01d5ecf6505cc0667a41194d5cb1f07d06e3540b
[lttng-tools.git] / src / common / ust-consumer / ust-consumer.c
1 /*
2 * Copyright (C) 2011 - Julien Desfossez <julien.desfossez@polymtl.ca>
3 * Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
4 *
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License, version 2 only,
7 * as published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License along
15 * with this program; if not, write to the Free Software Foundation, Inc.,
16 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
17 */
18
19 #define _LGPL_SOURCE
20 #include <assert.h>
21 #include <lttng/ust-ctl.h>
22 #include <poll.h>
23 #include <pthread.h>
24 #include <stdlib.h>
25 #include <string.h>
26 #include <sys/mman.h>
27 #include <sys/socket.h>
28 #include <sys/stat.h>
29 #include <sys/types.h>
30 #include <inttypes.h>
31 #include <unistd.h>
32 #include <urcu/list.h>
33 #include <signal.h>
34
35 #include <bin/lttng-consumerd/health-consumerd.h>
36 #include <common/common.h>
37 #include <common/sessiond-comm/sessiond-comm.h>
38 #include <common/relayd/relayd.h>
39 #include <common/compat/fcntl.h>
40 #include <common/compat/endian.h>
41 #include <common/consumer/consumer-metadata-cache.h>
42 #include <common/consumer/consumer-stream.h>
43 #include <common/consumer/consumer-timer.h>
44 #include <common/utils.h>
45 #include <common/index/index.h>
46
47 #include "ust-consumer.h"
48
49 #define INT_MAX_STR_LEN 12 /* includes \0 */
50
51 extern struct lttng_consumer_global_data consumer_data;
52 extern int consumer_poll_timeout;
53 extern volatile int consumer_quit;
54
55 /*
56 * Free channel object and all streams associated with it. This MUST be used
57 * only and only if the channel has _NEVER_ been added to the global channel
58 * hash table.
59 */
60 static void destroy_channel(struct lttng_consumer_channel *channel)
61 {
62 struct lttng_consumer_stream *stream, *stmp;
63
64 assert(channel);
65
66 DBG("UST consumer cleaning stream list");
67
68 cds_list_for_each_entry_safe(stream, stmp, &channel->streams.head,
69 send_node) {
70
71 health_code_update();
72
73 cds_list_del(&stream->send_node);
74 ustctl_destroy_stream(stream->ustream);
75 free(stream);
76 }
77
78 /*
79 * If a channel is available meaning that was created before the streams
80 * were, delete it.
81 */
82 if (channel->uchan) {
83 lttng_ustconsumer_del_channel(channel);
84 lttng_ustconsumer_free_channel(channel);
85 }
86 free(channel);
87 }
88
89 /*
90 * Add channel to internal consumer state.
91 *
92 * Returns 0 on success or else a negative value.
93 */
94 static int add_channel(struct lttng_consumer_channel *channel,
95 struct lttng_consumer_local_data *ctx)
96 {
97 int ret = 0;
98
99 assert(channel);
100 assert(ctx);
101
102 if (ctx->on_recv_channel != NULL) {
103 ret = ctx->on_recv_channel(channel);
104 if (ret == 0) {
105 ret = consumer_add_channel(channel, ctx);
106 } else if (ret < 0) {
107 /* Most likely an ENOMEM. */
108 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
109 goto error;
110 }
111 } else {
112 ret = consumer_add_channel(channel, ctx);
113 }
114
115 DBG("UST consumer channel added (key: %" PRIu64 ")", channel->key);
116
117 error:
118 return ret;
119 }
120
121 /*
122 * Allocate and return a consumer channel object.
123 */
124 static struct lttng_consumer_channel *allocate_channel(uint64_t session_id,
125 const char *pathname, const char *name, uid_t uid, gid_t gid,
126 uint64_t relayd_id, uint64_t key, enum lttng_event_output output,
127 uint64_t tracefile_size, uint64_t tracefile_count,
128 uint64_t session_id_per_pid, unsigned int monitor,
129 unsigned int live_timer_interval,
130 const char *root_shm_path, const char *shm_path)
131 {
132 assert(pathname);
133 assert(name);
134
135 return consumer_allocate_channel(key, session_id, pathname, name, uid,
136 gid, relayd_id, output, tracefile_size,
137 tracefile_count, session_id_per_pid, monitor,
138 live_timer_interval, root_shm_path, shm_path);
139 }
140
141 /*
142 * Allocate and return a consumer stream object. If _alloc_ret is not NULL, the
143 * error value if applicable is set in it else it is kept untouched.
144 *
145 * Return NULL on error else the newly allocated stream object.
146 */
147 static struct lttng_consumer_stream *allocate_stream(int cpu, int key,
148 struct lttng_consumer_channel *channel,
149 struct lttng_consumer_local_data *ctx, int *_alloc_ret)
150 {
151 int alloc_ret;
152 struct lttng_consumer_stream *stream = NULL;
153
154 assert(channel);
155 assert(ctx);
156
157 stream = consumer_allocate_stream(channel->key,
158 key,
159 LTTNG_CONSUMER_ACTIVE_STREAM,
160 channel->name,
161 channel->uid,
162 channel->gid,
163 channel->relayd_id,
164 channel->session_id,
165 cpu,
166 &alloc_ret,
167 channel->type,
168 channel->monitor);
169 if (stream == NULL) {
170 switch (alloc_ret) {
171 case -ENOENT:
172 /*
173 * We could not find the channel. Can happen if cpu hotplug
174 * happens while tearing down.
175 */
176 DBG3("Could not find channel");
177 break;
178 case -ENOMEM:
179 case -EINVAL:
180 default:
181 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
182 break;
183 }
184 goto error;
185 }
186
187 stream->chan = channel;
188
189 error:
190 if (_alloc_ret) {
191 *_alloc_ret = alloc_ret;
192 }
193 return stream;
194 }
195
196 /*
197 * Send the given stream pointer to the corresponding thread.
198 *
199 * Returns 0 on success else a negative value.
200 */
201 static int send_stream_to_thread(struct lttng_consumer_stream *stream,
202 struct lttng_consumer_local_data *ctx)
203 {
204 int ret;
205 struct lttng_pipe *stream_pipe;
206
207 /* Get the right pipe where the stream will be sent. */
208 if (stream->metadata_flag) {
209 ret = consumer_add_metadata_stream(stream);
210 if (ret) {
211 ERR("Consumer add metadata stream %" PRIu64 " failed.",
212 stream->key);
213 goto error;
214 }
215 stream_pipe = ctx->consumer_metadata_pipe;
216 } else {
217 ret = consumer_add_data_stream(stream);
218 if (ret) {
219 ERR("Consumer add stream %" PRIu64 " failed.",
220 stream->key);
221 goto error;
222 }
223 stream_pipe = ctx->consumer_data_pipe;
224 }
225
226 /*
227 * From this point on, the stream's ownership has been moved away from
228 * the channel and becomes globally visible.
229 */
230 stream->globally_visible = 1;
231
232 ret = lttng_pipe_write(stream_pipe, &stream, sizeof(stream));
233 if (ret < 0) {
234 ERR("Consumer write %s stream to pipe %d",
235 stream->metadata_flag ? "metadata" : "data",
236 lttng_pipe_get_writefd(stream_pipe));
237 if (stream->metadata_flag) {
238 consumer_del_stream_for_metadata(stream);
239 } else {
240 consumer_del_stream_for_data(stream);
241 }
242 }
243 error:
244 return ret;
245 }
246
247 static
248 int get_stream_shm_path(char *stream_shm_path, const char *shm_path, int cpu)
249 {
250 char cpu_nr[INT_MAX_STR_LEN]; /* int max len */
251 int ret;
252
253 strncpy(stream_shm_path, shm_path, PATH_MAX);
254 stream_shm_path[PATH_MAX - 1] = '\0';
255 ret = snprintf(cpu_nr, INT_MAX_STR_LEN, "%i", cpu);
256 if (ret < 0) {
257 PERROR("snprintf");
258 goto end;
259 }
260 strncat(stream_shm_path, cpu_nr,
261 PATH_MAX - strlen(stream_shm_path) - 1);
262 ret = 0;
263 end:
264 return ret;
265 }
266
267 /*
268 * Create streams for the given channel using liblttng-ust-ctl.
269 *
270 * Return 0 on success else a negative value.
271 */
272 static int create_ust_streams(struct lttng_consumer_channel *channel,
273 struct lttng_consumer_local_data *ctx)
274 {
275 int ret, cpu = 0;
276 struct ustctl_consumer_stream *ustream;
277 struct lttng_consumer_stream *stream;
278
279 assert(channel);
280 assert(ctx);
281
282 /*
283 * While a stream is available from ustctl. When NULL is returned, we've
284 * reached the end of the possible stream for the channel.
285 */
286 while ((ustream = ustctl_create_stream(channel->uchan, cpu))) {
287 int wait_fd;
288 int ust_metadata_pipe[2];
289
290 health_code_update();
291
292 if (channel->type == CONSUMER_CHANNEL_TYPE_METADATA && channel->monitor) {
293 ret = utils_create_pipe_cloexec_nonblock(ust_metadata_pipe);
294 if (ret < 0) {
295 ERR("Create ust metadata poll pipe");
296 goto error;
297 }
298 wait_fd = ust_metadata_pipe[0];
299 } else {
300 wait_fd = ustctl_stream_get_wait_fd(ustream);
301 }
302
303 /* Allocate consumer stream object. */
304 stream = allocate_stream(cpu, wait_fd, channel, ctx, &ret);
305 if (!stream) {
306 goto error_alloc;
307 }
308 stream->ustream = ustream;
309 /*
310 * Store it so we can save multiple function calls afterwards since
311 * this value is used heavily in the stream threads. This is UST
312 * specific so this is why it's done after allocation.
313 */
314 stream->wait_fd = wait_fd;
315
316 /*
317 * Increment channel refcount since the channel reference has now been
318 * assigned in the allocation process above.
319 */
320 if (stream->chan->monitor) {
321 uatomic_inc(&stream->chan->refcount);
322 }
323
324 /*
325 * Order is important this is why a list is used. On error, the caller
326 * should clean this list.
327 */
328 cds_list_add_tail(&stream->send_node, &channel->streams.head);
329
330 ret = ustctl_get_max_subbuf_size(stream->ustream,
331 &stream->max_sb_size);
332 if (ret < 0) {
333 ERR("ustctl_get_max_subbuf_size failed for stream %s",
334 stream->name);
335 goto error;
336 }
337
338 /* Do actions once stream has been received. */
339 if (ctx->on_recv_stream) {
340 ret = ctx->on_recv_stream(stream);
341 if (ret < 0) {
342 goto error;
343 }
344 }
345
346 DBG("UST consumer add stream %s (key: %" PRIu64 ") with relayd id %" PRIu64,
347 stream->name, stream->key, stream->relayd_stream_id);
348
349 /* Set next CPU stream. */
350 channel->streams.count = ++cpu;
351
352 /* Keep stream reference when creating metadata. */
353 if (channel->type == CONSUMER_CHANNEL_TYPE_METADATA) {
354 channel->metadata_stream = stream;
355 if (channel->monitor) {
356 /* Set metadata poll pipe if we created one */
357 memcpy(stream->ust_metadata_poll_pipe,
358 ust_metadata_pipe,
359 sizeof(ust_metadata_pipe));
360 }
361 }
362 }
363
364 return 0;
365
366 error:
367 error_alloc:
368 return ret;
369 }
370
371 /*
372 * create_posix_shm is never called concurrently within a process.
373 */
374 static
375 int create_posix_shm(void)
376 {
377 char tmp_name[NAME_MAX];
378 int shmfd, ret;
379
380 ret = snprintf(tmp_name, NAME_MAX, "/ust-shm-consumer-%d", getpid());
381 if (ret < 0) {
382 PERROR("snprintf");
383 return -1;
384 }
385 /*
386 * Allocate shm, and immediately unlink its shm oject, keeping
387 * only the file descriptor as a reference to the object.
388 * We specifically do _not_ use the / at the beginning of the
389 * pathname so that some OS implementations can keep it local to
390 * the process (POSIX leaves this implementation-defined).
