Fix: ust-consumer: flush empty packets on snapshot channel
[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 INT_MAX_STR_LEN 12 /* 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 lttng_ustconsumer_free_channel(channel);
86 }
87 free(channel);
88 }
89
90 /*
91 * Add channel to internal consumer state.
92 *
93 * Returns 0 on success or else a negative value.
94 */
95 static int add_channel(struct lttng_consumer_channel *channel,
96 struct lttng_consumer_local_data *ctx)
97 {
98 int ret = 0;
99
100 assert(channel);
101 assert(ctx);
102
103 if (ctx->on_recv_channel != NULL) {
104 ret = ctx->on_recv_channel(channel);
105 if (ret == 0) {
106 ret = consumer_add_channel(channel, ctx);
107 } else if (ret < 0) {
108 /* Most likely an ENOMEM. */
109 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
110 goto error;
111 }
112 } else {
113 ret = consumer_add_channel(channel, ctx);
114 }
115
116 DBG("UST consumer channel added (key: %" PRIu64 ")", channel->key);
117
118 error:
119 return ret;
120 }
121
122 /*
123 * Allocate and return a consumer channel object.
124 */
125 static struct lttng_consumer_channel *allocate_channel(uint64_t session_id,
126 const char *pathname, const char *name, uid_t uid, gid_t gid,
127 uint64_t relayd_id, uint64_t key, enum lttng_event_output output,
128 uint64_t tracefile_size, uint64_t tracefile_count,
129 uint64_t session_id_per_pid, unsigned int monitor,
130 unsigned int live_timer_interval,
131 const char *root_shm_path, const char *shm_path)
132 {
133 assert(pathname);
134 assert(name);
135
136 return consumer_allocate_channel(key, session_id, pathname, name, uid,
137 gid, relayd_id, output, tracefile_size,
138 tracefile_count, session_id_per_pid, monitor,
139 live_timer_interval, root_shm_path, shm_path);
140 }
141
142 /*
143 * Allocate and return a consumer stream object. If _alloc_ret is not NULL, the
144 * error value if applicable is set in it else it is kept untouched.
145 *
146 * Return NULL on error else the newly allocated stream object.
147 */
148 static struct lttng_consumer_stream *allocate_stream(int cpu, int key,
149 struct lttng_consumer_channel *channel,
150 struct lttng_consumer_local_data *ctx, int *_alloc_ret)
151 {
152 int alloc_ret;
153 struct lttng_consumer_stream *stream = NULL;
154
155 assert(channel);
156 assert(ctx);
157
158 stream = consumer_allocate_stream(channel->key,
159 key,
160 LTTNG_CONSUMER_ACTIVE_STREAM,
161 channel->name,
162 channel->uid,
163 channel->gid,
164 channel->relayd_id,
165 channel->session_id,
166 cpu,
167 &alloc_ret,
168 channel->type,
169 channel->monitor);
170 if (stream == NULL) {
171 switch (alloc_ret) {
172 case -ENOENT:
173 /*
174 * We could not find the channel. Can happen if cpu hotplug
175 * happens while tearing down.
176 */
177 DBG3("Could not find channel");
178 break;
179 case -ENOMEM:
180 case -EINVAL:
181 default:
182 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
183 break;
184 }
185 goto error;
186 }
187
188 stream->chan = channel;
189
190 error:
191 if (_alloc_ret) {
192 *_alloc_ret = alloc_ret;
193 }
194 return stream;
195 }
196
197 /*
198 * Send the given stream pointer to the corresponding thread.
199 *
200 * Returns 0 on success else a negative value.
201 */
202 static int send_stream_to_thread(struct lttng_consumer_stream *stream,
203 struct lttng_consumer_local_data *ctx)
204 {
205 int ret;
206 struct lttng_pipe *stream_pipe;
207
208 /* Get the right pipe where the stream will be sent. */
209 if (stream->metadata_flag) {
210 ret = consumer_add_metadata_stream(stream);
211 if (ret) {
212 ERR("Consumer add metadata stream %" PRIu64 " failed.",
213 stream->key);
214 goto error;
215 }
216 stream_pipe = ctx->consumer_metadata_pipe;
217 } else {
218 ret = consumer_add_data_stream(stream);
219 if (ret) {
220 ERR("Consumer add stream %" PRIu64 " failed.",
221 stream->key);
222 goto error;
223 }
224 stream_pipe = ctx->consumer_data_pipe;
225 }
226
227 /*
228 * From this point on, the stream's ownership has been moved away from
229 * the channel and becomes globally visible.
230 */
231 stream->globally_visible = 1;
232
233 ret = lttng_pipe_write(stream_pipe, &stream, sizeof(stream));
234 if (ret < 0) {
235 ERR("Consumer write %s stream to pipe %d",
236 stream->metadata_flag ? "metadata" : "data",
237 lttng_pipe_get_writefd(stream_pipe));
238 if (stream->metadata_flag) {
239 consumer_del_stream_for_metadata(stream);
240 } else {
241 consumer_del_stream_for_data(stream);
242 }
243 }
244 error:
245 return ret;
246 }
247
248 static
249 int get_stream_shm_path(char *stream_shm_path, const char *shm_path, int cpu)
250 {
251 char cpu_nr[INT_MAX_STR_LEN]; /* int max len */
252 int ret;
253
254 strncpy(stream_shm_path, shm_path, PATH_MAX);
255 stream_shm_path[PATH_MAX - 1] = '\0';
256 ret = snprintf(cpu_nr, INT_MAX_STR_LEN, "%i", cpu);
257 if (ret < 0) {
258 PERROR("snprintf");
259 goto end;
260 }
261 strncat(stream_shm_path, cpu_nr,
262 PATH_MAX - strlen(stream_shm_path) - 1);
263 ret = 0;
264 end:
265 return ret;
266 }
267
268 /*
269 * Create streams for the given channel using liblttng-ust-ctl.
270 *
271 * Return 0 on success else a negative value.
272 */
273 static int create_ust_streams(struct lttng_consumer_channel *channel,
274 struct lttng_consumer_local_data *ctx)
275 {
276 int ret, cpu = 0;
277 struct ustctl_consumer_stream *ustream;
278 struct lttng_consumer_stream *stream;
279
280 assert(channel);
281 assert(ctx);
282
283 /*
284 * While a stream is available from ustctl. When NULL is returned, we've
285 * reached the end of the possible stream for the channel.
286 */
287 while ((ustream = ustctl_create_stream(channel->uchan, cpu))) {
288 int wait_fd;
289 int ust_metadata_pipe[2];
290
291 health_code_update();
292
293 if (channel->type == CONSUMER_CHANNEL_TYPE_METADATA && channel->monitor) {
294 ret = utils_create_pipe_cloexec_nonblock(ust_metadata_pipe);
295 if (ret < 0) {
296 ERR("Create ust metadata poll pipe");
297 goto error;
298 }
299 wait_fd = ust_metadata_pipe[0];
300 } else {
301 wait_fd = ustctl_stream_get_wait_fd(ustream);
302 }
303
304 /* Allocate consumer stream object. */
305 stream = allocate_stream(cpu, wait_fd, channel, ctx, &ret);
306 if (!stream) {
307 goto error_alloc;
308 }
309 stream->ustream = ustream;
310 /*
311 * Store it so we can save multiple function calls afterwards since
312 * this value is used heavily in the stream threads. This is UST
313 * specific so this is why it's done after allocation.
314 */
315 stream->wait_fd = wait_fd;
316
317 /*
318 * Increment channel refcount since the channel reference has now been
319 * assigned in the allocation process above.
320 */
321 if (stream->chan->monitor) {
322 uatomic_inc(&stream->chan->refcount);
323 }
324
325 /*
326 * Order is important this is why a list is used. On error, the caller
327 * should clean this list.
328 */
329 cds_list_add_tail(&stream->send_node, &channel->streams.head);
330
331 ret = ustctl_get_max_subbuf_size(stream->ustream,
332 &stream->max_sb_size);
333 if (ret < 0) {
334 ERR("ustctl_get_max_subbuf_size failed for stream %s",
335 stream->name);
336 goto error;
337 }
338
339 /* Do actions once stream has been received. */
340 if (ctx->on_recv_stream) {
341 ret = ctx->on_recv_stream(stream);
342 if (ret < 0) {
343 goto error;
344 }
345 }
346
347 DBG("UST consumer add stream %s (key: %" PRIu64 ") with relayd id %" PRIu64,
348 stream->name, stream->key, stream->relayd_stream_id);
349
350 /* Set next CPU stream. */
351 channel->streams.count = ++cpu;
352
353 /* Keep stream reference when creating metadata. */
354 if (channel->type == CONSUMER_CHANNEL_TYPE_METADATA) {
355 channel->metadata_stream = stream;
356 if (channel->monitor) {
357 /* Set metadata poll pipe if we created one */
358 memcpy(stream->ust_metadata_poll_pipe,
359 ust_metadata_pipe,
360 sizeof(ust_metadata_pipe));
361 }
362 }
363 }
364
365 return 0;
366
367 error:
368 error_alloc:
369 return ret;
370 }
371
372 /*
373 * create_posix_shm is never called concurrently within a process.
374 */
375 static
376 int create_posix_shm(void)
377 {
378 char tmp_name[NAME_MAX];
379 int shmfd, ret;
380
381 ret = snprintf(tmp_name, NAME_MAX, "/ust-shm-consumer-%d", getpid());
382 if (ret < 0) {
383 PERROR("snprintf");
384 return -1;
385 }
386 /*
387 * Allocate shm, and immediately unlink its shm oject, keeping
388 * only the file descriptor as a reference to the object.
389 * We specifically do _not_ use the / at the beginning of the
390 * pathname so that some OS implementations can keep it local to
391 * the process (POSIX leaves this implementation-defined).
392 */
393 shmfd = shm_open(tmp_name, O_CREAT | O_EXCL | O_RDWR, 0700);
394 if (shmfd < 0) {
395 PERROR("shm_open");
396 goto error_shm_open;
397 }
398 ret = shm_unlink(tmp_name);
399 if (ret < 0 && errno != ENOENT) {
400 PERROR("shm_unlink");
401 goto error_shm_release;
402 }
403 return shmfd;
404
405 error_shm_release:
406 ret = close(shmfd);
407 if (ret) {
408 PERROR("close");
409 }
410 error_shm_open:
411 return -1;
412 }
413
414 static int open_ust_stream_fd(struct lttng_consumer_channel *channel,
415 struct ustctl_consumer_channel_attr *attr,
416 int cpu)
417 {
418 char shm_path[PATH_MAX];
419 int ret;
420
421 if (!channel->shm_path[0]) {
422 return create_posix_shm();
423 }
424 ret = get_stream_shm_path(shm_path, channel->shm_path, cpu);
425 if (ret) {
426 goto error_shm_path;
427 }
428 return run_as_open(shm_path,
429 O_RDWR | O_CREAT | O_EXCL, S_IRUSR | S_IWUSR,
430 channel->uid, channel->gid);
431
432 error_shm_path:
433 return -1;
434 }
435
436 /*
437 * Create an UST channel with the given attributes and send it to the session
438 * daemon using the ust ctl API.
