Fix: Use distinct variables for ret and discarded_events
[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 ustctl_flush_buffer(stream->ustream, 1);
771 }
772 error:
773 rcu_read_unlock();
774 return ret;
775 }
776
777 /*
778 * Close metadata stream wakeup_fd using the given key to retrieve the channel.
779 * RCU read side lock MUST be acquired before calling this function.
780 *
781 * Return 0 on success else an LTTng error code.
782 */
783 static int close_metadata(uint64_t chan_key)
784 {
785 int ret = 0;
786 struct lttng_consumer_channel *channel;
787
788 DBG("UST consumer close metadata key %" PRIu64, chan_key);
789
790 channel = consumer_find_channel(chan_key);
791 if (!channel) {
792 /*
793 * This is possible if the metadata thread has issue a delete because
794 * the endpoint point of the stream hung up. There is no way the
795 * session daemon can know about it thus use a DBG instead of an actual
796 * error.
797 */
798 DBG("UST consumer close metadata %" PRIu64 " not found", chan_key);
799 ret = LTTNG_ERR_UST_CHAN_NOT_FOUND;
800 goto error;
801 }
802
803 pthread_mutex_lock(&consumer_data.lock);
804 pthread_mutex_lock(&channel->lock);
805
806 if (cds_lfht_is_node_deleted(&channel->node.node)) {
807 goto error_unlock;
808 }
809
810 lttng_ustconsumer_close_metadata(channel);
811
812 error_unlock:
813 pthread_mutex_unlock(&channel->lock);
814 pthread_mutex_unlock(&consumer_data.lock);
815 error:
816 return ret;
817 }
818
819 /*
820 * RCU read side lock MUST be acquired before calling this function.
821 *
822 * Return 0 on success else an LTTng error code.
823 */
824 static int setup_metadata(struct lttng_consumer_local_data *ctx, uint64_t key)
825 {
826 int ret;
827 struct lttng_consumer_channel *metadata;
828
829 DBG("UST consumer setup metadata key %" PRIu64, key);
830
831 metadata = consumer_find_channel(key);
832 if (!metadata) {
833 ERR("UST consumer push metadata %" PRIu64 " not found", key);
834 ret = LTTNG_ERR_UST_CHAN_NOT_FOUND;
835 goto end;
836 }
837
838 /*
839 * In no monitor mode, the metadata channel has no stream(s) so skip the
840 * ownership transfer to the metadata thread.
841 */
842 if (!metadata->monitor) {
843 DBG("Metadata channel in no monitor");
844 ret = 0;
845 goto end;
846 }
847
848 /*
849 * Send metadata stream to relayd if one available. Availability is
850 * known if the stream is still in the list of the channel.
851 */
852 if (cds_list_empty(&metadata->streams.head)) {
853 ERR("Metadata channel key %" PRIu64 ", no stream available.", key);
854 ret = LTTCOMM_CONSUMERD_ERROR_METADATA;
855 goto error_no_stream;
856 }
857
858 /* Send metadata stream to relayd if needed. */
859 if (metadata->metadata_stream->net_seq_idx != (uint64_t) -1ULL) {
860 ret = consumer_send_relayd_stream(metadata->metadata_stream,
861 metadata->pathname);
862 if (ret < 0) {
863 ret = LTTCOMM_CONSUMERD_ERROR_METADATA;
864 goto error;
865 }
866 ret = consumer_send_relayd_streams_sent(
867 metadata->metadata_stream->net_seq_idx);
868 if (ret < 0) {
869 ret = LTTCOMM_CONSUMERD_RELAYD_FAIL;
870 goto error;
871 }
872 }
873
874 ret = send_streams_to_thread(metadata, ctx);
875 if (ret < 0) {
876 /*
877 * If we are unable to send the stream to the thread, there is
878 * a big problem so just stop everything.
879 */
880 ret = LTTCOMM_CONSUMERD_FATAL;
881 goto error;
882 }
883 /* List MUST be empty after or else it could be reused. */
884 assert(cds_list_empty(&metadata->streams.head));
885
886 ret = 0;
887 goto end;
888
889 error:
890 /*
891 * Delete metadata channel on error. At this point, the metadata stream can
892 * NOT be monitored by the metadata thread thus having the guarantee that
893 * the stream is still in the local stream list of the channel. This call
894 * will make sure to clean that list.
895 */
896 consumer_stream_destroy(metadata->metadata_stream, NULL);
897 cds_list_del(&metadata->metadata_stream->send_node);
898 metadata->metadata_stream = NULL;
899 error_no_stream:
900 end:
901 return ret;
902 }
903
904 /*
905 * Snapshot the whole metadata.
906 *
907 * Returns 0 on success, < 0 on error
908 */
909 static int snapshot_metadata(uint64_t key, char *path, uint64_t relayd_id,
910 struct lttng_consumer_local_data *ctx)
911 {
912 int ret = 0;
913 struct lttng_consumer_channel *metadata_channel;
914 struct lttng_consumer_stream *metadata_stream;
915
916 assert(path);
917 assert(ctx);
918
919 DBG("UST consumer snapshot metadata with key %" PRIu64 " at path %s",
920 key, path);
921
922 rcu_read_lock();
923
924 metadata_channel = consumer_find_channel(key);
925 if (!metadata_channel) {
926 ERR("UST snapshot metadata channel not found for key %" PRIu64,
927 key);
928 ret = -1;
929 goto error;
930 }
931 assert(!metadata_channel->monitor);
932
933 health_code_update();
934
935 /*
936 * Ask the sessiond if we have new metadata waiting and update the
937 * consumer metadata cache.
938 */
939 ret = lttng_ustconsumer_request_metadata(ctx, metadata_channel, 0, 1);
940 if (ret < 0) {
941 goto error;
942 }
943
944 health_code_update();
945
946 /*
947 * The metadata stream is NOT created in no monitor mode when the channel
948 * is created on a sessiond ask channel command.
949 */
950 ret = create_ust_streams(metadata_channel, ctx);
951 if (ret < 0) {
952 goto error;
953 }
954
955 metadata_stream = metadata_channel->metadata_stream;
956 assert(metadata_stream);
957
958 if (relayd_id != (uint64_t) -1ULL) {
959 metadata_stream->net_seq_idx = relayd_id;
960 ret = consumer_send_relayd_stream(metadata_stream, path);
961 if (ret < 0) {
962 goto error_stream;
963 }
964 } else {
965 ret = utils_create_stream_file(path, metadata_stream->name,
966 metadata_stream->chan->tracefile_size,
967 metadata_stream->tracefile_count_current,
968 metadata_stream->uid, metadata_stream->gid, NULL);
969 if (ret < 0) {
970 goto error_stream;
971 }
972 metadata_stream->out_fd = ret;
973 metadata_stream->tracefile_size_current = 0;
974 }
975
976 do {
977 health_code_update();
978
979 ret = lttng_consumer_read_subbuffer(metadata_stream, ctx);
980 if (ret < 0) {
981 goto error_stream;
982 }
983 } while (ret > 0);
984
985 error_stream:
986 /*
987 * Clean up the stream completly because the next snapshot will use a new
988 * metadata stream.
989 */
990 consumer_stream_destroy(metadata_stream, NULL);
991 cds_list_del(&metadata_stream->send_node);
992 metadata_channel->metadata_stream = NULL;
993
994 error:
995 rcu_read_unlock();
996 return ret;
997 }
998
999 /*
1000 * Take a snapshot of all the stream of a channel.
1001 *
1002 * Returns 0 on success, < 0 on error
1003 */
1004 static int snapshot_channel(uint64_t key, char *path, uint64_t relayd_id,
1005 uint64_t nb_packets_per_stream, struct lttng_consumer_local_data *ctx)
1006 {
1007 int ret;
1008 unsigned use_relayd = 0;
1009 unsigned long consumed_pos, produced_pos;
1010 struct lttng_consumer_channel *channel;
1011 struct lttng_consumer_stream *stream;
1012
1013 assert(path);
1014 assert(ctx);
1015
1016 rcu_read_lock();
1017
1018 if (relayd_id != (uint64_t) -1ULL) {
1019 use_relayd = 1;
1020 }
1021
1022 channel = consumer_find_channel(key);
1023 if (!channel) {
1024 ERR("UST snapshot channel not found for key %" PRIu64, key);
1025 ret = -1;
1026 goto error;
1027 }
1028 assert(!channel->monitor);
1029 DBG("UST consumer snapshot channel %" PRIu64, key);
1030
1031 cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
1032 /* Are we at a position _before_ the first available packet ? */
1033 bool before_first_packet = true;
1034
1035 health_code_update();
1036
1037 /* Lock stream because we are about to change its state. */
1038 pthread_mutex_lock(&stream->lock);
1039 stream->net_seq_idx = relayd_id;
1040
1041 if (use_relayd) {
1042 ret = consumer_send_relayd_stream(stream, path);
1043 if (ret < 0) {
1044 goto error_unlock;
1045 }
1046 } else {
1047 ret = utils_create_stream_file(path, stream->name,
1048 stream->chan->tracefile_size,
1049 stream->tracefile_count_current,
1050 stream->uid, stream->gid, NULL);
1051 if (ret < 0) {
1052 goto error_unlock;
1053 }
1054 stream->out_fd = ret;
1055 stream->tracefile_size_current = 0;
1056
1057 DBG("UST consumer snapshot stream %s/%s (%" PRIu64 ")", path,
1058 stream->name, stream->key);
1059 }
1060 if (relayd_id != -1ULL) {
1061 ret = consumer_send_relayd_streams_sent(relayd_id);
1062 if (ret < 0) {
1063 goto error_unlock;
1064 }
1065 }
1066
1067 ustctl_flush_buffer(stream->ustream, 1);
1068
1069 ret = lttng_ustconsumer_take_snapshot(stream);
1070 if (ret < 0) {
1071 ERR("Taking UST snapshot");
1072 goto error_unlock;
1073 }
1074
1075 ret = lttng_ustconsumer_get_produced_snapshot(stream, &produced_pos);
1076 if (ret < 0) {
1077 ERR("Produced UST snapshot position");
1078 goto error_unlock;
1079 }
1080
1081 ret = lttng_ustconsumer_get_consumed_snapshot(stream, &consumed_pos);
1082 if (ret < 0) {
1083 ERR("Consumerd UST snapshot position");
1084 goto error_unlock;
1085 }
1086
1087 /*
1088 * The original value is sent back if max stream size is larger than
1089 * the possible size of the snapshot. Also, we assume that the session
1090 * daemon should never send a maximum stream size that is lower than
1091 * subbuffer size.
