Fix: WARN() should print as WARN level, not ERR
[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 *
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License, version 2 only,
7 * as published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License along
15 * with this program; if not, write to the Free Software Foundation, Inc.,
16 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
17 */
18
19 #define _LGPL_SOURCE
20 #include <assert.h>
21 #include <lttng/ust-ctl.h>
22 #include <poll.h>
23 #include <pthread.h>
24 #include <stdlib.h>
25 #include <string.h>
26 #include <sys/mman.h>
27 #include <sys/socket.h>
28 #include <sys/stat.h>
29 #include <sys/types.h>
30 #include <inttypes.h>
31 #include <unistd.h>
32 #include <urcu/list.h>
33 #include <signal.h>
34
35 #include <bin/lttng-consumerd/health-consumerd.h>
36 #include <common/common.h>
37 #include <common/sessiond-comm/sessiond-comm.h>
38 #include <common/relayd/relayd.h>
39 #include <common/compat/fcntl.h>
40 #include <common/compat/endian.h>
41 #include <common/consumer/consumer-metadata-cache.h>
42 #include <common/consumer/consumer-stream.h>
43 #include <common/consumer/consumer-timer.h>
44 #include <common/utils.h>
45 #include <common/index/index.h>
46
47 #include "ust-consumer.h"
48
49 #define INT_MAX_STR_LEN 12 /* includes \0 */
50
51 extern struct lttng_consumer_global_data consumer_data;
52 extern int consumer_poll_timeout;
53 extern volatile int consumer_quit;
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_recursive_rmdir(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 /* Are we at a position _before_ the first available packet ? */
1080 bool before_first_packet = true;
1081
1082 health_code_update();
1083
1084 /* Lock stream because we are about to change its state. */
1085 pthread_mutex_lock(&stream->lock);
1086 stream->net_seq_idx = relayd_id;
1087
1088 if (use_relayd) {
1089 ret = consumer_send_relayd_stream(stream, path);
1090 if (ret < 0) {
1091 goto error_unlock;
1092 }
1093 } else {
1094 ret = utils_create_stream_file(path, stream->name,
1095 stream->chan->tracefile_size,
1096 stream->tracefile_count_current,
1097 stream->uid, stream->gid, NULL);
1098 if (ret < 0) {
1099 goto error_unlock;
1100 }
1101 stream->out_fd = ret;
1102 stream->tracefile_size_current = 0;
1103
1104 DBG("UST consumer snapshot stream %s/%s (%" PRIu64 ")", path,
1105 stream->name, stream->key);
1106 }
1107 if (relayd_id != -1ULL) {
1108 ret = consumer_send_relayd_streams_sent(relayd_id);
1109 if (ret < 0) {
1110 goto error_unlock;
1111 }
1112 }
1113
1114 ustctl_flush_buffer(stream->ustream, 1);
1115
1116 ret = lttng_ustconsumer_take_snapshot(stream);
1117 if (ret < 0) {
1118 ERR("Taking UST snapshot");
1119 goto error_unlock;
1120 }
1121
1122 ret = lttng_ustconsumer_get_produced_snapshot(stream, &produced_pos);
1123 if (ret < 0) {
1124 ERR("Produced UST snapshot position");
1125 goto error_unlock;
1126 }
1127
1128 ret = lttng_ustconsumer_get_consumed_snapshot(stream, &consumed_pos);
1129 if (ret < 0) {
1130 ERR("Consumerd UST snapshot position");
1131 goto error_unlock;
1132 }
1133
1134 /*
1135 * The original value is sent back if max stream size is larger than
1136 * the possible size of the snapshot. Also, we assume that the session
1137 * daemon should never send a maximum stream size that is lower than
1138 * subbuffer size.
1139 */
1140 consumed_pos = consumer_get_consume_start_pos(consumed_pos,
1141 produced_pos, nb_packets_per_stream,
1142 stream->max_sb_size);
1143
1144 while (consumed_pos < produced_pos) {
1145 ssize_t read_len;
1146 unsigned long len, padded_len;
1147 int lost_packet = 0;
1148
1149 health_code_update();
1150
1151 DBG("UST consumer taking snapshot at pos %lu", consumed_pos);
1152
1153 ret = ustctl_get_subbuf(stream->ustream, &consumed_pos);
1154 if (ret < 0) {
1155 if (ret != -EAGAIN) {
1156 PERROR("ustctl_get_subbuf snapshot");
1157 goto error_close_stream;
1158 }
1159 DBG("UST consumer get subbuf failed. Skipping it.");
1160 consumed_pos += stream->max_sb_size;
1161
1162 /*
1163 * Start accounting lost packets only when we
1164 * already have extracted packets (to match the
1165 * content of the final snapshot).
1166 */
1167 if (!before_first_packet) {
1168 lost_packet = 1;
1169 }
1170 continue;
1171 }
1172
1173 ret = ustctl_get_subbuf_size(stream->ustream, &len);
1174 if (ret < 0) {
1175 ERR("Snapshot ustctl_get_subbuf_size");
1176 goto error_put_subbuf;
1177 }
1178
1179 ret = ustctl_get_padded_subbuf_size(stream->ustream, &padded_len);
1180 if (ret < 0) {
1181 ERR("Snapshot ustctl_get_padded_subbuf_size");
1182 goto error_put_subbuf;
1183 }
1184
1185 read_len = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, len,
1186 padded_len - len, NULL);
1187 if (use_relayd) {
1188 if (read_len != len) {
1189 ret = -EPERM;
1190 goto error_put_subbuf;
1191 }
1192 } else {
1193 if (read_len != padded_len) {
1194 ret = -EPERM;
1195 goto error_put_subbuf;
1196 }
1197 }
1198
1199 ret = ustctl_put_subbuf(stream->ustream);
1200 if (ret < 0) {
1201 ERR("Snapshot ustctl_put_subbuf");
1202 goto error_close_stream;
1203 }
1204 consumed_pos += stream->max_sb_size;
1205
1206 /*
1207 * Only account lost packets located between
1208 * succesfully extracted packets (do not account before
1209 * and after since they are not visible in the
1210 * resulting snapshot).
1211 */
1212 stream->chan->lost_packets += lost_packet;
1213 lost_packet = 0;
1214 before_first_packet = false;
1215 }
1216
1217 /* Simply close the stream so we can use it on the next snapshot. */
1218 consumer_stream_close(stream);
1219 pthread_mutex_unlock(&stream->lock);
1220 }
1221
1222 rcu_read_unlock();
1223 return 0;
1224
1225 error_put_subbuf:
1226 if (ustctl_put_subbuf(stream->ustream) < 0) {
1227 ERR("Snapshot ustctl_put_subbuf");
1228 }
1229 error_close_stream:
1230 consumer_stream_close(stream);
1231 error_unlock:
1232 pthread_mutex_unlock(&stream->lock);
1233 error:
1234 rcu_read_unlock();
1235 return ret;
1236 }
1237
1238 /*
1239 * Receive the metadata updates from the sessiond. Supports receiving
1240 * overlapping metadata, but is needs to always belong to a contiguous
1241 * range starting from 0.
1242 * Be careful about the locks held when calling this function: it needs
1243 * the metadata cache flush to concurrently progress in order to
1244 * complete.
1245 */
1246 int lttng_ustconsumer_recv_metadata(int sock, uint64_t key, uint64_t offset,
1247 uint64_t len, uint64_t version,
1248 struct lttng_consumer_channel *channel, int timer, int wait)
1249 {
1250 int ret, ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1251 char *metadata_str;
1252
1253 DBG("UST consumer push metadata key %" PRIu64 " of len %" PRIu64, key, len);
1254
1255 metadata_str = zmalloc(len * sizeof(char));
1256 if (!metadata_str) {
1257 PERROR("zmalloc metadata string");
1258 ret_code = LTTCOMM_CONSUMERD_ENOMEM;
1259 goto end;
1260 }
1261
1262 health_code_update();
1263
1264 /* Receive metadata string. */
1265 ret = lttcomm_recv_unix_sock(sock, metadata_str, len);
1266 if (ret < 0) {
1267 /* Session daemon is dead so return gracefully. */
1268 ret_code = ret;
1269 goto end_free;
1270 }
1271
1272 health_code_update();
1273
1274 pthread_mutex_lock(&channel->metadata_cache->lock);
1275 ret = consumer_metadata_cache_write(channel, offset, len, version,
1276 metadata_str);
1277 if (ret < 0) {
1278 /* Unable to handle metadata. Notify session daemon. */
1279 ret_code = LTTCOMM_CONSUMERD_ERROR_METADATA;
1280 /*
1281 * Skip metadata flush on write error since the offset and len might
1282 * not have been updated which could create an infinite loop below when
1283 * waiting for the metadata cache to be flushed.
