Account the lost packets in snapshot mode
[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, struct lttng_consumer_channel *channel,
1201 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, metadata_str);
1229 if (ret < 0) {
1230 /* Unable to handle metadata. Notify session daemon. */
1231 ret_code = LTTCOMM_CONSUMERD_ERROR_METADATA;
1232 /*
1233 * Skip metadata flush on write error since the offset and len might
1234 * not have been updated which could create an infinite loop below when
1235 * waiting for the metadata cache to be flushed.
1236 */
1237 pthread_mutex_unlock(&channel->metadata_cache->lock);
1238 goto end_free;
1239 }
1240 pthread_mutex_unlock(&channel->metadata_cache->lock);
1241
1242 if (!wait) {
1243 goto end_free;
1244 }
1245 while (consumer_metadata_cache_flushed(channel, offset + len, timer)) {
1246 DBG("Waiting for metadata to be flushed");
1247
1248 health_code_update();
1249
1250 usleep(DEFAULT_METADATA_AVAILABILITY_WAIT_TIME);
1251 }
1252
1253 end_free:
1254 free(metadata_str);
1255 end:
1256 return ret_code;
1257 }
1258
1259 /*
1260 * Receive command from session daemon and process it.
1261 *
1262 * Return 1 on success else a negative value or 0.
1263 */
1264 int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
1265 int sock, struct pollfd *consumer_sockpoll)
1266 {
1267 ssize_t ret;
1268 enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1269 struct lttcomm_consumer_msg msg;
1270 struct lttng_consumer_channel *channel = NULL;
1271
1272 health_code_update();
1273
1274 ret = lttcomm_recv_unix_sock(sock, &msg, sizeof(msg));
1275 if (ret != sizeof(msg)) {
1276 DBG("Consumer received unexpected message size %zd (expects %zu)",
1277 ret, sizeof(msg));
1278 /*
1279 * The ret value might 0 meaning an orderly shutdown but this is ok
1280 * since the caller handles this.
1281 */
1282 if (ret > 0) {
1283 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD);
1284 ret = -1;
1285 }
1286 return ret;
1287 }
1288
1289 health_code_update();
1290
1291 /* deprecated */
1292 assert(msg.cmd_type != LTTNG_CONSUMER_STOP);
1293
1294 health_code_update();
1295
1296 /* relayd needs RCU read-side lock */
1297 rcu_read_lock();
1298
1299 switch (msg.cmd_type) {
1300 case LTTNG_CONSUMER_ADD_RELAYD_SOCKET:
1301 {
1302 /* Session daemon status message are handled in the following call. */
1303 ret = consumer_add_relayd_socket(msg.u.relayd_sock.net_index,
1304 msg.u.relayd_sock.type, ctx, sock, consumer_sockpoll,
1305 &msg.u.relayd_sock.sock, msg.u.relayd_sock.session_id,
1306 msg.u.relayd_sock.relayd_session_id);
1307 goto end_nosignal;
1308 }
1309 case LTTNG_CONSUMER_DESTROY_RELAYD:
1310 {
1311 uint64_t index = msg.u.destroy_relayd.net_seq_idx;
1312 struct consumer_relayd_sock_pair *relayd;
1313
1314 DBG("UST consumer destroying relayd %" PRIu64, index);
1315
1316 /* Get relayd reference if exists. */
1317 relayd = consumer_find_relayd(index);
1318 if (relayd == NULL) {
1319 DBG("Unable to find relayd %" PRIu64, index);
1320 ret_code = LTTCOMM_CONSUMERD_RELAYD_FAIL;
1321 }
1322
1323 /*
1324 * Each relayd socket pair has a refcount of stream attached to it
1325 * which tells if the relayd is still active or not depending on the
1326 * refcount value.
1327 *
1328 * This will set the destroy flag of the relayd object and destroy it
1329 * if the refcount reaches zero when called.
1330 *
1331 * The destroy can happen either here or when a stream fd hangs up.
1332 */
1333 if (relayd) {
1334 consumer_flag_relayd_for_destroy(relayd);
1335 }
1336
1337 goto end_msg_sessiond;
1338 }
1339 case LTTNG_CONSUMER_UPDATE_STREAM:
1340 {
1341 rcu_read_unlock();
1342 return -ENOSYS;
1343 }
1344 case LTTNG_CONSUMER_DATA_PENDING:
1345 {
1346 int ret, is_data_pending;
1347 uint64_t id = msg.u.data_pending.session_id;
1348
1349 DBG("UST consumer data pending command for id %" PRIu64, id);
1350
1351 is_data_pending = consumer_data_pending(id);
1352
1353 /* Send back returned value to session daemon */
1354 ret = lttcomm_send_unix_sock(sock, &is_data_pending,
1355 sizeof(is_data_pending));
1356 if (ret < 0) {
1357 DBG("Error when sending the data pending ret code: %d", ret);
1358 goto error_fatal;
1359 }
1360
1361 /*
1362 * No need to send back a status message since the data pending
1363 * returned value is the response.
1364 */
1365 break;
1366 }
1367 case LTTNG_CONSUMER_ASK_CHANNEL_CREATION:
1368 {
1369 int ret;
1370 struct ustctl_consumer_channel_attr attr;
1371
1372 /* Create a plain object and reserve a channel key. */
1373 channel = allocate_channel(msg.u.ask_channel.session_id,
1374 msg.u.ask_channel.pathname, msg.u.ask_channel.name,
1375 msg.u.ask_channel.uid, msg.u.ask_channel.gid,
1376 msg.u.ask_channel.relayd_id, msg.u.ask_channel.key,
1377 (enum lttng_event_output) msg.u.ask_channel.output,
1378 msg.u.ask_channel.tracefile_size,
1379 msg.u.ask_channel.tracefile_count,
1380 msg.u.ask_channel.session_id_per_pid,
1381 msg.u.ask_channel.monitor,
1382 msg.u.ask_channel.live_timer_interval,
1383 msg.u.ask_channel.root_shm_path,
1384 msg.u.ask_channel.shm_path);
1385 if (!channel) {
1386 goto end_channel_error;
1387 }
1388
1389 /*
1390 * Assign UST application UID to the channel. This value is ignored for
1391 * per PID buffers. This is specific to UST thus setting this after the
1392 * allocation.
