttng_event_output output,
129 uint64_t tracefile_size, uint64_t tracefile_count,
130 uint64_t session_id_per_pid, unsigned int monitor,
131 unsigned int live_timer_interval,
132 const char *root_shm_path, const char *shm_path)
133 {
134 assert(pathname);
135 assert(name);
136
137 return consumer_allocate_channel(key, session_id, chunk_id, pathname,
138 name, relayd_id, output, tracefile_size,
139 tracefile_count, session_id_per_pid, monitor,
140 live_timer_interval, root_shm_path, shm_path);
141 }
142
143 /*
144 * Allocate and return a consumer stream object. If _alloc_ret is not NULL, the
145 * error value if applicable is set in it else it is kept untouched.
146 *
147 * Return NULL on error else the newly allocated stream object.
148 */
149 static struct lttng_consumer_stream *allocate_stream(int cpu, int key,
150 struct lttng_consumer_channel *channel,
151 struct lttng_consumer_local_data *ctx, int *_alloc_ret)
152 {
153 int alloc_ret;
154 struct lttng_consumer_stream *stream = NULL;
155
156 assert(channel);
157 assert(ctx);
158
159 stream = consumer_allocate_stream(channel->key,
160 key,
161 channel->name,
162 channel->relayd_id,
163 channel->session_id,
164 channel->trace_chunk,
165 cpu,
166 &alloc_ret,
167 channel->type,
168 channel->monitor);
169 if (stream == NULL) {
170 switch (alloc_ret) {
171 case -ENOENT:
172 /*
173 * We could not find the channel. Can happen if cpu hotplug
174 * happens while tearing down.
175 */
176 DBG3("Could not find channel");
177 break;
178 case -ENOMEM:
179 case -EINVAL:
180 default:
181 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
182 break;
183 }
184 goto error;
185 }
186
187 consumer_stream_update_channel_attributes(stream, channel);
188 stream->chan = channel;
189
190 error:
191 if (_alloc_ret) {
192 *_alloc_ret = alloc_ret;
193 }
194 return stream;
195 }
196
197 /*
198 * Send the given stream pointer to the corresponding thread.
199 *
200 * Returns 0 on success else a negative value.
201 */
202 static int send_stream_to_thread(struct lttng_consumer_stream *stream,
203 struct lttng_consumer_local_data *ctx)
204 {
205 int ret;
206 struct lttng_pipe *stream_pipe;
207
208 /* Get the right pipe where the stream will be sent. */
209 if (stream->metadata_flag) {
210 consumer_add_metadata_stream(stream);
211 stream_pipe = ctx->consumer_metadata_pipe;
212 } else {
213 consumer_add_data_stream(stream);
214 stream_pipe = ctx->consumer_data_pipe;
215 }
216
217 /*
218 * From this point on, the stream's ownership has been moved away from
219 * the channel and it becomes globally visible. Hence, remove it from
220 * the local stream list to prevent the stream from being both local and
221 * global.
222 */
223 stream->globally_visible = 1;
224 cds_list_del(&stream->send_node);
225
226 ret = lttng_pipe_write(stream_pipe, &stream, sizeof(stream));
227 if (ret < 0) {
228 ERR("Consumer write %s stream to pipe %d",
229 stream->metadata_flag ? "metadata" : "data",
230 lttng_pipe_get_writefd(stream_pipe));
231 if (stream->metadata_flag) {
232 consumer_del_stream_for_metadata(stream);
233 } else {
234 consumer_del_stream_for_data(stream);
235 }
236 goto error;
237 }
238
239 error:
240 return ret;
241 }
242
243 static
244 int get_stream_shm_path(char *stream_shm_path, const char *shm_path, int cpu)
245 {
246 char cpu_nr[INT_MAX_STR_LEN]; /* int max len */
247 int ret;
248
249 strncpy(stream_shm_path, shm_path, PATH_MAX);
250 stream_shm_path[PATH_MAX - 1] = '\0';
251 ret = snprintf(cpu_nr, INT_MAX_STR_LEN, "%i", cpu);
252 if (ret < 0) {
253 PERROR("snprintf");
254 goto end;
255 }
256 strncat(stream_shm_path, cpu_nr,
257 PATH_MAX - strlen(stream_shm_path) - 1);
258 ret = 0;
259 end:
260 return ret;
261 }
262
263 /*
264 * Create streams for the given channel using liblttng-ust-ctl.
265 * The channel lock must be acquired by the caller.
266 *
267 * Return 0 on success else a negative value.
268 */
269 static int create_ust_streams(struct lttng_consumer_channel *channel,
270 struct lttng_consumer_local_data *ctx)
271 {
272 int ret, cpu = 0;
273 struct ustctl_consumer_stream *ustream;
274 struct lttng_consumer_stream *stream;
275 pthread_mutex_t *current_stream_lock = NULL;
276
277 assert(channel);
278 assert(ctx);
279
280 /*
281 * While a stream is available from ustctl. When NULL is returned, we've
282 * reached the end of the possible stream for the channel.
283 */
284 while ((ustream = ustctl_create_stream(channel->uchan, cpu))) {
285 int wait_fd;
286 int ust_metadata_pipe[2];
287
288 health_code_update();
289
290 if (channel->type == CONSUMER_CHANNEL_TYPE_METADATA && channel->monitor) {
291 ret = utils_create_pipe_cloexec_nonblock(ust_metadata_pipe);
292 if (ret < 0) {
293 ERR("Create ust metadata poll pipe");
294 goto error;
295 }
296 wait_fd = ust_metadata_pipe[0];
297 } else {
298 wait_fd = ustctl_stream_get_wait_fd(ustream);
299 }
300
301 /* Allocate consumer stream object. */
302 stream = allocate_stream(cpu, wait_fd, channel, ctx, &ret);
303 if (!stream) {
304 goto error_alloc;
305 }
306 stream->ustream = ustream;
307 /*
308 * Store it so we can save multiple function calls afterwards since
309 * this value is used heavily in the stream threads. This is UST
310 * specific so this is why it's done after allocation.
311 */
312 stream->wait_fd = wait_fd;
313
314 /*
315 * Increment channel refcount since the channel reference has now been
316 * assigned in the allocation process above.
317 */
318 if (stream->chan->monitor) {
319 uatomic_inc(&stream->chan->refcount);
320 }
321
322 pthread_mutex_lock(&stream->lock);
323 current_stream_lock = &stream->lock;
324 /*
325 * Order is important this is why a list is used. On error, the caller
326 * should clean this list.
327 */
328 cds_list_add_tail(&stream->send_node, &channel->streams.head);
329
330 ret = ustctl_get_max_subbuf_size(stream->ustream,
331 &stream->max_sb_size);
332 if (ret < 0) {
333 ERR("ustctl_get_max_subbuf_size failed for stream %s",
334 stream->name);
335 goto error;
336 }
337
338 /* Do actions once stream has been received. */
339 if (ctx->on_recv_stream) {
340 ret = ctx->on_recv_stream(stream);
341 if (ret < 0) {
342 goto error;
343 }
344 }
345
346 DBG("UST consumer add stream %s (key: %" PRIu64 ") with relayd id %" PRIu64,
347 stream->name, stream->key, stream->relayd_stream_id);
348
349 /* Set next CPU stream. */
350 channel->streams.count = ++cpu;
351
352 /* Keep stream reference when creating metadata. */
353 if (channel->type == CONSUMER_CHANNEL_TYPE_METADATA) {
354 channel->metadata_stream = stream;
355 if (channel->monitor) {
356 /* Set metadata poll pipe if we created one */
357 memcpy(stream->ust_metadata_poll_pipe,
358 ust_metadata_pipe,
359 sizeof(ust_metadata_pipe));
360 }
361 }
362 pthread_mutex_unlock(&stream->lock);
363 current_stream_lock = NULL;
364 }
365
366 return 0;
367
368 error:
369 error_alloc:
370 if (current_stream_lock) {
371 pthread_mutex_unlock(current_stream_lock);
372 }
373 return ret;
374 }
375
376 /*
377 * create_posix_shm is never called concurrently within a process.
378 */
379 static
380 int create_posix_shm(void)
381 {
382 char tmp_name[NAME_MAX];
383 int shmfd, ret;
384
385 ret = snprintf(tmp_name, NAME_MAX, "/ust-shm-consumer-%d", getpid());
386 if (ret < 0) {
387 PERROR("snprintf");
388 return -1;
389 }
390 /*
391 * Allocate shm, and immediately unlink its shm oject, keeping
392 * only the file descriptor as a reference to the object.
393 * We specifically do _not_ use the / at the beginning of the
394 * pathname so that some OS implementations can keep it local to
395 * the process (POSIX leaves this implementation-defined).
396 */
397 shmfd = shm_open(tmp_name, O_CREAT | O_EXCL | O_RDWR, 0700);
398 if (shmfd < 0) {
399 PERROR("shm_open");
400 goto error_shm_open;
401 }
402 ret = shm_unlink(tmp_name);
403 if (ret < 0 && errno != ENOENT) {
404 PERROR("shm_unlink");
405 goto error_shm_release;
406 }
407 return shmfd;
408
409 error_shm_release:
410 ret = close(shmfd);
411 if (ret) {
412 PERROR("close");
413 }
414 error_shm_open:
415 return -1;
416 }
417
418 static int open_ust_stream_fd(struct lttng_consumer_channel *channel, int cpu,
419 const struct lttng_credentials *session_credentials)
420 {
421 char shm_path[PATH_MAX];
422 int ret;
423
424 if (!channel->shm_path[0]) {
425 return create_posix_shm();
426 }
427 ret = get_stream_shm_path(shm_path, channel->shm_path, cpu);
428 if (ret) {
429 goto error_shm_path;
430 }
431 return run_as_open(shm_path,
432 O_RDWR | O_CREAT | O_EXCL, S_IRUSR | S_IWUSR,
433 session_credentials->uid, session_credentials->gid);
434
435 error_shm_path:
436 return -1;
437 }
438
439 /*
440 * Create an UST channel with the given attributes and send it to the session
441 * daemon using the ust ctl API.
442 *
443 * Return 0 on success or else a negative value.
444 */
445 static int create_ust_channel(struct lttng_consumer_channel *channel,
446 struct ustctl_consumer_channel_attr *attr,
447 struct ustctl_consumer_channel **ust_chanp)
448 {
449 int ret, nr_stream_fds, i, j;
450 int *stream_fds;
451 struct ustctl_consumer_channel *ust_channel;
452
453 assert(channel);
454 assert(attr);
455 assert(ust_chanp);
456 assert(channel->buffer_credentials.is_set);
457
458 DBG3("Creating channel to ustctl with attr: [overwrite: %d, "
459 "subbuf_size: %" PRIu64 ", num_subbuf: %" PRIu64 ", "
460 "switch_timer_interval: %u, read_timer_interval: %u, "
461 "output: %d, type: %d", attr->overwrite, attr->subbuf_size,
462 attr->num_subbuf, attr->switch_timer_interval,
463 attr->read_timer_interval, attr->output, attr->type);
464
465 if (channel->type == CONSUMER_CHANNEL_TYPE_METADATA)
466 nr_stream_fds = 1;
467 else
468 nr_stream_fds = ustctl_get_nr_stream_per_channel();
469 stream_fds = zmalloc(nr_stream_fds * sizeof(*stream_fds));
470 if (!stream_fds) {
471 ret = -1;
472 goto error_alloc;
473 }
474 for (i = 0; i < nr_stream_fds; i++) {
475 stream_fds[i] = open_ust_stream_fd(channel, i,
476 &channel->buffer_credentials.value);
477 if (stream_fds[i] < 0) {
478 ret = -1;
479 goto error_open;
480 }
481 }
482 ust_channel = ustctl_create_channel(attr, stream_fds, nr_stream_fds);
483 if (!ust_channel) {
484 ret = -1;
485 goto error_create;
486 }
487 channel->nr_stream_fds = nr_stream_fds;
488 channel->stream_fds = stream_fds;
489 *ust_chanp = ust_channel;
490
491 return 0;
492
493 error_create:
494 error_open:
495 for (j = i - 1; j >= 0; j--) {
496 int closeret;
497
498 closeret = close(stream_fds[j]);
499 if (closeret) {
500 PERROR("close");
501 }
502 if (channel->shm_path[0]) {
503 char shm_path[PATH_MAX];
504
505 closeret = get_stream_shm_path(shm_path,
506 channel->shm_path, j);
507 if (closeret) {
508 ERR("Cannot get stream shm path");
509 }
510 closeret = run_as_unlink(shm_path,
511 channel->buffer_credentials.value.uid,
512 channel->buffer_credentials.value.gid);
513 if (closeret) {
514 PERROR("unlink %s", shm_path);
515 }
516 }
517 }
518 /* Try to rmdir all directories under shm_path root. */
519 if (channel->root_shm_path[0]) {
520 (void) run_as_rmdir_recursive(channel->root_shm_path,
521 channel->buffer_credentials.value.uid,
522 channel->buffer_credentials.value.gid,
523 LTTNG_DIRECTORY_HANDLE_SKIP_NON_EMPTY_FLAG);
524 }
525 free(stream_fds);
526 error_alloc:
527 return ret;
528 }
529
530 /*
531 * Send a single given stream to the session daemon using the sock.
532 *
533 * Return 0 on success else a negative value.