1092 */
1093 consumed_pos = consumer_get_consume_start_pos(consumed_pos,
1094 produced_pos, nb_packets_per_stream,
1095 stream->max_sb_size);
1096
1097 while (consumed_pos < produced_pos) {
1098 ssize_t read_len;
1099 unsigned long len, padded_len;
1100 int lost_packet = 0;
1101
1102 health_code_update();
1103
1104 DBG("UST consumer taking snapshot at pos %lu", consumed_pos);
1105
1106 ret = ustctl_get_subbuf(stream->ustream, &consumed_pos);
1107 if (ret < 0) {
1108 if (ret != -EAGAIN) {
1109 PERROR("ustctl_get_subbuf snapshot");
1110 goto error_close_stream;
1111 }
1112 DBG("UST consumer get subbuf failed. Skipping it.");
1113 consumed_pos += stream->max_sb_size;
1114
1115 /*
1116 * Start accounting lost packets only when we
1117 * already have extracted packets (to match the
1118 * content of the final snapshot).
1119 */
1120 if (!before_first_packet) {
1121 lost_packet = 1;
1122 }
1123 continue;
1124 }
1125
1126 ret = ustctl_get_subbuf_size(stream->ustream, &len);
1127 if (ret < 0) {
1128 ERR("Snapshot ustctl_get_subbuf_size");
1129 goto error_put_subbuf;
1130 }
1131
1132 ret = ustctl_get_padded_subbuf_size(stream->ustream, &padded_len);
1133 if (ret < 0) {
1134 ERR("Snapshot ustctl_get_padded_subbuf_size");
1135 goto error_put_subbuf;
1136 }
1137
1138 read_len = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, len,
1139 padded_len - len, NULL);
1140 if (use_relayd) {
1141 if (read_len != len) {
1142 ret = -EPERM;
1143 goto error_put_subbuf;
1144 }
1145 } else {
1146 if (read_len != padded_len) {
1147 ret = -EPERM;
1148 goto error_put_subbuf;
1149 }
1150 }
1151
1152 ret = ustctl_put_subbuf(stream->ustream);
1153 if (ret < 0) {
1154 ERR("Snapshot ustctl_put_subbuf");
1155 goto error_close_stream;
1156 }
1157 consumed_pos += stream->max_sb_size;
1158
1159 /*
1160 * Only account lost packets located between
1161 * succesfully extracted packets (do not account before
1162 * and after since they are not visible in the
1163 * resulting snapshot).
1164 */
1165 stream->chan->lost_packets += lost_packet;
1166 lost_packet = 0;
1167 before_first_packet = false;
1168 }
1169
1170 /* Simply close the stream so we can use it on the next snapshot. */
1171 consumer_stream_close(stream);
1172 pthread_mutex_unlock(&stream->lock);
1173 }
1174
1175 rcu_read_unlock();
1176 return 0;
1177
1178 error_put_subbuf:
1179 if (ustctl_put_subbuf(stream->ustream) < 0) {
1180 ERR("Snapshot ustctl_put_subbuf");
1181 }
1182 error_close_stream:
1183 consumer_stream_close(stream);
1184 error_unlock:
1185 pthread_mutex_unlock(&stream->lock);
1186 error:
1187 rcu_read_unlock();
1188 return ret;
1189 }
1190
1191 /*
1192 * Receive the metadata updates from the sessiond. Supports receiving
1193 * overlapping metadata, but is needs to always belong to a contiguous
1194 * range starting from 0.
1195 * Be careful about the locks held when calling this function: it needs
1196 * the metadata cache flush to concurrently progress in order to
1197 * complete.
1198 */
1199 int lttng_ustconsumer_recv_metadata(int sock, uint64_t key, uint64_t offset,
1200 uint64_t len, uint64_t version,
1201 struct lttng_consumer_channel *channel, int timer, int wait)
1202 {
1203 int ret, ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1204 char *metadata_str;
1205
1206 DBG("UST consumer push metadata key %" PRIu64 " of len %" PRIu64, key, len);
1207
1208 metadata_str = zmalloc(len * sizeof(char));
1209 if (!metadata_str) {
1210 PERROR("zmalloc metadata string");
1211 ret_code = LTTCOMM_CONSUMERD_ENOMEM;
1212 goto end;
1213 }
1214
1215 health_code_update();
1216
1217 /* Receive metadata string. */
1218 ret = lttcomm_recv_unix_sock(sock, metadata_str, len);
1219 if (ret < 0) {
1220 /* Session daemon is dead so return gracefully. */
1221 ret_code = ret;
1222 goto end_free;
1223 }
1224
1225 health_code_update();
1226
1227 pthread_mutex_lock(&channel->metadata_cache->lock);
1228 ret = consumer_metadata_cache_write(channel, offset, len, version,
1229 metadata_str);
1230 if (ret < 0) {
1231 /* Unable to handle metadata. Notify session daemon. */
1232 ret_code = LTTCOMM_CONSUMERD_ERROR_METADATA;
1233 /*
1234 * Skip metadata flush on write error since the offset and len might
1235 * not have been updated which could create an infinite loop below when
1236 * waiting for the metadata cache to be flushed.
1237 */
1238 pthread_mutex_unlock(&channel->metadata_cache->lock);
1239 goto end_free;
1240 }
1241 pthread_mutex_unlock(&channel->metadata_cache->lock);
1242
1243 if (!wait) {
1244 goto end_free;
1245 }
1246 while (consumer_metadata_cache_flushed(channel, offset + len, timer)) {
1247 DBG("Waiting for metadata to be flushed");
1248
1249 health_code_update();
1250
1251 usleep(DEFAULT_METADATA_AVAILABILITY_WAIT_TIME);
1252 }
1253
1254 end_free:
1255 free(metadata_str);
1256 end:
1257 return ret_code;
1258 }
1259
1260 /*
1261 * Receive command from session daemon and process it.
1262 *
1263 * Return 1 on success else a negative value or 0.
1264 */
1265 int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
1266 int sock, struct pollfd *consumer_sockpoll)
1267 {
1268 ssize_t ret;
1269 enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1270 struct lttcomm_consumer_msg msg;
1271 struct lttng_consumer_channel *channel = NULL;
1272
1273 health_code_update();
1274
1275 ret = lttcomm_recv_unix_sock(sock, &msg, sizeof(msg));
1276 if (ret != sizeof(msg)) {
1277 DBG("Consumer received unexpected message size %zd (expects %zu)",
1278 ret, sizeof(msg));
1279 /*
1280 * The ret value might 0 meaning an orderly shutdown but this is ok
1281 * since the caller handles this.
1282 */
1283 if (ret > 0) {
1284 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD);
1285 ret = -1;
1286 }
1287 return ret;
1288 }
1289
1290 health_code_update();
1291
1292 /* deprecated */
1293 assert(msg.cmd_type != LTTNG_CONSUMER_STOP);
1294
1295 health_code_update();
1296
1297 /* relayd needs RCU read-side lock */
1298 rcu_read_lock();
1299
1300 switch (msg.cmd_type) {
1301 case LTTNG_CONSUMER_ADD_RELAYD_SOCKET:
1302 {
1303 /* Session daemon status message are handled in the following call. */
1304 ret = consumer_add_relayd_socket(msg.u.relayd_sock.net_index,
1305 msg.u.relayd_sock.type, ctx, sock, consumer_sockpoll,
1306 &msg.u.relayd_sock.sock, msg.u.relayd_sock.session_id,
1307 msg.u.relayd_sock.relayd_session_id);
1308 goto end_nosignal;
1309 }
1310 case LTTNG_CONSUMER_DESTROY_RELAYD:
1311 {
1312 uint64_t index = msg.u.destroy_relayd.net_seq_idx;
1313 struct consumer_relayd_sock_pair *relayd;
1314
1315 DBG("UST consumer destroying relayd %" PRIu64, index);
1316
1317 /* Get relayd reference if exists. */
1318 relayd = consumer_find_relayd(index);
1319 if (relayd == NULL) {
1320 DBG("Unable to find relayd %" PRIu64, index);
1321 ret_code = LTTCOMM_CONSUMERD_RELAYD_FAIL;
1322 }
1323
1324 /*
1325 * Each relayd socket pair has a refcount of stream attached to it
1326 * which tells if the relayd is still active or not depending on the
1327 * refcount value.