1393 */
1394 channel->ust_app_uid = msg.u.ask_channel.ust_app_uid;
1395
1396 /* Build channel attributes from received message. */
1397 attr.subbuf_size = msg.u.ask_channel.subbuf_size;
1398 attr.num_subbuf = msg.u.ask_channel.num_subbuf;
1399 attr.overwrite = msg.u.ask_channel.overwrite;
1400 attr.switch_timer_interval = msg.u.ask_channel.switch_timer_interval;
1401 attr.read_timer_interval = msg.u.ask_channel.read_timer_interval;
1402 attr.chan_id = msg.u.ask_channel.chan_id;
1403 memcpy(attr.uuid, msg.u.ask_channel.uuid, sizeof(attr.uuid));
1404
1405 /* Match channel buffer type to the UST abi. */
1406 switch (msg.u.ask_channel.output) {
1407 case LTTNG_EVENT_MMAP:
1408 default:
1409 attr.output = LTTNG_UST_MMAP;
1410 break;
1411 }
1412
1413 /* Translate and save channel type. */
1414 switch (msg.u.ask_channel.type) {
1415 case LTTNG_UST_CHAN_PER_CPU:
1416 channel->type = CONSUMER_CHANNEL_TYPE_DATA;
1417 attr.type = LTTNG_UST_CHAN_PER_CPU;
1418 /*
1419 * Set refcount to 1 for owner. Below, we will
1420 * pass ownership to the
1421 * consumer_thread_channel_poll() thread.
1422 */
1423 channel->refcount = 1;
1424 break;
1425 case LTTNG_UST_CHAN_METADATA:
1426 channel->type = CONSUMER_CHANNEL_TYPE_METADATA;
1427 attr.type = LTTNG_UST_CHAN_METADATA;
1428 break;
1429 default:
1430 assert(0);
1431 goto error_fatal;
1432 };
1433
1434 health_code_update();
1435
1436 ret = ask_channel(ctx, sock, channel, &attr);
1437 if (ret < 0) {
1438 goto end_channel_error;
1439 }
1440
1441 if (msg.u.ask_channel.type == LTTNG_UST_CHAN_METADATA) {
1442 ret = consumer_metadata_cache_allocate(channel);
1443 if (ret < 0) {
1444 ERR("Allocating metadata cache");
1445 goto end_channel_error;
1446 }
1447 consumer_timer_switch_start(channel, attr.switch_timer_interval);
1448 attr.switch_timer_interval = 0;
1449 } else {
1450 consumer_timer_live_start(channel,
1451 msg.u.ask_channel.live_timer_interval);
1452 }
1453
1454 health_code_update();
1455
1456 /*
1457 * Add the channel to the internal state AFTER all streams were created
1458 * and successfully sent to session daemon. This way, all streams must
1459 * be ready before this channel is visible to the threads.
1460 * If add_channel succeeds, ownership of the channel is
1461 * passed to consumer_thread_channel_poll().
1462 */
1463 ret = add_channel(channel, ctx);
1464 if (ret < 0) {
1465 if (msg.u.ask_channel.type == LTTNG_UST_CHAN_METADATA) {
1466 if (channel->switch_timer_enabled == 1) {
1467 consumer_timer_switch_stop(channel);
1468 }
1469 consumer_metadata_cache_destroy(channel);
1470 }
1471 if (channel->live_timer_enabled == 1) {
1472 consumer_timer_live_stop(channel);
1473 }
1474 goto end_channel_error;
1475 }
1476
1477 health_code_update();
1478
1479 /*
1480 * Channel and streams are now created. Inform the session daemon that
1481 * everything went well and should wait to receive the channel and
1482 * streams with ustctl API.
1483 */
1484 ret = consumer_send_status_channel(sock, channel);
1485 if (ret < 0) {
1486 /*
1487 * There is probably a problem on the socket.
1488 */
1489 goto error_fatal;
1490 }
1491
1492 break;
1493 }
1494 case LTTNG_CONSUMER_GET_CHANNEL:
1495 {
1496 int ret, relayd_err = 0;
1497 uint64_t key = msg.u.get_channel.key;
1498 struct lttng_consumer_channel *channel;
1499
1500 channel = consumer_find_channel(key);
1501 if (!channel) {
1502 ERR("UST consumer get channel key %" PRIu64 " not found", key);
1503 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
1504 goto end_msg_sessiond;
1505 }
1506
1507 health_code_update();
1508
1509 /* Send everything to sessiond. */
1510 ret = send_sessiond_channel(sock, channel, ctx, &relayd_err);
1511 if (ret < 0) {
1512 if (relayd_err) {
1513 /*
1514 * We were unable to send to the relayd the stream so avoid
1515 * sending back a fatal error to the thread since this is OK
1516 * and the consumer can continue its work. The above call
1517 * has sent the error status message to the sessiond.
1518 */
1519 goto end_nosignal;
1520 }
1521 /*
1522 * The communicaton was broken hence there is a bad state between
1523 * the consumer and sessiond so stop everything.
1524 */
1525 goto error_fatal;
1526 }
1527
1528 health_code_update();
1529
1530 /*
1531 * In no monitor mode, the streams ownership is kept inside the channel
1532 * so don't send them to the data thread.
1533 */
1534 if (!channel->monitor) {
1535 goto end_msg_sessiond;
1536 }
1537
1538 ret = send_streams_to_thread(channel, ctx);
1539 if (ret < 0) {
1540 /*
1541 * If we are unable to send the stream to the thread, there is
1542 * a big problem so just stop everything.
1543 */
1544 goto error_fatal;
1545 }
1546 /* List MUST be empty after or else it could be reused. */
1547 assert(cds_list_empty(&channel->streams.head));
1548 goto end_msg_sessiond;
1549 }
1550 case LTTNG_CONSUMER_DESTROY_CHANNEL:
1551 {
1552 uint64_t key = msg.u.destroy_channel.key;
1553
1554 /*
1555 * Only called if streams have not been sent to stream
1556 * manager thread. However, channel has been sent to
1557 * channel manager thread.
1558 */
1559 notify_thread_del_channel(ctx, key);
1560 goto end_msg_sessiond;
1561 }
1562 case LTTNG_CONSUMER_CLOSE_METADATA:
1563 {
1564 int ret;
1565
1566 ret = close_metadata(msg.u.close_metadata.key);
1567 if (ret != 0) {
1568 ret_code = ret;
1569 }
1570
1571 goto end_msg_sessiond;
1572 }
1573 case LTTNG_CONSUMER_FLUSH_CHANNEL:
1574 {
1575 int ret;
1576
1577 ret = flush_channel(msg.u.flush_channel.key);
1578 if (ret != 0) {
1579 ret_code = ret;
1580 }
1581
1582 goto end_msg_sessiond;
1583 }
1584 case LTTNG_CONSUMER_PUSH_METADATA:
1585 {
1586 int ret;
1587 uint64_t len = msg.u.push_metadata.len;
1588 uint64_t key = msg.u.push_metadata.key;
1589 uint64_t offset = msg.u.push_metadata.target_offset;
1590 struct lttng_consumer_channel *channel;
1591
1592 DBG("UST consumer push metadata key %" PRIu64 " of len %" PRIu64, key,
1593 len);
1594
1595 channel = consumer_find_channel(key);
1596 if (!channel) {
1597 /*
1598 * This is possible if the metadata creation on the consumer side
1599 * is in flight vis-a-vis a concurrent push metadata from the
1600 * session daemon. Simply return that the channel failed and the
1601 * session daemon will handle that message correctly considering
1602 * that this race is acceptable thus the DBG() statement here.
