Fix: consumer: dangling chunk on buffer allocation failure
[lttng-tools.git] / src / common / ust-consumer / ust-consumer.c
1 /*
2 * Copyright (C) 2011 Julien Desfossez <julien.desfossez@polymtl.ca>
3 * Copyright (C) 2011 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
4 * Copyright (C) 2017 Jérémie Galarneau <jeremie.galarneau@efficios.com>
5 *
6 * SPDX-License-Identifier: GPL-2.0-only
7 *
8 */
9
10 #define _LGPL_SOURCE
11 #include <assert.h>
12 #include <lttng/ust-ctl.h>
13 #include <poll.h>
14 #include <pthread.h>
15 #include <stdlib.h>
16 #include <string.h>
17 #include <sys/mman.h>
18 #include <sys/socket.h>
19 #include <sys/stat.h>
20 #include <sys/types.h>
21 #include <inttypes.h>
22 #include <unistd.h>
23 #include <urcu/list.h>
24 #include <signal.h>
25 #include <stdbool.h>
26 #include <stdint.h>
27
28 #include <bin/lttng-consumerd/health-consumerd.h>
29 #include <common/common.h>
30 #include <common/sessiond-comm/sessiond-comm.h>
31 #include <common/relayd/relayd.h>
32 #include <common/compat/fcntl.h>
33 #include <common/compat/endian.h>
34 #include <common/consumer/consumer-metadata-cache.h>
35 #include <common/consumer/consumer-stream.h>
36 #include <common/consumer/consumer-timer.h>
37 #include <common/utils.h>
38 #include <common/index/index.h>
39 #include <common/consumer/consumer.h>
40 #include <common/optional.h>
41
42 #include "ust-consumer.h"
43
44 #define INT_MAX_STR_LEN 12 /* includes \0 */
45
46 extern struct lttng_consumer_global_data consumer_data;
47 extern int consumer_poll_timeout;
48
49 /*
50 * Free channel object and all streams associated with it. This MUST be used
51 * only and only if the channel has _NEVER_ been added to the global channel
52 * hash table.
53 */
54 static void destroy_channel(struct lttng_consumer_channel *channel)
55 {
56 struct lttng_consumer_stream *stream, *stmp;
57
58 assert(channel);
59
60 DBG("UST consumer cleaning stream list");
61
62 cds_list_for_each_entry_safe(stream, stmp, &channel->streams.head,
63 send_node) {
64
65 health_code_update();
66
67 cds_list_del(&stream->send_node);
68 ustctl_destroy_stream(stream->ustream);
69 lttng_trace_chunk_put(stream->trace_chunk);
70 free(stream);
71 }
72
73 /*
74 * If a channel is available meaning that was created before the streams
75 * were, delete it.
76 */
77 if (channel->uchan) {
78 lttng_ustconsumer_del_channel(channel);
79 lttng_ustconsumer_free_channel(channel);
80 }
81
82 if (channel->trace_chunk) {
83 lttng_trace_chunk_put(channel->trace_chunk);
84 }
85
86 free(channel);
87 }
88
89 /*
90 * Add channel to internal consumer state.
91 *
92 * Returns 0 on success or else a negative value.
93 */
94 static int add_channel(struct lttng_consumer_channel *channel,
95 struct lttng_consumer_local_data *ctx)
96 {
97 int ret = 0;
98
99 assert(channel);
100 assert(ctx);
101
102 if (ctx->on_recv_channel != NULL) {
103 ret = ctx->on_recv_channel(channel);
104 if (ret == 0) {
105 ret = consumer_add_channel(channel, ctx);
106 } else if (ret < 0) {
107 /* Most likely an ENOMEM. */
108 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
109 goto error;
110 }
111 } else {
112 ret = consumer_add_channel(channel, ctx);
113 }
114
115 DBG("UST consumer channel added (key: %" PRIu64 ")", channel->key);
116
117 error:
118 return ret;
119 }
120
121 /*
122 * Allocate and return a consumer stream object. If _alloc_ret is not NULL, the
123 * error value if applicable is set in it else it is kept untouched.
124 *
125 * Return NULL on error else the newly allocated stream object.
126 */
127 static struct lttng_consumer_stream *allocate_stream(int cpu, int key,
128 struct lttng_consumer_channel *channel,
129 struct lttng_consumer_local_data *ctx, int *_alloc_ret)
130 {
131 int alloc_ret;
132 struct lttng_consumer_stream *stream = NULL;
133
134 assert(channel);
135 assert(ctx);
136
137 stream = consumer_stream_create(
138 channel,
139 channel->key,
140 key,
141 channel->name,
142 channel->relayd_id,
143 channel->session_id,
144 channel->trace_chunk,
145 cpu,
146 &alloc_ret,
147 channel->type,
148 channel->monitor);
149 if (stream == NULL) {
150 switch (alloc_ret) {
151 case -ENOENT:
152 /*
153 * We could not find the channel. Can happen if cpu hotplug
154 * happens while tearing down.
155 */
156 DBG3("Could not find channel");
157 break;
158 case -ENOMEM:
159 case -EINVAL:
160 default:
161 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
162 break;
163 }
164 goto error;
165 }
166
167 consumer_stream_update_channel_attributes(stream, channel);
168
169 error:
170 if (_alloc_ret) {
171 *_alloc_ret = alloc_ret;
172 }
173 return stream;
174 }
175
176 /*
177 * Send the given stream pointer to the corresponding thread.
178 *
179 * Returns 0 on success else a negative value.
180 */
181 static int send_stream_to_thread(struct lttng_consumer_stream *stream,
182 struct lttng_consumer_local_data *ctx)
183 {
184 int ret;
185 struct lttng_pipe *stream_pipe;
186
187 /* Get the right pipe where the stream will be sent. */
188 if (stream->metadata_flag) {
189 consumer_add_metadata_stream(stream);
190 stream_pipe = ctx->consumer_metadata_pipe;
191 } else {
192 consumer_add_data_stream(stream);
193 stream_pipe = ctx->consumer_data_pipe;
194 }
195
196 /*
197 * From this point on, the stream's ownership has been moved away from
198 * the channel and it becomes globally visible. Hence, remove it from
199 * the local stream list to prevent the stream from being both local and
200 * global.
201 */
202 stream->globally_visible = 1;
203 cds_list_del(&stream->send_node);
204
205 ret = lttng_pipe_write(stream_pipe, &stream, sizeof(stream));
206 if (ret < 0) {
207 ERR("Consumer write %s stream to pipe %d",
208 stream->metadata_flag ? "metadata" : "data",
209 lttng_pipe_get_writefd(stream_pipe));
210 if (stream->metadata_flag) {
211 consumer_del_stream_for_metadata(stream);
212 } else {
213 consumer_del_stream_for_data(stream);
214 }
215 goto error;
216 }
217
218 error:
219 return ret;
220 }
221
222 static
223 int get_stream_shm_path(char *stream_shm_path, const char *shm_path, int cpu)
224 {
225 char cpu_nr[INT_MAX_STR_LEN]; /* int max len */
226 int ret;
227
228 strncpy(stream_shm_path, shm_path, PATH_MAX);
229 stream_shm_path[PATH_MAX - 1] = '\0';
230 ret = snprintf(cpu_nr, INT_MAX_STR_LEN, "%i", cpu);
231 if (ret < 0) {
232 PERROR("snprintf");
233 goto end;
234 }
235 strncat(stream_shm_path, cpu_nr,
236 PATH_MAX - strlen(stream_shm_path) - 1);
237 ret = 0;
238 end:
239 return ret;
240 }
241
242 /*
243 * Create streams for the given channel using liblttng-ust-ctl.
244 * The channel lock must be acquired by the caller.
245 *
246 * Return 0 on success else a negative value.
247 */
248 static int create_ust_streams(struct lttng_consumer_channel *channel,
249 struct lttng_consumer_local_data *ctx)
250 {
251 int ret, cpu = 0;
252 struct ustctl_consumer_stream *ustream;
253 struct lttng_consumer_stream *stream;
254 pthread_mutex_t *current_stream_lock = NULL;
255
256 assert(channel);
257 assert(ctx);
258
259 /*
260 * While a stream is available from ustctl. When NULL is returned, we've
261 * reached the end of the possible stream for the channel.
262 */
263 while ((ustream = ustctl_create_stream(channel->uchan, cpu))) {
264 int wait_fd;
265 int ust_metadata_pipe[2];
266
267 health_code_update();
268
269 if (channel->type == CONSUMER_CHANNEL_TYPE_METADATA && channel->monitor) {
270 ret = utils_create_pipe_cloexec_nonblock(ust_metadata_pipe);
271 if (ret < 0) {
272 ERR("Create ust metadata poll pipe");
273 goto error;
274 }
275 wait_fd = ust_metadata_pipe[0];
276 } else {
277 wait_fd = ustctl_stream_get_wait_fd(ustream);
278 }
279
280 /* Allocate consumer stream object. */
281 stream = allocate_stream(cpu, wait_fd, channel, ctx, &ret);
282 if (!stream) {
283 goto error_alloc;
284 }
285 stream->ustream = ustream;
286 /*
287 * Store it so we can save multiple function calls afterwards since
288 * this value is used heavily in the stream threads. This is UST
289 * specific so this is why it's done after allocation.
290 */
291 stream->wait_fd = wait_fd;
292
293 /*
294 * Increment channel refcount since the channel reference has now been
295 * assigned in the allocation process above.
296 */
297 if (stream->chan->monitor) {
298 uatomic_inc(&stream->chan->refcount);
299 }
300
301 pthread_mutex_lock(&stream->lock);
302 current_stream_lock = &stream->lock;
303 /*
304 * Order is important this is why a list is used. On error, the caller
305 * should clean this list.
306 */
307 cds_list_add_tail(&stream->send_node, &channel->streams.head);
308
309 ret = ustctl_get_max_subbuf_size(stream->ustream,
310 &stream->max_sb_size);
311 if (ret < 0) {
312 ERR("ustctl_get_max_subbuf_size failed for stream %s",
313 stream->name);
314 goto error;
315 }
316
317 /* Do actions once stream has been received. */
318 if (ctx->on_recv_stream) {
319 ret = ctx->on_recv_stream(stream);
320 if (ret < 0) {
321 goto error;
322 }
323 }
324
325 DBG("UST consumer add stream %s (key: %" PRIu64 ") with relayd id %" PRIu64,
326 stream->name, stream->key, stream->relayd_stream_id);
327
328 /* Set next CPU stream. */
329 channel->streams.count = ++cpu;
330
331 /* Keep stream reference when creating metadata. */
332 if (channel->type == CONSUMER_CHANNEL_TYPE_METADATA) {
333 channel->metadata_stream = stream;
334 if (channel->monitor) {
335 /* Set metadata poll pipe if we created one */
336 memcpy(stream->ust_metadata_poll_pipe,
337 ust_metadata_pipe,
338 sizeof(ust_metadata_pipe));
339 }
340 }
341 pthread_mutex_unlock(&stream->lock);
342 current_stream_lock = NULL;
343 }
344
345 return 0;
346
347 error:
348 error_alloc:
349 if (current_stream_lock) {
350 pthread_mutex_unlock(current_stream_lock);
351 }
352 return ret;
353 }
354
355 /*
356 * create_posix_shm is never called concurrently within a process.
357 */
358 static
359 int create_posix_shm(void)
360 {
361 char tmp_name[NAME_MAX];
362 int shmfd, ret;
363
364 ret = snprintf(tmp_name, NAME_MAX, "/ust-shm-consumer-%d", getpid());
365 if (ret < 0) {
366 PERROR("snprintf");
367 return -1;
368 }
369 /*
370 * Allocate shm, and immediately unlink its shm oject, keeping
371 * only the file descriptor as a reference to the object.
372 * We specifically do _not_ use the / at the beginning of the
373 * pathname so that some OS implementations can keep it local to
374 * the process (POSIX leaves this implementation-defined).
375 */
376 shmfd = shm_open(tmp_name, O_CREAT | O_EXCL | O_RDWR, 0700);
377 if (shmfd < 0) {
378 PERROR("shm_open");
379 goto error_shm_open;
380 }
381 ret = shm_unlink(tmp_name);
382 if (ret < 0 && errno != ENOENT) {
383 PERROR("shm_unlink");
384 goto error_shm_release;
385 }
386 return shmfd;
387
388 error_shm_release:
389 ret = close(shmfd);
390 if (ret) {
391 PERROR("close");
392 }
393 error_shm_open:
394 return -1;
395 }
396
397 static int open_ust_stream_fd(struct lttng_consumer_channel *channel, int cpu,
398 const struct lttng_credentials *session_credentials)
399 {
400 char shm_path[PATH_MAX];
401 int ret;
402
403 if (!channel->shm_path[0]) {
404 return create_posix_shm();
405 }
406 ret = get_stream_shm_path(shm_path, channel->shm_path, cpu);
407 if (ret) {
408 goto error_shm_path;
409 }
410 return run_as_open(shm_path,
411 O_RDWR | O_CREAT | O_EXCL, S_IRUSR | S_IWUSR,
412 session_credentials->uid, session_credentials->gid);
413
414 error_shm_path:
415 return -1;
416 }
417
418 /*
419 * Create an UST channel with the given attributes and send it to the session
420 * daemon using the ust ctl API.
421 *
422 * Return 0 on success or else a negative value.
