Fix: consumer: rotation error return codes
[lttng-tools.git] / src / common / ust-consumer / ust-consumer.c
1 /*
2 * Copyright (C) 2011 - Julien Desfossez <julien.desfossez@polymtl.ca>
3 * Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
4 * Copyright (C) 2017 - Jérémie Galarneau <jeremie.galarneau@efficios.com>
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License, version 2 only,
8 * as published by the Free Software Foundation.
9 *
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
14 *
15 * You should have received a copy of the GNU General Public License along
16 * with this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
18 */
19
20 #define _LGPL_SOURCE
21 #include <assert.h>
22 #include <lttng/ust-ctl.h>
23 #include <poll.h>
24 #include <pthread.h>
25 #include <stdlib.h>
26 #include <string.h>
27 #include <sys/mman.h>
28 #include <sys/socket.h>
29 #include <sys/stat.h>
30 #include <sys/types.h>
31 #include <inttypes.h>
32 #include <unistd.h>
33 #include <urcu/list.h>
34 #include <signal.h>
35 #include <stdbool.h>
36
37 #include <bin/lttng-consumerd/health-consumerd.h>
38 #include <common/common.h>
39 #include <common/sessiond-comm/sessiond-comm.h>
40 #include <common/relayd/relayd.h>
41 #include <common/compat/fcntl.h>
42 #include <common/compat/endian.h>
43 #include <common/consumer/consumer-metadata-cache.h>
44 #include <common/consumer/consumer-stream.h>
45 #include <common/consumer/consumer-timer.h>
46 #include <common/utils.h>
47 #include <common/index/index.h>
48
49 #include "ust-consumer.h"
50
51 #define INT_MAX_STR_LEN 12 /* includes \0 */
52
53 extern struct lttng_consumer_global_data consumer_data;
54 extern int consumer_poll_timeout;
55
56 /*
57 * Free channel object and all streams associated with it. This MUST be used
58 * only and only if the channel has _NEVER_ been added to the global channel
59 * hash table.
60 */
61 static void destroy_channel(struct lttng_consumer_channel *channel)
62 {
63 struct lttng_consumer_stream *stream, *stmp;
64
65 assert(channel);
66
67 DBG("UST consumer cleaning stream list");
68
69 cds_list_for_each_entry_safe(stream, stmp, &channel->streams.head,
70 send_node) {
71
72 health_code_update();
73
74 cds_list_del(&stream->send_node);
75 ustctl_destroy_stream(stream->ustream);
76 free(stream);
77 }
78
79 /*
80 * If a channel is available meaning that was created before the streams
81 * were, delete it.
82 */
83 if (channel->uchan) {
84 lttng_ustconsumer_del_channel(channel);
85 lttng_ustconsumer_free_channel(channel);
86 }
87 free(channel);
88 }
89
90 /*
91 * Add channel to internal consumer state.
92 *
93 * Returns 0 on success or else a negative value.
94 */
95 static int add_channel(struct lttng_consumer_channel *channel,
96 struct lttng_consumer_local_data *ctx)
97 {
98 int ret = 0;
99
100 assert(channel);
101 assert(ctx);
102
103 if (ctx->on_recv_channel != NULL) {
104 ret = ctx->on_recv_channel(channel);
105 if (ret == 0) {
106 ret = consumer_add_channel(channel, ctx);
107 } else if (ret < 0) {
108 /* Most likely an ENOMEM. */
109 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
110 goto error;
111 }
112 } else {
113 ret = consumer_add_channel(channel, ctx);
114 }
115
116 DBG("UST consumer channel added (key: %" PRIu64 ")", channel->key);
117
118 error:
119 return ret;
120 }
121
122 /*
123 * Allocate and return a consumer channel object.
124 */
125 static struct lttng_consumer_channel *allocate_channel(uint64_t session_id,
126 const char *pathname, const char *name, uid_t uid, gid_t gid,
127 uint64_t relayd_id, uint64_t key, enum lttng_event_output output,
128 uint64_t tracefile_size, uint64_t tracefile_count,
129 uint64_t session_id_per_pid, unsigned int monitor,
130 unsigned int live_timer_interval,
131 const char *root_shm_path, const char *shm_path)
132 {
133 assert(pathname);
134 assert(name);
135
136 return consumer_allocate_channel(key, session_id, pathname, name, uid,
137 gid, relayd_id, output, tracefile_size,
138 tracefile_count, session_id_per_pid, monitor,
139 live_timer_interval, root_shm_path, shm_path);
140 }
141
142 /*
143 * Allocate and return a consumer stream object. If _alloc_ret is not NULL, the
144 * error value if applicable is set in it else it is kept untouched.
145 *
146 * Return NULL on error else the newly allocated stream object.
147 */
148 static struct lttng_consumer_stream *allocate_stream(int cpu, int key,
149 struct lttng_consumer_channel *channel,
150 struct lttng_consumer_local_data *ctx, int *_alloc_ret,
151 uint64_t trace_archive_id)
152 {
153 int alloc_ret;
154 struct lttng_consumer_stream *stream = NULL;
155
156 assert(channel);
157 assert(ctx);
158
159 stream = consumer_allocate_stream(channel->key,
160 key,
161 LTTNG_CONSUMER_ACTIVE_STREAM,
162 channel->name,
163 channel->uid,
164 channel->gid,
165 channel->relayd_id,
166 channel->session_id,
167 cpu,
168 &alloc_ret,
169 channel->type,
170 channel->monitor,
171 trace_archive_id);
172 if (stream == NULL) {
173 switch (alloc_ret) {
174 case -ENOENT:
175 /*
176 * We could not find the channel. Can happen if cpu hotplug
177 * happens while tearing down.
178 */
179 DBG3("Could not find channel");
180 break;
181 case -ENOMEM:
182 case -EINVAL:
183 default:
184 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
185 break;
186 }
187 goto error;
188 }
189
190 consumer_stream_update_channel_attributes(stream, channel);
191 stream->chan = channel;
192
193 error:
194 if (_alloc_ret) {
195 *_alloc_ret = alloc_ret;
196 }
197 return stream;
198 }
199
200 /*
201 * Send the given stream pointer to the corresponding thread.
202 *
203 * Returns 0 on success else a negative value.
204 */
205 static int send_stream_to_thread(struct lttng_consumer_stream *stream,
206 struct lttng_consumer_local_data *ctx)
207 {
208 int ret;
209 struct lttng_pipe *stream_pipe;
210
211 /* Get the right pipe where the stream will be sent. */
212 if (stream->metadata_flag) {
213 consumer_add_metadata_stream(stream);
214 stream_pipe = ctx->consumer_metadata_pipe;
215 } else {
216 consumer_add_data_stream(stream);
217 stream_pipe = ctx->consumer_data_pipe;
218 }
219
220 /*
221 * From this point on, the stream's ownership has been moved away from
222 * the channel and it becomes globally visible. Hence, remove it from
223 * the local stream list to prevent the stream from being both local and
224 * global.
225 */
226 stream->globally_visible = 1;
227 cds_list_del(&stream->send_node);
228
229 ret = lttng_pipe_write(stream_pipe, &stream, sizeof(stream));
230 if (ret < 0) {
231 ERR("Consumer write %s stream to pipe %d",
232 stream->metadata_flag ? "metadata" : "data",
233 lttng_pipe_get_writefd(stream_pipe));
234 if (stream->metadata_flag) {
235 consumer_del_stream_for_metadata(stream);
236 } else {
237 consumer_del_stream_for_data(stream);
238 }
239 goto error;
240 }
241
242 error:
243 return ret;
244 }
245
246 static
247 int get_stream_shm_path(char *stream_shm_path, const char *shm_path, int cpu)
248 {
249 char cpu_nr[INT_MAX_STR_LEN]; /* int max len */
250 int ret;
251
252 strncpy(stream_shm_path, shm_path, PATH_MAX);
253 stream_shm_path[PATH_MAX - 1] = '\0';
254 ret = snprintf(cpu_nr, INT_MAX_STR_LEN, "%i", cpu);
255 if (ret < 0) {
256 PERROR("snprintf");
257 goto end;
258 }
259 strncat(stream_shm_path, cpu_nr,
260 PATH_MAX - strlen(stream_shm_path) - 1);
261 ret = 0;
262 end:
263 return ret;
264 }
265
266 /*
267 * Create streams for the given channel using liblttng-ust-ctl.
268 *
269 * Return 0 on success else a negative value.
270 */
271 static int create_ust_streams(struct lttng_consumer_channel *channel,
272 struct lttng_consumer_local_data *ctx,
273 uint64_t trace_archive_id)
274 {
275 int ret, cpu = 0;
276 struct ustctl_consumer_stream *ustream;
277 struct lttng_consumer_stream *stream;
278
279 assert(channel);
280 assert(ctx);
281
282 /*
283 * While a stream is available from ustctl. When NULL is returned, we've
284 * reached the end of the possible stream for the channel.
285 */
286 while ((ustream = ustctl_create_stream(channel->uchan, cpu))) {
287 int wait_fd;
288 int ust_metadata_pipe[2];
289
290 health_code_update();
291
292 if (channel->type == CONSUMER_CHANNEL_TYPE_METADATA && channel->monitor) {
293 ret = utils_create_pipe_cloexec_nonblock(ust_metadata_pipe);
294 if (ret < 0) {
295 ERR("Create ust metadata poll pipe");
296 goto error;
297 }
298 wait_fd = ust_metadata_pipe[0];
299 } else {
300 wait_fd = ustctl_stream_get_wait_fd(ustream);
301 }
302
303 /* Allocate consumer stream object. */
304 stream = allocate_stream(cpu, wait_fd, channel, ctx, &ret,
305 trace_archive_id);
306 if (!stream) {
307 goto error_alloc;
308 }
309 stream->ustream = ustream;
310 /*
311 * Store it so we can save multiple function calls afterwards since
312 * this value is used heavily in the stream threads. This is UST
313 * specific so this is why it's done after allocation.
314 */
315 stream->wait_fd = wait_fd;
316
317 /*
318 * Increment channel refcount since the channel reference has now been
319 * assigned in the allocation process above.
320 */
321 if (stream->chan->monitor) {
322 uatomic_inc(&stream->chan->refcount);
323 }
324
325 /*
326 * Order is important this is why a list is used. On error, the caller
327 * should clean this list.
328 */
329 cds_list_add_tail(&stream->send_node, &channel->streams.head);
330
331 ret = ustctl_get_max_subbuf_size(stream->ustream,
332 &stream->max_sb_size);
333 if (ret < 0) {
334 ERR("ustctl_get_max_subbuf_size failed for stream %s",
335 stream->name);
336 goto error;
337 }
338
339 /* Do actions once stream has been received. */
340 if (ctx->on_recv_stream) {
341 ret = ctx->on_recv_stream(stream);
342 if (ret < 0) {
343 goto error;
344 }
345 }
346
347 DBG("UST consumer add stream %s (key: %" PRIu64 ") with relayd id %" PRIu64,
348 stream->name, stream->key, stream->relayd_stream_id);
349
350 /* Set next CPU stream. */
351 channel->streams.count = ++cpu;
352
353 /* Keep stream reference when creating metadata. */
354 if (channel->type == CONSUMER_CHANNEL_TYPE_METADATA) {
355 channel->metadata_stream = stream;
356 if (channel->monitor) {
357 /* Set metadata poll pipe if we created one */
358 memcpy(stream->ust_metadata_poll_pipe,
359 ust_metadata_pipe,
360 sizeof(ust_metadata_pipe));
361 }
362 }
363 }
364
365 return 0;
366
367 error:
368 error_alloc:
369 return ret;
370 }
371
372 /*
373 * create_posix_shm is never called concurrently within a process.
374 */
375 static
376 int create_posix_shm(void)
377 {
378 char tmp_name[NAME_MAX];
379 int shmfd, ret;
380
381 ret = snprintf(tmp_name, NAME_MAX, "/ust-shm-consumer-%d", getpid());
382 if (ret < 0) {
383 PERROR("snprintf");
384 return -1;
385 }
386 /*
387 * Allocate shm, and immediately unlink its shm oject, keeping
388 * only the file descriptor as a reference to the object.
389 * We specifically do _not_ use the / at the beginning of the
390 * pathname so that some OS implementations can keep it local to
391 * the process (POSIX leaves this implementation-defined).
392 */
393 shmfd = shm_open(tmp_name, O_CREAT | O_EXCL | O_RDWR, 0700);
394 if (shmfd < 0) {
395 PERROR("shm_open");
396 goto error_shm_open;
397 }
398 ret = shm_unlink(tmp_name);
399 if (ret < 0 && errno != ENOENT) {
400 PERROR("shm_unlink");
401 goto error_shm_release;
402 }
403 return shmfd;
404
405 error_shm_release:
406 ret = close(shmfd);
407 if (ret) {
408 PERROR("close");
409 }
410 error_shm_open:
411 return -1;
412 }
413
414 static int open_ust_stream_fd(struct lttng_consumer_channel *channel, int cpu)
415 {
416 char shm_path[PATH_MAX];
417 int ret;
418
419 if (!channel->shm_path[0]) {
420 return create_posix_shm();
421 }
422 ret = get_stream_shm_path(shm_path, channel->shm_path, cpu);
423 if (ret) {
424 goto error_shm_path;
425 }
426 return run_as_open(shm_path,
427 O_RDWR | O_CREAT | O_EXCL, S_IRUSR | S_IWUSR,
428 channel->uid, channel->gid);
429
430 error_shm_path:
431 return -1;
432 }
433
434 /*
435 * Create an UST channel with the given attributes and send it to the session
436 * daemon using the ust ctl API.
437 *
438 * Return 0 on success or else a negative value.
