Fix: sessiond vs consumerd push/get metadata deadlock
[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 _GNU_SOURCE
20 #define _LGPL_SOURCE
21 #include <assert.h>
22 #include <lttng/ust-ctl.h>
23 #include <poll.h>
24 #include <pthread.h>
25 #include <stdlib.h>
26 #include <string.h>
27 #include <sys/mman.h>
28 #include <sys/socket.h>
29 #include <sys/stat.h>
30 #include <sys/types.h>
31 #include <inttypes.h>
32 #include <unistd.h>
33 #include <urcu/list.h>
34 #include <signal.h>
35
36 #include <bin/lttng-consumerd/health-consumerd.h>
37 #include <common/common.h>
38 #include <common/sessiond-comm/sessiond-comm.h>
39 #include <common/relayd/relayd.h>
40 #include <common/compat/fcntl.h>
41 #include <common/compat/endian.h>
42 #include <common/consumer-metadata-cache.h>
43 #include <common/consumer-stream.h>
44 #include <common/consumer-timer.h>
45 #include <common/utils.h>
46 #include <common/index/index.h>
47
48 #include "ust-consumer.h"
49
50 #define UINT_MAX_STR_LEN 11 /* includes \0 */
51
52 extern struct lttng_consumer_global_data consumer_data;
53 extern int consumer_poll_timeout;
54 extern volatile int consumer_quit;
55
56 /*
57 * Free channel object and all streams associated with it. This MUST be used
58 * only and only if the channel has _NEVER_ been added to the global channel
59 * hash table.
60 */
61 static void destroy_channel(struct lttng_consumer_channel *channel)
62 {
63 struct lttng_consumer_stream *stream, *stmp;
64
65 assert(channel);
66
67 DBG("UST consumer cleaning stream list");
68
69 cds_list_for_each_entry_safe(stream, stmp, &channel->streams.head,
70 send_node) {
71
72 health_code_update();
73
74 cds_list_del(&stream->send_node);
75 ustctl_destroy_stream(stream->ustream);
76 free(stream);
77 }
78
79 /*
80 * If a channel is available meaning that was created before the streams
81 * were, delete it.
82 */
83 if (channel->uchan) {
84 lttng_ustconsumer_del_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[UINT_MAX_STR_LEN]; /* unsigned 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, UINT_MAX_STR_LEN, "%u", cpu);
256 if (ret != 1) {
257 ret = -1;
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
1033 health_code_update();
1034
1035 /* Lock stream because we are about to change its state. */
1036 pthread_mutex_lock(&stream->lock);
1037 stream->net_seq_idx = relayd_id;
1038
1039 if (use_relayd) {
1040 ret = consumer_send_relayd_stream(stream, path);
1041 if (ret < 0) {
1042 goto error_unlock;
1043 }
1044 } else {
1045 ret = utils_create_stream_file(path, stream->name,
1046 stream->chan->tracefile_size,
1047 stream->tracefile_count_current,
1048 stream->uid, stream->gid, NULL);
1049 if (ret < 0) {
1050 goto error_unlock;
1051 }
1052 stream->out_fd = ret;
1053 stream->tracefile_size_current = 0;
1054
1055 DBG("UST consumer snapshot stream %s/%s (%" PRIu64 ")", path,
1056 stream->name, stream->key);
1057 }
1058 if (relayd_id != -1ULL) {
1059 ret = consumer_send_relayd_streams_sent(relayd_id);
1060 if (ret < 0) {
1061 goto error_unlock;
1062 }
1063 }
1064
1065 ustctl_flush_buffer(stream->ustream, 1);
1066
1067 ret = lttng_ustconsumer_take_snapshot(stream);
1068 if (ret < 0) {
1069 ERR("Taking UST snapshot");
1070 goto error_unlock;
1071 }
1072
1073 ret = lttng_ustconsumer_get_produced_snapshot(stream, &produced_pos);
1074 if (ret < 0) {
1075 ERR("Produced UST snapshot position");
1076 goto error_unlock;
1077 }
1078
1079 ret = lttng_ustconsumer_get_consumed_snapshot(stream, &consumed_pos);
1080 if (ret < 0) {
1081 ERR("Consumerd UST snapshot position");
1082 goto error_unlock;
1083 }
1084
1085 /*
1086 * The original value is sent back if max stream size is larger than
1087 * the possible size of the snapshot. Also, we assume that the session
1088 * daemon should never send a maximum stream size that is lower than
1089 * subbuffer size.
1090 */
1091 consumed_pos = consumer_get_consume_start_pos(consumed_pos,
1092 produced_pos, nb_packets_per_stream,
1093 stream->max_sb_size);
1094
1095 while (consumed_pos < produced_pos) {
1096 ssize_t read_len;
1097 unsigned long len, padded_len;
1098
1099 health_code_update();
1100
1101 DBG("UST consumer taking snapshot at pos %lu", consumed_pos);
1102
1103 ret = ustctl_get_subbuf(stream->ustream, &consumed_pos);
1104 if (ret < 0) {
1105 if (ret != -EAGAIN) {
1106 PERROR("ustctl_get_subbuf snapshot");
1107 goto error_close_stream;
1108 }
1109 DBG("UST consumer get subbuf failed. Skipping it.");
1110 consumed_pos += stream->max_sb_size;
1111 continue;
1112 }
1113
1114 ret = ustctl_get_subbuf_size(stream->ustream, &len);
1115 if (ret < 0) {
1116 ERR("Snapshot ustctl_get_subbuf_size");
1117 goto error_put_subbuf;
1118 }
1119
1120 ret = ustctl_get_padded_subbuf_size(stream->ustream, &padded_len);
1121 if (ret < 0) {
1122 ERR("Snapshot ustctl_get_padded_subbuf_size");
1123 goto error_put_subbuf;
1124 }
1125
1126 read_len = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, len,
1127 padded_len - len, NULL);
1128 if (use_relayd) {
1129 if (read_len != len) {
1130 ret = -EPERM;
1131 goto error_put_subbuf;
1132 }
1133 } else {
1134 if (read_len != padded_len) {
1135 ret = -EPERM;
1136 goto error_put_subbuf;
1137 }
1138 }
1139
1140 ret = ustctl_put_subbuf(stream->ustream);
1141 if (ret < 0) {
1142 ERR("Snapshot ustctl_put_subbuf");
1143 goto error_close_stream;
1144 }
1145 consumed_pos += stream->max_sb_size;
1146 }
1147
1148 /* Simply close the stream so we can use it on the next snapshot. */
1149 consumer_stream_close(stream);
1150 pthread_mutex_unlock(&stream->lock);
1151 }
1152
1153 rcu_read_unlock();
1154 return 0;
1155
1156 error_put_subbuf:
1157 if (ustctl_put_subbuf(stream->ustream) < 0) {
1158 ERR("Snapshot ustctl_put_subbuf");
1159 }
1160 error_close_stream:
1161 consumer_stream_close(stream);
1162 error_unlock:
1163 pthread_mutex_unlock(&stream->lock);
1164 error:
1165 rcu_read_unlock();
1166 return ret;
1167 }
1168
1169 /*
1170 * Receive the metadata updates from the sessiond. Supports receiving
1171 * overlapping metadata, but is needs to always belong to a contiguous
1172 * range starting from 0.
