83080edde314285a434064e54b5db1699010570c
[lttng-tools.git] / src / common / ust-consumer / ust-consumer.c
1 /*
2 * Copyright (C) 2011 EfficiOS Inc.
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_init(&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_init(&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 metadata->metadata_stream = NULL;
979 send_streams_error:
980 error_no_stream:
981 end:
982 return ret;
983 }
984
985 /*
986 * Snapshot the whole metadata.
987 * RCU read-side lock must be held by the caller.
988 *
989 * Returns 0 on success, < 0 on error
990 */
991 static int snapshot_metadata(struct lttng_consumer_channel *metadata_channel,
992 uint64_t key, char *path, uint64_t relayd_id,
993 struct lttng_consumer_local_data *ctx)
994 {
995 int ret = 0;
996 struct lttng_consumer_stream *metadata_stream;
997
998 assert(path);
999 assert(ctx);
1000
1001 DBG("UST consumer snapshot metadata with key %" PRIu64 " at path %s",
1002 key, path);
1003
1004 rcu_read_lock();
1005
1006 assert(!metadata_channel->monitor);
1007
1008 health_code_update();
1009
1010 /*
1011 * Ask the sessiond if we have new metadata waiting and update the
1012 * consumer metadata cache.
1013 */
1014 ret = lttng_ustconsumer_request_metadata(ctx, metadata_channel, 0, 1);
1015 if (ret < 0) {
1016 goto error;
1017 }
1018
1019 health_code_update();
1020
1021 /*
1022 * The metadata stream is NOT created in no monitor mode when the channel
1023 * is created on a sessiond ask channel command.
1024 */
1025 ret = create_ust_streams(metadata_channel, ctx);
1026 if (ret < 0) {
1027 goto error;
1028 }
1029
1030 metadata_stream = metadata_channel->metadata_stream;
1031 assert(metadata_stream);
1032
1033 metadata_stream->read_subbuffer_ops.lock(metadata_stream);
1034 if (relayd_id != (uint64_t) -1ULL) {
1035 metadata_stream->net_seq_idx = relayd_id;
1036 ret = consumer_send_relayd_stream(metadata_stream, path);
1037 } else {
1038 ret = consumer_stream_create_output_files(metadata_stream,
1039 false);
1040 }
1041 if (ret < 0) {
1042 goto error_stream;
1043 }
1044
1045 do {
1046 health_code_update();
1047 ret = lttng_consumer_read_subbuffer(metadata_stream, ctx, true);
1048 if (ret < 0) {
1049 goto error_stream;
1050 }
1051 } while (ret > 0);
1052
1053 error_stream:
1054 metadata_stream->read_subbuffer_ops.unlock(metadata_stream);
1055 /*
1056 * Clean up the stream completely because the next snapshot will use a
1057 * new metadata stream.
1058 */
1059 consumer_stream_destroy(metadata_stream, NULL);
1060 metadata_channel->metadata_stream = NULL;
1061
1062 error:
1063 rcu_read_unlock();
1064 return ret;
1065 }
1066
1067 static
1068 int get_current_subbuf_addr(struct lttng_consumer_stream *stream,
1069 const char **addr)
1070 {
1071 int ret;
1072 unsigned long mmap_offset;
1073 const char *mmap_base;
1074
1075 mmap_base = ustctl_get_mmap_base(stream->ustream);
1076 if (!mmap_base) {
1077 ERR("Failed to get mmap base for stream `%s`",
1078 stream->name);
1079 ret = -EPERM;
1080 goto error;
1081 }
1082
1083 ret = ustctl_get_mmap_read_offset(stream->ustream, &mmap_offset);
1084 if (ret != 0) {
1085 ERR("Failed to get mmap offset for stream `%s`", stream->name);
1086 ret = -EINVAL;
1087 goto error;
1088 }
1089
1090 *addr = mmap_base + mmap_offset;
1091 error:
1092 return ret;
1093
1094 }
1095
1096 /*
1097 * Take a snapshot of all the stream of a channel.
1098 * RCU read-side lock and the channel lock must be held by the caller.
1099 *
1100 * Returns 0 on success, < 0 on error
1101 */
1102 static int snapshot_channel(struct lttng_consumer_channel *channel,
1103 uint64_t key, char *path, uint64_t relayd_id,
1104 uint64_t nb_packets_per_stream,
1105 struct lttng_consumer_local_data *ctx)
1106 {
1107 int ret;
1108 unsigned use_relayd = 0;
1109 unsigned long consumed_pos, produced_pos;
1110 struct lttng_consumer_stream *stream;
1111
1112 assert(path);
1113 assert(ctx);
1114
1115 rcu_read_lock();
1116
1117 if (relayd_id != (uint64_t) -1ULL) {
1118 use_relayd = 1;
1119 }
1120
1121 assert(!channel->monitor);
1122 DBG("UST consumer snapshot channel %" PRIu64, key);
1123
1124 cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
1125 health_code_update();
1126
1127 /* Lock stream because we are about to change its state. */
1128 pthread_mutex_lock(&stream->lock);
1129 assert(channel->trace_chunk);
1130 if (!lttng_trace_chunk_get(channel->trace_chunk)) {
1131 /*
1132 * Can't happen barring an internal error as the channel
1133 * holds a reference to the trace chunk.
1134 */
1135 ERR("Failed to acquire reference to channel's trace chunk");
1136 ret = -1;
1137 goto error_unlock;
1138 }
1139 assert(!stream->trace_chunk);
1140 stream->trace_chunk = channel->trace_chunk;
1141
1142 stream->net_seq_idx = relayd_id;
1143
1144 if (use_relayd) {
1145 ret = consumer_send_relayd_stream(stream, path);
1146 if (ret < 0) {
1147 goto error_unlock;
1148 }
1149 } else {
1150 ret = consumer_stream_create_output_files(stream,
1151 false);
1152 if (ret < 0) {
1153 goto error_unlock;
1154 }
1155 DBG("UST consumer snapshot stream (%" PRIu64 ")",
1156 stream->key);
1157 }
1158
1159 /*
1160 * If tracing is active, we want to perform a "full" buffer flush.
1161 * Else, if quiescent, it has already been done by the prior stop.
1162 */
1163 if (!stream->quiescent) {
1164 ustctl_flush_buffer(stream->ustream, 0);
1165 }
1166
1167 ret = lttng_ustconsumer_take_snapshot(stream);
1168 if (ret < 0) {
1169 ERR("Taking UST snapshot");
1170 goto error_unlock;
1171 }
1172
1173 ret = lttng_ustconsumer_get_produced_snapshot(stream, &produced_pos);
1174 if (ret < 0) {
1175 ERR("Produced UST snapshot position");
1176 goto error_unlock;
1177 }
1178
1179 ret = lttng_ustconsumer_get_consumed_snapshot(stream, &consumed_pos);
1180 if (ret < 0) {
1181 ERR("Consumerd UST snapshot position");
1182 goto error_unlock;
1183 }
1184
1185 /*
1186 * The original value is sent back if max stream size is larger than
1187 * the possible size of the snapshot. Also, we assume that the session
1188 * daemon should never send a maximum stream size that is lower than
1189 * subbuffer size.
1190 */
1191 consumed_pos = consumer_get_consume_start_pos(consumed_pos,
1192 produced_pos, nb_packets_per_stream,
1193 stream->max_sb_size);
1194
1195 while ((long) (consumed_pos - produced_pos) < 0) {
1196 ssize_t read_len;
1197 unsigned long len, padded_len;
1198 const char *subbuf_addr;
1199 struct lttng_buffer_view subbuf_view;
1200
1201 health_code_update();
1202
1203 DBG("UST consumer taking snapshot at pos %lu", consumed_pos);
1204
1205 ret = ustctl_get_subbuf(stream->ustream, &consumed_pos);
1206 if (ret < 0) {
1207 if (ret != -EAGAIN) {
1208 PERROR("ustctl_get_subbuf snapshot");
1209 goto error_close_stream;
1210 }
1211 DBG("UST consumer get subbuf failed. Skipping it.");
1212 consumed_pos += stream->max_sb_size;
1213 stream->chan->lost_packets++;
1214 continue;
1215 }
1216
1217 ret = ustctl_get_subbuf_size(stream->ustream, &len);
1218 if (ret < 0) {
1219 ERR("Snapshot ustctl_get_subbuf_size");
1220 goto error_put_subbuf;
1221 }
1222
1223 ret = ustctl_get_padded_subbuf_size(stream->ustream, &padded_len);
1224 if (ret < 0) {
1225 ERR("Snapshot ustctl_get_padded_subbuf_size");
1226 goto error_put_subbuf;
1227 }
1228
1229 ret = get_current_subbuf_addr(stream, &subbuf_addr);
1230 if (ret) {
1231 goto error_put_subbuf;
1232 }
1233
1234 subbuf_view = lttng_buffer_view_init(
1235 subbuf_addr, 0, padded_len);
1236 read_len = lttng_consumer_on_read_subbuffer_mmap(
1237 stream, &subbuf_view, padded_len - len);
1238 if (use_relayd) {
1239 if (read_len != len) {
1240 ret = -EPERM;
1241 goto error_put_subbuf;
1242 }
1243 } else {
1244 if (read_len != padded_len) {
1245 ret = -EPERM;
1246 goto error_put_subbuf;
1247 }
1248 }
1249
1250 ret = ustctl_put_subbuf(stream->ustream);
1251 if (ret < 0) {
1252 ERR("Snapshot ustctl_put_subbuf");
1253 goto error_close_stream;
1254 }
1255 consumed_pos += stream->max_sb_size;
1256 }
1257
1258 /* Simply close the stream so we can use it on the next snapshot. */
1259 consumer_stream_close(stream);
1260 pthread_mutex_unlock(&stream->lock);
1261 }
1262
1263 rcu_read_unlock();
1264 return 0;
1265
1266 error_put_subbuf:
1267 if (ustctl_put_subbuf(stream->ustream) < 0) {
1268 ERR("Snapshot ustctl_put_subbuf");
1269 }
1270 error_close_stream:
1271 consumer_stream_close(stream);
1272 error_unlock:
1273 pthread_mutex_unlock(&stream->lock);
1274 rcu_read_unlock();
1275 return ret;
1276 }
1277
1278 static
1279 void metadata_stream_reset_cache_consumed_position(
1280 struct lttng_consumer_stream *stream)
1281 {
1282 ASSERT_LOCKED(stream->lock);
1283
1284 DBG("Reset metadata cache of session %" PRIu64,
1285 stream->chan->session_id);
1286 stream->ust_metadata_pushed = 0;
1287 }
1288
1289 /*
1290 * Receive the metadata updates from the sessiond. Supports receiving
1291 * overlapping metadata, but is needs to always belong to a contiguous
1292 * range starting from 0.
