Clean-up: ust-consumer: simplify metadata cache unlock on error path
[lttng-tools.git] / src / common / ust-consumer / ust-consumer.c
1 /*
2 * Copyright (C) 2011 Julien Desfossez <julien.desfossez@polymtl.ca>
3 * Copyright (C) 2011 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
4 * Copyright (C) 2017 Jérémie Galarneau <jeremie.galarneau@efficios.com>
5 *
6 * SPDX-License-Identifier: GPL-2.0-only
7 *
8 */
9
10 #define _LGPL_SOURCE
11 #include <assert.h>
12 #include <lttng/ust-ctl.h>
13 #include <poll.h>
14 #include <pthread.h>
15 #include <stdlib.h>
16 #include <string.h>
17 #include <sys/mman.h>
18 #include <sys/socket.h>
19 #include <sys/stat.h>
20 #include <sys/types.h>
21 #include <inttypes.h>
22 #include <unistd.h>
23 #include <urcu/list.h>
24 #include <signal.h>
25 #include <stdbool.h>
26 #include <stdint.h>
27
28 #include <bin/lttng-consumerd/health-consumerd.h>
29 #include <common/common.h>
30 #include <common/sessiond-comm/sessiond-comm.h>
31 #include <common/relayd/relayd.h>
32 #include <common/compat/fcntl.h>
33 #include <common/compat/endian.h>
34 #include <common/consumer/consumer-metadata-cache.h>
35 #include <common/consumer/consumer-stream.h>
36 #include <common/consumer/consumer-timer.h>
37 #include <common/utils.h>
38 #include <common/index/index.h>
39 #include <common/consumer/consumer.h>
40 #include <common/optional.h>
41
42 #include "ust-consumer.h"
43
44 #define INT_MAX_STR_LEN 12 /* includes \0 */
45
46 extern struct lttng_consumer_global_data consumer_data;
47 extern int consumer_poll_timeout;
48
49 /*
50 * Free channel object and all streams associated with it. This MUST be used
51 * only and only if the channel has _NEVER_ been added to the global channel
52 * hash table.
53 */
54 static void destroy_channel(struct lttng_consumer_channel *channel)
55 {
56 struct lttng_consumer_stream *stream, *stmp;
57
58 assert(channel);
59
60 DBG("UST consumer cleaning stream list");
61
62 cds_list_for_each_entry_safe(stream, stmp, &channel->streams.head,
63 send_node) {
64
65 health_code_update();
66
67 cds_list_del(&stream->send_node);
68 ustctl_destroy_stream(stream->ustream);
69 lttng_trace_chunk_put(stream->trace_chunk);
70 free(stream);
71 }
72
73 /*
74 * If a channel is available meaning that was created before the streams
75 * were, delete it.
76 */
77 if (channel->uchan) {
78 lttng_ustconsumer_del_channel(channel);
79 lttng_ustconsumer_free_channel(channel);
80 }
81
82 if (channel->trace_chunk) {
83 lttng_trace_chunk_put(channel->trace_chunk);
84 }
85
86 free(channel);
87 }
88
89 /*
90 * Add channel to internal consumer state.
91 *
92 * Returns 0 on success or else a negative value.
93 */
94 static int add_channel(struct lttng_consumer_channel *channel,
95 struct lttng_consumer_local_data *ctx)
96 {
97 int ret = 0;
98
99 assert(channel);
100 assert(ctx);
101
102 if (ctx->on_recv_channel != NULL) {
103 ret = ctx->on_recv_channel(channel);
104 if (ret == 0) {
105 ret = consumer_add_channel(channel, ctx);
106 } else if (ret < 0) {
107 /* Most likely an ENOMEM. */
108 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
109 goto error;
110 }
111 } else {
112 ret = consumer_add_channel(channel, ctx);
113 }
114
115 DBG("UST consumer channel added (key: %" PRIu64 ")", channel->key);
116
117 error:
118 return ret;
119 }
120
121 /*
122 * Allocate and return a consumer stream object. If _alloc_ret is not NULL, the
123 * error value if applicable is set in it else it is kept untouched.
124 *
125 * Return NULL on error else the newly allocated stream object.
126 */
127 static struct lttng_consumer_stream *allocate_stream(int cpu, int key,
128 struct lttng_consumer_channel *channel,
129 struct lttng_consumer_local_data *ctx, int *_alloc_ret)
130 {
131 int alloc_ret;
132 struct lttng_consumer_stream *stream = NULL;
133
134 assert(channel);
135 assert(ctx);
136
137 stream = consumer_stream_create(
138 channel,
139 channel->key,
140 key,
141 channel->name,
142 channel->relayd_id,
143 channel->session_id,
144 channel->trace_chunk,
145 cpu,
146 &alloc_ret,
147 channel->type,
148 channel->monitor);
149 if (stream == NULL) {
150 switch (alloc_ret) {
151 case -ENOENT:
152 /*
153 * We could not find the channel. Can happen if cpu hotplug
154 * happens while tearing down.
155 */
156 DBG3("Could not find channel");
157 break;
158 case -ENOMEM:
159 case -EINVAL:
160 default:
161 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
162 break;
163 }
164 goto error;
165 }
166
167 consumer_stream_update_channel_attributes(stream, channel);
168
169 error:
170 if (_alloc_ret) {
171 *_alloc_ret = alloc_ret;
172 }
173 return stream;
174 }
175
176 /*
177 * Send the given stream pointer to the corresponding thread.
178 *
179 * Returns 0 on success else a negative value.
180 */
181 static int send_stream_to_thread(struct lttng_consumer_stream *stream,
182 struct lttng_consumer_local_data *ctx)
183 {
184 int ret;
185 struct lttng_pipe *stream_pipe;
186
187 /* Get the right pipe where the stream will be sent. */
188 if (stream->metadata_flag) {
189 consumer_add_metadata_stream(stream);
190 stream_pipe = ctx->consumer_metadata_pipe;
191 } else {
192 consumer_add_data_stream(stream);
193 stream_pipe = ctx->consumer_data_pipe;
194 }
195
196 /*
197 * From this point on, the stream's ownership has been moved away from
198 * the channel and it becomes globally visible. Hence, remove it from
199 * the local stream list to prevent the stream from being both local and
200 * global.
201 */
202 stream->globally_visible = 1;
203 cds_list_del(&stream->send_node);
204
205 ret = lttng_pipe_write(stream_pipe, &stream, sizeof(stream));
206 if (ret < 0) {
207 ERR("Consumer write %s stream to pipe %d",
208 stream->metadata_flag ? "metadata" : "data",
209 lttng_pipe_get_writefd(stream_pipe));
210 if (stream->metadata_flag) {
211 consumer_del_stream_for_metadata(stream);
212 } else {
213 consumer_del_stream_for_data(stream);
214 }
215 goto error;
216 }
217
218 error:
219 return ret;
220 }
221
222 static
223 int get_stream_shm_path(char *stream_shm_path, const char *shm_path, int cpu)
224 {
225 char cpu_nr[INT_MAX_STR_LEN]; /* int max len */
226 int ret;
227
228 strncpy(stream_shm_path, shm_path, PATH_MAX);
229 stream_shm_path[PATH_MAX - 1] = '\0';
230 ret = snprintf(cpu_nr, INT_MAX_STR_LEN, "%i", cpu);
231 if (ret < 0) {
232 PERROR("snprintf");
233 goto end;
234 }
235 strncat(stream_shm_path, cpu_nr,
236 PATH_MAX - strlen(stream_shm_path) - 1);
237 ret = 0;
238 end:
239 return ret;
240 }
241
242 /*
243 * Create streams for the given channel using liblttng-ust-ctl.
244 * The channel lock must be acquired by the caller.
245 *
246 * Return 0 on success else a negative value.
247 */
248 static int create_ust_streams(struct lttng_consumer_channel *channel,
249 struct lttng_consumer_local_data *ctx)
250 {
251 int ret, cpu = 0;
252 struct ustctl_consumer_stream *ustream;
253 struct lttng_consumer_stream *stream;
254 pthread_mutex_t *current_stream_lock = NULL;
255
256 assert(channel);
257 assert(ctx);
258
259 /*
260 * While a stream is available from ustctl. When NULL is returned, we've
261 * reached the end of the possible stream for the channel.
262 */
263 while ((ustream = ustctl_create_stream(channel->uchan, cpu))) {
264 int wait_fd;
265 int ust_metadata_pipe[2];
266
267 health_code_update();
268
269 if (channel->type == CONSUMER_CHANNEL_TYPE_METADATA && channel->monitor) {
270 ret = utils_create_pipe_cloexec_nonblock(ust_metadata_pipe);
271 if (ret < 0) {
272 ERR("Create ust metadata poll pipe");
273 goto error;
274 }
275 wait_fd = ust_metadata_pipe[0];
276 } else {
277 wait_fd = ustctl_stream_get_wait_fd(ustream);
278 }
279
280 /* Allocate consumer stream object. */
281 stream = allocate_stream(cpu, wait_fd, channel, ctx, &ret);
282 if (!stream) {
283 goto error_alloc;
284 }
285 stream->ustream = ustream;
286 /*
287 * Store it so we can save multiple function calls afterwards since
288 * this value is used heavily in the stream threads. This is UST
289 * specific so this is why it's done after allocation.
290 */
291 stream->wait_fd = wait_fd;
292
293 /*
294 * Increment channel refcount since the channel reference has now been
295 * assigned in the allocation process above.
296 */
297 if (stream->chan->monitor) {
298 uatomic_inc(&stream->chan->refcount);
299 }
300
301 pthread_mutex_lock(&stream->lock);
302 current_stream_lock = &stream->lock;
303 /*
304 * Order is important this is why a list is used. On error, the caller
305 * should clean this list.
306 */
307 cds_list_add_tail(&stream->send_node, &channel->streams.head);
308
309 ret = ustctl_get_max_subbuf_size(stream->ustream,
310 &stream->max_sb_size);
311 if (ret < 0) {
312 ERR("ustctl_get_max_subbuf_size failed for stream %s",
313 stream->name);
314 goto error;
315 }
316
317 /* Do actions once stream has been received. */
318 if (ctx->on_recv_stream) {
319 ret = ctx->on_recv_stream(stream);
320 if (ret < 0) {
321 goto error;
322 }
323 }
324
325 DBG("UST consumer add stream %s (key: %" PRIu64 ") with relayd id %" PRIu64,
326 stream->name, stream->key, stream->relayd_stream_id);
327
328 /* Set next CPU stream. */
329 channel->streams.count = ++cpu;
330
331 /* Keep stream reference when creating metadata. */
332 if (channel->type == CONSUMER_CHANNEL_TYPE_METADATA) {
333 channel->metadata_stream = stream;
334 if (channel->monitor) {
335 /* Set metadata poll pipe if we created one */
336 memcpy(stream->ust_metadata_poll_pipe,
337 ust_metadata_pipe,
338 sizeof(ust_metadata_pipe));
339 }
340 }
341 pthread_mutex_unlock(&stream->lock);
342 current_stream_lock = NULL;
343 }
344
345 return 0;
346
347 error:
348 error_alloc:
349 if (current_stream_lock) {
350 pthread_mutex_unlock(current_stream_lock);
351 }
352 return ret;
353 }
354
355 /*
356 * create_posix_shm is never called concurrently within a process.
357 */
358 static
359 int create_posix_shm(void)
360 {
361 char tmp_name[NAME_MAX];
362 int shmfd, ret;
363
364 ret = snprintf(tmp_name, NAME_MAX, "/ust-shm-consumer-%d", getpid());
365 if (ret < 0) {
366 PERROR("snprintf");
367 return -1;
368 }
369 /*
370 * Allocate shm, and immediately unlink its shm oject, keeping
371 * only the file descriptor as a reference to the object.
372 * We specifically do _not_ use the / at the beginning of the
373 * pathname so that some OS implementations can keep it local to
374 * the process (POSIX leaves this implementation-defined).
375 */
376 shmfd = shm_open(tmp_name, O_CREAT | O_EXCL | O_RDWR, 0700);
377 if (shmfd < 0) {
378 PERROR("shm_open");
379 goto error_shm_open;
380 }
381 ret = shm_unlink(tmp_name);
382 if (ret < 0 && errno != ENOENT) {
383 PERROR("shm_unlink");
384 goto error_shm_release;
385 }
386 return shmfd;
387
388 error_shm_release:
389 ret = close(shmfd);
390 if (ret) {
391 PERROR("close");
392 }
393 error_shm_open:
394 return -1;
395 }
396
397 static int open_ust_stream_fd(struct lttng_consumer_channel *channel, int cpu,
398 const struct lttng_credentials *session_credentials)
399 {
400 char shm_path[PATH_MAX];
401 int ret;
402
403 if (!channel->shm_path[0]) {
404 return create_posix_shm();
405 }
406 ret = get_stream_shm_path(shm_path, channel->shm_path, cpu);
407 if (ret) {
408 goto error_shm_path;
409 }
410 return run_as_open(shm_path,
411 O_RDWR | O_CREAT | O_EXCL, S_IRUSR | S_IWUSR,
412 lttng_credentials_get_uid(session_credentials),
413 lttng_credentials_get_gid(session_credentials));
414
415 error_shm_path:
416 return -1;
417 }
418
419 /*
420 * Create an UST channel with the given attributes and send it to the session
421 * daemon using the ust ctl API.
422 *
423 * Return 0 on success or else a negative value.