1173 * Be careful about the locks held when calling this function: it needs
1174 * the metadata cache flush to concurrently progress in order to
1175 * complete.
1176 */
1177 int lttng_ustconsumer_recv_metadata(int sock, uint64_t key, uint64_t offset,
1178 uint64_t len, struct lttng_consumer_channel *channel,
1179 int timer, int wait)
1180 {
1181 int ret, ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1182 char *metadata_str;
1183
1184 DBG("UST consumer push metadata key %" PRIu64 " of len %" PRIu64, key, len);
1185
1186 metadata_str = zmalloc(len * sizeof(char));
1187 if (!metadata_str) {
1188 PERROR("zmalloc metadata string");
1189 ret_code = LTTCOMM_CONSUMERD_ENOMEM;
1190 goto end;
1191 }
1192
1193 health_code_update();
1194
1195 /* Receive metadata string. */
1196 ret = lttcomm_recv_unix_sock(sock, metadata_str, len);
1197 if (ret < 0) {
1198 /* Session daemon is dead so return gracefully. */
1199 ret_code = ret;
1200 goto end_free;
1201 }
1202
1203 health_code_update();
1204
1205 pthread_mutex_lock(&channel->metadata_cache->lock);
1206 ret = consumer_metadata_cache_write(channel, offset, len, metadata_str);
1207 if (ret < 0) {
1208 /* Unable to handle metadata. Notify session daemon. */
1209 ret_code = LTTCOMM_CONSUMERD_ERROR_METADATA;
1210 /*
1211 * Skip metadata flush on write error since the offset and len might
1212 * not have been updated which could create an infinite loop below when
1213 * waiting for the metadata cache to be flushed.
1214 */
1215 pthread_mutex_unlock(&channel->metadata_cache->lock);
1216 goto end_free;
1217 }
1218 pthread_mutex_unlock(&channel->metadata_cache->lock);
1219
1220 if (!wait) {
1221 goto end_free;
1222 }
1223 while (consumer_metadata_cache_flushed(channel, offset + len, timer)) {
1224 DBG("Waiting for metadata to be flushed");
1225
1226 health_code_update();
1227
1228 usleep(DEFAULT_METADATA_AVAILABILITY_WAIT_TIME);
1229 }
1230
1231 end_free:
1232 free(metadata_str);
1233 end:
1234 return ret_code;
1235 }
1236
1237 /*
1238 * Receive command from session daemon and process it.
1239 *
1240 * Return 1 on success else a negative value or 0.
1241 */
1242 int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
1243 int sock, struct pollfd *consumer_sockpoll)
1244 {
1245 ssize_t ret;
1246 enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1247 struct lttcomm_consumer_msg msg;
1248 struct lttng_consumer_channel *channel = NULL;
1249
1250 health_code_update();
1251
1252 ret = lttcomm_recv_unix_sock(sock, &msg, sizeof(msg));
1253 if (ret != sizeof(msg)) {
1254 DBG("Consumer received unexpected message size %zd (expects %zu)",
1255 ret, sizeof(msg));
1256 /*
1257 * The ret value might 0 meaning an orderly shutdown but this is ok
1258 * since the caller handles this.
1259 */
1260 if (ret > 0) {
1261 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD);
1262 ret = -1;
1263 }
1264 return ret;
1265 }
1266
1267 health_code_update();
1268
1269 /* deprecated */
1270 assert(msg.cmd_type != LTTNG_CONSUMER_STOP);
1271
1272 health_code_update();
1273
1274 /* relayd needs RCU read-side lock */
1275 rcu_read_lock();
1276
1277 switch (msg.cmd_type) {
1278 case LTTNG_CONSUMER_ADD_RELAYD_SOCKET:
1279 {
1280 /* Session daemon status message are handled in the following call. */
1281 ret = consumer_add_relayd_socket(msg.u.relayd_sock.net_index,
1282 msg.u.relayd_sock.type, ctx, sock, consumer_sockpoll,
1283 &msg.u.relayd_sock.sock, msg.u.relayd_sock.session_id,
1284 msg.u.relayd_sock.relayd_session_id);
1285 goto end_nosignal;
1286 }
1287 case LTTNG_CONSUMER_DESTROY_RELAYD:
1288 {
1289 uint64_t index = msg.u.destroy_relayd.net_seq_idx;
1290 struct consumer_relayd_sock_pair *relayd;
1291
1292 DBG("UST consumer destroying relayd %" PRIu64, index);
1293
1294 /* Get relayd reference if exists. */
1295 relayd = consumer_find_relayd(index);
1296 if (relayd == NULL) {
1297 DBG("Unable to find relayd %" PRIu64, index);
1298 ret_code = LTTCOMM_CONSUMERD_RELAYD_FAIL;
1299 }
1300
1301 /*
1302 * Each relayd socket pair has a refcount of stream attached to it
1303 * which tells if the relayd is still active or not depending on the
1304 * refcount value.
1305 *
1306 * This will set the destroy flag of the relayd object and destroy it
1307 * if the refcount reaches zero when called.
1308 *
1309 * The destroy can happen either here or when a stream fd hangs up.
1310 */
1311 if (relayd) {
1312 consumer_flag_relayd_for_destroy(relayd);
1313 }
1314
1315 goto end_msg_sessiond;
1316 }
1317 case LTTNG_CONSUMER_UPDATE_STREAM:
1318 {
1319 rcu_read_unlock();
1320 return -ENOSYS;
1321 }
1322 case LTTNG_CONSUMER_DATA_PENDING:
1323 {
1324 int ret, is_data_pending;
1325 uint64_t id = msg.u.data_pending.session_id;
1326
1327 DBG("UST consumer data pending command for id %" PRIu64, id);
1328
1329 is_data_pending = consumer_data_pending(id);
1330
1331 /* Send back returned value to session daemon */
1332 ret = lttcomm_send_unix_sock(sock, &is_data_pending,
1333 sizeof(is_data_pending));
1334 if (ret < 0) {
1335 DBG("Error when sending the data pending ret code: %d", ret);
1336 goto error_fatal;
1337 }
1338
1339 /*
1340 * No need to send back a status message since the data pending
1341 * returned value is the response.
1342 */
1343 break;
1344 }
1345 case LTTNG_CONSUMER_ASK_CHANNEL_CREATION:
1346 {
1347 int ret;
1348 struct ustctl_consumer_channel_attr attr;
1349
1350 /* Create a plain object and reserve a channel key. */
1351 channel = allocate_channel(msg.u.ask_channel.session_id,
1352 msg.u.ask_channel.pathname, msg.u.ask_channel.name,
1353 msg.u.ask_channel.uid, msg.u.ask_channel.gid,
1354 msg.u.ask_channel.relayd_id, msg.u.ask_channel.key,
1355 (enum lttng_event_output) msg.u.ask_channel.output,
1356 msg.u.ask_channel.tracefile_size,
1357 msg.u.ask_channel.tracefile_count,
1358 msg.u.ask_channel.session_id_per_pid,
1359 msg.u.ask_channel.monitor,
1360 msg.u.ask_channel.live_timer_interval,
1361 msg.u.ask_channel.root_shm_path,
1362 msg.u.ask_channel.shm_path);
1363 if (!channel) {
1364 goto end_channel_error;
1365 }
1366
1367 /*
1368 * Assign UST application UID to the channel. This value is ignored for
1369 * per PID buffers. This is specific to UST thus setting this after the
1370 * allocation.
