Basic structure for the rotate command
[deliverable/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
36 #include <bin/lttng-consumerd/health-consumerd.h>
37 #include <common/common.h>
38 #include <common/sessiond-comm/sessiond-comm.h>
39 #include <common/relayd/relayd.h>
40 #include <common/compat/fcntl.h>
41 #include <common/compat/endian.h>
42 #include <common/consumer/consumer-metadata-cache.h>
43 #include <common/consumer/consumer-stream.h>
44 #include <common/consumer/consumer-timer.h>
45 #include <common/utils.h>
46 #include <common/index/index.h>
47
48 #include "ust-consumer.h"
49
50 #define INT_MAX_STR_LEN 12 /* includes \0 */
51
52 extern struct lttng_consumer_global_data consumer_data;
53 extern int consumer_poll_timeout;
54
55 /*
56 * Free channel object and all streams associated with it. This MUST be used
57 * only and only if the channel has _NEVER_ been added to the global channel
58 * hash table.
59 */
60 static void destroy_channel(struct lttng_consumer_channel *channel)
61 {
62 struct lttng_consumer_stream *stream, *stmp;
63
64 assert(channel);
65
66 DBG("UST consumer cleaning stream list");
67
68 cds_list_for_each_entry_safe(stream, stmp, &channel->streams.head,
69 send_node) {
70
71 health_code_update();
72
73 cds_list_del(&stream->send_node);
74 ustctl_destroy_stream(stream->ustream);
75 free(stream);
76 }
77
78 /*
79 * If a channel is available meaning that was created before the streams
80 * were, delete it.
81 */
82 if (channel->uchan) {
83 lttng_ustconsumer_del_channel(channel);
84 lttng_ustconsumer_free_channel(channel);
85 }
86 free(channel);
87 }
88
89 /*
90 * Add channel to internal consumer state.
91 *
92 * Returns 0 on success or else a negative value.
93 */
94 static int add_channel(struct lttng_consumer_channel *channel,
95 struct lttng_consumer_local_data *ctx)
96 {
97 int ret = 0;
98
99 assert(channel);
100 assert(ctx);
101
102 if (ctx->on_recv_channel != NULL) {
103 ret = ctx->on_recv_channel(channel);
104 if (ret == 0) {
105 ret = consumer_add_channel(channel, ctx);
106 } else if (ret < 0) {
107 /* Most likely an ENOMEM. */
108 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
109 goto error;
110 }
111 } else {
112 ret = consumer_add_channel(channel, ctx);
113 }
114
115 DBG("UST consumer channel added (key: %" PRIu64 ")", channel->key);
116
117 error:
118 return ret;
119 }
120
121 /*
122 * Allocate and return a consumer channel object.
123 */
124 static struct lttng_consumer_channel *allocate_channel(uint64_t session_id,
125 const char *pathname, const char *name, uid_t uid, gid_t gid,
126 uint64_t relayd_id, uint64_t key, enum lttng_event_output output,
127 uint64_t tracefile_size, uint64_t tracefile_count,
128 uint64_t session_id_per_pid, unsigned int monitor,
129 unsigned int live_timer_interval,
130 const char *root_shm_path, const char *shm_path)
131 {
132 assert(pathname);
133 assert(name);
134
135 return consumer_allocate_channel(key, session_id, pathname, name, uid,
136 gid, relayd_id, output, tracefile_size,
137 tracefile_count, session_id_per_pid, monitor,
138 live_timer_interval, root_shm_path, shm_path);
139 }
140
141 /*
142 * Allocate and return a consumer stream object. If _alloc_ret is not NULL, the
143 * error value if applicable is set in it else it is kept untouched.
144 *
145 * Return NULL on error else the newly allocated stream object.
146 */
147 static struct lttng_consumer_stream *allocate_stream(int cpu, int key,
148 struct lttng_consumer_channel *channel,
149 struct lttng_consumer_local_data *ctx, int *_alloc_ret)
150 {
151 int alloc_ret;
152 struct lttng_consumer_stream *stream = NULL;
153
154 assert(channel);
155 assert(ctx);
156
157 stream = consumer_allocate_stream(channel->key,
158 key,
159 LTTNG_CONSUMER_ACTIVE_STREAM,
160 channel->name,
161 channel->uid,
162 channel->gid,
163 channel->relayd_id,
164 channel->session_id,
165 cpu,
166 &alloc_ret,
167 channel->type,
168 channel->monitor);
169 if (stream == NULL) {
170 switch (alloc_ret) {
171 case -ENOENT:
172 /*
173 * We could not find the channel. Can happen if cpu hotplug
174 * happens while tearing down.
175 */
176 DBG3("Could not find channel");
177 break;
178 case -ENOMEM:
179 case -EINVAL:
180 default:
181 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
182 break;
183 }
184 goto error;
185 }
186
187 stream->chan = channel;
188
189 error:
190 if (_alloc_ret) {
191 *_alloc_ret = alloc_ret;
192 }
193 return stream;
194 }
195
196 /*
197 * Send the given stream pointer to the corresponding thread.
198 *
199 * Returns 0 on success else a negative value.
200 */
201 static int send_stream_to_thread(struct lttng_consumer_stream *stream,
202 struct lttng_consumer_local_data *ctx)
203 {
204 int ret;
205 struct lttng_pipe *stream_pipe;
206
207 /* Get the right pipe where the stream will be sent. */
208 if (stream->metadata_flag) {
209 ret = consumer_add_metadata_stream(stream);
210 if (ret) {
211 ERR("Consumer add metadata stream %" PRIu64 " failed.",
212 stream->key);
213 goto error;
214 }
215 stream_pipe = ctx->consumer_metadata_pipe;
216 } else {
217 ret = consumer_add_data_stream(stream);
218 if (ret) {
219 ERR("Consumer add stream %" PRIu64 " failed.",
220 stream->key);
221 goto error;
222 }
223 stream_pipe = ctx->consumer_data_pipe;
224 }
225
226 /*
227 * From this point on, the stream's ownership has been moved away from
228 * the channel and becomes globally visible.
229 */
230 stream->globally_visible = 1;
231
232 ret = lttng_pipe_write(stream_pipe, &stream, sizeof(stream));
233 if (ret < 0) {
234 ERR("Consumer write %s stream to pipe %d",
235 stream->metadata_flag ? "metadata" : "data",
236 lttng_pipe_get_writefd(stream_pipe));
237 if (stream->metadata_flag) {
238 consumer_del_stream_for_metadata(stream);
239 } else {
240 consumer_del_stream_for_data(stream);
241 }
242 }
243 error:
244 return ret;
245 }
246
247 static
248 int get_stream_shm_path(char *stream_shm_path, const char *shm_path, int cpu)
249 {
250 char cpu_nr[INT_MAX_STR_LEN]; /* int max len */
251 int ret;
252
253 strncpy(stream_shm_path, shm_path, PATH_MAX);
254 stream_shm_path[PATH_MAX - 1] = '\0';
255 ret = snprintf(cpu_nr, INT_MAX_STR_LEN, "%i", cpu);
256 if (ret < 0) {
257 PERROR("snprintf");
258 goto end;
259 }
260 strncat(stream_shm_path, cpu_nr,
261 PATH_MAX - strlen(stream_shm_path) - 1);
262 ret = 0;
263 end:
264 return ret;
265 }
266
267 /*
268 * Create streams for the given channel using liblttng-ust-ctl.
269 *
270 * Return 0 on success else a negative value.
271 */
272 static int create_ust_streams(struct lttng_consumer_channel *channel,
273 struct lttng_consumer_local_data *ctx)
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 if (!stream) {
306 goto error_alloc;
307 }
308 stream->ustream = ustream;
309 /*
310 * Store it so we can save multiple function calls afterwards since
311 * this value is used heavily in the stream threads. This is UST
312 * specific so this is why it's done after allocation.
313 */
314 stream->wait_fd = wait_fd;
315
316 /*
317 * Increment channel refcount since the channel reference has now been
318 * assigned in the allocation process above.
319 */
320 if (stream->chan->monitor) {
321 uatomic_inc(&stream->chan->refcount);
322 }
323
324 /*
325 * Order is important this is why a list is used. On error, the caller
326 * should clean this list.
327 */
328 cds_list_add_tail(&stream->send_node, &channel->streams.head);
329
330 ret = ustctl_get_max_subbuf_size(stream->ustream,
331 &stream->max_sb_size);
332 if (ret < 0) {
333 ERR("ustctl_get_max_subbuf_size failed for stream %s",
334 stream->name);
335 goto error;
336 }
337
338 /* Do actions once stream has been received. */
339 if (ctx->on_recv_stream) {
340 ret = ctx->on_recv_stream(stream);
341 if (ret < 0) {
342 goto error;
343 }
344 }
345
346 DBG("UST consumer add stream %s (key: %" PRIu64 ") with relayd id %" PRIu64,
347 stream->name, stream->key, stream->relayd_stream_id);
348
349 /* Set next CPU stream. */
350 channel->streams.count = ++cpu;
351
352 /* Keep stream reference when creating metadata. */
353 if (channel->type == CONSUMER_CHANNEL_TYPE_METADATA) {
354 channel->metadata_stream = stream;
355 if (channel->monitor) {
356 /* Set metadata poll pipe if we created one */
357 memcpy(stream->ust_metadata_poll_pipe,
358 ust_metadata_pipe,
359 sizeof(ust_metadata_pipe));
360 }
361 }
362 }
363
364 return 0;
365
366 error:
367 error_alloc:
368 return ret;
369 }
370
371 /*
372 * create_posix_shm is never called concurrently within a process.
373 */
374 static
375 int create_posix_shm(void)
376 {
377 char tmp_name[NAME_MAX];
378 int shmfd, ret;
379
380 ret = snprintf(tmp_name, NAME_MAX, "/ust-shm-consumer-%d", getpid());
381 if (ret < 0) {
382 PERROR("snprintf");
383 return -1;
384 }
385 /*
386 * Allocate shm, and immediately unlink its shm oject, keeping
387 * only the file descriptor as a reference to the object.
388 * We specifically do _not_ use the / at the beginning of the
389 * pathname so that some OS implementations can keep it local to
390 * the process (POSIX leaves this implementation-defined).
391 */
392 shmfd = shm_open(tmp_name, O_CREAT | O_EXCL | O_RDWR, 0700);
393 if (shmfd < 0) {
394 PERROR("shm_open");
395 goto error_shm_open;
396 }
397 ret = shm_unlink(tmp_name);
398 if (ret < 0 && errno != ENOENT) {
399 PERROR("shm_unlink");
400 goto error_shm_release;
401 }
402 return shmfd;
403
404 error_shm_release:
405 ret = close(shmfd);
406 if (ret) {
407 PERROR("close");
408 }
409 error_shm_open:
410 return -1;
411 }
412
413 static int open_ust_stream_fd(struct lttng_consumer_channel *channel,
414 struct ustctl_consumer_channel_attr *attr,
415 int cpu)
416 {
417 char shm_path[PATH_MAX];
418 int ret;
419
420 if (!channel->shm_path[0]) {
421 return create_posix_shm();
422 }
423 ret = get_stream_shm_path(shm_path, channel->shm_path, cpu);
424 if (ret) {
425 goto error_shm_path;
426 }
427 return run_as_open(shm_path,
428 O_RDWR | O_CREAT | O_EXCL, S_IRUSR | S_IWUSR,
429 channel->uid, channel->gid);
430
431 error_shm_path:
432 return -1;
433 }
434
435 /*
436 * Create an UST channel with the given attributes and send it to the session
437 * daemon using the ust ctl API.
438 *
439 * Return 0 on success or else a negative value.
440 */
441 static int create_ust_channel(struct lttng_consumer_channel *channel,
442 struct ustctl_consumer_channel_attr *attr,
443 struct ustctl_consumer_channel **ust_chanp)
444 {
445 int ret, nr_stream_fds, i, j;
446 int *stream_fds;
447 struct ustctl_consumer_channel *ust_channel;
448
449 assert(channel);
450 assert(attr);
451 assert(ust_chanp);
452
453 DBG3("Creating channel to ustctl with attr: [overwrite: %d, "
454 "subbuf_size: %" PRIu64 ", num_subbuf: %" PRIu64 ", "
455 "switch_timer_interval: %u, read_timer_interval: %u, "
456 "output: %d, type: %d", attr->overwrite, attr->subbuf_size,
457 attr->num_subbuf, attr->switch_timer_interval,
458 attr->read_timer_interval, attr->output, attr->type);
459
460 if (channel->type == CONSUMER_CHANNEL_TYPE_METADATA)
461 nr_stream_fds = 1;
462 else
463 nr_stream_fds = ustctl_get_nr_stream_per_channel();
464 stream_fds = zmalloc(nr_stream_fds * sizeof(*stream_fds));
465 if (!stream_fds) {
466 ret = -1;
467 goto error_alloc;
468 }
469 for (i = 0; i < nr_stream_fds; i++) {
470 stream_fds[i] = open_ust_stream_fd(channel, attr, i);
471 if (stream_fds[i] < 0) {
472 ret = -1;
473 goto error_open;
474 }
475 }
476 ust_channel = ustctl_create_channel(attr, stream_fds, nr_stream_fds);
477 if (!ust_channel) {
478 ret = -1;
479 goto error_create;
480 }
481 channel->nr_stream_fds = nr_stream_fds;
482 channel->stream_fds = stream_fds;
483 *ust_chanp = ust_channel;
484
485 return 0;
486
487 error_create:
488 error_open:
489 for (j = i - 1; j >= 0; j--) {
490 int closeret;
491
492 closeret = close(stream_fds[j]);
493 if (closeret) {
494 PERROR("close");
495 }
496 if (channel->shm_path[0]) {
497 char shm_path[PATH_MAX];
498
499 closeret = get_stream_shm_path(shm_path,
500 channel->shm_path, j);
501 if (closeret) {
502 ERR("Cannot get stream shm path");
503 }
504 closeret = run_as_unlink(shm_path,
505 channel->uid, channel->gid);
506 if (closeret) {
507 PERROR("unlink %s", shm_path);
508 }
509 }
510 }
511 /* Try to rmdir all directories under shm_path root. */
512 if (channel->root_shm_path[0]) {
513 (void) run_as_rmdir_recursive(channel->root_shm_path,
514 channel->uid, channel->gid);
515 }
516 free(stream_fds);
517 error_alloc:
518 return ret;
519 }
520
521 /*
522 * Send a single given stream to the session daemon using the sock.
