Fix: clarify man page lttng.1
[lttng-tools.git] / src / common / kernel-consumer / kernel-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 #include <assert.h>
21 #include <poll.h>
22 #include <pthread.h>
23 #include <stdlib.h>
24 #include <string.h>
25 #include <sys/mman.h>
26 #include <sys/socket.h>
27 #include <sys/types.h>
28 #include <inttypes.h>
29 #include <unistd.h>
30 #include <sys/stat.h>
31
32 #include <bin/lttng-consumerd/health-consumerd.h>
33 #include <common/common.h>
34 #include <common/kernel-ctl/kernel-ctl.h>
35 #include <common/sessiond-comm/sessiond-comm.h>
36 #include <common/sessiond-comm/relayd.h>
37 #include <common/compat/fcntl.h>
38 #include <common/compat/endian.h>
39 #include <common/pipe.h>
40 #include <common/relayd/relayd.h>
41 #include <common/utils.h>
42 #include <common/consumer-stream.h>
43 #include <common/index/index.h>
44 #include <common/consumer-timer.h>
45
46 #include "kernel-consumer.h"
47
48 extern struct lttng_consumer_global_data consumer_data;
49 extern int consumer_poll_timeout;
50 extern volatile int consumer_quit;
51
52 /*
53 * Take a snapshot for a specific fd
54 *
55 * Returns 0 on success, < 0 on error
56 */
57 int lttng_kconsumer_take_snapshot(struct lttng_consumer_stream *stream)
58 {
59 int ret = 0;
60 int infd = stream->wait_fd;
61
62 ret = kernctl_snapshot(infd);
63 if (ret != 0) {
64 perror("Getting sub-buffer snapshot.");
65 ret = -errno;
66 }
67
68 return ret;
69 }
70
71 /*
72 * Get the produced position
73 *
74 * Returns 0 on success, < 0 on error
75 */
76 int lttng_kconsumer_get_produced_snapshot(struct lttng_consumer_stream *stream,
77 unsigned long *pos)
78 {
79 int ret;
80 int infd = stream->wait_fd;
81
82 ret = kernctl_snapshot_get_produced(infd, pos);
83 if (ret != 0) {
84 perror("kernctl_snapshot_get_produced");
85 ret = -errno;
86 }
87
88 return ret;
89 }
90
91 /*
92 * Get the consumerd position
93 *
94 * Returns 0 on success, < 0 on error
95 */
96 int lttng_kconsumer_get_consumed_snapshot(struct lttng_consumer_stream *stream,
97 unsigned long *pos)
98 {
99 int ret;
100 int infd = stream->wait_fd;
101
102 ret = kernctl_snapshot_get_consumed(infd, pos);
103 if (ret != 0) {
104 perror("kernctl_snapshot_get_consumed");
105 ret = -errno;
106 }
107
108 return ret;
109 }
110
111 /*
112 * Take a snapshot of all the stream of a channel
113 *
114 * Returns 0 on success, < 0 on error
115 */
116 int lttng_kconsumer_snapshot_channel(uint64_t key, char *path,
117 uint64_t relayd_id, uint64_t max_stream_size,
118 struct lttng_consumer_local_data *ctx)
119 {
120 int ret;
121 unsigned long consumed_pos, produced_pos;
122 struct lttng_consumer_channel *channel;
123 struct lttng_consumer_stream *stream;
124
125 DBG("Kernel consumer snapshot channel %" PRIu64, key);
126
127 rcu_read_lock();
128
129 channel = consumer_find_channel(key);
130 if (!channel) {
131 ERR("No channel found for key %" PRIu64, key);
132 ret = -1;
133 goto end;
134 }
135
136 /* Splice is not supported yet for channel snapshot. */
137 if (channel->output != CONSUMER_CHANNEL_MMAP) {
138 ERR("Unsupported output %d", channel->output);
139 ret = -1;
140 goto end;
141 }
142
143 cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
144
145 health_code_update();
146
147 /*
148 * Lock stream because we are about to change its state.
149 */
150 pthread_mutex_lock(&stream->lock);
151
152 /*
153 * Assign the received relayd ID so we can use it for streaming. The streams
154 * are not visible to anyone so this is OK to change it.