1328 *
1329 * This will set the destroy flag of the relayd object and destroy it
1330 * if the refcount reaches zero when called.
1331 *
1332 * The destroy can happen either here or when a stream fd hangs up.
1333 */
1334 if (relayd) {
1335 consumer_flag_relayd_for_destroy(relayd);
1336 }
1337
1338 goto end_msg_sessiond;
1339 }
1340 case LTTNG_CONSUMER_UPDATE_STREAM:
1341 {
1342 rcu_read_unlock();
1343 return -ENOSYS;
1344 }
1345 case LTTNG_CONSUMER_DATA_PENDING:
1346 {
1347 int ret, is_data_pending;
1348 uint64_t id = msg.u.data_pending.session_id;
1349
1350 DBG("UST consumer data pending command for id %" PRIu64, id);
1351
1352 is_data_pending = consumer_data_pending(id);
1353
1354 /* Send back returned value to session daemon */
1355 ret = lttcomm_send_unix_sock(sock, &is_data_pending,
1356 sizeof(is_data_pending));
1357 if (ret < 0) {
1358 DBG("Error when sending the data pending ret code: %d", ret);
1359 goto error_fatal;
1360 }
1361
1362 /*
1363 * No need to send back a status message since the data pending
1364 * returned value is the response.
1365 */
1366 break;
1367 }
1368 case LTTNG_CONSUMER_ASK_CHANNEL_CREATION:
1369 {
1370 int ret;
1371 struct ustctl_consumer_channel_attr attr;
1372
1373 /* Create a plain object and reserve a channel key. */
1374 channel = allocate_channel(msg.u.ask_channel.session_id,
1375 msg.u.ask_channel.pathname, msg.u.ask_channel.name,
1376 msg.u.ask_channel.uid, msg.u.ask_channel.gid,
1377 msg.u.ask_channel.relayd_id, msg.u.ask_channel.key,
1378 (enum lttng_event_output) msg.u.ask_channel.output,
1379 msg.u.ask_channel.tracefile_size,
1380 msg.u.ask_channel.tracefile_count,
1381 msg.u.ask_channel.session_id_per_pid,
1382 msg.u.ask_channel.monitor,
1383 msg.u.ask_channel.live_timer_interval,
1384 msg.u.ask_channel.root_shm_path,
1385 msg.u.ask_channel.shm_path);
1386 if (!channel) {
1387 goto end_channel_error;
1388 }
1389
1390 /*
1391 * Assign UST application UID to the channel. This value is ignored for
1392 * per PID buffers. This is specific to UST thus setting this after the
1393 * allocation.
1394 */
1395 channel->ust_app_uid = msg.u.ask_channel.ust_app_uid;
1396
1397 /* Build channel attributes from received message. */
1398 attr.subbuf_size = msg.u.ask_channel.subbuf_size;
1399 attr.num_subbuf = msg.u.ask_channel.num_subbuf;
1400 attr.overwrite = msg.u.ask_channel.overwrite;
1401 attr.switch_timer_interval = msg.u.ask_channel.switch_timer_interval;
1402 attr.read_timer_interval = msg.u.ask_channel.read_timer_interval;
1403 attr.chan_id = msg.u.ask_channel.chan_id;
1404 memcpy(attr.uuid, msg.u.ask_channel.uuid, sizeof(attr.uuid));
1405
1406 /* Match channel buffer type to the UST abi. */
1407 switch (msg.u.ask_channel.output) {
1408 case LTTNG_EVENT_MMAP:
1409 default:
1410 attr.output = LTTNG_UST_MMAP;
1411 break;
1412 }
1413
1414 /* Translate and save channel type. */
1415 switch (msg.u.ask_channel.type) {
1416 case LTTNG_UST_CHAN_PER_CPU:
1417 channel->type = CONSUMER_CHANNEL_TYPE_DATA;
1418 attr.type = LTTNG_UST_CHAN_PER_CPU;
1419 /*
1420 * Set refcount to 1 for owner. Below, we will
1421 * pass ownership to the
1422 * consumer_thread_channel_poll() thread.
1423 */
1424 channel->refcount = 1;
1425 break;
1426 case LTTNG_UST_CHAN_METADATA:
1427 channel->type = CONSUMER_CHANNEL_TYPE_METADATA;
1428 attr.type = LTTNG_UST_CHAN_METADATA;
1429 break;
1430 default:
1431 assert(0);
1432 goto error_fatal;
1433 };
1434
1435 health_code_update();
1436
1437 ret = ask_channel(ctx, sock, channel, &attr);
1438 if (ret < 0) {
1439 goto end_channel_error;
1440 }
1441
1442 if (msg.u.ask_channel.type == LTTNG_UST_CHAN_METADATA) {
1443 ret = consumer_metadata_cache_allocate(channel);
1444 if (ret < 0) {
1445 ERR("Allocating metadata cache");
1446 goto end_channel_error;
1447 }
1448 consumer_timer_switch_start(channel, attr.switch_timer_interval);
1449 attr.switch_timer_interval = 0;
1450 } else {
1451 consumer_timer_live_start(channel,
1452 msg.u.ask_channel.live_timer_interval);
1453 }
1454
1455 health_code_update();
1456
1457 /*
1458 * Add the channel to the internal state AFTER all streams were created
1459 * and successfully sent to session daemon. This way, all streams must
1460 * be ready before this channel is visible to the threads.
1461 * If add_channel succeeds, ownership of the channel is
1462 * passed to consumer_thread_channel_poll().
1463 */
1464 ret = add_channel(channel, ctx);
1465 if (ret < 0) {
1466 if (msg.u.ask_channel.type == LTTNG_UST_CHAN_METADATA) {
1467 if (channel->switch_timer_enabled == 1) {
1468 consumer_timer_switch_stop(channel);
1469 }
1470 consumer_metadata_cache_destroy(channel);
1471 }
1472 if (channel->live_timer_enabled == 1) {
1473 consumer_timer_live_stop(channel);
1474 }
1475 goto end_channel_error;
1476 }
1477
1478 health_code_update();
1479
1480 /*
1481 * Channel and streams are now created. Inform the session daemon that
1482 * everything went well and should wait to receive the channel and
1483 * streams with ustctl API.
1484 */
1485 ret = consumer_send_status_channel(sock, channel);
1486 if (ret < 0) {
1487 /*
1488 * There is probably a problem on the socket.
1489 */
1490 goto error_fatal;
1491 }
1492
1493 break;
1494 }
1495 case LTTNG_CONSUMER_GET_CHANNEL:
1496 {
1497 int ret, relayd_err = 0;
1498 uint64_t key = msg.u.get_channel.key;
1499 struct lttng_consumer_channel *channel;
1500
1501 channel = consumer_find_channel(key);
1502 if (!channel) {
1503 ERR("UST consumer get channel key %" PRIu64 " not found", key);
1504 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
1505 goto end_msg_sessiond;
1506 }
1507
1508 health_code_update();
1509
1510 /* Send everything to sessiond. */
1511 ret = send_sessiond_channel(sock, channel, ctx, &relayd_err);
1512 if (ret < 0) {
1513 if (relayd_err) {
1514 /*
1515 * We were unable to send to the relayd the stream so avoid
1516 * sending back a fatal error to the thread since this is OK
1517 * and the consumer can continue its work. The above call
1518 * has sent the error status message to the sessiond.
1519 */
1520 goto end_nosignal;
1521 }
1522 /*
1523 * The communicaton was broken hence there is a bad state between
1524 * the consumer and sessiond so stop everything.
1525 */
1526 goto error_fatal;
1527 }
1528
1529 health_code_update();
1530
1531 /*
1532 * In no monitor mode, the streams ownership is kept inside the channel
1533 * so don't send them to the data thread.
1534 */
1535 if (!channel->monitor) {
1536 goto end_msg_sessiond;
1537 }
1538
1539 ret = send_streams_to_thread(channel, ctx);
1540 if (ret < 0) {
1541 /*
1542 * If we are unable to send the stream to the thread, there is
1543 * a big problem so just stop everything.
1544 */
1545 goto error_fatal;
1546 }
1547 /* List MUST be empty after or else it could be reused. */
1548 assert(cds_list_empty(&channel->streams.head));
1549 goto end_msg_sessiond;
1550 }
1551 case LTTNG_CONSUMER_DESTROY_CHANNEL:
1552 {
1553 uint64_t key = msg.u.destroy_channel.key;
1554
1555 /*
1556 * Only called if streams have not been sent to stream
1557 * manager thread. However, channel has been sent to
1558 * channel manager thread.