424 */
425 static int create_ust_channel(struct lttng_consumer_channel *channel,
426 struct ustctl_consumer_channel_attr *attr,
427 struct ustctl_consumer_channel **ust_chanp)
428 {
429 int ret, nr_stream_fds, i, j;
430 int *stream_fds;
431 struct ustctl_consumer_channel *ust_channel;
432
433 assert(channel);
434 assert(attr);
435 assert(ust_chanp);
436 assert(channel->buffer_credentials.is_set);
437
438 DBG3("Creating channel to ustctl with attr: [overwrite: %d, "
439 "subbuf_size: %" PRIu64 ", num_subbuf: %" PRIu64 ", "
440 "switch_timer_interval: %u, read_timer_interval: %u, "
441 "output: %d, type: %d", attr->overwrite, attr->subbuf_size,
442 attr->num_subbuf, attr->switch_timer_interval,
443 attr->read_timer_interval, attr->output, attr->type);
444
445 if (channel->type == CONSUMER_CHANNEL_TYPE_METADATA)
446 nr_stream_fds = 1;
447 else
448 nr_stream_fds = ustctl_get_nr_stream_per_channel();
449 stream_fds = zmalloc(nr_stream_fds * sizeof(*stream_fds));
450 if (!stream_fds) {
451 ret = -1;
452 goto error_alloc;
453 }
454 for (i = 0; i < nr_stream_fds; i++) {
455 stream_fds[i] = open_ust_stream_fd(channel, i,
456 &channel->buffer_credentials.value);
457 if (stream_fds[i] < 0) {
458 ret = -1;
459 goto error_open;
460 }
461 }
462 ust_channel = ustctl_create_channel(attr, stream_fds, nr_stream_fds);
463 if (!ust_channel) {
464 ret = -1;
465 goto error_create;
466 }
467 channel->nr_stream_fds = nr_stream_fds;
468 channel->stream_fds = stream_fds;
469 *ust_chanp = ust_channel;
470
471 return 0;
472
473 error_create:
474 error_open:
475 for (j = i - 1; j >= 0; j--) {
476 int closeret;
477
478 closeret = close(stream_fds[j]);
479 if (closeret) {
480 PERROR("close");
481 }
482 if (channel->shm_path[0]) {
483 char shm_path[PATH_MAX];
484
485 closeret = get_stream_shm_path(shm_path,
486 channel->shm_path, j);
487 if (closeret) {
488 ERR("Cannot get stream shm path");
489 }
490 closeret = run_as_unlink(shm_path,
491 lttng_credentials_get_uid(LTTNG_OPTIONAL_GET_PTR(
492 channel->buffer_credentials)),
493 lttng_credentials_get_gid(LTTNG_OPTIONAL_GET_PTR(
494 channel->buffer_credentials)));
495 if (closeret) {
496 PERROR("unlink %s", shm_path);
497 }
498 }
499 }
500 /* Try to rmdir all directories under shm_path root. */
501 if (channel->root_shm_path[0]) {
502 (void) run_as_rmdir_recursive(channel->root_shm_path,
503 lttng_credentials_get_uid(LTTNG_OPTIONAL_GET_PTR(
504 channel->buffer_credentials)),
505 lttng_credentials_get_gid(LTTNG_OPTIONAL_GET_PTR(
506 channel->buffer_credentials)),
507 LTTNG_DIRECTORY_HANDLE_SKIP_NON_EMPTY_FLAG);
508 }
509 free(stream_fds);
510 error_alloc:
511 return ret;
512 }
513
514 /*
515 * Send a single given stream to the session daemon using the sock.
516 *
517 * Return 0 on success else a negative value.
518 */
519 static int send_sessiond_stream(int sock, struct lttng_consumer_stream *stream)
520 {
521 int ret;
522
523 assert(stream);
524 assert(sock >= 0);
525
526 DBG("UST consumer sending stream %" PRIu64 " to sessiond", stream->key);
527
528 /* Send stream to session daemon. */
529 ret = ustctl_send_stream_to_sessiond(sock, stream->ustream);
530 if (ret < 0) {
531 goto error;
532 }
533
534 error:
535 return ret;
536 }
537
538 /*
539 * Send channel to sessiond and relayd if applicable.
540 *
541 * Return 0 on success or else a negative value.
542 */
543 static int send_channel_to_sessiond_and_relayd(int sock,
544 struct lttng_consumer_channel *channel,
545 struct lttng_consumer_local_data *ctx, int *relayd_error)
546 {
547 int ret, ret_code = LTTCOMM_CONSUMERD_SUCCESS;
548 struct lttng_consumer_stream *stream;
549 uint64_t net_seq_idx = -1ULL;
550
551 assert(channel);
552 assert(ctx);
553 assert(sock >= 0);
554
555 DBG("UST consumer sending channel %s to sessiond", channel->name);
556
557 if (channel->relayd_id != (uint64_t) -1ULL) {
558 cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
559
560 health_code_update();
561
562 /* Try to send the stream to the relayd if one is available. */
563 DBG("Sending stream %" PRIu64 " of channel \"%s\" to relayd",
564 stream->key, channel->name);
565 ret = consumer_send_relayd_stream(stream, stream->chan->pathname);
566 if (ret < 0) {
567 /*
568 * Flag that the relayd was the problem here probably due to a
569 * communicaton error on the socket.
570 */
571 if (relayd_error) {
572 *relayd_error = 1;
573 }
574 ret_code = LTTCOMM_CONSUMERD_RELAYD_FAIL;
575 }
576 if (net_seq_idx == -1ULL) {
577 net_seq_idx = stream->net_seq_idx;
578 }
579 }
580 }
581
582 /* Inform sessiond that we are about to send channel and streams. */
583 ret = consumer_send_status_msg(sock, ret_code);
584 if (ret < 0 || ret_code != LTTCOMM_CONSUMERD_SUCCESS) {
585 /*
586 * Either the session daemon is not responding or the relayd died so we
587 * stop now.
588 */
589 goto error;
590 }
591
592 /* Send channel to sessiond. */
593 ret = ustctl_send_channel_to_sessiond(sock, channel->uchan);
594 if (ret < 0) {
595 goto error;
596 }
597
598 ret = ustctl_channel_close_wakeup_fd(channel->uchan);
599 if (ret < 0) {
600 goto error;
601 }
602
603 /* The channel was sent successfully to the sessiond at this point. */
604 cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
605
606 health_code_update();
607
608 /* Send stream to session daemon. */
609 ret = send_sessiond_stream(sock, stream);
610 if (ret < 0) {
611 goto error;
612 }
613 }
614
615 /* Tell sessiond there is no more stream. */
616 ret = ustctl_send_stream_to_sessiond(sock, NULL);
617 if (ret < 0) {
618 goto error;
619 }
620
621 DBG("UST consumer NULL stream sent to sessiond");
622
623 return 0;
624
625 error:
626 if (ret_code != LTTCOMM_CONSUMERD_SUCCESS) {
627 ret = -1;
628 }
629 return ret;
630 }
631
632 /*
633 * Creates a channel and streams and add the channel it to the channel internal
634 * state. The created stream must ONLY be sent once the GET_CHANNEL command is
635 * received.
636 *
637 * Return 0 on success or else, a negative value is returned and the channel
638 * MUST be destroyed by consumer_del_channel().
639 */
640 static int ask_channel(struct lttng_consumer_local_data *ctx,
641 struct lttng_consumer_channel *channel,
642 struct ustctl_consumer_channel_attr *attr)
643 {
644 int ret;
645
646 assert(ctx);
647 assert(channel);
648 assert(attr);
649
650 /*
651 * This value is still used by the kernel consumer since for the kernel,
652 * the stream ownership is not IN the consumer so we need to have the
653 * number of left stream that needs to be initialized so we can know when
654 * to delete the channel (see consumer.c).
655 *
656 * As for the user space tracer now, the consumer creates and sends the
657 * stream to the session daemon which only sends them to the application
658 * once every stream of a channel is received making this value useless
659 * because we they will be added to the poll thread before the application
660 * receives them. This ensures that a stream can not hang up during
661 * initilization of a channel.
662 */
663 channel->nb_init_stream_left = 0;
664
665 /* The reply msg status is handled in the following call. */
666 ret = create_ust_channel(channel, attr, &channel->uchan);
667 if (ret < 0) {
668 goto end;
669 }
670
671 channel->wait_fd = ustctl_channel_get_wait_fd(channel->uchan);
672
673 /*
674 * For the snapshots (no monitor), we create the metadata streams
675 * on demand, not during the channel creation.
676 */
677 if (channel->type == CONSUMER_CHANNEL_TYPE_METADATA && !channel->monitor) {
678 ret = 0;
679 goto end;
680 }
681
682 /* Open all streams for this channel. */
683 pthread_mutex_lock(&channel->lock);
684 ret = create_ust_streams(channel, ctx);
685 pthread_mutex_unlock(&channel->lock);
686 if (ret < 0) {
687 goto end;
688 }
689
690 end:
691 return ret;
692 }
693
694 /*
695 * Send all stream of a channel to the right thread handling it.
696 *
697 * On error, return a negative value else 0 on success.
698 */
699 static int send_streams_to_thread(struct lttng_consumer_channel *channel,
700 struct lttng_consumer_local_data *ctx)
701 {
702 int ret = 0;
703 struct lttng_consumer_stream *stream, *stmp;
704
705 assert(channel);
706 assert(ctx);
707
708 /* Send streams to the corresponding thread. */
709 cds_list_for_each_entry_safe(stream, stmp, &channel->streams.head,
710 send_node) {
711
712 health_code_update();
713
714 /* Sending the stream to the thread. */
715 ret = send_stream_to_thread(stream, ctx);
716 if (ret < 0) {
717 /*
718 * If we are unable to send the stream to the thread, there is
719 * a big problem so just stop everything.
720 */
721 goto error;
722 }
723 }
724
725 error:
726 return ret;
727 }
728
729 /*
730 * Flush channel's streams using the given key to retrieve the channel.
731 *
732 * Return 0 on success else an LTTng error code.
733 */
734 static int flush_channel(uint64_t chan_key)
735 {
736 int ret = 0;
737 struct lttng_consumer_channel *channel;
738 struct lttng_consumer_stream *stream;
739 struct lttng_ht *ht;
740 struct lttng_ht_iter iter;
741
742 DBG("UST consumer flush channel key %" PRIu64, chan_key);
743
744 rcu_read_lock();
745 channel = consumer_find_channel(chan_key);
746 if (!channel) {
747 ERR("UST consumer flush channel %" PRIu64 " not found", chan_key);
748 ret = LTTNG_ERR_UST_CHAN_NOT_FOUND;
749 goto error;
750 }
751
752 ht = consumer_data.stream_per_chan_id_ht;
753
754 /* For each stream of the channel id, flush it. */
755 cds_lfht_for_each_entry_duplicate(ht->ht,
756 ht->hash_fct(&channel->key, lttng_ht_seed), ht->match_fct,
757 &channel->key, &iter.iter, stream, node_channel_id.node) {
758
759 health_code_update();
760
761 pthread_mutex_lock(&stream->lock);
762
763 /*
764 * Protect against concurrent teardown of a stream.
765 */
766 if (cds_lfht_is_node_deleted(&stream->node.node)) {
767 goto next;
768 }
769
770 if (!stream->quiescent) {
771 ustctl_flush_buffer(stream->ustream, 0);
772 stream->quiescent = true;
773 }
774 next:
775 pthread_mutex_unlock(&stream->lock);
776 }
777 error:
778 rcu_read_unlock();
779 return ret;
780 }
781
782 /*
783 * Clear quiescent state from channel's streams using the given key to
784 * retrieve the channel.
785 *
786 * Return 0 on success else an LTTng error code.
787 */
788 static int clear_quiescent_channel(uint64_t chan_key)
789 {
790 int ret = 0;
791 struct lttng_consumer_channel *channel;
792 struct lttng_consumer_stream *stream;
793 struct lttng_ht *ht;
794 struct lttng_ht_iter iter;
795
796 DBG("UST consumer clear quiescent channel key %" PRIu64, chan_key);
797
798 rcu_read_lock();
799 channel = consumer_find_channel(chan_key);
800 if (!channel) {
801 ERR("UST consumer clear quiescent channel %" PRIu64 " not found", chan_key);
802 ret = LTTNG_ERR_UST_CHAN_NOT_FOUND;
803 goto error;
804 }
805
806 ht = consumer_data.stream_per_chan_id_ht;
807
808 /* For each stream of the channel id, clear quiescent state. */
809 cds_lfht_for_each_entry_duplicate(ht->ht,
810 ht->hash_fct(&channel->key, lttng_ht_seed), ht->match_fct,
811 &channel->key, &iter.iter, stream, node_channel_id.node) {
812
813 health_code_update();
814
815 pthread_mutex_lock(&stream->lock);
816 stream->quiescent = false;
817 pthread_mutex_unlock(&stream->lock);
818 }
819 error:
820 rcu_read_unlock();
821 return ret;
822 }
823
824 /*
825 * Close metadata stream wakeup_fd using the given key to retrieve the channel.
826 *
827 * Return 0 on success else an LTTng error code.
828 */
829 static int close_metadata(uint64_t chan_key)
830 {
831 int ret = 0;
832 struct lttng_consumer_channel *channel;
833 unsigned int channel_monitor;
834
835 DBG("UST consumer close metadata key %" PRIu64, chan_key);
836
837 channel = consumer_find_channel(chan_key);
838 if (!channel) {
839 /*
840 * This is possible if the metadata thread has issue a delete because
841 * the endpoint point of the stream hung up. There is no way the
842 * session daemon can know about it thus use a DBG instead of an actual
843 * error.
844 */
845 DBG("UST consumer close metadata %" PRIu64 " not found", chan_key);
846 ret = LTTNG_ERR_UST_CHAN_NOT_FOUND;
847 goto error;
848 }
849
850 pthread_mutex_lock(&consumer_data.lock);
851 pthread_mutex_lock(&channel->lock);
852 channel_monitor = channel->monitor;
853 if (cds_lfht_is_node_deleted(&channel->node.node)) {
854 goto error_unlock;
855 }
856
857 lttng_ustconsumer_close_metadata(channel);
858 pthread_mutex_unlock(&channel->lock);
859 pthread_mutex_unlock(&consumer_data.lock);
860
861 /*
862 * The ownership of a metadata channel depends on the type of
863 * session to which it belongs. In effect, the monitor flag is checked
864 * to determine if this metadata channel is in "snapshot" mode or not.
865 *
866 * In the non-snapshot case, the metadata channel is created along with
867 * a single stream which will remain present until the metadata channel
868 * is destroyed (on the destruction of its session). In this case, the
869 * metadata stream in "monitored" by the metadata poll thread and holds
870 * the ownership of its channel.
871 *
872 * Closing the metadata will cause the metadata stream's "metadata poll
873 * pipe" to be closed. Closing this pipe will wake-up the metadata poll
874 * thread which will teardown the metadata stream which, in return,
875 * deletes the metadata channel.
876 *
877 * In the snapshot case, the metadata stream is created and destroyed
878 * on every snapshot record. Since the channel doesn't have an owner
879 * other than the session daemon, it is safe to destroy it immediately
880 * on reception of the CLOSE_METADATA command.