1293 * Be careful about the locks held when calling this function: it needs
1294 * the metadata cache flush to concurrently progress in order to
1295 * complete.
1296 */
1297 int lttng_ustconsumer_recv_metadata(int sock, uint64_t key, uint64_t offset,
1298 uint64_t len, uint64_t version,
1299 struct lttng_consumer_channel *channel, int timer, int wait)
1300 {
1301 int ret, ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1302 char *metadata_str;
1303 enum consumer_metadata_cache_write_status cache_write_status;
1304
1305 DBG("UST consumer push metadata key %" PRIu64 " of len %" PRIu64, key, len);
1306
1307 metadata_str = zmalloc(len * sizeof(char));
1308 if (!metadata_str) {
1309 PERROR("zmalloc metadata string");
1310 ret_code = LTTCOMM_CONSUMERD_ENOMEM;
1311 goto end;
1312 }
1313
1314 health_code_update();
1315
1316 /* Receive metadata string. */
1317 ret = lttcomm_recv_unix_sock(sock, metadata_str, len);
1318 if (ret < 0) {
1319 /* Session daemon is dead so return gracefully. */
1320 ret_code = ret;
1321 goto end_free;
1322 }
1323
1324 health_code_update();
1325
1326 pthread_mutex_lock(&channel->metadata_cache->lock);
1327 cache_write_status = consumer_metadata_cache_write(
1328 channel, offset, len, version, metadata_str);
1329 pthread_mutex_unlock(&channel->metadata_cache->lock);
1330 switch (cache_write_status) {
1331 case CONSUMER_METADATA_CACHE_WRITE_STATUS_NO_CHANGE:
1332 /*
1333 * The write entirely overlapped with existing contents of the
1334 * same metadata version (same content); there is nothing to do.
1335 */
1336 break;
1337 case CONSUMER_METADATA_CACHE_WRITE_STATUS_INVALIDATED:
1338 /*
1339 * The metadata cache was invalidated (previously pushed
1340 * content has been overwritten). Reset the stream's consumed
1341 * metadata position to ensure the metadata poll thread consumes
1342 * the whole cache.
1343 */
1344
1345 /*
1346 * channel::metadata_stream can be null when the metadata
1347 * channel is under a snapshot session type. No need to update
1348 * the stream position in that scenario.
1349 */
1350 if (channel->metadata_stream != NULL) {
1351 pthread_mutex_lock(&channel->metadata_stream->lock);
1352 metadata_stream_reset_cache_consumed_position(
1353 channel->metadata_stream);
1354 pthread_mutex_unlock(&channel->metadata_stream->lock);
1355 }
1356 /* Fall-through. */
1357 case CONSUMER_METADATA_CACHE_WRITE_STATUS_APPENDED_CONTENT:
1358 /*
1359 * In both cases, the metadata poll thread has new data to
1360 * consume.
1361 */
1362 ret = consumer_metadata_wakeup_pipe(channel);
1363 if (ret) {
1364 ret_code = LTTCOMM_CONSUMERD_ERROR_METADATA;
1365 goto end_free;
1366 }
1367 break;
1368 case CONSUMER_METADATA_CACHE_WRITE_STATUS_ERROR:
1369 /* Unable to handle metadata. Notify session daemon. */
1370 ret_code = LTTCOMM_CONSUMERD_ERROR_METADATA;
1371 /*
1372 * Skip metadata flush on write error since the offset and len might
1373 * not have been updated which could create an infinite loop below when
1374 * waiting for the metadata cache to be flushed.
1375 */
1376 goto end_free;
1377 default:
1378 abort();
1379 }
1380
1381 if (!wait) {
1382 goto end_free;
1383 }
1384 while (consumer_metadata_cache_flushed(channel, offset + len, timer)) {
1385 DBG("Waiting for metadata to be flushed");
1386
1387 health_code_update();
1388
1389 usleep(DEFAULT_METADATA_AVAILABILITY_WAIT_TIME);
1390 }
1391
1392 end_free:
1393 free(metadata_str);
1394 end:
1395 return ret_code;
1396 }
1397
1398 /*
1399 * Receive command from session daemon and process it.
1400 *
1401 * Return 1 on success else a negative value or 0.
1402 */
1403 int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
1404 int sock, struct pollfd *consumer_sockpoll)
1405 {
1406 ssize_t ret;
1407 enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1408 struct lttcomm_consumer_msg msg;
1409 struct lttng_consumer_channel *channel = NULL;
1410
1411 health_code_update();
1412
1413 ret = lttcomm_recv_unix_sock(sock, &msg, sizeof(msg));
1414 if (ret != sizeof(msg)) {
1415 DBG("Consumer received unexpected message size %zd (expects %zu)",
1416 ret, sizeof(msg));
1417 /*
1418 * The ret value might 0 meaning an orderly shutdown but this is ok
1419 * since the caller handles this.
1420 */
1421 if (ret > 0) {
1422 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD);
1423 ret = -1;
1424 }
1425 return ret;
1426 }
1427
1428 health_code_update();
1429
1430 /* deprecated */
1431 assert(msg.cmd_type != LTTNG_CONSUMER_STOP);
1432
1433 health_code_update();
1434
1435 /* relayd needs RCU read-side lock */
1436 rcu_read_lock();
1437
1438 switch (msg.cmd_type) {
1439 case LTTNG_CONSUMER_ADD_RELAYD_SOCKET:
1440 {
1441 uint32_t major = msg.u.relayd_sock.major;
1442 uint32_t minor = msg.u.relayd_sock.minor;
1443 enum lttcomm_sock_proto protocol =
1444 (enum lttcomm_sock_proto) msg.u.relayd_sock
1445 .relayd_socket_protocol;
1446
1447 /* Session daemon status message are handled in the following call. */
1448 consumer_add_relayd_socket(msg.u.relayd_sock.net_index,
1449 msg.u.relayd_sock.type, ctx, sock,
1450 consumer_sockpoll, msg.u.relayd_sock.session_id,
1451 msg.u.relayd_sock.relayd_session_id, major,
1452 minor, protocol);
1453 goto end_nosignal;
1454 }
1455 case LTTNG_CONSUMER_DESTROY_RELAYD:
1456 {
1457 uint64_t index = msg.u.destroy_relayd.net_seq_idx;
1458 struct consumer_relayd_sock_pair *relayd;
1459
1460 DBG("UST consumer destroying relayd %" PRIu64, index);
1461
1462 /* Get relayd reference if exists. */
1463 relayd = consumer_find_relayd(index);
1464 if (relayd == NULL) {
1465 DBG("Unable to find relayd %" PRIu64, index);
1466 ret_code = LTTCOMM_CONSUMERD_RELAYD_FAIL;
1467 }
1468
1469 /*
1470 * Each relayd socket pair has a refcount of stream attached to it
1471 * which tells if the relayd is still active or not depending on the
1472 * refcount value.
1473 *
1474 * This will set the destroy flag of the relayd object and destroy it
1475 * if the refcount reaches zero when called.
1476 *
1477 * The destroy can happen either here or when a stream fd hangs up.
1478 */
1479 if (relayd) {
1480 consumer_flag_relayd_for_destroy(relayd);
1481 }
1482
1483 goto end_msg_sessiond;
1484 }
1485 case LTTNG_CONSUMER_UPDATE_STREAM:
1486 {
1487 rcu_read_unlock();
1488 return -ENOSYS;
1489 }
1490 case LTTNG_CONSUMER_DATA_PENDING:
1491 {
1492 int ret, is_data_pending;
1493 uint64_t id = msg.u.data_pending.session_id;
1494
1495 DBG("UST consumer data pending command for id %" PRIu64, id);
1496
1497 is_data_pending = consumer_data_pending(id);
1498
1499 /* Send back returned value to session daemon */
1500 ret = lttcomm_send_unix_sock(sock, &is_data_pending,
1501 sizeof(is_data_pending));
1502 if (ret < 0) {
1503 DBG("Error when sending the data pending ret code: %d", ret);
1504 goto error_fatal;
1505 }
1506
1507 /*
1508 * No need to send back a status message since the data pending
1509 * returned value is the response.
1510 */
1511 break;
1512 }
1513 case LTTNG_CONSUMER_ASK_CHANNEL_CREATION:
1514 {
1515 int ret;
1516 struct ustctl_consumer_channel_attr attr;
1517 const uint64_t chunk_id = msg.u.ask_channel.chunk_id.value;
1518 const struct lttng_credentials buffer_credentials = {
1519 .uid = msg.u.ask_channel.buffer_credentials.uid,
1520 .gid = msg.u.ask_channel.buffer_credentials.gid,
1521 };
1522
1523 /* Create a plain object and reserve a channel key. */
1524 channel = consumer_allocate_channel(
1525 msg.u.ask_channel.key,
1526 msg.u.ask_channel.session_id,
1527 msg.u.ask_channel.chunk_id.is_set ?
1528 &chunk_id : NULL,
1529 msg.u.ask_channel.pathname,
1530 msg.u.ask_channel.name,
1531 msg.u.ask_channel.relayd_id,
1532 (enum lttng_event_output) msg.u.ask_channel.output,
1533 msg.u.ask_channel.tracefile_size,
1534 msg.u.ask_channel.tracefile_count,
1535 msg.u.ask_channel.session_id_per_pid,
1536 msg.u.ask_channel.monitor,
1537 msg.u.ask_channel.live_timer_interval,
1538 msg.u.ask_channel.is_live,
1539 msg.u.ask_channel.root_shm_path,
1540 msg.u.ask_channel.shm_path);
1541 if (!channel) {
1542 goto end_channel_error;
1543 }
1544
1545 LTTNG_OPTIONAL_SET(&channel->buffer_credentials,
1546 buffer_credentials);
1547
1548 /*
1549 * Assign UST application UID to the channel. This value is ignored for
1550 * per PID buffers. This is specific to UST thus setting this after the
1551 * allocation.
1552 */
1553 channel->ust_app_uid = msg.u.ask_channel.ust_app_uid;
1554
1555 /* Build channel attributes from received message. */
1556 attr.subbuf_size = msg.u.ask_channel.subbuf_size;
1557 attr.num_subbuf = msg.u.ask_channel.num_subbuf;
1558 attr.overwrite = msg.u.ask_channel.overwrite;
1559 attr.switch_timer_interval = msg.u.ask_channel.switch_timer_interval;
1560 attr.read_timer_interval = msg.u.ask_channel.read_timer_interval;
1561 attr.chan_id = msg.u.ask_channel.chan_id;
1562 memcpy(attr.uuid, msg.u.ask_channel.uuid, sizeof(attr.uuid));
1563 attr.blocking_timeout= msg.u.ask_channel.blocking_timeout;
1564
1565 /* Match channel buffer type to the UST abi. */
1566 switch (msg.u.ask_channel.output) {
1567 case LTTNG_EVENT_MMAP:
1568 default:
1569 attr.output = LTTNG_UST_MMAP;
1570 break;
1571 }
1572
1573 /* Translate and save channel type. */
1574 switch (msg.u.ask_channel.type) {
1575 case LTTNG_UST_CHAN_PER_CPU:
1576 channel->type = CONSUMER_CHANNEL_TYPE_DATA;
1577 attr.type = LTTNG_UST_CHAN_PER_CPU;
1578 /*
1579 * Set refcount to 1 for owner. Below, we will
1580 * pass ownership to the
1581 * consumer_thread_channel_poll() thread.