391 */
392 shmfd = shm_open(tmp_name, O_CREAT | O_EXCL | O_RDWR, 0700);
393 if (shmfd < 0) {
394 PERROR("shm_open");
395 goto error_shm_open;
396 }
397 ret = shm_unlink(tmp_name);
398 if (ret < 0 && errno != ENOENT) {
399 PERROR("shm_unlink");
400 goto error_shm_release;
401 }
402 return shmfd;
403
404 error_shm_release:
405 ret = close(shmfd);
406 if (ret) {
407 PERROR("close");
408 }
409 error_shm_open:
410 return -1;
411 }
412
413 static int open_ust_stream_fd(struct lttng_consumer_channel *channel,
414 struct ustctl_consumer_channel_attr *attr,
415 int cpu)
416 {
417 char shm_path[PATH_MAX];
418 int ret;
419
420 if (!channel->shm_path[0]) {
421 return create_posix_shm();
422 }
423 ret = get_stream_shm_path(shm_path, channel->shm_path, cpu);
424 if (ret) {
425 goto error_shm_path;
426 }
427 return run_as_open(shm_path,
428 O_RDWR | O_CREAT | O_EXCL, S_IRUSR | S_IWUSR,
429 channel->uid, channel->gid);
430
431 error_shm_path:
432 return -1;
433 }
434
435 /*
436 * Create an UST channel with the given attributes and send it to the session
437 * daemon using the ust ctl API.
438 *
439 * Return 0 on success or else a negative value.
440 */
441 static int create_ust_channel(struct lttng_consumer_channel *channel,
442 struct ustctl_consumer_channel_attr *attr,
443 struct ustctl_consumer_channel **ust_chanp)
444 {
445 int ret, nr_stream_fds, i, j;
446 int *stream_fds;
447 struct ustctl_consumer_channel *ust_channel;
448
449 assert(channel);
450 assert(attr);
451 assert(ust_chanp);
452
453 DBG3("Creating channel to ustctl with attr: [overwrite: %d, "
454 "subbuf_size: %" PRIu64 ", num_subbuf: %" PRIu64 ", "
455 "switch_timer_interval: %u, read_timer_interval: %u, "
456 "output: %d, type: %d", attr->overwrite, attr->subbuf_size,
457 attr->num_subbuf, attr->switch_timer_interval,
458 attr->read_timer_interval, attr->output, attr->type);
459
460 if (channel->type == CONSUMER_CHANNEL_TYPE_METADATA)
461 nr_stream_fds = 1;
462 else
463 nr_stream_fds = ustctl_get_nr_stream_per_channel();
464 stream_fds = zmalloc(nr_stream_fds * sizeof(*stream_fds));
465 if (!stream_fds) {
466 ret = -1;
467 goto error_alloc;
468 }
469 for (i = 0; i < nr_stream_fds; i++) {
470 stream_fds[i] = open_ust_stream_fd(channel, attr, i);
471 if (stream_fds[i] < 0) {
472 ret = -1;
473 goto error_open;
474 }
475 }
476 ust_channel = ustctl_create_channel(attr, stream_fds, nr_stream_fds);
477 if (!ust_channel) {
478 ret = -1;
479 goto error_create;
480 }
481 channel->nr_stream_fds = nr_stream_fds;
482 channel->stream_fds = stream_fds;
483 *ust_chanp = ust_channel;
484
485 return 0;
486
487 error_create:
488 error_open:
489 for (j = i - 1; j >= 0; j--) {
490 int closeret;
491
492 closeret = close(stream_fds[j]);
493 if (closeret) {
494 PERROR("close");
495 }
496 if (channel->shm_path[0]) {
497 char shm_path[PATH_MAX];
498
499 closeret = get_stream_shm_path(shm_path,
500 channel->shm_path, j);
501 if (closeret) {
502 ERR("Cannot get stream shm path");
503 }
504 closeret = run_as_unlink(shm_path,
505 channel->uid, channel->gid);
506 if (closeret) {
507 PERROR("unlink %s", shm_path);
508 }
509 }
510 }
511 /* Try to rmdir all directories under shm_path root. */
512 if (channel->root_shm_path[0]) {
513 (void) run_as_recursive_rmdir(channel->root_shm_path,
514 channel->uid, channel->gid);
515 }
516 free(stream_fds);
517 error_alloc:
518 return ret;
519 }
520
521 /*
522 * Send a single given stream to the session daemon using the sock.
523 *
524 * Return 0 on success else a negative value.
525 */
526 static int send_sessiond_stream(int sock, struct lttng_consumer_stream *stream)
527 {
528 int ret;
529
530 assert(stream);
531 assert(sock >= 0);
532
533 DBG("UST consumer sending stream %" PRIu64 " to sessiond", stream->key);
534
535 /* Send stream to session daemon. */
536 ret = ustctl_send_stream_to_sessiond(sock, stream->ustream);
537 if (ret < 0) {
538 goto error;
539 }
540
541 error:
542 return ret;
543 }
544
545 /*
546 * Send channel to sessiond.
547 *
548 * Return 0 on success or else a negative value.
549 */
550 static int send_sessiond_channel(int sock,
551 struct lttng_consumer_channel *channel,
552 struct lttng_consumer_local_data *ctx, int *relayd_error)
553 {
554 int ret, ret_code = LTTCOMM_CONSUMERD_SUCCESS;
555 struct lttng_consumer_stream *stream;
556 uint64_t net_seq_idx = -1ULL;
557
558 assert(channel);
559 assert(ctx);
560 assert(sock >= 0);
561
562 DBG("UST consumer sending channel %s to sessiond", channel->name);
563
564 if (channel->relayd_id != (uint64_t) -1ULL) {
565 cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
566
567 health_code_update();
568
569 /* Try to send the stream to the relayd if one is available. */
570 ret = consumer_send_relayd_stream(stream, stream->chan->pathname);
571 if (ret < 0) {
572 /*
573 * Flag that the relayd was the problem here probably due to a
574 * communicaton error on the socket.
575 */
576 if (relayd_error) {
577 *relayd_error = 1;
578 }
579 ret_code = LTTCOMM_CONSUMERD_RELAYD_FAIL;
580 }
581 if (net_seq_idx == -1ULL) {
582 net_seq_idx = stream->net_seq_idx;
583 }
584 }
585 }
586
587 /* Inform sessiond that we are about to send channel and streams. */
588 ret = consumer_send_status_msg(sock, ret_code);
589 if (ret < 0 || ret_code != LTTCOMM_CONSUMERD_SUCCESS) {
590 /*
591 * Either the session daemon is not responding or the relayd died so we
592 * stop now.
593 */
594 goto error;
595 }
596
597 /* Send channel to sessiond. */
598 ret = ustctl_send_channel_to_sessiond(sock, channel->uchan);
599 if (ret < 0) {
600 goto error;
601 }
602
603 ret = ustctl_channel_close_wakeup_fd(channel->uchan);
604 if (ret < 0) {
605 goto error;
606 }
607
608 /* The channel was sent successfully to the sessiond at this point. */
609 cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
610
611 health_code_update();
612
613 /* Send stream to session daemon. */
614 ret = send_sessiond_stream(sock, stream);
615 if (ret < 0) {
616 goto error;
617 }
618 }
619
620 /* Tell sessiond there is no more stream. */
621 ret = ustctl_send_stream_to_sessiond(sock, NULL);
622 if (ret < 0) {
623 goto error;
624 }
625
626 DBG("UST consumer NULL stream sent to sessiond");
627
628 return 0;
629
630 error:
631 if (ret_code != LTTCOMM_CONSUMERD_SUCCESS) {
632 ret = -1;
633 }
634 return ret;
635 }
636
637 /*
638 * Creates a channel and streams and add the channel it to the channel internal
639 * state. The created stream must ONLY be sent once the GET_CHANNEL command is
640 * received.
641 *
642 * Return 0 on success or else, a negative value is returned and the channel
643 * MUST be destroyed by consumer_del_channel().
644 */
645 static int ask_channel(struct lttng_consumer_local_data *ctx, int sock,
646 struct lttng_consumer_channel *channel,
647 struct ustctl_consumer_channel_attr *attr)
648 {
649 int ret;
650
651 assert(ctx);
652 assert(channel);
653 assert(attr);
654
655 /*
656 * This value is still used by the kernel consumer since for the kernel,
657 * the stream ownership is not IN the consumer so we need to have the
658 * number of left stream that needs to be initialized so we can know when
659 * to delete the channel (see consumer.c).
660 *
661 * As for the user space tracer now, the consumer creates and sends the
662 * stream to the session daemon which only sends them to the application
663 * once every stream of a channel is received making this value useless
664 * because we they will be added to the poll thread before the application
665 * receives them. This ensures that a stream can not hang up during
666 * initilization of a channel.
667 */
668 channel->nb_init_stream_left = 0;
669
670 /* The reply msg status is handled in the following call. */
671 ret = create_ust_channel(channel, attr, &channel->uchan);
672 if (ret < 0) {
673 goto end;
674 }
675
676 channel->wait_fd = ustctl_channel_get_wait_fd(channel->uchan);
677
678 /*
679 * For the snapshots (no monitor), we create the metadata streams
680 * on demand, not during the channel creation.
681 */
682 if (channel->type == CONSUMER_CHANNEL_TYPE_METADATA && !channel->monitor) {
683 ret = 0;
684 goto end;
685 }
686
687 /* Open all streams for this channel. */
688 ret = create_ust_streams(channel, ctx);
689 if (ret < 0) {
690 goto end;
691 }
692
693 end:
694 return ret;
695 }
696
697 /*
698 * Send all stream of a channel to the right thread handling it.
699 *
700 * On error, return a negative value else 0 on success.
701 */
702 static int send_streams_to_thread(struct lttng_consumer_channel *channel,
703 struct lttng_consumer_local_data *ctx)
704 {
705 int ret = 0;
706 struct lttng_consumer_stream *stream, *stmp;
707
708 assert(channel);
709 assert(ctx);
710
711 /* Send streams to the corresponding thread. */
712 cds_list_for_each_entry_safe(stream, stmp, &channel->streams.head,
713 send_node) {
714
715 health_code_update();
716
717 /* Sending the stream to the thread. */
718 ret = send_stream_to_thread(stream, ctx);
719 if (ret < 0) {
720 /*
721 * If we are unable to send the stream to the thread, there is
722 * a big problem so just stop everything.
723 */
724 /* Remove node from the channel stream list. */
725 cds_list_del(&stream->send_node);
726 goto error;
727 }
728
729 /* Remove node from the channel stream list. */
730 cds_list_del(&stream->send_node);
731
732 }
733
734 error:
735 return ret;
736 }
737
738 /*
739 * Flush channel's streams using the given key to retrieve the channel.
740 *
741 * Return 0 on success else an LTTng error code.
742 */
743 static int flush_channel(uint64_t chan_key)
744 {
745 int ret = 0;
746 struct lttng_consumer_channel *channel;
747 struct lttng_consumer_stream *stream;
748 struct lttng_ht *ht;
749 struct lttng_ht_iter iter;
750
751 DBG("UST consumer flush channel key %" PRIu64, chan_key);
752
753 rcu_read_lock();
754 channel = consumer_find_channel(chan_key);
755 if (!channel) {
756 ERR("UST consumer flush channel %" PRIu64 " not found", chan_key);
757 ret = LTTNG_ERR_UST_CHAN_NOT_FOUND;
758 goto error;
759 }
760
761 ht = consumer_data.stream_per_chan_id_ht;
762
763 /* For each stream of the channel id, flush it. */
764 cds_lfht_for_each_entry_duplicate(ht->ht,
765 ht->hash_fct(&channel->key, lttng_ht_seed), ht->match_fct,
766 &channel->key, &iter.iter, stream, node_channel_id.node) {
767
768 health_code_update();
769
770 pthread_mutex_lock(&stream->lock);
771 if (!stream->quiescent) {
772 ustctl_flush_buffer(stream->ustream, 0);
773 stream->quiescent = true;
774 }
775 pthread_mutex_unlock(&stream->lock);
776 }
777 error:
778 rcu_read_unlock();
779 return ret;
780 }
781
782 /*
783 * Clear quiescent state from channel's streams using the given key to
784 * retrieve the channel.
785 *
786 * Return 0 on success else an LTTng error code.