1559 */
1560 notify_thread_del_channel(ctx, key);
1561 goto end_msg_sessiond;
1562 }
1563 case LTTNG_CONSUMER_CLOSE_METADATA:
1564 {
1565 int ret;
1566
1567 ret = close_metadata(msg.u.close_metadata.key);
1568 if (ret != 0) {
1569 ret_code = ret;
1570 }
1571
1572 goto end_msg_sessiond;
1573 }
1574 case LTTNG_CONSUMER_FLUSH_CHANNEL:
1575 {
1576 int ret;
1577
1578 ret = flush_channel(msg.u.flush_channel.key);
1579 if (ret != 0) {
1580 ret_code = ret;
1581 }
1582
1583 goto end_msg_sessiond;
1584 }
1585 case LTTNG_CONSUMER_PUSH_METADATA:
1586 {
1587 int ret;
1588 uint64_t len = msg.u.push_metadata.len;
1589 uint64_t key = msg.u.push_metadata.key;
1590 uint64_t offset = msg.u.push_metadata.target_offset;
1591 uint64_t version = msg.u.push_metadata.version;
1592 struct lttng_consumer_channel *channel;
1593
1594 DBG("UST consumer push metadata key %" PRIu64 " of len %" PRIu64, key,
1595 len);
1596
1597 channel = consumer_find_channel(key);
1598 if (!channel) {
1599 /*
1600 * This is possible if the metadata creation on the consumer side
1601 * is in flight vis-a-vis a concurrent push metadata from the
1602 * session daemon. Simply return that the channel failed and the
1603 * session daemon will handle that message correctly considering
1604 * that this race is acceptable thus the DBG() statement here.
1605 */
1606 DBG("UST consumer push metadata %" PRIu64 " not found", key);
1607 ret_code = LTTCOMM_CONSUMERD_CHANNEL_FAIL;
1608 goto end_msg_sessiond;
1609 }
1610
1611 health_code_update();
1612
1613 if (!len) {
1614 /*
1615 * There is nothing to receive. We have simply
1616 * checked whether the channel can be found.
1617 */
1618 ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1619 goto end_msg_sessiond;
1620 }
1621
1622 /* Tell session daemon we are ready to receive the metadata. */
1623 ret = consumer_send_status_msg(sock, LTTCOMM_CONSUMERD_SUCCESS);
1624 if (ret < 0) {
1625 /* Somehow, the session daemon is not responding anymore. */
1626 goto error_fatal;
1627 }
1628
1629 health_code_update();
1630
1631 /* Wait for more data. */
1632 health_poll_entry();
1633 ret = lttng_consumer_poll_socket(consumer_sockpoll);
1634 health_poll_exit();
1635 if (ret) {
1636 goto error_fatal;
1637 }
1638
1639 health_code_update();
1640
1641 ret = lttng_ustconsumer_recv_metadata(sock, key, offset,
1642 len, version, channel, 0, 1);
1643 if (ret < 0) {
1644 /* error receiving from sessiond */
1645 goto error_fatal;
1646 } else {
1647 ret_code = ret;
1648 goto end_msg_sessiond;
1649 }
1650 }
1651 case LTTNG_CONSUMER_SETUP_METADATA:
1652 {
1653 int ret;
1654
1655 ret = setup_metadata(ctx, msg.u.setup_metadata.key);
1656 if (ret) {
1657 ret_code = ret;
1658 }
1659 goto end_msg_sessiond;
1660 }
1661 case LTTNG_CONSUMER_SNAPSHOT_CHANNEL:
1662 {
1663 if (msg.u.snapshot_channel.metadata) {
1664 ret = snapshot_metadata(msg.u.snapshot_channel.key,
1665 msg.u.snapshot_channel.pathname,
1666 msg.u.snapshot_channel.relayd_id,
1667 ctx);
1668 if (ret < 0) {
1669 ERR("Snapshot metadata failed");
1670 ret_code = LTTCOMM_CONSUMERD_ERROR_METADATA;
1671 }
1672 } else {
1673 ret = snapshot_channel(msg.u.snapshot_channel.key,
1674 msg.u.snapshot_channel.pathname,
1675 msg.u.snapshot_channel.relayd_id,
1676 msg.u.snapshot_channel.nb_packets_per_stream,
1677 ctx);
1678 if (ret < 0) {
1679 ERR("Snapshot channel failed");
1680 ret_code = LTTCOMM_CONSUMERD_CHANNEL_FAIL;
1681 }
1682 }
1683
1684 health_code_update();
1685 ret = consumer_send_status_msg(sock, ret_code);
1686 if (ret < 0) {
1687 /* Somehow, the session daemon is not responding anymore. */
1688 goto end_nosignal;
1689 }
1690 health_code_update();
1691 break;
1692 }
1693 case LTTNG_CONSUMER_DISCARDED_EVENTS:
1694 {
1695 int ret = 0;
1696 uint64_t discarded_events;
1697 struct lttng_ht_iter iter;
1698 struct lttng_ht *ht;
1699 struct lttng_consumer_stream *stream;
1700 uint64_t id = msg.u.discarded_events.session_id;
1701 uint64_t key = msg.u.discarded_events.channel_key;
1702
1703 DBG("UST consumer discarded events command for session id %"
1704 PRIu64, id);
1705 rcu_read_lock();
1706 pthread_mutex_lock(&consumer_data.lock);
1707
1708 ht = consumer_data.stream_list_ht;
1709
1710 /*
1711 * We only need a reference to the channel, but they are not
1712 * directly indexed, so we just use the first matching stream
1713 * to extract the information we need, we default to 0 if not
1714 * found (no events are dropped if the channel is not yet in
1715 * use).
1716 */
1717 discarded_events = 0;
1718 cds_lfht_for_each_entry_duplicate(ht->ht,
1719 ht->hash_fct(&id, lttng_ht_seed),
1720 ht->match_fct, &id,
1721 &iter.iter, stream, node_session_id.node) {
1722 if (stream->chan->key == key) {
1723 discarded_events = stream->chan->discarded_events;
1724 break;
1725 }
1726 }
1727 pthread_mutex_unlock(&consumer_data.lock);
1728 rcu_read_unlock();
1729
1730 DBG("UST consumer discarded events command for session id %"
1731 PRIu64 ", channel key %" PRIu64, id, key);
1732
1733 health_code_update();
1734
1735 /* Send back returned value to session daemon */
1736 ret = lttcomm_send_unix_sock(sock, &discarded_events, sizeof(discarded_events));
1737 if (ret < 0) {
1738 PERROR("send discarded events");
1739 goto error_fatal;
1740 }
1741
1742 break;
1743 }
1744 case LTTNG_CONSUMER_LOST_PACKETS:
1745 {
1746 int ret;
1747 uint64_t lost_packets;
1748 struct lttng_ht_iter iter;
1749 struct lttng_ht *ht;
1750 struct lttng_consumer_stream *stream;
1751 uint64_t id = msg.u.lost_packets.session_id;
1752 uint64_t key = msg.u.lost_packets.channel_key;
1753
1754 DBG("UST consumer lost packets command for session id %"
1755 PRIu64, id);
1756 rcu_read_lock();
1757 pthread_mutex_lock(&consumer_data.lock);
1758
1759 ht = consumer_data.stream_list_ht;
1760
1761 /*
1762 * We only need a reference to the channel, but they are not
1763 * directly indexed, so we just use the first matching stream
1764 * to extract the information we need, we default to 0 if not
1765 * found (no packets lost if the channel is not yet in use).
1766 */
1767 lost_packets = 0;
1768 cds_lfht_for_each_entry_duplicate(ht->ht,
1769 ht->hash_fct(&id, lttng_ht_seed),
1770 ht->match_fct, &id,
1771 &iter.iter, stream, node_session_id.node) {
1772 if (stream->chan->key == key) {
1773 lost_packets = stream->chan->lost_packets;
1774 break;
1775 }
1776 }
1777 pthread_mutex_unlock(&consumer_data.lock);
1778 rcu_read_unlock();
1779
1780 DBG("UST consumer lost packets command for session id %"
1781 PRIu64 ", channel key %" PRIu64, id, key);
1782
1783 health_code_update();
1784
1785 /* Send back returned value to session daemon */
1786 ret = lttcomm_send_unix_sock(sock, &lost_packets,
1787 sizeof(lost_packets));
1788 if (ret < 0) {
1789 PERROR("send lost packets");
1790 goto error_fatal;
1791 }
1792
1793 break;
1794 }
1795 default:
1796 break;
1797 }
1798
1799 end_nosignal:
1800 rcu_read_unlock();
1801
1802 health_code_update();
1803
1804 /*
1805 * Return 1 to indicate success since the 0 value can be a socket
1806 * shutdown during the recv() or send() call.
1807 */
1808 return 1;
1809
1810 end_msg_sessiond:
1811 /*
1812 * The returned value here is not useful since either way we'll return 1 to
1813 * the caller because the session daemon socket management is done
1814 * elsewhere. Returning a negative code or 0 will shutdown the consumer.