423 */
424 static int create_ust_channel(struct lttng_consumer_channel *channel,
425 struct ustctl_consumer_channel_attr *attr,
426 struct ustctl_consumer_channel **ust_chanp)
427 {
428 int ret, nr_stream_fds, i, j;
429 int *stream_fds;
430 struct ustctl_consumer_channel *ust_channel;
431
432 assert(channel);
433 assert(attr);
434 assert(ust_chanp);
435 assert(channel->buffer_credentials.is_set);
436
437 DBG3("Creating channel to ustctl with attr: [overwrite: %d, "
438 "subbuf_size: %" PRIu64 ", num_subbuf: %" PRIu64 ", "
439 "switch_timer_interval: %u, read_timer_interval: %u, "
440 "output: %d, type: %d", attr->overwrite, attr->subbuf_size,
441 attr->num_subbuf, attr->switch_timer_interval,
442 attr->read_timer_interval, attr->output, attr->type);
443
444 if (channel->type == CONSUMER_CHANNEL_TYPE_METADATA)
445 nr_stream_fds = 1;
446 else
447 nr_stream_fds = ustctl_get_nr_stream_per_channel();
448 stream_fds = zmalloc(nr_stream_fds * sizeof(*stream_fds));
449 if (!stream_fds) {
450 ret = -1;
451 goto error_alloc;
452 }
453 for (i = 0; i < nr_stream_fds; i++) {
454 stream_fds[i] = open_ust_stream_fd(channel, i,
455 &channel->buffer_credentials.value);
456 if (stream_fds[i] < 0) {
457 ret = -1;
458 goto error_open;
459 }
460 }
461 ust_channel = ustctl_create_channel(attr, stream_fds, nr_stream_fds);
462 if (!ust_channel) {
463 ret = -1;
464 goto error_create;
465 }
466 channel->nr_stream_fds = nr_stream_fds;
467 channel->stream_fds = stream_fds;
468 *ust_chanp = ust_channel;
469
470 return 0;
471
472 error_create:
473 error_open:
474 for (j = i - 1; j >= 0; j--) {
475 int closeret;
476
477 closeret = close(stream_fds[j]);
478 if (closeret) {
479 PERROR("close");
480 }
481 if (channel->shm_path[0]) {
482 char shm_path[PATH_MAX];
483
484 closeret = get_stream_shm_path(shm_path,
485 channel->shm_path, j);
486 if (closeret) {
487 ERR("Cannot get stream shm path");
488 }
489 closeret = run_as_unlink(shm_path,
490 channel->buffer_credentials.value.uid,
491 channel->buffer_credentials.value.gid);
492 if (closeret) {
493 PERROR("unlink %s", shm_path);
494 }
495 }
496 }
497 /* Try to rmdir all directories under shm_path root. */
498 if (channel->root_shm_path[0]) {
499 (void) run_as_rmdir_recursive(channel->root_shm_path,
500 channel->buffer_credentials.value.uid,
501 channel->buffer_credentials.value.gid,
502 LTTNG_DIRECTORY_HANDLE_SKIP_NON_EMPTY_FLAG);
503 }
504 free(stream_fds);
505 error_alloc:
506 return ret;
507 }
508
509 /*
510 * Send a single given stream to the session daemon using the sock.
511 *
512 * Return 0 on success else a negative value.
513 */
514 static int send_sessiond_stream(int sock, struct lttng_consumer_stream *stream)
515 {
516 int ret;
517
518 assert(stream);
519 assert(sock >= 0);
520
521 DBG("UST consumer sending stream %" PRIu64 " to sessiond", stream->key);
522
523 /* Send stream to session daemon. */
524 ret = ustctl_send_stream_to_sessiond(sock, stream->ustream);
525 if (ret < 0) {
526 goto error;
527 }
528
529 error:
530 return ret;
531 }
532
533 /*
534 * Send channel to sessiond and relayd if applicable.
535 *
536 * Return 0 on success or else a negative value.
537 */
538 static int send_channel_to_sessiond_and_relayd(int sock,
539 struct lttng_consumer_channel *channel,
540 struct lttng_consumer_local_data *ctx, int *relayd_error)
541 {
542 int ret, ret_code = LTTCOMM_CONSUMERD_SUCCESS;
543 struct lttng_consumer_stream *stream;
544 uint64_t net_seq_idx = -1ULL;
545
546 assert(channel);
547 assert(ctx);
548 assert(sock >= 0);
549
550 DBG("UST consumer sending channel %s to sessiond", channel->name);
551
552 if (channel->relayd_id != (uint64_t) -1ULL) {
553 cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
554
555 health_code_update();
556
557 /* Try to send the stream to the relayd if one is available. */
558 DBG("Sending stream %" PRIu64 " of channel \"%s\" to relayd",
559 stream->key, channel->name);
560 ret = consumer_send_relayd_stream(stream, stream->chan->pathname);
561 if (ret < 0) {
562 /*
563 * Flag that the relayd was the problem here probably due to a
564 * communicaton error on the socket.
565 */
566 if (relayd_error) {
567 *relayd_error = 1;
568 }
569 ret_code = LTTCOMM_CONSUMERD_RELAYD_FAIL;
570 }
571 if (net_seq_idx == -1ULL) {
572 net_seq_idx = stream->net_seq_idx;
573 }
574 }
575 }
576
577 /* Inform sessiond that we are about to send channel and streams. */
578 ret = consumer_send_status_msg(sock, ret_code);
579 if (ret < 0 || ret_code != LTTCOMM_CONSUMERD_SUCCESS) {
580 /*
581 * Either the session daemon is not responding or the relayd died so we
582 * stop now.
583 */
584 goto error;
585 }
586
587 /* Send channel to sessiond. */
588 ret = ustctl_send_channel_to_sessiond(sock, channel->uchan);
589 if (ret < 0) {
590 goto error;
591 }
592
593 ret = ustctl_channel_close_wakeup_fd(channel->uchan);
594 if (ret < 0) {
595 goto error;
596 }
597
598 /* The channel was sent successfully to the sessiond at this point. */
599 cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
600
601 health_code_update();
602
603 /* Send stream to session daemon. */
604 ret = send_sessiond_stream(sock, stream);
605 if (ret < 0) {
606 goto error;
607 }
608 }
609
610 /* Tell sessiond there is no more stream. */
611 ret = ustctl_send_stream_to_sessiond(sock, NULL);
612 if (ret < 0) {
613 goto error;
614 }
615
616 DBG("UST consumer NULL stream sent to sessiond");
617
618 return 0;
619
620 error:
621 if (ret_code != LTTCOMM_CONSUMERD_SUCCESS) {
622 ret = -1;
623 }
624 return ret;
625 }
626
627 /*
628 * Creates a channel and streams and add the channel it to the channel internal
629 * state. The created stream must ONLY be sent once the GET_CHANNEL command is
630 * received.
631 *
632 * Return 0 on success or else, a negative value is returned and the channel
633 * MUST be destroyed by consumer_del_channel().
634 */
635 static int ask_channel(struct lttng_consumer_local_data *ctx,
636 struct lttng_consumer_channel *channel,
637 struct ustctl_consumer_channel_attr *attr)
638 {
639 int ret;
640
641 assert(ctx);
642 assert(channel);
643 assert(attr);
644
645 /*
646 * This value is still used by the kernel consumer since for the kernel,
647 * the stream ownership is not IN the consumer so we need to have the
648 * number of left stream that needs to be initialized so we can know when
649 * to delete the channel (see consumer.c).
650 *
651 * As for the user space tracer now, the consumer creates and sends the
652 * stream to the session daemon which only sends them to the application
653 * once every stream of a channel is received making this value useless
654 * because we they will be added to the poll thread before the application
655 * receives them. This ensures that a stream can not hang up during
656 * initilization of a channel.
657 */
658 channel->nb_init_stream_left = 0;
659
660 /* The reply msg status is handled in the following call. */
661 ret = create_ust_channel(channel, attr, &channel->uchan);
662 if (ret < 0) {
663 goto end;
664 }
665
666 channel->wait_fd = ustctl_channel_get_wait_fd(channel->uchan);
667
668 /*
669 * For the snapshots (no monitor), we create the metadata streams
670 * on demand, not during the channel creation.
671 */
672 if (channel->type == CONSUMER_CHANNEL_TYPE_METADATA && !channel->monitor) {
673 ret = 0;
674 goto end;
675 }
676
677 /* Open all streams for this channel. */
678 pthread_mutex_lock(&channel->lock);
679 ret = create_ust_streams(channel, ctx);
680 pthread_mutex_unlock(&channel->lock);
681 if (ret < 0) {
682 goto end;
683 }
684
685 end:
686 return ret;
687 }
688
689 /*
690 * Send all stream of a channel to the right thread handling it.
691 *
692 * On error, return a negative value else 0 on success.
693 */
694 static int send_streams_to_thread(struct lttng_consumer_channel *channel,
695 struct lttng_consumer_local_data *ctx)
696 {
697 int ret = 0;
698 struct lttng_consumer_stream *stream, *stmp;
699
700 assert(channel);
701 assert(ctx);
702
703 /* Send streams to the corresponding thread. */
704 cds_list_for_each_entry_safe(stream, stmp, &channel->streams.head,
705 send_node) {
706
707 health_code_update();
708
709 /* Sending the stream to the thread. */
710 ret = send_stream_to_thread(stream, ctx);
711 if (ret < 0) {
712 /*
713 * If we are unable to send the stream to the thread, there is
714 * a big problem so just stop everything.
715 */
716 goto error;
717 }
718 }
719
720 error:
721 return ret;
722 }
723
724 /*
725 * Flush channel's streams using the given key to retrieve the channel.
726 *
727 * Return 0 on success else an LTTng error code.
728 */
729 static int flush_channel(uint64_t chan_key)
730 {
731 int ret = 0;
732 struct lttng_consumer_channel *channel;
733 struct lttng_consumer_stream *stream;
734 struct lttng_ht *ht;
735 struct lttng_ht_iter iter;
736
737 DBG("UST consumer flush channel key %" PRIu64, chan_key);
738
739 rcu_read_lock();
740 channel = consumer_find_channel(chan_key);
741 if (!channel) {
742 ERR("UST consumer flush channel %" PRIu64 " not found", chan_key);
743 ret = LTTNG_ERR_UST_CHAN_NOT_FOUND;
744 goto error;
745 }
746
747 ht = consumer_data.stream_per_chan_id_ht;
748
749 /* For each stream of the channel id, flush it. */
750 cds_lfht_for_each_entry_duplicate(ht->ht,
751 ht->hash_fct(&channel->key, lttng_ht_seed), ht->match_fct,
752 &channel->key, &iter.iter, stream, node_channel_id.node) {
753
754 health_code_update();
755
756 pthread_mutex_lock(&stream->lock);
757
758 /*
759 * Protect against concurrent teardown of a stream.
760 */
761 if (cds_lfht_is_node_deleted(&stream->node.node)) {
762 goto next;
763 }
764
765 if (!stream->quiescent) {
766 ustctl_flush_buffer(stream->ustream, 0);
767 stream->quiescent = true;
768 }
769 next:
770 pthread_mutex_unlock(&stream->lock);
771 }
772 error:
773 rcu_read_unlock();
774 return ret;
775 }
776
777 /*
778 * Clear quiescent state from channel's streams using the given key to
779 * retrieve the channel.
780 *
781 * Return 0 on success else an LTTng error code.
782 */
783 static int clear_quiescent_channel(uint64_t chan_key)
784 {
785 int ret = 0;
786 struct lttng_consumer_channel *channel;
787 struct lttng_consumer_stream *stream;
788 struct lttng_ht *ht;
789 struct lttng_ht_iter iter;
790
791 DBG("UST consumer clear quiescent channel key %" PRIu64, chan_key);
792
793 rcu_read_lock();
794 channel = consumer_find_channel(chan_key);
795 if (!channel) {
796 ERR("UST consumer clear quiescent channel %" PRIu64 " not found", chan_key);
797 ret = LTTNG_ERR_UST_CHAN_NOT_FOUND;
798 goto error;
799 }
800
801 ht = consumer_data.stream_per_chan_id_ht;
802
803 /* For each stream of the channel id, clear quiescent state. */
804 cds_lfht_for_each_entry_duplicate(ht->ht,
805 ht->hash_fct(&channel->key, lttng_ht_seed), ht->match_fct,
806 &channel->key, &iter.iter, stream, node_channel_id.node) {
807
808 health_code_update();
809
810 pthread_mutex_lock(&stream->lock);
811 stream->quiescent = false;
812 pthread_mutex_unlock(&stream->lock);
813 }
814 error:
815 rcu_read_unlock();
816 return ret;
817 }
818
819 /*
820 * Close metadata stream wakeup_fd using the given key to retrieve the channel.
821 *
822 * Return 0 on success else an LTTng error code.
823 */
824 static int close_metadata(uint64_t chan_key)
825 {
826 int ret = 0;
827 struct lttng_consumer_channel *channel;
828 unsigned int channel_monitor;
829
830 DBG("UST consumer close metadata key %" PRIu64, chan_key);
831
832 channel = consumer_find_channel(chan_key);
833 if (!channel) {
834 /*
835 * This is possible if the metadata thread has issue a delete because
836 * the endpoint point of the stream hung up. There is no way the
837 * session daemon can know about it thus use a DBG instead of an actual
838 * error.
839 */
840 DBG("UST consumer close metadata %" PRIu64 " not found", chan_key);
841 ret = LTTNG_ERR_UST_CHAN_NOT_FOUND;
842 goto error;
843 }
844
845 pthread_mutex_lock(&consumer_data.lock);
846 pthread_mutex_lock(&channel->lock);
847 channel_monitor = channel->monitor;
848 if (cds_lfht_is_node_deleted(&channel->node.node)) {
849 goto error_unlock;
850 }
851
852 lttng_ustconsumer_close_metadata(channel);
853 pthread_mutex_unlock(&channel->lock);
854 pthread_mutex_unlock(&consumer_data.lock);
855
856 /*
857 * The ownership of a metadata channel depends on the type of
858 * session to which it belongs. In effect, the monitor flag is checked
859 * to determine if this metadata channel is in "snapshot" mode or not.
860 *
861 * In the non-snapshot case, the metadata channel is created along with
862 * a single stream which will remain present until the metadata channel
863 * is destroyed (on the destruction of its session). In this case, the
864 * metadata stream in "monitored" by the metadata poll thread and holds
865 * the ownership of its channel.
866 *
867 * Closing the metadata will cause the metadata stream's "metadata poll
868 * pipe" to be closed. Closing this pipe will wake-up the metadata poll
869 * thread which will teardown the metadata stream which, in return,
870 * deletes the metadata channel.
871 *
872 * In the snapshot case, the metadata stream is created and destroyed
873 * on every snapshot record. Since the channel doesn't have an owner
874 * other than the session daemon, it is safe to destroy it immediately
875 * on reception of the CLOSE_METADATA command.
876 */
877 if (!channel_monitor) {
878 /*
879 * The channel and consumer_data locks must be
880 * released before this call since consumer_del_channel
881 * re-acquires the channel and consumer_data locks to teardown
882 * the channel and queue its reclamation by the "call_rcu"
883 * worker thread.