439 *
440 * Return 0 on success or else a negative value.
441 */
442 static int create_ust_channel(struct lttng_consumer_channel *channel,
443 struct ustctl_consumer_channel_attr *attr,
444 struct ustctl_consumer_channel **ust_chanp)
445 {
446 int ret, nr_stream_fds, i, j;
447 int *stream_fds;
448 struct ustctl_consumer_channel *ust_channel;
449
450 assert(channel);
451 assert(attr);
452 assert(ust_chanp);
453
454 DBG3("Creating channel to ustctl with attr: [overwrite: %d, "
455 "subbuf_size: %" PRIu64 ", num_subbuf: %" PRIu64 ", "
456 "switch_timer_interval: %u, read_timer_interval: %u, "
457 "output: %d, type: %d", attr->overwrite, attr->subbuf_size,
458 attr->num_subbuf, attr->switch_timer_interval,
459 attr->read_timer_interval, attr->output, attr->type);
460
461 if (channel->type == CONSUMER_CHANNEL_TYPE_METADATA)
462 nr_stream_fds = 1;
463 else
464 nr_stream_fds = ustctl_get_nr_stream_per_channel();
465 stream_fds = zmalloc(nr_stream_fds * sizeof(*stream_fds));
466 if (!stream_fds) {
467 ret = -1;
468 goto error_alloc;
469 }
470 for (i = 0; i < nr_stream_fds; i++) {
471 stream_fds[i] = open_ust_stream_fd(channel, attr, i);
472 if (stream_fds[i] < 0) {
473 ret = -1;
474 goto error_open;
475 }
476 }
477 ust_channel = ustctl_create_channel(attr, stream_fds, nr_stream_fds);
478 if (!ust_channel) {
479 ret = -1;
480 goto error_create;
481 }
482 channel->nr_stream_fds = nr_stream_fds;
483 channel->stream_fds = stream_fds;
484 *ust_chanp = ust_channel;
485
486 return 0;
487
488 error_create:
489 error_open:
490 for (j = i - 1; j >= 0; j--) {
491 int closeret;
492
493 closeret = close(stream_fds[j]);
494 if (closeret) {
495 PERROR("close");
496 }
497 if (channel->shm_path[0]) {
498 char shm_path[PATH_MAX];
499
500 closeret = get_stream_shm_path(shm_path,
501 channel->shm_path, j);
502 if (closeret) {
503 ERR("Cannot get stream shm path");
504 }
505 closeret = run_as_unlink(shm_path,
506 channel->uid, channel->gid);
507 if (closeret) {
508 PERROR("unlink %s", shm_path);
509 }
510 }
511 }
512 /* Try to rmdir all directories under shm_path root. */
513 if (channel->root_shm_path[0]) {
514 (void) run_as_recursive_rmdir(channel->root_shm_path,
515 channel->uid, channel->gid);
516 }
517 free(stream_fds);
518 error_alloc:
519 return ret;
520 }
521
522 /*
523 * Send a single given stream to the session daemon using the sock.
524 *
525 * Return 0 on success else a negative value.
526 */
527 static int send_sessiond_stream(int sock, struct lttng_consumer_stream *stream)
528 {
529 int ret;
530
531 assert(stream);
532 assert(sock >= 0);
533
534 DBG("UST consumer sending stream %" PRIu64 " to sessiond", stream->key);
535
536 /* Send stream to session daemon. */
537 ret = ustctl_send_stream_to_sessiond(sock, stream->ustream);
538 if (ret < 0) {
539 goto error;
540 }
541
542 error:
543 return ret;
544 }
545
546 /*
547 * Send channel to sessiond.
548 *
549 * Return 0 on success or else a negative value.
550 */
551 static int send_sessiond_channel(int sock,
552 struct lttng_consumer_channel *channel,
553 struct lttng_consumer_local_data *ctx, int *relayd_error)
554 {
555 int ret, ret_code = LTTCOMM_CONSUMERD_SUCCESS;
556 struct lttng_consumer_stream *stream;
557 uint64_t net_seq_idx = -1ULL;
558
559 assert(channel);
560 assert(ctx);
561 assert(sock >= 0);
562
563 DBG("UST consumer sending channel %s to sessiond", channel->name);
564
565 if (channel->relayd_id != (uint64_t) -1ULL) {
566 cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
567
568 health_code_update();
569
570 /* Try to send the stream to the relayd if one is available. */
571 ret = consumer_send_relayd_stream(stream, stream->chan->pathname);
572 if (ret < 0) {
573 /*
574 * Flag that the relayd was the problem here probably due to a
575 * communicaton error on the socket.
576 */
577 if (relayd_error) {
578 *relayd_error = 1;
579 }
580 ret_code = LTTCOMM_CONSUMERD_RELAYD_FAIL;
581 }
582 if (net_seq_idx == -1ULL) {
583 net_seq_idx = stream->net_seq_idx;
584 }
585 }
586 }
587
588 /* Inform sessiond that we are about to send channel and streams. */
589 ret = consumer_send_status_msg(sock, ret_code);
590 if (ret < 0 || ret_code != LTTCOMM_CONSUMERD_SUCCESS) {
591 /*
592 * Either the session daemon is not responding or the relayd died so we
593 * stop now.
594 */
595 goto error;
596 }
597
598 /* Send channel to sessiond. */
599 ret = ustctl_send_channel_to_sessiond(sock, channel->uchan);
600 if (ret < 0) {
601 goto error;
602 }
603
604 ret = ustctl_channel_close_wakeup_fd(channel->uchan);
605 if (ret < 0) {
606 goto error;
607 }
608
609 /* The channel was sent successfully to the sessiond at this point. */
610 cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
611
612 health_code_update();
613
614 /* Send stream to session daemon. */
615 ret = send_sessiond_stream(sock, stream);
616 if (ret < 0) {
617 goto error;
618 }
619 }
620
621 /* Tell sessiond there is no more stream. */
622 ret = ustctl_send_stream_to_sessiond(sock, NULL);
623 if (ret < 0) {
624 goto error;
625 }
626
627 DBG("UST consumer NULL stream sent to sessiond");
628
629 return 0;
630
631 error:
632 if (ret_code != LTTCOMM_CONSUMERD_SUCCESS) {
633 ret = -1;
634 }
635 return ret;
636 }
637
638 /*
639 * Creates a channel and streams and add the channel it to the channel internal
640 * state. The created stream must ONLY be sent once the GET_CHANNEL command is
641 * received.
642 *
643 * Return 0 on success or else, a negative value is returned and the channel
644 * MUST be destroyed by consumer_del_channel().
645 */
646 static int ask_channel(struct lttng_consumer_local_data *ctx, int sock,
647 struct lttng_consumer_channel *channel,
648 struct ustctl_consumer_channel_attr *attr)
649 {
650 int ret;
651
652 assert(ctx);
653 assert(channel);
654 assert(attr);
655
656 /*
657 * This value is still used by the kernel consumer since for the kernel,
658 * the stream ownership is not IN the consumer so we need to have the
659 * number of left stream that needs to be initialized so we can know when
660 * to delete the channel (see consumer.c).
661 *
662 * As for the user space tracer now, the consumer creates and sends the
663 * stream to the session daemon which only sends them to the application
664 * once every stream of a channel is received making this value useless
665 * because we they will be added to the poll thread before the application
666 * receives them. This ensures that a stream can not hang up during
667 * initilization of a channel.
668 */
669 channel->nb_init_stream_left = 0;
670
671 /* The reply msg status is handled in the following call. */
672 ret = create_ust_channel(channel, attr, &channel->uchan);
673 if (ret < 0) {
674 goto end;
675 }
676
677 channel->wait_fd = ustctl_channel_get_wait_fd(channel->uchan);
678
679 /*
680 * For the snapshots (no monitor), we create the metadata streams
681 * on demand, not during the channel creation.
682 */
683 if (channel->type == CONSUMER_CHANNEL_TYPE_METADATA && !channel->monitor) {
684 ret = 0;
685 goto end;
686 }
687
688 /* Open all streams for this channel. */
689 ret = create_ust_streams(channel, ctx);
690 if (ret < 0) {
691 goto end;
692 }
693
694 end:
695 return ret;
696 }
697
698 /*
699 * Send all stream of a channel to the right thread handling it.
700 *
701 * On error, return a negative value else 0 on success.
702 */
703 static int send_streams_to_thread(struct lttng_consumer_channel *channel,
704 struct lttng_consumer_local_data *ctx)
705 {
706 int ret = 0;
707 struct lttng_consumer_stream *stream, *stmp;
708
709 assert(channel);
710 assert(ctx);
711
712 /* Send streams to the corresponding thread. */
713 cds_list_for_each_entry_safe(stream, stmp, &channel->streams.head,
714 send_node) {
715
716 health_code_update();
717
718 /* Sending the stream to the thread. */
719 ret = send_stream_to_thread(stream, ctx);
720 if (ret < 0) {
721 /*
722 * If we are unable to send the stream to the thread, there is
723 * a big problem so just stop everything.
724 */
725 /* Remove node from the channel stream list. */
726 cds_list_del(&stream->send_node);
727 goto error;
728 }
729
730 /* Remove node from the channel stream list. */
731 cds_list_del(&stream->send_node);
732
733 }
734
735 error:
736 return ret;
737 }
738
739 /*
740 * Flush channel's streams using the given key to retrieve the channel.
741 *
742 * Return 0 on success else an LTTng error code.
743 */
744 static int flush_channel(uint64_t chan_key)
745 {
746 int ret = 0;
747 struct lttng_consumer_channel *channel;
748 struct lttng_consumer_stream *stream;
749 struct lttng_ht *ht;
750 struct lttng_ht_iter iter;
751
752 DBG("UST consumer flush channel key %" PRIu64, chan_key);
753
754 rcu_read_lock();
755 channel = consumer_find_channel(chan_key);
756 if (!channel) {
757 ERR("UST consumer flush channel %" PRIu64 " not found", chan_key);
758 ret = LTTNG_ERR_UST_CHAN_NOT_FOUND;
759 goto error;
760 }
761
762 ht = consumer_data.stream_per_chan_id_ht;
763
764 /* For each stream of the channel id, flush it. */
765 cds_lfht_for_each_entry_duplicate(ht->ht,
766 ht->hash_fct(&channel->key, lttng_ht_seed), ht->match_fct,
767 &channel->key, &iter.iter, stream, node_channel_id.node) {
768
769 health_code_update();
770
771 pthread_mutex_lock(&stream->lock);
772 if (!stream->quiescent) {
773 ustctl_flush_buffer(stream->ustream, 0);
774 stream->quiescent = true;
775 }
776 pthread_mutex_unlock(&stream->lock);
777 }
778 error:
779 rcu_read_unlock();
780 return ret;
781 }
782
783 /*
784 * Clear quiescent state from channel's streams using the given key to
785 * retrieve the channel.