881 */
882 if (!channel_monitor) {
883 /*
884 * The channel and consumer_data locks must be
885 * released before this call since consumer_del_channel
886 * re-acquires the channel and consumer_data locks to teardown
887 * the channel and queue its reclamation by the "call_rcu"
888 * worker thread.
889 */
890 consumer_del_channel(channel);
891 }
892
893 return ret;
894 error_unlock:
895 pthread_mutex_unlock(&channel->lock);
896 pthread_mutex_unlock(&consumer_data.lock);
897 error:
898 return ret;
899 }
900
901 /*
902 * RCU read side lock MUST be acquired before calling this function.
903 *
904 * Return 0 on success else an LTTng error code.
905 */
906 static int setup_metadata(struct lttng_consumer_local_data *ctx, uint64_t key)
907 {
908 int ret;
909 struct lttng_consumer_channel *metadata;
910
911 DBG("UST consumer setup metadata key %" PRIu64, key);
912
913 metadata = consumer_find_channel(key);
914 if (!metadata) {
915 ERR("UST consumer push metadata %" PRIu64 " not found", key);
916 ret = LTTNG_ERR_UST_CHAN_NOT_FOUND;
917 goto end;
918 }
919
920 /*
921 * In no monitor mode, the metadata channel has no stream(s) so skip the
922 * ownership transfer to the metadata thread.
923 */
924 if (!metadata->monitor) {
925 DBG("Metadata channel in no monitor");
926 ret = 0;
927 goto end;
928 }
929
930 /*
931 * Send metadata stream to relayd if one available. Availability is
932 * known if the stream is still in the list of the channel.
933 */
934 if (cds_list_empty(&metadata->streams.head)) {
935 ERR("Metadata channel key %" PRIu64 ", no stream available.", key);
936 ret = LTTCOMM_CONSUMERD_ERROR_METADATA;
937 goto error_no_stream;
938 }
939
940 /* Send metadata stream to relayd if needed. */
941 if (metadata->metadata_stream->net_seq_idx != (uint64_t) -1ULL) {
942 ret = consumer_send_relayd_stream(metadata->metadata_stream,
943 metadata->pathname);
944 if (ret < 0) {
945 ret = LTTCOMM_CONSUMERD_ERROR_METADATA;
946 goto error;
947 }
948 ret = consumer_send_relayd_streams_sent(
949 metadata->metadata_stream->net_seq_idx);
950 if (ret < 0) {
951 ret = LTTCOMM_CONSUMERD_RELAYD_FAIL;
952 goto error;
953 }
954 }
955
956 /*
957 * Ownership of metadata stream is passed along. Freeing is handled by
958 * the callee.
959 */
960 ret = send_streams_to_thread(metadata, ctx);
961 if (ret < 0) {
962 /*
963 * If we are unable to send the stream to the thread, there is
964 * a big problem so just stop everything.
965 */
966 ret = LTTCOMM_CONSUMERD_FATAL;
967 goto send_streams_error;
968 }
969 /* List MUST be empty after or else it could be reused. */
970 assert(cds_list_empty(&metadata->streams.head));
971
972 ret = 0;
973 goto end;
974
975 error:
976 /*
977 * Delete metadata channel on error. At this point, the metadata stream can
978 * NOT be monitored by the metadata thread thus having the guarantee that
979 * the stream is still in the local stream list of the channel. This call
980 * will make sure to clean that list.
981 */
982 consumer_stream_destroy(metadata->metadata_stream, NULL);
983 cds_list_del(&metadata->metadata_stream->send_node);
984 metadata->metadata_stream = NULL;
985 send_streams_error:
986 error_no_stream:
987 end:
988 return ret;
989 }
990
991 /*
992 * Snapshot the whole metadata.
993 * RCU read-side lock must be held by the caller.
994 *
995 * Returns 0 on success, < 0 on error
996 */
997 static int snapshot_metadata(struct lttng_consumer_channel *metadata_channel,
998 uint64_t key, char *path, uint64_t relayd_id,
999 struct lttng_consumer_local_data *ctx)
1000 {
1001 int ret = 0;
1002 struct lttng_consumer_stream *metadata_stream;
1003
1004 assert(path);
1005 assert(ctx);
1006
1007 DBG("UST consumer snapshot metadata with key %" PRIu64 " at path %s",
1008 key, path);
1009
1010 rcu_read_lock();
1011
1012 assert(!metadata_channel->monitor);
1013
1014 health_code_update();
1015
1016 /*
1017 * Ask the sessiond if we have new metadata waiting and update the
1018 * consumer metadata cache.
1019 */
1020 ret = lttng_ustconsumer_request_metadata(ctx, metadata_channel, 0, 1);
1021 if (ret < 0) {
1022 goto error;
1023 }
1024
1025 health_code_update();
1026
1027 /*
1028 * The metadata stream is NOT created in no monitor mode when the channel
1029 * is created on a sessiond ask channel command.
1030 */
1031 ret = create_ust_streams(metadata_channel, ctx);
1032 if (ret < 0) {
1033 goto error;
1034 }
1035
1036 metadata_stream = metadata_channel->metadata_stream;
1037 assert(metadata_stream);
1038
1039 pthread_mutex_lock(&metadata_stream->lock);
1040 if (relayd_id != (uint64_t) -1ULL) {
1041 metadata_stream->net_seq_idx = relayd_id;
1042 ret = consumer_send_relayd_stream(metadata_stream, path);
1043 } else {
1044 ret = consumer_stream_create_output_files(metadata_stream,
1045 false);
1046 }
1047 pthread_mutex_unlock(&metadata_stream->lock);
1048 if (ret < 0) {
1049 goto error_stream;
1050 }
1051
1052 do {
1053 health_code_update();
1054
1055 ret = lttng_consumer_read_subbuffer(metadata_stream, ctx, true);
1056 if (ret < 0) {
1057 goto error_stream;
1058 }
1059 } while (ret > 0);
1060
1061 error_stream:
1062 /*
1063 * Clean up the stream completly because the next snapshot will use a new
1064 * metadata stream.
1065 */
1066 consumer_stream_destroy(metadata_stream, NULL);
1067 cds_list_del(&metadata_stream->send_node);
1068 metadata_channel->metadata_stream = NULL;
1069
1070 error:
1071 rcu_read_unlock();
1072 return ret;
1073 }
1074
1075 static
1076 int get_current_subbuf_addr(struct lttng_consumer_stream *stream,
1077 const char **addr)
1078 {
1079 int ret;
1080 unsigned long mmap_offset;
1081 const char *mmap_base;
1082
1083 mmap_base = ustctl_get_mmap_base(stream->ustream);
1084 if (!mmap_base) {
1085 ERR("Failed to get mmap base for stream `%s`",
1086 stream->name);
1087 ret = -EPERM;
1088 goto error;
1089 }
1090
1091 ret = ustctl_get_mmap_read_offset(stream->ustream, &mmap_offset);
1092 if (ret != 0) {
1093 ERR("Failed to get mmap offset for stream `%s`", stream->name);
1094 ret = -EINVAL;
1095 goto error;
1096 }
1097
1098 *addr = mmap_base + mmap_offset;
1099 error:
1100 return ret;
1101
1102 }
1103
1104 /*
1105 * Take a snapshot of all the stream of a channel.
1106 * RCU read-side lock and the channel lock must be held by the caller.
1107 *
1108 * Returns 0 on success, < 0 on error
1109 */
1110 static int snapshot_channel(struct lttng_consumer_channel *channel,
1111 uint64_t key, char *path, uint64_t relayd_id,
1112 uint64_t nb_packets_per_stream,
1113 struct lttng_consumer_local_data *ctx)
1114 {
1115 int ret;
1116 unsigned use_relayd = 0;
1117 unsigned long consumed_pos, produced_pos;
1118 struct lttng_consumer_stream *stream;
1119
1120 assert(path);
1121 assert(ctx);
1122
1123 rcu_read_lock();
1124
1125 if (relayd_id != (uint64_t) -1ULL) {
1126 use_relayd = 1;
1127 }
1128
1129 assert(!channel->monitor);
1130 DBG("UST consumer snapshot channel %" PRIu64, key);
1131
1132 cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
1133 health_code_update();
1134
1135 /* Lock stream because we are about to change its state. */
1136 pthread_mutex_lock(&stream->lock);
1137 assert(channel->trace_chunk);
1138 if (!lttng_trace_chunk_get(channel->trace_chunk)) {
1139 /*
1140 * Can't happen barring an internal error as the channel
1141 * holds a reference to the trace chunk.
1142 */
1143 ERR("Failed to acquire reference to channel's trace chunk");
1144 ret = -1;
1145 goto error_unlock;
1146 }
1147 assert(!stream->trace_chunk);
1148 stream->trace_chunk = channel->trace_chunk;
1149
1150 stream->net_seq_idx = relayd_id;
1151
1152 if (use_relayd) {
1153 ret = consumer_send_relayd_stream(stream, path);
1154 if (ret < 0) {
1155 goto error_unlock;
1156 }
1157 } else {
1158 ret = consumer_stream_create_output_files(stream,
1159 false);
1160 if (ret < 0) {
1161 goto error_unlock;
1162 }
1163 DBG("UST consumer snapshot stream (%" PRIu64 ")",
1164 stream->key);
1165 }
1166
1167 /*
1168 * If tracing is active, we want to perform a "full" buffer flush.
1169 * Else, if quiescent, it has already been done by the prior stop.
1170 */
1171 if (!stream->quiescent) {
1172 ustctl_flush_buffer(stream->ustream, 0);
1173 }
1174
1175 ret = lttng_ustconsumer_take_snapshot(stream);
1176 if (ret < 0) {
1177 ERR("Taking UST snapshot");
1178 goto error_unlock;
1179 }
1180
1181 ret = lttng_ustconsumer_get_produced_snapshot(stream, &produced_pos);
1182 if (ret < 0) {
1183 ERR("Produced UST snapshot position");
1184 goto error_unlock;
1185 }
1186
1187 ret = lttng_ustconsumer_get_consumed_snapshot(stream, &consumed_pos);
1188 if (ret < 0) {
1189 ERR("Consumerd UST snapshot position");
1190 goto error_unlock;
1191 }
1192
1193 /*
1194 * The original value is sent back if max stream size is larger than
1195 * the possible size of the snapshot. Also, we assume that the session
1196 * daemon should never send a maximum stream size that is lower than
1197 * subbuffer size.
1198 */
1199 consumed_pos = consumer_get_consume_start_pos(consumed_pos,
1200 produced_pos, nb_packets_per_stream,
1201 stream->max_sb_size);
1202
1203 while ((long) (consumed_pos - produced_pos) < 0) {
1204 ssize_t read_len;
1205 unsigned long len, padded_len;
1206 const char *subbuf_addr;
1207 struct lttng_buffer_view subbuf_view;
1208
1209 health_code_update();
1210
1211 DBG("UST consumer taking snapshot at pos %lu", consumed_pos);
1212
1213 ret = ustctl_get_subbuf(stream->ustream, &consumed_pos);
1214 if (ret < 0) {
1215 if (ret != -EAGAIN) {
1216 PERROR("ustctl_get_subbuf snapshot");
1217 goto error_close_stream;
1218 }
1219 DBG("UST consumer get subbuf failed. Skipping it.");
1220 consumed_pos += stream->max_sb_size;
1221 stream->chan->lost_packets++;
1222 continue;
1223 }
1224
1225 ret = ustctl_get_subbuf_size(stream->ustream, &len);
1226 if (ret < 0) {
1227 ERR("Snapshot ustctl_get_subbuf_size");
1228 goto error_put_subbuf;
1229 }
1230
1231 ret = ustctl_get_padded_subbuf_size(stream->ustream, &padded_len);
1232 if (ret < 0) {
1233 ERR("Snapshot ustctl_get_padded_subbuf_size");
1234 goto error_put_subbuf;
1235 }
1236
1237 ret = get_current_subbuf_addr(stream, &subbuf_addr);
1238 if (ret) {
1239 goto error_put_subbuf;
1240 }
1241
1242 subbuf_view = lttng_buffer_view_init(
1243 subbuf_addr, 0, padded_len);
1244 read_len = lttng_consumer_on_read_subbuffer_mmap(
1245 stream, &subbuf_view, padded_len - len);
1246 if (use_relayd) {
1247 if (read_len != len) {
1248 ret = -EPERM;
1249 goto error_put_subbuf;
1250 }
1251 } else {
1252 if (read_len != padded_len) {
1253 ret = -EPERM;
1254 goto error_put_subbuf;
1255 }
1256 }
1257
1258 ret = ustctl_put_subbuf(stream->ustream);
1259 if (ret < 0) {
1260 ERR("Snapshot ustctl_put_subbuf");
1261 goto error_close_stream;
1262 }
1263 consumed_pos += stream->max_sb_size;
1264 }
1265
1266 /* Simply close the stream so we can use it on the next snapshot. */
1267 consumer_stream_close(stream);
1268 pthread_mutex_unlock(&stream->lock);
1269 }
1270
1271 rcu_read_unlock();
1272 return 0;
1273
1274 error_put_subbuf:
1275 if (ustctl_put_subbuf(stream->ustream) < 0) {
1276 ERR("Snapshot ustctl_put_subbuf");
1277 }
1278 error_close_stream:
1279 consumer_stream_close(stream);
1280 error_unlock:
1281 pthread_mutex_unlock(&stream->lock);
1282 rcu_read_unlock();
1283 return ret;
1284 }
1285
1286 /*
1287 * Receive the metadata updates from the sessiond. Supports receiving
1288 * overlapping metadata, but is needs to always belong to a contiguous
1289 * range starting from 0.
1290 * Be careful about the locks held when calling this function: it needs
1291 * the metadata cache flush to concurrently progress in order to
1292 * complete.