1582 */
1583 channel->refcount = 1;
1584 break;
1585 case LTTNG_UST_CHAN_METADATA:
1586 channel->type = CONSUMER_CHANNEL_TYPE_METADATA;
1587 attr.type = LTTNG_UST_CHAN_METADATA;
1588 break;
1589 default:
1590 assert(0);
1591 goto error_fatal;
1592 };
1593
1594 health_code_update();
1595
1596 ret = ask_channel(ctx, channel, &attr);
1597 if (ret < 0) {
1598 goto end_channel_error;
1599 }
1600
1601 if (msg.u.ask_channel.type == LTTNG_UST_CHAN_METADATA) {
1602 ret = consumer_metadata_cache_allocate(channel);
1603 if (ret < 0) {
1604 ERR("Allocating metadata cache");
1605 goto end_channel_error;
1606 }
1607 consumer_timer_switch_start(channel, attr.switch_timer_interval);
1608 attr.switch_timer_interval = 0;
1609 } else {
1610 int monitor_start_ret;
1611
1612 consumer_timer_live_start(channel,
1613 msg.u.ask_channel.live_timer_interval);
1614 monitor_start_ret = consumer_timer_monitor_start(
1615 channel,
1616 msg.u.ask_channel.monitor_timer_interval);
1617 if (monitor_start_ret < 0) {
1618 ERR("Starting channel monitoring timer failed");
1619 goto end_channel_error;
1620 }
1621 }
1622
1623 health_code_update();
1624
1625 /*
1626 * Add the channel to the internal state AFTER all streams were created
1627 * and successfully sent to session daemon. This way, all streams must
1628 * be ready before this channel is visible to the threads.
1629 * If add_channel succeeds, ownership of the channel is
1630 * passed to consumer_thread_channel_poll().
1631 */
1632 ret = add_channel(channel, ctx);
1633 if (ret < 0) {
1634 if (msg.u.ask_channel.type == LTTNG_UST_CHAN_METADATA) {
1635 if (channel->switch_timer_enabled == 1) {
1636 consumer_timer_switch_stop(channel);
1637 }
1638 consumer_metadata_cache_destroy(channel);
1639 }
1640 if (channel->live_timer_enabled == 1) {
1641 consumer_timer_live_stop(channel);
1642 }
1643 if (channel->monitor_timer_enabled == 1) {
1644 consumer_timer_monitor_stop(channel);
1645 }
1646 goto end_channel_error;
1647 }
1648
1649 health_code_update();
1650
1651 /*
1652 * Channel and streams are now created. Inform the session daemon that
1653 * everything went well and should wait to receive the channel and
1654 * streams with ustctl API.
1655 */
1656 ret = consumer_send_status_channel(sock, channel);
1657 if (ret < 0) {
1658 /*
1659 * There is probably a problem on the socket.
1660 */
1661 goto error_fatal;
1662 }
1663
1664 break;
1665 }
1666 case LTTNG_CONSUMER_GET_CHANNEL:
1667 {
1668 int ret, relayd_err = 0;
1669 uint64_t key = msg.u.get_channel.key;
1670 struct lttng_consumer_channel *channel;
1671
1672 channel = consumer_find_channel(key);
1673 if (!channel) {
1674 ERR("UST consumer get channel key %" PRIu64 " not found", key);
1675 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
1676 goto end_get_channel;
1677 }
1678
1679 health_code_update();
1680
1681 /* Send the channel to sessiond (and relayd, if applicable). */
1682 ret = send_channel_to_sessiond_and_relayd(sock, channel, ctx,
1683 &relayd_err);
1684 if (ret < 0) {
1685 if (relayd_err) {
1686 /*
1687 * We were unable to send to the relayd the stream so avoid
1688 * sending back a fatal error to the thread since this is OK
1689 * and the consumer can continue its work. The above call
1690 * has sent the error status message to the sessiond.
1691 */
1692 goto end_get_channel_nosignal;
1693 }
1694 /*
1695 * The communicaton was broken hence there is a bad state between
1696 * the consumer and sessiond so stop everything.
1697 */
1698 goto error_get_channel_fatal;
1699 }
1700
1701 health_code_update();
1702
1703 /*
1704 * In no monitor mode, the streams ownership is kept inside the channel
1705 * so don't send them to the data thread.
1706 */
1707 if (!channel->monitor) {
1708 goto end_get_channel;
1709 }
1710
1711 ret = send_streams_to_thread(channel, ctx);
1712 if (ret < 0) {
1713 /*
1714 * If we are unable to send the stream to the thread, there is
1715 * a big problem so just stop everything.
1716 */
1717 goto error_get_channel_fatal;
1718 }
1719 /* List MUST be empty after or else it could be reused. */
1720 assert(cds_list_empty(&channel->streams.head));
1721 end_get_channel:
1722 goto end_msg_sessiond;
1723 error_get_channel_fatal:
1724 goto error_fatal;
1725 end_get_channel_nosignal:
1726 goto end_nosignal;
1727 }
1728 case LTTNG_CONSUMER_DESTROY_CHANNEL:
1729 {
1730 uint64_t key = msg.u.destroy_channel.key;
1731
1732 /*
1733 * Only called if streams have not been sent to stream
1734 * manager thread. However, channel has been sent to
1735 * channel manager thread.
1736 */
1737 notify_thread_del_channel(ctx, key);
1738 goto end_msg_sessiond;
1739 }
1740 case LTTNG_CONSUMER_CLOSE_METADATA:
1741 {
1742 int ret;
1743
1744 ret = close_metadata(msg.u.close_metadata.key);
1745 if (ret != 0) {
1746 ret_code = ret;
1747 }
1748
1749 goto end_msg_sessiond;
1750 }
1751 case LTTNG_CONSUMER_FLUSH_CHANNEL:
1752 {
1753 int ret;
1754
1755 ret = flush_channel(msg.u.flush_channel.key);
1756 if (ret != 0) {
1757 ret_code = ret;
1758 }
1759
1760 goto end_msg_sessiond;
1761 }
1762 case LTTNG_CONSUMER_CLEAR_QUIESCENT_CHANNEL:
1763 {
1764 int ret;
1765
1766 ret = clear_quiescent_channel(
1767 msg.u.clear_quiescent_channel.key);
1768 if (ret != 0) {
1769 ret_code = ret;
1770 }
1771
1772 goto end_msg_sessiond;
1773 }
1774 case LTTNG_CONSUMER_PUSH_METADATA:
1775 {
1776 int ret;
1777 uint64_t len = msg.u.push_metadata.len;
1778 uint64_t key = msg.u.push_metadata.key;
1779 uint64_t offset = msg.u.push_metadata.target_offset;
1780 uint64_t version = msg.u.push_metadata.version;
1781 struct lttng_consumer_channel *channel;
1782
1783 DBG("UST consumer push metadata key %" PRIu64 " of len %" PRIu64, key,
1784 len);
1785
1786 channel = consumer_find_channel(key);
1787 if (!channel) {
1788 /*
1789 * This is possible if the metadata creation on the consumer side
1790 * is in flight vis-a-vis a concurrent push metadata from the
1791 * session daemon. Simply return that the channel failed and the
1792 * session daemon will handle that message correctly considering
1793 * that this race is acceptable thus the DBG() statement here.
1794 */
1795 DBG("UST consumer push metadata %" PRIu64 " not found", key);
1796 ret_code = LTTCOMM_CONSUMERD_CHANNEL_FAIL;
1797 goto end_push_metadata_msg_sessiond;
1798 }
1799
1800 health_code_update();
1801
1802 if (!len) {
1803 /*
1804 * There is nothing to receive. We have simply
1805 * checked whether the channel can be found.
1806 */
1807 ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1808 goto end_push_metadata_msg_sessiond;
1809 }
1810
1811 /* Tell session daemon we are ready to receive the metadata. */
1812 ret = consumer_send_status_msg(sock, LTTCOMM_CONSUMERD_SUCCESS);
1813 if (ret < 0) {
1814 /* Somehow, the session daemon is not responding anymore. */
1815 goto error_push_metadata_fatal;
1816 }
1817
1818 health_code_update();
1819
1820 /* Wait for more data. */
1821 health_poll_entry();
1822 ret = lttng_consumer_poll_socket(consumer_sockpoll);
1823 health_poll_exit();
1824 if (ret) {
1825 goto error_push_metadata_fatal;
1826 }
1827
1828 health_code_update();
1829
1830 ret = lttng_ustconsumer_recv_metadata(sock, key, offset,
1831 len, version, channel, 0, 1);
1832 if (ret < 0) {
1833 /* error receiving from sessiond */
1834 goto error_push_metadata_fatal;
1835 } else {
1836 ret_code = ret;
1837 goto end_push_metadata_msg_sessiond;
1838 }
1839 end_push_metadata_msg_sessiond:
1840 goto end_msg_sessiond;
1841 error_push_metadata_fatal:
1842 goto error_fatal;
1843 }
1844 case LTTNG_CONSUMER_SETUP_METADATA:
1845 {
1846 int ret;
1847
1848 ret = setup_metadata(ctx, msg.u.setup_metadata.key);
1849 if (ret) {
1850 ret_code = ret;
1851 }
1852 goto end_msg_sessiond;
1853 }
1854 case LTTNG_CONSUMER_SNAPSHOT_CHANNEL:
1855 {
1856 struct lttng_consumer_channel *channel;
1857 uint64_t key = msg.u.snapshot_channel.key;
1858
1859 channel = consumer_find_channel(key);
1860 if (!channel) {
1861 DBG("UST snapshot channel not found for key %" PRIu64, key);
1862 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
1863 } else {
1864 if (msg.u.snapshot_channel.metadata) {
1865 ret = snapshot_metadata(channel, key,
1866 msg.u.snapshot_channel.pathname,
1867 msg.u.snapshot_channel.relayd_id,
1868 ctx);
1869 if (ret < 0) {
1870 ERR("Snapshot metadata failed");
1871 ret_code = LTTCOMM_CONSUMERD_SNAPSHOT_FAILED;
1872 }
1873 } else {
1874 ret = snapshot_channel(channel, key,
1875 msg.u.snapshot_channel.pathname,
1876 msg.u.snapshot_channel.relayd_id,
1877 msg.u.snapshot_channel.nb_packets_per_stream,
1878 ctx);
1879 if (ret < 0) {
1880 ERR("Snapshot channel failed");
1881 ret_code = LTTCOMM_CONSUMERD_SNAPSHOT_FAILED;
1882 }
1883 }
1884 }
1885 health_code_update();
1886 ret = consumer_send_status_msg(sock, ret_code);
1887 if (ret < 0) {
1888 /* Somehow, the session daemon is not responding anymore. */
1889 goto end_nosignal;
1890 }
1891 health_code_update();
1892 break;
1893 }
1894 case LTTNG_CONSUMER_DISCARDED_EVENTS:
1895 {
1896 int ret = 0;
1897 uint64_t discarded_events;
1898 struct lttng_ht_iter iter;
1899 struct lttng_ht *ht;
1900 struct lttng_consumer_stream *stream;
1901 uint64_t id = msg.u.discarded_events.session_id;
1902 uint64_t key = msg.u.discarded_events.channel_key;
1903
1904 DBG("UST consumer discarded events command for session id %"
1905 PRIu64, id);
1906 rcu_read_lock();
1907 pthread_mutex_lock(&consumer_data.lock);
1908
1909 ht = consumer_data.stream_list_ht;
1910
1911 /*
1912 * We only need a reference to the channel, but they are not
1913 * directly indexed, so we just use the first matching stream
1914 * to extract the information we need, we default to 0 if not
1915 * found (no events are dropped if the channel is not yet in
1916 * use).
