Fix: consumer: snapshot error return code
[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 free(stream);
77 }
78
79 /*
80 * If a channel is available meaning that was created before the streams
81 * were, delete it.
82 */
83 if (channel->uchan) {
84 lttng_ustconsumer_del_channel(channel);
85 lttng_ustconsumer_free_channel(channel);
86 }
87 free(channel);
88 }
89
90 /*
91 * Add channel to internal consumer state.
92 *
93 * Returns 0 on success or else a negative value.
94 */
95 static int add_channel(struct lttng_consumer_channel *channel,
96 struct lttng_consumer_local_data *ctx)
97 {
98 int ret = 0;
99
100 assert(channel);
101 assert(ctx);
102
103 if (ctx->on_recv_channel != NULL) {
104 ret = ctx->on_recv_channel(channel);
105 if (ret == 0) {
106 ret = consumer_add_channel(channel, ctx);
107 } else if (ret < 0) {
108 /* Most likely an ENOMEM. */
109 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
110 goto error;
111 }
112 } else {
113 ret = consumer_add_channel(channel, ctx);
114 }
115
116 DBG("UST consumer channel added (key: %" PRIu64 ")", channel->key);
117
118 error:
119 return ret;
120 }
121
122 /*
123 * Allocate and return a consumer channel object.
124 */
125 static struct lttng_consumer_channel *allocate_channel(uint64_t session_id,
126 const char *pathname, const char *name, uid_t uid, gid_t gid,
127 uint64_t relayd_id, uint64_t key, enum lttng_event_output output,
128 uint64_t tracefile_size, uint64_t tracefile_count,
129 uint64_t session_id_per_pid, unsigned int monitor,
130 unsigned int live_timer_interval,
131 const char *root_shm_path, const char *shm_path)
132 {
133 assert(pathname);
134 assert(name);
135
136 return consumer_allocate_channel(key, session_id, pathname, name, uid,
137 gid, relayd_id, output, tracefile_size,
138 tracefile_count, session_id_per_pid, monitor,
139 live_timer_interval, root_shm_path, shm_path);
140 }
141
142 /*
143 * Allocate and return a consumer stream object. If _alloc_ret is not NULL, the
144 * error value if applicable is set in it else it is kept untouched.
145 *
146 * Return NULL on error else the newly allocated stream object.
147 */
148 static struct lttng_consumer_stream *allocate_stream(int cpu, int key,
149 struct lttng_consumer_channel *channel,
150 struct lttng_consumer_local_data *ctx, int *_alloc_ret,
151 uint64_t trace_archive_id)
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 LTTNG_CONSUMER_ACTIVE_STREAM,
162 channel->name,
163 channel->uid,
164 channel->gid,
165 channel->relayd_id,
166 channel->session_id,
167 cpu,
168 &alloc_ret,
169 channel->type,
170 channel->monitor,
171 trace_archive_id);
172 if (stream == NULL) {
173 switch (alloc_ret) {
174 case -ENOENT:
175 /*
176 * We could not find the channel. Can happen if cpu hotplug
177 * happens while tearing down.
178 */
179 DBG3("Could not find channel");
180 break;
181 case -ENOMEM:
182 case -EINVAL:
183 default:
184 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
185 break;
186 }
187 goto error;
188 }
189
190 consumer_stream_update_channel_attributes(stream, channel);
191 stream->chan = channel;
192
193 error:
194 if (_alloc_ret) {
195 *_alloc_ret = alloc_ret;
196 }
197 return stream;
198 }
199
200 /*
201 * Send the given stream pointer to the corresponding thread.
202 *
203 * Returns 0 on success else a negative value.
204 */
205 static int send_stream_to_thread(struct lttng_consumer_stream *stream,
206 struct lttng_consumer_local_data *ctx)
207 {
208 int ret;
209 struct lttng_pipe *stream_pipe;
210
211 /* Get the right pipe where the stream will be sent. */
212 if (stream->metadata_flag) {
213 consumer_add_metadata_stream(stream);
214 stream_pipe = ctx->consumer_metadata_pipe;
215 } else {
216 consumer_add_data_stream(stream);
217 stream_pipe = ctx->consumer_data_pipe;
218 }
219
220 /*
221 * From this point on, the stream's ownership has been moved away from
222 * the channel and it becomes globally visible. Hence, remove it from
223 * the local stream list to prevent the stream from being both local and
224 * global.
225 */
226 stream->globally_visible = 1;
227 cds_list_del(&stream->send_node);
228
229 ret = lttng_pipe_write(stream_pipe, &stream, sizeof(stream));
230 if (ret < 0) {
231 ERR("Consumer write %s stream to pipe %d",
232 stream->metadata_flag ? "metadata" : "data",
233 lttng_pipe_get_writefd(stream_pipe));
234 if (stream->metadata_flag) {
235 consumer_del_stream_for_metadata(stream);
236 } else {
237 consumer_del_stream_for_data(stream);
238 }
239 goto error;
240 }
241
242 error:
243 return ret;
244 }
245
246 static
247 int get_stream_shm_path(char *stream_shm_path, const char *shm_path, int cpu)
248 {
249 char cpu_nr[INT_MAX_STR_LEN]; /* int max len */
250 int ret;
251
252 strncpy(stream_shm_path, shm_path, PATH_MAX);
253 stream_shm_path[PATH_MAX - 1] = '\0';
254 ret = snprintf(cpu_nr, INT_MAX_STR_LEN, "%i", cpu);
255 if (ret < 0) {
256 PERROR("snprintf");
257 goto end;
258 }
259 strncat(stream_shm_path, cpu_nr,
260 PATH_MAX - strlen(stream_shm_path) - 1);
261 ret = 0;
262 end:
263 return ret;
264 }
265
266 /*
267 * Create streams for the given channel using liblttng-ust-ctl.
268 *
269 * Return 0 on success else a negative value.
270 */
271 static int create_ust_streams(struct lttng_consumer_channel *channel,
272 struct lttng_consumer_local_data *ctx,
273 uint64_t trace_archive_id)
274 {
275 int ret, cpu = 0;
276 struct ustctl_consumer_stream *ustream;
277 struct lttng_consumer_stream *stream;
278
279 assert(channel);
280 assert(ctx);
281
282 /*
283 * While a stream is available from ustctl. When NULL is returned, we've
284 * reached the end of the possible stream for the channel.
285 */
286 while ((ustream = ustctl_create_stream(channel->uchan, cpu))) {
287 int wait_fd;
288 int ust_metadata_pipe[2];
289
290 health_code_update();
291
292 if (channel->type == CONSUMER_CHANNEL_TYPE_METADATA && channel->monitor) {
293 ret = utils_create_pipe_cloexec_nonblock(ust_metadata_pipe);
294 if (ret < 0) {
295 ERR("Create ust metadata poll pipe");
296 goto error;
297 }
298 wait_fd = ust_metadata_pipe[0];
299 } else {
300 wait_fd = ustctl_stream_get_wait_fd(ustream);
301 }
302
303 /* Allocate consumer stream object. */
304 stream = allocate_stream(cpu, wait_fd, channel, ctx, &ret,
305 trace_archive_id);
306 if (!stream) {
307 goto error_alloc;
308 }
309 stream->ustream = ustream;
310 /*
311 * Store it so we can save multiple function calls afterwards since
312 * this value is used heavily in the stream threads. This is UST
313 * specific so this is why it's done after allocation.
314 */
315 stream->wait_fd = wait_fd;
316
317 /*
318 * Increment channel refcount since the channel reference has now been
319 * assigned in the allocation process above.
320 */
321 if (stream->chan->monitor) {
322 uatomic_inc(&stream->chan->refcount);
323 }
324
325 /*
326 * Order is important this is why a list is used. On error, the caller
327 * should clean this list.
328 */
329 cds_list_add_tail(&stream->send_node, &channel->streams.head);
330
331 ret = ustctl_get_max_subbuf_size(stream->ustream,
332 &stream->max_sb_size);
333 if (ret < 0) {
334 ERR("ustctl_get_max_subbuf_size failed for stream %s",
335 stream->name);
336 goto error;
337 }
338
339 /* Do actions once stream has been received. */
340 if (ctx->on_recv_stream) {
341 ret = ctx->on_recv_stream(stream);
342 if (ret < 0) {
343 goto error;
344 }
345 }
346
347 DBG("UST consumer add stream %s (key: %" PRIu64 ") with relayd id %" PRIu64,
348 stream->name, stream->key, stream->relayd_stream_id);
349
350 /* Set next CPU stream. */
351 channel->streams.count = ++cpu;
352
353 /* Keep stream reference when creating metadata. */
354 if (channel->type == CONSUMER_CHANNEL_TYPE_METADATA) {
355 channel->metadata_stream = stream;
356 if (channel->monitor) {
357 /* Set metadata poll pipe if we created one */
358 memcpy(stream->ust_metadata_poll_pipe,
359 ust_metadata_pipe,
360 sizeof(ust_metadata_pipe));
361 }
362 }
363 }
364
365 return 0;
366
367 error:
368 error_alloc:
369 return ret;
370 }
371
372 /*
373 * create_posix_shm is never called concurrently within a process.
374 */
375 static
376 int create_posix_shm(void)
377 {
378 char tmp_name[NAME_MAX];
379 int shmfd, ret;
380
381 ret = snprintf(tmp_name, NAME_MAX, "/ust-shm-consumer-%d", getpid());
382 if (ret < 0) {
383 PERROR("snprintf");
384 return -1;
385 }
386 /*
387 * Allocate shm, and immediately unlink its shm oject, keeping
388 * only the file descriptor as a reference to the object.
389 * We specifically do _not_ use the / at the beginning of the
390 * pathname so that some OS implementations can keep it local to
391 * the process (POSIX leaves this implementation-defined).
392 */
393 shmfd = shm_open(tmp_name, O_CREAT | O_EXCL | O_RDWR, 0700);
394 if (shmfd < 0) {
395 PERROR("shm_open");
396 goto error_shm_open;
397 }
398 ret = shm_unlink(tmp_name);
399 if (ret < 0 && errno != ENOENT) {
400 PERROR("shm_unlink");
401 goto error_shm_release;
402 }
403 return shmfd;
404
405 error_shm_release:
406 ret = close(shmfd);
407 if (ret) {
408 PERROR("close");
409 }
410 error_shm_open:
411 return -1;
412 }
413
414 static int open_ust_stream_fd(struct lttng_consumer_channel *channel, int cpu)
415 {
416 char shm_path[PATH_MAX];
417 int ret;
418
419 if (!channel->shm_path[0]) {
420 return create_posix_shm();
421 }
422 ret = get_stream_shm_path(shm_path, channel->shm_path, cpu);
423 if (ret) {
424 goto error_shm_path;
425 }
426 return run_as_open(shm_path,
427 O_RDWR | O_CREAT | O_EXCL, S_IRUSR | S_IWUSR,
428 channel->uid, channel->gid);
429
430 error_shm_path:
431 return -1;
432 }
433
434 /*
435 * Create an UST channel with the given attributes and send it to the session
436 * daemon using the ust ctl API.
437 *
438 * Return 0 on success or else a negative value.
439 */
440 static int create_ust_channel(struct lttng_consumer_channel *channel,
441 struct ustctl_consumer_channel_attr *attr,
442 struct ustctl_consumer_channel **ust_chanp)
443 {
444 int ret, nr_stream_fds, i, j;
445 int *stream_fds;
446 struct ustctl_consumer_channel *ust_channel;
447
448 assert(channel);
449 assert(attr);
450 assert(ust_chanp);
451
452 DBG3("Creating channel to ustctl with attr: [overwrite: %d, "
453 "subbuf_size: %" PRIu64 ", num_subbuf: %" PRIu64 ", "
454 "switch_timer_interval: %u, read_timer_interval: %u, "
455 "output: %d, type: %d", attr->overwrite, attr->subbuf_size,
456 attr->num_subbuf, attr->switch_timer_interval,
457 attr->read_timer_interval, attr->output, attr->type);
458
459 if (channel->type == CONSUMER_CHANNEL_TYPE_METADATA)
460 nr_stream_fds = 1;
461 else
462 nr_stream_fds = ustctl_get_nr_stream_per_channel();
463 stream_fds = zmalloc(nr_stream_fds * sizeof(*stream_fds));
464 if (!stream_fds) {
465 ret = -1;
466 goto error_alloc;
467 }
468 for (i = 0; i < nr_stream_fds; i++) {
469 stream_fds[i] = open_ust_stream_fd(channel, i);
470 if (stream_fds[i] < 0) {
471 ret = -1;
472 goto error_open;
473 }
474 }
475 ust_channel = ustctl_create_channel(attr, stream_fds, nr_stream_fds);
476 if (!ust_channel) {
477 ret = -1;
478 goto error_create;
479 }
480 channel->nr_stream_fds = nr_stream_fds;
481 channel->stream_fds = stream_fds;
482 *ust_chanp = ust_channel;
483
484 return 0;
485
486 error_create:
487 error_open:
488 for (j = i - 1; j >= 0; j--) {
489 int closeret;
490
491 closeret = close(stream_fds[j]);
492 if (closeret) {
493 PERROR("close");
494 }
495 if (channel->shm_path[0]) {
496 char shm_path[PATH_MAX];
497
498 closeret = get_stream_shm_path(shm_path,
499 channel->shm_path, j);
500 if (closeret) {
501 ERR("Cannot get stream shm path");
502 }
503 closeret = run_as_unlink(shm_path,
504 channel->uid, channel->gid);
505 if (closeret) {
506 PERROR("unlink %s", shm_path);
507 }
508 }
509 }
510 /* Try to rmdir all directories under shm_path root. */
511 if (channel->root_shm_path[0]) {
512 (void) run_as_rmdir_recursive(channel->root_shm_path,
513 channel->uid, channel->gid);
514 }
515 free(stream_fds);
516 error_alloc:
517 return ret;
518 }
519
520 /*
521 * Send a single given stream to the session daemon using the sock.