523 *
524 * Return 0 on success else a negative value.
525 */
526 static int send_sessiond_stream(int sock, struct lttng_consumer_stream *stream)
527 {
528 int ret;
529
530 assert(stream);
531 assert(sock >= 0);
532
533 DBG("UST consumer sending stream %" PRIu64 " to sessiond", stream->key);
534
535 /* Send stream to session daemon. */
536 ret = ustctl_send_stream_to_sessiond(sock, stream->ustream);
537 if (ret < 0) {
538 goto error;
539 }
540
541 error:
542 return ret;
543 }
544
545 /*
546 * Send channel to sessiond.
547 *
548 * Return 0 on success or else a negative value.
549 */
550 static int send_sessiond_channel(int sock,
551 struct lttng_consumer_channel *channel,
552 struct lttng_consumer_local_data *ctx, int *relayd_error)
553 {
554 int ret, ret_code = LTTCOMM_CONSUMERD_SUCCESS;
555 struct lttng_consumer_stream *stream;
556 uint64_t net_seq_idx = -1ULL;
557
558 assert(channel);
559 assert(ctx);
560 assert(sock >= 0);
561
562 DBG("UST consumer sending channel %s to sessiond", channel->name);
563
564 if (channel->relayd_id != (uint64_t) -1ULL) {
565 cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
566
567 health_code_update();
568
569 /* Try to send the stream to the relayd if one is available. */
570 ret = consumer_send_relayd_stream(stream, stream->chan->pathname);
571 if (ret < 0) {
572 /*
573 * Flag that the relayd was the problem here probably due to a
574 * communicaton error on the socket.
575 */
576 if (relayd_error) {
577 *relayd_error = 1;
578 }
579 ret_code = LTTCOMM_CONSUMERD_RELAYD_FAIL;
580 }
581 if (net_seq_idx == -1ULL) {
582 net_seq_idx = stream->net_seq_idx;
583 }
584 }
585 }
586
587 /* Inform sessiond that we are about to send channel and streams. */
588 ret = consumer_send_status_msg(sock, ret_code);
589 if (ret < 0 || ret_code != LTTCOMM_CONSUMERD_SUCCESS) {
590 /*
591 * Either the session daemon is not responding or the relayd died so we
592 * stop now.
593 */
594 goto error;
595 }
596
597 /* Send channel to sessiond. */
598 ret = ustctl_send_channel_to_sessiond(sock, channel->uchan);
599 if (ret < 0) {
600 goto error;
601 }
602
603 ret = ustctl_channel_close_wakeup_fd(channel->uchan);
604 if (ret < 0) {
605 goto error;
606 }
607
608 /* The channel was sent successfully to the sessiond at this point. */
609 cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
610
611 health_code_update();
612
613 /* Send stream to session daemon. */
614 ret = send_sessiond_stream(sock, stream);
615 if (ret < 0) {
616 goto error;
617 }
618 }
619
620 /* Tell sessiond there is no more stream. */
621 ret = ustctl_send_stream_to_sessiond(sock, NULL);
622 if (ret < 0) {
623 goto error;
624 }
625
626 DBG("UST consumer NULL stream sent to sessiond");
627
628 return 0;
629
630 error:
631 if (ret_code != LTTCOMM_CONSUMERD_SUCCESS) {
632 ret = -1;
633 }
634 return ret;
635 }
636
637 /*
638 * Creates a channel and streams and add the channel it to the channel internal
639 * state. The created stream must ONLY be sent once the GET_CHANNEL command is
640 * received.
641 *
642 * Return 0 on success or else, a negative value is returned and the channel
643 * MUST be destroyed by consumer_del_channel().
644 */
645 static int ask_channel(struct lttng_consumer_local_data *ctx, int sock,
646 struct lttng_consumer_channel *channel,
647 struct ustctl_consumer_channel_attr *attr)
648 {
649 int ret;
650
651 assert(ctx);
652 assert(channel);
653 assert(attr);
654
655 /*
656 * This value is still used by the kernel consumer since for the kernel,
657 * the stream ownership is not IN the consumer so we need to have the
658 * number of left stream that needs to be initialized so we can know when
659 * to delete the channel (see consumer.c).
660 *
661 * As for the user space tracer now, the consumer creates and sends the
662 * stream to the session daemon which only sends them to the application
663 * once every stream of a channel is received making this value useless
664 * because we they will be added to the poll thread before the application
665 * receives them. This ensures that a stream can not hang up during
666 * initilization of a channel.
667 */
668 channel->nb_init_stream_left = 0;
669
670 /* The reply msg status is handled in the following call. */
671 ret = create_ust_channel(channel, attr, &channel->uchan);
672 if (ret < 0) {
673 goto end;
674 }
675
676 channel->wait_fd = ustctl_channel_get_wait_fd(channel->uchan);
677
678 /*
679 * For the snapshots (no monitor), we create the metadata streams
680 * on demand, not during the channel creation.
681 */
682 if (channel->type == CONSUMER_CHANNEL_TYPE_METADATA && !channel->monitor) {
683 ret = 0;
684 goto end;
685 }
686
687 /* Open all streams for this channel. */
688 ret = create_ust_streams(channel, ctx);
689 if (ret < 0) {
690 goto end;
691 }
692
693 end:
694 return ret;
695 }
696
697 /*
698 * Send all stream of a channel to the right thread handling it.
699 *
700 * On error, return a negative value else 0 on success.
701 */
702 static int send_streams_to_thread(struct lttng_consumer_channel *channel,
703 struct lttng_consumer_local_data *ctx)
704 {
705 int ret = 0;
706 struct lttng_consumer_stream *stream, *stmp;
707
708 assert(channel);
709 assert(ctx);
710
711 /* Send streams to the corresponding thread. */
712 cds_list_for_each_entry_safe(stream, stmp, &channel->streams.head,
713 send_node) {
714
715 health_code_update();
716
717 /* Sending the stream to the thread. */
718 ret = send_stream_to_thread(stream, ctx);
719 if (ret < 0) {
720 /*
721 * If we are unable to send the stream to the thread, there is
722 * a big problem so just stop everything.
723 */
724 /* Remove node from the channel stream list. */
725 cds_list_del(&stream->send_node);
726 goto error;
727 }
728
729 /* Remove node from the channel stream list. */
730 cds_list_del(&stream->send_node);
731
732 }
733
734 error:
735 return ret;
736 }
737
738 /*
739 * Flush channel's streams using the given key to retrieve the channel.
740 *
741 * Return 0 on success else an LTTng error code.
742 */
743 static int flush_channel(uint64_t chan_key)
744 {
745 int ret = 0;
746 struct lttng_consumer_channel *channel;
747 struct lttng_consumer_stream *stream;
748 struct lttng_ht *ht;
749 struct lttng_ht_iter iter;
750
751 DBG("UST consumer flush channel key %" PRIu64, chan_key);
752
753 rcu_read_lock();
754 channel = consumer_find_channel(chan_key);
755 if (!channel) {
756 ERR("UST consumer flush channel %" PRIu64 " not found", chan_key);
757 ret = LTTNG_ERR_UST_CHAN_NOT_FOUND;
758 goto error;
759 }
760
761 ht = consumer_data.stream_per_chan_id_ht;
762
763 /* For each stream of the channel id, flush it. */
764 cds_lfht_for_each_entry_duplicate(ht->ht,
765 ht->hash_fct(&channel->key, lttng_ht_seed), ht->match_fct,
766 &channel->key, &iter.iter, stream, node_channel_id.node) {
767
768 health_code_update();
769
770 pthread_mutex_lock(&stream->lock);
771 if (!stream->quiescent) {
772 ustctl_flush_buffer(stream->ustream, 0);
773 stream->quiescent = true;
774 }
775 pthread_mutex_unlock(&stream->lock);
776 }
777 error:
778 rcu_read_unlock();
779 return ret;
780 }
781
782 /*
783 * Clear quiescent state from channel's streams using the given key to
784 * retrieve the channel.
785 *
786 * Return 0 on success else an LTTng error code.
787 */
788 static int clear_quiescent_channel(uint64_t chan_key)
789 {
790 int ret = 0;
791 struct lttng_consumer_channel *channel;
792 struct lttng_consumer_stream *stream;
793 struct lttng_ht *ht;
794 struct lttng_ht_iter iter;
795
796 DBG("UST consumer clear quiescent channel key %" PRIu64, chan_key);
797
798 rcu_read_lock();
799 channel = consumer_find_channel(chan_key);
800 if (!channel) {
801 ERR("UST consumer clear quiescent channel %" PRIu64 " not found", chan_key);
802 ret = LTTNG_ERR_UST_CHAN_NOT_FOUND;
803 goto error;
804 }
805
806 ht = consumer_data.stream_per_chan_id_ht;
807
808 /* For each stream of the channel id, clear quiescent state. */
809 cds_lfht_for_each_entry_duplicate(ht->ht,
810 ht->hash_fct(&channel->key, lttng_ht_seed), ht->match_fct,
811 &channel->key, &iter.iter, stream, node_channel_id.node) {
812
813 health_code_update();
814
815 pthread_mutex_lock(&stream->lock);
816 stream->quiescent = false;
817 pthread_mutex_unlock(&stream->lock);
818 }
819 error:
820 rcu_read_unlock();
821 return ret;
822 }
823
824 /*
825 * Close metadata stream wakeup_fd using the given key to retrieve the channel.
826 * RCU read side lock MUST be acquired before calling this function.
827 *
828 * Return 0 on success else an LTTng error code.
829 */
830 static int close_metadata(uint64_t chan_key)
831 {
832 int ret = 0;
833 struct lttng_consumer_channel *channel;
834
835 DBG("UST consumer close metadata key %" PRIu64, chan_key);
836
837 channel = consumer_find_channel(chan_key);
838 if (!channel) {
839 /*
840 * This is possible if the metadata thread has issue a delete because
841 * the endpoint point of the stream hung up. There is no way the
842 * session daemon can know about it thus use a DBG instead of an actual
843 * error.
844 */
845 DBG("UST consumer close metadata %" PRIu64 " not found", chan_key);
846 ret = LTTNG_ERR_UST_CHAN_NOT_FOUND;
847 goto error;
848 }
849
850 pthread_mutex_lock(&consumer_data.lock);
851 pthread_mutex_lock(&channel->lock);
852
853 if (cds_lfht_is_node_deleted(&channel->node.node)) {
854 goto error_unlock;
855 }
856
857 lttng_ustconsumer_close_metadata(channel);
858
859 error_unlock:
860 pthread_mutex_unlock(&channel->lock);
861 pthread_mutex_unlock(&consumer_data.lock);
862 error:
863 return ret;
864 }
865
866 /*
867 * RCU read side lock MUST be acquired before calling this function.
868 *
869 * Return 0 on success else an LTTng error code.
870 */
871 static int setup_metadata(struct lttng_consumer_local_data *ctx, uint64_t key)
872 {
873 int ret;
874 struct lttng_consumer_channel *metadata;
875
876 DBG("UST consumer setup metadata key %" PRIu64, key);
877
878 metadata = consumer_find_channel(key);
879 if (!metadata) {
880 ERR("UST consumer push metadata %" PRIu64 " not found", key);
881 ret = LTTNG_ERR_UST_CHAN_NOT_FOUND;
882 goto end;
883 }
884
885 /*
886 * In no monitor mode, the metadata channel has no stream(s) so skip the
887 * ownership transfer to the metadata thread.
888 */
889 if (!metadata->monitor) {
890 DBG("Metadata channel in no monitor");
891 ret = 0;
892 goto end;
893 }
894
895 /*
896 * Send metadata stream to relayd if one available. Availability is
897 * known if the stream is still in the list of the channel.
898 */
899 if (cds_list_empty(&metadata->streams.head)) {
900 ERR("Metadata channel key %" PRIu64 ", no stream available.", key);
901 ret = LTTCOMM_CONSUMERD_ERROR_METADATA;
902 goto error_no_stream;
903 }
904
905 /* Send metadata stream to relayd if needed. */
906 if (metadata->metadata_stream->net_seq_idx != (uint64_t) -1ULL) {
907 ret = consumer_send_relayd_stream(metadata->metadata_stream,
908 metadata->pathname);
909 if (ret < 0) {
910 ret = LTTCOMM_CONSUMERD_ERROR_METADATA;
911 goto error;
912 }
913 ret = consumer_send_relayd_streams_sent(
914 metadata->metadata_stream->net_seq_idx);
915 if (ret < 0) {
916 ret = LTTCOMM_CONSUMERD_RELAYD_FAIL;
917 goto error;
918 }
919 }
920
921 ret = send_streams_to_thread(metadata, ctx);
922 if (ret < 0) {
923 /*
924 * If we are unable to send the stream to the thread, there is
925 * a big problem so just stop everything.
926 */
927 ret = LTTCOMM_CONSUMERD_FATAL;
928 goto error;
929 }
930 /* List MUST be empty after or else it could be reused. */
931 assert(cds_list_empty(&metadata->streams.head));
932
933 ret = 0;
934 goto end;
935
936 error:
937 /*
938 * Delete metadata channel on error. At this point, the metadata stream can
939 * NOT be monitored by the metadata thread thus having the guarantee that
940 * the stream is still in the local stream list of the channel. This call
941 * will make sure to clean that list.
942 */
943 consumer_stream_destroy(metadata->metadata_stream, NULL);
944 cds_list_del(&metadata->metadata_stream->send_node);
945 metadata->metadata_stream = NULL;
946 error_no_stream:
947 end:
948 return ret;
949 }
950
951 /*
952 * Snapshot the whole metadata.
