Send session creation time to relay daemon when supported
[lttng-tools.git] / src / common / ust-consumer / ust-consumer.c
1 /*
2 * Copyright (C) 2011 - Julien Desfossez <julien.desfossez@polymtl.ca>
3 * Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
4 * Copyright (C) 2017 - Jérémie Galarneau <jeremie.galarneau@efficios.com>
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License, version 2 only,
8 * as published by the Free Software Foundation.
9 *
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
14 *
15 * You should have received a copy of the GNU General Public License along
16 * with this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
18 */
19
20 #define _LGPL_SOURCE
21 #include <assert.h>
22 #include <lttng/ust-ctl.h>
23 #include <poll.h>
24 #include <pthread.h>
25 #include <stdlib.h>
26 #include <string.h>
27 #include <sys/mman.h>
28 #include <sys/socket.h>
29 #include <sys/stat.h>
30 #include <sys/types.h>
31 #include <inttypes.h>
32 #include <unistd.h>
33 #include <urcu/list.h>
34 #include <signal.h>
35 #include <stdbool.h>
36
37 #include <bin/lttng-consumerd/health-consumerd.h>
38 #include <common/common.h>
39 #include <common/sessiond-comm/sessiond-comm.h>
40 #include <common/relayd/relayd.h>
41 #include <common/compat/fcntl.h>
42 #include <common/compat/endian.h>
43 #include <common/consumer/consumer-metadata-cache.h>
44 #include <common/consumer/consumer-stream.h>
45 #include <common/consumer/consumer-timer.h>
46 #include <common/utils.h>
47 #include <common/index/index.h>
48
49 #include "ust-consumer.h"
50
51 #define INT_MAX_STR_LEN 12 /* includes \0 */
52
53 extern struct lttng_consumer_global_data consumer_data;
54 extern int consumer_poll_timeout;
55
56 /*
57 * Free channel object and all streams associated with it. This MUST be used
58 * only and only if the channel has _NEVER_ been added to the global channel
59 * hash table.
60 */
61 static void destroy_channel(struct lttng_consumer_channel *channel)
62 {
63 struct lttng_consumer_stream *stream, *stmp;
64
65 assert(channel);
66
67 DBG("UST consumer cleaning stream list");
68
69 cds_list_for_each_entry_safe(stream, stmp, &channel->streams.head,
70 send_node) {
71
72 health_code_update();
73
74 cds_list_del(&stream->send_node);
75 ustctl_destroy_stream(stream->ustream);
76 lttng_trace_chunk_put(stream->trace_chunk);
77 free(stream);
78 }
79
80 /*
81 * If a channel is available meaning that was created before the streams
82 * were, delete it.
83 */
84 if (channel->uchan) {
85 lttng_ustconsumer_del_channel(channel);
86 lttng_ustconsumer_free_channel(channel);
87 }
88 free(channel);
89 }
90
91 /*
92 * Add channel to internal consumer state.
93 *
94 * Returns 0 on success or else a negative value.
95 */
96 static int add_channel(struct lttng_consumer_channel *channel,
97 struct lttng_consumer_local_data *ctx)
98 {
99 int ret = 0;
100
101 assert(channel);
102 assert(ctx);
103
104 if (ctx->on_recv_channel != NULL) {
105 ret = ctx->on_recv_channel(channel);
106 if (ret == 0) {
107 ret = consumer_add_channel(channel, ctx);
108 } else if (ret < 0) {
109 /* Most likely an ENOMEM. */
110 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
111 goto error;
112 }
113 } else {
114 ret = consumer_add_channel(channel, ctx);
115 }
116
117 DBG("UST consumer channel added (key: %" PRIu64 ")", channel->key);
118
119 error:
120 return ret;
121 }
122
123 /*
124 * Allocate and return a consumer channel object.
125 */
126 static struct lttng_consumer_channel *allocate_channel(uint64_t session_id,
127 const uint64_t *chunk_id, const char *pathname, const char *name,
128 uint64_t relayd_id, uint64_t key, enum lttng_event_output output,
129 uint64_t tracefile_size, uint64_t tracefile_count,
130 uint64_t session_id_per_pid, unsigned int monitor,
131 unsigned int live_timer_interval,
132 const char *root_shm_path, const char *shm_path)
133 {
134 assert(pathname);
135 assert(name);
136
137 return consumer_allocate_channel(key, session_id, chunk_id, pathname,
138 name, relayd_id, output, tracefile_size,
139 tracefile_count, session_id_per_pid, monitor,
140 live_timer_interval, root_shm_path, shm_path);
141 }
142
143 /*
144 * Allocate and return a consumer stream object. If _alloc_ret is not NULL, the
145 * error value if applicable is set in it else it is kept untouched.
146 *
147 * Return NULL on error else the newly allocated stream object.
148 */
149 static struct lttng_consumer_stream *allocate_stream(int cpu, int key,
150 struct lttng_consumer_channel *channel,
151 struct lttng_consumer_local_data *ctx, int *_alloc_ret)
152 {
153 int alloc_ret;
154 struct lttng_consumer_stream *stream = NULL;
155
156 assert(channel);
157 assert(ctx);
158
159 stream = consumer_allocate_stream(channel->key,
160 key,
161 channel->name,
162 channel->relayd_id,
163 channel->session_id,
164 channel->trace_chunk,
165 cpu,
166 &alloc_ret,
167 channel->type,
168 channel->monitor);
169 if (stream == NULL) {
170 switch (alloc_ret) {
171 case -ENOENT:
172 /*
173 * We could not find the channel. Can happen if cpu hotplug
174 * happens while tearing down.
175 */
176 DBG3("Could not find channel");
177 break;
178 case -ENOMEM:
179 case -EINVAL:
180 default:
181 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
182 break;
183 }
184 goto error;
185 }
186
187 consumer_stream_update_channel_attributes(stream, channel);
188 stream->chan = channel;
189
190 error:
191 if (_alloc_ret) {
192 *_alloc_ret = alloc_ret;
193 }
194 return stream;
195 }
196
197 /*
198 * Send the given stream pointer to the corresponding thread.
199 *
200 * Returns 0 on success else a negative value.
201 */
202 static int send_stream_to_thread(struct lttng_consumer_stream *stream,
203 struct lttng_consumer_local_data *ctx)
204 {
205 int ret;
206 struct lttng_pipe *stream_pipe;
207
208 /* Get the right pipe where the stream will be sent. */
209 if (stream->metadata_flag) {
210 consumer_add_metadata_stream(stream);
211 stream_pipe = ctx->consumer_metadata_pipe;
212 } else {
213 consumer_add_data_stream(stream);
214 stream_pipe = ctx->consumer_data_pipe;
215 }
216
217 /*
218 * From this point on, the stream's ownership has been moved away from
219 * the channel and it becomes globally visible. Hence, remove it from
220 * the local stream list to prevent the stream from being both local and
221 * global.
222 */
223 stream->globally_visible = 1;
224 cds_list_del(&stream->send_node);
225
226 ret = lttng_pipe_write(stream_pipe, &stream, sizeof(stream));
227 if (ret < 0) {
228 ERR("Consumer write %s stream to pipe %d",
229 stream->metadata_flag ? "metadata" : "data",
230 lttng_pipe_get_writefd(stream_pipe));
231 if (stream->metadata_flag) {
232 consumer_del_stream_for_metadata(stream);
233 } else {
234 consumer_del_stream_for_data(stream);
235 }
236 goto error;
237 }
238
239 error:
240 return ret;
241 }
242
243 static
244 int get_stream_shm_path(char *stream_shm_path, const char *shm_path, int cpu)
245 {
246 char cpu_nr[INT_MAX_STR_LEN]; /* int max len */
247 int ret;
248
249 strncpy(stream_shm_path, shm_path, PATH_MAX);
250 stream_shm_path[PATH_MAX - 1] = '\0';
251 ret = snprintf(cpu_nr, INT_MAX_STR_LEN, "%i", cpu);
252 if (ret < 0) {
253 PERROR("snprintf");
254 goto end;
255 }
256 strncat(stream_shm_path, cpu_nr,
257 PATH_MAX - strlen(stream_shm_path) - 1);
258 ret = 0;
259 end:
260 return ret;
261 }
262
263 /*
264 * Create streams for the given channel using liblttng-ust-ctl.
265 * The channel lock must be acquired by the caller.
266 *
267 * Return 0 on success else a negative value.
268 */
269 static int create_ust_streams(struct lttng_consumer_channel *channel,
270 struct lttng_consumer_local_data *ctx)
271 {
272 int ret, cpu = 0;
273 struct ustctl_consumer_stream *ustream;
274 struct lttng_consumer_stream *stream;
275 pthread_mutex_t *current_stream_lock = NULL;
276
277 assert(channel);
278 assert(ctx);
279
280 /*
281 * While a stream is available from ustctl. When NULL is returned, we've
282 * reached the end of the possible stream for the channel.
283 */
284 while ((ustream = ustctl_create_stream(channel->uchan, cpu))) {
285 int wait_fd;
286 int ust_metadata_pipe[2];
287
288 health_code_update();
289
290 if (channel->type == CONSUMER_CHANNEL_TYPE_METADATA && channel->monitor) {
291 ret = utils_create_pipe_cloexec_nonblock(ust_metadata_pipe);
292 if (ret < 0) {
293 ERR("Create ust metadata poll pipe");
294 goto error;
295 }
296 wait_fd = ust_metadata_pipe[0];
297 } else {
298 wait_fd = ustctl_stream_get_wait_fd(ustream);
299 }
300
301 /* Allocate consumer stream object. */
302 stream = allocate_stream(cpu, wait_fd, channel, ctx, &ret);
303 if (!stream) {
304 goto error_alloc;
305 }
306 stream->ustream = ustream;
307 /*
308 * Store it so we can save multiple function calls afterwards since
309 * this value is used heavily in the stream threads. This is UST
310 * specific so this is why it's done after allocation.
311 */
312 stream->wait_fd = wait_fd;
313
314 /*
315 * Increment channel refcount since the channel reference has now been
316 * assigned in the allocation process above.
317 */
318 if (stream->chan->monitor) {
319 uatomic_inc(&stream->chan->refcount);
320 }
321
322 pthread_mutex_lock(&stream->lock);
323 current_stream_lock = &stream->lock;
324 /*
325 * Order is important this is why a list is used. On error, the caller
326 * should clean this list.
327 */
328 cds_list_add_tail(&stream->send_node, &channel->streams.head);
329
330 ret = ustctl_get_max_subbuf_size(stream->ustream,
331 &stream->max_sb_size);
332 if (ret < 0) {
333 ERR("ustctl_get_max_subbuf_size failed for stream %s",
334 stream->name);
335 goto error;
336 }
337
338 /* Do actions once stream has been received. */
339 if (ctx->on_recv_stream) {
340 ret = ctx->on_recv_stream(stream);
341 if (ret < 0) {
342 goto error;
343 }
344 }
345
346 DBG("UST consumer add stream %s (key: %" PRIu64 ") with relayd id %" PRIu64,
347 stream->name, stream->key, stream->relayd_stream_id);
348
349 /* Set next CPU stream. */
350 channel->streams.count = ++cpu;
351
352 /* Keep stream reference when creating metadata. */
353 if (channel->type == CONSUMER_CHANNEL_TYPE_METADATA) {
354 channel->metadata_stream = stream;
355 if (channel->monitor) {
356 /* Set metadata poll pipe if we created one */
357 memcpy(stream->ust_metadata_poll_pipe,
358 ust_metadata_pipe,
359 sizeof(ust_metadata_pipe));
360 }
361 }
362 pthread_mutex_unlock(&stream->lock);
363 current_stream_lock = NULL;
364 }
365
366 return 0;
367
368 error:
369 error_alloc:
370 if (current_stream_lock) {
371 pthread_mutex_unlock(current_stream_lock);
372 }
373 return ret;
374 }
375
376 /*
377 * create_posix_shm is never called concurrently within a process.
378 */
379 static
380 int create_posix_shm(void)
381 {
382 char tmp_name[NAME_MAX];
383 int shmfd, ret;
384
385 ret = snprintf(tmp_name, NAME_MAX, "/ust-shm-consumer-%d", getpid());
386 if (ret < 0) {
387 PERROR("snprintf");
388 return -1;
389 }
390 /*
391 * Allocate shm, and immediately unlink its shm oject, keeping
392 * only the file descriptor as a reference to the object.
393 * We specifically do _not_ use the / at the beginning of the
394 * pathname so that some OS implementations can keep it local to
395 * the process (POSIX leaves this implementation-defined).
396 */
397 shmfd = shm_open(tmp_name, O_CREAT | O_EXCL | O_RDWR, 0700);
398 if (shmfd < 0) {
399 PERROR("shm_open");
400 goto error_shm_open;
401 }
402 ret = shm_unlink(tmp_name);
403 if (ret < 0 && errno != ENOENT) {
404 PERROR("shm_unlink");
405 goto error_shm_release;
406 }
407 return shmfd;
408
409 error_shm_release:
410 ret = close(shmfd);
411 if (ret) {
412 PERROR("close");
413 }
414 error_shm_open:
415 return -1;
416 }
417
418 static int open_ust_stream_fd(struct lttng_consumer_channel *channel, int cpu,
419 const struct lttng_credentials *session_credentials)
420 {
421 char shm_path[PATH_MAX];
422 int ret;
423
424 if (!channel->shm_path[0]) {
425 return create_posix_shm();
426 }
427 ret = get_stream_shm_path(shm_path, channel->shm_path, cpu);
428 if (ret) {
429 goto error_shm_path;
430 }
431 return run_as_open(shm_path,
432 O_RDWR | O_CREAT | O_EXCL, S_IRUSR | S_IWUSR,
433 session_credentials->uid, session_credentials->gid);
434
435 error_shm_path:
436 return -1;
437 }
438
439 /*
440 * Create an UST channel with the given attributes and send it to the session
441 * daemon using the ust ctl API.