786 *
787 * Return 0 on success else an LTTng error code.
788 */
789 static int clear_quiescent_channel(uint64_t chan_key)
790 {
791 int ret = 0;
792 struct lttng_consumer_channel *channel;
793 struct lttng_consumer_stream *stream;
794 struct lttng_ht *ht;
795 struct lttng_ht_iter iter;
796
797 DBG("UST consumer clear quiescent channel key %" PRIu64, chan_key);
798
799 rcu_read_lock();
800 channel = consumer_find_channel(chan_key);
801 if (!channel) {
802 ERR("UST consumer clear quiescent channel %" PRIu64 " not found", chan_key);
803 ret = LTTNG_ERR_UST_CHAN_NOT_FOUND;
804 goto error;
805 }
806
807 ht = consumer_data.stream_per_chan_id_ht;
808
809 /* For each stream of the channel id, clear quiescent state. */
810 cds_lfht_for_each_entry_duplicate(ht->ht,
811 ht->hash_fct(&channel->key, lttng_ht_seed), ht->match_fct,
812 &channel->key, &iter.iter, stream, node_channel_id.node) {
813
814 health_code_update();
815
816 pthread_mutex_lock(&stream->lock);
817 stream->quiescent = false;
818 pthread_mutex_unlock(&stream->lock);
819 }
820 error:
821 rcu_read_unlock();
822 return ret;
823 }
824
825 /*
826 * Close metadata stream wakeup_fd using the given key to retrieve the channel.
827 * RCU read side lock MUST be acquired before calling this function.
828 *
829 * Return 0 on success else an LTTng error code.
830 */
831 static int close_metadata(uint64_t chan_key)
832 {
833 int ret = 0;
834 struct lttng_consumer_channel *channel;
835
836 DBG("UST consumer close metadata key %" PRIu64, chan_key);
837
838 channel = consumer_find_channel(chan_key);
839 if (!channel) {
840 /*
841 * This is possible if the metadata thread has issue a delete because
842 * the endpoint point of the stream hung up. There is no way the
843 * session daemon can know about it thus use a DBG instead of an actual
844 * error.
845 */
846 DBG("UST consumer close metadata %" PRIu64 " not found", chan_key);
847 ret = LTTNG_ERR_UST_CHAN_NOT_FOUND;
848 goto error;
849 }
850
851 pthread_mutex_lock(&consumer_data.lock);
852 pthread_mutex_lock(&channel->lock);
853
854 if (cds_lfht_is_node_deleted(&channel->node.node)) {
855 goto error_unlock;
856 }
857
858 lttng_ustconsumer_close_metadata(channel);
859
860 error_unlock:
861 pthread_mutex_unlock(&channel->lock);
862 pthread_mutex_unlock(&consumer_data.lock);
863 error:
864 return ret;
865 }
866
867 /*
868 * RCU read side lock MUST be acquired before calling this function.
869 *
870 * Return 0 on success else an LTTng error code.
871 */
872 static int setup_metadata(struct lttng_consumer_local_data *ctx, uint64_t key)
873 {
874 int ret;
875 struct lttng_consumer_channel *metadata;
876
877 DBG("UST consumer setup metadata key %" PRIu64, key);
878
879 metadata = consumer_find_channel(key);
880 if (!metadata) {
881 ERR("UST consumer push metadata %" PRIu64 " not found", key);
882 ret = LTTNG_ERR_UST_CHAN_NOT_FOUND;
883 goto end;
884 }
885
886 /*
887 * In no monitor mode, the metadata channel has no stream(s) so skip the
888 * ownership transfer to the metadata thread.
889 */
890 if (!metadata->monitor) {
891 DBG("Metadata channel in no monitor");
892 ret = 0;
893 goto end;
894 }
895
896 /*
897 * Send metadata stream to relayd if one available. Availability is
898 * known if the stream is still in the list of the channel.
899 */
900 if (cds_list_empty(&metadata->streams.head)) {
901 ERR("Metadata channel key %" PRIu64 ", no stream available.", key);
902 ret = LTTCOMM_CONSUMERD_ERROR_METADATA;
903 goto error_no_stream;
904 }
905
906 /* Send metadata stream to relayd if needed. */
907 if (metadata->metadata_stream->net_seq_idx != (uint64_t) -1ULL) {
908 ret = consumer_send_relayd_stream(metadata->metadata_stream,
909 metadata->pathname);
910 if (ret < 0) {
911 ret = LTTCOMM_CONSUMERD_ERROR_METADATA;
912 goto error;
913 }
914 ret = consumer_send_relayd_streams_sent(
915 metadata->metadata_stream->net_seq_idx);
916 if (ret < 0) {
917 ret = LTTCOMM_CONSUMERD_RELAYD_FAIL;
918 goto error;
919 }
920 }
921
922 ret = send_streams_to_thread(metadata, ctx);
923 if (ret < 0) {
924 /*
925 * If we are unable to send the stream to the thread, there is
926 * a big problem so just stop everything.
927 */
928 ret = LTTCOMM_CONSUMERD_FATAL;
929 goto error;
930 }
931 /* List MUST be empty after or else it could be reused. */
932 assert(cds_list_empty(&metadata->streams.head));
933
934 ret = 0;
935 goto end;
936
937 error:
938 /*
939 * Delete metadata channel on error. At this point, the metadata stream can
940 * NOT be monitored by the metadata thread thus having the guarantee that
941 * the stream is still in the local stream list of the channel. This call
942 * will make sure to clean that list.
943 */
944 consumer_stream_destroy(metadata->metadata_stream, NULL);
945 cds_list_del(&metadata->metadata_stream->send_node);
946 metadata->metadata_stream = NULL;
947 error_no_stream:
948 end:
949 return ret;
950 }
951
952 /*
953 * Snapshot the whole metadata.
954 *
955 * Returns 0 on success, < 0 on error
956 */
957 static int snapshot_metadata(uint64_t key, char *path, uint64_t relayd_id,
958 struct lttng_consumer_local_data *ctx)
959 {
960 int ret = 0;
961 struct lttng_consumer_channel *metadata_channel;
962 struct lttng_consumer_stream *metadata_stream;
963
964 assert(path);
965 assert(ctx);
966
967 DBG("UST consumer snapshot metadata with key %" PRIu64 " at path %s",
968 key, path);
969
970 rcu_read_lock();
971
972 metadata_channel = consumer_find_channel(key);
973 if (!metadata_channel) {
974 ERR("UST snapshot metadata channel not found for key %" PRIu64,
975 key);
976 ret = -1;
977 goto error;
978 }
979 assert(!metadata_channel->monitor);
980
981 health_code_update();
982
983 /*
984 * Ask the sessiond if we have new metadata waiting and update the
985 * consumer metadata cache.
986 */
987 ret = lttng_ustconsumer_request_metadata(ctx, metadata_channel, 0, 1);
988 if (ret < 0) {
989 goto error;
990 }
991
992 health_code_update();
993
994 /*
995 * The metadata stream is NOT created in no monitor mode when the channel
996 * is created on a sessiond ask channel command.
997 */
998 ret = create_ust_streams(metadata_channel, ctx);
999 if (ret < 0) {
1000 goto error;
1001 }
1002
1003 metadata_stream = metadata_channel->metadata_stream;
1004 assert(metadata_stream);
1005
1006 if (relayd_id != (uint64_t) -1ULL) {
1007 metadata_stream->net_seq_idx = relayd_id;
1008 ret = consumer_send_relayd_stream(metadata_stream, path);
1009 if (ret < 0) {
1010 goto error_stream;
1011 }
1012 } else {
1013 ret = utils_create_stream_file(path, metadata_stream->name,
1014 metadata_stream->chan->tracefile_size,
1015 metadata_stream->tracefile_count_current,
1016 metadata_stream->uid, metadata_stream->gid, NULL);
1017 if (ret < 0) {
1018 goto error_stream;
1019 }
1020 metadata_stream->out_fd = ret;
1021 metadata_stream->tracefile_size_current = 0;
1022 }
1023
1024 do {
1025 health_code_update();
1026
1027 ret = lttng_consumer_read_subbuffer(metadata_stream, ctx);
1028 if (ret < 0) {
1029 goto error_stream;
1030 }
1031 } while (ret > 0);
1032
1033 error_stream:
1034 /*
1035 * Clean up the stream completly because the next snapshot will use a new
1036 * metadata stream.
1037 */
1038 consumer_stream_destroy(metadata_stream, NULL);
1039 cds_list_del(&metadata_stream->send_node);
1040 metadata_channel->metadata_stream = NULL;
1041
1042 error:
1043 rcu_read_unlock();
1044 return ret;
1045 }
1046
1047 /*
1048 * Take a snapshot of all the stream of a channel.
1049 *
1050 * Returns 0 on success, < 0 on error
1051 */
1052 static int snapshot_channel(uint64_t key, char *path, uint64_t relayd_id,
1053 uint64_t nb_packets_per_stream, struct lttng_consumer_local_data *ctx)
1054 {
1055 int ret;
1056 unsigned use_relayd = 0;
1057 unsigned long consumed_pos, produced_pos;
1058 struct lttng_consumer_channel *channel;
1059 struct lttng_consumer_stream *stream;
1060
1061 assert(path);
1062 assert(ctx);
1063
1064 rcu_read_lock();
1065
1066 if (relayd_id != (uint64_t) -1ULL) {
1067 use_relayd = 1;
1068 }
1069
1070 channel = consumer_find_channel(key);
1071 if (!channel) {
1072 ERR("UST snapshot channel not found for key %" PRIu64, key);
1073 ret = -1;
1074 goto error;
1075 }
1076 assert(!channel->monitor);
1077 DBG("UST consumer snapshot channel %" PRIu64, key);
1078
1079 cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
1080
1081 health_code_update();
1082
1083 /* Lock stream because we are about to change its state. */
1084 pthread_mutex_lock(&stream->lock);
1085 stream->net_seq_idx = relayd_id;
1086
1087 if (use_relayd) {
1088 ret = consumer_send_relayd_stream(stream, path);
1089 if (ret < 0) {
1090 goto error_unlock;
1091 }
1092 } else {
1093 ret = utils_create_stream_file(path, stream->name,
1094 stream->chan->tracefile_size,
1095 stream->tracefile_count_current,
1096 stream->uid, stream->gid, NULL);
1097 if (ret < 0) {
1098 goto error_unlock;
1099 }
1100 stream->out_fd = ret;
1101 stream->tracefile_size_current = 0;
1102
1103 DBG("UST consumer snapshot stream %s/%s (%" PRIu64 ")", path,
1104 stream->name, stream->key);
1105 }
1106 if (relayd_id != -1ULL) {
1107 ret = consumer_send_relayd_streams_sent(relayd_id);
1108 if (ret < 0) {
1109 goto error_unlock;
1110 }
1111 }
1112
1113 /*
1114 * If tracing is active, we want to perform a "full" buffer flush.