953 *
954 * Returns 0 on success, < 0 on error
955 */
956 static int snapshot_metadata(uint64_t key, char *path, uint64_t relayd_id,
957 struct lttng_consumer_local_data *ctx)
958 {
959 int ret = 0;
960 struct lttng_consumer_channel *metadata_channel;
961 struct lttng_consumer_stream *metadata_stream;
962
963 assert(path);
964 assert(ctx);
965
966 DBG("UST consumer snapshot metadata with key %" PRIu64 " at path %s",
967 key, path);
968
969 rcu_read_lock();
970
971 metadata_channel = consumer_find_channel(key);
972 if (!metadata_channel) {
973 ERR("UST snapshot metadata channel not found for key %" PRIu64,
974 key);
975 ret = -1;
976 goto error;
977 }
978 assert(!metadata_channel->monitor);
979
980 health_code_update();
981
982 /*
983 * Ask the sessiond if we have new metadata waiting and update the
984 * consumer metadata cache.
985 */
986 ret = lttng_ustconsumer_request_metadata(ctx, metadata_channel, 0, 1);
987 if (ret < 0) {
988 goto error;
989 }
990
991 health_code_update();
992
993 /*
994 * The metadata stream is NOT created in no monitor mode when the channel
995 * is created on a sessiond ask channel command.
996 */
997 ret = create_ust_streams(metadata_channel, ctx);
998 if (ret < 0) {
999 goto error;
1000 }
1001
1002 metadata_stream = metadata_channel->metadata_stream;
1003 assert(metadata_stream);
1004
1005 if (relayd_id != (uint64_t) -1ULL) {
1006 metadata_stream->net_seq_idx = relayd_id;
1007 ret = consumer_send_relayd_stream(metadata_stream, path);
1008 if (ret < 0) {
1009 goto error_stream;
1010 }
1011 } else {
1012 ret = utils_create_stream_file(path, metadata_stream->name,
1013 metadata_stream->chan->tracefile_size,
1014 metadata_stream->tracefile_count_current,
1015 metadata_stream->uid, metadata_stream->gid, NULL);
1016 if (ret < 0) {
1017 goto error_stream;
1018 }
1019 metadata_stream->out_fd = ret;
1020 metadata_stream->tracefile_size_current = 0;
1021 }
1022
1023 do {
1024 health_code_update();
1025
1026 ret = lttng_consumer_read_subbuffer(metadata_stream, ctx);
1027 if (ret < 0) {
1028 goto error_stream;
1029 }
1030 } while (ret > 0);
1031
1032 error_stream:
1033 /*
1034 * Clean up the stream completly because the next snapshot will use a new
1035 * metadata stream.
1036 */
1037 consumer_stream_destroy(metadata_stream, NULL);
1038 cds_list_del(&metadata_stream->send_node);
1039 metadata_channel->metadata_stream = NULL;
1040
1041 error:
1042 rcu_read_unlock();
1043 return ret;
1044 }
1045
1046 /*
1047 * Take a snapshot of all the stream of a channel.
1048 *
1049 * Returns 0 on success, < 0 on error
1050 */
1051 static int snapshot_channel(uint64_t key, char *path, uint64_t relayd_id,
1052 uint64_t nb_packets_per_stream, struct lttng_consumer_local_data *ctx)
1053 {
1054 int ret;
1055 unsigned use_relayd = 0;
1056 unsigned long consumed_pos, produced_pos;
1057 struct lttng_consumer_channel *channel;
1058 struct lttng_consumer_stream *stream;
1059
1060 assert(path);
1061 assert(ctx);
1062
1063 rcu_read_lock();
1064
1065 if (relayd_id != (uint64_t) -1ULL) {
1066 use_relayd = 1;
1067 }
1068
1069 channel = consumer_find_channel(key);
1070 if (!channel) {
1071 ERR("UST snapshot channel not found for key %" PRIu64, key);
1072 ret = -1;
1073 goto error;
1074 }
1075 assert(!channel->monitor);
1076 DBG("UST consumer snapshot channel %" PRIu64, key);
1077
1078 cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
1079 health_code_update();
1080
1081 /* Lock stream because we are about to change its state. */
1082 pthread_mutex_lock(&stream->lock);
1083 stream->net_seq_idx = relayd_id;
1084
1085 if (use_relayd) {
1086 ret = consumer_send_relayd_stream(stream, path);
1087 if (ret < 0) {
1088 goto error_unlock;
1089 }
1090 } else {
1091 ret = utils_create_stream_file(path, stream->name,
1092 stream->chan->tracefile_size,
1093 stream->tracefile_count_current,
1094 stream->uid, stream->gid, NULL);
1095 if (ret < 0) {
1096 goto error_unlock;
1097 }
1098 stream->out_fd = ret;
1099 stream->tracefile_size_current = 0;
1100
1101 DBG("UST consumer snapshot stream %s/%s (%" PRIu64 ")", path,
1102 stream->name, stream->key);
1103 }
1104
1105 /*
1106 * If tracing is active, we want to perform a "full" buffer flush.
1107 * Else, if quiescent, it has already been done by the prior stop.
1108 */
1109 if (!stream->quiescent) {
1110 ustctl_flush_buffer(stream->ustream, 0);
1111 }
1112
1113 ret = lttng_ustconsumer_take_snapshot(stream);
1114 if (ret < 0) {
1115 ERR("Taking UST snapshot");
1116 goto error_unlock;
1117 }
1118
1119 ret = lttng_ustconsumer_get_produced_snapshot(stream, &produced_pos);
1120 if (ret < 0) {
1121 ERR("Produced UST snapshot position");
1122 goto error_unlock;
1123 }
1124
1125 ret = lttng_ustconsumer_get_consumed_snapshot(stream, &consumed_pos);
1126 if (ret < 0) {
1127 ERR("Consumerd UST snapshot position");
1128 goto error_unlock;
1129 }
1130
1131 /*
1132 * The original value is sent back if max stream size is larger than
1133 * the possible size of the snapshot. Also, we assume that the session
1134 * daemon should never send a maximum stream size that is lower than
1135 * subbuffer size.
1136 */
1137 consumed_pos = consumer_get_consume_start_pos(consumed_pos,
1138 produced_pos, nb_packets_per_stream,
1139 stream->max_sb_size);
1140
1141 while (consumed_pos < produced_pos) {
1142 ssize_t read_len;
1143 unsigned long len, padded_len;
1144
1145 health_code_update();
1146
1147 DBG("UST consumer taking snapshot at pos %lu", consumed_pos);
1148
1149 ret = ustctl_get_subbuf(stream->ustream, &consumed_pos);
1150 if (ret < 0) {
1151 if (ret != -EAGAIN) {
1152 PERROR("ustctl_get_subbuf snapshot");
1153 goto error_close_stream;
1154 }
1155 DBG("UST consumer get subbuf failed. Skipping it.");
1156 consumed_pos += stream->max_sb_size;
1157 stream->chan->lost_packets++;
1158 continue;
1159 }
1160
1161 ret = ustctl_get_subbuf_size(stream->ustream, &len);
1162 if (ret < 0) {
1163 ERR("Snapshot ustctl_get_subbuf_size");
1164 goto error_put_subbuf;
1165 }
1166
1167 ret = ustctl_get_padded_subbuf_size(stream->ustream, &padded_len);
1168 if (ret < 0) {
1169 ERR("Snapshot ustctl_get_padded_subbuf_size");
1170 goto error_put_subbuf;
1171 }
1172
1173 read_len = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, len,
1174 padded_len - len, NULL);
1175 if (use_relayd) {
1176 if (read_len != len) {
1177 ret = -EPERM;
1178 goto error_put_subbuf;
1179 }
1180 } else {
1181 if (read_len != padded_len) {
1182 ret = -EPERM;
1183 goto error_put_subbuf;
1184 }
1185 }
1186
1187 ret = ustctl_put_subbuf(stream->ustream);
1188 if (ret < 0) {
1189 ERR("Snapshot ustctl_put_subbuf");
1190 goto error_close_stream;
1191 }
1192 consumed_pos += stream->max_sb_size;
1193 }
1194
1195 /* Simply close the stream so we can use it on the next snapshot. */
1196 consumer_stream_close(stream);
1197 pthread_mutex_unlock(&stream->lock);
1198 }
1199
1200 rcu_read_unlock();
1201 return 0;
1202
1203 error_put_subbuf:
1204 if (ustctl_put_subbuf(stream->ustream) < 0) {
1205 ERR("Snapshot ustctl_put_subbuf");
1206 }
1207 error_close_stream:
1208 consumer_stream_close(stream);
1209 error_unlock:
1210 pthread_mutex_unlock(&stream->lock);
1211 error:
1212 rcu_read_unlock();
1213 return ret;
1214 }
1215
1216 /*
1217 * When a channel has finished the rotation of all its streams, inform the
1218 * session daemon.
1219 */
1220 static
1221 int rotate_notify_sessiond(struct lttng_consumer_local_data *ctx,
1222 uint64_t key)
1223 {
1224 int ret;
1225
1226 do {
1227 ret = write(ctx->channel_rotate_pipe, &key, sizeof(key));
1228 } while (ret == -1 && errno == EINTR);
1229 if (ret == -1) {
1230 PERROR("write to the channel rotate pipe");
1231 } else {
1232 DBG("Sent channel rotation notification for channel key %"
1233 PRIu64, key);
1234 }
1235
1236 return ret;
1237 }
1238
1239 /*
1240 * Performs the stream rotation for the rotate session feature if needed.
1241 * It must be called with the stream and channel locks held.
1242 *
1243 * FIXME: find a way to lock the chan without deadlock. same for kernel.
1244 *
1245 * Return 0 on success, a negative number of error.
1246 */
1247 static
1248 int stream_rotation(struct lttng_consumer_local_data *ctx,
1249 struct lttng_consumer_stream *stream)
1250 {
1251 int ret;
1252 unsigned long consumed_pos;
1253
1254 if (!stream->rotate_position && !stream->rotate_ready) {
1255 ret = 0;
1256 goto end;
1257 }
1258
1259 /*
1260 * If we don't have the rotate_ready flag, check the consumed position
1261 * to determine if we need to rotate.
1262 */
1263 if (!stream->rotate_ready) {
1264 ret = lttng_ustconsumer_sample_snapshot_positions(stream);
1265 if (ret < 0) {
1266 ERR("Taking UST snapshot positions");
1267 goto error;
1268 }
1269
1270 ret = lttng_ustconsumer_get_consumed_snapshot(stream, &consumed_pos);
1271 if (ret < 0) {
1272 ERR("Produced UST snapshot position");
1273 goto error;
1274 }
1275
1276 fprintf(stderr, "packet %lu, pos %lu\n", stream->key, consumed_pos);
1277 /* Rotate position not reached yet. */
1278 if (consumed_pos < stream->rotate_position) {
1279 ret = 0;
1280 goto end;
1281 }
1282 fprintf(stderr, "Rotate position %lu (expected %lu) reached for stream %lu\n",
1283 consumed_pos, stream->rotate_position, stream->key);
1284 } else {
1285 fprintf(stderr, "Rotate position reached for stream %lu\n",
1286 stream->key);
1287 }
1288
1289 ret = close(stream->out_fd);
1290 if (ret < 0) {
1291 PERROR("Closing tracefile");
1292 goto error;
1293 }
1294
1295 fprintf(stderr, "Rotating stream %lu to %s/%s\n", stream->key,
1296 stream->chan->pathname, stream->name);
1297 ret = utils_create_stream_file(stream->chan->pathname, stream->name,
1298 stream->chan->tracefile_size, stream->tracefile_count_current,
1299 stream->uid, stream->gid, NULL);
1300 if (ret < 0) {
1301 goto error;
1302 }
1303 stream->out_fd = ret;
1304 stream->tracefile_size_current = 0;
1305
1306 if (!stream->metadata_flag) {
1307 struct lttng_index_file *index_file;
1308
1309 lttng_index_file_put(stream->index_file);
1310
1311 index_file = lttng_index_file_create(stream->chan->pathname,
1312 stream->name, stream->uid, stream->gid,
1313 stream->chan->tracefile_size,
1314 stream->tracefile_count_current,
1315 CTF_INDEX_MAJOR, CTF_INDEX_MINOR);
1316 if (!index_file) {
1317 goto error;
1318 }
1319 stream->index_file = index_file;
1320 stream->out_fd_offset = 0;
1321 } else {
1322 /*
1323 * Reset the position pushed from the metadata cache so it
1324 * will write from the beginning on the next push.
1325 */
1326 stream->ust_metadata_pushed = 0;
1327 /*
1328 * Wakeup the metadata thread so it dumps the metadata cache
1329 * to file again.
1330 */
1331 consumer_metadata_wakeup_pipe(stream->chan);
1332 }
1333
1334 stream->rotate_position = 0;
1335 stream->rotate_ready = 0;
1336
1337 if (--stream->chan->nr_stream_rotate_pending == 0) {
1338 rotate_notify_sessiond(ctx, stream->chan->key);
1339 fprintf(stderr, "SENT %lu\n", stream->chan->key);
1340 }
1341
1342 ret = 0;
1343 goto end;
1344
1345 error:
1346 ret = -1;
1347 end:
1348 return ret;
1349 }
1350
1351 /*
1352 * Sample the rotate position for all the streams of a channel.