439 */
440 static int create_ust_channel(struct lttng_consumer_channel *channel,
441 struct ustctl_consumer_channel_attr *attr,
442 struct ustctl_consumer_channel **ust_chanp)
443 {
444 int ret, nr_stream_fds, i, j;
445 int *stream_fds;
446 struct ustctl_consumer_channel *ust_channel;
447
448 assert(channel);
449 assert(attr);
450 assert(ust_chanp);
451
452 DBG3("Creating channel to ustctl with attr: [overwrite: %d, "
453 "subbuf_size: %" PRIu64 ", num_subbuf: %" PRIu64 ", "
454 "switch_timer_interval: %u, read_timer_interval: %u, "
455 "output: %d, type: %d", attr->overwrite, attr->subbuf_size,
456 attr->num_subbuf, attr->switch_timer_interval,
457 attr->read_timer_interval, attr->output, attr->type);
458
459 if (channel->type == CONSUMER_CHANNEL_TYPE_METADATA)
460 nr_stream_fds = 1;
461 else
462 nr_stream_fds = ustctl_get_nr_stream_per_channel();
463 stream_fds = zmalloc(nr_stream_fds * sizeof(*stream_fds));
464 if (!stream_fds) {
465 ret = -1;
466 goto error_alloc;
467 }
468 for (i = 0; i < nr_stream_fds; i++) {
469 stream_fds[i] = open_ust_stream_fd(channel, i);
470 if (stream_fds[i] < 0) {
471 ret = -1;
472 goto error_open;
473 }
474 }
475 ust_channel = ustctl_create_channel(attr, stream_fds, nr_stream_fds);
476 if (!ust_channel) {
477 ret = -1;
478 goto error_create;
479 }
480 channel->nr_stream_fds = nr_stream_fds;
481 channel->stream_fds = stream_fds;
482 *ust_chanp = ust_channel;
483
484 return 0;
485
486 error_create:
487 error_open:
488 for (j = i - 1; j >= 0; j--) {
489 int closeret;
490
491 closeret = close(stream_fds[j]);
492 if (closeret) {
493 PERROR("close");
494 }
495 if (channel->shm_path[0]) {
496 char shm_path[PATH_MAX];
497
498 closeret = get_stream_shm_path(shm_path,
499 channel->shm_path, j);
500 if (closeret) {
501 ERR("Cannot get stream shm path");
502 }
503 closeret = run_as_unlink(shm_path,
504 channel->uid, channel->gid);
505 if (closeret) {
506 PERROR("unlink %s", shm_path);
507 }
508 }
509 }
510 /* Try to rmdir all directories under shm_path root. */
511 if (channel->root_shm_path[0]) {
512 (void) run_as_rmdir_recursive(channel->root_shm_path,
513 channel->uid, channel->gid);
514 }
515 free(stream_fds);
516 error_alloc:
517 return ret;
518 }
519
520 /*
521 * Send a single given stream to the session daemon using the sock.
522 *
523 * Return 0 on success else a negative value.
524 */
525 static int send_sessiond_stream(int sock, struct lttng_consumer_stream *stream)
526 {
527 int ret;
528
529 assert(stream);
530 assert(sock >= 0);
531
532 DBG("UST consumer sending stream %" PRIu64 " to sessiond", stream->key);
533
534 /* Send stream to session daemon. */
535 ret = ustctl_send_stream_to_sessiond(sock, stream->ustream);
536 if (ret < 0) {
537 goto error;
538 }
539
540 error:
541 return ret;
542 }
543
544 /*
545 * Send channel to sessiond and relayd if applicable.
546 *
547 * Return 0 on success or else a negative value.
548 */
549 static int send_channel_to_sessiond_and_relayd(int sock,
550 struct lttng_consumer_channel *channel,
551 struct lttng_consumer_local_data *ctx, int *relayd_error)
552 {
553 int ret, ret_code = LTTCOMM_CONSUMERD_SUCCESS;
554 struct lttng_consumer_stream *stream;
555 uint64_t net_seq_idx = -1ULL;
556
557 assert(channel);
558 assert(ctx);
559 assert(sock >= 0);
560
561 DBG("UST consumer sending channel %s to sessiond", channel->name);
562
563 if (channel->relayd_id != (uint64_t) -1ULL) {
564 cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
565
566 health_code_update();
567
568 /* Try to send the stream to the relayd if one is available. */
569 DBG("Sending stream %" PRIu64 " of channel \"%s\" to relayd",
570 stream->key, channel->name);
571 ret = consumer_send_relayd_stream(stream, stream->chan->pathname);
572 if (ret < 0) {
573 /*
574 * Flag that the relayd was the problem here probably due to a
575 * communicaton error on the socket.
576 */
577 if (relayd_error) {
578 *relayd_error = 1;
579 }
580 ret_code = LTTCOMM_CONSUMERD_RELAYD_FAIL;
581 }
582 if (net_seq_idx == -1ULL) {
583 net_seq_idx = stream->net_seq_idx;
584 }
585 }
586 }
587
588 /* Inform sessiond that we are about to send channel and streams. */
589 ret = consumer_send_status_msg(sock, ret_code);
590 if (ret < 0 || ret_code != LTTCOMM_CONSUMERD_SUCCESS) {
591 /*
592 * Either the session daemon is not responding or the relayd died so we
593 * stop now.
594 */
595 goto error;
596 }
597
598 /* Send channel to sessiond. */
599 ret = ustctl_send_channel_to_sessiond(sock, channel->uchan);
600 if (ret < 0) {
601 goto error;
602 }
603
604 ret = ustctl_channel_close_wakeup_fd(channel->uchan);
605 if (ret < 0) {
606 goto error;
607 }
608
609 /* The channel was sent successfully to the sessiond at this point. */
610 cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
611
612 health_code_update();
613
614 /* Send stream to session daemon. */
615 ret = send_sessiond_stream(sock, stream);
616 if (ret < 0) {
617 goto error;
618 }
619 }
620
621 /* Tell sessiond there is no more stream. */
622 ret = ustctl_send_stream_to_sessiond(sock, NULL);
623 if (ret < 0) {
624 goto error;
625 }
626
627 DBG("UST consumer NULL stream sent to sessiond");
628
629 return 0;
630
631 error:
632 if (ret_code != LTTCOMM_CONSUMERD_SUCCESS) {
633 ret = -1;
634 }
635 return ret;
636 }
637
638 /*
639 * Creates a channel and streams and add the channel it to the channel internal
640 * state. The created stream must ONLY be sent once the GET_CHANNEL command is
641 * received.
642 *
643 * Return 0 on success or else, a negative value is returned and the channel
644 * MUST be destroyed by consumer_del_channel().
645 */
646 static int ask_channel(struct lttng_consumer_local_data *ctx,
647 struct lttng_consumer_channel *channel,
648 struct ustctl_consumer_channel_attr *attr,
649 uint64_t trace_archive_id)
650 {
651 int ret;
652
653 assert(ctx);
654 assert(channel);
655 assert(attr);
656
657 /*
658 * This value is still used by the kernel consumer since for the kernel,
659 * the stream ownership is not IN the consumer so we need to have the
660 * number of left stream that needs to be initialized so we can know when
661 * to delete the channel (see consumer.c).
662 *
663 * As for the user space tracer now, the consumer creates and sends the
664 * stream to the session daemon which only sends them to the application
665 * once every stream of a channel is received making this value useless
666 * because we they will be added to the poll thread before the application
667 * receives them. This ensures that a stream can not hang up during
668 * initilization of a channel.
669 */
670 channel->nb_init_stream_left = 0;
671
672 /* The reply msg status is handled in the following call. */
673 ret = create_ust_channel(channel, attr, &channel->uchan);
674 if (ret < 0) {
675 goto end;
676 }
677
678 channel->wait_fd = ustctl_channel_get_wait_fd(channel->uchan);
679
680 /*
681 * For the snapshots (no monitor), we create the metadata streams
682 * on demand, not during the channel creation.
683 */
684 if (channel->type == CONSUMER_CHANNEL_TYPE_METADATA && !channel->monitor) {
685 ret = 0;
686 goto end;
687 }
688
689 /* Open all streams for this channel. */
690 ret = create_ust_streams(channel, ctx, trace_archive_id);
691 if (ret < 0) {
692 goto end;
693 }
694
695 end:
696 return ret;
697 }
698
699 /*
700 * Send all stream of a channel to the right thread handling it.
701 *
702 * On error, return a negative value else 0 on success.
703 */
704 static int send_streams_to_thread(struct lttng_consumer_channel *channel,
705 struct lttng_consumer_local_data *ctx)
706 {
707 int ret = 0;
708 struct lttng_consumer_stream *stream, *stmp;
709
710 assert(channel);
711 assert(ctx);
712
713 /* Send streams to the corresponding thread. */
714 cds_list_for_each_entry_safe(stream, stmp, &channel->streams.head,
715 send_node) {
716
717 health_code_update();
718
719 /* Sending the stream to the thread. */
720 ret = send_stream_to_thread(stream, ctx);
721 if (ret < 0) {
722 /*
723 * If we are unable to send the stream to the thread, there is
724 * a big problem so just stop everything.
725 */
726 goto error;
727 }
728 }
729
730 error:
731 return ret;
732 }
733
734 /*
735 * Flush channel's streams using the given key to retrieve the channel.
736 *
737 * Return 0 on success else an LTTng error code.
738 */
739 static int flush_channel(uint64_t chan_key)
740 {
741 int ret = 0;
742 struct lttng_consumer_channel *channel;
743 struct lttng_consumer_stream *stream;
744 struct lttng_ht *ht;
745 struct lttng_ht_iter iter;
746
747 DBG("UST consumer flush channel key %" PRIu64, chan_key);
748
749 rcu_read_lock();
750 channel = consumer_find_channel(chan_key);
751 if (!channel) {
752 ERR("UST consumer flush channel %" PRIu64 " not found", chan_key);
753 ret = LTTNG_ERR_UST_CHAN_NOT_FOUND;
754 goto error;
755 }
756
757 ht = consumer_data.stream_per_chan_id_ht;
758
759 /* For each stream of the channel id, flush it. */
760 cds_lfht_for_each_entry_duplicate(ht->ht,
761 ht->hash_fct(&channel->key, lttng_ht_seed), ht->match_fct,
762 &channel->key, &iter.iter, stream, node_channel_id.node) {
763
764 health_code_update();
765
766 pthread_mutex_lock(&stream->lock);
767 if (!stream->quiescent) {
768 ustctl_flush_buffer(stream->ustream, 0);
769 stream->quiescent = true;
770 }
771 pthread_mutex_unlock(&stream->lock);
772 }
773 error:
774 rcu_read_unlock();
775 return ret;
776 }
777
778 /*
779 * Clear quiescent state from channel's streams using the given key to
780 * retrieve the channel.
781 *
782 * Return 0 on success else an LTTng error code.
783 */
784 static int clear_quiescent_channel(uint64_t chan_key)
785 {
786 int ret = 0;
787 struct lttng_consumer_channel *channel;
788 struct lttng_consumer_stream *stream;
789 struct lttng_ht *ht;
790 struct lttng_ht_iter iter;
791
792 DBG("UST consumer clear quiescent channel key %" PRIu64, chan_key);
793
794 rcu_read_lock();
795 channel = consumer_find_channel(chan_key);
796 if (!channel) {
797 ERR("UST consumer clear quiescent channel %" PRIu64 " not found", chan_key);
798 ret = LTTNG_ERR_UST_CHAN_NOT_FOUND;
799 goto error;
800 }
801
802 ht = consumer_data.stream_per_chan_id_ht;
803
804 /* For each stream of the channel id, clear quiescent state. */
805 cds_lfht_for_each_entry_duplicate(ht->ht,
806 ht->hash_fct(&channel->key, lttng_ht_seed), ht->match_fct,
807 &channel->key, &iter.iter, stream, node_channel_id.node) {
808
809 health_code_update();
810
811 pthread_mutex_lock(&stream->lock);
812 stream->quiescent = false;
813 pthread_mutex_unlock(&stream->lock);
814 }
815 error:
816 rcu_read_unlock();
817 return ret;
818 }
819
820 /*
821 * Close metadata stream wakeup_fd using the given key to retrieve the channel.
822 * RCU read side lock MUST be acquired before calling this function.
823 *
824 * Return 0 on success else an LTTng error code.
825 */
826 static int close_metadata(uint64_t chan_key)
827 {
828 int ret = 0;
829 struct lttng_consumer_channel *channel;
830 unsigned int channel_monitor;
831
832 DBG("UST consumer close metadata key %" PRIu64, chan_key);
833
834 channel = consumer_find_channel(chan_key);
835 if (!channel) {
836 /*
837 * This is possible if the metadata thread has issue a delete because
838 * the endpoint point of the stream hung up. There is no way the
839 * session daemon can know about it thus use a DBG instead of an actual
840 * error.
841 */
842 DBG("UST consumer close metadata %" PRIu64 " not found", chan_key);
843 ret = LTTNG_ERR_UST_CHAN_NOT_FOUND;
844 goto error;
845 }
846
847 pthread_mutex_lock(&consumer_data.lock);
848 pthread_mutex_lock(&channel->lock);
849 channel_monitor = channel->monitor;
850 if (cds_lfht_is_node_deleted(&channel->node.node)) {
851 goto error_unlock;
852 }
853
854 lttng_ustconsumer_close_metadata(channel);
855 pthread_mutex_unlock(&channel->lock);
856 pthread_mutex_unlock(&consumer_data.lock);
857
858 /*
859 * The ownership of a metadata channel depends on the type of
860 * session to which it belongs. In effect, the monitor flag is checked
861 * to determine if this metadata channel is in "snapshot" mode or not.
862 *
863 * In the non-snapshot case, the metadata channel is created along with
864 * a single stream which will remain present until the metadata channel
865 * is destroyed (on the destruction of its session). In this case, the
866 * metadata stream in "monitored" by the metadata poll thread and holds
867 * the ownership of its channel.