1293 */
1294 int lttng_ustconsumer_recv_metadata(int sock, uint64_t key, uint64_t offset,
1295 uint64_t len, uint64_t version,
1296 struct lttng_consumer_channel *channel, int timer, int wait)
1297 {
1298 int ret, ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1299 char *metadata_str;
1300
1301 DBG("UST consumer push metadata key %" PRIu64 " of len %" PRIu64, key, len);
1302
1303 metadata_str = zmalloc(len * sizeof(char));
1304 if (!metadata_str) {
1305 PERROR("zmalloc metadata string");
1306 ret_code = LTTCOMM_CONSUMERD_ENOMEM;
1307 goto end;
1308 }
1309
1310 health_code_update();
1311
1312 /* Receive metadata string. */
1313 ret = lttcomm_recv_unix_sock(sock, metadata_str, len);
1314 if (ret < 0) {
1315 /* Session daemon is dead so return gracefully. */
1316 ret_code = ret;
1317 goto end_free;
1318 }
1319
1320 health_code_update();
1321
1322 pthread_mutex_lock(&channel->metadata_cache->lock);
1323 ret = consumer_metadata_cache_write(channel, offset, len, version,
1324 metadata_str);
1325 pthread_mutex_unlock(&channel->metadata_cache->lock);
1326 if (ret < 0) {
1327 /* Unable to handle metadata. Notify session daemon. */
1328 ret_code = LTTCOMM_CONSUMERD_ERROR_METADATA;
1329 /*
1330 * Skip metadata flush on write error since the offset and len might
1331 * not have been updated which could create an infinite loop below when
1332 * waiting for the metadata cache to be flushed.
1333 */
1334 goto end_free;
1335 }
1336
1337 if (!wait) {
1338 goto end_free;
1339 }
1340 while (consumer_metadata_cache_flushed(channel, offset + len, timer)) {
1341 DBG("Waiting for metadata to be flushed");
1342
1343 health_code_update();
1344
1345 usleep(DEFAULT_METADATA_AVAILABILITY_WAIT_TIME);
1346 }
1347
1348 end_free:
1349 free(metadata_str);
1350 end:
1351 return ret_code;
1352 }
1353
1354 /*
1355 * Receive command from session daemon and process it.
1356 *
1357 * Return 1 on success else a negative value or 0.
1358 */
1359 int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
1360 int sock, struct pollfd *consumer_sockpoll)
1361 {
1362 ssize_t ret;
1363 enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1364 struct lttcomm_consumer_msg msg;
1365 struct lttng_consumer_channel *channel = NULL;
1366
1367 health_code_update();
1368
1369 ret = lttcomm_recv_unix_sock(sock, &msg, sizeof(msg));
1370 if (ret != sizeof(msg)) {
1371 DBG("Consumer received unexpected message size %zd (expects %zu)",
1372 ret, sizeof(msg));
1373 /*
1374 * The ret value might 0 meaning an orderly shutdown but this is ok
1375 * since the caller handles this.
1376 */
1377 if (ret > 0) {
1378 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD);
1379 ret = -1;
1380 }
1381 return ret;
1382 }
1383
1384 health_code_update();
1385
1386 /* deprecated */
1387 assert(msg.cmd_type != LTTNG_CONSUMER_STOP);
1388
1389 health_code_update();
1390
1391 /* relayd needs RCU read-side lock */
1392 rcu_read_lock();
1393
1394 switch (msg.cmd_type) {
1395 case LTTNG_CONSUMER_ADD_RELAYD_SOCKET:
1396 {
1397 /* Session daemon status message are handled in the following call. */
1398 consumer_add_relayd_socket(msg.u.relayd_sock.net_index,
1399 msg.u.relayd_sock.type, ctx, sock, consumer_sockpoll,
1400 &msg.u.relayd_sock.sock, msg.u.relayd_sock.session_id,
1401 msg.u.relayd_sock.relayd_session_id);
1402 goto end_nosignal;
1403 }
1404 case LTTNG_CONSUMER_DESTROY_RELAYD:
1405 {
1406 uint64_t index = msg.u.destroy_relayd.net_seq_idx;
1407 struct consumer_relayd_sock_pair *relayd;
1408
1409 DBG("UST consumer destroying relayd %" PRIu64, index);
1410
1411 /* Get relayd reference if exists. */
1412 relayd = consumer_find_relayd(index);
1413 if (relayd == NULL) {
1414 DBG("Unable to find relayd %" PRIu64, index);
1415 ret_code = LTTCOMM_CONSUMERD_RELAYD_FAIL;
1416 }
1417
1418 /*
1419 * Each relayd socket pair has a refcount of stream attached to it
1420 * which tells if the relayd is still active or not depending on the
1421 * refcount value.
1422 *
1423 * This will set the destroy flag of the relayd object and destroy it
1424 * if the refcount reaches zero when called.
1425 *
1426 * The destroy can happen either here or when a stream fd hangs up.
1427 */
1428 if (relayd) {
1429 consumer_flag_relayd_for_destroy(relayd);
1430 }
1431
1432 goto end_msg_sessiond;
1433 }
1434 case LTTNG_CONSUMER_UPDATE_STREAM:
1435 {
1436 rcu_read_unlock();
1437 return -ENOSYS;
1438 }
1439 case LTTNG_CONSUMER_DATA_PENDING:
1440 {
1441 int ret, is_data_pending;
1442 uint64_t id = msg.u.data_pending.session_id;
1443
1444 DBG("UST consumer data pending command for id %" PRIu64, id);
1445
1446 is_data_pending = consumer_data_pending(id);
1447
1448 /* Send back returned value to session daemon */
1449 ret = lttcomm_send_unix_sock(sock, &is_data_pending,
1450 sizeof(is_data_pending));
1451 if (ret < 0) {
1452 DBG("Error when sending the data pending ret code: %d", ret);
1453 goto error_fatal;
1454 }
1455
1456 /*
1457 * No need to send back a status message since the data pending
1458 * returned value is the response.
1459 */
1460 break;
1461 }
1462 case LTTNG_CONSUMER_ASK_CHANNEL_CREATION:
1463 {
1464 int ret;
1465 struct ustctl_consumer_channel_attr attr;
1466 const uint64_t chunk_id = msg.u.ask_channel.chunk_id.value;
1467 const struct lttng_credentials buffer_credentials = {
1468 .uid = LTTNG_OPTIONAL_INIT_VALUE(msg.u.ask_channel.buffer_credentials.uid),
1469 .gid = LTTNG_OPTIONAL_INIT_VALUE(msg.u.ask_channel.buffer_credentials.gid),
1470 };
1471
1472 /* Create a plain object and reserve a channel key. */
1473 channel = consumer_allocate_channel(
1474 msg.u.ask_channel.key,
1475 msg.u.ask_channel.session_id,
1476 msg.u.ask_channel.chunk_id.is_set ?
1477 &chunk_id : NULL,
1478 msg.u.ask_channel.pathname,
1479 msg.u.ask_channel.name,
1480 msg.u.ask_channel.relayd_id,
1481 (enum lttng_event_output) msg.u.ask_channel.output,
1482 msg.u.ask_channel.tracefile_size,
1483 msg.u.ask_channel.tracefile_count,
1484 msg.u.ask_channel.session_id_per_pid,
1485 msg.u.ask_channel.monitor,
1486 msg.u.ask_channel.live_timer_interval,
1487 msg.u.ask_channel.is_live,
1488 msg.u.ask_channel.root_shm_path,
1489 msg.u.ask_channel.shm_path);
1490 if (!channel) {
1491 goto end_channel_error;
1492 }
1493
1494 LTTNG_OPTIONAL_SET(&channel->buffer_credentials,
1495 buffer_credentials);
1496
1497 /*
1498 * Assign UST application UID to the channel. This value is ignored for
1499 * per PID buffers. This is specific to UST thus setting this after the
1500 * allocation.
1501 */
1502 channel->ust_app_uid = msg.u.ask_channel.ust_app_uid;
1503
1504 /* Build channel attributes from received message. */
1505 attr.subbuf_size = msg.u.ask_channel.subbuf_size;
1506 attr.num_subbuf = msg.u.ask_channel.num_subbuf;
1507 attr.overwrite = msg.u.ask_channel.overwrite;
1508 attr.switch_timer_interval = msg.u.ask_channel.switch_timer_interval;
1509 attr.read_timer_interval = msg.u.ask_channel.read_timer_interval;
1510 attr.chan_id = msg.u.ask_channel.chan_id;
1511 memcpy(attr.uuid, msg.u.ask_channel.uuid, sizeof(attr.uuid));
1512 attr.blocking_timeout= msg.u.ask_channel.blocking_timeout;
1513
1514 /* Match channel buffer type to the UST abi. */
1515 switch (msg.u.ask_channel.output) {
1516 case LTTNG_EVENT_MMAP:
1517 default:
1518 attr.output = LTTNG_UST_MMAP;
1519 break;
1520 }
1521
1522 /* Translate and save channel type. */
1523 switch (msg.u.ask_channel.type) {
1524 case LTTNG_UST_CHAN_PER_CPU:
1525 channel->type = CONSUMER_CHANNEL_TYPE_DATA;
1526 attr.type = LTTNG_UST_CHAN_PER_CPU;
1527 /*
1528 * Set refcount to 1 for owner. Below, we will
1529 * pass ownership to the
1530 * consumer_thread_channel_poll() thread.
1531 */
1532 channel->refcount = 1;
1533 break;
1534 case LTTNG_UST_CHAN_METADATA:
1535 channel->type = CONSUMER_CHANNEL_TYPE_METADATA;
1536 attr.type = LTTNG_UST_CHAN_METADATA;
1537 break;
1538 default:
1539 assert(0);
1540 goto error_fatal;
1541 };
1542
1543 health_code_update();
1544
1545 ret = ask_channel(ctx, channel, &attr);
1546 if (ret < 0) {
1547 goto end_channel_error;
1548 }
1549
1550 if (msg.u.ask_channel.type == LTTNG_UST_CHAN_METADATA) {
1551 ret = consumer_metadata_cache_allocate(channel);
1552 if (ret < 0) {
1553 ERR("Allocating metadata cache");
1554 goto end_channel_error;
1555 }
1556 consumer_timer_switch_start(channel, attr.switch_timer_interval);
1557 attr.switch_timer_interval = 0;
1558 } else {
1559 int monitor_start_ret;
1560
1561 consumer_timer_live_start(channel,
1562 msg.u.ask_channel.live_timer_interval);
1563 monitor_start_ret = consumer_timer_monitor_start(
1564 channel,
1565 msg.u.ask_channel.monitor_timer_interval);
1566 if (monitor_start_ret < 0) {
1567 ERR("Starting channel monitoring timer failed");
1568 goto end_channel_error;
1569 }
1570 }
1571
1572 health_code_update();
1573
1574 /*
1575 * Add the channel to the internal state AFTER all streams were created
1576 * and successfully sent to session daemon. This way, all streams must
1577 * be ready before this channel is visible to the threads.
1578 * If add_channel succeeds, ownership of the channel is
1579 * passed to consumer_thread_channel_poll().
1580 */
1581 ret = add_channel(channel, ctx);
1582 if (ret < 0) {
1583 if (msg.u.ask_channel.type == LTTNG_UST_CHAN_METADATA) {
1584 if (channel->switch_timer_enabled == 1) {
1585 consumer_timer_switch_stop(channel);
1586 }
1587 consumer_metadata_cache_destroy(channel);
1588 }
1589 if (channel->live_timer_enabled == 1) {
1590 consumer_timer_live_stop(channel);
1591 }
1592 if (channel->monitor_timer_enabled == 1) {
1593 consumer_timer_monitor_stop(channel);
1594 }
1595 goto end_channel_error;
1596 }
1597
1598 health_code_update();
1599
1600 /*
1601 * Channel and streams are now created. Inform the session daemon that
1602 * everything went well and should wait to receive the channel and
1603 * streams with ustctl API.
1604 */
1605 ret = consumer_send_status_channel(sock, channel);
1606 if (ret < 0) {
1607 /*
1608 * There is probably a problem on the socket.
1609 */
1610 goto error_fatal;
1611 }
1612
1613 break;
1614 }
1615 case LTTNG_CONSUMER_GET_CHANNEL:
1616 {
1617 int ret, relayd_err = 0;
1618 uint64_t key = msg.u.get_channel.key;
1619 struct lttng_consumer_channel *channel;
1620
1621 channel = consumer_find_channel(key);
1622 if (!channel) {
1623 ERR("UST consumer get channel key %" PRIu64 " not found", key);
1624 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
1625 goto end_get_channel;
1626 }
1627
1628 health_code_update();
1629
1630 /* Send the channel to sessiond (and relayd, if applicable). */
1631 ret = send_channel_to_sessiond_and_relayd(sock, channel, ctx,
1632 &relayd_err);
1633 if (ret < 0) {
1634 if (relayd_err) {
1635 /*
1636 * We were unable to send to the relayd the stream so avoid
1637 * sending back a fatal error to the thread since this is OK
1638 * and the consumer can continue its work. The above call
1639 * has sent the error status message to the sessiond.
1640 */
1641 goto end_get_channel_nosignal;
1642 }
1643 /*
1644 * The communicaton was broken hence there is a bad state between
1645 * the consumer and sessiond so stop everything.
1646 */
1647 goto error_get_channel_fatal;
1648 }
1649
1650 health_code_update();
1651
1652 /*
1653 * In no monitor mode, the streams ownership is kept inside the channel
1654 * so don't send them to the data thread.
1655 */
1656 if (!channel->monitor) {
1657 goto end_get_channel;
1658 }
1659
1660 ret = send_streams_to_thread(channel, ctx);
1661 if (ret < 0) {
1662 /*
1663 * If we are unable to send the stream to the thread, there is
1664 * a big problem so just stop everything.
1665 */
1666 goto error_get_channel_fatal;
1667 }
1668 /* List MUST be empty after or else it could be reused. */
1669 assert(cds_list_empty(&channel->streams.head));
1670 end_get_channel:
1671 goto end_msg_sessiond;
1672 error_get_channel_fatal:
1673 goto error_fatal;
1674 end_get_channel_nosignal:
1675 goto end_nosignal;
1676 }
1677 case LTTNG_CONSUMER_DESTROY_CHANNEL:
1678 {
1679 uint64_t key = msg.u.destroy_channel.key;
1680
1681 /*
1682 * Only called if streams have not been sent to stream
1683 * manager thread. However, channel has been sent to
1684 * channel manager thread.