522 *
523 * Return 0 on success else a negative value.
524 */
525 static int send_sessiond_stream(int sock, struct lttng_consumer_stream *stream)
526 {
527 int ret;
528
529 assert(stream);
530 assert(sock >= 0);
531
532 DBG("UST consumer sending stream %" PRIu64 " to sessiond", stream->key);
533
534 /* Send stream to session daemon. */
535 ret = ustctl_send_stream_to_sessiond(sock, stream->ustream);
536 if (ret < 0) {
537 goto error;
538 }
539
540 error:
541 return ret;
542 }
543
544 /*
545 * Send channel to sessiond and relayd if applicable.
546 *
547 * Return 0 on success or else a negative value.
548 */
549 static int send_channel_to_sessiond_and_relayd(int sock,
550 struct lttng_consumer_channel *channel,
551 struct lttng_consumer_local_data *ctx, int *relayd_error)
552 {
553 int ret, ret_code = LTTCOMM_CONSUMERD_SUCCESS;
554 struct lttng_consumer_stream *stream;
555 uint64_t net_seq_idx = -1ULL;
556
557 assert(channel);
558 assert(ctx);
559 assert(sock >= 0);
560
561 DBG("UST consumer sending channel %s to sessiond", channel->name);
562
563 if (channel->relayd_id != (uint64_t) -1ULL) {
564 cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
565
566 health_code_update();
567
568 /* Try to send the stream to the relayd if one is available. */
569 DBG("Sending stream %" PRIu64 " of channel \"%s\" to relayd",
570 stream->key, channel->name);
571 ret = consumer_send_relayd_stream(stream, stream->chan->pathname);
572 if (ret < 0) {
573 /*
574 * Flag that the relayd was the problem here probably due to a
575 * communicaton error on the socket.
576 */
577 if (relayd_error) {
578 *relayd_error = 1;
579 }
580 ret_code = LTTCOMM_CONSUMERD_RELAYD_FAIL;
581 }
582 if (net_seq_idx == -1ULL) {
583 net_seq_idx = stream->net_seq_idx;
584 }
585 }
586 }
587
588 /* Inform sessiond that we are about to send channel and streams. */
589 ret = consumer_send_status_msg(sock, ret_code);
590 if (ret < 0 || ret_code != LTTCOMM_CONSUMERD_SUCCESS) {
591 /*
592 * Either the session daemon is not responding or the relayd died so we
593 * stop now.
594 */
595 goto error;
596 }
597
598 /* Send channel to sessiond. */
599 ret = ustctl_send_channel_to_sessiond(sock, channel->uchan);
600 if (ret < 0) {
601 goto error;
602 }
603
604 ret = ustctl_channel_close_wakeup_fd(channel->uchan);
605 if (ret < 0) {
606 goto error;
607 }
608
609 /* The channel was sent successfully to the sessiond at this point. */
610 cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
611
612 health_code_update();
613
614 /* Send stream to session daemon. */
615 ret = send_sessiond_stream(sock, stream);
616 if (ret < 0) {
617 goto error;
618 }
619 }
620
621 /* Tell sessiond there is no more stream. */
622 ret = ustctl_send_stream_to_sessiond(sock, NULL);
623 if (ret < 0) {
624 goto error;
625 }
626
627 DBG("UST consumer NULL stream sent to sessiond");
628
629 return 0;
630
631 error:
632 if (ret_code != LTTCOMM_CONSUMERD_SUCCESS) {
633 ret = -1;
634 }
635 return ret;
636 }
637
638 /*
639 * Creates a channel and streams and add the channel it to the channel internal
640 * state. The created stream must ONLY be sent once the GET_CHANNEL command is
641 * received.
642 *
643 * Return 0 on success or else, a negative value is returned and the channel
644 * MUST be destroyed by consumer_del_channel().
645 */
646 static int ask_channel(struct lttng_consumer_local_data *ctx,
647 struct lttng_consumer_channel *channel,
648 struct ustctl_consumer_channel_attr *attr,
649 uint64_t trace_archive_id)
650 {
651 int ret;
652
653 assert(ctx);
654 assert(channel);
655 assert(attr);
656
657 /*
658 * This value is still used by the kernel consumer since for the kernel,
659 * the stream ownership is not IN the consumer so we need to have the
660 * number of left stream that needs to be initialized so we can know when
661 * to delete the channel (see consumer.c).
662 *
663 * As for the user space tracer now, the consumer creates and sends the
664 * stream to the session daemon which only sends them to the application
665 * once every stream of a channel is received making this value useless
666 * because we they will be added to the poll thread before the application
667 * receives them. This ensures that a stream can not hang up during
668 * initilization of a channel.
669 */
670 channel->nb_init_stream_left = 0;
671
672 /* The reply msg status is handled in the following call. */
673 ret = create_ust_channel(channel, attr, &channel->uchan);
674 if (ret < 0) {
675 goto end;
676 }
677
678 channel->wait_fd = ustctl_channel_get_wait_fd(channel->uchan);
679
680 /*
681 * For the snapshots (no monitor), we create the metadata streams
682 * on demand, not during the channel creation.
683 */
684 if (channel->type == CONSUMER_CHANNEL_TYPE_METADATA && !channel->monitor) {
685 ret = 0;
686 goto end;
687 }
688
689 /* Open all streams for this channel. */
690 ret = create_ust_streams(channel, ctx, trace_archive_id);
691 if (ret < 0) {
692 goto end;
693 }
694
695 end:
696 return ret;
697 }
698
699 /*
700 * Send all stream of a channel to the right thread handling it.
701 *
702 * On error, return a negative value else 0 on success.
703 */
704 static int send_streams_to_thread(struct lttng_consumer_channel *channel,
705 struct lttng_consumer_local_data *ctx)
706 {
707 int ret = 0;
708 struct lttng_consumer_stream *stream, *stmp;
709
710 assert(channel);
711 assert(ctx);
712
713 /* Send streams to the corresponding thread. */
714 cds_list_for_each_entry_safe(stream, stmp, &channel->streams.head,
715 send_node) {
716
717 health_code_update();
718
719 /* Sending the stream to the thread. */
720 ret = send_stream_to_thread(stream, ctx);
721 if (ret < 0) {
722 /*
723 * If we are unable to send the stream to the thread, there is
724 * a big problem so just stop everything.
725 */
726 goto error;
727 }
728 }
729
730 error:
731 return ret;
732 }
733
734 /*
735 * Flush channel's streams using the given key to retrieve the channel.
736 *
737 * Return 0 on success else an LTTng error code.
738 */
739 static int flush_channel(uint64_t chan_key)
740 {
741 int ret = 0;
742 struct lttng_consumer_channel *channel;
743 struct lttng_consumer_stream *stream;
744 struct lttng_ht *ht;
745 struct lttng_ht_iter iter;
746
747 DBG("UST consumer flush channel key %" PRIu64, chan_key);
748
749 rcu_read_lock();
750 channel = consumer_find_channel(chan_key);
751 if (!channel) {
752 ERR("UST consumer flush channel %" PRIu64 " not found", chan_key);
753 ret = LTTNG_ERR_UST_CHAN_NOT_FOUND;
754 goto error;
755 }
756
757 ht = consumer_data.stream_per_chan_id_ht;
758
759 /* For each stream of the channel id, flush it. */
760 cds_lfht_for_each_entry_duplicate(ht->ht,
761 ht->hash_fct(&channel->key, lttng_ht_seed), ht->match_fct,
762 &channel->key, &iter.iter, stream, node_channel_id.node) {
763
764 health_code_update();
765
766 pthread_mutex_lock(&stream->lock);
767 if (!stream->quiescent) {
768 ustctl_flush_buffer(stream->ustream, 0);
769 stream->quiescent = true;
770 }
771 pthread_mutex_unlock(&stream->lock);
772 }
773 error:
774 rcu_read_unlock();
775 return ret;
776 }
777
778 /*
779 * Clear quiescent state from channel's streams using the given key to
780 * retrieve the channel.
781 *
782 * Return 0 on success else an LTTng error code.
783 */
784 static int clear_quiescent_channel(uint64_t chan_key)
785 {
786 int ret = 0;
787 struct lttng_consumer_channel *channel;
788 struct lttng_consumer_stream *stream;
789 struct lttng_ht *ht;
790 struct lttng_ht_iter iter;
791
792 DBG("UST consumer clear quiescent channel key %" PRIu64, chan_key);
793
794 rcu_read_lock();
795 channel = consumer_find_channel(chan_key);
796 if (!channel) {
797 ERR("UST consumer clear quiescent channel %" PRIu64 " not found", chan_key);
798 ret = LTTNG_ERR_UST_CHAN_NOT_FOUND;
799 goto error;
800 }
801
802 ht = consumer_data.stream_per_chan_id_ht;
803
804 /* For each stream of the channel id, clear quiescent state. */
805 cds_lfht_for_each_entry_duplicate(ht->ht,
806 ht->hash_fct(&channel->key, lttng_ht_seed), ht->match_fct,
807 &channel->key, &iter.iter, stream, node_channel_id.node) {
808
809 health_code_update();
810
811 pthread_mutex_lock(&stream->lock);
812 stream->quiescent = false;
813 pthread_mutex_unlock(&stream->lock);
814 }
815 error:
816 rcu_read_unlock();
817 return ret;
818 }
819
820 /*
821 * Close metadata stream wakeup_fd using the given key to retrieve the channel.
822 * RCU read side lock MUST be acquired before calling this function.
823 *
824 * Return 0 on success else an LTTng error code.
825 */
826 static int close_metadata(uint64_t chan_key)
827 {
828 int ret = 0;
829 struct lttng_consumer_channel *channel;
830 unsigned int channel_monitor;
831
832 DBG("UST consumer close metadata key %" PRIu64, chan_key);
833
834 channel = consumer_find_channel(chan_key);
835 if (!channel) {
836 /*
837 * This is possible if the metadata thread has issue a delete because
838 * the endpoint point of the stream hung up. There is no way the
839 * session daemon can know about it thus use a DBG instead of an actual
840 * error.
841 */
842 DBG("UST consumer close metadata %" PRIu64 " not found", chan_key);
843 ret = LTTNG_ERR_UST_CHAN_NOT_FOUND;
844 goto error;
845 }
846
847 pthread_mutex_lock(&consumer_data.lock);
848 pthread_mutex_lock(&channel->lock);
849 channel_monitor = channel->monitor;
850 if (cds_lfht_is_node_deleted(&channel->node.node)) {
851 goto error_unlock;
852 }
853
854 lttng_ustconsumer_close_metadata(channel);
855 pthread_mutex_unlock(&channel->lock);
856 pthread_mutex_unlock(&consumer_data.lock);
857
858 /*
859 * The ownership of a metadata channel depends on the type of
860 * session to which it belongs. In effect, the monitor flag is checked
861 * to determine if this metadata channel is in "snapshot" mode or not.
862 *
863 * In the non-snapshot case, the metadata channel is created along with
864 * a single stream which will remain present until the metadata channel
865 * is destroyed (on the destruction of its session). In this case, the
866 * metadata stream in "monitored" by the metadata poll thread and holds
867 * the ownership of its channel.
868 *
869 * Closing the metadata will cause the metadata stream's "metadata poll
870 * pipe" to be closed. Closing this pipe will wake-up the metadata poll
871 * thread which will teardown the metadata stream which, in return,
872 * deletes the metadata channel.
873 *
874 * In the snapshot case, the metadata stream is created and destroyed
875 * on every snapshot record. Since the channel doesn't have an owner
876 * other than the session daemon, it is safe to destroy it immediately
877 * on reception of the CLOSE_METADATA command.
878 */
879 if (!channel_monitor) {
880 /*
881 * The channel and consumer_data locks must be
882 * released before this call since consumer_del_channel
883 * re-acquires the channel and consumer_data locks to teardown
884 * the channel and queue its reclamation by the "call_rcu"
885 * worker thread.
886 */
887 consumer_del_channel(channel);
888 }
889
890 return ret;
891 error_unlock:
892 pthread_mutex_unlock(&channel->lock);
893 pthread_mutex_unlock(&consumer_data.lock);
894 error:
895 return ret;
896 }
897
898 /*
899 * RCU read side lock MUST be acquired before calling this function.
900 *
901 * Return 0 on success else an LTTng error code.