442 *
443 * Return 0 on success or else a negative value.
444 */
445 static int create_ust_channel(struct lttng_consumer_channel *channel,
446 struct ustctl_consumer_channel_attr *attr,
447 struct ustctl_consumer_channel **ust_chanp)
448 {
449 int ret, nr_stream_fds, i, j;
450 int *stream_fds;
451 struct ustctl_consumer_channel *ust_channel;
452
453 assert(channel);
454 assert(attr);
455 assert(ust_chanp);
456 assert(channel->buffer_credentials.is_set);
457
458 DBG3("Creating channel to ustctl with attr: [overwrite: %d, "
459 "subbuf_size: %" PRIu64 ", num_subbuf: %" PRIu64 ", "
460 "switch_timer_interval: %u, read_timer_interval: %u, "
461 "output: %d, type: %d", attr->overwrite, attr->subbuf_size,
462 attr->num_subbuf, attr->switch_timer_interval,
463 attr->read_timer_interval, attr->output, attr->type);
464
465 if (channel->type == CONSUMER_CHANNEL_TYPE_METADATA)
466 nr_stream_fds = 1;
467 else
468 nr_stream_fds = ustctl_get_nr_stream_per_channel();
469 stream_fds = zmalloc(nr_stream_fds * sizeof(*stream_fds));
470 if (!stream_fds) {
471 ret = -1;
472 goto error_alloc;
473 }
474 for (i = 0; i < nr_stream_fds; i++) {
475 stream_fds[i] = open_ust_stream_fd(channel, i,
476 &channel->buffer_credentials.value);
477 if (stream_fds[i] < 0) {
478 ret = -1;
479 goto error_open;
480 }
481 }
482 ust_channel = ustctl_create_channel(attr, stream_fds, nr_stream_fds);
483 if (!ust_channel) {
484 ret = -1;
485 goto error_create;
486 }
487 channel->nr_stream_fds = nr_stream_fds;
488 channel->stream_fds = stream_fds;
489 *ust_chanp = ust_channel;
490
491 return 0;
492
493 error_create:
494 error_open:
495 for (j = i - 1; j >= 0; j--) {
496 int closeret;
497
498 closeret = close(stream_fds[j]);
499 if (closeret) {
500 PERROR("close");
501 }
502 if (channel->shm_path[0]) {
503 char shm_path[PATH_MAX];
504
505 closeret = get_stream_shm_path(shm_path,
506 channel->shm_path, j);
507 if (closeret) {
508 ERR("Cannot get stream shm path");
509 }
510 closeret = run_as_unlink(shm_path,
511 channel->buffer_credentials.value.uid,
512 channel->buffer_credentials.value.gid);
513 if (closeret) {
514 PERROR("unlink %s", shm_path);
515 }
516 }
517 }
518 /* Try to rmdir all directories under shm_path root. */
519 if (channel->root_shm_path[0]) {
520 (void) run_as_rmdir_recursive(channel->root_shm_path,
521 channel->buffer_credentials.value.uid,
522 channel->buffer_credentials.value.gid);
523 }
524 free(stream_fds);
525 error_alloc:
526 return ret;
527 }
528
529 /*
530 * Send a single given stream to the session daemon using the sock.
531 *
532 * Return 0 on success else a negative value.
533 */
534 static int send_sessiond_stream(int sock, struct lttng_consumer_stream *stream)
535 {
536 int ret;
537
538 assert(stream);
539 assert(sock >= 0);
540
541 DBG("UST consumer sending stream %" PRIu64 " to sessiond", stream->key);
542
543 /* Send stream to session daemon. */
544 ret = ustctl_send_stream_to_sessiond(sock, stream->ustream);
545 if (ret < 0) {
546 goto error;
547 }
548
549 error:
550 return ret;
551 }
552
553 /*
554 * Send channel to sessiond and relayd if applicable.
555 *
556 * Return 0 on success or else a negative value.
557 */
558 static int send_channel_to_sessiond_and_relayd(int sock,
559 struct lttng_consumer_channel *channel,
560 struct lttng_consumer_local_data *ctx, int *relayd_error)
561 {
562 int ret, ret_code = LTTCOMM_CONSUMERD_SUCCESS;
563 struct lttng_consumer_stream *stream;
564 uint64_t net_seq_idx = -1ULL;
565
566 assert(channel);
567 assert(ctx);
568 assert(sock >= 0);
569
570 DBG("UST consumer sending channel %s to sessiond", channel->name);
571
572 if (channel->relayd_id != (uint64_t) -1ULL) {
573 cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
574
575 health_code_update();
576
577 /* Try to send the stream to the relayd if one is available. */
578 DBG("Sending stream %" PRIu64 " of channel \"%s\" to relayd",
579 stream->key, channel->name);
580 ret = consumer_send_relayd_stream(stream, stream->chan->pathname);
581 if (ret < 0) {
582 /*
583 * Flag that the relayd was the problem here probably due to a
584 * communicaton error on the socket.
585 */
586 if (relayd_error) {
587 *relayd_error = 1;
588 }
589 ret_code = LTTCOMM_CONSUMERD_RELAYD_FAIL;
590 }
591 if (net_seq_idx == -1ULL) {
592 net_seq_idx = stream->net_seq_idx;
593 }
594 }
595 }
596
597 /* Inform sessiond that we are about to send channel and streams. */
598 ret = consumer_send_status_msg(sock, ret_code);
599 if (ret < 0 || ret_code != LTTCOMM_CONSUMERD_SUCCESS) {
600 /*
601 * Either the session daemon is not responding or the relayd died so we
602 * stop now.
603 */
604 goto error;
605 }
606
607 /* Send channel to sessiond. */
608 ret = ustctl_send_channel_to_sessiond(sock, channel->uchan);
609 if (ret < 0) {
610 goto error;
611 }
612
613 ret = ustctl_channel_close_wakeup_fd(channel->uchan);
614 if (ret < 0) {
615 goto error;
616 }
617
618 /* The channel was sent successfully to the sessiond at this point. */
619 cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
620
621 health_code_update();
622
623 /* Send stream to session daemon. */
624 ret = send_sessiond_stream(sock, stream);
625 if (ret < 0) {
626 goto error;
627 }
628 }
629
630 /* Tell sessiond there is no more stream. */
631 ret = ustctl_send_stream_to_sessiond(sock, NULL);
632 if (ret < 0) {
633 goto error;
634 }
635
636 DBG("UST consumer NULL stream sent to sessiond");
637
638 return 0;
639
640 error:
641 if (ret_code != LTTCOMM_CONSUMERD_SUCCESS) {
642 ret = -1;
643 }
644 return ret;
645 }
646
647 /*
648 * Creates a channel and streams and add the channel it to the channel internal
649 * state. The created stream must ONLY be sent once the GET_CHANNEL command is
650 * received.
651 *
652 * Return 0 on success or else, a negative value is returned and the channel
653 * MUST be destroyed by consumer_del_channel().
654 */
655 static int ask_channel(struct lttng_consumer_local_data *ctx,
656 struct lttng_consumer_channel *channel,
657 struct ustctl_consumer_channel_attr *attr)
658 {
659 int ret;
660
661 assert(ctx);
662 assert(channel);
663 assert(attr);
664
665 /*
666 * This value is still used by the kernel consumer since for the kernel,
667 * the stream ownership is not IN the consumer so we need to have the
668 * number of left stream that needs to be initialized so we can know when
669 * to delete the channel (see consumer.c).
670 *
671 * As for the user space tracer now, the consumer creates and sends the
672 * stream to the session daemon which only sends them to the application
673 * once every stream of a channel is received making this value useless
674 * because we they will be added to the poll thread before the application
675 * receives them. This ensures that a stream can not hang up during
676 * initilization of a channel.
677 */
678 channel->nb_init_stream_left = 0;
679
680 /* The reply msg status is handled in the following call. */
681 ret = create_ust_channel(channel, attr, &channel->uchan);
682 if (ret < 0) {
683 goto end;
684 }
685
686 channel->wait_fd = ustctl_channel_get_wait_fd(channel->uchan);
687
688 /*
689 * For the snapshots (no monitor), we create the metadata streams
690 * on demand, not during the channel creation.
691 */
692 if (channel->type == CONSUMER_CHANNEL_TYPE_METADATA && !channel->monitor) {
693 ret = 0;
694 goto end;
695 }
696
697 /* Open all streams for this channel. */
698 pthread_mutex_lock(&channel->lock);
699 ret = create_ust_streams(channel, ctx);
700 pthread_mutex_unlock(&channel->lock);
701 if (ret < 0) {
702 goto end;
703 }
704
705 end:
706 return ret;
707 }
708
709 /*
710 * Send all stream of a channel to the right thread handling it.
711 *
712 * On error, return a negative value else 0 on success.
713 */
714 static int send_streams_to_thread(struct lttng_consumer_channel *channel,
715 struct lttng_consumer_local_data *ctx)
716 {
717 int ret = 0;
718 struct lttng_consumer_stream *stream, *stmp;
719
720 assert(channel);
721 assert(ctx);
722
723 /* Send streams to the corresponding thread. */
724 cds_list_for_each_entry_safe(stream, stmp, &channel->streams.head,
725 send_node) {
726
727 health_code_update();
728
729 /* Sending the stream to the thread. */
730 ret = send_stream_to_thread(stream, ctx);
731 if (ret < 0) {
732 /*
733 * If we are unable to send the stream to the thread, there is
734 * a big problem so just stop everything.
735 */
736 goto error;
737 }
738 }
739
740 error:
741 return ret;
742 }
743
744 /*
745 * Flush channel's streams using the given key to retrieve the channel.
746 *
747 * Return 0 on success else an LTTng error code.
748 */
749 static int flush_channel(uint64_t chan_key)
750 {
751 int ret = 0;
752 struct lttng_consumer_channel *channel;
753 struct lttng_consumer_stream *stream;
754 struct lttng_ht *ht;
755 struct lttng_ht_iter iter;
756
757 DBG("UST consumer flush channel key %" PRIu64, chan_key);
758
759 rcu_read_lock();
760 channel = consumer_find_channel(chan_key);
761 if (!channel) {
762 ERR("UST consumer flush channel %" PRIu64 " not found", chan_key);
763 ret = LTTNG_ERR_UST_CHAN_NOT_FOUND;
764 goto error;
765 }
766
767 ht = consumer_data.stream_per_chan_id_ht;
768
769 /* For each stream of the channel id, flush it. */
770 cds_lfht_for_each_entry_duplicate(ht->ht,
771 ht->hash_fct(&channel->key, lttng_ht_seed), ht->match_fct,
772 &channel->key, &iter.iter, stream, node_channel_id.node) {
773
774 health_code_update();
775
776 pthread_mutex_lock(&stream->lock);
777 if (!stream->quiescent) {
778 ustctl_flush_buffer(stream->ustream, 0);
779 stream->quiescent = true;
780 }
781 pthread_mutex_unlock(&stream->lock);
782 }
783 error:
784 rcu_read_unlock();
785 return ret;
786 }
787
788 /*
789 * Clear quiescent state from channel's streams using the given key to
790 * retrieve the channel.
791 *
792 * Return 0 on success else an LTTng error code.
793 */
794 static int clear_quiescent_channel(uint64_t chan_key)
795 {
796 int ret = 0;
797 struct lttng_consumer_channel *channel;
798 struct lttng_consumer_stream *stream;
799 struct lttng_ht *ht;
800 struct lttng_ht_iter iter;
801
802 DBG("UST consumer clear quiescent channel key %" PRIu64, chan_key);
803
804 rcu_read_lock();
805 channel = consumer_find_channel(chan_key);
806 if (!channel) {
807 ERR("UST consumer clear quiescent channel %" PRIu64 " not found", chan_key);
808 ret = LTTNG_ERR_UST_CHAN_NOT_FOUND;
809 goto error;
810 }
811
812 ht = consumer_data.stream_per_chan_id_ht;
813
814 /* For each stream of the channel id, clear quiescent state. */
815 cds_lfht_for_each_entry_duplicate(ht->ht,
816 ht->hash_fct(&channel->key, lttng_ht_seed), ht->match_fct,
817 &channel->key, &iter.iter, stream, node_channel_id.node) {
818
819 health_code_update();
820
821 pthread_mutex_lock(&stream->lock);
822 stream->quiescent = false;
823 pthread_mutex_unlock(&stream->lock);
824 }
825 error:
826 rcu_read_unlock();
827 return ret;
828 }
829
830 /*
831 * Close metadata stream wakeup_fd using the given key to retrieve the channel.
832 *
833 * Return 0 on success else an LTTng error code.
834 */
835 static int close_metadata(uint64_t chan_key)
836 {
837 int ret = 0;
838 struct lttng_consumer_channel *channel;
839 unsigned int channel_monitor;
840
841 DBG("UST consumer close metadata key %" PRIu64, chan_key);
842
843 channel = consumer_find_channel(chan_key);
844 if (!channel) {
845 /*
846 * This is possible if the metadata thread has issue a delete because
847 * the endpoint point of the stream hung up. There is no way the
848 * session daemon can know about it thus use a DBG instead of an actual
849 * error.
850 */
851 DBG("UST consumer close metadata %" PRIu64 " not found", chan_key);
852 ret = LTTNG_ERR_UST_CHAN_NOT_FOUND;
853 goto error;
854 }
855
856 pthread_mutex_lock(&consumer_data.lock);
857 pthread_mutex_lock(&channel->lock);
858 channel_monitor = channel->monitor;
859 if (cds_lfht_is_node_deleted(&channel->node.node)) {
860 goto error_unlock;
861 }
862
863 lttng_ustconsumer_close_metadata(channel);
864 pthread_mutex_unlock(&channel->lock);
865 pthread_mutex_unlock(&consumer_data.lock);
866
867 /*
868 * The ownership of a metadata channel depends on the type of
869 * session to which it belongs. In effect, the monitor flag is checked
870 * to determine if this metadata channel is in "snapshot" mode or not.