1685 */
1686 notify_thread_del_channel(ctx, key);
1687 goto end_msg_sessiond;
1688 }
1689 case LTTNG_CONSUMER_CLOSE_METADATA:
1690 {
1691 int ret;
1692
1693 ret = close_metadata(msg.u.close_metadata.key);
1694 if (ret != 0) {
1695 ret_code = ret;
1696 }
1697
1698 goto end_msg_sessiond;
1699 }
1700 case LTTNG_CONSUMER_FLUSH_CHANNEL:
1701 {
1702 int ret;
1703
1704 ret = flush_channel(msg.u.flush_channel.key);
1705 if (ret != 0) {
1706 ret_code = ret;
1707 }
1708
1709 goto end_msg_sessiond;
1710 }
1711 case LTTNG_CONSUMER_CLEAR_QUIESCENT_CHANNEL:
1712 {
1713 int ret;
1714
1715 ret = clear_quiescent_channel(
1716 msg.u.clear_quiescent_channel.key);
1717 if (ret != 0) {
1718 ret_code = ret;
1719 }
1720
1721 goto end_msg_sessiond;
1722 }
1723 case LTTNG_CONSUMER_PUSH_METADATA:
1724 {
1725 int ret;
1726 uint64_t len = msg.u.push_metadata.len;
1727 uint64_t key = msg.u.push_metadata.key;
1728 uint64_t offset = msg.u.push_metadata.target_offset;
1729 uint64_t version = msg.u.push_metadata.version;
1730 struct lttng_consumer_channel *channel;
1731
1732 DBG("UST consumer push metadata key %" PRIu64 " of len %" PRIu64, key,
1733 len);
1734
1735 channel = consumer_find_channel(key);
1736 if (!channel) {
1737 /*
1738 * This is possible if the metadata creation on the consumer side
1739 * is in flight vis-a-vis a concurrent push metadata from the
1740 * session daemon. Simply return that the channel failed and the
1741 * session daemon will handle that message correctly considering
1742 * that this race is acceptable thus the DBG() statement here.
1743 */
1744 DBG("UST consumer push metadata %" PRIu64 " not found", key);
1745 ret_code = LTTCOMM_CONSUMERD_CHANNEL_FAIL;
1746 goto end_push_metadata_msg_sessiond;
1747 }
1748
1749 health_code_update();
1750
1751 if (!len) {
1752 /*
1753 * There is nothing to receive. We have simply
1754 * checked whether the channel can be found.
1755 */
1756 ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1757 goto end_push_metadata_msg_sessiond;
1758 }
1759
1760 /* Tell session daemon we are ready to receive the metadata. */
1761 ret = consumer_send_status_msg(sock, LTTCOMM_CONSUMERD_SUCCESS);
1762 if (ret < 0) {
1763 /* Somehow, the session daemon is not responding anymore. */
1764 goto error_push_metadata_fatal;
1765 }
1766
1767 health_code_update();
1768
1769 /* Wait for more data. */
1770 health_poll_entry();
1771 ret = lttng_consumer_poll_socket(consumer_sockpoll);
1772 health_poll_exit();
1773 if (ret) {
1774 goto error_push_metadata_fatal;
1775 }
1776
1777 health_code_update();
1778
1779 ret = lttng_ustconsumer_recv_metadata(sock, key, offset,
1780 len, version, channel, 0, 1);
1781 if (ret < 0) {
1782 /* error receiving from sessiond */
1783 goto error_push_metadata_fatal;
1784 } else {
1785 ret_code = ret;
1786 goto end_push_metadata_msg_sessiond;
1787 }
1788 end_push_metadata_msg_sessiond:
1789 goto end_msg_sessiond;
1790 error_push_metadata_fatal:
1791 goto error_fatal;
1792 }
1793 case LTTNG_CONSUMER_SETUP_METADATA:
1794 {
1795 int ret;
1796
1797 ret = setup_metadata(ctx, msg.u.setup_metadata.key);
1798 if (ret) {
1799 ret_code = ret;
1800 }
1801 goto end_msg_sessiond;
1802 }
1803 case LTTNG_CONSUMER_SNAPSHOT_CHANNEL:
1804 {
1805 struct lttng_consumer_channel *channel;
1806 uint64_t key = msg.u.snapshot_channel.key;
1807
1808 channel = consumer_find_channel(key);
1809 if (!channel) {
1810 DBG("UST snapshot channel not found for key %" PRIu64, key);
1811 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
1812 } else {
1813 if (msg.u.snapshot_channel.metadata) {
1814 ret = snapshot_metadata(channel, key,
1815 msg.u.snapshot_channel.pathname,
1816 msg.u.snapshot_channel.relayd_id,
1817 ctx);
1818 if (ret < 0) {
1819 ERR("Snapshot metadata failed");
1820 ret_code = LTTCOMM_CONSUMERD_SNAPSHOT_FAILED;
1821 }
1822 } else {
1823 ret = snapshot_channel(channel, key,
1824 msg.u.snapshot_channel.pathname,
1825 msg.u.snapshot_channel.relayd_id,
1826 msg.u.snapshot_channel.nb_packets_per_stream,
1827 ctx);
1828 if (ret < 0) {
1829 ERR("Snapshot channel failed");
1830 ret_code = LTTCOMM_CONSUMERD_SNAPSHOT_FAILED;
1831 }
1832 }
1833 }
1834 health_code_update();
1835 ret = consumer_send_status_msg(sock, ret_code);
1836 if (ret < 0) {
1837 /* Somehow, the session daemon is not responding anymore. */
1838 goto end_nosignal;
1839 }
1840 health_code_update();
1841 break;
1842 }
1843 case LTTNG_CONSUMER_DISCARDED_EVENTS:
1844 {
1845 int ret = 0;
1846 uint64_t discarded_events;
1847 struct lttng_ht_iter iter;
1848 struct lttng_ht *ht;
1849 struct lttng_consumer_stream *stream;
1850 uint64_t id = msg.u.discarded_events.session_id;
1851 uint64_t key = msg.u.discarded_events.channel_key;
1852
1853 DBG("UST consumer discarded events command for session id %"
1854 PRIu64, id);
1855 rcu_read_lock();
1856 pthread_mutex_lock(&consumer_data.lock);
1857
1858 ht = consumer_data.stream_list_ht;
1859
1860 /*
1861 * We only need a reference to the channel, but they are not
1862 * directly indexed, so we just use the first matching stream
1863 * to extract the information we need, we default to 0 if not
1864 * found (no events are dropped if the channel is not yet in
1865 * use).
1866 */
1867 discarded_events = 0;
1868 cds_lfht_for_each_entry_duplicate(ht->ht,
1869 ht->hash_fct(&id, lttng_ht_seed),
1870 ht->match_fct, &id,
1871 &iter.iter, stream, node_session_id.node) {
1872 if (stream->chan->key == key) {
1873 discarded_events = stream->chan->discarded_events;
1874 break;
1875 }
1876 }
1877 pthread_mutex_unlock(&consumer_data.lock);
1878 rcu_read_unlock();
1879
1880 DBG("UST consumer discarded events command for session id %"
1881 PRIu64 ", channel key %" PRIu64, id, key);
1882
1883 health_code_update();
1884
1885 /* Send back returned value to session daemon */
1886 ret = lttcomm_send_unix_sock(sock, &discarded_events, sizeof(discarded_events));
1887 if (ret < 0) {
1888 PERROR("send discarded events");
1889 goto error_fatal;
1890 }
1891
1892 break;
1893 }
1894 case LTTNG_CONSUMER_LOST_PACKETS:
1895 {
1896 int ret;
1897 uint64_t lost_packets;
1898 struct lttng_ht_iter iter;
1899 struct lttng_ht *ht;
1900 struct lttng_consumer_stream *stream;
1901 uint64_t id = msg.u.lost_packets.session_id;
1902 uint64_t key = msg.u.lost_packets.channel_key;
1903
1904 DBG("UST consumer lost packets 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 packets lost if the channel is not yet in use).
1916 */
1917 lost_packets = 0;
1918 cds_lfht_for_each_entry_duplicate(ht->ht,
1919 ht->hash_fct(&id, lttng_ht_seed),
1920 ht->match_fct, &id,
1921 &iter.iter, stream, node_session_id.node) {
1922 if (stream->chan->key == key) {
1923 lost_packets = stream->chan->lost_packets;
1924 break;
1925 }
1926 }
1927 pthread_mutex_unlock(&consumer_data.lock);
1928 rcu_read_unlock();
1929
1930 DBG("UST consumer lost packets command for session id %"
1931 PRIu64 ", channel key %" PRIu64, id, key);
1932
1933 health_code_update();
1934
1935 /* Send back returned value to session daemon */
1936 ret = lttcomm_send_unix_sock(sock, &lost_packets,
1937 sizeof(lost_packets));
1938 if (ret < 0) {
1939 PERROR("send lost packets");
1940 goto error_fatal;
1941 }
1942
1943 break;
1944 }
1945 case LTTNG_CONSUMER_SET_CHANNEL_MONITOR_PIPE:
1946 {
1947 int channel_monitor_pipe;
1948
1949 ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1950 /* Successfully received the command's type. */
1951 ret = consumer_send_status_msg(sock, ret_code);
1952 if (ret < 0) {
1953 goto error_fatal;
1954 }
1955
1956 ret = lttcomm_recv_fds_unix_sock(sock, &channel_monitor_pipe,
1957 1);
1958 if (ret != sizeof(channel_monitor_pipe)) {
1959 ERR("Failed to receive channel monitor pipe");
1960 goto error_fatal;
1961 }
1962
1963 DBG("Received channel monitor pipe (%d)", channel_monitor_pipe);
1964 ret = consumer_timer_thread_set_channel_monitor_pipe(
1965 channel_monitor_pipe);
1966 if (!ret) {
1967 int flags;
1968
1969 ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1970 /* Set the pipe as non-blocking. */
1971 ret = fcntl(channel_monitor_pipe, F_GETFL, 0);
1972 if (ret == -1) {
1973 PERROR("fcntl get flags of the channel monitoring pipe");
1974 goto error_fatal;
1975 }
1976 flags = ret;
1977
1978 ret = fcntl(channel_monitor_pipe, F_SETFL,
1979 flags | O_NONBLOCK);
1980 if (ret == -1) {
1981 PERROR("fcntl set O_NONBLOCK flag of the channel monitoring pipe");
1982 goto error_fatal;
1983 }
1984 DBG("Channel monitor pipe set as non-blocking");
1985 } else {
1986 ret_code = LTTCOMM_CONSUMERD_ALREADY_SET;
1987 }
1988 goto end_msg_sessiond;
1989 }
1990 case LTTNG_CONSUMER_ROTATE_CHANNEL:
1991 {
1992 struct lttng_consumer_channel *channel;
1993 uint64_t key = msg.u.rotate_channel.key;
1994
1995 channel = consumer_find_channel(key);
1996 if (!channel) {
1997 DBG("Channel %" PRIu64 " not found", key);
1998 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
1999 } else {
2000 /*
2001 * Sample the rotate position of all the streams in
2002 * this channel.
2003 */
2004 ret = lttng_consumer_rotate_channel(channel, key,
2005 msg.u.rotate_channel.relayd_id,
2006 msg.u.rotate_channel.metadata,
2007 ctx);
2008 if (ret < 0) {
2009 ERR("Rotate channel failed");
2010 ret_code = LTTCOMM_CONSUMERD_ROTATION_FAIL;
2011 }
2012
2013 health_code_update();
2014 }
2015 ret = consumer_send_status_msg(sock, ret_code);
2016 if (ret < 0) {
2017 /* Somehow, the session daemon is not responding anymore. */
2018 goto end_rotate_channel_nosignal;
2019 }
2020
2021 /*
2022 * Rotate the streams that are ready right now.
2023 * FIXME: this is a second consecutive iteration over the
2024 * streams in a channel, there is probably a better way to
2025 * handle this, but it needs to be after the
2026 * consumer_send_status_msg() call.
2027 */
2028 if (channel) {
2029 ret = lttng_consumer_rotate_ready_streams(
2030 channel, key, ctx);
2031 if (ret < 0) {
2032 ERR("Rotate channel failed");
2033 }
2034 }
2035 break;
2036 end_rotate_channel_nosignal:
2037 goto end_nosignal;
2038 }
2039 case LTTNG_CONSUMER_CLEAR_CHANNEL:
2040 {
2041 struct lttng_consumer_channel *channel;
2042 uint64_t key = msg.u.clear_channel.key;
2043
2044 channel = consumer_find_channel(key);
2045 if (!channel) {
2046 DBG("Channel %" PRIu64 " not found", key);
2047 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
2048 } else {
2049 ret = lttng_consumer_clear_channel(channel);
2050 if (ret) {
2051 ERR("Clear channel failed key %" PRIu64, key);
2052 ret_code = ret;
2053 }
2054
2055 health_code_update();
2056 }
2057 ret = consumer_send_status_msg(sock, ret_code);
2058 if (ret < 0) {
2059 /* Somehow, the session daemon is not responding anymore. */
2060 goto end_nosignal;
2061 }
2062 break;
2063 }
2064 case LTTNG_CONSUMER_INIT:
2065 {
2066 ret_code = lttng_consumer_init_command(ctx,
2067 msg.u.init.sessiond_uuid);
2068 health_code_update();
2069 ret = consumer_send_status_msg(sock, ret_code);
2070 if (ret < 0) {
2071 /* Somehow, the session daemon is not responding anymore. */
2072 goto end_nosignal;
2073 }
2074 break;
2075 }
2076 case LTTNG_CONSUMER_CREATE_TRACE_CHUNK:
2077 {
2078 const struct lttng_credentials credentials = {
2079 .uid = LTTNG_OPTIONAL_INIT_VALUE(msg.u.create_trace_chunk.credentials.value.uid),
2080 .gid = LTTNG_OPTIONAL_INIT_VALUE(msg.u.create_trace_chunk.credentials.value.gid),
2081 };
2082 const bool is_local_trace =
2083 !msg.u.create_trace_chunk.relayd_id.is_set;
2084 const uint64_t relayd_id =
2085 msg.u.create_trace_chunk.relayd_id.value;
2086 const char *chunk_override_name =
2087 *msg.u.create_trace_chunk.override_name ?
2088 msg.u.create_trace_chunk.override_name :
2089 NULL;
2090 struct lttng_directory_handle *chunk_directory_handle = NULL;
2091
2092 /*
2093 * The session daemon will only provide a chunk directory file
2094 * descriptor for local traces.
2095 */
2096 if (is_local_trace) {
2097 int chunk_dirfd;
2098
2099 /* Acnowledge the reception of the command. */
2100 ret = consumer_send_status_msg(sock,
2101 LTTCOMM_CONSUMERD_SUCCESS);
2102 if (ret < 0) {
2103 /* Somehow, the session daemon is not responding anymore. */
2104 goto end_nosignal;
2105 }
2106
2107 /*
2108 * Receive trace chunk domain dirfd.
