Fix: lost packet accounting always lost on snapshot
[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 health_code_update();
1080
1081 /* Lock stream because we are about to change its state. */
1082 pthread_mutex_lock(&stream->lock);
1083 stream->net_seq_idx = relayd_id;
1084
1085 if (use_relayd) {
1086 ret = consumer_send_relayd_stream(stream, path);
1087 if (ret < 0) {
1088 goto error_unlock;
1089 }
1090 } else {
1091 ret = utils_create_stream_file(path, stream->name,
1092 stream->chan->tracefile_size,
1093 stream->tracefile_count_current,
1094 stream->uid, stream->gid, NULL);
1095 if (ret < 0) {
1096 goto error_unlock;
1097 }
1098 stream->out_fd = ret;
1099 stream->tracefile_size_current = 0;
1100
1101 DBG("UST consumer snapshot stream %s/%s (%" PRIu64 ")", path,
1102 stream->name, stream->key);
1103 }
1104 if (relayd_id != -1ULL) {
1105 ret = consumer_send_relayd_streams_sent(relayd_id);
1106 if (ret < 0) {
1107 goto error_unlock;
1108 }
1109 }
1110
1111 /*
1112 * If tracing is active, we want to perform a "full" buffer flush.
1113 * Else, if quiescent, it has already been done by the prior stop.
1114 */
1115 if (!stream->quiescent) {
1116 ustctl_flush_buffer(stream->ustream, 0);
1117 }
1118
1119 ret = lttng_ustconsumer_take_snapshot(stream);
1120 if (ret < 0) {
1121 ERR("Taking UST snapshot");
1122 goto error_unlock;
1123 }
1124
1125 ret = lttng_ustconsumer_get_produced_snapshot(stream, &produced_pos);
1126 if (ret < 0) {
1127 ERR("Produced UST snapshot position");
1128 goto error_unlock;
1129 }
1130
1131 ret = lttng_ustconsumer_get_consumed_snapshot(stream, &consumed_pos);
1132 if (ret < 0) {
1133 ERR("Consumerd UST snapshot position");
1134 goto error_unlock;
1135 }
1136
1137 /*
1138 * The original value is sent back if max stream size is larger than
1139 * the possible size of the snapshot. Also, we assume that the session
1140 * daemon should never send a maximum stream size that is lower than
1141 * subbuffer size.
1142 */
1143 consumed_pos = consumer_get_consume_start_pos(consumed_pos,
1144 produced_pos, nb_packets_per_stream,
1145 stream->max_sb_size);
1146
1147 while (consumed_pos < produced_pos) {
1148 ssize_t read_len;
1149 unsigned long len, padded_len;
1150
1151 health_code_update();
1152
1153 DBG("UST consumer taking snapshot at pos %lu", consumed_pos);
1154
1155 ret = ustctl_get_subbuf(stream->ustream, &consumed_pos);
1156 if (ret < 0) {
1157 if (ret != -EAGAIN) {
1158 PERROR("ustctl_get_subbuf snapshot");
1159 goto error_close_stream;
1160 }
1161 DBG("UST consumer get subbuf failed. Skipping it.");
1162 consumed_pos += stream->max_sb_size;
1163 stream->chan->lost_packets++;
1164 continue;
1165 }
1166
1167 ret = ustctl_get_subbuf_size(stream->ustream, &len);
1168 if (ret < 0) {
1169 ERR("Snapshot ustctl_get_subbuf_size");
1170 goto error_put_subbuf;
1171 }
1172
1173 ret = ustctl_get_padded_subbuf_size(stream->ustream, &padded_len);
1174 if (ret < 0) {
1175 ERR("Snapshot ustctl_get_padded_subbuf_size");
1176 goto error_put_subbuf;
1177 }
1178
1179 read_len = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, len,
1180 padded_len - len, NULL);
1181 if (use_relayd) {
1182 if (read_len != len) {
1183 ret = -EPERM;
1184 goto error_put_subbuf;
1185 }
1186 } else {
1187 if (read_len != padded_len) {
1188 ret = -EPERM;
1189 goto error_put_subbuf;
1190 }
1191 }
1192
1193 ret = ustctl_put_subbuf(stream->ustream);
1194 if (ret < 0) {
1195 ERR("Snapshot ustctl_put_subbuf");
1196 goto error_close_stream;
1197 }
1198 consumed_pos += stream->max_sb_size;
1199 }
1200
1201 /* Simply close the stream so we can use it on the next snapshot. */
1202 consumer_stream_close(stream);
1203 pthread_mutex_unlock(&stream->lock);
1204 }
1205
1206 rcu_read_unlock();
1207 return 0;
1208
1209 error_put_subbuf:
1210 if (ustctl_put_subbuf(stream->ustream) < 0) {
1211 ERR("Snapshot ustctl_put_subbuf");
1212 }
1213 error_close_stream:
1214 consumer_stream_close(stream);
1215 error_unlock:
1216 pthread_mutex_unlock(&stream->lock);
1217 error:
1218 rcu_read_unlock();
1219 return ret;
1220 }
1221
1222 /*
1223 * Receive the metadata updates from the sessiond. Supports receiving
1224 * overlapping metadata, but is needs to always belong to a contiguous
1225 * range starting from 0.
1226 * Be careful about the locks held when calling this function: it needs
1227 * the metadata cache flush to concurrently progress in order to
1228 * complete.
1229 */
1230 int lttng_ustconsumer_recv_metadata(int sock, uint64_t key, uint64_t offset,
1231 uint64_t len, uint64_t version,
1232 struct lttng_consumer_channel *channel, int timer, int wait)
1233 {
1234 int ret, ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1235 char *metadata_str;
1236
1237 DBG("UST consumer push metadata key %" PRIu64 " of len %" PRIu64, key, len);
1238
1239 metadata_str = zmalloc(len * sizeof(char));
1240 if (!metadata_str) {
1241 PERROR("zmalloc metadata string");
1242 ret_code = LTTCOMM_CONSUMERD_ENOMEM;
1243 goto end;
1244 }
1245
1246 health_code_update();
1247
1248 /* Receive metadata string. */
1249 ret = lttcomm_recv_unix_sock(sock, metadata_str, len);
1250 if (ret < 0) {
1251 /* Session daemon is dead so return gracefully. */
1252 ret_code = ret;
1253 goto end_free;
1254 }
1255
1256 health_code_update();
1257
1258 pthread_mutex_lock(&channel->metadata_cache->lock);
1259 ret = consumer_metadata_cache_write(channel, offset, len, version,
1260 metadata_str);
1261 if (ret < 0) {
1262 /* Unable to handle metadata. Notify session daemon. */
1263 ret_code = LTTCOMM_CONSUMERD_ERROR_METADATA;
1264 /*
1265 * Skip metadata flush on write error since the offset and len might
1266 * not have been updated which could create an infinite loop below when
1267 * waiting for the metadata cache to be flushed.
1268 */
1269 pthread_mutex_unlock(&channel->metadata_cache->lock);
1270 goto end_free;
1271 }
1272 pthread_mutex_unlock(&channel->metadata_cache->lock);
1273
1274 if (!wait) {
1275 goto end_free;
1276 }
1277 while (consumer_metadata_cache_flushed(channel, offset + len, timer)) {
1278 DBG("Waiting for metadata to be flushed");
1279
1280 health_code_update();
1281
1282 usleep(DEFAULT_METADATA_AVAILABILITY_WAIT_TIME);
1283 }
1284
1285 end_free:
1286 free(metadata_str);
1287 end:
1288 return ret_code;
1289 }
1290
1291 /*
1292 * Receive command from session daemon and process it.
1293 *
1294 * Return 1 on success else a negative value or 0.
1295 */
1296 int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
1297 int sock, struct pollfd *consumer_sockpoll)
1298 {
1299 ssize_t ret;
1300 enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1301 struct lttcomm_consumer_msg msg;
1302 struct lttng_consumer_channel *channel = NULL;
1303
1304 health_code_update();
1305
1306 ret = lttcomm_recv_unix_sock(sock, &msg, sizeof(msg));
1307 if (ret != sizeof(msg)) {
1308 DBG("Consumer received unexpected message size %zd (expects %zu)",
1309 ret, sizeof(msg));
1310 /*
1311 * The ret value might 0 meaning an orderly shutdown but this is ok
1312 * since the caller handles this.