1603 */
1604 DBG("UST consumer push metadata %" PRIu64 " not found", key);
1605 ret_code = LTTCOMM_CONSUMERD_CHANNEL_FAIL;
1606 goto end_msg_sessiond;
1607 }
1608
1609 health_code_update();
1610
1611 if (!len) {
1612 /*
1613 * There is nothing to receive. We have simply
1614 * checked whether the channel can be found.
1615 */
1616 ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1617 goto end_msg_sessiond;
1618 }
1619
1620 /* Tell session daemon we are ready to receive the metadata. */
1621 ret = consumer_send_status_msg(sock, LTTCOMM_CONSUMERD_SUCCESS);
1622 if (ret < 0) {
1623 /* Somehow, the session daemon is not responding anymore. */
1624 goto error_fatal;
1625 }
1626
1627 health_code_update();
1628
1629 /* Wait for more data. */
1630 health_poll_entry();
1631 ret = lttng_consumer_poll_socket(consumer_sockpoll);
1632 health_poll_exit();
1633 if (ret) {
1634 goto error_fatal;
1635 }
1636
1637 health_code_update();
1638
1639 ret = lttng_ustconsumer_recv_metadata(sock, key, offset,
1640 len, channel, 0, 1);
1641 if (ret < 0) {
1642 /* error receiving from sessiond */
1643 goto error_fatal;
1644 } else {
1645 ret_code = ret;
1646 goto end_msg_sessiond;
1647 }
1648 }
1649 case LTTNG_CONSUMER_SETUP_METADATA:
1650 {
1651 int ret;
1652
1653 ret = setup_metadata(ctx, msg.u.setup_metadata.key);
1654 if (ret) {
1655 ret_code = ret;
1656 }
1657 goto end_msg_sessiond;
1658 }
1659 case LTTNG_CONSUMER_SNAPSHOT_CHANNEL:
1660 {
1661 if (msg.u.snapshot_channel.metadata) {
1662 ret = snapshot_metadata(msg.u.snapshot_channel.key,
1663 msg.u.snapshot_channel.pathname,
1664 msg.u.snapshot_channel.relayd_id,
1665 ctx);
1666 if (ret < 0) {
1667 ERR("Snapshot metadata failed");
1668 ret_code = LTTCOMM_CONSUMERD_ERROR_METADATA;
1669 }
1670 } else {
1671 ret = snapshot_channel(msg.u.snapshot_channel.key,
1672 msg.u.snapshot_channel.pathname,
1673 msg.u.snapshot_channel.relayd_id,
1674 msg.u.snapshot_channel.nb_packets_per_stream,
1675 ctx);
1676 if (ret < 0) {
1677 ERR("Snapshot channel failed");
1678 ret_code = LTTCOMM_CONSUMERD_CHANNEL_FAIL;
1679 }
1680 }
1681
1682 health_code_update();
1683 ret = consumer_send_status_msg(sock, ret_code);
1684 if (ret < 0) {
1685 /* Somehow, the session daemon is not responding anymore. */
1686 goto end_nosignal;
1687 }
1688 health_code_update();
1689 break;
1690 }
1691 case LTTNG_CONSUMER_DISCARDED_EVENTS:
1692 {
1693 uint64_t ret;
1694 struct lttng_ht_iter iter;
1695 struct lttng_ht *ht;
1696 struct lttng_consumer_stream *stream;
1697 uint64_t id = msg.u.discarded_events.session_id;
1698 uint64_t key = msg.u.discarded_events.channel_key;
1699
1700 DBG("UST consumer discarded events command for session id %"
1701 PRIu64, id);
1702 rcu_read_lock();
1703 pthread_mutex_lock(&consumer_data.lock);
1704
1705 ht = consumer_data.stream_list_ht;
1706
1707 /*
1708 * We only need a reference to the channel, but they are not
1709 * directly indexed, so we just use the first matching stream
1710 * to extract the information we need, we default to 0 if not
1711 * found (no events are dropped if the channel is not yet in
1712 * use).
1713 */
1714 ret = 0;
1715 cds_lfht_for_each_entry_duplicate(ht->ht,
1716 ht->hash_fct(&id, lttng_ht_seed),
1717 ht->match_fct, &id,
1718 &iter.iter, stream, node_session_id.node) {
1719 if (stream->chan->key == key) {
1720 ret = stream->chan->discarded_events;
1721 break;
1722 }
1723 }
1724 pthread_mutex_unlock(&consumer_data.lock);
1725 rcu_read_unlock();
1726
1727 DBG("UST consumer discarded events command for session id %"
1728 PRIu64 ", channel key %" PRIu64, id, key);
1729
1730 health_code_update();
1731
1732 /* Send back returned value to session daemon */
1733 ret = lttcomm_send_unix_sock(sock, &ret, sizeof(ret));
1734 if (ret < 0) {
1735 PERROR("send discarded events");
1736 goto error_fatal;
1737 }
1738
1739 break;
1740 }
1741 case LTTNG_CONSUMER_LOST_PACKETS:
1742 {
1743 int ret;
1744 uint64_t lost_packets;
1745 struct lttng_ht_iter iter;
1746 struct lttng_ht *ht;
1747 struct lttng_consumer_stream *stream;
1748 uint64_t id = msg.u.lost_packets.session_id;
1749 uint64_t key = msg.u.lost_packets.channel_key;
1750
1751 DBG("UST consumer lost packets command for session id %"
1752 PRIu64, id);
1753 rcu_read_lock();
1754 pthread_mutex_lock(&consumer_data.lock);
1755
1756 ht = consumer_data.stream_list_ht;
1757
1758 /*
1759 * We only need a reference to the channel, but they are not
1760 * directly indexed, so we just use the first matching stream
1761 * to extract the information we need, we default to 0 if not
1762 * found (no packets lost if the channel is not yet in use).
1763 */
1764 lost_packets = 0;
1765 cds_lfht_for_each_entry_duplicate(ht->ht,
1766 ht->hash_fct(&id, lttng_ht_seed),
1767 ht->match_fct, &id,
1768 &iter.iter, stream, node_session_id.node) {
1769 if (stream->chan->key == key) {
1770 lost_packets = stream->chan->lost_packets;
1771 break;
1772 }
1773 }
1774 pthread_mutex_unlock(&consumer_data.lock);
1775 rcu_read_unlock();
1776
1777 DBG("UST consumer lost packets command for session id %"
1778 PRIu64 ", channel key %" PRIu64, id, key);
1779
1780 health_code_update();
1781
1782 /* Send back returned value to session daemon */
1783 ret = lttcomm_send_unix_sock(sock, &lost_packets,
1784 sizeof(lost_packets));
1785 if (ret < 0) {
1786 PERROR("send lost packets");
1787 goto error_fatal;
1788 }
1789
1790 break;
1791 }
1792 default:
1793 break;
1794 }
1795
1796 end_nosignal:
1797 rcu_read_unlock();
1798
1799 health_code_update();
1800
1801 /*
1802 * Return 1 to indicate success since the 0 value can be a socket
1803 * shutdown during the recv() or send() call.