787 */
788 static int clear_quiescent_channel(uint64_t chan_key)
789 {
790 int ret = 0;
791 struct lttng_consumer_channel *channel;
792 struct lttng_consumer_stream *stream;
793 struct lttng_ht *ht;
794 struct lttng_ht_iter iter;
795
796 DBG("UST consumer clear quiescent channel key %" PRIu64, chan_key);
797
798 rcu_read_lock();
799 channel = consumer_find_channel(chan_key);
800 if (!channel) {
801 ERR("UST consumer clear quiescent channel %" PRIu64 " not found", chan_key);
802 ret = LTTNG_ERR_UST_CHAN_NOT_FOUND;
803 goto error;
804 }
805
806 ht = consumer_data.stream_per_chan_id_ht;
807
808 /* For each stream of the channel id, clear quiescent state. */
809 cds_lfht_for_each_entry_duplicate(ht->ht,
810 ht->hash_fct(&channel->key, lttng_ht_seed), ht->match_fct,
811 &channel->key, &iter.iter, stream, node_channel_id.node) {
812
813 health_code_update();
814
815 pthread_mutex_lock(&stream->lock);
816 stream->quiescent = false;
817 pthread_mutex_unlock(&stream->lock);
818 }
819 error:
820 rcu_read_unlock();
821 return ret;
822 }
823
824 /*
825 * Close metadata stream wakeup_fd using the given key to retrieve the channel.
826 * RCU read side lock MUST be acquired before calling this function.
827 *
828 * Return 0 on success else an LTTng error code.
829 */
830 static int close_metadata(uint64_t chan_key)
831 {
832 int ret = 0;
833 struct lttng_consumer_channel *channel;
834
835 DBG("UST consumer close metadata key %" PRIu64, chan_key);
836
837 channel = consumer_find_channel(chan_key);
838 if (!channel) {
839 /*
840 * This is possible if the metadata thread has issue a delete because
841 * the endpoint point of the stream hung up. There is no way the
842 * session daemon can know about it thus use a DBG instead of an actual
843 * error.
844 */
845 DBG("UST consumer close metadata %" PRIu64 " not found", chan_key);
846 ret = LTTNG_ERR_UST_CHAN_NOT_FOUND;
847 goto error;
848 }
849
850 pthread_mutex_lock(&consumer_data.lock);
851 pthread_mutex_lock(&channel->lock);
852
853 if (cds_lfht_is_node_deleted(&channel->node.node)) {
854 goto error_unlock;
855 }
856
857 lttng_ustconsumer_close_metadata(channel);
858
859 error_unlock:
860 pthread_mutex_unlock(&channel->lock);
861 pthread_mutex_unlock(&consumer_data.lock);
862 error:
863 return ret;
864 }
865
866 /*
867 * RCU read side lock MUST be acquired before calling this function.
868 *
869 * Return 0 on success else an LTTng error code.
870 */
871 static int setup_metadata(struct lttng_consumer_local_data *ctx, uint64_t key)
872 {
873 int ret;
874 struct lttng_consumer_channel *metadata;
875
876 DBG("UST consumer setup metadata key %" PRIu64, key);
877
878 metadata = consumer_find_channel(key);
879 if (!metadata) {
880 ERR("UST consumer push metadata %" PRIu64 " not found", key);
881 ret = LTTNG_ERR_UST_CHAN_NOT_FOUND;
882 goto end;
883 }
884
885 /*
886 * In no monitor mode, the metadata channel has no stream(s) so skip the
887 * ownership transfer to the metadata thread.
888 */
889 if (!metadata->monitor) {
890 DBG("Metadata channel in no monitor");
891 ret = 0;
892 goto end;
893 }
894
895 /*
896 * Send metadata stream to relayd if one available. Availability is
897 * known if the stream is still in the list of the channel.
898 */
899 if (cds_list_empty(&metadata->streams.head)) {
900 ERR("Metadata channel key %" PRIu64 ", no stream available.", key);
901 ret = LTTCOMM_CONSUMERD_ERROR_METADATA;
902 goto error_no_stream;
903 }
904
905 /* Send metadata stream to relayd if needed. */
906 if (metadata->metadata_stream->net_seq_idx != (uint64_t) -1ULL) {
907 ret = consumer_send_relayd_stream(metadata->metadata_stream,
908 metadata->pathname);
909 if (ret < 0) {
910 ret = LTTCOMM_CONSUMERD_ERROR_METADATA;
911 goto error;
912 }
913 ret = consumer_send_relayd_streams_sent(
914 metadata->metadata_stream->net_seq_idx);
915 if (ret < 0) {
916 ret = LTTCOMM_CONSUMERD_RELAYD_FAIL;
917 goto error;
918 }
919 }
920
921 ret = send_streams_to_thread(metadata, ctx);
922 if (ret < 0) {
923 /*
924 * If we are unable to send the stream to the thread, there is
925 * a big problem so just stop everything.
926 */
927 ret = LTTCOMM_CONSUMERD_FATAL;
928 goto error;
929 }
930 /* List MUST be empty after or else it could be reused. */
931 assert(cds_list_empty(&metadata->streams.head));
932
933 ret = 0;
934 goto end;
935
936 error:
937 /*
938 * Delete metadata channel on error. At this point, the metadata stream can
939 * NOT be monitored by the metadata thread thus having the guarantee that
940 * the stream is still in the local stream list of the channel. This call
941 * will make sure to clean that list.
942 */
943 consumer_stream_destroy(metadata->metadata_stream, NULL);
944 cds_list_del(&metadata->metadata_stream->send_node);
945 metadata->metadata_stream = NULL;
946 error_no_stream:
947 end:
948 return ret;
949 }
950
951 /*
952 * Snapshot the whole metadata.
953 *
954 * Returns 0 on success, < 0 on error
955 */
956 static int snapshot_metadata(uint64_t key, char *path, uint64_t relayd_id,
957 struct lttng_consumer_local_data *ctx)
958 {
959 int ret = 0;
960 struct lttng_consumer_channel *metadata_channel;
961 struct lttng_consumer_stream *metadata_stream;
962
963 assert(path);
964 assert(ctx);
965
966 DBG("UST consumer snapshot metadata with key %" PRIu64 " at path %s",
967 key, path);
968
969 rcu_read_lock();
970
971 metadata_channel = consumer_find_channel(key);
972 if (!metadata_channel) {
973 ERR("UST snapshot metadata channel not found for key %" PRIu64,
974 key);
975 ret = -1;
976 goto error;
977 }
978 assert(!metadata_channel->monitor);
979
980 health_code_update();
981
982 /*
983 * Ask the sessiond if we have new metadata waiting and update the
984 * consumer metadata cache.
985 */
986 ret = lttng_ustconsumer_request_metadata(ctx, metadata_channel, 0, 1);
987 if (ret < 0) {
988 goto error;
989 }
990
991 health_code_update();
992
993 /*
994 * The metadata stream is NOT created in no monitor mode when the channel
995 * is created on a sessiond ask channel command.
996 */
997 ret = create_ust_streams(metadata_channel, ctx);
998 if (ret < 0) {
999 goto error;
1000 }
1001
1002 metadata_stream = metadata_channel->metadata_stream;
1003 assert(metadata_stream);
1004
1005 if (relayd_id != (uint64_t) -1ULL) {
1006 metadata_stream->net_seq_idx = relayd_id;
1007 ret = consumer_send_relayd_stream(metadata_stream, path);
1008 if (ret < 0) {
1009 goto error_stream;
1010 }
1011 } else {
1012 ret = utils_create_stream_file(path, metadata_stream->name,
1013 metadata_stream->chan->tracefile_size,
1014 metadata_stream->tracefile_count_current,
1015 metadata_stream->uid, metadata_stream->gid, NULL);
1016 if (ret < 0) {
1017 goto error_stream;
1018 }
1019 metadata_stream->out_fd = ret;
1020 metadata_stream->tracefile_size_current = 0;
1021 }
1022
1023 do {
1024 health_code_update();
1025
1026 ret = lttng_consumer_read_subbuffer(metadata_stream, ctx);
1027 if (ret < 0) {
1028 goto error_stream;
1029 }
1030 } while (ret > 0);
1031
1032 error_stream:
1033 /*
1034 * Clean up the stream completly because the next snapshot will use a new
1035 * metadata stream.
1036 */
1037 consumer_stream_destroy(metadata_stream, NULL);
1038 cds_list_del(&metadata_stream->send_node);
1039 metadata_channel->metadata_stream = NULL;
1040
1041 error:
1042 rcu_read_unlock();
1043 return ret;
1044 }
1045
1046 /*
1047 * Take a snapshot of all the stream of a channel.
1048 *
1049 * Returns 0 on success, < 0 on error
1050 */
1051 static int snapshot_channel(uint64_t key, char *path, uint64_t relayd_id,
1052 uint64_t nb_packets_per_stream, struct lttng_consumer_local_data *ctx)
1053 {
1054 int ret;
1055 unsigned use_relayd = 0;
1056 unsigned long consumed_pos, produced_pos;
1057 struct lttng_consumer_channel *channel;
1058 struct lttng_consumer_stream *stream;
1059
1060 assert(path);
1061 assert(ctx);
1062
1063 rcu_read_lock();
1064
1065 if (relayd_id != (uint64_t) -1ULL) {
1066 use_relayd = 1;
1067 }
1068
1069 channel = consumer_find_channel(key);
1070 if (!channel) {
1071 ERR("UST snapshot channel not found for key %" PRIu64, key);
1072 ret = -1;
1073 goto error;
1074 }
1075 assert(!channel->monitor);
1076 DBG("UST consumer snapshot channel %" PRIu64, key);
1077
1078 cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
1079 /* Are we at a position _before_ the first available packet ? */
1080 bool before_first_packet = true;
1081
1082 health_code_update();
1083
1084 /* Lock stream because we are about to change its state. */
1085 pthread_mutex_lock(&stream->lock);
1086 stream->net_seq_idx = relayd_id;
1087
1088 if (use_relayd) {
1089 ret = consumer_send_relayd_stream(stream, path);
1090 if (ret < 0) {
1091 goto error_unlock;
1092 }
1093 } else {
1094 ret = utils_create_stream_file(path, stream->name,
1095 stream->chan->tracefile_size,
1096 stream->tracefile_count_current,
1097 stream->uid, stream->gid, NULL);
1098 if (ret < 0) {
1099 goto error_unlock;
1100 }
1101 stream->out_fd = ret;
1102 stream->tracefile_size_current = 0;
1103
1104 DBG("UST consumer snapshot stream %s/%s (%" PRIu64 ")", path,
1105 stream->name, stream->key);
1106 }
1107 if (relayd_id != -1ULL) {
1108 ret = consumer_send_relayd_streams_sent(relayd_id);
1109 if (ret < 0) {
1110 goto error_unlock;
1111 }
1112 }
1113
1114 /*
1115 * If tracing is active, we want to perform a "full" buffer flush.
1116 * Else, if quiescent, it has already been done by the prior stop.
1117 */
1118 if (!stream->quiescent) {
1119 ustctl_flush_buffer(stream->ustream, 0);
1120 }
1121
1122 ret = lttng_ustconsumer_take_snapshot(stream);
1123 if (ret < 0) {
1124 ERR("Taking UST snapshot");
1125 goto error_unlock;
1126 }
1127
1128 ret = lttng_ustconsumer_get_produced_snapshot(stream, &produced_pos);
1129 if (ret < 0) {
1130 ERR("Produced UST snapshot position");
1131 goto error_unlock;
1132 }
1133
1134 ret = lttng_ustconsumer_get_consumed_snapshot(stream, &consumed_pos);
1135 if (ret < 0) {
1136 ERR("Consumerd UST snapshot position");
1137 goto error_unlock;
1138 }
1139
1140 /*
1141 * The original value is sent back if max stream size is larger than
1142 * the possible size of the snapshot. Also, we assume that the session
1143 * daemon should never send a maximum stream size that is lower than
1144 * subbuffer size.
1145 */
1146 consumed_pos = consumer_get_consume_start_pos(consumed_pos,
1147 produced_pos, nb_packets_per_stream,
1148 stream->max_sb_size);
1149
1150 while (consumed_pos < produced_pos) {
1151 ssize_t read_len;
1152 unsigned long len, padded_len;
1153 int lost_packet = 0;
1154
1155 health_code_update();
1156
1157 DBG("UST consumer taking snapshot at pos %lu", consumed_pos);
1158
1159 ret = ustctl_get_subbuf(stream->ustream, &consumed_pos);
1160 if (ret < 0) {
1161 if (ret != -EAGAIN) {
1162 PERROR("ustctl_get_subbuf snapshot");
1163 goto error_close_stream;
1164 }
1165 DBG("UST consumer get subbuf failed. Skipping it.");
1166 consumed_pos += stream->max_sb_size;
1167
1168 /*
1169 * Start accounting lost packets only when we
1170 * already have extracted packets (to match the
1171 * content of the final snapshot).