1353 *
1354 * Returns 0 on success, < 0 on error
1355 */
1356 static
1357 int lttng_ustconsumer_rotate_channel(uint64_t key, char *path,
1358 uint64_t relayd_id, uint32_t metadata,
1359 struct lttng_consumer_local_data *ctx)
1360 {
1361 int ret;
1362 struct lttng_consumer_channel *channel;
1363 struct lttng_consumer_stream *stream;
1364 struct lttng_ht_iter iter;
1365 struct lttng_ht *ht = consumer_data.stream_per_chan_id_ht;
1366
1367 /* FIXME: metadata param is useless */
1368
1369 DBG("UST consumer sample rotate position for channel %" PRIu64, key);
1370
1371 rcu_read_lock();
1372
1373 channel = consumer_find_channel(key);
1374 if (!channel) {
1375 ERR("No channel found for key %" PRIu64, key);
1376 ret = -1;
1377 goto end;
1378 }
1379 pthread_mutex_lock(&channel->lock);
1380 snprintf(channel->pathname, PATH_MAX, "%s", path);
1381
1382 cds_lfht_for_each_entry_duplicate(ht->ht,
1383 ht->hash_fct(&channel->key, lttng_ht_seed),
1384 ht->match_fct, &channel->key, &iter.iter,
1385 stream, node_channel_id.node) {
1386 uint64_t consumed_pos;
1387 health_code_update();
1388
1389 /*
1390 * Lock stream because we are about to change its state.
1391 */
1392 pthread_mutex_lock(&stream->lock);
1393 ret = lttng_ustconsumer_sample_snapshot_positions(stream);
1394 if (ret < 0) {
1395 ERR("Taking UST snapshot positions");
1396 goto end_unlock;
1397 }
1398
1399 ret = lttng_ustconsumer_get_produced_snapshot(stream,
1400 &stream->rotate_position);
1401 if (ret < 0) {
1402 ERR("Produced UST snapshot position");
1403 goto end_unlock;
1404 }
1405 fprintf(stderr, "Stream %lu should rotate after %lu to %s\n",
1406 stream->key, stream->rotate_position,
1407 channel->pathname);
1408 lttng_ustconsumer_get_consumed_snapshot(stream,
1409 &consumed_pos);
1410 fprintf(stderr, "consumed %lu\n", consumed_pos);
1411 if (consumed_pos == stream->rotate_position) {
1412 stream->rotate_ready = 1;
1413 fprintf(stderr, "Stream %lu ready to rotate to %s\n",
1414 stream->key, channel->pathname);
1415 }
1416 channel->nr_stream_rotate_pending++;
1417
1418 ustctl_flush_buffer(stream->ustream, 1);
1419
1420 pthread_mutex_unlock(&stream->lock);
1421 }
1422
1423 ret = 0;
1424 goto end_unlock_channel;
1425
1426 end_unlock:
1427 pthread_mutex_unlock(&stream->lock);
1428 end_unlock_channel:
1429 pthread_mutex_unlock(&channel->lock);
1430 end:
1431 rcu_read_unlock();
1432 return ret;
1433 }
1434
1435 /*
1436 * Rotate all the ready streams.
1437 *
1438 * This is especially important for low throughput streams that have already
1439 * been consumed, we cannot wait for their next packet to perform the
1440 * rotation.
1441 *
1442 * Returns 0 on success, < 0 on error
1443 */
1444 static
1445 int lttng_ustconsumer_rotate_ready_streams(uint64_t key,
1446 struct lttng_consumer_local_data *ctx)
1447 {
1448 int ret;
1449 struct lttng_consumer_channel *channel;
1450 struct lttng_consumer_stream *stream;
1451 struct lttng_ht_iter iter;
1452 struct lttng_ht *ht = consumer_data.stream_per_chan_id_ht;
1453
1454 rcu_read_lock();
1455
1456 channel = consumer_find_channel(key);
1457 if (!channel) {
1458 ERR("No channel found for key %" PRIu64, key);
1459 ret = -1;
1460 goto end;
1461 }
1462 cds_lfht_for_each_entry_duplicate(ht->ht,
1463 ht->hash_fct(&channel->key, lttng_ht_seed),
1464 ht->match_fct, &channel->key, &iter.iter,
1465 stream, node_channel_id.node) {
1466 health_code_update();
1467
1468 /*
1469 * Lock stream because we are about to change its state.
1470 */
1471 pthread_mutex_lock(&stream->lock);
1472 if (stream->rotate_ready == 0) {
1473 pthread_mutex_unlock(&stream->lock);
1474 continue;
1475 }
1476 ret = stream_rotation(ctx, stream);
1477 if (ret < 0) {
1478 pthread_mutex_unlock(&stream->lock);
1479 ERR("Stream rotation error");
1480 goto end;
1481 }
1482
1483 pthread_mutex_unlock(&stream->lock);
1484 }
1485
1486 ret = 0;
1487
1488 end:
1489 rcu_read_unlock();
1490 return ret;
1491 }
1492
1493 /*
1494 * Receive the metadata updates from the sessiond. Supports receiving
1495 * overlapping metadata, but is needs to always belong to a contiguous
1496 * range starting from 0.
1497 * Be careful about the locks held when calling this function: it needs
1498 * the metadata cache flush to concurrently progress in order to
1499 * complete.
1500 */
1501 int lttng_ustconsumer_recv_metadata(int sock, uint64_t key, uint64_t offset,
1502 uint64_t len, uint64_t version,
1503 struct lttng_consumer_channel *channel, int timer, int wait)
1504 {
1505 int ret, ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1506 char *metadata_str;
1507
1508 DBG("UST consumer push metadata key %" PRIu64 " of len %" PRIu64, key, len);
1509
1510 metadata_str = zmalloc(len * sizeof(char));
1511 if (!metadata_str) {
1512 PERROR("zmalloc metadata string");
1513 ret_code = LTTCOMM_CONSUMERD_ENOMEM;
1514 goto end;
1515 }
1516
1517 health_code_update();
1518
1519 /* Receive metadata string. */
1520 ret = lttcomm_recv_unix_sock(sock, metadata_str, len);
1521 if (ret < 0) {
1522 /* Session daemon is dead so return gracefully. */
1523 ret_code = ret;
1524 goto end_free;
1525 }
1526
1527 health_code_update();
1528
1529 pthread_mutex_lock(&channel->metadata_cache->lock);
1530 ret = consumer_metadata_cache_write(channel, offset, len, version,
1531 metadata_str);
1532 if (ret < 0) {
1533 /* Unable to handle metadata. Notify session daemon. */
1534 ret_code = LTTCOMM_CONSUMERD_ERROR_METADATA;
1535 /*
1536 * Skip metadata flush on write error since the offset and len might
1537 * not have been updated which could create an infinite loop below when
1538 * waiting for the metadata cache to be flushed.
1539 */
1540 pthread_mutex_unlock(&channel->metadata_cache->lock);
1541 goto end_free;
1542 }
1543 pthread_mutex_unlock(&channel->metadata_cache->lock);
1544
1545 if (!wait) {
1546 goto end_free;
1547 }
1548 while (consumer_metadata_cache_flushed(channel, offset + len, timer)) {
1549 DBG("Waiting for metadata to be flushed");
1550
1551 health_code_update();
1552
1553 usleep(DEFAULT_METADATA_AVAILABILITY_WAIT_TIME);
1554 }
1555
1556 end_free:
1557 free(metadata_str);
1558 end:
1559 return ret_code;
1560 }
1561
1562 /*
1563 * Receive command from session daemon and process it.
1564 *
1565 * Return 1 on success else a negative value or 0.
1566 */
1567 int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
1568 int sock, struct pollfd *consumer_sockpoll)
1569 {
1570 ssize_t ret;
1571 enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1572 struct lttcomm_consumer_msg msg;
1573 struct lttng_consumer_channel *channel = NULL;
1574
1575 health_code_update();
1576
1577 ret = lttcomm_recv_unix_sock(sock, &msg, sizeof(msg));
1578 if (ret != sizeof(msg)) {
1579 DBG("Consumer received unexpected message size %zd (expects %zu)",
1580 ret, sizeof(msg));
1581 /*
1582 * The ret value might 0 meaning an orderly shutdown but this is ok
1583 * since the caller handles this.
1584 */
1585 if (ret > 0) {
1586 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD);
1587 ret = -1;
1588 }
1589 return ret;
1590 }
1591
1592 health_code_update();
1593
1594 /* deprecated */
1595 assert(msg.cmd_type != LTTNG_CONSUMER_STOP);
1596
1597 health_code_update();
1598
1599 /* relayd needs RCU read-side lock */
1600 rcu_read_lock();
1601
1602 switch (msg.cmd_type) {
1603 case LTTNG_CONSUMER_ADD_RELAYD_SOCKET:
1604 {
1605 /* Session daemon status message are handled in the following call. */
1606 consumer_add_relayd_socket(msg.u.relayd_sock.net_index,
1607 msg.u.relayd_sock.type, ctx, sock, consumer_sockpoll,
1608 &msg.u.relayd_sock.sock, msg.u.relayd_sock.session_id,
1609 msg.u.relayd_sock.relayd_session_id);
1610 goto end_nosignal;
1611 }
1612 case LTTNG_CONSUMER_DESTROY_RELAYD:
1613 {
1614 uint64_t index = msg.u.destroy_relayd.net_seq_idx;
1615 struct consumer_relayd_sock_pair *relayd;
1616
1617 DBG("UST consumer destroying relayd %" PRIu64, index);
1618
1619 /* Get relayd reference if exists. */
1620 relayd = consumer_find_relayd(index);
1621 if (relayd == NULL) {
1622 DBG("Unable to find relayd %" PRIu64, index);
1623 ret_code = LTTCOMM_CONSUMERD_RELAYD_FAIL;
1624 }
1625
1626 /*
1627 * Each relayd socket pair has a refcount of stream attached to it
1628 * which tells if the relayd is still active or not depending on the
1629 * refcount value.
1630 *
1631 * This will set the destroy flag of the relayd object and destroy it
1632 * if the refcount reaches zero when called.
1633 *
1634 * The destroy can happen either here or when a stream fd hangs up.
1635 */
1636 if (relayd) {
1637 consumer_flag_relayd_for_destroy(relayd);
1638 }
1639
1640 goto end_msg_sessiond;
1641 }
1642 case LTTNG_CONSUMER_UPDATE_STREAM:
1643 {
1644 rcu_read_unlock();
1645 return -ENOSYS;
1646 }
1647 case LTTNG_CONSUMER_DATA_PENDING:
1648 {
1649 int ret, is_data_pending;
1650 uint64_t id = msg.u.data_pending.session_id;
1651
1652 DBG("UST consumer data pending command for id %" PRIu64, id);
1653
1654 is_data_pending = consumer_data_pending(id);
1655
1656 /* Send back returned value to session daemon */
1657 ret = lttcomm_send_unix_sock(sock, &is_data_pending,
1658 sizeof(is_data_pending));
1659 if (ret < 0) {
1660 DBG("Error when sending the data pending ret code: %d", ret);
1661 goto error_fatal;
1662 }
1663
1664 /*
1665 * No need to send back a status message since the data pending
1666 * returned value is the response.
1667 */
1668 break;
1669 }
1670 case LTTNG_CONSUMER_ASK_CHANNEL_CREATION:
1671 {
1672 int ret;
1673 struct ustctl_consumer_channel_attr attr;
1674
1675 /* Create a plain object and reserve a channel key. */
1676 channel = allocate_channel(msg.u.ask_channel.session_id,
1677 msg.u.ask_channel.pathname, msg.u.ask_channel.name,
1678 msg.u.ask_channel.uid, msg.u.ask_channel.gid,
1679 msg.u.ask_channel.relayd_id, msg.u.ask_channel.key,
1680 (enum lttng_event_output) msg.u.ask_channel.output,
1681 msg.u.ask_channel.tracefile_size,
1682 msg.u.ask_channel.tracefile_count,
1683 msg.u.ask_channel.session_id_per_pid,
1684 msg.u.ask_channel.monitor,
1685 msg.u.ask_channel.live_timer_interval,
1686 msg.u.ask_channel.root_shm_path,
1687 msg.u.ask_channel.shm_path);
1688 if (!channel) {
1689 goto end_channel_error;
1690 }
1691
1692 /*
1693 * Assign UST application UID to the channel. This value is ignored for
1694 * per PID buffers. This is specific to UST thus setting this after the
1695 * allocation.
1696 */
1697 channel->ust_app_uid = msg.u.ask_channel.ust_app_uid;
1698
1699 /* Build channel attributes from received message. */
1700 attr.subbuf_size = msg.u.ask_channel.subbuf_size;
1701 attr.num_subbuf = msg.u.ask_channel.num_subbuf;
1702 attr.overwrite = msg.u.ask_channel.overwrite;
1703 attr.switch_timer_interval = msg.u.ask_channel.switch_timer_interval;
1704 attr.read_timer_interval = msg.u.ask_channel.read_timer_interval;
1705 attr.chan_id = msg.u.ask_channel.chan_id;
1706 memcpy(attr.uuid, msg.u.ask_channel.uuid, sizeof(attr.uuid));
1707 attr.blocking_timeout= msg.u.ask_channel.blocking_timeout;
1708
1709 /* Match channel buffer type to the UST abi. */
1710 switch (msg.u.ask_channel.output) {
1711 case LTTNG_EVENT_MMAP:
1712 default:
1713 attr.output = LTTNG_UST_MMAP;
1714 break;
1715 }
1716
1717 /* Translate and save channel type. */
1718 switch (msg.u.ask_channel.type) {
1719 case LTTNG_UST_CHAN_PER_CPU:
1720 channel->type = CONSUMER_CHANNEL_TYPE_DATA;
1721 attr.type = LTTNG_UST_CHAN_PER_CPU;
1722 /*
1723 * Set refcount to 1 for owner. Below, we will
1724 * pass ownership to the
1725 * consumer_thread_channel_poll() thread.