1804 */
1805 return 1;
1806
1807 end_msg_sessiond:
1808 /*
1809 * The returned value here is not useful since either way we'll return 1 to
1810 * the caller because the session daemon socket management is done
1811 * elsewhere. Returning a negative code or 0 will shutdown the consumer.
1812 */
1813 ret = consumer_send_status_msg(sock, ret_code);
1814 if (ret < 0) {
1815 goto error_fatal;
1816 }
1817 rcu_read_unlock();
1818
1819 health_code_update();
1820
1821 return 1;
1822 end_channel_error:
1823 if (channel) {
1824 /*
1825 * Free channel here since no one has a reference to it. We don't
1826 * free after that because a stream can store this pointer.
1827 */
1828 destroy_channel(channel);
1829 }
1830 /* We have to send a status channel message indicating an error. */
1831 ret = consumer_send_status_channel(sock, NULL);
1832 if (ret < 0) {
1833 /* Stop everything if session daemon can not be notified. */
1834 goto error_fatal;
1835 }
1836 rcu_read_unlock();
1837
1838 health_code_update();
1839
1840 return 1;
1841 error_fatal:
1842 rcu_read_unlock();
1843 /* This will issue a consumer stop. */
1844 return -1;
1845 }
1846
1847 /*
1848 * Wrapper over the mmap() read offset from ust-ctl library. Since this can be
1849 * compiled out, we isolate it in this library.
1850 */
1851 int lttng_ustctl_get_mmap_read_offset(struct lttng_consumer_stream *stream,
1852 unsigned long *off)
1853 {
1854 assert(stream);
1855 assert(stream->ustream);
1856
1857 return ustctl_get_mmap_read_offset(stream->ustream, off);
1858 }
1859
1860 /*
1861 * Wrapper over the mmap() read offset from ust-ctl library. Since this can be
1862 * compiled out, we isolate it in this library.
1863 */
1864 void *lttng_ustctl_get_mmap_base(struct lttng_consumer_stream *stream)
1865 {
1866 assert(stream);
1867 assert(stream->ustream);
1868
1869 return ustctl_get_mmap_base(stream->ustream);
1870 }
1871
1872 /*
1873 * Take a snapshot for a specific fd
1874 *
1875 * Returns 0 on success, < 0 on error
1876 */
1877 int lttng_ustconsumer_take_snapshot(struct lttng_consumer_stream *stream)
1878 {
1879 assert(stream);
1880 assert(stream->ustream);
1881
1882 return ustctl_snapshot(stream->ustream);
1883 }
1884
1885 /*
1886 * Get the produced position
1887 *
1888 * Returns 0 on success, < 0 on error
1889 */
1890 int lttng_ustconsumer_get_produced_snapshot(
1891 struct lttng_consumer_stream *stream, unsigned long *pos)
1892 {
1893 assert(stream);
1894 assert(stream->ustream);
1895 assert(pos);
1896
1897 return ustctl_snapshot_get_produced(stream->ustream, pos);
1898 }
1899
1900 /*
1901 * Get the consumed position
1902 *
1903 * Returns 0 on success, < 0 on error
1904 */
1905 int lttng_ustconsumer_get_consumed_snapshot(
1906 struct lttng_consumer_stream *stream, unsigned long *pos)
1907 {
1908 assert(stream);
1909 assert(stream->ustream);
1910 assert(pos);
1911
1912 return ustctl_snapshot_get_consumed(stream->ustream, pos);
1913 }
1914
1915 void lttng_ustconsumer_flush_buffer(struct lttng_consumer_stream *stream,
1916 int producer)
1917 {
1918 assert(stream);
1919 assert(stream->ustream);
1920
1921 ustctl_flush_buffer(stream->ustream, producer);
1922 }
1923
1924 int lttng_ustconsumer_get_current_timestamp(
1925 struct lttng_consumer_stream *stream, uint64_t *ts)
1926 {
1927 assert(stream);
1928 assert(stream->ustream);
1929 assert(ts);
1930
1931 return ustctl_get_current_timestamp(stream->ustream, ts);
1932 }
1933
1934 int lttng_ustconsumer_get_sequence_number(
1935 struct lttng_consumer_stream *stream, uint64_t *seq)
1936 {
1937 assert(stream);
1938 assert(stream->ustream);
1939 assert(seq);
1940
1941 return ustctl_get_sequence_number(stream->ustream, seq);
1942 }
1943
1944 /*
1945 * Called when the stream signal the consumer that it has hang up.
1946 */
1947 void lttng_ustconsumer_on_stream_hangup(struct lttng_consumer_stream *stream)
1948 {
1949 assert(stream);
1950 assert(stream->ustream);
1951
1952 ustctl_flush_buffer(stream->ustream, 0);
1953 stream->hangup_flush_done = 1;
1954 }
1955
1956 void lttng_ustconsumer_del_channel(struct lttng_consumer_channel *chan)
1957 {
1958 int i;
1959
1960 assert(chan);
1961 assert(chan->uchan);
1962
1963 if (chan->switch_timer_enabled == 1) {
1964 consumer_timer_switch_stop(chan);
1965 }
1966 for (i = 0; i < chan->nr_stream_fds; i++) {
1967 int ret;
1968
1969 ret = close(chan->stream_fds[i]);
1970 if (ret) {
1971 PERROR("close");
1972 }
1973 if (chan->shm_path[0]) {
1974 char shm_path[PATH_MAX];
1975
1976 ret = get_stream_shm_path(shm_path, chan->shm_path, i);
1977 if (ret) {
1978 ERR("Cannot get stream shm path");
1979 }
1980 ret = run_as_unlink(shm_path, chan->uid, chan->gid);
1981 if (ret) {
1982 PERROR("unlink %s", shm_path);
1983 }
1984 }
1985 }
1986 /* Try to rmdir all directories under shm_path root. */
1987 if (chan->root_shm_path[0]) {
1988 (void) run_as_recursive_rmdir(chan->root_shm_path,
1989 chan->uid, chan->gid);
1990 }
1991 }
1992
1993 void lttng_ustconsumer_free_channel(struct lttng_consumer_channel *chan)
1994 {
1995 assert(chan);
1996 assert(chan->uchan);
1997
1998 consumer_metadata_cache_destroy(chan);
1999 ustctl_destroy_channel(chan->uchan);
2000 free(chan->stream_fds);
2001 }
2002
2003 void lttng_ustconsumer_del_stream(struct lttng_consumer_stream *stream)
2004 {
2005 assert(stream);
2006 assert(stream->ustream);
2007
2008 if (stream->chan->switch_timer_enabled == 1) {
2009 consumer_timer_switch_stop(stream->chan);
2010 }
2011 ustctl_destroy_stream(stream->ustream);
2012 }
2013
2014 int lttng_ustconsumer_get_wakeup_fd(struct lttng_consumer_stream *stream)
2015 {
2016 assert(stream);
2017 assert(stream->ustream);
2018
2019 return ustctl_stream_get_wakeup_fd(stream->ustream);
2020 }
2021
2022 int lttng_ustconsumer_close_wakeup_fd(struct lttng_consumer_stream *stream)
2023 {
2024 assert(stream);
2025 assert(stream->ustream);
2026
2027 return ustctl_stream_close_wakeup_fd(stream->ustream);
2028 }
2029
2030 /*
2031 * Populate index values of a UST stream. Values are set in big endian order.