1313 */
1314 if (ret > 0) {
1315 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD);
1316 ret = -1;
1317 }
1318 return ret;
1319 }
1320
1321 health_code_update();
1322
1323 /* deprecated */
1324 assert(msg.cmd_type != LTTNG_CONSUMER_STOP);
1325
1326 health_code_update();
1327
1328 /* relayd needs RCU read-side lock */
1329 rcu_read_lock();
1330
1331 switch (msg.cmd_type) {
1332 case LTTNG_CONSUMER_ADD_RELAYD_SOCKET:
1333 {
1334 /* Session daemon status message are handled in the following call. */
1335 ret = consumer_add_relayd_socket(msg.u.relayd_sock.net_index,
1336 msg.u.relayd_sock.type, ctx, sock, consumer_sockpoll,
1337 &msg.u.relayd_sock.sock, msg.u.relayd_sock.session_id,
1338 msg.u.relayd_sock.relayd_session_id);
1339 goto end_nosignal;
1340 }
1341 case LTTNG_CONSUMER_DESTROY_RELAYD:
1342 {
1343 uint64_t index = msg.u.destroy_relayd.net_seq_idx;
1344 struct consumer_relayd_sock_pair *relayd;
1345
1346 DBG("UST consumer destroying relayd %" PRIu64, index);
1347
1348 /* Get relayd reference if exists. */
1349 relayd = consumer_find_relayd(index);
1350 if (relayd == NULL) {
1351 DBG("Unable to find relayd %" PRIu64, index);
1352 ret_code = LTTCOMM_CONSUMERD_RELAYD_FAIL;
1353 }
1354
1355 /*
1356 * Each relayd socket pair has a refcount of stream attached to it
1357 * which tells if the relayd is still active or not depending on the
1358 * refcount value.
1359 *
1360 * This will set the destroy flag of the relayd object and destroy it
1361 * if the refcount reaches zero when called.
1362 *
1363 * The destroy can happen either here or when a stream fd hangs up.
1364 */
1365 if (relayd) {
1366 consumer_flag_relayd_for_destroy(relayd);
1367 }
1368
1369 goto end_msg_sessiond;
1370 }
1371 case LTTNG_CONSUMER_UPDATE_STREAM:
1372 {
1373 rcu_read_unlock();
1374 return -ENOSYS;
1375 }
1376 case LTTNG_CONSUMER_DATA_PENDING:
1377 {
1378 int ret, is_data_pending;
1379 uint64_t id = msg.u.data_pending.session_id;
1380
1381 DBG("UST consumer data pending command for id %" PRIu64, id);
1382
1383 is_data_pending = consumer_data_pending(id);
1384
1385 /* Send back returned value to session daemon */
1386 ret = lttcomm_send_unix_sock(sock, &is_data_pending,
1387 sizeof(is_data_pending));
1388 if (ret < 0) {
1389 DBG("Error when sending the data pending ret code: %d", ret);
1390 goto error_fatal;
1391 }
1392
1393 /*
1394 * No need to send back a status message since the data pending
1395 * returned value is the response.
1396 */
1397 break;
1398 }
1399 case LTTNG_CONSUMER_ASK_CHANNEL_CREATION:
1400 {
1401 int ret;
1402 struct ustctl_consumer_channel_attr attr;
1403
1404 /* Create a plain object and reserve a channel key. */
1405 channel = allocate_channel(msg.u.ask_channel.session_id,
1406 msg.u.ask_channel.pathname, msg.u.ask_channel.name,
1407 msg.u.ask_channel.uid, msg.u.ask_channel.gid,
1408 msg.u.ask_channel.relayd_id, msg.u.ask_channel.key,
1409 (enum lttng_event_output) msg.u.ask_channel.output,
1410 msg.u.ask_channel.tracefile_size,
1411 msg.u.ask_channel.tracefile_count,
1412 msg.u.ask_channel.session_id_per_pid,
1413 msg.u.ask_channel.monitor,
1414 msg.u.ask_channel.live_timer_interval,
1415 msg.u.ask_channel.root_shm_path,
1416 msg.u.ask_channel.shm_path);
1417 if (!channel) {
1418 goto end_channel_error;
1419 }
1420
1421 /*
1422 * Assign UST application UID to the channel. This value is ignored for
1423 * per PID buffers. This is specific to UST thus setting this after the
1424 * allocation.
1425 */
1426 channel->ust_app_uid = msg.u.ask_channel.ust_app_uid;
1427
1428 /* Build channel attributes from received message. */
1429 attr.subbuf_size = msg.u.ask_channel.subbuf_size;
1430 attr.num_subbuf = msg.u.ask_channel.num_subbuf;
1431 attr.overwrite = msg.u.ask_channel.overwrite;
1432 attr.switch_timer_interval = msg.u.ask_channel.switch_timer_interval;
1433 attr.read_timer_interval = msg.u.ask_channel.read_timer_interval;
1434 attr.chan_id = msg.u.ask_channel.chan_id;
1435 memcpy(attr.uuid, msg.u.ask_channel.uuid, sizeof(attr.uuid));
1436
1437 /* Match channel buffer type to the UST abi. */
1438 switch (msg.u.ask_channel.output) {
1439 case LTTNG_EVENT_MMAP:
1440 default:
1441 attr.output = LTTNG_UST_MMAP;
1442 break;
1443 }
1444
1445 /* Translate and save channel type. */
1446 switch (msg.u.ask_channel.type) {
1447 case LTTNG_UST_CHAN_PER_CPU:
1448 channel->type = CONSUMER_CHANNEL_TYPE_DATA;
1449 attr.type = LTTNG_UST_CHAN_PER_CPU;
1450 /*
1451 * Set refcount to 1 for owner. Below, we will
1452 * pass ownership to the
1453 * consumer_thread_channel_poll() thread.
1454 */
1455 channel->refcount = 1;
1456 break;
1457 case LTTNG_UST_CHAN_METADATA:
1458 channel->type = CONSUMER_CHANNEL_TYPE_METADATA;
1459 attr.type = LTTNG_UST_CHAN_METADATA;
1460 break;
1461 default:
1462 assert(0);
1463 goto error_fatal;
1464 };
1465
1466 health_code_update();
1467
1468 ret = ask_channel(ctx, sock, channel, &attr);
1469 if (ret < 0) {
1470 goto end_channel_error;
1471 }
1472
1473 if (msg.u.ask_channel.type == LTTNG_UST_CHAN_METADATA) {
1474 ret = consumer_metadata_cache_allocate(channel);
1475 if (ret < 0) {
1476 ERR("Allocating metadata cache");
1477 goto end_channel_error;
1478 }
1479 consumer_timer_switch_start(channel, attr.switch_timer_interval);
1480 attr.switch_timer_interval = 0;
1481 } else {
1482 consumer_timer_live_start(channel,
1483 msg.u.ask_channel.live_timer_interval);
1484 }
1485
1486 health_code_update();
1487
1488 /*
1489 * Add the channel to the internal state AFTER all streams were created
1490 * and successfully sent to session daemon. This way, all streams must
1491 * be ready before this channel is visible to the threads.
1492 * If add_channel succeeds, ownership of the channel is
1493 * passed to consumer_thread_channel_poll().
1494 */
1495 ret = add_channel(channel, ctx);
1496 if (ret < 0) {
1497 if (msg.u.ask_channel.type == LTTNG_UST_CHAN_METADATA) {
1498 if (channel->switch_timer_enabled == 1) {
1499 consumer_timer_switch_stop(channel);
1500 }
1501 consumer_metadata_cache_destroy(channel);
1502 }
1503 if (channel->live_timer_enabled == 1) {
1504 consumer_timer_live_stop(channel);
1505 }
1506 goto end_channel_error;
1507 }
1508
1509 health_code_update();
1510
1511 /*
1512 * Channel and streams are now created. Inform the session daemon that
1513 * everything went well and should wait to receive the channel and
1514 * streams with ustctl API.