2109 */
2110 ret = lttcomm_recv_fds_unix_sock(sock, &chunk_dirfd, 1);
2111 if (ret != sizeof(chunk_dirfd)) {
2112 ERR("Failed to receive trace chunk domain directory file descriptor");
2113 goto error_fatal;
2114 }
2115
2116 DBG("Received trace chunk domain directory fd (%d)",
2117 chunk_dirfd);
2118 chunk_directory_handle = lttng_directory_handle_create_from_dirfd(
2119 chunk_dirfd);
2120 if (!chunk_directory_handle) {
2121 ERR("Failed to initialize chunk domain directory handle from directory file descriptor");
2122 if (close(chunk_dirfd)) {
2123 PERROR("Failed to close chunk directory file descriptor");
2124 }
2125 goto error_fatal;
2126 }
2127 }
2128
2129 ret_code = lttng_consumer_create_trace_chunk(
2130 !is_local_trace ? &relayd_id : NULL,
2131 msg.u.create_trace_chunk.session_id,
2132 msg.u.create_trace_chunk.chunk_id,
2133 (time_t) msg.u.create_trace_chunk
2134 .creation_timestamp,
2135 chunk_override_name,
2136 msg.u.create_trace_chunk.credentials.is_set ?
2137 &credentials :
2138 NULL,
2139 chunk_directory_handle);
2140 lttng_directory_handle_put(chunk_directory_handle);
2141 goto end_msg_sessiond;
2142 }
2143 case LTTNG_CONSUMER_CLOSE_TRACE_CHUNK:
2144 {
2145 enum lttng_trace_chunk_command_type close_command =
2146 msg.u.close_trace_chunk.close_command.value;
2147 const uint64_t relayd_id =
2148 msg.u.close_trace_chunk.relayd_id.value;
2149 struct lttcomm_consumer_close_trace_chunk_reply reply;
2150 char closed_trace_chunk_path[LTTNG_PATH_MAX];
2151 int ret;
2152
2153 ret_code = lttng_consumer_close_trace_chunk(
2154 msg.u.close_trace_chunk.relayd_id.is_set ?
2155 &relayd_id :
2156 NULL,
2157 msg.u.close_trace_chunk.session_id,
2158 msg.u.close_trace_chunk.chunk_id,
2159 (time_t) msg.u.close_trace_chunk.close_timestamp,
2160 msg.u.close_trace_chunk.close_command.is_set ?
2161 &close_command :
2162 NULL, closed_trace_chunk_path);
2163 reply.ret_code = ret_code;
2164 reply.path_length = strlen(closed_trace_chunk_path) + 1;
2165 ret = lttcomm_send_unix_sock(sock, &reply, sizeof(reply));
2166 if (ret != sizeof(reply)) {
2167 goto error_fatal;
2168 }
2169 ret = lttcomm_send_unix_sock(sock, closed_trace_chunk_path,
2170 reply.path_length);
2171 if (ret != reply.path_length) {
2172 goto error_fatal;
2173 }
2174 goto end_nosignal;
2175 }
2176 case LTTNG_CONSUMER_TRACE_CHUNK_EXISTS:
2177 {
2178 const uint64_t relayd_id =
2179 msg.u.trace_chunk_exists.relayd_id.value;
2180
2181 ret_code = lttng_consumer_trace_chunk_exists(
2182 msg.u.trace_chunk_exists.relayd_id.is_set ?
2183 &relayd_id : NULL,
2184 msg.u.trace_chunk_exists.session_id,
2185 msg.u.trace_chunk_exists.chunk_id);
2186 goto end_msg_sessiond;
2187 }
2188 case LTTNG_CONSUMER_OPEN_CHANNEL_PACKETS:
2189 {
2190 const uint64_t key = msg.u.open_channel_packets.key;
2191 struct lttng_consumer_channel *channel =
2192 consumer_find_channel(key);
2193
2194 if (channel) {
2195 pthread_mutex_lock(&channel->lock);
2196 ret_code = lttng_consumer_open_channel_packets(channel);
2197 pthread_mutex_unlock(&channel->lock);
2198 } else {
2199 /*
2200 * The channel could have disappeared in per-pid
2201 * buffering mode.
2202 */
2203 DBG("Channel %" PRIu64 " not found", key);
2204 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
2205 }
2206
2207 health_code_update();
2208 goto end_msg_sessiond;
2209 }
2210 default:
2211 break;
2212 }
2213
2214 end_nosignal:
2215 /*
2216 * Return 1 to indicate success since the 0 value can be a socket
2217 * shutdown during the recv() or send() call.
2218 */
2219 ret = 1;
2220 goto end;
2221
2222 end_msg_sessiond:
2223 /*
2224 * The returned value here is not useful since either way we'll return 1 to
2225 * the caller because the session daemon socket management is done
2226 * elsewhere. Returning a negative code or 0 will shutdown the consumer.
2227 */
2228 ret = consumer_send_status_msg(sock, ret_code);
2229 if (ret < 0) {
2230 goto error_fatal;
2231 }
2232 ret = 1;
2233 goto end;
2234
2235 end_channel_error:
2236 if (channel) {
2237 /*
2238 * Free channel here since no one has a reference to it. We don't
2239 * free after that because a stream can store this pointer.
2240 */
2241 destroy_channel(channel);
2242 }
2243 /* We have to send a status channel message indicating an error. */
2244 ret = consumer_send_status_channel(sock, NULL);
2245 if (ret < 0) {
2246 /* Stop everything if session daemon can not be notified. */
2247 goto error_fatal;
2248 }
2249 ret = 1;
2250 goto end;
2251
2252 error_fatal:
2253 /* This will issue a consumer stop. */
2254 ret = -1;
2255 goto end;
2256
2257 end:
2258 rcu_read_unlock();
2259 health_code_update();
2260 return ret;
2261 }
2262
2263 void lttng_ustctl_flush_buffer(struct lttng_consumer_stream *stream,
2264 int producer_active)
2265 {
2266 assert(stream);
2267 assert(stream->ustream);
2268
2269 ustctl_flush_buffer(stream->ustream, producer_active);
2270 }
2271
2272 /*
2273 * Take a snapshot for a specific stream.
2274 *
2275 * Returns 0 on success, < 0 on error
2276 */
2277 int lttng_ustconsumer_take_snapshot(struct lttng_consumer_stream *stream)
2278 {
2279 assert(stream);
2280 assert(stream->ustream);
2281
2282 return ustctl_snapshot(stream->ustream);
2283 }
2284
2285 /*
2286 * Sample consumed and produced positions for a specific stream.
2287 *
2288 * Returns 0 on success, < 0 on error.
2289 */
2290 int lttng_ustconsumer_sample_snapshot_positions(
2291 struct lttng_consumer_stream *stream)
2292 {
2293 assert(stream);
2294 assert(stream->ustream);
2295
2296 return ustctl_snapshot_sample_positions(stream->ustream);
2297 }
2298
2299 /*
2300 * Get the produced position
2301 *
2302 * Returns 0 on success, < 0 on error
2303 */
2304 int lttng_ustconsumer_get_produced_snapshot(
2305 struct lttng_consumer_stream *stream, unsigned long *pos)
2306 {
2307 assert(stream);
2308 assert(stream->ustream);
2309 assert(pos);
2310
2311 return ustctl_snapshot_get_produced(stream->ustream, pos);
2312 }
2313
2314 /*
2315 * Get the consumed position
2316 *
2317 * Returns 0 on success, < 0 on error
2318 */
2319 int lttng_ustconsumer_get_consumed_snapshot(
2320 struct lttng_consumer_stream *stream, unsigned long *pos)
2321 {
2322 assert(stream);
2323 assert(stream->ustream);
2324 assert(pos);
2325
2326 return ustctl_snapshot_get_consumed(stream->ustream, pos);
2327 }
2328
2329 void lttng_ustconsumer_flush_buffer(struct lttng_consumer_stream *stream,
2330 int producer)
2331 {
2332 assert(stream);
2333 assert(stream->ustream);
2334
2335 ustctl_flush_buffer(stream->ustream, producer);
2336 }
2337
2338 void lttng_ustconsumer_clear_buffer(struct lttng_consumer_stream *stream)
2339 {
2340 assert(stream);
2341 assert(stream->ustream);
2342
2343 ustctl_clear_buffer(stream->ustream);
2344 }
2345
2346 int lttng_ustconsumer_get_current_timestamp(
2347 struct lttng_consumer_stream *stream, uint64_t *ts)
2348 {
2349 assert(stream);
2350 assert(stream->ustream);
2351 assert(ts);
2352
2353 return ustctl_get_current_timestamp(stream->ustream, ts);
2354 }
2355
2356 int lttng_ustconsumer_get_sequence_number(
2357 struct lttng_consumer_stream *stream, uint64_t *seq)
2358 {
2359 assert(stream);
2360 assert(stream->ustream);
2361 assert(seq);
2362
2363 return ustctl_get_sequence_number(stream->ustream, seq);
2364 }
2365
2366 /*
2367 * Called when the stream signals the consumer that it has hung up.
2368 */
2369 void lttng_ustconsumer_on_stream_hangup(struct lttng_consumer_stream *stream)
2370 {
2371 assert(stream);
2372 assert(stream->ustream);
2373
2374 pthread_mutex_lock(&stream->lock);
2375 if (!stream->quiescent) {
2376 ustctl_flush_buffer(stream->ustream, 0);
2377 stream->quiescent = true;
2378 }
2379 pthread_mutex_unlock(&stream->lock);
2380 stream->hangup_flush_done = 1;
2381 }
2382
2383 void lttng_ustconsumer_del_channel(struct lttng_consumer_channel *chan)
2384 {
2385 int i;
2386
2387 assert(chan);
2388 assert(chan->uchan);
2389 assert(chan->buffer_credentials.is_set);
2390
2391 if (chan->switch_timer_enabled == 1) {
2392 consumer_timer_switch_stop(chan);
2393 }
2394 for (i = 0; i < chan->nr_stream_fds; i++) {
2395 int ret;
2396
2397 ret = close(chan->stream_fds[i]);
2398 if (ret) {
2399 PERROR("close");
2400 }
2401 if (chan->shm_path[0]) {
2402 char shm_path[PATH_MAX];
2403
2404 ret = get_stream_shm_path(shm_path, chan->shm_path, i);
2405 if (ret) {
2406 ERR("Cannot get stream shm path");
2407 }
2408 ret = run_as_unlink(shm_path,
2409 lttng_credentials_get_uid(LTTNG_OPTIONAL_GET_PTR(
2410 chan->buffer_credentials)),
2411 lttng_credentials_get_gid(LTTNG_OPTIONAL_GET_PTR(
2412 chan->buffer_credentials)));
2413 if (ret) {
2414 PERROR("unlink %s", shm_path);
2415 }
2416 }
2417 }
2418 }
2419
2420 void lttng_ustconsumer_free_channel(struct lttng_consumer_channel *chan)
2421 {
2422 assert(chan);
2423 assert(chan->uchan);
2424 assert(chan->buffer_credentials.is_set);
2425
2426 consumer_metadata_cache_destroy(chan);
2427 ustctl_destroy_channel(chan->uchan);
2428 /* Try to rmdir all directories under shm_path root. */
2429 if (chan->root_shm_path[0]) {
2430 (void) run_as_rmdir_recursive(chan->root_shm_path,
2431 lttng_credentials_get_uid(LTTNG_OPTIONAL_GET_PTR(
2432 chan->buffer_credentials)),
2433 lttng_credentials_get_gid(LTTNG_OPTIONAL_GET_PTR(
2434 chan->buffer_credentials)),
2435 LTTNG_DIRECTORY_HANDLE_SKIP_NON_EMPTY_FLAG);
2436 }
2437 free(chan->stream_fds);
2438 }
2439
2440 void lttng_ustconsumer_del_stream(struct lttng_consumer_stream *stream)
2441 {
2442 assert(stream);
2443 assert(stream->ustream);
2444
2445 if (stream->chan->switch_timer_enabled == 1) {
2446 consumer_timer_switch_stop(stream->chan);
2447 }
2448 ustctl_destroy_stream(stream->ustream);
2449 }
2450
2451 int lttng_ustconsumer_get_wakeup_fd(struct lttng_consumer_stream *stream)
2452 {
2453 assert(stream);
2454 assert(stream->ustream);
2455
2456 return ustctl_stream_get_wakeup_fd(stream->ustream);
2457 }
2458
2459 int lttng_ustconsumer_close_wakeup_fd(struct lttng_consumer_stream *stream)
2460 {
2461 assert(stream);
2462 assert(stream->ustream);
2463
2464 return ustctl_stream_close_wakeup_fd(stream->ustream);
2465 }
2466
2467 static
2468 void metadata_stream_reset_cache_consumed_position(
2469 struct lttng_consumer_stream *stream)
2470 {
2471 DBG("Reset metadata cache of session %" PRIu64,
2472 stream->chan->session_id);
2473 stream->ust_metadata_pushed = 0;
2474 }
2475
2476 /*
2477 * Write up to one packet from the metadata cache to the channel.
2478 *
2479 * Returns the number of bytes pushed from the cache into the ring buffer, or a
2480 * negative value on error.
2481 */
2482 static
2483 int commit_one_metadata_packet(struct lttng_consumer_stream *stream)
2484 {
2485 ssize_t write_len;
2486 int ret;
2487
2488 pthread_mutex_lock(&stream->chan->metadata_cache->lock);
2489 if (stream->chan->metadata_cache->max_offset ==
2490 stream->ust_metadata_pushed) {
2491 /*
2492 * In the context of a user space metadata channel, a
2493 * change in version can be detected in two ways:
2494 * 1) During the pre-consume of the `read_subbuffer` loop,
2495 * 2) When populating the metadata ring buffer (i.e. here).
2496 *
2497 * This function is invoked when there is no metadata
2498 * available in the ring-buffer. If all data was consumed
2499 * up to the size of the metadata cache, there is no metadata
2500 * to insert in the ring-buffer.
2501 *
2502 * However, the metadata version could still have changed (a
2503 * regeneration without any new data will yield the same cache
2504 * size).
2505 *
2506 * The cache's version is checked for a version change and the
2507 * consumed position is reset if one occurred.
2508 *
2509 * This check is only necessary for the user space domain as
2510 * it has to manage the cache explicitly. If this reset was not
2511 * performed, no metadata would be consumed (and no reset would
2512 * occur as part of the pre-consume) until the metadata size
2513 * exceeded the cache size.