1815 */
1816 ret = consumer_send_status_msg(sock, ret_code);
1817 if (ret < 0) {
1818 goto error_fatal;
1819 }
1820 rcu_read_unlock();
1821
1822 health_code_update();
1823
1824 return 1;
1825 end_channel_error:
1826 if (channel) {
1827 /*
1828 * Free channel here since no one has a reference to it. We don't
1829 * free after that because a stream can store this pointer.
1830 */
1831 destroy_channel(channel);
1832 }
1833 /* We have to send a status channel message indicating an error. */
1834 ret = consumer_send_status_channel(sock, NULL);
1835 if (ret < 0) {
1836 /* Stop everything if session daemon can not be notified. */
1837 goto error_fatal;
1838 }
1839 rcu_read_unlock();
1840
1841 health_code_update();
1842
1843 return 1;
1844 error_fatal:
1845 rcu_read_unlock();
1846 /* This will issue a consumer stop. */
1847 return -1;
1848 }
1849
1850 /*
1851 * Wrapper over the mmap() read offset from ust-ctl library. Since this can be
1852 * compiled out, we isolate it in this library.
1853 */
1854 int lttng_ustctl_get_mmap_read_offset(struct lttng_consumer_stream *stream,
1855 unsigned long *off)
1856 {
1857 assert(stream);
1858 assert(stream->ustream);
1859
1860 return ustctl_get_mmap_read_offset(stream->ustream, off);
1861 }
1862
1863 /*
1864 * Wrapper over the mmap() read offset from ust-ctl library. Since this can be
1865 * compiled out, we isolate it in this library.
1866 */
1867 void *lttng_ustctl_get_mmap_base(struct lttng_consumer_stream *stream)
1868 {
1869 assert(stream);
1870 assert(stream->ustream);
1871
1872 return ustctl_get_mmap_base(stream->ustream);
1873 }
1874
1875 /*
1876 * Take a snapshot for a specific fd
1877 *
1878 * Returns 0 on success, < 0 on error
1879 */
1880 int lttng_ustconsumer_take_snapshot(struct lttng_consumer_stream *stream)
1881 {
1882 assert(stream);
1883 assert(stream->ustream);
1884
1885 return ustctl_snapshot(stream->ustream);
1886 }
1887
1888 /*
1889 * Get the produced position
1890 *
1891 * Returns 0 on success, < 0 on error
1892 */
1893 int lttng_ustconsumer_get_produced_snapshot(
1894 struct lttng_consumer_stream *stream, unsigned long *pos)
1895 {
1896 assert(stream);
1897 assert(stream->ustream);
1898 assert(pos);
1899
1900 return ustctl_snapshot_get_produced(stream->ustream, pos);
1901 }
1902
1903 /*
1904 * Get the consumed position
1905 *
1906 * Returns 0 on success, < 0 on error
1907 */
1908 int lttng_ustconsumer_get_consumed_snapshot(
1909 struct lttng_consumer_stream *stream, unsigned long *pos)
1910 {
1911 assert(stream);
1912 assert(stream->ustream);
1913 assert(pos);
1914
1915 return ustctl_snapshot_get_consumed(stream->ustream, pos);
1916 }
1917
1918 void lttng_ustconsumer_flush_buffer(struct lttng_consumer_stream *stream,
1919 int producer)
1920 {
1921 assert(stream);
1922 assert(stream->ustream);
1923
1924 ustctl_flush_buffer(stream->ustream, producer);
1925 }
1926
1927 int lttng_ustconsumer_get_current_timestamp(
1928 struct lttng_consumer_stream *stream, uint64_t *ts)
1929 {
1930 assert(stream);
1931 assert(stream->ustream);
1932 assert(ts);
1933
1934 return ustctl_get_current_timestamp(stream->ustream, ts);
1935 }
1936
1937 int lttng_ustconsumer_get_sequence_number(
1938 struct lttng_consumer_stream *stream, uint64_t *seq)
1939 {
1940 assert(stream);
1941 assert(stream->ustream);
1942 assert(seq);
1943
1944 return ustctl_get_sequence_number(stream->ustream, seq);
1945 }
1946
1947 /*
1948 * Called when the stream signal the consumer that it has hang up.
1949 */
1950 void lttng_ustconsumer_on_stream_hangup(struct lttng_consumer_stream *stream)
1951 {
1952 assert(stream);
1953 assert(stream->ustream);
1954
1955 ustctl_flush_buffer(stream->ustream, 0);
1956 stream->hangup_flush_done = 1;
1957 }
1958
1959 void lttng_ustconsumer_del_channel(struct lttng_consumer_channel *chan)
1960 {
1961 int i;
1962
1963 assert(chan);
1964 assert(chan->uchan);
1965
1966 if (chan->switch_timer_enabled == 1) {
1967 consumer_timer_switch_stop(chan);
1968 }
1969 for (i = 0; i < chan->nr_stream_fds; i++) {
1970 int ret;
1971
1972 ret = close(chan->stream_fds[i]);
1973 if (ret) {
1974 PERROR("close");
1975 }
1976 if (chan->shm_path[0]) {
1977 char shm_path[PATH_MAX];
1978
1979 ret = get_stream_shm_path(shm_path, chan->shm_path, i);
1980 if (ret) {
1981 ERR("Cannot get stream shm path");
1982 }
1983 ret = run_as_unlink(shm_path, chan->uid, chan->gid);
1984 if (ret) {
1985 PERROR("unlink %s", shm_path);
1986 }
1987 }
1988 }
1989 }
1990
1991 void lttng_ustconsumer_free_channel(struct lttng_consumer_channel *chan)
1992 {
1993 assert(chan);
1994 assert(chan->uchan);
1995
1996 consumer_metadata_cache_destroy(chan);
1997 ustctl_destroy_channel(chan->uchan);
1998 /* Try to rmdir all directories under shm_path root. */
1999 if (chan->root_shm_path[0]) {
2000 (void) run_as_recursive_rmdir(chan->root_shm_path,
2001 chan->uid, chan->gid);
2002 }
2003 free(chan->stream_fds);
2004 }
2005
2006 void lttng_ustconsumer_del_stream(struct lttng_consumer_stream *stream)
2007 {
2008 assert(stream);
2009 assert(stream->ustream);
2010
2011 if (stream->chan->switch_timer_enabled == 1) {
2012 consumer_timer_switch_stop(stream->chan);
2013 }
2014 ustctl_destroy_stream(stream->ustream);
2015 }
2016
2017 int lttng_ustconsumer_get_wakeup_fd(struct lttng_consumer_stream *stream)
2018 {
2019 assert(stream);
2020 assert(stream->ustream);
2021
2022 return ustctl_stream_get_wakeup_fd(stream->ustream);
2023 }
2024
2025 int lttng_ustconsumer_close_wakeup_fd(struct lttng_consumer_stream *stream)
2026 {
2027 assert(stream);
2028 assert(stream->ustream);
2029
2030 return ustctl_stream_close_wakeup_fd(stream->ustream);
2031 }
2032
2033 /*
2034 * Populate index values of a UST stream. Values are set in big endian order.
2035 *
2036 * Return 0 on success or else a negative value.
2037 */
2038 static int get_index_values(struct ctf_packet_index *index,
2039 struct ustctl_consumer_stream *ustream)
2040 {
2041 int ret;
2042
2043 ret = ustctl_get_timestamp_begin(ustream, &index->timestamp_begin);
2044 if (ret < 0) {
2045 PERROR("ustctl_get_timestamp_begin");
2046 goto error;
2047 }
2048 index->timestamp_begin = htobe64(index->timestamp_begin);
2049
2050 ret = ustctl_get_timestamp_end(ustream, &index->timestamp_end);
2051 if (ret < 0) {
2052 PERROR("ustctl_get_timestamp_end");
2053 goto error;
2054 }
2055 index->timestamp_end = htobe64(index->timestamp_end);
2056
2057 ret = ustctl_get_events_discarded(ustream, &index->events_discarded);
2058 if (ret < 0) {
2059 PERROR("ustctl_get_events_discarded");
2060 goto error;
2061 }
2062 index->events_discarded = htobe64(index->events_discarded);
2063
2064 ret = ustctl_get_content_size(ustream, &index->content_size);
2065 if (ret < 0) {
2066 PERROR("ustctl_get_content_size");
2067 goto error;
2068 }
2069 index->content_size = htobe64(index->content_size);
2070
2071 ret = ustctl_get_packet_size(ustream, &index->packet_size);
2072 if (ret < 0) {
2073 PERROR("ustctl_get_packet_size");
2074 goto error;
2075 }
2076 index->packet_size = htobe64(index->packet_size);
2077
2078 ret = ustctl_get_stream_id(ustream, &index->stream_id);
2079 if (ret < 0) {
2080 PERROR("ustctl_get_stream_id");
2081 goto error;
2082 }
2083 index->stream_id = htobe64(index->stream_id);
2084
2085 ret = ustctl_get_instance_id(ustream, &index->stream_instance_id);
2086 if (ret < 0) {
2087 PERROR("ustctl_get_instance_id");
2088 goto error;
2089 }
2090 index->stream_instance_id = htobe64(index->stream_instance_id);
2091
2092 ret = ustctl_get_sequence_number(ustream, &index->packet_seq_num);
2093 if (ret < 0) {
2094 PERROR("ustctl_get_sequence_number");
2095 goto error;
2096 }
2097 index->packet_seq_num = htobe64(index->packet_seq_num);
2098
2099 error:
2100 return ret;
2101 }
2102
2103 static
2104 void metadata_stream_reset_cache(struct lttng_consumer_stream *stream,
2105 struct consumer_metadata_cache *cache)
2106 {
2107 DBG("Metadata stream update to version %" PRIu64,
2108 cache->version);
2109 stream->ust_metadata_pushed = 0;
2110 stream->metadata_version = cache->version;
2111 stream->reset_metadata_flag = 1;
2112 }
2113
2114 /*
2115 * Check if the version of the metadata stream and metadata cache match.