868 *
869 * Closing the metadata will cause the metadata stream's "metadata poll
870 * pipe" to be closed. Closing this pipe will wake-up the metadata poll
871 * thread which will teardown the metadata stream which, in return,
872 * deletes the metadata channel.
873 *
874 * In the snapshot case, the metadata stream is created and destroyed
875 * on every snapshot record. Since the channel doesn't have an owner
876 * other than the session daemon, it is safe to destroy it immediately
877 * on reception of the CLOSE_METADATA command.
878 */
879 if (!channel_monitor) {
880 /*
881 * The channel and consumer_data locks must be
882 * released before this call since consumer_del_channel
883 * re-acquires the channel and consumer_data locks to teardown
884 * the channel and queue its reclamation by the "call_rcu"
885 * worker thread.
886 */
887 consumer_del_channel(channel);
888 }
889
890 return ret;
891 error_unlock:
892 pthread_mutex_unlock(&channel->lock);
893 pthread_mutex_unlock(&consumer_data.lock);
894 error:
895 return ret;
896 }
897
898 /*
899 * RCU read side lock MUST be acquired before calling this function.
900 *
901 * Return 0 on success else an LTTng error code.
902 */
903 static int setup_metadata(struct lttng_consumer_local_data *ctx, uint64_t key)
904 {
905 int ret;
906 struct lttng_consumer_channel *metadata;
907
908 DBG("UST consumer setup metadata key %" PRIu64, key);
909
910 metadata = consumer_find_channel(key);
911 if (!metadata) {
912 ERR("UST consumer push metadata %" PRIu64 " not found", key);
913 ret = LTTNG_ERR_UST_CHAN_NOT_FOUND;
914 goto end;
915 }
916
917 /*
918 * In no monitor mode, the metadata channel has no stream(s) so skip the
919 * ownership transfer to the metadata thread.
920 */
921 if (!metadata->monitor) {
922 DBG("Metadata channel in no monitor");
923 ret = 0;
924 goto end;
925 }
926
927 /*
928 * Send metadata stream to relayd if one available. Availability is
929 * known if the stream is still in the list of the channel.
930 */
931 if (cds_list_empty(&metadata->streams.head)) {
932 ERR("Metadata channel key %" PRIu64 ", no stream available.", key);
933 ret = LTTCOMM_CONSUMERD_ERROR_METADATA;
934 goto error_no_stream;
935 }
936
937 /* Send metadata stream to relayd if needed. */
938 if (metadata->metadata_stream->net_seq_idx != (uint64_t) -1ULL) {
939 ret = consumer_send_relayd_stream(metadata->metadata_stream,
940 metadata->pathname);
941 if (ret < 0) {
942 ret = LTTCOMM_CONSUMERD_ERROR_METADATA;
943 goto error;
944 }
945 ret = consumer_send_relayd_streams_sent(
946 metadata->metadata_stream->net_seq_idx);
947 if (ret < 0) {
948 ret = LTTCOMM_CONSUMERD_RELAYD_FAIL;
949 goto error;
950 }
951 }
952
953 /*
954 * Ownership of metadata stream is passed along. Freeing is handled by
955 * the callee.
956 */
957 ret = send_streams_to_thread(metadata, ctx);
958 if (ret < 0) {
959 /*
960 * If we are unable to send the stream to the thread, there is
961 * a big problem so just stop everything.
962 */
963 ret = LTTCOMM_CONSUMERD_FATAL;
964 goto send_streams_error;
965 }
966 /* List MUST be empty after or else it could be reused. */
967 assert(cds_list_empty(&metadata->streams.head));
968
969 ret = 0;
970 goto end;
971
972 error:
973 /*
974 * Delete metadata channel on error. At this point, the metadata stream can
975 * NOT be monitored by the metadata thread thus having the guarantee that
976 * the stream is still in the local stream list of the channel. This call
977 * will make sure to clean that list.
978 */
979 consumer_stream_destroy(metadata->metadata_stream, NULL);
980 cds_list_del(&metadata->metadata_stream->send_node);
981 metadata->metadata_stream = NULL;
982 send_streams_error:
983 error_no_stream:
984 end:
985 return ret;
986 }
987
988 /*
989 * Snapshot the whole metadata.
990 *
991 * Returns 0 on success, < 0 on error
992 */
993 static int snapshot_metadata(uint64_t key, char *path, uint64_t relayd_id,
994 struct lttng_consumer_local_data *ctx,
995 uint64_t trace_archive_id)
996 {
997 int ret = 0;
998 struct lttng_consumer_channel *metadata_channel;
999 struct lttng_consumer_stream *metadata_stream;
1000
1001 assert(path);
1002 assert(ctx);
1003
1004 DBG("UST consumer snapshot metadata with key %" PRIu64 " at path %s",
1005 key, path);
1006
1007 rcu_read_lock();
1008
1009 metadata_channel = consumer_find_channel(key);
1010 if (!metadata_channel) {
1011 ERR("UST snapshot metadata channel not found for key %" PRIu64,
1012 key);
1013 ret = -1;
1014 goto error;
1015 }
1016 assert(!metadata_channel->monitor);
1017
1018 health_code_update();
1019
1020 /*
1021 * Ask the sessiond if we have new metadata waiting and update the
1022 * consumer metadata cache.
1023 */
1024 ret = lttng_ustconsumer_request_metadata(ctx, metadata_channel, 0, 1);
1025 if (ret < 0) {
1026 goto error;
1027 }
1028
1029 health_code_update();
1030
1031 /*
1032 * The metadata stream is NOT created in no monitor mode when the channel
1033 * is created on a sessiond ask channel command.
1034 */
1035 ret = create_ust_streams(metadata_channel, ctx, trace_archive_id);
1036 if (ret < 0) {
1037 goto error;
1038 }
1039
1040 metadata_stream = metadata_channel->metadata_stream;
1041 assert(metadata_stream);
1042
1043 if (relayd_id != (uint64_t) -1ULL) {
1044 metadata_stream->net_seq_idx = relayd_id;
1045 ret = consumer_send_relayd_stream(metadata_stream, path);
1046 if (ret < 0) {
1047 goto error_stream;
1048 }
1049 } else {
1050 ret = utils_create_stream_file(path, metadata_stream->name,
1051 metadata_stream->chan->tracefile_size,
1052 metadata_stream->tracefile_count_current,
1053 metadata_stream->uid, metadata_stream->gid, NULL);
1054 if (ret < 0) {
1055 goto error_stream;
1056 }
1057 metadata_stream->out_fd = ret;
1058 metadata_stream->tracefile_size_current = 0;
1059 }
1060
1061 do {
1062 health_code_update();
1063
1064 ret = lttng_consumer_read_subbuffer(metadata_stream, ctx);
1065 if (ret < 0) {
1066 goto error_stream;
1067 }
1068 } while (ret > 0);
1069
1070 error_stream:
1071 /*
1072 * Clean up the stream completly because the next snapshot will use a new
1073 * metadata stream.
1074 */
1075 consumer_stream_destroy(metadata_stream, NULL);
1076 cds_list_del(&metadata_stream->send_node);
1077 metadata_channel->metadata_stream = NULL;
1078
1079 error:
1080 rcu_read_unlock();
1081 return ret;
1082 }
1083
1084 /*
1085 * Take a snapshot of all the stream of a channel.
1086 *
1087 * Returns 0 on success, < 0 on error
1088 */
1089 static int snapshot_channel(uint64_t key, char *path, uint64_t relayd_id,
1090 uint64_t nb_packets_per_stream,
1091 struct lttng_consumer_local_data *ctx)
1092 {
1093 int ret;
1094 unsigned use_relayd = 0;
1095 unsigned long consumed_pos, produced_pos;
1096 struct lttng_consumer_channel *channel;
1097 struct lttng_consumer_stream *stream;
1098
1099 assert(path);
1100 assert(ctx);
1101
1102 rcu_read_lock();
1103
1104 if (relayd_id != (uint64_t) -1ULL) {
1105 use_relayd = 1;
1106 }
1107
1108 channel = consumer_find_channel(key);
1109 if (!channel) {
1110 ERR("UST snapshot channel not found for key %" PRIu64, key);
1111 ret = -1;
1112 goto error;
1113 }
1114 assert(!channel->monitor);
1115 DBG("UST consumer snapshot channel %" PRIu64, key);
1116
1117 cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
1118 health_code_update();
1119
1120 /* Lock stream because we are about to change its state. */
1121 pthread_mutex_lock(&stream->lock);
1122 stream->net_seq_idx = relayd_id;
1123
1124 if (use_relayd) {
1125 ret = consumer_send_relayd_stream(stream, path);
1126 if (ret < 0) {
1127 goto error_unlock;
1128 }
1129 } else {
1130 ret = utils_create_stream_file(path, stream->name,
1131 stream->chan->tracefile_size,
1132 stream->tracefile_count_current,
1133 stream->uid, stream->gid, NULL);
1134 if (ret < 0) {
1135 goto error_unlock;
1136 }
1137 stream->out_fd = ret;
1138 stream->tracefile_size_current = 0;
1139
1140 DBG("UST consumer snapshot stream %s/%s (%" PRIu64 ")", path,
1141 stream->name, stream->key);
1142 }
1143
1144 /*
1145 * If tracing is active, we want to perform a "full" buffer flush.
1146 * Else, if quiescent, it has already been done by the prior stop.
1147 */
1148 if (!stream->quiescent) {
1149 ustctl_flush_buffer(stream->ustream, 0);
1150 }
1151
1152 ret = lttng_ustconsumer_take_snapshot(stream);
1153 if (ret < 0) {
1154 ERR("Taking UST snapshot");
1155 goto error_unlock;
1156 }
1157
1158 ret = lttng_ustconsumer_get_produced_snapshot(stream, &produced_pos);
1159 if (ret < 0) {
1160 ERR("Produced UST snapshot position");
1161 goto error_unlock;
1162 }
1163
1164 ret = lttng_ustconsumer_get_consumed_snapshot(stream, &consumed_pos);
1165 if (ret < 0) {
1166 ERR("Consumerd UST snapshot position");
1167 goto error_unlock;
1168 }
1169
1170 /*
1171 * The original value is sent back if max stream size is larger than
1172 * the possible size of the snapshot. Also, we assume that the session
1173 * daemon should never send a maximum stream size that is lower than
1174 * subbuffer size.
1175 */
1176 consumed_pos = consumer_get_consume_start_pos(consumed_pos,
1177 produced_pos, nb_packets_per_stream,
1178 stream->max_sb_size);
1179
1180 while (consumed_pos < produced_pos) {
1181 ssize_t read_len;
1182 unsigned long len, padded_len;
1183
1184 health_code_update();
1185
1186 DBG("UST consumer taking snapshot at pos %lu", consumed_pos);
1187
1188 ret = ustctl_get_subbuf(stream->ustream, &consumed_pos);
1189 if (ret < 0) {
1190 if (ret != -EAGAIN) {
1191 PERROR("ustctl_get_subbuf snapshot");
1192 goto error_close_stream;
1193 }
1194 DBG("UST consumer get subbuf failed. Skipping it.");
1195 consumed_pos += stream->max_sb_size;
1196 stream->chan->lost_packets++;
1197 continue;
1198 }
1199
1200 ret = ustctl_get_subbuf_size(stream->ustream, &len);
1201 if (ret < 0) {
1202 ERR("Snapshot ustctl_get_subbuf_size");
1203 goto error_put_subbuf;
1204 }
1205
1206 ret = ustctl_get_padded_subbuf_size(stream->ustream, &padded_len);
1207 if (ret < 0) {
1208 ERR("Snapshot ustctl_get_padded_subbuf_size");
1209 goto error_put_subbuf;
1210 }
1211
1212 read_len = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, len,
1213 padded_len - len, NULL);
1214 if (use_relayd) {
1215 if (read_len != len) {
1216 ret = -EPERM;
1217 goto error_put_subbuf;
1218 }
1219 } else {
1220 if (read_len != padded_len) {
1221 ret = -EPERM;
1222 goto error_put_subbuf;
1223 }
1224 }
1225
1226 ret = ustctl_put_subbuf(stream->ustream);
1227 if (ret < 0) {
1228 ERR("Snapshot ustctl_put_subbuf");
1229 goto error_close_stream;
1230 }
1231 consumed_pos += stream->max_sb_size;
1232 }
1233
1234 /* Simply close the stream so we can use it on the next snapshot. */
1235 consumer_stream_close(stream);
1236 pthread_mutex_unlock(&stream->lock);
1237 }
1238
1239 rcu_read_unlock();
1240 return 0;
1241
1242 error_put_subbuf:
1243 if (ustctl_put_subbuf(stream->ustream) < 0) {
1244 ERR("Snapshot ustctl_put_subbuf");
1245 }
1246 error_close_stream:
1247 consumer_stream_close(stream);
1248 error_unlock:
1249 pthread_mutex_unlock(&stream->lock);
1250 error:
1251 rcu_read_unlock();
1252 return ret;
1253 }
1254
1255 /*
1256 * Receive the metadata updates from the sessiond. Supports receiving
1257 * overlapping metadata, but is needs to always belong to a contiguous
1258 * range starting from 0.
1259 * Be careful about the locks held when calling this function: it needs
1260 * the metadata cache flush to concurrently progress in order to
1261 * complete.