1172 */
1173 if (!before_first_packet) {
1174 lost_packet = 1;
1175 }
1176 continue;
1177 }
1178
1179 ret = ustctl_get_subbuf_size(stream->ustream, &len);
1180 if (ret < 0) {
1181 ERR("Snapshot ustctl_get_subbuf_size");
1182 goto error_put_subbuf;
1183 }
1184
1185 ret = ustctl_get_padded_subbuf_size(stream->ustream, &padded_len);
1186 if (ret < 0) {
1187 ERR("Snapshot ustctl_get_padded_subbuf_size");
1188 goto error_put_subbuf;
1189 }
1190
1191 read_len = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, len,
1192 padded_len - len, NULL);
1193 if (use_relayd) {
1194 if (read_len != len) {
1195 ret = -EPERM;
1196 goto error_put_subbuf;
1197 }
1198 } else {
1199 if (read_len != padded_len) {
1200 ret = -EPERM;
1201 goto error_put_subbuf;
1202 }
1203 }
1204
1205 ret = ustctl_put_subbuf(stream->ustream);
1206 if (ret < 0) {
1207 ERR("Snapshot ustctl_put_subbuf");
1208 goto error_close_stream;
1209 }
1210 consumed_pos += stream->max_sb_size;
1211
1212 /*
1213 * Only account lost packets located between
1214 * succesfully extracted packets (do not account before
1215 * and after since they are not visible in the
1216 * resulting snapshot).
1217 */
1218 stream->chan->lost_packets += lost_packet;
1219 lost_packet = 0;
1220 before_first_packet = false;
1221 }
1222
1223 /* Simply close the stream so we can use it on the next snapshot. */
1224 consumer_stream_close(stream);
1225 pthread_mutex_unlock(&stream->lock);
1226 }
1227
1228 rcu_read_unlock();
1229 return 0;
1230
1231 error_put_subbuf:
1232 if (ustctl_put_subbuf(stream->ustream) < 0) {
1233 ERR("Snapshot ustctl_put_subbuf");
1234 }
1235 error_close_stream:
1236 consumer_stream_close(stream);
1237 error_unlock:
1238 pthread_mutex_unlock(&stream->lock);
1239 error:
1240 rcu_read_unlock();
1241 return ret;
1242 }
1243
1244 /*
1245 * Receive the metadata updates from the sessiond. Supports receiving
1246 * overlapping metadata, but is needs to always belong to a contiguous
1247 * range starting from 0.
1248 * Be careful about the locks held when calling this function: it needs
1249 * the metadata cache flush to concurrently progress in order to
1250 * complete.
1251 */
1252 int lttng_ustconsumer_recv_metadata(int sock, uint64_t key, uint64_t offset,
1253 uint64_t len, uint64_t version,
1254 struct lttng_consumer_channel *channel, int timer, int wait)
1255 {
1256 int ret, ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1257 char *metadata_str;
1258
1259 DBG("UST consumer push metadata key %" PRIu64 " of len %" PRIu64, key, len);
1260
1261 metadata_str = zmalloc(len * sizeof(char));
1262 if (!metadata_str) {
1263 PERROR("zmalloc metadata string");
1264 ret_code = LTTCOMM_CONSUMERD_ENOMEM;
1265 goto end;
1266 }
1267
1268 health_code_update();
1269
1270 /* Receive metadata string. */
1271 ret = lttcomm_recv_unix_sock(sock, metadata_str, len);
1272 if (ret < 0) {
1273 /* Session daemon is dead so return gracefully. */
1274 ret_code = ret;
1275 goto end_free;
1276 }
1277
1278 health_code_update();
1279
1280 pthread_mutex_lock(&channel->metadata_cache->lock);
1281 ret = consumer_metadata_cache_write(channel, offset, len, version,
1282 metadata_str);
1283 if (ret < 0) {
1284 /* Unable to handle metadata. Notify session daemon. */
1285 ret_code = LTTCOMM_CONSUMERD_ERROR_METADATA;
1286 /*
1287 * Skip metadata flush on write error since the offset and len might
1288 * not have been updated which could create an infinite loop below when
1289 * waiting for the metadata cache to be flushed.
1290 */
1291 pthread_mutex_unlock(&channel->metadata_cache->lock);
1292 goto end_free;
1293 }
1294 pthread_mutex_unlock(&channel->metadata_cache->lock);
1295
1296 if (!wait) {
1297 goto end_free;
1298 }
1299 while (consumer_metadata_cache_flushed(channel, offset + len, timer)) {
1300 DBG("Waiting for metadata to be flushed");
1301
1302 health_code_update();
1303
1304 usleep(DEFAULT_METADATA_AVAILABILITY_WAIT_TIME);
1305 }
1306
1307 end_free:
1308 free(metadata_str);
1309 end:
1310 return ret_code;
1311 }
1312
1313 /*
1314 * Receive command from session daemon and process it.
1315 *
1316 * Return 1 on success else a negative value or 0.
1317 */
1318 int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
1319 int sock, struct pollfd *consumer_sockpoll)
1320 {
1321 ssize_t ret;
1322 enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1323 struct lttcomm_consumer_msg msg;
1324 struct lttng_consumer_channel *channel = NULL;
1325
1326 health_code_update();
1327
1328 ret = lttcomm_recv_unix_sock(sock, &msg, sizeof(msg));
1329 if (ret != sizeof(msg)) {
1330 DBG("Consumer received unexpected message size %zd (expects %zu)",
1331 ret, sizeof(msg));
1332 /*
1333 * The ret value might 0 meaning an orderly shutdown but this is ok
1334 * since the caller handles this.
1335 */
1336 if (ret > 0) {
1337 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD);
1338 ret = -1;
1339 }
1340 return ret;
1341 }
1342
1343 health_code_update();
1344
1345 /* deprecated */
1346 assert(msg.cmd_type != LTTNG_CONSUMER_STOP);
1347
1348 health_code_update();
1349
1350 /* relayd needs RCU read-side lock */
1351 rcu_read_lock();
1352
1353 switch (msg.cmd_type) {
1354 case LTTNG_CONSUMER_ADD_RELAYD_SOCKET:
1355 {
1356 /* Session daemon status message are handled in the following call. */
1357 ret = consumer_add_relayd_socket(msg.u.relayd_sock.net_index,
1358 msg.u.relayd_sock.type, ctx, sock, consumer_sockpoll,
1359 &msg.u.relayd_sock.sock, msg.u.relayd_sock.session_id,
1360 msg.u.relayd_sock.relayd_session_id);
1361 goto end_nosignal;
1362 }
1363 case LTTNG_CONSUMER_DESTROY_RELAYD:
1364 {
1365 uint64_t index = msg.u.destroy_relayd.net_seq_idx;
1366 struct consumer_relayd_sock_pair *relayd;
1367
1368 DBG("UST consumer destroying relayd %" PRIu64, index);
1369
1370 /* Get relayd reference if exists. */
1371 relayd = consumer_find_relayd(index);
1372 if (relayd == NULL) {
1373 DBG("Unable to find relayd %" PRIu64, index);
1374 ret_code = LTTCOMM_CONSUMERD_RELAYD_FAIL;
1375 }
1376
1377 /*
1378 * Each relayd socket pair has a refcount of stream attached to it
1379 * which tells if the relayd is still active or not depending on the
1380 * refcount value.
1381 *
1382 * This will set the destroy flag of the relayd object and destroy it
1383 * if the refcount reaches zero when called.
1384 *
1385 * The destroy can happen either here or when a stream fd hangs up.
1386 */
1387 if (relayd) {
1388 consumer_flag_relayd_for_destroy(relayd);
1389 }
1390
1391 goto end_msg_sessiond;
1392 }
1393 case LTTNG_CONSUMER_UPDATE_STREAM:
1394 {
1395 rcu_read_unlock();
1396 return -ENOSYS;
1397 }
1398 case LTTNG_CONSUMER_DATA_PENDING:
1399 {
1400 int ret, is_data_pending;
1401 uint64_t id = msg.u.data_pending.session_id;
1402
1403 DBG("UST consumer data pending command for id %" PRIu64, id);
1404
1405 is_data_pending = consumer_data_pending(id);
1406
1407 /* Send back returned value to session daemon */
1408 ret = lttcomm_send_unix_sock(sock, &is_data_pending,
1409 sizeof(is_data_pending));
1410 if (ret < 0) {
1411 DBG("Error when sending the data pending ret code: %d", ret);
1412 goto error_fatal;
1413 }
1414
1415 /*
1416 * No need to send back a status message since the data pending
1417 * returned value is the response.
1418 */
1419 break;
1420 }
1421 case LTTNG_CONSUMER_ASK_CHANNEL_CREATION:
1422 {
1423 int ret;
1424 struct ustctl_consumer_channel_attr attr;
1425
1426 /* Create a plain object and reserve a channel key. */
1427 channel = allocate_channel(msg.u.ask_channel.session_id,
1428 msg.u.ask_channel.pathname, msg.u.ask_channel.name,
1429 msg.u.ask_channel.uid, msg.u.ask_channel.gid,
1430 msg.u.ask_channel.relayd_id, msg.u.ask_channel.key,
1431 (enum lttng_event_output) msg.u.ask_channel.output,
1432 msg.u.ask_channel.tracefile_size,
1433 msg.u.ask_channel.tracefile_count,
1434 msg.u.ask_channel.session_id_per_pid,
1435 msg.u.ask_channel.monitor,
1436 msg.u.ask_channel.live_timer_interval,
1437 msg.u.ask_channel.root_shm_path,
1438 msg.u.ask_channel.shm_path);
1439 if (!channel) {
1440 goto end_channel_error;
1441 }
1442
1443 /*
1444 * Assign UST application UID to the channel. This value is ignored for
1445 * per PID buffers. This is specific to UST thus setting this after the
1446 * allocation.
1447 */
1448 channel->ust_app_uid = msg.u.ask_channel.ust_app_uid;
1449
1450 /* Build channel attributes from received message. */
1451 attr.subbuf_size = msg.u.ask_channel.subbuf_size;
1452 attr.num_subbuf = msg.u.ask_channel.num_subbuf;
1453 attr.overwrite = msg.u.ask_channel.overwrite;
1454 attr.switch_timer_interval = msg.u.ask_channel.switch_timer_interval;
1455 attr.read_timer_interval = msg.u.ask_channel.read_timer_interval;
1456 attr.chan_id = msg.u.ask_channel.chan_id;
1457 memcpy(attr.uuid, msg.u.ask_channel.uuid, sizeof(attr.uuid));
1458
1459 /* Match channel buffer type to the UST abi. */
1460 switch (msg.u.ask_channel.output) {
1461 case LTTNG_EVENT_MMAP:
1462 default:
1463 attr.output = LTTNG_UST_MMAP;
1464 break;
1465 }
1466
1467 /* Translate and save channel type. */
1468 switch (msg.u.ask_channel.type) {
1469 case LTTNG_UST_CHAN_PER_CPU:
1470 channel->type = CONSUMER_CHANNEL_TYPE_DATA;
1471 attr.type = LTTNG_UST_CHAN_PER_CPU;
1472 /*
1473 * Set refcount to 1 for owner. Below, we will
1474 * pass ownership to the
1475 * consumer_thread_channel_poll() thread.
1476 */
1477 channel->refcount = 1;
1478 break;
1479 case LTTNG_UST_CHAN_METADATA:
1480 channel->type = CONSUMER_CHANNEL_TYPE_METADATA;
1481 attr.type = LTTNG_UST_CHAN_METADATA;
1482 break;
1483 default:
1484 assert(0);
1485 goto error_fatal;
1486 };
1487
1488 health_code_update();
1489
1490 ret = ask_channel(ctx, sock, channel, &attr);
1491 if (ret < 0) {
1492 goto end_channel_error;
1493 }
1494
1495 if (msg.u.ask_channel.type == LTTNG_UST_CHAN_METADATA) {
1496 ret = consumer_metadata_cache_allocate(channel);
1497 if (ret < 0) {
1498 ERR("Allocating metadata cache");
1499 goto end_channel_error;
1500 }
1501 consumer_timer_switch_start(channel, attr.switch_timer_interval);
1502 attr.switch_timer_interval = 0;
1503 } else {
1504 consumer_timer_live_start(channel,
1505 msg.u.ask_channel.live_timer_interval);
1506 }
1507
1508 health_code_update();
1509
1510 /*
1511 * Add the channel to the internal state AFTER all streams were created
1512 * and successfully sent to session daemon. This way, all streams must
1513 * be ready before this channel is visible to the threads.