534 */
535 static int send_sessiond_stream(int sock, struct lttng_consumer_stream *stream)
536 {
537 int ret;
538
539 assert(stream);
540 assert(sock >= 0);
541
542 DBG("UST consumer sending stream %" PRIu64 " to sessiond", stream->key);
543
544 /* Send stream to session daemon. */
545 ret = ustctl_send_stream_to_sessiond(sock, stream->ustream);
546 if (ret < 0) {
547 goto error;
548 }
549
550 error:
551 return ret;
552 }
553
554 /*
555 * Send channel to sessiond and relayd if applicable.
556 *
557 * Return 0 on success or else a negative value.
558 */
559 static int send_channel_to_sessiond_and_relayd(int sock,
560 struct lttng_consumer_channel *channel,
561 struct lttng_consumer_local_data *ctx, int *relayd_error)
562 {
563 int ret, ret_code = LTTCOMM_CONSUMERD_SUCCESS;
564 struct lttng_consumer_stream *stream;
565 uint64_t net_seq_idx = -1ULL;
566
567 assert(channel);
568 assert(ctx);
569 assert(sock >= 0);
570
571 DBG("UST consumer sending channel %s to sessiond", channel->name);
572
573 if (channel->relayd_id != (uint64_t) -1ULL) {
574 cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
575
576 health_code_update();
577
578 /* Try to send the stream to the relayd if one is available. */
579 DBG("Sending stream %" PRIu64 " of channel \"%s\" to relayd",
580 stream->key, channel->name);
581 ret = consumer_send_relayd_stream(stream, stream->chan->pathname);
582 if (ret < 0) {
583 /*
584 * Flag that the relayd was the problem here probably due to a
585 * communicaton error on the socket.
586 */
587 if (relayd_error) {
588 *relayd_error = 1;
589 }
590 ret_code = LTTCOMM_CONSUMERD_RELAYD_FAIL;
591 }
592 if (net_seq_idx == -1ULL) {
593 net_seq_idx = stream->net_seq_idx;
594 }
595 }
596 }
597
598 /* Inform sessiond that we are about to send channel and streams. */
599 ret = consumer_send_status_msg(sock, ret_code);
600 if (ret < 0 || ret_code != LTTCOMM_CONSUMERD_SUCCESS) {
601 /*
602 * Either the session daemon is not responding or the relayd died so we
603 * stop now.
604 */
605 goto error;
606 }
607
608 /* Send channel to sessiond. */
609 ret = ustctl_send_channel_to_sessiond(sock, channel->uchan);
610 if (ret < 0) {
611 goto error;
612 }
613
614 ret = ustctl_channel_close_wakeup_fd(channel->uchan);
615 if (ret < 0) {
616 goto error;
617 }
618
619 /* The channel was sent successfully to the sessiond at this point. */
620 cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
621
622 health_code_update();
623
624 /* Send stream to session daemon. */
625 ret = send_sessiond_stream(sock, stream);
626 if (ret < 0) {
627 goto error;
628 }
629 }
630
631 /* Tell sessiond there is no more stream. */
632 ret = ustctl_send_stream_to_sessiond(sock, NULL);
633 if (ret < 0) {
634 goto error;
635 }
636
637 DBG("UST consumer NULL stream sent to sessiond");
638
639 return 0;
640
641 error:
642 if (ret_code != LTTCOMM_CONSUMERD_SUCCESS) {
643 ret = -1;
644 }
645 return ret;
646 }
647
648 /*
649 * Creates a channel and streams and add the channel it to the channel internal
650 * state. The created stream must ONLY be sent once the GET_CHANNEL command is
651 * received.
652 *
653 * Return 0 on success or else, a negative value is returned and the channel
654 * MUST be destroyed by consumer_del_channel().
655 */
656 static int ask_channel(struct lttng_consumer_local_data *ctx,
657 struct lttng_consumer_channel *channel,
658 struct ustctl_consumer_channel_attr *attr)
659 {
660 int ret;
661
662 assert(ctx);
663 assert(channel);
664 assert(attr);
665
666 /*
667 * This value is still used by the kernel consumer since for the kernel,
668 * the stream ownership is not IN the consumer so we need to have the
669 * number of left stream that needs to be initialized so we can know when
670 * to delete the channel (see consumer.c).
671 *
672 * As for the user space tracer now, the consumer creates and sends the
673 * stream to the session daemon which only sends them to the application
674 * once every stream of a channel is received making this value useless
675 * because we they will be added to the poll thread before the application
676 * receives them. This ensures that a stream can not hang up during
677 * initilization of a channel.
678 */
679 channel->nb_init_stream_left = 0;
680
681 /* The reply msg status is handled in the following call. */
682 ret = create_ust_channel(channel, attr, &channel->uchan);
683 if (ret < 0) {
684 goto end;
685 }
686
687 channel->wait_fd = ustctl_channel_get_wait_fd(channel->uchan);
688
689 /*
690 * For the snapshots (no monitor), we create the metadata streams
691 * on demand, not during the channel creation.
692 */
693 if (channel->type == CONSUMER_CHANNEL_TYPE_METADATA && !channel->monitor) {
694 ret = 0;
695 goto end;
696 }
697
698 /* Open all streams for this channel. */
699 pthread_mutex_lock(&channel->lock);
700 ret = create_ust_streams(channel, ctx);
701 pthread_mutex_unlock(&channel->lock);
702 if (ret < 0) {
703 goto end;
704 }
705
706 end:
707 return ret;
708 }
709
710 /*
711 * Send all stream of a channel to the right thread handling it.
712 *
713 * On error, return a negative value else 0 on success.
714 */
715 static int send_streams_to_thread(struct lttng_consumer_channel *channel,
716 struct lttng_consumer_local_data *ctx)
717 {
718 int ret = 0;
719 struct lttng_consumer_stream *stream, *stmp;
720
721 assert(channel);
722 assert(ctx);
723
724 /* Send streams to the corresponding thread. */
725 cds_list_for_each_entry_safe(stream, stmp, &channel->streams.head,
726 send_node) {
727
728 health_code_update();
729
730 /* Sending the stream to the thread. */
731 ret = send_stream_to_thread(stream, ctx);
732 if (ret < 0) {
733 /*
734 * If we are unable to send the stream to the thread, there is
735 * a big problem so just stop everything.
736 */
737 goto error;
738 }
739 }
740
741 error:
742 return ret;
743 }
744
745 /*
746 * Flush channel's streams using the given key to retrieve the channel.
747 *
748 * Return 0 on success else an LTTng error code.
749 */
750 static int flush_channel(uint64_t chan_key)
751 {
752 int ret = 0;
753 struct lttng_consumer_channel *channel;
754 struct lttng_consumer_stream *stream;
755 struct lttng_ht *ht;
756 struct lttng_ht_iter iter;
757
758 DBG("UST consumer flush channel key %" PRIu64, chan_key);
759
760 rcu_read_lock();
761 channel = consumer_find_channel(chan_key);
762 if (!channel) {
763 ERR("UST consumer flush channel %" PRIu64 " not found", chan_key);
764 ret = LTTNG_ERR_UST_CHAN_NOT_FOUND;
765 goto error;
766 }
767
768 ht = consumer_data.stream_per_chan_id_ht;
769
770 /* For each stream of the channel id, flush it. */
771 cds_lfht_for_each_entry_duplicate(ht->ht,
772 ht->hash_fct(&channel->key, lttng_ht_seed), ht->match_fct,
773 &channel->key, &iter.iter, stream, node_channel_id.node) {
774
775 health_code_update();
776
777 pthread_mutex_lock(&stream->lock);
778 if (!stream->quiescent) {
779 ustctl_flush_buffer(stream->ustream, 0);
780 stream->quiescent = true;
781 }
782 pthread_mutex_unlock(&stream->lock);
783 }
784 error:
785 rcu_read_unlock();
786 return ret;
787 }
788
789 /*
790 * Clear quiescent state from channel's streams using the given key to
791 * retrieve the channel.
792 *
793 * Return 0 on success else an LTTng error code.
794 */
795 static int clear_quiescent_channel(uint64_t chan_key)
796 {
797 int ret = 0;
798 struct lttng_consumer_channel *channel;
799 struct lttng_consumer_stream *stream;
800 struct lttng_ht *ht;
801 struct lttng_ht_iter iter;
802
803 DBG("UST consumer clear quiescent channel key %" PRIu64, chan_key);
804
805 rcu_read_lock();
806 channel = consumer_find_channel(chan_key);
807 if (!channel) {
808 ERR("UST consumer clear quiescent channel %" PRIu64 " not found", chan_key);
809 ret = LTTNG_ERR_UST_CHAN_NOT_FOUND;
810 goto error;
811 }
812
813 ht = consumer_data.stream_per_chan_id_ht;
814
815 /* For each stream of the channel id, clear quiescent state. */
816 cds_lfht_for_each_entry_duplicate(ht->ht,
817 ht->hash_fct(&channel->key, lttng_ht_seed), ht->match_fct,
818 &channel->key, &iter.iter, stream, node_channel_id.node) {
819
820 health_code_update();
821
822 pthread_mutex_lock(&stream->lock);
823 stream->quiescent = false;
824 pthread_mutex_unlock(&stream->lock);
825 }
826 error:
827 rcu_read_unlock();
828 return ret;
829 }
830
831 /*
832 * Close metadata stream wakeup_fd using the given key to retrieve the channel.
833 *
834 * Return 0 on success else an LTTng error code.
835 */
836 static int close_metadata(uint64_t chan_key)
837 {
838 int ret = 0;
839 struct lttng_consumer_channel *channel;
840 unsigned int channel_monitor;
841
842 DBG("UST consumer close metadata key %" PRIu64, chan_key);
843
844 channel = consumer_find_channel(chan_key);
845 if (!channel) {
846 /*
847 * This is possible if the metadata thread has issue a delete because
848 * the endpoint point of the stream hung up. There is no way the
849 * session daemon can know about it thus use a DBG instead of an actual
850 * error.
851 */
852 DBG("UST consumer close metadata %" PRIu64 " not found", chan_key);
853 ret = LTTNG_ERR_UST_CHAN_NOT_FOUND;
854 goto error;
855 }
856
857 pthread_mutex_lock(&consumer_data.lock);
858 pthread_mutex_lock(&channel->lock);
859 channel_monitor = channel->monitor;
860 if (cds_lfht_is_node_deleted(&channel->node.node)) {
861 goto error_unlock;
862 }
863
864 lttng_ustconsumer_close_metadata(channel);
865 pthread_mutex_unlock(&channel->lock);
866 pthread_mutex_unlock(&consumer_data.lock);
867
868 /*
869 * The ownership of a metadata channel depends on the type of
870 * session to which it belongs. In effect, the monitor flag is checked
871 * to determine if this metadata channel is in "snapshot" mode or not.
872 *
873 * In the non-snapshot case, the metadata channel is created along with
874 * a single stream which will remain present until the metadata channel
875 * is destroyed (on the destruction of its session). In this case, the
876 * metadata stream in "monitored" by the metadata poll thread and holds
877 * the ownership of its channel.
878 *
879 * Closing the metadata will cause the metadata stream's "metadata poll
880 * pipe" to be closed. Closing this pipe will wake-up the metadata poll
881 * thread which will teardown the metadata stream which, in return,
882 * deletes the metadata channel.
883 *
884 * In the snapshot case, the metadata stream is created and destroyed
885 * on every snapshot record. Since the channel doesn't have an owner
886 * other than the session daemon, it is safe to destroy it immediately
887 * on reception of the CLOSE_METADATA command.
888 */
889 if (!channel_monitor) {
890 /*
891 * The channel and consumer_data locks must be
892 * released before this call since consumer_del_channel
893 * re-acquires the channel and consumer_data locks to teardown
894 * the channel and queue its reclamation by the "call_rcu"
895 * worker thread.
896 */
897 consumer_del_channel(channel);
898 }
899
900 return ret;
901 error_unlock:
902 pthread_mutex_unlock(&channel->lock);
903 pthread_mutex_unlock(&consumer_data.lock);
904 error:
905 return ret;
906 }
907
908 /*
909 * RCU read side lock MUST be acquired before calling this function.
910 *
911 * Return 0 on success else an LTTng error code.
912 */
913 static int setup_metadata(struct lttng_consumer_local_data *ctx, uint64_t key)
914 {
915 int ret;
916 struct lttng_consumer_channel *metadata;
917
918 DBG("UST consumer setup metadata key %" PRIu64, key);
919
920 metadata = consumer_find_channel(key);
921 if (!metadata) {
922 ERR("UST consumer push metadata %" PRIu64 " not found", key);
923 ret = LTTNG_ERR_UST_CHAN_NOT_FOUND;
924 goto end;
925 }
926
927 /*
928 * In no monitor mode, the metadata channel has no stream(s) so skip the
929 * ownership transfer to the metadata thread.
930 */
931 if (!metadata->monitor) {
932 DBG("Metadata channel in no monitor");
933 ret = 0;
934 goto end;
935 }
936
937 /*
938 * Send metadata stream to relayd if one available. Availability is
939 * known if the stream is still in the list of the channel.
940 */
941 if (cds_list_empty(&metadata->streams.head)) {
942 ERR("Metadata channel key %" PRIu64 ", no stream available.", key);
943 ret = LTTCOMM_CONSUMERD_ERROR_METADATA;
944 goto error_no_stream;
945 }
946
947 /* Send metadata stream to relayd if needed. */
948 if (metadata->metadata_stream->net_seq_idx != (uint64_t) -1ULL) {
949 ret = consumer_send_relayd_stream(metadata->metadata_stream,
950 metadata->pathname);
951 if (ret < 0) {
952 ret = LTTCOMM_CONSUMERD_ERROR_METADATA;
953 goto error;
954 }
955 ret = consumer_send_relayd_streams_sent(
956 metadata->metadata_stream->net_seq_idx);
957 if (ret < 0) {
958 ret = LTTCOMM_CONSUMERD_RELAYD_FAIL;
959 goto error;
960 }
961 }
962
963 /*
964 * Ownership of metadata stream is passed along. Freeing is handled by
965 * the callee.