1262 */
1263 int lttng_ustconsumer_recv_metadata(int sock, uint64_t key, uint64_t offset,
1264 uint64_t len, uint64_t version,
1265 struct lttng_consumer_channel *channel, int timer, int wait)
1266 {
1267 int ret, ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1268 char *metadata_str;
1269
1270 DBG("UST consumer push metadata key %" PRIu64 " of len %" PRIu64, key, len);
1271
1272 metadata_str = zmalloc(len * sizeof(char));
1273 if (!metadata_str) {
1274 PERROR("zmalloc metadata string");
1275 ret_code = LTTCOMM_CONSUMERD_ENOMEM;
1276 goto end;
1277 }
1278
1279 health_code_update();
1280
1281 /* Receive metadata string. */
1282 ret = lttcomm_recv_unix_sock(sock, metadata_str, len);
1283 if (ret < 0) {
1284 /* Session daemon is dead so return gracefully. */
1285 ret_code = ret;
1286 goto end_free;
1287 }
1288
1289 health_code_update();
1290
1291 pthread_mutex_lock(&channel->metadata_cache->lock);
1292 ret = consumer_metadata_cache_write(channel, offset, len, version,
1293 metadata_str);
1294 if (ret < 0) {
1295 /* Unable to handle metadata. Notify session daemon. */
1296 ret_code = LTTCOMM_CONSUMERD_ERROR_METADATA;
1297 /*
1298 * Skip metadata flush on write error since the offset and len might
1299 * not have been updated which could create an infinite loop below when
1300 * waiting for the metadata cache to be flushed.
1301 */
1302 pthread_mutex_unlock(&channel->metadata_cache->lock);
1303 goto end_free;
1304 }
1305 pthread_mutex_unlock(&channel->metadata_cache->lock);
1306
1307 if (!wait) {
1308 goto end_free;
1309 }
1310 while (consumer_metadata_cache_flushed(channel, offset + len, timer)) {
1311 DBG("Waiting for metadata to be flushed");
1312
1313 health_code_update();
1314
1315 usleep(DEFAULT_METADATA_AVAILABILITY_WAIT_TIME);
1316 }
1317
1318 end_free:
1319 free(metadata_str);
1320 end:
1321 return ret_code;
1322 }
1323
1324 /*
1325 * Receive command from session daemon and process it.
1326 *
1327 * Return 1 on success else a negative value or 0.
1328 */
1329 int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
1330 int sock, struct pollfd *consumer_sockpoll)
1331 {
1332 ssize_t ret;
1333 enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1334 struct lttcomm_consumer_msg msg;
1335 struct lttng_consumer_channel *channel = NULL;
1336
1337 health_code_update();
1338
1339 ret = lttcomm_recv_unix_sock(sock, &msg, sizeof(msg));
1340 if (ret != sizeof(msg)) {
1341 DBG("Consumer received unexpected message size %zd (expects %zu)",
1342 ret, sizeof(msg));
1343 /*
1344 * The ret value might 0 meaning an orderly shutdown but this is ok
1345 * since the caller handles this.
1346 */
1347 if (ret > 0) {
1348 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD);
1349 ret = -1;
1350 }
1351 return ret;
1352 }
1353
1354 health_code_update();
1355
1356 /* deprecated */
1357 assert(msg.cmd_type != LTTNG_CONSUMER_STOP);
1358
1359 health_code_update();
1360
1361 /* relayd needs RCU read-side lock */
1362 rcu_read_lock();
1363
1364 switch (msg.cmd_type) {
1365 case LTTNG_CONSUMER_ADD_RELAYD_SOCKET:
1366 {
1367 /* Session daemon status message are handled in the following call. */
1368 consumer_add_relayd_socket(msg.u.relayd_sock.net_index,
1369 msg.u.relayd_sock.type, ctx, sock, consumer_sockpoll,
1370 &msg.u.relayd_sock.sock, msg.u.relayd_sock.session_id,
1371 msg.u.relayd_sock.relayd_session_id);
1372 goto end_nosignal;
1373 }
1374 case LTTNG_CONSUMER_DESTROY_RELAYD:
1375 {
1376 uint64_t index = msg.u.destroy_relayd.net_seq_idx;
1377 struct consumer_relayd_sock_pair *relayd;
1378
1379 DBG("UST consumer destroying relayd %" PRIu64, index);
1380
1381 /* Get relayd reference if exists. */
1382 relayd = consumer_find_relayd(index);
1383 if (relayd == NULL) {
1384 DBG("Unable to find relayd %" PRIu64, index);
1385 ret_code = LTTCOMM_CONSUMERD_RELAYD_FAIL;
1386 }
1387
1388 /*
1389 * Each relayd socket pair has a refcount of stream attached to it
1390 * which tells if the relayd is still active or not depending on the
1391 * refcount value.
1392 *
1393 * This will set the destroy flag of the relayd object and destroy it
1394 * if the refcount reaches zero when called.
1395 *
1396 * The destroy can happen either here or when a stream fd hangs up.
1397 */
1398 if (relayd) {
1399 consumer_flag_relayd_for_destroy(relayd);
1400 }
1401
1402 goto end_msg_sessiond;
1403 }
1404 case LTTNG_CONSUMER_UPDATE_STREAM:
1405 {
1406 rcu_read_unlock();
1407 return -ENOSYS;
1408 }
1409 case LTTNG_CONSUMER_DATA_PENDING:
1410 {
1411 int ret, is_data_pending;
1412 uint64_t id = msg.u.data_pending.session_id;
1413
1414 DBG("UST consumer data pending command for id %" PRIu64, id);
1415
1416 is_data_pending = consumer_data_pending(id);
1417
1418 /* Send back returned value to session daemon */
1419 ret = lttcomm_send_unix_sock(sock, &is_data_pending,
1420 sizeof(is_data_pending));
1421 if (ret < 0) {
1422 DBG("Error when sending the data pending ret code: %d", ret);
1423 goto error_fatal;
1424 }
1425
1426 /*
1427 * No need to send back a status message since the data pending
1428 * returned value is the response.
1429 */
1430 break;
1431 }
1432 case LTTNG_CONSUMER_ASK_CHANNEL_CREATION:
1433 {
1434 int ret;
1435 struct ustctl_consumer_channel_attr attr;
1436
1437 /* Create a plain object and reserve a channel key. */
1438 channel = allocate_channel(msg.u.ask_channel.session_id,
1439 msg.u.ask_channel.pathname, msg.u.ask_channel.name,
1440 msg.u.ask_channel.uid, msg.u.ask_channel.gid,
1441 msg.u.ask_channel.relayd_id, msg.u.ask_channel.key,
1442 (enum lttng_event_output) msg.u.ask_channel.output,
1443 msg.u.ask_channel.tracefile_size,
1444 msg.u.ask_channel.tracefile_count,
1445 msg.u.ask_channel.session_id_per_pid,
1446 msg.u.ask_channel.monitor,
1447 msg.u.ask_channel.live_timer_interval,
1448 msg.u.ask_channel.root_shm_path,
1449 msg.u.ask_channel.shm_path);
1450 if (!channel) {
1451 goto end_channel_error;
1452 }
1453
1454 /*
1455 * Assign UST application UID to the channel. This value is ignored for
1456 * per PID buffers. This is specific to UST thus setting this after the
1457 * allocation.
1458 */
1459 channel->ust_app_uid = msg.u.ask_channel.ust_app_uid;
1460
1461 /* Build channel attributes from received message. */
1462 attr.subbuf_size = msg.u.ask_channel.subbuf_size;
1463 attr.num_subbuf = msg.u.ask_channel.num_subbuf;
1464 attr.overwrite = msg.u.ask_channel.overwrite;
1465 attr.switch_timer_interval = msg.u.ask_channel.switch_timer_interval;
1466 attr.read_timer_interval = msg.u.ask_channel.read_timer_interval;
1467 attr.chan_id = msg.u.ask_channel.chan_id;
1468 memcpy(attr.uuid, msg.u.ask_channel.uuid, sizeof(attr.uuid));
1469 attr.blocking_timeout= msg.u.ask_channel.blocking_timeout;
1470
1471 /* Match channel buffer type to the UST abi. */
1472 switch (msg.u.ask_channel.output) {
1473 case LTTNG_EVENT_MMAP:
1474 default:
1475 attr.output = LTTNG_UST_MMAP;
1476 break;
1477 }
1478
1479 /* Translate and save channel type. */
1480 switch (msg.u.ask_channel.type) {
1481 case LTTNG_UST_CHAN_PER_CPU:
1482 channel->type = CONSUMER_CHANNEL_TYPE_DATA;
1483 attr.type = LTTNG_UST_CHAN_PER_CPU;
1484 /*
1485 * Set refcount to 1 for owner. Below, we will
1486 * pass ownership to the
1487 * consumer_thread_channel_poll() thread.
1488 */
1489 channel->refcount = 1;
1490 break;
1491 case LTTNG_UST_CHAN_METADATA:
1492 channel->type = CONSUMER_CHANNEL_TYPE_METADATA;
1493 attr.type = LTTNG_UST_CHAN_METADATA;
1494 break;
1495 default:
1496 assert(0);
1497 goto error_fatal;
1498 };
1499
1500 health_code_update();
1501
1502 ret = ask_channel(ctx, channel, &attr,
1503 msg.u.ask_channel.trace_archive_id);
1504 if (ret < 0) {
1505 goto end_channel_error;
1506 }
1507
1508 if (msg.u.ask_channel.type == LTTNG_UST_CHAN_METADATA) {
1509 ret = consumer_metadata_cache_allocate(channel);
1510 if (ret < 0) {
1511 ERR("Allocating metadata cache");
1512 goto end_channel_error;
1513 }
1514 consumer_timer_switch_start(channel, attr.switch_timer_interval);
1515 attr.switch_timer_interval = 0;
1516 } else {
1517 int monitor_start_ret;
1518
1519 consumer_timer_live_start(channel,
1520 msg.u.ask_channel.live_timer_interval);
1521 monitor_start_ret = consumer_timer_monitor_start(
1522 channel,
1523 msg.u.ask_channel.monitor_timer_interval);
1524 if (monitor_start_ret < 0) {
1525 ERR("Starting channel monitoring timer failed");
1526 goto end_channel_error;
1527 }
1528 }
1529
1530 health_code_update();
1531
1532 /*
1533 * Add the channel to the internal state AFTER all streams were created
1534 * and successfully sent to session daemon. This way, all streams must
1535 * be ready before this channel is visible to the threads.
1536 * If add_channel succeeds, ownership of the channel is
1537 * passed to consumer_thread_channel_poll().
1538 */
1539 ret = add_channel(channel, ctx);
1540 if (ret < 0) {
1541 if (msg.u.ask_channel.type == LTTNG_UST_CHAN_METADATA) {
1542 if (channel->switch_timer_enabled == 1) {
1543 consumer_timer_switch_stop(channel);
1544 }
1545 consumer_metadata_cache_destroy(channel);
1546 }
1547 if (channel->live_timer_enabled == 1) {
1548 consumer_timer_live_stop(channel);
1549 }
1550 if (channel->monitor_timer_enabled == 1) {
1551 consumer_timer_monitor_stop(channel);
1552 }
1553 goto end_channel_error;
1554 }
1555
1556 health_code_update();
1557
1558 /*
1559 * Channel and streams are now created. Inform the session daemon that
1560 * everything went well and should wait to receive the channel and
1561 * streams with ustctl API.
1562 */
1563 ret = consumer_send_status_channel(sock, channel);
1564 if (ret < 0) {
1565 /*
1566 * There is probably a problem on the socket.
1567 */
1568 goto error_fatal;
1569 }
1570
1571 break;
1572 }
1573 case LTTNG_CONSUMER_GET_CHANNEL:
1574 {
1575 int ret, relayd_err = 0;
1576 uint64_t key = msg.u.get_channel.key;
1577 struct lttng_consumer_channel *channel;
1578
1579 channel = consumer_find_channel(key);
1580 if (!channel) {
1581 ERR("UST consumer get channel key %" PRIu64 " not found", key);
1582 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
1583 goto end_msg_sessiond;
1584 }
1585
1586 health_code_update();
1587
1588 /* Send the channel to sessiond (and relayd, if applicable). */
1589 ret = send_channel_to_sessiond_and_relayd(sock, channel, ctx,
1590 &relayd_err);
1591 if (ret < 0) {
1592 if (relayd_err) {
1593 /*
1594 * We were unable to send to the relayd the stream so avoid
1595 * sending back a fatal error to the thread since this is OK
1596 * and the consumer can continue its work. The above call
1597 * has sent the error status message to the sessiond.
1598 */
1599 goto end_nosignal;
1600 }
1601 /*
1602 * The communicaton was broken hence there is a bad state between
1603 * the consumer and sessiond so stop everything.
1604 */
1605 goto error_fatal;
1606 }
1607
1608 health_code_update();
1609
1610 /*
1611 * In no monitor mode, the streams ownership is kept inside the channel
1612 * so don't send them to the data thread.
1613 */
1614 if (!channel->monitor) {
1615 goto end_msg_sessiond;
1616 }
1617
1618 ret = send_streams_to_thread(channel, ctx);
1619 if (ret < 0) {
1620 /*
1621 * If we are unable to send the stream to the thread, there is
1622 * a big problem so just stop everything.
1623 */
1624 goto error_fatal;
1625 }
1626 /* List MUST be empty after or else it could be reused. */
1627 assert(cds_list_empty(&channel->streams.head));
1628 goto end_msg_sessiond;
1629 }
1630 case LTTNG_CONSUMER_DESTROY_CHANNEL:
1631 {
1632 uint64_t key = msg.u.destroy_channel.key;
1633
1634 /*
1635 * Only called if streams have not been sent to stream
1636 * manager thread. However, channel has been sent to
1637 * channel manager thread.