1284 */
1285 pthread_mutex_unlock(&channel->metadata_cache->lock);
1286 goto end_free;
1287 }
1288 pthread_mutex_unlock(&channel->metadata_cache->lock);
1289
1290 if (!wait) {
1291 goto end_free;
1292 }
1293 while (consumer_metadata_cache_flushed(channel, offset + len, timer)) {
1294 DBG("Waiting for metadata to be flushed");
1295
1296 health_code_update();
1297
1298 usleep(DEFAULT_METADATA_AVAILABILITY_WAIT_TIME);
1299 }
1300
1301 end_free:
1302 free(metadata_str);
1303 end:
1304 return ret_code;
1305 }
1306
1307 /*
1308 * Receive command from session daemon and process it.
1309 *
1310 * Return 1 on success else a negative value or 0.
1311 */
1312 int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
1313 int sock, struct pollfd *consumer_sockpoll)
1314 {
1315 ssize_t ret;
1316 enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1317 struct lttcomm_consumer_msg msg;
1318 struct lttng_consumer_channel *channel = NULL;
1319
1320 health_code_update();
1321
1322 ret = lttcomm_recv_unix_sock(sock, &msg, sizeof(msg));
1323 if (ret != sizeof(msg)) {
1324 DBG("Consumer received unexpected message size %zd (expects %zu)",
1325 ret, sizeof(msg));
1326 /*
1327 * The ret value might 0 meaning an orderly shutdown but this is ok
1328 * since the caller handles this.
1329 */
1330 if (ret > 0) {
1331 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD);
1332 ret = -1;
1333 }
1334 return ret;
1335 }
1336
1337 health_code_update();
1338
1339 /* deprecated */
1340 assert(msg.cmd_type != LTTNG_CONSUMER_STOP);
1341
1342 health_code_update();
1343
1344 /* relayd needs RCU read-side lock */
1345 rcu_read_lock();
1346
1347 switch (msg.cmd_type) {
1348 case LTTNG_CONSUMER_ADD_RELAYD_SOCKET:
1349 {
1350 /* Session daemon status message are handled in the following call. */
1351 ret = consumer_add_relayd_socket(msg.u.relayd_sock.net_index,
1352 msg.u.relayd_sock.type, ctx, sock, consumer_sockpoll,
1353 &msg.u.relayd_sock.sock, msg.u.relayd_sock.session_id,
1354 msg.u.relayd_sock.relayd_session_id);
1355 goto end_nosignal;
1356 }
1357 case LTTNG_CONSUMER_DESTROY_RELAYD:
1358 {
1359 uint64_t index = msg.u.destroy_relayd.net_seq_idx;
1360 struct consumer_relayd_sock_pair *relayd;
1361
1362 DBG("UST consumer destroying relayd %" PRIu64, index);
1363
1364 /* Get relayd reference if exists. */
1365 relayd = consumer_find_relayd(index);
1366 if (relayd == NULL) {
1367 DBG("Unable to find relayd %" PRIu64, index);
1368 ret_code = LTTCOMM_CONSUMERD_RELAYD_FAIL;
1369 }
1370
1371 /*
1372 * Each relayd socket pair has a refcount of stream attached to it
1373 * which tells if the relayd is still active or not depending on the
1374 * refcount value.
1375 *
1376 * This will set the destroy flag of the relayd object and destroy it
1377 * if the refcount reaches zero when called.
1378 *
1379 * The destroy can happen either here or when a stream fd hangs up.
1380 */
1381 if (relayd) {
1382 consumer_flag_relayd_for_destroy(relayd);
1383 }
1384
1385 goto end_msg_sessiond;
1386 }
1387 case LTTNG_CONSUMER_UPDATE_STREAM:
1388 {
1389 rcu_read_unlock();
1390 return -ENOSYS;
1391 }
1392 case LTTNG_CONSUMER_DATA_PENDING:
1393 {
1394 int ret, is_data_pending;
1395 uint64_t id = msg.u.data_pending.session_id;
1396
1397 DBG("UST consumer data pending command for id %" PRIu64, id);
1398
1399 is_data_pending = consumer_data_pending(id);
1400
1401 /* Send back returned value to session daemon */
1402 ret = lttcomm_send_unix_sock(sock, &is_data_pending,
1403 sizeof(is_data_pending));
1404 if (ret < 0) {
1405 DBG("Error when sending the data pending ret code: %d", ret);
1406 goto error_fatal;
1407 }
1408
1409 /*
1410 * No need to send back a status message since the data pending
1411 * returned value is the response.
1412 */
1413 break;
1414 }
1415 case LTTNG_CONSUMER_ASK_CHANNEL_CREATION:
1416 {
1417 int ret;
1418 struct ustctl_consumer_channel_attr attr;
1419
1420 /* Create a plain object and reserve a channel key. */
1421 channel = allocate_channel(msg.u.ask_channel.session_id,
1422 msg.u.ask_channel.pathname, msg.u.ask_channel.name,
1423 msg.u.ask_channel.uid, msg.u.ask_channel.gid,
1424 msg.u.ask_channel.relayd_id, msg.u.ask_channel.key,
1425 (enum lttng_event_output) msg.u.ask_channel.output,
1426 msg.u.ask_channel.tracefile_size,
1427 msg.u.ask_channel.tracefile_count,
1428 msg.u.ask_channel.session_id_per_pid,
1429 msg.u.ask_channel.monitor,
1430 msg.u.ask_channel.live_timer_interval,
1431 msg.u.ask_channel.root_shm_path,
1432 msg.u.ask_channel.shm_path);
1433 if (!channel) {
1434 goto end_channel_error;
1435 }
1436
1437 /*
1438 * Assign UST application UID to the channel. This value is ignored for
1439 * per PID buffers. This is specific to UST thus setting this after the
1440 * allocation.
1441 */
1442 channel->ust_app_uid = msg.u.ask_channel.ust_app_uid;
1443
1444 /* Build channel attributes from received message. */
1445 attr.subbuf_size = msg.u.ask_channel.subbuf_size;
1446 attr.num_subbuf = msg.u.ask_channel.num_subbuf;
1447 attr.overwrite = msg.u.ask_channel.overwrite;
1448 attr.switch_timer_interval = msg.u.ask_channel.switch_timer_interval;
1449 attr.read_timer_interval = msg.u.ask_channel.read_timer_interval;
1450 attr.chan_id = msg.u.ask_channel.chan_id;
1451 memcpy(attr.uuid, msg.u.ask_channel.uuid, sizeof(attr.uuid));
1452
1453 /* Match channel buffer type to the UST abi. */
1454 switch (msg.u.ask_channel.output) {
1455 case LTTNG_EVENT_MMAP:
1456 default:
1457 attr.output = LTTNG_UST_MMAP;
1458 break;
1459 }
1460
1461 /* Translate and save channel type. */
1462 switch (msg.u.ask_channel.type) {
1463 case LTTNG_UST_CHAN_PER_CPU:
1464 channel->type = CONSUMER_CHANNEL_TYPE_DATA;
1465 attr.type = LTTNG_UST_CHAN_PER_CPU;
1466 /*
1467 * Set refcount to 1 for owner. Below, we will
1468 * pass ownership to the
1469 * consumer_thread_channel_poll() thread.