966 */
967 ret = send_streams_to_thread(metadata, ctx);
968 if (ret < 0) {
969 /*
970 * If we are unable to send the stream to the thread, there is
971 * a big problem so just stop everything.
972 */
973 ret = LTTCOMM_CONSUMERD_FATAL;
974 goto send_streams_error;
975 }
976 /* List MUST be empty after or else it could be reused. */
977 assert(cds_list_empty(&metadata->streams.head));
978
979 ret = 0;
980 goto end;
981
982 error:
983 /*
984 * Delete metadata channel on error. At this point, the metadata stream can
985 * NOT be monitored by the metadata thread thus having the guarantee that
986 * the stream is still in the local stream list of the channel. This call
987 * will make sure to clean that list.
988 */
989 consumer_stream_destroy(metadata->metadata_stream, NULL);
990 cds_list_del(&metadata->metadata_stream->send_node);
991 metadata->metadata_stream = NULL;
992 send_streams_error:
993 error_no_stream:
994 end:
995 return ret;
996 }
997
998 /*
999 * Snapshot the whole metadata.
1000 * RCU read-side lock must be held by the caller.
1001 *
1002 * Returns 0 on success, < 0 on error
1003 */
1004 static int snapshot_metadata(struct lttng_consumer_channel *metadata_channel,
1005 uint64_t key, char *path, uint64_t relayd_id,
1006 struct lttng_consumer_local_data *ctx)
1007 {
1008 int ret = 0;
1009 struct lttng_consumer_stream *metadata_stream;
1010
1011 assert(path);
1012 assert(ctx);
1013
1014 DBG("UST consumer snapshot metadata with key %" PRIu64 " at path %s",
1015 key, path);
1016
1017 rcu_read_lock();
1018
1019 assert(!metadata_channel->monitor);
1020
1021 health_code_update();
1022
1023 /*
1024 * Ask the sessiond if we have new metadata waiting and update the
1025 * consumer metadata cache.
1026 */
1027 ret = lttng_ustconsumer_request_metadata(ctx, metadata_channel, 0, 1);
1028 if (ret < 0) {
1029 goto error;
1030 }
1031
1032 health_code_update();
1033
1034 /*
1035 * The metadata stream is NOT created in no monitor mode when the channel
1036 * is created on a sessiond ask channel command.
1037 */
1038 ret = create_ust_streams(metadata_channel, ctx);
1039 if (ret < 0) {
1040 goto error;
1041 }
1042
1043 metadata_stream = metadata_channel->metadata_stream;
1044 assert(metadata_stream);
1045
1046 pthread_mutex_lock(&metadata_stream->lock);
1047 if (relayd_id != (uint64_t) -1ULL) {
1048 metadata_stream->net_seq_idx = relayd_id;
1049 ret = consumer_send_relayd_stream(metadata_stream, path);
1050 } else {
1051 ret = consumer_stream_create_output_files(metadata_stream,
1052 false);
1053 }
1054 pthread_mutex_unlock(&metadata_stream->lock);
1055 if (ret < 0) {
1056 goto error_stream;
1057 }
1058
1059 do {
1060 health_code_update();
1061
1062 ret = lttng_consumer_read_subbuffer(metadata_stream, ctx);
1063 if (ret < 0) {
1064 goto error_stream;
1065 }
1066 } while (ret > 0);
1067
1068 error_stream:
1069 /*
1070 * Clean up the stream completly because the next snapshot will use a new
1071 * metadata stream.
1072 */
1073 pthread_mutex_lock(&metadata_stream->lock);
1074 consumer_stream_destroy(metadata_stream, NULL);
1075 cds_list_del(&metadata_stream->send_node);
1076 metadata_channel->metadata_stream = NULL;
1077
1078 error:
1079 rcu_read_unlock();
1080 return ret;
1081 }
1082
1083 /*
1084 * Take a snapshot of all the stream of a channel.
1085 * RCU read-side lock and the channel lock must be held by the caller.
1086 *
1087 * Returns 0 on success, < 0 on error
1088 */
1089 static int snapshot_channel(struct lttng_consumer_channel *channel,
1090 uint64_t key, char *path, uint64_t relayd_id,
1091 uint64_t nb_packets_per_stream,
1092 struct lttng_consumer_local_data *ctx)
1093 {
1094 int ret;
1095 unsigned use_relayd = 0;
1096 unsigned long consumed_pos, produced_pos;
1097 struct lttng_consumer_stream *stream;
1098
1099 assert(path);
1100 assert(ctx);
1101
1102 rcu_read_lock();
1103
1104 if (relayd_id != (uint64_t) -1ULL) {
1105 use_relayd = 1;
1106 }
1107
1108 assert(!channel->monitor);
1109 DBG("UST consumer snapshot channel %" PRIu64, key);
1110
1111 cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
1112 health_code_update();
1113
1114 /* Lock stream because we are about to change its state. */
1115 pthread_mutex_lock(&stream->lock);
1116 assert(channel->trace_chunk);
1117 if (!lttng_trace_chunk_get(channel->trace_chunk)) {
1118 /*
1119 * Can't happen barring an internal error as the channel
1120 * holds a reference to the trace chunk.
1121 */
1122 ERR("Failed to acquire reference to channel's trace chunk");
1123 ret = -1;
1124 goto error_unlock;
1125 }
1126 assert(!stream->trace_chunk);
1127 stream->trace_chunk = channel->trace_chunk;
1128
1129 stream->net_seq_idx = relayd_id;
1130
1131 if (use_relayd) {
1132 ret = consumer_send_relayd_stream(stream, path);
1133 if (ret < 0) {
1134 goto error_unlock;
1135 }
1136 } else {
1137 ret = consumer_stream_create_output_files(stream,
1138 false);
1139 if (ret < 0) {
1140 goto error_unlock;
1141 }
1142 DBG("UST consumer snapshot stream (%" PRIu64 ")",
1143 stream->key);
1144 }
1145
1146 /*
1147 * If tracing is active, we want to perform a "full" buffer flush.
1148 * Else, if quiescent, it has already been done by the prior stop.
1149 */
1150 if (!stream->quiescent) {
1151 ustctl_flush_buffer(stream->ustream, 0);
1152 }
1153
1154 ret = lttng_ustconsumer_take_snapshot(stream);
1155 if (ret < 0) {
1156 ERR("Taking UST snapshot");
1157 goto error_unlock;
1158 }
1159
1160 ret = lttng_ustconsumer_get_produced_snapshot(stream, &produced_pos);
1161 if (ret < 0) {
1162 ERR("Produced UST snapshot position");
1163 goto error_unlock;
1164 }
1165
1166 ret = lttng_ustconsumer_get_consumed_snapshot(stream, &consumed_pos);
1167 if (ret < 0) {
1168 ERR("Consumerd UST snapshot position");
1169 goto error_unlock;
1170 }
1171
1172 /*
1173 * The original value is sent back if max stream size is larger than
1174 * the possible size of the snapshot. Also, we assume that the session
1175 * daemon should never send a maximum stream size that is lower than
1176 * subbuffer size.
1177 */
1178 consumed_pos = consumer_get_consume_start_pos(consumed_pos,
1179 produced_pos, nb_packets_per_stream,
1180 stream->max_sb_size);
1181
1182 while ((long) (consumed_pos - produced_pos) < 0) {
1183 ssize_t read_len;
1184 unsigned long len, padded_len;
1185
1186 health_code_update();
1187
1188 DBG("UST consumer taking snapshot at pos %lu", consumed_pos);
1189
1190 ret = ustctl_get_subbuf(stream->ustream, &consumed_pos);
1191 if (ret < 0) {
1192 if (ret != -EAGAIN) {
1193 PERROR("ustctl_get_subbuf snapshot");
1194 goto error_close_stream;
1195 }
1196 DBG("UST consumer get subbuf failed. Skipping it.");
1197 consumed_pos += stream->max_sb_size;
1198 stream->chan->lost_packets++;
1199 continue;
1200 }
1201
1202 ret = ustctl_get_subbuf_size(stream->ustream, &len);
1203 if (ret < 0) {
1204 ERR("Snapshot ustctl_get_subbuf_size");
1205 goto error_put_subbuf;
1206 }
1207
1208 ret = ustctl_get_padded_subbuf_size(stream->ustream, &padded_len);
1209 if (ret < 0) {
1210 ERR("Snapshot ustctl_get_padded_subbuf_size");
1211 goto error_put_subbuf;
1212 }
1213
1214 read_len = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, len,
1215 padded_len - len, NULL);
1216 if (use_relayd) {
1217 if (read_len != len) {
1218 ret = -EPERM;
1219 goto error_put_subbuf;
1220 }
1221 } else {
1222 if (read_len != padded_len) {
1223 ret = -EPERM;
1224 goto error_put_subbuf;
1225 }
1226 }
1227
1228 ret = ustctl_put_subbuf(stream->ustream);
1229 if (ret < 0) {
1230 ERR("Snapshot ustctl_put_subbuf");
1231 goto error_close_stream;
1232 }
1233 consumed_pos += stream->max_sb_size;
1234 }
1235
1236 /* Simply close the stream so we can use it on the next snapshot. */
1237 consumer_stream_close(stream);
1238 pthread_mutex_unlock(&stream->lock);
1239 }
1240
1241 rcu_read_unlock();
1242 return 0;
1243
1244 error_put_subbuf:
1245 if (ustctl_put_subbuf(stream->ustream) < 0) {
1246 ERR("Snapshot ustctl_put_subbuf");
1247 }
1248 error_close_stream:
1249 consumer_stream_close(stream);
1250 error_unlock:
1251 pthread_mutex_unlock(&stream->lock);
1252 rcu_read_unlock();
1253 return ret;
1254 }
1255
1256 /*
1257 * Receive the metadata updates from the sessiond. Supports receiving
1258 * overlapping metadata, but is needs to always belong to a contiguous
1259 * range starting from 0.
1260 * Be careful about the locks held when calling this function: it needs
1261 * the metadata cache flush to concurrently progress in order to
1262 * complete.
1263 */
1264 int lttng_ustconsumer_recv_metadata(int sock, uint64_t key, uint64_t offset,
1265 uint64_t len, uint64_t version,
1266 struct lttng_consumer_channel *channel, int timer, int wait)
1267 {
1268 int ret, ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1269 char *metadata_str;
1270
1271 DBG("UST consumer push metadata key %" PRIu64 " of len %" PRIu64, key, len);
1272
1273 metadata_str = zmalloc(len * sizeof(char));
1274 if (!metadata_str) {
1275 PERROR("zmalloc metadata string");
1276 ret_code = LTTCOMM_CONSUMERD_ENOMEM;
1277 goto end;
1278 }
1279
1280 health_code_update();
1281
1282 /* Receive metadata string. */
1283 ret = lttcomm_recv_unix_sock(sock, metadata_str, len);
1284 if (ret < 0) {
1285 /* Session daemon is dead so return gracefully. */
1286 ret_code = ret;
1287 goto end_free;
1288 }
1289
1290 health_code_update();
1291
1292 pthread_mutex_lock(&channel->metadata_cache->lock);
1293 ret = consumer_metadata_cache_write(channel, offset, len, version,
1294 metadata_str);
1295 if (ret < 0) {
1296 /* Unable to handle metadata. Notify session daemon. */
1297 ret_code = LTTCOMM_CONSUMERD_ERROR_METADATA;
1298 /*
1299 * Skip metadata flush on write error since the offset and len might
1300 * not have been updated which could create an infinite loop below when
1301 * waiting for the metadata cache to be flushed.
1302 */
1303 pthread_mutex_unlock(&channel->metadata_cache->lock);
1304 goto end_free;
1305 }
1306 pthread_mutex_unlock(&channel->metadata_cache->lock);
1307
1308 if (!wait) {
1309 goto end_free;
1310 }
1311 while (consumer_metadata_cache_flushed(channel, offset + len, timer)) {
1312 DBG("Waiting for metadata to be flushed");
1313
1314 health_code_update();
1315
1316 usleep(DEFAULT_METADATA_AVAILABILITY_WAIT_TIME);
1317 }
1318
1319 end_free:
1320 free(metadata_str);
1321 end:
1322 return ret_code;
1323 }
1324
1325 /*
1326 * Receive command from session daemon and process it.
1327 *
1328 * Return 1 on success else a negative value or 0.
1329 */
1330 int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
1331 int sock, struct pollfd *consumer_sockpoll)
1332 {
1333 ssize_t ret;
1334 enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1335 struct lttcomm_consumer_msg msg;
1336 struct lttng_consumer_channel *channel = NULL;
1337
1338 health_code_update();
1339
1340 ret = lttcomm_recv_unix_sock(sock, &msg, sizeof(msg));
1341 if (ret != sizeof(msg)) {
1342 DBG("Consumer received unexpected message size %zd (expects %zu)",
1343 ret, sizeof(msg));
1344 /*
1345 * The ret value might 0 meaning an orderly shutdown but this is ok
1346 * since the caller handles this.