1115 * Else, if quiescent, it has already been done by the prior stop.
1116 */
1117 if (!stream->quiescent) {
1118 ustctl_flush_buffer(stream->ustream, 0);
1119 }
1120
1121 ret = lttng_ustconsumer_take_snapshot(stream);
1122 if (ret < 0) {
1123 ERR("Taking UST snapshot");
1124 goto error_unlock;
1125 }
1126
1127 ret = lttng_ustconsumer_get_produced_snapshot(stream, &produced_pos);
1128 if (ret < 0) {
1129 ERR("Produced UST snapshot position");
1130 goto error_unlock;
1131 }
1132
1133 ret = lttng_ustconsumer_get_consumed_snapshot(stream, &consumed_pos);
1134 if (ret < 0) {
1135 ERR("Consumerd UST snapshot position");
1136 goto error_unlock;
1137 }
1138
1139 /*
1140 * The original value is sent back if max stream size is larger than
1141 * the possible size of the snapshot. Also, we assume that the session
1142 * daemon should never send a maximum stream size that is lower than
1143 * subbuffer size.
1144 */
1145 consumed_pos = consumer_get_consume_start_pos(consumed_pos,
1146 produced_pos, nb_packets_per_stream,
1147 stream->max_sb_size);
1148
1149 while (consumed_pos < produced_pos) {
1150 ssize_t read_len;
1151 unsigned long len, padded_len;
1152
1153 health_code_update();
1154
1155 DBG("UST consumer taking snapshot at pos %lu", consumed_pos);
1156
1157 ret = ustctl_get_subbuf(stream->ustream, &consumed_pos);
1158 if (ret < 0) {
1159 if (ret != -EAGAIN) {
1160 PERROR("ustctl_get_subbuf snapshot");
1161 goto error_close_stream;
1162 }
1163 DBG("UST consumer get subbuf failed. Skipping it.");
1164 consumed_pos += stream->max_sb_size;
1165 continue;
1166 }
1167
1168 ret = ustctl_get_subbuf_size(stream->ustream, &len);
1169 if (ret < 0) {
1170 ERR("Snapshot ustctl_get_subbuf_size");
1171 goto error_put_subbuf;
1172 }
1173
1174 ret = ustctl_get_padded_subbuf_size(stream->ustream, &padded_len);
1175 if (ret < 0) {
1176 ERR("Snapshot ustctl_get_padded_subbuf_size");
1177 goto error_put_subbuf;
1178 }
1179
1180 read_len = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, len,
1181 padded_len - len, NULL);
1182 if (use_relayd) {
1183 if (read_len != len) {
1184 ret = -EPERM;
1185 goto error_put_subbuf;
1186 }
1187 } else {
1188 if (read_len != padded_len) {
1189 ret = -EPERM;
1190 goto error_put_subbuf;
1191 }
1192 }
1193
1194 ret = ustctl_put_subbuf(stream->ustream);
1195 if (ret < 0) {
1196 ERR("Snapshot ustctl_put_subbuf");
1197 goto error_close_stream;
1198 }
1199 consumed_pos += stream->max_sb_size;
1200 }
1201
1202 /* Simply close the stream so we can use it on the next snapshot. */
1203 consumer_stream_close(stream);
1204 pthread_mutex_unlock(&stream->lock);
1205 }
1206
1207 rcu_read_unlock();
1208 return 0;
1209
1210 error_put_subbuf:
1211 if (ustctl_put_subbuf(stream->ustream) < 0) {
1212 ERR("Snapshot ustctl_put_subbuf");
1213 }
1214 error_close_stream:
1215 consumer_stream_close(stream);
1216 error_unlock:
1217 pthread_mutex_unlock(&stream->lock);
1218 error:
1219 rcu_read_unlock();
1220 return ret;
1221 }
1222
1223 /*
1224 * Receive the metadata updates from the sessiond. Supports receiving
1225 * overlapping metadata, but is needs to always belong to a contiguous
1226 * range starting from 0.
1227 * Be careful about the locks held when calling this function: it needs
1228 * the metadata cache flush to concurrently progress in order to
1229 * complete.
1230 */
1231 int lttng_ustconsumer_recv_metadata(int sock, uint64_t key, uint64_t offset,
1232 uint64_t len, struct lttng_consumer_channel *channel,
1233 int timer, int wait)
1234 {
1235 int ret, ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1236 char *metadata_str;
1237
1238 DBG("UST consumer push metadata key %" PRIu64 " of len %" PRIu64, key, len);
1239
1240 metadata_str = zmalloc(len * sizeof(char));
1241 if (!metadata_str) {
1242 PERROR("zmalloc metadata string");
1243 ret_code = LTTCOMM_CONSUMERD_ENOMEM;
1244 goto end;
1245 }
1246
1247 health_code_update();
1248
1249 /* Receive metadata string. */
1250 ret = lttcomm_recv_unix_sock(sock, metadata_str, len);
1251 if (ret < 0) {
1252 /* Session daemon is dead so return gracefully. */
1253 ret_code = ret;
1254 goto end_free;
1255 }
1256
1257 health_code_update();
1258
1259 pthread_mutex_lock(&channel->metadata_cache->lock);
1260 ret = consumer_metadata_cache_write(channel, offset, len, metadata_str);
1261 if (ret < 0) {
1262 /* Unable to handle metadata. Notify session daemon. */
1263 ret_code = LTTCOMM_CONSUMERD_ERROR_METADATA;
1264 /*
1265 * Skip metadata flush on write error since the offset and len might
1266 * not have been updated which could create an infinite loop below when
1267 * waiting for the metadata cache to be flushed.
1268 */
1269 pthread_mutex_unlock(&channel->metadata_cache->lock);
1270 goto end_free;
1271 }
1272 pthread_mutex_unlock(&channel->metadata_cache->lock);
1273
1274 if (!wait) {
1275 goto end_free;
1276 }
1277 while (consumer_metadata_cache_flushed(channel, offset + len, timer)) {
1278 DBG("Waiting for metadata to be flushed");
1279
1280 health_code_update();
1281
1282 usleep(DEFAULT_METADATA_AVAILABILITY_WAIT_TIME);
1283 }
1284
1285 end_free:
1286 free(metadata_str);
1287 end:
1288 return ret_code;
1289 }
1290
1291 /*
1292 * Receive command from session daemon and process it.
1293 *
1294 * Return 1 on success else a negative value or 0.
1295 */
1296 int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
1297 int sock, struct pollfd *consumer_sockpoll)
1298 {
1299 ssize_t ret;
1300 enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1301 struct lttcomm_consumer_msg msg;
1302 struct lttng_consumer_channel *channel = NULL;
1303
1304 health_code_update();
1305
1306 ret = lttcomm_recv_unix_sock(sock, &msg, sizeof(msg));
1307 if (ret != sizeof(msg)) {
1308 DBG("Consumer received unexpected message size %zd (expects %zu)",
1309 ret, sizeof(msg));
1310 /*
1311 * The ret value might 0 meaning an orderly shutdown but this is ok
1312 * since the caller handles this.
1313 */
1314 if (ret > 0) {
1315 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD);
1316 ret = -1;
1317 }
1318 return ret;
1319 }
1320
1321 health_code_update();
1322
1323 /* deprecated */
1324 assert(msg.cmd_type != LTTNG_CONSUMER_STOP);
1325
1326 health_code_update();
1327
1328 /* relayd needs RCU read-side lock */
1329 rcu_read_lock();
1330
1331 switch (msg.cmd_type) {
1332 case LTTNG_CONSUMER_ADD_RELAYD_SOCKET:
1333 {
1334 /* Session daemon status message are handled in the following call. */
1335 ret = consumer_add_relayd_socket(msg.u.relayd_sock.net_index,
1336 msg.u.relayd_sock.type, ctx, sock, consumer_sockpoll,
1337 &msg.u.relayd_sock.sock, msg.u.relayd_sock.session_id,
1338 msg.u.relayd_sock.relayd_session_id);
1339 goto end_nosignal;
1340 }
1341 case LTTNG_CONSUMER_DESTROY_RELAYD:
1342 {
1343 uint64_t index = msg.u.destroy_relayd.net_seq_idx;
1344 struct consumer_relayd_sock_pair *relayd;
1345
1346 DBG("UST consumer destroying relayd %" PRIu64, index);
1347
1348 /* Get relayd reference if exists. */
1349 relayd = consumer_find_relayd(index);
1350 if (relayd == NULL) {
1351 DBG("Unable to find relayd %" PRIu64, index);
1352 ret_code = LTTCOMM_CONSUMERD_RELAYD_FAIL;
1353 }
1354
1355 /*
1356 * Each relayd socket pair has a refcount of stream attached to it
1357 * which tells if the relayd is still active or not depending on the
1358 * refcount value.
1359 *
1360 * This will set the destroy flag of the relayd object and destroy it
1361 * if the refcount reaches zero when called.
1362 *
1363 * The destroy can happen either here or when a stream fd hangs up.
1364 */
1365 if (relayd) {
1366 consumer_flag_relayd_for_destroy(relayd);
1367 }
1368
1369 goto end_msg_sessiond;
1370 }
1371 case LTTNG_CONSUMER_UPDATE_STREAM:
1372 {
1373 rcu_read_unlock();
1374 return -ENOSYS;
1375 }
1376 case LTTNG_CONSUMER_DATA_PENDING:
1377 {
1378 int ret, is_data_pending;
1379 uint64_t id = msg.u.data_pending.session_id;
1380
1381 DBG("UST consumer data pending command for id %" PRIu64, id);
1382
1383 is_data_pending = consumer_data_pending(id);
1384
1385 /* Send back returned value to session daemon */
1386 ret = lttcomm_send_unix_sock(sock, &is_data_pending,
1387 sizeof(is_data_pending));
1388 if (ret < 0) {
1389 DBG("Error when sending the data pending ret code: %d", ret);
1390 goto error_fatal;
1391 }
1392
1393 /*
1394 * No need to send back a status message since the data pending
1395 * returned value is the response.
1396 */
1397 break;
1398 }
1399 case LTTNG_CONSUMER_ASK_CHANNEL_CREATION:
1400 {
1401 int ret;
1402 struct ustctl_consumer_channel_attr attr;
1403
1404 /* Create a plain object and reserve a channel key. */
1405 channel = allocate_channel(msg.u.ask_channel.session_id,
1406 msg.u.ask_channel.pathname, msg.u.ask_channel.name,
1407 msg.u.ask_channel.uid, msg.u.ask_channel.gid,
1408 msg.u.ask_channel.relayd_id, msg.u.ask_channel.key,
1409 (enum lttng_event_output) msg.u.ask_channel.output,
1410 msg.u.ask_channel.tracefile_size,
1411 msg.u.ask_channel.tracefile_count,
1412 msg.u.ask_channel.session_id_per_pid,
1413 msg.u.ask_channel.monitor,
1414 msg.u.ask_channel.live_timer_interval,
1415 msg.u.ask_channel.root_shm_path,
1416 msg.u.ask_channel.shm_path);
1417 if (!channel) {
1418 goto end_channel_error;
1419 }
1420
1421 /*
1422 * Assign UST application UID to the channel. This value is ignored for
1423 * per PID buffers. This is specific to UST thus setting this after the
1424 * allocation.
