ust-consumer: fix: metadata stream lock taken before destroy
[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
779 /*
780 * Protect against concurrent teardown of a stream.
781 */
782 if (cds_lfht_is_node_deleted(&stream->node.node)) {
783 goto next;
784 }
785
786 if (!stream->quiescent) {
787 ustctl_flush_buffer(stream->ustream, 0);
788 stream->quiescent = true;
789 }
790 next:
791 pthread_mutex_unlock(&stream->lock);
792 }
793 error:
794 rcu_read_unlock();
795 return ret;
796 }
797
798 /*
799 * Clear quiescent state from channel's streams using the given key to
800 * retrieve the channel.
801 *
802 * Return 0 on success else an LTTng error code.
803 */
804 static int clear_quiescent_channel(uint64_t chan_key)
805 {
806 int ret = 0;
807 struct lttng_consumer_channel *channel;
808 struct lttng_consumer_stream *stream;
809 struct lttng_ht *ht;
810 struct lttng_ht_iter iter;
811
812 DBG("UST consumer clear quiescent channel key %" PRIu64, chan_key);
813
814 rcu_read_lock();
815 channel = consumer_find_channel(chan_key);
816 if (!channel) {
817 ERR("UST consumer clear quiescent channel %" PRIu64 " not found", chan_key);
818 ret = LTTNG_ERR_UST_CHAN_NOT_FOUND;
819 goto error;
820 }
821
822 ht = consumer_data.stream_per_chan_id_ht;
823
824 /* For each stream of the channel id, clear quiescent state. */
825 cds_lfht_for_each_entry_duplicate(ht->ht,
826 ht->hash_fct(&channel->key, lttng_ht_seed), ht->match_fct,
827 &channel->key, &iter.iter, stream, node_channel_id.node) {
828
829 health_code_update();
830
831 pthread_mutex_lock(&stream->lock);
832 stream->quiescent = false;
833 pthread_mutex_unlock(&stream->lock);
834 }
835 error:
836 rcu_read_unlock();
837 return ret;
838 }
839
840 /*
841 * Close metadata stream wakeup_fd using the given key to retrieve the channel.
842 *
843 * Return 0 on success else an LTTng error code.
844 */
845 static int close_metadata(uint64_t chan_key)
846 {
847 int ret = 0;
848 struct lttng_consumer_channel *channel;
849 unsigned int channel_monitor;
850
851 DBG("UST consumer close metadata key %" PRIu64, chan_key);
852
853 channel = consumer_find_channel(chan_key);
854 if (!channel) {
855 /*
856 * This is possible if the metadata thread has issue a delete because
857 * the endpoint point of the stream hung up. There is no way the
858 * session daemon can know about it thus use a DBG instead of an actual
859 * error.
860 */
861 DBG("UST consumer close metadata %" PRIu64 " not found", chan_key);
862 ret = LTTNG_ERR_UST_CHAN_NOT_FOUND;
863 goto error;
864 }
865
866 pthread_mutex_lock(&consumer_data.lock);
867 pthread_mutex_lock(&channel->lock);
868 channel_monitor = channel->monitor;
869 if (cds_lfht_is_node_deleted(&channel->node.node)) {
870 goto error_unlock;
871 }
872
873 lttng_ustconsumer_close_metadata(channel);
874 pthread_mutex_unlock(&channel->lock);
875 pthread_mutex_unlock(&consumer_data.lock);
876
877 /*
878 * The ownership of a metadata channel depends on the type of
879 * session to which it belongs. In effect, the monitor flag is checked
880 * to determine if this metadata channel is in "snapshot" mode or not.
881 *
882 * In the non-snapshot case, the metadata channel is created along with
883 * a single stream which will remain present until the metadata channel
884 * is destroyed (on the destruction of its session). In this case, the
885 * metadata stream in "monitored" by the metadata poll thread and holds
886 * the ownership of its channel.
887 *
888 * Closing the metadata will cause the metadata stream's "metadata poll
889 * pipe" to be closed. Closing this pipe will wake-up the metadata poll
890 * thread which will teardown the metadata stream which, in return,
891 * deletes the metadata channel.
892 *
893 * In the snapshot case, the metadata stream is created and destroyed
894 * on every snapshot record. Since the channel doesn't have an owner
895 * other than the session daemon, it is safe to destroy it immediately
896 * on reception of the CLOSE_METADATA command.
897 */
898 if (!channel_monitor) {
899 /*
900 * The channel and consumer_data locks must be
901 * released before this call since consumer_del_channel
902 * re-acquires the channel and consumer_data locks to teardown
903 * the channel and queue its reclamation by the "call_rcu"
904 * worker thread.
905 */
906 consumer_del_channel(channel);
907 }
908
909 return ret;
910 error_unlock:
911 pthread_mutex_unlock(&channel->lock);
912 pthread_mutex_unlock(&consumer_data.lock);
913 error:
914 return ret;
915 }
916
917 /*
918 * RCU read side lock MUST be acquired before calling this function.
919 *
920 * Return 0 on success else an LTTng error code.
921 */
922 static int setup_metadata(struct lttng_consumer_local_data *ctx, uint64_t key)
923 {
924 int ret;
925 struct lttng_consumer_channel *metadata;
926
927 DBG("UST consumer setup metadata key %" PRIu64, key);
928
929 metadata = consumer_find_channel(key);
930 if (!metadata) {
931 ERR("UST consumer push metadata %" PRIu64 " not found", key);
932 ret = LTTNG_ERR_UST_CHAN_NOT_FOUND;
933 goto end;
934 }
935
936 /*
937 * In no monitor mode, the metadata channel has no stream(s) so skip the
938 * ownership transfer to the metadata thread.
939 */
940 if (!metadata->monitor) {
941 DBG("Metadata channel in no monitor");
942 ret = 0;
943 goto end;
944 }
945
946 /*
947 * Send metadata stream to relayd if one available. Availability is
948 * known if the stream is still in the list of the channel.
949 */
950 if (cds_list_empty(&metadata->streams.head)) {
951 ERR("Metadata channel key %" PRIu64 ", no stream available.", key);
952 ret = LTTCOMM_CONSUMERD_ERROR_METADATA;
953 goto error_no_stream;
954 }
955
956 /* Send metadata stream to relayd if needed. */
957 if (metadata->metadata_stream->net_seq_idx != (uint64_t) -1ULL) {
958 ret = consumer_send_relayd_stream(metadata->metadata_stream,
959 metadata->pathname);
960 if (ret < 0) {
961 ret = LTTCOMM_CONSUMERD_ERROR_METADATA;
962 goto error;
963 }
964 ret = consumer_send_relayd_streams_sent(
965 metadata->metadata_stream->net_seq_idx);
966 if (ret < 0) {
967 ret = LTTCOMM_CONSUMERD_RELAYD_FAIL;
968 goto error;
969 }
970 }
971
972 /*
973 * Ownership of metadata stream is passed along. Freeing is handled by
974 * the callee.
975 */
976 ret = send_streams_to_thread(metadata, ctx);
977 if (ret < 0) {
978 /*
979 * If we are unable to send the stream to the thread, there is
980 * a big problem so just stop everything.
981 */
982 ret = LTTCOMM_CONSUMERD_FATAL;
983 goto send_streams_error;
984 }
985 /* List MUST be empty after or else it could be reused. */
986 assert(cds_list_empty(&metadata->streams.head));
987
988 ret = 0;
989 goto end;
990
991 error:
992 /*
993 * Delete metadata channel on error. At this point, the metadata stream can
994 * NOT be monitored by the metadata thread thus having the guarantee that
995 * the stream is still in the local stream list of the channel. This call
996 * will make sure to clean that list.
997 */
998 consumer_stream_destroy(metadata->metadata_stream, NULL);
999 cds_list_del(&metadata->metadata_stream->send_node);
1000 metadata->metadata_stream = NULL;
1001 send_streams_error:
1002 error_no_stream:
1003 end:
1004 return ret;
1005 }
1006
1007 /*
1008 * Snapshot the whole metadata.
1009 * RCU read-side lock must be held by the caller.
1010 *
1011 * Returns 0 on success, < 0 on error
1012 */
1013 static int snapshot_metadata(struct lttng_consumer_channel *metadata_channel,
1014 uint64_t key, char *path, uint64_t relayd_id,
1015 struct lttng_consumer_local_data *ctx)
1016 {
1017 int ret = 0;
1018 struct lttng_consumer_stream *metadata_stream;
1019
1020 assert(path);
1021 assert(ctx);
1022
1023 DBG("UST consumer snapshot metadata with key %" PRIu64 " at path %s",
1024 key, path);
1025
1026 rcu_read_lock();
1027
1028 assert(!metadata_channel->monitor);
1029
1030 health_code_update();
1031
1032 /*
1033 * Ask the sessiond if we have new metadata waiting and update the
1034 * consumer metadata cache.
1035 */
1036 ret = lttng_ustconsumer_request_metadata(ctx, metadata_channel, 0, 1);
1037 if (ret < 0) {
1038 goto error;
1039 }
1040
1041 health_code_update();
1042
1043 /*
1044 * The metadata stream is NOT created in no monitor mode when the channel
1045 * is created on a sessiond ask channel command.
1046 */
1047 ret = create_ust_streams(metadata_channel, ctx);
1048 if (ret < 0) {
1049 goto error;
1050 }
1051
1052 metadata_stream = metadata_channel->metadata_stream;
1053 assert(metadata_stream);
1054
1055 pthread_mutex_lock(&metadata_stream->lock);
1056 if (relayd_id != (uint64_t) -1ULL) {
1057 metadata_stream->net_seq_idx = relayd_id;
1058 ret = consumer_send_relayd_stream(metadata_stream, path);
1059 } else {
1060 ret = consumer_stream_create_output_files(metadata_stream,
1061 false);
1062 }
1063 pthread_mutex_unlock(&metadata_stream->lock);
1064 if (ret < 0) {
1065 goto error_stream;
1066 }
1067
1068 do {
1069 health_code_update();
1070
1071 ret = lttng_consumer_read_subbuffer(metadata_stream, ctx);
1072 if (ret < 0) {
1073 goto error_stream;
1074 }
1075 } while (ret > 0);
1076
1077 error_stream:
1078 /*
1079 * Clean up the stream completly because the next snapshot will use a new
1080 * metadata stream.
1081 */
1082 consumer_stream_destroy(metadata_stream, NULL);
1083 cds_list_del(&metadata_stream->send_node);
1084 metadata_channel->metadata_stream = NULL;
1085
1086 error:
1087 rcu_read_unlock();
1088 return ret;
1089 }
1090
1091 /*
1092 * Take a snapshot of all the stream of a channel.
1093 * RCU read-side lock and the channel lock must be held by the caller.