1726 */
1727 channel->refcount = 1;
1728 break;
1729 case LTTNG_UST_CHAN_METADATA:
1730 channel->type = CONSUMER_CHANNEL_TYPE_METADATA;
1731 attr.type = LTTNG_UST_CHAN_METADATA;
1732 break;
1733 default:
1734 assert(0);
1735 goto error_fatal;
1736 };
1737
1738 health_code_update();
1739
1740 ret = ask_channel(ctx, sock, channel, &attr);
1741 if (ret < 0) {
1742 goto end_channel_error;
1743 }
1744
1745 if (msg.u.ask_channel.type == LTTNG_UST_CHAN_METADATA) {
1746 ret = consumer_metadata_cache_allocate(channel);
1747 if (ret < 0) {
1748 ERR("Allocating metadata cache");
1749 goto end_channel_error;
1750 }
1751 consumer_timer_switch_start(channel, attr.switch_timer_interval);
1752 attr.switch_timer_interval = 0;
1753 } else {
1754 int monitor_start_ret;
1755
1756 consumer_timer_live_start(channel,
1757 msg.u.ask_channel.live_timer_interval);
1758 monitor_start_ret = consumer_timer_monitor_start(
1759 channel,
1760 msg.u.ask_channel.monitor_timer_interval);
1761 if (monitor_start_ret < 0) {
1762 ERR("Starting channel monitoring timer failed");
1763 goto end_channel_error;
1764 }
1765 }
1766
1767 health_code_update();
1768
1769 /*
1770 * Add the channel to the internal state AFTER all streams were created
1771 * and successfully sent to session daemon. This way, all streams must
1772 * be ready before this channel is visible to the threads.
1773 * If add_channel succeeds, ownership of the channel is
1774 * passed to consumer_thread_channel_poll().
1775 */
1776 ret = add_channel(channel, ctx);
1777 if (ret < 0) {
1778 if (msg.u.ask_channel.type == LTTNG_UST_CHAN_METADATA) {
1779 if (channel->switch_timer_enabled == 1) {
1780 consumer_timer_switch_stop(channel);
1781 }
1782 consumer_metadata_cache_destroy(channel);
1783 }
1784 if (channel->live_timer_enabled == 1) {
1785 consumer_timer_live_stop(channel);
1786 }
1787 if (channel->monitor_timer_enabled == 1) {
1788 consumer_timer_monitor_stop(channel);
1789 }
1790 goto end_channel_error;
1791 }
1792
1793 health_code_update();
1794
1795 /*
1796 * Channel and streams are now created. Inform the session daemon that
1797 * everything went well and should wait to receive the channel and
1798 * streams with ustctl API.
1799 */
1800 ret = consumer_send_status_channel(sock, channel);
1801 if (ret < 0) {
1802 /*
1803 * There is probably a problem on the socket.
1804 */
1805 goto error_fatal;
1806 }
1807
1808 break;
1809 }
1810 case LTTNG_CONSUMER_GET_CHANNEL:
1811 {
1812 int ret, relayd_err = 0;
1813 uint64_t key = msg.u.get_channel.key;
1814 struct lttng_consumer_channel *channel;
1815
1816 channel = consumer_find_channel(key);
1817 if (!channel) {
1818 ERR("UST consumer get channel key %" PRIu64 " not found", key);
1819 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
1820 goto end_msg_sessiond;
1821 }
1822
1823 health_code_update();
1824
1825 /* Send everything to sessiond. */
1826 ret = send_sessiond_channel(sock, channel, ctx, &relayd_err);
1827 if (ret < 0) {
1828 if (relayd_err) {
1829 /*
1830 * We were unable to send to the relayd the stream so avoid
1831 * sending back a fatal error to the thread since this is OK
1832 * and the consumer can continue its work. The above call
1833 * has sent the error status message to the sessiond.
1834 */
1835 goto end_nosignal;
1836 }
1837 /*
1838 * The communicaton was broken hence there is a bad state between
1839 * the consumer and sessiond so stop everything.
1840 */
1841 goto error_fatal;
1842 }
1843
1844 health_code_update();
1845
1846 /*
1847 * In no monitor mode, the streams ownership is kept inside the channel
1848 * so don't send them to the data thread.
1849 */
1850 if (!channel->monitor) {
1851 goto end_msg_sessiond;
1852 }
1853
1854 ret = send_streams_to_thread(channel, ctx);
1855 if (ret < 0) {
1856 /*
1857 * If we are unable to send the stream to the thread, there is
1858 * a big problem so just stop everything.
1859 */
1860 goto error_fatal;
1861 }
1862 /* List MUST be empty after or else it could be reused. */
1863 assert(cds_list_empty(&channel->streams.head));
1864 goto end_msg_sessiond;
1865 }
1866 case LTTNG_CONSUMER_DESTROY_CHANNEL:
1867 {
1868 uint64_t key = msg.u.destroy_channel.key;
1869
1870 /*
1871 * Only called if streams have not been sent to stream
1872 * manager thread. However, channel has been sent to
1873 * channel manager thread.
1874 */
1875 notify_thread_del_channel(ctx, key);
1876 goto end_msg_sessiond;
1877 }
1878 case LTTNG_CONSUMER_CLOSE_METADATA:
1879 {
1880 int ret;
1881
1882 ret = close_metadata(msg.u.close_metadata.key);
1883 if (ret != 0) {
1884 ret_code = ret;
1885 }
1886
1887 goto end_msg_sessiond;
1888 }
1889 case LTTNG_CONSUMER_FLUSH_CHANNEL:
1890 {
1891 int ret;
1892
1893 ret = flush_channel(msg.u.flush_channel.key);
1894 if (ret != 0) {
1895 ret_code = ret;
1896 }
1897
1898 goto end_msg_sessiond;
1899 }
1900 case LTTNG_CONSUMER_CLEAR_QUIESCENT_CHANNEL:
1901 {
1902 int ret;
1903
1904 ret = clear_quiescent_channel(
1905 msg.u.clear_quiescent_channel.key);
1906 if (ret != 0) {
1907 ret_code = ret;
1908 }
1909
1910 goto end_msg_sessiond;
1911 }
1912 case LTTNG_CONSUMER_PUSH_METADATA:
1913 {
1914 int ret;
1915 uint64_t len = msg.u.push_metadata.len;
1916 uint64_t key = msg.u.push_metadata.key;
1917 uint64_t offset = msg.u.push_metadata.target_offset;
1918 uint64_t version = msg.u.push_metadata.version;
1919 struct lttng_consumer_channel *channel;
1920
1921 DBG("UST consumer push metadata key %" PRIu64 " of len %" PRIu64, key,
1922 len);
1923
1924 channel = consumer_find_channel(key);
1925 if (!channel) {
1926 /*
1927 * This is possible if the metadata creation on the consumer side
1928 * is in flight vis-a-vis a concurrent push metadata from the
1929 * session daemon. Simply return that the channel failed and the
1930 * session daemon will handle that message correctly considering
1931 * that this race is acceptable thus the DBG() statement here.
1932 */
1933 DBG("UST consumer push metadata %" PRIu64 " not found", key);
1934 ret_code = LTTCOMM_CONSUMERD_CHANNEL_FAIL;
1935 goto end_msg_sessiond;
1936 }
1937
1938 health_code_update();
1939
1940 if (!len) {
1941 /*
1942 * There is nothing to receive. We have simply
1943 * checked whether the channel can be found.
1944 */
1945 ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1946 goto end_msg_sessiond;
1947 }
1948
1949 /* Tell session daemon we are ready to receive the metadata. */
1950 ret = consumer_send_status_msg(sock, LTTCOMM_CONSUMERD_SUCCESS);
1951 if (ret < 0) {
1952 /* Somehow, the session daemon is not responding anymore. */
1953 goto error_fatal;
1954 }
1955
1956 health_code_update();
1957
1958 /* Wait for more data. */
1959 health_poll_entry();
1960 ret = lttng_consumer_poll_socket(consumer_sockpoll);
1961 health_poll_exit();
1962 if (ret) {
1963 goto error_fatal;
1964 }
1965
1966 health_code_update();
1967
1968 ret = lttng_ustconsumer_recv_metadata(sock, key, offset,
1969 len, version, channel, 0, 1);
1970 if (ret < 0) {
1971 /* error receiving from sessiond */
1972 goto error_fatal;
1973 } else {
1974 ret_code = ret;
1975 goto end_msg_sessiond;
1976 }
1977 }
1978 case LTTNG_CONSUMER_SETUP_METADATA:
1979 {
1980 int ret;
1981
1982 ret = setup_metadata(ctx, msg.u.setup_metadata.key);
1983 if (ret) {
1984 ret_code = ret;
1985 }
1986 goto end_msg_sessiond;
1987 }
1988 case LTTNG_CONSUMER_SNAPSHOT_CHANNEL:
1989 {
1990 if (msg.u.snapshot_channel.metadata) {
1991 ret = snapshot_metadata(msg.u.snapshot_channel.key,
1992 msg.u.snapshot_channel.pathname,
1993 msg.u.snapshot_channel.relayd_id,
1994 ctx);
1995 if (ret < 0) {
1996 ERR("Snapshot metadata failed");
1997 ret_code = LTTCOMM_CONSUMERD_ERROR_METADATA;
1998 }
1999 } else {
2000 ret = snapshot_channel(msg.u.snapshot_channel.key,
2001 msg.u.snapshot_channel.pathname,
2002 msg.u.snapshot_channel.relayd_id,
2003 msg.u.snapshot_channel.nb_packets_per_stream,
2004 ctx);
2005 if (ret < 0) {
2006 ERR("Snapshot channel failed");
2007 ret_code = LTTCOMM_CONSUMERD_CHANNEL_FAIL;
2008 }
2009 }
2010
2011 health_code_update();
2012 ret = consumer_send_status_msg(sock, ret_code);
2013 if (ret < 0) {
2014 /* Somehow, the session daemon is not responding anymore. */
2015 goto end_nosignal;
2016 }
2017 health_code_update();
2018 break;
2019 }
2020 case LTTNG_CONSUMER_DISCARDED_EVENTS:
2021 {
2022 int ret = 0;
2023 uint64_t discarded_events;
2024 struct lttng_ht_iter iter;
2025 struct lttng_ht *ht;
2026 struct lttng_consumer_stream *stream;
2027 uint64_t id = msg.u.discarded_events.session_id;
2028 uint64_t key = msg.u.discarded_events.channel_key;
2029
2030 DBG("UST consumer discarded events command for session id %"
2031 PRIu64, id);
2032 rcu_read_lock();
2033 pthread_mutex_lock(&consumer_data.lock);
2034
2035 ht = consumer_data.stream_list_ht;
2036
2037 /*
2038 * We only need a reference to the channel, but they are not
2039 * directly indexed, so we just use the first matching stream
2040 * to extract the information we need, we default to 0 if not
2041 * found (no events are dropped if the channel is not yet in
2042 * use).
2043 */
2044 discarded_events = 0;
2045 cds_lfht_for_each_entry_duplicate(ht->ht,
2046 ht->hash_fct(&id, lttng_ht_seed),
2047 ht->match_fct, &id,
2048 &iter.iter, stream, node_session_id.node) {
2049 if (stream->chan->key == key) {
2050 discarded_events = stream->chan->discarded_events;
2051 break;
2052 }
2053 }
2054 pthread_mutex_unlock(&consumer_data.lock);
2055 rcu_read_unlock();
2056
2057 DBG("UST consumer discarded events command for session id %"
2058 PRIu64 ", channel key %" PRIu64, id, key);
2059
2060 health_code_update();
2061
2062 /* Send back returned value to session daemon */
2063 ret = lttcomm_send_unix_sock(sock, &discarded_events, sizeof(discarded_events));
2064 if (ret < 0) {
2065 PERROR("send discarded events");
2066 goto error_fatal;
2067 }
2068
2069 break;
2070 }
2071 case LTTNG_CONSUMER_LOST_PACKETS:
2072 {
2073 int ret;
2074 uint64_t lost_packets;
2075 struct lttng_ht_iter iter;
2076 struct lttng_ht *ht;
2077 struct lttng_consumer_stream *stream;
2078 uint64_t id = msg.u.lost_packets.session_id;
2079 uint64_t key = msg.u.lost_packets.channel_key;
2080
2081 DBG("UST consumer lost packets command for session id %"
2082 PRIu64, id);
2083 rcu_read_lock();
2084 pthread_mutex_lock(&consumer_data.lock);
2085
2086 ht = consumer_data.stream_list_ht;
2087
2088 /*
2089 * We only need a reference to the channel, but they are not
2090 * directly indexed, so we just use the first matching stream
2091 * to extract the information we need, we default to 0 if not
2092 * found (no packets lost if the channel is not yet in use).