1347 */
1348 if (ret > 0) {
1349 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD);
1350 ret = -1;
1351 }
1352 return ret;
1353 }
1354
1355 health_code_update();
1356
1357 /* deprecated */
1358 assert(msg.cmd_type != LTTNG_CONSUMER_STOP);
1359
1360 health_code_update();
1361
1362 /* relayd needs RCU read-side lock */
1363 rcu_read_lock();
1364
1365 switch (msg.cmd_type) {
1366 case LTTNG_CONSUMER_ADD_RELAYD_SOCKET:
1367 {
1368 /* Session daemon status message are handled in the following call. */
1369 consumer_add_relayd_socket(msg.u.relayd_sock.net_index,
1370 msg.u.relayd_sock.type, ctx, sock, consumer_sockpoll,
1371 &msg.u.relayd_sock.sock, msg.u.relayd_sock.session_id,
1372 msg.u.relayd_sock.relayd_session_id);
1373 goto end_nosignal;
1374 }
1375 case LTTNG_CONSUMER_DESTROY_RELAYD:
1376 {
1377 uint64_t index = msg.u.destroy_relayd.net_seq_idx;
1378 struct consumer_relayd_sock_pair *relayd;
1379
1380 DBG("UST consumer destroying relayd %" PRIu64, index);
1381
1382 /* Get relayd reference if exists. */
1383 relayd = consumer_find_relayd(index);
1384 if (relayd == NULL) {
1385 DBG("Unable to find relayd %" PRIu64, index);
1386 ret_code = LTTCOMM_CONSUMERD_RELAYD_FAIL;
1387 }
1388
1389 /*
1390 * Each relayd socket pair has a refcount of stream attached to it
1391 * which tells if the relayd is still active or not depending on the
1392 * refcount value.
1393 *
1394 * This will set the destroy flag of the relayd object and destroy it
1395 * if the refcount reaches zero when called.
1396 *
1397 * The destroy can happen either here or when a stream fd hangs up.
1398 */
1399 if (relayd) {
1400 consumer_flag_relayd_for_destroy(relayd);
1401 }
1402
1403 goto end_msg_sessiond;
1404 }
1405 case LTTNG_CONSUMER_UPDATE_STREAM:
1406 {
1407 rcu_read_unlock();
1408 return -ENOSYS;
1409 }
1410 case LTTNG_CONSUMER_DATA_PENDING:
1411 {
1412 int ret, is_data_pending;
1413 uint64_t id = msg.u.data_pending.session_id;
1414
1415 DBG("UST consumer data pending command for id %" PRIu64, id);
1416
1417 is_data_pending = consumer_data_pending(id);
1418
1419 /* Send back returned value to session daemon */
1420 ret = lttcomm_send_unix_sock(sock, &is_data_pending,
1421 sizeof(is_data_pending));
1422 if (ret < 0) {
1423 DBG("Error when sending the data pending ret code: %d", ret);
1424 goto error_fatal;
1425 }
1426
1427 /*
1428 * No need to send back a status message since the data pending
1429 * returned value is the response.
1430 */
1431 break;
1432 }
1433 case LTTNG_CONSUMER_ASK_CHANNEL_CREATION:
1434 {
1435 int ret;
1436 struct ustctl_consumer_channel_attr attr;
1437 const uint64_t chunk_id = msg.u.ask_channel.chunk_id.value;
1438 const struct lttng_credentials buffer_credentials = {
1439 .uid = msg.u.ask_channel.buffer_credentials.uid,
1440 .gid = msg.u.ask_channel.buffer_credentials.gid,
1441 };
1442
1443 /* Create a plain object and reserve a channel key. */
1444 channel = allocate_channel(msg.u.ask_channel.session_id,
1445 msg.u.ask_channel.chunk_id.is_set ?
1446 &chunk_id : NULL,
1447 msg.u.ask_channel.pathname,
1448 msg.u.ask_channel.name,
1449 msg.u.ask_channel.relayd_id,
1450 msg.u.ask_channel.key,
1451 (enum lttng_event_output) msg.u.ask_channel.output,
1452 msg.u.ask_channel.tracefile_size,
1453 msg.u.ask_channel.tracefile_count,
1454 msg.u.ask_channel.session_id_per_pid,
1455 msg.u.ask_channel.monitor,
1456 msg.u.ask_channel.live_timer_interval,
1457 msg.u.ask_channel.root_shm_path,
1458 msg.u.ask_channel.shm_path);
1459 if (!channel) {
1460 goto end_channel_error;
1461 }
1462
1463 LTTNG_OPTIONAL_SET(&channel->buffer_credentials,
1464 buffer_credentials);
1465
1466 /*
1467 * Assign UST application UID to the channel. This value is ignored for
1468 * per PID buffers. This is specific to UST thus setting this after the
1469 * allocation.
1470 */
1471 channel->ust_app_uid = msg.u.ask_channel.ust_app_uid;
1472
1473 /* Build channel attributes from received message. */
1474 attr.subbuf_size = msg.u.ask_channel.subbuf_size;
1475 attr.num_subbuf = msg.u.ask_channel.num_subbuf;
1476 attr.overwrite = msg.u.ask_channel.overwrite;
1477 attr.switch_timer_interval = msg.u.ask_channel.switch_timer_interval;
1478 attr.read_timer_interval = msg.u.ask_channel.read_timer_interval;
1479 attr.chan_id = msg.u.ask_channel.chan_id;
1480 memcpy(attr.uuid, msg.u.ask_channel.uuid, sizeof(attr.uuid));
1481 attr.blocking_timeout= msg.u.ask_channel.blocking_timeout;
1482
1483 /* Match channel buffer type to the UST abi. */
1484 switch (msg.u.ask_channel.output) {
1485 case LTTNG_EVENT_MMAP:
1486 default:
1487 attr.output = LTTNG_UST_MMAP;
1488 break;
1489 }
1490
1491 /* Translate and save channel type. */
1492 switch (msg.u.ask_channel.type) {
1493 case LTTNG_UST_CHAN_PER_CPU:
1494 channel->type = CONSUMER_CHANNEL_TYPE_DATA;
1495 attr.type = LTTNG_UST_CHAN_PER_CPU;
1496 /*
1497 * Set refcount to 1 for owner. Below, we will
1498 * pass ownership to the
1499 * consumer_thread_channel_poll() thread.
1500 */
1501 channel->refcount = 1;
1502 break;
1503 case LTTNG_UST_CHAN_METADATA:
1504 channel->type = CONSUMER_CHANNEL_TYPE_METADATA;
1505 attr.type = LTTNG_UST_CHAN_METADATA;
1506 break;
1507 default:
1508 assert(0);
1509 goto error_fatal;
1510 };
1511
1512 health_code_update();
1513
1514 ret = ask_channel(ctx, channel, &attr);
1515 if (ret < 0) {
1516 goto end_channel_error;
1517 }
1518
1519 if (msg.u.ask_channel.type == LTTNG_UST_CHAN_METADATA) {
1520 ret = consumer_metadata_cache_allocate(channel);
1521 if (ret < 0) {
1522 ERR("Allocating metadata cache");
1523 goto end_channel_error;
1524 }
1525 consumer_timer_switch_start(channel, attr.switch_timer_interval);
1526 attr.switch_timer_interval = 0;
1527 } else {
1528 int monitor_start_ret;
1529
1530 consumer_timer_live_start(channel,
1531 msg.u.ask_channel.live_timer_interval);
1532 monitor_start_ret = consumer_timer_monitor_start(
1533 channel,
1534 msg.u.ask_channel.monitor_timer_interval);
1535 if (monitor_start_ret < 0) {
1536 ERR("Starting channel monitoring timer failed");
1537 goto end_channel_error;
1538 }
1539 }
1540
1541 health_code_update();
1542
1543 /*
1544 * Add the channel to the internal state AFTER all streams were created
1545 * and successfully sent to session daemon. This way, all streams must
1546 * be ready before this channel is visible to the threads.
1547 * If add_channel succeeds, ownership of the channel is
1548 * passed to consumer_thread_channel_poll().
1549 */
1550 ret = add_channel(channel, ctx);
1551 if (ret < 0) {
1552 if (msg.u.ask_channel.type == LTTNG_UST_CHAN_METADATA) {
1553 if (channel->switch_timer_enabled == 1) {
1554 consumer_timer_switch_stop(channel);
1555 }
1556 consumer_metadata_cache_destroy(channel);
1557 }
1558 if (channel->live_timer_enabled == 1) {
1559 consumer_timer_live_stop(channel);
1560 }
1561 if (channel->monitor_timer_enabled == 1) {
1562 consumer_timer_monitor_stop(channel);
1563 }
1564 goto end_channel_error;
1565 }
1566
1567 health_code_update();
1568
1569 /*
1570 * Channel and streams are now created. Inform the session daemon that
1571 * everything went well and should wait to receive the channel and
1572 * streams with ustctl API.
1573 */
1574 ret = consumer_send_status_channel(sock, channel);
1575 if (ret < 0) {
1576 /*
1577 * There is probably a problem on the socket.
1578 */
1579 goto error_fatal;
1580 }
1581
1582 break;
1583 }
1584 case LTTNG_CONSUMER_GET_CHANNEL:
1585 {
1586 int ret, relayd_err = 0;
1587 uint64_t key = msg.u.get_channel.key;
1588 struct lttng_consumer_channel *channel;
1589
1590 channel = consumer_find_channel(key);
1591 if (!channel) {
1592 ERR("UST consumer get channel key %" PRIu64 " not found", key);
1593 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
1594 goto end_get_channel;
1595 }
1596
1597 health_code_update();
1598
1599 /* Send the channel to sessiond (and relayd, if applicable). */
1600 ret = send_channel_to_sessiond_and_relayd(sock, channel, ctx,
1601 &relayd_err);
1602 if (ret < 0) {
1603 if (relayd_err) {
1604 /*
1605 * We were unable to send to the relayd the stream so avoid
1606 * sending back a fatal error to the thread since this is OK
1607 * and the consumer can continue its work. The above call
1608 * has sent the error status message to the sessiond.
1609 */
1610 goto end_get_channel_nosignal;
1611 }
1612 /*
1613 * The communicaton was broken hence there is a bad state between
1614 * the consumer and sessiond so stop everything.
1615 */
1616 goto error_get_channel_fatal;
1617 }
1618
1619 health_code_update();
1620
1621 /*
1622 * In no monitor mode, the streams ownership is kept inside the channel
1623 * so don't send them to the data thread.
1624 */
1625 if (!channel->monitor) {
1626 goto end_get_channel;
1627 }
1628
1629 ret = send_streams_to_thread(channel, ctx);
1630 if (ret < 0) {
1631 /*
1632 * If we are unable to send the stream to the thread, there is
1633 * a big problem so just stop everything.
1634 */
1635 goto error_get_channel_fatal;
1636 }
1637 /* List MUST be empty after or else it could be reused. */
1638 assert(cds_list_empty(&channel->streams.head));
1639 end_get_channel:
1640 goto end_msg_sessiond;
1641 error_get_channel_fatal:
1642 goto error_fatal;
1643 end_get_channel_nosignal:
1644 goto end_nosignal;
1645 }
1646 case LTTNG_CONSUMER_DESTROY_CHANNEL:
1647 {
1648 uint64_t key = msg.u.destroy_channel.key;
1649
1650 /*
1651 * Only called if streams have not been sent to stream
1652 * manager thread. However, channel has been sent to
1653 * channel manager thread.
1654 */
1655 notify_thread_del_channel(ctx, key);
1656 goto end_msg_sessiond;
1657 }
1658 case LTTNG_CONSUMER_CLOSE_METADATA:
1659 {
1660 int ret;
1661
1662 ret = close_metadata(msg.u.close_metadata.key);
1663 if (ret != 0) {
1664 ret_code = ret;
1665 }
1666
1667 goto end_msg_sessiond;
1668 }
1669 case LTTNG_CONSUMER_FLUSH_CHANNEL:
1670 {
1671 int ret;
1672
1673 ret = flush_channel(msg.u.flush_channel.key);
1674 if (ret != 0) {
1675 ret_code = ret;
1676 }
1677
1678 goto end_msg_sessiond;
1679 }
1680 case LTTNG_CONSUMER_CLEAR_QUIESCENT_CHANNEL:
1681 {
1682 int ret;
1683
1684 ret = clear_quiescent_channel(
1685 msg.u.clear_quiescent_channel.key);
1686 if (ret != 0) {
1687 ret_code = ret;
1688 }
1689
1690 goto end_msg_sessiond;
1691 }
1692 case LTTNG_CONSUMER_PUSH_METADATA:
1693 {
1694 int ret;
1695 uint64_t len = msg.u.push_metadata.len;
1696 uint64_t key = msg.u.push_metadata.key;
1697 uint64_t offset = msg.u.push_metadata.target_offset;
1698 uint64_t version = msg.u.push_metadata.version;
1699 struct lttng_consumer_channel *channel;
1700
1701 DBG("UST consumer push metadata key %" PRIu64 " of len %" PRIu64, key,
1702 len);
1703
1704 channel = consumer_find_channel(key);
1705 if (!channel) {
1706 /*
1707 * This is possible if the metadata creation on the consumer side
1708 * is in flight vis-a-vis a concurrent push metadata from the
1709 * session daemon. Simply return that the channel failed and the
1710 * session daemon will handle that message correctly considering
1711 * that this race is acceptable thus the DBG() statement here.