155 */
156 stream->net_seq_idx = relayd_id;
157 channel->relayd_id = relayd_id;
158 if (relayd_id != (uint64_t) -1ULL) {
159 ret = consumer_send_relayd_stream(stream, path);
160 if (ret < 0) {
161 ERR("sending stream to relayd");
162 goto end_unlock;
163 }
164 } else {
165 ret = utils_create_stream_file(path, stream->name,
166 stream->chan->tracefile_size,
167 stream->tracefile_count_current,
168 stream->uid, stream->gid, NULL);
169 if (ret < 0) {
170 ERR("utils_create_stream_file");
171 goto end_unlock;
172 }
173
174 stream->out_fd = ret;
175 stream->tracefile_size_current = 0;
176
177 DBG("Kernel consumer snapshot stream %s/%s (%" PRIu64 ")",
178 path, stream->name, stream->key);
179 }
180 if (relayd_id != -1ULL) {
181 ret = consumer_send_relayd_streams_sent(relayd_id);
182 if (ret < 0) {
183 ERR("sending streams sent to relayd");
184 goto end_unlock;
185 }
186 }
187
188 ret = kernctl_buffer_flush(stream->wait_fd);
189 if (ret < 0) {
190 ERR("Failed to flush kernel stream");
191 ret = -errno;
192 goto end_unlock;
193 }
194
195 ret = lttng_kconsumer_take_snapshot(stream);
196 if (ret < 0) {
197 ERR("Taking kernel snapshot");
198 goto end_unlock;
199 }
200
201 ret = lttng_kconsumer_get_produced_snapshot(stream, &produced_pos);
202 if (ret < 0) {
203 ERR("Produced kernel snapshot position");
204 goto end_unlock;
205 }
206
207 ret = lttng_kconsumer_get_consumed_snapshot(stream, &consumed_pos);
208 if (ret < 0) {
209 ERR("Consumerd kernel snapshot position");
210 goto end_unlock;
211 }
212
213 if (stream->max_sb_size == 0) {
214 ret = kernctl_get_max_subbuf_size(stream->wait_fd,
215 &stream->max_sb_size);
216 if (ret < 0) {
217 ERR("Getting kernel max_sb_size");
218 ret = -errno;
219 goto end_unlock;
220 }
221 }
222
223 /*
224 * The original value is sent back if max stream size is larger than
225 * the possible size of the snapshot. Also, we asume that the session
226 * daemon should never send a maximum stream size that is lower than
227 * subbuffer size.
228 */
229 consumed_pos = consumer_get_consumed_maxsize(consumed_pos,
230 produced_pos, max_stream_size);
231
232 while (consumed_pos < produced_pos) {
233 ssize_t read_len;
234 unsigned long len, padded_len;
235
236 health_code_update();
237
238 DBG("Kernel consumer taking snapshot at pos %lu", consumed_pos);
239
240 ret = kernctl_get_subbuf(stream->wait_fd, &consumed_pos);
241 if (ret < 0) {
242 if (errno != EAGAIN) {
243 PERROR("kernctl_get_subbuf snapshot");
244 ret = -errno;
245 goto end_unlock;
246 }
247 DBG("Kernel consumer get subbuf failed. Skipping it.");
248 consumed_pos += stream->max_sb_size;
249 continue;
250 }
251
252 ret = kernctl_get_subbuf_size(stream->wait_fd, &len);
253 if (ret < 0) {
254 ERR("Snapshot kernctl_get_subbuf_size");
255 ret = -errno;
256 goto error_put_subbuf;
257 }
258
259 ret = kernctl_get_padded_subbuf_size(stream->wait_fd, &padded_len);
260 if (ret < 0) {
261 ERR("Snapshot kernctl_get_padded_subbuf_size");
262 ret = -errno;
263 goto error_put_subbuf;
264 }
265
266 read_len = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, len,
267 padded_len - len, NULL);
268 /*
269 * We write the padded len in local tracefiles but the data len
270 * when using a relay. Display the error but continue processing
271 * to try to release the subbuffer.