2032 *
2033 * Return 0 on success or else a negative value.
2034 */
2035 static int get_index_values(struct ctf_packet_index *index,
2036 struct ustctl_consumer_stream *ustream)
2037 {
2038 int ret;
2039
2040 ret = ustctl_get_timestamp_begin(ustream, &index->timestamp_begin);
2041 if (ret < 0) {
2042 PERROR("ustctl_get_timestamp_begin");
2043 goto error;
2044 }
2045 index->timestamp_begin = htobe64(index->timestamp_begin);
2046
2047 ret = ustctl_get_timestamp_end(ustream, &index->timestamp_end);
2048 if (ret < 0) {
2049 PERROR("ustctl_get_timestamp_end");
2050 goto error;
2051 }
2052 index->timestamp_end = htobe64(index->timestamp_end);
2053
2054 ret = ustctl_get_events_discarded(ustream, &index->events_discarded);
2055 if (ret < 0) {
2056 PERROR("ustctl_get_events_discarded");
2057 goto error;
2058 }
2059 index->events_discarded = htobe64(index->events_discarded);
2060
2061 ret = ustctl_get_content_size(ustream, &index->content_size);
2062 if (ret < 0) {
2063 PERROR("ustctl_get_content_size");
2064 goto error;
2065 }
2066 index->content_size = htobe64(index->content_size);
2067
2068 ret = ustctl_get_packet_size(ustream, &index->packet_size);
2069 if (ret < 0) {
2070 PERROR("ustctl_get_packet_size");
2071 goto error;
2072 }
2073 index->packet_size = htobe64(index->packet_size);
2074
2075 ret = ustctl_get_stream_id(ustream, &index->stream_id);
2076 if (ret < 0) {
2077 PERROR("ustctl_get_stream_id");
2078 goto error;
2079 }
2080 index->stream_id = htobe64(index->stream_id);
2081
2082 error:
2083 return ret;
2084 }
2085
2086 /*
2087 * Write up to one packet from the metadata cache to the channel.
2088 *
2089 * Returns the number of bytes pushed in the cache, or a negative value
2090 * on error.
2091 */
2092 static
2093 int commit_one_metadata_packet(struct lttng_consumer_stream *stream)
2094 {
2095 ssize_t write_len;
2096 int ret;
2097
2098 pthread_mutex_lock(&stream->chan->metadata_cache->lock);
2099 if (stream->chan->metadata_cache->max_offset
2100 == stream->ust_metadata_pushed) {
2101 ret = 0;
2102 goto end;
2103 }
2104
2105 write_len = ustctl_write_one_packet_to_channel(stream->chan->uchan,
2106 &stream->chan->metadata_cache->data[stream->ust_metadata_pushed],
2107 stream->chan->metadata_cache->max_offset
2108 - stream->ust_metadata_pushed);
2109 assert(write_len != 0);
2110 if (write_len < 0) {
2111 ERR("Writing one metadata packet");
2112 ret = -1;
2113 goto end;
2114 }
2115 stream->ust_metadata_pushed += write_len;
2116
2117 assert(stream->chan->metadata_cache->max_offset >=
2118 stream->ust_metadata_pushed);
2119 ret = write_len;
2120
2121 end:
2122 pthread_mutex_unlock(&stream->chan->metadata_cache->lock);
2123 return ret;
2124 }
2125
2126
2127 /*
2128 * Sync metadata meaning request them to the session daemon and snapshot to the
2129 * metadata thread can consumer them.
2130 *
2131 * Metadata stream lock is held here, but we need to release it when
2132 * interacting with sessiond, else we cause a deadlock with live
2133 * awaiting on metadata to be pushed out.
2134 *
2135 * Return 0 if new metadatda is available, EAGAIN if the metadata stream
2136 * is empty or a negative value on error.
2137 */
2138 int lttng_ustconsumer_sync_metadata(struct lttng_consumer_local_data *ctx,
2139 struct lttng_consumer_stream *metadata)
2140 {
2141 int ret;
2142 int retry = 0;
2143
2144 assert(ctx);
2145 assert(metadata);
2146
2147 pthread_mutex_unlock(&metadata->lock);
2148 /*
2149 * Request metadata from the sessiond, but don't wait for the flush
2150 * because we locked the metadata thread.
2151 */
2152 ret = lttng_ustconsumer_request_metadata(ctx, metadata->chan, 0, 0);
2153 if (ret < 0) {
2154 goto end;
2155 }
2156 pthread_mutex_lock(&metadata->lock);
2157
2158 ret = commit_one_metadata_packet(metadata);
2159 if (ret <= 0) {
2160 goto end;
2161 } else if (ret > 0) {
2162 retry = 1;
2163 }
2164
2165 ustctl_flush_buffer(metadata->ustream, 1);
2166 ret = ustctl_snapshot(metadata->ustream);
2167 if (ret < 0) {
2168 if (errno != EAGAIN) {
2169 ERR("Sync metadata, taking UST snapshot");
2170 goto end;
2171 }
2172 DBG("No new metadata when syncing them.");
2173 /* No new metadata, exit. */
2174 ret = ENODATA;
2175 goto end;
2176 }
2177
2178 /*
2179 * After this flush, we still need to extract metadata.
2180 */
2181 if (retry) {
2182 ret = EAGAIN;
2183 }
2184
2185 end:
2186 return ret;
2187 }
2188
2189 /*
2190 * Return 0 on success else a negative value.
2191 */
2192 static int notify_if_more_data(struct lttng_consumer_stream *stream,
2193 struct lttng_consumer_local_data *ctx)
2194 {
2195 int ret;
2196 struct ustctl_consumer_stream *ustream;
2197
2198 assert(stream);
2199 assert(ctx);
2200
2201 ustream = stream->ustream;
2202
2203 /*
2204 * First, we are going to check if there is a new subbuffer available
2205 * before reading the stream wait_fd.