1917 */
1918 discarded_events = 0;
1919 cds_lfht_for_each_entry_duplicate(ht->ht,
1920 ht->hash_fct(&id, lttng_ht_seed),
1921 ht->match_fct, &id,
1922 &iter.iter, stream, node_session_id.node) {
1923 if (stream->chan->key == key) {
1924 discarded_events = stream->chan->discarded_events;
1925 break;
1926 }
1927 }
1928 pthread_mutex_unlock(&consumer_data.lock);
1929 rcu_read_unlock();
1930
1931 DBG("UST consumer discarded events command for session id %"
1932 PRIu64 ", channel key %" PRIu64, id, key);
1933
1934 health_code_update();
1935
1936 /* Send back returned value to session daemon */
1937 ret = lttcomm_send_unix_sock(sock, &discarded_events, sizeof(discarded_events));
1938 if (ret < 0) {
1939 PERROR("send discarded events");
1940 goto error_fatal;
1941 }
1942
1943 break;
1944 }
1945 case LTTNG_CONSUMER_LOST_PACKETS:
1946 {
1947 int ret;
1948 uint64_t lost_packets;
1949 struct lttng_ht_iter iter;
1950 struct lttng_ht *ht;
1951 struct lttng_consumer_stream *stream;
1952 uint64_t id = msg.u.lost_packets.session_id;
1953 uint64_t key = msg.u.lost_packets.channel_key;
1954
1955 DBG("UST consumer lost packets command for session id %"
1956 PRIu64, id);
1957 rcu_read_lock();
1958 pthread_mutex_lock(&consumer_data.lock);
1959
1960 ht = consumer_data.stream_list_ht;
1961
1962 /*
1963 * We only need a reference to the channel, but they are not
1964 * directly indexed, so we just use the first matching stream
1965 * to extract the information we need, we default to 0 if not
1966 * found (no packets lost if the channel is not yet in use).
1967 */
1968 lost_packets = 0;
1969 cds_lfht_for_each_entry_duplicate(ht->ht,
1970 ht->hash_fct(&id, lttng_ht_seed),
1971 ht->match_fct, &id,
1972 &iter.iter, stream, node_session_id.node) {
1973 if (stream->chan->key == key) {
1974 lost_packets = stream->chan->lost_packets;
1975 break;
1976 }
1977 }
1978 pthread_mutex_unlock(&consumer_data.lock);
1979 rcu_read_unlock();
1980
1981 DBG("UST consumer lost packets command for session id %"
1982 PRIu64 ", channel key %" PRIu64, id, key);
1983
1984 health_code_update();
1985
1986 /* Send back returned value to session daemon */
1987 ret = lttcomm_send_unix_sock(sock, &lost_packets,
1988 sizeof(lost_packets));
1989 if (ret < 0) {
1990 PERROR("send lost packets");
1991 goto error_fatal;
1992 }
1993
1994 break;
1995 }
1996 case LTTNG_CONSUMER_SET_CHANNEL_MONITOR_PIPE:
1997 {
1998 int channel_monitor_pipe;
1999
2000 ret_code = LTTCOMM_CONSUMERD_SUCCESS;
2001 /* Successfully received the command's type. */
2002 ret = consumer_send_status_msg(sock, ret_code);
2003 if (ret < 0) {
2004 goto error_fatal;
2005 }
2006
2007 ret = lttcomm_recv_fds_unix_sock(sock, &channel_monitor_pipe,
2008 1);
2009 if (ret != sizeof(channel_monitor_pipe)) {
2010 ERR("Failed to receive channel monitor pipe");
2011 goto error_fatal;
2012 }
2013
2014 DBG("Received channel monitor pipe (%d)", channel_monitor_pipe);
2015 ret = consumer_timer_thread_set_channel_monitor_pipe(
2016 channel_monitor_pipe);
2017 if (!ret) {
2018 int flags;
2019
2020 ret_code = LTTCOMM_CONSUMERD_SUCCESS;
2021 /* Set the pipe as non-blocking. */
2022 ret = fcntl(channel_monitor_pipe, F_GETFL, 0);
2023 if (ret == -1) {
2024 PERROR("fcntl get flags of the channel monitoring pipe");
2025 goto error_fatal;
2026 }
2027 flags = ret;
2028
2029 ret = fcntl(channel_monitor_pipe, F_SETFL,
2030 flags | O_NONBLOCK);
2031 if (ret == -1) {
2032 PERROR("fcntl set O_NONBLOCK flag of the channel monitoring pipe");
2033 goto error_fatal;
2034 }
2035 DBG("Channel monitor pipe set as non-blocking");
2036 } else {
2037 ret_code = LTTCOMM_CONSUMERD_ALREADY_SET;
2038 }
2039 goto end_msg_sessiond;
2040 }
2041 case LTTNG_CONSUMER_ROTATE_CHANNEL:
2042 {
2043 struct lttng_consumer_channel *channel;
2044 uint64_t key = msg.u.rotate_channel.key;
2045
2046 channel = consumer_find_channel(key);
2047 if (!channel) {
2048 DBG("Channel %" PRIu64 " not found", key);
2049 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
2050 } else {
2051 /*
2052 * Sample the rotate position of all the streams in
2053 * this channel.
2054 */
2055 ret = lttng_consumer_rotate_channel(channel, key,
2056 msg.u.rotate_channel.relayd_id,
2057 msg.u.rotate_channel.metadata,
2058 ctx);
2059 if (ret < 0) {
2060 ERR("Rotate channel failed");
2061 ret_code = LTTCOMM_CONSUMERD_ROTATION_FAIL;
2062 }
2063
2064 health_code_update();
2065 }
2066 ret = consumer_send_status_msg(sock, ret_code);
2067 if (ret < 0) {
2068 /* Somehow, the session daemon is not responding anymore. */
2069 goto end_rotate_channel_nosignal;
2070 }
2071
2072 /*
2073 * Rotate the streams that are ready right now.
2074 * FIXME: this is a second consecutive iteration over the
2075 * streams in a channel, there is probably a better way to
2076 * handle this, but it needs to be after the
2077 * consumer_send_status_msg() call.
2078 */
2079 if (channel) {
2080 ret = lttng_consumer_rotate_ready_streams(
2081 channel, key, ctx);
2082 if (ret < 0) {
2083 ERR("Rotate channel failed");
2084 }
2085 }
2086 break;
2087 end_rotate_channel_nosignal:
2088 goto end_nosignal;
2089 }
2090 case LTTNG_CONSUMER_CLEAR_CHANNEL:
2091 {
2092 struct lttng_consumer_channel *channel;
2093 uint64_t key = msg.u.clear_channel.key;
2094
2095 channel = consumer_find_channel(key);
2096 if (!channel) {
2097 DBG("Channel %" PRIu64 " not found", key);
2098 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
2099 } else {
2100 ret = lttng_consumer_clear_channel(channel);
2101 if (ret) {
2102 ERR("Clear channel failed key %" PRIu64, key);
2103 ret_code = ret;
2104 }
2105
2106 health_code_update();
2107 }
2108 ret = consumer_send_status_msg(sock, ret_code);
2109 if (ret < 0) {
2110 /* Somehow, the session daemon is not responding anymore. */
2111 goto end_nosignal;
2112 }
2113 break;
2114 }
2115 case LTTNG_CONSUMER_INIT:
2116 {
2117 ret_code = lttng_consumer_init_command(ctx,
2118 msg.u.init.sessiond_uuid);
2119 health_code_update();
2120 ret = consumer_send_status_msg(sock, ret_code);
2121 if (ret < 0) {
2122 /* Somehow, the session daemon is not responding anymore. */
2123 goto end_nosignal;
2124 }
2125 break;
2126 }
2127 case LTTNG_CONSUMER_CREATE_TRACE_CHUNK:
2128 {
2129 const struct lttng_credentials credentials = {
2130 .uid = msg.u.create_trace_chunk.credentials.value.uid,
2131 .gid = msg.u.create_trace_chunk.credentials.value.gid,
2132 };
2133 const bool is_local_trace =
2134 !msg.u.create_trace_chunk.relayd_id.is_set;
2135 const uint64_t relayd_id =
2136 msg.u.create_trace_chunk.relayd_id.value;
2137 const char *chunk_override_name =
2138 *msg.u.create_trace_chunk.override_name ?
2139 msg.u.create_trace_chunk.override_name :
2140 NULL;
2141 struct lttng_directory_handle *chunk_directory_handle = NULL;
2142
2143 /*
2144 * The session daemon will only provide a chunk directory file
2145 * descriptor for local traces.
2146 */
2147 if (is_local_trace) {
2148 int chunk_dirfd;
2149
2150 /* Acnowledge the reception of the command. */
2151 ret = consumer_send_status_msg(sock,
2152 LTTCOMM_CONSUMERD_SUCCESS);
2153 if (ret < 0) {
2154 /* Somehow, the session daemon is not responding anymore. */
2155 goto end_nosignal;
2156 }
2157
2158 /*
2159 * Receive trace chunk domain dirfd.