1514 * If add_channel succeeds, ownership of the channel is
1515 * passed to consumer_thread_channel_poll().
1516 */
1517 ret = add_channel(channel, ctx);
1518 if (ret < 0) {
1519 if (msg.u.ask_channel.type == LTTNG_UST_CHAN_METADATA) {
1520 if (channel->switch_timer_enabled == 1) {
1521 consumer_timer_switch_stop(channel);
1522 }
1523 consumer_metadata_cache_destroy(channel);
1524 }
1525 if (channel->live_timer_enabled == 1) {
1526 consumer_timer_live_stop(channel);
1527 }
1528 goto end_channel_error;
1529 }
1530
1531 health_code_update();
1532
1533 /*
1534 * Channel and streams are now created. Inform the session daemon that
1535 * everything went well and should wait to receive the channel and
1536 * streams with ustctl API.
1537 */
1538 ret = consumer_send_status_channel(sock, channel);
1539 if (ret < 0) {
1540 /*
1541 * There is probably a problem on the socket.
1542 */
1543 goto error_fatal;
1544 }
1545
1546 break;
1547 }
1548 case LTTNG_CONSUMER_GET_CHANNEL:
1549 {
1550 int ret, relayd_err = 0;
1551 uint64_t key = msg.u.get_channel.key;
1552 struct lttng_consumer_channel *channel;
1553
1554 channel = consumer_find_channel(key);
1555 if (!channel) {
1556 ERR("UST consumer get channel key %" PRIu64 " not found", key);
1557 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
1558 goto end_msg_sessiond;
1559 }
1560
1561 health_code_update();
1562
1563 /* Send everything to sessiond. */
1564 ret = send_sessiond_channel(sock, channel, ctx, &relayd_err);
1565 if (ret < 0) {
1566 if (relayd_err) {
1567 /*
1568 * We were unable to send to the relayd the stream so avoid
1569 * sending back a fatal error to the thread since this is OK
1570 * and the consumer can continue its work. The above call
1571 * has sent the error status message to the sessiond.
1572 */
1573 goto end_nosignal;
1574 }
1575 /*
1576 * The communicaton was broken hence there is a bad state between
1577 * the consumer and sessiond so stop everything.
1578 */
1579 goto error_fatal;
1580 }
1581
1582 health_code_update();
1583
1584 /*
1585 * In no monitor mode, the streams ownership is kept inside the channel
1586 * so don't send them to the data thread.
1587 */
1588 if (!channel->monitor) {
1589 goto end_msg_sessiond;
1590 }
1591
1592 ret = send_streams_to_thread(channel, ctx);
1593 if (ret < 0) {
1594 /*
1595 * If we are unable to send the stream to the thread, there is
1596 * a big problem so just stop everything.
1597 */
1598 goto error_fatal;
1599 }
1600 /* List MUST be empty after or else it could be reused. */
1601 assert(cds_list_empty(&channel->streams.head));
1602 goto end_msg_sessiond;
1603 }
1604 case LTTNG_CONSUMER_DESTROY_CHANNEL:
1605 {
1606 uint64_t key = msg.u.destroy_channel.key;
1607
1608 /*
1609 * Only called if streams have not been sent to stream
1610 * manager thread. However, channel has been sent to
1611 * channel manager thread.
1612 */
1613 notify_thread_del_channel(ctx, key);
1614 goto end_msg_sessiond;
1615 }
1616 case LTTNG_CONSUMER_CLOSE_METADATA:
1617 {
1618 int ret;
1619
1620 ret = close_metadata(msg.u.close_metadata.key);
1621 if (ret != 0) {
1622 ret_code = ret;
1623 }
1624
1625 goto end_msg_sessiond;
1626 }
1627 case LTTNG_CONSUMER_FLUSH_CHANNEL:
1628 {
1629 int ret;
1630
1631 ret = flush_channel(msg.u.flush_channel.key);
1632 if (ret != 0) {
1633 ret_code = ret;
1634 }
1635
1636 goto end_msg_sessiond;
1637 }
1638 case LTTNG_CONSUMER_CLEAR_QUIESCENT_CHANNEL:
1639 {
1640 int ret;
1641
1642 ret = clear_quiescent_channel(
1643 msg.u.clear_quiescent_channel.key);
1644 if (ret != 0) {
1645 ret_code = ret;
1646 }
1647
1648 goto end_msg_sessiond;
1649 }
1650 case LTTNG_CONSUMER_PUSH_METADATA:
1651 {
1652 int ret;
1653 uint64_t len = msg.u.push_metadata.len;
1654 uint64_t key = msg.u.push_metadata.key;
1655 uint64_t offset = msg.u.push_metadata.target_offset;
1656 uint64_t version = msg.u.push_metadata.version;
1657 struct lttng_consumer_channel *channel;
1658
1659 DBG("UST consumer push metadata key %" PRIu64 " of len %" PRIu64, key,
1660 len);
1661
1662 channel = consumer_find_channel(key);
1663 if (!channel) {
1664 /*
1665 * This is possible if the metadata creation on the consumer side
1666 * is in flight vis-a-vis a concurrent push metadata from the
1667 * session daemon. Simply return that the channel failed and the
1668 * session daemon will handle that message correctly considering
1669 * that this race is acceptable thus the DBG() statement here.
1670 */
1671 DBG("UST consumer push metadata %" PRIu64 " not found", key);
1672 ret_code = LTTCOMM_CONSUMERD_CHANNEL_FAIL;
1673 goto end_msg_sessiond;
1674 }
1675
1676 health_code_update();
1677
1678 if (!len) {
1679 /*
1680 * There is nothing to receive. We have simply
1681 * checked whether the channel can be found.
1682 */
1683 ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1684 goto end_msg_sessiond;
1685 }
1686
1687 /* Tell session daemon we are ready to receive the metadata. */
1688 ret = consumer_send_status_msg(sock, LTTCOMM_CONSUMERD_SUCCESS);
1689 if (ret < 0) {
1690 /* Somehow, the session daemon is not responding anymore. */
1691 goto error_fatal;
1692 }
1693
1694 health_code_update();
1695
1696 /* Wait for more data. */
1697 health_poll_entry();
1698 ret = lttng_consumer_poll_socket(consumer_sockpoll);
1699 health_poll_exit();
1700 if (ret) {
1701 goto error_fatal;
1702 }
1703
1704 health_code_update();
1705
1706 ret = lttng_ustconsumer_recv_metadata(sock, key, offset,
1707 len, version, channel, 0, 1);
1708 if (ret < 0) {
1709 /* error receiving from sessiond */
1710 goto error_fatal;
1711 } else {
1712 ret_code = ret;
1713 goto end_msg_sessiond;
1714 }
1715 }
1716 case LTTNG_CONSUMER_SETUP_METADATA:
1717 {
1718 int ret;
1719
1720 ret = setup_metadata(ctx, msg.u.setup_metadata.key);
1721 if (ret) {
1722 ret_code = ret;
1723 }
1724 goto end_msg_sessiond;
1725 }
1726 case LTTNG_CONSUMER_SNAPSHOT_CHANNEL:
1727 {
1728 if (msg.u.snapshot_channel.metadata) {
1729 ret = snapshot_metadata(msg.u.snapshot_channel.key,
1730 msg.u.snapshot_channel.pathname,
1731 msg.u.snapshot_channel.relayd_id,
1732 ctx);
1733 if (ret < 0) {
1734 ERR("Snapshot metadata failed");
1735 ret_code = LTTCOMM_CONSUMERD_ERROR_METADATA;
1736 }
1737 } else {
1738 ret = snapshot_channel(msg.u.snapshot_channel.key,
1739 msg.u.snapshot_channel.pathname,
1740 msg.u.snapshot_channel.relayd_id,
1741 msg.u.snapshot_channel.nb_packets_per_stream,
1742 ctx);
1743 if (ret < 0) {
1744 ERR("Snapshot channel failed");
1745 ret_code = LTTCOMM_CONSUMERD_CHANNEL_FAIL;
1746 }
1747 }
1748
1749 health_code_update();
1750 ret = consumer_send_status_msg(sock, ret_code);
1751 if (ret < 0) {
1752 /* Somehow, the session daemon is not responding anymore. */
1753 goto end_nosignal;
1754 }
1755 health_code_update();
1756 break;
1757 }
1758 case LTTNG_CONSUMER_DISCARDED_EVENTS:
1759 {
1760 int ret = 0;
1761 uint64_t discarded_events;
1762 struct lttng_ht_iter iter;
1763 struct lttng_ht *ht;
1764 struct lttng_consumer_stream *stream;
1765 uint64_t id = msg.u.discarded_events.session_id;
1766 uint64_t key = msg.u.discarded_events.channel_key;
1767
1768 DBG("UST consumer discarded events command for session id %"
1769 PRIu64, id);
1770 rcu_read_lock();
1771 pthread_mutex_lock(&consumer_data.lock);
1772
1773 ht = consumer_data.stream_list_ht;
1774
1775 /*
1776 * We only need a reference to the channel, but they are not
1777 * directly indexed, so we just use the first matching stream
1778 * to extract the information we need, we default to 0 if not
1779 * found (no events are dropped if the channel is not yet in
1780 * use).
1781 */
1782 discarded_events = 0;
1783 cds_lfht_for_each_entry_duplicate(ht->ht,
1784 ht->hash_fct(&id, lttng_ht_seed),
1785 ht->match_fct, &id,
1786 &iter.iter, stream, node_session_id.node) {
1787 if (stream->chan->key == key) {
1788 discarded_events = stream->chan->discarded_events;
1789 break;
1790 }
1791 }
1792 pthread_mutex_unlock(&consumer_data.lock);
1793 rcu_read_unlock();
1794
1795 DBG("UST consumer discarded events command for session id %"
1796 PRIu64 ", channel key %" PRIu64, id, key);
1797
1798 health_code_update();
1799
1800 /* Send back returned value to session daemon */
1801 ret = lttcomm_send_unix_sock(sock, &discarded_events, sizeof(discarded_events));
1802 if (ret < 0) {
1803 PERROR("send discarded events");
1804 goto error_fatal;
1805 }
1806
1807 break;
1808 }
1809 case LTTNG_CONSUMER_LOST_PACKETS:
1810 {
1811 int ret;
1812 uint64_t lost_packets;
1813 struct lttng_ht_iter iter;
1814 struct lttng_ht *ht;
1815 struct lttng_consumer_stream *stream;
1816 uint64_t id = msg.u.lost_packets.session_id;
1817 uint64_t key = msg.u.lost_packets.channel_key;
1818
1819 DBG("UST consumer lost packets command for session id %"
1820 PRIu64, id);
1821 rcu_read_lock();
1822 pthread_mutex_lock(&consumer_data.lock);
1823
1824 ht = consumer_data.stream_list_ht;
1825
1826 /*
1827 * We only need a reference to the channel, but they are not
1828 * directly indexed, so we just use the first matching stream
1829 * to extract the information we need, we default to 0 if not
1830 * found (no packets lost if the channel is not yet in use).
1831 */
1832 lost_packets = 0;
1833 cds_lfht_for_each_entry_duplicate(ht->ht,
1834 ht->hash_fct(&id, lttng_ht_seed),
1835 ht->match_fct, &id,
1836 &iter.iter, stream, node_session_id.node) {
1837 if (stream->chan->key == key) {
1838 lost_packets = stream->chan->lost_packets;
1839 break;
1840 }
1841 }
1842 pthread_mutex_unlock(&consumer_data.lock);
1843 rcu_read_unlock();
1844
1845 DBG("UST consumer lost packets command for session id %"
1846 PRIu64 ", channel key %" PRIu64, id, key);
1847
1848 health_code_update();
1849
1850 /* Send back returned value to session daemon */
1851 ret = lttcomm_send_unix_sock(sock, &lost_packets,
1852 sizeof(lost_packets));
1853 if (ret < 0) {
1854 PERROR("send lost packets");
1855 goto error_fatal;
1856 }
1857
1858 break;
1859 }
1860 default:
1861 break;
1862 }
1863
1864 end_nosignal:
1865 rcu_read_unlock();
1866
1867 health_code_update();
1868
1869 /*
1870 * Return 1 to indicate success since the 0 value can be a socket
1871 * shutdown during the recv() or send() call.