272 */
273 if (relayd_id != (uint64_t) -1ULL) {
274 if (read_len != len) {
275 ERR("Error sending to the relay (ret: %zd != len: %lu)",
276 read_len, len);
277 }
278 } else {
279 if (read_len != padded_len) {
280 ERR("Error writing to tracefile (ret: %zd != len: %lu)",
281 read_len, padded_len);
282 }
283 }
284
285 ret = kernctl_put_subbuf(stream->wait_fd);
286 if (ret < 0) {
287 ERR("Snapshot kernctl_put_subbuf");
288 ret = -errno;
289 goto end_unlock;
290 }
291 consumed_pos += stream->max_sb_size;
292 }
293
294 if (relayd_id == (uint64_t) -1ULL) {
295 if (stream->out_fd >= 0) {
296 ret = close(stream->out_fd);
297 if (ret < 0) {
298 PERROR("Kernel consumer snapshot close out_fd");
299 goto end_unlock;
300 }
301 stream->out_fd = -1;
302 }
303 } else {
304 close_relayd_stream(stream);
305 stream->net_seq_idx = (uint64_t) -1ULL;
306 }
307 pthread_mutex_unlock(&stream->lock);
308 }
309
310 /* All good! */
311 ret = 0;
312 goto end;
313
314 error_put_subbuf:
315 ret = kernctl_put_subbuf(stream->wait_fd);
316 if (ret < 0) {
317 ret = -errno;
318 ERR("Snapshot kernctl_put_subbuf error path");
319 }
320 end_unlock:
321 pthread_mutex_unlock(&stream->lock);
322 end:
323 rcu_read_unlock();
324 return ret;
325 }
326
327 /*
328 * Read the whole metadata available for a snapshot.
329 *
330 * Returns 0 on success, < 0 on error
331 */
332 int lttng_kconsumer_snapshot_metadata(uint64_t key, char *path,
333 uint64_t relayd_id, struct lttng_consumer_local_data *ctx)
334 {
335 int ret, use_relayd = 0;
336 ssize_t ret_read;
337 struct lttng_consumer_channel *metadata_channel;
338 struct lttng_consumer_stream *metadata_stream;
339
340 assert(ctx);
341
342 DBG("Kernel consumer snapshot metadata with key %" PRIu64 " at path %s",
343 key, path);
344
345 rcu_read_lock();
346
347 metadata_channel = consumer_find_channel(key);
348 if (!metadata_channel) {
349 ERR("Kernel snapshot metadata not found for key %" PRIu64, key);
350 ret = -1;
351 goto error;
352 }
353
354 metadata_stream = metadata_channel->metadata_stream;
355 assert(metadata_stream);
356
357 /* Flag once that we have a valid relayd for the stream. */
358 if (relayd_id != (uint64_t) -1ULL) {
359 use_relayd = 1;
360 }
361
362 if (use_relayd) {
363 ret = consumer_send_relayd_stream(metadata_stream, path);
364 if (ret < 0) {
365 goto error;
366 }
367 } else {
368 ret = utils_create_stream_file(path, metadata_stream->name,
369 metadata_stream->chan->tracefile_size,
370 metadata_stream->tracefile_count_current,
371 metadata_stream->uid, metadata_stream->gid, NULL);
372 if (ret < 0) {
373 goto error;
374 }
375 metadata_stream->out_fd = ret;
376 }
377
378 do {
379 health_code_update();
380
381 ret_read = lttng_kconsumer_read_subbuffer(metadata_stream, ctx);
382 if (ret_read < 0) {
383 if (ret_read != -EAGAIN) {
384 ERR("Kernel snapshot reading metadata subbuffer (ret: %zd)",
385 ret_read);
386 goto error;
387 }
388 /* ret_read is negative at this point so we will exit the loop. */
389 continue;
390 }
391 } while (ret_read >= 0);
392
393 if (use_relayd) {
394 close_relayd_stream(metadata_stream);
395 metadata_stream->net_seq_idx = (uint64_t) -1ULL;
396 } else {
397 if (metadata_stream->out_fd >= 0) {
398 ret = close(metadata_stream->out_fd);
399 if (ret < 0) {
400 PERROR("Kernel consumer snapshot metadata close out_fd");
401 /*
402 * Don't go on error here since the snapshot was successful at this
403 * point but somehow the close failed.
404 */
405 }
406 metadata_stream->out_fd = -1;
407 }
408 }
409
410 ret = 0;
411
412 cds_list_del(&metadata_stream->send_node);
413 consumer_stream_destroy(metadata_stream, NULL);
414 metadata_channel->metadata_stream = NULL;
415 error:
416 rcu_read_unlock();
417 return ret;
418 }
419
420 /*
421 * Receive command from session daemon and process it.