1470 */
1471 channel->refcount = 1;
1472 break;
1473 case LTTNG_UST_CHAN_METADATA:
1474 channel->type = CONSUMER_CHANNEL_TYPE_METADATA;
1475 attr.type = LTTNG_UST_CHAN_METADATA;
1476 break;
1477 default:
1478 assert(0);
1479 goto error_fatal;
1480 };
1481
1482 health_code_update();
1483
1484 ret = ask_channel(ctx, sock, channel, &attr);
1485 if (ret < 0) {
1486 goto end_channel_error;
1487 }
1488
1489 if (msg.u.ask_channel.type == LTTNG_UST_CHAN_METADATA) {
1490 ret = consumer_metadata_cache_allocate(channel);
1491 if (ret < 0) {
1492 ERR("Allocating metadata cache");
1493 goto end_channel_error;
1494 }
1495 consumer_timer_switch_start(channel, attr.switch_timer_interval);
1496 attr.switch_timer_interval = 0;
1497 } else {
1498 consumer_timer_live_start(channel,
1499 msg.u.ask_channel.live_timer_interval);
1500 }
1501
1502 health_code_update();
1503
1504 /*
1505 * Add the channel to the internal state AFTER all streams were created
1506 * and successfully sent to session daemon. This way, all streams must
1507 * be ready before this channel is visible to the threads.
1508 * If add_channel succeeds, ownership of the channel is
1509 * passed to consumer_thread_channel_poll().
1510 */
1511 ret = add_channel(channel, ctx);
1512 if (ret < 0) {
1513 if (msg.u.ask_channel.type == LTTNG_UST_CHAN_METADATA) {
1514 if (channel->switch_timer_enabled == 1) {
1515 consumer_timer_switch_stop(channel);
1516 }
1517 consumer_metadata_cache_destroy(channel);
1518 }
1519 if (channel->live_timer_enabled == 1) {
1520 consumer_timer_live_stop(channel);
1521 }
1522 goto end_channel_error;
1523 }
1524
1525 health_code_update();
1526
1527 /*
1528 * Channel and streams are now created. Inform the session daemon that
1529 * everything went well and should wait to receive the channel and
1530 * streams with ustctl API.
1531 */
1532 ret = consumer_send_status_channel(sock, channel);
1533 if (ret < 0) {
1534 /*
1535 * There is probably a problem on the socket.
1536 */
1537 goto error_fatal;
1538 }
1539
1540 break;
1541 }
1542 case LTTNG_CONSUMER_GET_CHANNEL:
1543 {
1544 int ret, relayd_err = 0;
1545 uint64_t key = msg.u.get_channel.key;
1546 struct lttng_consumer_channel *channel;
1547
1548 channel = consumer_find_channel(key);
1549 if (!channel) {
1550 ERR("UST consumer get channel key %" PRIu64 " not found", key);
1551 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
1552 goto end_msg_sessiond;
1553 }
1554
1555 health_code_update();
1556
1557 /* Send everything to sessiond. */
1558 ret = send_sessiond_channel(sock, channel, ctx, &relayd_err);
1559 if (ret < 0) {
1560 if (relayd_err) {
1561 /*
1562 * We were unable to send to the relayd the stream so avoid
1563 * sending back a fatal error to the thread since this is OK
1564 * and the consumer can continue its work. The above call
1565 * has sent the error status message to the sessiond.
1566 */
1567 goto end_nosignal;
1568 }
1569 /*
1570 * The communicaton was broken hence there is a bad state between
1571 * the consumer and sessiond so stop everything.
1572 */
1573 goto error_fatal;
1574 }
1575
1576 health_code_update();
1577
1578 /*
1579 * In no monitor mode, the streams ownership is kept inside the channel
1580 * so don't send them to the data thread.
1581 */
1582 if (!channel->monitor) {
1583 goto end_msg_sessiond;
1584 }
1585
1586 ret = send_streams_to_thread(channel, ctx);
1587 if (ret < 0) {
1588 /*
1589 * If we are unable to send the stream to the thread, there is
1590 * a big problem so just stop everything.
1591 */
1592 goto error_fatal;
1593 }
1594 /* List MUST be empty after or else it could be reused. */
1595 assert(cds_list_empty(&channel->streams.head));
1596 goto end_msg_sessiond;
1597 }
1598 case LTTNG_CONSUMER_DESTROY_CHANNEL:
1599 {
1600 uint64_t key = msg.u.destroy_channel.key;
1601
1602 /*
1603 * Only called if streams have not been sent to stream
1604 * manager thread. However, channel has been sent to
1605 * channel manager thread.
1606 */
1607 notify_thread_del_channel(ctx, key);
1608 goto end_msg_sessiond;
1609 }
1610 case LTTNG_CONSUMER_CLOSE_METADATA:
1611 {
1612 int ret;
1613
1614 ret = close_metadata(msg.u.close_metadata.key);
1615 if (ret != 0) {
1616 ret_code = ret;
1617 }
1618
1619 goto end_msg_sessiond;
1620 }
1621 case LTTNG_CONSUMER_FLUSH_CHANNEL:
1622 {
1623 int ret;
1624
1625 ret = flush_channel(msg.u.flush_channel.key);
1626 if (ret != 0) {
1627 ret_code = ret;
1628 }
1629
1630 goto end_msg_sessiond;
1631 }
1632 case LTTNG_CONSUMER_CLEAR_QUIESCENT_CHANNEL:
1633 {
1634 int ret;
1635
1636 ret = clear_quiescent_channel(
1637 msg.u.clear_quiescent_channel.key);
1638 if (ret != 0) {
1639 ret_code = ret;
1640 }
1641
1642 goto end_msg_sessiond;
1643 }
1644 case LTTNG_CONSUMER_PUSH_METADATA:
1645 {
1646 int ret;
1647 uint64_t len = msg.u.push_metadata.len;
1648 uint64_t key = msg.u.push_metadata.key;
1649 uint64_t offset = msg.u.push_metadata.target_offset;
1650 uint64_t version = msg.u.push_metadata.version;
1651 struct lttng_consumer_channel *channel;
1652
1653 DBG("UST consumer push metadata key %" PRIu64 " of len %" PRIu64, key,
1654 len);
1655
1656 channel = consumer_find_channel(key);
1657 if (!channel) {
1658 /*
1659 * This is possible if the metadata creation on the consumer side
1660 * is in flight vis-a-vis a concurrent push metadata from the
1661 * session daemon. Simply return that the channel failed and the
1662 * session daemon will handle that message correctly considering
1663 * that this race is acceptable thus the DBG() statement here.
1664 */
1665 DBG("UST consumer push metadata %" PRIu64 " not found", key);
1666 ret_code = LTTCOMM_CONSUMERD_CHANNEL_FAIL;
1667 goto end_msg_sessiond;
1668 }
1669
1670 health_code_update();
1671
1672 if (!len) {
1673 /*
1674 * There is nothing to receive. We have simply
1675 * checked whether the channel can be found.
1676 */
1677 ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1678 goto end_msg_sessiond;
1679 }
1680
1681 /* Tell session daemon we are ready to receive the metadata. */
1682 ret = consumer_send_status_msg(sock, LTTCOMM_CONSUMERD_SUCCESS);
1683 if (ret < 0) {
1684 /* Somehow, the session daemon is not responding anymore. */
1685 goto error_fatal;
1686 }
1687
1688 health_code_update();
1689
1690 /* Wait for more data. */
1691 health_poll_entry();
1692 ret = lttng_consumer_poll_socket(consumer_sockpoll);
1693 health_poll_exit();
1694 if (ret) {
1695 goto error_fatal;
1696 }
1697
1698 health_code_update();
1699
1700 ret = lttng_ustconsumer_recv_metadata(sock, key, offset,
1701 len, version, channel, 0, 1);
1702 if (ret < 0) {
1703 /* error receiving from sessiond */
1704 goto error_fatal;
1705 } else {
1706 ret_code = ret;
1707 goto end_msg_sessiond;
1708 }
1709 }
1710 case LTTNG_CONSUMER_SETUP_METADATA:
1711 {
1712 int ret;
1713
1714 ret = setup_metadata(ctx, msg.u.setup_metadata.key);
1715 if (ret) {
1716 ret_code = ret;
1717 }
1718 goto end_msg_sessiond;
1719 }
1720 case LTTNG_CONSUMER_SNAPSHOT_CHANNEL:
1721 {
1722 if (msg.u.snapshot_channel.metadata) {
1723 ret = snapshot_metadata(msg.u.snapshot_channel.key,
1724 msg.u.snapshot_channel.pathname,
1725 msg.u.snapshot_channel.relayd_id,
1726 ctx);
1727 if (ret < 0) {
1728 ERR("Snapshot metadata failed");
1729 ret_code = LTTCOMM_CONSUMERD_ERROR_METADATA;
1730 }
1731 } else {
1732 ret = snapshot_channel(msg.u.snapshot_channel.key,
1733 msg.u.snapshot_channel.pathname,
1734 msg.u.snapshot_channel.relayd_id,
1735 msg.u.snapshot_channel.nb_packets_per_stream,
1736 ctx);
1737 if (ret < 0) {
1738 ERR("Snapshot channel failed");
1739 ret_code = LTTCOMM_CONSUMERD_CHANNEL_FAIL;
1740 }
1741 }
1742
1743 health_code_update();
1744 ret = consumer_send_status_msg(sock, ret_code);
1745 if (ret < 0) {
1746 /* Somehow, the session daemon is not responding anymore. */
1747 goto end_nosignal;
1748 }
1749 health_code_update();
1750 break;
1751 }
1752 case LTTNG_CONSUMER_DISCARDED_EVENTS:
1753 {
1754 int ret = 0;
1755 uint64_t discarded_events;
1756 struct lttng_ht_iter iter;
1757 struct lttng_ht *ht;
1758 struct lttng_consumer_stream *stream;
1759 uint64_t id = msg.u.discarded_events.session_id;
1760 uint64_t key = msg.u.discarded_events.channel_key;
1761
1762 DBG("UST consumer discarded events command for session id %"
1763 PRIu64, id);
1764 rcu_read_lock();
1765 pthread_mutex_lock(&consumer_data.lock);
1766
1767 ht = consumer_data.stream_list_ht;
1768
1769 /*
1770 * We only need a reference to the channel, but they are not
1771 * directly indexed, so we just use the first matching stream
1772 * to extract the information we need, we default to 0 if not
1773 * found (no events are dropped if the channel is not yet in
1774 * use).