1094 *
1095 * Returns 0 on success, < 0 on error
1096 */
1097 static int snapshot_channel(struct lttng_consumer_channel *channel,
1098 uint64_t key, char *path, uint64_t relayd_id,
1099 uint64_t nb_packets_per_stream,
1100 struct lttng_consumer_local_data *ctx)
1101 {
1102 int ret;
1103 unsigned use_relayd = 0;
1104 unsigned long consumed_pos, produced_pos;
1105 struct lttng_consumer_stream *stream;
1106
1107 assert(path);
1108 assert(ctx);
1109
1110 rcu_read_lock();
1111
1112 if (relayd_id != (uint64_t) -1ULL) {
1113 use_relayd = 1;
1114 }
1115
1116 assert(!channel->monitor);
1117 DBG("UST consumer snapshot channel %" PRIu64, key);
1118
1119 cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
1120 health_code_update();
1121
1122 /* Lock stream because we are about to change its state. */
1123 pthread_mutex_lock(&stream->lock);
1124 assert(channel->trace_chunk);
1125 if (!lttng_trace_chunk_get(channel->trace_chunk)) {
1126 /*
1127 * Can't happen barring an internal error as the channel
1128 * holds a reference to the trace chunk.
1129 */
1130 ERR("Failed to acquire reference to channel's trace chunk");
1131 ret = -1;
1132 goto error_unlock;
1133 }
1134 assert(!stream->trace_chunk);
1135 stream->trace_chunk = channel->trace_chunk;
1136
1137 stream->net_seq_idx = relayd_id;
1138
1139 if (use_relayd) {
1140 ret = consumer_send_relayd_stream(stream, path);
1141 if (ret < 0) {
1142 goto error_unlock;
1143 }
1144 } else {
1145 ret = consumer_stream_create_output_files(stream,
1146 false);
1147 if (ret < 0) {
1148 goto error_unlock;
1149 }
1150 DBG("UST consumer snapshot stream (%" PRIu64 ")",
1151 stream->key);
1152 }
1153
1154 /*
1155 * If tracing is active, we want to perform a "full" buffer flush.
1156 * Else, if quiescent, it has already been done by the prior stop.
1157 */
1158 if (!stream->quiescent) {
1159 ustctl_flush_buffer(stream->ustream, 0);
1160 }
1161
1162 ret = lttng_ustconsumer_take_snapshot(stream);
1163 if (ret < 0) {
1164 ERR("Taking UST snapshot");
1165 goto error_unlock;
1166 }
1167
1168 ret = lttng_ustconsumer_get_produced_snapshot(stream, &produced_pos);
1169 if (ret < 0) {
1170 ERR("Produced UST snapshot position");
1171 goto error_unlock;
1172 }
1173
1174 ret = lttng_ustconsumer_get_consumed_snapshot(stream, &consumed_pos);
1175 if (ret < 0) {
1176 ERR("Consumerd UST snapshot position");
1177 goto error_unlock;
1178 }
1179
1180 /*
1181 * The original value is sent back if max stream size is larger than
1182 * the possible size of the snapshot. Also, we assume that the session
1183 * daemon should never send a maximum stream size that is lower than
1184 * subbuffer size.
1185 */
1186 consumed_pos = consumer_get_consume_start_pos(consumed_pos,
1187 produced_pos, nb_packets_per_stream,
1188 stream->max_sb_size);
1189
1190 while ((long) (consumed_pos - produced_pos) < 0) {
1191 ssize_t read_len;
1192 unsigned long len, padded_len;
1193
1194 health_code_update();
1195
1196 DBG("UST consumer taking snapshot at pos %lu", consumed_pos);
1197
1198 ret = ustctl_get_subbuf(stream->ustream, &consumed_pos);
1199 if (ret < 0) {
1200 if (ret != -EAGAIN) {
1201 PERROR("ustctl_get_subbuf snapshot");
1202 goto error_close_stream;
1203 }
1204 DBG("UST consumer get subbuf failed. Skipping it.");
1205 consumed_pos += stream->max_sb_size;
1206 stream->chan->lost_packets++;
1207 continue;
1208 }
1209
1210 ret = ustctl_get_subbuf_size(stream->ustream, &len);
1211 if (ret < 0) {
1212 ERR("Snapshot ustctl_get_subbuf_size");
1213 goto error_put_subbuf;
1214 }
1215
1216 ret = ustctl_get_padded_subbuf_size(stream->ustream, &padded_len);
1217 if (ret < 0) {
1218 ERR("Snapshot ustctl_get_padded_subbuf_size");
1219 goto error_put_subbuf;
1220 }
1221
1222 read_len = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, len,
1223 padded_len - len, NULL);
1224 if (use_relayd) {
1225 if (read_len != len) {
1226 ret = -EPERM;
1227 goto error_put_subbuf;
1228 }
1229 } else {
1230 if (read_len != padded_len) {
1231 ret = -EPERM;
1232 goto error_put_subbuf;
1233 }
1234 }
1235
1236 ret = ustctl_put_subbuf(stream->ustream);
1237 if (ret < 0) {
1238 ERR("Snapshot ustctl_put_subbuf");
1239 goto error_close_stream;
1240 }
1241 consumed_pos += stream->max_sb_size;
1242 }
1243
1244 /* Simply close the stream so we can use it on the next snapshot. */
1245 consumer_stream_close(stream);
1246 pthread_mutex_unlock(&stream->lock);
1247 }
1248
1249 rcu_read_unlock();
1250 return 0;
1251
1252 error_put_subbuf:
1253 if (ustctl_put_subbuf(stream->ustream) < 0) {
1254 ERR("Snapshot ustctl_put_subbuf");
1255 }
1256 error_close_stream:
1257 consumer_stream_close(stream);
1258 error_unlock:
1259 pthread_mutex_unlock(&stream->lock);
1260 rcu_read_unlock();
1261 return ret;
1262 }
1263
1264 /*
1265 * Receive the metadata updates from the sessiond. Supports receiving
1266 * overlapping metadata, but is needs to always belong to a contiguous
1267 * range starting from 0.
1268 * Be careful about the locks held when calling this function: it needs
1269 * the metadata cache flush to concurrently progress in order to
1270 * complete.
1271 */
1272 int lttng_ustconsumer_recv_metadata(int sock, uint64_t key, uint64_t offset,
1273 uint64_t len, uint64_t version,
1274 struct lttng_consumer_channel *channel, int timer, int wait)
1275 {
1276 int ret, ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1277 char *metadata_str;
1278
1279 DBG("UST consumer push metadata key %" PRIu64 " of len %" PRIu64, key, len);
1280
1281 metadata_str = zmalloc(len * sizeof(char));
1282 if (!metadata_str) {
1283 PERROR("zmalloc metadata string");
1284 ret_code = LTTCOMM_CONSUMERD_ENOMEM;
1285 goto end;
1286 }
1287
1288 health_code_update();
1289
1290 /* Receive metadata string. */
1291 ret = lttcomm_recv_unix_sock(sock, metadata_str, len);
1292 if (ret < 0) {
1293 /* Session daemon is dead so return gracefully. */
1294 ret_code = ret;
1295 goto end_free;
1296 }
1297
1298 health_code_update();
1299
1300 pthread_mutex_lock(&channel->metadata_cache->lock);
1301 ret = consumer_metadata_cache_write(channel, offset, len, version,
1302 metadata_str);
1303 if (ret < 0) {
1304 /* Unable to handle metadata. Notify session daemon. */
1305 ret_code = LTTCOMM_CONSUMERD_ERROR_METADATA;
1306 /*
1307 * Skip metadata flush on write error since the offset and len might
1308 * not have been updated which could create an infinite loop below when
1309 * waiting for the metadata cache to be flushed.
1310 */
1311 pthread_mutex_unlock(&channel->metadata_cache->lock);
1312 goto end_free;
1313 }
1314 pthread_mutex_unlock(&channel->metadata_cache->lock);
1315
1316 if (!wait) {
1317 goto end_free;
1318 }
1319 while (consumer_metadata_cache_flushed(channel, offset + len, timer)) {
1320 DBG("Waiting for metadata to be flushed");
1321
1322 health_code_update();
1323
1324 usleep(DEFAULT_METADATA_AVAILABILITY_WAIT_TIME);
1325 }
1326
1327 end_free:
1328 free(metadata_str);
1329 end:
1330 return ret_code;
1331 }
1332
1333 /*
1334 * Receive command from session daemon and process it.
1335 *
1336 * Return 1 on success else a negative value or 0.
1337 */
1338 int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
1339 int sock, struct pollfd *consumer_sockpoll)
1340 {
1341 ssize_t ret;
1342 enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1343 struct lttcomm_consumer_msg msg;
1344 struct lttng_consumer_channel *channel = NULL;
1345
1346 health_code_update();
1347
1348 ret = lttcomm_recv_unix_sock(sock, &msg, sizeof(msg));
1349 if (ret != sizeof(msg)) {
1350 DBG("Consumer received unexpected message size %zd (expects %zu)",
1351 ret, sizeof(msg));
1352 /*
1353 * The ret value might 0 meaning an orderly shutdown but this is ok
1354 * since the caller handles this.
1355 */
1356 if (ret > 0) {
1357 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD);
1358 ret = -1;
1359 }
1360 return ret;
1361 }
1362
1363 health_code_update();
1364
1365 /* deprecated */
1366 assert(msg.cmd_type != LTTNG_CONSUMER_STOP);
1367
1368 health_code_update();
1369
1370 /* relayd needs RCU read-side lock */
1371 rcu_read_lock();
1372
1373 switch (msg.cmd_type) {
1374 case LTTNG_CONSUMER_ADD_RELAYD_SOCKET:
1375 {
1376 /* Session daemon status message are handled in the following call. */
1377 consumer_add_relayd_socket(msg.u.relayd_sock.net_index,
1378 msg.u.relayd_sock.type, ctx, sock, consumer_sockpoll,
1379 &msg.u.relayd_sock.sock, msg.u.relayd_sock.session_id,
1380 msg.u.relayd_sock.relayd_session_id);
1381 goto end_nosignal;
1382 }
1383 case LTTNG_CONSUMER_DESTROY_RELAYD:
1384 {
1385 uint64_t index = msg.u.destroy_relayd.net_seq_idx;
1386 struct consumer_relayd_sock_pair *relayd;
1387
1388 DBG("UST consumer destroying relayd %" PRIu64, index);
1389
1390 /* Get relayd reference if exists. */
1391 relayd = consumer_find_relayd(index);
1392 if (relayd == NULL) {
1393 DBG("Unable to find relayd %" PRIu64, index);
1394 ret_code = LTTCOMM_CONSUMERD_RELAYD_FAIL;
1395 }
1396
1397 /*
1398 * Each relayd socket pair has a refcount of stream attached to it
1399 * which tells if the relayd is still active or not depending on the
1400 * refcount value.
1401 *
1402 * This will set the destroy flag of the relayd object and destroy it
1403 * if the refcount reaches zero when called.
1404 *
1405 * The destroy can happen either here or when a stream fd hangs up.