1712 */
1713 DBG("UST consumer push metadata %" PRIu64 " not found", key);
1714 ret_code = LTTCOMM_CONSUMERD_CHANNEL_FAIL;
1715 goto end_msg_sessiond;
1716 }
1717
1718 health_code_update();
1719
1720 if (!len) {
1721 /*
1722 * There is nothing to receive. We have simply
1723 * checked whether the channel can be found.
1724 */
1725 ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1726 goto end_msg_sessiond;
1727 }
1728
1729 /* Tell session daemon we are ready to receive the metadata. */
1730 ret = consumer_send_status_msg(sock, LTTCOMM_CONSUMERD_SUCCESS);
1731 if (ret < 0) {
1732 /* Somehow, the session daemon is not responding anymore. */
1733 goto error_fatal;
1734 }
1735
1736 health_code_update();
1737
1738 /* Wait for more data. */
1739 health_poll_entry();
1740 ret = lttng_consumer_poll_socket(consumer_sockpoll);
1741 health_poll_exit();
1742 if (ret) {
1743 goto error_fatal;
1744 }
1745
1746 health_code_update();
1747
1748 ret = lttng_ustconsumer_recv_metadata(sock, key, offset,
1749 len, version, channel, 0, 1);
1750 if (ret < 0) {
1751 /* error receiving from sessiond */
1752 goto error_fatal;
1753 } else {
1754 ret_code = ret;
1755 goto end_msg_sessiond;
1756 }
1757 }
1758 case LTTNG_CONSUMER_SETUP_METADATA:
1759 {
1760 int ret;
1761
1762 ret = setup_metadata(ctx, msg.u.setup_metadata.key);
1763 if (ret) {
1764 ret_code = ret;
1765 }
1766 goto end_msg_sessiond;
1767 }
1768 case LTTNG_CONSUMER_SNAPSHOT_CHANNEL:
1769 {
1770 struct lttng_consumer_channel *channel;
1771 uint64_t key = msg.u.snapshot_channel.key;
1772
1773 channel = consumer_find_channel(key);
1774 if (!channel) {
1775 DBG("UST snapshot channel not found for key %" PRIu64, key);
1776 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
1777 } else {
1778 if (msg.u.snapshot_channel.metadata) {
1779 ret = snapshot_metadata(channel, key,
1780 msg.u.snapshot_channel.pathname,
1781 msg.u.snapshot_channel.relayd_id,
1782 ctx);
1783 if (ret < 0) {
1784 ERR("Snapshot metadata failed");
1785 ret_code = LTTCOMM_CONSUMERD_SNAPSHOT_FAILED;
1786 }
1787 } else {
1788 ret = snapshot_channel(channel, key,
1789 msg.u.snapshot_channel.pathname,
1790 msg.u.snapshot_channel.relayd_id,
1791 msg.u.snapshot_channel.nb_packets_per_stream,
1792 ctx);
1793 if (ret < 0) {
1794 ERR("Snapshot channel failed");
1795 ret_code = LTTCOMM_CONSUMERD_SNAPSHOT_FAILED;
1796 }
1797 }
1798 }
1799 health_code_update();
1800 ret = consumer_send_status_msg(sock, ret_code);
1801 if (ret < 0) {
1802 /* Somehow, the session daemon is not responding anymore. */
1803 goto end_nosignal;
1804 }
1805 health_code_update();
1806 break;
1807 }
1808 case LTTNG_CONSUMER_DISCARDED_EVENTS:
1809 {
1810 int ret = 0;
1811 uint64_t discarded_events;
1812 struct lttng_ht_iter iter;
1813 struct lttng_ht *ht;
1814 struct lttng_consumer_stream *stream;
1815 uint64_t id = msg.u.discarded_events.session_id;
1816 uint64_t key = msg.u.discarded_events.channel_key;
1817
1818 DBG("UST consumer discarded events command for session id %"
1819 PRIu64, id);
1820 rcu_read_lock();
1821 pthread_mutex_lock(&consumer_data.lock);
1822
1823 ht = consumer_data.stream_list_ht;
1824
1825 /*
1826 * We only need a reference to the channel, but they are not
1827 * directly indexed, so we just use the first matching stream
1828 * to extract the information we need, we default to 0 if not
1829 * found (no events are dropped if the channel is not yet in
1830 * use).
1831 */
1832 discarded_events = 0;
1833 cds_lfht_for_each_entry_duplicate(ht->ht,
1834 ht->hash_fct(&id, lttng_ht_seed),
1835 ht->match_fct, &id,
1836 &iter.iter, stream, node_session_id.node) {
1837 if (stream->chan->key == key) {
1838 discarded_events = stream->chan->discarded_events;
1839 break;
1840 }
1841 }
1842 pthread_mutex_unlock(&consumer_data.lock);
1843 rcu_read_unlock();
1844
1845 DBG("UST consumer discarded events command for session id %"
1846 PRIu64 ", channel key %" PRIu64, id, key);
1847
1848 health_code_update();
1849
1850 /* Send back returned value to session daemon */
1851 ret = lttcomm_send_unix_sock(sock, &discarded_events, sizeof(discarded_events));
1852 if (ret < 0) {
1853 PERROR("send discarded events");
1854 goto error_fatal;
1855 }
1856
1857 break;
1858 }
1859 case LTTNG_CONSUMER_LOST_PACKETS:
1860 {
1861 int ret;
1862 uint64_t lost_packets;
1863 struct lttng_ht_iter iter;
1864 struct lttng_ht *ht;
1865 struct lttng_consumer_stream *stream;
1866 uint64_t id = msg.u.lost_packets.session_id;
1867 uint64_t key = msg.u.lost_packets.channel_key;
1868
1869 DBG("UST consumer lost packets command for session id %"
1870 PRIu64, id);
1871 rcu_read_lock();
1872 pthread_mutex_lock(&consumer_data.lock);
1873
1874 ht = consumer_data.stream_list_ht;
1875
1876 /*
1877 * We only need a reference to the channel, but they are not
1878 * directly indexed, so we just use the first matching stream
1879 * to extract the information we need, we default to 0 if not
1880 * found (no packets lost if the channel is not yet in use).
1881 */
1882 lost_packets = 0;
1883 cds_lfht_for_each_entry_duplicate(ht->ht,
1884 ht->hash_fct(&id, lttng_ht_seed),
1885 ht->match_fct, &id,
1886 &iter.iter, stream, node_session_id.node) {
1887 if (stream->chan->key == key) {
1888 lost_packets = stream->chan->lost_packets;
1889 break;
1890 }
1891 }
1892 pthread_mutex_unlock(&consumer_data.lock);
1893 rcu_read_unlock();
1894
1895 DBG("UST consumer lost packets command for session id %"
1896 PRIu64 ", channel key %" PRIu64, id, key);
1897
1898 health_code_update();
1899
1900 /* Send back returned value to session daemon */
1901 ret = lttcomm_send_unix_sock(sock, &lost_packets,
1902 sizeof(lost_packets));
1903 if (ret < 0) {
1904 PERROR("send lost packets");
1905 goto error_fatal;
1906 }
1907
1908 break;
1909 }
1910 case LTTNG_CONSUMER_SET_CHANNEL_MONITOR_PIPE:
1911 {
1912 int channel_monitor_pipe;
1913
1914 ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1915 /* Successfully received the command's type. */
1916 ret = consumer_send_status_msg(sock, ret_code);
1917 if (ret < 0) {
1918 goto error_fatal;
1919 }
1920
1921 ret = lttcomm_recv_fds_unix_sock(sock, &channel_monitor_pipe,
1922 1);
1923 if (ret != sizeof(channel_monitor_pipe)) {
1924 ERR("Failed to receive channel monitor pipe");
1925 goto error_fatal;
1926 }
1927
1928 DBG("Received channel monitor pipe (%d)", channel_monitor_pipe);
1929 ret = consumer_timer_thread_set_channel_monitor_pipe(
1930 channel_monitor_pipe);
1931 if (!ret) {
1932 int flags;
1933
1934 ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1935 /* Set the pipe as non-blocking. */
1936 ret = fcntl(channel_monitor_pipe, F_GETFL, 0);
1937 if (ret == -1) {
1938 PERROR("fcntl get flags of the channel monitoring pipe");
1939 goto error_fatal;
1940 }
1941 flags = ret;
1942
1943 ret = fcntl(channel_monitor_pipe, F_SETFL,
1944 flags | O_NONBLOCK);
1945 if (ret == -1) {
1946 PERROR("fcntl set O_NONBLOCK flag of the channel monitoring pipe");
1947 goto error_fatal;
1948 }
1949 DBG("Channel monitor pipe set as non-blocking");
1950 } else {
1951 ret_code = LTTCOMM_CONSUMERD_ALREADY_SET;
1952 }
1953 goto end_msg_sessiond;
1954 }
1955 case LTTNG_CONSUMER_ROTATE_CHANNEL:
1956 {
1957 struct lttng_consumer_channel *channel;
1958 uint64_t key = msg.u.rotate_channel.key;
1959
1960 channel = consumer_find_channel(key);
1961 if (!channel) {
1962 DBG("Channel %" PRIu64 " not found", key);
1963 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
1964 } else {
1965 /*
1966 * Sample the rotate position of all the streams in
1967 * this channel.
1968 */
1969 ret = lttng_consumer_rotate_channel(channel, key,
1970 msg.u.rotate_channel.relayd_id,
1971 msg.u.rotate_channel.metadata,
1972 ctx);
1973 if (ret < 0) {
1974 ERR("Rotate channel failed");
1975 ret_code = LTTCOMM_CONSUMERD_ROTATION_FAIL;
1976 }
1977
1978 health_code_update();
1979 }
1980 ret = consumer_send_status_msg(sock, ret_code);
1981 if (ret < 0) {
1982 /* Somehow, the session daemon is not responding anymore. */
1983 goto end_nosignal;
1984 }
1985
1986 /*
1987 * Rotate the streams that are ready right now.
1988 * FIXME: this is a second consecutive iteration over the
1989 * streams in a channel, there is probably a better way to
1990 * handle this, but it needs to be after the
1991 * consumer_send_status_msg() call.
1992 */
1993 if (channel) {
1994 ret = lttng_consumer_rotate_ready_streams(
1995 channel, key, ctx);
1996 if (ret < 0) {
1997 ERR("Rotate channel failed");
1998 }
1999 }
2000 break;
2001 }
2002 case LTTNG_CONSUMER_INIT:
2003 {
2004 ret_code = lttng_consumer_init_command(ctx,
2005 msg.u.init.sessiond_uuid);
2006 health_code_update();
2007 ret = consumer_send_status_msg(sock, ret_code);
2008 if (ret < 0) {
2009 /* Somehow, the session daemon is not responding anymore. */
2010 goto end_nosignal;
2011 }
2012 break;
2013 }
2014 case LTTNG_CONSUMER_CREATE_TRACE_CHUNK:
2015 {
2016 const struct lttng_credentials credentials = {
2017 .uid = msg.u.create_trace_chunk.credentials.value.uid,
2018 .gid = msg.u.create_trace_chunk.credentials.value.gid,
2019 };
2020 const bool is_local_trace =
2021 !msg.u.create_trace_chunk.relayd_id.is_set;
2022 const uint64_t relayd_id =
2023 msg.u.create_trace_chunk.relayd_id.value;
2024 const char *chunk_override_name =
2025 *msg.u.create_trace_chunk.override_name ?
2026 msg.u.create_trace_chunk.override_name :
2027 NULL;
2028 LTTNG_OPTIONAL(struct lttng_directory_handle) chunk_directory_handle =
2029 LTTNG_OPTIONAL_INIT;
2030
2031 /*
2032 * The session daemon will only provide a chunk directory file
2033 * descriptor for local traces.
2034 */
2035 if (is_local_trace) {
2036 int chunk_dirfd;
2037
2038 /* Acnowledge the reception of the command. */
2039 ret = consumer_send_status_msg(sock,
2040 LTTCOMM_CONSUMERD_SUCCESS);
2041 if (ret < 0) {
2042 /* Somehow, the session daemon is not responding anymore. */
2043 goto end_nosignal;
2044 }
2045
2046 ret = lttcomm_recv_fds_unix_sock(sock, &chunk_dirfd, 1);
2047 if (ret != sizeof(chunk_dirfd)) {
2048 ERR("Failed to receive trace chunk directory file descriptor");
2049 goto error_fatal;
2050 }
2051
2052 DBG("Received trace chunk directory fd (%d)",
2053 chunk_dirfd);
2054 ret = lttng_directory_handle_init_from_dirfd(
2055 &chunk_directory_handle.value,
2056 chunk_dirfd);
2057 if (ret) {
2058 ERR("Failed to initialize chunk directory handle from directory file descriptor");
2059 if (close(chunk_dirfd)) {
2060 PERROR("Failed to close chunk directory file descriptor");
2061 }
2062 goto error_fatal;
2063 }
2064 chunk_directory_handle.is_set = true;
2065 }
2066
2067 ret_code = lttng_consumer_create_trace_chunk(
2068 !is_local_trace ? &relayd_id : NULL,
2069 msg.u.create_trace_chunk.session_id,
2070 msg.u.create_trace_chunk.chunk_id,
2071 (time_t) msg.u.create_trace_chunk
2072 .creation_timestamp,
2073 chunk_override_name,
2074 msg.u.create_trace_chunk.credentials.is_set ?
2075 &credentials :
2076 NULL,
2077 chunk_directory_handle.is_set ?