2160 */
2161 ret = lttcomm_recv_fds_unix_sock(sock, &chunk_dirfd, 1);
2162 if (ret != sizeof(chunk_dirfd)) {
2163 ERR("Failed to receive trace chunk domain directory file descriptor");
2164 goto error_fatal;
2165 }
2166
2167 DBG("Received trace chunk domain directory fd (%d)",
2168 chunk_dirfd);
2169 chunk_directory_handle = lttng_directory_handle_create_from_dirfd(
2170 chunk_dirfd);
2171 if (!chunk_directory_handle) {
2172 ERR("Failed to initialize chunk domain directory handle from directory file descriptor");
2173 if (close(chunk_dirfd)) {
2174 PERROR("Failed to close chunk directory file descriptor");
2175 }
2176 goto error_fatal;
2177 }
2178 }
2179
2180 ret_code = lttng_consumer_create_trace_chunk(
2181 !is_local_trace ? &relayd_id : NULL,
2182 msg.u.create_trace_chunk.session_id,
2183 msg.u.create_trace_chunk.chunk_id,
2184 (time_t) msg.u.create_trace_chunk
2185 .creation_timestamp,
2186 chunk_override_name,
2187 msg.u.create_trace_chunk.credentials.is_set ?
2188 &credentials :
2189 NULL,
2190 chunk_directory_handle);
2191 lttng_directory_handle_put(chunk_directory_handle);
2192 goto end_msg_sessiond;
2193 }
2194 case LTTNG_CONSUMER_CLOSE_TRACE_CHUNK:
2195 {
2196 enum lttng_trace_chunk_command_type close_command =
2197 msg.u.close_trace_chunk.close_command.value;
2198 const uint64_t relayd_id =
2199 msg.u.close_trace_chunk.relayd_id.value;
2200 struct lttcomm_consumer_close_trace_chunk_reply reply;
2201 char closed_trace_chunk_path[LTTNG_PATH_MAX] = {};
2202 int ret;
2203
2204 ret_code = lttng_consumer_close_trace_chunk(
2205 msg.u.close_trace_chunk.relayd_id.is_set ?
2206 &relayd_id :
2207 NULL,
2208 msg.u.close_trace_chunk.session_id,
2209 msg.u.close_trace_chunk.chunk_id,
2210 (time_t) msg.u.close_trace_chunk.close_timestamp,
2211 msg.u.close_trace_chunk.close_command.is_set ?
2212 &close_command :
2213 NULL, closed_trace_chunk_path);
2214 reply.ret_code = ret_code;
2215 reply.path_length = strlen(closed_trace_chunk_path) + 1;
2216 ret = lttcomm_send_unix_sock(sock, &reply, sizeof(reply));
2217 if (ret != sizeof(reply)) {
2218 goto error_fatal;
2219 }
2220 ret = lttcomm_send_unix_sock(sock, closed_trace_chunk_path,
2221 reply.path_length);
2222 if (ret != reply.path_length) {
2223 goto error_fatal;
2224 }
2225 goto end_nosignal;
2226 }
2227 case LTTNG_CONSUMER_TRACE_CHUNK_EXISTS:
2228 {
2229 const uint64_t relayd_id =
2230 msg.u.trace_chunk_exists.relayd_id.value;
2231
2232 ret_code = lttng_consumer_trace_chunk_exists(
2233 msg.u.trace_chunk_exists.relayd_id.is_set ?
2234 &relayd_id : NULL,
2235 msg.u.trace_chunk_exists.session_id,
2236 msg.u.trace_chunk_exists.chunk_id);
2237 goto end_msg_sessiond;
2238 }
2239 case LTTNG_CONSUMER_OPEN_CHANNEL_PACKETS:
2240 {
2241 const uint64_t key = msg.u.open_channel_packets.key;
2242 struct lttng_consumer_channel *channel =
2243 consumer_find_channel(key);
2244
2245 if (channel) {
2246 pthread_mutex_lock(&channel->lock);
2247 ret_code = lttng_consumer_open_channel_packets(channel);
2248 pthread_mutex_unlock(&channel->lock);
2249 } else {
2250 /*
2251 * The channel could have disappeared in per-pid
2252 * buffering mode.
2253 */
2254 DBG("Channel %" PRIu64 " not found", key);
2255 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
2256 }
2257
2258 health_code_update();
2259 goto end_msg_sessiond;
2260 }
2261 default:
2262 break;
2263 }
2264
2265 end_nosignal:
2266 /*
2267 * Return 1 to indicate success since the 0 value can be a socket
2268 * shutdown during the recv() or send() call.
2269 */
2270 ret = 1;
2271 goto end;
2272
2273 end_msg_sessiond:
2274 /*
2275 * The returned value here is not useful since either way we'll return 1 to
2276 * the caller because the session daemon socket management is done
2277 * elsewhere. Returning a negative code or 0 will shutdown the consumer.
2278 */
2279 ret = consumer_send_status_msg(sock, ret_code);
2280 if (ret < 0) {
2281 goto error_fatal;
2282 }
2283 ret = 1;
2284 goto end;
2285
2286 end_channel_error:
2287 if (channel) {
2288 /*
2289 * Free channel here since no one has a reference to it. We don't
2290 * free after that because a stream can store this pointer.
2291 */
2292 destroy_channel(channel);
2293 }
2294 /* We have to send a status channel message indicating an error. */
2295 ret = consumer_send_status_channel(sock, NULL);
2296 if (ret < 0) {
2297 /* Stop everything if session daemon can not be notified. */
2298 goto error_fatal;
2299 }
2300 ret = 1;
2301 goto end;
2302
2303 error_fatal:
2304 /* This will issue a consumer stop. */
2305 ret = -1;
2306 goto end;
2307
2308 end:
2309 rcu_read_unlock();
2310 health_code_update();
2311 return ret;
2312 }
2313
2314 void lttng_ustctl_flush_buffer(struct lttng_consumer_stream *stream,
2315 int producer_active)
2316 {
2317 assert(stream);
2318 assert(stream->ustream);
2319
2320 ustctl_flush_buffer(stream->ustream, producer_active);
2321 }
2322
2323 /*
2324 * Take a snapshot for a specific stream.
2325 *
2326 * Returns 0 on success, < 0 on error
2327 */
2328 int lttng_ustconsumer_take_snapshot(struct lttng_consumer_stream *stream)
2329 {
2330 assert(stream);
2331 assert(stream->ustream);
2332
2333 return ustctl_snapshot(stream->ustream);
2334 }
2335
2336 /*
2337 * Sample consumed and produced positions for a specific stream.
2338 *
2339 * Returns 0 on success, < 0 on error.
2340 */
2341 int lttng_ustconsumer_sample_snapshot_positions(
2342 struct lttng_consumer_stream *stream)
2343 {
2344 assert(stream);
2345 assert(stream->ustream);
2346
2347 return ustctl_snapshot_sample_positions(stream->ustream);
2348 }
2349
2350 /*
2351 * Get the produced position
2352 *
2353 * Returns 0 on success, < 0 on error
2354 */
2355 int lttng_ustconsumer_get_produced_snapshot(
2356 struct lttng_consumer_stream *stream, unsigned long *pos)
2357 {
2358 assert(stream);
2359 assert(stream->ustream);
2360 assert(pos);
2361
2362 return ustctl_snapshot_get_produced(stream->ustream, pos);
2363 }
2364
2365 /*
2366 * Get the consumed position
2367 *
2368 * Returns 0 on success, < 0 on error
2369 */
2370 int lttng_ustconsumer_get_consumed_snapshot(
2371 struct lttng_consumer_stream *stream, unsigned long *pos)
2372 {
2373 assert(stream);
2374 assert(stream->ustream);
2375 assert(pos);
2376
2377 return ustctl_snapshot_get_consumed(stream->ustream, pos);
2378 }
2379
2380 void lttng_ustconsumer_flush_buffer(struct lttng_consumer_stream *stream,
2381 int producer)
2382 {
2383 assert(stream);
2384 assert(stream->ustream);
2385
2386 ustctl_flush_buffer(stream->ustream, producer);
2387 }
2388
2389 void lttng_ustconsumer_clear_buffer(struct lttng_consumer_stream *stream)
2390 {
2391 assert(stream);
2392 assert(stream->ustream);
2393
2394 ustctl_clear_buffer(stream->ustream);
2395 }
2396
2397 int lttng_ustconsumer_get_current_timestamp(
2398 struct lttng_consumer_stream *stream, uint64_t *ts)
2399 {
2400 assert(stream);
2401 assert(stream->ustream);
2402 assert(ts);
2403
2404 return ustctl_get_current_timestamp(stream->ustream, ts);
2405 }
2406
2407 int lttng_ustconsumer_get_sequence_number(
2408 struct lttng_consumer_stream *stream, uint64_t *seq)
2409 {
2410 assert(stream);
2411 assert(stream->ustream);
2412 assert(seq);
2413
2414 return ustctl_get_sequence_number(stream->ustream, seq);
2415 }
2416
2417 /*
2418 * Called when the stream signals the consumer that it has hung up.
2419 */
2420 void lttng_ustconsumer_on_stream_hangup(struct lttng_consumer_stream *stream)
2421 {
2422 assert(stream);
2423 assert(stream->ustream);
2424
2425 pthread_mutex_lock(&stream->lock);
2426 if (!stream->quiescent) {
2427 ustctl_flush_buffer(stream->ustream, 0);
2428 stream->quiescent = true;
2429 }
2430 pthread_mutex_unlock(&stream->lock);
2431 stream->hangup_flush_done = 1;
2432 }
2433
2434 void lttng_ustconsumer_del_channel(struct lttng_consumer_channel *chan)
2435 {
2436 int i;
2437
2438 assert(chan);
2439 assert(chan->uchan);
2440 assert(chan->buffer_credentials.is_set);
2441
2442 if (chan->switch_timer_enabled == 1) {
2443 consumer_timer_switch_stop(chan);
2444 }
2445 for (i = 0; i < chan->nr_stream_fds; i++) {
2446 int ret;
2447
2448 ret = close(chan->stream_fds[i]);
2449 if (ret) {
2450 PERROR("close");
2451 }
2452 if (chan->shm_path[0]) {
2453 char shm_path[PATH_MAX];
2454
2455 ret = get_stream_shm_path(shm_path, chan->shm_path, i);
2456 if (ret) {
2457 ERR("Cannot get stream shm path");
2458 }
2459 ret = run_as_unlink(shm_path,
2460 chan->buffer_credentials.value.uid,
2461 chan->buffer_credentials.value.gid);
2462 if (ret) {
2463 PERROR("unlink %s", shm_path);
2464 }
2465 }
2466 }
2467 }
2468
2469 void lttng_ustconsumer_free_channel(struct lttng_consumer_channel *chan)
2470 {
2471 assert(chan);
2472 assert(chan->uchan);
2473 assert(chan->buffer_credentials.is_set);
2474
2475 consumer_metadata_cache_destroy(chan);
2476 ustctl_destroy_channel(chan->uchan);
2477 /* Try to rmdir all directories under shm_path root. */
2478 if (chan->root_shm_path[0]) {
2479 (void) run_as_rmdir_recursive(chan->root_shm_path,
2480 chan->buffer_credentials.value.uid,
2481 chan->buffer_credentials.value.gid,
2482 LTTNG_DIRECTORY_HANDLE_SKIP_NON_EMPTY_FLAG);
2483 }
2484 free(chan->stream_fds);
2485 }
2486
2487 void lttng_ustconsumer_del_stream(struct lttng_consumer_stream *stream)
2488 {
2489 assert(stream);
2490 assert(stream->ustream);
2491
2492 if (stream->chan->switch_timer_enabled == 1) {
2493 consumer_timer_switch_stop(stream->chan);
2494 }
2495 ustctl_destroy_stream(stream->ustream);
2496 }
2497
2498 int lttng_ustconsumer_get_wakeup_fd(struct lttng_consumer_stream *stream)
2499 {
2500 assert(stream);
2501 assert(stream->ustream);
2502
2503 return ustctl_stream_get_wakeup_fd(stream->ustream);
2504 }
2505
2506 int lttng_ustconsumer_close_wakeup_fd(struct lttng_consumer_stream *stream)
2507 {
2508 assert(stream);
2509 assert(stream->ustream);
2510
2511 return ustctl_stream_close_wakeup_fd(stream->ustream);
2512 }
2513
2514 /*
2515 * Write up to one packet from the metadata cache to the channel.