1371 */
1372 channel->ust_app_uid = msg.u.ask_channel.ust_app_uid;
1373
1374 /* Build channel attributes from received message. */
1375 attr.subbuf_size = msg.u.ask_channel.subbuf_size;
1376 attr.num_subbuf = msg.u.ask_channel.num_subbuf;
1377 attr.overwrite = msg.u.ask_channel.overwrite;
1378 attr.switch_timer_interval = msg.u.ask_channel.switch_timer_interval;
1379 attr.read_timer_interval = msg.u.ask_channel.read_timer_interval;
1380 attr.chan_id = msg.u.ask_channel.chan_id;
1381 memcpy(attr.uuid, msg.u.ask_channel.uuid, sizeof(attr.uuid));
1382
1383 /* Match channel buffer type to the UST abi. */
1384 switch (msg.u.ask_channel.output) {
1385 case LTTNG_EVENT_MMAP:
1386 default:
1387 attr.output = LTTNG_UST_MMAP;
1388 break;
1389 }
1390
1391 /* Translate and save channel type. */
1392 switch (msg.u.ask_channel.type) {
1393 case LTTNG_UST_CHAN_PER_CPU:
1394 channel->type = CONSUMER_CHANNEL_TYPE_DATA;
1395 attr.type = LTTNG_UST_CHAN_PER_CPU;
1396 /*
1397 * Set refcount to 1 for owner. Below, we will
1398 * pass ownership to the
1399 * consumer_thread_channel_poll() thread.
1400 */
1401 channel->refcount = 1;
1402 break;
1403 case LTTNG_UST_CHAN_METADATA:
1404 channel->type = CONSUMER_CHANNEL_TYPE_METADATA;
1405 attr.type = LTTNG_UST_CHAN_METADATA;
1406 break;
1407 default:
1408 assert(0);
1409 goto error_fatal;
1410 };
1411
1412 health_code_update();
1413
1414 ret = ask_channel(ctx, sock, channel, &attr);
1415 if (ret < 0) {
1416 goto end_channel_error;
1417 }
1418
1419 if (msg.u.ask_channel.type == LTTNG_UST_CHAN_METADATA) {
1420 ret = consumer_metadata_cache_allocate(channel);
1421 if (ret < 0) {
1422 ERR("Allocating metadata cache");
1423 goto end_channel_error;
1424 }
1425 consumer_timer_switch_start(channel, attr.switch_timer_interval);
1426 attr.switch_timer_interval = 0;
1427 } else {
1428 consumer_timer_live_start(channel,
1429 msg.u.ask_channel.live_timer_interval);
1430 }
1431
1432 health_code_update();
1433
1434 /*
1435 * Add the channel to the internal state AFTER all streams were created
1436 * and successfully sent to session daemon. This way, all streams must
1437 * be ready before this channel is visible to the threads.
1438 * If add_channel succeeds, ownership of the channel is
1439 * passed to consumer_thread_channel_poll().
1440 */
1441 ret = add_channel(channel, ctx);
1442 if (ret < 0) {
1443 if (msg.u.ask_channel.type == LTTNG_UST_CHAN_METADATA) {
1444 if (channel->switch_timer_enabled == 1) {
1445 consumer_timer_switch_stop(channel);
1446 }
1447 consumer_metadata_cache_destroy(channel);
1448 }
1449 if (channel->live_timer_enabled == 1) {
1450 consumer_timer_live_stop(channel);
1451 }
1452 goto end_channel_error;
1453 }
1454
1455 health_code_update();
1456
1457 /*
1458 * Channel and streams are now created. Inform the session daemon that
1459 * everything went well and should wait to receive the channel and
1460 * streams with ustctl API.
1461 */
1462 ret = consumer_send_status_channel(sock, channel);
1463 if (ret < 0) {
1464 /*
1465 * There is probably a problem on the socket.
1466 */
1467 goto error_fatal;
1468 }
1469
1470 break;
1471 }
1472 case LTTNG_CONSUMER_GET_CHANNEL:
1473 {
1474 int ret, relayd_err = 0;
1475 uint64_t key = msg.u.get_channel.key;
1476 struct lttng_consumer_channel *channel;
1477
1478 channel = consumer_find_channel(key);
1479 if (!channel) {
1480 ERR("UST consumer get channel key %" PRIu64 " not found", key);
1481 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
1482 goto end_msg_sessiond;
1483 }
1484
1485 health_code_update();
1486
1487 /* Send everything to sessiond. */
1488 ret = send_sessiond_channel(sock, channel, ctx, &relayd_err);
1489 if (ret < 0) {
1490 if (relayd_err) {
1491 /*
1492 * We were unable to send to the relayd the stream so avoid
1493 * sending back a fatal error to the thread since this is OK
1494 * and the consumer can continue its work. The above call
1495 * has sent the error status message to the sessiond.
1496 */
1497 goto end_nosignal;
1498 }
1499 /*
1500 * The communicaton was broken hence there is a bad state between
1501 * the consumer and sessiond so stop everything.
1502 */
1503 goto error_fatal;
1504 }
1505
1506 health_code_update();
1507
1508 /*
1509 * In no monitor mode, the streams ownership is kept inside the channel
1510 * so don't send them to the data thread.
1511 */
1512 if (!channel->monitor) {
1513 goto end_msg_sessiond;
1514 }
1515
1516 ret = send_streams_to_thread(channel, ctx);
1517 if (ret < 0) {
1518 /*
1519 * If we are unable to send the stream to the thread, there is
1520 * a big problem so just stop everything.
1521 */
1522 goto error_fatal;
1523 }
1524 /* List MUST be empty after or else it could be reused. */
1525 assert(cds_list_empty(&channel->streams.head));
1526 goto end_msg_sessiond;
1527 }
1528 case LTTNG_CONSUMER_DESTROY_CHANNEL:
1529 {
1530 uint64_t key = msg.u.destroy_channel.key;
1531
1532 /*
1533 * Only called if streams have not been sent to stream
1534 * manager thread. However, channel has been sent to
1535 * channel manager thread.
1536 */
1537 notify_thread_del_channel(ctx, key);
1538 goto end_msg_sessiond;
1539 }
1540 case LTTNG_CONSUMER_CLOSE_METADATA:
1541 {
1542 int ret;
1543
1544 ret = close_metadata(msg.u.close_metadata.key);
1545 if (ret != 0) {
1546 ret_code = ret;
1547 }
1548
1549 goto end_msg_sessiond;
1550 }
1551 case LTTNG_CONSUMER_FLUSH_CHANNEL:
1552 {
1553 int ret;
1554
1555 ret = flush_channel(msg.u.flush_channel.key);
1556 if (ret != 0) {
1557 ret_code = ret;
1558 }
1559
1560 goto end_msg_sessiond;
1561 }
1562 case LTTNG_CONSUMER_PUSH_METADATA:
1563 {
1564 int ret;
1565 uint64_t len = msg.u.push_metadata.len;
1566 uint64_t key = msg.u.push_metadata.key;
1567 uint64_t offset = msg.u.push_metadata.target_offset;
1568 struct lttng_consumer_channel *channel;
1569
1570 DBG("UST consumer push metadata key %" PRIu64 " of len %" PRIu64, key,
1571 len);
1572
1573 channel = consumer_find_channel(key);
1574 if (!channel) {
1575 /*
1576 * This is possible if the metadata creation on the consumer side
1577 * is in flight vis-a-vis a concurrent push metadata from the
1578 * session daemon. Simply return that the channel failed and the
1579 * session daemon will handle that message correctly considering
1580 * that this race is acceptable thus the DBG() statement here.