2078 &chunk_directory_handle.value :
2079 NULL);
2080
2081 if (chunk_directory_handle.is_set) {
2082 lttng_directory_handle_fini(
2083 &chunk_directory_handle.value);
2084 }
2085 goto end_msg_sessiond;
2086 }
2087 case LTTNG_CONSUMER_CLOSE_TRACE_CHUNK:
2088 {
2089 enum lttng_trace_chunk_command_type close_command =
2090 msg.u.close_trace_chunk.close_command.value;
2091 const uint64_t relayd_id =
2092 msg.u.close_trace_chunk.relayd_id.value;
2093
2094 ret_code = lttng_consumer_close_trace_chunk(
2095 msg.u.close_trace_chunk.relayd_id.is_set ?
2096 &relayd_id :
2097 NULL,
2098 msg.u.close_trace_chunk.session_id,
2099 msg.u.close_trace_chunk.chunk_id,
2100 (time_t) msg.u.close_trace_chunk.close_timestamp,
2101 msg.u.close_trace_chunk.close_command.is_set ?
2102 &close_command :
2103 NULL);
2104 goto end_msg_sessiond;
2105 }
2106 case LTTNG_CONSUMER_TRACE_CHUNK_EXISTS:
2107 {
2108 const uint64_t relayd_id =
2109 msg.u.trace_chunk_exists.relayd_id.value;
2110
2111 ret_code = lttng_consumer_trace_chunk_exists(
2112 msg.u.trace_chunk_exists.relayd_id.is_set ?
2113 &relayd_id : NULL,
2114 msg.u.trace_chunk_exists.session_id,
2115 msg.u.trace_chunk_exists.chunk_id);
2116 goto end_msg_sessiond;
2117 }
2118 default:
2119 break;
2120 }
2121
2122 end_nosignal:
2123 /*
2124 * Return 1 to indicate success since the 0 value can be a socket
2125 * shutdown during the recv() or send() call.
2126 */
2127 ret = 1;
2128 goto end;
2129
2130 end_msg_sessiond:
2131 /*
2132 * The returned value here is not useful since either way we'll return 1 to
2133 * the caller because the session daemon socket management is done
2134 * elsewhere. Returning a negative code or 0 will shutdown the consumer.
2135 */
2136 ret = consumer_send_status_msg(sock, ret_code);
2137 if (ret < 0) {
2138 goto error_fatal;
2139 }
2140 ret = 1;
2141 goto end;
2142
2143 end_channel_error:
2144 if (channel) {
2145 pthread_mutex_unlock(&channel->lock);
2146 /*
2147 * Free channel here since no one has a reference to it. We don't
2148 * free after that because a stream can store this pointer.
2149 */
2150 destroy_channel(channel);
2151 }
2152 /* We have to send a status channel message indicating an error. */
2153 ret = consumer_send_status_channel(sock, NULL);
2154 if (ret < 0) {
2155 /* Stop everything if session daemon can not be notified. */
2156 goto error_fatal;
2157 }
2158 ret = 1;
2159 goto end;
2160
2161 error_fatal:
2162 /* This will issue a consumer stop. */
2163 ret = -1;
2164 goto end;
2165
2166 end:
2167 rcu_read_unlock();
2168 health_code_update();
2169 return ret;
2170 }
2171
2172 /*
2173 * Wrapper over the mmap() read offset from ust-ctl library. Since this can be
2174 * compiled out, we isolate it in this library.
2175 */
2176 int lttng_ustctl_get_mmap_read_offset(struct lttng_consumer_stream *stream,
2177 unsigned long *off)
2178 {
2179 assert(stream);
2180 assert(stream->ustream);
2181
2182 return ustctl_get_mmap_read_offset(stream->ustream, off);
2183 }
2184
2185 /*
2186 * Wrapper over the mmap() read offset from ust-ctl library. Since this can be
2187 * compiled out, we isolate it in this library.
2188 */
2189 void *lttng_ustctl_get_mmap_base(struct lttng_consumer_stream *stream)
2190 {
2191 assert(stream);
2192 assert(stream->ustream);
2193
2194 return ustctl_get_mmap_base(stream->ustream);
2195 }
2196
2197 void lttng_ustctl_flush_buffer(struct lttng_consumer_stream *stream,
2198 int producer_active)
2199 {
2200 assert(stream);
2201 assert(stream->ustream);
2202
2203 ustctl_flush_buffer(stream->ustream, producer_active);
2204 }
2205
2206 /*
2207 * Take a snapshot for a specific stream.
2208 *
2209 * Returns 0 on success, < 0 on error
2210 */
2211 int lttng_ustconsumer_take_snapshot(struct lttng_consumer_stream *stream)
2212 {
2213 assert(stream);
2214 assert(stream->ustream);
2215
2216 return ustctl_snapshot(stream->ustream);
2217 }
2218
2219 /*
2220 * Sample consumed and produced positions for a specific stream.
2221 *
2222 * Returns 0 on success, < 0 on error.
2223 */
2224 int lttng_ustconsumer_sample_snapshot_positions(
2225 struct lttng_consumer_stream *stream)
2226 {
2227 assert(stream);
2228 assert(stream->ustream);
2229
2230 return ustctl_snapshot_sample_positions(stream->ustream);
2231 }
2232
2233 /*
2234 * Get the produced position
2235 *
2236 * Returns 0 on success, < 0 on error
2237 */
2238 int lttng_ustconsumer_get_produced_snapshot(
2239 struct lttng_consumer_stream *stream, unsigned long *pos)
2240 {
2241 assert(stream);
2242 assert(stream->ustream);
2243 assert(pos);
2244
2245 return ustctl_snapshot_get_produced(stream->ustream, pos);
2246 }
2247
2248 /*
2249 * Get the consumed position
2250 *
2251 * Returns 0 on success, < 0 on error
2252 */
2253 int lttng_ustconsumer_get_consumed_snapshot(
2254 struct lttng_consumer_stream *stream, unsigned long *pos)
2255 {
2256 assert(stream);
2257 assert(stream->ustream);
2258 assert(pos);
2259
2260 return ustctl_snapshot_get_consumed(stream->ustream, pos);
2261 }
2262
2263 void lttng_ustconsumer_flush_buffer(struct lttng_consumer_stream *stream,
2264 int producer)
2265 {
2266 assert(stream);
2267 assert(stream->ustream);
2268
2269 ustctl_flush_buffer(stream->ustream, producer);
2270 }
2271
2272 int lttng_ustconsumer_get_current_timestamp(
2273 struct lttng_consumer_stream *stream, uint64_t *ts)
2274 {
2275 assert(stream);
2276 assert(stream->ustream);
2277 assert(ts);
2278
2279 return ustctl_get_current_timestamp(stream->ustream, ts);
2280 }
2281
2282 int lttng_ustconsumer_get_sequence_number(
2283 struct lttng_consumer_stream *stream, uint64_t *seq)
2284 {
2285 assert(stream);
2286 assert(stream->ustream);
2287 assert(seq);
2288
2289 return ustctl_get_sequence_number(stream->ustream, seq);
2290 }
2291
2292 /*
2293 * Called when the stream signals the consumer that it has hung up.
2294 */
2295 void lttng_ustconsumer_on_stream_hangup(struct lttng_consumer_stream *stream)
2296 {
2297 assert(stream);
2298 assert(stream->ustream);
2299
2300 pthread_mutex_lock(&stream->lock);
2301 if (!stream->quiescent) {
2302 ustctl_flush_buffer(stream->ustream, 0);
2303 stream->quiescent = true;
2304 }
2305 pthread_mutex_unlock(&stream->lock);
2306 stream->hangup_flush_done = 1;
2307 }
2308
2309 void lttng_ustconsumer_del_channel(struct lttng_consumer_channel *chan)
2310 {
2311 int i;
2312
2313 assert(chan);
2314 assert(chan->uchan);
2315 assert(chan->buffer_credentials.is_set);
2316
2317 if (chan->switch_timer_enabled == 1) {
2318 consumer_timer_switch_stop(chan);
2319 }
2320 for (i = 0; i < chan->nr_stream_fds; i++) {
2321 int ret;
2322
2323 ret = close(chan->stream_fds[i]);
2324 if (ret) {
2325 PERROR("close");
2326 }
2327 if (chan->shm_path[0]) {
2328 char shm_path[PATH_MAX];
2329
2330 ret = get_stream_shm_path(shm_path, chan->shm_path, i);
2331 if (ret) {
2332 ERR("Cannot get stream shm path");
2333 }
2334 ret = run_as_unlink(shm_path,
2335 chan->buffer_credentials.value.uid,
2336 chan->buffer_credentials.value.gid);
2337 if (ret) {
2338 PERROR("unlink %s", shm_path);
2339 }
2340 }
2341 }
2342 }
2343
2344 void lttng_ustconsumer_free_channel(struct lttng_consumer_channel *chan)
2345 {
2346 assert(chan);
2347 assert(chan->uchan);
2348 assert(chan->buffer_credentials.is_set);
2349
2350 consumer_metadata_cache_destroy(chan);
2351 ustctl_destroy_channel(chan->uchan);
2352 /* Try to rmdir all directories under shm_path root. */
2353 if (chan->root_shm_path[0]) {
2354 (void) run_as_rmdir_recursive(chan->root_shm_path,
2355 chan->buffer_credentials.value.uid,
2356 chan->buffer_credentials.value.gid,
2357 LTTNG_DIRECTORY_HANDLE_SKIP_NON_EMPTY_FLAG);
2358 }
2359 free(chan->stream_fds);
2360 }
2361
2362 void lttng_ustconsumer_del_stream(struct lttng_consumer_stream *stream)
2363 {
2364 assert(stream);
2365 assert(stream->ustream);
2366
2367 if (stream->chan->switch_timer_enabled == 1) {
2368 consumer_timer_switch_stop(stream->chan);
2369 }
2370 ustctl_destroy_stream(stream->ustream);
2371 }
2372
2373 int lttng_ustconsumer_get_wakeup_fd(struct lttng_consumer_stream *stream)
2374 {
2375 assert(stream);
2376 assert(stream->ustream);
2377
2378 return ustctl_stream_get_wakeup_fd(stream->ustream);
2379 }
2380
2381 int lttng_ustconsumer_close_wakeup_fd(struct lttng_consumer_stream *stream)
2382 {
2383 assert(stream);
2384 assert(stream->ustream);
2385
2386 return ustctl_stream_close_wakeup_fd(stream->ustream);
2387 }
2388
2389 /*
2390 * Populate index values of a UST stream. Values are set in big endian order.
2391 *
2392 * Return 0 on success or else a negative value.
2393 */
2394 static int get_index_values(struct ctf_packet_index *index,
2395 struct ustctl_consumer_stream *ustream)
2396 {
2397 int ret;
2398
2399 ret = ustctl_get_timestamp_begin(ustream, &index->timestamp_begin);
2400 if (ret < 0) {
2401 PERROR("ustctl_get_timestamp_begin");
2402 goto error;
2403 }
2404 index->timestamp_begin = htobe64(index->timestamp_begin);
2405
2406 ret = ustctl_get_timestamp_end(ustream, &index->timestamp_end);
2407 if (ret < 0) {
2408 PERROR("ustctl_get_timestamp_end");
2409 goto error;
2410 }
2411 index->timestamp_end = htobe64(index->timestamp_end);
2412
2413 ret = ustctl_get_events_discarded(ustream, &index->events_discarded);
2414 if (ret < 0) {
2415 PERROR("ustctl_get_events_discarded");
2416 goto error;
2417 }
2418 index->events_discarded = htobe64(index->events_discarded);
2419
2420 ret = ustctl_get_content_size(ustream, &index->content_size);
2421 if (ret < 0) {
2422 PERROR("ustctl_get_content_size");
2423 goto error;
2424 }
2425 index->content_size = htobe64(index->content_size);
2426
2427 ret = ustctl_get_packet_size(ustream, &index->packet_size);
2428 if (ret < 0) {
2429 PERROR("ustctl_get_packet_size");
2430 goto error;
2431 }
2432 index->packet_size = htobe64(index->packet_size);
2433
2434 ret = ustctl_get_stream_id(ustream, &index->stream_id);
2435 if (ret < 0) {
2436 PERROR("ustctl_get_stream_id");
2437 goto error;
2438 }
2439 index->stream_id = htobe64(index->stream_id);
2440
2441 ret = ustctl_get_instance_id(ustream, &index->stream_instance_id);
2442 if (ret < 0) {
2443 PERROR("ustctl_get_instance_id");
2444 goto error;
2445 }
2446 index->stream_instance_id = htobe64(index->stream_instance_id);
2447
2448 ret = ustctl_get_sequence_number(ustream, &index->packet_seq_num);
2449 if (ret < 0) {
2450 PERROR("ustctl_get_sequence_number");
2451 goto error;
2452 }
2453 index->packet_seq_num = htobe64(index->packet_seq_num);
2454
2455 error:
2456 return ret;
2457 }
2458
2459 static
2460 void metadata_stream_reset_cache(struct lttng_consumer_stream *stream,
2461 struct consumer_metadata_cache *cache)
2462 {
2463 DBG("Metadata stream update to version %" PRIu64,
2464 cache->version);
2465 stream->ust_metadata_pushed = 0;
2466 stream->metadata_version = cache->version;
2467 stream->reset_metadata_flag = 1;
2468 }
2469
2470 /*
2471 * Check if the version of the metadata stream and metadata cache match.
2472 * If the cache got updated, reset the metadata stream.
2473 * The stream lock and metadata cache lock MUST be held.