2093 */
2094 lost_packets = 0;
2095 cds_lfht_for_each_entry_duplicate(ht->ht,
2096 ht->hash_fct(&id, lttng_ht_seed),
2097 ht->match_fct, &id,
2098 &iter.iter, stream, node_session_id.node) {
2099 if (stream->chan->key == key) {
2100 lost_packets = stream->chan->lost_packets;
2101 break;
2102 }
2103 }
2104 pthread_mutex_unlock(&consumer_data.lock);
2105 rcu_read_unlock();
2106
2107 DBG("UST consumer lost packets command for session id %"
2108 PRIu64 ", channel key %" PRIu64, id, key);
2109
2110 health_code_update();
2111
2112 /* Send back returned value to session daemon */
2113 ret = lttcomm_send_unix_sock(sock, &lost_packets,
2114 sizeof(lost_packets));
2115 if (ret < 0) {
2116 PERROR("send lost packets");
2117 goto error_fatal;
2118 }
2119
2120 break;
2121 }
2122 case LTTNG_CONSUMER_SET_CHANNEL_MONITOR_PIPE:
2123 {
2124 int channel_monitor_pipe;
2125
2126 ret_code = LTTCOMM_CONSUMERD_SUCCESS;
2127 /* Successfully received the command's type. */
2128 ret = consumer_send_status_msg(sock, ret_code);
2129 if (ret < 0) {
2130 goto error_fatal;
2131 }
2132
2133 ret = lttcomm_recv_fds_unix_sock(sock, &channel_monitor_pipe,
2134 1);
2135 if (ret != sizeof(channel_monitor_pipe)) {
2136 ERR("Failed to receive channel monitor pipe");
2137 goto error_fatal;
2138 }
2139
2140 DBG("Received channel monitor pipe (%d)", channel_monitor_pipe);
2141 ret = consumer_timer_thread_set_channel_monitor_pipe(
2142 channel_monitor_pipe);
2143 if (!ret) {
2144 int flags;
2145
2146 ret_code = LTTCOMM_CONSUMERD_SUCCESS;
2147 /* Set the pipe as non-blocking. */
2148 ret = fcntl(channel_monitor_pipe, F_GETFL, 0);
2149 if (ret == -1) {
2150 PERROR("fcntl get flags of the channel monitoring pipe");
2151 goto error_fatal;
2152 }
2153 flags = ret;
2154
2155 ret = fcntl(channel_monitor_pipe, F_SETFL,
2156 flags | O_NONBLOCK);
2157 if (ret == -1) {
2158 PERROR("fcntl set O_NONBLOCK flag of the channel monitoring pipe");
2159 goto error_fatal;
2160 }
2161 DBG("Channel monitor pipe set as non-blocking");
2162 } else {
2163 ret_code = LTTCOMM_CONSUMERD_ALREADY_SET;
2164 }
2165 goto end_msg_sessiond;
2166 }
2167 case LTTNG_CONSUMER_SET_CHANNEL_ROTATE_PIPE:
2168 {
2169 int channel_rotate_pipe;
2170 int flags;
2171
2172 ret_code = LTTCOMM_CONSUMERD_SUCCESS;
2173 /* Successfully received the command's type. */
2174 ret = consumer_send_status_msg(sock, ret_code);
2175 if (ret < 0) {
2176 goto error_fatal;
2177 }
2178
2179 ret = lttcomm_recv_fds_unix_sock(sock, &channel_rotate_pipe,
2180 1);
2181 if (ret != sizeof(channel_rotate_pipe)) {
2182 ERR("Failed to receive channel rotate pipe");
2183 goto error_fatal;
2184 }
2185
2186 DBG("Received channel rotate pipe (%d)", channel_rotate_pipe);
2187 ctx->channel_rotate_pipe = channel_rotate_pipe;
2188 /* Set the pipe as non-blocking. */
2189 ret = fcntl(channel_rotate_pipe, F_GETFL, 0);
2190 if (ret == -1) {
2191 PERROR("fcntl get flags of the channel rotate pipe");
2192 goto error_fatal;
2193 }
2194 flags = ret;
2195
2196 ret = fcntl(channel_rotate_pipe, F_SETFL,
2197 flags | O_NONBLOCK);
2198 if (ret == -1) {
2199 PERROR("fcntl set O_NONBLOCK flag of the channel rotate pipe");
2200 goto error_fatal;
2201 }
2202 DBG("Channel rotate pipe set as non-blocking");
2203 ret_code = LTTCOMM_CONSUMERD_SUCCESS;
2204 ret = consumer_send_status_msg(sock, ret_code);
2205 if (ret < 0) {
2206 goto error_fatal;
2207 }
2208 break;
2209 }
2210 case LTTNG_CONSUMER_ROTATE_CHANNEL:
2211 {
2212 ret = lttng_ustconsumer_rotate_channel(msg.u.rotate_channel.key,
2213 msg.u.rotate_channel.pathname,
2214 msg.u.rotate_channel.relayd_id,
2215 msg.u.rotate_channel.metadata,
2216 ctx);
2217 if (ret < 0) {
2218 ERR("Rotate channel failed");
2219 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
2220 }
2221
2222 health_code_update();
2223
2224 ret = consumer_send_status_msg(sock, ret_code);
2225 if (ret < 0) {
2226 /* Somehow, the session daemon is not responding anymore. */
2227 goto end_nosignal;
2228 }
2229
2230 /*
2231 * Rotate the streams that are ready right now.
2232 * FIXME: this is a second consecutive iteration over the
2233 * streams in a channel, there is probably a better way to
2234 * handle this, but it needs to be after the
2235 * consumer_send_status_msg() call.
2236 */
2237 ret = lttng_ustconsumer_rotate_ready_streams(
2238 msg.u.rotate_channel.key, ctx);
2239 if (ret < 0) {
2240 ERR("Rotate channel failed");
2241 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
2242 }
2243 break;
2244 }
2245 default:
2246 break;
2247 }
2248
2249 end_nosignal:
2250 rcu_read_unlock();
2251
2252 health_code_update();
2253
2254 /*
2255 * Return 1 to indicate success since the 0 value can be a socket
2256 * shutdown during the recv() or send() call.
2257 */
2258 return 1;
2259
2260 end_msg_sessiond:
2261 /*
2262 * The returned value here is not useful since either way we'll return 1 to
2263 * the caller because the session daemon socket management is done
2264 * elsewhere. Returning a negative code or 0 will shutdown the consumer.
2265 */
2266 ret = consumer_send_status_msg(sock, ret_code);
2267 if (ret < 0) {
2268 goto error_fatal;
2269 }
2270 rcu_read_unlock();
2271
2272 health_code_update();
2273
2274 return 1;
2275 end_channel_error:
2276 if (channel) {
2277 /*
2278 * Free channel here since no one has a reference to it. We don't
2279 * free after that because a stream can store this pointer.
2280 */
2281 destroy_channel(channel);
2282 }
2283 /* We have to send a status channel message indicating an error. */
2284 ret = consumer_send_status_channel(sock, NULL);
2285 if (ret < 0) {
2286 /* Stop everything if session daemon can not be notified. */
2287 goto error_fatal;
2288 }
2289 rcu_read_unlock();
2290
2291 health_code_update();
2292
2293 return 1;
2294 error_fatal:
2295 rcu_read_unlock();
2296 /* This will issue a consumer stop. */
2297 return -1;
2298 }
2299
2300 /*
2301 * Wrapper over the mmap() read offset from ust-ctl library. Since this can be
2302 * compiled out, we isolate it in this library.
2303 */
2304 int lttng_ustctl_get_mmap_read_offset(struct lttng_consumer_stream *stream,
2305 unsigned long *off)
2306 {
2307 assert(stream);
2308 assert(stream->ustream);
2309
2310 return ustctl_get_mmap_read_offset(stream->ustream, off);
2311 }
2312
2313 /*
2314 * Wrapper over the mmap() read offset from ust-ctl library. Since this can be
2315 * compiled out, we isolate it in this library.
2316 */
2317 void *lttng_ustctl_get_mmap_base(struct lttng_consumer_stream *stream)
2318 {
2319 assert(stream);
2320 assert(stream->ustream);
2321
2322 return ustctl_get_mmap_base(stream->ustream);
2323 }
2324
2325 /*
2326 * Take a snapshot for a specific stream.
2327 *
2328 * Returns 0 on success, < 0 on error
2329 */
2330 int lttng_ustconsumer_take_snapshot(struct lttng_consumer_stream *stream)
2331 {
2332 assert(stream);
2333 assert(stream->ustream);
2334
2335 return ustctl_snapshot(stream->ustream);
2336 }
2337
2338 /*
2339 * Sample consumed and produced positions for a specific stream.
2340 *
2341 * Returns 0 on success, < 0 on error.
2342 */
2343 int lttng_ustconsumer_sample_snapshot_positions(
2344 struct lttng_consumer_stream *stream)
2345 {
2346 assert(stream);
2347 assert(stream->ustream);
2348
2349 return ustctl_snapshot_sample_positions(stream->ustream);
2350 }
2351
2352 /*
2353 * Get the produced position
2354 *
2355 * Returns 0 on success, < 0 on error
2356 */
2357 int lttng_ustconsumer_get_produced_snapshot(
2358 struct lttng_consumer_stream *stream, unsigned long *pos)
2359 {
2360 assert(stream);
2361 assert(stream->ustream);
2362 assert(pos);
2363
2364 return ustctl_snapshot_get_produced(stream->ustream, pos);
2365 }
2366
2367 /*
2368 * Get the consumed position
2369 *
2370 * Returns 0 on success, < 0 on error
2371 */
2372 int lttng_ustconsumer_get_consumed_snapshot(
2373 struct lttng_consumer_stream *stream, unsigned long *pos)
2374 {
2375 assert(stream);
2376 assert(stream->ustream);
2377 assert(pos);
2378
2379 return ustctl_snapshot_get_consumed(stream->ustream, pos);
2380 }
2381
2382 void lttng_ustconsumer_flush_buffer(struct lttng_consumer_stream *stream,
2383 int producer)
2384 {
2385 assert(stream);
2386 assert(stream->ustream);
2387
2388 ustctl_flush_buffer(stream->ustream, producer);
2389 }
2390
2391 int lttng_ustconsumer_get_current_timestamp(
2392 struct lttng_consumer_stream *stream, uint64_t *ts)
2393 {
2394 assert(stream);
2395 assert(stream->ustream);
2396 assert(ts);
2397
2398 return ustctl_get_current_timestamp(stream->ustream, ts);
2399 }
2400
2401 int lttng_ustconsumer_get_sequence_number(
2402 struct lttng_consumer_stream *stream, uint64_t *seq)
2403 {
2404 assert(stream);
2405 assert(stream->ustream);
2406 assert(seq);
2407
2408 return ustctl_get_sequence_number(stream->ustream, seq);
2409 }
2410
2411 /*
2412 * Called when the stream signals the consumer that it has hung up.
2413 */
2414 void lttng_ustconsumer_on_stream_hangup(struct lttng_consumer_stream *stream)
2415 {
2416 assert(stream);
2417 assert(stream->ustream);
2418
2419 pthread_mutex_lock(&stream->lock);
2420 if (!stream->quiescent) {
2421 ustctl_flush_buffer(stream->ustream, 0);
2422 stream->quiescent = true;
2423 }
2424 pthread_mutex_unlock(&stream->lock);
2425 stream->hangup_flush_done = 1;
2426 }
2427
2428 void lttng_ustconsumer_del_channel(struct lttng_consumer_channel *chan)
2429 {
2430 int i;
2431
2432 assert(chan);
2433 assert(chan->uchan);
2434
2435 if (chan->switch_timer_enabled == 1) {
2436 consumer_timer_switch_stop(chan);
2437 }
2438 for (i = 0; i < chan->nr_stream_fds; i++) {
2439 int ret;
2440
2441 ret = close(chan->stream_fds[i]);
2442 if (ret) {
2443 PERROR("close");
2444 }
2445 if (chan->shm_path[0]) {
2446 char shm_path[PATH_MAX];
2447
2448 ret = get_stream_shm_path(shm_path, chan->shm_path, i);
2449 if (ret) {
2450 ERR("Cannot get stream shm path");
2451 }
2452 ret = run_as_unlink(shm_path, chan->uid, chan->gid);
2453 if (ret) {
2454 PERROR("unlink %s", shm_path);
2455 }
2456 }
2457 }
2458 }
2459
2460 void lttng_ustconsumer_free_channel(struct lttng_consumer_channel *chan)
2461 {
2462 assert(chan);
2463 assert(chan->uchan);
2464
2465 consumer_metadata_cache_destroy(chan);
2466 ustctl_destroy_channel(chan->uchan);
2467 /* Try to rmdir all directories under shm_path root. */
2468 if (chan->root_shm_path[0]) {
2469 (void) run_as_rmdir_recursive(chan->root_shm_path,
2470 chan->uid, chan->gid);
2471 }
2472 free(chan->stream_fds);
2473 }
2474
2475 void lttng_ustconsumer_del_stream(struct lttng_consumer_stream *stream)
2476 {
2477 assert(stream);
2478 assert(stream->ustream);
2479
2480 if (stream->chan->switch_timer_enabled == 1) {
2481 consumer_timer_switch_stop(stream->chan);
2482 }
2483 ustctl_destroy_stream(stream->ustream);
2484 }
2485
2486 int lttng_ustconsumer_get_wakeup_fd(struct lttng_consumer_stream *stream)
2487 {
2488 assert(stream);
2489 assert(stream->ustream);
2490
2491 return ustctl_stream_get_wakeup_fd(stream->ustream);
2492 }
2493
2494 int lttng_ustconsumer_close_wakeup_fd(struct lttng_consumer_stream *stream)
2495 {
2496 assert(stream);
2497 assert(stream->ustream);
2498
2499 return ustctl_stream_close_wakeup_fd(stream->ustream);
2500 }
2501
2502 /*
2503 * Populate index values of a UST stream. Values are set in big endian order.
2504 *
2505 * Return 0 on success or else a negative value.
2506 */
2507 static int get_index_values(struct ctf_packet_index *index,
2508 struct ustctl_consumer_stream *ustream)
2509 {
2510 int ret;
2511
2512 ret = ustctl_get_timestamp_begin(ustream, &index->timestamp_begin);
2513 if (ret < 0) {
2514 PERROR("ustctl_get_timestamp_begin");
2515 goto error;
2516 }
2517 index->timestamp_begin = htobe64(index->timestamp_begin);
2518
2519 ret = ustctl_get_timestamp_end(ustream, &index->timestamp_end);
2520 if (ret < 0) {
2521 PERROR("ustctl_get_timestamp_end");
2522 goto error;
2523 }
2524 index->timestamp_end = htobe64(index->timestamp_end);
2525
2526 ret = ustctl_get_events_discarded(ustream, &index->events_discarded);
2527 if (ret < 0) {
2528 PERROR("ustctl_get_events_discarded");
2529 goto error;
2530 }
2531 index->events_discarded = htobe64(index->events_discarded);
2532
2533 ret = ustctl_get_content_size(ustream, &index->content_size);
2534 if (ret < 0) {
2535 PERROR("ustctl_get_content_size");
2536 goto error;
2537 }
2538 index->content_size = htobe64(index->content_size);
2539
2540 ret = ustctl_get_packet_size(ustream, &index->packet_size);
2541 if (ret < 0) {
2542 PERROR("ustctl_get_packet_size");
2543 goto error;
2544 }
2545 index->packet_size = htobe64(index->packet_size);
2546
2547 ret = ustctl_get_stream_id(ustream, &index->stream_id);
2548 if (ret < 0) {
2549 PERROR("ustctl_get_stream_id");
2550 goto error;
2551 }
2552 index->stream_id = htobe64(index->stream_id);
2553
2554 ret = ustctl_get_instance_id(ustream, &index->stream_instance_id);
2555 if (ret < 0) {
2556 PERROR("ustctl_get_instance_id");
2557 goto error;
2558 }
2559 index->stream_instance_id = htobe64(index->stream_instance_id);
2560
2561 ret = ustctl_get_sequence_number(ustream, &index->packet_seq_num);
2562 if (ret < 0) {
2563 PERROR("ustctl_get_sequence_number");
2564 goto error;
2565 }
2566 index->packet_seq_num = htobe64(index->packet_seq_num);
2567
2568 error:
2569 return ret;
2570 }
2571
2572 static
2573 void metadata_stream_reset_cache(struct lttng_consumer_stream *stream,
2574 struct consumer_metadata_cache *cache)
2575 {
2576 DBG("Metadata stream update to version %" PRIu64,
2577 cache->version);
2578 stream->ust_metadata_pushed = 0;
2579 stream->metadata_version = cache->version;
2580 stream->reset_metadata_flag = 1;
2581 }
2582
2583 /*
2584 * Check if the version of the metadata stream and metadata cache match.