1406 */
1407 if (relayd) {
1408 consumer_flag_relayd_for_destroy(relayd);
1409 }
1410
1411 goto end_msg_sessiond;
1412 }
1413 case LTTNG_CONSUMER_UPDATE_STREAM:
1414 {
1415 rcu_read_unlock();
1416 return -ENOSYS;
1417 }
1418 case LTTNG_CONSUMER_DATA_PENDING:
1419 {
1420 int ret, is_data_pending;
1421 uint64_t id = msg.u.data_pending.session_id;
1422
1423 DBG("UST consumer data pending command for id %" PRIu64, id);
1424
1425 is_data_pending = consumer_data_pending(id);
1426
1427 /* Send back returned value to session daemon */
1428 ret = lttcomm_send_unix_sock(sock, &is_data_pending,
1429 sizeof(is_data_pending));
1430 if (ret < 0) {
1431 DBG("Error when sending the data pending ret code: %d", ret);
1432 goto error_fatal;
1433 }
1434
1435 /*
1436 * No need to send back a status message since the data pending
1437 * returned value is the response.
1438 */
1439 break;
1440 }
1441 case LTTNG_CONSUMER_ASK_CHANNEL_CREATION:
1442 {
1443 int ret;
1444 struct ustctl_consumer_channel_attr attr;
1445 const uint64_t chunk_id = msg.u.ask_channel.chunk_id.value;
1446 const struct lttng_credentials buffer_credentials = {
1447 .uid = msg.u.ask_channel.buffer_credentials.uid,
1448 .gid = msg.u.ask_channel.buffer_credentials.gid,
1449 };
1450
1451 /* Create a plain object and reserve a channel key. */
1452 channel = allocate_channel(msg.u.ask_channel.session_id,
1453 msg.u.ask_channel.chunk_id.is_set ?
1454 &chunk_id : NULL,
1455 msg.u.ask_channel.pathname,
1456 msg.u.ask_channel.name,
1457 msg.u.ask_channel.relayd_id,
1458 msg.u.ask_channel.key,
1459 (enum lttng_event_output) msg.u.ask_channel.output,
1460 msg.u.ask_channel.tracefile_size,
1461 msg.u.ask_channel.tracefile_count,
1462 msg.u.ask_channel.session_id_per_pid,
1463 msg.u.ask_channel.monitor,
1464 msg.u.ask_channel.live_timer_interval,
1465 msg.u.ask_channel.root_shm_path,
1466 msg.u.ask_channel.shm_path);
1467 if (!channel) {
1468 goto end_channel_error;
1469 }
1470
1471 LTTNG_OPTIONAL_SET(&channel->buffer_credentials,
1472 buffer_credentials);
1473
1474 /*
1475 * Assign UST application UID to the channel. This value is ignored for
1476 * per PID buffers. This is specific to UST thus setting this after the
1477 * allocation.
1478 */
1479 channel->ust_app_uid = msg.u.ask_channel.ust_app_uid;
1480
1481 /* Build channel attributes from received message. */
1482 attr.subbuf_size = msg.u.ask_channel.subbuf_size;
1483 attr.num_subbuf = msg.u.ask_channel.num_subbuf;
1484 attr.overwrite = msg.u.ask_channel.overwrite;
1485 attr.switch_timer_interval = msg.u.ask_channel.switch_timer_interval;
1486 attr.read_timer_interval = msg.u.ask_channel.read_timer_interval;
1487 attr.chan_id = msg.u.ask_channel.chan_id;
1488 memcpy(attr.uuid, msg.u.ask_channel.uuid, sizeof(attr.uuid));
1489 attr.blocking_timeout= msg.u.ask_channel.blocking_timeout;
1490
1491 /* Match channel buffer type to the UST abi. */
1492 switch (msg.u.ask_channel.output) {
1493 case LTTNG_EVENT_MMAP:
1494 default:
1495 attr.output = LTTNG_UST_MMAP;
1496 break;
1497 }
1498
1499 /* Translate and save channel type. */
1500 switch (msg.u.ask_channel.type) {
1501 case LTTNG_UST_CHAN_PER_CPU:
1502 channel->type = CONSUMER_CHANNEL_TYPE_DATA;
1503 attr.type = LTTNG_UST_CHAN_PER_CPU;
1504 /*
1505 * Set refcount to 1 for owner. Below, we will
1506 * pass ownership to the
1507 * consumer_thread_channel_poll() thread.
1508 */
1509 channel->refcount = 1;
1510 break;
1511 case LTTNG_UST_CHAN_METADATA:
1512 channel->type = CONSUMER_CHANNEL_TYPE_METADATA;
1513 attr.type = LTTNG_UST_CHAN_METADATA;
1514 break;
1515 default:
1516 assert(0);
1517 goto error_fatal;
1518 };
1519
1520 health_code_update();
1521
1522 ret = ask_channel(ctx, channel, &attr);
1523 if (ret < 0) {
1524 goto end_channel_error;
1525 }
1526
1527 if (msg.u.ask_channel.type == LTTNG_UST_CHAN_METADATA) {
1528 ret = consumer_metadata_cache_allocate(channel);
1529 if (ret < 0) {
1530 ERR("Allocating metadata cache");
1531 goto end_channel_error;
1532 }
1533 consumer_timer_switch_start(channel, attr.switch_timer_interval);
1534 attr.switch_timer_interval = 0;
1535 } else {
1536 int monitor_start_ret;
1537
1538 consumer_timer_live_start(channel,
1539 msg.u.ask_channel.live_timer_interval);
1540 monitor_start_ret = consumer_timer_monitor_start(
1541 channel,
1542 msg.u.ask_channel.monitor_timer_interval);
1543 if (monitor_start_ret < 0) {
1544 ERR("Starting channel monitoring timer failed");
1545 goto end_channel_error;
1546 }
1547 }
1548
1549 health_code_update();
1550
1551 /*
1552 * Add the channel to the internal state AFTER all streams were created
1553 * and successfully sent to session daemon. This way, all streams must
1554 * be ready before this channel is visible to the threads.
1555 * If add_channel succeeds, ownership of the channel is
1556 * passed to consumer_thread_channel_poll().
1557 */
1558 ret = add_channel(channel, ctx);
1559 if (ret < 0) {
1560 if (msg.u.ask_channel.type == LTTNG_UST_CHAN_METADATA) {
1561 if (channel->switch_timer_enabled == 1) {
1562 consumer_timer_switch_stop(channel);
1563 }
1564 consumer_metadata_cache_destroy(channel);
1565 }
1566 if (channel->live_timer_enabled == 1) {
1567 consumer_timer_live_stop(channel);
1568 }
1569 if (channel->monitor_timer_enabled == 1) {
1570 consumer_timer_monitor_stop(channel);
1571 }
1572 goto end_channel_error;
1573 }
1574
1575 health_code_update();
1576
1577 /*
1578 * Channel and streams are now created. Inform the session daemon that
1579 * everything went well and should wait to receive the channel and
1580 * streams with ustctl API.
1581 */
1582 ret = consumer_send_status_channel(sock, channel);
1583 if (ret < 0) {
1584 /*
1585 * There is probably a problem on the socket.
1586 */
1587 goto error_fatal;
1588 }
1589
1590 break;
1591 }
1592 case LTTNG_CONSUMER_GET_CHANNEL:
1593 {
1594 int ret, relayd_err = 0;
1595 uint64_t key = msg.u.get_channel.key;
1596 struct lttng_consumer_channel *channel;
1597
1598 channel = consumer_find_channel(key);
1599 if (!channel) {
1600 ERR("UST consumer get channel key %" PRIu64 " not found", key);
1601 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
1602 goto end_get_channel;
1603 }
1604
1605 health_code_update();
1606
1607 /* Send the channel to sessiond (and relayd, if applicable). */
1608 ret = send_channel_to_sessiond_and_relayd(sock, channel, ctx,
1609 &relayd_err);
1610 if (ret < 0) {
1611 if (relayd_err) {
1612 /*
1613 * We were unable to send to the relayd the stream so avoid
1614 * sending back a fatal error to the thread since this is OK
1615 * and the consumer can continue its work. The above call
1616 * has sent the error status message to the sessiond.
1617 */
1618 goto end_get_channel_nosignal;
1619 }
1620 /*
1621 * The communicaton was broken hence there is a bad state between
1622 * the consumer and sessiond so stop everything.
1623 */
1624 goto error_get_channel_fatal;
1625 }
1626
1627 health_code_update();
1628
1629 /*
1630 * In no monitor mode, the streams ownership is kept inside the channel
1631 * so don't send them to the data thread.
1632 */
1633 if (!channel->monitor) {
1634 goto end_get_channel;
1635 }
1636
1637 ret = send_streams_to_thread(channel, ctx);
1638 if (ret < 0) {
1639 /*
1640 * If we are unable to send the stream to the thread, there is
1641 * a big problem so just stop everything.
1642 */
1643 goto error_get_channel_fatal;
1644 }
1645 /* List MUST be empty after or else it could be reused. */
1646 assert(cds_list_empty(&channel->streams.head));
1647 end_get_channel:
1648 goto end_msg_sessiond;
1649 error_get_channel_fatal:
1650 goto error_fatal;
1651 end_get_channel_nosignal:
1652 goto end_nosignal;
1653 }
1654 case LTTNG_CONSUMER_DESTROY_CHANNEL:
1655 {
1656 uint64_t key = msg.u.destroy_channel.key;
1657
1658 /*
1659 * Only called if streams have not been sent to stream
1660 * manager thread. However, channel has been sent to
1661 * channel manager thread.
1662 */
1663 notify_thread_del_channel(ctx, key);
1664 goto end_msg_sessiond;
1665 }
1666 case LTTNG_CONSUMER_CLOSE_METADATA:
1667 {
1668 int ret;
1669
1670 ret = close_metadata(msg.u.close_metadata.key);
1671 if (ret != 0) {
1672 ret_code = ret;
1673 }
1674
1675 goto end_msg_sessiond;
1676 }
1677 case LTTNG_CONSUMER_FLUSH_CHANNEL:
1678 {
1679 int ret;
1680
1681 ret = flush_channel(msg.u.flush_channel.key);
1682 if (ret != 0) {
1683 ret_code = ret;
1684 }
1685
1686 goto end_msg_sessiond;
1687 }
1688 case LTTNG_CONSUMER_CLEAR_QUIESCENT_CHANNEL:
1689 {
1690 int ret;
1691
1692 ret = clear_quiescent_channel(
1693 msg.u.clear_quiescent_channel.key);
1694 if (ret != 0) {
1695 ret_code = ret;
1696 }
1697
1698 goto end_msg_sessiond;
1699 }
1700 case LTTNG_CONSUMER_PUSH_METADATA:
1701 {
1702 int ret;
1703 uint64_t len = msg.u.push_metadata.len;
1704 uint64_t key = msg.u.push_metadata.key;
1705 uint64_t offset = msg.u.push_metadata.target_offset;
1706 uint64_t version = msg.u.push_metadata.version;
1707 struct lttng_consumer_channel *channel;
1708
1709 DBG("UST consumer push metadata key %" PRIu64 " of len %" PRIu64, key,
1710 len);
1711
1712 channel = consumer_find_channel(key);
1713 if (!channel) {
1714 /*
1715 * This is possible if the metadata creation on the consumer side
1716 * is in flight vis-a-vis a concurrent push metadata from the
1717 * session daemon. Simply return that the channel failed and the
1718 * session daemon will handle that message correctly considering
1719 * that this race is acceptable thus the DBG() statement here.