884 */
885 consumer_del_channel(channel);
886 }
887
888 return ret;
889 error_unlock:
890 pthread_mutex_unlock(&channel->lock);
891 pthread_mutex_unlock(&consumer_data.lock);
892 error:
893 return ret;
894 }
895
896 /*
897 * RCU read side lock MUST be acquired before calling this function.
898 *
899 * Return 0 on success else an LTTng error code.
900 */
901 static int setup_metadata(struct lttng_consumer_local_data *ctx, uint64_t key)
902 {
903 int ret;
904 struct lttng_consumer_channel *metadata;
905
906 DBG("UST consumer setup metadata key %" PRIu64, key);
907
908 metadata = consumer_find_channel(key);
909 if (!metadata) {
910 ERR("UST consumer push metadata %" PRIu64 " not found", key);
911 ret = LTTNG_ERR_UST_CHAN_NOT_FOUND;
912 goto end;
913 }
914
915 /*
916 * In no monitor mode, the metadata channel has no stream(s) so skip the
917 * ownership transfer to the metadata thread.
918 */
919 if (!metadata->monitor) {
920 DBG("Metadata channel in no monitor");
921 ret = 0;
922 goto end;
923 }
924
925 /*
926 * Send metadata stream to relayd if one available. Availability is
927 * known if the stream is still in the list of the channel.
928 */
929 if (cds_list_empty(&metadata->streams.head)) {
930 ERR("Metadata channel key %" PRIu64 ", no stream available.", key);
931 ret = LTTCOMM_CONSUMERD_ERROR_METADATA;
932 goto error_no_stream;
933 }
934
935 /* Send metadata stream to relayd if needed. */
936 if (metadata->metadata_stream->net_seq_idx != (uint64_t) -1ULL) {
937 ret = consumer_send_relayd_stream(metadata->metadata_stream,
938 metadata->pathname);
939 if (ret < 0) {
940 ret = LTTCOMM_CONSUMERD_ERROR_METADATA;
941 goto error;
942 }
943 ret = consumer_send_relayd_streams_sent(
944 metadata->metadata_stream->net_seq_idx);
945 if (ret < 0) {
946 ret = LTTCOMM_CONSUMERD_RELAYD_FAIL;
947 goto error;
948 }
949 }
950
951 /*
952 * Ownership of metadata stream is passed along. Freeing is handled by
953 * the callee.
954 */
955 ret = send_streams_to_thread(metadata, ctx);
956 if (ret < 0) {
957 /*
958 * If we are unable to send the stream to the thread, there is
959 * a big problem so just stop everything.
960 */
961 ret = LTTCOMM_CONSUMERD_FATAL;
962 goto send_streams_error;
963 }
964 /* List MUST be empty after or else it could be reused. */
965 assert(cds_list_empty(&metadata->streams.head));
966
967 ret = 0;
968 goto end;
969
970 error:
971 /*
972 * Delete metadata channel on error. At this point, the metadata stream can
973 * NOT be monitored by the metadata thread thus having the guarantee that
974 * the stream is still in the local stream list of the channel. This call
975 * will make sure to clean that list.
976 */
977 consumer_stream_destroy(metadata->metadata_stream, NULL);
978 cds_list_del(&metadata->metadata_stream->send_node);
979 metadata->metadata_stream = NULL;
980 send_streams_error:
981 error_no_stream:
982 end:
983 return ret;
984 }
985
986 /*
987 * Snapshot the whole metadata.
988 * RCU read-side lock must be held by the caller.
989 *
990 * Returns 0 on success, < 0 on error
991 */
992 static int snapshot_metadata(struct lttng_consumer_channel *metadata_channel,
993 uint64_t key, char *path, uint64_t relayd_id,
994 struct lttng_consumer_local_data *ctx)
995 {
996 int ret = 0;
997 struct lttng_consumer_stream *metadata_stream;
998
999 assert(path);
1000 assert(ctx);
1001
1002 DBG("UST consumer snapshot metadata with key %" PRIu64 " at path %s",
1003 key, path);
1004
1005 rcu_read_lock();
1006
1007 assert(!metadata_channel->monitor);
1008
1009 health_code_update();
1010
1011 /*
1012 * Ask the sessiond if we have new metadata waiting and update the
1013 * consumer metadata cache.
1014 */
1015 ret = lttng_ustconsumer_request_metadata(ctx, metadata_channel, 0, 1);
1016 if (ret < 0) {
1017 goto error;
1018 }
1019
1020 health_code_update();
1021
1022 /*
1023 * The metadata stream is NOT created in no monitor mode when the channel
1024 * is created on a sessiond ask channel command.
1025 */
1026 ret = create_ust_streams(metadata_channel, ctx);
1027 if (ret < 0) {
1028 goto error;
1029 }
1030
1031 metadata_stream = metadata_channel->metadata_stream;
1032 assert(metadata_stream);
1033
1034 pthread_mutex_lock(&metadata_stream->lock);
1035 if (relayd_id != (uint64_t) -1ULL) {
1036 metadata_stream->net_seq_idx = relayd_id;
1037 ret = consumer_send_relayd_stream(metadata_stream, path);
1038 } else {
1039 ret = consumer_stream_create_output_files(metadata_stream,
1040 false);
1041 }
1042 pthread_mutex_unlock(&metadata_stream->lock);
1043 if (ret < 0) {
1044 goto error_stream;
1045 }
1046
1047 do {
1048 health_code_update();
1049
1050 ret = lttng_consumer_read_subbuffer(metadata_stream, ctx, true);
1051 if (ret < 0) {
1052 goto error_stream;
1053 }
1054 } while (ret > 0);
1055
1056 error_stream:
1057 /*
1058 * Clean up the stream completly because the next snapshot will use a new
1059 * metadata stream.
1060 */
1061 consumer_stream_destroy(metadata_stream, NULL);
1062 cds_list_del(&metadata_stream->send_node);
1063 metadata_channel->metadata_stream = NULL;
1064
1065 error:
1066 rcu_read_unlock();
1067 return ret;
1068 }
1069
1070 static
1071 int get_current_subbuf_addr(struct lttng_consumer_stream *stream,
1072 const char **addr)
1073 {
1074 int ret;
1075 unsigned long mmap_offset;
1076 const char *mmap_base;
1077
1078 mmap_base = ustctl_get_mmap_base(stream->ustream);
1079 if (!mmap_base) {
1080 ERR("Failed to get mmap base for stream `%s`",
1081 stream->name);
1082 ret = -EPERM;
1083 goto error;
1084 }
1085
1086 ret = ustctl_get_mmap_read_offset(stream->ustream, &mmap_offset);
1087 if (ret != 0) {
1088 ERR("Failed to get mmap offset for stream `%s`", stream->name);
1089 ret = -EINVAL;
1090 goto error;
1091 }
1092
1093 *addr = mmap_base + mmap_offset;
1094 error:
1095 return ret;
1096
1097 }
1098
1099 /*
1100 * Take a snapshot of all the stream of a channel.
1101 * RCU read-side lock and the channel lock must be held by the caller.
1102 *
1103 * Returns 0 on success, < 0 on error
1104 */
1105 static int snapshot_channel(struct lttng_consumer_channel *channel,
1106 uint64_t key, char *path, uint64_t relayd_id,
1107 uint64_t nb_packets_per_stream,
1108 struct lttng_consumer_local_data *ctx)
1109 {
1110 int ret;
1111 unsigned use_relayd = 0;
1112 unsigned long consumed_pos, produced_pos;
1113 struct lttng_consumer_stream *stream;
1114
1115 assert(path);
1116 assert(ctx);
1117
1118 rcu_read_lock();
1119
1120 if (relayd_id != (uint64_t) -1ULL) {
1121 use_relayd = 1;
1122 }
1123
1124 assert(!channel->monitor);
1125 DBG("UST consumer snapshot channel %" PRIu64, key);
1126
1127 cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
1128 health_code_update();
1129
1130 /* Lock stream because we are about to change its state. */
1131 pthread_mutex_lock(&stream->lock);
1132 assert(channel->trace_chunk);
1133 if (!lttng_trace_chunk_get(channel->trace_chunk)) {
1134 /*
1135 * Can't happen barring an internal error as the channel
1136 * holds a reference to the trace chunk.
1137 */
1138 ERR("Failed to acquire reference to channel's trace chunk");
1139 ret = -1;
1140 goto error_unlock;
1141 }
1142 assert(!stream->trace_chunk);
1143 stream->trace_chunk = channel->trace_chunk;
1144
1145 stream->net_seq_idx = relayd_id;
1146
1147 if (use_relayd) {
1148 ret = consumer_send_relayd_stream(stream, path);
1149 if (ret < 0) {
1150 goto error_unlock;
1151 }
1152 } else {
1153 ret = consumer_stream_create_output_files(stream,
1154 false);
1155 if (ret < 0) {
1156 goto error_unlock;
1157 }
1158 DBG("UST consumer snapshot stream (%" PRIu64 ")",
1159 stream->key);
1160 }
1161
1162 /*
1163 * If tracing is active, we want to perform a "full" buffer flush.
1164 * Else, if quiescent, it has already been done by the prior stop.
1165 */
1166 if (!stream->quiescent) {
1167 ustctl_flush_buffer(stream->ustream, 0);
1168 }
1169
1170 ret = lttng_ustconsumer_take_snapshot(stream);
1171 if (ret < 0) {
1172 ERR("Taking UST snapshot");
1173 goto error_unlock;
1174 }
1175
1176 ret = lttng_ustconsumer_get_produced_snapshot(stream, &produced_pos);
1177 if (ret < 0) {
1178 ERR("Produced UST snapshot position");
1179 goto error_unlock;
1180 }
1181
1182 ret = lttng_ustconsumer_get_consumed_snapshot(stream, &consumed_pos);
1183 if (ret < 0) {
1184 ERR("Consumerd UST snapshot position");
1185 goto error_unlock;
1186 }
1187
1188 /*
1189 * The original value is sent back if max stream size is larger than
1190 * the possible size of the snapshot. Also, we assume that the session
1191 * daemon should never send a maximum stream size that is lower than
1192 * subbuffer size.
1193 */
1194 consumed_pos = consumer_get_consume_start_pos(consumed_pos,
1195 produced_pos, nb_packets_per_stream,
1196 stream->max_sb_size);
1197
1198 while ((long) (consumed_pos - produced_pos) < 0) {
1199 ssize_t read_len;
1200 unsigned long len, padded_len;
1201 const char *subbuf_addr;
1202 struct lttng_buffer_view subbuf_view;
1203
1204 health_code_update();
1205
1206 DBG("UST consumer taking snapshot at pos %lu", consumed_pos);
1207
1208 ret = ustctl_get_subbuf(stream->ustream, &consumed_pos);
1209 if (ret < 0) {
1210 if (ret != -EAGAIN) {
1211 PERROR("ustctl_get_subbuf snapshot");
1212 goto error_close_stream;
1213 }
1214 DBG("UST consumer get subbuf failed. Skipping it.");
1215 consumed_pos += stream->max_sb_size;
1216 stream->chan->lost_packets++;
1217 continue;
1218 }
1219
1220 ret = ustctl_get_subbuf_size(stream->ustream, &len);
1221 if (ret < 0) {
1222 ERR("Snapshot ustctl_get_subbuf_size");
1223 goto error_put_subbuf;
1224 }
1225
1226 ret = ustctl_get_padded_subbuf_size(stream->ustream, &padded_len);
1227 if (ret < 0) {
1228 ERR("Snapshot ustctl_get_padded_subbuf_size");
1229 goto error_put_subbuf;
1230 }
1231
1232 ret = get_current_subbuf_addr(stream, &subbuf_addr);
1233 if (ret) {
1234 goto error_put_subbuf;
1235 }
1236
1237 subbuf_view = lttng_buffer_view_init(
1238 subbuf_addr, 0, padded_len);
1239 read_len = lttng_consumer_on_read_subbuffer_mmap(
1240 stream, &subbuf_view, padded_len - len);
1241 if (use_relayd) {
1242 if (read_len != len) {
1243 ret = -EPERM;
1244 goto error_put_subbuf;
1245 }
1246 } else {
1247 if (read_len != padded_len) {
1248 ret = -EPERM;
1249 goto error_put_subbuf;
1250 }
1251 }
1252
1253 ret = ustctl_put_subbuf(stream->ustream);
1254 if (ret < 0) {
1255 ERR("Snapshot ustctl_put_subbuf");
1256 goto error_close_stream;
1257 }
1258 consumed_pos += stream->max_sb_size;
1259 }
1260
1261 /* Simply close the stream so we can use it on the next snapshot. */
1262 consumer_stream_close(stream);
1263 pthread_mutex_unlock(&stream->lock);
1264 }
1265
1266 rcu_read_unlock();
1267 return 0;
1268
1269 error_put_subbuf:
1270 if (ustctl_put_subbuf(stream->ustream) < 0) {
1271 ERR("Snapshot ustctl_put_subbuf");
1272 }
1273 error_close_stream:
1274 consumer_stream_close(stream);
1275 error_unlock:
1276 pthread_mutex_unlock(&stream->lock);
1277 rcu_read_unlock();
1278 return ret;
1279 }
1280
1281 /*
1282 * Receive the metadata updates from the sessiond. Supports receiving
1283 * overlapping metadata, but is needs to always belong to a contiguous
1284 * range starting from 0.
1285 * Be careful about the locks held when calling this function: it needs
1286 * the metadata cache flush to concurrently progress in order to
1287 * complete.