1515 */
1516 ret = consumer_send_status_channel(sock, channel);
1517 if (ret < 0) {
1518 /*
1519 * There is probably a problem on the socket.
1520 */
1521 goto error_fatal;
1522 }
1523
1524 break;
1525 }
1526 case LTTNG_CONSUMER_GET_CHANNEL:
1527 {
1528 int ret, relayd_err = 0;
1529 uint64_t key = msg.u.get_channel.key;
1530 struct lttng_consumer_channel *channel;
1531
1532 channel = consumer_find_channel(key);
1533 if (!channel) {
1534 ERR("UST consumer get channel key %" PRIu64 " not found", key);
1535 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
1536 goto end_msg_sessiond;
1537 }
1538
1539 health_code_update();
1540
1541 /* Send everything to sessiond. */
1542 ret = send_sessiond_channel(sock, channel, ctx, &relayd_err);
1543 if (ret < 0) {
1544 if (relayd_err) {
1545 /*
1546 * We were unable to send to the relayd the stream so avoid
1547 * sending back a fatal error to the thread since this is OK
1548 * and the consumer can continue its work. The above call
1549 * has sent the error status message to the sessiond.
1550 */
1551 goto end_nosignal;
1552 }
1553 /*
1554 * The communicaton was broken hence there is a bad state between
1555 * the consumer and sessiond so stop everything.
1556 */
1557 goto error_fatal;
1558 }
1559
1560 health_code_update();
1561
1562 /*
1563 * In no monitor mode, the streams ownership is kept inside the channel
1564 * so don't send them to the data thread.
1565 */
1566 if (!channel->monitor) {
1567 goto end_msg_sessiond;
1568 }
1569
1570 ret = send_streams_to_thread(channel, ctx);
1571 if (ret < 0) {
1572 /*
1573 * If we are unable to send the stream to the thread, there is
1574 * a big problem so just stop everything.
1575 */
1576 goto error_fatal;
1577 }
1578 /* List MUST be empty after or else it could be reused. */
1579 assert(cds_list_empty(&channel->streams.head));
1580 goto end_msg_sessiond;
1581 }
1582 case LTTNG_CONSUMER_DESTROY_CHANNEL:
1583 {
1584 uint64_t key = msg.u.destroy_channel.key;
1585
1586 /*
1587 * Only called if streams have not been sent to stream
1588 * manager thread. However, channel has been sent to
1589 * channel manager thread.
1590 */
1591 notify_thread_del_channel(ctx, key);
1592 goto end_msg_sessiond;
1593 }
1594 case LTTNG_CONSUMER_CLOSE_METADATA:
1595 {
1596 int ret;
1597
1598 ret = close_metadata(msg.u.close_metadata.key);
1599 if (ret != 0) {
1600 ret_code = ret;
1601 }
1602
1603 goto end_msg_sessiond;
1604 }
1605 case LTTNG_CONSUMER_FLUSH_CHANNEL:
1606 {
1607 int ret;
1608
1609 ret = flush_channel(msg.u.flush_channel.key);
1610 if (ret != 0) {
1611 ret_code = ret;
1612 }
1613
1614 goto end_msg_sessiond;
1615 }
1616 case LTTNG_CONSUMER_CLEAR_QUIESCENT_CHANNEL:
1617 {
1618 int ret;
1619
1620 ret = clear_quiescent_channel(
1621 msg.u.clear_quiescent_channel.key);
1622 if (ret != 0) {
1623 ret_code = ret;
1624 }
1625
1626 goto end_msg_sessiond;
1627 }
1628 case LTTNG_CONSUMER_PUSH_METADATA:
1629 {
1630 int ret;
1631 uint64_t len = msg.u.push_metadata.len;
1632 uint64_t key = msg.u.push_metadata.key;
1633 uint64_t offset = msg.u.push_metadata.target_offset;
1634 uint64_t version = msg.u.push_metadata.version;
1635 struct lttng_consumer_channel *channel;
1636
1637 DBG("UST consumer push metadata key %" PRIu64 " of len %" PRIu64, key,
1638 len);
1639
1640 channel = consumer_find_channel(key);
1641 if (!channel) {
1642 /*
1643 * This is possible if the metadata creation on the consumer side
1644 * is in flight vis-a-vis a concurrent push metadata from the
1645 * session daemon. Simply return that the channel failed and the
1646 * session daemon will handle that message correctly considering
1647 * that this race is acceptable thus the DBG() statement here.
1648 */
1649 DBG("UST consumer push metadata %" PRIu64 " not found", key);
1650 ret_code = LTTCOMM_CONSUMERD_CHANNEL_FAIL;
1651 goto end_msg_sessiond;
1652 }
1653
1654 health_code_update();
1655
1656 if (!len) {
1657 /*
1658 * There is nothing to receive. We have simply
1659 * checked whether the channel can be found.
1660 */
1661 ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1662 goto end_msg_sessiond;
1663 }
1664
1665 /* Tell session daemon we are ready to receive the metadata. */
1666 ret = consumer_send_status_msg(sock, LTTCOMM_CONSUMERD_SUCCESS);
1667 if (ret < 0) {
1668 /* Somehow, the session daemon is not responding anymore. */
1669 goto error_fatal;
1670 }
1671
1672 health_code_update();
1673
1674 /* Wait for more data. */
1675 health_poll_entry();
1676 ret = lttng_consumer_poll_socket(consumer_sockpoll);
1677 health_poll_exit();
1678 if (ret) {
1679 goto error_fatal;
1680 }
1681
1682 health_code_update();
1683
1684 ret = lttng_ustconsumer_recv_metadata(sock, key, offset,
1685 len, version, channel, 0, 1);
1686 if (ret < 0) {
1687 /* error receiving from sessiond */
1688 goto error_fatal;
1689 } else {
1690 ret_code = ret;
1691 goto end_msg_sessiond;
1692 }
1693 }
1694 case LTTNG_CONSUMER_SETUP_METADATA:
1695 {
1696 int ret;
1697
1698 ret = setup_metadata(ctx, msg.u.setup_metadata.key);
1699 if (ret) {
1700 ret_code = ret;
1701 }
1702 goto end_msg_sessiond;
1703 }
1704 case LTTNG_CONSUMER_SNAPSHOT_CHANNEL:
1705 {
1706 if (msg.u.snapshot_channel.metadata) {
1707 ret = snapshot_metadata(msg.u.snapshot_channel.key,
1708 msg.u.snapshot_channel.pathname,
1709 msg.u.snapshot_channel.relayd_id,
1710 ctx);
1711 if (ret < 0) {
1712 ERR("Snapshot metadata failed");
1713 ret_code = LTTCOMM_CONSUMERD_ERROR_METADATA;
1714 }
1715 } else {
1716 ret = snapshot_channel(msg.u.snapshot_channel.key,
1717 msg.u.snapshot_channel.pathname,
1718 msg.u.snapshot_channel.relayd_id,
1719 msg.u.snapshot_channel.nb_packets_per_stream,
1720 ctx);
1721 if (ret < 0) {
1722 ERR("Snapshot channel failed");
1723 ret_code = LTTCOMM_CONSUMERD_CHANNEL_FAIL;
1724 }
1725 }
1726
1727 health_code_update();
1728 ret = consumer_send_status_msg(sock, ret_code);
1729 if (ret < 0) {
1730 /* Somehow, the session daemon is not responding anymore. */
1731 goto end_nosignal;
1732 }
1733 health_code_update();
1734 break;
1735 }
1736 case LTTNG_CONSUMER_DISCARDED_EVENTS:
1737 {
1738 int ret = 0;
1739 uint64_t discarded_events;
1740 struct lttng_ht_iter iter;
1741 struct lttng_ht *ht;
1742 struct lttng_consumer_stream *stream;
1743 uint64_t id = msg.u.discarded_events.session_id;
1744 uint64_t key = msg.u.discarded_events.channel_key;
1745
1746 DBG("UST consumer discarded events command for session id %"
1747 PRIu64, id);
1748 rcu_read_lock();
1749 pthread_mutex_lock(&consumer_data.lock);
1750
1751 ht = consumer_data.stream_list_ht;
1752
1753 /*
1754 * We only need a reference to the channel, but they are not
1755 * directly indexed, so we just use the first matching stream
1756 * to extract the information we need, we default to 0 if not
1757 * found (no events are dropped if the channel is not yet in
1758 * use).