1425 */
1426 channel->ust_app_uid = msg.u.ask_channel.ust_app_uid;
1427
1428 /* Build channel attributes from received message. */
1429 attr.subbuf_size = msg.u.ask_channel.subbuf_size;
1430 attr.num_subbuf = msg.u.ask_channel.num_subbuf;
1431 attr.overwrite = msg.u.ask_channel.overwrite;
1432 attr.switch_timer_interval = msg.u.ask_channel.switch_timer_interval;
1433 attr.read_timer_interval = msg.u.ask_channel.read_timer_interval;
1434 attr.chan_id = msg.u.ask_channel.chan_id;
1435 memcpy(attr.uuid, msg.u.ask_channel.uuid, sizeof(attr.uuid));
1436
1437 /* Match channel buffer type to the UST abi. */
1438 switch (msg.u.ask_channel.output) {
1439 case LTTNG_EVENT_MMAP:
1440 default:
1441 attr.output = LTTNG_UST_MMAP;
1442 break;
1443 }
1444
1445 /* Translate and save channel type. */
1446 switch (msg.u.ask_channel.type) {
1447 case LTTNG_UST_CHAN_PER_CPU:
1448 channel->type = CONSUMER_CHANNEL_TYPE_DATA;
1449 attr.type = LTTNG_UST_CHAN_PER_CPU;
1450 /*
1451 * Set refcount to 1 for owner. Below, we will
1452 * pass ownership to the
1453 * consumer_thread_channel_poll() thread.
1454 */
1455 channel->refcount = 1;
1456 break;
1457 case LTTNG_UST_CHAN_METADATA:
1458 channel->type = CONSUMER_CHANNEL_TYPE_METADATA;
1459 attr.type = LTTNG_UST_CHAN_METADATA;
1460 break;
1461 default:
1462 assert(0);
1463 goto error_fatal;
1464 };
1465
1466 health_code_update();
1467
1468 ret = ask_channel(ctx, sock, channel, &attr);
1469 if (ret < 0) {
1470 goto end_channel_error;
1471 }
1472
1473 if (msg.u.ask_channel.type == LTTNG_UST_CHAN_METADATA) {
1474 ret = consumer_metadata_cache_allocate(channel);
1475 if (ret < 0) {
1476 ERR("Allocating metadata cache");
1477 goto end_channel_error;
1478 }
1479 consumer_timer_switch_start(channel, attr.switch_timer_interval);
1480 attr.switch_timer_interval = 0;
1481 } else {
1482 consumer_timer_live_start(channel,
1483 msg.u.ask_channel.live_timer_interval);
1484 }
1485
1486 health_code_update();
1487
1488 /*
1489 * Add the channel to the internal state AFTER all streams were created
1490 * and successfully sent to session daemon. This way, all streams must
1491 * be ready before this channel is visible to the threads.
1492 * If add_channel succeeds, ownership of the channel is
1493 * passed to consumer_thread_channel_poll().
1494 */
1495 ret = add_channel(channel, ctx);
1496 if (ret < 0) {
1497 if (msg.u.ask_channel.type == LTTNG_UST_CHAN_METADATA) {
1498 if (channel->switch_timer_enabled == 1) {
1499 consumer_timer_switch_stop(channel);
1500 }
1501 consumer_metadata_cache_destroy(channel);
1502 }
1503 if (channel->live_timer_enabled == 1) {
1504 consumer_timer_live_stop(channel);
1505 }
1506 goto end_channel_error;
1507 }
1508
1509 health_code_update();
1510
1511 /*
1512 * Channel and streams are now created. Inform the session daemon that
1513 * everything went well and should wait to receive the channel and
1514 * streams with ustctl API.
1515 */
1516 ret = consumer_send_status_channel(sock, channel);
1517 if (ret < 0) {
1518 /*
1519 * There is probably a problem on the socket.
1520 */
1521 goto error_fatal;
1522 }
1523
1524 break;
1525 }
1526 case LTTNG_CONSUMER_GET_CHANNEL:
1527 {
1528 int ret, relayd_err = 0;
1529 uint64_t key = msg.u.get_channel.key;
1530 struct lttng_consumer_channel *channel;
1531
1532 channel = consumer_find_channel(key);
1533 if (!channel) {
1534 ERR("UST consumer get channel key %" PRIu64 " not found", key);
1535 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
1536 goto end_msg_sessiond;
1537 }
1538
1539 health_code_update();
1540
1541 /* Send everything to sessiond. */
1542 ret = send_sessiond_channel(sock, channel, ctx, &relayd_err);
1543 if (ret < 0) {
1544 if (relayd_err) {
1545 /*
1546 * We were unable to send to the relayd the stream so avoid
1547 * sending back a fatal error to the thread since this is OK
1548 * and the consumer can continue its work. The above call
1549 * has sent the error status message to the sessiond.
1550 */
1551 goto end_nosignal;
1552 }
1553 /*
1554 * The communicaton was broken hence there is a bad state between
1555 * the consumer and sessiond so stop everything.
1556 */
1557 goto error_fatal;
1558 }
1559
1560 health_code_update();
1561
1562 /*
1563 * In no monitor mode, the streams ownership is kept inside the channel
1564 * so don't send them to the data thread.
1565 */
1566 if (!channel->monitor) {
1567 goto end_msg_sessiond;
1568 }
1569
1570 ret = send_streams_to_thread(channel, ctx);
1571 if (ret < 0) {
1572 /*
1573 * If we are unable to send the stream to the thread, there is
1574 * a big problem so just stop everything.
1575 */
1576 goto error_fatal;
1577 }
1578 /* List MUST be empty after or else it could be reused. */
1579 assert(cds_list_empty(&channel->streams.head));
1580 goto end_msg_sessiond;
1581 }
1582 case LTTNG_CONSUMER_DESTROY_CHANNEL:
1583 {
1584 uint64_t key = msg.u.destroy_channel.key;
1585
1586 /*
1587 * Only called if streams have not been sent to stream
1588 * manager thread. However, channel has been sent to
1589 * channel manager thread.
1590 */
1591 notify_thread_del_channel(ctx, key);
1592 goto end_msg_sessiond;
1593 }
1594 case LTTNG_CONSUMER_CLOSE_METADATA:
1595 {
1596 int ret;
1597
1598 ret = close_metadata(msg.u.close_metadata.key);
1599 if (ret != 0) {
1600 ret_code = ret;
1601 }
1602
1603 goto end_msg_sessiond;
1604 }
1605 case LTTNG_CONSUMER_FLUSH_CHANNEL:
1606 {
1607 int ret;
1608
1609 ret = flush_channel(msg.u.flush_channel.key);
1610 if (ret != 0) {
1611 ret_code = ret;
1612 }
1613
1614 goto end_msg_sessiond;
1615 }
1616 case LTTNG_CONSUMER_CLEAR_QUIESCENT_CHANNEL:
1617 {
1618 int ret;
1619
1620 ret = clear_quiescent_channel(
1621 msg.u.clear_quiescent_channel.key);
1622 if (ret != 0) {
1623 ret_code = ret;
1624 }
1625
1626 goto end_msg_sessiond;
1627 }
1628 case LTTNG_CONSUMER_PUSH_METADATA:
1629 {
1630 int ret;
1631 uint64_t len = msg.u.push_metadata.len;
1632 uint64_t key = msg.u.push_metadata.key;
1633 uint64_t offset = msg.u.push_metadata.target_offset;
1634 struct lttng_consumer_channel *channel;
1635
1636 DBG("UST consumer push metadata key %" PRIu64 " of len %" PRIu64, key,
1637 len);
1638
1639 channel = consumer_find_channel(key);
1640 if (!channel) {
1641 /*
1642 * This is possible if the metadata creation on the consumer side
1643 * is in flight vis-a-vis a concurrent push metadata from the
1644 * session daemon. Simply return that the channel failed and the
1645 * session daemon will handle that message correctly considering
1646 * that this race is acceptable thus the DBG() statement here.
1647 */
1648 DBG("UST consumer push metadata %" PRIu64 " not found", key);
1649 ret_code = LTTCOMM_CONSUMERD_CHANNEL_FAIL;
1650 goto end_msg_sessiond;
1651 }
1652
1653 health_code_update();
1654
1655 if (!len) {
1656 /*
1657 * There is nothing to receive. We have simply
1658 * checked whether the channel can be found.
1659 */
1660 ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1661 goto end_msg_sessiond;
1662 }
1663
1664 /* Tell session daemon we are ready to receive the metadata. */
1665 ret = consumer_send_status_msg(sock, LTTCOMM_CONSUMERD_SUCCESS);
1666 if (ret < 0) {
1667 /* Somehow, the session daemon is not responding anymore. */
1668 goto error_fatal;
1669 }
1670
1671 health_code_update();
1672
1673 /* Wait for more data. */
1674 health_poll_entry();
1675 ret = lttng_consumer_poll_socket(consumer_sockpoll);
1676 health_poll_exit();
1677 if (ret) {
1678 goto error_fatal;
1679 }
1680
1681 health_code_update();
1682
1683 ret = lttng_ustconsumer_recv_metadata(sock, key, offset,
1684 len, channel, 0, 1);
1685 if (ret < 0) {
1686 /* error receiving from sessiond */
1687 goto error_fatal;
1688 } else {
1689 ret_code = ret;
1690 goto end_msg_sessiond;
1691 }
1692 }
1693 case LTTNG_CONSUMER_SETUP_METADATA:
1694 {
1695 int ret;
1696
1697 ret = setup_metadata(ctx, msg.u.setup_metadata.key);
1698 if (ret) {
1699 ret_code = ret;
1700 }
1701 goto end_msg_sessiond;
1702 }
1703 case LTTNG_CONSUMER_SNAPSHOT_CHANNEL:
1704 {
1705 if (msg.u.snapshot_channel.metadata) {
1706 ret = snapshot_metadata(msg.u.snapshot_channel.key,
1707 msg.u.snapshot_channel.pathname,
1708 msg.u.snapshot_channel.relayd_id,
1709 ctx);
1710 if (ret < 0) {
1711 ERR("Snapshot metadata failed");
1712 ret_code = LTTCOMM_CONSUMERD_ERROR_METADATA;
1713 }
1714 } else {
1715 ret = snapshot_channel(msg.u.snapshot_channel.key,
1716 msg.u.snapshot_channel.pathname,
1717 msg.u.snapshot_channel.relayd_id,
1718 msg.u.snapshot_channel.nb_packets_per_stream,
1719 ctx);
1720 if (ret < 0) {
1721 ERR("Snapshot channel failed");
1722 ret_code = LTTCOMM_CONSUMERD_CHANNEL_FAIL;
1723 }
1724 }
1725
1726 health_code_update();
1727 ret = consumer_send_status_msg(sock, ret_code);
1728 if (ret < 0) {
1729 /* Somehow, the session daemon is not responding anymore. */
1730 goto end_nosignal;
1731 }
1732 health_code_update();
1733 break;
1734 }
1735 default:
1736 break;
1737 }
1738
1739 end_nosignal:
1740 rcu_read_unlock();
1741
1742 health_code_update();
1743
1744 /*
1745 * Return 1 to indicate success since the 0 value can be a socket
1746 * shutdown during the recv() or send() call.