902 */
903 static int setup_metadata(struct lttng_consumer_local_data *ctx, uint64_t key)
904 {
905 int ret;
906 struct lttng_consumer_channel *metadata;
907
908 DBG("UST consumer setup metadata key %" PRIu64, key);
909
910 metadata = consumer_find_channel(key);
911 if (!metadata) {
912 ERR("UST consumer push metadata %" PRIu64 " not found", key);
913 ret = LTTNG_ERR_UST_CHAN_NOT_FOUND;
914 goto end;
915 }
916
917 /*
918 * In no monitor mode, the metadata channel has no stream(s) so skip the
919 * ownership transfer to the metadata thread.
920 */
921 if (!metadata->monitor) {
922 DBG("Metadata channel in no monitor");
923 ret = 0;
924 goto end;
925 }
926
927 /*
928 * Send metadata stream to relayd if one available. Availability is
929 * known if the stream is still in the list of the channel.
930 */
931 if (cds_list_empty(&metadata->streams.head)) {
932 ERR("Metadata channel key %" PRIu64 ", no stream available.", key);
933 ret = LTTCOMM_CONSUMERD_ERROR_METADATA;
934 goto error_no_stream;
935 }
936
937 /* Send metadata stream to relayd if needed. */
938 if (metadata->metadata_stream->net_seq_idx != (uint64_t) -1ULL) {
939 ret = consumer_send_relayd_stream(metadata->metadata_stream,
940 metadata->pathname);
941 if (ret < 0) {
942 ret = LTTCOMM_CONSUMERD_ERROR_METADATA;
943 goto error;
944 }
945 ret = consumer_send_relayd_streams_sent(
946 metadata->metadata_stream->net_seq_idx);
947 if (ret < 0) {
948 ret = LTTCOMM_CONSUMERD_RELAYD_FAIL;
949 goto error;
950 }
951 }
952
953 /*
954 * Ownership of metadata stream is passed along. Freeing is handled by
955 * the callee.
956 */
957 ret = send_streams_to_thread(metadata, ctx);
958 if (ret < 0) {
959 /*
960 * If we are unable to send the stream to the thread, there is
961 * a big problem so just stop everything.
962 */
963 ret = LTTCOMM_CONSUMERD_FATAL;
964 goto send_streams_error;
965 }
966 /* List MUST be empty after or else it could be reused. */
967 assert(cds_list_empty(&metadata->streams.head));
968
969 ret = 0;
970 goto end;
971
972 error:
973 /*
974 * Delete metadata channel on error. At this point, the metadata stream can
975 * NOT be monitored by the metadata thread thus having the guarantee that
976 * the stream is still in the local stream list of the channel. This call
977 * will make sure to clean that list.
978 */
979 consumer_stream_destroy(metadata->metadata_stream, NULL);
980 cds_list_del(&metadata->metadata_stream->send_node);
981 metadata->metadata_stream = NULL;
982 send_streams_error:
983 error_no_stream:
984 end:
985 return ret;
986 }
987
988 /*
989 * Snapshot the whole metadata.
990 * RCU read-side lock must be held across this function to ensure existence of
991 * metadata_channel.
992 *
993 * Returns 0 on success, < 0 on error
994 */
995 static int snapshot_metadata(struct lttng_consumer_channel *metadata_channel,
996 uint64_t key, char *path, uint64_t relayd_id,
997 struct lttng_consumer_local_data *ctx,
998 uint64_t trace_archive_id)
999 {
1000 int ret = 0;
1001 struct lttng_consumer_stream *metadata_stream;
1002
1003 assert(path);
1004 assert(ctx);
1005
1006 DBG("UST consumer snapshot metadata with key %" PRIu64 " at path %s",
1007 key, path);
1008
1009 rcu_read_lock();
1010
1011 assert(!metadata_channel->monitor);
1012
1013 health_code_update();
1014
1015 /*
1016 * Ask the sessiond if we have new metadata waiting and update the
1017 * consumer metadata cache.
1018 */
1019 ret = lttng_ustconsumer_request_metadata(ctx, metadata_channel, 0, 1);
1020 if (ret < 0) {
1021 goto error;
1022 }
1023
1024 health_code_update();
1025
1026 /*
1027 * The metadata stream is NOT created in no monitor mode when the channel
1028 * is created on a sessiond ask channel command.
1029 */
1030 ret = create_ust_streams(metadata_channel, ctx, trace_archive_id);
1031 if (ret < 0) {
1032 goto error;
1033 }
1034
1035 metadata_stream = metadata_channel->metadata_stream;
1036 assert(metadata_stream);
1037
1038 if (relayd_id != (uint64_t) -1ULL) {
1039 metadata_stream->net_seq_idx = relayd_id;
1040 ret = consumer_send_relayd_stream(metadata_stream, path);
1041 if (ret < 0) {
1042 goto error_stream;
1043 }
1044 } else {
1045 ret = utils_create_stream_file(path, metadata_stream->name,
1046 metadata_stream->chan->tracefile_size,
1047 metadata_stream->tracefile_count_current,
1048 metadata_stream->uid, metadata_stream->gid, NULL);
1049 if (ret < 0) {
1050 goto error_stream;
1051 }
1052 metadata_stream->out_fd = ret;
1053 metadata_stream->tracefile_size_current = 0;
1054 }
1055
1056 do {
1057 health_code_update();
1058
1059 ret = lttng_consumer_read_subbuffer(metadata_stream, ctx);
1060 if (ret < 0) {
1061 goto error_stream;
1062 }
1063 } while (ret > 0);
1064
1065 error_stream:
1066 /*
1067 * Clean up the stream completly because the next snapshot will use a new
1068 * metadata stream.
1069 */
1070 consumer_stream_destroy(metadata_stream, NULL);
1071 cds_list_del(&metadata_stream->send_node);
1072 metadata_channel->metadata_stream = NULL;
1073
1074 error:
1075 rcu_read_unlock();
1076 return ret;
1077 }
1078
1079 /*
1080 * Take a snapshot of all the stream of a channel.
1081 * RCU read-side lock must be held across this function to ensure existence of
1082 * channel.
1083 *
1084 * Returns 0 on success, < 0 on error
1085 */
1086 static int snapshot_channel(struct lttng_consumer_channel *channel,
1087 uint64_t key, char *path, uint64_t relayd_id,
1088 uint64_t nb_packets_per_stream,
1089 struct lttng_consumer_local_data *ctx)
1090 {
1091 int ret;
1092 unsigned use_relayd = 0;
1093 unsigned long consumed_pos, produced_pos;
1094 struct lttng_consumer_stream *stream;
1095
1096 assert(path);
1097 assert(ctx);
1098
1099 rcu_read_lock();
1100
1101 if (relayd_id != (uint64_t) -1ULL) {
1102 use_relayd = 1;
1103 }
1104
1105 assert(!channel->monitor);
1106 DBG("UST consumer snapshot channel %" PRIu64, key);
1107
1108 cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
1109 health_code_update();
1110
1111 /* Lock stream because we are about to change its state. */
1112 pthread_mutex_lock(&stream->lock);
1113 stream->net_seq_idx = relayd_id;
1114
1115 if (use_relayd) {
1116 ret = consumer_send_relayd_stream(stream, path);
1117 if (ret < 0) {
1118 goto error_unlock;
1119 }
1120 } else {
1121 ret = utils_create_stream_file(path, stream->name,
1122 stream->chan->tracefile_size,
1123 stream->tracefile_count_current,
1124 stream->uid, stream->gid, NULL);
1125 if (ret < 0) {
1126 goto error_unlock;
1127 }
1128 stream->out_fd = ret;
1129 stream->tracefile_size_current = 0;
1130
1131 DBG("UST consumer snapshot stream %s/%s (%" PRIu64 ")", path,
1132 stream->name, stream->key);
1133 }
1134
1135 /*
1136 * If tracing is active, we want to perform a "full" buffer flush.
1137 * Else, if quiescent, it has already been done by the prior stop.
1138 */
1139 if (!stream->quiescent) {
1140 ustctl_flush_buffer(stream->ustream, 0);
1141 }
1142
1143 ret = lttng_ustconsumer_take_snapshot(stream);
1144 if (ret < 0) {
1145 ERR("Taking UST snapshot");
1146 goto error_unlock;
1147 }
1148
1149 ret = lttng_ustconsumer_get_produced_snapshot(stream, &produced_pos);
1150 if (ret < 0) {
1151 ERR("Produced UST snapshot position");
1152 goto error_unlock;
1153 }
1154
1155 ret = lttng_ustconsumer_get_consumed_snapshot(stream, &consumed_pos);
1156 if (ret < 0) {
1157 ERR("Consumerd UST snapshot position");
1158 goto error_unlock;
1159 }
1160
1161 /*
1162 * The original value is sent back if max stream size is larger than
1163 * the possible size of the snapshot. Also, we assume that the session
1164 * daemon should never send a maximum stream size that is lower than
1165 * subbuffer size.
1166 */
1167 consumed_pos = consumer_get_consume_start_pos(consumed_pos,
1168 produced_pos, nb_packets_per_stream,
1169 stream->max_sb_size);
1170
1171 while (consumed_pos < produced_pos) {
1172 ssize_t read_len;
1173 unsigned long len, padded_len;
1174
1175 health_code_update();
1176
1177 DBG("UST consumer taking snapshot at pos %lu", consumed_pos);
1178
1179 ret = ustctl_get_subbuf(stream->ustream, &consumed_pos);
1180 if (ret < 0) {
1181 if (ret != -EAGAIN) {
1182 PERROR("ustctl_get_subbuf snapshot");
1183 goto error_close_stream;
1184 }
1185 DBG("UST consumer get subbuf failed. Skipping it.");
1186 consumed_pos += stream->max_sb_size;
1187 stream->chan->lost_packets++;
1188 continue;
1189 }
1190
1191 ret = ustctl_get_subbuf_size(stream->ustream, &len);
1192 if (ret < 0) {
1193 ERR("Snapshot ustctl_get_subbuf_size");
1194 goto error_put_subbuf;
1195 }
1196
1197 ret = ustctl_get_padded_subbuf_size(stream->ustream, &padded_len);
1198 if (ret < 0) {
1199 ERR("Snapshot ustctl_get_padded_subbuf_size");
1200 goto error_put_subbuf;
1201 }
1202
1203 read_len = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, len,
1204 padded_len - len, NULL);
1205 if (use_relayd) {
1206 if (read_len != len) {
1207 ret = -EPERM;
1208 goto error_put_subbuf;
1209 }
1210 } else {
1211 if (read_len != padded_len) {
1212 ret = -EPERM;
1213 goto error_put_subbuf;
1214 }
1215 }
1216
1217 ret = ustctl_put_subbuf(stream->ustream);
1218 if (ret < 0) {
1219 ERR("Snapshot ustctl_put_subbuf");
1220 goto error_close_stream;
1221 }
1222 consumed_pos += stream->max_sb_size;
1223 }
1224
1225 /* Simply close the stream so we can use it on the next snapshot. */
1226 consumer_stream_close(stream);
1227 pthread_mutex_unlock(&stream->lock);
1228 }
1229
1230 rcu_read_unlock();
1231 return 0;
1232
1233 error_put_subbuf:
1234 if (ustctl_put_subbuf(stream->ustream) < 0) {
1235 ERR("Snapshot ustctl_put_subbuf");
1236 }
1237 error_close_stream:
1238 consumer_stream_close(stream);
1239 error_unlock:
1240 pthread_mutex_unlock(&stream->lock);
1241 rcu_read_unlock();
1242 return ret;
1243 }
1244
1245 /*
1246 * Receive the metadata updates from the sessiond. Supports receiving
1247 * overlapping metadata, but is needs to always belong to a contiguous
1248 * range starting from 0.
1249 * Be careful about the locks held when calling this function: it needs
1250 * the metadata cache flush to concurrently progress in order to
1251 * complete.
1252 */
1253 int lttng_ustconsumer_recv_metadata(int sock, uint64_t key, uint64_t offset,
1254 uint64_t len, uint64_t version,
1255 struct lttng_consumer_channel *channel, int timer, int wait)
1256 {
1257 int ret, ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1258 char *metadata_str;
1259
1260 DBG("UST consumer push metadata key %" PRIu64 " of len %" PRIu64, key, len);
1261
1262 metadata_str = zmalloc(len * sizeof(char));
1263 if (!metadata_str) {
1264 PERROR("zmalloc metadata string");
1265 ret_code = LTTCOMM_CONSUMERD_ENOMEM;
1266 goto end;
1267 }
1268
1269 health_code_update();
1270
1271 /* Receive metadata string. */
1272 ret = lttcomm_recv_unix_sock(sock, metadata_str, len);
1273 if (ret < 0) {
1274 /* Session daemon is dead so return gracefully. */
1275 ret_code = ret;
1276 goto end_free;
1277 }
1278
1279 health_code_update();
1280
1281 pthread_mutex_lock(&channel->metadata_cache->lock);
1282 ret = consumer_metadata_cache_write(channel, offset, len, version,
1283 metadata_str);
1284 if (ret < 0) {
1285 /* Unable to handle metadata. Notify session daemon. */
1286 ret_code = LTTCOMM_CONSUMERD_ERROR_METADATA;
1287 /*
1288 * Skip metadata flush on write error since the offset and len might
1289 * not have been updated which could create an infinite loop below when
1290 * waiting for the metadata cache to be flushed.