1775 */
1776 discarded_events = 0;
1777 cds_lfht_for_each_entry_duplicate(ht->ht,
1778 ht->hash_fct(&id, lttng_ht_seed),
1779 ht->match_fct, &id,
1780 &iter.iter, stream, node_session_id.node) {
1781 if (stream->chan->key == key) {
1782 discarded_events = stream->chan->discarded_events;
1783 break;
1784 }
1785 }
1786 pthread_mutex_unlock(&consumer_data.lock);
1787 rcu_read_unlock();
1788
1789 DBG("UST consumer discarded events command for session id %"
1790 PRIu64 ", channel key %" PRIu64, id, key);
1791
1792 health_code_update();
1793
1794 /* Send back returned value to session daemon */
1795 ret = lttcomm_send_unix_sock(sock, &discarded_events, sizeof(discarded_events));
1796 if (ret < 0) {
1797 PERROR("send discarded events");
1798 goto error_fatal;
1799 }
1800
1801 break;
1802 }
1803 case LTTNG_CONSUMER_LOST_PACKETS:
1804 {
1805 int ret;
1806 uint64_t lost_packets;
1807 struct lttng_ht_iter iter;
1808 struct lttng_ht *ht;
1809 struct lttng_consumer_stream *stream;
1810 uint64_t id = msg.u.lost_packets.session_id;
1811 uint64_t key = msg.u.lost_packets.channel_key;
1812
1813 DBG("UST consumer lost packets command for session id %"
1814 PRIu64, id);
1815 rcu_read_lock();
1816 pthread_mutex_lock(&consumer_data.lock);
1817
1818 ht = consumer_data.stream_list_ht;
1819
1820 /*
1821 * We only need a reference to the channel, but they are not
1822 * directly indexed, so we just use the first matching stream
1823 * to extract the information we need, we default to 0 if not
1824 * found (no packets lost if the channel is not yet in use).
1825 */
1826 lost_packets = 0;
1827 cds_lfht_for_each_entry_duplicate(ht->ht,
1828 ht->hash_fct(&id, lttng_ht_seed),
1829 ht->match_fct, &id,
1830 &iter.iter, stream, node_session_id.node) {
1831 if (stream->chan->key == key) {
1832 lost_packets = stream->chan->lost_packets;
1833 break;
1834 }
1835 }
1836 pthread_mutex_unlock(&consumer_data.lock);
1837 rcu_read_unlock();
1838
1839 DBG("UST consumer lost packets command for session id %"
1840 PRIu64 ", channel key %" PRIu64, id, key);
1841
1842 health_code_update();
1843
1844 /* Send back returned value to session daemon */
1845 ret = lttcomm_send_unix_sock(sock, &lost_packets,
1846 sizeof(lost_packets));
1847 if (ret < 0) {
1848 PERROR("send lost packets");
1849 goto error_fatal;
1850 }
1851
1852 break;
1853 }
1854 default:
1855 break;
1856 }
1857
1858 end_nosignal:
1859 rcu_read_unlock();
1860
1861 health_code_update();
1862
1863 /*
1864 * Return 1 to indicate success since the 0 value can be a socket
1865 * shutdown during the recv() or send() call.
1866 */
1867 return 1;
1868
1869 end_msg_sessiond:
1870 /*
1871 * The returned value here is not useful since either way we'll return 1 to
1872 * the caller because the session daemon socket management is done
1873 * elsewhere. Returning a negative code or 0 will shutdown the consumer.
1874 */
1875 ret = consumer_send_status_msg(sock, ret_code);
1876 if (ret < 0) {
1877 goto error_fatal;
1878 }
1879 rcu_read_unlock();
1880
1881 health_code_update();
1882
1883 return 1;
1884 end_channel_error:
1885 if (channel) {
1886 /*
1887 * Free channel here since no one has a reference to it. We don't
1888 * free after that because a stream can store this pointer.
1889 */
1890 destroy_channel(channel);
1891 }
1892 /* We have to send a status channel message indicating an error. */
1893 ret = consumer_send_status_channel(sock, NULL);
1894 if (ret < 0) {
1895 /* Stop everything if session daemon can not be notified. */
1896 goto error_fatal;
1897 }
1898 rcu_read_unlock();
1899
1900 health_code_update();
1901
1902 return 1;
1903 error_fatal:
1904 rcu_read_unlock();
1905 /* This will issue a consumer stop. */
1906 return -1;
1907 }
1908
1909 /*
1910 * Wrapper over the mmap() read offset from ust-ctl library. Since this can be
1911 * compiled out, we isolate it in this library.
1912 */
1913 int lttng_ustctl_get_mmap_read_offset(struct lttng_consumer_stream *stream,
1914 unsigned long *off)
1915 {
1916 assert(stream);
1917 assert(stream->ustream);
1918
1919 return ustctl_get_mmap_read_offset(stream->ustream, off);
1920 }
1921
1922 /*
1923 * Wrapper over the mmap() read offset from ust-ctl library. Since this can be
1924 * compiled out, we isolate it in this library.
1925 */
1926 void *lttng_ustctl_get_mmap_base(struct lttng_consumer_stream *stream)
1927 {
1928 assert(stream);
1929 assert(stream->ustream);
1930
1931 return ustctl_get_mmap_base(stream->ustream);
1932 }
1933
1934 /*
1935 * Take a snapshot for a specific fd
1936 *
1937 * Returns 0 on success, < 0 on error
1938 */
1939 int lttng_ustconsumer_take_snapshot(struct lttng_consumer_stream *stream)
1940 {
1941 assert(stream);
1942 assert(stream->ustream);
1943
1944 return ustctl_snapshot(stream->ustream);
1945 }
1946
1947 /*
1948 * Get the produced position
1949 *
1950 * Returns 0 on success, < 0 on error
1951 */
1952 int lttng_ustconsumer_get_produced_snapshot(
1953 struct lttng_consumer_stream *stream, unsigned long *pos)
1954 {
1955 assert(stream);
1956 assert(stream->ustream);
1957 assert(pos);
1958
1959 return ustctl_snapshot_get_produced(stream->ustream, pos);
1960 }
1961
1962 /*
1963 * Get the consumed position
1964 *
1965 * Returns 0 on success, < 0 on error
1966 */
1967 int lttng_ustconsumer_get_consumed_snapshot(
1968 struct lttng_consumer_stream *stream, unsigned long *pos)
1969 {
1970 assert(stream);
1971 assert(stream->ustream);
1972 assert(pos);
1973
1974 return ustctl_snapshot_get_consumed(stream->ustream, pos);
1975 }
1976
1977 void lttng_ustconsumer_flush_buffer(struct lttng_consumer_stream *stream,
1978 int producer)
1979 {
1980 assert(stream);
1981 assert(stream->ustream);
1982
1983 ustctl_flush_buffer(stream->ustream, producer);
1984 }
1985
1986 int lttng_ustconsumer_get_current_timestamp(
1987 struct lttng_consumer_stream *stream, uint64_t *ts)
1988 {
1989 assert(stream);
1990 assert(stream->ustream);
1991 assert(ts);
1992
1993 return ustctl_get_current_timestamp(stream->ustream, ts);
1994 }
1995
1996 int lttng_ustconsumer_get_sequence_number(
1997 struct lttng_consumer_stream *stream, uint64_t *seq)
1998 {
1999 assert(stream);
2000 assert(stream->ustream);
2001 assert(seq);
2002
2003 return ustctl_get_sequence_number(stream->ustream, seq);
2004 }
2005
2006 /*
2007 * Called when the stream signals the consumer that it has hung up.