1638 */
1639 notify_thread_del_channel(ctx, key);
1640 goto end_msg_sessiond;
1641 }
1642 case LTTNG_CONSUMER_CLOSE_METADATA:
1643 {
1644 int ret;
1645
1646 ret = close_metadata(msg.u.close_metadata.key);
1647 if (ret != 0) {
1648 ret_code = ret;
1649 }
1650
1651 goto end_msg_sessiond;
1652 }
1653 case LTTNG_CONSUMER_FLUSH_CHANNEL:
1654 {
1655 int ret;
1656
1657 ret = flush_channel(msg.u.flush_channel.key);
1658 if (ret != 0) {
1659 ret_code = ret;
1660 }
1661
1662 goto end_msg_sessiond;
1663 }
1664 case LTTNG_CONSUMER_CLEAR_QUIESCENT_CHANNEL:
1665 {
1666 int ret;
1667
1668 ret = clear_quiescent_channel(
1669 msg.u.clear_quiescent_channel.key);
1670 if (ret != 0) {
1671 ret_code = ret;
1672 }
1673
1674 goto end_msg_sessiond;
1675 }
1676 case LTTNG_CONSUMER_PUSH_METADATA:
1677 {
1678 int ret;
1679 uint64_t len = msg.u.push_metadata.len;
1680 uint64_t key = msg.u.push_metadata.key;
1681 uint64_t offset = msg.u.push_metadata.target_offset;
1682 uint64_t version = msg.u.push_metadata.version;
1683 struct lttng_consumer_channel *channel;
1684
1685 DBG("UST consumer push metadata key %" PRIu64 " of len %" PRIu64, key,
1686 len);
1687
1688 channel = consumer_find_channel(key);
1689 if (!channel) {
1690 /*
1691 * This is possible if the metadata creation on the consumer side
1692 * is in flight vis-a-vis a concurrent push metadata from the
1693 * session daemon. Simply return that the channel failed and the
1694 * session daemon will handle that message correctly considering
1695 * that this race is acceptable thus the DBG() statement here.
1696 */
1697 DBG("UST consumer push metadata %" PRIu64 " not found", key);
1698 ret_code = LTTCOMM_CONSUMERD_CHANNEL_FAIL;
1699 goto end_msg_sessiond;
1700 }
1701
1702 health_code_update();
1703
1704 if (!len) {
1705 /*
1706 * There is nothing to receive. We have simply
1707 * checked whether the channel can be found.
1708 */
1709 ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1710 goto end_msg_sessiond;
1711 }
1712
1713 /* Tell session daemon we are ready to receive the metadata. */
1714 ret = consumer_send_status_msg(sock, LTTCOMM_CONSUMERD_SUCCESS);
1715 if (ret < 0) {
1716 /* Somehow, the session daemon is not responding anymore. */
1717 goto error_fatal;
1718 }
1719
1720 health_code_update();
1721
1722 /* Wait for more data. */
1723 health_poll_entry();
1724 ret = lttng_consumer_poll_socket(consumer_sockpoll);
1725 health_poll_exit();
1726 if (ret) {
1727 goto error_fatal;
1728 }
1729
1730 health_code_update();
1731
1732 ret = lttng_ustconsumer_recv_metadata(sock, key, offset,
1733 len, version, channel, 0, 1);
1734 if (ret < 0) {
1735 /* error receiving from sessiond */
1736 goto error_fatal;
1737 } else {
1738 ret_code = ret;
1739 goto end_msg_sessiond;
1740 }
1741 }
1742 case LTTNG_CONSUMER_SETUP_METADATA:
1743 {
1744 int ret;
1745
1746 ret = setup_metadata(ctx, msg.u.setup_metadata.key);
1747 if (ret) {
1748 ret_code = ret;
1749 }
1750 goto end_msg_sessiond;
1751 }
1752 case LTTNG_CONSUMER_SNAPSHOT_CHANNEL:
1753 {
1754 if (msg.u.snapshot_channel.metadata) {
1755 ret = snapshot_metadata(msg.u.snapshot_channel.key,
1756 msg.u.snapshot_channel.pathname,
1757 msg.u.snapshot_channel.relayd_id,
1758 ctx,
1759 msg.u.snapshot_channel.trace_archive_id);
1760 if (ret < 0) {
1761 ERR("Snapshot metadata failed");
1762 ret_code = LTTCOMM_CONSUMERD_ERROR_METADATA;
1763 }
1764 } else {
1765 ret = snapshot_channel(msg.u.snapshot_channel.key,
1766 msg.u.snapshot_channel.pathname,
1767 msg.u.snapshot_channel.relayd_id,
1768 msg.u.snapshot_channel.nb_packets_per_stream,
1769 ctx);
1770 if (ret < 0) {
1771 ERR("Snapshot channel failed");
1772 ret_code = LTTCOMM_CONSUMERD_CHANNEL_FAIL;
1773 }
1774 }
1775
1776 health_code_update();
1777 ret = consumer_send_status_msg(sock, ret_code);
1778 if (ret < 0) {
1779 /* Somehow, the session daemon is not responding anymore. */
1780 goto end_nosignal;
1781 }
1782 health_code_update();
1783 break;
1784 }
1785 case LTTNG_CONSUMER_DISCARDED_EVENTS:
1786 {
1787 int ret = 0;
1788 uint64_t discarded_events;
1789 struct lttng_ht_iter iter;
1790 struct lttng_ht *ht;
1791 struct lttng_consumer_stream *stream;
1792 uint64_t id = msg.u.discarded_events.session_id;
1793 uint64_t key = msg.u.discarded_events.channel_key;
1794
1795 DBG("UST consumer discarded events command for session id %"
1796 PRIu64, id);
1797 rcu_read_lock();
1798 pthread_mutex_lock(&consumer_data.lock);
1799
1800 ht = consumer_data.stream_list_ht;
1801
1802 /*
1803 * We only need a reference to the channel, but they are not
1804 * directly indexed, so we just use the first matching stream
1805 * to extract the information we need, we default to 0 if not
1806 * found (no events are dropped if the channel is not yet in
1807 * use).
1808 */
1809 discarded_events = 0;
1810 cds_lfht_for_each_entry_duplicate(ht->ht,
1811 ht->hash_fct(&id, lttng_ht_seed),
1812 ht->match_fct, &id,
1813 &iter.iter, stream, node_session_id.node) {
1814 if (stream->chan->key == key) {
1815 discarded_events = stream->chan->discarded_events;
1816 break;
1817 }
1818 }
1819 pthread_mutex_unlock(&consumer_data.lock);
1820 rcu_read_unlock();
1821
1822 DBG("UST consumer discarded events command for session id %"
1823 PRIu64 ", channel key %" PRIu64, id, key);
1824
1825 health_code_update();
1826
1827 /* Send back returned value to session daemon */
1828 ret = lttcomm_send_unix_sock(sock, &discarded_events, sizeof(discarded_events));
1829 if (ret < 0) {
1830 PERROR("send discarded events");
1831 goto error_fatal;
1832 }
1833
1834 break;
1835 }
1836 case LTTNG_CONSUMER_LOST_PACKETS:
1837 {
1838 int ret;
1839 uint64_t lost_packets;
1840 struct lttng_ht_iter iter;
1841 struct lttng_ht *ht;
1842 struct lttng_consumer_stream *stream;
1843 uint64_t id = msg.u.lost_packets.session_id;
1844 uint64_t key = msg.u.lost_packets.channel_key;
1845
1846 DBG("UST consumer lost packets command for session id %"
1847 PRIu64, id);
1848 rcu_read_lock();
1849 pthread_mutex_lock(&consumer_data.lock);
1850
1851 ht = consumer_data.stream_list_ht;
1852
1853 /*
1854 * We only need a reference to the channel, but they are not
1855 * directly indexed, so we just use the first matching stream
1856 * to extract the information we need, we default to 0 if not
1857 * found (no packets lost if the channel is not yet in use).
1858 */
1859 lost_packets = 0;
1860 cds_lfht_for_each_entry_duplicate(ht->ht,
1861 ht->hash_fct(&id, lttng_ht_seed),
1862 ht->match_fct, &id,
1863 &iter.iter, stream, node_session_id.node) {
1864 if (stream->chan->key == key) {
1865 lost_packets = stream->chan->lost_packets;
1866 break;
1867 }
1868 }
1869 pthread_mutex_unlock(&consumer_data.lock);
1870 rcu_read_unlock();
1871
1872 DBG("UST consumer lost packets command for session id %"
1873 PRIu64 ", channel key %" PRIu64, id, key);
1874
1875 health_code_update();
1876
1877 /* Send back returned value to session daemon */
1878 ret = lttcomm_send_unix_sock(sock, &lost_packets,
1879 sizeof(lost_packets));
1880 if (ret < 0) {
1881 PERROR("send lost packets");
1882 goto error_fatal;
1883 }
1884
1885 break;
1886 }
1887 case LTTNG_CONSUMER_SET_CHANNEL_MONITOR_PIPE:
1888 {
1889 int channel_monitor_pipe;
1890
1891 ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1892 /* Successfully received the command's type. */
1893 ret = consumer_send_status_msg(sock, ret_code);
1894 if (ret < 0) {
1895 goto error_fatal;
1896 }
1897
1898 ret = lttcomm_recv_fds_unix_sock(sock, &channel_monitor_pipe,
1899 1);
1900 if (ret != sizeof(channel_monitor_pipe)) {
1901 ERR("Failed to receive channel monitor pipe");
1902 goto error_fatal;
1903 }
1904
1905 DBG("Received channel monitor pipe (%d)", channel_monitor_pipe);
1906 ret = consumer_timer_thread_set_channel_monitor_pipe(
1907 channel_monitor_pipe);
1908 if (!ret) {
1909 int flags;
1910
1911 ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1912 /* Set the pipe as non-blocking. */
1913 ret = fcntl(channel_monitor_pipe, F_GETFL, 0);
1914 if (ret == -1) {
1915 PERROR("fcntl get flags of the channel monitoring pipe");
1916 goto error_fatal;
1917 }
1918 flags = ret;
1919
1920 ret = fcntl(channel_monitor_pipe, F_SETFL,
1921 flags | O_NONBLOCK);
1922 if (ret == -1) {
1923 PERROR("fcntl set O_NONBLOCK flag of the channel monitoring pipe");
1924 goto error_fatal;
1925 }
1926 DBG("Channel monitor pipe set as non-blocking");
1927 } else {
1928 ret_code = LTTCOMM_CONSUMERD_ALREADY_SET;
1929 }
1930 goto end_msg_sessiond;
1931 }
1932 case LTTNG_CONSUMER_ROTATE_CHANNEL:
1933 {
1934 struct lttng_consumer_channel *channel;
1935 uint64_t key = msg.u.rotate_channel.key;
1936
1937 channel = consumer_find_channel(key);
1938 if (!channel) {
1939 DBG("Channel %" PRIu64 " not found", key);
1940 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
1941 } else {
1942 /*
1943 * Sample the rotate position of all the streams in
1944 * this channel.
1945 */
1946 ret = lttng_consumer_rotate_channel(channel, key,
1947 msg.u.rotate_channel.pathname,
1948 msg.u.rotate_channel.relayd_id,
1949 msg.u.rotate_channel.metadata,
1950 msg.u.rotate_channel.new_chunk_id,
1951 ctx);
1952 if (ret < 0) {
1953 ERR("Rotate channel failed");
1954 ret_code = LTTCOMM_CONSUMERD_ROTATION_FAIL;
1955 }
1956
1957 health_code_update();
1958 }
1959 ret = consumer_send_status_msg(sock, ret_code);
1960 if (ret < 0) {
1961 /* Somehow, the session daemon is not responding anymore. */
1962 goto end_nosignal;
1963 }
1964
1965 /*
1966 * Rotate the streams that are ready right now.
1967 * FIXME: this is a second consecutive iteration over the
1968 * streams in a channel, there is probably a better way to
1969 * handle this, but it needs to be after the
1970 * consumer_send_status_msg() call.
1971 */
1972 if (channel) {
1973 ret = lttng_consumer_rotate_ready_streams(
1974 channel, key, ctx);
1975 if (ret < 0) {
1976 ERR("Rotate channel failed");
1977 }
1978 }
1979 break;
1980 }
1981 case LTTNG_CONSUMER_ROTATE_RENAME:
1982 {
1983 DBG("Consumer rename session %" PRIu64 " after rotation",
1984 msg.u.rotate_rename.session_id);
1985 ret = lttng_consumer_rotate_rename(msg.u.rotate_rename.old_path,
1986 msg.u.rotate_rename.new_path,
1987 msg.u.rotate_rename.uid,
1988 msg.u.rotate_rename.gid,
1989 msg.u.rotate_rename.relayd_id);
1990 if (ret < 0) {
1991 ERR("Rotate rename failed");
1992 ret_code = LTTCOMM_CONSUMERD_ROTATE_RENAME_FAILED;
1993 }
1994
1995 health_code_update();
1996
1997 ret = consumer_send_status_msg(sock, ret_code);
1998 if (ret < 0) {
1999 /* Somehow, the session daemon is not responding anymore. */
2000 goto end_nosignal;
2001 }
2002 break;
2003 }
2004 case LTTNG_CONSUMER_CHECK_ROTATION_PENDING_LOCAL:
2005 {
2006 int pending;
2007 uint32_t pending_reply;
2008
2009 DBG("Perform local check of pending rotation for session id %" PRIu64,
2010 msg.u.check_rotation_pending_local.session_id);
2011 pending = lttng_consumer_check_rotation_pending_local(
2012 msg.u.check_rotation_pending_local.session_id,
2013 msg.u.check_rotation_pending_local.chunk_id);
2014 if (pending < 0) {
2015 ERR("Local rotation pending check failed with code %i", pending);
2016 ret_code = LTTCOMM_CONSUMERD_ROTATION_PENDING_LOCAL_FAILED;
2017 } else {
2018 pending_reply = !!pending;
2019 }
2020
2021 health_code_update();
2022
2023 ret = consumer_send_status_msg(sock, ret_code);
2024 if (ret < 0) {
2025 /* Somehow, the session daemon is not responding anymore. */
2026 goto end_nosignal;
2027 }
2028
2029 if (pending < 0) {
2030 /*
2031 * An error occured while running the command;
2032 * don't send the 'pending' flag as the sessiond
2033 * will not read it.