1747 */
1748 return 1;
1749
1750 end_msg_sessiond:
1751 /*
1752 * The returned value here is not useful since either way we'll return 1 to
1753 * the caller because the session daemon socket management is done
1754 * elsewhere. Returning a negative code or 0 will shutdown the consumer.
1755 */
1756 ret = consumer_send_status_msg(sock, ret_code);
1757 if (ret < 0) {
1758 goto error_fatal;
1759 }
1760 rcu_read_unlock();
1761
1762 health_code_update();
1763
1764 return 1;
1765 end_channel_error:
1766 if (channel) {
1767 /*
1768 * Free channel here since no one has a reference to it. We don't
1769 * free after that because a stream can store this pointer.
1770 */
1771 destroy_channel(channel);
1772 }
1773 /* We have to send a status channel message indicating an error. */
1774 ret = consumer_send_status_channel(sock, NULL);
1775 if (ret < 0) {
1776 /* Stop everything if session daemon can not be notified. */
1777 goto error_fatal;
1778 }
1779 rcu_read_unlock();
1780
1781 health_code_update();
1782
1783 return 1;
1784 error_fatal:
1785 rcu_read_unlock();
1786 /* This will issue a consumer stop. */
1787 return -1;
1788 }
1789
1790 /*
1791 * Wrapper over the mmap() read offset from ust-ctl library. Since this can be
1792 * compiled out, we isolate it in this library.
1793 */
1794 int lttng_ustctl_get_mmap_read_offset(struct lttng_consumer_stream *stream,
1795 unsigned long *off)
1796 {
1797 assert(stream);
1798 assert(stream->ustream);
1799
1800 return ustctl_get_mmap_read_offset(stream->ustream, off);
1801 }
1802
1803 /*
1804 * Wrapper over the mmap() read offset from ust-ctl library. Since this can be
1805 * compiled out, we isolate it in this library.
1806 */
1807 void *lttng_ustctl_get_mmap_base(struct lttng_consumer_stream *stream)
1808 {
1809 assert(stream);
1810 assert(stream->ustream);
1811
1812 return ustctl_get_mmap_base(stream->ustream);
1813 }
1814
1815 /*
1816 * Take a snapshot for a specific fd
1817 *
1818 * Returns 0 on success, < 0 on error
1819 */
1820 int lttng_ustconsumer_take_snapshot(struct lttng_consumer_stream *stream)
1821 {
1822 assert(stream);
1823 assert(stream->ustream);
1824
1825 return ustctl_snapshot(stream->ustream);
1826 }
1827
1828 /*
1829 * Get the produced position
1830 *
1831 * Returns 0 on success, < 0 on error
1832 */
1833 int lttng_ustconsumer_get_produced_snapshot(
1834 struct lttng_consumer_stream *stream, unsigned long *pos)
1835 {
1836 assert(stream);
1837 assert(stream->ustream);
1838 assert(pos);
1839
1840 return ustctl_snapshot_get_produced(stream->ustream, pos);
1841 }
1842
1843 /*
1844 * Get the consumed position
1845 *
1846 * Returns 0 on success, < 0 on error
1847 */
1848 int lttng_ustconsumer_get_consumed_snapshot(
1849 struct lttng_consumer_stream *stream, unsigned long *pos)
1850 {
1851 assert(stream);
1852 assert(stream->ustream);
1853 assert(pos);
1854
1855 return ustctl_snapshot_get_consumed(stream->ustream, pos);
1856 }
1857
1858 void lttng_ustconsumer_flush_buffer(struct lttng_consumer_stream *stream,
1859 int producer)
1860 {
1861 assert(stream);
1862 assert(stream->ustream);
1863
1864 ustctl_flush_buffer(stream->ustream, producer);
1865 }
1866
1867 int lttng_ustconsumer_get_current_timestamp(
1868 struct lttng_consumer_stream *stream, uint64_t *ts)
1869 {
1870 assert(stream);
1871 assert(stream->ustream);
1872 assert(ts);
1873
1874 return ustctl_get_current_timestamp(stream->ustream, ts);
1875 }
1876
1877 /*
1878 * Called when the stream signals the consumer that it has hung up.
1879 */
1880 void lttng_ustconsumer_on_stream_hangup(struct lttng_consumer_stream *stream)
1881 {
1882 assert(stream);
1883 assert(stream->ustream);
1884
1885 pthread_mutex_lock(&stream->lock);
1886 if (!stream->quiescent) {
1887 ustctl_flush_buffer(stream->ustream, 0);
1888 stream->quiescent = true;
1889 }
1890 pthread_mutex_unlock(&stream->lock);
1891 stream->hangup_flush_done = 1;
1892 }
1893
1894 void lttng_ustconsumer_del_channel(struct lttng_consumer_channel *chan)
1895 {
1896 int i;
1897
1898 assert(chan);
1899 assert(chan->uchan);
1900
1901 if (chan->switch_timer_enabled == 1) {
1902 consumer_timer_switch_stop(chan);
1903 }
1904 for (i = 0; i < chan->nr_stream_fds; i++) {
1905 int ret;
1906
1907 ret = close(chan->stream_fds[i]);
1908 if (ret) {
1909 PERROR("close");
1910 }
1911 if (chan->shm_path[0]) {
1912 char shm_path[PATH_MAX];
1913
1914 ret = get_stream_shm_path(shm_path, chan->shm_path, i);
1915 if (ret) {
1916 ERR("Cannot get stream shm path");
1917 }
1918 ret = run_as_unlink(shm_path, chan->uid, chan->gid);
1919 if (ret) {
1920 PERROR("unlink %s", shm_path);
1921 }
1922 }
1923 }
1924 }
1925
1926 void lttng_ustconsumer_free_channel(struct lttng_consumer_channel *chan)
1927 {
1928 assert(chan);
1929 assert(chan->uchan);
1930
1931 consumer_metadata_cache_destroy(chan);
1932 ustctl_destroy_channel(chan->uchan);
1933 /* Try to rmdir all directories under shm_path root. */
1934 if (chan->root_shm_path[0]) {
1935 (void) run_as_recursive_rmdir(chan->root_shm_path,
1936 chan->uid, chan->gid);
1937 }
1938 free(chan->stream_fds);
1939 }
1940
1941 void lttng_ustconsumer_del_stream(struct lttng_consumer_stream *stream)
1942 {
1943 assert(stream);
1944 assert(stream->ustream);
1945
1946 if (stream->chan->switch_timer_enabled == 1) {
1947 consumer_timer_switch_stop(stream->chan);
1948 }
1949 ustctl_destroy_stream(stream->ustream);
1950 }
1951
1952 int lttng_ustconsumer_get_wakeup_fd(struct lttng_consumer_stream *stream)
1953 {
1954 assert(stream);
1955 assert(stream->ustream);
1956
1957 return ustctl_stream_get_wakeup_fd(stream->ustream);
1958 }
1959
1960 int lttng_ustconsumer_close_wakeup_fd(struct lttng_consumer_stream *stream)
1961 {
1962 assert(stream);
1963 assert(stream->ustream);
1964
1965 return ustctl_stream_close_wakeup_fd(stream->ustream);
1966 }
1967
1968 /*
1969 * Populate index values of a UST stream. Values are set in big endian order.
1970 *
1971 * Return 0 on success or else a negative value.
1972 */
1973 static int get_index_values(struct ctf_packet_index *index,
1974 struct ustctl_consumer_stream *ustream)
1975 {
1976 int ret;
1977
1978 ret = ustctl_get_timestamp_begin(ustream, &index->timestamp_begin);
1979 if (ret < 0) {
1980 PERROR("ustctl_get_timestamp_begin");
1981 goto error;
1982 }
1983 index->timestamp_begin = htobe64(index->timestamp_begin);
1984
1985 ret = ustctl_get_timestamp_end(ustream, &index->timestamp_end);
1986 if (ret < 0) {
1987 PERROR("ustctl_get_timestamp_end");
1988 goto error;
1989 }
1990 index->timestamp_end = htobe64(index->timestamp_end);
1991
1992 ret = ustctl_get_events_discarded(ustream, &index->events_discarded);
1993 if (ret < 0) {
1994 PERROR("ustctl_get_events_discarded");
1995 goto error;
1996 }
1997 index->events_discarded = htobe64(index->events_discarded);
1998
1999 ret = ustctl_get_content_size(ustream, &index->content_size);
2000 if (ret < 0) {
2001 PERROR("ustctl_get_content_size");
2002 goto error;
2003 }
2004 index->content_size = htobe64(index->content_size);
2005
2006 ret = ustctl_get_packet_size(ustream, &index->packet_size);
2007 if (ret < 0) {
2008 PERROR("ustctl_get_packet_size");
2009 goto error;
2010 }
2011 index->packet_size = htobe64(index->packet_size);
2012
2013 ret = ustctl_get_stream_id(ustream, &index->stream_id);
2014 if (ret < 0) {
2015 PERROR("ustctl_get_stream_id");
2016 goto error;
2017 }
2018 index->stream_id = htobe64(index->stream_id);
2019
2020 error:
2021 return ret;
2022 }
2023
2024 /*
2025 * Write up to one packet from the metadata cache to the channel.
2026 *
2027 * Returns the number of bytes pushed in the cache, or a negative value
2028 * on error.
2029 */
2030 static
2031 int commit_one_metadata_packet(struct lttng_consumer_stream *stream)
2032 {
2033 ssize_t write_len;
2034 int ret;
2035
2036 pthread_mutex_lock(&stream->chan->metadata_cache->lock);
2037 if (stream->chan->metadata_cache->max_offset
2038 == stream->ust_metadata_pushed) {
2039 ret = 0;
2040 goto end;
2041 }
2042
2043 write_len = ustctl_write_one_packet_to_channel(stream->chan->uchan,
2044 &stream->chan->metadata_cache->data[stream->ust_metadata_pushed],
2045 stream->chan->metadata_cache->max_offset
2046 - stream->ust_metadata_pushed);
2047 assert(write_len != 0);
2048 if (write_len < 0) {
2049 ERR("Writing one metadata packet");
2050 ret = -1;
2051 goto end;
2052 }
2053 stream->ust_metadata_pushed += write_len;
2054
2055 assert(stream->chan->metadata_cache->max_offset >=
2056 stream->ust_metadata_pushed);
2057 ret = write_len;
2058
2059 end:
2060 pthread_mutex_unlock(&stream->chan->metadata_cache->lock);
2061 return ret;
2062 }
2063
2064
2065 /*
2066 * Sync metadata meaning request them to the session daemon and snapshot to the
2067 * metadata thread can consumer them.