2008 */
2009 void lttng_ustconsumer_on_stream_hangup(struct lttng_consumer_stream *stream)
2010 {
2011 assert(stream);
2012 assert(stream->ustream);
2013
2014 pthread_mutex_lock(&stream->lock);
2015 if (!stream->quiescent) {
2016 ustctl_flush_buffer(stream->ustream, 0);
2017 stream->quiescent = true;
2018 }
2019 pthread_mutex_unlock(&stream->lock);
2020 stream->hangup_flush_done = 1;
2021 }
2022
2023 void lttng_ustconsumer_del_channel(struct lttng_consumer_channel *chan)
2024 {
2025 int i;
2026
2027 assert(chan);
2028 assert(chan->uchan);
2029
2030 if (chan->switch_timer_enabled == 1) {
2031 consumer_timer_switch_stop(chan);
2032 }
2033 for (i = 0; i < chan->nr_stream_fds; i++) {
2034 int ret;
2035
2036 ret = close(chan->stream_fds[i]);
2037 if (ret) {
2038 PERROR("close");
2039 }
2040 if (chan->shm_path[0]) {
2041 char shm_path[PATH_MAX];
2042
2043 ret = get_stream_shm_path(shm_path, chan->shm_path, i);
2044 if (ret) {
2045 ERR("Cannot get stream shm path");
2046 }
2047 ret = run_as_unlink(shm_path, chan->uid, chan->gid);
2048 if (ret) {
2049 PERROR("unlink %s", shm_path);
2050 }
2051 }
2052 }
2053 }
2054
2055 void lttng_ustconsumer_free_channel(struct lttng_consumer_channel *chan)
2056 {
2057 assert(chan);
2058 assert(chan->uchan);
2059
2060 consumer_metadata_cache_destroy(chan);
2061 ustctl_destroy_channel(chan->uchan);
2062 /* Try to rmdir all directories under shm_path root. */
2063 if (chan->root_shm_path[0]) {
2064 (void) run_as_recursive_rmdir(chan->root_shm_path,
2065 chan->uid, chan->gid);
2066 }
2067 free(chan->stream_fds);
2068 }
2069
2070 void lttng_ustconsumer_del_stream(struct lttng_consumer_stream *stream)
2071 {
2072 assert(stream);
2073 assert(stream->ustream);
2074
2075 if (stream->chan->switch_timer_enabled == 1) {
2076 consumer_timer_switch_stop(stream->chan);
2077 }
2078 ustctl_destroy_stream(stream->ustream);
2079 }
2080
2081 int lttng_ustconsumer_get_wakeup_fd(struct lttng_consumer_stream *stream)
2082 {
2083 assert(stream);
2084 assert(stream->ustream);
2085
2086 return ustctl_stream_get_wakeup_fd(stream->ustream);
2087 }
2088
2089 int lttng_ustconsumer_close_wakeup_fd(struct lttng_consumer_stream *stream)
2090 {
2091 assert(stream);
2092 assert(stream->ustream);
2093
2094 return ustctl_stream_close_wakeup_fd(stream->ustream);
2095 }
2096
2097 /*
2098 * Populate index values of a UST stream. Values are set in big endian order.
2099 *
2100 * Return 0 on success or else a negative value.
2101 */
2102 static int get_index_values(struct ctf_packet_index *index,
2103 struct ustctl_consumer_stream *ustream)
2104 {
2105 int ret;
2106
2107 ret = ustctl_get_timestamp_begin(ustream, &index->timestamp_begin);
2108 if (ret < 0) {
2109 PERROR("ustctl_get_timestamp_begin");
2110 goto error;
2111 }
2112 index->timestamp_begin = htobe64(index->timestamp_begin);
2113
2114 ret = ustctl_get_timestamp_end(ustream, &index->timestamp_end);
2115 if (ret < 0) {
2116 PERROR("ustctl_get_timestamp_end");
2117 goto error;
2118 }
2119 index->timestamp_end = htobe64(index->timestamp_end);
2120
2121 ret = ustctl_get_events_discarded(ustream, &index->events_discarded);
2122 if (ret < 0) {
2123 PERROR("ustctl_get_events_discarded");
2124 goto error;
2125 }
2126 index->events_discarded = htobe64(index->events_discarded);
2127
2128 ret = ustctl_get_content_size(ustream, &index->content_size);
2129 if (ret < 0) {
2130 PERROR("ustctl_get_content_size");
2131 goto error;
2132 }
2133 index->content_size = htobe64(index->content_size);
2134
2135 ret = ustctl_get_packet_size(ustream, &index->packet_size);
2136 if (ret < 0) {
2137 PERROR("ustctl_get_packet_size");
2138 goto error;
2139 }
2140 index->packet_size = htobe64(index->packet_size);
2141
2142 ret = ustctl_get_stream_id(ustream, &index->stream_id);
2143 if (ret < 0) {
2144 PERROR("ustctl_get_stream_id");
2145 goto error;
2146 }
2147 index->stream_id = htobe64(index->stream_id);
2148
2149 ret = ustctl_get_instance_id(ustream, &index->stream_instance_id);
2150 if (ret < 0) {
2151 PERROR("ustctl_get_instance_id");
2152 goto error;
2153 }
2154 index->stream_instance_id = htobe64(index->stream_instance_id);
2155
2156 ret = ustctl_get_sequence_number(ustream, &index->packet_seq_num);
2157 if (ret < 0) {
2158 PERROR("ustctl_get_sequence_number");
2159 goto error;
2160 }
2161 index->packet_seq_num = htobe64(index->packet_seq_num);
2162
2163 error:
2164 return ret;
2165 }
2166
2167 static
2168 void metadata_stream_reset_cache(struct lttng_consumer_stream *stream,
2169 struct consumer_metadata_cache *cache)
2170 {
2171 DBG("Metadata stream update to version %" PRIu64,
2172 cache->version);
2173 stream->ust_metadata_pushed = 0;
2174 stream->metadata_version = cache->version;
2175 stream->reset_metadata_flag = 1;
2176 }
2177
2178 /*
2179 * Check if the version of the metadata stream and metadata cache match.
2180 * If the cache got updated, reset the metadata stream.
2181 * The stream lock and metadata cache lock MUST be held.
2182 * Return 0 on success, a negative value on error.
2183 */
2184 static
2185 int metadata_stream_check_version(struct lttng_consumer_stream *stream)
2186 {
2187 int ret = 0;
2188 struct consumer_metadata_cache *cache = stream->chan->metadata_cache;
2189
2190 if (cache->version == stream->metadata_version) {
2191 goto end;
2192 }
2193 metadata_stream_reset_cache(stream, cache);
2194
2195 end:
2196 return ret;
2197 }
2198
2199 /*
2200 * Write up to one packet from the metadata cache to the channel.
2201 *
2202 * Returns the number of bytes pushed in the cache, or a negative value
2203 * on error.