1581 */
1582 DBG("UST consumer push metadata %" PRIu64 " not found", key);
1583 ret_code = LTTCOMM_CONSUMERD_CHANNEL_FAIL;
1584 goto end_msg_sessiond;
1585 }
1586
1587 health_code_update();
1588
1589 if (!len) {
1590 /*
1591 * There is nothing to receive. We have simply
1592 * checked whether the channel can be found.
1593 */
1594 ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1595 goto end_msg_sessiond;
1596 }
1597
1598 /* Tell session daemon we are ready to receive the metadata. */
1599 ret = consumer_send_status_msg(sock, LTTCOMM_CONSUMERD_SUCCESS);
1600 if (ret < 0) {
1601 /* Somehow, the session daemon is not responding anymore. */
1602 goto error_fatal;
1603 }
1604
1605 health_code_update();
1606
1607 /* Wait for more data. */
1608 health_poll_entry();
1609 ret = lttng_consumer_poll_socket(consumer_sockpoll);
1610 health_poll_exit();
1611 if (ret) {
1612 goto error_fatal;
1613 }
1614
1615 health_code_update();
1616
1617 ret = lttng_ustconsumer_recv_metadata(sock, key, offset,
1618 len, channel, 0, 1);
1619 if (ret < 0) {
1620 /* error receiving from sessiond */
1621 goto error_fatal;
1622 } else {
1623 ret_code = ret;
1624 goto end_msg_sessiond;
1625 }
1626 }
1627 case LTTNG_CONSUMER_SETUP_METADATA:
1628 {
1629 int ret;
1630
1631 ret = setup_metadata(ctx, msg.u.setup_metadata.key);
1632 if (ret) {
1633 ret_code = ret;
1634 }
1635 goto end_msg_sessiond;
1636 }
1637 case LTTNG_CONSUMER_SNAPSHOT_CHANNEL:
1638 {
1639 if (msg.u.snapshot_channel.metadata) {
1640 ret = snapshot_metadata(msg.u.snapshot_channel.key,
1641 msg.u.snapshot_channel.pathname,
1642 msg.u.snapshot_channel.relayd_id,
1643 ctx);
1644 if (ret < 0) {
1645 ERR("Snapshot metadata failed");
1646 ret_code = LTTCOMM_CONSUMERD_ERROR_METADATA;
1647 }
1648 } else {
1649 ret = snapshot_channel(msg.u.snapshot_channel.key,
1650 msg.u.snapshot_channel.pathname,
1651 msg.u.snapshot_channel.relayd_id,
1652 msg.u.snapshot_channel.nb_packets_per_stream,
1653 ctx);
1654 if (ret < 0) {
1655 ERR("Snapshot channel failed");
1656 ret_code = LTTCOMM_CONSUMERD_CHANNEL_FAIL;
1657 }
1658 }
1659
1660 health_code_update();
1661 ret = consumer_send_status_msg(sock, ret_code);
1662 if (ret < 0) {
1663 /* Somehow, the session daemon is not responding anymore. */
1664 goto end_nosignal;
1665 }
1666 health_code_update();
1667 break;
1668 }
1669 default:
1670 break;
1671 }
1672
1673 end_nosignal:
1674 rcu_read_unlock();
1675
1676 health_code_update();
1677
1678 /*
1679 * Return 1 to indicate success since the 0 value can be a socket
1680 * shutdown during the recv() or send() call.
1681 */
1682 return 1;
1683
1684 end_msg_sessiond:
1685 /*
1686 * The returned value here is not useful since either way we'll return 1 to
1687 * the caller because the session daemon socket management is done
1688 * elsewhere. Returning a negative code or 0 will shutdown the consumer.
1689 */
1690 ret = consumer_send_status_msg(sock, ret_code);
1691 if (ret < 0) {
1692 goto error_fatal;
1693 }
1694 rcu_read_unlock();
1695
1696 health_code_update();
1697
1698 return 1;
1699 end_channel_error:
1700 if (channel) {
1701 /*
1702 * Free channel here since no one has a reference to it. We don't
1703 * free after that because a stream can store this pointer.
1704 */
1705 destroy_channel(channel);
1706 }
1707 /* We have to send a status channel message indicating an error. */
1708 ret = consumer_send_status_channel(sock, NULL);
1709 if (ret < 0) {
1710 /* Stop everything if session daemon can not be notified. */
1711 goto error_fatal;
1712 }
1713 rcu_read_unlock();
1714
1715 health_code_update();
1716
1717 return 1;
1718 error_fatal:
1719 rcu_read_unlock();
1720 /* This will issue a consumer stop. */
1721 return -1;
1722 }
1723
1724 /*
1725 * Wrapper over the mmap() read offset from ust-ctl library. Since this can be
1726 * compiled out, we isolate it in this library.
1727 */
1728 int lttng_ustctl_get_mmap_read_offset(struct lttng_consumer_stream *stream,
1729 unsigned long *off)
1730 {
1731 assert(stream);
1732 assert(stream->ustream);
1733
1734 return ustctl_get_mmap_read_offset(stream->ustream, off);
1735 }
1736
1737 /*
1738 * Wrapper over the mmap() read offset from ust-ctl library. Since this can be
1739 * compiled out, we isolate it in this library.
1740 */
1741 void *lttng_ustctl_get_mmap_base(struct lttng_consumer_stream *stream)
1742 {
1743 assert(stream);
1744 assert(stream->ustream);
1745
1746 return ustctl_get_mmap_base(stream->ustream);
1747 }
1748
1749 /*
1750 * Take a snapshot for a specific fd
1751 *
1752 * Returns 0 on success, < 0 on error
1753 */
1754 int lttng_ustconsumer_take_snapshot(struct lttng_consumer_stream *stream)
1755 {
1756 assert(stream);
1757 assert(stream->ustream);
1758
1759 return ustctl_snapshot(stream->ustream);
1760 }
1761
1762 /*
1763 * Get the produced position
1764 *
1765 * Returns 0 on success, < 0 on error
1766 */
1767 int lttng_ustconsumer_get_produced_snapshot(
1768 struct lttng_consumer_stream *stream, unsigned long *pos)
1769 {
1770 assert(stream);
1771 assert(stream->ustream);
1772 assert(pos);
1773
1774 return ustctl_snapshot_get_produced(stream->ustream, pos);
1775 }
1776
1777 /*
1778 * Get the consumed position
1779 *
1780 * Returns 0 on success, < 0 on error
1781 */
1782 int lttng_ustconsumer_get_consumed_snapshot(
1783 struct lttng_consumer_stream *stream, unsigned long *pos)
1784 {
1785 assert(stream);
1786 assert(stream->ustream);
1787 assert(pos);
1788
1789 return ustctl_snapshot_get_consumed(stream->ustream, pos);
1790 }
1791
1792 void lttng_ustconsumer_flush_buffer(struct lttng_consumer_stream *stream,
1793 int producer)
1794 {
1795 assert(stream);
1796 assert(stream->ustream);
1797
1798 ustctl_flush_buffer(stream->ustream, producer);
1799 }
1800
1801 int lttng_ustconsumer_get_current_timestamp(
1802 struct lttng_consumer_stream *stream, uint64_t *ts)
1803 {
1804 assert(stream);
1805 assert(stream->ustream);
1806 assert(ts);
1807
1808 return ustctl_get_current_timestamp(stream->ustream, ts);
1809 }
1810
1811 /*
1812 * Called when the stream signal the consumer that it has hang up.