1288 */
1289 int lttng_ustconsumer_recv_metadata(int sock, uint64_t key, uint64_t offset,
1290 uint64_t len, uint64_t version,
1291 struct lttng_consumer_channel *channel, int timer, int wait)
1292 {
1293 int ret, ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1294 char *metadata_str;
1295
1296 DBG("UST consumer push metadata key %" PRIu64 " of len %" PRIu64, key, len);
1297
1298 metadata_str = zmalloc(len * sizeof(char));
1299 if (!metadata_str) {
1300 PERROR("zmalloc metadata string");
1301 ret_code = LTTCOMM_CONSUMERD_ENOMEM;
1302 goto end;
1303 }
1304
1305 health_code_update();
1306
1307 /* Receive metadata string. */
1308 ret = lttcomm_recv_unix_sock(sock, metadata_str, len);
1309 if (ret < 0) {
1310 /* Session daemon is dead so return gracefully. */
1311 ret_code = ret;
1312 goto end_free;
1313 }
1314
1315 health_code_update();
1316
1317 pthread_mutex_lock(&channel->metadata_cache->lock);
1318 ret = consumer_metadata_cache_write(channel, offset, len, version,
1319 metadata_str);
1320 if (ret < 0) {
1321 /* Unable to handle metadata. Notify session daemon. */
1322 ret_code = LTTCOMM_CONSUMERD_ERROR_METADATA;
1323 /*
1324 * Skip metadata flush on write error since the offset and len might
1325 * not have been updated which could create an infinite loop below when
1326 * waiting for the metadata cache to be flushed.
1327 */
1328 pthread_mutex_unlock(&channel->metadata_cache->lock);
1329 goto end_free;
1330 }
1331 pthread_mutex_unlock(&channel->metadata_cache->lock);
1332
1333 if (!wait) {
1334 goto end_free;
1335 }
1336 while (consumer_metadata_cache_flushed(channel, offset + len, timer)) {
1337 DBG("Waiting for metadata to be flushed");
1338
1339 health_code_update();
1340
1341 usleep(DEFAULT_METADATA_AVAILABILITY_WAIT_TIME);
1342 }
1343
1344 end_free:
1345 free(metadata_str);
1346 end:
1347 return ret_code;
1348 }
1349
1350 /*
1351 * Receive command from session daemon and process it.
1352 *
1353 * Return 1 on success else a negative value or 0.
1354 */
1355 int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
1356 int sock, struct pollfd *consumer_sockpoll)
1357 {
1358 ssize_t ret;
1359 enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1360 struct lttcomm_consumer_msg msg;
1361 struct lttng_consumer_channel *channel = NULL;
1362
1363 health_code_update();
1364
1365 ret = lttcomm_recv_unix_sock(sock, &msg, sizeof(msg));
1366 if (ret != sizeof(msg)) {
1367 DBG("Consumer received unexpected message size %zd (expects %zu)",
1368 ret, sizeof(msg));
1369 /*
1370 * The ret value might 0 meaning an orderly shutdown but this is ok
1371 * since the caller handles this.
1372 */
1373 if (ret > 0) {
1374 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD);
1375 ret = -1;
1376 }
1377 return ret;
1378 }
1379
1380 health_code_update();
1381
1382 /* deprecated */
1383 assert(msg.cmd_type != LTTNG_CONSUMER_STOP);
1384
1385 health_code_update();
1386
1387 /* relayd needs RCU read-side lock */
1388 rcu_read_lock();
1389
1390 switch (msg.cmd_type) {
1391 case LTTNG_CONSUMER_ADD_RELAYD_SOCKET:
1392 {
1393 /* Session daemon status message are handled in the following call. */
1394 consumer_add_relayd_socket(msg.u.relayd_sock.net_index,
1395 msg.u.relayd_sock.type, ctx, sock, consumer_sockpoll,
1396 &msg.u.relayd_sock.sock, msg.u.relayd_sock.session_id,
1397 msg.u.relayd_sock.relayd_session_id);
1398 goto end_nosignal;
1399 }
1400 case LTTNG_CONSUMER_DESTROY_RELAYD:
1401 {
1402 uint64_t index = msg.u.destroy_relayd.net_seq_idx;
1403 struct consumer_relayd_sock_pair *relayd;
1404
1405 DBG("UST consumer destroying relayd %" PRIu64, index);
1406
1407 /* Get relayd reference if exists. */
1408 relayd = consumer_find_relayd(index);
1409 if (relayd == NULL) {
1410 DBG("Unable to find relayd %" PRIu64, index);
1411 ret_code = LTTCOMM_CONSUMERD_RELAYD_FAIL;
1412 }
1413
1414 /*
1415 * Each relayd socket pair has a refcount of stream attached to it
1416 * which tells if the relayd is still active or not depending on the
1417 * refcount value.
1418 *
1419 * This will set the destroy flag of the relayd object and destroy it
1420 * if the refcount reaches zero when called.
1421 *
1422 * The destroy can happen either here or when a stream fd hangs up.
1423 */
1424 if (relayd) {
1425 consumer_flag_relayd_for_destroy(relayd);
1426 }
1427
1428 goto end_msg_sessiond;
1429 }
1430 case LTTNG_CONSUMER_UPDATE_STREAM:
1431 {
1432 rcu_read_unlock();
1433 return -ENOSYS;
1434 }
1435 case LTTNG_CONSUMER_DATA_PENDING:
1436 {
1437 int ret, is_data_pending;
1438 uint64_t id = msg.u.data_pending.session_id;
1439
1440 DBG("UST consumer data pending command for id %" PRIu64, id);
1441
1442 is_data_pending = consumer_data_pending(id);
1443
1444 /* Send back returned value to session daemon */
1445 ret = lttcomm_send_unix_sock(sock, &is_data_pending,
1446 sizeof(is_data_pending));
1447 if (ret < 0) {
1448 DBG("Error when sending the data pending ret code: %d", ret);
1449 goto error_fatal;
1450 }
1451
1452 /*
1453 * No need to send back a status message since the data pending
1454 * returned value is the response.
1455 */
1456 break;
1457 }
1458 case LTTNG_CONSUMER_ASK_CHANNEL_CREATION:
1459 {
1460 int ret;
1461 struct ustctl_consumer_channel_attr attr;
1462 const uint64_t chunk_id = msg.u.ask_channel.chunk_id.value;
1463 const struct lttng_credentials buffer_credentials = {
1464 .uid = msg.u.ask_channel.buffer_credentials.uid,
1465 .gid = msg.u.ask_channel.buffer_credentials.gid,
1466 };
1467
1468 /* Create a plain object and reserve a channel key. */
1469 channel = consumer_allocate_channel(
1470 msg.u.ask_channel.key,
1471 msg.u.ask_channel.session_id,
1472 msg.u.ask_channel.chunk_id.is_set ?
1473 &chunk_id : NULL,
1474 msg.u.ask_channel.pathname,
1475 msg.u.ask_channel.name,
1476 msg.u.ask_channel.relayd_id,
1477 (enum lttng_event_output) msg.u.ask_channel.output,
1478 msg.u.ask_channel.tracefile_size,
1479 msg.u.ask_channel.tracefile_count,
1480 msg.u.ask_channel.session_id_per_pid,
1481 msg.u.ask_channel.monitor,
1482 msg.u.ask_channel.live_timer_interval,
1483 msg.u.ask_channel.is_live,
1484 msg.u.ask_channel.root_shm_path,
1485 msg.u.ask_channel.shm_path);
1486 if (!channel) {
1487 goto end_channel_error;
1488 }
1489
1490 LTTNG_OPTIONAL_SET(&channel->buffer_credentials,
1491 buffer_credentials);
1492
1493 /*
1494 * Assign UST application UID to the channel. This value is ignored for
1495 * per PID buffers. This is specific to UST thus setting this after the
1496 * allocation.
1497 */
1498 channel->ust_app_uid = msg.u.ask_channel.ust_app_uid;
1499
1500 /* Build channel attributes from received message. */
1501 attr.subbuf_size = msg.u.ask_channel.subbuf_size;
1502 attr.num_subbuf = msg.u.ask_channel.num_subbuf;
1503 attr.overwrite = msg.u.ask_channel.overwrite;
1504 attr.switch_timer_interval = msg.u.ask_channel.switch_timer_interval;
1505 attr.read_timer_interval = msg.u.ask_channel.read_timer_interval;
1506 attr.chan_id = msg.u.ask_channel.chan_id;
1507 memcpy(attr.uuid, msg.u.ask_channel.uuid, sizeof(attr.uuid));
1508 attr.blocking_timeout= msg.u.ask_channel.blocking_timeout;
1509
1510 /* Match channel buffer type to the UST abi. */
1511 switch (msg.u.ask_channel.output) {
1512 case LTTNG_EVENT_MMAP:
1513 default:
1514 attr.output = LTTNG_UST_MMAP;
1515 break;
1516 }
1517
1518 /* Translate and save channel type. */
1519 switch (msg.u.ask_channel.type) {
1520 case LTTNG_UST_CHAN_PER_CPU:
1521 channel->type = CONSUMER_CHANNEL_TYPE_DATA;
1522 attr.type = LTTNG_UST_CHAN_PER_CPU;
1523 /*
1524 * Set refcount to 1 for owner. Below, we will
1525 * pass ownership to the
1526 * consumer_thread_channel_poll() thread.
1527 */
1528 channel->refcount = 1;
1529 break;
1530 case LTTNG_UST_CHAN_METADATA:
1531 channel->type = CONSUMER_CHANNEL_TYPE_METADATA;
1532 attr.type = LTTNG_UST_CHAN_METADATA;
1533 break;
1534 default:
1535 assert(0);
1536 goto error_fatal;
1537 };
1538
1539 health_code_update();
1540
1541 ret = ask_channel(ctx, channel, &attr);
1542 if (ret < 0) {
1543 goto end_channel_error;
1544 }
1545
1546 if (msg.u.ask_channel.type == LTTNG_UST_CHAN_METADATA) {
1547 ret = consumer_metadata_cache_allocate(channel);
1548 if (ret < 0) {
1549 ERR("Allocating metadata cache");
1550 goto end_channel_error;
1551 }
1552 consumer_timer_switch_start(channel, attr.switch_timer_interval);
1553 attr.switch_timer_interval = 0;
1554 } else {
1555 int monitor_start_ret;
1556
1557 consumer_timer_live_start(channel,
1558 msg.u.ask_channel.live_timer_interval);
1559 monitor_start_ret = consumer_timer_monitor_start(
1560 channel,
1561 msg.u.ask_channel.monitor_timer_interval);
1562 if (monitor_start_ret < 0) {
1563 ERR("Starting channel monitoring timer failed");
1564 goto end_channel_error;
1565 }
1566 }
1567
1568 health_code_update();
1569
1570 /*
1571 * Add the channel to the internal state AFTER all streams were created
1572 * and successfully sent to session daemon. This way, all streams must
1573 * be ready before this channel is visible to the threads.
1574 * If add_channel succeeds, ownership of the channel is
1575 * passed to consumer_thread_channel_poll().
1576 */
1577 ret = add_channel(channel, ctx);
1578 if (ret < 0) {
1579 if (msg.u.ask_channel.type == LTTNG_UST_CHAN_METADATA) {
1580 if (channel->switch_timer_enabled == 1) {
1581 consumer_timer_switch_stop(channel);
1582 }
1583 consumer_metadata_cache_destroy(channel);
1584 }
1585 if (channel->live_timer_enabled == 1) {
1586 consumer_timer_live_stop(channel);
1587 }
1588 if (channel->monitor_timer_enabled == 1) {
1589 consumer_timer_monitor_stop(channel);
1590 }
1591 goto end_channel_error;
1592 }
1593
1594 health_code_update();
1595
1596 /*
1597 * Channel and streams are now created. Inform the session daemon that
1598 * everything went well and should wait to receive the channel and
1599 * streams with ustctl API.
1600 */
1601 ret = consumer_send_status_channel(sock, channel);
1602 if (ret < 0) {
1603 /*
1604 * There is probably a problem on the socket.
1605 */
1606 goto error_fatal;
1607 }
1608
1609 break;
1610 }
1611 case LTTNG_CONSUMER_GET_CHANNEL:
1612 {
1613 int ret, relayd_err = 0;
1614 uint64_t key = msg.u.get_channel.key;
1615 struct lttng_consumer_channel *channel;
1616
1617 channel = consumer_find_channel(key);
1618 if (!channel) {
1619 ERR("UST consumer get channel key %" PRIu64 " not found", key);
1620 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
1621 goto end_get_channel;
1622 }
1623
1624 health_code_update();
1625
1626 /* Send the channel to sessiond (and relayd, if applicable). */
1627 ret = send_channel_to_sessiond_and_relayd(sock, channel, ctx,
1628 &relayd_err);
1629 if (ret < 0) {
1630 if (relayd_err) {
1631 /*
1632 * We were unable to send to the relayd the stream so avoid
1633 * sending back a fatal error to the thread since this is OK
1634 * and the consumer can continue its work. The above call
1635 * has sent the error status message to the sessiond.
1636 */
1637 goto end_get_channel_nosignal;
1638 }
1639 /*
1640 * The communicaton was broken hence there is a bad state between
1641 * the consumer and sessiond so stop everything.
1642 */
1643 goto error_get_channel_fatal;
1644 }
1645
1646 health_code_update();
1647
1648 /*
1649 * In no monitor mode, the streams ownership is kept inside the channel
1650 * so don't send them to the data thread.
1651 */
1652 if (!channel->monitor) {
1653 goto end_get_channel;
1654 }
1655
1656 ret = send_streams_to_thread(channel, ctx);
1657 if (ret < 0) {
1658 /*
1659 * If we are unable to send the stream to the thread, there is
1660 * a big problem so just stop everything.
1661 */
1662 goto error_get_channel_fatal;
1663 }
1664 /* List MUST be empty after or else it could be reused. */
1665 assert(cds_list_empty(&channel->streams.head));
1666 end_get_channel:
1667 goto end_msg_sessiond;
1668 error_get_channel_fatal:
1669 goto error_fatal;
1670 end_get_channel_nosignal:
1671 goto end_nosignal;
1672 }
1673 case LTTNG_CONSUMER_DESTROY_CHANNEL:
1674 {
1675 uint64_t key = msg.u.destroy_channel.key;
1676
1677 /*
1678 * Only called if streams have not been sent to stream
1679 * manager thread. However, channel has been sent to
1680 * channel manager thread.