2204 */
2205 static
2206 int commit_one_metadata_packet(struct lttng_consumer_stream *stream)
2207 {
2208 ssize_t write_len;
2209 int ret;
2210
2211 pthread_mutex_lock(&stream->chan->metadata_cache->lock);
2212 ret = metadata_stream_check_version(stream);
2213 if (ret < 0) {
2214 goto end;
2215 }
2216 if (stream->chan->metadata_cache->max_offset
2217 == stream->ust_metadata_pushed) {
2218 ret = 0;
2219 goto end;
2220 }
2221
2222 write_len = ustctl_write_one_packet_to_channel(stream->chan->uchan,
2223 &stream->chan->metadata_cache->data[stream->ust_metadata_pushed],
2224 stream->chan->metadata_cache->max_offset
2225 - stream->ust_metadata_pushed);
2226 assert(write_len != 0);
2227 if (write_len < 0) {
2228 ERR("Writing one metadata packet");
2229 ret = -1;
2230 goto end;
2231 }
2232 stream->ust_metadata_pushed += write_len;
2233
2234 assert(stream->chan->metadata_cache->max_offset >=
2235 stream->ust_metadata_pushed);
2236 ret = write_len;
2237
2238 end:
2239 pthread_mutex_unlock(&stream->chan->metadata_cache->lock);
2240 return ret;
2241 }
2242
2243
2244 /*
2245 * Sync metadata meaning request them to the session daemon and snapshot to the
2246 * metadata thread can consumer them.
2247 *
2248 * Metadata stream lock is held here, but we need to release it when
2249 * interacting with sessiond, else we cause a deadlock with live
2250 * awaiting on metadata to be pushed out.
2251 *
2252 * Return 0 if new metadatda is available, EAGAIN if the metadata stream
2253 * is empty or a negative value on error.
2254 */
2255 int lttng_ustconsumer_sync_metadata(struct lttng_consumer_local_data *ctx,
2256 struct lttng_consumer_stream *metadata)
2257 {
2258 int ret;
2259 int retry = 0;
2260
2261 assert(ctx);
2262 assert(metadata);
2263
2264 pthread_mutex_unlock(&metadata->lock);
2265 /*
2266 * Request metadata from the sessiond, but don't wait for the flush
2267 * because we locked the metadata thread.
2268 */
2269 ret = lttng_ustconsumer_request_metadata(ctx, metadata->chan, 0, 0);
2270 if (ret < 0) {
2271 goto end;
2272 }
2273 pthread_mutex_lock(&metadata->lock);
2274
2275 ret = commit_one_metadata_packet(metadata);
2276 if (ret <= 0) {
2277 goto end;
2278 } else if (ret > 0) {
2279 retry = 1;
2280 }
2281
2282 ustctl_flush_buffer(metadata->ustream, 1);
2283 ret = ustctl_snapshot(metadata->ustream);
2284 if (ret < 0) {
2285 if (errno != EAGAIN) {
2286 ERR("Sync metadata, taking UST snapshot");
2287 goto end;
2288 }
2289 DBG("No new metadata when syncing them.");
2290 /* No new metadata, exit. */
2291 ret = ENODATA;
2292 goto end;
2293 }
2294
2295 /*
2296 * After this flush, we still need to extract metadata.
2297 */
2298 if (retry) {
2299 ret = EAGAIN;
2300 }
2301
2302 end:
2303 return ret;
2304 }
2305
2306 /*
2307 * Return 0 on success else a negative value.
2308 */
2309 static int notify_if_more_data(struct lttng_consumer_stream *stream,
2310 struct lttng_consumer_local_data *ctx)
2311 {
2312 int ret;
2313 struct ustctl_consumer_stream *ustream;
2314
2315 assert(stream);
2316 assert(ctx);
2317
2318 ustream = stream->ustream;
2319
2320 /*
2321 * First, we are going to check if there is a new subbuffer available
2322 * before reading the stream wait_fd.
2323 */
2324 /* Get the next subbuffer */
2325 ret = ustctl_get_next_subbuf(ustream);
2326 if (ret) {
2327 /* No more data found, flag the stream. */
2328 stream->has_data = 0;
2329 ret = 0;
2330 goto end;
2331 }
2332
2333 ret = ustctl_put_subbuf(ustream);
2334 assert(!ret);
2335
2336 /* This stream still has data. Flag it and wake up the data thread. */
2337 stream->has_data = 1;
2338
2339 if (stream->monitor && !stream->hangup_flush_done && !ctx->has_wakeup) {
2340 ssize_t writelen;
2341
2342 writelen = lttng_pipe_write(ctx->consumer_wakeup_pipe, "!", 1);
2343 if (writelen < 0 && errno != EAGAIN && errno != EWOULDBLOCK) {
2344 ret = writelen;
2345 goto end;
2346 }
2347
2348 /* The wake up pipe has been notified. */
2349 ctx->has_wakeup = 1;
2350 }
2351 ret = 0;
2352
2353 end:
2354 return ret;
2355 }
2356
2357 static
2358 int update_stream_stats(struct lttng_consumer_stream *stream)
2359 {
2360 int ret;
2361 uint64_t seq, discarded;
2362
2363 ret = ustctl_get_sequence_number(stream->ustream, &seq);
2364 if (ret < 0) {
2365 PERROR("ustctl_get_sequence_number");
2366 goto end;
2367 }
2368 /*
2369 * Start the sequence when we extract the first packet in case we don't
2370 * start at 0 (for example if a consumer is not connected to the
2371 * session immediately after the beginning).
2372 */
2373 if (stream->last_sequence_number == -1ULL) {
2374 stream->last_sequence_number = seq;
2375 } else if (seq > stream->last_sequence_number) {
2376 stream->chan->lost_packets += seq -
2377 stream->last_sequence_number - 1;
2378 } else {
2379 /* seq <= last_sequence_number */
2380 ERR("Sequence number inconsistent : prev = %" PRIu64
2381 ", current = %" PRIu64,
2382 stream->last_sequence_number, seq);
2383 ret = -1;
2384 goto end;
2385 }
2386 stream->last_sequence_number = seq;
2387
2388 ret = ustctl_get_events_discarded(stream->ustream, &discarded);
2389 if (ret < 0) {
2390 PERROR("kernctl_get_events_discarded");
2391 goto end;
2392 }
2393 if (discarded < stream->last_discarded_events) {
2394 /*
2395 * Overflow has occured. We assume only one wrap-around
2396 * has occured.
2397 */
2398 stream->chan->discarded_events +=
2399 (1ULL << (CAA_BITS_PER_LONG - 1)) -
2400 stream->last_discarded_events + discarded;
2401 } else {
2402 stream->chan->discarded_events += discarded -
2403 stream->last_discarded_events;
2404 }
2405 stream->last_discarded_events = discarded;
2406 ret = 0;
2407
2408 end:
2409 return ret;
2410 }
2411
2412 /*
2413 * Read subbuffer from the given stream.
2414 *
2415 * Stream lock MUST be acquired.
2416 *
2417 * Return 0 on success else a negative value.
2418 */
2419 int lttng_ustconsumer_read_subbuffer(struct lttng_consumer_stream *stream,
2420 struct lttng_consumer_local_data *ctx)
2421 {
2422 unsigned long len, subbuf_size, padding;
2423 int err, write_index = 1;
2424 long ret = 0;
2425 struct ustctl_consumer_stream *ustream;
2426 struct ctf_packet_index index;
2427
2428 assert(stream);
2429 assert(stream->ustream);
2430 assert(ctx);
2431
2432 DBG("In UST read_subbuffer (wait_fd: %d, name: %s)", stream->wait_fd,
2433 stream->name);
2434
2435 /* Ease our life for what's next. */
2436 ustream = stream->ustream;
2437
2438 /*
2439 * We can consume the 1 byte written into the wait_fd by UST. Don't trigger
2440 * error if we cannot read this one byte (read returns 0), or if the error
2441 * is EAGAIN or EWOULDBLOCK.
2442 *
2443 * This is only done when the stream is monitored by a thread, before the
2444 * flush is done after a hangup and if the stream is not flagged with data
2445 * since there might be nothing to consume in the wait fd but still have
2446 * data available flagged by the consumer wake up pipe.
2447 */
2448 if (stream->monitor && !stream->hangup_flush_done && !stream->has_data) {
2449 char dummy;
2450 ssize_t readlen;
2451
2452 readlen = lttng_read(stream->wait_fd, &dummy, 1);
2453 if (readlen < 0 && errno != EAGAIN && errno != EWOULDBLOCK) {
2454 ret = readlen;
2455 goto end;
2456 }
2457 }
2458
2459 retry:
2460 /* Get the next subbuffer */
2461 err = ustctl_get_next_subbuf(ustream);
2462 if (err != 0) {
2463 /*
2464 * Populate metadata info if the existing info has
2465 * already been read.