871 *
872 * In the non-snapshot case, the metadata channel is created along with
873 * a single stream which will remain present until the metadata channel
874 * is destroyed (on the destruction of its session). In this case, the
875 * metadata stream in "monitored" by the metadata poll thread and holds
876 * the ownership of its channel.
877 *
878 * Closing the metadata will cause the metadata stream's "metadata poll
879 * pipe" to be closed. Closing this pipe will wake-up the metadata poll
880 * thread which will teardown the metadata stream which, in return,
881 * deletes the metadata channel.
882 *
883 * In the snapshot case, the metadata stream is created and destroyed
884 * on every snapshot record. Since the channel doesn't have an owner
885 * other than the session daemon, it is safe to destroy it immediately
886 * on reception of the CLOSE_METADATA command.
887 */
888 if (!channel_monitor) {
889 /*
890 * The channel and consumer_data locks must be
891 * released before this call since consumer_del_channel
892 * re-acquires the channel and consumer_data locks to teardown
893 * the channel and queue its reclamation by the "call_rcu"
894 * worker thread.
895 */
896 consumer_del_channel(channel);
897 }
898
899 return ret;
900 error_unlock:
901 pthread_mutex_unlock(&channel->lock);
902 pthread_mutex_unlock(&consumer_data.lock);
903 error:
904 return ret;
905 }
906
907 /*
908 * RCU read side lock MUST be acquired before calling this function.
909 *
910 * Return 0 on success else an LTTng error code.
911 */
912 static int setup_metadata(struct lttng_consumer_local_data *ctx, uint64_t key)
913 {
914 int ret;
915 struct lttng_consumer_channel *metadata;
916
917 DBG("UST consumer setup metadata key %" PRIu64, key);
918
919 metadata = consumer_find_channel(key);
920 if (!metadata) {
921 ERR("UST consumer push metadata %" PRIu64 " not found", key);
922 ret = LTTNG_ERR_UST_CHAN_NOT_FOUND;
923 goto end;
924 }
925
926 /*
927 * In no monitor mode, the metadata channel has no stream(s) so skip the
928 * ownership transfer to the metadata thread.
929 */
930 if (!metadata->monitor) {
931 DBG("Metadata channel in no monitor");
932 ret = 0;
933 goto end;
934 }
935
936 /*
937 * Send metadata stream to relayd if one available. Availability is
938 * known if the stream is still in the list of the channel.
939 */
940 if (cds_list_empty(&metadata->streams.head)) {
941 ERR("Metadata channel key %" PRIu64 ", no stream available.", key);
942 ret = LTTCOMM_CONSUMERD_ERROR_METADATA;
943 goto error_no_stream;
944 }
945
946 /* Send metadata stream to relayd if needed. */
947 if (metadata->metadata_stream->net_seq_idx != (uint64_t) -1ULL) {
948 ret = consumer_send_relayd_stream(metadata->metadata_stream,
949 metadata->pathname);
950 if (ret < 0) {
951 ret = LTTCOMM_CONSUMERD_ERROR_METADATA;
952 goto error;
953 }
954 ret = consumer_send_relayd_streams_sent(
955 metadata->metadata_stream->net_seq_idx);
956 if (ret < 0) {
957 ret = LTTCOMM_CONSUMERD_RELAYD_FAIL;
958 goto error;
959 }
960 }
961
962 /*
963 * Ownership of metadata stream is passed along. Freeing is handled by
964 * the callee.
965 */
966 ret = send_streams_to_thread(metadata, ctx);
967 if (ret < 0) {
968 /*
969 * If we are unable to send the stream to the thread, there is
970 * a big problem so just stop everything.
971 */
972 ret = LTTCOMM_CONSUMERD_FATAL;
973 goto send_streams_error;
974 }
975 /* List MUST be empty after or else it could be reused. */
976 assert(cds_list_empty(&metadata->streams.head));
977
978 ret = 0;
979 goto end;
980
981 error:
982 /*
983 * Delete metadata channel on error. At this point, the metadata stream can
984 * NOT be monitored by the metadata thread thus having the guarantee that
985 * the stream is still in the local stream list of the channel. This call
986 * will make sure to clean that list.
987 */
988 consumer_stream_destroy(metadata->metadata_stream, NULL);
989 cds_list_del(&metadata->metadata_stream->send_node);
990 metadata->metadata_stream = NULL;
991 send_streams_error:
992 error_no_stream:
993 end:
994 return ret;
995 }
996
997 /*
998 * Snapshot the whole metadata.
999 * RCU read-side lock must be held by the caller.
1000 *
1001 * Returns 0 on success, < 0 on error
1002 */
1003 static int snapshot_metadata(struct lttng_consumer_channel *metadata_channel,
1004 uint64_t key, char *path, uint64_t relayd_id,
1005 struct lttng_consumer_local_data *ctx)
1006 {
1007 int ret = 0;
1008 struct lttng_consumer_stream *metadata_stream;
1009
1010 assert(path);
1011 assert(ctx);
1012
1013 DBG("UST consumer snapshot metadata with key %" PRIu64 " at path %s",
1014 key, path);
1015
1016 rcu_read_lock();
1017
1018 assert(!metadata_channel->monitor);
1019
1020 health_code_update();
1021
1022 /*
1023 * Ask the sessiond if we have new metadata waiting and update the
1024 * consumer metadata cache.
1025 */
1026 ret = lttng_ustconsumer_request_metadata(ctx, metadata_channel, 0, 1);
1027 if (ret < 0) {
1028 goto error;
1029 }
1030
1031 health_code_update();
1032
1033 /*
1034 * The metadata stream is NOT created in no monitor mode when the channel
1035 * is created on a sessiond ask channel command.
1036 */
1037 ret = create_ust_streams(metadata_channel, ctx);
1038 if (ret < 0) {
1039 goto error;
1040 }
1041
1042 metadata_stream = metadata_channel->metadata_stream;
1043 assert(metadata_stream);
1044
1045 pthread_mutex_lock(&metadata_stream->lock);
1046 if (relayd_id != (uint64_t) -1ULL) {
1047 metadata_stream->net_seq_idx = relayd_id;
1048 ret = consumer_send_relayd_stream(metadata_stream, path);
1049 } else {
1050 ret = consumer_stream_create_output_files(metadata_stream,
1051 false);
1052 }
1053 pthread_mutex_unlock(&metadata_stream->lock);
1054 if (ret < 0) {
1055 goto error_stream;
1056 }
1057
1058 do {
1059 health_code_update();
1060
1061 ret = lttng_consumer_read_subbuffer(metadata_stream, ctx);
1062 if (ret < 0) {
1063 goto error_stream;
1064 }
1065 } while (ret > 0);
1066
1067 error_stream:
1068 /*
1069 * Clean up the stream completly because the next snapshot will use a new
1070 * metadata stream.
1071 */
1072 pthread_mutex_lock(&metadata_stream->lock);
1073 consumer_stream_destroy(metadata_stream, NULL);
1074 cds_list_del(&metadata_stream->send_node);
1075 metadata_channel->metadata_stream = NULL;
1076
1077 error:
1078 rcu_read_unlock();
1079 return ret;
1080 }
1081
1082 /*
1083 * Take a snapshot of all the stream of a channel.
1084 * RCU read-side lock and the channel lock must be held by the caller.
1085 *
1086 * Returns 0 on success, < 0 on error
1087 */
1088 static int snapshot_channel(struct lttng_consumer_channel *channel,
1089 uint64_t key, char *path, uint64_t relayd_id,
1090 uint64_t nb_packets_per_stream,
1091 struct lttng_consumer_local_data *ctx)
1092 {
1093 int ret;
1094 unsigned use_relayd = 0;
1095 unsigned long consumed_pos, produced_pos;
1096 struct lttng_consumer_stream *stream;
1097
1098 assert(path);
1099 assert(ctx);
1100
1101 rcu_read_lock();
1102
1103 if (relayd_id != (uint64_t) -1ULL) {
1104 use_relayd = 1;
1105 }
1106
1107 assert(!channel->monitor);
1108 DBG("UST consumer snapshot channel %" PRIu64, key);
1109
1110 cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
1111 health_code_update();
1112
1113 /* Lock stream because we are about to change its state. */
1114 pthread_mutex_lock(&stream->lock);
1115 assert(channel->trace_chunk);
1116 if (!lttng_trace_chunk_get(channel->trace_chunk)) {
1117 /*
1118 * Can't happen barring an internal error as the channel
1119 * holds a reference to the trace chunk.
1120 */
1121 ERR("Failed to acquire reference to channel's trace chunk");
1122 ret = -1;
1123 goto error_unlock;
1124 }
1125 assert(!stream->trace_chunk);
1126 stream->trace_chunk = channel->trace_chunk;
1127
1128 stream->net_seq_idx = relayd_id;
1129
1130 if (use_relayd) {
1131 ret = consumer_send_relayd_stream(stream, path);
1132 if (ret < 0) {
1133 goto error_unlock;
1134 }
1135 } else {
1136 ret = consumer_stream_create_output_files(stream,
1137 false);
1138 if (ret < 0) {
1139 goto error_unlock;
1140 }
1141 DBG("UST consumer snapshot stream (%" PRIu64 ")",
1142 stream->key);
1143 }
1144
1145 /*
1146 * If tracing is active, we want to perform a "full" buffer flush.
1147 * Else, if quiescent, it has already been done by the prior stop.
1148 */
1149 if (!stream->quiescent) {
1150 ustctl_flush_buffer(stream->ustream, 0);
1151 }
1152
1153 ret = lttng_ustconsumer_take_snapshot(stream);
1154 if (ret < 0) {
1155 ERR("Taking UST snapshot");
1156 goto error_unlock;
1157 }
1158
1159 ret = lttng_ustconsumer_get_produced_snapshot(stream, &produced_pos);
1160 if (ret < 0) {
1161 ERR("Produced UST snapshot position");
1162 goto error_unlock;
1163 }
1164
1165 ret = lttng_ustconsumer_get_consumed_snapshot(stream, &consumed_pos);
1166 if (ret < 0) {
1167 ERR("Consumerd UST snapshot position");
1168 goto error_unlock;
1169 }
1170
1171 /*
1172 * The original value is sent back if max stream size is larger than
1173 * the possible size of the snapshot. Also, we assume that the session
1174 * daemon should never send a maximum stream size that is lower than
1175 * subbuffer size.
1176 */
1177 consumed_pos = consumer_get_consume_start_pos(consumed_pos,
1178 produced_pos, nb_packets_per_stream,
1179 stream->max_sb_size);
1180
1181 while ((long) (consumed_pos - produced_pos) < 0) {
1182 ssize_t read_len;
1183 unsigned long len, padded_len;
1184
1185 health_code_update();
1186
1187 DBG("UST consumer taking snapshot at pos %lu", consumed_pos);
1188
1189 ret = ustctl_get_subbuf(stream->ustream, &consumed_pos);
1190 if (ret < 0) {
1191 if (ret != -EAGAIN) {
1192 PERROR("ustctl_get_subbuf snapshot");
1193 goto error_close_stream;
1194 }
1195 DBG("UST consumer get subbuf failed. Skipping it.");
1196 consumed_pos += stream->max_sb_size;
1197 stream->chan->lost_packets++;
1198 continue;
1199 }
1200
1201 ret = ustctl_get_subbuf_size(stream->ustream, &len);
1202 if (ret < 0) {
1203 ERR("Snapshot ustctl_get_subbuf_size");
1204 goto error_put_subbuf;
1205 }
1206
1207 ret = ustctl_get_padded_subbuf_size(stream->ustream, &padded_len);
1208 if (ret < 0) {
1209 ERR("Snapshot ustctl_get_padded_subbuf_size");
1210 goto error_put_subbuf;
1211 }
1212
1213 read_len = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, len,
1214 padded_len - len, NULL);
1215 if (use_relayd) {
1216 if (read_len != len) {
1217 ret = -EPERM;
1218 goto error_put_subbuf;
1219 }
1220 } else {
1221 if (read_len != padded_len) {
1222 ret = -EPERM;
1223 goto error_put_subbuf;
1224 }
1225 }
1226
1227 ret = ustctl_put_subbuf(stream->ustream);
1228 if (ret < 0) {
1229 ERR("Snapshot ustctl_put_subbuf");
1230 goto error_close_stream;
1231 }
1232 consumed_pos += stream->max_sb_size;
1233 }
1234
1235 /* Simply close the stream so we can use it on the next snapshot. */
1236 consumer_stream_close(stream);
1237 pthread_mutex_unlock(&stream->lock);
1238 }
1239
1240 rcu_read_unlock();
1241 return 0;
1242
1243 error_put_subbuf:
1244 if (ustctl_put_subbuf(stream->ustream) < 0) {
1245 ERR("Snapshot ustctl_put_subbuf");
1246 }
1247 error_close_stream:
1248 consumer_stream_close(stream);
1249 error_unlock:
1250 pthread_mutex_unlock(&stream->lock);
1251 rcu_read_unlock();
1252 return ret;
1253 }
1254
1255 /*
1256 * Receive the metadata updates from the sessiond. Supports receiving
1257 * overlapping metadata, but is needs to always belong to a contiguous
1258 * range starting from 0.
1259 * Be careful about the locks held when calling this function: it needs
1260 * the metadata cache flush to concurrently progress in order to
1261 * complete.