1681 */
1682 notify_thread_del_channel(ctx, key);
1683 goto end_msg_sessiond;
1684 }
1685 case LTTNG_CONSUMER_CLOSE_METADATA:
1686 {
1687 int ret;
1688
1689 ret = close_metadata(msg.u.close_metadata.key);
1690 if (ret != 0) {
1691 ret_code = ret;
1692 }
1693
1694 goto end_msg_sessiond;
1695 }
1696 case LTTNG_CONSUMER_FLUSH_CHANNEL:
1697 {
1698 int ret;
1699
1700 ret = flush_channel(msg.u.flush_channel.key);
1701 if (ret != 0) {
1702 ret_code = ret;
1703 }
1704
1705 goto end_msg_sessiond;
1706 }
1707 case LTTNG_CONSUMER_CLEAR_QUIESCENT_CHANNEL:
1708 {
1709 int ret;
1710
1711 ret = clear_quiescent_channel(
1712 msg.u.clear_quiescent_channel.key);
1713 if (ret != 0) {
1714 ret_code = ret;
1715 }
1716
1717 goto end_msg_sessiond;
1718 }
1719 case LTTNG_CONSUMER_PUSH_METADATA:
1720 {
1721 int ret;
1722 uint64_t len = msg.u.push_metadata.len;
1723 uint64_t key = msg.u.push_metadata.key;
1724 uint64_t offset = msg.u.push_metadata.target_offset;
1725 uint64_t version = msg.u.push_metadata.version;
1726 struct lttng_consumer_channel *channel;
1727
1728 DBG("UST consumer push metadata key %" PRIu64 " of len %" PRIu64, key,
1729 len);
1730
1731 channel = consumer_find_channel(key);
1732 if (!channel) {
1733 /*
1734 * This is possible if the metadata creation on the consumer side
1735 * is in flight vis-a-vis a concurrent push metadata from the
1736 * session daemon. Simply return that the channel failed and the
1737 * session daemon will handle that message correctly considering
1738 * that this race is acceptable thus the DBG() statement here.
1739 */
1740 DBG("UST consumer push metadata %" PRIu64 " not found", key);
1741 ret_code = LTTCOMM_CONSUMERD_CHANNEL_FAIL;
1742 goto end_push_metadata_msg_sessiond;
1743 }
1744
1745 health_code_update();
1746
1747 if (!len) {
1748 /*
1749 * There is nothing to receive. We have simply
1750 * checked whether the channel can be found.
1751 */
1752 ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1753 goto end_push_metadata_msg_sessiond;
1754 }
1755
1756 /* Tell session daemon we are ready to receive the metadata. */
1757 ret = consumer_send_status_msg(sock, LTTCOMM_CONSUMERD_SUCCESS);
1758 if (ret < 0) {
1759 /* Somehow, the session daemon is not responding anymore. */
1760 goto error_push_metadata_fatal;
1761 }
1762
1763 health_code_update();
1764
1765 /* Wait for more data. */
1766 health_poll_entry();
1767 ret = lttng_consumer_poll_socket(consumer_sockpoll);
1768 health_poll_exit();
1769 if (ret) {
1770 goto error_push_metadata_fatal;
1771 }
1772
1773 health_code_update();
1774
1775 ret = lttng_ustconsumer_recv_metadata(sock, key, offset,
1776 len, version, channel, 0, 1);
1777 if (ret < 0) {
1778 /* error receiving from sessiond */
1779 goto error_push_metadata_fatal;
1780 } else {
1781 ret_code = ret;
1782 goto end_push_metadata_msg_sessiond;
1783 }
1784 end_push_metadata_msg_sessiond:
1785 goto end_msg_sessiond;
1786 error_push_metadata_fatal:
1787 goto error_fatal;
1788 }
1789 case LTTNG_CONSUMER_SETUP_METADATA:
1790 {
1791 int ret;
1792
1793 ret = setup_metadata(ctx, msg.u.setup_metadata.key);
1794 if (ret) {
1795 ret_code = ret;
1796 }
1797 goto end_msg_sessiond;
1798 }
1799 case LTTNG_CONSUMER_SNAPSHOT_CHANNEL:
1800 {
1801 struct lttng_consumer_channel *channel;
1802 uint64_t key = msg.u.snapshot_channel.key;
1803
1804 channel = consumer_find_channel(key);
1805 if (!channel) {
1806 DBG("UST snapshot channel not found for key %" PRIu64, key);
1807 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
1808 } else {
1809 if (msg.u.snapshot_channel.metadata) {
1810 ret = snapshot_metadata(channel, key,
1811 msg.u.snapshot_channel.pathname,
1812 msg.u.snapshot_channel.relayd_id,
1813 ctx);
1814 if (ret < 0) {
1815 ERR("Snapshot metadata failed");
1816 ret_code = LTTCOMM_CONSUMERD_SNAPSHOT_FAILED;
1817 }
1818 } else {
1819 ret = snapshot_channel(channel, key,
1820 msg.u.snapshot_channel.pathname,
1821 msg.u.snapshot_channel.relayd_id,
1822 msg.u.snapshot_channel.nb_packets_per_stream,
1823 ctx);
1824 if (ret < 0) {
1825 ERR("Snapshot channel failed");
1826 ret_code = LTTCOMM_CONSUMERD_SNAPSHOT_FAILED;
1827 }
1828 }
1829 }
1830 health_code_update();
1831 ret = consumer_send_status_msg(sock, ret_code);
1832 if (ret < 0) {
1833 /* Somehow, the session daemon is not responding anymore. */
1834 goto end_nosignal;
1835 }
1836 health_code_update();
1837 break;
1838 }
1839 case LTTNG_CONSUMER_DISCARDED_EVENTS:
1840 {
1841 int ret = 0;
1842 uint64_t discarded_events;
1843 struct lttng_ht_iter iter;
1844 struct lttng_ht *ht;
1845 struct lttng_consumer_stream *stream;
1846 uint64_t id = msg.u.discarded_events.session_id;
1847 uint64_t key = msg.u.discarded_events.channel_key;
1848
1849 DBG("UST consumer discarded events command for session id %"
1850 PRIu64, id);
1851 rcu_read_lock();
1852 pthread_mutex_lock(&consumer_data.lock);
1853
1854 ht = consumer_data.stream_list_ht;
1855
1856 /*
1857 * We only need a reference to the channel, but they are not
1858 * directly indexed, so we just use the first matching stream
1859 * to extract the information we need, we default to 0 if not
1860 * found (no events are dropped if the channel is not yet in
1861 * use).
1862 */
1863 discarded_events = 0;
1864 cds_lfht_for_each_entry_duplicate(ht->ht,
1865 ht->hash_fct(&id, lttng_ht_seed),
1866 ht->match_fct, &id,
1867 &iter.iter, stream, node_session_id.node) {
1868 if (stream->chan->key == key) {
1869 discarded_events = stream->chan->discarded_events;
1870 break;
1871 }
1872 }
1873 pthread_mutex_unlock(&consumer_data.lock);
1874 rcu_read_unlock();
1875
1876 DBG("UST consumer discarded events command for session id %"
1877 PRIu64 ", channel key %" PRIu64, id, key);
1878
1879 health_code_update();
1880
1881 /* Send back returned value to session daemon */
1882 ret = lttcomm_send_unix_sock(sock, &discarded_events, sizeof(discarded_events));
1883 if (ret < 0) {
1884 PERROR("send discarded events");
1885 goto error_fatal;
1886 }
1887
1888 break;
1889 }
1890 case LTTNG_CONSUMER_LOST_PACKETS:
1891 {
1892 int ret;
1893 uint64_t lost_packets;
1894 struct lttng_ht_iter iter;
1895 struct lttng_ht *ht;
1896 struct lttng_consumer_stream *stream;
1897 uint64_t id = msg.u.lost_packets.session_id;
1898 uint64_t key = msg.u.lost_packets.channel_key;
1899
1900 DBG("UST consumer lost packets command for session id %"
1901 PRIu64, id);
1902 rcu_read_lock();
1903 pthread_mutex_lock(&consumer_data.lock);
1904
1905 ht = consumer_data.stream_list_ht;
1906
1907 /*
1908 * We only need a reference to the channel, but they are not
1909 * directly indexed, so we just use the first matching stream
1910 * to extract the information we need, we default to 0 if not
1911 * found (no packets lost if the channel is not yet in use).
1912 */
1913 lost_packets = 0;
1914 cds_lfht_for_each_entry_duplicate(ht->ht,
1915 ht->hash_fct(&id, lttng_ht_seed),
1916 ht->match_fct, &id,
1917 &iter.iter, stream, node_session_id.node) {
1918 if (stream->chan->key == key) {
1919 lost_packets = stream->chan->lost_packets;
1920 break;
1921 }
1922 }
1923 pthread_mutex_unlock(&consumer_data.lock);
1924 rcu_read_unlock();
1925
1926 DBG("UST consumer lost packets command for session id %"
1927 PRIu64 ", channel key %" PRIu64, id, key);
1928
1929 health_code_update();
1930
1931 /* Send back returned value to session daemon */
1932 ret = lttcomm_send_unix_sock(sock, &lost_packets,
1933 sizeof(lost_packets));
1934 if (ret < 0) {
1935 PERROR("send lost packets");
1936 goto error_fatal;
1937 }
1938
1939 break;
1940 }
1941 case LTTNG_CONSUMER_SET_CHANNEL_MONITOR_PIPE:
1942 {
1943 int channel_monitor_pipe;
1944
1945 ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1946 /* Successfully received the command's type. */
1947 ret = consumer_send_status_msg(sock, ret_code);
1948 if (ret < 0) {
1949 goto error_fatal;
1950 }
1951
1952 ret = lttcomm_recv_fds_unix_sock(sock, &channel_monitor_pipe,
1953 1);
1954 if (ret != sizeof(channel_monitor_pipe)) {
1955 ERR("Failed to receive channel monitor pipe");
1956 goto error_fatal;
1957 }
1958
1959 DBG("Received channel monitor pipe (%d)", channel_monitor_pipe);
1960 ret = consumer_timer_thread_set_channel_monitor_pipe(
1961 channel_monitor_pipe);
1962 if (!ret) {
1963 int flags;
1964
1965 ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1966 /* Set the pipe as non-blocking. */
1967 ret = fcntl(channel_monitor_pipe, F_GETFL, 0);
1968 if (ret == -1) {
1969 PERROR("fcntl get flags of the channel monitoring pipe");
1970 goto error_fatal;
1971 }
1972 flags = ret;
1973
1974 ret = fcntl(channel_monitor_pipe, F_SETFL,
1975 flags | O_NONBLOCK);
1976 if (ret == -1) {
1977 PERROR("fcntl set O_NONBLOCK flag of the channel monitoring pipe");
1978 goto error_fatal;
1979 }
1980 DBG("Channel monitor pipe set as non-blocking");
1981 } else {
1982 ret_code = LTTCOMM_CONSUMERD_ALREADY_SET;
1983 }
1984 goto end_msg_sessiond;
1985 }
1986 case LTTNG_CONSUMER_ROTATE_CHANNEL:
1987 {
1988 struct lttng_consumer_channel *channel;
1989 uint64_t key = msg.u.rotate_channel.key;
1990
1991 channel = consumer_find_channel(key);
1992 if (!channel) {
1993 DBG("Channel %" PRIu64 " not found", key);
1994 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
1995 } else {
1996 /*
1997 * Sample the rotate position of all the streams in
1998 * this channel.
1999 */
2000 ret = lttng_consumer_rotate_channel(channel, key,
2001 msg.u.rotate_channel.relayd_id,
2002 msg.u.rotate_channel.metadata,
2003 ctx);
2004 if (ret < 0) {
2005 ERR("Rotate channel failed");
2006 ret_code = LTTCOMM_CONSUMERD_ROTATION_FAIL;
2007 }
2008
2009 health_code_update();
2010 }
2011 ret = consumer_send_status_msg(sock, ret_code);
2012 if (ret < 0) {
2013 /* Somehow, the session daemon is not responding anymore. */
2014 goto end_rotate_channel_nosignal;
2015 }
2016
2017 /*
2018 * Rotate the streams that are ready right now.
2019 * FIXME: this is a second consecutive iteration over the
2020 * streams in a channel, there is probably a better way to
2021 * handle this, but it needs to be after the
2022 * consumer_send_status_msg() call.
2023 */
2024 if (channel) {
2025 ret = lttng_consumer_rotate_ready_streams(
2026 channel, key, ctx);
2027 if (ret < 0) {
2028 ERR("Rotate channel failed");
2029 }
2030 }
2031 break;
2032 end_rotate_channel_nosignal:
2033 goto end_nosignal;
2034 }
2035 case LTTNG_CONSUMER_CLEAR_CHANNEL:
2036 {
2037 struct lttng_consumer_channel *channel;
2038 uint64_t key = msg.u.clear_channel.key;
2039
2040 channel = consumer_find_channel(key);
2041 if (!channel) {
2042 DBG("Channel %" PRIu64 " not found", key);
2043 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
2044 } else {
2045 ret = lttng_consumer_clear_channel(channel);
2046 if (ret) {
2047 ERR("Clear channel failed key %" PRIu64, key);
2048 ret_code = ret;
2049 }
2050
2051 health_code_update();
2052 }
2053 ret = consumer_send_status_msg(sock, ret_code);
2054 if (ret < 0) {
2055 /* Somehow, the session daemon is not responding anymore. */
2056 goto end_nosignal;
2057 }
2058 break;
2059 }
2060 case LTTNG_CONSUMER_INIT:
2061 {
2062 ret_code = lttng_consumer_init_command(ctx,
2063 msg.u.init.sessiond_uuid);
2064 health_code_update();
2065 ret = consumer_send_status_msg(sock, ret_code);
2066 if (ret < 0) {
2067 /* Somehow, the session daemon is not responding anymore. */
2068 goto end_nosignal;
2069 }
2070 break;
2071 }
2072 case LTTNG_CONSUMER_CREATE_TRACE_CHUNK:
2073 {
2074 const struct lttng_credentials credentials = {
2075 .uid = msg.u.create_trace_chunk.credentials.value.uid,
2076 .gid = msg.u.create_trace_chunk.credentials.value.gid,
2077 };
2078 const bool is_local_trace =
2079 !msg.u.create_trace_chunk.relayd_id.is_set;
2080 const uint64_t relayd_id =
2081 msg.u.create_trace_chunk.relayd_id.value;
2082 const char *chunk_override_name =
2083 *msg.u.create_trace_chunk.override_name ?
2084 msg.u.create_trace_chunk.override_name :
2085 NULL;
2086 struct lttng_directory_handle *chunk_directory_handle = NULL;
2087
2088 /*
2089 * The session daemon will only provide a chunk directory file
2090 * descriptor for local traces.