1813 */
1814 void lttng_ustconsumer_on_stream_hangup(struct lttng_consumer_stream *stream)
1815 {
1816 assert(stream);
1817 assert(stream->ustream);
1818
1819 ustctl_flush_buffer(stream->ustream, 0);
1820 stream->hangup_flush_done = 1;
1821 }
1822
1823 void lttng_ustconsumer_del_channel(struct lttng_consumer_channel *chan)
1824 {
1825 int i;
1826
1827 assert(chan);
1828 assert(chan->uchan);
1829
1830 if (chan->switch_timer_enabled == 1) {
1831 consumer_timer_switch_stop(chan);
1832 }
1833 consumer_metadata_cache_destroy(chan);
1834 ustctl_destroy_channel(chan->uchan);
1835 for (i = 0; i < chan->nr_stream_fds; i++) {
1836 int ret;
1837
1838 ret = close(chan->stream_fds[i]);
1839 if (ret) {
1840 PERROR("close");
1841 }
1842 if (chan->shm_path[0]) {
1843 char shm_path[PATH_MAX];
1844
1845 ret = get_stream_shm_path(shm_path, chan->shm_path, i);
1846 if (ret) {
1847 ERR("Cannot get stream shm path");
1848 }
1849 ret = run_as_unlink(shm_path, chan->uid, chan->gid);
1850 if (ret) {
1851 PERROR("unlink %s", shm_path);
1852 }
1853 }
1854 }
1855 free(chan->stream_fds);
1856 /* Try to rmdir all directories under shm_path root. */
1857 if (chan->root_shm_path[0]) {
1858 (void) run_as_recursive_rmdir(chan->root_shm_path,
1859 chan->uid, chan->gid);
1860 }
1861 }
1862
1863 void lttng_ustconsumer_del_stream(struct lttng_consumer_stream *stream)
1864 {
1865 assert(stream);
1866 assert(stream->ustream);
1867
1868 if (stream->chan->switch_timer_enabled == 1) {
1869 consumer_timer_switch_stop(stream->chan);
1870 }
1871 ustctl_destroy_stream(stream->ustream);
1872 }
1873
1874 int lttng_ustconsumer_get_wakeup_fd(struct lttng_consumer_stream *stream)
1875 {
1876 assert(stream);
1877 assert(stream->ustream);
1878
1879 return ustctl_stream_get_wakeup_fd(stream->ustream);
1880 }
1881
1882 int lttng_ustconsumer_close_wakeup_fd(struct lttng_consumer_stream *stream)
1883 {
1884 assert(stream);
1885 assert(stream->ustream);
1886
1887 return ustctl_stream_close_wakeup_fd(stream->ustream);
1888 }
1889
1890 /*
1891 * Populate index values of a UST stream. Values are set in big endian order.
1892 *
1893 * Return 0 on success or else a negative value.
1894 */
1895 static int get_index_values(struct ctf_packet_index *index,
1896 struct ustctl_consumer_stream *ustream)
1897 {
1898 int ret;
1899
1900 ret = ustctl_get_timestamp_begin(ustream, &index->timestamp_begin);
1901 if (ret < 0) {
1902 PERROR("ustctl_get_timestamp_begin");
1903 goto error;
1904 }
1905 index->timestamp_begin = htobe64(index->timestamp_begin);
1906
1907 ret = ustctl_get_timestamp_end(ustream, &index->timestamp_end);
1908 if (ret < 0) {
1909 PERROR("ustctl_get_timestamp_end");
1910 goto error;
1911 }
1912 index->timestamp_end = htobe64(index->timestamp_end);
1913
1914 ret = ustctl_get_events_discarded(ustream, &index->events_discarded);
1915 if (ret < 0) {
1916 PERROR("ustctl_get_events_discarded");
1917 goto error;
1918 }
1919 index->events_discarded = htobe64(index->events_discarded);
1920
1921 ret = ustctl_get_content_size(ustream, &index->content_size);
1922 if (ret < 0) {
1923 PERROR("ustctl_get_content_size");
1924 goto error;
1925 }
1926 index->content_size = htobe64(index->content_size);
1927
1928 ret = ustctl_get_packet_size(ustream, &index->packet_size);
1929 if (ret < 0) {
1930 PERROR("ustctl_get_packet_size");
1931 goto error;
1932 }
1933 index->packet_size = htobe64(index->packet_size);
1934
1935 ret = ustctl_get_stream_id(ustream, &index->stream_id);
1936 if (ret < 0) {
1937 PERROR("ustctl_get_stream_id");
1938 goto error;
1939 }
1940 index->stream_id = htobe64(index->stream_id);
1941
1942 error:
1943 return ret;
1944 }
1945
1946 /*
1947 * Write up to one packet from the metadata cache to the channel.
1948 *
1949 * Returns the number of bytes pushed in the cache, or a negative value
1950 * on error.
1951 */
1952 static
1953 int commit_one_metadata_packet(struct lttng_consumer_stream *stream)
1954 {
1955 ssize_t write_len;
1956 int ret;
1957
1958 pthread_mutex_lock(&stream->chan->metadata_cache->lock);
1959 if (stream->chan->metadata_cache->max_offset
1960 == stream->ust_metadata_pushed) {
1961 ret = 0;
1962 goto end;
1963 }
1964
1965 write_len = ustctl_write_one_packet_to_channel(stream->chan->uchan,
1966 &stream->chan->metadata_cache->data[stream->ust_metadata_pushed],
1967 stream->chan->metadata_cache->max_offset
1968 - stream->ust_metadata_pushed);
1969 assert(write_len != 0);
1970 if (write_len < 0) {
1971 ERR("Writing one metadata packet");
1972 ret = -1;
1973 goto end;
1974 }
1975 stream->ust_metadata_pushed += write_len;
1976
1977 assert(stream->chan->metadata_cache->max_offset >=
1978 stream->ust_metadata_pushed);
1979 ret = write_len;
1980
1981 end:
1982 pthread_mutex_unlock(&stream->chan->metadata_cache->lock);
1983 return ret;
1984 }
1985
1986
1987 /*
1988 * Sync metadata meaning request them to the session daemon and snapshot to the
1989 * metadata thread can consumer them.
1990 *
1991 * Metadata stream lock is held here, but we need to release it when
1992 * interacting with sessiond, else we cause a deadlock with live
1993 * awaiting on metadata to be pushed out.