1262 */
1263 int lttng_ustconsumer_recv_metadata(int sock, uint64_t key, uint64_t offset,
1264 uint64_t len, uint64_t version,
1265 struct lttng_consumer_channel *channel, int timer, int wait)
1266 {
1267 int ret, ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1268 char *metadata_str;
1269
1270 DBG("UST consumer push metadata key %" PRIu64 " of len %" PRIu64, key, len);
1271
1272 metadata_str = zmalloc(len * sizeof(char));
1273 if (!metadata_str) {
1274 PERROR("zmalloc metadata string");
1275 ret_code = LTTCOMM_CONSUMERD_ENOMEM;
1276 goto end;
1277 }
1278
1279 health_code_update();
1280
1281 /* Receive metadata string. */
1282 ret = lttcomm_recv_unix_sock(sock, metadata_str, len);
1283 if (ret < 0) {
1284 /* Session daemon is dead so return gracefully. */
1285 ret_code = ret;
1286 goto end_free;
1287 }
1288
1289 health_code_update();
1290
1291 pthread_mutex_lock(&channel->metadata_cache->lock);
1292 ret = consumer_metadata_cache_write(channel, offset, len, version,
1293 metadata_str);
1294 if (ret < 0) {
1295 /* Unable to handle metadata. Notify session daemon. */
1296 ret_code = LTTCOMM_CONSUMERD_ERROR_METADATA;
1297 /*
1298 * Skip metadata flush on write error since the offset and len might
1299 * not have been updated which could create an infinite loop below when
1300 * waiting for the metadata cache to be flushed.
1301 */
1302 pthread_mutex_unlock(&channel->metadata_cache->lock);
1303 goto end_free;
1304 }
1305 pthread_mutex_unlock(&channel->metadata_cache->lock);
1306
1307 if (!wait) {
1308 goto end_free;
1309 }
1310 while (consumer_metadata_cache_flushed(channel, offset + len, timer)) {
1311 DBG("Waiting for metadata to be flushed");
1312
1313 health_code_update();
1314
1315 usleep(DEFAULT_METADATA_AVAILABILITY_WAIT_TIME);
1316 }
1317
1318 end_free:
1319 free(metadata_str);
1320 end:
1321 return ret_code;
1322 }
1323
1324 /*
1325 * Receive command from session daemon and process it.
1326 *
1327 * Return 1 on success else a negative value or 0.
1328 */
1329 int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
1330 int sock, struct pollfd *consumer_sockpoll)
1331 {
1332 ssize_t ret;
1333 enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1334 struct lttcomm_consumer_msg msg;
1335 struct lttng_consumer_channel *channel = NULL;
1336
1337 health_code_update();
1338
1339 ret = lttcomm_recv_unix_sock(sock, &msg, sizeof(msg));
1340 if (ret != sizeof(msg)) {
1341 DBG("Consumer received unexpected message size %zd (expects %zu)",
1342 ret, sizeof(msg));
1343 /*
1344 * The ret value might 0 meaning an orderly shutdown but this is ok
1345 * since the caller handles this.
1346 */
1347 if (ret > 0) {
1348 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD);
1349 ret = -1;
1350 }
1351 return ret;
1352 }
1353
1354 health_code_update();
1355
1356 /* deprecated */
1357 assert(msg.cmd_type != LTTNG_CONSUMER_STOP);
1358
1359 health_code_update();
1360
1361 /* relayd needs RCU read-side lock */
1362 rcu_read_lock();
1363
1364 switch (msg.cmd_type) {
1365 case LTTNG_CONSUMER_ADD_RELAYD_SOCKET:
1366 {
1367 /* Session daemon status message are handled in the following call. */
1368 consumer_add_relayd_socket(msg.u.relayd_sock.net_index,
1369 msg.u.relayd_sock.type, ctx, sock, consumer_sockpoll,
1370 &msg.u.relayd_sock.sock, msg.u.relayd_sock.session_id,
1371 msg.u.relayd_sock.relayd_session_id);
1372 goto end_nosignal;
1373 }
1374 case LTTNG_CONSUMER_DESTROY_RELAYD:
1375 {
1376 uint64_t index = msg.u.destroy_relayd.net_seq_idx;
1377 struct consumer_relayd_sock_pair *relayd;
1378
1379 DBG("UST consumer destroying relayd %" PRIu64, index);
1380
1381 /* Get relayd reference if exists. */
1382 relayd = consumer_find_relayd(index);
1383 if (relayd == NULL) {
1384 DBG("Unable to find relayd %" PRIu64, index);
1385 ret_code = LTTCOMM_CONSUMERD_RELAYD_FAIL;
1386 }
1387
1388 /*
1389 * Each relayd socket pair has a refcount of stream attached to it
1390 * which tells if the relayd is still active or not depending on the
1391 * refcount value.
1392 *
1393 * This will set the destroy flag of the relayd object and destroy it
1394 * if the refcount reaches zero when called.
1395 *
1396 * The destroy can happen either here or when a stream fd hangs up.
1397 */
1398 if (relayd) {
1399 consumer_flag_relayd_for_destroy(relayd);
1400 }
1401
1402 goto end_msg_sessiond;
1403 }
1404 case LTTNG_CONSUMER_UPDATE_STREAM:
1405 {
1406 rcu_read_unlock();
1407 return -ENOSYS;
1408 }
1409 case LTTNG_CONSUMER_DATA_PENDING:
1410 {
1411 int ret, is_data_pending;
1412 uint64_t id = msg.u.data_pending.session_id;
1413
1414 DBG("UST consumer data pending command for id %" PRIu64, id);
1415
1416 is_data_pending = consumer_data_pending(id);
1417
1418 /* Send back returned value to session daemon */
1419 ret = lttcomm_send_unix_sock(sock, &is_data_pending,
1420 sizeof(is_data_pending));
1421 if (ret < 0) {
1422 DBG("Error when sending the data pending ret code: %d", ret);
1423 goto error_fatal;
1424 }
1425
1426 /*
1427 * No need to send back a status message since the data pending
1428 * returned value is the response.
1429 */
1430 break;
1431 }
1432 case LTTNG_CONSUMER_ASK_CHANNEL_CREATION:
1433 {
1434 int ret;
1435 struct ustctl_consumer_channel_attr attr;
1436 const uint64_t chunk_id = msg.u.ask_channel.chunk_id.value;
1437 const struct lttng_credentials buffer_credentials = {
1438 .uid = msg.u.ask_channel.buffer_credentials.uid,
1439 .gid = msg.u.ask_channel.buffer_credentials.gid,
1440 };
1441
1442 /* Create a plain object and reserve a channel key. */
1443 channel = allocate_channel(msg.u.ask_channel.session_id,
1444 msg.u.ask_channel.chunk_id.is_set ?
1445 &chunk_id : NULL,
1446 msg.u.ask_channel.pathname,
1447 msg.u.ask_channel.name,
1448 msg.u.ask_channel.relayd_id,
1449 msg.u.ask_channel.key,
1450 (enum lttng_event_output) msg.u.ask_channel.output,
1451 msg.u.ask_channel.tracefile_size,
1452 msg.u.ask_channel.tracefile_count,
1453 msg.u.ask_channel.session_id_per_pid,
1454 msg.u.ask_channel.monitor,
1455 msg.u.ask_channel.live_timer_interval,
1456 msg.u.ask_channel.root_shm_path,
1457 msg.u.ask_channel.shm_path);
1458 if (!channel) {
1459 goto end_channel_error;
1460 }
1461
1462 LTTNG_OPTIONAL_SET(&channel->buffer_credentials,
1463 buffer_credentials);
1464
1465 /*
1466 * Assign UST application UID to the channel. This value is ignored for
1467 * per PID buffers. This is specific to UST thus setting this after the
1468 * allocation.
1469 */
1470 channel->ust_app_uid = msg.u.ask_channel.ust_app_uid;
1471
1472 /* Build channel attributes from received message. */
1473 attr.subbuf_size = msg.u.ask_channel.subbuf_size;
1474 attr.num_subbuf = msg.u.ask_channel.num_subbuf;
1475 attr.overwrite = msg.u.ask_channel.overwrite;
1476 attr.switch_timer_interval = msg.u.ask_channel.switch_timer_interval;
1477 attr.read_timer_interval = msg.u.ask_channel.read_timer_interval;
1478 attr.chan_id = msg.u.ask_channel.chan_id;
1479 memcpy(attr.uuid, msg.u.ask_channel.uuid, sizeof(attr.uuid));
1480 attr.blocking_timeout= msg.u.ask_channel.blocking_timeout;
1481
1482 /* Match channel buffer type to the UST abi. */
1483 switch (msg.u.ask_channel.output) {
1484 case LTTNG_EVENT_MMAP:
1485 default:
1486 attr.output = LTTNG_UST_MMAP;
1487 break;
1488 }
1489
1490 /* Translate and save channel type. */
1491 switch (msg.u.ask_channel.type) {
1492 case LTTNG_UST_CHAN_PER_CPU:
1493 channel->type = CONSUMER_CHANNEL_TYPE_DATA;
1494 attr.type = LTTNG_UST_CHAN_PER_CPU;
1495 /*
1496 * Set refcount to 1 for owner. Below, we will
1497 * pass ownership to the
1498 * consumer_thread_channel_poll() thread.
1499 */
1500 channel->refcount = 1;
1501 break;
1502 case LTTNG_UST_CHAN_METADATA:
1503 channel->type = CONSUMER_CHANNEL_TYPE_METADATA;
1504 attr.type = LTTNG_UST_CHAN_METADATA;
1505 break;
1506 default:
1507 assert(0);
1508 goto error_fatal;
1509 };
1510
1511 health_code_update();
1512
1513 ret = ask_channel(ctx, channel, &attr);
1514 if (ret < 0) {
1515 goto end_channel_error;
1516 }
1517
1518 if (msg.u.ask_channel.type == LTTNG_UST_CHAN_METADATA) {
1519 ret = consumer_metadata_cache_allocate(channel);
1520 if (ret < 0) {
1521 ERR("Allocating metadata cache");
1522 goto end_channel_error;
1523 }
1524 consumer_timer_switch_start(channel, attr.switch_timer_interval);
1525 attr.switch_timer_interval = 0;
1526 } else {
1527 int monitor_start_ret;
1528
1529 consumer_timer_live_start(channel,
1530 msg.u.ask_channel.live_timer_interval);
1531 monitor_start_ret = consumer_timer_monitor_start(
1532 channel,
1533 msg.u.ask_channel.monitor_timer_interval);
1534 if (monitor_start_ret < 0) {
1535 ERR("Starting channel monitoring timer failed");
1536 goto end_channel_error;
1537 }
1538 }
1539
1540 health_code_update();
1541
1542 /*
1543 * Add the channel to the internal state AFTER all streams were created
1544 * and successfully sent to session daemon. This way, all streams must
1545 * be ready before this channel is visible to the threads.
1546 * If add_channel succeeds, ownership of the channel is
1547 * passed to consumer_thread_channel_poll().
1548 */
1549 ret = add_channel(channel, ctx);
1550 if (ret < 0) {
1551 if (msg.u.ask_channel.type == LTTNG_UST_CHAN_METADATA) {
1552 if (channel->switch_timer_enabled == 1) {
1553 consumer_timer_switch_stop(channel);
1554 }
1555 consumer_metadata_cache_destroy(channel);
1556 }
1557 if (channel->live_timer_enabled == 1) {
1558 consumer_timer_live_stop(channel);
1559 }
1560 if (channel->monitor_timer_enabled == 1) {
1561 consumer_timer_monitor_stop(channel);
1562 }
1563 goto end_channel_error;
1564 }
1565
1566 health_code_update();
1567
1568 /*
1569 * Channel and streams are now created. Inform the session daemon that
1570 * everything went well and should wait to receive the channel and
1571 * streams with ustctl API.
1572 */
1573 ret = consumer_send_status_channel(sock, channel);
1574 if (ret < 0) {
1575 /*
1576 * There is probably a problem on the socket.
1577 */
1578 goto error_fatal;
1579 }
1580
1581 break;
1582 }
1583 case LTTNG_CONSUMER_GET_CHANNEL:
1584 {
1585 int ret, relayd_err = 0;
1586 uint64_t key = msg.u.get_channel.key;
1587 struct lttng_consumer_channel *channel;
1588
1589 channel = consumer_find_channel(key);
1590 if (!channel) {
1591 ERR("UST consumer get channel key %" PRIu64 " not found", key);
1592 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
1593 goto end_msg_sessiond;
1594 }
1595
1596 health_code_update();
1597
1598 /* Send the channel to sessiond (and relayd, if applicable). */
1599 ret = send_channel_to_sessiond_and_relayd(sock, channel, ctx,
1600 &relayd_err);
1601 if (ret < 0) {
1602 if (relayd_err) {
1603 /*
1604 * We were unable to send to the relayd the stream so avoid
1605 * sending back a fatal error to the thread since this is OK
1606 * and the consumer can continue its work. The above call
1607 * has sent the error status message to the sessiond.
1608 */
1609 goto end_nosignal;
1610 }
1611 /*
1612 * The communicaton was broken hence there is a bad state between
1613 * the consumer and sessiond so stop everything.
1614 */
1615 goto error_fatal;
1616 }
1617
1618 health_code_update();
1619
1620 /*
1621 * In no monitor mode, the streams ownership is kept inside the channel
1622 * so don't send them to the data thread.
1623 */
1624 if (!channel->monitor) {
1625 goto end_msg_sessiond;
1626 }
1627
1628 ret = send_streams_to_thread(channel, ctx);
1629 if (ret < 0) {
1630 /*
1631 * If we are unable to send the stream to the thread, there is
1632 * a big problem so just stop everything.
1633 */
1634 goto error_fatal;
1635 }
1636 /* List MUST be empty after or else it could be reused. */
1637 assert(cds_list_empty(&channel->streams.head));
1638 goto end_msg_sessiond;
1639 }
1640 case LTTNG_CONSUMER_DESTROY_CHANNEL:
1641 {
1642 uint64_t key = msg.u.destroy_channel.key;
1643
1644 /*
1645 * Only called if streams have not been sent to stream
1646 * manager thread. However, channel has been sent to
1647 * channel manager thread.