2514 */
2515 if (stream->metadata_version !=
2516 stream->chan->metadata_cache->version) {
2517 metadata_stream_reset_cache_consumed_position(stream);
2518 consumer_stream_metadata_set_version(stream,
2519 stream->chan->metadata_cache->version);
2520 } else {
2521 ret = 0;
2522 goto end;
2523 }
2524 }
2525
2526 write_len = ustctl_write_one_packet_to_channel(stream->chan->uchan,
2527 &stream->chan->metadata_cache->data[stream->ust_metadata_pushed],
2528 stream->chan->metadata_cache->max_offset
2529 - stream->ust_metadata_pushed);
2530 assert(write_len != 0);
2531 if (write_len < 0) {
2532 ERR("Writing one metadata packet");
2533 ret = write_len;
2534 goto end;
2535 }
2536 stream->ust_metadata_pushed += write_len;
2537
2538 assert(stream->chan->metadata_cache->max_offset >=
2539 stream->ust_metadata_pushed);
2540 ret = write_len;
2541
2542 /*
2543 * Switch packet (but don't open the next one) on every commit of
2544 * a metadata packet. Since the subbuffer is fully filled (with padding,
2545 * if needed), the stream is "quiescent" after this commit.
2546 */
2547 ustctl_flush_buffer(stream->ustream, 1);
2548 stream->quiescent = true;
2549 end:
2550 pthread_mutex_unlock(&stream->chan->metadata_cache->lock);
2551 return ret;
2552 }
2553
2554
2555 /*
2556 * Sync metadata meaning request them to the session daemon and snapshot to the
2557 * metadata thread can consumer them.
2558 *
2559 * Metadata stream lock is held here, but we need to release it when
2560 * interacting with sessiond, else we cause a deadlock with live
2561 * awaiting on metadata to be pushed out.
2562 *
2563 * The RCU read side lock must be held by the caller.
2564 */
2565 enum sync_metadata_status lttng_ustconsumer_sync_metadata(
2566 struct lttng_consumer_local_data *ctx,
2567 struct lttng_consumer_stream *metadata_stream)
2568 {
2569 int ret;
2570 enum sync_metadata_status status;
2571 struct lttng_consumer_channel *metadata_channel;
2572
2573 assert(ctx);
2574 assert(metadata_stream);
2575
2576 metadata_channel = metadata_stream->chan;
2577 pthread_mutex_unlock(&metadata_stream->lock);
2578 /*
2579 * Request metadata from the sessiond, but don't wait for the flush
2580 * because we locked the metadata thread.
2581 */
2582 ret = lttng_ustconsumer_request_metadata(ctx, metadata_channel, 0, 0);
2583 pthread_mutex_lock(&metadata_stream->lock);
2584 if (ret < 0) {
2585 status = SYNC_METADATA_STATUS_ERROR;
2586 goto end;
2587 }
2588
2589 /*
2590 * The metadata stream and channel can be deleted while the
2591 * metadata stream lock was released. The streamed is checked
2592 * for deletion before we use it further.
2593 *
2594 * Note that it is safe to access a logically-deleted stream since its
2595 * existence is still guaranteed by the RCU read side lock. However,
2596 * it should no longer be used. The close/deletion of the metadata
2597 * channel and stream already guarantees that all metadata has been
2598 * consumed. Therefore, there is nothing left to do in this function.
2599 */
2600 if (consumer_stream_is_deleted(metadata_stream)) {
2601 DBG("Metadata stream %" PRIu64 " was deleted during the metadata synchronization",
2602 metadata_stream->key);
2603 status = SYNC_METADATA_STATUS_NO_DATA;
2604 goto end;
2605 }
2606
2607 ret = commit_one_metadata_packet(metadata_stream);
2608 if (ret < 0) {
2609 status = SYNC_METADATA_STATUS_ERROR;
2610 goto end;
2611 } else if (ret > 0) {
2612 status = SYNC_METADATA_STATUS_NEW_DATA;
2613 } else /* ret == 0 */ {
2614 status = SYNC_METADATA_STATUS_NO_DATA;
2615 goto end;
2616 }
2617
2618 ret = ustctl_snapshot(metadata_stream->ustream);
2619 if (ret < 0) {
2620 ERR("Failed to take a snapshot of the metadata ring-buffer positions, ret = %d", ret);
2621 status = SYNC_METADATA_STATUS_ERROR;
2622 goto end;
2623 }
2624
2625 end:
2626 return status;
2627 }
2628
2629 /*
2630 * Return 0 on success else a negative value.
2631 */
2632 static int notify_if_more_data(struct lttng_consumer_stream *stream,
2633 struct lttng_consumer_local_data *ctx)
2634 {
2635 int ret;
2636 struct ustctl_consumer_stream *ustream;
2637
2638 assert(stream);
2639 assert(ctx);
2640
2641 ustream = stream->ustream;
2642
2643 /*
2644 * First, we are going to check if there is a new subbuffer available
2645 * before reading the stream wait_fd.
2646 */
2647 /* Get the next subbuffer */
2648 ret = ustctl_get_next_subbuf(ustream);
2649 if (ret) {
2650 /* No more data found, flag the stream. */
2651 stream->has_data = 0;
2652 ret = 0;
2653 goto end;
2654 }
2655
2656 ret = ustctl_put_subbuf(ustream);
2657 assert(!ret);
2658
2659 /* This stream still has data. Flag it and wake up the data thread. */
2660 stream->has_data = 1;
2661
2662 if (stream->monitor && !stream->hangup_flush_done && !ctx->has_wakeup) {
2663 ssize_t writelen;
2664
2665 writelen = lttng_pipe_write(ctx->consumer_wakeup_pipe, "!", 1);
2666 if (writelen < 0 && errno != EAGAIN && errno != EWOULDBLOCK) {
2667 ret = writelen;
2668 goto end;
2669 }
2670
2671 /* The wake up pipe has been notified. */
2672 ctx->has_wakeup = 1;
2673 }
2674 ret = 0;
2675
2676 end:
2677 return ret;
2678 }
2679
2680 static int consumer_stream_ust_on_wake_up(struct lttng_consumer_stream *stream)
2681 {
2682 int ret = 0;
2683
2684 /*
2685 * We can consume the 1 byte written into the wait_fd by
2686 * UST. Don't trigger error if we cannot read this one byte
2687 * (read returns 0), or if the error is EAGAIN or EWOULDBLOCK.
2688 *
2689 * This is only done when the stream is monitored by a thread,
2690 * before the flush is done after a hangup and if the stream
2691 * is not flagged with data since there might be nothing to
2692 * consume in the wait fd but still have data available
2693 * flagged by the consumer wake up pipe.
2694 */
2695 if (stream->monitor && !stream->hangup_flush_done && !stream->has_data) {
2696 char dummy;
2697 ssize_t readlen;
2698
2699 readlen = lttng_read(stream->wait_fd, &dummy, 1);
2700 if (readlen < 0 && errno != EAGAIN && errno != EWOULDBLOCK) {
2701 ret = readlen;
2702 }
2703 }
2704
2705 return ret;
2706 }
2707
2708 static int extract_common_subbuffer_info(struct lttng_consumer_stream *stream,
2709 struct stream_subbuffer *subbuf)
2710 {
2711 int ret;
2712
2713 ret = ustctl_get_subbuf_size(
2714 stream->ustream, &subbuf->info.data.subbuf_size);
2715 if (ret) {
2716 goto end;
2717 }
2718
2719 ret = ustctl_get_padded_subbuf_size(
2720 stream->ustream, &subbuf->info.data.padded_subbuf_size);
2721 if (ret) {
2722 goto end;
2723 }
2724
2725 end:
2726 return ret;
2727 }
2728
2729 static int extract_metadata_subbuffer_info(struct lttng_consumer_stream *stream,
2730 struct stream_subbuffer *subbuf)
2731 {
2732 int ret;
2733
2734 ret = extract_common_subbuffer_info(stream, subbuf);
2735 if (ret) {
2736 goto end;
2737 }
2738
2739 subbuf->info.metadata.version = stream->metadata_version;
2740
2741 end:
2742 return ret;
2743 }
2744
2745 static int extract_data_subbuffer_info(struct lttng_consumer_stream *stream,
2746 struct stream_subbuffer *subbuf)
2747 {
2748 int ret;
2749
2750 ret = extract_common_subbuffer_info(stream, subbuf);
2751 if (ret) {
2752 goto end;
2753 }
2754
2755 ret = ustctl_get_packet_size(
2756 stream->ustream, &subbuf->info.data.packet_size);
2757 if (ret < 0) {
2758 PERROR("Failed to get sub-buffer packet size");
2759 goto end;
2760 }
2761
2762 ret = ustctl_get_content_size(
2763 stream->ustream, &subbuf->info.data.content_size);
2764 if (ret < 0) {
2765 PERROR("Failed to get sub-buffer content size");
2766 goto end;
2767 }
2768
2769 ret = ustctl_get_timestamp_begin(
2770 stream->ustream, &subbuf->info.data.timestamp_begin);
2771 if (ret < 0) {
2772 PERROR("Failed to get sub-buffer begin timestamp");
2773 goto end;
2774 }
2775
2776 ret = ustctl_get_timestamp_end(
2777 stream->ustream, &subbuf->info.data.timestamp_end);
2778 if (ret < 0) {
2779 PERROR("Failed to get sub-buffer end timestamp");
2780 goto end;
2781 }
2782
2783 ret = ustctl_get_events_discarded(
2784 stream->ustream, &subbuf->info.data.events_discarded);
2785 if (ret) {
2786 PERROR("Failed to get sub-buffer events discarded count");
2787 goto end;
2788 }
2789
2790 ret = ustctl_get_sequence_number(stream->ustream,
2791 &subbuf->info.data.sequence_number.value);
2792 if (ret) {
2793 /* May not be supported by older LTTng-modules. */
2794 if (ret != -ENOTTY) {
2795 PERROR("Failed to get sub-buffer sequence number");
2796 goto end;
2797 }
2798 } else {
2799 subbuf->info.data.sequence_number.is_set = true;
2800 }
2801
2802 ret = ustctl_get_stream_id(
2803 stream->ustream, &subbuf->info.data.stream_id);
2804 if (ret < 0) {
2805 PERROR("Failed to get stream id");
2806 goto end;
2807 }
2808
2809 ret = ustctl_get_instance_id(stream->ustream,
2810 &subbuf->info.data.stream_instance_id.value);
2811 if (ret) {
2812 /* May not be supported by older LTTng-modules. */
2813 if (ret != -ENOTTY) {
2814 PERROR("Failed to get stream instance id");
2815 goto end;
2816 }
2817 } else {
2818 subbuf->info.data.stream_instance_id.is_set = true;
2819 }
2820 end:
2821 return ret;
2822 }
2823
2824 static int get_next_subbuffer_common(struct lttng_consumer_stream *stream,
2825 struct stream_subbuffer *subbuffer)
2826 {
2827 int ret;
2828 const char *addr;
2829
2830 ret = stream->read_subbuffer_ops.extract_subbuffer_info(
2831 stream, subbuffer);
2832 if (ret) {
2833 goto end;
2834 }
2835
2836 ret = get_current_subbuf_addr(stream, &addr);
2837 if (ret) {
2838 goto end;
2839 }
2840
2841 subbuffer->buffer.buffer = lttng_buffer_view_init(
2842 addr, 0, subbuffer->info.data.padded_subbuf_size);
2843 assert(subbuffer->buffer.buffer.data != NULL);
2844 end:
2845 return ret;
2846 }
2847
2848 static int get_next_subbuffer(struct lttng_consumer_stream *stream,
2849 struct stream_subbuffer *subbuffer)
2850 {
2851 int ret;
2852
2853 ret = ustctl_get_next_subbuf(stream->ustream);
2854 if (ret) {
2855 goto end;
2856 }
2857
2858 ret = get_next_subbuffer_common(stream, subbuffer);
2859 if (ret) {
2860 goto end;
2861 }
2862 end:
2863 return ret;
2864 }
2865
2866 static int get_next_subbuffer_metadata(struct lttng_consumer_stream *stream,
2867 struct stream_subbuffer *subbuffer)
2868 {
2869 int ret;
2870 bool cache_empty;
2871 bool got_subbuffer;
2872 bool coherent;
2873 bool buffer_empty;
2874 unsigned long consumed_pos, produced_pos;
2875
2876 do {
2877 ret = ustctl_get_next_subbuf(stream->ustream);
2878 if (ret == 0) {
2879 got_subbuffer = true;
2880 } else {
2881 got_subbuffer = false;
2882 if (ret != -EAGAIN) {
2883 /* Fatal error. */
2884 goto end;
2885 }
2886 }
2887
2888 /*
2889 * Determine if the cache is empty and ensure that a sub-buffer
2890 * is made available if the cache is not empty.
2891 */
2892 if (!got_subbuffer) {
2893 ret = commit_one_metadata_packet(stream);
2894 if (ret < 0 && ret != -ENOBUFS) {
2895 goto end;
2896 } else if (ret == 0) {
2897 /* Not an error, the cache is empty. */
2898 cache_empty = true;
2899 ret = -ENODATA;
2900 goto end;
2901 } else {
2902 cache_empty = false;
2903 }
2904 } else {
2905 pthread_mutex_lock(&stream->chan->metadata_cache->lock);
2906 cache_empty = stream->chan->metadata_cache->max_offset ==
2907 stream->ust_metadata_pushed;
2908 pthread_mutex_unlock(&stream->chan->metadata_cache->lock);
2909 }
2910 } while (!got_subbuffer);
2911
2912 /* Populate sub-buffer infos and view. */
2913 ret = get_next_subbuffer_common(stream, subbuffer);
2914 if (ret) {
2915 goto end;
2916 }
2917
2918 ret = lttng_ustconsumer_sample_snapshot_positions(stream);
2919 if (ret < 0) {
2920 /*
2921 * -EAGAIN is not expected since we got a sub-buffer and haven't
2922 * pushed the consumption position yet (on put_next).