2068 *
2069 * Metadata stream lock is held here, but we need to release it when
2070 * interacting with sessiond, else we cause a deadlock with live
2071 * awaiting on metadata to be pushed out.
2072 *
2073 * Return 0 if new metadatda is available, EAGAIN if the metadata stream
2074 * is empty or a negative value on error.
2075 */
2076 int lttng_ustconsumer_sync_metadata(struct lttng_consumer_local_data *ctx,
2077 struct lttng_consumer_stream *metadata)
2078 {
2079 int ret;
2080 int retry = 0;
2081
2082 assert(ctx);
2083 assert(metadata);
2084
2085 pthread_mutex_unlock(&metadata->lock);
2086 /*
2087 * Request metadata from the sessiond, but don't wait for the flush
2088 * because we locked the metadata thread.
2089 */
2090 ret = lttng_ustconsumer_request_metadata(ctx, metadata->chan, 0, 0);
2091 if (ret < 0) {
2092 goto end;
2093 }
2094 pthread_mutex_lock(&metadata->lock);
2095
2096 ret = commit_one_metadata_packet(metadata);
2097 if (ret <= 0) {
2098 goto end;
2099 } else if (ret > 0) {
2100 retry = 1;
2101 }
2102
2103 ustctl_flush_buffer(metadata->ustream, 1);
2104 ret = ustctl_snapshot(metadata->ustream);
2105 if (ret < 0) {
2106 if (errno != EAGAIN) {
2107 ERR("Sync metadata, taking UST snapshot");
2108 goto end;
2109 }
2110 DBG("No new metadata when syncing them.");
2111 /* No new metadata, exit. */
2112 ret = ENODATA;
2113 goto end;
2114 }
2115
2116 /*
2117 * After this flush, we still need to extract metadata.
2118 */
2119 if (retry) {
2120 ret = EAGAIN;
2121 }
2122
2123 end:
2124 return ret;
2125 }
2126
2127 /*
2128 * Return 0 on success else a negative value.
2129 */
2130 static int notify_if_more_data(struct lttng_consumer_stream *stream,
2131 struct lttng_consumer_local_data *ctx)
2132 {
2133 int ret;
2134 struct ustctl_consumer_stream *ustream;
2135
2136 assert(stream);
2137 assert(ctx);
2138
2139 ustream = stream->ustream;
2140
2141 /*
2142 * First, we are going to check if there is a new subbuffer available
2143 * before reading the stream wait_fd.
2144 */
2145 /* Get the next subbuffer */
2146 ret = ustctl_get_next_subbuf(ustream);
2147 if (ret) {
2148 /* No more data found, flag the stream. */
2149 stream->has_data = 0;
2150 ret = 0;
2151 goto end;
2152 }
2153
2154 ret = ustctl_put_subbuf(ustream);
2155 assert(!ret);
2156
2157 /* This stream still has data. Flag it and wake up the data thread. */
2158 stream->has_data = 1;
2159
2160 if (stream->monitor && !stream->hangup_flush_done && !ctx->has_wakeup) {
2161 ssize_t writelen;
2162
2163 writelen = lttng_pipe_write(ctx->consumer_wakeup_pipe, "!", 1);
2164 if (writelen < 0 && errno != EAGAIN && errno != EWOULDBLOCK) {
2165 ret = writelen;
2166 goto end;
2167 }
2168
2169 /* The wake up pipe has been notified. */
2170 ctx->has_wakeup = 1;
2171 }
2172 ret = 0;
2173
2174 end:
2175 return ret;
2176 }
2177
2178 /*
2179 * Read subbuffer from the given stream.
2180 *
2181 * Stream lock MUST be acquired.
2182 *
2183 * Return 0 on success else a negative value.
2184 */
2185 int lttng_ustconsumer_read_subbuffer(struct lttng_consumer_stream *stream,
2186 struct lttng_consumer_local_data *ctx)
2187 {
2188 unsigned long len, subbuf_size, padding;
2189 int err, write_index = 1;
2190 long ret = 0;
2191 struct ustctl_consumer_stream *ustream;
2192 struct ctf_packet_index index;
2193
2194 assert(stream);
2195 assert(stream->ustream);
2196 assert(ctx);
2197
2198 DBG("In UST read_subbuffer (wait_fd: %d, name: %s)", stream->wait_fd,
2199 stream->name);
2200
2201 /* Ease our life for what's next. */
2202 ustream = stream->ustream;
2203
2204 /*
2205 * We can consume the 1 byte written into the wait_fd by UST. Don't trigger
2206 * error if we cannot read this one byte (read returns 0), or if the error
2207 * is EAGAIN or EWOULDBLOCK.
2208 *
2209 * This is only done when the stream is monitored by a thread, before the
2210 * flush is done after a hangup and if the stream is not flagged with data
2211 * since there might be nothing to consume in the wait fd but still have
2212 * data available flagged by the consumer wake up pipe.
2213 */
2214 if (stream->monitor && !stream->hangup_flush_done && !stream->has_data) {
2215 char dummy;
2216 ssize_t readlen;
2217
2218 readlen = lttng_read(stream->wait_fd, &dummy, 1);
2219 if (readlen < 0 && errno != EAGAIN && errno != EWOULDBLOCK) {
2220 ret = readlen;
2221 goto end;
2222 }
2223 }
2224
2225 retry:
2226 /* Get the next subbuffer */
2227 err = ustctl_get_next_subbuf(ustream);
2228 if (err != 0) {
2229 /*
2230 * Populate metadata info if the existing info has
2231 * already been read.
2232 */
2233 if (stream->metadata_flag) {
2234 ret = commit_one_metadata_packet(stream);
2235 if (ret <= 0) {
2236 goto end;
2237 }
2238 ustctl_flush_buffer(stream->ustream, 1);
2239 goto retry;
2240 }
2241
2242 ret = err; /* ustctl_get_next_subbuf returns negative, caller expect positive. */
2243 /*
2244 * This is a debug message even for single-threaded consumer,
2245 * because poll() have more relaxed criterions than get subbuf,
2246 * so get_subbuf may fail for short race windows where poll()
2247 * would issue wakeups.
2248 */
2249 DBG("Reserving sub buffer failed (everything is normal, "
2250 "it is due to concurrency) [ret: %d]", err);
2251 goto end;
2252 }
2253 assert(stream->chan->output == CONSUMER_CHANNEL_MMAP);
2254
2255 if (!stream->metadata_flag) {
2256 index.offset = htobe64(stream->out_fd_offset);
2257 ret = get_index_values(&index, ustream);
2258 if (ret < 0) {
2259 goto end;
2260 }
2261 } else {
2262 write_index = 0;
2263 }
2264
2265 /* Get the full padded subbuffer size */
2266 err = ustctl_get_padded_subbuf_size(ustream, &len);
2267 assert(err == 0);
2268
2269 /* Get subbuffer data size (without padding) */
2270 err = ustctl_get_subbuf_size(ustream, &subbuf_size);
2271 assert(err == 0);
2272
2273 /* Make sure we don't get a subbuffer size bigger than the padded */
2274 assert(len >= subbuf_size);
2275
2276 padding = len - subbuf_size;
2277 /* write the subbuffer to the tracefile */
2278 ret = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, subbuf_size, padding, &index);
2279 /*
2280 * The mmap operation should write subbuf_size amount of data when network
2281 * streaming or the full padding (len) size when we are _not_ streaming.
2282 */
2283 if ((ret != subbuf_size && stream->net_seq_idx != (uint64_t) -1ULL) ||
2284 (ret != len && stream->net_seq_idx == (uint64_t) -1ULL)) {
2285 /*
2286 * Display the error but continue processing to try to release the
2287 * subbuffer. This is a DBG statement since any unexpected kill or
2288 * signal, the application gets unregistered, relayd gets closed or
2289 * anything that affects the buffer lifetime will trigger this error.
2290 * So, for the sake of the user, don't print this error since it can
2291 * happen and it is OK with the code flow.
2292 */
2293 DBG("Error writing to tracefile "
2294 "(ret: %ld != len: %lu != subbuf_size: %lu)",
2295 ret, len, subbuf_size);
2296 write_index = 0;
2297 }
2298 err = ustctl_put_next_subbuf(ustream);
2299 assert(err == 0);
2300
2301 /*
2302 * This will consumer the byte on the wait_fd if and only if there is not
2303 * next subbuffer to be acquired.
2304 */
2305 if (!stream->metadata_flag) {
2306 ret = notify_if_more_data(stream, ctx);
2307 if (ret < 0) {
2308 goto end;
2309 }
2310 }
2311
2312 /* Write index if needed. */
2313 if (!write_index) {
2314 goto end;
2315 }
2316
2317 if (stream->chan->live_timer_interval && !stream->metadata_flag) {
2318 /*
2319 * In live, block until all the metadata is sent.
2320 */
2321 pthread_mutex_lock(&stream->metadata_timer_lock);
2322 assert(!stream->missed_metadata_flush);
2323 stream->waiting_on_metadata = true;
2324 pthread_mutex_unlock(&stream->metadata_timer_lock);
2325
2326 err = consumer_stream_sync_metadata(ctx, stream->session_id);
2327
2328 pthread_mutex_lock(&stream->metadata_timer_lock);
2329 stream->waiting_on_metadata = false;
2330 if (stream->missed_metadata_flush) {
2331 stream->missed_metadata_flush = false;
2332 pthread_mutex_unlock(&stream->metadata_timer_lock);
2333 (void) consumer_flush_ust_index(stream);
2334 } else {
2335 pthread_mutex_unlock(&stream->metadata_timer_lock);
2336 }
2337
2338 if (err < 0) {
2339 goto end;
2340 }
2341 }
2342
2343 assert(!stream->metadata_flag);
2344 err = consumer_stream_write_index(stream, &index);
2345 if (err < 0) {
2346 goto end;
2347 }
2348
2349 end:
2350 return ret;
2351 }
2352
2353 /*
2354 * Called when a stream is created.
2355 *
2356 * Return 0 on success or else a negative value.
2357 */
2358 int lttng_ustconsumer_on_recv_stream(struct lttng_consumer_stream *stream)
2359 {
2360 int ret;
2361
2362 assert(stream);
2363
2364 /* Don't create anything if this is set for streaming. */
2365 if (stream->net_seq_idx == (uint64_t) -1ULL && stream->chan->monitor) {
2366 ret = utils_create_stream_file(stream->chan->pathname, stream->name,
2367 stream->chan->tracefile_size, stream->tracefile_count_current,
2368 stream->uid, stream->gid, NULL);
2369 if (ret < 0) {
2370 goto error;
2371 }
2372 stream->out_fd = ret;
2373 stream->tracefile_size_current = 0;
2374
2375 if (!stream->metadata_flag) {
2376 ret = index_create_file(stream->chan->pathname,
2377 stream->name, stream->uid, stream->gid,
2378 stream->chan->tracefile_size,
2379 stream->tracefile_count_current);
2380 if (ret < 0) {
2381 goto error;
2382 }
2383 stream->index_fd = ret;
2384 }
2385 }
2386 ret = 0;
2387
2388 error:
2389 return ret;
2390 }
2391
2392 /*
2393 * Check if data is still being extracted from the buffers for a specific
2394 * stream. Consumer data lock MUST be acquired before calling this function
2395 * and the stream lock.