1872 */
1873 return 1;
1874
1875 end_msg_sessiond:
1876 /*
1877 * The returned value here is not useful since either way we'll return 1 to
1878 * the caller because the session daemon socket management is done
1879 * elsewhere. Returning a negative code or 0 will shutdown the consumer.
1880 */
1881 ret = consumer_send_status_msg(sock, ret_code);
1882 if (ret < 0) {
1883 goto error_fatal;
1884 }
1885 rcu_read_unlock();
1886
1887 health_code_update();
1888
1889 return 1;
1890 end_channel_error:
1891 if (channel) {
1892 /*
1893 * Free channel here since no one has a reference to it. We don't
1894 * free after that because a stream can store this pointer.
1895 */
1896 destroy_channel(channel);
1897 }
1898 /* We have to send a status channel message indicating an error. */
1899 ret = consumer_send_status_channel(sock, NULL);
1900 if (ret < 0) {
1901 /* Stop everything if session daemon can not be notified. */
1902 goto error_fatal;
1903 }
1904 rcu_read_unlock();
1905
1906 health_code_update();
1907
1908 return 1;
1909 error_fatal:
1910 rcu_read_unlock();
1911 /* This will issue a consumer stop. */
1912 return -1;
1913 }
1914
1915 /*
1916 * Wrapper over the mmap() read offset from ust-ctl library. Since this can be
1917 * compiled out, we isolate it in this library.
1918 */
1919 int lttng_ustctl_get_mmap_read_offset(struct lttng_consumer_stream *stream,
1920 unsigned long *off)
1921 {
1922 assert(stream);
1923 assert(stream->ustream);
1924
1925 return ustctl_get_mmap_read_offset(stream->ustream, off);
1926 }
1927
1928 /*
1929 * Wrapper over the mmap() read offset from ust-ctl library. Since this can be
1930 * compiled out, we isolate it in this library.
1931 */
1932 void *lttng_ustctl_get_mmap_base(struct lttng_consumer_stream *stream)
1933 {
1934 assert(stream);
1935 assert(stream->ustream);
1936
1937 return ustctl_get_mmap_base(stream->ustream);
1938 }
1939
1940 /*
1941 * Take a snapshot for a specific fd
1942 *
1943 * Returns 0 on success, < 0 on error
1944 */
1945 int lttng_ustconsumer_take_snapshot(struct lttng_consumer_stream *stream)
1946 {
1947 assert(stream);
1948 assert(stream->ustream);
1949
1950 return ustctl_snapshot(stream->ustream);
1951 }
1952
1953 /*
1954 * Get the produced position
1955 *
1956 * Returns 0 on success, < 0 on error
1957 */
1958 int lttng_ustconsumer_get_produced_snapshot(
1959 struct lttng_consumer_stream *stream, unsigned long *pos)
1960 {
1961 assert(stream);
1962 assert(stream->ustream);
1963 assert(pos);
1964
1965 return ustctl_snapshot_get_produced(stream->ustream, pos);
1966 }
1967
1968 /*
1969 * Get the consumed position
1970 *
1971 * Returns 0 on success, < 0 on error
1972 */
1973 int lttng_ustconsumer_get_consumed_snapshot(
1974 struct lttng_consumer_stream *stream, unsigned long *pos)
1975 {
1976 assert(stream);
1977 assert(stream->ustream);
1978 assert(pos);
1979
1980 return ustctl_snapshot_get_consumed(stream->ustream, pos);
1981 }
1982
1983 void lttng_ustconsumer_flush_buffer(struct lttng_consumer_stream *stream,
1984 int producer)
1985 {
1986 assert(stream);
1987 assert(stream->ustream);
1988
1989 ustctl_flush_buffer(stream->ustream, producer);
1990 }
1991
1992 int lttng_ustconsumer_get_current_timestamp(
1993 struct lttng_consumer_stream *stream, uint64_t *ts)
1994 {
1995 assert(stream);
1996 assert(stream->ustream);
1997 assert(ts);
1998
1999 return ustctl_get_current_timestamp(stream->ustream, ts);
2000 }
2001
2002 int lttng_ustconsumer_get_sequence_number(
2003 struct lttng_consumer_stream *stream, uint64_t *seq)
2004 {
2005 assert(stream);
2006 assert(stream->ustream);
2007 assert(seq);
2008
2009 return ustctl_get_sequence_number(stream->ustream, seq);
2010 }
2011
2012 /*
2013 * Called when the stream signals the consumer that it has hung up.
2014 */
2015 void lttng_ustconsumer_on_stream_hangup(struct lttng_consumer_stream *stream)
2016 {
2017 assert(stream);
2018 assert(stream->ustream);
2019
2020 pthread_mutex_lock(&stream->lock);
2021 if (!stream->quiescent) {
2022 ustctl_flush_buffer(stream->ustream, 0);
2023 stream->quiescent = true;
2024 }
2025 pthread_mutex_unlock(&stream->lock);
2026 stream->hangup_flush_done = 1;
2027 }
2028
2029 void lttng_ustconsumer_del_channel(struct lttng_consumer_channel *chan)
2030 {
2031 int i;
2032
2033 assert(chan);
2034 assert(chan->uchan);
2035
2036 if (chan->switch_timer_enabled == 1) {
2037 consumer_timer_switch_stop(chan);
2038 }
2039 for (i = 0; i < chan->nr_stream_fds; i++) {
2040 int ret;
2041
2042 ret = close(chan->stream_fds[i]);
2043 if (ret) {
2044 PERROR("close");
2045 }
2046 if (chan->shm_path[0]) {
2047 char shm_path[PATH_MAX];
2048
2049 ret = get_stream_shm_path(shm_path, chan->shm_path, i);
2050 if (ret) {
2051 ERR("Cannot get stream shm path");
2052 }
2053 ret = run_as_unlink(shm_path, chan->uid, chan->gid);
2054 if (ret) {
2055 PERROR("unlink %s", shm_path);
2056 }
2057 }
2058 }
2059 }
2060
2061 void lttng_ustconsumer_free_channel(struct lttng_consumer_channel *chan)
2062 {
2063 assert(chan);
2064 assert(chan->uchan);
2065
2066 consumer_metadata_cache_destroy(chan);
2067 ustctl_destroy_channel(chan->uchan);
2068 /* Try to rmdir all directories under shm_path root. */
2069 if (chan->root_shm_path[0]) {
2070 (void) run_as_recursive_rmdir(chan->root_shm_path,
2071 chan->uid, chan->gid);
2072 }
2073 free(chan->stream_fds);
2074 }
2075
2076 void lttng_ustconsumer_del_stream(struct lttng_consumer_stream *stream)
2077 {
2078 assert(stream);
2079 assert(stream->ustream);
2080
2081 if (stream->chan->switch_timer_enabled == 1) {
2082 consumer_timer_switch_stop(stream->chan);
2083 }
2084 ustctl_destroy_stream(stream->ustream);
2085 }
2086
2087 int lttng_ustconsumer_get_wakeup_fd(struct lttng_consumer_stream *stream)
2088 {
2089 assert(stream);
2090 assert(stream->ustream);
2091
2092 return ustctl_stream_get_wakeup_fd(stream->ustream);
2093 }
2094
2095 int lttng_ustconsumer_close_wakeup_fd(struct lttng_consumer_stream *stream)
2096 {
2097 assert(stream);
2098 assert(stream->ustream);
2099
2100 return ustctl_stream_close_wakeup_fd(stream->ustream);
2101 }
2102
2103 /*
2104 * Populate index values of a UST stream. Values are set in big endian order.
2105 *
2106 * Return 0 on success or else a negative value.
2107 */
2108 static int get_index_values(struct ctf_packet_index *index,
2109 struct ustctl_consumer_stream *ustream)
2110 {
2111 int ret;
2112
2113 ret = ustctl_get_timestamp_begin(ustream, &index->timestamp_begin);
2114 if (ret < 0) {
2115 PERROR("ustctl_get_timestamp_begin");
2116 goto error;
2117 }
2118 index->timestamp_begin = htobe64(index->timestamp_begin);
2119
2120 ret = ustctl_get_timestamp_end(ustream, &index->timestamp_end);
2121 if (ret < 0) {
2122 PERROR("ustctl_get_timestamp_end");
2123 goto error;
2124 }
2125 index->timestamp_end = htobe64(index->timestamp_end);
2126
2127 ret = ustctl_get_events_discarded(ustream, &index->events_discarded);
2128 if (ret < 0) {
2129 PERROR("ustctl_get_events_discarded");
2130 goto error;
2131 }
2132 index->events_discarded = htobe64(index->events_discarded);
2133
2134 ret = ustctl_get_content_size(ustream, &index->content_size);
2135 if (ret < 0) {
2136 PERROR("ustctl_get_content_size");
2137 goto error;
2138 }
2139 index->content_size = htobe64(index->content_size);
2140
2141 ret = ustctl_get_packet_size(ustream, &index->packet_size);
2142 if (ret < 0) {
2143 PERROR("ustctl_get_packet_size");
2144 goto error;
2145 }
2146 index->packet_size = htobe64(index->packet_size);
2147
2148 ret = ustctl_get_stream_id(ustream, &index->stream_id);
2149 if (ret < 0) {
2150 PERROR("ustctl_get_stream_id");
2151 goto error;
2152 }
2153 index->stream_id = htobe64(index->stream_id);
2154
2155 ret = ustctl_get_instance_id(ustream, &index->stream_instance_id);
2156 if (ret < 0) {
2157 PERROR("ustctl_get_instance_id");
2158 goto error;
2159 }
2160 index->stream_instance_id = htobe64(index->stream_instance_id);
2161
2162 ret = ustctl_get_sequence_number(ustream, &index->packet_seq_num);
2163 if (ret < 0) {
2164 PERROR("ustctl_get_sequence_number");
2165 goto error;
2166 }
2167 index->packet_seq_num = htobe64(index->packet_seq_num);
2168
2169 error:
2170 return ret;
2171 }
2172
2173 static
2174 void metadata_stream_reset_cache(struct lttng_consumer_stream *stream,
2175 struct consumer_metadata_cache *cache)
2176 {
2177 DBG("Metadata stream update to version %" PRIu64,
2178 cache->version);
2179 stream->ust_metadata_pushed = 0;
2180 stream->metadata_version = cache->version;
2181 stream->reset_metadata_flag = 1;
2182 }
2183
2184 /*
2185 * Check if the version of the metadata stream and metadata cache match.
2186 * If the cache got updated, reset the metadata stream.
2187 * The stream lock and metadata cache lock MUST be held.
2188 * Return 0 on success, a negative value on error.
2189 */
2190 static
2191 int metadata_stream_check_version(struct lttng_consumer_stream *stream)
2192 {
2193 int ret = 0;
2194 struct consumer_metadata_cache *cache = stream->chan->metadata_cache;
2195
2196 if (cache->version == stream->metadata_version) {
2197 goto end;
2198 }
2199 metadata_stream_reset_cache(stream, cache);
2200
2201 end:
2202 return ret;
2203 }
2204
2205 /*
2206 * Write up to one packet from the metadata cache to the channel.
2207 *
2208 * Returns the number of bytes pushed in the cache, or a negative value
2209 * on error.