2116 * If the cache got updated, reset the metadata stream.
2117 * The stream lock and metadata cache lock MUST be held.
2118 * Return 0 on success, a negative value on error.
2119 */
2120 static
2121 int metadata_stream_check_version(struct lttng_consumer_stream *stream)
2122 {
2123 int ret = 0;
2124 struct consumer_metadata_cache *cache = stream->chan->metadata_cache;
2125
2126 if (cache->version == stream->metadata_version) {
2127 goto end;
2128 }
2129 metadata_stream_reset_cache(stream, cache);
2130
2131 end:
2132 return ret;
2133 }
2134
2135 /*
2136 * Write up to one packet from the metadata cache to the channel.
2137 *
2138 * Returns the number of bytes pushed in the cache, or a negative value
2139 * on error.
2140 */
2141 static
2142 int commit_one_metadata_packet(struct lttng_consumer_stream *stream)
2143 {
2144 ssize_t write_len;
2145 int ret;
2146
2147 pthread_mutex_lock(&stream->chan->metadata_cache->lock);
2148 ret = metadata_stream_check_version(stream);
2149 if (ret < 0) {
2150 goto end;
2151 }
2152 if (stream->chan->metadata_cache->max_offset
2153 == stream->ust_metadata_pushed) {
2154 ret = 0;
2155 goto end;
2156 }
2157
2158 write_len = ustctl_write_one_packet_to_channel(stream->chan->uchan,
2159 &stream->chan->metadata_cache->data[stream->ust_metadata_pushed],
2160 stream->chan->metadata_cache->max_offset
2161 - stream->ust_metadata_pushed);
2162 assert(write_len != 0);
2163 if (write_len < 0) {
2164 ERR("Writing one metadata packet");
2165 ret = -1;
2166 goto end;
2167 }
2168 stream->ust_metadata_pushed += write_len;
2169
2170 assert(stream->chan->metadata_cache->max_offset >=
2171 stream->ust_metadata_pushed);
2172 ret = write_len;
2173
2174 end:
2175 pthread_mutex_unlock(&stream->chan->metadata_cache->lock);
2176 return ret;
2177 }
2178
2179
2180 /*
2181 * Sync metadata meaning request them to the session daemon and snapshot to the
2182 * metadata thread can consumer them.
2183 *
2184 * Metadata stream lock is held here, but we need to release it when
2185 * interacting with sessiond, else we cause a deadlock with live
2186 * awaiting on metadata to be pushed out.
2187 *
2188 * Return 0 if new metadatda is available, EAGAIN if the metadata stream
2189 * is empty or a negative value on error.
2190 */
2191 int lttng_ustconsumer_sync_metadata(struct lttng_consumer_local_data *ctx,
2192 struct lttng_consumer_stream *metadata)
2193 {
2194 int ret;
2195 int retry = 0;
2196
2197 assert(ctx);
2198 assert(metadata);
2199
2200 pthread_mutex_unlock(&metadata->lock);
2201 /*
2202 * Request metadata from the sessiond, but don't wait for the flush
2203 * because we locked the metadata thread.
2204 */
2205 ret = lttng_ustconsumer_request_metadata(ctx, metadata->chan, 0, 0);
2206 if (ret < 0) {
2207 goto end;
2208 }
2209 pthread_mutex_lock(&metadata->lock);
2210
2211 ret = commit_one_metadata_packet(metadata);
2212 if (ret <= 0) {
2213 goto end;
2214 } else if (ret > 0) {
2215 retry = 1;
2216 }
2217
2218 ustctl_flush_buffer(metadata->ustream, 1);
2219 ret = ustctl_snapshot(metadata->ustream);
2220 if (ret < 0) {
2221 if (errno != EAGAIN) {
2222 ERR("Sync metadata, taking UST snapshot");
2223 goto end;
2224 }
2225 DBG("No new metadata when syncing them.");
2226 /* No new metadata, exit. */
2227 ret = ENODATA;
2228 goto end;
2229 }
2230
2231 /*
2232 * After this flush, we still need to extract metadata.
2233 */
2234 if (retry) {
2235 ret = EAGAIN;
2236 }
2237
2238 end:
2239 return ret;
2240 }
2241
2242 /*
2243 * Return 0 on success else a negative value.
2244 */
2245 static int notify_if_more_data(struct lttng_consumer_stream *stream,
2246 struct lttng_consumer_local_data *ctx)
2247 {
2248 int ret;
2249 struct ustctl_consumer_stream *ustream;
2250
2251 assert(stream);
2252 assert(ctx);
2253
2254 ustream = stream->ustream;
2255
2256 /*
2257 * First, we are going to check if there is a new subbuffer available
2258 * before reading the stream wait_fd.
2259 */
2260 /* Get the next subbuffer */
2261 ret = ustctl_get_next_subbuf(ustream);
2262 if (ret) {
2263 /* No more data found, flag the stream. */
2264 stream->has_data = 0;
2265 ret = 0;
2266 goto end;
2267 }
2268
2269 ret = ustctl_put_subbuf(ustream);
2270 assert(!ret);
2271
2272 /* This stream still has data. Flag it and wake up the data thread. */
2273 stream->has_data = 1;
2274
2275 if (stream->monitor && !stream->hangup_flush_done && !ctx->has_wakeup) {
2276 ssize_t writelen;
2277
2278 writelen = lttng_pipe_write(ctx->consumer_wakeup_pipe, "!", 1);
2279 if (writelen < 0 && errno != EAGAIN && errno != EWOULDBLOCK) {
2280 ret = writelen;
2281 goto end;
2282 }
2283
2284 /* The wake up pipe has been notified. */
2285 ctx->has_wakeup = 1;
2286 }
2287 ret = 0;
2288
2289 end:
2290 return ret;
2291 }
2292
2293 static
2294 int update_stream_stats(struct lttng_consumer_stream *stream)
2295 {
2296 int ret;
2297 uint64_t seq, discarded;
2298
2299 ret = ustctl_get_sequence_number(stream->ustream, &seq);
2300 if (ret < 0) {
2301 PERROR("ustctl_get_sequence_number");
2302 goto end;
2303 }
2304 /*
2305 * Start the sequence when we extract the first packet in case we don't
2306 * start at 0 (for example if a consumer is not connected to the
2307 * session immediately after the beginning).
2308 */
2309 if (stream->last_sequence_number == -1ULL) {
2310 stream->last_sequence_number = seq;
2311 } else if (seq > stream->last_sequence_number) {
2312 stream->chan->lost_packets += seq -
2313 stream->last_sequence_number - 1;
2314 } else {
2315 /* seq <= last_sequence_number */
2316 ERR("Sequence number inconsistent : prev = %" PRIu64
2317 ", current = %" PRIu64,
2318 stream->last_sequence_number, seq);
2319 ret = -1;
2320 goto end;
2321 }
2322 stream->last_sequence_number = seq;
2323
2324 ret = ustctl_get_events_discarded(stream->ustream, &discarded);
2325 if (ret < 0) {
2326 PERROR("kernctl_get_events_discarded");
2327 goto end;
2328 }
2329 if (discarded < stream->last_discarded_events) {
2330 /*
2331 * Overflow has occured. We assume only one wrap-around
2332 * has occured.
2333 */
2334 stream->chan->discarded_events +=
2335 (1ULL << (CAA_BITS_PER_LONG - 1)) -
2336 stream->last_discarded_events + discarded;
2337 } else {
2338 stream->chan->discarded_events += discarded -
2339 stream->last_discarded_events;
2340 }
2341 stream->last_discarded_events = discarded;
2342 ret = 0;
2343
2344 end:
2345 return ret;
2346 }
2347
2348 /*
2349 * Read subbuffer from the given stream.
2350 *
2351 * Stream lock MUST be acquired.
2352 *
2353 * Return 0 on success else a negative value.