1720 */
1721 DBG("UST consumer push metadata %" PRIu64 " not found", key);
1722 ret_code = LTTCOMM_CONSUMERD_CHANNEL_FAIL;
1723 goto end_push_metadata_msg_sessiond;
1724 }
1725
1726 health_code_update();
1727
1728 if (!len) {
1729 /*
1730 * There is nothing to receive. We have simply
1731 * checked whether the channel can be found.
1732 */
1733 ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1734 goto end_push_metadata_msg_sessiond;
1735 }
1736
1737 /* Tell session daemon we are ready to receive the metadata. */
1738 ret = consumer_send_status_msg(sock, LTTCOMM_CONSUMERD_SUCCESS);
1739 if (ret < 0) {
1740 /* Somehow, the session daemon is not responding anymore. */
1741 goto error_push_metadata_fatal;
1742 }
1743
1744 health_code_update();
1745
1746 /* Wait for more data. */
1747 health_poll_entry();
1748 ret = lttng_consumer_poll_socket(consumer_sockpoll);
1749 health_poll_exit();
1750 if (ret) {
1751 goto error_push_metadata_fatal;
1752 }
1753
1754 health_code_update();
1755
1756 ret = lttng_ustconsumer_recv_metadata(sock, key, offset,
1757 len, version, channel, 0, 1);
1758 if (ret < 0) {
1759 /* error receiving from sessiond */
1760 goto error_push_metadata_fatal;
1761 } else {
1762 ret_code = ret;
1763 goto end_push_metadata_msg_sessiond;
1764 }
1765 end_push_metadata_msg_sessiond:
1766 goto end_msg_sessiond;
1767 error_push_metadata_fatal:
1768 goto error_fatal;
1769 }
1770 case LTTNG_CONSUMER_SETUP_METADATA:
1771 {
1772 int ret;
1773
1774 ret = setup_metadata(ctx, msg.u.setup_metadata.key);
1775 if (ret) {
1776 ret_code = ret;
1777 }
1778 goto end_msg_sessiond;
1779 }
1780 case LTTNG_CONSUMER_SNAPSHOT_CHANNEL:
1781 {
1782 struct lttng_consumer_channel *channel;
1783 uint64_t key = msg.u.snapshot_channel.key;
1784
1785 channel = consumer_find_channel(key);
1786 if (!channel) {
1787 DBG("UST snapshot channel not found for key %" PRIu64, key);
1788 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
1789 } else {
1790 if (msg.u.snapshot_channel.metadata) {
1791 ret = snapshot_metadata(channel, key,
1792 msg.u.snapshot_channel.pathname,
1793 msg.u.snapshot_channel.relayd_id,
1794 ctx);
1795 if (ret < 0) {
1796 ERR("Snapshot metadata failed");
1797 ret_code = LTTCOMM_CONSUMERD_SNAPSHOT_FAILED;
1798 }
1799 } else {
1800 ret = snapshot_channel(channel, key,
1801 msg.u.snapshot_channel.pathname,
1802 msg.u.snapshot_channel.relayd_id,
1803 msg.u.snapshot_channel.nb_packets_per_stream,
1804 ctx);
1805 if (ret < 0) {
1806 ERR("Snapshot channel failed");
1807 ret_code = LTTCOMM_CONSUMERD_SNAPSHOT_FAILED;
1808 }
1809 }
1810 }
1811 health_code_update();
1812 ret = consumer_send_status_msg(sock, ret_code);
1813 if (ret < 0) {
1814 /* Somehow, the session daemon is not responding anymore. */
1815 goto end_nosignal;
1816 }
1817 health_code_update();
1818 break;
1819 }
1820 case LTTNG_CONSUMER_DISCARDED_EVENTS:
1821 {
1822 int ret = 0;
1823 uint64_t discarded_events;
1824 struct lttng_ht_iter iter;
1825 struct lttng_ht *ht;
1826 struct lttng_consumer_stream *stream;
1827 uint64_t id = msg.u.discarded_events.session_id;
1828 uint64_t key = msg.u.discarded_events.channel_key;
1829
1830 DBG("UST consumer discarded events command for session id %"
1831 PRIu64, id);
1832 rcu_read_lock();
1833 pthread_mutex_lock(&consumer_data.lock);
1834
1835 ht = consumer_data.stream_list_ht;
1836
1837 /*
1838 * We only need a reference to the channel, but they are not
1839 * directly indexed, so we just use the first matching stream
1840 * to extract the information we need, we default to 0 if not
1841 * found (no events are dropped if the channel is not yet in
1842 * use).
1843 */
1844 discarded_events = 0;
1845 cds_lfht_for_each_entry_duplicate(ht->ht,
1846 ht->hash_fct(&id, lttng_ht_seed),
1847 ht->match_fct, &id,
1848 &iter.iter, stream, node_session_id.node) {
1849 if (stream->chan->key == key) {
1850 discarded_events = stream->chan->discarded_events;
1851 break;
1852 }
1853 }
1854 pthread_mutex_unlock(&consumer_data.lock);
1855 rcu_read_unlock();
1856
1857 DBG("UST consumer discarded events command for session id %"
1858 PRIu64 ", channel key %" PRIu64, id, key);
1859
1860 health_code_update();
1861
1862 /* Send back returned value to session daemon */
1863 ret = lttcomm_send_unix_sock(sock, &discarded_events, sizeof(discarded_events));
1864 if (ret < 0) {
1865 PERROR("send discarded events");
1866 goto error_fatal;
1867 }
1868
1869 break;
1870 }
1871 case LTTNG_CONSUMER_LOST_PACKETS:
1872 {
1873 int ret;
1874 uint64_t lost_packets;
1875 struct lttng_ht_iter iter;
1876 struct lttng_ht *ht;
1877 struct lttng_consumer_stream *stream;
1878 uint64_t id = msg.u.lost_packets.session_id;
1879 uint64_t key = msg.u.lost_packets.channel_key;
1880
1881 DBG("UST consumer lost packets command for session id %"
1882 PRIu64, id);
1883 rcu_read_lock();
1884 pthread_mutex_lock(&consumer_data.lock);
1885
1886 ht = consumer_data.stream_list_ht;
1887
1888 /*
1889 * We only need a reference to the channel, but they are not
1890 * directly indexed, so we just use the first matching stream
1891 * to extract the information we need, we default to 0 if not
1892 * found (no packets lost if the channel is not yet in use).
1893 */
1894 lost_packets = 0;
1895 cds_lfht_for_each_entry_duplicate(ht->ht,
1896 ht->hash_fct(&id, lttng_ht_seed),
1897 ht->match_fct, &id,
1898 &iter.iter, stream, node_session_id.node) {
1899 if (stream->chan->key == key) {
1900 lost_packets = stream->chan->lost_packets;
1901 break;
1902 }
1903 }
1904 pthread_mutex_unlock(&consumer_data.lock);
1905 rcu_read_unlock();
1906
1907 DBG("UST consumer lost packets command for session id %"
1908 PRIu64 ", channel key %" PRIu64, id, key);
1909
1910 health_code_update();
1911
1912 /* Send back returned value to session daemon */
1913 ret = lttcomm_send_unix_sock(sock, &lost_packets,
1914 sizeof(lost_packets));
1915 if (ret < 0) {
1916 PERROR("send lost packets");
1917 goto error_fatal;
1918 }
1919
1920 break;
1921 }
1922 case LTTNG_CONSUMER_SET_CHANNEL_MONITOR_PIPE:
1923 {
1924 int channel_monitor_pipe;
1925
1926 ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1927 /* Successfully received the command's type. */
1928 ret = consumer_send_status_msg(sock, ret_code);
1929 if (ret < 0) {
1930 goto error_fatal;
1931 }
1932
1933 ret = lttcomm_recv_fds_unix_sock(sock, &channel_monitor_pipe,
1934 1);
1935 if (ret != sizeof(channel_monitor_pipe)) {
1936 ERR("Failed to receive channel monitor pipe");
1937 goto error_fatal;
1938 }
1939
1940 DBG("Received channel monitor pipe (%d)", channel_monitor_pipe);
1941 ret = consumer_timer_thread_set_channel_monitor_pipe(
1942 channel_monitor_pipe);
1943 if (!ret) {
1944 int flags;
1945
1946 ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1947 /* Set the pipe as non-blocking. */
1948 ret = fcntl(channel_monitor_pipe, F_GETFL, 0);
1949 if (ret == -1) {
1950 PERROR("fcntl get flags of the channel monitoring pipe");
1951 goto error_fatal;
1952 }
1953 flags = ret;
1954
1955 ret = fcntl(channel_monitor_pipe, F_SETFL,
1956 flags | O_NONBLOCK);
1957 if (ret == -1) {
1958 PERROR("fcntl set O_NONBLOCK flag of the channel monitoring pipe");
1959 goto error_fatal;
1960 }
1961 DBG("Channel monitor pipe set as non-blocking");
1962 } else {
1963 ret_code = LTTCOMM_CONSUMERD_ALREADY_SET;
1964 }
1965 goto end_msg_sessiond;
1966 }
1967 case LTTNG_CONSUMER_ROTATE_CHANNEL:
1968 {
1969 struct lttng_consumer_channel *channel;
1970 uint64_t key = msg.u.rotate_channel.key;
1971
1972 channel = consumer_find_channel(key);
1973 if (!channel) {
1974 DBG("Channel %" PRIu64 " not found", key);
1975 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
1976 } else {
1977 /*
1978 * Sample the rotate position of all the streams in
1979 * this channel.
1980 */
1981 ret = lttng_consumer_rotate_channel(channel, key,
1982 msg.u.rotate_channel.relayd_id,
1983 msg.u.rotate_channel.metadata,
1984 ctx);
1985 if (ret < 0) {
1986 ERR("Rotate channel failed");
1987 ret_code = LTTCOMM_CONSUMERD_ROTATION_FAIL;
1988 }
1989
1990 health_code_update();
1991 }
1992 ret = consumer_send_status_msg(sock, ret_code);
1993 if (ret < 0) {
1994 /* Somehow, the session daemon is not responding anymore. */
1995 goto end_rotate_channel_nosignal;
1996 }
1997
1998 /*
1999 * Rotate the streams that are ready right now.
2000 * FIXME: this is a second consecutive iteration over the
2001 * streams in a channel, there is probably a better way to
2002 * handle this, but it needs to be after the
2003 * consumer_send_status_msg() call.