1648 */
1649 notify_thread_del_channel(ctx, key);
1650 goto end_msg_sessiond;
1651 }
1652 case LTTNG_CONSUMER_CLOSE_METADATA:
1653 {
1654 int ret;
1655
1656 ret = close_metadata(msg.u.close_metadata.key);
1657 if (ret != 0) {
1658 ret_code = ret;
1659 }
1660
1661 goto end_msg_sessiond;
1662 }
1663 case LTTNG_CONSUMER_FLUSH_CHANNEL:
1664 {
1665 int ret;
1666
1667 ret = flush_channel(msg.u.flush_channel.key);
1668 if (ret != 0) {
1669 ret_code = ret;
1670 }
1671
1672 goto end_msg_sessiond;
1673 }
1674 case LTTNG_CONSUMER_CLEAR_QUIESCENT_CHANNEL:
1675 {
1676 int ret;
1677
1678 ret = clear_quiescent_channel(
1679 msg.u.clear_quiescent_channel.key);
1680 if (ret != 0) {
1681 ret_code = ret;
1682 }
1683
1684 goto end_msg_sessiond;
1685 }
1686 case LTTNG_CONSUMER_PUSH_METADATA:
1687 {
1688 int ret;
1689 uint64_t len = msg.u.push_metadata.len;
1690 uint64_t key = msg.u.push_metadata.key;
1691 uint64_t offset = msg.u.push_metadata.target_offset;
1692 uint64_t version = msg.u.push_metadata.version;
1693 struct lttng_consumer_channel *channel;
1694
1695 DBG("UST consumer push metadata key %" PRIu64 " of len %" PRIu64, key,
1696 len);
1697
1698 channel = consumer_find_channel(key);
1699 if (!channel) {
1700 /*
1701 * This is possible if the metadata creation on the consumer side
1702 * is in flight vis-a-vis a concurrent push metadata from the
1703 * session daemon. Simply return that the channel failed and the
1704 * session daemon will handle that message correctly considering
1705 * that this race is acceptable thus the DBG() statement here.
1706 */
1707 DBG("UST consumer push metadata %" PRIu64 " not found", key);
1708 ret_code = LTTCOMM_CONSUMERD_CHANNEL_FAIL;
1709 goto end_msg_sessiond;
1710 }
1711
1712 health_code_update();
1713
1714 if (!len) {
1715 /*
1716 * There is nothing to receive. We have simply
1717 * checked whether the channel can be found.
1718 */
1719 ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1720 goto end_msg_sessiond;
1721 }
1722
1723 /* Tell session daemon we are ready to receive the metadata. */
1724 ret = consumer_send_status_msg(sock, LTTCOMM_CONSUMERD_SUCCESS);
1725 if (ret < 0) {
1726 /* Somehow, the session daemon is not responding anymore. */
1727 goto error_fatal;
1728 }
1729
1730 health_code_update();
1731
1732 /* Wait for more data. */
1733 health_poll_entry();
1734 ret = lttng_consumer_poll_socket(consumer_sockpoll);
1735 health_poll_exit();
1736 if (ret) {
1737 goto error_fatal;
1738 }
1739
1740 health_code_update();
1741
1742 ret = lttng_ustconsumer_recv_metadata(sock, key, offset,
1743 len, version, channel, 0, 1);
1744 if (ret < 0) {
1745 /* error receiving from sessiond */
1746 goto error_fatal;
1747 } else {
1748 ret_code = ret;
1749 goto end_msg_sessiond;
1750 }
1751 }
1752 case LTTNG_CONSUMER_SETUP_METADATA:
1753 {
1754 int ret;
1755
1756 ret = setup_metadata(ctx, msg.u.setup_metadata.key);
1757 if (ret) {
1758 ret_code = ret;
1759 }
1760 goto end_msg_sessiond;
1761 }
1762 case LTTNG_CONSUMER_SNAPSHOT_CHANNEL:
1763 {
1764 struct lttng_consumer_channel *channel;
1765 uint64_t key = msg.u.snapshot_channel.key;
1766
1767 channel = consumer_find_channel(key);
1768 if (!channel) {
1769 DBG("UST snapshot channel not found for key %" PRIu64, key);
1770 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
1771 } else {
1772 if (msg.u.snapshot_channel.metadata) {
1773 ret = snapshot_metadata(channel, key,
1774 msg.u.snapshot_channel.pathname,
1775 msg.u.snapshot_channel.relayd_id,
1776 ctx);
1777 if (ret < 0) {
1778 ERR("Snapshot metadata failed");
1779 ret_code = LTTCOMM_CONSUMERD_SNAPSHOT_FAILED;
1780 }
1781 } else {
1782 ret = snapshot_channel(channel, key,
1783 msg.u.snapshot_channel.pathname,
1784 msg.u.snapshot_channel.relayd_id,
1785 msg.u.snapshot_channel.nb_packets_per_stream,
1786 ctx);
1787 if (ret < 0) {
1788 ERR("Snapshot channel failed");
1789 ret_code = LTTCOMM_CONSUMERD_SNAPSHOT_FAILED;
1790 }
1791 }
1792 }
1793 health_code_update();
1794 ret = consumer_send_status_msg(sock, ret_code);
1795 if (ret < 0) {
1796 /* Somehow, the session daemon is not responding anymore. */
1797 goto end_nosignal;
1798 }
1799 health_code_update();
1800 break;
1801 }
1802 case LTTNG_CONSUMER_DISCARDED_EVENTS:
1803 {
1804 int ret = 0;
1805 uint64_t discarded_events;
1806 struct lttng_ht_iter iter;
1807 struct lttng_ht *ht;
1808 struct lttng_consumer_stream *stream;
1809 uint64_t id = msg.u.discarded_events.session_id;
1810 uint64_t key = msg.u.discarded_events.channel_key;
1811
1812 DBG("UST consumer discarded events command for session id %"
1813 PRIu64, id);
1814 rcu_read_lock();
1815 pthread_mutex_lock(&consumer_data.lock);
1816
1817 ht = consumer_data.stream_list_ht;
1818
1819 /*
1820 * We only need a reference to the channel, but they are not
1821 * directly indexed, so we just use the first matching stream
1822 * to extract the information we need, we default to 0 if not
1823 * found (no events are dropped if the channel is not yet in
1824 * use).
1825 */
1826 discarded_events = 0;
1827 cds_lfht_for_each_entry_duplicate(ht->ht,
1828 ht->hash_fct(&id, lttng_ht_seed),
1829 ht->match_fct, &id,
1830 &iter.iter, stream, node_session_id.node) {
1831 if (stream->chan->key == key) {
1832 discarded_events = stream->chan->discarded_events;
1833 break;
1834 }
1835 }
1836 pthread_mutex_unlock(&consumer_data.lock);
1837 rcu_read_unlock();
1838
1839 DBG("UST consumer discarded events command for session id %"
1840 PRIu64 ", channel key %" PRIu64, id, key);
1841
1842 health_code_update();
1843
1844 /* Send back returned value to session daemon */
1845 ret = lttcomm_send_unix_sock(sock, &discarded_events, sizeof(discarded_events));
1846 if (ret < 0) {
1847 PERROR("send discarded events");
1848 goto error_fatal;
1849 }
1850
1851 break;
1852 }
1853 case LTTNG_CONSUMER_LOST_PACKETS:
1854 {
1855 int ret;
1856 uint64_t lost_packets;
1857 struct lttng_ht_iter iter;
1858 struct lttng_ht *ht;
1859 struct lttng_consumer_stream *stream;
1860 uint64_t id = msg.u.lost_packets.session_id;
1861 uint64_t key = msg.u.lost_packets.channel_key;
1862
1863 DBG("UST consumer lost packets command for session id %"
1864 PRIu64, id);
1865 rcu_read_lock();
1866 pthread_mutex_lock(&consumer_data.lock);
1867
1868 ht = consumer_data.stream_list_ht;
1869
1870 /*
1871 * We only need a reference to the channel, but they are not
1872 * directly indexed, so we just use the first matching stream
1873 * to extract the information we need, we default to 0 if not
1874 * found (no packets lost if the channel is not yet in use).
1875 */
1876 lost_packets = 0;
1877 cds_lfht_for_each_entry_duplicate(ht->ht,
1878 ht->hash_fct(&id, lttng_ht_seed),
1879 ht->match_fct, &id,
1880 &iter.iter, stream, node_session_id.node) {
1881 if (stream->chan->key == key) {
1882 lost_packets = stream->chan->lost_packets;
1883 break;
1884 }
1885 }
1886 pthread_mutex_unlock(&consumer_data.lock);
1887 rcu_read_unlock();
1888
1889 DBG("UST consumer lost packets command for session id %"
1890 PRIu64 ", channel key %" PRIu64, id, key);
1891
1892 health_code_update();
1893
1894 /* Send back returned value to session daemon */
1895 ret = lttcomm_send_unix_sock(sock, &lost_packets,
1896 sizeof(lost_packets));
1897 if (ret < 0) {
1898 PERROR("send lost packets");
1899 goto error_fatal;
1900 }
1901
1902 break;
1903 }
1904 case LTTNG_CONSUMER_SET_CHANNEL_MONITOR_PIPE:
1905 {
1906 int channel_monitor_pipe;
1907
1908 ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1909 /* Successfully received the command's type. */
1910 ret = consumer_send_status_msg(sock, ret_code);
1911 if (ret < 0) {
1912 goto error_fatal;
1913 }
1914
1915 ret = lttcomm_recv_fds_unix_sock(sock, &channel_monitor_pipe,
1916 1);
1917 if (ret != sizeof(channel_monitor_pipe)) {
1918 ERR("Failed to receive channel monitor pipe");
1919 goto error_fatal;
1920 }
1921
1922 DBG("Received channel monitor pipe (%d)", channel_monitor_pipe);
1923 ret = consumer_timer_thread_set_channel_monitor_pipe(
1924 channel_monitor_pipe);
1925 if (!ret) {
1926 int flags;
1927
1928 ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1929 /* Set the pipe as non-blocking. */
1930 ret = fcntl(channel_monitor_pipe, F_GETFL, 0);
1931 if (ret == -1) {
1932 PERROR("fcntl get flags of the channel monitoring pipe");
1933 goto error_fatal;
1934 }
1935 flags = ret;
1936
1937 ret = fcntl(channel_monitor_pipe, F_SETFL,
1938 flags | O_NONBLOCK);
1939 if (ret == -1) {
1940 PERROR("fcntl set O_NONBLOCK flag of the channel monitoring pipe");
1941 goto error_fatal;
1942 }
1943 DBG("Channel monitor pipe set as non-blocking");
1944 } else {
1945 ret_code = LTTCOMM_CONSUMERD_ALREADY_SET;
1946 }
1947 goto end_msg_sessiond;
1948 }
1949 case LTTNG_CONSUMER_ROTATE_CHANNEL:
1950 {
1951 struct lttng_consumer_channel *channel;
1952 uint64_t key = msg.u.rotate_channel.key;
1953
1954 channel = consumer_find_channel(key);
1955 if (!channel) {
1956 DBG("Channel %" PRIu64 " not found", key);
1957 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
1958 } else {
1959 /*
1960 * Sample the rotate position of all the streams in
1961 * this channel.
1962 */
1963 ret = lttng_consumer_rotate_channel(channel, key,
1964 msg.u.rotate_channel.relayd_id,
1965 msg.u.rotate_channel.metadata,
1966 ctx);
1967 if (ret < 0) {
1968 ERR("Rotate channel failed");
1969 ret_code = LTTCOMM_CONSUMERD_ROTATION_FAIL;
1970 }
1971
1972 health_code_update();
1973 }
1974 ret = consumer_send_status_msg(sock, ret_code);
1975 if (ret < 0) {
1976 /* Somehow, the session daemon is not responding anymore. */
1977 goto end_nosignal;
1978 }
1979
1980 /*
1981 * Rotate the streams that are ready right now.
1982 * FIXME: this is a second consecutive iteration over the
1983 * streams in a channel, there is probably a better way to
1984 * handle this, but it needs to be after the
1985 * consumer_send_status_msg() call.
1986 */
1987 if (channel) {
1988 ret = lttng_consumer_rotate_ready_streams(
1989 channel, key, ctx);
1990 if (ret < 0) {
1991 ERR("Rotate channel failed");
1992 }
1993 }
1994 break;
1995 }
1996 case LTTNG_CONSUMER_INIT:
1997 {
1998 ret_code = lttng_consumer_init_command(ctx,
1999 msg.u.init.sessiond_uuid);
2000 health_code_update();
2001 ret = consumer_send_status_msg(sock, ret_code);
2002 if (ret < 0) {
2003 /* Somehow, the session daemon is not responding anymore. */
2004 goto end_nosignal;
2005 }
2006 break;
2007 }
2008 case LTTNG_CONSUMER_CREATE_TRACE_CHUNK:
2009 {
2010 const struct lttng_credentials credentials = {
2011 .uid = msg.u.create_trace_chunk.credentials.value.uid,
2012 .gid = msg.u.create_trace_chunk.credentials.value.gid,
2013 };
2014 const bool is_local_trace =
2015 !msg.u.create_trace_chunk.relayd_id.is_set;
2016 const uint64_t relayd_id =
2017 msg.u.create_trace_chunk.relayd_id.value;
2018 const char *chunk_override_name =
2019 *msg.u.create_trace_chunk.override_name ?
2020 msg.u.create_trace_chunk.override_name :
2021 NULL;
2022 LTTNG_OPTIONAL(struct lttng_directory_handle) chunk_directory_handle =
2023 LTTNG_OPTIONAL_INIT;
2024
2025 /*
2026 * The session daemon will only provide a chunk directory file
2027 * descriptor for local traces.