1994 *
1995 * Return 0 if new metadatda is available, EAGAIN if the metadata stream
1996 * is empty or a negative value on error.
1997 */
1998 int lttng_ustconsumer_sync_metadata(struct lttng_consumer_local_data *ctx,
1999 struct lttng_consumer_stream *metadata)
2000 {
2001 int ret;
2002 int retry = 0;
2003
2004 assert(ctx);
2005 assert(metadata);
2006
2007 pthread_mutex_unlock(&metadata->lock);
2008 /*
2009 * Request metadata from the sessiond, but don't wait for the flush
2010 * because we locked the metadata thread.
2011 */
2012 ret = lttng_ustconsumer_request_metadata(ctx, metadata->chan, 0, 0);
2013 if (ret < 0) {
2014 goto end;
2015 }
2016 pthread_mutex_lock(&metadata->lock);
2017
2018 ret = commit_one_metadata_packet(metadata);
2019 if (ret <= 0) {
2020 goto end;
2021 } else if (ret > 0) {
2022 retry = 1;
2023 }
2024
2025 ustctl_flush_buffer(metadata->ustream, 1);
2026 ret = ustctl_snapshot(metadata->ustream);
2027 if (ret < 0) {
2028 if (errno != EAGAIN) {
2029 ERR("Sync metadata, taking UST snapshot");
2030 goto end;
2031 }
2032 DBG("No new metadata when syncing them.");
2033 /* No new metadata, exit. */
2034 ret = ENODATA;
2035 goto end;
2036 }
2037
2038 /*
2039 * After this flush, we still need to extract metadata.
2040 */
2041 if (retry) {
2042 ret = EAGAIN;
2043 }
2044
2045 end:
2046 return ret;
2047 }
2048
2049 /*
2050 * Return 0 on success else a negative value.
2051 */
2052 static int notify_if_more_data(struct lttng_consumer_stream *stream,
2053 struct lttng_consumer_local_data *ctx)
2054 {
2055 int ret;
2056 struct ustctl_consumer_stream *ustream;
2057
2058 assert(stream);
2059 assert(ctx);
2060
2061 ustream = stream->ustream;
2062
2063 /*
2064 * First, we are going to check if there is a new subbuffer available
2065 * before reading the stream wait_fd.
2066 */
2067 /* Get the next subbuffer */
2068 ret = ustctl_get_next_subbuf(ustream);
2069 if (ret) {
2070 /* No more data found, flag the stream. */
2071 stream->has_data = 0;
2072 ret = 0;
2073 goto end;
2074 }
2075
2076 ret = ustctl_put_subbuf(ustream);
2077 assert(!ret);
2078
2079 /* This stream still has data. Flag it and wake up the data thread. */
2080 stream->has_data = 1;
2081
2082 if (stream->monitor && !stream->hangup_flush_done && !ctx->has_wakeup) {
2083 ssize_t writelen;
2084
2085 writelen = lttng_pipe_write(ctx->consumer_wakeup_pipe, "!", 1);
2086 if (writelen < 0 && errno != EAGAIN && errno != EWOULDBLOCK) {
2087 ret = writelen;
2088 goto end;
2089 }
2090
2091 /* The wake up pipe has been notified. */
2092 ctx->has_wakeup = 1;
2093 }
2094 ret = 0;
2095
2096 end:
2097 return ret;
2098 }
2099
2100 /*
2101 * Read subbuffer from the given stream.
2102 *
2103 * Stream lock MUST be acquired.
2104 *
2105 * Return 0 on success else a negative value.
2106 */
2107 int lttng_ustconsumer_read_subbuffer(struct lttng_consumer_stream *stream,
2108 struct lttng_consumer_local_data *ctx)
2109 {
2110 unsigned long len, subbuf_size, padding;
2111 int err, write_index = 1;
2112 long ret = 0;
2113 struct ustctl_consumer_stream *ustream;
2114 struct ctf_packet_index index;
2115
2116 assert(stream);
2117 assert(stream->ustream);
2118 assert(ctx);
2119
2120 DBG("In UST read_subbuffer (wait_fd: %d, name: %s)", stream->wait_fd,
2121 stream->name);
2122
2123 /* Ease our life for what's next. */
2124 ustream = stream->ustream;
2125
2126 /*
2127 * We can consume the 1 byte written into the wait_fd by UST. Don't trigger
2128 * error if we cannot read this one byte (read returns 0), or if the error
2129 * is EAGAIN or EWOULDBLOCK.
2130 *
2131 * This is only done when the stream is monitored by a thread, before the
2132 * flush is done after a hangup and if the stream is not flagged with data
2133 * since there might be nothing to consume in the wait fd but still have
2134 * data available flagged by the consumer wake up pipe.
2135 */
2136 if (stream->monitor && !stream->hangup_flush_done && !stream->has_data) {
2137 char dummy;
2138 ssize_t readlen;
2139
2140 readlen = lttng_read(stream->wait_fd, &dummy, 1);
2141 if (readlen < 0 && errno != EAGAIN && errno != EWOULDBLOCK) {
2142 ret = readlen;
2143 goto end;
2144 }
2145 }
2146
2147 retry:
2148 /* Get the next subbuffer */
2149 err = ustctl_get_next_subbuf(ustream);
2150 if (err != 0) {
2151 /*
2152 * Populate metadata info if the existing info has
2153 * already been read.
2154 */
2155 if (stream->metadata_flag) {
2156 ret = commit_one_metadata_packet(stream);
2157 if (ret <= 0) {
2158 goto end;
2159 }
2160 ustctl_flush_buffer(stream->ustream, 1);
2161 goto retry;
2162 }
2163
2164 ret = err; /* ustctl_get_next_subbuf returns negative, caller expect positive. */
2165 /*
2166 * This is a debug message even for single-threaded consumer,
2167 * because poll() have more relaxed criterions than get subbuf,
2168 * so get_subbuf may fail for short race windows where poll()
2169 * would issue wakeups.
2170 */
2171 DBG("Reserving sub buffer failed (everything is normal, "
2172 "it is due to concurrency) [ret: %d]", err);
2173 goto end;
2174 }
2175 assert(stream->chan->output == CONSUMER_CHANNEL_MMAP);
2176
2177 if (!stream->metadata_flag) {
2178 index.offset = htobe64(stream->out_fd_offset);
2179 ret = get_index_values(&index, ustream);
2180 if (ret < 0) {
2181 goto end;
2182 }
2183 } else {
2184 write_index = 0;
2185 }
2186
2187 /* Get the full padded subbuffer size */
2188 err = ustctl_get_padded_subbuf_size(ustream, &len);
2189 assert(err == 0);
2190
2191 /* Get subbuffer data size (without padding) */
2192 err = ustctl_get_subbuf_size(ustream, &subbuf_size);
2193 assert(err == 0);
2194
2195 /* Make sure we don't get a subbuffer size bigger than the padded */
2196 assert(len >= subbuf_size);
2197
2198 padding = len - subbuf_size;
2199 /* write the subbuffer to the tracefile */
2200 ret = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, subbuf_size, padding, &index);
2201 /*
2202 * The mmap operation should write subbuf_size amount of data when network
2203 * streaming or the full padding (len) size when we are _not_ streaming.