2091 */
2092 if (is_local_trace) {
2093 int chunk_dirfd;
2094
2095 /* Acnowledge the reception of the command. */
2096 ret = consumer_send_status_msg(sock,
2097 LTTCOMM_CONSUMERD_SUCCESS);
2098 if (ret < 0) {
2099 /* Somehow, the session daemon is not responding anymore. */
2100 goto end_nosignal;
2101 }
2102
2103 /*
2104 * Receive trace chunk domain dirfd.
2105 */
2106 ret = lttcomm_recv_fds_unix_sock(sock, &chunk_dirfd, 1);
2107 if (ret != sizeof(chunk_dirfd)) {
2108 ERR("Failed to receive trace chunk domain directory file descriptor");
2109 goto error_fatal;
2110 }
2111
2112 DBG("Received trace chunk domain directory fd (%d)",
2113 chunk_dirfd);
2114 chunk_directory_handle = lttng_directory_handle_create_from_dirfd(
2115 chunk_dirfd);
2116 if (!chunk_directory_handle) {
2117 ERR("Failed to initialize chunk domain directory handle from directory file descriptor");
2118 if (close(chunk_dirfd)) {
2119 PERROR("Failed to close chunk directory file descriptor");
2120 }
2121 goto error_fatal;
2122 }
2123 }
2124
2125 ret_code = lttng_consumer_create_trace_chunk(
2126 !is_local_trace ? &relayd_id : NULL,
2127 msg.u.create_trace_chunk.session_id,
2128 msg.u.create_trace_chunk.chunk_id,
2129 (time_t) msg.u.create_trace_chunk
2130 .creation_timestamp,
2131 chunk_override_name,
2132 msg.u.create_trace_chunk.credentials.is_set ?
2133 &credentials :
2134 NULL,
2135 chunk_directory_handle);
2136 lttng_directory_handle_put(chunk_directory_handle);
2137 goto end_msg_sessiond;
2138 }
2139 case LTTNG_CONSUMER_CLOSE_TRACE_CHUNK:
2140 {
2141 enum lttng_trace_chunk_command_type close_command =
2142 msg.u.close_trace_chunk.close_command.value;
2143 const uint64_t relayd_id =
2144 msg.u.close_trace_chunk.relayd_id.value;
2145 struct lttcomm_consumer_close_trace_chunk_reply reply;
2146 char closed_trace_chunk_path[LTTNG_PATH_MAX];
2147 int ret;
2148
2149 ret_code = lttng_consumer_close_trace_chunk(
2150 msg.u.close_trace_chunk.relayd_id.is_set ?
2151 &relayd_id :
2152 NULL,
2153 msg.u.close_trace_chunk.session_id,
2154 msg.u.close_trace_chunk.chunk_id,
2155 (time_t) msg.u.close_trace_chunk.close_timestamp,
2156 msg.u.close_trace_chunk.close_command.is_set ?
2157 &close_command :
2158 NULL, closed_trace_chunk_path);
2159 reply.ret_code = ret_code;
2160 reply.path_length = strlen(closed_trace_chunk_path) + 1;
2161 ret = lttcomm_send_unix_sock(sock, &reply, sizeof(reply));
2162 if (ret != sizeof(reply)) {
2163 goto error_fatal;
2164 }
2165 ret = lttcomm_send_unix_sock(sock, closed_trace_chunk_path,
2166 reply.path_length);
2167 if (ret != reply.path_length) {
2168 goto error_fatal;
2169 }
2170 goto end_nosignal;
2171 }
2172 case LTTNG_CONSUMER_TRACE_CHUNK_EXISTS:
2173 {
2174 const uint64_t relayd_id =
2175 msg.u.trace_chunk_exists.relayd_id.value;
2176
2177 ret_code = lttng_consumer_trace_chunk_exists(
2178 msg.u.trace_chunk_exists.relayd_id.is_set ?
2179 &relayd_id : NULL,
2180 msg.u.trace_chunk_exists.session_id,
2181 msg.u.trace_chunk_exists.chunk_id);
2182 goto end_msg_sessiond;
2183 }
2184 default:
2185 break;
2186 }
2187
2188 end_nosignal:
2189 /*
2190 * Return 1 to indicate success since the 0 value can be a socket
2191 * shutdown during the recv() or send() call.
2192 */
2193 ret = 1;
2194 goto end;
2195
2196 end_msg_sessiond:
2197 /*
2198 * The returned value here is not useful since either way we'll return 1 to
2199 * the caller because the session daemon socket management is done
2200 * elsewhere. Returning a negative code or 0 will shutdown the consumer.
2201 */
2202 ret = consumer_send_status_msg(sock, ret_code);
2203 if (ret < 0) {
2204 goto error_fatal;
2205 }
2206 ret = 1;
2207 goto end;
2208
2209 end_channel_error:
2210 if (channel) {
2211 /*
2212 * Free channel here since no one has a reference to it. We don't
2213 * free after that because a stream can store this pointer.
2214 */
2215 destroy_channel(channel);
2216 }
2217 /* We have to send a status channel message indicating an error. */
2218 ret = consumer_send_status_channel(sock, NULL);
2219 if (ret < 0) {
2220 /* Stop everything if session daemon can not be notified. */
2221 goto error_fatal;
2222 }
2223 ret = 1;
2224 goto end;
2225
2226 error_fatal:
2227 /* This will issue a consumer stop. */
2228 ret = -1;
2229 goto end;
2230
2231 end:
2232 rcu_read_unlock();
2233 health_code_update();
2234 return ret;
2235 }
2236
2237 void lttng_ustctl_flush_buffer(struct lttng_consumer_stream *stream,
2238 int producer_active)
2239 {
2240 assert(stream);
2241 assert(stream->ustream);
2242
2243 ustctl_flush_buffer(stream->ustream, producer_active);
2244 }
2245
2246 /*
2247 * Take a snapshot for a specific stream.
2248 *
2249 * Returns 0 on success, < 0 on error
2250 */
2251 int lttng_ustconsumer_take_snapshot(struct lttng_consumer_stream *stream)
2252 {
2253 assert(stream);
2254 assert(stream->ustream);
2255
2256 return ustctl_snapshot(stream->ustream);
2257 }
2258
2259 /*
2260 * Sample consumed and produced positions for a specific stream.
2261 *
2262 * Returns 0 on success, < 0 on error.
2263 */
2264 int lttng_ustconsumer_sample_snapshot_positions(
2265 struct lttng_consumer_stream *stream)
2266 {
2267 assert(stream);
2268 assert(stream->ustream);
2269
2270 return ustctl_snapshot_sample_positions(stream->ustream);
2271 }
2272
2273 /*
2274 * Get the produced position
2275 *
2276 * Returns 0 on success, < 0 on error
2277 */
2278 int lttng_ustconsumer_get_produced_snapshot(
2279 struct lttng_consumer_stream *stream, unsigned long *pos)
2280 {
2281 assert(stream);
2282 assert(stream->ustream);
2283 assert(pos);
2284
2285 return ustctl_snapshot_get_produced(stream->ustream, pos);
2286 }
2287
2288 /*
2289 * Get the consumed position
2290 *
2291 * Returns 0 on success, < 0 on error
2292 */
2293 int lttng_ustconsumer_get_consumed_snapshot(
2294 struct lttng_consumer_stream *stream, unsigned long *pos)
2295 {
2296 assert(stream);
2297 assert(stream->ustream);
2298 assert(pos);
2299
2300 return ustctl_snapshot_get_consumed(stream->ustream, pos);
2301 }
2302
2303 void lttng_ustconsumer_flush_buffer(struct lttng_consumer_stream *stream,
2304 int producer)
2305 {
2306 assert(stream);
2307 assert(stream->ustream);
2308
2309 ustctl_flush_buffer(stream->ustream, producer);
2310 }
2311
2312 void lttng_ustconsumer_clear_buffer(struct lttng_consumer_stream *stream)
2313 {
2314 assert(stream);
2315 assert(stream->ustream);
2316
2317 ustctl_clear_buffer(stream->ustream);
2318 }
2319
2320 int lttng_ustconsumer_get_current_timestamp(
2321 struct lttng_consumer_stream *stream, uint64_t *ts)
2322 {
2323 assert(stream);
2324 assert(stream->ustream);
2325 assert(ts);
2326
2327 return ustctl_get_current_timestamp(stream->ustream, ts);
2328 }
2329
2330 int lttng_ustconsumer_get_sequence_number(
2331 struct lttng_consumer_stream *stream, uint64_t *seq)
2332 {
2333 assert(stream);
2334 assert(stream->ustream);
2335 assert(seq);
2336
2337 return ustctl_get_sequence_number(stream->ustream, seq);
2338 }
2339
2340 /*
2341 * Called when the stream signals the consumer that it has hung up.
2342 */
2343 void lttng_ustconsumer_on_stream_hangup(struct lttng_consumer_stream *stream)
2344 {
2345 assert(stream);
2346 assert(stream->ustream);
2347
2348 pthread_mutex_lock(&stream->lock);
2349 if (!stream->quiescent) {
2350 ustctl_flush_buffer(stream->ustream, 0);
2351 stream->quiescent = true;
2352 }
2353 pthread_mutex_unlock(&stream->lock);
2354 stream->hangup_flush_done = 1;
2355 }
2356
2357 void lttng_ustconsumer_del_channel(struct lttng_consumer_channel *chan)
2358 {
2359 int i;
2360
2361 assert(chan);
2362 assert(chan->uchan);
2363 assert(chan->buffer_credentials.is_set);
2364
2365 if (chan->switch_timer_enabled == 1) {
2366 consumer_timer_switch_stop(chan);
2367 }
2368 for (i = 0; i < chan->nr_stream_fds; i++) {
2369 int ret;
2370
2371 ret = close(chan->stream_fds[i]);
2372 if (ret) {
2373 PERROR("close");
2374 }
2375 if (chan->shm_path[0]) {
2376 char shm_path[PATH_MAX];
2377
2378 ret = get_stream_shm_path(shm_path, chan->shm_path, i);
2379 if (ret) {
2380 ERR("Cannot get stream shm path");
2381 }
2382 ret = run_as_unlink(shm_path,
2383 chan->buffer_credentials.value.uid,
2384 chan->buffer_credentials.value.gid);
2385 if (ret) {
2386 PERROR("unlink %s", shm_path);
2387 }
2388 }
2389 }
2390 }
2391
2392 void lttng_ustconsumer_free_channel(struct lttng_consumer_channel *chan)
2393 {
2394 assert(chan);
2395 assert(chan->uchan);
2396 assert(chan->buffer_credentials.is_set);
2397
2398 consumer_metadata_cache_destroy(chan);
2399 ustctl_destroy_channel(chan->uchan);
2400 /* Try to rmdir all directories under shm_path root. */
2401 if (chan->root_shm_path[0]) {
2402 (void) run_as_rmdir_recursive(chan->root_shm_path,
2403 chan->buffer_credentials.value.uid,
2404 chan->buffer_credentials.value.gid,
2405 LTTNG_DIRECTORY_HANDLE_SKIP_NON_EMPTY_FLAG);
2406 }
2407 free(chan->stream_fds);
2408 }
2409
2410 void lttng_ustconsumer_del_stream(struct lttng_consumer_stream *stream)
2411 {
2412 assert(stream);
2413 assert(stream->ustream);
2414
2415 if (stream->chan->switch_timer_enabled == 1) {
2416 consumer_timer_switch_stop(stream->chan);
2417 }
2418 ustctl_destroy_stream(stream->ustream);
2419 }
2420
2421 int lttng_ustconsumer_get_wakeup_fd(struct lttng_consumer_stream *stream)
2422 {
2423 assert(stream);
2424 assert(stream->ustream);
2425
2426 return ustctl_stream_get_wakeup_fd(stream->ustream);
2427 }
2428
2429 int lttng_ustconsumer_close_wakeup_fd(struct lttng_consumer_stream *stream)
2430 {
2431 assert(stream);
2432 assert(stream->ustream);
2433
2434 return ustctl_stream_close_wakeup_fd(stream->ustream);
2435 }
2436
2437 static
2438 void metadata_stream_reset_cache_consumed_position(
2439 struct lttng_consumer_stream *stream)
2440 {
2441 DBG("Reset metadata cache of session %" PRIu64,
2442 stream->chan->session_id);
2443 stream->ust_metadata_pushed = 0;
2444 }
2445
2446 /*
2447 * Write up to one packet from the metadata cache to the channel.
2448 *
2449 * Returns the number of bytes pushed in the cache, or a negative value
2450 * on error.
2451 */
2452 static
2453 int commit_one_metadata_packet(struct lttng_consumer_stream *stream)
2454 {
2455 ssize_t write_len;
2456 int ret;
2457
2458 pthread_mutex_lock(&stream->chan->metadata_cache->lock);
2459 if (stream->chan->metadata_cache->max_offset ==
2460 stream->ust_metadata_pushed) {
2461 /*
2462 * In the context of a user space metadata channel, a
2463 * change in version can be detected in two ways:
2464 * 1) During the pre-consume of the `read_subbuffer` loop,
2465 * 2) When populating the metadata ring buffer (i.e. here).
2466 *
2467 * This function is invoked when there is no metadata
2468 * available in the ring-buffer. If all data was consumed
2469 * up to the size of the metadata cache, there is no metadata
2470 * to insert in the ring-buffer.
2471 *
2472 * However, the metadata version could still have changed (a
2473 * regeneration without any new data will yield the same cache
2474 * size).
2475 *
2476 * The cache's version is checked for a version change and the
2477 * consumed position is reset if one occurred.
2478 *
2479 * This check is only necessary for the user space domain as
2480 * it has to manage the cache explicitly. If this reset was not
2481 * performed, no metadata would be consumed (and no reset would
2482 * occur as part of the pre-consume) until the metadata size
2483 * exceeded the cache size.