2210 */
2211 static
2212 int commit_one_metadata_packet(struct lttng_consumer_stream *stream)
2213 {
2214 ssize_t write_len;
2215 int ret;
2216
2217 pthread_mutex_lock(&stream->chan->metadata_cache->lock);
2218 ret = metadata_stream_check_version(stream);
2219 if (ret < 0) {
2220 goto end;
2221 }
2222 if (stream->chan->metadata_cache->max_offset
2223 == stream->ust_metadata_pushed) {
2224 ret = 0;
2225 goto end;
2226 }
2227
2228 write_len = ustctl_write_one_packet_to_channel(stream->chan->uchan,
2229 &stream->chan->metadata_cache->data[stream->ust_metadata_pushed],
2230 stream->chan->metadata_cache->max_offset
2231 - stream->ust_metadata_pushed);
2232 assert(write_len != 0);
2233 if (write_len < 0) {
2234 ERR("Writing one metadata packet");
2235 ret = -1;
2236 goto end;
2237 }
2238 stream->ust_metadata_pushed += write_len;
2239
2240 assert(stream->chan->metadata_cache->max_offset >=
2241 stream->ust_metadata_pushed);
2242 ret = write_len;
2243
2244 end:
2245 pthread_mutex_unlock(&stream->chan->metadata_cache->lock);
2246 return ret;
2247 }
2248
2249
2250 /*
2251 * Sync metadata meaning request them to the session daemon and snapshot to the
2252 * metadata thread can consumer them.
2253 *
2254 * Metadata stream lock is held here, but we need to release it when
2255 * interacting with sessiond, else we cause a deadlock with live
2256 * awaiting on metadata to be pushed out.
2257 *
2258 * Return 0 if new metadatda is available, EAGAIN if the metadata stream
2259 * is empty or a negative value on error.
2260 */
2261 int lttng_ustconsumer_sync_metadata(struct lttng_consumer_local_data *ctx,
2262 struct lttng_consumer_stream *metadata)
2263 {
2264 int ret;
2265 int retry = 0;
2266
2267 assert(ctx);
2268 assert(metadata);
2269
2270 pthread_mutex_unlock(&metadata->lock);
2271 /*
2272 * Request metadata from the sessiond, but don't wait for the flush
2273 * because we locked the metadata thread.
2274 */
2275 ret = lttng_ustconsumer_request_metadata(ctx, metadata->chan, 0, 0);
2276 if (ret < 0) {
2277 goto end;
2278 }
2279 pthread_mutex_lock(&metadata->lock);
2280
2281 ret = commit_one_metadata_packet(metadata);
2282 if (ret <= 0) {
2283 goto end;
2284 } else if (ret > 0) {
2285 retry = 1;
2286 }
2287
2288 ustctl_flush_buffer(metadata->ustream, 1);
2289 ret = ustctl_snapshot(metadata->ustream);
2290 if (ret < 0) {
2291 if (errno != EAGAIN) {
2292 ERR("Sync metadata, taking UST snapshot");
2293 goto end;
2294 }
2295 DBG("No new metadata when syncing them.");
2296 /* No new metadata, exit. */
2297 ret = ENODATA;
2298 goto end;
2299 }
2300
2301 /*
2302 * After this flush, we still need to extract metadata.
2303 */
2304 if (retry) {
2305 ret = EAGAIN;
2306 }
2307
2308 end:
2309 return ret;
2310 }
2311
2312 /*
2313 * Return 0 on success else a negative value.
2314 */
2315 static int notify_if_more_data(struct lttng_consumer_stream *stream,
2316 struct lttng_consumer_local_data *ctx)
2317 {
2318 int ret;
2319 struct ustctl_consumer_stream *ustream;
2320
2321 assert(stream);
2322 assert(ctx);
2323
2324 ustream = stream->ustream;
2325
2326 /*
2327 * First, we are going to check if there is a new subbuffer available
2328 * before reading the stream wait_fd.
2329 */
2330 /* Get the next subbuffer */
2331 ret = ustctl_get_next_subbuf(ustream);
2332 if (ret) {
2333 /* No more data found, flag the stream. */
2334 stream->has_data = 0;
2335 ret = 0;
2336 goto end;
2337 }
2338
2339 ret = ustctl_put_subbuf(ustream);
2340 assert(!ret);
2341
2342 /* This stream still has data. Flag it and wake up the data thread. */
2343 stream->has_data = 1;
2344
2345 if (stream->monitor && !stream->hangup_flush_done && !ctx->has_wakeup) {
2346 ssize_t writelen;
2347
2348 writelen = lttng_pipe_write(ctx->consumer_wakeup_pipe, "!", 1);
2349 if (writelen < 0 && errno != EAGAIN && errno != EWOULDBLOCK) {
2350 ret = writelen;
2351 goto end;
2352 }
2353
2354 /* The wake up pipe has been notified. */
2355 ctx->has_wakeup = 1;
2356 }
2357 ret = 0;
2358
2359 end:
2360 return ret;
2361 }
2362
2363 static
2364 int update_stream_stats(struct lttng_consumer_stream *stream)
2365 {
2366 int ret;
2367 uint64_t seq, discarded;
2368
2369 ret = ustctl_get_sequence_number(stream->ustream, &seq);
2370 if (ret < 0) {
2371 PERROR("ustctl_get_sequence_number");
2372 goto end;
2373 }
2374 /*
2375 * Start the sequence when we extract the first packet in case we don't
2376 * start at 0 (for example if a consumer is not connected to the
2377 * session immediately after the beginning).
2378 */
2379 if (stream->last_sequence_number == -1ULL) {
2380 stream->last_sequence_number = seq;
2381 } else if (seq > stream->last_sequence_number) {
2382 stream->chan->lost_packets += seq -
2383 stream->last_sequence_number - 1;
2384 } else {
2385 /* seq <= last_sequence_number */
2386 ERR("Sequence number inconsistent : prev = %" PRIu64
2387 ", current = %" PRIu64,
2388 stream->last_sequence_number, seq);
2389 ret = -1;
2390 goto end;
2391 }
2392 stream->last_sequence_number = seq;
2393
2394 ret = ustctl_get_events_discarded(stream->ustream, &discarded);
2395 if (ret < 0) {
2396 PERROR("kernctl_get_events_discarded");
2397 goto end;
2398 }
2399 if (discarded < stream->last_discarded_events) {
2400 /*
2401 * Overflow has occurred. We assume only one wrap-around
2402 * has occurred.
2403 */
2404 stream->chan->discarded_events +=
2405 (1ULL << (CAA_BITS_PER_LONG - 1)) -
2406 stream->last_discarded_events + discarded;
2407 } else {
2408 stream->chan->discarded_events += discarded -
2409 stream->last_discarded_events;
2410 }
2411 stream->last_discarded_events = discarded;
2412 ret = 0;
2413
2414 end:
2415 return ret;
2416 }
2417
2418 /*
2419 * Read subbuffer from the given stream.
2420 *
2421 * Stream lock MUST be acquired.
2422 *
2423 * Return 0 on success else a negative value.
2424 */
2425 int lttng_ustconsumer_read_subbuffer(struct lttng_consumer_stream *stream,
2426 struct lttng_consumer_local_data *ctx)
2427 {
2428 unsigned long len, subbuf_size, padding;
2429 int err, write_index = 1;
2430 long ret = 0;
2431 struct ustctl_consumer_stream *ustream;
2432 struct ctf_packet_index index;
2433
2434 assert(stream);
2435 assert(stream->ustream);
2436 assert(ctx);
2437
2438 DBG("In UST read_subbuffer (wait_fd: %d, name: %s)", stream->wait_fd,
2439 stream->name);
2440
2441 /* Ease our life for what's next. */
2442 ustream = stream->ustream;
2443
2444 /*
2445 * We can consume the 1 byte written into the wait_fd by UST. Don't trigger
2446 * error if we cannot read this one byte (read returns 0), or if the error
2447 * is EAGAIN or EWOULDBLOCK.
2448 *
2449 * This is only done when the stream is monitored by a thread, before the
2450 * flush is done after a hangup and if the stream is not flagged with data
2451 * since there might be nothing to consume in the wait fd but still have
2452 * data available flagged by the consumer wake up pipe.
2453 */
2454 if (stream->monitor && !stream->hangup_flush_done && !stream->has_data) {
2455 char dummy;
2456 ssize_t readlen;
2457
2458 readlen = lttng_read(stream->wait_fd, &dummy, 1);
2459 if (readlen < 0 && errno != EAGAIN && errno != EWOULDBLOCK) {
2460 ret = readlen;
2461 goto end;
2462 }
2463 }
2464
2465 retry:
2466 /* Get the next subbuffer */
2467 err = ustctl_get_next_subbuf(ustream);
2468 if (err != 0) {
2469 /*
2470 * Populate metadata info if the existing info has
2471 * already been read.
2472 */
2473 if (stream->metadata_flag) {
2474 ret = commit_one_metadata_packet(stream);
2475 if (ret <= 0) {
2476 goto end;
2477 }
2478 ustctl_flush_buffer(stream->ustream, 1);
2479 goto retry;
2480 }
2481
2482 ret = err; /* ustctl_get_next_subbuf returns negative, caller expect positive. */
2483 /*
2484 * This is a debug message even for single-threaded consumer,
2485 * because poll() have more relaxed criterions than get subbuf,
2486 * so get_subbuf may fail for short race windows where poll()
2487 * would issue wakeups.
2488 */
2489 DBG("Reserving sub buffer failed (everything is normal, "
2490 "it is due to concurrency) [ret: %d]", err);
2491 goto end;
2492 }
2493 assert(stream->chan->output == CONSUMER_CHANNEL_MMAP);
2494
2495 if (!stream->metadata_flag) {
2496 index.offset = htobe64(stream->out_fd_offset);
2497 ret = get_index_values(&index, ustream);
2498 if (ret < 0) {
2499 goto end;
2500 }
2501
2502 /* Update the stream's sequence and discarded events count. */
2503 ret = update_stream_stats(stream);
2504 if (ret < 0) {
2505 PERROR("kernctl_get_events_discarded");
2506 goto end;
2507 }
2508 } else {
2509 write_index = 0;
2510 }
2511
2512 /* Get the full padded subbuffer size */
2513 err = ustctl_get_padded_subbuf_size(ustream, &len);
2514 assert(err == 0);
2515
2516 /* Get subbuffer data size (without padding) */
2517 err = ustctl_get_subbuf_size(ustream, &subbuf_size);
2518 assert(err == 0);
2519
2520 /* Make sure we don't get a subbuffer size bigger than the padded */
2521 assert(len >= subbuf_size);
2522
2523 padding = len - subbuf_size;
2524 /* write the subbuffer to the tracefile */
2525 ret = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, subbuf_size, padding, &index);
2526 /*
2527 * The mmap operation should write subbuf_size amount of data when network
2528 * streaming or the full padding (len) size when we are _not_ streaming.
2529 */
2530 if ((ret != subbuf_size && stream->net_seq_idx != (uint64_t) -1ULL) ||
2531 (ret != len && stream->net_seq_idx == (uint64_t) -1ULL)) {
2532 /*
2533 * Display the error but continue processing to try to release the
2534 * subbuffer. This is a DBG statement since any unexpected kill or
2535 * signal, the application gets unregistered, relayd gets closed or
2536 * anything that affects the buffer lifetime will trigger this error.
2537 * So, for the sake of the user, don't print this error since it can
2538 * happen and it is OK with the code flow.
2539 */
2540 DBG("Error writing to tracefile "
2541 "(ret: %ld != len: %lu != subbuf_size: %lu)",
2542 ret, len, subbuf_size);
2543 write_index = 0;
2544 }
2545 err = ustctl_put_next_subbuf(ustream);
2546 assert(err == 0);
2547
2548 /*
2549 * This will consumer the byte on the wait_fd if and only if there is not
2550 * next subbuffer to be acquired.
2551 */
2552 if (!stream->metadata_flag) {
2553 ret = notify_if_more_data(stream, ctx);
2554 if (ret < 0) {
2555 goto end;
2556 }
2557 }
2558
2559 /* Write index if needed. */
2560 if (!write_index) {
2561 goto end;
2562 }
2563
2564 if (stream->chan->live_timer_interval && !stream->metadata_flag) {
2565 /*
2566 * In live, block until all the metadata is sent.
2567 */
2568 pthread_mutex_lock(&stream->metadata_timer_lock);
2569 assert(!stream->missed_metadata_flush);
2570 stream->waiting_on_metadata = true;
2571 pthread_mutex_unlock(&stream->metadata_timer_lock);
2572
2573 err = consumer_stream_sync_metadata(ctx, stream->session_id);
2574
2575 pthread_mutex_lock(&stream->metadata_timer_lock);
2576 stream->waiting_on_metadata = false;
2577 if (stream->missed_metadata_flush) {
2578 stream->missed_metadata_flush = false;
2579 pthread_mutex_unlock(&stream->metadata_timer_lock);
2580 (void) consumer_flush_ust_index(stream);
2581 } else {
2582 pthread_mutex_unlock(&stream->metadata_timer_lock);
2583 }
2584
2585 if (err < 0) {
2586 goto end;
2587 }
2588 }
2589
2590 assert(!stream->metadata_flag);
2591 err = consumer_stream_write_index(stream, &index);
2592 if (err < 0) {
2593 goto end;
2594 }
2595
2596 end:
2597 return ret;
2598 }
2599
2600 /*
2601 * Called when a stream is created.