2923 */
2924 PERROR("Failed to take a snapshot of metadata buffer positions");
2925 goto end;
2926 }
2927
2928 ret = lttng_ustconsumer_get_consumed_snapshot(stream, &consumed_pos);
2929 if (ret) {
2930 PERROR("Failed to get metadata consumed position");
2931 goto end;
2932 }
2933
2934 ret = lttng_ustconsumer_get_produced_snapshot(stream, &produced_pos);
2935 if (ret) {
2936 PERROR("Failed to get metadata produced position");
2937 goto end;
2938 }
2939
2940 /* Last sub-buffer of the ring buffer ? */
2941 buffer_empty = (consumed_pos + stream->max_sb_size) == produced_pos;
2942
2943 /*
2944 * The sessiond registry lock ensures that coherent units of metadata
2945 * are pushed to the consumer daemon at once. Hence, if a sub-buffer is
2946 * acquired, the cache is empty, and it is the only available sub-buffer
2947 * available, it is safe to assume that it is "coherent".
2948 */
2949 coherent = got_subbuffer && cache_empty && buffer_empty;
2950
2951 LTTNG_OPTIONAL_SET(&subbuffer->info.metadata.coherent, coherent);
2952 end:
2953 return ret;
2954 }
2955
2956 static int put_next_subbuffer(struct lttng_consumer_stream *stream,
2957 struct stream_subbuffer *subbuffer)
2958 {
2959 const int ret = ustctl_put_next_subbuf(stream->ustream);
2960
2961 assert(ret == 0);
2962 return ret;
2963 }
2964
2965 static int signal_metadata(struct lttng_consumer_stream *stream,
2966 struct lttng_consumer_local_data *ctx)
2967 {
2968 ASSERT_LOCKED(stream->metadata_rdv_lock);
2969 return pthread_cond_broadcast(&stream->metadata_rdv) ? -errno : 0;
2970 }
2971
2972 static int lttng_ustconsumer_set_stream_ops(
2973 struct lttng_consumer_stream *stream)
2974 {
2975 int ret = 0;
2976
2977 stream->read_subbuffer_ops.on_wake_up = consumer_stream_ust_on_wake_up;
2978 if (stream->metadata_flag) {
2979 stream->read_subbuffer_ops.get_next_subbuffer =
2980 get_next_subbuffer_metadata;
2981 stream->read_subbuffer_ops.extract_subbuffer_info =
2982 extract_metadata_subbuffer_info;
2983 stream->read_subbuffer_ops.reset_metadata =
2984 metadata_stream_reset_cache_consumed_position;
2985 if (stream->chan->is_live) {
2986 stream->read_subbuffer_ops.on_sleep = signal_metadata;
2987 ret = consumer_stream_enable_metadata_bucketization(
2988 stream);
2989 if (ret) {
2990 goto end;
2991 }
2992 }
2993 } else {
2994 stream->read_subbuffer_ops.get_next_subbuffer =
2995 get_next_subbuffer;
2996 stream->read_subbuffer_ops.extract_subbuffer_info =
2997 extract_data_subbuffer_info;
2998 stream->read_subbuffer_ops.on_sleep = notify_if_more_data;
2999 if (stream->chan->is_live) {
3000 stream->read_subbuffer_ops.send_live_beacon =
3001 consumer_flush_ust_index;
3002 }
3003 }
3004
3005 stream->read_subbuffer_ops.put_next_subbuffer = put_next_subbuffer;
3006 end:
3007 return ret;
3008 }
3009
3010 /*
3011 * Called when a stream is created.
3012 *
3013 * Return 0 on success or else a negative value.
3014 */
3015 int lttng_ustconsumer_on_recv_stream(struct lttng_consumer_stream *stream)
3016 {
3017 int ret;
3018
3019 assert(stream);
3020
3021 /*
3022 * Don't create anything if this is set for streaming or if there is
3023 * no current trace chunk on the parent channel.
3024 */
3025 if (stream->net_seq_idx == (uint64_t) -1ULL && stream->chan->monitor &&
3026 stream->chan->trace_chunk) {
3027 ret = consumer_stream_create_output_files(stream, true);
3028 if (ret) {
3029 goto error;
3030 }
3031 }
3032
3033 lttng_ustconsumer_set_stream_ops(stream);
3034 ret = 0;
3035
3036 error:
3037 return ret;
3038 }
3039
3040 /*
3041 * Check if data is still being extracted from the buffers for a specific
3042 * stream. Consumer data lock MUST be acquired before calling this function
3043 * and the stream lock.
3044 *
3045 * Return 1 if the traced data are still getting read else 0 meaning that the
3046 * data is available for trace viewer reading.
3047 */
3048 int lttng_ustconsumer_data_pending(struct lttng_consumer_stream *stream)
3049 {
3050 int ret;
3051
3052 assert(stream);
3053 assert(stream->ustream);
3054
3055 DBG("UST consumer checking data pending");
3056
3057 if (stream->endpoint_status != CONSUMER_ENDPOINT_ACTIVE) {
3058 ret = 0;
3059 goto end;
3060 }
3061
3062 if (stream->chan->type == CONSUMER_CHANNEL_TYPE_METADATA) {
3063 uint64_t contiguous, pushed;
3064
3065 /* Ease our life a bit. */
3066 contiguous = stream->chan->metadata_cache->max_offset;
3067 pushed = stream->ust_metadata_pushed;
3068
3069 /*
3070 * We can simply check whether all contiguously available data
3071 * has been pushed to the ring buffer, since the push operation
3072 * is performed within get_next_subbuf(), and because both
3073 * get_next_subbuf() and put_next_subbuf() are issued atomically
3074 * thanks to the stream lock within
3075 * lttng_ustconsumer_read_subbuffer(). This basically means that
3076 * whetnever ust_metadata_pushed is incremented, the associated
3077 * metadata has been consumed from the metadata stream.
3078 */
3079 DBG("UST consumer metadata pending check: contiguous %" PRIu64 " vs pushed %" PRIu64,
3080 contiguous, pushed);
3081 assert(((int64_t) (contiguous - pushed)) >= 0);
3082 if ((contiguous != pushed) ||
3083 (((int64_t) contiguous - pushed) > 0 || contiguous == 0)) {
3084 ret = 1; /* Data is pending */
3085 goto end;
3086 }
3087 } else {
3088 ret = ustctl_get_next_subbuf(stream->ustream);
3089 if (ret == 0) {
3090 /*
3091 * There is still data so let's put back this
3092 * subbuffer.
3093 */
3094 ret = ustctl_put_subbuf(stream->ustream);
3095 assert(ret == 0);
3096 ret = 1; /* Data is pending */
3097 goto end;
3098 }
3099 }
3100
3101 /* Data is NOT pending so ready to be read. */
3102 ret = 0;
3103
3104 end:
3105 return ret;
3106 }
3107
3108 /*
3109 * Stop a given metadata channel timer if enabled and close the wait fd which
3110 * is the poll pipe of the metadata stream.
3111 *
3112 * This MUST be called with the metadata channel lock acquired.
3113 */
3114 void lttng_ustconsumer_close_metadata(struct lttng_consumer_channel *metadata)
3115 {
3116 int ret;
3117
3118 assert(metadata);
3119 assert(metadata->type == CONSUMER_CHANNEL_TYPE_METADATA);
3120
3121 DBG("Closing metadata channel key %" PRIu64, metadata->key);
3122
3123 if (metadata->switch_timer_enabled == 1) {
3124 consumer_timer_switch_stop(metadata);
3125 }
3126
3127 if (!metadata->metadata_stream) {
3128 goto end;
3129 }
3130
3131 /*
3132 * Closing write side so the thread monitoring the stream wakes up if any
3133 * and clean the metadata stream.
3134 */
3135 if (metadata->metadata_stream->ust_metadata_poll_pipe[1] >= 0) {
3136 ret = close(metadata->metadata_stream->ust_metadata_poll_pipe[1]);
3137 if (ret < 0) {
3138 PERROR("closing metadata pipe write side");
3139 }
3140 metadata->metadata_stream->ust_metadata_poll_pipe[1] = -1;
3141 }
3142
3143 end:
3144 return;
3145 }
3146
3147 /*
3148 * Close every metadata stream wait fd of the metadata hash table. This
3149 * function MUST be used very carefully so not to run into a race between the
3150 * metadata thread handling streams and this function closing their wait fd.
3151 *
3152 * For UST, this is used when the session daemon hangs up. Its the metadata
3153 * producer so calling this is safe because we are assured that no state change
3154 * can occur in the metadata thread for the streams in the hash table.
3155 */
3156 void lttng_ustconsumer_close_all_metadata(struct lttng_ht *metadata_ht)
3157 {
3158 struct lttng_ht_iter iter;
3159 struct lttng_consumer_stream *stream;
3160
3161 assert(metadata_ht);
3162 assert(metadata_ht->ht);
3163
3164 DBG("UST consumer closing all metadata streams");
3165
3166 rcu_read_lock();
3167 cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream,
3168 node.node) {
3169
3170 health_code_update();
3171
3172 pthread_mutex_lock(&stream->chan->lock);
3173 lttng_ustconsumer_close_metadata(stream->chan);
3174 pthread_mutex_unlock(&stream->chan->lock);
3175
3176 }
3177 rcu_read_unlock();
3178 }
3179
3180 void lttng_ustconsumer_close_stream_wakeup(struct lttng_consumer_stream *stream)
3181 {
3182 int ret;
3183
3184 ret = ustctl_stream_close_wakeup_fd(stream->ustream);
3185 if (ret < 0) {
3186 ERR("Unable to close wakeup fd");
3187 }
3188 }
3189
3190 /*
3191 * Please refer to consumer-timer.c before adding any lock within this
3192 * function or any of its callees. Timers have a very strict locking
3193 * semantic with respect to teardown. Failure to respect this semantic
3194 * introduces deadlocks.
3195 *
3196 * DON'T hold the metadata lock when calling this function, else this
3197 * can cause deadlock involving consumer awaiting for metadata to be
3198 * pushed out due to concurrent interaction with the session daemon.
3199 */
3200 int lttng_ustconsumer_request_metadata(struct lttng_consumer_local_data *ctx,
3201 struct lttng_consumer_channel *channel, int timer, int wait)
3202 {
3203 struct lttcomm_metadata_request_msg request;
3204 struct lttcomm_consumer_msg msg;
3205 enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS;
3206 uint64_t len, key, offset, version;
3207 int ret;
3208
3209 assert(channel);
3210 assert(channel->metadata_cache);
3211
3212 memset(&request, 0, sizeof(request));
3213
3214 /* send the metadata request to sessiond */
3215 switch (consumer_data.type) {
3216 case LTTNG_CONSUMER64_UST:
3217 request.bits_per_long = 64;
3218 break;
3219 case LTTNG_CONSUMER32_UST:
3220 request.bits_per_long = 32;
3221 break;
3222 default:
3223 request.bits_per_long = 0;
3224 break;
3225 }
3226
3227 request.session_id = channel->session_id;
3228 request.session_id_per_pid = channel->session_id_per_pid;
3229 /*
3230 * Request the application UID here so the metadata of that application can
3231 * be sent back. The channel UID corresponds to the user UID of the session
3232 * used for the rights on the stream file(s).
3233 */
3234 request.uid = channel->ust_app_uid;
3235 request.key = channel->key;
3236
3237 DBG("Sending metadata request to sessiond, session id %" PRIu64
3238 ", per-pid %" PRIu64 ", app UID %u and channel key %" PRIu64,
3239 request.session_id, request.session_id_per_pid, request.uid,
3240 request.key);
3241
3242 pthread_mutex_lock(&ctx->metadata_socket_lock);
3243
3244 health_code_update();
3245
3246 ret = lttcomm_send_unix_sock(ctx->consumer_metadata_socket, &request,
3247 sizeof(request));
3248 if (ret < 0) {
3249 ERR("Asking metadata to sessiond");
3250 goto end;
3251 }
3252
3253 health_code_update();
3254
3255 /* Receive the metadata from sessiond */
3256 ret = lttcomm_recv_unix_sock(ctx->consumer_metadata_socket, &msg,
3257 sizeof(msg));
3258 if (ret != sizeof(msg)) {
3259 DBG("Consumer received unexpected message size %d (expects %zu)",
3260 ret, sizeof(msg));
3261 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD);
3262 /*
3263 * The ret value might 0 meaning an orderly shutdown but this is ok
3264 * since the caller handles this.
3265 */
3266 goto end;
3267 }
3268
3269 health_code_update();
3270
3271 if (msg.cmd_type == LTTNG_ERR_UND) {
3272 /* No registry found */
3273 (void) consumer_send_status_msg(ctx->consumer_metadata_socket,
3274 ret_code);
3275 ret = 0;
3276 goto end;
3277 } else if (msg.cmd_type != LTTNG_CONSUMER_PUSH_METADATA) {
3278 ERR("Unexpected cmd_type received %d", msg.cmd_type);
3279 ret = -1;
3280 goto end;
3281 }
3282
3283 len = msg.u.push_metadata.len;
3284 key = msg.u.push_metadata.key;
3285 offset = msg.u.push_metadata.target_offset;
3286 version = msg.u.push_metadata.version;
3287
3288 assert(key == channel->key);
3289 if (len == 0) {
3290 DBG("No new metadata to receive for key %" PRIu64, key);
3291 }
3292
3293 health_code_update();
3294
3295 /* Tell session daemon we are ready to receive the metadata. */
3296 ret = consumer_send_status_msg(ctx->consumer_metadata_socket,
3297 LTTCOMM_CONSUMERD_SUCCESS);
3298 if (ret < 0 || len == 0) {
3299 /*
3300 * Somehow, the session daemon is not responding anymore or there is
3301 * nothing to receive.
3302 */
3303 goto end;
3304 }
3305
3306 health_code_update();
3307
3308 ret = lttng_ustconsumer_recv_metadata(ctx->consumer_metadata_socket,
3309 key, offset, len, version, channel, timer, wait);
3310 if (ret >= 0) {
3311 /*
3312 * Only send the status msg if the sessiond is alive meaning a positive
3313 * ret code.
3314 */
3315 (void) consumer_send_status_msg(ctx->consumer_metadata_socket, ret);
3316 }
3317 ret = 0;
3318
3319 end:
3320 health_code_update();
3321
3322 pthread_mutex_unlock(&ctx->metadata_socket_lock);
3323 return ret;
3324 }
3325
3326 /*
3327 * Return the ustctl call for the get stream id.
3328 */
3329 int lttng_ustconsumer_get_stream_id(struct lttng_consumer_stream *stream,
3330 uint64_t *stream_id)
3331 {
3332 assert(stream);
3333 assert(stream_id);
3334
3335 return ustctl_get_stream_id(stream->ustream, stream_id);
3336 }
This page took 0.141165 seconds and 5 git commands to generate.