2354 */
2355 int lttng_ustconsumer_read_subbuffer(struct lttng_consumer_stream *stream,
2356 struct lttng_consumer_local_data *ctx)
2357 {
2358 unsigned long len, subbuf_size, padding;
2359 int err, write_index = 1;
2360 long ret = 0;
2361 struct ustctl_consumer_stream *ustream;
2362 struct ctf_packet_index index;
2363
2364 assert(stream);
2365 assert(stream->ustream);
2366 assert(ctx);
2367
2368 DBG("In UST read_subbuffer (wait_fd: %d, name: %s)", stream->wait_fd,
2369 stream->name);
2370
2371 /* Ease our life for what's next. */
2372 ustream = stream->ustream;
2373
2374 /*
2375 * We can consume the 1 byte written into the wait_fd by UST. Don't trigger
2376 * error if we cannot read this one byte (read returns 0), or if the error
2377 * is EAGAIN or EWOULDBLOCK.
2378 *
2379 * This is only done when the stream is monitored by a thread, before the
2380 * flush is done after a hangup and if the stream is not flagged with data
2381 * since there might be nothing to consume in the wait fd but still have
2382 * data available flagged by the consumer wake up pipe.
2383 */
2384 if (stream->monitor && !stream->hangup_flush_done && !stream->has_data) {
2385 char dummy;
2386 ssize_t readlen;
2387
2388 readlen = lttng_read(stream->wait_fd, &dummy, 1);
2389 if (readlen < 0 && errno != EAGAIN && errno != EWOULDBLOCK) {
2390 ret = readlen;
2391 goto end;
2392 }
2393 }
2394
2395 retry:
2396 /* Get the next subbuffer */
2397 err = ustctl_get_next_subbuf(ustream);
2398 if (err != 0) {
2399 /*
2400 * Populate metadata info if the existing info has
2401 * already been read.
2402 */
2403 if (stream->metadata_flag) {
2404 ret = commit_one_metadata_packet(stream);
2405 if (ret <= 0) {
2406 goto end;
2407 }
2408 ustctl_flush_buffer(stream->ustream, 1);
2409 goto retry;
2410 }
2411
2412 ret = err; /* ustctl_get_next_subbuf returns negative, caller expect positive. */
2413 /*
2414 * This is a debug message even for single-threaded consumer,
2415 * because poll() have more relaxed criterions than get subbuf,
2416 * so get_subbuf may fail for short race windows where poll()
2417 * would issue wakeups.
2418 */
2419 DBG("Reserving sub buffer failed (everything is normal, "
2420 "it is due to concurrency) [ret: %d]", err);
2421 goto end;
2422 }
2423 assert(stream->chan->output == CONSUMER_CHANNEL_MMAP);
2424
2425 if (!stream->metadata_flag) {
2426 index.offset = htobe64(stream->out_fd_offset);
2427 ret = get_index_values(&index, ustream);
2428 if (ret < 0) {
2429 goto end;
2430 }
2431
2432 /* Update the stream's sequence and discarded events count. */
2433 ret = update_stream_stats(stream);
2434 if (ret < 0) {
2435 PERROR("kernctl_get_events_discarded");
2436 goto end;
2437 }
2438 } else {
2439 write_index = 0;
2440 }
2441
2442 /* Get the full padded subbuffer size */
2443 err = ustctl_get_padded_subbuf_size(ustream, &len);
2444 assert(err == 0);
2445
2446 /* Get subbuffer data size (without padding) */
2447 err = ustctl_get_subbuf_size(ustream, &subbuf_size);
2448 assert(err == 0);
2449
2450 /* Make sure we don't get a subbuffer size bigger than the padded */
2451 assert(len >= subbuf_size);
2452
2453 padding = len - subbuf_size;
2454 /* write the subbuffer to the tracefile */
2455 ret = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, subbuf_size, padding, &index);
2456 /*
2457 * The mmap operation should write subbuf_size amount of data when network
2458 * streaming or the full padding (len) size when we are _not_ streaming.
2459 */
2460 if ((ret != subbuf_size && stream->net_seq_idx != (uint64_t) -1ULL) ||
2461 (ret != len && stream->net_seq_idx == (uint64_t) -1ULL)) {
2462 /*
2463 * Display the error but continue processing to try to release the
2464 * subbuffer. This is a DBG statement since any unexpected kill or
2465 * signal, the application gets unregistered, relayd gets closed or
2466 * anything that affects the buffer lifetime will trigger this error.
2467 * So, for the sake of the user, don't print this error since it can
2468 * happen and it is OK with the code flow.
2469 */
2470 DBG("Error writing to tracefile "
2471 "(ret: %ld != len: %lu != subbuf_size: %lu)",
2472 ret, len, subbuf_size);
2473 write_index = 0;
2474 }
2475 err = ustctl_put_next_subbuf(ustream);
2476 assert(err == 0);
2477
2478 /*
2479 * This will consumer the byte on the wait_fd if and only if there is not
2480 * next subbuffer to be acquired.
2481 */
2482 if (!stream->metadata_flag) {
2483 ret = notify_if_more_data(stream, ctx);
2484 if (ret < 0) {
2485 goto end;
2486 }
2487 }
2488
2489 /* Write index if needed. */
2490 if (!write_index) {
2491 goto end;
2492 }
2493
2494 if (stream->chan->live_timer_interval && !stream->metadata_flag) {
2495 /*
2496 * In live, block until all the metadata is sent.
2497 */
2498 pthread_mutex_lock(&stream->metadata_timer_lock);
2499 assert(!stream->missed_metadata_flush);
2500 stream->waiting_on_metadata = true;
2501 pthread_mutex_unlock(&stream->metadata_timer_lock);
2502
2503 err = consumer_stream_sync_metadata(ctx, stream->session_id);
2504
2505 pthread_mutex_lock(&stream->metadata_timer_lock);
2506 stream->waiting_on_metadata = false;
2507 if (stream->missed_metadata_flush) {
2508 stream->missed_metadata_flush = false;
2509 pthread_mutex_unlock(&stream->metadata_timer_lock);
2510 (void) consumer_flush_ust_index(stream);
2511 } else {
2512 pthread_mutex_unlock(&stream->metadata_timer_lock);
2513 }
2514
2515 if (err < 0) {
2516 goto end;
2517 }
2518 }
2519
2520 assert(!stream->metadata_flag);
2521 err = consumer_stream_write_index(stream, &index);
2522 if (err < 0) {
2523 goto end;
2524 }
2525
2526 end:
2527 return ret;
2528 }
2529
2530 /*
2531 * Called when a stream is created.
2532 *
2533 * Return 0 on success or else a negative value.
2534 */
2535 int lttng_ustconsumer_on_recv_stream(struct lttng_consumer_stream *stream)
2536 {
2537 int ret;
2538
2539 assert(stream);
2540
2541 /* Don't create anything if this is set for streaming. */
2542 if (stream->net_seq_idx == (uint64_t) -1ULL && stream->chan->monitor) {
2543 ret = utils_create_stream_file(stream->chan->pathname, stream->name,
2544 stream->chan->tracefile_size, stream->tracefile_count_current,
2545 stream->uid, stream->gid, NULL);
2546 if (ret < 0) {
2547 goto error;
2548 }
2549 stream->out_fd = ret;
2550 stream->tracefile_size_current = 0;
2551
2552 if (!stream->metadata_flag) {
2553 ret = index_create_file(stream->chan->pathname,
2554 stream->name, stream->uid, stream->gid,
2555 stream->chan->tracefile_size,
2556 stream->tracefile_count_current);
2557 if (ret < 0) {
2558 goto error;
2559 }
2560 stream->index_fd = ret;
2561 }
2562 }
2563 ret = 0;
2564
2565 error:
2566 return ret;
2567 }
2568
2569 /*
2570 * Check if data is still being extracted from the buffers for a specific
2571 * stream. Consumer data lock MUST be acquired before calling this function
2572 * and the stream lock.
2573 *
2574 * Return 1 if the traced data are still getting read else 0 meaning that the
2575 * data is available for trace viewer reading.
2576 */
2577 int lttng_ustconsumer_data_pending(struct lttng_consumer_stream *stream)
2578 {
2579 int ret;
2580
2581 assert(stream);
2582 assert(stream->ustream);
2583
2584 DBG("UST consumer checking data pending");
2585
2586 if (stream->endpoint_status != CONSUMER_ENDPOINT_ACTIVE) {
2587 ret = 0;
2588 goto end;
2589 }
2590
2591 if (stream->chan->type == CONSUMER_CHANNEL_TYPE_METADATA) {
2592 uint64_t contiguous, pushed;
2593
2594 /* Ease our life a bit. */
2595 contiguous = stream->chan->metadata_cache->max_offset;
2596 pushed = stream->ust_metadata_pushed;
2597
2598 /*
2599 * We can simply check whether all contiguously available data
2600 * has been pushed to the ring buffer, since the push operation
2601 * is performed within get_next_subbuf(), and because both
2602 * get_next_subbuf() and put_next_subbuf() are issued atomically
2603 * thanks to the stream lock within
2604 * lttng_ustconsumer_read_subbuffer(). This basically means that
2605 * whetnever ust_metadata_pushed is incremented, the associated
2606 * metadata has been consumed from the metadata stream.