1291 */
1292 pthread_mutex_unlock(&channel->metadata_cache->lock);
1293 goto end_free;
1294 }
1295 pthread_mutex_unlock(&channel->metadata_cache->lock);
1296
1297 if (!wait) {
1298 goto end_free;
1299 }
1300 while (consumer_metadata_cache_flushed(channel, offset + len, timer)) {
1301 DBG("Waiting for metadata to be flushed");
1302
1303 health_code_update();
1304
1305 usleep(DEFAULT_METADATA_AVAILABILITY_WAIT_TIME);
1306 }
1307
1308 end_free:
1309 free(metadata_str);
1310 end:
1311 return ret_code;
1312 }
1313
1314 /*
1315 * Receive command from session daemon and process it.
1316 *
1317 * Return 1 on success else a negative value or 0.
1318 */
1319 int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
1320 int sock, struct pollfd *consumer_sockpoll)
1321 {
1322 ssize_t ret;
1323 enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1324 struct lttcomm_consumer_msg msg;
1325 struct lttng_consumer_channel *channel = NULL;
1326
1327 health_code_update();
1328
1329 ret = lttcomm_recv_unix_sock(sock, &msg, sizeof(msg));
1330 if (ret != sizeof(msg)) {
1331 DBG("Consumer received unexpected message size %zd (expects %zu)",
1332 ret, sizeof(msg));
1333 /*
1334 * The ret value might 0 meaning an orderly shutdown but this is ok
1335 * since the caller handles this.
1336 */
1337 if (ret > 0) {
1338 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD);
1339 ret = -1;
1340 }
1341 return ret;
1342 }
1343
1344 health_code_update();
1345
1346 /* deprecated */
1347 assert(msg.cmd_type != LTTNG_CONSUMER_STOP);
1348
1349 health_code_update();
1350
1351 /* relayd needs RCU read-side lock */
1352 rcu_read_lock();
1353
1354 switch (msg.cmd_type) {
1355 case LTTNG_CONSUMER_ADD_RELAYD_SOCKET:
1356 {
1357 /* Session daemon status message are handled in the following call. */
1358 consumer_add_relayd_socket(msg.u.relayd_sock.net_index,
1359 msg.u.relayd_sock.type, ctx, sock, consumer_sockpoll,
1360 &msg.u.relayd_sock.sock, msg.u.relayd_sock.session_id,
1361 msg.u.relayd_sock.relayd_session_id);
1362 goto end_nosignal;
1363 }
1364 case LTTNG_CONSUMER_DESTROY_RELAYD:
1365 {
1366 uint64_t index = msg.u.destroy_relayd.net_seq_idx;
1367 struct consumer_relayd_sock_pair *relayd;
1368
1369 DBG("UST consumer destroying relayd %" PRIu64, index);
1370
1371 /* Get relayd reference if exists. */
1372 relayd = consumer_find_relayd(index);
1373 if (relayd == NULL) {
1374 DBG("Unable to find relayd %" PRIu64, index);
1375 ret_code = LTTCOMM_CONSUMERD_RELAYD_FAIL;
1376 }
1377
1378 /*
1379 * Each relayd socket pair has a refcount of stream attached to it
1380 * which tells if the relayd is still active or not depending on the
1381 * refcount value.
1382 *
1383 * This will set the destroy flag of the relayd object and destroy it
1384 * if the refcount reaches zero when called.
1385 *
1386 * The destroy can happen either here or when a stream fd hangs up.
1387 */
1388 if (relayd) {
1389 consumer_flag_relayd_for_destroy(relayd);
1390 }
1391
1392 goto end_msg_sessiond;
1393 }
1394 case LTTNG_CONSUMER_UPDATE_STREAM:
1395 {
1396 rcu_read_unlock();
1397 return -ENOSYS;
1398 }
1399 case LTTNG_CONSUMER_DATA_PENDING:
1400 {
1401 int ret, is_data_pending;
1402 uint64_t id = msg.u.data_pending.session_id;
1403
1404 DBG("UST consumer data pending command for id %" PRIu64, id);
1405
1406 is_data_pending = consumer_data_pending(id);
1407
1408 /* Send back returned value to session daemon */
1409 ret = lttcomm_send_unix_sock(sock, &is_data_pending,
1410 sizeof(is_data_pending));
1411 if (ret < 0) {
1412 DBG("Error when sending the data pending ret code: %d", ret);
1413 goto error_fatal;
1414 }
1415
1416 /*
1417 * No need to send back a status message since the data pending
1418 * returned value is the response.
1419 */
1420 break;
1421 }
1422 case LTTNG_CONSUMER_ASK_CHANNEL_CREATION:
1423 {
1424 int ret;
1425 struct ustctl_consumer_channel_attr attr;
1426
1427 /* Create a plain object and reserve a channel key. */
1428 channel = allocate_channel(msg.u.ask_channel.session_id,
1429 msg.u.ask_channel.pathname, msg.u.ask_channel.name,
1430 msg.u.ask_channel.uid, msg.u.ask_channel.gid,
1431 msg.u.ask_channel.relayd_id, msg.u.ask_channel.key,
1432 (enum lttng_event_output) msg.u.ask_channel.output,
1433 msg.u.ask_channel.tracefile_size,
1434 msg.u.ask_channel.tracefile_count,
1435 msg.u.ask_channel.session_id_per_pid,
1436 msg.u.ask_channel.monitor,
1437 msg.u.ask_channel.live_timer_interval,
1438 msg.u.ask_channel.root_shm_path,
1439 msg.u.ask_channel.shm_path);
1440 if (!channel) {
1441 goto end_channel_error;
1442 }
1443
1444 /*
1445 * Assign UST application UID to the channel. This value is ignored for
1446 * per PID buffers. This is specific to UST thus setting this after the
1447 * allocation.
1448 */
1449 channel->ust_app_uid = msg.u.ask_channel.ust_app_uid;
1450
1451 /* Build channel attributes from received message. */
1452 attr.subbuf_size = msg.u.ask_channel.subbuf_size;
1453 attr.num_subbuf = msg.u.ask_channel.num_subbuf;
1454 attr.overwrite = msg.u.ask_channel.overwrite;
1455 attr.switch_timer_interval = msg.u.ask_channel.switch_timer_interval;
1456 attr.read_timer_interval = msg.u.ask_channel.read_timer_interval;
1457 attr.chan_id = msg.u.ask_channel.chan_id;
1458 memcpy(attr.uuid, msg.u.ask_channel.uuid, sizeof(attr.uuid));
1459 attr.blocking_timeout= msg.u.ask_channel.blocking_timeout;
1460
1461 /* Match channel buffer type to the UST abi. */
1462 switch (msg.u.ask_channel.output) {
1463 case LTTNG_EVENT_MMAP:
1464 default:
1465 attr.output = LTTNG_UST_MMAP;
1466 break;
1467 }
1468
1469 /* Translate and save channel type. */
1470 switch (msg.u.ask_channel.type) {
1471 case LTTNG_UST_CHAN_PER_CPU:
1472 channel->type = CONSUMER_CHANNEL_TYPE_DATA;
1473 attr.type = LTTNG_UST_CHAN_PER_CPU;
1474 /*
1475 * Set refcount to 1 for owner. Below, we will
1476 * pass ownership to the
1477 * consumer_thread_channel_poll() thread.
1478 */
1479 channel->refcount = 1;
1480 break;
1481 case LTTNG_UST_CHAN_METADATA:
1482 channel->type = CONSUMER_CHANNEL_TYPE_METADATA;
1483 attr.type = LTTNG_UST_CHAN_METADATA;
1484 break;
1485 default:
1486 assert(0);
1487 goto error_fatal;
1488 };
1489
1490 health_code_update();
1491
1492 ret = ask_channel(ctx, channel, &attr,
1493 msg.u.ask_channel.trace_archive_id);
1494 if (ret < 0) {
1495 goto end_channel_error;
1496 }
1497
1498 if (msg.u.ask_channel.type == LTTNG_UST_CHAN_METADATA) {
1499 ret = consumer_metadata_cache_allocate(channel);
1500 if (ret < 0) {
1501 ERR("Allocating metadata cache");
1502 goto end_channel_error;
1503 }
1504 consumer_timer_switch_start(channel, attr.switch_timer_interval);
1505 attr.switch_timer_interval = 0;
1506 } else {
1507 int monitor_start_ret;
1508
1509 consumer_timer_live_start(channel,
1510 msg.u.ask_channel.live_timer_interval);
1511 monitor_start_ret = consumer_timer_monitor_start(
1512 channel,
1513 msg.u.ask_channel.monitor_timer_interval);
1514 if (monitor_start_ret < 0) {
1515 ERR("Starting channel monitoring timer failed");
1516 goto end_channel_error;
1517 }
1518 }
1519
1520 health_code_update();
1521
1522 /*
1523 * Add the channel to the internal state AFTER all streams were created
1524 * and successfully sent to session daemon. This way, all streams must
1525 * be ready before this channel is visible to the threads.
1526 * If add_channel succeeds, ownership of the channel is
1527 * passed to consumer_thread_channel_poll().
1528 */
1529 ret = add_channel(channel, ctx);
1530 if (ret < 0) {
1531 if (msg.u.ask_channel.type == LTTNG_UST_CHAN_METADATA) {
1532 if (channel->switch_timer_enabled == 1) {
1533 consumer_timer_switch_stop(channel);
1534 }
1535 consumer_metadata_cache_destroy(channel);
1536 }
1537 if (channel->live_timer_enabled == 1) {
1538 consumer_timer_live_stop(channel);
1539 }
1540 if (channel->monitor_timer_enabled == 1) {
1541 consumer_timer_monitor_stop(channel);
1542 }
1543 goto end_channel_error;
1544 }
1545
1546 health_code_update();
1547
1548 /*
1549 * Channel and streams are now created. Inform the session daemon that
1550 * everything went well and should wait to receive the channel and
1551 * streams with ustctl API.
1552 */
1553 ret = consumer_send_status_channel(sock, channel);
1554 if (ret < 0) {
1555 /*
1556 * There is probably a problem on the socket.
1557 */
1558 goto error_fatal;
1559 }
1560
1561 break;
1562 }
1563 case LTTNG_CONSUMER_GET_CHANNEL:
1564 {
1565 int ret, relayd_err = 0;
1566 uint64_t key = msg.u.get_channel.key;
1567 struct lttng_consumer_channel *channel;
1568
1569 channel = consumer_find_channel(key);
1570 if (!channel) {
1571 ERR("UST consumer get channel key %" PRIu64 " not found", key);
1572 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
1573 goto end_msg_sessiond;
1574 }
1575
1576 health_code_update();
1577
1578 /* Send the channel to sessiond (and relayd, if applicable). */
1579 ret = send_channel_to_sessiond_and_relayd(sock, channel, ctx,
1580 &relayd_err);
1581 if (ret < 0) {
1582 if (relayd_err) {
1583 /*
1584 * We were unable to send to the relayd the stream so avoid
1585 * sending back a fatal error to the thread since this is OK
1586 * and the consumer can continue its work. The above call
1587 * has sent the error status message to the sessiond.
1588 */
1589 goto end_nosignal;
1590 }
1591 /*
1592 * The communicaton was broken hence there is a bad state between
1593 * the consumer and sessiond so stop everything.
1594 */
1595 goto error_fatal;
1596 }
1597
1598 health_code_update();
1599
1600 /*
1601 * In no monitor mode, the streams ownership is kept inside the channel
1602 * so don't send them to the data thread.
1603 */
1604 if (!channel->monitor) {
1605 goto end_msg_sessiond;
1606 }
1607
1608 ret = send_streams_to_thread(channel, ctx);
1609 if (ret < 0) {
1610 /*
1611 * If we are unable to send the stream to the thread, there is
1612 * a big problem so just stop everything.
1613 */
1614 goto error_fatal;
1615 }
1616 /* List MUST be empty after or else it could be reused. */
1617 assert(cds_list_empty(&channel->streams.head));
1618 goto end_msg_sessiond;
1619 }
1620 case LTTNG_CONSUMER_DESTROY_CHANNEL:
1621 {
1622 uint64_t key = msg.u.destroy_channel.key;
1623
1624 /*
1625 * Only called if streams have not been sent to stream
1626 * manager thread. However, channel has been sent to
1627 * channel manager thread.