2004 */
2005 if (channel) {
2006 ret = lttng_consumer_rotate_ready_streams(
2007 channel, key, ctx);
2008 if (ret < 0) {
2009 ERR("Rotate channel failed");
2010 }
2011 }
2012 break;
2013 end_rotate_channel_nosignal:
2014 goto end_nosignal;
2015 }
2016 case LTTNG_CONSUMER_INIT:
2017 {
2018 ret_code = lttng_consumer_init_command(ctx,
2019 msg.u.init.sessiond_uuid);
2020 health_code_update();
2021 ret = consumer_send_status_msg(sock, ret_code);
2022 if (ret < 0) {
2023 /* Somehow, the session daemon is not responding anymore. */
2024 goto end_nosignal;
2025 }
2026 break;
2027 }
2028 case LTTNG_CONSUMER_CREATE_TRACE_CHUNK:
2029 {
2030 const struct lttng_credentials credentials = {
2031 .uid = msg.u.create_trace_chunk.credentials.value.uid,
2032 .gid = msg.u.create_trace_chunk.credentials.value.gid,
2033 };
2034 const bool is_local_trace =
2035 !msg.u.create_trace_chunk.relayd_id.is_set;
2036 const uint64_t relayd_id =
2037 msg.u.create_trace_chunk.relayd_id.value;
2038 const char *chunk_override_name =
2039 *msg.u.create_trace_chunk.override_name ?
2040 msg.u.create_trace_chunk.override_name :
2041 NULL;
2042 LTTNG_OPTIONAL(struct lttng_directory_handle) chunk_directory_handle =
2043 LTTNG_OPTIONAL_INIT;
2044
2045 /*
2046 * The session daemon will only provide a chunk directory file
2047 * descriptor for local traces.
2048 */
2049 if (is_local_trace) {
2050 int chunk_dirfd;
2051
2052 /* Acnowledge the reception of the command. */
2053 ret = consumer_send_status_msg(sock,
2054 LTTCOMM_CONSUMERD_SUCCESS);
2055 if (ret < 0) {
2056 /* Somehow, the session daemon is not responding anymore. */
2057 goto end_nosignal;
2058 }
2059
2060 ret = lttcomm_recv_fds_unix_sock(sock, &chunk_dirfd, 1);
2061 if (ret != sizeof(chunk_dirfd)) {
2062 ERR("Failed to receive trace chunk directory file descriptor");
2063 goto error_fatal;
2064 }
2065
2066 DBG("Received trace chunk directory fd (%d)",
2067 chunk_dirfd);
2068 ret = lttng_directory_handle_init_from_dirfd(
2069 &chunk_directory_handle.value,
2070 chunk_dirfd);
2071 if (ret) {
2072 ERR("Failed to initialize chunk directory handle from directory file descriptor");
2073 if (close(chunk_dirfd)) {
2074 PERROR("Failed to close chunk directory file descriptor");
2075 }
2076 goto error_fatal;
2077 }
2078 chunk_directory_handle.is_set = true;
2079 }
2080
2081 ret_code = lttng_consumer_create_trace_chunk(
2082 !is_local_trace ? &relayd_id : NULL,
2083 msg.u.create_trace_chunk.session_id,
2084 msg.u.create_trace_chunk.chunk_id,
2085 (time_t) msg.u.create_trace_chunk
2086 .creation_timestamp,
2087 chunk_override_name,
2088 msg.u.create_trace_chunk.credentials.is_set ?
2089 &credentials :
2090 NULL,
2091 chunk_directory_handle.is_set ?
2092 &chunk_directory_handle.value :
2093 NULL);
2094
2095 if (chunk_directory_handle.is_set) {
2096 lttng_directory_handle_fini(
2097 &chunk_directory_handle.value);
2098 }
2099 goto end_msg_sessiond;
2100 }
2101 case LTTNG_CONSUMER_CLOSE_TRACE_CHUNK:
2102 {
2103 enum lttng_trace_chunk_command_type close_command =
2104 msg.u.close_trace_chunk.close_command.value;
2105 const uint64_t relayd_id =
2106 msg.u.close_trace_chunk.relayd_id.value;
2107 struct lttcomm_consumer_close_trace_chunk_reply reply;
2108 char closed_trace_chunk_path[LTTNG_PATH_MAX];
2109 int ret;
2110
2111 ret_code = lttng_consumer_close_trace_chunk(
2112 msg.u.close_trace_chunk.relayd_id.is_set ?
2113 &relayd_id :
2114 NULL,
2115 msg.u.close_trace_chunk.session_id,
2116 msg.u.close_trace_chunk.chunk_id,
2117 (time_t) msg.u.close_trace_chunk.close_timestamp,
2118 msg.u.close_trace_chunk.close_command.is_set ?
2119 &close_command :
2120 NULL, closed_trace_chunk_path);
2121 reply.ret_code = ret_code;
2122 reply.path_length = strlen(closed_trace_chunk_path) + 1;
2123 ret = lttcomm_send_unix_sock(sock, &reply, sizeof(reply));
2124 if (ret != sizeof(reply)) {
2125 goto error_fatal;
2126 }
2127 ret = lttcomm_send_unix_sock(sock, closed_trace_chunk_path,
2128 reply.path_length);
2129 if (ret != reply.path_length) {
2130 goto error_fatal;
2131 }
2132 goto end_nosignal;
2133 }
2134 case LTTNG_CONSUMER_TRACE_CHUNK_EXISTS:
2135 {
2136 const uint64_t relayd_id =
2137 msg.u.trace_chunk_exists.relayd_id.value;
2138
2139 ret_code = lttng_consumer_trace_chunk_exists(
2140 msg.u.trace_chunk_exists.relayd_id.is_set ?
2141 &relayd_id : NULL,
2142 msg.u.trace_chunk_exists.session_id,
2143 msg.u.trace_chunk_exists.chunk_id);
2144 goto end_msg_sessiond;
2145 }
2146 default:
2147 break;
2148 }
2149
2150 end_nosignal:
2151 /*
2152 * Return 1 to indicate success since the 0 value can be a socket
2153 * shutdown during the recv() or send() call.
2154 */
2155 ret = 1;
2156 goto end;
2157
2158 end_msg_sessiond:
2159 /*
2160 * The returned value here is not useful since either way we'll return 1 to
2161 * the caller because the session daemon socket management is done
2162 * elsewhere. Returning a negative code or 0 will shutdown the consumer.
2163 */
2164 ret = consumer_send_status_msg(sock, ret_code);
2165 if (ret < 0) {
2166 goto error_fatal;
2167 }
2168 ret = 1;
2169 goto end;
2170
2171 end_channel_error:
2172 if (channel) {
2173 pthread_mutex_unlock(&channel->lock);
2174 /*
2175 * Free channel here since no one has a reference to it. We don't
2176 * free after that because a stream can store this pointer.
2177 */
2178 destroy_channel(channel);
2179 }
2180 /* We have to send a status channel message indicating an error. */
2181 ret = consumer_send_status_channel(sock, NULL);
2182 if (ret < 0) {
2183 /* Stop everything if session daemon can not be notified. */
2184 goto error_fatal;
2185 }
2186 ret = 1;
2187 goto end;
2188
2189 error_fatal:
2190 /* This will issue a consumer stop. */
2191 ret = -1;
2192 goto end;
2193
2194 end:
2195 rcu_read_unlock();
2196 health_code_update();
2197 return ret;
2198 }
2199
2200 /*
2201 * Wrapper over the mmap() read offset from ust-ctl library. Since this can be
2202 * compiled out, we isolate it in this library.
2203 */
2204 int lttng_ustctl_get_mmap_read_offset(struct lttng_consumer_stream *stream,
2205 unsigned long *off)
2206 {
2207 assert(stream);
2208 assert(stream->ustream);
2209
2210 return ustctl_get_mmap_read_offset(stream->ustream, off);
2211 }
2212
2213 /*
2214 * Wrapper over the mmap() read offset from ust-ctl library. Since this can be
2215 * compiled out, we isolate it in this library.
2216 */
2217 void *lttng_ustctl_get_mmap_base(struct lttng_consumer_stream *stream)
2218 {
2219 assert(stream);
2220 assert(stream->ustream);
2221
2222 return ustctl_get_mmap_base(stream->ustream);
2223 }
2224
2225 void lttng_ustctl_flush_buffer(struct lttng_consumer_stream *stream,
2226 int producer_active)
2227 {
2228 assert(stream);
2229 assert(stream->ustream);
2230
2231 ustctl_flush_buffer(stream->ustream, producer_active);
2232 }
2233
2234 /*
2235 * Take a snapshot for a specific stream.
2236 *
2237 * Returns 0 on success, < 0 on error
2238 */
2239 int lttng_ustconsumer_take_snapshot(struct lttng_consumer_stream *stream)
2240 {
2241 assert(stream);
2242 assert(stream->ustream);
2243
2244 return ustctl_snapshot(stream->ustream);
2245 }
2246
2247 /*
2248 * Sample consumed and produced positions for a specific stream.
2249 *
2250 * Returns 0 on success, < 0 on error.
2251 */
2252 int lttng_ustconsumer_sample_snapshot_positions(
2253 struct lttng_consumer_stream *stream)
2254 {
2255 assert(stream);
2256 assert(stream->ustream);
2257
2258 return ustctl_snapshot_sample_positions(stream->ustream);
2259 }
2260
2261 /*
2262 * Get the produced position
2263 *
2264 * Returns 0 on success, < 0 on error
2265 */
2266 int lttng_ustconsumer_get_produced_snapshot(
2267 struct lttng_consumer_stream *stream, unsigned long *pos)
2268 {
2269 assert(stream);
2270 assert(stream->ustream);
2271 assert(pos);
2272
2273 return ustctl_snapshot_get_produced(stream->ustream, pos);
2274 }
2275
2276 /*
2277 * Get the consumed position
2278 *
2279 * Returns 0 on success, < 0 on error
2280 */
2281 int lttng_ustconsumer_get_consumed_snapshot(
2282 struct lttng_consumer_stream *stream, unsigned long *pos)
2283 {
2284 assert(stream);
2285 assert(stream->ustream);
2286 assert(pos);
2287
2288 return ustctl_snapshot_get_consumed(stream->ustream, pos);
2289 }
2290
2291 void lttng_ustconsumer_flush_buffer(struct lttng_consumer_stream *stream,
2292 int producer)
2293 {
2294 assert(stream);
2295 assert(stream->ustream);
2296
2297 ustctl_flush_buffer(stream->ustream, producer);
2298 }
2299
2300 int lttng_ustconsumer_get_current_timestamp(
2301 struct lttng_consumer_stream *stream, uint64_t *ts)
2302 {
2303 assert(stream);
2304 assert(stream->ustream);
2305 assert(ts);
2306
2307 return ustctl_get_current_timestamp(stream->ustream, ts);
2308 }
2309
2310 int lttng_ustconsumer_get_sequence_number(
2311 struct lttng_consumer_stream *stream, uint64_t *seq)
2312 {
2313 assert(stream);
2314 assert(stream->ustream);
2315 assert(seq);
2316
2317 return ustctl_get_sequence_number(stream->ustream, seq);
2318 }
2319
2320 /*
2321 * Called when the stream signals the consumer that it has hung up.