2204 */
2205 if ((ret != subbuf_size && stream->net_seq_idx != (uint64_t) -1ULL) ||
2206 (ret != len && stream->net_seq_idx == (uint64_t) -1ULL)) {
2207 /*
2208 * Display the error but continue processing to try to release the
2209 * subbuffer. This is a DBG statement since any unexpected kill or
2210 * signal, the application gets unregistered, relayd gets closed or
2211 * anything that affects the buffer lifetime will trigger this error.
2212 * So, for the sake of the user, don't print this error since it can
2213 * happen and it is OK with the code flow.
2214 */
2215 DBG("Error writing to tracefile "
2216 "(ret: %ld != len: %lu != subbuf_size: %lu)",
2217 ret, len, subbuf_size);
2218 write_index = 0;
2219 }
2220 err = ustctl_put_next_subbuf(ustream);
2221 assert(err == 0);
2222
2223 /*
2224 * This will consumer the byte on the wait_fd if and only if there is not
2225 * next subbuffer to be acquired.
2226 */
2227 if (!stream->metadata_flag) {
2228 ret = notify_if_more_data(stream, ctx);
2229 if (ret < 0) {
2230 goto end;
2231 }
2232 }
2233
2234 /* Write index if needed. */
2235 if (!write_index) {
2236 goto end;
2237 }
2238
2239 if (stream->chan->live_timer_interval && !stream->metadata_flag) {
2240 /*
2241 * In live, block until all the metadata is sent.
2242 */
2243 pthread_mutex_lock(&stream->metadata_timer_lock);
2244 assert(!stream->missed_metadata_flush);
2245 stream->waiting_on_metadata = true;
2246 pthread_mutex_unlock(&stream->metadata_timer_lock);
2247
2248 err = consumer_stream_sync_metadata(ctx, stream->session_id);
2249
2250 pthread_mutex_lock(&stream->metadata_timer_lock);
2251 stream->waiting_on_metadata = false;
2252 if (stream->missed_metadata_flush) {
2253 stream->missed_metadata_flush = false;
2254 pthread_mutex_unlock(&stream->metadata_timer_lock);
2255 (void) consumer_flush_ust_index(stream);
2256 } else {
2257 pthread_mutex_unlock(&stream->metadata_timer_lock);
2258 }
2259
2260 if (err < 0) {
2261 goto end;
2262 }
2263 }
2264
2265 assert(!stream->metadata_flag);
2266 err = consumer_stream_write_index(stream, &index);
2267 if (err < 0) {
2268 goto end;
2269 }
2270
2271 end:
2272 return ret;
2273 }
2274
2275 /*
2276 * Called when a stream is created.
2277 *
2278 * Return 0 on success or else a negative value.
2279 */
2280 int lttng_ustconsumer_on_recv_stream(struct lttng_consumer_stream *stream)
2281 {
2282 int ret;
2283
2284 assert(stream);
2285
2286 /* Don't create anything if this is set for streaming. */
2287 if (stream->net_seq_idx == (uint64_t) -1ULL && stream->chan->monitor) {
2288 ret = utils_create_stream_file(stream->chan->pathname, stream->name,
2289 stream->chan->tracefile_size, stream->tracefile_count_current,
2290 stream->uid, stream->gid, NULL);
2291 if (ret < 0) {
2292 goto error;
2293 }
2294 stream->out_fd = ret;
2295 stream->tracefile_size_current = 0;
2296
2297 if (!stream->metadata_flag) {
2298 ret = index_create_file(stream->chan->pathname,
2299 stream->name, stream->uid, stream->gid,
2300 stream->chan->tracefile_size,
2301 stream->tracefile_count_current);
2302 if (ret < 0) {
2303 goto error;
2304 }
2305 stream->index_fd = ret;
2306 }
2307 }
2308 ret = 0;
2309
2310 error:
2311 return ret;
2312 }
2313
2314 /*
2315 * Check if data is still being extracted from the buffers for a specific
2316 * stream. Consumer data lock MUST be acquired before calling this function
2317 * and the stream lock.
2318 *
2319 * Return 1 if the traced data are still getting read else 0 meaning that the
2320 * data is available for trace viewer reading.
2321 */
2322 int lttng_ustconsumer_data_pending(struct lttng_consumer_stream *stream)
2323 {
2324 int ret;
2325
2326 assert(stream);
2327 assert(stream->ustream);
2328
2329 DBG("UST consumer checking data pending");
2330
2331 if (stream->endpoint_status != CONSUMER_ENDPOINT_ACTIVE) {
2332 ret = 0;
2333 goto end;
2334 }
2335
2336 if (stream->chan->type == CONSUMER_CHANNEL_TYPE_METADATA) {
2337 uint64_t contiguous, pushed;
2338
2339 /* Ease our life a bit. */
2340 contiguous = stream->chan->metadata_cache->max_offset;
2341 pushed = stream->ust_metadata_pushed;
2342
2343 /*
2344 * We can simply check whether all contiguously available data
2345 * has been pushed to the ring buffer, since the push operation
2346 * is performed within get_next_subbuf(), and because both
2347 * get_next_subbuf() and put_next_subbuf() are issued atomically
2348 * thanks to the stream lock within
2349 * lttng_ustconsumer_read_subbuffer(). This basically means that
2350 * whetnever ust_metadata_pushed is incremented, the associated
2351 * metadata has been consumed from the metadata stream.
2352 */
2353 DBG("UST consumer metadata pending check: contiguous %" PRIu64 " vs pushed %" PRIu64,
2354 contiguous, pushed);
2355 assert(((int64_t) (contiguous - pushed)) >= 0);
2356 if ((contiguous != pushed) ||
2357 (((int64_t) contiguous - pushed) > 0 || contiguous == 0)) {
2358 ret = 1; /* Data is pending */
2359 goto end;
2360 }
2361 } else {
2362 ret = ustctl_get_next_subbuf(stream->ustream);
2363 if (ret == 0) {
2364 /*
2365 * There is still data so let's put back this
2366 * subbuffer.
2367 */
2368 ret = ustctl_put_subbuf(stream->ustream);
2369 assert(ret == 0);
2370 ret = 1; /* Data is pending */
2371 goto end;
2372 }
2373 }
2374
2375 /* Data is NOT pending so ready to be read. */
2376 ret = 0;
2377
2378 end:
2379 return ret;
2380 }
2381
2382 /*
2383 * Stop a given metadata channel timer if enabled and close the wait fd which
2384 * is the poll pipe of the metadata stream.
2385 *
2386 * This MUST be called with the metadata channel acquired.
2387 */
2388 void lttng_ustconsumer_close_metadata(struct lttng_consumer_channel *metadata)
2389 {
2390 int ret;
2391
2392 assert(metadata);
2393 assert(metadata->type == CONSUMER_CHANNEL_TYPE_METADATA);
2394
2395 DBG("Closing metadata channel key %" PRIu64, metadata->key);
2396
2397 if (metadata->switch_timer_enabled == 1) {
2398 consumer_timer_switch_stop(metadata);
2399 }
2400
2401 if (!metadata->metadata_stream) {
2402 goto end;
2403 }
2404
2405 /*
2406 * Closing write side so the thread monitoring the stream wakes up if any
2407 * and clean the metadata stream.