1628 */
1629 notify_thread_del_channel(ctx, key);
1630 goto end_msg_sessiond;
1631 }
1632 case LTTNG_CONSUMER_CLOSE_METADATA:
1633 {
1634 int ret;
1635
1636 ret = close_metadata(msg.u.close_metadata.key);
1637 if (ret != 0) {
1638 ret_code = ret;
1639 }
1640
1641 goto end_msg_sessiond;
1642 }
1643 case LTTNG_CONSUMER_FLUSH_CHANNEL:
1644 {
1645 int ret;
1646
1647 ret = flush_channel(msg.u.flush_channel.key);
1648 if (ret != 0) {
1649 ret_code = ret;
1650 }
1651
1652 goto end_msg_sessiond;
1653 }
1654 case LTTNG_CONSUMER_CLEAR_QUIESCENT_CHANNEL:
1655 {
1656 int ret;
1657
1658 ret = clear_quiescent_channel(
1659 msg.u.clear_quiescent_channel.key);
1660 if (ret != 0) {
1661 ret_code = ret;
1662 }
1663
1664 goto end_msg_sessiond;
1665 }
1666 case LTTNG_CONSUMER_PUSH_METADATA:
1667 {
1668 int ret;
1669 uint64_t len = msg.u.push_metadata.len;
1670 uint64_t key = msg.u.push_metadata.key;
1671 uint64_t offset = msg.u.push_metadata.target_offset;
1672 uint64_t version = msg.u.push_metadata.version;
1673 struct lttng_consumer_channel *channel;
1674
1675 DBG("UST consumer push metadata key %" PRIu64 " of len %" PRIu64, key,
1676 len);
1677
1678 channel = consumer_find_channel(key);
1679 if (!channel) {
1680 /*
1681 * This is possible if the metadata creation on the consumer side
1682 * is in flight vis-a-vis a concurrent push metadata from the
1683 * session daemon. Simply return that the channel failed and the
1684 * session daemon will handle that message correctly considering
1685 * that this race is acceptable thus the DBG() statement here.
1686 */
1687 DBG("UST consumer push metadata %" PRIu64 " not found", key);
1688 ret_code = LTTCOMM_CONSUMERD_CHANNEL_FAIL;
1689 goto end_msg_sessiond;
1690 }
1691
1692 health_code_update();
1693
1694 if (!len) {
1695 /*
1696 * There is nothing to receive. We have simply
1697 * checked whether the channel can be found.
1698 */
1699 ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1700 goto end_msg_sessiond;
1701 }
1702
1703 /* Tell session daemon we are ready to receive the metadata. */
1704 ret = consumer_send_status_msg(sock, LTTCOMM_CONSUMERD_SUCCESS);
1705 if (ret < 0) {
1706 /* Somehow, the session daemon is not responding anymore. */
1707 goto error_fatal;
1708 }
1709
1710 health_code_update();
1711
1712 /* Wait for more data. */
1713 health_poll_entry();
1714 ret = lttng_consumer_poll_socket(consumer_sockpoll);
1715 health_poll_exit();
1716 if (ret) {
1717 goto error_fatal;
1718 }
1719
1720 health_code_update();
1721
1722 ret = lttng_ustconsumer_recv_metadata(sock, key, offset,
1723 len, version, channel, 0, 1);
1724 if (ret < 0) {
1725 /* error receiving from sessiond */
1726 goto error_fatal;
1727 } else {
1728 ret_code = ret;
1729 goto end_msg_sessiond;
1730 }
1731 }
1732 case LTTNG_CONSUMER_SETUP_METADATA:
1733 {
1734 int ret;
1735
1736 ret = setup_metadata(ctx, msg.u.setup_metadata.key);
1737 if (ret) {
1738 ret_code = ret;
1739 }
1740 goto end_msg_sessiond;
1741 }
1742 case LTTNG_CONSUMER_SNAPSHOT_CHANNEL:
1743 {
1744 struct lttng_consumer_channel *channel;
1745 uint64_t key = msg.u.snapshot_channel.key;
1746
1747 channel = consumer_find_channel(key);
1748 if (!channel) {
1749 DBG("UST snapshot channel not found for key %" PRIu64, key);
1750 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
1751 } else {
1752 if (msg.u.snapshot_channel.metadata) {
1753 ret = snapshot_metadata(channel, key,
1754 msg.u.snapshot_channel.pathname,
1755 msg.u.snapshot_channel.relayd_id,
1756 ctx,
1757 msg.u.snapshot_channel.trace_archive_id);
1758 if (ret < 0) {
1759 ERR("Snapshot metadata failed");
1760 ret_code = LTTCOMM_CONSUMERD_SNAPSHOT_FAILED;
1761 }
1762 } else {
1763 ret = snapshot_channel(channel, key,
1764 msg.u.snapshot_channel.pathname,
1765 msg.u.snapshot_channel.relayd_id,
1766 msg.u.snapshot_channel.nb_packets_per_stream,
1767 ctx);
1768 if (ret < 0) {
1769 ERR("Snapshot channel failed");
1770 ret_code = LTTCOMM_CONSUMERD_SNAPSHOT_FAILED;
1771 }
1772 }
1773 }
1774 health_code_update();
1775 ret = consumer_send_status_msg(sock, ret_code);
1776 if (ret < 0) {
1777 /* Somehow, the session daemon is not responding anymore. */
1778 goto end_nosignal;
1779 }
1780 health_code_update();
1781 break;
1782 }
1783 case LTTNG_CONSUMER_DISCARDED_EVENTS:
1784 {
1785 int ret = 0;
1786 uint64_t discarded_events;
1787 struct lttng_ht_iter iter;
1788 struct lttng_ht *ht;
1789 struct lttng_consumer_stream *stream;
1790 uint64_t id = msg.u.discarded_events.session_id;
1791 uint64_t key = msg.u.discarded_events.channel_key;
1792
1793 DBG("UST consumer discarded events command for session id %"
1794 PRIu64, id);
1795 rcu_read_lock();
1796 pthread_mutex_lock(&consumer_data.lock);
1797
1798 ht = consumer_data.stream_list_ht;
1799
1800 /*
1801 * We only need a reference to the channel, but they are not
1802 * directly indexed, so we just use the first matching stream
1803 * to extract the information we need, we default to 0 if not
1804 * found (no events are dropped if the channel is not yet in
1805 * use).
1806 */
1807 discarded_events = 0;
1808 cds_lfht_for_each_entry_duplicate(ht->ht,
1809 ht->hash_fct(&id, lttng_ht_seed),
1810 ht->match_fct, &id,
1811 &iter.iter, stream, node_session_id.node) {
1812 if (stream->chan->key == key) {
1813 discarded_events = stream->chan->discarded_events;
1814 break;
1815 }
1816 }
1817 pthread_mutex_unlock(&consumer_data.lock);
1818 rcu_read_unlock();
1819
1820 DBG("UST consumer discarded events command for session id %"
1821 PRIu64 ", channel key %" PRIu64, id, key);
1822
1823 health_code_update();
1824
1825 /* Send back returned value to session daemon */
1826 ret = lttcomm_send_unix_sock(sock, &discarded_events, sizeof(discarded_events));
1827 if (ret < 0) {
1828 PERROR("send discarded events");
1829 goto error_fatal;
1830 }
1831
1832 break;
1833 }
1834 case LTTNG_CONSUMER_LOST_PACKETS:
1835 {
1836 int ret;
1837 uint64_t lost_packets;
1838 struct lttng_ht_iter iter;
1839 struct lttng_ht *ht;
1840 struct lttng_consumer_stream *stream;
1841 uint64_t id = msg.u.lost_packets.session_id;
1842 uint64_t key = msg.u.lost_packets.channel_key;
1843
1844 DBG("UST consumer lost packets command for session id %"
1845 PRIu64, id);
1846 rcu_read_lock();
1847 pthread_mutex_lock(&consumer_data.lock);
1848
1849 ht = consumer_data.stream_list_ht;
1850
1851 /*
1852 * We only need a reference to the channel, but they are not
1853 * directly indexed, so we just use the first matching stream
1854 * to extract the information we need, we default to 0 if not
1855 * found (no packets lost if the channel is not yet in use).
1856 */
1857 lost_packets = 0;
1858 cds_lfht_for_each_entry_duplicate(ht->ht,
1859 ht->hash_fct(&id, lttng_ht_seed),
1860 ht->match_fct, &id,
1861 &iter.iter, stream, node_session_id.node) {
1862 if (stream->chan->key == key) {
1863 lost_packets = stream->chan->lost_packets;
1864 break;
1865 }
1866 }
1867 pthread_mutex_unlock(&consumer_data.lock);
1868 rcu_read_unlock();
1869
1870 DBG("UST consumer lost packets command for session id %"
1871 PRIu64 ", channel key %" PRIu64, id, key);
1872
1873 health_code_update();
1874
1875 /* Send back returned value to session daemon */
1876 ret = lttcomm_send_unix_sock(sock, &lost_packets,
1877 sizeof(lost_packets));
1878 if (ret < 0) {
1879 PERROR("send lost packets");
1880 goto error_fatal;
1881 }
1882
1883 break;
1884 }
1885 case LTTNG_CONSUMER_SET_CHANNEL_MONITOR_PIPE:
1886 {
1887 int channel_monitor_pipe;
1888
1889 ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1890 /* Successfully received the command's type. */
1891 ret = consumer_send_status_msg(sock, ret_code);
1892 if (ret < 0) {
1893 goto error_fatal;
1894 }
1895
1896 ret = lttcomm_recv_fds_unix_sock(sock, &channel_monitor_pipe,
1897 1);
1898 if (ret != sizeof(channel_monitor_pipe)) {
1899 ERR("Failed to receive channel monitor pipe");
1900 goto error_fatal;
1901 }
1902
1903 DBG("Received channel monitor pipe (%d)", channel_monitor_pipe);
1904 ret = consumer_timer_thread_set_channel_monitor_pipe(
1905 channel_monitor_pipe);
1906 if (!ret) {
1907 int flags;
1908
1909 ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1910 /* Set the pipe as non-blocking. */
1911 ret = fcntl(channel_monitor_pipe, F_GETFL, 0);
1912 if (ret == -1) {
1913 PERROR("fcntl get flags of the channel monitoring pipe");
1914 goto error_fatal;
1915 }
1916 flags = ret;
1917
1918 ret = fcntl(channel_monitor_pipe, F_SETFL,
1919 flags | O_NONBLOCK);
1920 if (ret == -1) {
1921 PERROR("fcntl set O_NONBLOCK flag of the channel monitoring pipe");
1922 goto error_fatal;
1923 }
1924 DBG("Channel monitor pipe set as non-blocking");
1925 } else {
1926 ret_code = LTTCOMM_CONSUMERD_ALREADY_SET;
1927 }
1928 goto end_msg_sessiond;
1929 }
1930 case LTTNG_CONSUMER_ROTATE_CHANNEL:
1931 {
1932 struct lttng_consumer_channel *channel;
1933 uint64_t key = msg.u.rotate_channel.key;
1934
1935 channel = consumer_find_channel(key);
1936 if (!channel) {
1937 DBG("Channel %" PRIu64 " not found", key);
1938 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
1939 } else {
1940 /*
1941 * Sample the rotate position of all the streams in
1942 * this channel.
1943 */
1944 ret = lttng_consumer_rotate_channel(channel, key,
1945 msg.u.rotate_channel.pathname,
1946 msg.u.rotate_channel.relayd_id,
1947 msg.u.rotate_channel.metadata,
1948 msg.u.rotate_channel.new_chunk_id,
1949 ctx);
1950 if (ret < 0) {
1951 ERR("Rotate channel failed");
1952 ret_code = LTTCOMM_CONSUMERD_ROTATION_FAIL;
1953 }
1954
1955 health_code_update();
1956 }
1957 ret = consumer_send_status_msg(sock, ret_code);
1958 if (ret < 0) {
1959 /* Somehow, the session daemon is not responding anymore. */
1960 goto end_nosignal;
1961 }
1962
1963 /*
1964 * Rotate the streams that are ready right now.
1965 * FIXME: this is a second consecutive iteration over the
1966 * streams in a channel, there is probably a better way to
1967 * handle this, but it needs to be after the
1968 * consumer_send_status_msg() call.
1969 */
1970 if (channel) {
1971 ret = lttng_consumer_rotate_ready_streams(
1972 channel, key, ctx);
1973 if (ret < 0) {
1974 ERR("Rotate channel failed");
1975 }
1976 }
1977 break;
1978 }
1979 case LTTNG_CONSUMER_ROTATE_RENAME:
1980 {
1981 DBG("Consumer rename session %" PRIu64 " after rotation",
1982 msg.u.rotate_rename.session_id);
1983 ret = lttng_consumer_rotate_rename(msg.u.rotate_rename.old_path,
1984 msg.u.rotate_rename.new_path,
1985 msg.u.rotate_rename.uid,
1986 msg.u.rotate_rename.gid,
1987 msg.u.rotate_rename.relayd_id);
1988 if (ret < 0) {
1989 ERR("Rotate rename failed");
1990 ret_code = LTTCOMM_CONSUMERD_ROTATE_RENAME_FAILED;
1991 }
1992
1993 health_code_update();
1994
1995 ret = consumer_send_status_msg(sock, ret_code);
1996 if (ret < 0) {
1997 /* Somehow, the session daemon is not responding anymore. */
1998 goto end_nosignal;
1999 }
2000 break;
2001 }
2002 case LTTNG_CONSUMER_CHECK_ROTATION_PENDING_LOCAL:
2003 {
2004 int pending;
2005 uint32_t pending_reply;
2006
2007 DBG("Perform local check of pending rotation for session id %" PRIu64,
2008 msg.u.check_rotation_pending_local.session_id);
2009 pending = lttng_consumer_check_rotation_pending_local(
2010 msg.u.check_rotation_pending_local.session_id,
2011 msg.u.check_rotation_pending_local.chunk_id);
2012 if (pending < 0) {
2013 ERR("Local rotation pending check failed with code %i", pending);
2014 ret_code = LTTCOMM_CONSUMERD_ROTATION_PENDING_LOCAL_FAILED;
2015 } else {
2016 pending_reply = !!pending;
2017 }
2018
2019 health_code_update();
2020
2021 ret = consumer_send_status_msg(sock, ret_code);
2022 if (ret < 0) {
2023 /* Somehow, the session daemon is not responding anymore. */
2024 goto end_nosignal;
2025 }
2026
2027 if (pending < 0) {
2028 /*
2029 * An error occured while running the command;
2030 * don't send the 'pending' flag as the sessiond
2031 * will not read it.