2484 */
2485 if (stream->metadata_version !=
2486 stream->chan->metadata_cache->version) {
2487 metadata_stream_reset_cache_consumed_position(stream);
2488 consumer_stream_metadata_set_version(stream,
2489 stream->chan->metadata_cache->version);
2490 } else {
2491 ret = 0;
2492 goto end;
2493 }
2494 }
2495
2496 write_len = ustctl_write_one_packet_to_channel(stream->chan->uchan,
2497 &stream->chan->metadata_cache->data[stream->ust_metadata_pushed],
2498 stream->chan->metadata_cache->max_offset
2499 - stream->ust_metadata_pushed);
2500 assert(write_len != 0);
2501 if (write_len < 0) {
2502 ERR("Writing one metadata packet");
2503 ret = write_len;
2504 goto end;
2505 }
2506 stream->ust_metadata_pushed += write_len;
2507
2508 assert(stream->chan->metadata_cache->max_offset >=
2509 stream->ust_metadata_pushed);
2510 ret = write_len;
2511
2512 /*
2513 * Switch packet (but don't open the next one) on every commit of
2514 * a metadata packet. Since the subbuffer is fully filled (with padding,
2515 * if needed), the stream is "quiescent" after this commit.
2516 */
2517 ustctl_flush_buffer(stream->ustream, 1);
2518 stream->quiescent = true;
2519 end:
2520 pthread_mutex_unlock(&stream->chan->metadata_cache->lock);
2521 return ret;
2522 }
2523
2524
2525 /*
2526 * Sync metadata meaning request them to the session daemon and snapshot to the
2527 * metadata thread can consumer them.
2528 *
2529 * Metadata stream lock is held here, but we need to release it when
2530 * interacting with sessiond, else we cause a deadlock with live
2531 * awaiting on metadata to be pushed out.
2532 *
2533 * The RCU read side lock must be held by the caller.
2534 *
2535 * Return 0 if new metadatda is available, EAGAIN if the metadata stream
2536 * is empty or a negative value on error.
2537 */
2538 int lttng_ustconsumer_sync_metadata(struct lttng_consumer_local_data *ctx,
2539 struct lttng_consumer_stream *metadata_stream)
2540 {
2541 int ret;
2542 int retry = 0;
2543 struct lttng_consumer_channel *metadata_channel;
2544
2545 assert(ctx);
2546 assert(metadata_stream);
2547
2548 metadata_channel = metadata_stream->chan;
2549 pthread_mutex_unlock(&metadata_stream->lock);
2550 /*
2551 * Request metadata from the sessiond, but don't wait for the flush
2552 * because we locked the metadata thread.
2553 */
2554 ret = lttng_ustconsumer_request_metadata(ctx, metadata_channel, 0, 0);
2555 pthread_mutex_lock(&metadata_stream->lock);
2556 if (ret < 0) {
2557 goto end;
2558 }
2559
2560 /*
2561 * The metadata stream and channel can be deleted while the
2562 * metadata stream lock was released. The streamed is checked
2563 * for deletion before we use it further.
2564 *
2565 * Note that it is safe to access a logically-deleted stream since its
2566 * existence is still guaranteed by the RCU read side lock. However,
2567 * it should no longer be used. The close/deletion of the metadata
2568 * channel and stream already guarantees that all metadata has been
2569 * consumed. Therefore, there is nothing left to do in this function.
2570 */
2571 if (consumer_stream_is_deleted(metadata_stream)) {
2572 DBG("Metadata stream %" PRIu64 " was deleted during the metadata synchronization",
2573 metadata_stream->key);
2574 ret = 0;
2575 goto end;
2576 }
2577
2578 ret = commit_one_metadata_packet(metadata_stream);
2579 if (ret <= 0) {
2580 goto end;
2581 } else if (ret > 0) {
2582 retry = 1;
2583 }
2584
2585 ret = ustctl_snapshot(metadata_stream->ustream);
2586 if (ret < 0) {
2587 if (errno != EAGAIN) {
2588 ERR("Sync metadata, taking UST snapshot");
2589 goto end;
2590 }
2591 DBG("No new metadata when syncing them.");
2592 /* No new metadata, exit. */
2593 ret = ENODATA;
2594 goto end;
2595 }
2596
2597 /*
2598 * After this flush, we still need to extract metadata.
2599 */
2600 if (retry) {
2601 ret = EAGAIN;
2602 }
2603
2604 end:
2605 return ret;
2606 }
2607
2608 /*
2609 * Return 0 on success else a negative value.
2610 */
2611 static int notify_if_more_data(struct lttng_consumer_stream *stream,
2612 struct lttng_consumer_local_data *ctx)
2613 {
2614 int ret;
2615 struct ustctl_consumer_stream *ustream;
2616
2617 assert(stream);
2618 assert(ctx);
2619
2620 ustream = stream->ustream;
2621
2622 /*
2623 * First, we are going to check if there is a new subbuffer available
2624 * before reading the stream wait_fd.
2625 */
2626 /* Get the next subbuffer */
2627 ret = ustctl_get_next_subbuf(ustream);
2628 if (ret) {
2629 /* No more data found, flag the stream. */
2630 stream->has_data = 0;
2631 ret = 0;
2632 goto end;
2633 }
2634
2635 ret = ustctl_put_subbuf(ustream);
2636 assert(!ret);
2637
2638 /* This stream still has data. Flag it and wake up the data thread. */
2639 stream->has_data = 1;
2640
2641 if (stream->monitor && !stream->hangup_flush_done && !ctx->has_wakeup) {
2642 ssize_t writelen;
2643
2644 writelen = lttng_pipe_write(ctx->consumer_wakeup_pipe, "!", 1);
2645 if (writelen < 0 && errno != EAGAIN && errno != EWOULDBLOCK) {
2646 ret = writelen;
2647 goto end;
2648 }
2649
2650 /* The wake up pipe has been notified. */
2651 ctx->has_wakeup = 1;
2652 }
2653 ret = 0;
2654
2655 end:
2656 return ret;
2657 }
2658
2659 static int consumer_stream_ust_on_wake_up(struct lttng_consumer_stream *stream)
2660 {
2661 int ret = 0;
2662
2663 /*
2664 * We can consume the 1 byte written into the wait_fd by
2665 * UST. Don't trigger error if we cannot read this one byte
2666 * (read returns 0), or if the error is EAGAIN or EWOULDBLOCK.
2667 *
2668 * This is only done when the stream is monitored by a thread,
2669 * before the flush is done after a hangup and if the stream
2670 * is not flagged with data since there might be nothing to
2671 * consume in the wait fd but still have data available
2672 * flagged by the consumer wake up pipe.
2673 */
2674 if (stream->monitor && !stream->hangup_flush_done && !stream->has_data) {
2675 char dummy;
2676 ssize_t readlen;
2677
2678 readlen = lttng_read(stream->wait_fd, &dummy, 1);
2679 if (readlen < 0 && errno != EAGAIN && errno != EWOULDBLOCK) {
2680 ret = readlen;
2681 }
2682 }
2683
2684 return ret;
2685 }
2686
2687 static int extract_common_subbuffer_info(struct lttng_consumer_stream *stream,
2688 struct stream_subbuffer *subbuf)
2689 {
2690 int ret;
2691
2692 ret = ustctl_get_subbuf_size(
2693 stream->ustream, &subbuf->info.data.subbuf_size);
2694 if (ret) {
2695 goto end;
2696 }
2697
2698 ret = ustctl_get_padded_subbuf_size(
2699 stream->ustream, &subbuf->info.data.padded_subbuf_size);
2700 if (ret) {
2701 goto end;
2702 }
2703
2704 end:
2705 return ret;
2706 }
2707
2708 static int extract_metadata_subbuffer_info(struct lttng_consumer_stream *stream,
2709 struct stream_subbuffer *subbuf)
2710 {
2711 int ret;
2712
2713 ret = extract_common_subbuffer_info(stream, subbuf);
2714 if (ret) {
2715 goto end;
2716 }
2717
2718 subbuf->info.metadata.version = stream->metadata_version;
2719
2720 end:
2721 return ret;
2722 }
2723
2724 static int extract_data_subbuffer_info(struct lttng_consumer_stream *stream,
2725 struct stream_subbuffer *subbuf)
2726 {
2727 int ret;
2728
2729 ret = extract_common_subbuffer_info(stream, subbuf);
2730 if (ret) {
2731 goto end;
2732 }
2733
2734 ret = ustctl_get_packet_size(
2735 stream->ustream, &subbuf->info.data.packet_size);
2736 if (ret < 0) {
2737 PERROR("Failed to get sub-buffer packet size");
2738 goto end;
2739 }
2740
2741 ret = ustctl_get_content_size(
2742 stream->ustream, &subbuf->info.data.content_size);
2743 if (ret < 0) {
2744 PERROR("Failed to get sub-buffer content size");
2745 goto end;
2746 }
2747
2748 ret = ustctl_get_timestamp_begin(
2749 stream->ustream, &subbuf->info.data.timestamp_begin);
2750 if (ret < 0) {
2751 PERROR("Failed to get sub-buffer begin timestamp");
2752 goto end;
2753 }
2754
2755 ret = ustctl_get_timestamp_end(
2756 stream->ustream, &subbuf->info.data.timestamp_end);
2757 if (ret < 0) {
2758 PERROR("Failed to get sub-buffer end timestamp");
2759 goto end;
2760 }
2761
2762 ret = ustctl_get_events_discarded(
2763 stream->ustream, &subbuf->info.data.events_discarded);
2764 if (ret) {
2765 PERROR("Failed to get sub-buffer events discarded count");
2766 goto end;
2767 }
2768
2769 ret = ustctl_get_sequence_number(stream->ustream,
2770 &subbuf->info.data.sequence_number.value);
2771 if (ret) {
2772 /* May not be supported by older LTTng-modules. */
2773 if (ret != -ENOTTY) {
2774 PERROR("Failed to get sub-buffer sequence number");
2775 goto end;
2776 }
2777 } else {
2778 subbuf->info.data.sequence_number.is_set = true;
2779 }
2780
2781 ret = ustctl_get_stream_id(
2782 stream->ustream, &subbuf->info.data.stream_id);
2783 if (ret < 0) {
2784 PERROR("Failed to get stream id");
2785 goto end;
2786 }
2787
2788 ret = ustctl_get_instance_id(stream->ustream,
2789 &subbuf->info.data.stream_instance_id.value);
2790 if (ret) {
2791 /* May not be supported by older LTTng-modules. */
2792 if (ret != -ENOTTY) {
2793 PERROR("Failed to get stream instance id");
2794 goto end;
2795 }
2796 } else {
2797 subbuf->info.data.stream_instance_id.is_set = true;
2798 }
2799 end:
2800 return ret;
2801 }
2802
2803 static int get_next_subbuffer_common(struct lttng_consumer_stream *stream,
2804 struct stream_subbuffer *subbuffer)
2805 {
2806 int ret;
2807 const char *addr;
2808
2809 ret = stream->read_subbuffer_ops.extract_subbuffer_info(
2810 stream, subbuffer);
2811 if (ret) {
2812 goto end;
2813 }
2814
2815 ret = get_current_subbuf_addr(stream, &addr);
2816 if (ret) {
2817 goto end;
2818 }
2819
2820 subbuffer->buffer.buffer = lttng_buffer_view_init(
2821 addr, 0, subbuffer->info.data.padded_subbuf_size);
2822 assert(subbuffer->buffer.buffer.data != NULL);
2823 end:
2824 return ret;
2825 }
2826
2827 static int get_next_subbuffer(struct lttng_consumer_stream *stream,
2828 struct stream_subbuffer *subbuffer)
2829 {
2830 int ret;
2831
2832 ret = ustctl_get_next_subbuf(stream->ustream);
2833 if (ret) {
2834 goto end;
2835 }
2836
2837 ret = get_next_subbuffer_common(stream, subbuffer);
2838 if (ret) {
2839 goto end;
2840 }
2841 end:
2842 return ret;
2843 }
2844
2845 static int get_next_subbuffer_metadata(struct lttng_consumer_stream *stream,
2846 struct stream_subbuffer *subbuffer)
2847 {
2848 int ret;
2849 bool cache_empty;
2850 bool got_subbuffer;
2851 bool coherent;
2852 bool buffer_empty;
2853 unsigned long consumed_pos, produced_pos;
2854
2855 do {
2856 ret = ustctl_get_next_subbuf(stream->ustream);
2857 if (ret == 0) {
2858 got_subbuffer = true;
2859 } else {
2860 got_subbuffer = false;
2861 if (ret != -EAGAIN) {
2862 /* Fatal error. */
2863 goto end;
2864 }
2865 }
2866
2867 /*
2868 * Determine if the cache is empty and ensure that a sub-buffer
2869 * is made available if the cache is not empty.
2870 */
2871 if (!got_subbuffer) {
2872 ret = commit_one_metadata_packet(stream);
2873 if (ret < 0 && ret != -ENOBUFS) {
2874 goto end;
2875 } else if (ret == 0) {
2876 /* Not an error, the cache is empty. */
2877 cache_empty = true;
2878 ret = -ENODATA;
2879 goto end;
2880 } else {
2881 cache_empty = false;
2882 }
2883 } else {
2884 pthread_mutex_lock(&stream->chan->metadata_cache->lock);
2885 cache_empty = stream->chan->metadata_cache->max_offset ==
2886 stream->ust_metadata_pushed;
2887 pthread_mutex_unlock(&stream->chan->metadata_cache->lock);
2888 }
2889 } while (!got_subbuffer);
2890
2891 /* Populate sub-buffer infos and view. */
2892 ret = get_next_subbuffer_common(stream, subbuffer);
2893 if (ret) {
2894 goto end;
2895 }
2896
2897 ret = lttng_ustconsumer_sample_snapshot_positions(stream);
2898 if (ret < 0) {
2899 /*
2900 * -EAGAIN is not expected since we got a sub-buffer and haven't
2901 * pushed the consumption position yet (on put_next).
2902 */
2903 PERROR("Failed to take a snapshot of metadata buffer positions");
2904 goto end;
2905 }
2906
2907 ret = lttng_ustconsumer_get_consumed_snapshot(stream, &consumed_pos);
2908 if (ret) {
2909 PERROR("Failed to get metadata consumed position");
2910 goto end;
2911 }
2912
2913 ret = lttng_ustconsumer_get_produced_snapshot(stream, &produced_pos);
2914 if (ret) {
2915 PERROR("Failed to get metadata produced position");
2916 goto end;
2917 }
2918
2919 /* Last sub-buffer of the ring buffer ? */
2920 buffer_empty = (consumed_pos + stream->max_sb_size) == produced_pos;
2921
2922 /*
2923 * The sessiond registry lock ensures that coherent units of metadata
2924 * are pushed to the consumer daemon at once. Hence, if a sub-buffer is
2925 * acquired, the cache is empty, and it is the only available sub-buffer
2926 * available, it is safe to assume that it is "coherent".