2602 *
2603 * Return 0 on success or else a negative value.
2604 */
2605 int lttng_ustconsumer_on_recv_stream(struct lttng_consumer_stream *stream)
2606 {
2607 int ret;
2608
2609 assert(stream);
2610
2611 /* Don't create anything if this is set for streaming. */
2612 if (stream->net_seq_idx == (uint64_t) -1ULL && stream->chan->monitor) {
2613 ret = utils_create_stream_file(stream->chan->pathname, stream->name,
2614 stream->chan->tracefile_size, stream->tracefile_count_current,
2615 stream->uid, stream->gid, NULL);
2616 if (ret < 0) {
2617 goto error;
2618 }
2619 stream->out_fd = ret;
2620 stream->tracefile_size_current = 0;
2621
2622 if (!stream->metadata_flag) {
2623 ret = index_create_file(stream->chan->pathname,
2624 stream->name, stream->uid, stream->gid,
2625 stream->chan->tracefile_size,
2626 stream->tracefile_count_current);
2627 if (ret < 0) {
2628 goto error;
2629 }
2630 stream->index_fd = ret;
2631 }
2632 }
2633 ret = 0;
2634
2635 error:
2636 return ret;
2637 }
2638
2639 /*
2640 * Check if data is still being extracted from the buffers for a specific
2641 * stream. Consumer data lock MUST be acquired before calling this function
2642 * and the stream lock.
2643 *
2644 * Return 1 if the traced data are still getting read else 0 meaning that the
2645 * data is available for trace viewer reading.
2646 */
2647 int lttng_ustconsumer_data_pending(struct lttng_consumer_stream *stream)
2648 {
2649 int ret;
2650
2651 assert(stream);
2652 assert(stream->ustream);
2653
2654 DBG("UST consumer checking data pending");
2655
2656 if (stream->endpoint_status != CONSUMER_ENDPOINT_ACTIVE) {
2657 ret = 0;
2658 goto end;
2659 }
2660
2661 if (stream->chan->type == CONSUMER_CHANNEL_TYPE_METADATA) {
2662 uint64_t contiguous, pushed;
2663
2664 /* Ease our life a bit. */
2665 contiguous = stream->chan->metadata_cache->max_offset;
2666 pushed = stream->ust_metadata_pushed;
2667
2668 /*
2669 * We can simply check whether all contiguously available data
2670 * has been pushed to the ring buffer, since the push operation
2671 * is performed within get_next_subbuf(), and because both
2672 * get_next_subbuf() and put_next_subbuf() are issued atomically
2673 * thanks to the stream lock within
2674 * lttng_ustconsumer_read_subbuffer(). This basically means that
2675 * whetnever ust_metadata_pushed is incremented, the associated
2676 * metadata has been consumed from the metadata stream.
2677 */
2678 DBG("UST consumer metadata pending check: contiguous %" PRIu64 " vs pushed %" PRIu64,
2679 contiguous, pushed);
2680 assert(((int64_t) (contiguous - pushed)) >= 0);
2681 if ((contiguous != pushed) ||
2682 (((int64_t) contiguous - pushed) > 0 || contiguous == 0)) {
2683 ret = 1; /* Data is pending */
2684 goto end;
2685 }
2686 } else {
2687 ret = ustctl_get_next_subbuf(stream->ustream);
2688 if (ret == 0) {
2689 /*
2690 * There is still data so let's put back this
2691 * subbuffer.
2692 */
2693 ret = ustctl_put_subbuf(stream->ustream);
2694 assert(ret == 0);
2695 ret = 1; /* Data is pending */
2696 goto end;
2697 }
2698 }
2699
2700 /* Data is NOT pending so ready to be read. */
2701 ret = 0;
2702
2703 end:
2704 return ret;
2705 }
2706
2707 /*
2708 * Stop a given metadata channel timer if enabled and close the wait fd which
2709 * is the poll pipe of the metadata stream.
2710 *
2711 * This MUST be called with the metadata channel acquired.
2712 */
2713 void lttng_ustconsumer_close_metadata(struct lttng_consumer_channel *metadata)
2714 {
2715 int ret;
2716
2717 assert(metadata);
2718 assert(metadata->type == CONSUMER_CHANNEL_TYPE_METADATA);
2719
2720 DBG("Closing metadata channel key %" PRIu64, metadata->key);
2721
2722 if (metadata->switch_timer_enabled == 1) {
2723 consumer_timer_switch_stop(metadata);
2724 }
2725
2726 if (!metadata->metadata_stream) {
2727 goto end;
2728 }
2729
2730 /*
2731 * Closing write side so the thread monitoring the stream wakes up if any
2732 * and clean the metadata stream.
2733 */
2734 if (metadata->metadata_stream->ust_metadata_poll_pipe[1] >= 0) {
2735 ret = close(metadata->metadata_stream->ust_metadata_poll_pipe[1]);
2736 if (ret < 0) {
2737 PERROR("closing metadata pipe write side");
2738 }
2739 metadata->metadata_stream->ust_metadata_poll_pipe[1] = -1;
2740 }
2741
2742 end:
2743 return;
2744 }
2745
2746 /*
2747 * Close every metadata stream wait fd of the metadata hash table. This
2748 * function MUST be used very carefully so not to run into a race between the
2749 * metadata thread handling streams and this function closing their wait fd.
2750 *
2751 * For UST, this is used when the session daemon hangs up. Its the metadata
2752 * producer so calling this is safe because we are assured that no state change
2753 * can occur in the metadata thread for the streams in the hash table.
2754 */
2755 void lttng_ustconsumer_close_all_metadata(struct lttng_ht *metadata_ht)
2756 {
2757 struct lttng_ht_iter iter;
2758 struct lttng_consumer_stream *stream;
2759
2760 assert(metadata_ht);
2761 assert(metadata_ht->ht);
2762
2763 DBG("UST consumer closing all metadata streams");
2764
2765 rcu_read_lock();
2766 cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream,
2767 node.node) {
2768
2769 health_code_update();
2770
2771 pthread_mutex_lock(&stream->chan->lock);
2772 lttng_ustconsumer_close_metadata(stream->chan);
2773 pthread_mutex_unlock(&stream->chan->lock);
2774
2775 }
2776 rcu_read_unlock();
2777 }
2778
2779 void lttng_ustconsumer_close_stream_wakeup(struct lttng_consumer_stream *stream)
2780 {
2781 int ret;
2782
2783 ret = ustctl_stream_close_wakeup_fd(stream->ustream);
2784 if (ret < 0) {
2785 ERR("Unable to close wakeup fd");
2786 }
2787 }
2788
2789 /*
2790 * Please refer to consumer-timer.c before adding any lock within this
2791 * function or any of its callees. Timers have a very strict locking
2792 * semantic with respect to teardown. Failure to respect this semantic
2793 * introduces deadlocks.
2794 *
2795 * DON'T hold the metadata lock when calling this function, else this
2796 * can cause deadlock involving consumer awaiting for metadata to be
2797 * pushed out due to concurrent interaction with the session daemon.
2798 */
2799 int lttng_ustconsumer_request_metadata(struct lttng_consumer_local_data *ctx,
2800 struct lttng_consumer_channel *channel, int timer, int wait)
2801 {
2802 struct lttcomm_metadata_request_msg request;
2803 struct lttcomm_consumer_msg msg;
2804 enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS;
2805 uint64_t len, key, offset, version;
2806 int ret;
2807
2808 assert(channel);
2809 assert(channel->metadata_cache);
2810
2811 memset(&request, 0, sizeof(request));
2812
2813 /* send the metadata request to sessiond */
2814 switch (consumer_data.type) {
2815 case LTTNG_CONSUMER64_UST:
2816 request.bits_per_long = 64;
2817 break;
2818 case LTTNG_CONSUMER32_UST:
2819 request.bits_per_long = 32;
2820 break;
2821 default:
2822 request.bits_per_long = 0;
2823 break;
2824 }
2825
2826 request.session_id = channel->session_id;
2827 request.session_id_per_pid = channel->session_id_per_pid;
2828 /*
2829 * Request the application UID here so the metadata of that application can
2830 * be sent back. The channel UID corresponds to the user UID of the session
2831 * used for the rights on the stream file(s).
2832 */
2833 request.uid = channel->ust_app_uid;
2834 request.key = channel->key;
2835
2836 DBG("Sending metadata request to sessiond, session id %" PRIu64
2837 ", per-pid %" PRIu64 ", app UID %u and channek key %" PRIu64,
2838 request.session_id, request.session_id_per_pid, request.uid,
2839 request.key);
2840
2841 pthread_mutex_lock(&ctx->metadata_socket_lock);
2842
2843 health_code_update();
2844
2845 ret = lttcomm_send_unix_sock(ctx->consumer_metadata_socket, &request,
2846 sizeof(request));
2847 if (ret < 0) {
2848 ERR("Asking metadata to sessiond");
2849 goto end;
2850 }
2851
2852 health_code_update();
2853
2854 /* Receive the metadata from sessiond */
2855 ret = lttcomm_recv_unix_sock(ctx->consumer_metadata_socket, &msg,
2856 sizeof(msg));
2857 if (ret != sizeof(msg)) {
2858 DBG("Consumer received unexpected message size %d (expects %zu)",
2859 ret, sizeof(msg));
2860 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD);
2861 /*
2862 * The ret value might 0 meaning an orderly shutdown but this is ok
2863 * since the caller handles this.
2864 */
2865 goto end;
2866 }
2867
2868 health_code_update();
2869
2870 if (msg.cmd_type == LTTNG_ERR_UND) {
2871 /* No registry found */
2872 (void) consumer_send_status_msg(ctx->consumer_metadata_socket,
2873 ret_code);
2874 ret = 0;
2875 goto end;
2876 } else if (msg.cmd_type != LTTNG_CONSUMER_PUSH_METADATA) {
2877 ERR("Unexpected cmd_type received %d", msg.cmd_type);
2878 ret = -1;
2879 goto end;
2880 }
2881
2882 len = msg.u.push_metadata.len;
2883 key = msg.u.push_metadata.key;
2884 offset = msg.u.push_metadata.target_offset;
2885 version = msg.u.push_metadata.version;
2886
2887 assert(key == channel->key);
2888 if (len == 0) {
2889 DBG("No new metadata to receive for key %" PRIu64, key);
2890 }
2891
2892 health_code_update();
2893
2894 /* Tell session daemon we are ready to receive the metadata. */
2895 ret = consumer_send_status_msg(ctx->consumer_metadata_socket,
2896 LTTCOMM_CONSUMERD_SUCCESS);
2897 if (ret < 0 || len == 0) {
2898 /*
2899 * Somehow, the session daemon is not responding anymore or there is
2900 * nothing to receive.
2901 */
2902 goto end;
2903 }
2904
2905 health_code_update();
2906
2907 ret = lttng_ustconsumer_recv_metadata(ctx->consumer_metadata_socket,
2908 key, offset, len, version, channel, timer, wait);
2909 if (ret >= 0) {
2910 /*
2911 * Only send the status msg if the sessiond is alive meaning a positive
2912 * ret code.
2913 */
2914 (void) consumer_send_status_msg(ctx->consumer_metadata_socket, ret);
2915 }
2916 ret = 0;
2917
2918 end:
2919 health_code_update();
2920
2921 pthread_mutex_unlock(&ctx->metadata_socket_lock);
2922 return ret;
2923 }
2924
2925 /*
2926 * Return the ustctl call for the get stream id.
2927 */
2928 int lttng_ustconsumer_get_stream_id(struct lttng_consumer_stream *stream,
2929 uint64_t *stream_id)
2930 {
2931 assert(stream);
2932 assert(stream_id);
2933
2934 return ustctl_get_stream_id(stream->ustream, stream_id);
2935 }
This page took 0.15662 seconds and 3 git commands to generate.