2028 */
2029 if (is_local_trace) {
2030 int chunk_dirfd;
2031
2032 /* Acnowledge the reception of the command. */
2033 ret = consumer_send_status_msg(sock,
2034 LTTCOMM_CONSUMERD_SUCCESS);
2035 if (ret < 0) {
2036 /* Somehow, the session daemon is not responding anymore. */
2037 goto end_nosignal;
2038 }
2039
2040 ret = lttcomm_recv_fds_unix_sock(sock, &chunk_dirfd, 1);
2041 if (ret != sizeof(chunk_dirfd)) {
2042 ERR("Failed to receive trace chunk directory file descriptor");
2043 goto error_fatal;
2044 }
2045
2046 DBG("Received trace chunk directory fd (%d)",
2047 chunk_dirfd);
2048 ret = lttng_directory_handle_init_from_dirfd(
2049 &chunk_directory_handle.value,
2050 chunk_dirfd);
2051 if (ret) {
2052 ERR("Failed to initialize chunk directory handle from directory file descriptor");
2053 if (close(chunk_dirfd)) {
2054 PERROR("Failed to close chunk directory file descriptor");
2055 }
2056 goto error_fatal;
2057 }
2058 chunk_directory_handle.is_set = true;
2059 }
2060
2061 ret_code = lttng_consumer_create_trace_chunk(
2062 !is_local_trace ? &relayd_id : NULL,
2063 msg.u.create_trace_chunk.session_id,
2064 msg.u.create_trace_chunk.chunk_id,
2065 (time_t) msg.u.create_trace_chunk
2066 .creation_timestamp,
2067 chunk_override_name,
2068 msg.u.create_trace_chunk.credentials.is_set ?
2069 &credentials :
2070 NULL,
2071 chunk_directory_handle.is_set ?
2072 &chunk_directory_handle.value :
2073 NULL);
2074
2075 if (chunk_directory_handle.is_set) {
2076 lttng_directory_handle_fini(
2077 &chunk_directory_handle.value);
2078 }
2079 goto end_msg_sessiond;
2080 }
2081 case LTTNG_CONSUMER_CLOSE_TRACE_CHUNK:
2082 {
2083 enum lttng_trace_chunk_command_type close_command =
2084 msg.u.close_trace_chunk.close_command.value;
2085 const uint64_t relayd_id =
2086 msg.u.close_trace_chunk.relayd_id.value;
2087
2088 ret_code = lttng_consumer_close_trace_chunk(
2089 msg.u.close_trace_chunk.relayd_id.is_set ?
2090 &relayd_id :
2091 NULL,
2092 msg.u.close_trace_chunk.session_id,
2093 msg.u.close_trace_chunk.chunk_id,
2094 (time_t) msg.u.close_trace_chunk.close_timestamp,
2095 msg.u.close_trace_chunk.close_command.is_set ?
2096 &close_command :
2097 NULL);
2098 goto end_msg_sessiond;
2099 }
2100 case LTTNG_CONSUMER_TRACE_CHUNK_EXISTS:
2101 {
2102 const uint64_t relayd_id =
2103 msg.u.trace_chunk_exists.relayd_id.value;
2104
2105 ret_code = lttng_consumer_trace_chunk_exists(
2106 msg.u.trace_chunk_exists.relayd_id.is_set ?
2107 &relayd_id : NULL,
2108 msg.u.trace_chunk_exists.session_id,
2109 msg.u.trace_chunk_exists.chunk_id);
2110 goto end_msg_sessiond;
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 pthread_mutex_unlock(&channel->lock);
2145 /*
2146 * Free channel here since no one has a reference to it. We don't
2147 * free after that because a stream can store this pointer.
2148 */
2149 destroy_channel(channel);
2150 }
2151 /* We have to send a status channel message indicating an error. */
2152 ret = consumer_send_status_channel(sock, NULL);
2153 if (ret < 0) {
2154 /* Stop everything if session daemon can not be notified. */
2155 goto error_fatal;
2156 }
2157 rcu_read_unlock();
2158
2159 health_code_update();
2160
2161 return 1;
2162 error_fatal:
2163 rcu_read_unlock();
2164 /* This will issue a consumer stop. */
2165 return -1;
2166 }
2167
2168 /*
2169 * Wrapper over the mmap() read offset from ust-ctl library. Since this can be
2170 * compiled out, we isolate it in this library.
2171 */
2172 int lttng_ustctl_get_mmap_read_offset(struct lttng_consumer_stream *stream,
2173 unsigned long *off)
2174 {
2175 assert(stream);
2176 assert(stream->ustream);
2177
2178 return ustctl_get_mmap_read_offset(stream->ustream, off);
2179 }
2180
2181 /*
2182 * Wrapper over the mmap() read offset from ust-ctl library. Since this can be
2183 * compiled out, we isolate it in this library.
2184 */
2185 void *lttng_ustctl_get_mmap_base(struct lttng_consumer_stream *stream)
2186 {
2187 assert(stream);
2188 assert(stream->ustream);
2189
2190 return ustctl_get_mmap_base(stream->ustream);
2191 }
2192
2193 void lttng_ustctl_flush_buffer(struct lttng_consumer_stream *stream,
2194 int producer_active)
2195 {
2196 assert(stream);
2197 assert(stream->ustream);
2198
2199 ustctl_flush_buffer(stream->ustream, producer_active);
2200 }
2201
2202 /*
2203 * Take a snapshot for a specific stream.
2204 *
2205 * Returns 0 on success, < 0 on error
2206 */
2207 int lttng_ustconsumer_take_snapshot(struct lttng_consumer_stream *stream)
2208 {
2209 assert(stream);
2210 assert(stream->ustream);
2211
2212 return ustctl_snapshot(stream->ustream);
2213 }
2214
2215 /*
2216 * Sample consumed and produced positions for a specific stream.
2217 *
2218 * Returns 0 on success, < 0 on error.
2219 */
2220 int lttng_ustconsumer_sample_snapshot_positions(
2221 struct lttng_consumer_stream *stream)
2222 {
2223 assert(stream);
2224 assert(stream->ustream);
2225
2226 return ustctl_snapshot_sample_positions(stream->ustream);
2227 }
2228
2229 /*
2230 * Get the produced position
2231 *
2232 * Returns 0 on success, < 0 on error
2233 */
2234 int lttng_ustconsumer_get_produced_snapshot(
2235 struct lttng_consumer_stream *stream, unsigned long *pos)
2236 {
2237 assert(stream);
2238 assert(stream->ustream);
2239 assert(pos);
2240
2241 return ustctl_snapshot_get_produced(stream->ustream, pos);
2242 }
2243
2244 /*
2245 * Get the consumed position
2246 *
2247 * Returns 0 on success, < 0 on error
2248 */
2249 int lttng_ustconsumer_get_consumed_snapshot(
2250 struct lttng_consumer_stream *stream, unsigned long *pos)
2251 {
2252 assert(stream);
2253 assert(stream->ustream);
2254 assert(pos);
2255
2256 return ustctl_snapshot_get_consumed(stream->ustream, pos);
2257 }
2258
2259 void lttng_ustconsumer_flush_buffer(struct lttng_consumer_stream *stream,
2260 int producer)
2261 {
2262 assert(stream);
2263 assert(stream->ustream);
2264
2265 ustctl_flush_buffer(stream->ustream, producer);
2266 }
2267
2268 int lttng_ustconsumer_get_current_timestamp(
2269 struct lttng_consumer_stream *stream, uint64_t *ts)
2270 {
2271 assert(stream);
2272 assert(stream->ustream);
2273 assert(ts);
2274
2275 return ustctl_get_current_timestamp(stream->ustream, ts);
2276 }
2277
2278 int lttng_ustconsumer_get_sequence_number(
2279 struct lttng_consumer_stream *stream, uint64_t *seq)
2280 {
2281 assert(stream);
2282 assert(stream->ustream);
2283 assert(seq);
2284
2285 return ustctl_get_sequence_number(stream->ustream, seq);
2286 }
2287
2288 /*
2289 * Called when the stream signals the consumer that it has hung up.
2290 */
2291 void lttng_ustconsumer_on_stream_hangup(struct lttng_consumer_stream *stream)
2292 {
2293 assert(stream);
2294 assert(stream->ustream);
2295
2296 pthread_mutex_lock(&stream->lock);
2297 if (!stream->quiescent) {
2298 ustctl_flush_buffer(stream->ustream, 0);
2299 stream->quiescent = true;
2300 }
2301 pthread_mutex_unlock(&stream->lock);
2302 stream->hangup_flush_done = 1;
2303 }
2304
2305 void lttng_ustconsumer_del_channel(struct lttng_consumer_channel *chan)
2306 {
2307 int i;
2308
2309 assert(chan);
2310 assert(chan->uchan);
2311 assert(chan->buffer_credentials.is_set);
2312
2313 if (chan->switch_timer_enabled == 1) {
2314 consumer_timer_switch_stop(chan);
2315 }
2316 for (i = 0; i < chan->nr_stream_fds; i++) {
2317 int ret;
2318
2319 ret = close(chan->stream_fds[i]);
2320 if (ret) {
2321 PERROR("close");
2322 }
2323 if (chan->shm_path[0]) {
2324 char shm_path[PATH_MAX];
2325
2326 ret = get_stream_shm_path(shm_path, chan->shm_path, i);
2327 if (ret) {
2328 ERR("Cannot get stream shm path");
2329 }
2330 ret = run_as_unlink(shm_path,
2331 chan->buffer_credentials.value.uid,
2332 chan->buffer_credentials.value.gid);
2333 if (ret) {
2334 PERROR("unlink %s", shm_path);
2335 }
2336 }
2337 }
2338 }
2339
2340 void lttng_ustconsumer_free_channel(struct lttng_consumer_channel *chan)
2341 {
2342 assert(chan);
2343 assert(chan->uchan);
2344 assert(chan->buffer_credentials.is_set);
2345
2346 consumer_metadata_cache_destroy(chan);
2347 ustctl_destroy_channel(chan->uchan);
2348 /* Try to rmdir all directories under shm_path root. */
2349 if (chan->root_shm_path[0]) {
2350 (void) run_as_rmdir_recursive(chan->root_shm_path,
2351 chan->buffer_credentials.value.uid,
2352 chan->buffer_credentials.value.gid);
2353 }
2354 free(chan->stream_fds);
2355 }
2356
2357 void lttng_ustconsumer_del_stream(struct lttng_consumer_stream *stream)
2358 {
2359 assert(stream);
2360 assert(stream->ustream);
2361
2362 if (stream->chan->switch_timer_enabled == 1) {
2363 consumer_timer_switch_stop(stream->chan);
2364 }
2365 ustctl_destroy_stream(stream->ustream);
2366 }
2367
2368 int lttng_ustconsumer_get_wakeup_fd(struct lttng_consumer_stream *stream)
2369 {
2370 assert(stream);
2371 assert(stream->ustream);
2372
2373 return ustctl_stream_get_wakeup_fd(stream->ustream);
2374 }
2375
2376 int lttng_ustconsumer_close_wakeup_fd(struct lttng_consumer_stream *stream)
2377 {
2378 assert(stream);
2379 assert(stream->ustream);
2380
2381 return ustctl_stream_close_wakeup_fd(stream->ustream);
2382 }
2383
2384 /*
2385 * Populate index values of a UST stream. Values are set in big endian order.
2386 *
2387 * Return 0 on success or else a negative value.
2388 */
2389 static int get_index_values(struct ctf_packet_index *index,
2390 struct ustctl_consumer_stream *ustream)
2391 {
2392 int ret;
2393
2394 ret = ustctl_get_timestamp_begin(ustream, &index->timestamp_begin);
2395 if (ret < 0) {
2396 PERROR("ustctl_get_timestamp_begin");
2397 goto error;
2398 }
2399 index->timestamp_begin = htobe64(index->timestamp_begin);
2400
2401 ret = ustctl_get_timestamp_end(ustream, &index->timestamp_end);
2402 if (ret < 0) {
2403 PERROR("ustctl_get_timestamp_end");
2404 goto error;
2405 }
2406 index->timestamp_end = htobe64(index->timestamp_end);
2407
2408 ret = ustctl_get_events_discarded(ustream, &index->events_discarded);
2409 if (ret < 0) {
2410 PERROR("ustctl_get_events_discarded");
2411 goto error;
2412 }
2413 index->events_discarded = htobe64(index->events_discarded);
2414
2415 ret = ustctl_get_content_size(ustream, &index->content_size);
2416 if (ret < 0) {
2417 PERROR("ustctl_get_content_size");
2418 goto error;
2419 }
2420 index->content_size = htobe64(index->content_size);
2421
2422 ret = ustctl_get_packet_size(ustream, &index->packet_size);
2423 if (ret < 0) {
2424 PERROR("ustctl_get_packet_size");
2425 goto error;
2426 }
2427 index->packet_size = htobe64(index->packet_size);
2428
2429 ret = ustctl_get_stream_id(ustream, &index->stream_id);
2430 if (ret < 0) {
2431 PERROR("ustctl_get_stream_id");
2432 goto error;
2433 }
2434 index->stream_id = htobe64(index->stream_id);
2435
2436 ret = ustctl_get_instance_id(ustream, &index->stream_instance_id);
2437 if (ret < 0) {
2438 PERROR("ustctl_get_instance_id");
2439 goto error;
2440 }
2441 index->stream_instance_id = htobe64(index->stream_instance_id);
2442
2443 ret = ustctl_get_sequence_number(ustream, &index->packet_seq_num);
2444 if (ret < 0) {
2445 PERROR("ustctl_get_sequence_number");
2446 goto error;
2447 }
2448 index->packet_seq_num = htobe64(index->packet_seq_num);
2449
2450 error:
2451 return ret;
2452 }
2453
2454 static
2455 void metadata_stream_reset_cache(struct lttng_consumer_stream *stream,
2456 struct consumer_metadata_cache *cache)
2457 {
2458 DBG("Metadata stream update to version %" PRIu64,
2459 cache->version);
2460 stream->ust_metadata_pushed = 0;
2461 stream->metadata_version = cache->version;
2462 stream->reset_metadata_flag = 1;
2463 }
2464
2465 /*
2466 * Check if the version of the metadata stream and metadata cache match.