2396 *
2397 * Return 1 if the traced data are still getting read else 0 meaning that the
2398 * data is available for trace viewer reading.
2399 */
2400 int lttng_ustconsumer_data_pending(struct lttng_consumer_stream *stream)
2401 {
2402 int ret;
2403
2404 assert(stream);
2405 assert(stream->ustream);
2406
2407 DBG("UST consumer checking data pending");
2408
2409 if (stream->endpoint_status != CONSUMER_ENDPOINT_ACTIVE) {
2410 ret = 0;
2411 goto end;
2412 }
2413
2414 if (stream->chan->type == CONSUMER_CHANNEL_TYPE_METADATA) {
2415 uint64_t contiguous, pushed;
2416
2417 /* Ease our life a bit. */
2418 contiguous = stream->chan->metadata_cache->max_offset;
2419 pushed = stream->ust_metadata_pushed;
2420
2421 /*
2422 * We can simply check whether all contiguously available data
2423 * has been pushed to the ring buffer, since the push operation
2424 * is performed within get_next_subbuf(), and because both
2425 * get_next_subbuf() and put_next_subbuf() are issued atomically
2426 * thanks to the stream lock within
2427 * lttng_ustconsumer_read_subbuffer(). This basically means that
2428 * whetnever ust_metadata_pushed is incremented, the associated
2429 * metadata has been consumed from the metadata stream.
2430 */
2431 DBG("UST consumer metadata pending check: contiguous %" PRIu64 " vs pushed %" PRIu64,
2432 contiguous, pushed);
2433 assert(((int64_t) (contiguous - pushed)) >= 0);
2434 if ((contiguous != pushed) ||
2435 (((int64_t) contiguous - pushed) > 0 || contiguous == 0)) {
2436 ret = 1; /* Data is pending */
2437 goto end;
2438 }
2439 } else {
2440 ret = ustctl_get_next_subbuf(stream->ustream);
2441 if (ret == 0) {
2442 /*
2443 * There is still data so let's put back this
2444 * subbuffer.
2445 */
2446 ret = ustctl_put_subbuf(stream->ustream);
2447 assert(ret == 0);
2448 ret = 1; /* Data is pending */
2449 goto end;
2450 }
2451 }
2452
2453 /* Data is NOT pending so ready to be read. */
2454 ret = 0;
2455
2456 end:
2457 return ret;
2458 }
2459
2460 /*
2461 * Stop a given metadata channel timer if enabled and close the wait fd which
2462 * is the poll pipe of the metadata stream.
2463 *
2464 * This MUST be called with the metadata channel acquired.
2465 */
2466 void lttng_ustconsumer_close_metadata(struct lttng_consumer_channel *metadata)
2467 {
2468 int ret;
2469
2470 assert(metadata);
2471 assert(metadata->type == CONSUMER_CHANNEL_TYPE_METADATA);
2472
2473 DBG("Closing metadata channel key %" PRIu64, metadata->key);
2474
2475 if (metadata->switch_timer_enabled == 1) {
2476 consumer_timer_switch_stop(metadata);
2477 }
2478
2479 if (!metadata->metadata_stream) {
2480 goto end;
2481 }
2482
2483 /*
2484 * Closing write side so the thread monitoring the stream wakes up if any
2485 * and clean the metadata stream.
2486 */
2487 if (metadata->metadata_stream->ust_metadata_poll_pipe[1] >= 0) {
2488 ret = close(metadata->metadata_stream->ust_metadata_poll_pipe[1]);
2489 if (ret < 0) {
2490 PERROR("closing metadata pipe write side");
2491 }
2492 metadata->metadata_stream->ust_metadata_poll_pipe[1] = -1;
2493 }
2494
2495 end:
2496 return;
2497 }
2498
2499 /*
2500 * Close every metadata stream wait fd of the metadata hash table. This
2501 * function MUST be used very carefully so not to run into a race between the
2502 * metadata thread handling streams and this function closing their wait fd.
2503 *
2504 * For UST, this is used when the session daemon hangs up. Its the metadata
2505 * producer so calling this is safe because we are assured that no state change
2506 * can occur in the metadata thread for the streams in the hash table.
2507 */
2508 void lttng_ustconsumer_close_all_metadata(struct lttng_ht *metadata_ht)
2509 {
2510 struct lttng_ht_iter iter;
2511 struct lttng_consumer_stream *stream;
2512
2513 assert(metadata_ht);
2514 assert(metadata_ht->ht);
2515
2516 DBG("UST consumer closing all metadata streams");
2517
2518 rcu_read_lock();
2519 cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream,
2520 node.node) {
2521
2522 health_code_update();
2523
2524 pthread_mutex_lock(&stream->chan->lock);
2525 lttng_ustconsumer_close_metadata(stream->chan);
2526 pthread_mutex_unlock(&stream->chan->lock);
2527
2528 }
2529 rcu_read_unlock();
2530 }
2531
2532 void lttng_ustconsumer_close_stream_wakeup(struct lttng_consumer_stream *stream)
2533 {
2534 int ret;
2535
2536 ret = ustctl_stream_close_wakeup_fd(stream->ustream);
2537 if (ret < 0) {
2538 ERR("Unable to close wakeup fd");
2539 }
2540 }
2541
2542 /*
2543 * Please refer to consumer-timer.c before adding any lock within this
2544 * function or any of its callees. Timers have a very strict locking
2545 * semantic with respect to teardown. Failure to respect this semantic
2546 * introduces deadlocks.
2547 *
2548 * DON'T hold the metadata lock when calling this function, else this
2549 * can cause deadlock involving consumer awaiting for metadata to be
2550 * pushed out due to concurrent interaction with the session daemon.
2551 */
2552 int lttng_ustconsumer_request_metadata(struct lttng_consumer_local_data *ctx,
2553 struct lttng_consumer_channel *channel, int timer, int wait)
2554 {
2555 struct lttcomm_metadata_request_msg request;
2556 struct lttcomm_consumer_msg msg;
2557 enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS;
2558 uint64_t len, key, offset;
2559 int ret;
2560
2561 assert(channel);
2562 assert(channel->metadata_cache);
2563
2564 memset(&request, 0, sizeof(request));
2565
2566 /* send the metadata request to sessiond */
2567 switch (consumer_data.type) {
2568 case LTTNG_CONSUMER64_UST:
2569 request.bits_per_long = 64;
2570 break;
2571 case LTTNG_CONSUMER32_UST:
2572 request.bits_per_long = 32;
2573 break;
2574 default:
2575 request.bits_per_long = 0;
2576 break;
2577 }
2578
2579 request.session_id = channel->session_id;
2580 request.session_id_per_pid = channel->session_id_per_pid;
2581 /*
2582 * Request the application UID here so the metadata of that application can
2583 * be sent back. The channel UID corresponds to the user UID of the session
2584 * used for the rights on the stream file(s).
2585 */
2586 request.uid = channel->ust_app_uid;
2587 request.key = channel->key;
2588
2589 DBG("Sending metadata request to sessiond, session id %" PRIu64
2590 ", per-pid %" PRIu64 ", app UID %u and channek key %" PRIu64,
2591 request.session_id, request.session_id_per_pid, request.uid,
2592 request.key);
2593
2594 pthread_mutex_lock(&ctx->metadata_socket_lock);
2595
2596 health_code_update();
2597
2598 ret = lttcomm_send_unix_sock(ctx->consumer_metadata_socket, &request,
2599 sizeof(request));
2600 if (ret < 0) {
2601 ERR("Asking metadata to sessiond");
2602 goto end;
2603 }
2604
2605 health_code_update();
2606
2607 /* Receive the metadata from sessiond */
2608 ret = lttcomm_recv_unix_sock(ctx->consumer_metadata_socket, &msg,
2609 sizeof(msg));
2610 if (ret != sizeof(msg)) {
2611 DBG("Consumer received unexpected message size %d (expects %zu)",
2612 ret, sizeof(msg));
2613 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD);
2614 /*
2615 * The ret value might 0 meaning an orderly shutdown but this is ok
2616 * since the caller handles this.
2617 */
2618 goto end;
2619 }
2620
2621 health_code_update();
2622
2623 if (msg.cmd_type == LTTNG_ERR_UND) {
2624 /* No registry found */
2625 (void) consumer_send_status_msg(ctx->consumer_metadata_socket,
2626 ret_code);
2627 ret = 0;
2628 goto end;
2629 } else if (msg.cmd_type != LTTNG_CONSUMER_PUSH_METADATA) {
2630 ERR("Unexpected cmd_type received %d", msg.cmd_type);
2631 ret = -1;
2632 goto end;
2633 }
2634
2635 len = msg.u.push_metadata.len;
2636 key = msg.u.push_metadata.key;
2637 offset = msg.u.push_metadata.target_offset;
2638
2639 assert(key == channel->key);
2640 if (len == 0) {
2641 DBG("No new metadata to receive for key %" PRIu64, key);
2642 }
2643
2644 health_code_update();
2645
2646 /* Tell session daemon we are ready to receive the metadata. */
2647 ret = consumer_send_status_msg(ctx->consumer_metadata_socket,
2648 LTTCOMM_CONSUMERD_SUCCESS);
2649 if (ret < 0 || len == 0) {
2650 /*
2651 * Somehow, the session daemon is not responding anymore or there is
2652 * nothing to receive.
2653 */
2654 goto end;
2655 }
2656
2657 health_code_update();
2658
2659 ret = lttng_ustconsumer_recv_metadata(ctx->consumer_metadata_socket,
2660 key, offset, len, channel, timer, wait);
2661 if (ret >= 0) {
2662 /*
2663 * Only send the status msg if the sessiond is alive meaning a positive
2664 * ret code.
2665 */
2666 (void) consumer_send_status_msg(ctx->consumer_metadata_socket, ret);
2667 }
2668 ret = 0;
2669
2670 end:
2671 health_code_update();
2672
2673 pthread_mutex_unlock(&ctx->metadata_socket_lock);
2674 return ret;
2675 }
2676
2677 /*
2678 * Return the ustctl call for the get stream id.
2679 */
2680 int lttng_ustconsumer_get_stream_id(struct lttng_consumer_stream *stream,
2681 uint64_t *stream_id)
2682 {
2683 assert(stream);
2684 assert(stream_id);
2685
2686 return ustctl_get_stream_id(stream->ustream, stream_id);
2687 }
This page took 0.141123 seconds and 4 git commands to generate.