2322 */
2323 void lttng_ustconsumer_on_stream_hangup(struct lttng_consumer_stream *stream)
2324 {
2325 assert(stream);
2326 assert(stream->ustream);
2327
2328 pthread_mutex_lock(&stream->lock);
2329 if (!stream->quiescent) {
2330 ustctl_flush_buffer(stream->ustream, 0);
2331 stream->quiescent = true;
2332 }
2333 pthread_mutex_unlock(&stream->lock);
2334 stream->hangup_flush_done = 1;
2335 }
2336
2337 void lttng_ustconsumer_del_channel(struct lttng_consumer_channel *chan)
2338 {
2339 int i;
2340
2341 assert(chan);
2342 assert(chan->uchan);
2343 assert(chan->buffer_credentials.is_set);
2344
2345 if (chan->switch_timer_enabled == 1) {
2346 consumer_timer_switch_stop(chan);
2347 }
2348 for (i = 0; i < chan->nr_stream_fds; i++) {
2349 int ret;
2350
2351 ret = close(chan->stream_fds[i]);
2352 if (ret) {
2353 PERROR("close");
2354 }
2355 if (chan->shm_path[0]) {
2356 char shm_path[PATH_MAX];
2357
2358 ret = get_stream_shm_path(shm_path, chan->shm_path, i);
2359 if (ret) {
2360 ERR("Cannot get stream shm path");
2361 }
2362 ret = run_as_unlink(shm_path,
2363 chan->buffer_credentials.value.uid,
2364 chan->buffer_credentials.value.gid);
2365 if (ret) {
2366 PERROR("unlink %s", shm_path);
2367 }
2368 }
2369 }
2370 }
2371
2372 void lttng_ustconsumer_free_channel(struct lttng_consumer_channel *chan)
2373 {
2374 assert(chan);
2375 assert(chan->uchan);
2376 assert(chan->buffer_credentials.is_set);
2377
2378 consumer_metadata_cache_destroy(chan);
2379 ustctl_destroy_channel(chan->uchan);
2380 /* Try to rmdir all directories under shm_path root. */
2381 if (chan->root_shm_path[0]) {
2382 (void) run_as_rmdir_recursive(chan->root_shm_path,
2383 chan->buffer_credentials.value.uid,
2384 chan->buffer_credentials.value.gid,
2385 LTTNG_DIRECTORY_HANDLE_SKIP_NON_EMPTY_FLAG);
2386 }
2387 free(chan->stream_fds);
2388 }
2389
2390 void lttng_ustconsumer_del_stream(struct lttng_consumer_stream *stream)
2391 {
2392 assert(stream);
2393 assert(stream->ustream);
2394
2395 if (stream->chan->switch_timer_enabled == 1) {
2396 consumer_timer_switch_stop(stream->chan);
2397 }
2398 ustctl_destroy_stream(stream->ustream);
2399 }
2400
2401 int lttng_ustconsumer_get_wakeup_fd(struct lttng_consumer_stream *stream)
2402 {
2403 assert(stream);
2404 assert(stream->ustream);
2405
2406 return ustctl_stream_get_wakeup_fd(stream->ustream);
2407 }
2408
2409 int lttng_ustconsumer_close_wakeup_fd(struct lttng_consumer_stream *stream)
2410 {
2411 assert(stream);
2412 assert(stream->ustream);
2413
2414 return ustctl_stream_close_wakeup_fd(stream->ustream);
2415 }
2416
2417 /*
2418 * Populate index values of a UST stream. Values are set in big endian order.
2419 *
2420 * Return 0 on success or else a negative value.
2421 */
2422 static int get_index_values(struct ctf_packet_index *index,
2423 struct ustctl_consumer_stream *ustream)
2424 {
2425 int ret;
2426 uint64_t packet_size, content_size, timestamp_begin, timestamp_end,
2427 events_discarded, stream_id, stream_instance_id,
2428 packet_seq_num;
2429
2430 ret = ustctl_get_timestamp_begin(ustream, &timestamp_begin);
2431 if (ret < 0) {
2432 PERROR("ustctl_get_timestamp_begin");
2433 goto error;
2434 }
2435
2436 ret = ustctl_get_timestamp_end(ustream, &timestamp_end);
2437 if (ret < 0) {
2438 PERROR("ustctl_get_timestamp_end");
2439 goto error;
2440 }
2441
2442 ret = ustctl_get_events_discarded(ustream, &events_discarded);
2443 if (ret < 0) {
2444 PERROR("ustctl_get_events_discarded");
2445 goto error;
2446 }
2447
2448 ret = ustctl_get_content_size(ustream, &content_size);
2449 if (ret < 0) {
2450 PERROR("ustctl_get_content_size");
2451 goto error;
2452 }
2453
2454 ret = ustctl_get_packet_size(ustream, &packet_size);
2455 if (ret < 0) {
2456 PERROR("ustctl_get_packet_size");
2457 goto error;
2458 }
2459
2460 ret = ustctl_get_stream_id(ustream, &stream_id);
2461 if (ret < 0) {
2462 PERROR("ustctl_get_stream_id");
2463 goto error;
2464 }
2465
2466 ret = ustctl_get_instance_id(ustream, &stream_instance_id);
2467 if (ret < 0) {
2468 PERROR("ustctl_get_instance_id");
2469 goto error;
2470 }
2471
2472 ret = ustctl_get_sequence_number(ustream, &packet_seq_num);
2473 if (ret < 0) {
2474 PERROR("ustctl_get_sequence_number");
2475 goto error;
2476 }
2477
2478 *index = (typeof(*index)) {
2479 .offset = index->offset,
2480 .packet_size = htobe64(packet_size),
2481 .content_size = htobe64(content_size),
2482 .timestamp_begin = htobe64(timestamp_begin),
2483 .timestamp_end = htobe64(timestamp_end),
2484 .events_discarded = htobe64(events_discarded),
2485 .stream_id = htobe64(stream_id),
2486 .stream_instance_id = htobe64(stream_instance_id),
2487 .packet_seq_num = htobe64(packet_seq_num),
2488 };
2489
2490 error:
2491 return ret;
2492 }
2493
2494 static
2495 void metadata_stream_reset_cache(struct lttng_consumer_stream *stream,
2496 struct consumer_metadata_cache *cache)
2497 {
2498 DBG("Metadata stream update to version %" PRIu64,
2499 cache->version);
2500 stream->ust_metadata_pushed = 0;
2501 stream->metadata_version = cache->version;
2502 stream->reset_metadata_flag = 1;
2503 }
2504
2505 /*
2506 * Check if the version of the metadata stream and metadata cache match.
2507 * If the cache got updated, reset the metadata stream.
2508 * The stream lock and metadata cache lock MUST be held.
2509 * Return 0 on success, a negative value on error.
2510 */
2511 static
2512 int metadata_stream_check_version(struct lttng_consumer_stream *stream)
2513 {
2514 int ret = 0;
2515 struct consumer_metadata_cache *cache = stream->chan->metadata_cache;
2516
2517 if (cache->version == stream->metadata_version) {
2518 goto end;
2519 }
2520 metadata_stream_reset_cache(stream, cache);
2521
2522 end:
2523 return ret;
2524 }
2525
2526 /*
2527 * Write up to one packet from the metadata cache to the channel.
2528 *
2529 * Returns the number of bytes pushed in the cache, or a negative value
2530 * on error.
2531 */
2532 static
2533 int commit_one_metadata_packet(struct lttng_consumer_stream *stream)
2534 {
2535 ssize_t write_len;
2536 int ret;
2537
2538 pthread_mutex_lock(&stream->chan->metadata_cache->lock);
2539 ret = metadata_stream_check_version(stream);
2540 if (ret < 0) {
2541 goto end;
2542 }
2543 if (stream->chan->metadata_cache->max_offset
2544 == stream->ust_metadata_pushed) {
2545 ret = 0;
2546 goto end;
2547 }
2548
2549 write_len = ustctl_write_one_packet_to_channel(stream->chan->uchan,
2550 &stream->chan->metadata_cache->data[stream->ust_metadata_pushed],
2551 stream->chan->metadata_cache->max_offset
2552 - stream->ust_metadata_pushed);
2553 assert(write_len != 0);
2554 if (write_len < 0) {
2555 ERR("Writing one metadata packet");
2556 ret = -1;
2557 goto end;
2558 }
2559 stream->ust_metadata_pushed += write_len;
2560
2561 assert(stream->chan->metadata_cache->max_offset >=
2562 stream->ust_metadata_pushed);
2563 ret = write_len;
2564
2565 /*
2566 * Switch packet (but don't open the next one) on every commit of
2567 * a metadata packet. Since the subbuffer is fully filled (with padding,
2568 * if needed), the stream is "quiescent" after this commit.
2569 */
2570 ustctl_flush_buffer(stream->ustream, 1);
2571 stream->quiescent = true;
2572 end:
2573 pthread_mutex_unlock(&stream->chan->metadata_cache->lock);
2574 return ret;
2575 }
2576
2577
2578 /*
2579 * Sync metadata meaning request them to the session daemon and snapshot to the
2580 * metadata thread can consumer them.
2581 *
2582 * Metadata stream lock is held here, but we need to release it when
2583 * interacting with sessiond, else we cause a deadlock with live
2584 * awaiting on metadata to be pushed out.
2585 *
2586 * Return 0 if new metadatda is available, EAGAIN if the metadata stream
2587 * is empty or a negative value on error.
2588 */
2589 int lttng_ustconsumer_sync_metadata(struct lttng_consumer_local_data *ctx,
2590 struct lttng_consumer_stream *metadata)
2591 {
2592 int ret;
2593 int retry = 0;
2594
2595 assert(ctx);
2596 assert(metadata);
2597
2598 pthread_mutex_unlock(&metadata->lock);
2599 /*
2600 * Request metadata from the sessiond, but don't wait for the flush
2601 * because we locked the metadata thread.
2602 */
2603 ret = lttng_ustconsumer_request_metadata(ctx, metadata->chan, 0, 0);
2604 pthread_mutex_lock(&metadata->lock);
2605 if (ret < 0) {
2606 goto end;
2607 }
2608
2609 ret = commit_one_metadata_packet(metadata);
2610 if (ret <= 0) {
2611 goto end;
2612 } else if (ret > 0) {
2613 retry = 1;
2614 }
2615
2616 ret = ustctl_snapshot(metadata->ustream);
2617 if (ret < 0) {
2618 if (errno != EAGAIN) {
2619 ERR("Sync metadata, taking UST snapshot");
2620 goto end;
2621 }
2622 DBG("No new metadata when syncing them.");
2623 /* No new metadata, exit. */
2624 ret = ENODATA;
2625 goto end;
2626 }
2627
2628 /*
2629 * After this flush, we still need to extract metadata.
2630 */
2631 if (retry) {
2632 ret = EAGAIN;
2633 }
2634
2635 end:
2636 return ret;
2637 }
2638
2639 /*
2640 * Return 0 on success else a negative value.
2641 */
2642 static int notify_if_more_data(struct lttng_consumer_stream *stream,
2643 struct lttng_consumer_local_data *ctx)
2644 {
2645 int ret;
2646 struct ustctl_consumer_stream *ustream;
2647
2648 assert(stream);
2649 assert(ctx);
2650
2651 ustream = stream->ustream;
2652
2653 /*
2654 * First, we are going to check if there is a new subbuffer available
2655 * before reading the stream wait_fd.