2032 */
2033 break;
2034 }
2035
2036 /* Send back returned value to session daemon */
2037 ret = lttcomm_send_unix_sock(sock, &pending_reply,
2038 sizeof(pending_reply));
2039 if (ret < 0) {
2040 PERROR("Failed to send rotation pending return code");
2041 goto error_fatal;
2042 }
2043 break;
2044 }
2045 case LTTNG_CONSUMER_CHECK_ROTATION_PENDING_RELAY:
2046 {
2047 int pending;
2048 uint32_t pending_reply;
2049
2050 DBG("Perform relayd check of pending rotation for session id %" PRIu64,
2051 msg.u.check_rotation_pending_relay.session_id);
2052 pending = lttng_consumer_check_rotation_pending_relay(
2053 msg.u.check_rotation_pending_relay.session_id,
2054 msg.u.check_rotation_pending_relay.relayd_id,
2055 msg.u.check_rotation_pending_relay.chunk_id);
2056 if (pending < 0) {
2057 ERR("Relayd rotation pending check failed with code %i", pending);
2058 ret_code = LTTCOMM_CONSUMERD_ROTATION_PENDING_RELAY_FAILED;
2059 } else {
2060 pending_reply = !!pending;
2061 }
2062
2063 health_code_update();
2064
2065 ret = consumer_send_status_msg(sock, ret_code);
2066 if (ret < 0) {
2067 /* Somehow, the session daemon is not responding anymore. */
2068 goto end_nosignal;
2069 }
2070
2071 if (pending < 0) {
2072 /*
2073 * An error occured while running the command;
2074 * don't send the 'pending' flag as the sessiond
2075 * will not read it.
2076 */
2077 break;
2078 }
2079
2080 /* Send back returned value to session daemon */
2081 ret = lttcomm_send_unix_sock(sock, &pending_reply,
2082 sizeof(pending_reply));
2083 if (ret < 0) {
2084 PERROR("Failed to send rotation pending return code");
2085 goto error_fatal;
2086 }
2087 break;
2088 }
2089 case LTTNG_CONSUMER_MKDIR:
2090 {
2091 DBG("Consumer mkdir %s in session %" PRIu64,
2092 msg.u.mkdir.path,
2093 msg.u.mkdir.session_id);
2094 ret = lttng_consumer_mkdir(msg.u.mkdir.path,
2095 msg.u.mkdir.uid,
2096 msg.u.mkdir.gid,
2097 msg.u.mkdir.relayd_id);
2098 if (ret < 0) {
2099 ERR("consumer mkdir failed");
2100 ret_code = LTTCOMM_CONSUMERD_MKDIR_FAILED;
2101 }
2102
2103 health_code_update();
2104
2105 ret = consumer_send_status_msg(sock, ret_code);
2106 if (ret < 0) {
2107 /* Somehow, the session daemon is not responding anymore. */
2108 goto end_nosignal;
2109 }
2110 break;
2111 }
2112 default:
2113 break;
2114 }
2115
2116 end_nosignal:
2117 rcu_read_unlock();
2118
2119 health_code_update();
2120
2121 /*
2122 * Return 1 to indicate success since the 0 value can be a socket
2123 * shutdown during the recv() or send() call.
2124 */
2125 return 1;
2126
2127 end_msg_sessiond:
2128 /*
2129 * The returned value here is not useful since either way we'll return 1 to
2130 * the caller because the session daemon socket management is done
2131 * elsewhere. Returning a negative code or 0 will shutdown the consumer.
2132 */
2133 ret = consumer_send_status_msg(sock, ret_code);
2134 if (ret < 0) {
2135 goto error_fatal;
2136 }
2137 rcu_read_unlock();
2138
2139 health_code_update();
2140
2141 return 1;
2142 end_channel_error:
2143 if (channel) {
2144 /*
2145 * Free channel here since no one has a reference to it. We don't
2146 * free after that because a stream can store this pointer.
2147 */
2148 destroy_channel(channel);
2149 }
2150 /* We have to send a status channel message indicating an error. */
2151 ret = consumer_send_status_channel(sock, NULL);
2152 if (ret < 0) {
2153 /* Stop everything if session daemon can not be notified. */
2154 goto error_fatal;
2155 }
2156 rcu_read_unlock();
2157
2158 health_code_update();
2159
2160 return 1;
2161 error_fatal:
2162 rcu_read_unlock();
2163 /* This will issue a consumer stop. */
2164 return -1;
2165 }
2166
2167 /*
2168 * Wrapper over the mmap() read offset from ust-ctl library. Since this can be
2169 * compiled out, we isolate it in this library.
2170 */
2171 int lttng_ustctl_get_mmap_read_offset(struct lttng_consumer_stream *stream,
2172 unsigned long *off)
2173 {
2174 assert(stream);
2175 assert(stream->ustream);
2176
2177 return ustctl_get_mmap_read_offset(stream->ustream, off);
2178 }
2179
2180 /*
2181 * Wrapper over the mmap() read offset from ust-ctl library. Since this can be
2182 * compiled out, we isolate it in this library.
2183 */
2184 void *lttng_ustctl_get_mmap_base(struct lttng_consumer_stream *stream)
2185 {
2186 assert(stream);
2187 assert(stream->ustream);
2188
2189 return ustctl_get_mmap_base(stream->ustream);
2190 }
2191
2192 void lttng_ustctl_flush_buffer(struct lttng_consumer_stream *stream,
2193 int producer_active)
2194 {
2195 assert(stream);
2196 assert(stream->ustream);
2197
2198 ustctl_flush_buffer(stream->ustream, producer_active);
2199 }
2200
2201 /*
2202 * Take a snapshot for a specific stream.
2203 *
2204 * Returns 0 on success, < 0 on error
2205 */
2206 int lttng_ustconsumer_take_snapshot(struct lttng_consumer_stream *stream)
2207 {
2208 assert(stream);
2209 assert(stream->ustream);
2210
2211 return ustctl_snapshot(stream->ustream);
2212 }
2213
2214 /*
2215 * Sample consumed and produced positions for a specific stream.
2216 *
2217 * Returns 0 on success, < 0 on error.
2218 */
2219 int lttng_ustconsumer_sample_snapshot_positions(
2220 struct lttng_consumer_stream *stream)
2221 {
2222 assert(stream);
2223 assert(stream->ustream);
2224
2225 return ustctl_snapshot_sample_positions(stream->ustream);
2226 }
2227
2228 /*
2229 * Get the produced position
2230 *
2231 * Returns 0 on success, < 0 on error
2232 */
2233 int lttng_ustconsumer_get_produced_snapshot(
2234 struct lttng_consumer_stream *stream, unsigned long *pos)
2235 {
2236 assert(stream);
2237 assert(stream->ustream);
2238 assert(pos);
2239
2240 return ustctl_snapshot_get_produced(stream->ustream, pos);
2241 }
2242
2243 /*
2244 * Get the consumed position
2245 *
2246 * Returns 0 on success, < 0 on error
2247 */
2248 int lttng_ustconsumer_get_consumed_snapshot(
2249 struct lttng_consumer_stream *stream, unsigned long *pos)
2250 {
2251 assert(stream);
2252 assert(stream->ustream);
2253 assert(pos);
2254
2255 return ustctl_snapshot_get_consumed(stream->ustream, pos);
2256 }
2257
2258 void lttng_ustconsumer_flush_buffer(struct lttng_consumer_stream *stream,
2259 int producer)
2260 {
2261 assert(stream);
2262 assert(stream->ustream);
2263
2264 ustctl_flush_buffer(stream->ustream, producer);
2265 }
2266
2267 int lttng_ustconsumer_get_current_timestamp(
2268 struct lttng_consumer_stream *stream, uint64_t *ts)
2269 {
2270 assert(stream);
2271 assert(stream->ustream);
2272 assert(ts);
2273
2274 return ustctl_get_current_timestamp(stream->ustream, ts);
2275 }
2276
2277 int lttng_ustconsumer_get_sequence_number(
2278 struct lttng_consumer_stream *stream, uint64_t *seq)
2279 {
2280 assert(stream);
2281 assert(stream->ustream);
2282 assert(seq);
2283
2284 return ustctl_get_sequence_number(stream->ustream, seq);
2285 }
2286
2287 /*
2288 * Called when the stream signals the consumer that it has hung up.
2289 */
2290 void lttng_ustconsumer_on_stream_hangup(struct lttng_consumer_stream *stream)
2291 {
2292 assert(stream);
2293 assert(stream->ustream);
2294
2295 pthread_mutex_lock(&stream->lock);
2296 if (!stream->quiescent) {
2297 ustctl_flush_buffer(stream->ustream, 0);
2298 stream->quiescent = true;
2299 }
2300 pthread_mutex_unlock(&stream->lock);
2301 stream->hangup_flush_done = 1;
2302 }
2303
2304 void lttng_ustconsumer_del_channel(struct lttng_consumer_channel *chan)
2305 {
2306 int i;
2307
2308 assert(chan);
2309 assert(chan->uchan);
2310
2311 if (chan->switch_timer_enabled == 1) {
2312 consumer_timer_switch_stop(chan);
2313 }
2314 for (i = 0; i < chan->nr_stream_fds; i++) {
2315 int ret;
2316
2317 ret = close(chan->stream_fds[i]);
2318 if (ret) {
2319 PERROR("close");
2320 }
2321 if (chan->shm_path[0]) {
2322 char shm_path[PATH_MAX];
2323
2324 ret = get_stream_shm_path(shm_path, chan->shm_path, i);
2325 if (ret) {
2326 ERR("Cannot get stream shm path");
2327 }
2328 ret = run_as_unlink(shm_path, chan->uid, chan->gid);
2329 if (ret) {
2330 PERROR("unlink %s", shm_path);
2331 }
2332 }
2333 }
2334 }
2335
2336 void lttng_ustconsumer_free_channel(struct lttng_consumer_channel *chan)
2337 {
2338 assert(chan);
2339 assert(chan->uchan);
2340
2341 consumer_metadata_cache_destroy(chan);
2342 ustctl_destroy_channel(chan->uchan);
2343 /* Try to rmdir all directories under shm_path root. */
2344 if (chan->root_shm_path[0]) {
2345 (void) run_as_rmdir_recursive(chan->root_shm_path,
2346 chan->uid, chan->gid);
2347 }
2348 free(chan->stream_fds);
2349 }
2350
2351 void lttng_ustconsumer_del_stream(struct lttng_consumer_stream *stream)
2352 {
2353 assert(stream);
2354 assert(stream->ustream);
2355
2356 if (stream->chan->switch_timer_enabled == 1) {
2357 consumer_timer_switch_stop(stream->chan);
2358 }
2359 ustctl_destroy_stream(stream->ustream);
2360 }
2361
2362 int lttng_ustconsumer_get_wakeup_fd(struct lttng_consumer_stream *stream)
2363 {
2364 assert(stream);
2365 assert(stream->ustream);
2366
2367 return ustctl_stream_get_wakeup_fd(stream->ustream);
2368 }
2369
2370 int lttng_ustconsumer_close_wakeup_fd(struct lttng_consumer_stream *stream)
2371 {
2372 assert(stream);
2373 assert(stream->ustream);
2374
2375 return ustctl_stream_close_wakeup_fd(stream->ustream);
2376 }
2377
2378 /*
2379 * Populate index values of a UST stream. Values are set in big endian order.
2380 *
2381 * Return 0 on success or else a negative value.