2034 */
2035 break;
2036 }
2037
2038 /* Send back returned value to session daemon */
2039 ret = lttcomm_send_unix_sock(sock, &pending_reply,
2040 sizeof(pending_reply));
2041 if (ret < 0) {
2042 PERROR("Failed to send rotation pending return code");
2043 goto error_fatal;
2044 }
2045 break;
2046 }
2047 case LTTNG_CONSUMER_CHECK_ROTATION_PENDING_RELAY:
2048 {
2049 int pending;
2050 uint32_t pending_reply;
2051
2052 DBG("Perform relayd check of pending rotation for session id %" PRIu64,
2053 msg.u.check_rotation_pending_relay.session_id);
2054 pending = lttng_consumer_check_rotation_pending_relay(
2055 msg.u.check_rotation_pending_relay.session_id,
2056 msg.u.check_rotation_pending_relay.relayd_id,
2057 msg.u.check_rotation_pending_relay.chunk_id);
2058 if (pending < 0) {
2059 ERR("Relayd rotation pending check failed with code %i", pending);
2060 ret_code = LTTCOMM_CONSUMERD_ROTATION_PENDING_RELAY_FAILED;
2061 } else {
2062 pending_reply = !!pending;
2063 }
2064
2065 health_code_update();
2066
2067 ret = consumer_send_status_msg(sock, ret_code);
2068 if (ret < 0) {
2069 /* Somehow, the session daemon is not responding anymore. */
2070 goto end_nosignal;
2071 }
2072
2073 if (pending < 0) {
2074 /*
2075 * An error occured while running the command;
2076 * don't send the 'pending' flag as the sessiond
2077 * will not read it.
2078 */
2079 break;
2080 }
2081
2082 /* Send back returned value to session daemon */
2083 ret = lttcomm_send_unix_sock(sock, &pending_reply,
2084 sizeof(pending_reply));
2085 if (ret < 0) {
2086 PERROR("Failed to send rotation pending return code");
2087 goto error_fatal;
2088 }
2089 break;
2090 }
2091 case LTTNG_CONSUMER_MKDIR:
2092 {
2093 DBG("Consumer mkdir %s in session %" PRIu64,
2094 msg.u.mkdir.path,
2095 msg.u.mkdir.session_id);
2096 ret = lttng_consumer_mkdir(msg.u.mkdir.path,
2097 msg.u.mkdir.uid,
2098 msg.u.mkdir.gid,
2099 msg.u.mkdir.relayd_id);
2100 if (ret < 0) {
2101 ERR("consumer mkdir failed");
2102 ret_code = LTTCOMM_CONSUMERD_MKDIR_FAILED;
2103 }
2104
2105 health_code_update();
2106
2107 ret = consumer_send_status_msg(sock, ret_code);
2108 if (ret < 0) {
2109 /* Somehow, the session daemon is not responding anymore. */
2110 goto end_nosignal;
2111 }
2112 break;
2113 }
2114 default:
2115 break;
2116 }
2117
2118 end_nosignal:
2119 rcu_read_unlock();
2120
2121 health_code_update();
2122
2123 /*
2124 * Return 1 to indicate success since the 0 value can be a socket
2125 * shutdown during the recv() or send() call.
2126 */
2127 return 1;
2128
2129 end_msg_sessiond:
2130 /*
2131 * The returned value here is not useful since either way we'll return 1 to
2132 * the caller because the session daemon socket management is done
2133 * elsewhere. Returning a negative code or 0 will shutdown the consumer.
2134 */
2135 ret = consumer_send_status_msg(sock, ret_code);
2136 if (ret < 0) {
2137 goto error_fatal;
2138 }
2139 rcu_read_unlock();
2140
2141 health_code_update();
2142
2143 return 1;
2144 end_channel_error:
2145 if (channel) {
2146 /*
2147 * Free channel here since no one has a reference to it. We don't
2148 * free after that because a stream can store this pointer.
2149 */
2150 destroy_channel(channel);
2151 }
2152 /* We have to send a status channel message indicating an error. */
2153 ret = consumer_send_status_channel(sock, NULL);
2154 if (ret < 0) {
2155 /* Stop everything if session daemon can not be notified. */
2156 goto error_fatal;
2157 }
2158 rcu_read_unlock();
2159
2160 health_code_update();
2161
2162 return 1;
2163 error_fatal:
2164 rcu_read_unlock();
2165 /* This will issue a consumer stop. */
2166 return -1;
2167 }
2168
2169 /*
2170 * Wrapper over the mmap() read offset from ust-ctl library. Since this can be
2171 * compiled out, we isolate it in this library.
2172 */
2173 int lttng_ustctl_get_mmap_read_offset(struct lttng_consumer_stream *stream,
2174 unsigned long *off)
2175 {
2176 assert(stream);
2177 assert(stream->ustream);
2178
2179 return ustctl_get_mmap_read_offset(stream->ustream, off);
2180 }
2181
2182 /*
2183 * Wrapper over the mmap() read offset from ust-ctl library. Since this can be
2184 * compiled out, we isolate it in this library.
2185 */
2186 void *lttng_ustctl_get_mmap_base(struct lttng_consumer_stream *stream)
2187 {
2188 assert(stream);
2189 assert(stream->ustream);
2190
2191 return ustctl_get_mmap_base(stream->ustream);
2192 }
2193
2194 void lttng_ustctl_flush_buffer(struct lttng_consumer_stream *stream,
2195 int producer_active)
2196 {
2197 assert(stream);
2198 assert(stream->ustream);
2199
2200 ustctl_flush_buffer(stream->ustream, producer_active);
2201 }
2202
2203 /*
2204 * Take a snapshot for a specific stream.
2205 *
2206 * Returns 0 on success, < 0 on error
2207 */
2208 int lttng_ustconsumer_take_snapshot(struct lttng_consumer_stream *stream)
2209 {
2210 assert(stream);
2211 assert(stream->ustream);
2212
2213 return ustctl_snapshot(stream->ustream);
2214 }
2215
2216 /*
2217 * Sample consumed and produced positions for a specific stream.
2218 *
2219 * Returns 0 on success, < 0 on error.
2220 */
2221 int lttng_ustconsumer_sample_snapshot_positions(
2222 struct lttng_consumer_stream *stream)
2223 {
2224 assert(stream);
2225 assert(stream->ustream);
2226
2227 return ustctl_snapshot_sample_positions(stream->ustream);
2228 }
2229
2230 /*
2231 * Get the produced position
2232 *
2233 * Returns 0 on success, < 0 on error
2234 */
2235 int lttng_ustconsumer_get_produced_snapshot(
2236 struct lttng_consumer_stream *stream, unsigned long *pos)
2237 {
2238 assert(stream);
2239 assert(stream->ustream);
2240 assert(pos);
2241
2242 return ustctl_snapshot_get_produced(stream->ustream, pos);
2243 }
2244
2245 /*
2246 * Get the consumed position
2247 *
2248 * Returns 0 on success, < 0 on error
2249 */
2250 int lttng_ustconsumer_get_consumed_snapshot(
2251 struct lttng_consumer_stream *stream, unsigned long *pos)
2252 {
2253 assert(stream);
2254 assert(stream->ustream);
2255 assert(pos);
2256
2257 return ustctl_snapshot_get_consumed(stream->ustream, pos);
2258 }
2259
2260 void lttng_ustconsumer_flush_buffer(struct lttng_consumer_stream *stream,
2261 int producer)
2262 {
2263 assert(stream);
2264 assert(stream->ustream);
2265
2266 ustctl_flush_buffer(stream->ustream, producer);
2267 }
2268
2269 int lttng_ustconsumer_get_current_timestamp(
2270 struct lttng_consumer_stream *stream, uint64_t *ts)
2271 {
2272 assert(stream);
2273 assert(stream->ustream);
2274 assert(ts);
2275
2276 return ustctl_get_current_timestamp(stream->ustream, ts);
2277 }
2278
2279 int lttng_ustconsumer_get_sequence_number(
2280 struct lttng_consumer_stream *stream, uint64_t *seq)
2281 {
2282 assert(stream);
2283 assert(stream->ustream);
2284 assert(seq);
2285
2286 return ustctl_get_sequence_number(stream->ustream, seq);
2287 }
2288
2289 /*
2290 * Called when the stream signals the consumer that it has hung up.
2291 */
2292 void lttng_ustconsumer_on_stream_hangup(struct lttng_consumer_stream *stream)
2293 {
2294 assert(stream);
2295 assert(stream->ustream);
2296
2297 pthread_mutex_lock(&stream->lock);
2298 if (!stream->quiescent) {
2299 ustctl_flush_buffer(stream->ustream, 0);
2300 stream->quiescent = true;
2301 }
2302 pthread_mutex_unlock(&stream->lock);
2303 stream->hangup_flush_done = 1;
2304 }
2305
2306 void lttng_ustconsumer_del_channel(struct lttng_consumer_channel *chan)
2307 {
2308 int i;
2309
2310 assert(chan);
2311 assert(chan->uchan);
2312
2313 if (chan->switch_timer_enabled == 1) {
2314 consumer_timer_switch_stop(chan);
2315 }
2316 for (i = 0; i < chan->nr_stream_fds; i++) {
2317 int ret;
2318
2319 ret = close(chan->stream_fds[i]);
2320 if (ret) {
2321 PERROR("close");
2322 }
2323 if (chan->shm_path[0]) {
2324 char shm_path[PATH_MAX];
2325
2326 ret = get_stream_shm_path(shm_path, chan->shm_path, i);
2327 if (ret) {
2328 ERR("Cannot get stream shm path");
2329 }
2330 ret = run_as_unlink(shm_path, chan->uid, chan->gid);
2331 if (ret) {
2332 PERROR("unlink %s", shm_path);
2333 }
2334 }
2335 }
2336 }
2337
2338 void lttng_ustconsumer_free_channel(struct lttng_consumer_channel *chan)
2339 {
2340 assert(chan);
2341 assert(chan->uchan);
2342
2343 consumer_metadata_cache_destroy(chan);
2344 ustctl_destroy_channel(chan->uchan);
2345 /* Try to rmdir all directories under shm_path root. */
2346 if (chan->root_shm_path[0]) {
2347 (void) run_as_rmdir_recursive(chan->root_shm_path,
2348 chan->uid, chan->gid);
2349 }
2350 free(chan->stream_fds);
2351 }
2352
2353 void lttng_ustconsumer_del_stream(struct lttng_consumer_stream *stream)
2354 {
2355 assert(stream);
2356 assert(stream->ustream);
2357
2358 if (stream->chan->switch_timer_enabled == 1) {
2359 consumer_timer_switch_stop(stream->chan);
2360 }
2361 ustctl_destroy_stream(stream->ustream);
2362 }
2363
2364 int lttng_ustconsumer_get_wakeup_fd(struct lttng_consumer_stream *stream)
2365 {
2366 assert(stream);
2367 assert(stream->ustream);
2368
2369 return ustctl_stream_get_wakeup_fd(stream->ustream);
2370 }
2371
2372 int lttng_ustconsumer_close_wakeup_fd(struct lttng_consumer_stream *stream)
2373 {
2374 assert(stream);
2375 assert(stream->ustream);
2376
2377 return ustctl_stream_close_wakeup_fd(stream->ustream);
2378 }
2379
2380 /*
2381 * Populate index values of a UST stream. Values are set in big endian order.
2382 *
2383 * Return 0 on success or else a negative value.
2384 */
2385 static int get_index_values(struct ctf_packet_index *index,
2386 struct ustctl_consumer_stream *ustream)
2387 {
2388 int ret;
2389
2390 ret = ustctl_get_timestamp_begin(ustream, &index->timestamp_begin);
2391 if (ret < 0) {
2392 PERROR("ustctl_get_timestamp_begin");
2393 goto error;
2394 }
2395 index->timestamp_begin = htobe64(index->timestamp_begin);
2396
2397 ret = ustctl_get_timestamp_end(ustream, &index->timestamp_end);
2398 if (ret < 0) {
2399 PERROR("ustctl_get_timestamp_end");
2400 goto error;
2401 }
2402 index->timestamp_end = htobe64(index->timestamp_end);
2403
2404 ret = ustctl_get_events_discarded(ustream, &index->events_discarded);
2405 if (ret < 0) {
2406 PERROR("ustctl_get_events_discarded");
2407 goto error;
2408 }
2409 index->events_discarded = htobe64(index->events_discarded);
2410
2411 ret = ustctl_get_content_size(ustream, &index->content_size);
2412 if (ret < 0) {
2413 PERROR("ustctl_get_content_size");
2414 goto error;
2415 }
2416 index->content_size = htobe64(index->content_size);
2417
2418 ret = ustctl_get_packet_size(ustream, &index->packet_size);
2419 if (ret < 0) {
2420 PERROR("ustctl_get_packet_size");
2421 goto error;
2422 }
2423 index->packet_size = htobe64(index->packet_size);
2424
2425 ret = ustctl_get_stream_id(ustream, &index->stream_id);
2426 if (ret < 0) {
2427 PERROR("ustctl_get_stream_id");
2428 goto error;
2429 }
2430 index->stream_id = htobe64(index->stream_id);
2431
2432 ret = ustctl_get_instance_id(ustream, &index->stream_instance_id);
2433 if (ret < 0) {
2434 PERROR("ustctl_get_instance_id");
2435 goto error;
2436 }
2437 index->stream_instance_id = htobe64(index->stream_instance_id);
2438
2439 ret = ustctl_get_sequence_number(ustream, &index->packet_seq_num);
2440 if (ret < 0) {
2441 PERROR("ustctl_get_sequence_number");
2442 goto error;
2443 }
2444 index->packet_seq_num = htobe64(index->packet_seq_num);
2445
2446 error:
2447 return ret;
2448 }
2449
2450 static
2451 void metadata_stream_reset_cache(struct lttng_consumer_stream *stream,
2452 struct consumer_metadata_cache *cache)
2453 {
2454 DBG("Metadata stream update to version %" PRIu64,
2455 cache->version);
2456 stream->ust_metadata_pushed = 0;
2457 stream->metadata_version = cache->version;
2458 stream->reset_metadata_flag = 1;
2459 }
2460
2461 /*
2462 * Check if the version of the metadata stream and metadata cache match.