2656 */
2657 /* Get the next subbuffer */
2658 ret = ustctl_get_next_subbuf(ustream);
2659 if (ret) {
2660 /* No more data found, flag the stream. */
2661 stream->has_data = 0;
2662 ret = 0;
2663 goto end;
2664 }
2665
2666 ret = ustctl_put_subbuf(ustream);
2667 assert(!ret);
2668
2669 /* This stream still has data. Flag it and wake up the data thread. */
2670 stream->has_data = 1;
2671
2672 if (stream->monitor && !stream->hangup_flush_done && !ctx->has_wakeup) {
2673 ssize_t writelen;
2674
2675 writelen = lttng_pipe_write(ctx->consumer_wakeup_pipe, "!", 1);
2676 if (writelen < 0 && errno != EAGAIN && errno != EWOULDBLOCK) {
2677 ret = writelen;
2678 goto end;
2679 }
2680
2681 /* The wake up pipe has been notified. */
2682 ctx->has_wakeup = 1;
2683 }
2684 ret = 0;
2685
2686 end:
2687 return ret;
2688 }
2689
2690 static
2691 int update_stream_stats(struct lttng_consumer_stream *stream)
2692 {
2693 int ret;
2694 uint64_t seq, discarded;
2695
2696 ret = ustctl_get_sequence_number(stream->ustream, &seq);
2697 if (ret < 0) {
2698 PERROR("ustctl_get_sequence_number");
2699 goto end;
2700 }
2701 /*
2702 * Start the sequence when we extract the first packet in case we don't
2703 * start at 0 (for example if a consumer is not connected to the
2704 * session immediately after the beginning).
2705 */
2706 if (stream->last_sequence_number == -1ULL) {
2707 stream->last_sequence_number = seq;
2708 } else if (seq > stream->last_sequence_number) {
2709 stream->chan->lost_packets += seq -
2710 stream->last_sequence_number - 1;
2711 } else {
2712 /* seq <= last_sequence_number */
2713 ERR("Sequence number inconsistent : prev = %" PRIu64
2714 ", current = %" PRIu64,
2715 stream->last_sequence_number, seq);
2716 ret = -1;
2717 goto end;
2718 }
2719 stream->last_sequence_number = seq;
2720
2721 ret = ustctl_get_events_discarded(stream->ustream, &discarded);
2722 if (ret < 0) {
2723 PERROR("kernctl_get_events_discarded");
2724 goto end;
2725 }
2726 if (discarded < stream->last_discarded_events) {
2727 /*
2728 * Overflow has occurred. We assume only one wrap-around
2729 * has occurred.
2730 */
2731 stream->chan->discarded_events +=
2732 (1ULL << (CAA_BITS_PER_LONG - 1)) -
2733 stream->last_discarded_events + discarded;
2734 } else {
2735 stream->chan->discarded_events += discarded -
2736 stream->last_discarded_events;
2737 }
2738 stream->last_discarded_events = discarded;
2739 ret = 0;
2740
2741 end:
2742 return ret;
2743 }
2744
2745 /*
2746 * Read subbuffer from the given stream.
2747 *
2748 * Stream and channel locks MUST be acquired by the caller.
2749 *
2750 * Return 0 on success else a negative value.
2751 */
2752 int lttng_ustconsumer_read_subbuffer(struct lttng_consumer_stream *stream,
2753 struct lttng_consumer_local_data *ctx)
2754 {
2755 unsigned long len, subbuf_size, padding;
2756 int err, write_index = 1, rotation_ret;
2757 long ret = 0;
2758 struct ustctl_consumer_stream *ustream;
2759 struct ctf_packet_index index;
2760
2761 assert(stream);
2762 assert(stream->ustream);
2763 assert(ctx);
2764
2765 DBG("In UST read_subbuffer (wait_fd: %d, name: %s)", stream->wait_fd,
2766 stream->name);
2767
2768 /* Ease our life for what's next. */
2769 ustream = stream->ustream;
2770
2771 /*
2772 * We can consume the 1 byte written into the wait_fd by UST. Don't trigger
2773 * error if we cannot read this one byte (read returns 0), or if the error
2774 * is EAGAIN or EWOULDBLOCK.
2775 *
2776 * This is only done when the stream is monitored by a thread, before the
2777 * flush is done after a hangup and if the stream is not flagged with data
2778 * since there might be nothing to consume in the wait fd but still have
2779 * data available flagged by the consumer wake up pipe.
2780 */
2781 if (stream->monitor && !stream->hangup_flush_done && !stream->has_data) {
2782 char dummy;
2783 ssize_t readlen;
2784
2785 readlen = lttng_read(stream->wait_fd, &dummy, 1);
2786 if (readlen < 0 && errno != EAGAIN && errno != EWOULDBLOCK) {
2787 ret = readlen;
2788 goto error;
2789 }
2790 }
2791
2792 /*
2793 * If the stream was flagged to be ready for rotation before we extract the
2794 * next packet, rotate it now.
2795 */
2796 if (stream->rotate_ready) {
2797 DBG("Rotate stream before extracting data");
2798 rotation_ret = lttng_consumer_rotate_stream(ctx, stream);
2799 if (rotation_ret < 0) {
2800 ERR("Stream rotation error");
2801 ret = -1;
2802 goto error;
2803 }
2804 }
2805
2806 retry:
2807 /* Get the next subbuffer */
2808 err = ustctl_get_next_subbuf(ustream);
2809 if (err != 0) {
2810 /*
2811 * Populate metadata info if the existing info has
2812 * already been read.
2813 */
2814 if (stream->metadata_flag) {
2815 ret = commit_one_metadata_packet(stream);
2816 if (ret <= 0) {
2817 goto error;
2818 }
2819 goto retry;
2820 }
2821
2822 ret = err; /* ustctl_get_next_subbuf returns negative, caller expect positive. */
2823 /*
2824 * This is a debug message even for single-threaded consumer,
2825 * because poll() have more relaxed criterions than get subbuf,
2826 * so get_subbuf may fail for short race windows where poll()
2827 * would issue wakeups.
2828 */
2829 DBG("Reserving sub buffer failed (everything is normal, "
2830 "it is due to concurrency) [ret: %d]", err);
2831 goto error;
2832 }
2833 assert(stream->chan->output == CONSUMER_CHANNEL_MMAP);
2834
2835 if (!stream->metadata_flag) {
2836 index.offset = htobe64(stream->out_fd_offset);
2837 ret = get_index_values(&index, ustream);
2838 if (ret < 0) {
2839 err = ustctl_put_subbuf(ustream);
2840 assert(err == 0);
2841 goto error;
2842 }
2843
2844 /* Update the stream's sequence and discarded events count. */
2845 ret = update_stream_stats(stream);
2846 if (ret < 0) {
2847 PERROR("kernctl_get_events_discarded");
2848 err = ustctl_put_subbuf(ustream);
2849 assert(err == 0);
2850 goto error;
2851 }
2852 } else {
2853 write_index = 0;
2854 }
2855
2856 /* Get the full padded subbuffer size */
2857 err = ustctl_get_padded_subbuf_size(ustream, &len);
2858 assert(err == 0);
2859
2860 /* Get subbuffer data size (without padding) */
2861 err = ustctl_get_subbuf_size(ustream, &subbuf_size);
2862 assert(err == 0);
2863
2864 /* Make sure we don't get a subbuffer size bigger than the padded */
2865 assert(len >= subbuf_size);
2866
2867 padding = len - subbuf_size;
2868
2869 /* write the subbuffer to the tracefile */
2870 ret = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, subbuf_size, padding, &index);
2871 /*
2872 * The mmap operation should write subbuf_size amount of data when network
2873 * streaming or the full padding (len) size when we are _not_ streaming.
2874 */
2875 if ((ret != subbuf_size && stream->net_seq_idx != (uint64_t) -1ULL) ||
2876 (ret != len && stream->net_seq_idx == (uint64_t) -1ULL)) {
2877 /*
2878 * Display the error but continue processing to try to release the
2879 * subbuffer. This is a DBG statement since any unexpected kill or
2880 * signal, the application gets unregistered, relayd gets closed or
2881 * anything that affects the buffer lifetime will trigger this error.
2882 * So, for the sake of the user, don't print this error since it can
2883 * happen and it is OK with the code flow.
2884 */
2885 DBG("Error writing to tracefile "
2886 "(ret: %ld != len: %lu != subbuf_size: %lu)",
2887 ret, len, subbuf_size);
2888 write_index = 0;
2889 }
2890 err = ustctl_put_next_subbuf(ustream);
2891 assert(err == 0);
2892
2893 /*
2894 * This will consumer the byte on the wait_fd if and only if there is not
2895 * next subbuffer to be acquired.
2896 */
2897 if (!stream->metadata_flag) {
2898 ret = notify_if_more_data(stream, ctx);
2899 if (ret < 0) {
2900 goto error;
2901 }
2902 }
2903
2904 /* Write index if needed. */
2905 if (!write_index) {
2906 goto rotate;
2907 }
2908
2909 if (stream->chan->live_timer_interval && !stream->metadata_flag) {
2910 /*
2911 * In live, block until all the metadata is sent.
2912 */
2913 pthread_mutex_lock(&stream->metadata_timer_lock);
2914 assert(!stream->missed_metadata_flush);
2915 stream->waiting_on_metadata = true;
2916 pthread_mutex_unlock(&stream->metadata_timer_lock);
2917
2918 err = consumer_stream_sync_metadata(ctx, stream->session_id);
2919
2920 pthread_mutex_lock(&stream->metadata_timer_lock);
2921 stream->waiting_on_metadata = false;
2922 if (stream->missed_metadata_flush) {
2923 stream->missed_metadata_flush = false;
2924 pthread_mutex_unlock(&stream->metadata_timer_lock);
2925 (void) consumer_flush_ust_index(stream);
2926 } else {
2927 pthread_mutex_unlock(&stream->metadata_timer_lock);
2928 }
2929
2930 if (err < 0) {
2931 goto error;
2932 }
2933 }
2934
2935 assert(!stream->metadata_flag);
2936 err = consumer_stream_write_index(stream, &index);
2937 if (err < 0) {
2938 goto error;
2939 }
2940
2941 rotate:
2942 /*
2943 * After extracting the packet, we check if the stream is now ready to be
2944 * rotated and perform the action immediately.
2945 */
2946 rotation_ret = lttng_consumer_stream_is_rotate_ready(stream);
2947 if (rotation_ret == 1) {
2948 rotation_ret = lttng_consumer_rotate_stream(ctx, stream);
2949 if (rotation_ret < 0) {
2950 ERR("Stream rotation error");
2951 ret = -1;
2952 goto error;
2953 }
2954 } else if (rotation_ret < 0) {
2955 ERR("Checking if stream is ready to rotate");
2956 ret = -1;
2957 goto error;
2958 }
2959 error:
2960 return ret;
2961 }
2962
2963 /*
2964 * Called when a stream is created.
2965 *
2966 * Return 0 on success or else a negative value.
2967 */
2968 int lttng_ustconsumer_on_recv_stream(struct lttng_consumer_stream *stream)
2969 {
2970 int ret;
2971
2972 assert(stream);
2973
2974 /*
2975 * Don't create anything if this is set for streaming or if there is
2976 * no current trace chunk on the parent channel.