2408 */
2409 if (metadata->metadata_stream->ust_metadata_poll_pipe[1] >= 0) {
2410 ret = close(metadata->metadata_stream->ust_metadata_poll_pipe[1]);
2411 if (ret < 0) {
2412 PERROR("closing metadata pipe write side");
2413 }
2414 metadata->metadata_stream->ust_metadata_poll_pipe[1] = -1;
2415 }
2416
2417 end:
2418 return;
2419 }
2420
2421 /*
2422 * Close every metadata stream wait fd of the metadata hash table. This
2423 * function MUST be used very carefully so not to run into a race between the
2424 * metadata thread handling streams and this function closing their wait fd.
2425 *
2426 * For UST, this is used when the session daemon hangs up. Its the metadata
2427 * producer so calling this is safe because we are assured that no state change
2428 * can occur in the metadata thread for the streams in the hash table.
2429 */
2430 void lttng_ustconsumer_close_all_metadata(struct lttng_ht *metadata_ht)
2431 {
2432 struct lttng_ht_iter iter;
2433 struct lttng_consumer_stream *stream;
2434
2435 assert(metadata_ht);
2436 assert(metadata_ht->ht);
2437
2438 DBG("UST consumer closing all metadata streams");
2439
2440 rcu_read_lock();
2441 cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream,
2442 node.node) {
2443
2444 health_code_update();
2445
2446 pthread_mutex_lock(&stream->chan->lock);
2447 lttng_ustconsumer_close_metadata(stream->chan);
2448 pthread_mutex_unlock(&stream->chan->lock);
2449
2450 }
2451 rcu_read_unlock();
2452 }
2453
2454 void lttng_ustconsumer_close_stream_wakeup(struct lttng_consumer_stream *stream)
2455 {
2456 int ret;
2457
2458 ret = ustctl_stream_close_wakeup_fd(stream->ustream);
2459 if (ret < 0) {
2460 ERR("Unable to close wakeup fd");
2461 }
2462 }
2463
2464 /*
2465 * Please refer to consumer-timer.c before adding any lock within this
2466 * function or any of its callees. Timers have a very strict locking
2467 * semantic with respect to teardown. Failure to respect this semantic
2468 * introduces deadlocks.
2469 *
2470 * DON'T hold the metadata lock when calling this function, else this
2471 * can cause deadlock involving consumer awaiting for metadata to be
2472 * pushed out due to concurrent interaction with the session daemon.
2473 */
2474 int lttng_ustconsumer_request_metadata(struct lttng_consumer_local_data *ctx,
2475 struct lttng_consumer_channel *channel, int timer, int wait)
2476 {
2477 struct lttcomm_metadata_request_msg request;
2478 struct lttcomm_consumer_msg msg;
2479 enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS;
2480 uint64_t len, key, offset;
2481 int ret;
2482
2483 assert(channel);
2484 assert(channel->metadata_cache);
2485
2486 memset(&request, 0, sizeof(request));
2487
2488 /* send the metadata request to sessiond */
2489 switch (consumer_data.type) {
2490 case LTTNG_CONSUMER64_UST:
2491 request.bits_per_long = 64;
2492 break;
2493 case LTTNG_CONSUMER32_UST:
2494 request.bits_per_long = 32;
2495 break;
2496 default:
2497 request.bits_per_long = 0;
2498 break;
2499 }
2500
2501 request.session_id = channel->session_id;
2502 request.session_id_per_pid = channel->session_id_per_pid;
2503 /*
2504 * Request the application UID here so the metadata of that application can
2505 * be sent back. The channel UID corresponds to the user UID of the session
2506 * used for the rights on the stream file(s).
2507 */
2508 request.uid = channel->ust_app_uid;
2509 request.key = channel->key;
2510
2511 DBG("Sending metadata request to sessiond, session id %" PRIu64
2512 ", per-pid %" PRIu64 ", app UID %u and channek key %" PRIu64,
2513 request.session_id, request.session_id_per_pid, request.uid,
2514 request.key);
2515
2516 pthread_mutex_lock(&ctx->metadata_socket_lock);
2517
2518 health_code_update();
2519
2520 ret = lttcomm_send_unix_sock(ctx->consumer_metadata_socket, &request,
2521 sizeof(request));
2522 if (ret < 0) {
2523 ERR("Asking metadata to sessiond");
2524 goto end;
2525 }
2526
2527 health_code_update();
2528
2529 /* Receive the metadata from sessiond */
2530 ret = lttcomm_recv_unix_sock(ctx->consumer_metadata_socket, &msg,
2531 sizeof(msg));
2532 if (ret != sizeof(msg)) {
2533 DBG("Consumer received unexpected message size %d (expects %zu)",
2534 ret, sizeof(msg));
2535 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD);
2536 /*
2537 * The ret value might 0 meaning an orderly shutdown but this is ok
2538 * since the caller handles this.
2539 */
2540 goto end;
2541 }
2542
2543 health_code_update();
2544
2545 if (msg.cmd_type == LTTNG_ERR_UND) {
2546 /* No registry found */
2547 (void) consumer_send_status_msg(ctx->consumer_metadata_socket,
2548 ret_code);
2549 ret = 0;
2550 goto end;
2551 } else if (msg.cmd_type != LTTNG_CONSUMER_PUSH_METADATA) {
2552 ERR("Unexpected cmd_type received %d", msg.cmd_type);
2553 ret = -1;
2554 goto end;
2555 }
2556
2557 len = msg.u.push_metadata.len;
2558 key = msg.u.push_metadata.key;
2559 offset = msg.u.push_metadata.target_offset;
2560
2561 assert(key == channel->key);
2562 if (len == 0) {
2563 DBG("No new metadata to receive for key %" PRIu64, key);
2564 }
2565
2566 health_code_update();
2567
2568 /* Tell session daemon we are ready to receive the metadata. */
2569 ret = consumer_send_status_msg(ctx->consumer_metadata_socket,
2570 LTTCOMM_CONSUMERD_SUCCESS);
2571 if (ret < 0 || len == 0) {
2572 /*
2573 * Somehow, the session daemon is not responding anymore or there is
2574 * nothing to receive.
2575 */
2576 goto end;
2577 }
2578
2579 health_code_update();
2580
2581 ret = lttng_ustconsumer_recv_metadata(ctx->consumer_metadata_socket,
2582 key, offset, len, channel, timer, wait);
2583 if (ret >= 0) {
2584 /*
2585 * Only send the status msg if the sessiond is alive meaning a positive
2586 * ret code.
2587 */
2588 (void) consumer_send_status_msg(ctx->consumer_metadata_socket, ret);
2589 }
2590 ret = 0;
2591
2592 end:
2593 health_code_update();
2594
2595 pthread_mutex_unlock(&ctx->metadata_socket_lock);
2596 return ret;
2597 }
2598
2599 /*
2600 * Return the ustctl call for the get stream id.
2601 */
2602 int lttng_ustconsumer_get_stream_id(struct lttng_consumer_stream *stream,
2603 uint64_t *stream_id)
2604 {
2605 assert(stream);
2606 assert(stream_id);
2607
2608 return ustctl_get_stream_id(stream->ustream, stream_id);
2609 }
This page took 0.147584 seconds and 4 git commands to generate.