1759 */
1760 discarded_events = 0;
1761 cds_lfht_for_each_entry_duplicate(ht->ht,
1762 ht->hash_fct(&id, lttng_ht_seed),
1763 ht->match_fct, &id,
1764 &iter.iter, stream, node_session_id.node) {
1765 if (stream->chan->key == key) {
1766 discarded_events = stream->chan->discarded_events;
1767 break;
1768 }
1769 }
1770 pthread_mutex_unlock(&consumer_data.lock);
1771 rcu_read_unlock();
1772
1773 DBG("UST consumer discarded events command for session id %"
1774 PRIu64 ", channel key %" PRIu64, id, key);
1775
1776 health_code_update();
1777
1778 /* Send back returned value to session daemon */
1779 ret = lttcomm_send_unix_sock(sock, &discarded_events, sizeof(discarded_events));
1780 if (ret < 0) {
1781 PERROR("send discarded events");
1782 goto error_fatal;
1783 }
1784
1785 break;
1786 }
1787 case LTTNG_CONSUMER_LOST_PACKETS:
1788 {
1789 int ret;
1790 uint64_t lost_packets;
1791 struct lttng_ht_iter iter;
1792 struct lttng_ht *ht;
1793 struct lttng_consumer_stream *stream;
1794 uint64_t id = msg.u.lost_packets.session_id;
1795 uint64_t key = msg.u.lost_packets.channel_key;
1796
1797 DBG("UST consumer lost packets command for session id %"
1798 PRIu64, id);
1799 rcu_read_lock();
1800 pthread_mutex_lock(&consumer_data.lock);
1801
1802 ht = consumer_data.stream_list_ht;
1803
1804 /*
1805 * We only need a reference to the channel, but they are not
1806 * directly indexed, so we just use the first matching stream
1807 * to extract the information we need, we default to 0 if not
1808 * found (no packets lost if the channel is not yet in use).
1809 */
1810 lost_packets = 0;
1811 cds_lfht_for_each_entry_duplicate(ht->ht,
1812 ht->hash_fct(&id, lttng_ht_seed),
1813 ht->match_fct, &id,
1814 &iter.iter, stream, node_session_id.node) {
1815 if (stream->chan->key == key) {
1816 lost_packets = stream->chan->lost_packets;
1817 break;
1818 }
1819 }
1820 pthread_mutex_unlock(&consumer_data.lock);
1821 rcu_read_unlock();
1822
1823 DBG("UST consumer lost packets command for session id %"
1824 PRIu64 ", channel key %" PRIu64, id, key);
1825
1826 health_code_update();
1827
1828 /* Send back returned value to session daemon */
1829 ret = lttcomm_send_unix_sock(sock, &lost_packets,
1830 sizeof(lost_packets));
1831 if (ret < 0) {
1832 PERROR("send lost packets");
1833 goto error_fatal;
1834 }
1835
1836 break;
1837 }
1838 default:
1839 break;
1840 }
1841
1842 end_nosignal:
1843 rcu_read_unlock();
1844
1845 health_code_update();
1846
1847 /*
1848 * Return 1 to indicate success since the 0 value can be a socket
1849 * shutdown during the recv() or send() call.
1850 */
1851 return 1;
1852
1853 end_msg_sessiond:
1854 /*
1855 * The returned value here is not useful since either way we'll return 1 to
1856 * the caller because the session daemon socket management is done
1857 * elsewhere. Returning a negative code or 0 will shutdown the consumer.
1858 */
1859 ret = consumer_send_status_msg(sock, ret_code);
1860 if (ret < 0) {
1861 goto error_fatal;
1862 }
1863 rcu_read_unlock();
1864
1865 health_code_update();
1866
1867 return 1;
1868 end_channel_error:
1869 if (channel) {
1870 /*
1871 * Free channel here since no one has a reference to it. We don't
1872 * free after that because a stream can store this pointer.
1873 */
1874 destroy_channel(channel);
1875 }
1876 /* We have to send a status channel message indicating an error. */
1877 ret = consumer_send_status_channel(sock, NULL);
1878 if (ret < 0) {
1879 /* Stop everything if session daemon can not be notified. */
1880 goto error_fatal;
1881 }
1882 rcu_read_unlock();
1883
1884 health_code_update();
1885
1886 return 1;
1887 error_fatal:
1888 rcu_read_unlock();
1889 /* This will issue a consumer stop. */
1890 return -1;
1891 }
1892
1893 /*
1894 * Wrapper over the mmap() read offset from ust-ctl library. Since this can be
1895 * compiled out, we isolate it in this library.
1896 */
1897 int lttng_ustctl_get_mmap_read_offset(struct lttng_consumer_stream *stream,
1898 unsigned long *off)
1899 {
1900 assert(stream);
1901 assert(stream->ustream);
1902
1903 return ustctl_get_mmap_read_offset(stream->ustream, off);
1904 }
1905
1906 /*
1907 * Wrapper over the mmap() read offset from ust-ctl library. Since this can be
1908 * compiled out, we isolate it in this library.
1909 */
1910 void *lttng_ustctl_get_mmap_base(struct lttng_consumer_stream *stream)
1911 {
1912 assert(stream);
1913 assert(stream->ustream);
1914
1915 return ustctl_get_mmap_base(stream->ustream);
1916 }
1917
1918 /*
1919 * Take a snapshot for a specific fd
1920 *
1921 * Returns 0 on success, < 0 on error
1922 */
1923 int lttng_ustconsumer_take_snapshot(struct lttng_consumer_stream *stream)
1924 {
1925 assert(stream);
1926 assert(stream->ustream);
1927
1928 return ustctl_snapshot(stream->ustream);
1929 }
1930
1931 /*
1932 * Get the produced position
1933 *
1934 * Returns 0 on success, < 0 on error
1935 */
1936 int lttng_ustconsumer_get_produced_snapshot(
1937 struct lttng_consumer_stream *stream, unsigned long *pos)
1938 {
1939 assert(stream);
1940 assert(stream->ustream);
1941 assert(pos);
1942
1943 return ustctl_snapshot_get_produced(stream->ustream, pos);
1944 }
1945
1946 /*
1947 * Get the consumed position
1948 *
1949 * Returns 0 on success, < 0 on error
1950 */
1951 int lttng_ustconsumer_get_consumed_snapshot(
1952 struct lttng_consumer_stream *stream, unsigned long *pos)
1953 {
1954 assert(stream);
1955 assert(stream->ustream);
1956 assert(pos);
1957
1958 return ustctl_snapshot_get_consumed(stream->ustream, pos);
1959 }
1960
1961 void lttng_ustconsumer_flush_buffer(struct lttng_consumer_stream *stream,
1962 int producer)
1963 {
1964 assert(stream);
1965 assert(stream->ustream);
1966
1967 ustctl_flush_buffer(stream->ustream, producer);
1968 }
1969
1970 int lttng_ustconsumer_get_current_timestamp(
1971 struct lttng_consumer_stream *stream, uint64_t *ts)
1972 {
1973 assert(stream);
1974 assert(stream->ustream);
1975 assert(ts);
1976
1977 return ustctl_get_current_timestamp(stream->ustream, ts);
1978 }
1979
1980 int lttng_ustconsumer_get_sequence_number(
1981 struct lttng_consumer_stream *stream, uint64_t *seq)
1982 {
1983 assert(stream);
1984 assert(stream->ustream);
1985 assert(seq);
1986
1987 return ustctl_get_sequence_number(stream->ustream, seq);
1988 }
1989
1990 /*
1991 * Called when the stream signals the consumer that it has hung up.