2206 */
2207 /* Get the next subbuffer */
2208 ret = ustctl_get_next_subbuf(ustream);
2209 if (ret) {
2210 /* No more data found, flag the stream. */
2211 stream->has_data = 0;
2212 ret = 0;
2213 goto end;
2214 }
2215
2216 ret = ustctl_put_subbuf(ustream);
2217 assert(!ret);
2218
2219 /* This stream still has data. Flag it and wake up the data thread. */
2220 stream->has_data = 1;
2221
2222 if (stream->monitor && !stream->hangup_flush_done && !ctx->has_wakeup) {
2223 ssize_t writelen;
2224
2225 writelen = lttng_pipe_write(ctx->consumer_wakeup_pipe, "!", 1);
2226 if (writelen < 0 && errno != EAGAIN && errno != EWOULDBLOCK) {
2227 ret = writelen;
2228 goto end;
2229 }
2230
2231 /* The wake up pipe has been notified. */
2232 ctx->has_wakeup = 1;
2233 }
2234 ret = 0;
2235
2236 end:
2237 return ret;
2238 }
2239
2240 static
2241 int update_stream_stats(struct lttng_consumer_stream *stream)
2242 {
2243 int ret;
2244 uint64_t seq, discarded;
2245
2246 ret = ustctl_get_sequence_number(stream->ustream, &seq);
2247 if (ret < 0) {
2248 PERROR("ustctl_get_sequence_number");
2249 goto end;
2250 }
2251 /*
2252 * Start the sequence when we extract the first packet in case we don't
2253 * start at 0 (for example if a consumer is not connected to the
2254 * session immediately after the beginning).
2255 */
2256 if (stream->last_sequence_number == -1ULL) {
2257 stream->last_sequence_number = seq;
2258 } else if (seq > stream->last_sequence_number) {
2259 stream->chan->lost_packets += seq -
2260 stream->last_sequence_number - 1;
2261 } else {
2262 /* seq <= last_sequence_number */
2263 ERR("Sequence number inconsistent : prev = %" PRIu64
2264 ", current = %" PRIu64,
2265 stream->last_sequence_number, seq);
2266 ret = -1;
2267 goto end;
2268 }
2269 stream->last_sequence_number = seq;
2270
2271 ret = ustctl_get_events_discarded(stream->ustream, &discarded);
2272 if (ret < 0) {
2273 PERROR("kernctl_get_events_discarded");
2274 goto end;
2275 }
2276 if (discarded < stream->last_discarded_events) {
2277 /*
2278 * Overflow has occured. We assume only one wrap-around
2279 * has occured.
2280 */
2281 stream->chan->discarded_events +=
2282 (1ULL << (CAA_BITS_PER_LONG - 1)) -
2283 stream->last_discarded_events + discarded;
2284 } else {
2285 stream->chan->discarded_events += discarded -
2286 stream->last_discarded_events;
2287 }
2288 stream->last_discarded_events = discarded;
2289 ret = 0;
2290
2291 end:
2292 return ret;
2293 }
2294
2295 /*
2296 * Read subbuffer from the given stream.
2297 *
2298 * Stream lock MUST be acquired.
2299 *
2300 * Return 0 on success else a negative value.
2301 */
2302 int lttng_ustconsumer_read_subbuffer(struct lttng_consumer_stream *stream,
2303 struct lttng_consumer_local_data *ctx)
2304 {
2305 unsigned long len, subbuf_size, padding;
2306 int err, write_index = 1;
2307 long ret = 0;
2308 struct ustctl_consumer_stream *ustream;
2309 struct ctf_packet_index index;
2310
2311 assert(stream);
2312 assert(stream->ustream);
2313 assert(ctx);
2314
2315 DBG("In UST read_subbuffer (wait_fd: %d, name: %s)", stream->wait_fd,
2316 stream->name);
2317
2318 /* Ease our life for what's next. */
2319 ustream = stream->ustream;
2320
2321 /*
2322 * We can consume the 1 byte written into the wait_fd by UST. Don't trigger
2323 * error if we cannot read this one byte (read returns 0), or if the error
2324 * is EAGAIN or EWOULDBLOCK.
2325 *
2326 * This is only done when the stream is monitored by a thread, before the
2327 * flush is done after a hangup and if the stream is not flagged with data
2328 * since there might be nothing to consume in the wait fd but still have
2329 * data available flagged by the consumer wake up pipe.
2330 */
2331 if (stream->monitor && !stream->hangup_flush_done && !stream->has_data) {
2332 char dummy;
2333 ssize_t readlen;
2334
2335 readlen = lttng_read(stream->wait_fd, &dummy, 1);
2336 if (readlen < 0 && errno != EAGAIN && errno != EWOULDBLOCK) {
2337 ret = readlen;
2338 goto end;
2339 }
2340 }
2341
2342 retry:
2343 /* Get the next subbuffer */
2344 err = ustctl_get_next_subbuf(ustream);
2345 if (err != 0) {
2346 /*
2347 * Populate metadata info if the existing info has
2348 * already been read.
2349 */
2350 if (stream->metadata_flag) {
2351 ret = commit_one_metadata_packet(stream);
2352 if (ret <= 0) {
2353 goto end;
2354 }
2355 ustctl_flush_buffer(stream->ustream, 1);
2356 goto retry;
2357 }
2358
2359 ret = err; /* ustctl_get_next_subbuf returns negative, caller expect positive. */
2360 /*
2361 * This is a debug message even for single-threaded consumer,
2362 * because poll() have more relaxed criterions than get subbuf,
2363 * so get_subbuf may fail for short race windows where poll()
2364 * would issue wakeups.
2365 */
2366 DBG("Reserving sub buffer failed (everything is normal, "
2367 "it is due to concurrency) [ret: %d]", err);
2368 goto end;
2369 }
2370 assert(stream->chan->output == CONSUMER_CHANNEL_MMAP);
2371
2372 if (!stream->metadata_flag) {
2373 index.offset = htobe64(stream->out_fd_offset);
2374 ret = get_index_values(&index, ustream);
2375 if (ret < 0) {
2376 goto end;
2377 }
2378
2379 /* Update the stream's sequence and discarded events count. */
2380 ret = update_stream_stats(stream);
2381 if (ret < 0) {
2382 PERROR("kernctl_get_events_discarded");
2383 goto end;
2384 }
2385 } else {
2386 write_index = 0;
2387 }
2388
2389 /* Get the full padded subbuffer size */
2390 err = ustctl_get_padded_subbuf_size(ustream, &len);
2391 assert(err == 0);
2392
2393 /* Get subbuffer data size (without padding) */
2394 err = ustctl_get_subbuf_size(ustream, &subbuf_size);
2395 assert(err == 0);
2396
2397 /* Make sure we don't get a subbuffer size bigger than the padded */
2398 assert(len >= subbuf_size);
2399
2400 padding = len - subbuf_size;
2401 /* write the subbuffer to the tracefile */
2402 ret = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, subbuf_size, padding, &index);
2403 /*
2404 * The mmap operation should write subbuf_size amount of data when network
2405 * streaming or the full padding (len) size when we are _not_ streaming.
2406 */
2407 if ((ret != subbuf_size && stream->net_seq_idx != (uint64_t) -1ULL) ||
2408 (ret != len && stream->net_seq_idx == (uint64_t) -1ULL)) {
2409 /*
2410 * Display the error but continue processing to try to release the
2411 * subbuffer. This is a DBG statement since any unexpected kill or
2412 * signal, the application gets unregistered, relayd gets closed or
2413 * anything that affects the buffer lifetime will trigger this error.
2414 * So, for the sake of the user, don't print this error since it can
2415 * happen and it is OK with the code flow.