2467 * If the cache got updated, reset the metadata stream.
2468 * The stream lock and metadata cache lock MUST be held.
2469 * Return 0 on success, a negative value on error.
2470 */
2471 static
2472 int metadata_stream_check_version(struct lttng_consumer_stream *stream)
2473 {
2474 int ret = 0;
2475 struct consumer_metadata_cache *cache = stream->chan->metadata_cache;
2476
2477 if (cache->version == stream->metadata_version) {
2478 goto end;
2479 }
2480 metadata_stream_reset_cache(stream, cache);
2481
2482 end:
2483 return ret;
2484 }
2485
2486 /*
2487 * Write up to one packet from the metadata cache to the channel.
2488 *
2489 * Returns the number of bytes pushed in the cache, or a negative value
2490 * on error.
2491 */
2492 static
2493 int commit_one_metadata_packet(struct lttng_consumer_stream *stream)
2494 {
2495 ssize_t write_len;
2496 int ret;
2497
2498 pthread_mutex_lock(&stream->chan->metadata_cache->lock);
2499 ret = metadata_stream_check_version(stream);
2500 if (ret < 0) {
2501 goto end;
2502 }
2503 if (stream->chan->metadata_cache->max_offset
2504 == stream->ust_metadata_pushed) {
2505 ret = 0;
2506 goto end;
2507 }
2508
2509 write_len = ustctl_write_one_packet_to_channel(stream->chan->uchan,
2510 &stream->chan->metadata_cache->data[stream->ust_metadata_pushed],
2511 stream->chan->metadata_cache->max_offset
2512 - stream->ust_metadata_pushed);
2513 assert(write_len != 0);
2514 if (write_len < 0) {
2515 ERR("Writing one metadata packet");
2516 ret = -1;
2517 goto end;
2518 }
2519 stream->ust_metadata_pushed += write_len;
2520
2521 assert(stream->chan->metadata_cache->max_offset >=
2522 stream->ust_metadata_pushed);
2523 ret = write_len;
2524
2525 /*
2526 * Switch packet (but don't open the next one) on every commit of
2527 * a metadata packet. Since the subbuffer is fully filled (with padding,
2528 * if needed), the stream is "quiescent" after this commit.
2529 */
2530 ustctl_flush_buffer(stream->ustream, 1);
2531 stream->quiescent = true;
2532 end:
2533 pthread_mutex_unlock(&stream->chan->metadata_cache->lock);
2534 return ret;
2535 }
2536
2537
2538 /*
2539 * Sync metadata meaning request them to the session daemon and snapshot to the
2540 * metadata thread can consumer them.
2541 *
2542 * Metadata stream lock is held here, but we need to release it when
2543 * interacting with sessiond, else we cause a deadlock with live
2544 * awaiting on metadata to be pushed out.
2545 *
2546 * Return 0 if new metadatda is available, EAGAIN if the metadata stream
2547 * is empty or a negative value on error.
2548 */
2549 int lttng_ustconsumer_sync_metadata(struct lttng_consumer_local_data *ctx,
2550 struct lttng_consumer_stream *metadata)
2551 {
2552 int ret;
2553 int retry = 0;
2554
2555 assert(ctx);
2556 assert(metadata);
2557
2558 pthread_mutex_unlock(&metadata->lock);
2559 /*
2560 * Request metadata from the sessiond, but don't wait for the flush
2561 * because we locked the metadata thread.
2562 */
2563 ret = lttng_ustconsumer_request_metadata(ctx, metadata->chan, 0, 0);
2564 pthread_mutex_lock(&metadata->lock);
2565 if (ret < 0) {
2566 goto end;
2567 }
2568
2569 ret = commit_one_metadata_packet(metadata);
2570 if (ret <= 0) {
2571 goto end;
2572 } else if (ret > 0) {
2573 retry = 1;
2574 }
2575
2576 ret = ustctl_snapshot(metadata->ustream);
2577 if (ret < 0) {
2578 if (errno != EAGAIN) {
2579 ERR("Sync metadata, taking UST snapshot");
2580 goto end;
2581 }
2582 DBG("No new metadata when syncing them.");
2583 /* No new metadata, exit. */
2584 ret = ENODATA;
2585 goto end;
2586 }
2587
2588 /*
2589 * After this flush, we still need to extract metadata.
2590 */
2591 if (retry) {
2592 ret = EAGAIN;
2593 }
2594
2595 end:
2596 return ret;
2597 }
2598
2599 /*
2600 * Return 0 on success else a negative value.
2601 */
2602 static int notify_if_more_data(struct lttng_consumer_stream *stream,
2603 struct lttng_consumer_local_data *ctx)
2604 {
2605 int ret;
2606 struct ustctl_consumer_stream *ustream;
2607
2608 assert(stream);
2609 assert(ctx);
2610
2611 ustream = stream->ustream;
2612
2613 /*
2614 * First, we are going to check if there is a new subbuffer available
2615 * before reading the stream wait_fd.
2616 */
2617 /* Get the next subbuffer */
2618 ret = ustctl_get_next_subbuf(ustream);
2619 if (ret) {
2620 /* No more data found, flag the stream. */
2621 stream->has_data = 0;
2622 ret = 0;
2623 goto end;
2624 }
2625
2626 ret = ustctl_put_subbuf(ustream);
2627 assert(!ret);
2628
2629 /* This stream still has data. Flag it and wake up the data thread. */
2630 stream->has_data = 1;
2631
2632 if (stream->monitor && !stream->hangup_flush_done && !ctx->has_wakeup) {
2633 ssize_t writelen;
2634
2635 writelen = lttng_pipe_write(ctx->consumer_wakeup_pipe, "!", 1);
2636 if (writelen < 0 && errno != EAGAIN && errno != EWOULDBLOCK) {
2637 ret = writelen;
2638 goto end;
2639 }
2640
2641 /* The wake up pipe has been notified. */
2642 ctx->has_wakeup = 1;
2643 }
2644 ret = 0;
2645
2646 end:
2647 return ret;
2648 }
2649
2650 static
2651 int update_stream_stats(struct lttng_consumer_stream *stream)
2652 {
2653 int ret;
2654 uint64_t seq, discarded;
2655
2656 ret = ustctl_get_sequence_number(stream->ustream, &seq);
2657 if (ret < 0) {
2658 PERROR("ustctl_get_sequence_number");
2659 goto end;
2660 }
2661 /*
2662 * Start the sequence when we extract the first packet in case we don't
2663 * start at 0 (for example if a consumer is not connected to the
2664 * session immediately after the beginning).
2665 */
2666 if (stream->last_sequence_number == -1ULL) {
2667 stream->last_sequence_number = seq;
2668 } else if (seq > stream->last_sequence_number) {
2669 stream->chan->lost_packets += seq -
2670 stream->last_sequence_number - 1;
2671 } else {
2672 /* seq <= last_sequence_number */
2673 ERR("Sequence number inconsistent : prev = %" PRIu64
2674 ", current = %" PRIu64,
2675 stream->last_sequence_number, seq);
2676 ret = -1;
2677 goto end;
2678 }
2679 stream->last_sequence_number = seq;
2680
2681 ret = ustctl_get_events_discarded(stream->ustream, &discarded);
2682 if (ret < 0) {
2683 PERROR("kernctl_get_events_discarded");
2684 goto end;
2685 }
2686 if (discarded < stream->last_discarded_events) {
2687 /*
2688 * Overflow has occurred. We assume only one wrap-around
2689 * has occurred.
2690 */
2691 stream->chan->discarded_events +=
2692 (1ULL << (CAA_BITS_PER_LONG - 1)) -
2693 stream->last_discarded_events + discarded;
2694 } else {
2695 stream->chan->discarded_events += discarded -
2696 stream->last_discarded_events;
2697 }
2698 stream->last_discarded_events = discarded;
2699 ret = 0;
2700
2701 end:
2702 return ret;
2703 }
2704
2705 /*
2706 * Read subbuffer from the given stream.
2707 *
2708 * Stream and channel locks MUST be acquired by the caller.
2709 *
2710 * Return 0 on success else a negative value.
2711 */
2712 int lttng_ustconsumer_read_subbuffer(struct lttng_consumer_stream *stream,
2713 struct lttng_consumer_local_data *ctx)
2714 {
2715 unsigned long len, subbuf_size, padding;
2716 int err, write_index = 1, rotation_ret;
2717 long ret = 0;
2718 struct ustctl_consumer_stream *ustream;
2719 struct ctf_packet_index index;
2720
2721 assert(stream);
2722 assert(stream->ustream);
2723 assert(ctx);
2724
2725 DBG("In UST read_subbuffer (wait_fd: %d, name: %s)", stream->wait_fd,
2726 stream->name);
2727
2728 /* Ease our life for what's next. */
2729 ustream = stream->ustream;
2730
2731 /*
2732 * We can consume the 1 byte written into the wait_fd by UST. Don't trigger
2733 * error if we cannot read this one byte (read returns 0), or if the error
2734 * is EAGAIN or EWOULDBLOCK.
2735 *
2736 * This is only done when the stream is monitored by a thread, before the
2737 * flush is done after a hangup and if the stream is not flagged with data
2738 * since there might be nothing to consume in the wait fd but still have
2739 * data available flagged by the consumer wake up pipe.
2740 */
2741 if (stream->monitor && !stream->hangup_flush_done && !stream->has_data) {
2742 char dummy;
2743 ssize_t readlen;
2744
2745 readlen = lttng_read(stream->wait_fd, &dummy, 1);
2746 if (readlen < 0 && errno != EAGAIN && errno != EWOULDBLOCK) {
2747 ret = readlen;
2748 goto error;
2749 }
2750 }
2751
2752 /*
2753 * If the stream was flagged to be ready for rotation before we extract the
2754 * next packet, rotate it now.
2755 */
2756 if (stream->rotate_ready) {
2757 DBG("Rotate stream before extracting data");
2758 rotation_ret = lttng_consumer_rotate_stream(ctx, stream);
2759 if (rotation_ret < 0) {
2760 ERR("Stream rotation error");
2761 ret = -1;
2762 goto error;
2763 }
2764 }
2765
2766 retry:
2767 /* Get the next subbuffer */
2768 err = ustctl_get_next_subbuf(ustream);
2769 if (err != 0) {
2770 /*
2771 * Populate metadata info if the existing info has
2772 * already been read.
2773 */
2774 if (stream->metadata_flag) {
2775 ret = commit_one_metadata_packet(stream);
2776 if (ret <= 0) {
2777 goto error;
2778 }
2779 goto retry;
2780 }
2781
2782 ret = err; /* ustctl_get_next_subbuf returns negative, caller expect positive. */
2783 /*
2784 * This is a debug message even for single-threaded consumer,
2785 * because poll() have more relaxed criterions than get subbuf,
2786 * so get_subbuf may fail for short race windows where poll()
2787 * would issue wakeups.
2788 */
2789 DBG("Reserving sub buffer failed (everything is normal, "
2790 "it is due to concurrency) [ret: %d]", err);
2791 goto error;
2792 }
2793 assert(stream->chan->output == CONSUMER_CHANNEL_MMAP);
2794
2795 if (!stream->metadata_flag) {
2796 index.offset = htobe64(stream->out_fd_offset);
2797 ret = get_index_values(&index, ustream);
2798 if (ret < 0) {
2799 err = ustctl_put_subbuf(ustream);
2800 assert(err == 0);
2801 goto error;
2802 }
2803
2804 /* Update the stream's sequence and discarded events count. */
2805 ret = update_stream_stats(stream);
2806 if (ret < 0) {
2807 PERROR("kernctl_get_events_discarded");
2808 err = ustctl_put_subbuf(ustream);
2809 assert(err == 0);
2810 goto error;
2811 }
2812 } else {
2813 write_index = 0;
2814 }
2815
2816 /* Get the full padded subbuffer size */
2817 err = ustctl_get_padded_subbuf_size(ustream, &len);
2818 assert(err == 0);
2819
2820 /* Get subbuffer data size (without padding) */
2821 err = ustctl_get_subbuf_size(ustream, &subbuf_size);
2822 assert(err == 0);
2823
2824 /* Make sure we don't get a subbuffer size bigger than the padded */
2825 assert(len >= subbuf_size);
2826
2827 padding = len - subbuf_size;
2828
2829 /* write the subbuffer to the tracefile */
2830 ret = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, subbuf_size, padding, &index);
2831 /*
2832 * The mmap operation should write subbuf_size amount of data when network
2833 * streaming or the full padding (len) size when we are _not_ streaming.
2834 */
2835 if ((ret != subbuf_size && stream->net_seq_idx != (uint64_t) -1ULL) ||
2836 (ret != len && stream->net_seq_idx == (uint64_t) -1ULL)) {
2837 /*
2838 * Display the error but continue processing to try to release the
2839 * subbuffer. This is a DBG statement since any unexpected kill or
2840 * signal, the application gets unregistered, relayd gets closed or
2841 * anything that affects the buffer lifetime will trigger this error.
2842 * So, for the sake of the user, don't print this error since it can
2843 * happen and it is OK with the code flow.