2466 */
2467 if (stream->metadata_flag) {
2468 ret = commit_one_metadata_packet(stream);
2469 if (ret <= 0) {
2470 goto end;
2471 }
2472 ustctl_flush_buffer(stream->ustream, 1);
2473 goto retry;
2474 }
2475
2476 ret = err; /* ustctl_get_next_subbuf returns negative, caller expect positive. */
2477 /*
2478 * This is a debug message even for single-threaded consumer,
2479 * because poll() have more relaxed criterions than get subbuf,
2480 * so get_subbuf may fail for short race windows where poll()
2481 * would issue wakeups.
2482 */
2483 DBG("Reserving sub buffer failed (everything is normal, "
2484 "it is due to concurrency) [ret: %d]", err);
2485 goto end;
2486 }
2487 assert(stream->chan->output == CONSUMER_CHANNEL_MMAP);
2488
2489 if (!stream->metadata_flag) {
2490 index.offset = htobe64(stream->out_fd_offset);
2491 ret = get_index_values(&index, ustream);
2492 if (ret < 0) {
2493 goto end;
2494 }
2495
2496 /* Update the stream's sequence and discarded events count. */
2497 ret = update_stream_stats(stream);
2498 if (ret < 0) {
2499 PERROR("kernctl_get_events_discarded");
2500 goto end;
2501 }
2502 } else {
2503 write_index = 0;
2504 }
2505
2506 /* Get the full padded subbuffer size */
2507 err = ustctl_get_padded_subbuf_size(ustream, &len);
2508 assert(err == 0);
2509
2510 /* Get subbuffer data size (without padding) */
2511 err = ustctl_get_subbuf_size(ustream, &subbuf_size);
2512 assert(err == 0);
2513
2514 /* Make sure we don't get a subbuffer size bigger than the padded */
2515 assert(len >= subbuf_size);
2516
2517 padding = len - subbuf_size;
2518 /* write the subbuffer to the tracefile */
2519 ret = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, subbuf_size, padding, &index);
2520 /*
2521 * The mmap operation should write subbuf_size amount of data when network
2522 * streaming or the full padding (len) size when we are _not_ streaming.
2523 */
2524 if ((ret != subbuf_size && stream->net_seq_idx != (uint64_t) -1ULL) ||
2525 (ret != len && stream->net_seq_idx == (uint64_t) -1ULL)) {
2526 /*
2527 * Display the error but continue processing to try to release the
2528 * subbuffer. This is a DBG statement since any unexpected kill or
2529 * signal, the application gets unregistered, relayd gets closed or
2530 * anything that affects the buffer lifetime will trigger this error.
2531 * So, for the sake of the user, don't print this error since it can
2532 * happen and it is OK with the code flow.
2533 */
2534 DBG("Error writing to tracefile "
2535 "(ret: %ld != len: %lu != subbuf_size: %lu)",
2536 ret, len, subbuf_size);
2537 write_index = 0;
2538 }
2539 err = ustctl_put_next_subbuf(ustream);
2540 assert(err == 0);
2541
2542 /*
2543 * This will consumer the byte on the wait_fd if and only if there is not
2544 * next subbuffer to be acquired.
2545 */
2546 if (!stream->metadata_flag) {
2547 ret = notify_if_more_data(stream, ctx);
2548 if (ret < 0) {
2549 goto end;
2550 }
2551 }
2552
2553 /* Write index if needed. */
2554 if (!write_index) {
2555 goto end;
2556 }
2557
2558 if (stream->chan->live_timer_interval && !stream->metadata_flag) {
2559 /*
2560 * In live, block until all the metadata is sent.
2561 */
2562 pthread_mutex_lock(&stream->metadata_timer_lock);
2563 assert(!stream->missed_metadata_flush);
2564 stream->waiting_on_metadata = true;
2565 pthread_mutex_unlock(&stream->metadata_timer_lock);
2566
2567 err = consumer_stream_sync_metadata(ctx, stream->session_id);
2568
2569 pthread_mutex_lock(&stream->metadata_timer_lock);
2570 stream->waiting_on_metadata = false;
2571 if (stream->missed_metadata_flush) {
2572 stream->missed_metadata_flush = false;
2573 pthread_mutex_unlock(&stream->metadata_timer_lock);
2574 (void) consumer_flush_ust_index(stream);
2575 } else {
2576 pthread_mutex_unlock(&stream->metadata_timer_lock);
2577 }
2578
2579 if (err < 0) {
2580 goto end;
2581 }
2582 }
2583
2584 assert(!stream->metadata_flag);
2585 err = consumer_stream_write_index(stream, &index);
2586 if (err < 0) {
2587 goto end;
2588 }
2589
2590 end:
2591 return ret;
2592 }
2593
2594 /*
2595 * Called when a stream is created.
2596 *
2597 * Return 0 on success or else a negative value.
2598 */
2599 int lttng_ustconsumer_on_recv_stream(struct lttng_consumer_stream *stream)
2600 {
2601 int ret;
2602
2603 assert(stream);
2604
2605 /* Don't create anything if this is set for streaming. */
2606 if (stream->net_seq_idx == (uint64_t) -1ULL && stream->chan->monitor) {
2607 ret = utils_create_stream_file(stream->chan->pathname, stream->name,
2608 stream->chan->tracefile_size, stream->tracefile_count_current,
2609 stream->uid, stream->gid, NULL);
2610 if (ret < 0) {
2611 goto error;
2612 }
2613 stream->out_fd = ret;
2614 stream->tracefile_size_current = 0;
2615
2616 if (!stream->metadata_flag) {
2617 ret = index_create_file(stream->chan->pathname,
2618 stream->name, stream->uid, stream->gid,
2619 stream->chan->tracefile_size,
2620 stream->tracefile_count_current);
2621 if (ret < 0) {
2622 goto error;
2623 }
2624 stream->index_fd = ret;
2625 }
2626 }
2627 ret = 0;
2628
2629 error:
2630 return ret;
2631 }
2632
2633 /*
2634 * Check if data is still being extracted from the buffers for a specific
2635 * stream. Consumer data lock MUST be acquired before calling this function
2636 * and the stream lock.
2637 *
2638 * Return 1 if the traced data are still getting read else 0 meaning that the
2639 * data is available for trace viewer reading.
2640 */
2641 int lttng_ustconsumer_data_pending(struct lttng_consumer_stream *stream)
2642 {
2643 int ret;
2644
2645 assert(stream);
2646 assert(stream->ustream);
2647
2648 DBG("UST consumer checking data pending");
2649
2650 if (stream->endpoint_status != CONSUMER_ENDPOINT_ACTIVE) {
2651 ret = 0;
2652 goto end;
2653 }
2654
2655 if (stream->chan->type == CONSUMER_CHANNEL_TYPE_METADATA) {
2656 uint64_t contiguous, pushed;
2657
2658 /* Ease our life a bit. */
2659 contiguous = stream->chan->metadata_cache->max_offset;
2660 pushed = stream->ust_metadata_pushed;
2661
2662 /*
2663 * We can simply check whether all contiguously available data
2664 * has been pushed to the ring buffer, since the push operation
2665 * is performed within get_next_subbuf(), and because both
2666 * get_next_subbuf() and put_next_subbuf() are issued atomically
2667 * thanks to the stream lock within
2668 * lttng_ustconsumer_read_subbuffer(). This basically means that
2669 * whetnever ust_metadata_pushed is incremented, the associated
2670 * metadata has been consumed from the metadata stream.
2671 */
2672 DBG("UST consumer metadata pending check: contiguous %" PRIu64 " vs pushed %" PRIu64,
2673 contiguous, pushed);
2674 assert(((int64_t) (contiguous - pushed)) >= 0);
2675 if ((contiguous != pushed) ||
2676 (((int64_t) contiguous - pushed) > 0 || contiguous == 0)) {
2677 ret = 1; /* Data is pending */
2678 goto end;
2679 }
2680 } else {
2681 ret = ustctl_get_next_subbuf(stream->ustream);
2682 if (ret == 0) {
2683 /*
2684 * There is still data so let's put back this
2685 * subbuffer.