2416 */
2417 DBG("Error writing to tracefile "
2418 "(ret: %ld != len: %lu != subbuf_size: %lu)",
2419 ret, len, subbuf_size);
2420 write_index = 0;
2421 }
2422 err = ustctl_put_next_subbuf(ustream);
2423 assert(err == 0);
2424
2425 /*
2426 * This will consumer the byte on the wait_fd if and only if there is not
2427 * next subbuffer to be acquired.
2428 */
2429 if (!stream->metadata_flag) {
2430 ret = notify_if_more_data(stream, ctx);
2431 if (ret < 0) {
2432 goto end;
2433 }
2434 }
2435
2436 /* Write index if needed. */
2437 if (!write_index) {
2438 goto end;
2439 }
2440
2441 if (stream->chan->live_timer_interval && !stream->metadata_flag) {
2442 /*
2443 * In live, block until all the metadata is sent.
2444 */
2445 pthread_mutex_lock(&stream->metadata_timer_lock);
2446 assert(!stream->missed_metadata_flush);
2447 stream->waiting_on_metadata = true;
2448 pthread_mutex_unlock(&stream->metadata_timer_lock);
2449
2450 err = consumer_stream_sync_metadata(ctx, stream->session_id);
2451
2452 pthread_mutex_lock(&stream->metadata_timer_lock);
2453 stream->waiting_on_metadata = false;
2454 if (stream->missed_metadata_flush) {
2455 stream->missed_metadata_flush = false;
2456 pthread_mutex_unlock(&stream->metadata_timer_lock);
2457 (void) consumer_flush_ust_index(stream);
2458 } else {
2459 pthread_mutex_unlock(&stream->metadata_timer_lock);
2460 }
2461
2462 if (err < 0) {
2463 goto end;
2464 }
2465 }
2466
2467 assert(!stream->metadata_flag);
2468 err = consumer_stream_write_index(stream, &index);
2469 if (err < 0) {
2470 goto end;
2471 }
2472
2473 end:
2474 return ret;
2475 }
2476
2477 /*
2478 * Called when a stream is created.
2479 *
2480 * Return 0 on success or else a negative value.
2481 */
2482 int lttng_ustconsumer_on_recv_stream(struct lttng_consumer_stream *stream)
2483 {
2484 int ret;
2485
2486 assert(stream);
2487
2488 /* Don't create anything if this is set for streaming. */
2489 if (stream->net_seq_idx == (uint64_t) -1ULL && stream->chan->monitor) {
2490 ret = utils_create_stream_file(stream->chan->pathname, stream->name,
2491 stream->chan->tracefile_size, stream->tracefile_count_current,
2492 stream->uid, stream->gid, NULL);
2493 if (ret < 0) {
2494 goto error;
2495 }
2496 stream->out_fd = ret;
2497 stream->tracefile_size_current = 0;
2498
2499 if (!stream->metadata_flag) {
2500 ret = index_create_file(stream->chan->pathname,
2501 stream->name, stream->uid, stream->gid,
2502 stream->chan->tracefile_size,
2503 stream->tracefile_count_current);
2504 if (ret < 0) {
2505 goto error;
2506 }
2507 stream->index_fd = ret;
2508 }
2509 }
2510 ret = 0;
2511
2512 error:
2513 return ret;
2514 }
2515
2516 /*
2517 * Check if data is still being extracted from the buffers for a specific
2518 * stream. Consumer data lock MUST be acquired before calling this function
2519 * and the stream lock.
2520 *
2521 * Return 1 if the traced data are still getting read else 0 meaning that the
2522 * data is available for trace viewer reading.
2523 */
2524 int lttng_ustconsumer_data_pending(struct lttng_consumer_stream *stream)
2525 {
2526 int ret;
2527
2528 assert(stream);
2529 assert(stream->ustream);
2530
2531 DBG("UST consumer checking data pending");
2532
2533 if (stream->endpoint_status != CONSUMER_ENDPOINT_ACTIVE) {
2534 ret = 0;
2535 goto end;
2536 }
2537
2538 if (stream->chan->type == CONSUMER_CHANNEL_TYPE_METADATA) {
2539 uint64_t contiguous, pushed;
2540
2541 /* Ease our life a bit. */
2542 contiguous = stream->chan->metadata_cache->max_offset;
2543 pushed = stream->ust_metadata_pushed;
2544
2545 /*
2546 * We can simply check whether all contiguously available data
2547 * has been pushed to the ring buffer, since the push operation
2548 * is performed within get_next_subbuf(), and because both
2549 * get_next_subbuf() and put_next_subbuf() are issued atomically
2550 * thanks to the stream lock within
2551 * lttng_ustconsumer_read_subbuffer(). This basically means that
2552 * whetnever ust_metadata_pushed is incremented, the associated
2553 * metadata has been consumed from the metadata stream.
2554 */
2555 DBG("UST consumer metadata pending check: contiguous %" PRIu64 " vs pushed %" PRIu64,
2556 contiguous, pushed);
2557 assert(((int64_t) (contiguous - pushed)) >= 0);
2558 if ((contiguous != pushed) ||
2559 (((int64_t) contiguous - pushed) > 0 || contiguous == 0)) {
2560 ret = 1; /* Data is pending */
2561 goto end;
2562 }
2563 } else {
2564 ret = ustctl_get_next_subbuf(stream->ustream);
2565 if (ret == 0) {
2566 /*
2567 * There is still data so let's put back this
2568 * subbuffer.
2569 */
2570 ret = ustctl_put_subbuf(stream->ustream);
2571 assert(ret == 0);
2572 ret = 1; /* Data is pending */
2573 goto end;
2574 }
2575 }
2576
2577 /* Data is NOT pending so ready to be read. */
2578 ret = 0;
2579
2580 end:
2581 return ret;
2582 }
2583
2584 /*
2585 * Stop a given metadata channel timer if enabled and close the wait fd which
2586 * is the poll pipe of the metadata stream.
2587 *
2588 * This MUST be called with the metadata channel acquired.
2589 */
2590 void lttng_ustconsumer_close_metadata(struct lttng_consumer_channel *metadata)
2591 {
2592 int ret;
2593
2594 assert(metadata);
2595 assert(metadata->type == CONSUMER_CHANNEL_TYPE_METADATA);
2596
2597 DBG("Closing metadata channel key %" PRIu64, metadata->key);
2598
2599 if (metadata->switch_timer_enabled == 1) {
2600 consumer_timer_switch_stop(metadata);
2601 }
2602
2603 if (!metadata->metadata_stream) {
2604 goto end;
2605 }
2606
2607 /*
2608 * Closing write side so the thread monitoring the stream wakes up if any
2609 * and clean the metadata stream.