2382 */
2383 static int get_index_values(struct ctf_packet_index *index,
2384 struct ustctl_consumer_stream *ustream)
2385 {
2386 int ret;
2387
2388 ret = ustctl_get_timestamp_begin(ustream, &index->timestamp_begin);
2389 if (ret < 0) {
2390 PERROR("ustctl_get_timestamp_begin");
2391 goto error;
2392 }
2393 index->timestamp_begin = htobe64(index->timestamp_begin);
2394
2395 ret = ustctl_get_timestamp_end(ustream, &index->timestamp_end);
2396 if (ret < 0) {
2397 PERROR("ustctl_get_timestamp_end");
2398 goto error;
2399 }
2400 index->timestamp_end = htobe64(index->timestamp_end);
2401
2402 ret = ustctl_get_events_discarded(ustream, &index->events_discarded);
2403 if (ret < 0) {
2404 PERROR("ustctl_get_events_discarded");
2405 goto error;
2406 }
2407 index->events_discarded = htobe64(index->events_discarded);
2408
2409 ret = ustctl_get_content_size(ustream, &index->content_size);
2410 if (ret < 0) {
2411 PERROR("ustctl_get_content_size");
2412 goto error;
2413 }
2414 index->content_size = htobe64(index->content_size);
2415
2416 ret = ustctl_get_packet_size(ustream, &index->packet_size);
2417 if (ret < 0) {
2418 PERROR("ustctl_get_packet_size");
2419 goto error;
2420 }
2421 index->packet_size = htobe64(index->packet_size);
2422
2423 ret = ustctl_get_stream_id(ustream, &index->stream_id);
2424 if (ret < 0) {
2425 PERROR("ustctl_get_stream_id");
2426 goto error;
2427 }
2428 index->stream_id = htobe64(index->stream_id);
2429
2430 ret = ustctl_get_instance_id(ustream, &index->stream_instance_id);
2431 if (ret < 0) {
2432 PERROR("ustctl_get_instance_id");
2433 goto error;
2434 }
2435 index->stream_instance_id = htobe64(index->stream_instance_id);
2436
2437 ret = ustctl_get_sequence_number(ustream, &index->packet_seq_num);
2438 if (ret < 0) {
2439 PERROR("ustctl_get_sequence_number");
2440 goto error;
2441 }
2442 index->packet_seq_num = htobe64(index->packet_seq_num);
2443
2444 error:
2445 return ret;
2446 }
2447
2448 static
2449 void metadata_stream_reset_cache(struct lttng_consumer_stream *stream,
2450 struct consumer_metadata_cache *cache)
2451 {
2452 DBG("Metadata stream update to version %" PRIu64,
2453 cache->version);
2454 stream->ust_metadata_pushed = 0;
2455 stream->metadata_version = cache->version;
2456 stream->reset_metadata_flag = 1;
2457 }
2458
2459 /*
2460 * Check if the version of the metadata stream and metadata cache match.
2461 * If the cache got updated, reset the metadata stream.
2462 * The stream lock and metadata cache lock MUST be held.
2463 * Return 0 on success, a negative value on error.
2464 */
2465 static
2466 int metadata_stream_check_version(struct lttng_consumer_stream *stream)
2467 {
2468 int ret = 0;
2469 struct consumer_metadata_cache *cache = stream->chan->metadata_cache;
2470
2471 if (cache->version == stream->metadata_version) {
2472 goto end;
2473 }
2474 metadata_stream_reset_cache(stream, cache);
2475
2476 end:
2477 return ret;
2478 }
2479
2480 /*
2481 * Write up to one packet from the metadata cache to the channel.
2482 *
2483 * Returns the number of bytes pushed in the cache, or a negative value
2484 * on error.
2485 */
2486 static
2487 int commit_one_metadata_packet(struct lttng_consumer_stream *stream)
2488 {
2489 ssize_t write_len;
2490 int ret;
2491
2492 pthread_mutex_lock(&stream->chan->metadata_cache->lock);
2493 ret = metadata_stream_check_version(stream);
2494 if (ret < 0) {
2495 goto end;
2496 }
2497 if (stream->chan->metadata_cache->max_offset
2498 == stream->ust_metadata_pushed) {
2499 ret = 0;
2500 goto end;
2501 }
2502
2503 write_len = ustctl_write_one_packet_to_channel(stream->chan->uchan,
2504 &stream->chan->metadata_cache->data[stream->ust_metadata_pushed],
2505 stream->chan->metadata_cache->max_offset
2506 - stream->ust_metadata_pushed);
2507 assert(write_len != 0);
2508 if (write_len < 0) {
2509 ERR("Writing one metadata packet");
2510 ret = -1;
2511 goto end;
2512 }
2513 stream->ust_metadata_pushed += write_len;
2514
2515 assert(stream->chan->metadata_cache->max_offset >=
2516 stream->ust_metadata_pushed);
2517 ret = write_len;
2518
2519 end:
2520 pthread_mutex_unlock(&stream->chan->metadata_cache->lock);
2521 return ret;
2522 }
2523
2524
2525 /*
2526 * Sync metadata meaning request them to the session daemon and snapshot to the
2527 * metadata thread can consumer them.
2528 *
2529 * Metadata stream lock is held here, but we need to release it when
2530 * interacting with sessiond, else we cause a deadlock with live
2531 * awaiting on metadata to be pushed out.
2532 *
2533 * Return 0 if new metadatda is available, EAGAIN if the metadata stream
2534 * is empty or a negative value on error.
2535 */
2536 int lttng_ustconsumer_sync_metadata(struct lttng_consumer_local_data *ctx,
2537 struct lttng_consumer_stream *metadata)
2538 {
2539 int ret;
2540 int retry = 0;
2541
2542 assert(ctx);
2543 assert(metadata);
2544
2545 pthread_mutex_unlock(&metadata->lock);
2546 /*
2547 * Request metadata from the sessiond, but don't wait for the flush
2548 * because we locked the metadata thread.
2549 */
2550 ret = lttng_ustconsumer_request_metadata(ctx, metadata->chan, 0, 0);
2551 pthread_mutex_lock(&metadata->lock);
2552 if (ret < 0) {
2553 goto end;
2554 }
2555
2556 ret = commit_one_metadata_packet(metadata);
2557 if (ret <= 0) {
2558 goto end;
2559 } else if (ret > 0) {
2560 retry = 1;
2561 }
2562
2563 ustctl_flush_buffer(metadata->ustream, 1);
2564 ret = ustctl_snapshot(metadata->ustream);
2565 if (ret < 0) {
2566 if (errno != EAGAIN) {
2567 ERR("Sync metadata, taking UST snapshot");
2568 goto end;
2569 }
2570 DBG("No new metadata when syncing them.");
2571 /* No new metadata, exit. */
2572 ret = ENODATA;
2573 goto end;
2574 }
2575
2576 /*
2577 * After this flush, we still need to extract metadata.
2578 */
2579 if (retry) {
2580 ret = EAGAIN;
2581 }
2582
2583 end:
2584 return ret;
2585 }
2586
2587 /*
2588 * Return 0 on success else a negative value.
2589 */
2590 static int notify_if_more_data(struct lttng_consumer_stream *stream,
2591 struct lttng_consumer_local_data *ctx)
2592 {
2593 int ret;
2594 struct ustctl_consumer_stream *ustream;
2595
2596 assert(stream);
2597 assert(ctx);
2598
2599 ustream = stream->ustream;
2600
2601 /*
2602 * First, we are going to check if there is a new subbuffer available
2603 * before reading the stream wait_fd.
2604 */
2605 /* Get the next subbuffer */
2606 ret = ustctl_get_next_subbuf(ustream);
2607 if (ret) {
2608 /* No more data found, flag the stream. */
2609 stream->has_data = 0;
2610 ret = 0;
2611 goto end;
2612 }
2613
2614 ret = ustctl_put_subbuf(ustream);
2615 assert(!ret);
2616
2617 /* This stream still has data. Flag it and wake up the data thread. */
2618 stream->has_data = 1;
2619
2620 if (stream->monitor && !stream->hangup_flush_done && !ctx->has_wakeup) {
2621 ssize_t writelen;
2622
2623 writelen = lttng_pipe_write(ctx->consumer_wakeup_pipe, "!", 1);
2624 if (writelen < 0 && errno != EAGAIN && errno != EWOULDBLOCK) {
2625 ret = writelen;
2626 goto end;
2627 }
2628
2629 /* The wake up pipe has been notified. */
2630 ctx->has_wakeup = 1;
2631 }
2632 ret = 0;
2633
2634 end:
2635 return ret;
2636 }
2637
2638 static
2639 int update_stream_stats(struct lttng_consumer_stream *stream)
2640 {
2641 int ret;
2642 uint64_t seq, discarded;
2643
2644 ret = ustctl_get_sequence_number(stream->ustream, &seq);
2645 if (ret < 0) {
2646 PERROR("ustctl_get_sequence_number");
2647 goto end;
2648 }
2649 /*
2650 * Start the sequence when we extract the first packet in case we don't
2651 * start at 0 (for example if a consumer is not connected to the
2652 * session immediately after the beginning).
2653 */
2654 if (stream->last_sequence_number == -1ULL) {
2655 stream->last_sequence_number = seq;
2656 } else if (seq > stream->last_sequence_number) {
2657 stream->chan->lost_packets += seq -
2658 stream->last_sequence_number - 1;
2659 } else {
2660 /* seq <= last_sequence_number */
2661 ERR("Sequence number inconsistent : prev = %" PRIu64
2662 ", current = %" PRIu64,
2663 stream->last_sequence_number, seq);
2664 ret = -1;
2665 goto end;
2666 }
2667 stream->last_sequence_number = seq;
2668
2669 ret = ustctl_get_events_discarded(stream->ustream, &discarded);
2670 if (ret < 0) {
2671 PERROR("kernctl_get_events_discarded");
2672 goto end;
2673 }
2674 if (discarded < stream->last_discarded_events) {
2675 /*
2676 * Overflow has occurred. We assume only one wrap-around
2677 * has occurred.
2678 */
2679 stream->chan->discarded_events +=
2680 (1ULL << (CAA_BITS_PER_LONG - 1)) -
2681 stream->last_discarded_events + discarded;
2682 } else {
2683 stream->chan->discarded_events += discarded -
2684 stream->last_discarded_events;
2685 }
2686 stream->last_discarded_events = discarded;
2687 ret = 0;
2688
2689 end:
2690 return ret;
2691 }
2692
2693 /*
2694 * Read subbuffer from the given stream.
2695 *
2696 * Stream lock MUST be acquired.
2697 *
2698 * Return 0 on success else a negative value.
2699 */
2700 int lttng_ustconsumer_read_subbuffer(struct lttng_consumer_stream *stream,
2701 struct lttng_consumer_local_data *ctx, bool *rotated)
2702 {
2703 unsigned long len, subbuf_size, padding;
2704 int err, write_index = 1, rotation_ret;
2705 long ret = 0;
2706 struct ustctl_consumer_stream *ustream;
2707 struct ctf_packet_index index;
2708
2709 assert(stream);
2710 assert(stream->ustream);
2711 assert(ctx);
2712
2713 DBG("In UST read_subbuffer (wait_fd: %d, name: %s)", stream->wait_fd,
2714 stream->name);
2715
2716 /* Ease our life for what's next. */
2717 ustream = stream->ustream;
2718
2719 /*
2720 * We can consume the 1 byte written into the wait_fd by UST. Don't trigger
2721 * error if we cannot read this one byte (read returns 0), or if the error
2722 * is EAGAIN or EWOULDBLOCK.
2723 *
2724 * This is only done when the stream is monitored by a thread, before the
2725 * flush is done after a hangup and if the stream is not flagged with data
2726 * since there might be nothing to consume in the wait fd but still have
2727 * data available flagged by the consumer wake up pipe.
2728 */
2729 if (stream->monitor && !stream->hangup_flush_done && !stream->has_data) {
2730 char dummy;
2731 ssize_t readlen;
2732
2733 readlen = lttng_read(stream->wait_fd, &dummy, 1);
2734 if (readlen < 0 && errno != EAGAIN && errno != EWOULDBLOCK) {
2735 ret = readlen;
2736 goto error;
2737 }
2738 }
2739
2740 /*
2741 * If the stream was flagged to be ready for rotation before we extract the
2742 * next packet, rotate it now.
2743 */
2744 if (stream->rotate_ready) {
2745 DBG("Rotate stream before extracting data");
2746 rotation_ret = lttng_consumer_rotate_stream(ctx, stream, rotated);
2747 if (rotation_ret < 0) {
2748 ERR("Stream rotation error");
2749 ret = -1;
2750 goto error;
2751 }
2752 }
2753
2754 retry:
2755 /* Get the next subbuffer */
2756 err = ustctl_get_next_subbuf(ustream);
2757 if (err != 0) {
2758 /*
2759 * Populate metadata info if the existing info has
2760 * already been read.
2761 */
2762 if (stream->metadata_flag) {
2763 ret = commit_one_metadata_packet(stream);
2764 if (ret <= 0) {
2765 goto error;
2766 }
2767 ustctl_flush_buffer(stream->ustream, 1);
2768 goto retry;
2769 }
2770
2771 ret = err; /* ustctl_get_next_subbuf returns negative, caller expect positive. */
2772 /*
2773 * This is a debug message even for single-threaded consumer,
2774 * because poll() have more relaxed criterions than get subbuf,
2775 * so get_subbuf may fail for short race windows where poll()
2776 * would issue wakeups.