2844 */
2845 DBG("Error writing to tracefile "
2846 "(ret: %ld != len: %lu != subbuf_size: %lu)",
2847 ret, len, subbuf_size);
2848 write_index = 0;
2849 }
2850 err = ustctl_put_next_subbuf(ustream);
2851 assert(err == 0);
2852
2853 /*
2854 * This will consumer the byte on the wait_fd if and only if there is not
2855 * next subbuffer to be acquired.
2856 */
2857 if (!stream->metadata_flag) {
2858 ret = notify_if_more_data(stream, ctx);
2859 if (ret < 0) {
2860 goto error;
2861 }
2862 }
2863
2864 /* Write index if needed. */
2865 if (!write_index) {
2866 goto rotate;
2867 }
2868
2869 if (stream->chan->live_timer_interval && !stream->metadata_flag) {
2870 /*
2871 * In live, block until all the metadata is sent.
2872 */
2873 pthread_mutex_lock(&stream->metadata_timer_lock);
2874 assert(!stream->missed_metadata_flush);
2875 stream->waiting_on_metadata = true;
2876 pthread_mutex_unlock(&stream->metadata_timer_lock);
2877
2878 err = consumer_stream_sync_metadata(ctx, stream->session_id);
2879
2880 pthread_mutex_lock(&stream->metadata_timer_lock);
2881 stream->waiting_on_metadata = false;
2882 if (stream->missed_metadata_flush) {
2883 stream->missed_metadata_flush = false;
2884 pthread_mutex_unlock(&stream->metadata_timer_lock);
2885 (void) consumer_flush_ust_index(stream);
2886 } else {
2887 pthread_mutex_unlock(&stream->metadata_timer_lock);
2888 }
2889
2890 if (err < 0) {
2891 goto error;
2892 }
2893 }
2894
2895 assert(!stream->metadata_flag);
2896 err = consumer_stream_write_index(stream, &index);
2897 if (err < 0) {
2898 goto error;
2899 }
2900
2901 rotate:
2902 /*
2903 * After extracting the packet, we check if the stream is now ready to be
2904 * rotated and perform the action immediately.
2905 */
2906 rotation_ret = lttng_consumer_stream_is_rotate_ready(stream);
2907 if (rotation_ret == 1) {
2908 rotation_ret = lttng_consumer_rotate_stream(ctx, stream);
2909 if (rotation_ret < 0) {
2910 ERR("Stream rotation error");
2911 ret = -1;
2912 goto error;
2913 }
2914 } else if (rotation_ret < 0) {
2915 ERR("Checking if stream is ready to rotate");
2916 ret = -1;
2917 goto error;
2918 }
2919 error:
2920 return ret;
2921 }
2922
2923 /*
2924 * Called when a stream is created.
2925 *
2926 * Return 0 on success or else a negative value.
2927 */
2928 int lttng_ustconsumer_on_recv_stream(struct lttng_consumer_stream *stream)
2929 {
2930 int ret;
2931
2932 assert(stream);
2933
2934 /*
2935 * Don't create anything if this is set for streaming or if there is
2936 * no current trace chunk on the parent channel.
2937 */
2938 if (stream->net_seq_idx == (uint64_t) -1ULL && stream->chan->monitor &&
2939 stream->chan->trace_chunk) {
2940 ret = consumer_stream_create_output_files(stream, true);
2941 if (ret) {
2942 goto error;
2943 }
2944 }
2945 ret = 0;
2946
2947 error:
2948 return ret;
2949 }
2950
2951 /*
2952 * Check if data is still being extracted from the buffers for a specific
2953 * stream. Consumer data lock MUST be acquired before calling this function
2954 * and the stream lock.
2955 *
2956 * Return 1 if the traced data are still getting read else 0 meaning that the
2957 * data is available for trace viewer reading.
2958 */
2959 int lttng_ustconsumer_data_pending(struct lttng_consumer_stream *stream)
2960 {
2961 int ret;
2962
2963 assert(stream);
2964 assert(stream->ustream);
2965
2966 DBG("UST consumer checking data pending");
2967
2968 if (stream->endpoint_status != CONSUMER_ENDPOINT_ACTIVE) {
2969 ret = 0;
2970 goto end;
2971 }
2972
2973 if (stream->chan->type == CONSUMER_CHANNEL_TYPE_METADATA) {
2974 uint64_t contiguous, pushed;
2975
2976 /* Ease our life a bit. */
2977 contiguous = stream->chan->metadata_cache->max_offset;
2978 pushed = stream->ust_metadata_pushed;
2979
2980 /*
2981 * We can simply check whether all contiguously available data
2982 * has been pushed to the ring buffer, since the push operation
2983 * is performed within get_next_subbuf(), and because both
2984 * get_next_subbuf() and put_next_subbuf() are issued atomically
2985 * thanks to the stream lock within
2986 * lttng_ustconsumer_read_subbuffer(). This basically means that
2987 * whetnever ust_metadata_pushed is incremented, the associated
2988 * metadata has been consumed from the metadata stream.
2989 */
2990 DBG("UST consumer metadata pending check: contiguous %" PRIu64 " vs pushed %" PRIu64,
2991 contiguous, pushed);
2992 assert(((int64_t) (contiguous - pushed)) >= 0);
2993 if ((contiguous != pushed) ||
2994 (((int64_t) contiguous - pushed) > 0 || contiguous == 0)) {
2995 ret = 1; /* Data is pending */
2996 goto end;
2997 }
2998 } else {
2999 ret = ustctl_get_next_subbuf(stream->ustream);
3000 if (ret == 0) {
3001 /*
3002 * There is still data so let's put back this
3003 * subbuffer.
3004 */
3005 ret = ustctl_put_subbuf(stream->ustream);
3006 assert(ret == 0);
3007 ret = 1; /* Data is pending */
3008 goto end;
3009 }
3010 }
3011
3012 /* Data is NOT pending so ready to be read. */
3013 ret = 0;
3014
3015 end:
3016 return ret;
3017 }
3018
3019 /*
3020 * Stop a given metadata channel timer if enabled and close the wait fd which
3021 * is the poll pipe of the metadata stream.
3022 *
3023 * This MUST be called with the metadata channel acquired.
3024 */
3025 void lttng_ustconsumer_close_metadata(struct lttng_consumer_channel *metadata)
3026 {
3027 int ret;
3028
3029 assert(metadata);
3030 assert(metadata->type == CONSUMER_CHANNEL_TYPE_METADATA);
3031
3032 DBG("Closing metadata channel key %" PRIu64, metadata->key);
3033
3034 if (metadata->switch_timer_enabled == 1) {
3035 consumer_timer_switch_stop(metadata);
3036 }
3037
3038 if (!metadata->metadata_stream) {
3039 goto end;
3040 }
3041
3042 /*
3043 * Closing write side so the thread monitoring the stream wakes up if any
3044 * and clean the metadata stream.
3045 */
3046 if (metadata->metadata_stream->ust_metadata_poll_pipe[1] >= 0) {
3047 ret = close(metadata->metadata_stream->ust_metadata_poll_pipe[1]);
3048 if (ret < 0) {
3049 PERROR("closing metadata pipe write side");
3050 }
3051 metadata->metadata_stream->ust_metadata_poll_pipe[1] = -1;
3052 }
3053
3054 end:
3055 return;
3056 }
3057
3058 /*
3059 * Close every metadata stream wait fd of the metadata hash table. This
3060 * function MUST be used very carefully so not to run into a race between the
3061 * metadata thread handling streams and this function closing their wait fd.
3062 *
3063 * For UST, this is used when the session daemon hangs up. Its the metadata
3064 * producer so calling this is safe because we are assured that no state change
3065 * can occur in the metadata thread for the streams in the hash table.
3066 */
3067 void lttng_ustconsumer_close_all_metadata(struct lttng_ht *metadata_ht)
3068 {
3069 struct lttng_ht_iter iter;
3070 struct lttng_consumer_stream *stream;
3071
3072 assert(metadata_ht);
3073 assert(metadata_ht->ht);
3074
3075 DBG("UST consumer closing all metadata streams");
3076
3077 rcu_read_lock();
3078 cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream,
3079 node.node) {
3080
3081 health_code_update();
3082
3083 pthread_mutex_lock(&stream->chan->lock);
3084 lttng_ustconsumer_close_metadata(stream->chan);
3085 pthread_mutex_unlock(&stream->chan->lock);
3086
3087 }
3088 rcu_read_unlock();
3089 }
3090
3091 void lttng_ustconsumer_close_stream_wakeup(struct lttng_consumer_stream *stream)
3092 {
3093 int ret;
3094
3095 ret = ustctl_stream_close_wakeup_fd(stream->ustream);
3096 if (ret < 0) {
3097 ERR("Unable to close wakeup fd");
3098 }
3099 }
3100
3101 /*
3102 * Please refer to consumer-timer.c before adding any lock within this
3103 * function or any of its callees. Timers have a very strict locking
3104 * semantic with respect to teardown. Failure to respect this semantic
3105 * introduces deadlocks.
3106 *
3107 * DON'T hold the metadata lock when calling this function, else this
3108 * can cause deadlock involving consumer awaiting for metadata to be
3109 * pushed out due to concurrent interaction with the session daemon.
3110 */
3111 int lttng_ustconsumer_request_metadata(struct lttng_consumer_local_data *ctx,
3112 struct lttng_consumer_channel *channel, int timer, int wait)
3113 {
3114 struct lttcomm_metadata_request_msg request;
3115 struct lttcomm_consumer_msg msg;
3116 enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS;
3117 uint64_t len, key, offset, version;
3118 int ret;
3119
3120 assert(channel);
3121 assert(channel->metadata_cache);
3122
3123 memset(&request, 0, sizeof(request));
3124
3125 /* send the metadata request to sessiond */
3126 switch (consumer_data.type) {
3127 case LTTNG_CONSUMER64_UST:
3128 request.bits_per_long = 64;
3129 break;
3130 case LTTNG_CONSUMER32_UST:
3131 request.bits_per_long = 32;
3132 break;
3133 default:
3134 request.bits_per_long = 0;
3135 break;
3136 }
3137
3138 request.session_id = channel->session_id;
3139 request.session_id_per_pid = channel->session_id_per_pid;
3140 /*
3141 * Request the application UID here so the metadata of that application can
3142 * be sent back. The channel UID corresponds to the user UID of the session
3143 * used for the rights on the stream file(s).
3144 */
3145 request.uid = channel->ust_app_uid;
3146 request.key = channel->key;
3147
3148 DBG("Sending metadata request to sessiond, session id %" PRIu64
3149 ", per-pid %" PRIu64 ", app UID %u and channel key %" PRIu64,
3150 request.session_id, request.session_id_per_pid, request.uid,
3151 request.key);
3152
3153 pthread_mutex_lock(&ctx->metadata_socket_lock);
3154
3155 health_code_update();
3156
3157 ret = lttcomm_send_unix_sock(ctx->consumer_metadata_socket, &request,
3158 sizeof(request));
3159 if (ret < 0) {
3160 ERR("Asking metadata to sessiond");
3161 goto end;
3162 }
3163
3164 health_code_update();
3165
3166 /* Receive the metadata from sessiond */
3167 ret = lttcomm_recv_unix_sock(ctx->consumer_metadata_socket, &msg,
3168 sizeof(msg));
3169 if (ret != sizeof(msg)) {
3170 DBG("Consumer received unexpected message size %d (expects %zu)",
3171 ret, sizeof(msg));
3172 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD);
3173 /*
3174 * The ret value might 0 meaning an orderly shutdown but this is ok
3175 * since the caller handles this.
3176 */
3177 goto end;
3178 }
3179
3180 health_code_update();
3181
3182 if (msg.cmd_type == LTTNG_ERR_UND) {
3183 /* No registry found */
3184 (void) consumer_send_status_msg(ctx->consumer_metadata_socket,
3185 ret_code);
3186 ret = 0;
3187 goto end;
3188 } else if (msg.cmd_type != LTTNG_CONSUMER_PUSH_METADATA) {
3189 ERR("Unexpected cmd_type received %d", msg.cmd_type);
3190 ret = -1;
3191 goto end;
3192 }
3193
3194 len = msg.u.push_metadata.len;
3195 key = msg.u.push_metadata.key;
3196 offset = msg.u.push_metadata.target_offset;
3197 version = msg.u.push_metadata.version;
3198
3199 assert(key == channel->key);
3200 if (len == 0) {
3201 DBG("No new metadata to receive for key %" PRIu64, key);
3202 }
3203
3204 health_code_update();
3205
3206 /* Tell session daemon we are ready to receive the metadata. */
3207 ret = consumer_send_status_msg(ctx->consumer_metadata_socket,
3208 LTTCOMM_CONSUMERD_SUCCESS);
3209 if (ret < 0 || len == 0) {
3210 /*
3211 * Somehow, the session daemon is not responding anymore or there is
3212 * nothing to receive.
3213 */
3214 goto end;
3215 }
3216
3217 health_code_update();
3218
3219 ret = lttng_ustconsumer_recv_metadata(ctx->consumer_metadata_socket,
3220 key, offset, len, version, channel, timer, wait);
3221 if (ret >= 0) {
3222 /*
3223 * Only send the status msg if the sessiond is alive meaning a positive
3224 * ret code.
3225 */
3226 (void) consumer_send_status_msg(ctx->consumer_metadata_socket, ret);
3227 }
3228 ret = 0;
3229
3230 end:
3231 health_code_update();
3232
3233 pthread_mutex_unlock(&ctx->metadata_socket_lock);
3234 return ret;
3235 }
3236
3237 /*
3238 * Return the ustctl call for the get stream id.
3239 */
3240 int lttng_ustconsumer_get_stream_id(struct lttng_consumer_stream *stream,
3241 uint64_t *stream_id)
3242 {
3243 assert(stream);
3244 assert(stream_id);
3245
3246 return ustctl_get_stream_id(stream->ustream, stream_id);
3247 }
This page took 0.101819 seconds and 4 git commands to generate.