1992 */
1993 void lttng_ustconsumer_on_stream_hangup(struct lttng_consumer_stream *stream)
1994 {
1995 assert(stream);
1996 assert(stream->ustream);
1997
1998 pthread_mutex_lock(&stream->lock);
1999 if (!stream->quiescent) {
2000 ustctl_flush_buffer(stream->ustream, 0);
2001 stream->quiescent = true;
2002 }
2003 pthread_mutex_unlock(&stream->lock);
2004 stream->hangup_flush_done = 1;
2005 }
2006
2007 void lttng_ustconsumer_del_channel(struct lttng_consumer_channel *chan)
2008 {
2009 int i;
2010
2011 assert(chan);
2012 assert(chan->uchan);
2013
2014 if (chan->switch_timer_enabled == 1) {
2015 consumer_timer_switch_stop(chan);
2016 }
2017 for (i = 0; i < chan->nr_stream_fds; i++) {
2018 int ret;
2019
2020 ret = close(chan->stream_fds[i]);
2021 if (ret) {
2022 PERROR("close");
2023 }
2024 if (chan->shm_path[0]) {
2025 char shm_path[PATH_MAX];
2026
2027 ret = get_stream_shm_path(shm_path, chan->shm_path, i);
2028 if (ret) {
2029 ERR("Cannot get stream shm path");
2030 }
2031 ret = run_as_unlink(shm_path, chan->uid, chan->gid);
2032 if (ret) {
2033 PERROR("unlink %s", shm_path);
2034 }
2035 }
2036 }
2037 }
2038
2039 void lttng_ustconsumer_free_channel(struct lttng_consumer_channel *chan)
2040 {
2041 assert(chan);
2042 assert(chan->uchan);
2043
2044 consumer_metadata_cache_destroy(chan);
2045 ustctl_destroy_channel(chan->uchan);
2046 /* Try to rmdir all directories under shm_path root. */
2047 if (chan->root_shm_path[0]) {
2048 (void) run_as_recursive_rmdir(chan->root_shm_path,
2049 chan->uid, chan->gid);
2050 }
2051 free(chan->stream_fds);
2052 }
2053
2054 void lttng_ustconsumer_del_stream(struct lttng_consumer_stream *stream)
2055 {
2056 assert(stream);
2057 assert(stream->ustream);
2058
2059 if (stream->chan->switch_timer_enabled == 1) {
2060 consumer_timer_switch_stop(stream->chan);
2061 }
2062 ustctl_destroy_stream(stream->ustream);
2063 }
2064
2065 int lttng_ustconsumer_get_wakeup_fd(struct lttng_consumer_stream *stream)
2066 {
2067 assert(stream);
2068 assert(stream->ustream);
2069
2070 return ustctl_stream_get_wakeup_fd(stream->ustream);
2071 }
2072
2073 int lttng_ustconsumer_close_wakeup_fd(struct lttng_consumer_stream *stream)
2074 {
2075 assert(stream);
2076 assert(stream->ustream);
2077
2078 return ustctl_stream_close_wakeup_fd(stream->ustream);
2079 }
2080
2081 /*
2082 * Populate index values of a UST stream. Values are set in big endian order.
2083 *
2084 * Return 0 on success or else a negative value.
2085 */
2086 static int get_index_values(struct ctf_packet_index *index,
2087 struct ustctl_consumer_stream *ustream)
2088 {
2089 int ret;
2090
2091 ret = ustctl_get_timestamp_begin(ustream, &index->timestamp_begin);
2092 if (ret < 0) {
2093 PERROR("ustctl_get_timestamp_begin");
2094 goto error;
2095 }
2096 index->timestamp_begin = htobe64(index->timestamp_begin);
2097
2098 ret = ustctl_get_timestamp_end(ustream, &index->timestamp_end);
2099 if (ret < 0) {
2100 PERROR("ustctl_get_timestamp_end");
2101 goto error;
2102 }
2103 index->timestamp_end = htobe64(index->timestamp_end);
2104
2105 ret = ustctl_get_events_discarded(ustream, &index->events_discarded);
2106 if (ret < 0) {
2107 PERROR("ustctl_get_events_discarded");
2108 goto error;
2109 }
2110 index->events_discarded = htobe64(index->events_discarded);
2111
2112 ret = ustctl_get_content_size(ustream, &index->content_size);
2113 if (ret < 0) {
2114 PERROR("ustctl_get_content_size");
2115 goto error;
2116 }
2117 index->content_size = htobe64(index->content_size);
2118
2119 ret = ustctl_get_packet_size(ustream, &index->packet_size);
2120 if (ret < 0) {
2121 PERROR("ustctl_get_packet_size");
2122 goto error;
2123 }
2124 index->packet_size = htobe64(index->packet_size);
2125
2126 ret = ustctl_get_stream_id(ustream, &index->stream_id);
2127 if (ret < 0) {
2128 PERROR("ustctl_get_stream_id");
2129 goto error;
2130 }
2131 index->stream_id = htobe64(index->stream_id);
2132
2133 ret = ustctl_get_instance_id(ustream, &index->stream_instance_id);
2134 if (ret < 0) {
2135 PERROR("ustctl_get_instance_id");
2136 goto error;
2137 }
2138 index->stream_instance_id = htobe64(index->stream_instance_id);
2139
2140 ret = ustctl_get_sequence_number(ustream, &index->packet_seq_num);
2141 if (ret < 0) {
2142 PERROR("ustctl_get_sequence_number");
2143 goto error;
2144 }
2145 index->packet_seq_num = htobe64(index->packet_seq_num);
2146
2147 error:
2148 return ret;
2149 }
2150
2151 static
2152 void metadata_stream_reset_cache(struct lttng_consumer_stream *stream,
2153 struct consumer_metadata_cache *cache)
2154 {
2155 DBG("Metadata stream update to version %" PRIu64,
2156 cache->version);
2157 stream->ust_metadata_pushed = 0;
2158 stream->metadata_version = cache->version;
2159 stream->reset_metadata_flag = 1;
2160 }
2161
2162 /*
2163 * Check if the version of the metadata stream and metadata cache match.
2164 * If the cache got updated, reset the metadata stream.
2165 * The stream lock and metadata cache lock MUST be held.
2166 * Return 0 on success, a negative value on error.
2167 */
2168 static
2169 int metadata_stream_check_version(struct lttng_consumer_stream *stream)
2170 {
2171 int ret = 0;
2172 struct consumer_metadata_cache *cache = stream->chan->metadata_cache;
2173
2174 if (cache->version == stream->metadata_version) {
2175 goto end;
2176 }
2177 metadata_stream_reset_cache(stream, cache);
2178
2179 end:
2180 return ret;
2181 }
2182
2183 /*
2184 * Write up to one packet from the metadata cache to the channel.
2185 *
2186 * Returns the number of bytes pushed in the cache, or a negative value
2187 * on error.
2188 */
2189 static
2190 int commit_one_metadata_packet(struct lttng_consumer_stream *stream)
2191 {
2192 ssize_t write_len;
2193 int ret;
2194
2195 pthread_mutex_lock(&stream->chan->metadata_cache->lock);
2196 ret = metadata_stream_check_version(stream);
2197 if (ret < 0) {
2198 goto end;
2199 }
2200 if (stream->chan->metadata_cache->max_offset
2201 == stream->ust_metadata_pushed) {
2202 ret = 0;
2203 goto end;
2204 }
2205
2206 write_len = ustctl_write_one_packet_to_channel(stream->chan->uchan,
2207 &stream->chan->metadata_cache->data[stream->ust_metadata_pushed],
2208 stream->chan->metadata_cache->max_offset
2209 - stream->ust_metadata_pushed);
2210 assert(write_len != 0);
2211 if (write_len < 0) {
2212 ERR("Writing one metadata packet");
2213 ret = -1;
2214 goto end;
2215 }
2216 stream->ust_metadata_pushed += write_len;
2217
2218 assert(stream->chan->metadata_cache->max_offset >=
2219 stream->ust_metadata_pushed);
2220 ret = write_len;
2221
2222 end:
2223 pthread_mutex_unlock(&stream->chan->metadata_cache->lock);
2224 return ret;
2225 }
2226
2227
2228 /*
2229 * Sync metadata meaning request them to the session daemon and snapshot to the
2230 * metadata thread can consumer them.