2610 */
2611 if (metadata->metadata_stream->ust_metadata_poll_pipe[1] >= 0) {
2612 ret = close(metadata->metadata_stream->ust_metadata_poll_pipe[1]);
2613 if (ret < 0) {
2614 PERROR("closing metadata pipe write side");
2615 }
2616 metadata->metadata_stream->ust_metadata_poll_pipe[1] = -1;
2617 }
2618
2619 end:
2620 return;
2621 }
2622
2623 /*
2624 * Close every metadata stream wait fd of the metadata hash table. This
2625 * function MUST be used very carefully so not to run into a race between the
2626 * metadata thread handling streams and this function closing their wait fd.
2627 *
2628 * For UST, this is used when the session daemon hangs up. Its the metadata
2629 * producer so calling this is safe because we are assured that no state change
2630 * can occur in the metadata thread for the streams in the hash table.
2631 */
2632 void lttng_ustconsumer_close_all_metadata(struct lttng_ht *metadata_ht)
2633 {
2634 struct lttng_ht_iter iter;
2635 struct lttng_consumer_stream *stream;
2636
2637 assert(metadata_ht);
2638 assert(metadata_ht->ht);
2639
2640 DBG("UST consumer closing all metadata streams");
2641
2642 rcu_read_lock();
2643 cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream,
2644 node.node) {
2645
2646 health_code_update();
2647
2648 pthread_mutex_lock(&stream->chan->lock);
2649 lttng_ustconsumer_close_metadata(stream->chan);
2650 pthread_mutex_unlock(&stream->chan->lock);
2651
2652 }
2653 rcu_read_unlock();
2654 }
2655
2656 void lttng_ustconsumer_close_stream_wakeup(struct lttng_consumer_stream *stream)
2657 {
2658 int ret;
2659
2660 ret = ustctl_stream_close_wakeup_fd(stream->ustream);
2661 if (ret < 0) {
2662 ERR("Unable to close wakeup fd");
2663 }
2664 }
2665
2666 /*
2667 * Please refer to consumer-timer.c before adding any lock within this
2668 * function or any of its callees. Timers have a very strict locking
2669 * semantic with respect to teardown. Failure to respect this semantic
2670 * introduces deadlocks.
2671 *
2672 * DON'T hold the metadata lock when calling this function, else this
2673 * can cause deadlock involving consumer awaiting for metadata to be
2674 * pushed out due to concurrent interaction with the session daemon.
2675 */
2676 int lttng_ustconsumer_request_metadata(struct lttng_consumer_local_data *ctx,
2677 struct lttng_consumer_channel *channel, int timer, int wait)
2678 {
2679 struct lttcomm_metadata_request_msg request;
2680 struct lttcomm_consumer_msg msg;
2681 enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS;
2682 uint64_t len, key, offset;
2683 int ret;
2684
2685 assert(channel);
2686 assert(channel->metadata_cache);
2687
2688 memset(&request, 0, sizeof(request));
2689
2690 /* send the metadata request to sessiond */
2691 switch (consumer_data.type) {
2692 case LTTNG_CONSUMER64_UST:
2693 request.bits_per_long = 64;
2694 break;
2695 case LTTNG_CONSUMER32_UST:
2696 request.bits_per_long = 32;
2697 break;
2698 default:
2699 request.bits_per_long = 0;
2700 break;
2701 }
2702
2703 request.session_id = channel->session_id;
2704 request.session_id_per_pid = channel->session_id_per_pid;
2705 /*
2706 * Request the application UID here so the metadata of that application can
2707 * be sent back. The channel UID corresponds to the user UID of the session
2708 * used for the rights on the stream file(s).
2709 */
2710 request.uid = channel->ust_app_uid;
2711 request.key = channel->key;
2712
2713 DBG("Sending metadata request to sessiond, session id %" PRIu64
2714 ", per-pid %" PRIu64 ", app UID %u and channek key %" PRIu64,
2715 request.session_id, request.session_id_per_pid, request.uid,
2716 request.key);
2717
2718 pthread_mutex_lock(&ctx->metadata_socket_lock);
2719
2720 health_code_update();
2721
2722 ret = lttcomm_send_unix_sock(ctx->consumer_metadata_socket, &request,
2723 sizeof(request));
2724 if (ret < 0) {
2725 ERR("Asking metadata to sessiond");
2726 goto end;
2727 }
2728
2729 health_code_update();
2730
2731 /* Receive the metadata from sessiond */
2732 ret = lttcomm_recv_unix_sock(ctx->consumer_metadata_socket, &msg,
2733 sizeof(msg));
2734 if (ret != sizeof(msg)) {
2735 DBG("Consumer received unexpected message size %d (expects %zu)",
2736 ret, sizeof(msg));
2737 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD);
2738 /*
2739 * The ret value might 0 meaning an orderly shutdown but this is ok
2740 * since the caller handles this.
2741 */
2742 goto end;
2743 }
2744
2745 health_code_update();
2746
2747 if (msg.cmd_type == LTTNG_ERR_UND) {
2748 /* No registry found */
2749 (void) consumer_send_status_msg(ctx->consumer_metadata_socket,
2750 ret_code);
2751 ret = 0;
2752 goto end;
2753 } else if (msg.cmd_type != LTTNG_CONSUMER_PUSH_METADATA) {
2754 ERR("Unexpected cmd_type received %d", msg.cmd_type);
2755 ret = -1;
2756 goto end;
2757 }
2758
2759 len = msg.u.push_metadata.len;
2760 key = msg.u.push_metadata.key;
2761 offset = msg.u.push_metadata.target_offset;
2762
2763 assert(key == channel->key);
2764 if (len == 0) {
2765 DBG("No new metadata to receive for key %" PRIu64, key);
2766 }
2767
2768 health_code_update();
2769
2770 /* Tell session daemon we are ready to receive the metadata. */
2771 ret = consumer_send_status_msg(ctx->consumer_metadata_socket,
2772 LTTCOMM_CONSUMERD_SUCCESS);
2773 if (ret < 0 || len == 0) {
2774 /*
2775 * Somehow, the session daemon is not responding anymore or there is
2776 * nothing to receive.
2777 */
2778 goto end;
2779 }
2780
2781 health_code_update();
2782
2783 ret = lttng_ustconsumer_recv_metadata(ctx->consumer_metadata_socket,
2784 key, offset, len, channel, timer, wait);
2785 if (ret >= 0) {
2786 /*
2787 * Only send the status msg if the sessiond is alive meaning a positive
2788 * ret code.
2789 */
2790 (void) consumer_send_status_msg(ctx->consumer_metadata_socket, ret);
2791 }
2792 ret = 0;
2793
2794 end:
2795 health_code_update();
2796
2797 pthread_mutex_unlock(&ctx->metadata_socket_lock);
2798 return ret;
2799 }
2800
2801 /*
2802 * Return the ustctl call for the get stream id.
2803 */
2804 int lttng_ustconsumer_get_stream_id(struct lttng_consumer_stream *stream,
2805 uint64_t *stream_id)
2806 {
2807 assert(stream);
2808 assert(stream_id);
2809
2810 return ustctl_get_stream_id(stream->ustream, stream_id);
2811 }
This page took 0.11822 seconds and 6 git commands to generate.