2516 *
2517 * Returns the number of bytes pushed from the cache into the ring buffer, or a
2518 * negative value on error.
2519 */
2520 static
2521 int commit_one_metadata_packet(struct lttng_consumer_stream *stream)
2522 {
2523 ssize_t write_len;
2524 int ret;
2525
2526 pthread_mutex_lock(&stream->chan->metadata_cache->lock);
2527 if (stream->chan->metadata_cache->max_offset ==
2528 stream->ust_metadata_pushed) {
2529 /*
2530 * In the context of a user space metadata channel, a
2531 * change in version can be detected in two ways:
2532 * 1) During the pre-consume of the `read_subbuffer` loop,
2533 * 2) When populating the metadata ring buffer (i.e. here).
2534 *
2535 * This function is invoked when there is no metadata
2536 * available in the ring-buffer. If all data was consumed
2537 * up to the size of the metadata cache, there is no metadata
2538 * to insert in the ring-buffer.
2539 *
2540 * However, the metadata version could still have changed (a
2541 * regeneration without any new data will yield the same cache
2542 * size).
2543 *
2544 * The cache's version is checked for a version change and the
2545 * consumed position is reset if one occurred.
2546 *
2547 * This check is only necessary for the user space domain as
2548 * it has to manage the cache explicitly. If this reset was not
2549 * performed, no metadata would be consumed (and no reset would
2550 * occur as part of the pre-consume) until the metadata size
2551 * exceeded the cache size.
2552 */
2553 if (stream->metadata_version !=
2554 stream->chan->metadata_cache->version) {
2555 metadata_stream_reset_cache_consumed_position(stream);
2556 consumer_stream_metadata_set_version(stream,
2557 stream->chan->metadata_cache->version);
2558 } else {
2559 ret = 0;
2560 goto end;
2561 }
2562 }
2563
2564 write_len = ustctl_write_one_packet_to_channel(stream->chan->uchan,
2565 &stream->chan->metadata_cache->data[stream->ust_metadata_pushed],
2566 stream->chan->metadata_cache->max_offset
2567 - stream->ust_metadata_pushed);
2568 assert(write_len != 0);
2569 if (write_len < 0) {
2570 ERR("Writing one metadata packet");
2571 ret = write_len;
2572 goto end;
2573 }
2574 stream->ust_metadata_pushed += write_len;
2575
2576 assert(stream->chan->metadata_cache->max_offset >=
2577 stream->ust_metadata_pushed);
2578 ret = write_len;
2579
2580 /*
2581 * Switch packet (but don't open the next one) on every commit of
2582 * a metadata packet. Since the subbuffer is fully filled (with padding,
2583 * if needed), the stream is "quiescent" after this commit.
2584 */
2585 ustctl_flush_buffer(stream->ustream, 1);
2586 stream->quiescent = true;
2587 end:
2588 pthread_mutex_unlock(&stream->chan->metadata_cache->lock);
2589 return ret;
2590 }
2591
2592
2593 /*
2594 * Sync metadata meaning request them to the session daemon and snapshot to the
2595 * metadata thread can consumer them.
2596 *
2597 * Metadata stream lock is held here, but we need to release it when
2598 * interacting with sessiond, else we cause a deadlock with live
2599 * awaiting on metadata to be pushed out.
2600 *
2601 * The RCU read side lock must be held by the caller.
2602 */
2603 enum sync_metadata_status lttng_ustconsumer_sync_metadata(
2604 struct lttng_consumer_local_data *ctx,
2605 struct lttng_consumer_stream *metadata_stream)
2606 {
2607 int ret;
2608 enum sync_metadata_status status;
2609 struct lttng_consumer_channel *metadata_channel;
2610
2611 assert(ctx);
2612 assert(metadata_stream);
2613
2614 metadata_channel = metadata_stream->chan;
2615 pthread_mutex_unlock(&metadata_stream->lock);
2616 /*
2617 * Request metadata from the sessiond, but don't wait for the flush
2618 * because we locked the metadata thread.
2619 */
2620 ret = lttng_ustconsumer_request_metadata(ctx, metadata_channel, 0, 0);
2621 pthread_mutex_lock(&metadata_stream->lock);
2622 if (ret < 0) {
2623 status = SYNC_METADATA_STATUS_ERROR;
2624 goto end;
2625 }
2626
2627 /*
2628 * The metadata stream and channel can be deleted while the
2629 * metadata stream lock was released. The streamed is checked
2630 * for deletion before we use it further.
2631 *
2632 * Note that it is safe to access a logically-deleted stream since its
2633 * existence is still guaranteed by the RCU read side lock. However,
2634 * it should no longer be used. The close/deletion of the metadata
2635 * channel and stream already guarantees that all metadata has been
2636 * consumed. Therefore, there is nothing left to do in this function.
2637 */
2638 if (consumer_stream_is_deleted(metadata_stream)) {
2639 DBG("Metadata stream %" PRIu64 " was deleted during the metadata synchronization",
2640 metadata_stream->key);
2641 status = SYNC_METADATA_STATUS_NO_DATA;
2642 goto end;
2643 }
2644
2645 ret = commit_one_metadata_packet(metadata_stream);
2646 if (ret < 0) {
2647 status = SYNC_METADATA_STATUS_ERROR;
2648 goto end;
2649 } else if (ret > 0) {
2650 status = SYNC_METADATA_STATUS_NEW_DATA;
2651 } else /* ret == 0 */ {
2652 status = SYNC_METADATA_STATUS_NO_DATA;
2653 goto end;
2654 }
2655
2656 ret = ustctl_snapshot(metadata_stream->ustream);
2657 if (ret < 0) {
2658 ERR("Failed to take a snapshot of the metadata ring-buffer positions, ret = %d", ret);
2659 status = SYNC_METADATA_STATUS_ERROR;
2660 goto end;
2661 }
2662
2663 end:
2664 return status;
2665 }
2666
2667 /*
2668 * Return 0 on success else a negative value.
2669 */
2670 static int notify_if_more_data(struct lttng_consumer_stream *stream,
2671 struct lttng_consumer_local_data *ctx)
2672 {
2673 int ret;
2674 struct ustctl_consumer_stream *ustream;
2675
2676 assert(stream);
2677 assert(ctx);
2678
2679 ustream = stream->ustream;
2680
2681 /*
2682 * First, we are going to check if there is a new subbuffer available
2683 * before reading the stream wait_fd.
2684 */
2685 /* Get the next subbuffer */
2686 ret = ustctl_get_next_subbuf(ustream);
2687 if (ret) {
2688 /* No more data found, flag the stream. */
2689 stream->has_data = 0;
2690 ret = 0;
2691 goto end;
2692 }
2693
2694 ret = ustctl_put_subbuf(ustream);
2695 assert(!ret);
2696
2697 /* This stream still has data. Flag it and wake up the data thread. */
2698 stream->has_data = 1;
2699
2700 if (stream->monitor && !stream->hangup_flush_done && !ctx->has_wakeup) {
2701 ssize_t writelen;
2702
2703 writelen = lttng_pipe_write(ctx->consumer_wakeup_pipe, "!", 1);
2704 if (writelen < 0 && errno != EAGAIN && errno != EWOULDBLOCK) {
2705 ret = writelen;
2706 goto end;
2707 }
2708
2709 /* The wake up pipe has been notified. */
2710 ctx->has_wakeup = 1;
2711 }
2712 ret = 0;
2713
2714 end:
2715 return ret;
2716 }
2717
2718 static int consumer_stream_ust_on_wake_up(struct lttng_consumer_stream *stream)
2719 {
2720 int ret = 0;
2721
2722 /*
2723 * We can consume the 1 byte written into the wait_fd by
2724 * UST. Don't trigger error if we cannot read this one byte
2725 * (read returns 0), or if the error is EAGAIN or EWOULDBLOCK.
2726 *
2727 * This is only done when the stream is monitored by a thread,
2728 * before the flush is done after a hangup and if the stream
2729 * is not flagged with data since there might be nothing to
2730 * consume in the wait fd but still have data available
2731 * flagged by the consumer wake up pipe.