2231 *
2232 * Metadata stream lock is held here, but we need to release it when
2233 * interacting with sessiond, else we cause a deadlock with live
2234 * awaiting on metadata to be pushed out.
2235 *
2236 * Return 0 if new metadatda is available, EAGAIN if the metadata stream
2237 * is empty or a negative value on error.
2238 */
2239 int lttng_ustconsumer_sync_metadata(struct lttng_consumer_local_data *ctx,
2240 struct lttng_consumer_stream *metadata)
2241 {
2242 int ret;
2243 int retry = 0;
2244
2245 assert(ctx);
2246 assert(metadata);
2247
2248 pthread_mutex_unlock(&metadata->lock);
2249 /*
2250 * Request metadata from the sessiond, but don't wait for the flush
2251 * because we locked the metadata thread.
2252 */
2253 ret = lttng_ustconsumer_request_metadata(ctx, metadata->chan, 0, 0);
2254 pthread_mutex_lock(&metadata->lock);
2255 if (ret < 0) {
2256 goto end;
2257 }
2258
2259 ret = commit_one_metadata_packet(metadata);
2260 if (ret <= 0) {
2261 goto end;
2262 } else if (ret > 0) {
2263 retry = 1;
2264 }
2265
2266 ustctl_flush_buffer(metadata->ustream, 1);
2267 ret = ustctl_snapshot(metadata->ustream);
2268 if (ret < 0) {
2269 if (errno != EAGAIN) {
2270 ERR("Sync metadata, taking UST snapshot");
2271 goto end;
2272 }
2273 DBG("No new metadata when syncing them.");
2274 /* No new metadata, exit. */
2275 ret = ENODATA;
2276 goto end;
2277 }
2278
2279 /*
2280 * After this flush, we still need to extract metadata.
2281 */
2282 if (retry) {
2283 ret = EAGAIN;
2284 }
2285
2286 end:
2287 return ret;
2288 }
2289
2290 /*
2291 * Return 0 on success else a negative value.
2292 */
2293 static int notify_if_more_data(struct lttng_consumer_stream *stream,
2294 struct lttng_consumer_local_data *ctx)
2295 {
2296 int ret;
2297 struct ustctl_consumer_stream *ustream;
2298
2299 assert(stream);
2300 assert(ctx);
2301
2302 ustream = stream->ustream;
2303
2304 /*
2305 * First, we are going to check if there is a new subbuffer available
2306 * before reading the stream wait_fd.
2307 */
2308 /* Get the next subbuffer */
2309 ret = ustctl_get_next_subbuf(ustream);
2310 if (ret) {
2311 /* No more data found, flag the stream. */
2312 stream->has_data = 0;
2313 ret = 0;
2314 goto end;
2315 }
2316
2317 ret = ustctl_put_subbuf(ustream);
2318 assert(!ret);
2319
2320 /* This stream still has data. Flag it and wake up the data thread. */
2321 stream->has_data = 1;
2322
2323 if (stream->monitor && !stream->hangup_flush_done && !ctx->has_wakeup) {
2324 ssize_t writelen;
2325
2326 writelen = lttng_pipe_write(ctx->consumer_wakeup_pipe, "!", 1);
2327 if (writelen < 0 && errno != EAGAIN && errno != EWOULDBLOCK) {
2328 ret = writelen;
2329 goto end;
2330 }
2331
2332 /* The wake up pipe has been notified. */
2333 ctx->has_wakeup = 1;
2334 }
2335 ret = 0;
2336
2337 end:
2338 return ret;
2339 }
2340
2341 static
2342 int update_stream_stats(struct lttng_consumer_stream *stream)
2343 {
2344 int ret;
2345 uint64_t seq, discarded;
2346
2347 ret = ustctl_get_sequence_number(stream->ustream, &seq);
2348 if (ret < 0) {
2349 PERROR("ustctl_get_sequence_number");
2350 goto end;
2351 }
2352 /*
2353 * Start the sequence when we extract the first packet in case we don't
2354 * start at 0 (for example if a consumer is not connected to the
2355 * session immediately after the beginning).
2356 */
2357 if (stream->last_sequence_number == -1ULL) {
2358 stream->last_sequence_number = seq;
2359 } else if (seq > stream->last_sequence_number) {
2360 stream->chan->lost_packets += seq -
2361 stream->last_sequence_number - 1;
2362 } else {
2363 /* seq <= last_sequence_number */
2364 ERR("Sequence number inconsistent : prev = %" PRIu64
2365 ", current = %" PRIu64,
2366 stream->last_sequence_number, seq);
2367 ret = -1;
2368 goto end;
2369 }
2370 stream->last_sequence_number = seq;
2371
2372 ret = ustctl_get_events_discarded(stream->ustream, &discarded);
2373 if (ret < 0) {
2374 PERROR("kernctl_get_events_discarded");
2375 goto end;
2376 }
2377 if (discarded < stream->last_discarded_events) {
2378 /*
2379 * Overflow has occurred. We assume only one wrap-around
2380 * has occurred.
2381 */
2382 stream->chan->discarded_events +=
2383 (1ULL << (CAA_BITS_PER_LONG - 1)) -
2384 stream->last_discarded_events + discarded;
2385 } else {
2386 stream->chan->discarded_events += discarded -
2387 stream->last_discarded_events;
2388 }
2389 stream->last_discarded_events = discarded;
2390 ret = 0;
2391
2392 end:
2393 return ret;
2394 }
2395
2396 /*
2397 * Read subbuffer from the given stream.
2398 *
2399 * Stream lock MUST be acquired.
2400 *
2401 * Return 0 on success else a negative value.
2402 */
2403 int lttng_ustconsumer_read_subbuffer(struct lttng_consumer_stream *stream,
2404 struct lttng_consumer_local_data *ctx)
2405 {
2406 unsigned long len, subbuf_size, padding;
2407 int err, write_index = 1;
2408 long ret = 0;
2409 struct ustctl_consumer_stream *ustream;
2410 struct ctf_packet_index index;
2411
2412 assert(stream);
2413 assert(stream->ustream);
2414 assert(ctx);
2415
2416 DBG("In UST read_subbuffer (wait_fd: %d, name: %s)", stream->wait_fd,
2417 stream->name);
2418
2419 /* Ease our life for what's next. */
2420 ustream = stream->ustream;
2421
2422 /*
2423 * We can consume the 1 byte written into the wait_fd by UST. Don't trigger
2424 * error if we cannot read this one byte (read returns 0), or if the error
2425 * is EAGAIN or EWOULDBLOCK.
2426 *
2427 * This is only done when the stream is monitored by a thread, before the
2428 * flush is done after a hangup and if the stream is not flagged with data
2429 * since there might be nothing to consume in the wait fd but still have
2430 * data available flagged by the consumer wake up pipe.
2431 */
2432 if (stream->monitor && !stream->hangup_flush_done && !stream->has_data) {
2433 char dummy;
2434 ssize_t readlen;
2435
2436 readlen = lttng_read(stream->wait_fd, &dummy, 1);
2437 if (readlen < 0 && errno != EAGAIN && errno != EWOULDBLOCK) {
2438 ret = readlen;
2439 goto end;
2440 }
2441 }
2442
2443 retry:
2444 /* Get the next subbuffer */
2445 err = ustctl_get_next_subbuf(ustream);
2446 if (err != 0) {
2447 /*
2448 * Populate metadata info if the existing info has
2449 * already been read.
2450 */
2451 if (stream->metadata_flag) {
2452 ret = commit_one_metadata_packet(stream);
2453 if (ret <= 0) {
2454 goto end;
2455 }
2456 ustctl_flush_buffer(stream->ustream, 1);
2457 goto retry;
2458 }
2459
2460 ret = err; /* ustctl_get_next_subbuf returns negative, caller expect positive. */
2461 /*
2462 * This is a debug message even for single-threaded consumer,
2463 * because poll() have more relaxed criterions than get subbuf,
2464 * so get_subbuf may fail for short race windows where poll()
2465 * would issue wakeups.