2927 */
2928 coherent = got_subbuffer && cache_empty && buffer_empty;
2929
2930 LTTNG_OPTIONAL_SET(&subbuffer->info.metadata.coherent, coherent);
2931 end:
2932 return ret;
2933 }
2934
2935 static int put_next_subbuffer(struct lttng_consumer_stream *stream,
2936 struct stream_subbuffer *subbuffer)
2937 {
2938 const int ret = ustctl_put_next_subbuf(stream->ustream);
2939
2940 assert(ret == 0);
2941 return ret;
2942 }
2943
2944 static int signal_metadata(struct lttng_consumer_stream *stream,
2945 struct lttng_consumer_local_data *ctx)
2946 {
2947 return pthread_cond_broadcast(&stream->metadata_rdv) ? -errno : 0;
2948 }
2949
2950 static int lttng_ustconsumer_set_stream_ops(
2951 struct lttng_consumer_stream *stream)
2952 {
2953 int ret = 0;
2954
2955 stream->read_subbuffer_ops.on_wake_up = consumer_stream_ust_on_wake_up;
2956 if (stream->metadata_flag) {
2957 stream->read_subbuffer_ops.get_next_subbuffer =
2958 get_next_subbuffer_metadata;
2959 stream->read_subbuffer_ops.extract_subbuffer_info =
2960 extract_metadata_subbuffer_info;
2961 stream->read_subbuffer_ops.reset_metadata =
2962 metadata_stream_reset_cache_consumed_position;
2963 if (stream->chan->is_live) {
2964 stream->read_subbuffer_ops.on_sleep = signal_metadata;
2965 ret = consumer_stream_enable_metadata_bucketization(
2966 stream);
2967 if (ret) {
2968 goto end;
2969 }
2970 }
2971 } else {
2972 stream->read_subbuffer_ops.get_next_subbuffer =
2973 get_next_subbuffer;
2974 stream->read_subbuffer_ops.extract_subbuffer_info =
2975 extract_data_subbuffer_info;
2976 stream->read_subbuffer_ops.on_sleep = notify_if_more_data;
2977 if (stream->chan->is_live) {
2978 stream->read_subbuffer_ops.send_live_beacon =
2979 consumer_flush_ust_index;
2980 }
2981 }
2982
2983 stream->read_subbuffer_ops.put_next_subbuffer = put_next_subbuffer;
2984 end:
2985 return ret;
2986 }
2987
2988 /*
2989 * Called when a stream is created.
2990 *
2991 * Return 0 on success or else a negative value.
2992 */
2993 int lttng_ustconsumer_on_recv_stream(struct lttng_consumer_stream *stream)
2994 {
2995 int ret;
2996
2997 assert(stream);
2998
2999 /*
3000 * Don't create anything if this is set for streaming or if there is
3001 * no current trace chunk on the parent channel.
3002 */
3003 if (stream->net_seq_idx == (uint64_t) -1ULL && stream->chan->monitor &&
3004 stream->chan->trace_chunk) {
3005 ret = consumer_stream_create_output_files(stream, true);
3006 if (ret) {
3007 goto error;
3008 }
3009 }
3010
3011 lttng_ustconsumer_set_stream_ops(stream);
3012 ret = 0;
3013
3014 error:
3015 return ret;
3016 }
3017
3018 /*
3019 * Check if data is still being extracted from the buffers for a specific
3020 * stream. Consumer data lock MUST be acquired before calling this function
3021 * and the stream lock.
3022 *
3023 * Return 1 if the traced data are still getting read else 0 meaning that the
3024 * data is available for trace viewer reading.
3025 */
3026 int lttng_ustconsumer_data_pending(struct lttng_consumer_stream *stream)
3027 {
3028 int ret;
3029
3030 assert(stream);
3031 assert(stream->ustream);
3032
3033 DBG("UST consumer checking data pending");
3034
3035 if (stream->endpoint_status != CONSUMER_ENDPOINT_ACTIVE) {
3036 ret = 0;
3037 goto end;
3038 }
3039
3040 if (stream->chan->type == CONSUMER_CHANNEL_TYPE_METADATA) {
3041 uint64_t contiguous, pushed;
3042
3043 /* Ease our life a bit. */
3044 contiguous = stream->chan->metadata_cache->max_offset;
3045 pushed = stream->ust_metadata_pushed;
3046
3047 /*
3048 * We can simply check whether all contiguously available data
3049 * has been pushed to the ring buffer, since the push operation
3050 * is performed within get_next_subbuf(), and because both
3051 * get_next_subbuf() and put_next_subbuf() are issued atomically
3052 * thanks to the stream lock within
3053 * lttng_ustconsumer_read_subbuffer(). This basically means that
3054 * whetnever ust_metadata_pushed is incremented, the associated
3055 * metadata has been consumed from the metadata stream.
3056 */
3057 DBG("UST consumer metadata pending check: contiguous %" PRIu64 " vs pushed %" PRIu64,
3058 contiguous, pushed);
3059 assert(((int64_t) (contiguous - pushed)) >= 0);
3060 if ((contiguous != pushed) ||
3061 (((int64_t) contiguous - pushed) > 0 || contiguous == 0)) {
3062 ret = 1; /* Data is pending */
3063 goto end;
3064 }
3065 } else {
3066 ret = ustctl_get_next_subbuf(stream->ustream);
3067 if (ret == 0) {
3068 /*
3069 * There is still data so let's put back this
3070 * subbuffer.
3071 */
3072 ret = ustctl_put_subbuf(stream->ustream);
3073 assert(ret == 0);
3074 ret = 1; /* Data is pending */
3075 goto end;
3076 }
3077 }
3078
3079 /* Data is NOT pending so ready to be read. */
3080 ret = 0;
3081
3082 end:
3083 return ret;
3084 }
3085
3086 /*
3087 * Stop a given metadata channel timer if enabled and close the wait fd which
3088 * is the poll pipe of the metadata stream.
3089 *
3090 * This MUST be called with the metadata channel lock acquired.
3091 */
3092 void lttng_ustconsumer_close_metadata(struct lttng_consumer_channel *metadata)
3093 {
3094 int ret;
3095
3096 assert(metadata);
3097 assert(metadata->type == CONSUMER_CHANNEL_TYPE_METADATA);
3098
3099 DBG("Closing metadata channel key %" PRIu64, metadata->key);
3100
3101 if (metadata->switch_timer_enabled == 1) {
3102 consumer_timer_switch_stop(metadata);
3103 }
3104
3105 if (!metadata->metadata_stream) {
3106 goto end;
3107 }
3108
3109 /*
3110 * Closing write side so the thread monitoring the stream wakes up if any
3111 * and clean the metadata stream.
3112 */
3113 if (metadata->metadata_stream->ust_metadata_poll_pipe[1] >= 0) {
3114 ret = close(metadata->metadata_stream->ust_metadata_poll_pipe[1]);
3115 if (ret < 0) {
3116 PERROR("closing metadata pipe write side");
3117 }
3118 metadata->metadata_stream->ust_metadata_poll_pipe[1] = -1;
3119 }
3120
3121 end:
3122 return;
3123 }
3124
3125 /*
3126 * Close every metadata stream wait fd of the metadata hash table. This
3127 * function MUST be used very carefully so not to run into a race between the
3128 * metadata thread handling streams and this function closing their wait fd.
3129 *
3130 * For UST, this is used when the session daemon hangs up. Its the metadata
3131 * producer so calling this is safe because we are assured that no state change
3132 * can occur in the metadata thread for the streams in the hash table.
3133 */
3134 void lttng_ustconsumer_close_all_metadata(struct lttng_ht *metadata_ht)
3135 {
3136 struct lttng_ht_iter iter;
3137 struct lttng_consumer_stream *stream;
3138
3139 assert(metadata_ht);
3140 assert(metadata_ht->ht);
3141
3142 DBG("UST consumer closing all metadata streams");
3143
3144 rcu_read_lock();
3145 cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream,
3146 node.node) {
3147
3148 health_code_update();
3149
3150 pthread_mutex_lock(&stream->chan->lock);
3151 lttng_ustconsumer_close_metadata(stream->chan);
3152 pthread_mutex_unlock(&stream->chan->lock);
3153
3154 }
3155 rcu_read_unlock();
3156 }
3157
3158 void lttng_ustconsumer_close_stream_wakeup(struct lttng_consumer_stream *stream)
3159 {
3160 int ret;
3161
3162 ret = ustctl_stream_close_wakeup_fd(stream->ustream);
3163 if (ret < 0) {
3164 ERR("Unable to close wakeup fd");
3165 }
3166 }
3167
3168 /*
3169 * Please refer to consumer-timer.c before adding any lock within this
3170 * function or any of its callees. Timers have a very strict locking
3171 * semantic with respect to teardown. Failure to respect this semantic
3172 * introduces deadlocks.
3173 *
3174 * DON'T hold the metadata lock when calling this function, else this
3175 * can cause deadlock involving consumer awaiting for metadata to be
3176 * pushed out due to concurrent interaction with the session daemon.
3177 */
3178 int lttng_ustconsumer_request_metadata(struct lttng_consumer_local_data *ctx,
3179 struct lttng_consumer_channel *channel, int timer, int wait)
3180 {
3181 struct lttcomm_metadata_request_msg request;
3182 struct lttcomm_consumer_msg msg;
3183 enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS;
3184 uint64_t len, key, offset, version;
3185 int ret;
3186
3187 assert(channel);
3188 assert(channel->metadata_cache);
3189
3190 memset(&request, 0, sizeof(request));
3191
3192 /* send the metadata request to sessiond */
3193 switch (consumer_data.type) {
3194 case LTTNG_CONSUMER64_UST:
3195 request.bits_per_long = 64;
3196 break;
3197 case LTTNG_CONSUMER32_UST:
3198 request.bits_per_long = 32;
3199 break;
3200 default:
3201 request.bits_per_long = 0;
3202 break;
3203 }
3204
3205 request.session_id = channel->session_id;
3206 request.session_id_per_pid = channel->session_id_per_pid;
3207 /*
3208 * Request the application UID here so the metadata of that application can
3209 * be sent back. The channel UID corresponds to the user UID of the session
3210 * used for the rights on the stream file(s).
3211 */
3212 request.uid = channel->ust_app_uid;
3213 request.key = channel->key;
3214
3215 DBG("Sending metadata request to sessiond, session id %" PRIu64
3216 ", per-pid %" PRIu64 ", app UID %u and channel key %" PRIu64,
3217 request.session_id, request.session_id_per_pid, request.uid,
3218 request.key);
3219
3220 pthread_mutex_lock(&ctx->metadata_socket_lock);
3221
3222 health_code_update();
3223
3224 ret = lttcomm_send_unix_sock(ctx->consumer_metadata_socket, &request,
3225 sizeof(request));
3226 if (ret < 0) {
3227 ERR("Asking metadata to sessiond");
3228 goto end;
3229 }
3230
3231 health_code_update();
3232
3233 /* Receive the metadata from sessiond */
3234 ret = lttcomm_recv_unix_sock(ctx->consumer_metadata_socket, &msg,
3235 sizeof(msg));
3236 if (ret != sizeof(msg)) {
3237 DBG("Consumer received unexpected message size %d (expects %zu)",
3238 ret, sizeof(msg));
3239 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD);
3240 /*
3241 * The ret value might 0 meaning an orderly shutdown but this is ok
3242 * since the caller handles this.
3243 */
3244 goto end;
3245 }
3246
3247 health_code_update();
3248
3249 if (msg.cmd_type == LTTNG_ERR_UND) {
3250 /* No registry found */
3251 (void) consumer_send_status_msg(ctx->consumer_metadata_socket,
3252 ret_code);
3253 ret = 0;
3254 goto end;
3255 } else if (msg.cmd_type != LTTNG_CONSUMER_PUSH_METADATA) {
3256 ERR("Unexpected cmd_type received %d", msg.cmd_type);
3257 ret = -1;
3258 goto end;
3259 }
3260
3261 len = msg.u.push_metadata.len;
3262 key = msg.u.push_metadata.key;
3263 offset = msg.u.push_metadata.target_offset;
3264 version = msg.u.push_metadata.version;
3265
3266 assert(key == channel->key);
3267 if (len == 0) {
3268 DBG("No new metadata to receive for key %" PRIu64, key);
3269 }
3270
3271 health_code_update();
3272
3273 /* Tell session daemon we are ready to receive the metadata. */
3274 ret = consumer_send_status_msg(ctx->consumer_metadata_socket,
3275 LTTCOMM_CONSUMERD_SUCCESS);
3276 if (ret < 0 || len == 0) {
3277 /*
3278 * Somehow, the session daemon is not responding anymore or there is
3279 * nothing to receive.
3280 */
3281 goto end;
3282 }
3283
3284 health_code_update();
3285
3286 ret = lttng_ustconsumer_recv_metadata(ctx->consumer_metadata_socket,
3287 key, offset, len, version, channel, timer, wait);
3288 if (ret >= 0) {
3289 /*
3290 * Only send the status msg if the sessiond is alive meaning a positive
3291 * ret code.
3292 */
3293 (void) consumer_send_status_msg(ctx->consumer_metadata_socket, ret);
3294 }
3295 ret = 0;
3296
3297 end:
3298 health_code_update();
3299
3300 pthread_mutex_unlock(&ctx->metadata_socket_lock);
3301 return ret;
3302 }
3303
3304 /*
3305 * Return the ustctl call for the get stream id.
3306 */
3307 int lttng_ustconsumer_get_stream_id(struct lttng_consumer_stream *stream,
3308 uint64_t *stream_id)
3309 {
3310 assert(stream);
3311 assert(stream_id);
3312
3313 return ustctl_get_stream_id(stream->ustream, stream_id);
3314 }
This page took 0.182308 seconds and 5 git commands to generate.