2686 */
2687 ret = ustctl_put_subbuf(stream->ustream);
2688 assert(ret == 0);
2689 ret = 1; /* Data is pending */
2690 goto end;
2691 }
2692 }
2693
2694 /* Data is NOT pending so ready to be read. */
2695 ret = 0;
2696
2697 end:
2698 return ret;
2699 }
2700
2701 /*
2702 * Stop a given metadata channel timer if enabled and close the wait fd which
2703 * is the poll pipe of the metadata stream.
2704 *
2705 * This MUST be called with the metadata channel acquired.
2706 */
2707 void lttng_ustconsumer_close_metadata(struct lttng_consumer_channel *metadata)
2708 {
2709 int ret;
2710
2711 assert(metadata);
2712 assert(metadata->type == CONSUMER_CHANNEL_TYPE_METADATA);
2713
2714 DBG("Closing metadata channel key %" PRIu64, metadata->key);
2715
2716 if (metadata->switch_timer_enabled == 1) {
2717 consumer_timer_switch_stop(metadata);
2718 }
2719
2720 if (!metadata->metadata_stream) {
2721 goto end;
2722 }
2723
2724 /*
2725 * Closing write side so the thread monitoring the stream wakes up if any
2726 * and clean the metadata stream.
2727 */
2728 if (metadata->metadata_stream->ust_metadata_poll_pipe[1] >= 0) {
2729 ret = close(metadata->metadata_stream->ust_metadata_poll_pipe[1]);
2730 if (ret < 0) {
2731 PERROR("closing metadata pipe write side");
2732 }
2733 metadata->metadata_stream->ust_metadata_poll_pipe[1] = -1;
2734 }
2735
2736 end:
2737 return;
2738 }
2739
2740 /*
2741 * Close every metadata stream wait fd of the metadata hash table. This
2742 * function MUST be used very carefully so not to run into a race between the
2743 * metadata thread handling streams and this function closing their wait fd.
2744 *
2745 * For UST, this is used when the session daemon hangs up. Its the metadata
2746 * producer so calling this is safe because we are assured that no state change
2747 * can occur in the metadata thread for the streams in the hash table.
2748 */
2749 void lttng_ustconsumer_close_all_metadata(struct lttng_ht *metadata_ht)
2750 {
2751 struct lttng_ht_iter iter;
2752 struct lttng_consumer_stream *stream;
2753
2754 assert(metadata_ht);
2755 assert(metadata_ht->ht);
2756
2757 DBG("UST consumer closing all metadata streams");
2758
2759 rcu_read_lock();
2760 cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream,
2761 node.node) {
2762
2763 health_code_update();
2764
2765 pthread_mutex_lock(&stream->chan->lock);
2766 lttng_ustconsumer_close_metadata(stream->chan);
2767 pthread_mutex_unlock(&stream->chan->lock);
2768
2769 }
2770 rcu_read_unlock();
2771 }
2772
2773 void lttng_ustconsumer_close_stream_wakeup(struct lttng_consumer_stream *stream)
2774 {
2775 int ret;
2776
2777 ret = ustctl_stream_close_wakeup_fd(stream->ustream);
2778 if (ret < 0) {
2779 ERR("Unable to close wakeup fd");
2780 }
2781 }
2782
2783 /*
2784 * Please refer to consumer-timer.c before adding any lock within this
2785 * function or any of its callees. Timers have a very strict locking
2786 * semantic with respect to teardown. Failure to respect this semantic
2787 * introduces deadlocks.
2788 *
2789 * DON'T hold the metadata lock when calling this function, else this
2790 * can cause deadlock involving consumer awaiting for metadata to be
2791 * pushed out due to concurrent interaction with the session daemon.
2792 */
2793 int lttng_ustconsumer_request_metadata(struct lttng_consumer_local_data *ctx,
2794 struct lttng_consumer_channel *channel, int timer, int wait)
2795 {
2796 struct lttcomm_metadata_request_msg request;
2797 struct lttcomm_consumer_msg msg;
2798 enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS;
2799 uint64_t len, key, offset, version;
2800 int ret;
2801
2802 assert(channel);
2803 assert(channel->metadata_cache);
2804
2805 memset(&request, 0, sizeof(request));
2806
2807 /* send the metadata request to sessiond */
2808 switch (consumer_data.type) {
2809 case LTTNG_CONSUMER64_UST:
2810 request.bits_per_long = 64;
2811 break;
2812 case LTTNG_CONSUMER32_UST:
2813 request.bits_per_long = 32;
2814 break;
2815 default:
2816 request.bits_per_long = 0;
2817 break;
2818 }
2819
2820 request.session_id = channel->session_id;
2821 request.session_id_per_pid = channel->session_id_per_pid;
2822 /*
2823 * Request the application UID here so the metadata of that application can
2824 * be sent back. The channel UID corresponds to the user UID of the session
2825 * used for the rights on the stream file(s).
2826 */
2827 request.uid = channel->ust_app_uid;
2828 request.key = channel->key;
2829
2830 DBG("Sending metadata request to sessiond, session id %" PRIu64
2831 ", per-pid %" PRIu64 ", app UID %u and channek key %" PRIu64,
2832 request.session_id, request.session_id_per_pid, request.uid,
2833 request.key);
2834
2835 pthread_mutex_lock(&ctx->metadata_socket_lock);
2836
2837 health_code_update();
2838
2839 ret = lttcomm_send_unix_sock(ctx->consumer_metadata_socket, &request,
2840 sizeof(request));
2841 if (ret < 0) {
2842 ERR("Asking metadata to sessiond");
2843 goto end;
2844 }
2845
2846 health_code_update();
2847
2848 /* Receive the metadata from sessiond */
2849 ret = lttcomm_recv_unix_sock(ctx->consumer_metadata_socket, &msg,
2850 sizeof(msg));
2851 if (ret != sizeof(msg)) {
2852 DBG("Consumer received unexpected message size %d (expects %zu)",
2853 ret, sizeof(msg));
2854 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD);
2855 /*
2856 * The ret value might 0 meaning an orderly shutdown but this is ok
2857 * since the caller handles this.
2858 */
2859 goto end;
2860 }
2861
2862 health_code_update();
2863
2864 if (msg.cmd_type == LTTNG_ERR_UND) {
2865 /* No registry found */
2866 (void) consumer_send_status_msg(ctx->consumer_metadata_socket,
2867 ret_code);
2868 ret = 0;
2869 goto end;
2870 } else if (msg.cmd_type != LTTNG_CONSUMER_PUSH_METADATA) {
2871 ERR("Unexpected cmd_type received %d", msg.cmd_type);
2872 ret = -1;
2873 goto end;
2874 }
2875
2876 len = msg.u.push_metadata.len;
2877 key = msg.u.push_metadata.key;
2878 offset = msg.u.push_metadata.target_offset;
2879 version = msg.u.push_metadata.version;
2880
2881 assert(key == channel->key);
2882 if (len == 0) {
2883 DBG("No new metadata to receive for key %" PRIu64, key);
2884 }
2885
2886 health_code_update();
2887
2888 /* Tell session daemon we are ready to receive the metadata. */
2889 ret = consumer_send_status_msg(ctx->consumer_metadata_socket,
2890 LTTCOMM_CONSUMERD_SUCCESS);
2891 if (ret < 0 || len == 0) {
2892 /*
2893 * Somehow, the session daemon is not responding anymore or there is
2894 * nothing to receive.
2895 */
2896 goto end;
2897 }
2898
2899 health_code_update();
2900
2901 ret = lttng_ustconsumer_recv_metadata(ctx->consumer_metadata_socket,
2902 key, offset, len, version, channel, timer, wait);
2903 if (ret >= 0) {
2904 /*
2905 * Only send the status msg if the sessiond is alive meaning a positive
2906 * ret code.
2907 */
2908 (void) consumer_send_status_msg(ctx->consumer_metadata_socket, ret);
2909 }
2910 ret = 0;
2911
2912 end:
2913 health_code_update();
2914
2915 pthread_mutex_unlock(&ctx->metadata_socket_lock);
2916 return ret;
2917 }
2918
2919 /*
2920 * Return the ustctl call for the get stream id.
2921 */
2922 int lttng_ustconsumer_get_stream_id(struct lttng_consumer_stream *stream,
2923 uint64_t *stream_id)
2924 {
2925 assert(stream);
2926 assert(stream_id);
2927
2928 return ustctl_get_stream_id(stream->ustream, stream_id);
2929 }
This page took 0.161818 seconds and 4 git commands to generate.