2466 */
2467 DBG("Reserving sub buffer failed (everything is normal, "
2468 "it is due to concurrency) [ret: %d]", err);
2469 goto end;
2470 }
2471 assert(stream->chan->output == CONSUMER_CHANNEL_MMAP);
2472
2473 if (!stream->metadata_flag) {
2474 index.offset = htobe64(stream->out_fd_offset);
2475 ret = get_index_values(&index, ustream);
2476 if (ret < 0) {
2477 err = ustctl_put_subbuf(ustream);
2478 assert(err == 0);
2479 goto end;
2480 }
2481
2482 /* Update the stream's sequence and discarded events count. */
2483 ret = update_stream_stats(stream);
2484 if (ret < 0) {
2485 PERROR("kernctl_get_events_discarded");
2486 err = ustctl_put_subbuf(ustream);
2487 assert(err == 0);
2488 goto end;
2489 }
2490 } else {
2491 write_index = 0;
2492 }
2493
2494 /* Get the full padded subbuffer size */
2495 err = ustctl_get_padded_subbuf_size(ustream, &len);
2496 assert(err == 0);
2497
2498 /* Get subbuffer data size (without padding) */
2499 err = ustctl_get_subbuf_size(ustream, &subbuf_size);
2500 assert(err == 0);
2501
2502 /* Make sure we don't get a subbuffer size bigger than the padded */
2503 assert(len >= subbuf_size);
2504
2505 padding = len - subbuf_size;
2506 /* write the subbuffer to the tracefile */
2507 ret = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, subbuf_size, padding, &index);
2508 /*
2509 * The mmap operation should write subbuf_size amount of data when network
2510 * streaming or the full padding (len) size when we are _not_ streaming.
2511 */
2512 if ((ret != subbuf_size && stream->net_seq_idx != (uint64_t) -1ULL) ||
2513 (ret != len && stream->net_seq_idx == (uint64_t) -1ULL)) {
2514 /*
2515 * Display the error but continue processing to try to release the
2516 * subbuffer. This is a DBG statement since any unexpected kill or
2517 * signal, the application gets unregistered, relayd gets closed or
2518 * anything that affects the buffer lifetime will trigger this error.
2519 * So, for the sake of the user, don't print this error since it can
2520 * happen and it is OK with the code flow.
2521 */
2522 DBG("Error writing to tracefile "
2523 "(ret: %ld != len: %lu != subbuf_size: %lu)",
2524 ret, len, subbuf_size);
2525 write_index = 0;
2526 }
2527 err = ustctl_put_next_subbuf(ustream);
2528 assert(err == 0);
2529
2530 /*
2531 * This will consumer the byte on the wait_fd if and only if there is not
2532 * next subbuffer to be acquired.
2533 */
2534 if (!stream->metadata_flag) {
2535 ret = notify_if_more_data(stream, ctx);
2536 if (ret < 0) {
2537 goto end;
2538 }
2539 }
2540
2541 /* Write index if needed. */
2542 if (!write_index) {
2543 goto end;
2544 }
2545
2546 if (stream->chan->live_timer_interval && !stream->metadata_flag) {
2547 /*
2548 * In live, block until all the metadata is sent.
2549 */
2550 pthread_mutex_lock(&stream->metadata_timer_lock);
2551 assert(!stream->missed_metadata_flush);
2552 stream->waiting_on_metadata = true;
2553 pthread_mutex_unlock(&stream->metadata_timer_lock);
2554
2555 err = consumer_stream_sync_metadata(ctx, stream->session_id);
2556
2557 pthread_mutex_lock(&stream->metadata_timer_lock);
2558 stream->waiting_on_metadata = false;
2559 if (stream->missed_metadata_flush) {
2560 stream->missed_metadata_flush = false;
2561 pthread_mutex_unlock(&stream->metadata_timer_lock);
2562 (void) consumer_flush_ust_index(stream);
2563 } else {
2564 pthread_mutex_unlock(&stream->metadata_timer_lock);
2565 }
2566
2567 if (err < 0) {
2568 goto end;
2569 }
2570 }
2571
2572 assert(!stream->metadata_flag);
2573 err = consumer_stream_write_index(stream, &index);
2574 if (err < 0) {
2575 goto end;
2576 }
2577
2578 end:
2579 return ret;
2580 }
2581
2582 /*
2583 * Called when a stream is created.
2584 *
2585 * Return 0 on success or else a negative value.
2586 */
2587 int lttng_ustconsumer_on_recv_stream(struct lttng_consumer_stream *stream)
2588 {
2589 int ret;
2590
2591 assert(stream);
2592
2593 /* Don't create anything if this is set for streaming. */
2594 if (stream->net_seq_idx == (uint64_t) -1ULL && stream->chan->monitor) {
2595 ret = utils_create_stream_file(stream->chan->pathname, stream->name,
2596 stream->chan->tracefile_size, stream->tracefile_count_current,
2597 stream->uid, stream->gid, NULL);
2598 if (ret < 0) {
2599 goto error;
2600 }
2601 stream->out_fd = ret;
2602 stream->tracefile_size_current = 0;
2603
2604 if (!stream->metadata_flag) {
2605 struct lttng_index_file *index_file;
2606
2607 index_file = lttng_index_file_create(stream->chan->pathname,
2608 stream->name, stream->uid, stream->gid,
2609 stream->chan->tracefile_size,
2610 stream->tracefile_count_current,
2611 CTF_INDEX_MAJOR, CTF_INDEX_MINOR);
2612 if (!index_file) {
2613 goto error;
2614 }
2615 stream->index_file = index_file;
2616 }
2617 }
2618 ret = 0;
2619
2620 error:
2621 return ret;
2622 }
2623
2624 /*
2625 * Check if data is still being extracted from the buffers for a specific
2626 * stream. Consumer data lock MUST be acquired before calling this function
2627 * and the stream lock.
2628 *
2629 * Return 1 if the traced data are still getting read else 0 meaning that the
2630 * data is available for trace viewer reading.
2631 */
2632 int lttng_ustconsumer_data_pending(struct lttng_consumer_stream *stream)
2633 {
2634 int ret;
2635
2636 assert(stream);
2637 assert(stream->ustream);
2638
2639 DBG("UST consumer checking data pending");
2640
2641 if (stream->endpoint_status != CONSUMER_ENDPOINT_ACTIVE) {
2642 ret = 0;
2643 goto end;
2644 }
2645
2646 if (stream->chan->type == CONSUMER_CHANNEL_TYPE_METADATA) {
2647 uint64_t contiguous, pushed;
2648
2649 /* Ease our life a bit. */
2650 contiguous = stream->chan->metadata_cache->max_offset;
2651 pushed = stream->ust_metadata_pushed;
2652
2653 /*
2654 * We can simply check whether all contiguously available data
2655 * has been pushed to the ring buffer, since the push operation
2656 * is performed within get_next_subbuf(), and because both
2657 * get_next_subbuf() and put_next_subbuf() are issued atomically
2658 * thanks to the stream lock within
2659 * lttng_ustconsumer_read_subbuffer(). This basically means that
2660 * whetnever ust_metadata_pushed is incremented, the associated
2661 * metadata has been consumed from the metadata stream.
2662 */
2663 DBG("UST consumer metadata pending check: contiguous %" PRIu64 " vs pushed %" PRIu64,
2664 contiguous, pushed);
2665 assert(((int64_t) (contiguous - pushed)) >= 0);
2666 if ((contiguous != pushed) ||
2667 (((int64_t) contiguous - pushed) > 0 || contiguous == 0)) {
2668 ret = 1; /* Data is pending */
2669 goto end;
2670 }
2671 } else {
2672 ret = ustctl_get_next_subbuf(stream->ustream);
2673 if (ret == 0) {
2674 /*
2675 * There is still data so let's put back this
2676 * subbuffer.
2677 */
2678 ret = ustctl_put_subbuf(stream->ustream);
2679 assert(ret == 0);
2680 ret = 1; /* Data is pending */
2681 goto end;
2682 }
2683 }
2684
2685 /* Data is NOT pending so ready to be read. */
2686 ret = 0;
2687
2688 end:
2689 return ret;
2690 }
2691
2692 /*
2693 * Stop a given metadata channel timer if enabled and close the wait fd which
2694 * is the poll pipe of the metadata stream.