2732 */
2733 if (stream->monitor && !stream->hangup_flush_done && !stream->has_data) {
2734 char dummy;
2735 ssize_t readlen;
2736
2737 readlen = lttng_read(stream->wait_fd, &dummy, 1);
2738 if (readlen < 0 && errno != EAGAIN && errno != EWOULDBLOCK) {
2739 ret = readlen;
2740 }
2741 }
2742
2743 return ret;
2744 }
2745
2746 static int extract_common_subbuffer_info(struct lttng_consumer_stream *stream,
2747 struct stream_subbuffer *subbuf)
2748 {
2749 int ret;
2750
2751 ret = ustctl_get_subbuf_size(
2752 stream->ustream, &subbuf->info.data.subbuf_size);
2753 if (ret) {
2754 goto end;
2755 }
2756
2757 ret = ustctl_get_padded_subbuf_size(
2758 stream->ustream, &subbuf->info.data.padded_subbuf_size);
2759 if (ret) {
2760 goto end;
2761 }
2762
2763 end:
2764 return ret;
2765 }
2766
2767 static int extract_metadata_subbuffer_info(struct lttng_consumer_stream *stream,
2768 struct stream_subbuffer *subbuf)
2769 {
2770 int ret;
2771
2772 ret = extract_common_subbuffer_info(stream, subbuf);
2773 if (ret) {
2774 goto end;
2775 }
2776
2777 subbuf->info.metadata.version = stream->metadata_version;
2778
2779 end:
2780 return ret;
2781 }
2782
2783 static int extract_data_subbuffer_info(struct lttng_consumer_stream *stream,
2784 struct stream_subbuffer *subbuf)
2785 {
2786 int ret;
2787
2788 ret = extract_common_subbuffer_info(stream, subbuf);
2789 if (ret) {
2790 goto end;
2791 }
2792
2793 ret = ustctl_get_packet_size(
2794 stream->ustream, &subbuf->info.data.packet_size);
2795 if (ret < 0) {
2796 PERROR("Failed to get sub-buffer packet size");
2797 goto end;
2798 }
2799
2800 ret = ustctl_get_content_size(
2801 stream->ustream, &subbuf->info.data.content_size);
2802 if (ret < 0) {
2803 PERROR("Failed to get sub-buffer content size");
2804 goto end;
2805 }
2806
2807 ret = ustctl_get_timestamp_begin(
2808 stream->ustream, &subbuf->info.data.timestamp_begin);
2809 if (ret < 0) {
2810 PERROR("Failed to get sub-buffer begin timestamp");
2811 goto end;
2812 }
2813
2814 ret = ustctl_get_timestamp_end(
2815 stream->ustream, &subbuf->info.data.timestamp_end);
2816 if (ret < 0) {
2817 PERROR("Failed to get sub-buffer end timestamp");
2818 goto end;
2819 }
2820
2821 ret = ustctl_get_events_discarded(
2822 stream->ustream, &subbuf->info.data.events_discarded);
2823 if (ret) {
2824 PERROR("Failed to get sub-buffer events discarded count");
2825 goto end;
2826 }
2827
2828 ret = ustctl_get_sequence_number(stream->ustream,
2829 &subbuf->info.data.sequence_number.value);
2830 if (ret) {
2831 /* May not be supported by older LTTng-modules. */
2832 if (ret != -ENOTTY) {
2833 PERROR("Failed to get sub-buffer sequence number");
2834 goto end;
2835 }
2836 } else {
2837 subbuf->info.data.sequence_number.is_set = true;
2838 }
2839
2840 ret = ustctl_get_stream_id(
2841 stream->ustream, &subbuf->info.data.stream_id);
2842 if (ret < 0) {
2843 PERROR("Failed to get stream id");
2844 goto end;
2845 }
2846
2847 ret = ustctl_get_instance_id(stream->ustream,
2848 &subbuf->info.data.stream_instance_id.value);
2849 if (ret) {
2850 /* May not be supported by older LTTng-modules. */
2851 if (ret != -ENOTTY) {
2852 PERROR("Failed to get stream instance id");
2853 goto end;
2854 }
2855 } else {
2856 subbuf->info.data.stream_instance_id.is_set = true;
2857 }
2858 end:
2859 return ret;
2860 }
2861
2862 static int get_next_subbuffer_common(struct lttng_consumer_stream *stream,
2863 struct stream_subbuffer *subbuffer)
2864 {
2865 int ret;
2866 const char *addr;
2867
2868 ret = stream->read_subbuffer_ops.extract_subbuffer_info(
2869 stream, subbuffer);
2870 if (ret) {
2871 goto end;
2872 }
2873
2874 ret = get_current_subbuf_addr(stream, &addr);
2875 if (ret) {
2876 goto end;
2877 }
2878
2879 subbuffer->buffer.buffer = lttng_buffer_view_init(
2880 addr, 0, subbuffer->info.data.padded_subbuf_size);
2881 assert(subbuffer->buffer.buffer.data != NULL);
2882 end:
2883 return ret;
2884 }
2885
2886 static int get_next_subbuffer(struct lttng_consumer_stream *stream,
2887 struct stream_subbuffer *subbuffer)
2888 {
2889 int ret;
2890
2891 ret = ustctl_get_next_subbuf(stream->ustream);
2892 if (ret) {
2893 goto end;
2894 }
2895
2896 ret = get_next_subbuffer_common(stream, subbuffer);
2897 if (ret) {
2898 goto end;
2899 }
2900 end:
2901 return ret;
2902 }
2903
2904 static int get_next_subbuffer_metadata(struct lttng_consumer_stream *stream,
2905 struct stream_subbuffer *subbuffer)
2906 {
2907 int ret;
2908 bool cache_empty;
2909 bool got_subbuffer;
2910 bool coherent;
2911 bool buffer_empty;
2912 unsigned long consumed_pos, produced_pos;
2913
2914 do {
2915 ret = ustctl_get_next_subbuf(stream->ustream);
2916 if (ret == 0) {
2917 got_subbuffer = true;
2918 } else {
2919 got_subbuffer = false;
2920 if (ret != -EAGAIN) {
2921 /* Fatal error. */
2922 goto end;
2923 }
2924 }
2925
2926 /*
2927 * Determine if the cache is empty and ensure that a sub-buffer
2928 * is made available if the cache is not empty.
2929 */
2930 if (!got_subbuffer) {
2931 ret = commit_one_metadata_packet(stream);
2932 if (ret < 0 && ret != -ENOBUFS) {
2933 goto end;
2934 } else if (ret == 0) {
2935 /* Not an error, the cache is empty. */
2936 cache_empty = true;
2937 ret = -ENODATA;
2938 goto end;
2939 } else {
2940 cache_empty = false;
2941 }
2942 } else {
2943 pthread_mutex_lock(&stream->chan->metadata_cache->lock);
2944 cache_empty = stream->chan->metadata_cache->max_offset ==
2945 stream->ust_metadata_pushed;
2946 pthread_mutex_unlock(&stream->chan->metadata_cache->lock);
2947 }
2948 } while (!got_subbuffer);
2949
2950 /* Populate sub-buffer infos and view. */
2951 ret = get_next_subbuffer_common(stream, subbuffer);
2952 if (ret) {
2953 goto end;
2954 }
2955
2956 ret = lttng_ustconsumer_sample_snapshot_positions(stream);
2957 if (ret < 0) {
2958 /*
2959 * -EAGAIN is not expected since we got a sub-buffer and haven't
2960 * pushed the consumption position yet (on put_next).
2961 */
2962 PERROR("Failed to take a snapshot of metadata buffer positions");
2963 goto end;
2964 }
2965
2966 ret = lttng_ustconsumer_get_consumed_snapshot(stream, &consumed_pos);
2967 if (ret) {
2968 PERROR("Failed to get metadata consumed position");
2969 goto end;
2970 }
2971
2972 ret = lttng_ustconsumer_get_produced_snapshot(stream, &produced_pos);
2973 if (ret) {
2974 PERROR("Failed to get metadata produced position");
2975 goto end;
2976 }
2977
2978 /* Last sub-buffer of the ring buffer ? */
2979 buffer_empty = (consumed_pos + stream->max_sb_size) == produced_pos;
2980
2981 /*
2982 * The sessiond registry lock ensures that coherent units of metadata
2983 * are pushed to the consumer daemon at once. Hence, if a sub-buffer is
2984 * acquired, the cache is empty, and it is the only available sub-buffer
2985 * available, it is safe to assume that it is "coherent".
2986 */
2987 coherent = got_subbuffer && cache_empty && buffer_empty;
2988
2989 LTTNG_OPTIONAL_SET(&subbuffer->info.metadata.coherent, coherent);
2990 end:
2991 return ret;
2992 }
2993
2994 static int put_next_subbuffer(struct lttng_consumer_stream *stream,
2995 struct stream_subbuffer *subbuffer)
2996 {
2997 const int ret = ustctl_put_next_subbuf(stream->ustream);
2998
2999 assert(ret == 0);
3000 return ret;
3001 }
3002
3003 static int signal_metadata(struct lttng_consumer_stream *stream,
3004 struct lttng_consumer_local_data *ctx)
3005 {
3006 return pthread_cond_broadcast(&stream->metadata_rdv) ? -errno : 0;
3007 }
3008
3009 static int lttng_ustconsumer_set_stream_ops(
3010 struct lttng_consumer_stream *stream)
3011 {
3012 int ret = 0;
3013
3014 stream->read_subbuffer_ops.on_wake_up = consumer_stream_ust_on_wake_up;
3015 if (stream->metadata_flag) {
3016 stream->read_subbuffer_ops.get_next_subbuffer =
3017 get_next_subbuffer_metadata;
3018 stream->read_subbuffer_ops.extract_subbuffer_info =
3019 extract_metadata_subbuffer_info;
3020 stream->read_subbuffer_ops.reset_metadata =
3021 metadata_stream_reset_cache_consumed_position;
3022 if (stream->chan->is_live) {
3023 stream->read_subbuffer_ops.on_sleep = signal_metadata;
3024 ret = consumer_stream_enable_metadata_bucketization(
3025 stream);
3026 if (ret) {
3027 goto end;
3028 }
3029 }
3030 } else {
3031 stream->read_subbuffer_ops.get_next_subbuffer =
3032 get_next_subbuffer;
3033 stream->read_subbuffer_ops.extract_subbuffer_info =
3034 extract_data_subbuffer_info;
3035 stream->read_subbuffer_ops.on_sleep = notify_if_more_data;
3036 if (stream->chan->is_live) {
3037 stream->read_subbuffer_ops.send_live_beacon =
3038 consumer_flush_ust_index;
3039 }
3040 }
3041
3042 stream->read_subbuffer_ops.put_next_subbuffer = put_next_subbuffer;
3043 end:
3044 return ret;
3045 }
3046
3047 /*
3048 * Called when a stream is created.
3049 *
3050 * Return 0 on success or else a negative value.
3051 */
3052 int lttng_ustconsumer_on_recv_stream(struct lttng_consumer_stream *stream)
3053 {
3054 int ret;
3055
3056 assert(stream);
3057
3058 /*
3059 * Don't create anything if this is set for streaming or if there is
3060 * no current trace chunk on the parent channel.
3061 */
3062 if (stream->net_seq_idx == (uint64_t) -1ULL && stream->chan->monitor &&
3063 stream->chan->trace_chunk) {
3064 ret = consumer_stream_create_output_files(stream, true);
3065 if (ret) {
3066 goto error;
3067 }
3068 }
3069
3070 lttng_ustconsumer_set_stream_ops(stream);
3071 ret = 0;
3072
3073 error:
3074 return ret;
3075 }
3076
3077 /*
3078 * Check if data is still being extracted from the buffers for a specific
3079 * stream. Consumer data lock MUST be acquired before calling this function
3080 * and the stream lock.