2474 * Return 0 on success, a negative value on error.
2475 */
2476 static
2477 int metadata_stream_check_version(struct lttng_consumer_stream *stream)
2478 {
2479 int ret = 0;
2480 struct consumer_metadata_cache *cache = stream->chan->metadata_cache;
2481
2482 if (cache->version == stream->metadata_version) {
2483 goto end;
2484 }
2485 metadata_stream_reset_cache(stream, cache);
2486
2487 end:
2488 return ret;
2489 }
2490
2491 /*
2492 * Write up to one packet from the metadata cache to the channel.
2493 *
2494 * Returns the number of bytes pushed in the cache, or a negative value
2495 * on error.
2496 */
2497 static
2498 int commit_one_metadata_packet(struct lttng_consumer_stream *stream)
2499 {
2500 ssize_t write_len;
2501 int ret;
2502
2503 pthread_mutex_lock(&stream->chan->metadata_cache->lock);
2504 ret = metadata_stream_check_version(stream);
2505 if (ret < 0) {
2506 goto end;
2507 }
2508 if (stream->chan->metadata_cache->max_offset
2509 == stream->ust_metadata_pushed) {
2510 ret = 0;
2511 goto end;
2512 }
2513
2514 write_len = ustctl_write_one_packet_to_channel(stream->chan->uchan,
2515 &stream->chan->metadata_cache->data[stream->ust_metadata_pushed],
2516 stream->chan->metadata_cache->max_offset
2517 - stream->ust_metadata_pushed);
2518 assert(write_len != 0);
2519 if (write_len < 0) {
2520 ERR("Writing one metadata packet");
2521 ret = -1;
2522 goto end;
2523 }
2524 stream->ust_metadata_pushed += write_len;
2525
2526 assert(stream->chan->metadata_cache->max_offset >=
2527 stream->ust_metadata_pushed);
2528 ret = write_len;
2529
2530 /*
2531 * Switch packet (but don't open the next one) on every commit of
2532 * a metadata packet. Since the subbuffer is fully filled (with padding,
2533 * if needed), the stream is "quiescent" after this commit.
2534 */
2535 ustctl_flush_buffer(stream->ustream, 1);
2536 stream->quiescent = true;
2537 end:
2538 pthread_mutex_unlock(&stream->chan->metadata_cache->lock);
2539 return ret;
2540 }
2541
2542
2543 /*
2544 * Sync metadata meaning request them to the session daemon and snapshot to the
2545 * metadata thread can consumer them.
2546 *
2547 * Metadata stream lock is held here, but we need to release it when
2548 * interacting with sessiond, else we cause a deadlock with live
2549 * awaiting on metadata to be pushed out.
2550 *
2551 * Return 0 if new metadatda is available, EAGAIN if the metadata stream
2552 * is empty or a negative value on error.
2553 */
2554 int lttng_ustconsumer_sync_metadata(struct lttng_consumer_local_data *ctx,
2555 struct lttng_consumer_stream *metadata)
2556 {
2557 int ret;
2558 int retry = 0;
2559
2560 assert(ctx);
2561 assert(metadata);
2562
2563 pthread_mutex_unlock(&metadata->lock);
2564 /*
2565 * Request metadata from the sessiond, but don't wait for the flush
2566 * because we locked the metadata thread.
2567 */
2568 ret = lttng_ustconsumer_request_metadata(ctx, metadata->chan, 0, 0);
2569 pthread_mutex_lock(&metadata->lock);
2570 if (ret < 0) {
2571 goto end;
2572 }
2573
2574 ret = commit_one_metadata_packet(metadata);
2575 if (ret <= 0) {
2576 goto end;
2577 } else if (ret > 0) {
2578 retry = 1;
2579 }
2580
2581 ret = ustctl_snapshot(metadata->ustream);
2582 if (ret < 0) {
2583 if (errno != EAGAIN) {
2584 ERR("Sync metadata, taking UST snapshot");
2585 goto end;
2586 }
2587 DBG("No new metadata when syncing them.");
2588 /* No new metadata, exit. */
2589 ret = ENODATA;
2590 goto end;
2591 }
2592
2593 /*
2594 * After this flush, we still need to extract metadata.
2595 */
2596 if (retry) {
2597 ret = EAGAIN;
2598 }
2599
2600 end:
2601 return ret;
2602 }
2603
2604 /*
2605 * Return 0 on success else a negative value.
2606 */
2607 static int notify_if_more_data(struct lttng_consumer_stream *stream,
2608 struct lttng_consumer_local_data *ctx)
2609 {
2610 int ret;
2611 struct ustctl_consumer_stream *ustream;
2612
2613 assert(stream);
2614 assert(ctx);
2615
2616 ustream = stream->ustream;
2617
2618 /*
2619 * First, we are going to check if there is a new subbuffer available
2620 * before reading the stream wait_fd.
2621 */
2622 /* Get the next subbuffer */
2623 ret = ustctl_get_next_subbuf(ustream);
2624 if (ret) {
2625 /* No more data found, flag the stream. */
2626 stream->has_data = 0;
2627 ret = 0;
2628 goto end;
2629 }
2630
2631 ret = ustctl_put_subbuf(ustream);
2632 assert(!ret);
2633
2634 /* This stream still has data. Flag it and wake up the data thread. */
2635 stream->has_data = 1;
2636
2637 if (stream->monitor && !stream->hangup_flush_done && !ctx->has_wakeup) {
2638 ssize_t writelen;
2639
2640 writelen = lttng_pipe_write(ctx->consumer_wakeup_pipe, "!", 1);
2641 if (writelen < 0 && errno != EAGAIN && errno != EWOULDBLOCK) {
2642 ret = writelen;
2643 goto end;
2644 }
2645
2646 /* The wake up pipe has been notified. */
2647 ctx->has_wakeup = 1;
2648 }
2649 ret = 0;
2650
2651 end:
2652 return ret;
2653 }
2654
2655 static
2656 int update_stream_stats(struct lttng_consumer_stream *stream)
2657 {
2658 int ret;
2659 uint64_t seq, discarded;
2660
2661 ret = ustctl_get_sequence_number(stream->ustream, &seq);
2662 if (ret < 0) {
2663 PERROR("ustctl_get_sequence_number");
2664 goto end;
2665 }
2666 /*
2667 * Start the sequence when we extract the first packet in case we don't
2668 * start at 0 (for example if a consumer is not connected to the
2669 * session immediately after the beginning).
2670 */
2671 if (stream->last_sequence_number == -1ULL) {
2672 stream->last_sequence_number = seq;
2673 } else if (seq > stream->last_sequence_number) {
2674 stream->chan->lost_packets += seq -
2675 stream->last_sequence_number - 1;
2676 } else {
2677 /* seq <= last_sequence_number */
2678 ERR("Sequence number inconsistent : prev = %" PRIu64
2679 ", current = %" PRIu64,
2680 stream->last_sequence_number, seq);
2681 ret = -1;
2682 goto end;
2683 }
2684 stream->last_sequence_number = seq;
2685
2686 ret = ustctl_get_events_discarded(stream->ustream, &discarded);
2687 if (ret < 0) {
2688 PERROR("kernctl_get_events_discarded");
2689 goto end;
2690 }
2691 if (discarded < stream->last_discarded_events) {
2692 /*
2693 * Overflow has occurred. We assume only one wrap-around
2694 * has occurred.
2695 */
2696 stream->chan->discarded_events +=
2697 (1ULL << (CAA_BITS_PER_LONG - 1)) -
2698 stream->last_discarded_events + discarded;
2699 } else {
2700 stream->chan->discarded_events += discarded -
2701 stream->last_discarded_events;
2702 }
2703 stream->last_discarded_events = discarded;
2704 ret = 0;
2705
2706 end:
2707 return ret;
2708 }
2709
2710 /*
2711 * Read subbuffer from the given stream.
2712 *
2713 * Stream and channel locks MUST be acquired by the caller.
2714 *
2715 * Return 0 on success else a negative value.
2716 */
2717 int lttng_ustconsumer_read_subbuffer(struct lttng_consumer_stream *stream,
2718 struct lttng_consumer_local_data *ctx)
2719 {
2720 unsigned long len, subbuf_size, padding;
2721 int err, write_index = 1, rotation_ret;
2722 long ret = 0;
2723 struct ustctl_consumer_stream *ustream;
2724 struct ctf_packet_index index;
2725
2726 assert(stream);
2727 assert(stream->ustream);
2728 assert(ctx);
2729
2730 DBG("In UST read_subbuffer (wait_fd: %d, name: %s)", stream->wait_fd,
2731 stream->name);
2732
2733 /* Ease our life for what's next. */
2734 ustream = stream->ustream;
2735
2736 /*
2737 * We can consume the 1 byte written into the wait_fd by UST. Don't trigger
2738 * error if we cannot read this one byte (read returns 0), or if the error
2739 * is EAGAIN or EWOULDBLOCK.
2740 *
2741 * This is only done when the stream is monitored by a thread, before the
2742 * flush is done after a hangup and if the stream is not flagged with data
2743 * since there might be nothing to consume in the wait fd but still have
2744 * data available flagged by the consumer wake up pipe.
2745 */
2746 if (stream->monitor && !stream->hangup_flush_done && !stream->has_data) {
2747 char dummy;
2748 ssize_t readlen;
2749
2750 readlen = lttng_read(stream->wait_fd, &dummy, 1);
2751 if (readlen < 0 && errno != EAGAIN && errno != EWOULDBLOCK) {
2752 ret = readlen;
2753 goto error;
2754 }
2755 }
2756
2757 /*
2758 * If the stream was flagged to be ready for rotation before we extract the
2759 * next packet, rotate it now.
2760 */
2761 if (stream->rotate_ready) {
2762 DBG("Rotate stream before extracting data");
2763 rotation_ret = lttng_consumer_rotate_stream(ctx, stream);
2764 if (rotation_ret < 0) {
2765 ERR("Stream rotation error");
2766 ret = -1;
2767 goto error;
2768 }
2769 }
2770
2771 retry:
2772 /* Get the next subbuffer */
2773 err = ustctl_get_next_subbuf(ustream);
2774 if (err != 0) {
2775 /*
2776 * Populate metadata info if the existing info has
2777 * already been read.
2778 */
2779 if (stream->metadata_flag) {
2780 ret = commit_one_metadata_packet(stream);
2781 if (ret <= 0) {
2782 goto error;
2783 }
2784 goto retry;
2785 }
2786
2787 ret = err; /* ustctl_get_next_subbuf returns negative, caller expect positive. */
2788 /*
2789 * This is a debug message even for single-threaded consumer,
2790 * because poll() have more relaxed criterions than get subbuf,
2791 * so get_subbuf may fail for short race windows where poll()
2792 * would issue wakeups.
2793 */
2794 DBG("Reserving sub buffer failed (everything is normal, "
2795 "it is due to concurrency) [ret: %d]", err);
2796 goto error;
2797 }
2798 assert(stream->chan->output == CONSUMER_CHANNEL_MMAP);
2799
2800 if (!stream->metadata_flag) {
2801 index.offset = htobe64(stream->out_fd_offset);
2802 ret = get_index_values(&index, ustream);
2803 if (ret < 0) {
2804 err = ustctl_put_subbuf(ustream);
2805 assert(err == 0);
2806 goto error;
2807 }
2808
2809 /* Update the stream's sequence and discarded events count. */
2810 ret = update_stream_stats(stream);
2811 if (ret < 0) {
2812 PERROR("kernctl_get_events_discarded");
2813 err = ustctl_put_subbuf(ustream);
2814 assert(err == 0);
2815 goto error;
2816 }
2817 } else {
2818 write_index = 0;
2819 }
2820
2821 /* Get the full padded subbuffer size */
2822 err = ustctl_get_padded_subbuf_size(ustream, &len);
2823 assert(err == 0);
2824
2825 /* Get subbuffer data size (without padding) */
2826 err = ustctl_get_subbuf_size(ustream, &subbuf_size);
2827 assert(err == 0);
2828
2829 /* Make sure we don't get a subbuffer size bigger than the padded */
2830 assert(len >= subbuf_size);
2831
2832 padding = len - subbuf_size;
2833
2834 /* write the subbuffer to the tracefile */
2835 ret = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, subbuf_size, padding, &index);
2836 /*
2837 * The mmap operation should write subbuf_size amount of data when network
2838 * streaming or the full padding (len) size when we are _not_ streaming.
2839 */
2840 if ((ret != subbuf_size && stream->net_seq_idx != (uint64_t) -1ULL) ||
2841 (ret != len && stream->net_seq_idx == (uint64_t) -1ULL)) {
2842 /*
2843 * Display the error but continue processing to try to release the
2844 * subbuffer. This is a DBG statement since any unexpected kill or
2845 * signal, the application gets unregistered, relayd gets closed or
2846 * anything that affects the buffer lifetime will trigger this error.
2847 * So, for the sake of the user, don't print this error since it can
2848 * happen and it is OK with the code flow.
2849 */
2850 DBG("Error writing to tracefile "
2851 "(ret: %ld != len: %lu != subbuf_size: %lu)",
2852 ret, len, subbuf_size);
2853 write_index = 0;
2854 }
2855 err = ustctl_put_next_subbuf(ustream);
2856 assert(err == 0);
2857
2858 /*
2859 * This will consumer the byte on the wait_fd if and only if there is not
2860 * next subbuffer to be acquired.