2585 * If the cache got updated, reset the metadata stream.
2586 * The stream lock and metadata cache lock MUST be held.
2587 * Return 0 on success, a negative value on error.
2588 */
2589 static
2590 int metadata_stream_check_version(struct lttng_consumer_stream *stream)
2591 {
2592 int ret = 0;
2593 struct consumer_metadata_cache *cache = stream->chan->metadata_cache;
2594
2595 if (cache->version == stream->metadata_version) {
2596 goto end;
2597 }
2598 metadata_stream_reset_cache(stream, cache);
2599
2600 end:
2601 return ret;
2602 }
2603
2604 /*
2605 * Write up to one packet from the metadata cache to the channel.
2606 *
2607 * Returns the number of bytes pushed in the cache, or a negative value
2608 * on error.
2609 */
2610 static
2611 int commit_one_metadata_packet(struct lttng_consumer_stream *stream)
2612 {
2613 ssize_t write_len;
2614 int ret;
2615
2616 pthread_mutex_lock(&stream->chan->metadata_cache->lock);
2617 ret = metadata_stream_check_version(stream);
2618 if (ret < 0) {
2619 goto end;
2620 }
2621 if (stream->chan->metadata_cache->max_offset
2622 == stream->ust_metadata_pushed) {
2623 ret = 0;
2624 goto end;
2625 }
2626
2627 write_len = ustctl_write_one_packet_to_channel(stream->chan->uchan,
2628 &stream->chan->metadata_cache->data[stream->ust_metadata_pushed],
2629 stream->chan->metadata_cache->max_offset
2630 - stream->ust_metadata_pushed);
2631 assert(write_len != 0);
2632 if (write_len < 0) {
2633 ERR("Writing one metadata packet");
2634 ret = -1;
2635 goto end;
2636 }
2637 stream->ust_metadata_pushed += write_len;
2638
2639 assert(stream->chan->metadata_cache->max_offset >=
2640 stream->ust_metadata_pushed);
2641 ret = write_len;
2642
2643 end:
2644 pthread_mutex_unlock(&stream->chan->metadata_cache->lock);
2645 return ret;
2646 }
2647
2648
2649 /*
2650 * Sync metadata meaning request them to the session daemon and snapshot to the
2651 * metadata thread can consumer them.
2652 *
2653 * Metadata stream lock is held here, but we need to release it when
2654 * interacting with sessiond, else we cause a deadlock with live
2655 * awaiting on metadata to be pushed out.
2656 *
2657 * Return 0 if new metadatda is available, EAGAIN if the metadata stream
2658 * is empty or a negative value on error.
2659 */
2660 int lttng_ustconsumer_sync_metadata(struct lttng_consumer_local_data *ctx,
2661 struct lttng_consumer_stream *metadata)
2662 {
2663 int ret;
2664 int retry = 0;
2665
2666 assert(ctx);
2667 assert(metadata);
2668
2669 pthread_mutex_unlock(&metadata->lock);
2670 /*
2671 * Request metadata from the sessiond, but don't wait for the flush
2672 * because we locked the metadata thread.
2673 */
2674 ret = lttng_ustconsumer_request_metadata(ctx, metadata->chan, 0, 0);
2675 pthread_mutex_lock(&metadata->lock);
2676 if (ret < 0) {
2677 goto end;
2678 }
2679
2680 ret = commit_one_metadata_packet(metadata);
2681 if (ret <= 0) {
2682 goto end;
2683 } else if (ret > 0) {
2684 retry = 1;
2685 }
2686
2687 ustctl_flush_buffer(metadata->ustream, 1);
2688 ret = ustctl_snapshot(metadata->ustream);
2689 if (ret < 0) {
2690 if (errno != EAGAIN) {
2691 ERR("Sync metadata, taking UST snapshot");
2692 goto end;
2693 }
2694 DBG("No new metadata when syncing them.");
2695 /* No new metadata, exit. */
2696 ret = ENODATA;
2697 goto end;
2698 }
2699
2700 /*
2701 * After this flush, we still need to extract metadata.
2702 */
2703 if (retry) {
2704 ret = EAGAIN;
2705 }
2706
2707 end:
2708 return ret;
2709 }
2710
2711 /*
2712 * Return 0 on success else a negative value.
2713 */
2714 static int notify_if_more_data(struct lttng_consumer_stream *stream,
2715 struct lttng_consumer_local_data *ctx)
2716 {
2717 int ret;
2718 struct ustctl_consumer_stream *ustream;
2719
2720 assert(stream);
2721 assert(ctx);
2722
2723 ustream = stream->ustream;
2724
2725 /*
2726 * First, we are going to check if there is a new subbuffer available
2727 * before reading the stream wait_fd.
2728 */
2729 /* Get the next subbuffer */
2730 ret = ustctl_get_next_subbuf(ustream);
2731 if (ret) {
2732 /* No more data found, flag the stream. */
2733 stream->has_data = 0;
2734 ret = 0;
2735 goto end;
2736 }
2737
2738 ret = ustctl_put_subbuf(ustream);
2739 assert(!ret);
2740
2741 /* This stream still has data. Flag it and wake up the data thread. */
2742 stream->has_data = 1;
2743
2744 if (stream->monitor && !stream->hangup_flush_done && !ctx->has_wakeup) {
2745 ssize_t writelen;
2746
2747 writelen = lttng_pipe_write(ctx->consumer_wakeup_pipe, "!", 1);
2748 if (writelen < 0 && errno != EAGAIN && errno != EWOULDBLOCK) {
2749 ret = writelen;
2750 goto end;
2751 }
2752
2753 /* The wake up pipe has been notified. */
2754 ctx->has_wakeup = 1;
2755 }
2756 ret = 0;
2757
2758 end:
2759 return ret;
2760 }
2761
2762 static
2763 int update_stream_stats(struct lttng_consumer_stream *stream)
2764 {
2765 int ret;
2766 uint64_t seq, discarded;
2767
2768 ret = ustctl_get_sequence_number(stream->ustream, &seq);
2769 if (ret < 0) {
2770 PERROR("ustctl_get_sequence_number");
2771 goto end;
2772 }
2773 /*
2774 * Start the sequence when we extract the first packet in case we don't
2775 * start at 0 (for example if a consumer is not connected to the
2776 * session immediately after the beginning).
2777 */
2778 if (stream->last_sequence_number == -1ULL) {
2779 stream->last_sequence_number = seq;
2780 } else if (seq > stream->last_sequence_number) {
2781 stream->chan->lost_packets += seq -
2782 stream->last_sequence_number - 1;
2783 } else {
2784 /* seq <= last_sequence_number */
2785 ERR("Sequence number inconsistent : prev = %" PRIu64
2786 ", current = %" PRIu64,
2787 stream->last_sequence_number, seq);
2788 ret = -1;
2789 goto end;
2790 }
2791 stream->last_sequence_number = seq;
2792
2793 ret = ustctl_get_events_discarded(stream->ustream, &discarded);
2794 if (ret < 0) {
2795 PERROR("kernctl_get_events_discarded");
2796 goto end;
2797 }
2798 if (discarded < stream->last_discarded_events) {
2799 /*
2800 * Overflow has occurred. We assume only one wrap-around
2801 * has occurred.
2802 */
2803 stream->chan->discarded_events +=
2804 (1ULL << (CAA_BITS_PER_LONG - 1)) -
2805 stream->last_discarded_events + discarded;
2806 } else {
2807 stream->chan->discarded_events += discarded -
2808 stream->last_discarded_events;
2809 }
2810 stream->last_discarded_events = discarded;
2811 ret = 0;
2812
2813 end:
2814 return ret;
2815 }
2816
2817 /*
2818 * Read subbuffer from the given stream.
2819 *
2820 * Stream lock MUST be acquired.
2821 *
2822 * Return 0 on success else a negative value.
2823 */
2824 int lttng_ustconsumer_read_subbuffer(struct lttng_consumer_stream *stream,
2825 struct lttng_consumer_local_data *ctx)
2826 {
2827 unsigned long len, subbuf_size, padding;
2828 int err, write_index = 1, rotation_ret;
2829 long ret = 0;
2830 struct ustctl_consumer_stream *ustream;
2831 struct ctf_packet_index index;
2832
2833 assert(stream);
2834 assert(stream->ustream);
2835 assert(ctx);
2836
2837 DBG("In UST read_subbuffer (wait_fd: %d, name: %s)", stream->wait_fd,
2838 stream->name);
2839
2840 /* Ease our life for what's next. */
2841 ustream = stream->ustream;
2842
2843 /*
2844 * We can consume the 1 byte written into the wait_fd by UST. Don't trigger
2845 * error if we cannot read this one byte (read returns 0), or if the error
2846 * is EAGAIN or EWOULDBLOCK.
2847 *
2848 * This is only done when the stream is monitored by a thread, before the
2849 * flush is done after a hangup and if the stream is not flagged with data
2850 * since there might be nothing to consume in the wait fd but still have
2851 * data available flagged by the consumer wake up pipe.
2852 */
2853 if (stream->monitor && !stream->hangup_flush_done && !stream->has_data) {
2854 char dummy;
2855 ssize_t readlen;
2856
2857 readlen = lttng_read(stream->wait_fd, &dummy, 1);
2858 if (readlen < 0 && errno != EAGAIN && errno != EWOULDBLOCK) {
2859 ret = readlen;
2860 goto end;
2861 }
2862 }
2863
2864 retry:
2865 /* Get the next subbuffer */
2866 err = ustctl_get_next_subbuf(ustream);
2867 if (err != 0) {
2868 /*
2869 * Populate metadata info if the existing info has
2870 * already been read.
2871 */
2872 if (stream->metadata_flag) {
2873 ret = commit_one_metadata_packet(stream);
2874 if (ret <= 0) {
2875 goto end;
2876 }
2877 ustctl_flush_buffer(stream->ustream, 1);
2878 goto retry;
2879 }
2880
2881 ret = err; /* ustctl_get_next_subbuf returns negative, caller expect positive. */
2882 /*
2883 * This is a debug message even for single-threaded consumer,
2884 * because poll() have more relaxed criterions than get subbuf,
2885 * so get_subbuf may fail for short race windows where poll()
2886 * would issue wakeups.
2887 */
2888 DBG("Reserving sub buffer failed (everything is normal, "
2889 "it is due to concurrency) [ret: %d]", err);
2890 goto end;
2891 }
2892 assert(stream->chan->output == CONSUMER_CHANNEL_MMAP);
2893
2894 if (!stream->metadata_flag) {
2895 index.offset = htobe64(stream->out_fd_offset);
2896 ret = get_index_values(&index, ustream);
2897 if (ret < 0) {
2898 err = ustctl_put_subbuf(ustream);
2899 assert(err == 0);
2900 goto end;
2901 }
2902
2903 /* Update the stream's sequence and discarded events count. */
2904 ret = update_stream_stats(stream);
2905 if (ret < 0) {
2906 PERROR("kernctl_get_events_discarded");
2907 err = ustctl_put_subbuf(ustream);
2908 assert(err == 0);
2909 goto end;
2910 }
2911 } else {
2912 write_index = 0;
2913 }
2914
2915 /* Get the full padded subbuffer size */
2916 err = ustctl_get_padded_subbuf_size(ustream, &len);
2917 assert(err == 0);
2918
2919 /* Get subbuffer data size (without padding) */
2920 err = ustctl_get_subbuf_size(ustream, &subbuf_size);
2921 assert(err == 0);
2922
2923 /* Make sure we don't get a subbuffer size bigger than the padded */
2924 assert(len >= subbuf_size);
2925
2926 padding = len - subbuf_size;
2927 /* write the subbuffer to the tracefile */
2928 ret = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, subbuf_size, padding, &index);
2929 /*
2930 * The mmap operation should write subbuf_size amount of data when network
2931 * streaming or the full padding (len) size when we are _not_ streaming.
2932 */
2933 if ((ret != subbuf_size && stream->net_seq_idx != (uint64_t) -1ULL) ||
2934 (ret != len && stream->net_seq_idx == (uint64_t) -1ULL)) {
2935 /*
2936 * Display the error but continue processing to try to release the
2937 * subbuffer. This is a DBG statement since any unexpected kill or
2938 * signal, the application gets unregistered, relayd gets closed or
2939 * anything that affects the buffer lifetime will trigger this error.
2940 * So, for the sake of the user, don't print this error since it can
2941 * happen and it is OK with the code flow.
2942 */
2943 DBG("Error writing to tracefile "
2944 "(ret: %ld != len: %lu != subbuf_size: %lu)",
2945 ret, len, subbuf_size);
2946 write_index = 0;
2947 }
2948 err = ustctl_put_next_subbuf(ustream);
2949 assert(err == 0);
2950
2951 /*
2952 * This will consumer the byte on the wait_fd if and only if there is not
2953 * next subbuffer to be acquired.