3081 *
3082 * Return 1 if the traced data are still getting read else 0 meaning that the
3083 * data is available for trace viewer reading.
3084 */
3085 int lttng_ustconsumer_data_pending(struct lttng_consumer_stream *stream)
3086 {
3087 int ret;
3088
3089 assert(stream);
3090 assert(stream->ustream);
3091 ASSERT_LOCKED(stream->lock);
3092
3093 DBG("UST consumer checking data pending");
3094
3095 if (stream->endpoint_status != CONSUMER_ENDPOINT_ACTIVE) {
3096 ret = 0;
3097 goto end;
3098 }
3099
3100 if (stream->chan->type == CONSUMER_CHANNEL_TYPE_METADATA) {
3101 uint64_t contiguous, pushed;
3102
3103 /* Ease our life a bit. */
3104 pthread_mutex_lock(&stream->chan->metadata_cache->lock);
3105 contiguous = stream->chan->metadata_cache->max_offset;
3106 pthread_mutex_unlock(&stream->chan->metadata_cache->lock);
3107 pushed = stream->ust_metadata_pushed;
3108
3109 /*
3110 * We can simply check whether all contiguously available data
3111 * has been pushed to the ring buffer, since the push operation
3112 * is performed within get_next_subbuf(), and because both
3113 * get_next_subbuf() and put_next_subbuf() are issued atomically
3114 * thanks to the stream lock within
3115 * lttng_ustconsumer_read_subbuffer(). This basically means that
3116 * whetnever ust_metadata_pushed is incremented, the associated
3117 * metadata has been consumed from the metadata stream.
3118 */
3119 DBG("UST consumer metadata pending check: contiguous %" PRIu64 " vs pushed %" PRIu64,
3120 contiguous, pushed);
3121 assert(((int64_t) (contiguous - pushed)) >= 0);
3122 if ((contiguous != pushed) ||
3123 (((int64_t) contiguous - pushed) > 0 || contiguous == 0)) {
3124 ret = 1; /* Data is pending */
3125 goto end;
3126 }
3127 } else {
3128 ret = ustctl_get_next_subbuf(stream->ustream);
3129 if (ret == 0) {
3130 /*
3131 * There is still data so let's put back this
3132 * subbuffer.
3133 */
3134 ret = ustctl_put_subbuf(stream->ustream);
3135 assert(ret == 0);
3136 ret = 1; /* Data is pending */
3137 goto end;
3138 }
3139 }
3140
3141 /* Data is NOT pending so ready to be read. */
3142 ret = 0;
3143
3144 end:
3145 return ret;
3146 }
3147
3148 /*
3149 * Stop a given metadata channel timer if enabled and close the wait fd which
3150 * is the poll pipe of the metadata stream.
3151 *
3152 * This MUST be called with the metadata channel lock acquired.
3153 */
3154 void lttng_ustconsumer_close_metadata(struct lttng_consumer_channel *metadata)
3155 {
3156 int ret;
3157
3158 assert(metadata);
3159 assert(metadata->type == CONSUMER_CHANNEL_TYPE_METADATA);
3160
3161 DBG("Closing metadata channel key %" PRIu64, metadata->key);
3162
3163 if (metadata->switch_timer_enabled == 1) {
3164 consumer_timer_switch_stop(metadata);
3165 }
3166
3167 if (!metadata->metadata_stream) {
3168 goto end;
3169 }
3170
3171 /*
3172 * Closing write side so the thread monitoring the stream wakes up if any
3173 * and clean the metadata stream.
3174 */
3175 if (metadata->metadata_stream->ust_metadata_poll_pipe[1] >= 0) {
3176 ret = close(metadata->metadata_stream->ust_metadata_poll_pipe[1]);
3177 if (ret < 0) {
3178 PERROR("closing metadata pipe write side");
3179 }
3180 metadata->metadata_stream->ust_metadata_poll_pipe[1] = -1;
3181 }
3182
3183 end:
3184 return;
3185 }
3186
3187 /*
3188 * Close every metadata stream wait fd of the metadata hash table. This
3189 * function MUST be used very carefully so not to run into a race between the
3190 * metadata thread handling streams and this function closing their wait fd.
3191 *
3192 * For UST, this is used when the session daemon hangs up. Its the metadata
3193 * producer so calling this is safe because we are assured that no state change
3194 * can occur in the metadata thread for the streams in the hash table.
3195 */
3196 void lttng_ustconsumer_close_all_metadata(struct lttng_ht *metadata_ht)
3197 {
3198 struct lttng_ht_iter iter;
3199 struct lttng_consumer_stream *stream;
3200
3201 assert(metadata_ht);
3202 assert(metadata_ht->ht);
3203
3204 DBG("UST consumer closing all metadata streams");
3205
3206 rcu_read_lock();
3207 cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream,
3208 node.node) {
3209
3210 health_code_update();
3211
3212 pthread_mutex_lock(&stream->chan->lock);
3213 lttng_ustconsumer_close_metadata(stream->chan);
3214 pthread_mutex_unlock(&stream->chan->lock);
3215
3216 }
3217 rcu_read_unlock();
3218 }
3219
3220 void lttng_ustconsumer_close_stream_wakeup(struct lttng_consumer_stream *stream)
3221 {
3222 int ret;
3223
3224 ret = ustctl_stream_close_wakeup_fd(stream->ustream);
3225 if (ret < 0) {
3226 ERR("Unable to close wakeup fd");
3227 }
3228 }
3229
3230 /*
3231 * Please refer to consumer-timer.c before adding any lock within this
3232 * function or any of its callees. Timers have a very strict locking
3233 * semantic with respect to teardown. Failure to respect this semantic
3234 * introduces deadlocks.
3235 *
3236 * DON'T hold the metadata lock when calling this function, else this
3237 * can cause deadlock involving consumer awaiting for metadata to be
3238 * pushed out due to concurrent interaction with the session daemon.
3239 */
3240 int lttng_ustconsumer_request_metadata(struct lttng_consumer_local_data *ctx,
3241 struct lttng_consumer_channel *channel, int timer, int wait)
3242 {
3243 struct lttcomm_metadata_request_msg request;
3244 struct lttcomm_consumer_msg msg;
3245 enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS;
3246 uint64_t len, key, offset, version;
3247 int ret;
3248
3249 assert(channel);
3250 assert(channel->metadata_cache);
3251
3252 memset(&request, 0, sizeof(request));
3253
3254 /* send the metadata request to sessiond */
3255 switch (consumer_data.type) {
3256 case LTTNG_CONSUMER64_UST:
3257 request.bits_per_long = 64;
3258 break;
3259 case LTTNG_CONSUMER32_UST:
3260 request.bits_per_long = 32;
3261 break;
3262 default:
3263 request.bits_per_long = 0;
3264 break;
3265 }
3266
3267 request.session_id = channel->session_id;
3268 request.session_id_per_pid = channel->session_id_per_pid;
3269 /*
3270 * Request the application UID here so the metadata of that application can
3271 * be sent back. The channel UID corresponds to the user UID of the session
3272 * used for the rights on the stream file(s).
3273 */
3274 request.uid = channel->ust_app_uid;
3275 request.key = channel->key;
3276
3277 DBG("Sending metadata request to sessiond, session id %" PRIu64
3278 ", per-pid %" PRIu64 ", app UID %u and channel key %" PRIu64,
3279 request.session_id, request.session_id_per_pid, request.uid,
3280 request.key);
3281
3282 pthread_mutex_lock(&ctx->metadata_socket_lock);
3283
3284 health_code_update();
3285
3286 ret = lttcomm_send_unix_sock(ctx->consumer_metadata_socket, &request,
3287 sizeof(request));
3288 if (ret < 0) {
3289 ERR("Asking metadata to sessiond");
3290 goto end;
3291 }
3292
3293 health_code_update();
3294
3295 /* Receive the metadata from sessiond */
3296 ret = lttcomm_recv_unix_sock(ctx->consumer_metadata_socket, &msg,
3297 sizeof(msg));
3298 if (ret != sizeof(msg)) {
3299 DBG("Consumer received unexpected message size %d (expects %zu)",
3300 ret, sizeof(msg));
3301 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD);
3302 /*
3303 * The ret value might 0 meaning an orderly shutdown but this is ok
3304 * since the caller handles this.
3305 */
3306 goto end;
3307 }
3308
3309 health_code_update();
3310
3311 if (msg.cmd_type == LTTNG_ERR_UND) {
3312 /* No registry found */
3313 (void) consumer_send_status_msg(ctx->consumer_metadata_socket,
3314 ret_code);
3315 ret = 0;
3316 goto end;
3317 } else if (msg.cmd_type != LTTNG_CONSUMER_PUSH_METADATA) {
3318 ERR("Unexpected cmd_type received %d", msg.cmd_type);
3319 ret = -1;
3320 goto end;
3321 }
3322
3323 len = msg.u.push_metadata.len;
3324 key = msg.u.push_metadata.key;
3325 offset = msg.u.push_metadata.target_offset;
3326 version = msg.u.push_metadata.version;
3327
3328 assert(key == channel->key);
3329 if (len == 0) {
3330 DBG("No new metadata to receive for key %" PRIu64, key);
3331 }
3332
3333 health_code_update();
3334
3335 /* Tell session daemon we are ready to receive the metadata. */
3336 ret = consumer_send_status_msg(ctx->consumer_metadata_socket,
3337 LTTCOMM_CONSUMERD_SUCCESS);
3338 if (ret < 0 || len == 0) {
3339 /*
3340 * Somehow, the session daemon is not responding anymore or there is
3341 * nothing to receive.
3342 */
3343 goto end;
3344 }
3345
3346 health_code_update();
3347
3348 ret = lttng_ustconsumer_recv_metadata(ctx->consumer_metadata_socket,
3349 key, offset, len, version, channel, timer, wait);
3350 if (ret >= 0) {
3351 /*
3352 * Only send the status msg if the sessiond is alive meaning a positive
3353 * ret code.
3354 */
3355 (void) consumer_send_status_msg(ctx->consumer_metadata_socket, ret);
3356 }
3357 ret = 0;
3358
3359 end:
3360 health_code_update();
3361
3362 pthread_mutex_unlock(&ctx->metadata_socket_lock);
3363 return ret;
3364 }
3365
3366 /*
3367 * Return the ustctl call for the get stream id.
3368 */
3369 int lttng_ustconsumer_get_stream_id(struct lttng_consumer_stream *stream,
3370 uint64_t *stream_id)
3371 {
3372 assert(stream);
3373 assert(stream_id);
3374
3375 return ustctl_get_stream_id(stream->ustream, stream_id);
3376 }
This page took 0.092977 seconds and 3 git commands to generate.