2977 */
2978 if (stream->net_seq_idx == (uint64_t) -1ULL && stream->chan->monitor &&
2979 stream->chan->trace_chunk) {
2980 ret = consumer_stream_create_output_files(stream, true);
2981 if (ret) {
2982 goto error;
2983 }
2984 }
2985 ret = 0;
2986
2987 error:
2988 return ret;
2989 }
2990
2991 /*
2992 * Check if data is still being extracted from the buffers for a specific
2993 * stream. Consumer data lock MUST be acquired before calling this function
2994 * and the stream lock.
2995 *
2996 * Return 1 if the traced data are still getting read else 0 meaning that the
2997 * data is available for trace viewer reading.
2998 */
2999 int lttng_ustconsumer_data_pending(struct lttng_consumer_stream *stream)
3000 {
3001 int ret;
3002
3003 assert(stream);
3004 assert(stream->ustream);
3005
3006 DBG("UST consumer checking data pending");
3007
3008 if (stream->endpoint_status != CONSUMER_ENDPOINT_ACTIVE) {
3009 ret = 0;
3010 goto end;
3011 }
3012
3013 if (stream->chan->type == CONSUMER_CHANNEL_TYPE_METADATA) {
3014 uint64_t contiguous, pushed;
3015
3016 /* Ease our life a bit. */
3017 contiguous = stream->chan->metadata_cache->max_offset;
3018 pushed = stream->ust_metadata_pushed;
3019
3020 /*
3021 * We can simply check whether all contiguously available data
3022 * has been pushed to the ring buffer, since the push operation
3023 * is performed within get_next_subbuf(), and because both
3024 * get_next_subbuf() and put_next_subbuf() are issued atomically
3025 * thanks to the stream lock within
3026 * lttng_ustconsumer_read_subbuffer(). This basically means that
3027 * whetnever ust_metadata_pushed is incremented, the associated
3028 * metadata has been consumed from the metadata stream.
3029 */
3030 DBG("UST consumer metadata pending check: contiguous %" PRIu64 " vs pushed %" PRIu64,
3031 contiguous, pushed);
3032 assert(((int64_t) (contiguous - pushed)) >= 0);
3033 if ((contiguous != pushed) ||
3034 (((int64_t) contiguous - pushed) > 0 || contiguous == 0)) {
3035 ret = 1; /* Data is pending */
3036 goto end;
3037 }
3038 } else {
3039 ret = ustctl_get_next_subbuf(stream->ustream);
3040 if (ret == 0) {
3041 /*
3042 * There is still data so let's put back this
3043 * subbuffer.
3044 */
3045 ret = ustctl_put_subbuf(stream->ustream);
3046 assert(ret == 0);
3047 ret = 1; /* Data is pending */
3048 goto end;
3049 }
3050 }
3051
3052 /* Data is NOT pending so ready to be read. */
3053 ret = 0;
3054
3055 end:
3056 return ret;
3057 }
3058
3059 /*
3060 * Stop a given metadata channel timer if enabled and close the wait fd which
3061 * is the poll pipe of the metadata stream.
3062 *
3063 * This MUST be called with the metadata channel lock acquired.
3064 */
3065 void lttng_ustconsumer_close_metadata(struct lttng_consumer_channel *metadata)
3066 {
3067 int ret;
3068
3069 assert(metadata);
3070 assert(metadata->type == CONSUMER_CHANNEL_TYPE_METADATA);
3071
3072 DBG("Closing metadata channel key %" PRIu64, metadata->key);
3073
3074 if (metadata->switch_timer_enabled == 1) {
3075 consumer_timer_switch_stop(metadata);
3076 }
3077
3078 if (!metadata->metadata_stream) {
3079 goto end;
3080 }
3081
3082 /*
3083 * Closing write side so the thread monitoring the stream wakes up if any
3084 * and clean the metadata stream.
3085 */
3086 if (metadata->metadata_stream->ust_metadata_poll_pipe[1] >= 0) {
3087 ret = close(metadata->metadata_stream->ust_metadata_poll_pipe[1]);
3088 if (ret < 0) {
3089 PERROR("closing metadata pipe write side");
3090 }
3091 metadata->metadata_stream->ust_metadata_poll_pipe[1] = -1;
3092 }
3093
3094 end:
3095 return;
3096 }
3097
3098 /*
3099 * Close every metadata stream wait fd of the metadata hash table. This
3100 * function MUST be used very carefully so not to run into a race between the
3101 * metadata thread handling streams and this function closing their wait fd.
3102 *
3103 * For UST, this is used when the session daemon hangs up. Its the metadata
3104 * producer so calling this is safe because we are assured that no state change
3105 * can occur in the metadata thread for the streams in the hash table.
3106 */
3107 void lttng_ustconsumer_close_all_metadata(struct lttng_ht *metadata_ht)
3108 {
3109 struct lttng_ht_iter iter;
3110 struct lttng_consumer_stream *stream;
3111
3112 assert(metadata_ht);
3113 assert(metadata_ht->ht);
3114
3115 DBG("UST consumer closing all metadata streams");
3116
3117 rcu_read_lock();
3118 cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream,
3119 node.node) {
3120
3121 health_code_update();
3122
3123 pthread_mutex_lock(&stream->chan->lock);
3124 lttng_ustconsumer_close_metadata(stream->chan);
3125 pthread_mutex_unlock(&stream->chan->lock);
3126
3127 }
3128 rcu_read_unlock();
3129 }
3130
3131 void lttng_ustconsumer_close_stream_wakeup(struct lttng_consumer_stream *stream)
3132 {
3133 int ret;
3134
3135 ret = ustctl_stream_close_wakeup_fd(stream->ustream);
3136 if (ret < 0) {
3137 ERR("Unable to close wakeup fd");
3138 }
3139 }
3140
3141 /*
3142 * Please refer to consumer-timer.c before adding any lock within this
3143 * function or any of its callees. Timers have a very strict locking
3144 * semantic with respect to teardown. Failure to respect this semantic
3145 * introduces deadlocks.
3146 *
3147 * DON'T hold the metadata lock when calling this function, else this
3148 * can cause deadlock involving consumer awaiting for metadata to be
3149 * pushed out due to concurrent interaction with the session daemon.
3150 */
3151 int lttng_ustconsumer_request_metadata(struct lttng_consumer_local_data *ctx,
3152 struct lttng_consumer_channel *channel, int timer, int wait)
3153 {
3154 struct lttcomm_metadata_request_msg request;
3155 struct lttcomm_consumer_msg msg;
3156 enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS;
3157 uint64_t len, key, offset, version;
3158 int ret;
3159
3160 assert(channel);
3161 assert(channel->metadata_cache);
3162
3163 memset(&request, 0, sizeof(request));
3164
3165 /* send the metadata request to sessiond */
3166 switch (consumer_data.type) {
3167 case LTTNG_CONSUMER64_UST:
3168 request.bits_per_long = 64;
3169 break;
3170 case LTTNG_CONSUMER32_UST:
3171 request.bits_per_long = 32;
3172 break;
3173 default:
3174 request.bits_per_long = 0;
3175 break;
3176 }
3177
3178 request.session_id = channel->session_id;
3179 request.session_id_per_pid = channel->session_id_per_pid;
3180 /*
3181 * Request the application UID here so the metadata of that application can
3182 * be sent back. The channel UID corresponds to the user UID of the session
3183 * used for the rights on the stream file(s).
3184 */
3185 request.uid = channel->ust_app_uid;
3186 request.key = channel->key;
3187
3188 DBG("Sending metadata request to sessiond, session id %" PRIu64
3189 ", per-pid %" PRIu64 ", app UID %u and channel key %" PRIu64,
3190 request.session_id, request.session_id_per_pid, request.uid,
3191 request.key);
3192
3193 pthread_mutex_lock(&ctx->metadata_socket_lock);
3194
3195 health_code_update();
3196
3197 ret = lttcomm_send_unix_sock(ctx->consumer_metadata_socket, &request,
3198 sizeof(request));
3199 if (ret < 0) {
3200 ERR("Asking metadata to sessiond");
3201 goto end;
3202 }
3203
3204 health_code_update();
3205
3206 /* Receive the metadata from sessiond */
3207 ret = lttcomm_recv_unix_sock(ctx->consumer_metadata_socket, &msg,
3208 sizeof(msg));
3209 if (ret != sizeof(msg)) {
3210 DBG("Consumer received unexpected message size %d (expects %zu)",
3211 ret, sizeof(msg));
3212 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD);
3213 /*
3214 * The ret value might 0 meaning an orderly shutdown but this is ok
3215 * since the caller handles this.
3216 */
3217 goto end;
3218 }
3219
3220 health_code_update();
3221
3222 if (msg.cmd_type == LTTNG_ERR_UND) {
3223 /* No registry found */
3224 (void) consumer_send_status_msg(ctx->consumer_metadata_socket,
3225 ret_code);
3226 ret = 0;
3227 goto end;
3228 } else if (msg.cmd_type != LTTNG_CONSUMER_PUSH_METADATA) {
3229 ERR("Unexpected cmd_type received %d", msg.cmd_type);
3230 ret = -1;
3231 goto end;
3232 }
3233
3234 len = msg.u.push_metadata.len;
3235 key = msg.u.push_metadata.key;
3236 offset = msg.u.push_metadata.target_offset;
3237 version = msg.u.push_metadata.version;
3238
3239 assert(key == channel->key);
3240 if (len == 0) {
3241 DBG("No new metadata to receive for key %" PRIu64, key);
3242 }
3243
3244 health_code_update();
3245
3246 /* Tell session daemon we are ready to receive the metadata. */
3247 ret = consumer_send_status_msg(ctx->consumer_metadata_socket,
3248 LTTCOMM_CONSUMERD_SUCCESS);
3249 if (ret < 0 || len == 0) {
3250 /*
3251 * Somehow, the session daemon is not responding anymore or there is
3252 * nothing to receive.
3253 */
3254 goto end;
3255 }
3256
3257 health_code_update();
3258
3259 ret = lttng_ustconsumer_recv_metadata(ctx->consumer_metadata_socket,
3260 key, offset, len, version, channel, timer, wait);
3261 if (ret >= 0) {
3262 /*
3263 * Only send the status msg if the sessiond is alive meaning a positive
3264 * ret code.
3265 */
3266 (void) consumer_send_status_msg(ctx->consumer_metadata_socket, ret);
3267 }
3268 ret = 0;
3269
3270 end:
3271 health_code_update();
3272
3273 pthread_mutex_unlock(&ctx->metadata_socket_lock);
3274 return ret;
3275 }
3276
3277 /*
3278 * Return the ustctl call for the get stream id.
3279 */
3280 int lttng_ustconsumer_get_stream_id(struct lttng_consumer_stream *stream,
3281 uint64_t *stream_id)
3282 {
3283 assert(stream);
3284 assert(stream_id);
3285
3286 return ustctl_get_stream_id(stream->ustream, stream_id);
3287 }
This page took 0.102349 seconds and 4 git commands to generate.