2954 */
2955 if (!stream->metadata_flag) {
2956 ret = notify_if_more_data(stream, ctx);
2957 if (ret < 0) {
2958 goto end;
2959 }
2960 }
2961
2962 /* Write index if needed. */
2963 if (!write_index) {
2964 goto end;
2965 }
2966
2967 if (stream->chan->live_timer_interval && !stream->metadata_flag) {
2968 /*
2969 * In live, block until all the metadata is sent.
2970 */
2971 pthread_mutex_lock(&stream->metadata_timer_lock);
2972 assert(!stream->missed_metadata_flush);
2973 stream->waiting_on_metadata = true;
2974 pthread_mutex_unlock(&stream->metadata_timer_lock);
2975
2976 err = consumer_stream_sync_metadata(ctx, stream->session_id);
2977
2978 pthread_mutex_lock(&stream->metadata_timer_lock);
2979 stream->waiting_on_metadata = false;
2980 if (stream->missed_metadata_flush) {
2981 stream->missed_metadata_flush = false;
2982 pthread_mutex_unlock(&stream->metadata_timer_lock);
2983 (void) consumer_flush_ust_index(stream);
2984 } else {
2985 pthread_mutex_unlock(&stream->metadata_timer_lock);
2986 }
2987
2988 if (err < 0) {
2989 goto end;
2990 }
2991 }
2992
2993 assert(!stream->metadata_flag);
2994 err = consumer_stream_write_index(stream, &index);
2995 if (err < 0) {
2996 goto end;
2997 }
2998
2999 end:
3000 /* FIXME: do we need this lock, it causes deadlocks when called
3001 * at the same time with lttng_ustconsumer_rotate_channel ? */
3002 // pthread_mutex_lock(&stream->chan->lock);
3003 rotation_ret = stream_rotation(ctx, stream);
3004 if (rotation_ret < 0) {
3005 // pthread_mutex_unlock(&stream->chan->lock);
3006 ret = -1;
3007 ERR("Stream rotation error");
3008 goto end;
3009 }
3010 // pthread_mutex_unlock(&stream->chan->lock);
3011 return ret;
3012 }
3013
3014 /*
3015 * Called when a stream is created.
3016 *
3017 * Return 0 on success or else a negative value.
3018 */
3019 int lttng_ustconsumer_on_recv_stream(struct lttng_consumer_stream *stream)
3020 {
3021 int ret;
3022
3023 assert(stream);
3024
3025 /* Don't create anything if this is set for streaming. */
3026 if (stream->net_seq_idx == (uint64_t) -1ULL && stream->chan->monitor) {
3027 ret = utils_create_stream_file(stream->chan->pathname, stream->name,
3028 stream->chan->tracefile_size, stream->tracefile_count_current,
3029 stream->uid, stream->gid, NULL);
3030 if (ret < 0) {
3031 goto error;
3032 }
3033 stream->out_fd = ret;
3034 stream->tracefile_size_current = 0;
3035
3036 if (!stream->metadata_flag) {
3037 struct lttng_index_file *index_file;
3038
3039 index_file = lttng_index_file_create(stream->chan->pathname,
3040 stream->name, stream->uid, stream->gid,
3041 stream->chan->tracefile_size,
3042 stream->tracefile_count_current,
3043 CTF_INDEX_MAJOR, CTF_INDEX_MINOR);
3044 if (!index_file) {
3045 goto error;
3046 }
3047 assert(!stream->index_file);
3048 stream->index_file = index_file;
3049 }
3050 }
3051 ret = 0;
3052
3053 error:
3054 return ret;
3055 }
3056
3057 /*
3058 * Check if data is still being extracted from the buffers for a specific
3059 * stream. Consumer data lock MUST be acquired before calling this function
3060 * and the stream lock.
3061 *
3062 * Return 1 if the traced data are still getting read else 0 meaning that the
3063 * data is available for trace viewer reading.
3064 */
3065 int lttng_ustconsumer_data_pending(struct lttng_consumer_stream *stream)
3066 {
3067 int ret;
3068
3069 assert(stream);
3070 assert(stream->ustream);
3071
3072 DBG("UST consumer checking data pending");
3073
3074 if (stream->endpoint_status != CONSUMER_ENDPOINT_ACTIVE) {
3075 ret = 0;
3076 goto end;
3077 }
3078
3079 if (stream->chan->type == CONSUMER_CHANNEL_TYPE_METADATA) {
3080 uint64_t contiguous, pushed;
3081
3082 /* Ease our life a bit. */
3083 contiguous = stream->chan->metadata_cache->max_offset;
3084 pushed = stream->ust_metadata_pushed;
3085
3086 /*
3087 * We can simply check whether all contiguously available data
3088 * has been pushed to the ring buffer, since the push operation
3089 * is performed within get_next_subbuf(), and because both
3090 * get_next_subbuf() and put_next_subbuf() are issued atomically
3091 * thanks to the stream lock within
3092 * lttng_ustconsumer_read_subbuffer(). This basically means that
3093 * whetnever ust_metadata_pushed is incremented, the associated
3094 * metadata has been consumed from the metadata stream.
3095 */
3096 DBG("UST consumer metadata pending check: contiguous %" PRIu64 " vs pushed %" PRIu64,
3097 contiguous, pushed);
3098 assert(((int64_t) (contiguous - pushed)) >= 0);
3099 if ((contiguous != pushed) ||
3100 (((int64_t) contiguous - pushed) > 0 || contiguous == 0)) {
3101 ret = 1; /* Data is pending */
3102 goto end;
3103 }
3104 } else {
3105 ret = ustctl_get_next_subbuf(stream->ustream);
3106 if (ret == 0) {
3107 /*
3108 * There is still data so let's put back this
3109 * subbuffer.
3110 */
3111 ret = ustctl_put_subbuf(stream->ustream);
3112 assert(ret == 0);
3113 ret = 1; /* Data is pending */
3114 goto end;
3115 }
3116 }
3117
3118 /* Data is NOT pending so ready to be read. */
3119 ret = 0;
3120
3121 end:
3122 return ret;
3123 }
3124
3125 /*
3126 * Stop a given metadata channel timer if enabled and close the wait fd which
3127 * is the poll pipe of the metadata stream.
3128 *
3129 * This MUST be called with the metadata channel acquired.
3130 */
3131 void lttng_ustconsumer_close_metadata(struct lttng_consumer_channel *metadata)
3132 {
3133 int ret;
3134
3135 assert(metadata);
3136 assert(metadata->type == CONSUMER_CHANNEL_TYPE_METADATA);
3137
3138 DBG("Closing metadata channel key %" PRIu64, metadata->key);
3139
3140 if (metadata->switch_timer_enabled == 1) {
3141 consumer_timer_switch_stop(metadata);
3142 }
3143
3144 if (!metadata->metadata_stream) {
3145 goto end;
3146 }
3147
3148 /*
3149 * Closing write side so the thread monitoring the stream wakes up if any
3150 * and clean the metadata stream.
3151 */
3152 if (metadata->metadata_stream->ust_metadata_poll_pipe[1] >= 0) {
3153 ret = close(metadata->metadata_stream->ust_metadata_poll_pipe[1]);
3154 if (ret < 0) {
3155 PERROR("closing metadata pipe write side");
3156 }
3157 metadata->metadata_stream->ust_metadata_poll_pipe[1] = -1;
3158 }
3159
3160 end:
3161 return;
3162 }
3163
3164 /*
3165 * Close every metadata stream wait fd of the metadata hash table. This
3166 * function MUST be used very carefully so not to run into a race between the
3167 * metadata thread handling streams and this function closing their wait fd.
3168 *
3169 * For UST, this is used when the session daemon hangs up. Its the metadata
3170 * producer so calling this is safe because we are assured that no state change
3171 * can occur in the metadata thread for the streams in the hash table.
3172 */
3173 void lttng_ustconsumer_close_all_metadata(struct lttng_ht *metadata_ht)
3174 {
3175 struct lttng_ht_iter iter;
3176 struct lttng_consumer_stream *stream;
3177
3178 assert(metadata_ht);
3179 assert(metadata_ht->ht);
3180
3181 DBG("UST consumer closing all metadata streams");
3182
3183 rcu_read_lock();
3184 cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream,
3185 node.node) {
3186
3187 health_code_update();
3188
3189 pthread_mutex_lock(&stream->chan->lock);
3190 lttng_ustconsumer_close_metadata(stream->chan);
3191 pthread_mutex_unlock(&stream->chan->lock);
3192
3193 }
3194 rcu_read_unlock();
3195 }
3196
3197 void lttng_ustconsumer_close_stream_wakeup(struct lttng_consumer_stream *stream)
3198 {
3199 int ret;
3200
3201 ret = ustctl_stream_close_wakeup_fd(stream->ustream);
3202 if (ret < 0) {
3203 ERR("Unable to close wakeup fd");
3204 }
3205 }
3206
3207 /*
3208 * Please refer to consumer-timer.c before adding any lock within this
3209 * function or any of its callees. Timers have a very strict locking
3210 * semantic with respect to teardown. Failure to respect this semantic
3211 * introduces deadlocks.
3212 *
3213 * DON'T hold the metadata lock when calling this function, else this
3214 * can cause deadlock involving consumer awaiting for metadata to be
3215 * pushed out due to concurrent interaction with the session daemon.
3216 */
3217 int lttng_ustconsumer_request_metadata(struct lttng_consumer_local_data *ctx,
3218 struct lttng_consumer_channel *channel, int timer, int wait)
3219 {
3220 struct lttcomm_metadata_request_msg request;
3221 struct lttcomm_consumer_msg msg;
3222 enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS;
3223 uint64_t len, key, offset, version;
3224 int ret;
3225
3226 assert(channel);
3227 assert(channel->metadata_cache);
3228
3229 memset(&request, 0, sizeof(request));
3230
3231 /* send the metadata request to sessiond */
3232 switch (consumer_data.type) {
3233 case LTTNG_CONSUMER64_UST:
3234 request.bits_per_long = 64;
3235 break;
3236 case LTTNG_CONSUMER32_UST:
3237 request.bits_per_long = 32;
3238 break;
3239 default:
3240 request.bits_per_long = 0;
3241 break;
3242 }
3243
3244 request.session_id = channel->session_id;
3245 request.session_id_per_pid = channel->session_id_per_pid;
3246 /*
3247 * Request the application UID here so the metadata of that application can
3248 * be sent back. The channel UID corresponds to the user UID of the session
3249 * used for the rights on the stream file(s).
3250 */
3251 request.uid = channel->ust_app_uid;
3252 request.key = channel->key;
3253
3254 DBG("Sending metadata request to sessiond, session id %" PRIu64
3255 ", per-pid %" PRIu64 ", app UID %u and channek key %" PRIu64,
3256 request.session_id, request.session_id_per_pid, request.uid,
3257 request.key);
3258
3259 pthread_mutex_lock(&ctx->metadata_socket_lock);
3260
3261 health_code_update();
3262
3263 ret = lttcomm_send_unix_sock(ctx->consumer_metadata_socket, &request,
3264 sizeof(request));
3265 if (ret < 0) {
3266 ERR("Asking metadata to sessiond");
3267 goto end;
3268 }
3269
3270 health_code_update();
3271
3272 /* Receive the metadata from sessiond */
3273 ret = lttcomm_recv_unix_sock(ctx->consumer_metadata_socket, &msg,
3274 sizeof(msg));
3275 if (ret != sizeof(msg)) {
3276 DBG("Consumer received unexpected message size %d (expects %zu)",
3277 ret, sizeof(msg));
3278 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD);
3279 /*
3280 * The ret value might 0 meaning an orderly shutdown but this is ok
3281 * since the caller handles this.
3282 */
3283 goto end;
3284 }
3285
3286 health_code_update();
3287
3288 if (msg.cmd_type == LTTNG_ERR_UND) {
3289 /* No registry found */
3290 (void) consumer_send_status_msg(ctx->consumer_metadata_socket,
3291 ret_code);
3292 ret = 0;
3293 goto end;
3294 } else if (msg.cmd_type != LTTNG_CONSUMER_PUSH_METADATA) {
3295 ERR("Unexpected cmd_type received %d", msg.cmd_type);
3296 ret = -1;
3297 goto end;
3298 }
3299
3300 len = msg.u.push_metadata.len;
3301 key = msg.u.push_metadata.key;
3302 offset = msg.u.push_metadata.target_offset;
3303 version = msg.u.push_metadata.version;
3304
3305 assert(key == channel->key);
3306 if (len == 0) {
3307 DBG("No new metadata to receive for key %" PRIu64, key);
3308 }
3309
3310 health_code_update();
3311
3312 /* Tell session daemon we are ready to receive the metadata. */
3313 ret = consumer_send_status_msg(ctx->consumer_metadata_socket,
3314 LTTCOMM_CONSUMERD_SUCCESS);
3315 if (ret < 0 || len == 0) {
3316 /*
3317 * Somehow, the session daemon is not responding anymore or there is
3318 * nothing to receive.
3319 */
3320 goto end;
3321 }
3322
3323 health_code_update();
3324
3325 ret = lttng_ustconsumer_recv_metadata(ctx->consumer_metadata_socket,
3326 key, offset, len, version, channel, timer, wait);
3327 if (ret >= 0) {
3328 /*
3329 * Only send the status msg if the sessiond is alive meaning a positive
3330 * ret code.
3331 */
3332 (void) consumer_send_status_msg(ctx->consumer_metadata_socket, ret);
3333 }
3334 ret = 0;
3335
3336 end:
3337 health_code_update();
3338
3339 pthread_mutex_unlock(&ctx->metadata_socket_lock);
3340 return ret;
3341 }
3342
3343 /*
3344 * Return the ustctl call for the get stream id.
3345 */
3346 int lttng_ustconsumer_get_stream_id(struct lttng_consumer_stream *stream,
3347 uint64_t *stream_id)
3348 {
3349 assert(stream);
3350 assert(stream_id);
3351
3352 return ustctl_get_stream_id(stream->ustream, stream_id);
3353 }
This page took 0.139631 seconds and 6 git commands to generate.