2607 */
2608 DBG("UST consumer metadata pending check: contiguous %" PRIu64 " vs pushed %" PRIu64,
2609 contiguous, pushed);
2610 assert(((int64_t) (contiguous - pushed)) >= 0);
2611 if ((contiguous != pushed) ||
2612 (((int64_t) contiguous - pushed) > 0 || contiguous == 0)) {
2613 ret = 1; /* Data is pending */
2614 goto end;
2615 }
2616 } else {
2617 ret = ustctl_get_next_subbuf(stream->ustream);
2618 if (ret == 0) {
2619 /*
2620 * There is still data so let's put back this
2621 * subbuffer.
2622 */
2623 ret = ustctl_put_subbuf(stream->ustream);
2624 assert(ret == 0);
2625 ret = 1; /* Data is pending */
2626 goto end;
2627 }
2628 }
2629
2630 /* Data is NOT pending so ready to be read. */
2631 ret = 0;
2632
2633 end:
2634 return ret;
2635 }
2636
2637 /*
2638 * Stop a given metadata channel timer if enabled and close the wait fd which
2639 * is the poll pipe of the metadata stream.
2640 *
2641 * This MUST be called with the metadata channel acquired.
2642 */
2643 void lttng_ustconsumer_close_metadata(struct lttng_consumer_channel *metadata)
2644 {
2645 int ret;
2646
2647 assert(metadata);
2648 assert(metadata->type == CONSUMER_CHANNEL_TYPE_METADATA);
2649
2650 DBG("Closing metadata channel key %" PRIu64, metadata->key);
2651
2652 if (metadata->switch_timer_enabled == 1) {
2653 consumer_timer_switch_stop(metadata);
2654 }
2655
2656 if (!metadata->metadata_stream) {
2657 goto end;
2658 }
2659
2660 /*
2661 * Closing write side so the thread monitoring the stream wakes up if any
2662 * and clean the metadata stream.
2663 */
2664 if (metadata->metadata_stream->ust_metadata_poll_pipe[1] >= 0) {
2665 ret = close(metadata->metadata_stream->ust_metadata_poll_pipe[1]);
2666 if (ret < 0) {
2667 PERROR("closing metadata pipe write side");
2668 }
2669 metadata->metadata_stream->ust_metadata_poll_pipe[1] = -1;
2670 }
2671
2672 end:
2673 return;
2674 }
2675
2676 /*
2677 * Close every metadata stream wait fd of the metadata hash table. This
2678 * function MUST be used very carefully so not to run into a race between the
2679 * metadata thread handling streams and this function closing their wait fd.
2680 *
2681 * For UST, this is used when the session daemon hangs up. Its the metadata
2682 * producer so calling this is safe because we are assured that no state change
2683 * can occur in the metadata thread for the streams in the hash table.
2684 */
2685 void lttng_ustconsumer_close_all_metadata(struct lttng_ht *metadata_ht)
2686 {
2687 struct lttng_ht_iter iter;
2688 struct lttng_consumer_stream *stream;
2689
2690 assert(metadata_ht);
2691 assert(metadata_ht->ht);
2692
2693 DBG("UST consumer closing all metadata streams");
2694
2695 rcu_read_lock();
2696 cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream,
2697 node.node) {
2698
2699 health_code_update();
2700
2701 pthread_mutex_lock(&stream->chan->lock);
2702 lttng_ustconsumer_close_metadata(stream->chan);
2703 pthread_mutex_unlock(&stream->chan->lock);
2704
2705 }
2706 rcu_read_unlock();
2707 }
2708
2709 void lttng_ustconsumer_close_stream_wakeup(struct lttng_consumer_stream *stream)
2710 {
2711 int ret;
2712
2713 ret = ustctl_stream_close_wakeup_fd(stream->ustream);
2714 if (ret < 0) {
2715 ERR("Unable to close wakeup fd");
2716 }
2717 }
2718
2719 /*
2720 * Please refer to consumer-timer.c before adding any lock within this
2721 * function or any of its callees. Timers have a very strict locking
2722 * semantic with respect to teardown. Failure to respect this semantic
2723 * introduces deadlocks.
2724 *
2725 * DON'T hold the metadata lock when calling this function, else this
2726 * can cause deadlock involving consumer awaiting for metadata to be
2727 * pushed out due to concurrent interaction with the session daemon.
2728 */
2729 int lttng_ustconsumer_request_metadata(struct lttng_consumer_local_data *ctx,
2730 struct lttng_consumer_channel *channel, int timer, int wait)
2731 {
2732 struct lttcomm_metadata_request_msg request;
2733 struct lttcomm_consumer_msg msg;
2734 enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS;
2735 uint64_t len, key, offset, version;
2736 int ret;
2737
2738 assert(channel);
2739 assert(channel->metadata_cache);
2740
2741 memset(&request, 0, sizeof(request));
2742
2743 /* send the metadata request to sessiond */
2744 switch (consumer_data.type) {
2745 case LTTNG_CONSUMER64_UST:
2746 request.bits_per_long = 64;
2747 break;
2748 case LTTNG_CONSUMER32_UST:
2749 request.bits_per_long = 32;
2750 break;
2751 default:
2752 request.bits_per_long = 0;
2753 break;
2754 }
2755
2756 request.session_id = channel->session_id;
2757 request.session_id_per_pid = channel->session_id_per_pid;
2758 /*
2759 * Request the application UID here so the metadata of that application can
2760 * be sent back. The channel UID corresponds to the user UID of the session
2761 * used for the rights on the stream file(s).
2762 */
2763 request.uid = channel->ust_app_uid;
2764 request.key = channel->key;
2765
2766 DBG("Sending metadata request to sessiond, session id %" PRIu64
2767 ", per-pid %" PRIu64 ", app UID %u and channek key %" PRIu64,
2768 request.session_id, request.session_id_per_pid, request.uid,
2769 request.key);
2770
2771 pthread_mutex_lock(&ctx->metadata_socket_lock);
2772
2773 health_code_update();
2774
2775 ret = lttcomm_send_unix_sock(ctx->consumer_metadata_socket, &request,
2776 sizeof(request));
2777 if (ret < 0) {
2778 ERR("Asking metadata to sessiond");
2779 goto end;
2780 }
2781
2782 health_code_update();
2783
2784 /* Receive the metadata from sessiond */
2785 ret = lttcomm_recv_unix_sock(ctx->consumer_metadata_socket, &msg,
2786 sizeof(msg));
2787 if (ret != sizeof(msg)) {
2788 DBG("Consumer received unexpected message size %d (expects %zu)",
2789 ret, sizeof(msg));
2790 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD);
2791 /*
2792 * The ret value might 0 meaning an orderly shutdown but this is ok
2793 * since the caller handles this.
2794 */
2795 goto end;
2796 }
2797
2798 health_code_update();
2799
2800 if (msg.cmd_type == LTTNG_ERR_UND) {
2801 /* No registry found */
2802 (void) consumer_send_status_msg(ctx->consumer_metadata_socket,
2803 ret_code);
2804 ret = 0;
2805 goto end;
2806 } else if (msg.cmd_type != LTTNG_CONSUMER_PUSH_METADATA) {
2807 ERR("Unexpected cmd_type received %d", msg.cmd_type);
2808 ret = -1;
2809 goto end;
2810 }
2811
2812 len = msg.u.push_metadata.len;
2813 key = msg.u.push_metadata.key;
2814 offset = msg.u.push_metadata.target_offset;
2815 version = msg.u.push_metadata.version;
2816
2817 assert(key == channel->key);
2818 if (len == 0) {
2819 DBG("No new metadata to receive for key %" PRIu64, key);
2820 }
2821
2822 health_code_update();
2823
2824 /* Tell session daemon we are ready to receive the metadata. */
2825 ret = consumer_send_status_msg(ctx->consumer_metadata_socket,
2826 LTTCOMM_CONSUMERD_SUCCESS);
2827 if (ret < 0 || len == 0) {
2828 /*
2829 * Somehow, the session daemon is not responding anymore or there is
2830 * nothing to receive.
2831 */
2832 goto end;
2833 }
2834
2835 health_code_update();
2836
2837 ret = lttng_ustconsumer_recv_metadata(ctx->consumer_metadata_socket,
2838 key, offset, len, version, channel, timer, wait);
2839 if (ret >= 0) {
2840 /*
2841 * Only send the status msg if the sessiond is alive meaning a positive
2842 * ret code.
2843 */
2844 (void) consumer_send_status_msg(ctx->consumer_metadata_socket, ret);
2845 }
2846 ret = 0;
2847
2848 end:
2849 health_code_update();
2850
2851 pthread_mutex_unlock(&ctx->metadata_socket_lock);
2852 return ret;
2853 }
2854
2855 /*
2856 * Return the ustctl call for the get stream id.
2857 */
2858 int lttng_ustconsumer_get_stream_id(struct lttng_consumer_stream *stream,
2859 uint64_t *stream_id)
2860 {
2861 assert(stream);
2862 assert(stream_id);
2863
2864 return ustctl_get_stream_id(stream->ustream, stream_id);
2865 }
This page took 0.124077 seconds and 5 git commands to generate.