2463 * If the cache got updated, reset the metadata stream.
2464 * The stream lock and metadata cache lock MUST be held.
2465 * Return 0 on success, a negative value on error.
2466 */
2467 static
2468 int metadata_stream_check_version(struct lttng_consumer_stream *stream)
2469 {
2470 int ret = 0;
2471 struct consumer_metadata_cache *cache = stream->chan->metadata_cache;
2472
2473 if (cache->version == stream->metadata_version) {
2474 goto end;
2475 }
2476 metadata_stream_reset_cache(stream, cache);
2477
2478 end:
2479 return ret;
2480 }
2481
2482 /*
2483 * Write up to one packet from the metadata cache to the channel.
2484 *
2485 * Returns the number of bytes pushed in the cache, or a negative value
2486 * on error.
2487 */
2488 static
2489 int commit_one_metadata_packet(struct lttng_consumer_stream *stream)
2490 {
2491 ssize_t write_len;
2492 int ret;
2493
2494 pthread_mutex_lock(&stream->chan->metadata_cache->lock);
2495 ret = metadata_stream_check_version(stream);
2496 if (ret < 0) {
2497 goto end;
2498 }
2499 if (stream->chan->metadata_cache->max_offset
2500 == stream->ust_metadata_pushed) {
2501 ret = 0;
2502 goto end;
2503 }
2504
2505 write_len = ustctl_write_one_packet_to_channel(stream->chan->uchan,
2506 &stream->chan->metadata_cache->data[stream->ust_metadata_pushed],
2507 stream->chan->metadata_cache->max_offset
2508 - stream->ust_metadata_pushed);
2509 assert(write_len != 0);
2510 if (write_len < 0) {
2511 ERR("Writing one metadata packet");
2512 ret = -1;
2513 goto end;
2514 }
2515 stream->ust_metadata_pushed += write_len;
2516
2517 assert(stream->chan->metadata_cache->max_offset >=
2518 stream->ust_metadata_pushed);
2519 ret = write_len;
2520
2521 end:
2522 pthread_mutex_unlock(&stream->chan->metadata_cache->lock);
2523 return ret;
2524 }
2525
2526
2527 /*
2528 * Sync metadata meaning request them to the session daemon and snapshot to the
2529 * metadata thread can consumer them.
2530 *
2531 * Metadata stream lock is held here, but we need to release it when
2532 * interacting with sessiond, else we cause a deadlock with live
2533 * awaiting on metadata to be pushed out.
2534 *
2535 * Return 0 if new metadatda is available, EAGAIN if the metadata stream
2536 * is empty or a negative value on error.
2537 */
2538 int lttng_ustconsumer_sync_metadata(struct lttng_consumer_local_data *ctx,
2539 struct lttng_consumer_stream *metadata)
2540 {
2541 int ret;
2542 int retry = 0;
2543
2544 assert(ctx);
2545 assert(metadata);
2546
2547 pthread_mutex_unlock(&metadata->lock);
2548 /*
2549 * Request metadata from the sessiond, but don't wait for the flush
2550 * because we locked the metadata thread.
2551 */
2552 ret = lttng_ustconsumer_request_metadata(ctx, metadata->chan, 0, 0);
2553 pthread_mutex_lock(&metadata->lock);
2554 if (ret < 0) {
2555 goto end;
2556 }
2557
2558 ret = commit_one_metadata_packet(metadata);
2559 if (ret <= 0) {
2560 goto end;
2561 } else if (ret > 0) {
2562 retry = 1;
2563 }
2564
2565 ustctl_flush_buffer(metadata->ustream, 1);
2566 ret = ustctl_snapshot(metadata->ustream);
2567 if (ret < 0) {
2568 if (errno != EAGAIN) {
2569 ERR("Sync metadata, taking UST snapshot");
2570 goto end;
2571 }
2572 DBG("No new metadata when syncing them.");
2573 /* No new metadata, exit. */
2574 ret = ENODATA;
2575 goto end;
2576 }
2577
2578 /*
2579 * After this flush, we still need to extract metadata.
2580 */
2581 if (retry) {
2582 ret = EAGAIN;
2583 }
2584
2585 end:
2586 return ret;
2587 }
2588
2589 /*
2590 * Return 0 on success else a negative value.
2591 */
2592 static int notify_if_more_data(struct lttng_consumer_stream *stream,
2593 struct lttng_consumer_local_data *ctx)
2594 {
2595 int ret;
2596 struct ustctl_consumer_stream *ustream;
2597
2598 assert(stream);
2599 assert(ctx);
2600
2601 ustream = stream->ustream;
2602
2603 /*
2604 * First, we are going to check if there is a new subbuffer available
2605 * before reading the stream wait_fd.
2606 */
2607 /* Get the next subbuffer */
2608 ret = ustctl_get_next_subbuf(ustream);
2609 if (ret) {
2610 /* No more data found, flag the stream. */
2611 stream->has_data = 0;
2612 ret = 0;
2613 goto end;
2614 }
2615
2616 ret = ustctl_put_subbuf(ustream);
2617 assert(!ret);
2618
2619 /* This stream still has data. Flag it and wake up the data thread. */
2620 stream->has_data = 1;
2621
2622 if (stream->monitor && !stream->hangup_flush_done && !ctx->has_wakeup) {
2623 ssize_t writelen;
2624
2625 writelen = lttng_pipe_write(ctx->consumer_wakeup_pipe, "!", 1);
2626 if (writelen < 0 && errno != EAGAIN && errno != EWOULDBLOCK) {
2627 ret = writelen;
2628 goto end;
2629 }
2630
2631 /* The wake up pipe has been notified. */
2632 ctx->has_wakeup = 1;
2633 }
2634 ret = 0;
2635
2636 end:
2637 return ret;
2638 }
2639
2640 static
2641 int update_stream_stats(struct lttng_consumer_stream *stream)
2642 {
2643 int ret;
2644 uint64_t seq, discarded;
2645
2646 ret = ustctl_get_sequence_number(stream->ustream, &seq);
2647 if (ret < 0) {
2648 PERROR("ustctl_get_sequence_number");
2649 goto end;
2650 }
2651 /*
2652 * Start the sequence when we extract the first packet in case we don't
2653 * start at 0 (for example if a consumer is not connected to the
2654 * session immediately after the beginning).
2655 */
2656 if (stream->last_sequence_number == -1ULL) {
2657 stream->last_sequence_number = seq;
2658 } else if (seq > stream->last_sequence_number) {
2659 stream->chan->lost_packets += seq -
2660 stream->last_sequence_number - 1;
2661 } else {
2662 /* seq <= last_sequence_number */
2663 ERR("Sequence number inconsistent : prev = %" PRIu64
2664 ", current = %" PRIu64,
2665 stream->last_sequence_number, seq);
2666 ret = -1;
2667 goto end;
2668 }
2669 stream->last_sequence_number = seq;
2670
2671 ret = ustctl_get_events_discarded(stream->ustream, &discarded);
2672 if (ret < 0) {
2673 PERROR("kernctl_get_events_discarded");
2674 goto end;
2675 }
2676 if (discarded < stream->last_discarded_events) {
2677 /*
2678 * Overflow has occurred. We assume only one wrap-around
2679 * has occurred.
2680 */
2681 stream->chan->discarded_events +=
2682 (1ULL << (CAA_BITS_PER_LONG - 1)) -
2683 stream->last_discarded_events + discarded;
2684 } else {
2685 stream->chan->discarded_events += discarded -
2686 stream->last_discarded_events;
2687 }
2688 stream->last_discarded_events = discarded;
2689 ret = 0;
2690
2691 end:
2692 return ret;
2693 }
2694
2695 /*
2696 * Read subbuffer from the given stream.
2697 *
2698 * Stream lock MUST be acquired.
2699 *
2700 * Return 0 on success else a negative value.
2701 */
2702 int lttng_ustconsumer_read_subbuffer(struct lttng_consumer_stream *stream,
2703 struct lttng_consumer_local_data *ctx, bool *rotated)
2704 {
2705 unsigned long len, subbuf_size, padding;
2706 int err, write_index = 1, rotation_ret;
2707 long ret = 0;
2708 struct ustctl_consumer_stream *ustream;
2709 struct ctf_packet_index index;
2710
2711 assert(stream);
2712 assert(stream->ustream);
2713 assert(ctx);
2714
2715 DBG("In UST read_subbuffer (wait_fd: %d, name: %s)", stream->wait_fd,
2716 stream->name);
2717
2718 /* Ease our life for what's next. */
2719 ustream = stream->ustream;
2720
2721 /*
2722 * We can consume the 1 byte written into the wait_fd by UST. Don't trigger
2723 * error if we cannot read this one byte (read returns 0), or if the error
2724 * is EAGAIN or EWOULDBLOCK.
2725 *
2726 * This is only done when the stream is monitored by a thread, before the
2727 * flush is done after a hangup and if the stream is not flagged with data
2728 * since there might be nothing to consume in the wait fd but still have
2729 * data available flagged by the consumer wake up pipe.
2730 */
2731 if (stream->monitor && !stream->hangup_flush_done && !stream->has_data) {
2732 char dummy;
2733 ssize_t readlen;
2734
2735 readlen = lttng_read(stream->wait_fd, &dummy, 1);
2736 if (readlen < 0 && errno != EAGAIN && errno != EWOULDBLOCK) {
2737 ret = readlen;
2738 goto error;
2739 }
2740 }
2741
2742 /*
2743 * If the stream was flagged to be ready for rotation before we extract the
2744 * next packet, rotate it now.
2745 */
2746 if (stream->rotate_ready) {
2747 DBG("Rotate stream before extracting data");
2748 rotation_ret = lttng_consumer_rotate_stream(ctx, stream, rotated);
2749 if (rotation_ret < 0) {
2750 ERR("Stream rotation error");
2751 ret = -1;
2752 goto error;
2753 }
2754 }
2755
2756 retry:
2757 /* Get the next subbuffer */
2758 err = ustctl_get_next_subbuf(ustream);
2759 if (err != 0) {
2760 /*
2761 * Populate metadata info if the existing info has
2762 * already been read.
2763 */
2764 if (stream->metadata_flag) {
2765 ret = commit_one_metadata_packet(stream);
2766 if (ret <= 0) {
2767 goto error;
2768 }
2769 ustctl_flush_buffer(stream->ustream, 1);
2770 goto retry;
2771 }
2772
2773 ret = err; /* ustctl_get_next_subbuf returns negative, caller expect positive. */
2774 /*
2775 * This is a debug message even for single-threaded consumer,
2776 * because poll() have more relaxed criterions than get subbuf,
2777 * so get_subbuf may fail for short race windows where poll()
2778 * would issue wakeups.
2779 */
2780 DBG("Reserving sub buffer failed (everything is normal, "
2781 "it is due to concurrency) [ret: %d]", err);
2782 goto error;
2783 }
2784 assert(stream->chan->output == CONSUMER_CHANNEL_MMAP);
2785
2786 if (!stream->metadata_flag) {
2787 index.offset = htobe64(stream->out_fd_offset);
2788 ret = get_index_values(&index, ustream);
2789 if (ret < 0) {
2790 err = ustctl_put_subbuf(ustream);
2791 assert(err == 0);
2792 goto error;
2793 }
2794
2795 /* Update the stream's sequence and discarded events count. */
2796 ret = update_stream_stats(stream);
2797 if (ret < 0) {
2798 PERROR("kernctl_get_events_discarded");
2799 err = ustctl_put_subbuf(ustream);
2800 assert(err == 0);
2801 goto error;
2802 }
2803 } else {
2804 write_index = 0;
2805 }
2806
2807 /* Get the full padded subbuffer size */
2808 err = ustctl_get_padded_subbuf_size(ustream, &len);
2809 assert(err == 0);
2810
2811 /* Get subbuffer data size (without padding) */
2812 err = ustctl_get_subbuf_size(ustream, &subbuf_size);
2813 assert(err == 0);
2814
2815 /* Make sure we don't get a subbuffer size bigger than the padded */
2816 assert(len >= subbuf_size);
2817
2818 padding = len - subbuf_size;
2819
2820 /* write the subbuffer to the tracefile */
2821 ret = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, subbuf_size, padding, &index);
2822 /*
2823 * The mmap operation should write subbuf_size amount of data when network
2824 * streaming or the full padding (len) size when we are _not_ streaming.
2825 */
2826 if ((ret != subbuf_size && stream->net_seq_idx != (uint64_t) -1ULL) ||
2827 (ret != len && stream->net_seq_idx == (uint64_t) -1ULL)) {
2828 /*
2829 * Display the error but continue processing to try to release the
2830 * subbuffer. This is a DBG statement since any unexpected kill or
2831 * signal, the application gets unregistered, relayd gets closed or
2832 * anything that affects the buffer lifetime will trigger this error.
2833 * So, for the sake of the user, don't print this error since it can
2834 * happen and it is OK with the code flow.
2835 */
2836 DBG("Error writing to tracefile "
2837 "(ret: %ld != len: %lu != subbuf_size: %lu)",
2838 ret, len, subbuf_size);
2839 write_index = 0;
2840 }
2841 err = ustctl_put_next_subbuf(ustream);
2842 assert(err == 0);
2843
2844 /*
2845 * This will consumer the byte on the wait_fd if and only if there is not
2846 * next subbuffer to be acquired.
2847 */
2848 if (!stream->metadata_flag) {
2849 ret = notify_if_more_data(stream, ctx);
2850 if (ret < 0) {
2851 goto error;
2852 }
2853 }
2854
2855 /* Write index if needed. */
2856 if (!write_index) {
2857 goto rotate;
2858 }
2859
2860 if (stream->chan->live_timer_interval && !stream->metadata_flag) {
2861 /*
2862 * In live, block until all the metadata is sent.