2861 */
2862 if (!stream->metadata_flag) {
2863 ret = notify_if_more_data(stream, ctx);
2864 if (ret < 0) {
2865 goto error;
2866 }
2867 }
2868
2869 /* Write index if needed. */
2870 if (!write_index) {
2871 goto rotate;
2872 }
2873
2874 if (stream->chan->live_timer_interval && !stream->metadata_flag) {
2875 /*
2876 * In live, block until all the metadata is sent.
2877 */
2878 pthread_mutex_lock(&stream->metadata_timer_lock);
2879 assert(!stream->missed_metadata_flush);
2880 stream->waiting_on_metadata = true;
2881 pthread_mutex_unlock(&stream->metadata_timer_lock);
2882
2883 err = consumer_stream_sync_metadata(ctx, stream->session_id);
2884
2885 pthread_mutex_lock(&stream->metadata_timer_lock);
2886 stream->waiting_on_metadata = false;
2887 if (stream->missed_metadata_flush) {
2888 stream->missed_metadata_flush = false;
2889 pthread_mutex_unlock(&stream->metadata_timer_lock);
2890 (void) consumer_flush_ust_index(stream);
2891 } else {
2892 pthread_mutex_unlock(&stream->metadata_timer_lock);
2893 }
2894
2895 if (err < 0) {
2896 goto error;
2897 }
2898 }
2899
2900 assert(!stream->metadata_flag);
2901 err = consumer_stream_write_index(stream, &index);
2902 if (err < 0) {
2903 goto error;
2904 }
2905
2906 rotate:
2907 /*
2908 * After extracting the packet, we check if the stream is now ready to be
2909 * rotated and perform the action immediately.
2910 */
2911 rotation_ret = lttng_consumer_stream_is_rotate_ready(stream);
2912 if (rotation_ret == 1) {
2913 rotation_ret = lttng_consumer_rotate_stream(ctx, stream);
2914 if (rotation_ret < 0) {
2915 ERR("Stream rotation error");
2916 ret = -1;
2917 goto error;
2918 }
2919 } else if (rotation_ret < 0) {
2920 ERR("Checking if stream is ready to rotate");
2921 ret = -1;
2922 goto error;
2923 }
2924 error:
2925 return ret;
2926 }
2927
2928 /*
2929 * Called when a stream is created.
2930 *
2931 * Return 0 on success or else a negative value.
2932 */
2933 int lttng_ustconsumer_on_recv_stream(struct lttng_consumer_stream *stream)
2934 {
2935 int ret;
2936
2937 assert(stream);
2938
2939 /*
2940 * Don't create anything if this is set for streaming or if there is
2941 * no current trace chunk on the parent channel.
2942 */
2943 if (stream->net_seq_idx == (uint64_t) -1ULL && stream->chan->monitor &&
2944 stream->chan->trace_chunk) {
2945 ret = consumer_stream_create_output_files(stream, true);
2946 if (ret) {
2947 goto error;
2948 }
2949 }
2950 ret = 0;
2951
2952 error:
2953 return ret;
2954 }
2955
2956 /*
2957 * Check if data is still being extracted from the buffers for a specific
2958 * stream. Consumer data lock MUST be acquired before calling this function
2959 * and the stream lock.
2960 *
2961 * Return 1 if the traced data are still getting read else 0 meaning that the
2962 * data is available for trace viewer reading.
2963 */
2964 int lttng_ustconsumer_data_pending(struct lttng_consumer_stream *stream)
2965 {
2966 int ret;
2967
2968 assert(stream);
2969 assert(stream->ustream);
2970
2971 DBG("UST consumer checking data pending");
2972
2973 if (stream->endpoint_status != CONSUMER_ENDPOINT_ACTIVE) {
2974 ret = 0;
2975 goto end;
2976 }
2977
2978 if (stream->chan->type == CONSUMER_CHANNEL_TYPE_METADATA) {
2979 uint64_t contiguous, pushed;
2980
2981 /* Ease our life a bit. */
2982 contiguous = stream->chan->metadata_cache->max_offset;
2983 pushed = stream->ust_metadata_pushed;
2984
2985 /*
2986 * We can simply check whether all contiguously available data
2987 * has been pushed to the ring buffer, since the push operation
2988 * is performed within get_next_subbuf(), and because both
2989 * get_next_subbuf() and put_next_subbuf() are issued atomically
2990 * thanks to the stream lock within
2991 * lttng_ustconsumer_read_subbuffer(). This basically means that
2992 * whetnever ust_metadata_pushed is incremented, the associated
2993 * metadata has been consumed from the metadata stream.
2994 */
2995 DBG("UST consumer metadata pending check: contiguous %" PRIu64 " vs pushed %" PRIu64,
2996 contiguous, pushed);
2997 assert(((int64_t) (contiguous - pushed)) >= 0);
2998 if ((contiguous != pushed) ||
2999 (((int64_t) contiguous - pushed) > 0 || contiguous == 0)) {
3000 ret = 1; /* Data is pending */
3001 goto end;
3002 }
3003 } else {
3004 ret = ustctl_get_next_subbuf(stream->ustream);
3005 if (ret == 0) {
3006 /*
3007 * There is still data so let's put back this
3008 * subbuffer.
3009 */
3010 ret = ustctl_put_subbuf(stream->ustream);
3011 assert(ret == 0);
3012 ret = 1; /* Data is pending */
3013 goto end;
3014 }
3015 }
3016
3017 /* Data is NOT pending so ready to be read. */
3018 ret = 0;
3019
3020 end:
3021 return ret;
3022 }
3023
3024 /*
3025 * Stop a given metadata channel timer if enabled and close the wait fd which
3026 * is the poll pipe of the metadata stream.
3027 *
3028 * This MUST be called with the metadata channel acquired.
3029 */
3030 void lttng_ustconsumer_close_metadata(struct lttng_consumer_channel *metadata)
3031 {
3032 int ret;
3033
3034 assert(metadata);
3035 assert(metadata->type == CONSUMER_CHANNEL_TYPE_METADATA);
3036
3037 DBG("Closing metadata channel key %" PRIu64, metadata->key);
3038
3039 if (metadata->switch_timer_enabled == 1) {
3040 consumer_timer_switch_stop(metadata);
3041 }
3042
3043 if (!metadata->metadata_stream) {
3044 goto end;
3045 }
3046
3047 /*
3048 * Closing write side so the thread monitoring the stream wakes up if any
3049 * and clean the metadata stream.
3050 */
3051 if (metadata->metadata_stream->ust_metadata_poll_pipe[1] >= 0) {
3052 ret = close(metadata->metadata_stream->ust_metadata_poll_pipe[1]);
3053 if (ret < 0) {
3054 PERROR("closing metadata pipe write side");
3055 }
3056 metadata->metadata_stream->ust_metadata_poll_pipe[1] = -1;
3057 }
3058
3059 end:
3060 return;
3061 }
3062
3063 /*
3064 * Close every metadata stream wait fd of the metadata hash table. This
3065 * function MUST be used very carefully so not to run into a race between the
3066 * metadata thread handling streams and this function closing their wait fd.
3067 *
3068 * For UST, this is used when the session daemon hangs up. Its the metadata
3069 * producer so calling this is safe because we are assured that no state change
3070 * can occur in the metadata thread for the streams in the hash table.
3071 */
3072 void lttng_ustconsumer_close_all_metadata(struct lttng_ht *metadata_ht)
3073 {
3074 struct lttng_ht_iter iter;
3075 struct lttng_consumer_stream *stream;
3076
3077 assert(metadata_ht);
3078 assert(metadata_ht->ht);
3079
3080 DBG("UST consumer closing all metadata streams");
3081
3082 rcu_read_lock();
3083 cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream,
3084 node.node) {
3085
3086 health_code_update();
3087
3088 pthread_mutex_lock(&stream->chan->lock);
3089 lttng_ustconsumer_close_metadata(stream->chan);
3090 pthread_mutex_unlock(&stream->chan->lock);
3091
3092 }
3093 rcu_read_unlock();
3094 }
3095
3096 void lttng_ustconsumer_close_stream_wakeup(struct lttng_consumer_stream *stream)
3097 {
3098 int ret;
3099
3100 ret = ustctl_stream_close_wakeup_fd(stream->ustream);
3101 if (ret < 0) {
3102 ERR("Unable to close wakeup fd");
3103 }
3104 }
3105
3106 /*
3107 * Please refer to consumer-timer.c before adding any lock within this
3108 * function or any of its callees. Timers have a very strict locking
3109 * semantic with respect to teardown. Failure to respect this semantic
3110 * introduces deadlocks.
3111 *
3112 * DON'T hold the metadata lock when calling this function, else this
3113 * can cause deadlock involving consumer awaiting for metadata to be
3114 * pushed out due to concurrent interaction with the session daemon.
3115 */
3116 int lttng_ustconsumer_request_metadata(struct lttng_consumer_local_data *ctx,
3117 struct lttng_consumer_channel *channel, int timer, int wait)
3118 {
3119 struct lttcomm_metadata_request_msg request;
3120 struct lttcomm_consumer_msg msg;
3121 enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS;
3122 uint64_t len, key, offset, version;
3123 int ret;
3124
3125 assert(channel);
3126 assert(channel->metadata_cache);
3127
3128 memset(&request, 0, sizeof(request));
3129
3130 /* send the metadata request to sessiond */
3131 switch (consumer_data.type) {
3132 case LTTNG_CONSUMER64_UST:
3133 request.bits_per_long = 64;
3134 break;
3135 case LTTNG_CONSUMER32_UST:
3136 request.bits_per_long = 32;
3137 break;
3138 default:
3139 request.bits_per_long = 0;
3140 break;
3141 }
3142
3143 request.session_id = channel->session_id;
3144 request.session_id_per_pid = channel->session_id_per_pid;
3145 /*
3146 * Request the application UID here so the metadata of that application can
3147 * be sent back. The channel UID corresponds to the user UID of the session
3148 * used for the rights on the stream file(s).
3149 */
3150 request.uid = channel->ust_app_uid;
3151 request.key = channel->key;
3152
3153 DBG("Sending metadata request to sessiond, session id %" PRIu64
3154 ", per-pid %" PRIu64 ", app UID %u and channel key %" PRIu64,
3155 request.session_id, request.session_id_per_pid, request.uid,
3156 request.key);
3157
3158 pthread_mutex_lock(&ctx->metadata_socket_lock);
3159
3160 health_code_update();
3161
3162 ret = lttcomm_send_unix_sock(ctx->consumer_metadata_socket, &request,
3163 sizeof(request));
3164 if (ret < 0) {
3165 ERR("Asking metadata to sessiond");
3166 goto end;
3167 }
3168
3169 health_code_update();
3170
3171 /* Receive the metadata from sessiond */
3172 ret = lttcomm_recv_unix_sock(ctx->consumer_metadata_socket, &msg,
3173 sizeof(msg));
3174 if (ret != sizeof(msg)) {
3175 DBG("Consumer received unexpected message size %d (expects %zu)",
3176 ret, sizeof(msg));
3177 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD);
3178 /*
3179 * The ret value might 0 meaning an orderly shutdown but this is ok
3180 * since the caller handles this.
3181 */
3182 goto end;
3183 }
3184
3185 health_code_update();
3186
3187 if (msg.cmd_type == LTTNG_ERR_UND) {
3188 /* No registry found */
3189 (void) consumer_send_status_msg(ctx->consumer_metadata_socket,
3190 ret_code);
3191 ret = 0;
3192 goto end;
3193 } else if (msg.cmd_type != LTTNG_CONSUMER_PUSH_METADATA) {
3194 ERR("Unexpected cmd_type received %d", msg.cmd_type);
3195 ret = -1;
3196 goto end;
3197 }
3198
3199 len = msg.u.push_metadata.len;
3200 key = msg.u.push_metadata.key;
3201 offset = msg.u.push_metadata.target_offset;
3202 version = msg.u.push_metadata.version;
3203
3204 assert(key == channel->key);
3205 if (len == 0) {
3206 DBG("No new metadata to receive for key %" PRIu64, key);
3207 }
3208
3209 health_code_update();
3210
3211 /* Tell session daemon we are ready to receive the metadata. */
3212 ret = consumer_send_status_msg(ctx->consumer_metadata_socket,
3213 LTTCOMM_CONSUMERD_SUCCESS);
3214 if (ret < 0 || len == 0) {
3215 /*
3216 * Somehow, the session daemon is not responding anymore or there is
3217 * nothing to receive.
3218 */
3219 goto end;
3220 }
3221
3222 health_code_update();
3223
3224 ret = lttng_ustconsumer_recv_metadata(ctx->consumer_metadata_socket,
3225 key, offset, len, version, channel, timer, wait);
3226 if (ret >= 0) {
3227 /*
3228 * Only send the status msg if the sessiond is alive meaning a positive
3229 * ret code.
3230 */
3231 (void) consumer_send_status_msg(ctx->consumer_metadata_socket, ret);
3232 }
3233 ret = 0;
3234
3235 end:
3236 health_code_update();
3237
3238 pthread_mutex_unlock(&ctx->metadata_socket_lock);
3239 return ret;
3240 }
3241
3242 /*
3243 * Return the ustctl call for the get stream id.
3244 */
3245 int lttng_ustconsumer_get_stream_id(struct lttng_consumer_stream *stream,
3246 uint64_t *stream_id)
3247 {
3248 assert(stream);
3249 assert(stream_id);
3250
3251 return ustctl_get_stream_id(stream->ustream, stream_id);
3252 }
This page took 0.142282 seconds and 3 git commands to generate.