2695 *
2696 * This MUST be called with the metadata channel acquired.
2697 */
2698 void lttng_ustconsumer_close_metadata(struct lttng_consumer_channel *metadata)
2699 {
2700 int ret;
2701
2702 assert(metadata);
2703 assert(metadata->type == CONSUMER_CHANNEL_TYPE_METADATA);
2704
2705 DBG("Closing metadata channel key %" PRIu64, metadata->key);
2706
2707 if (metadata->switch_timer_enabled == 1) {
2708 consumer_timer_switch_stop(metadata);
2709 }
2710
2711 if (!metadata->metadata_stream) {
2712 goto end;
2713 }
2714
2715 /*
2716 * Closing write side so the thread monitoring the stream wakes up if any
2717 * and clean the metadata stream.
2718 */
2719 if (metadata->metadata_stream->ust_metadata_poll_pipe[1] >= 0) {
2720 ret = close(metadata->metadata_stream->ust_metadata_poll_pipe[1]);
2721 if (ret < 0) {
2722 PERROR("closing metadata pipe write side");
2723 }
2724 metadata->metadata_stream->ust_metadata_poll_pipe[1] = -1;
2725 }
2726
2727 end:
2728 return;
2729 }
2730
2731 /*
2732 * Close every metadata stream wait fd of the metadata hash table. This
2733 * function MUST be used very carefully so not to run into a race between the
2734 * metadata thread handling streams and this function closing their wait fd.
2735 *
2736 * For UST, this is used when the session daemon hangs up. Its the metadata
2737 * producer so calling this is safe because we are assured that no state change
2738 * can occur in the metadata thread for the streams in the hash table.
2739 */
2740 void lttng_ustconsumer_close_all_metadata(struct lttng_ht *metadata_ht)
2741 {
2742 struct lttng_ht_iter iter;
2743 struct lttng_consumer_stream *stream;
2744
2745 assert(metadata_ht);
2746 assert(metadata_ht->ht);
2747
2748 DBG("UST consumer closing all metadata streams");
2749
2750 rcu_read_lock();
2751 cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream,
2752 node.node) {
2753
2754 health_code_update();
2755
2756 pthread_mutex_lock(&stream->chan->lock);
2757 lttng_ustconsumer_close_metadata(stream->chan);
2758 pthread_mutex_unlock(&stream->chan->lock);
2759
2760 }
2761 rcu_read_unlock();
2762 }
2763
2764 void lttng_ustconsumer_close_stream_wakeup(struct lttng_consumer_stream *stream)
2765 {
2766 int ret;
2767
2768 ret = ustctl_stream_close_wakeup_fd(stream->ustream);
2769 if (ret < 0) {
2770 ERR("Unable to close wakeup fd");
2771 }
2772 }
2773
2774 /*
2775 * Please refer to consumer-timer.c before adding any lock within this
2776 * function or any of its callees. Timers have a very strict locking
2777 * semantic with respect to teardown. Failure to respect this semantic
2778 * introduces deadlocks.
2779 *
2780 * DON'T hold the metadata lock when calling this function, else this
2781 * can cause deadlock involving consumer awaiting for metadata to be
2782 * pushed out due to concurrent interaction with the session daemon.
2783 */
2784 int lttng_ustconsumer_request_metadata(struct lttng_consumer_local_data *ctx,
2785 struct lttng_consumer_channel *channel, int timer, int wait)
2786 {
2787 struct lttcomm_metadata_request_msg request;
2788 struct lttcomm_consumer_msg msg;
2789 enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS;
2790 uint64_t len, key, offset, version;
2791 int ret;
2792
2793 assert(channel);
2794 assert(channel->metadata_cache);
2795
2796 memset(&request, 0, sizeof(request));
2797
2798 /* send the metadata request to sessiond */
2799 switch (consumer_data.type) {
2800 case LTTNG_CONSUMER64_UST:
2801 request.bits_per_long = 64;
2802 break;
2803 case LTTNG_CONSUMER32_UST:
2804 request.bits_per_long = 32;
2805 break;
2806 default:
2807 request.bits_per_long = 0;
2808 break;
2809 }
2810
2811 request.session_id = channel->session_id;
2812 request.session_id_per_pid = channel->session_id_per_pid;
2813 /*
2814 * Request the application UID here so the metadata of that application can
2815 * be sent back. The channel UID corresponds to the user UID of the session
2816 * used for the rights on the stream file(s).
2817 */
2818 request.uid = channel->ust_app_uid;
2819 request.key = channel->key;
2820
2821 DBG("Sending metadata request to sessiond, session id %" PRIu64
2822 ", per-pid %" PRIu64 ", app UID %u and channek key %" PRIu64,
2823 request.session_id, request.session_id_per_pid, request.uid,
2824 request.key);
2825
2826 pthread_mutex_lock(&ctx->metadata_socket_lock);
2827
2828 health_code_update();
2829
2830 ret = lttcomm_send_unix_sock(ctx->consumer_metadata_socket, &request,
2831 sizeof(request));
2832 if (ret < 0) {
2833 ERR("Asking metadata to sessiond");
2834 goto end;
2835 }
2836
2837 health_code_update();
2838
2839 /* Receive the metadata from sessiond */
2840 ret = lttcomm_recv_unix_sock(ctx->consumer_metadata_socket, &msg,
2841 sizeof(msg));
2842 if (ret != sizeof(msg)) {
2843 DBG("Consumer received unexpected message size %d (expects %zu)",
2844 ret, sizeof(msg));
2845 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD);
2846 /*
2847 * The ret value might 0 meaning an orderly shutdown but this is ok
2848 * since the caller handles this.
2849 */
2850 goto end;
2851 }
2852
2853 health_code_update();
2854
2855 if (msg.cmd_type == LTTNG_ERR_UND) {
2856 /* No registry found */
2857 (void) consumer_send_status_msg(ctx->consumer_metadata_socket,
2858 ret_code);
2859 ret = 0;
2860 goto end;
2861 } else if (msg.cmd_type != LTTNG_CONSUMER_PUSH_METADATA) {
2862 ERR("Unexpected cmd_type received %d", msg.cmd_type);
2863 ret = -1;
2864 goto end;
2865 }
2866
2867 len = msg.u.push_metadata.len;
2868 key = msg.u.push_metadata.key;
2869 offset = msg.u.push_metadata.target_offset;
2870 version = msg.u.push_metadata.version;
2871
2872 assert(key == channel->key);
2873 if (len == 0) {
2874 DBG("No new metadata to receive for key %" PRIu64, key);
2875 }
2876
2877 health_code_update();
2878
2879 /* Tell session daemon we are ready to receive the metadata. */
2880 ret = consumer_send_status_msg(ctx->consumer_metadata_socket,
2881 LTTCOMM_CONSUMERD_SUCCESS);
2882 if (ret < 0 || len == 0) {
2883 /*
2884 * Somehow, the session daemon is not responding anymore or there is
2885 * nothing to receive.
2886 */
2887 goto end;
2888 }
2889
2890 health_code_update();
2891
2892 ret = lttng_ustconsumer_recv_metadata(ctx->consumer_metadata_socket,
2893 key, offset, len, version, channel, timer, wait);
2894 if (ret >= 0) {
2895 /*
2896 * Only send the status msg if the sessiond is alive meaning a positive
2897 * ret code.
2898 */
2899 (void) consumer_send_status_msg(ctx->consumer_metadata_socket, ret);
2900 }
2901 ret = 0;
2902
2903 end:
2904 health_code_update();
2905
2906 pthread_mutex_unlock(&ctx->metadata_socket_lock);
2907 return ret;
2908 }
2909
2910 /*
2911 * Return the ustctl call for the get stream id.
2912 */
2913 int lttng_ustconsumer_get_stream_id(struct lttng_consumer_stream *stream,
2914 uint64_t *stream_id)
2915 {
2916 assert(stream);
2917 assert(stream_id);
2918
2919 return ustctl_get_stream_id(stream->ustream, stream_id);
2920 }
This page took 0.092851 seconds and 4 git commands to generate.