2863 */
2864 pthread_mutex_lock(&stream->metadata_timer_lock);
2865 assert(!stream->missed_metadata_flush);
2866 stream->waiting_on_metadata = true;
2867 pthread_mutex_unlock(&stream->metadata_timer_lock);
2868
2869 err = consumer_stream_sync_metadata(ctx, stream->session_id);
2870
2871 pthread_mutex_lock(&stream->metadata_timer_lock);
2872 stream->waiting_on_metadata = false;
2873 if (stream->missed_metadata_flush) {
2874 stream->missed_metadata_flush = false;
2875 pthread_mutex_unlock(&stream->metadata_timer_lock);
2876 (void) consumer_flush_ust_index(stream);
2877 } else {
2878 pthread_mutex_unlock(&stream->metadata_timer_lock);
2879 }
2880
2881 if (err < 0) {
2882 goto error;
2883 }
2884 }
2885
2886 assert(!stream->metadata_flag);
2887 err = consumer_stream_write_index(stream, &index);
2888 if (err < 0) {
2889 goto error;
2890 }
2891
2892 rotate:
2893 /*
2894 * After extracting the packet, we check if the stream is now ready to be
2895 * rotated and perform the action immediately.
2896 */
2897 rotation_ret = lttng_consumer_stream_is_rotate_ready(stream);
2898 if (rotation_ret == 1) {
2899 rotation_ret = lttng_consumer_rotate_stream(ctx, stream, rotated);
2900 if (rotation_ret < 0) {
2901 ERR("Stream rotation error");
2902 ret = -1;
2903 goto error;
2904 }
2905 } else if (rotation_ret < 0) {
2906 ERR("Checking if stream is ready to rotate");
2907 ret = -1;
2908 goto error;
2909 }
2910 error:
2911 return ret;
2912 }
2913
2914 /*
2915 * Called when a stream is created.
2916 *
2917 * Return 0 on success or else a negative value.
2918 */
2919 int lttng_ustconsumer_on_recv_stream(struct lttng_consumer_stream *stream)
2920 {
2921 int ret;
2922
2923 assert(stream);
2924
2925 /* Don't create anything if this is set for streaming. */
2926 if (stream->net_seq_idx == (uint64_t) -1ULL && stream->chan->monitor) {
2927 ret = utils_create_stream_file(stream->chan->pathname, stream->name,
2928 stream->chan->tracefile_size, stream->tracefile_count_current,
2929 stream->uid, stream->gid, NULL);
2930 if (ret < 0) {
2931 goto error;
2932 }
2933 stream->out_fd = ret;
2934 stream->tracefile_size_current = 0;
2935
2936 if (!stream->metadata_flag) {
2937 struct lttng_index_file *index_file;
2938
2939 index_file = lttng_index_file_create(stream->chan->pathname,
2940 stream->name, stream->uid, stream->gid,
2941 stream->chan->tracefile_size,
2942 stream->tracefile_count_current,
2943 CTF_INDEX_MAJOR, CTF_INDEX_MINOR);
2944 if (!index_file) {
2945 goto error;
2946 }
2947 assert(!stream->index_file);
2948 stream->index_file = index_file;
2949 }
2950 }
2951 ret = 0;
2952
2953 error:
2954 return ret;
2955 }
2956
2957 /*
2958 * Check if data is still being extracted from the buffers for a specific
2959 * stream. Consumer data lock MUST be acquired before calling this function
2960 * and the stream lock.
2961 *
2962 * Return 1 if the traced data are still getting read else 0 meaning that the
2963 * data is available for trace viewer reading.
2964 */
2965 int lttng_ustconsumer_data_pending(struct lttng_consumer_stream *stream)
2966 {
2967 int ret;
2968
2969 assert(stream);
2970 assert(stream->ustream);
2971
2972 DBG("UST consumer checking data pending");
2973
2974 if (stream->endpoint_status != CONSUMER_ENDPOINT_ACTIVE) {
2975 ret = 0;
2976 goto end;
2977 }
2978
2979 if (stream->chan->type == CONSUMER_CHANNEL_TYPE_METADATA) {
2980 uint64_t contiguous, pushed;
2981
2982 /* Ease our life a bit. */
2983 contiguous = stream->chan->metadata_cache->max_offset;
2984 pushed = stream->ust_metadata_pushed;
2985
2986 /*
2987 * We can simply check whether all contiguously available data
2988 * has been pushed to the ring buffer, since the push operation
2989 * is performed within get_next_subbuf(), and because both
2990 * get_next_subbuf() and put_next_subbuf() are issued atomically
2991 * thanks to the stream lock within
2992 * lttng_ustconsumer_read_subbuffer(). This basically means that
2993 * whetnever ust_metadata_pushed is incremented, the associated
2994 * metadata has been consumed from the metadata stream.
2995 */
2996 DBG("UST consumer metadata pending check: contiguous %" PRIu64 " vs pushed %" PRIu64,
2997 contiguous, pushed);
2998 assert(((int64_t) (contiguous - pushed)) >= 0);
2999 if ((contiguous != pushed) ||
3000 (((int64_t) contiguous - pushed) > 0 || contiguous == 0)) {
3001 ret = 1; /* Data is pending */
3002 goto end;
3003 }
3004 } else {
3005 ret = ustctl_get_next_subbuf(stream->ustream);
3006 if (ret == 0) {
3007 /*
3008 * There is still data so let's put back this
3009 * subbuffer.
3010 */
3011 ret = ustctl_put_subbuf(stream->ustream);
3012 assert(ret == 0);
3013 ret = 1; /* Data is pending */
3014 goto end;
3015 }
3016 }
3017
3018 /* Data is NOT pending so ready to be read. */
3019 ret = 0;
3020
3021 end:
3022 return ret;
3023 }
3024
3025 /*
3026 * Stop a given metadata channel timer if enabled and close the wait fd which
3027 * is the poll pipe of the metadata stream.
3028 *
3029 * This MUST be called with the metadata channel acquired.
3030 */
3031 void lttng_ustconsumer_close_metadata(struct lttng_consumer_channel *metadata)
3032 {
3033 int ret;
3034
3035 assert(metadata);
3036 assert(metadata->type == CONSUMER_CHANNEL_TYPE_METADATA);
3037
3038 DBG("Closing metadata channel key %" PRIu64, metadata->key);
3039
3040 if (metadata->switch_timer_enabled == 1) {
3041 consumer_timer_switch_stop(metadata);
3042 }
3043
3044 if (!metadata->metadata_stream) {
3045 goto end;
3046 }
3047
3048 /*
3049 * Closing write side so the thread monitoring the stream wakes up if any
3050 * and clean the metadata stream.
3051 */
3052 if (metadata->metadata_stream->ust_metadata_poll_pipe[1] >= 0) {
3053 ret = close(metadata->metadata_stream->ust_metadata_poll_pipe[1]);
3054 if (ret < 0) {
3055 PERROR("closing metadata pipe write side");
3056 }
3057 metadata->metadata_stream->ust_metadata_poll_pipe[1] = -1;
3058 }
3059
3060 end:
3061 return;
3062 }
3063
3064 /*
3065 * Close every metadata stream wait fd of the metadata hash table. This
3066 * function MUST be used very carefully so not to run into a race between the
3067 * metadata thread handling streams and this function closing their wait fd.
3068 *
3069 * For UST, this is used when the session daemon hangs up. Its the metadata
3070 * producer so calling this is safe because we are assured that no state change
3071 * can occur in the metadata thread for the streams in the hash table.
3072 */
3073 void lttng_ustconsumer_close_all_metadata(struct lttng_ht *metadata_ht)
3074 {
3075 struct lttng_ht_iter iter;
3076 struct lttng_consumer_stream *stream;
3077
3078 assert(metadata_ht);
3079 assert(metadata_ht->ht);
3080
3081 DBG("UST consumer closing all metadata streams");
3082
3083 rcu_read_lock();
3084 cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream,
3085 node.node) {
3086
3087 health_code_update();
3088
3089 pthread_mutex_lock(&stream->chan->lock);
3090 lttng_ustconsumer_close_metadata(stream->chan);
3091 pthread_mutex_unlock(&stream->chan->lock);
3092
3093 }
3094 rcu_read_unlock();
3095 }
3096
3097 void lttng_ustconsumer_close_stream_wakeup(struct lttng_consumer_stream *stream)
3098 {
3099 int ret;
3100
3101 ret = ustctl_stream_close_wakeup_fd(stream->ustream);
3102 if (ret < 0) {
3103 ERR("Unable to close wakeup fd");
3104 }
3105 }
3106
3107 /*
3108 * Please refer to consumer-timer.c before adding any lock within this
3109 * function or any of its callees. Timers have a very strict locking
3110 * semantic with respect to teardown. Failure to respect this semantic
3111 * introduces deadlocks.
3112 *
3113 * DON'T hold the metadata lock when calling this function, else this
3114 * can cause deadlock involving consumer awaiting for metadata to be
3115 * pushed out due to concurrent interaction with the session daemon.
3116 */
3117 int lttng_ustconsumer_request_metadata(struct lttng_consumer_local_data *ctx,
3118 struct lttng_consumer_channel *channel, int timer, int wait)
3119 {
3120 struct lttcomm_metadata_request_msg request;
3121 struct lttcomm_consumer_msg msg;
3122 enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS;
3123 uint64_t len, key, offset, version;
3124 int ret;
3125
3126 assert(channel);
3127 assert(channel->metadata_cache);
3128
3129 memset(&request, 0, sizeof(request));
3130
3131 /* send the metadata request to sessiond */
3132 switch (consumer_data.type) {
3133 case LTTNG_CONSUMER64_UST:
3134 request.bits_per_long = 64;
3135 break;
3136 case LTTNG_CONSUMER32_UST:
3137 request.bits_per_long = 32;
3138 break;
3139 default:
3140 request.bits_per_long = 0;
3141 break;
3142 }
3143
3144 request.session_id = channel->session_id;
3145 request.session_id_per_pid = channel->session_id_per_pid;
3146 /*
3147 * Request the application UID here so the metadata of that application can
3148 * be sent back. The channel UID corresponds to the user UID of the session
3149 * used for the rights on the stream file(s).
3150 */
3151 request.uid = channel->ust_app_uid;
3152 request.key = channel->key;
3153
3154 DBG("Sending metadata request to sessiond, session id %" PRIu64
3155 ", per-pid %" PRIu64 ", app UID %u and channel key %" PRIu64,
3156 request.session_id, request.session_id_per_pid, request.uid,
3157 request.key);
3158
3159 pthread_mutex_lock(&ctx->metadata_socket_lock);
3160
3161 health_code_update();
3162
3163 ret = lttcomm_send_unix_sock(ctx->consumer_metadata_socket, &request,
3164 sizeof(request));
3165 if (ret < 0) {
3166 ERR("Asking metadata to sessiond");
3167 goto end;
3168 }
3169
3170 health_code_update();
3171
3172 /* Receive the metadata from sessiond */
3173 ret = lttcomm_recv_unix_sock(ctx->consumer_metadata_socket, &msg,
3174 sizeof(msg));
3175 if (ret != sizeof(msg)) {
3176 DBG("Consumer received unexpected message size %d (expects %zu)",
3177 ret, sizeof(msg));
3178 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD);
3179 /*
3180 * The ret value might 0 meaning an orderly shutdown but this is ok
3181 * since the caller handles this.
3182 */
3183 goto end;
3184 }
3185
3186 health_code_update();
3187
3188 if (msg.cmd_type == LTTNG_ERR_UND) {
3189 /* No registry found */
3190 (void) consumer_send_status_msg(ctx->consumer_metadata_socket,
3191 ret_code);
3192 ret = 0;
3193 goto end;
3194 } else if (msg.cmd_type != LTTNG_CONSUMER_PUSH_METADATA) {
3195 ERR("Unexpected cmd_type received %d", msg.cmd_type);
3196 ret = -1;
3197 goto end;
3198 }
3199
3200 len = msg.u.push_metadata.len;
3201 key = msg.u.push_metadata.key;
3202 offset = msg.u.push_metadata.target_offset;
3203 version = msg.u.push_metadata.version;
3204
3205 assert(key == channel->key);
3206 if (len == 0) {
3207 DBG("No new metadata to receive for key %" PRIu64, key);
3208 }
3209
3210 health_code_update();
3211
3212 /* Tell session daemon we are ready to receive the metadata. */
3213 ret = consumer_send_status_msg(ctx->consumer_metadata_socket,
3214 LTTCOMM_CONSUMERD_SUCCESS);
3215 if (ret < 0 || len == 0) {
3216 /*
3217 * Somehow, the session daemon is not responding anymore or there is
3218 * nothing to receive.
3219 */
3220 goto end;
3221 }
3222
3223 health_code_update();
3224
3225 ret = lttng_ustconsumer_recv_metadata(ctx->consumer_metadata_socket,
3226 key, offset, len, version, channel, timer, wait);
3227 if (ret >= 0) {
3228 /*
3229 * Only send the status msg if the sessiond is alive meaning a positive
3230 * ret code.
3231 */
3232 (void) consumer_send_status_msg(ctx->consumer_metadata_socket, ret);
3233 }
3234 ret = 0;
3235
3236 end:
3237 health_code_update();
3238
3239 pthread_mutex_unlock(&ctx->metadata_socket_lock);
3240 return ret;
3241 }
3242
3243 /*
3244 * Return the ustctl call for the get stream id.
3245 */
3246 int lttng_ustconsumer_get_stream_id(struct lttng_consumer_stream *stream,
3247 uint64_t *stream_id)
3248 {
3249 assert(stream);
3250 assert(stream_id);
3251
3252 return ustctl_get_stream_id(stream->ustream, stream_id);
3253 }
This page took 0.13506 seconds and 6 git commands to generate.