2777 */
2778 DBG("Reserving sub buffer failed (everything is normal, "
2779 "it is due to concurrency) [ret: %d]", err);
2780 goto error;
2781 }
2782 assert(stream->chan->output == CONSUMER_CHANNEL_MMAP);
2783
2784 if (!stream->metadata_flag) {
2785 index.offset = htobe64(stream->out_fd_offset);
2786 ret = get_index_values(&index, ustream);
2787 if (ret < 0) {
2788 err = ustctl_put_subbuf(ustream);
2789 assert(err == 0);
2790 goto error;
2791 }
2792
2793 /* Update the stream's sequence and discarded events count. */
2794 ret = update_stream_stats(stream);
2795 if (ret < 0) {
2796 PERROR("kernctl_get_events_discarded");
2797 err = ustctl_put_subbuf(ustream);
2798 assert(err == 0);
2799 goto error;
2800 }
2801 } else {
2802 write_index = 0;
2803 }
2804
2805 /* Get the full padded subbuffer size */
2806 err = ustctl_get_padded_subbuf_size(ustream, &len);
2807 assert(err == 0);
2808
2809 /* Get subbuffer data size (without padding) */
2810 err = ustctl_get_subbuf_size(ustream, &subbuf_size);
2811 assert(err == 0);
2812
2813 /* Make sure we don't get a subbuffer size bigger than the padded */
2814 assert(len >= subbuf_size);
2815
2816 padding = len - subbuf_size;
2817
2818 /* write the subbuffer to the tracefile */
2819 ret = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, subbuf_size, padding, &index);
2820 /*
2821 * The mmap operation should write subbuf_size amount of data when network
2822 * streaming or the full padding (len) size when we are _not_ streaming.
2823 */
2824 if ((ret != subbuf_size && stream->net_seq_idx != (uint64_t) -1ULL) ||
2825 (ret != len && stream->net_seq_idx == (uint64_t) -1ULL)) {
2826 /*
2827 * Display the error but continue processing to try to release the
2828 * subbuffer. This is a DBG statement since any unexpected kill or
2829 * signal, the application gets unregistered, relayd gets closed or
2830 * anything that affects the buffer lifetime will trigger this error.
2831 * So, for the sake of the user, don't print this error since it can
2832 * happen and it is OK with the code flow.
2833 */
2834 DBG("Error writing to tracefile "
2835 "(ret: %ld != len: %lu != subbuf_size: %lu)",
2836 ret, len, subbuf_size);
2837 write_index = 0;
2838 }
2839 err = ustctl_put_next_subbuf(ustream);
2840 assert(err == 0);
2841
2842 /*
2843 * This will consumer the byte on the wait_fd if and only if there is not
2844 * next subbuffer to be acquired.
2845 */
2846 if (!stream->metadata_flag) {
2847 ret = notify_if_more_data(stream, ctx);
2848 if (ret < 0) {
2849 goto error;
2850 }
2851 }
2852
2853 /* Write index if needed. */
2854 if (!write_index) {
2855 goto rotate;
2856 }
2857
2858 if (stream->chan->live_timer_interval && !stream->metadata_flag) {
2859 /*
2860 * In live, block until all the metadata is sent.
2861 */
2862 pthread_mutex_lock(&stream->metadata_timer_lock);
2863 assert(!stream->missed_metadata_flush);
2864 stream->waiting_on_metadata = true;
2865 pthread_mutex_unlock(&stream->metadata_timer_lock);
2866
2867 err = consumer_stream_sync_metadata(ctx, stream->session_id);
2868
2869 pthread_mutex_lock(&stream->metadata_timer_lock);
2870 stream->waiting_on_metadata = false;
2871 if (stream->missed_metadata_flush) {
2872 stream->missed_metadata_flush = false;
2873 pthread_mutex_unlock(&stream->metadata_timer_lock);
2874 (void) consumer_flush_ust_index(stream);
2875 } else {
2876 pthread_mutex_unlock(&stream->metadata_timer_lock);
2877 }
2878
2879 if (err < 0) {
2880 goto error;
2881 }
2882 }
2883
2884 assert(!stream->metadata_flag);
2885 err = consumer_stream_write_index(stream, &index);
2886 if (err < 0) {
2887 goto error;
2888 }
2889
2890 rotate:
2891 /*
2892 * After extracting the packet, we check if the stream is now ready to be
2893 * rotated and perform the action immediately.
2894 */
2895 rotation_ret = lttng_consumer_stream_is_rotate_ready(stream);
2896 if (rotation_ret == 1) {
2897 rotation_ret = lttng_consumer_rotate_stream(ctx, stream, rotated);
2898 if (rotation_ret < 0) {
2899 ERR("Stream rotation error");
2900 ret = -1;
2901 goto error;
2902 }
2903 } else if (rotation_ret < 0) {
2904 ERR("Checking if stream is ready to rotate");
2905 ret = -1;
2906 goto error;
2907 }
2908 error:
2909 return ret;
2910 }
2911
2912 /*
2913 * Called when a stream is created.
2914 *
2915 * Return 0 on success or else a negative value.
2916 */
2917 int lttng_ustconsumer_on_recv_stream(struct lttng_consumer_stream *stream)
2918 {
2919 int ret;
2920
2921 assert(stream);
2922
2923 /* Don't create anything if this is set for streaming. */
2924 if (stream->net_seq_idx == (uint64_t) -1ULL && stream->chan->monitor) {
2925 ret = utils_create_stream_file(stream->chan->pathname, stream->name,
2926 stream->chan->tracefile_size, stream->tracefile_count_current,
2927 stream->uid, stream->gid, NULL);
2928 if (ret < 0) {
2929 goto error;
2930 }
2931 stream->out_fd = ret;
2932 stream->tracefile_size_current = 0;
2933
2934 if (!stream->metadata_flag) {
2935 struct lttng_index_file *index_file;
2936
2937 index_file = lttng_index_file_create(stream->chan->pathname,
2938 stream->name, stream->uid, stream->gid,
2939 stream->chan->tracefile_size,
2940 stream->tracefile_count_current,
2941 CTF_INDEX_MAJOR, CTF_INDEX_MINOR);
2942 if (!index_file) {
2943 goto error;
2944 }
2945 assert(!stream->index_file);
2946 stream->index_file = index_file;
2947 }
2948 }
2949 ret = 0;
2950
2951 error:
2952 return ret;
2953 }
2954
2955 /*
2956 * Check if data is still being extracted from the buffers for a specific
2957 * stream. Consumer data lock MUST be acquired before calling this function
2958 * and the stream lock.
2959 *
2960 * Return 1 if the traced data are still getting read else 0 meaning that the
2961 * data is available for trace viewer reading.
2962 */
2963 int lttng_ustconsumer_data_pending(struct lttng_consumer_stream *stream)
2964 {
2965 int ret;
2966
2967 assert(stream);
2968 assert(stream->ustream);
2969
2970 DBG("UST consumer checking data pending");
2971
2972 if (stream->endpoint_status != CONSUMER_ENDPOINT_ACTIVE) {
2973 ret = 0;
2974 goto end;
2975 }
2976
2977 if (stream->chan->type == CONSUMER_CHANNEL_TYPE_METADATA) {
2978 uint64_t contiguous, pushed;
2979
2980 /* Ease our life a bit. */
2981 contiguous = stream->chan->metadata_cache->max_offset;
2982 pushed = stream->ust_metadata_pushed;
2983
2984 /*
2985 * We can simply check whether all contiguously available data
2986 * has been pushed to the ring buffer, since the push operation
2987 * is performed within get_next_subbuf(), and because both
2988 * get_next_subbuf() and put_next_subbuf() are issued atomically
2989 * thanks to the stream lock within
2990 * lttng_ustconsumer_read_subbuffer(). This basically means that
2991 * whetnever ust_metadata_pushed is incremented, the associated
2992 * metadata has been consumed from the metadata stream.
2993 */
2994 DBG("UST consumer metadata pending check: contiguous %" PRIu64 " vs pushed %" PRIu64,
2995 contiguous, pushed);
2996 assert(((int64_t) (contiguous - pushed)) >= 0);
2997 if ((contiguous != pushed) ||
2998 (((int64_t) contiguous - pushed) > 0 || contiguous == 0)) {
2999 ret = 1; /* Data is pending */
3000 goto end;
3001 }
3002 } else {
3003 ret = ustctl_get_next_subbuf(stream->ustream);
3004 if (ret == 0) {
3005 /*
3006 * There is still data so let's put back this
3007 * subbuffer.
3008 */
3009 ret = ustctl_put_subbuf(stream->ustream);
3010 assert(ret == 0);
3011 ret = 1; /* Data is pending */
3012 goto end;
3013 }
3014 }
3015
3016 /* Data is NOT pending so ready to be read. */
3017 ret = 0;
3018
3019 end:
3020 return ret;
3021 }
3022
3023 /*
3024 * Stop a given metadata channel timer if enabled and close the wait fd which
3025 * is the poll pipe of the metadata stream.
3026 *
3027 * This MUST be called with the metadata channel acquired.
3028 */
3029 void lttng_ustconsumer_close_metadata(struct lttng_consumer_channel *metadata)
3030 {
3031 int ret;
3032
3033 assert(metadata);
3034 assert(metadata->type == CONSUMER_CHANNEL_TYPE_METADATA);
3035
3036 DBG("Closing metadata channel key %" PRIu64, metadata->key);
3037
3038 if (metadata->switch_timer_enabled == 1) {
3039 consumer_timer_switch_stop(metadata);
3040 }
3041
3042 if (!metadata->metadata_stream) {
3043 goto end;
3044 }
3045
3046 /*
3047 * Closing write side so the thread monitoring the stream wakes up if any
3048 * and clean the metadata stream.
3049 */
3050 if (metadata->metadata_stream->ust_metadata_poll_pipe[1] >= 0) {
3051 ret = close(metadata->metadata_stream->ust_metadata_poll_pipe[1]);
3052 if (ret < 0) {
3053 PERROR("closing metadata pipe write side");
3054 }
3055 metadata->metadata_stream->ust_metadata_poll_pipe[1] = -1;
3056 }
3057
3058 end:
3059 return;
3060 }
3061
3062 /*
3063 * Close every metadata stream wait fd of the metadata hash table. This
3064 * function MUST be used very carefully so not to run into a race between the
3065 * metadata thread handling streams and this function closing their wait fd.
3066 *
3067 * For UST, this is used when the session daemon hangs up. Its the metadata
3068 * producer so calling this is safe because we are assured that no state change
3069 * can occur in the metadata thread for the streams in the hash table.
3070 */
3071 void lttng_ustconsumer_close_all_metadata(struct lttng_ht *metadata_ht)
3072 {
3073 struct lttng_ht_iter iter;
3074 struct lttng_consumer_stream *stream;
3075
3076 assert(metadata_ht);
3077 assert(metadata_ht->ht);
3078
3079 DBG("UST consumer closing all metadata streams");
3080
3081 rcu_read_lock();
3082 cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream,
3083 node.node) {
3084
3085 health_code_update();
3086
3087 pthread_mutex_lock(&stream->chan->lock);
3088 lttng_ustconsumer_close_metadata(stream->chan);
3089 pthread_mutex_unlock(&stream->chan->lock);
3090
3091 }
3092 rcu_read_unlock();
3093 }
3094
3095 void lttng_ustconsumer_close_stream_wakeup(struct lttng_consumer_stream *stream)
3096 {
3097 int ret;
3098
3099 ret = ustctl_stream_close_wakeup_fd(stream->ustream);
3100 if (ret < 0) {
3101 ERR("Unable to close wakeup fd");
3102 }
3103 }
3104
3105 /*
3106 * Please refer to consumer-timer.c before adding any lock within this
3107 * function or any of its callees. Timers have a very strict locking
3108 * semantic with respect to teardown. Failure to respect this semantic
3109 * introduces deadlocks.
3110 *
3111 * DON'T hold the metadata lock when calling this function, else this
3112 * can cause deadlock involving consumer awaiting for metadata to be
3113 * pushed out due to concurrent interaction with the session daemon.
3114 */
3115 int lttng_ustconsumer_request_metadata(struct lttng_consumer_local_data *ctx,
3116 struct lttng_consumer_channel *channel, int timer, int wait)
3117 {
3118 struct lttcomm_metadata_request_msg request;
3119 struct lttcomm_consumer_msg msg;
3120 enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS;
3121 uint64_t len, key, offset, version;
3122 int ret;
3123
3124 assert(channel);
3125 assert(channel->metadata_cache);
3126
3127 memset(&request, 0, sizeof(request));
3128
3129 /* send the metadata request to sessiond */
3130 switch (consumer_data.type) {
3131 case LTTNG_CONSUMER64_UST:
3132 request.bits_per_long = 64;
3133 break;
3134 case LTTNG_CONSUMER32_UST:
3135 request.bits_per_long = 32;
3136 break;
3137 default:
3138 request.bits_per_long = 0;
3139 break;
3140 }
3141
3142 request.session_id = channel->session_id;
3143 request.session_id_per_pid = channel->session_id_per_pid;
3144 /*
3145 * Request the application UID here so the metadata of that application can
3146 * be sent back. The channel UID corresponds to the user UID of the session
3147 * used for the rights on the stream file(s).
3148 */
3149 request.uid = channel->ust_app_uid;
3150 request.key = channel->key;
3151
3152 DBG("Sending metadata request to sessiond, session id %" PRIu64
3153 ", per-pid %" PRIu64 ", app UID %u and channel key %" PRIu64,
3154 request.session_id, request.session_id_per_pid, request.uid,
3155 request.key);
3156
3157 pthread_mutex_lock(&ctx->metadata_socket_lock);
3158
3159 health_code_update();
3160
3161 ret = lttcomm_send_unix_sock(ctx->consumer_metadata_socket, &request,
3162 sizeof(request));
3163 if (ret < 0) {
3164 ERR("Asking metadata to sessiond");
3165 goto end;
3166 }
3167
3168 healt