422 *
423 * Return 1 on success else a negative value or 0.
424 */
425 int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
426 int sock, struct pollfd *consumer_sockpoll)
427 {
428 ssize_t ret;
429 enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS;
430 struct lttcomm_consumer_msg msg;
431
432 health_code_update();
433
434 ret = lttcomm_recv_unix_sock(sock, &msg, sizeof(msg));
435 if (ret != sizeof(msg)) {
436 if (ret > 0) {
437 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD);
438 ret = -1;
439 }
440 return ret;
441 }
442
443 health_code_update();
444
445 /* Deprecated command */
446 assert(msg.cmd_type != LTTNG_CONSUMER_STOP);
447
448 health_code_update();
449
450 /* relayd needs RCU read-side protection */
451 rcu_read_lock();
452
453 switch (msg.cmd_type) {
454 case LTTNG_CONSUMER_ADD_RELAYD_SOCKET:
455 {
456 /* Session daemon status message are handled in the following call. */
457 ret = consumer_add_relayd_socket(msg.u.relayd_sock.net_index,
458 msg.u.relayd_sock.type, ctx, sock, consumer_sockpoll,
459 &msg.u.relayd_sock.sock, msg.u.relayd_sock.session_id,
460 msg.u.relayd_sock.relayd_session_id);
461 goto end_nosignal;
462 }
463 case LTTNG_CONSUMER_ADD_CHANNEL:
464 {
465 struct lttng_consumer_channel *new_channel;
466 int ret_recv;
467
468 health_code_update();
469
470 /* First send a status message before receiving the fds. */
471 ret = consumer_send_status_msg(sock, ret_code);
472 if (ret < 0) {
473 /* Somehow, the session daemon is not responding anymore. */
474 goto error_fatal;
475 }
476
477 health_code_update();
478
479 DBG("consumer_add_channel %" PRIu64, msg.u.channel.channel_key);
480 new_channel = consumer_allocate_channel(msg.u.channel.channel_key,
481 msg.u.channel.session_id, msg.u.channel.pathname,
482 msg.u.channel.name, msg.u.channel.uid, msg.u.channel.gid,
483 msg.u.channel.relayd_id, msg.u.channel.output,
484 msg.u.channel.tracefile_size,
485 msg.u.channel.tracefile_count, 0,
486 msg.u.channel.monitor,
487 msg.u.channel.live_timer_interval);
488 if (new_channel == NULL) {
489 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
490 goto end_nosignal;
491 }
492 new_channel->nb_init_stream_left = msg.u.channel.nb_init_streams;
493 switch (msg.u.channel.output) {
494 case LTTNG_EVENT_SPLICE:
495 new_channel->output = CONSUMER_CHANNEL_SPLICE;
496 break;
497 case LTTNG_EVENT_MMAP:
498 new_channel->output = CONSUMER_CHANNEL_MMAP;
499 break;
500 default:
501 ERR("Channel output unknown %d", msg.u.channel.output);
502 goto end_nosignal;
503 }
504
505 /* Translate and save channel type. */
506 switch (msg.u.channel.type) {
507 case CONSUMER_CHANNEL_TYPE_DATA:
508 case CONSUMER_CHANNEL_TYPE_METADATA:
509 new_channel->type = msg.u.channel.type;
510 break;
511 default:
512 assert(0);
513 goto end_nosignal;
514 };
515
516 health_code_update();
517
518 if (ctx->on_recv_channel != NULL) {
519 ret_recv = ctx->on_recv_channel(new_channel);
520 if (ret_recv == 0) {
521 ret = consumer_add_channel(new_channel, ctx);
522 } else if (ret_recv < 0) {
523 goto end_nosignal;
524 }
525 } else {
526 ret = consumer_add_channel(new_channel, ctx);
527 }
528 if (CONSUMER_CHANNEL_TYPE_DATA) {
529 consumer_timer_live_start(new_channel,
530 msg.u.channel.live_timer_interval);
531 }
532
533 health_code_update();
534
535 /* If we received an error in add_channel, we need to report it. */
536 if (ret < 0) {
537 ret = consumer_send_status_msg(sock, ret);
538 if (ret < 0) {
539 goto error_fatal;
540 }
541 goto end_nosignal;
542 }
543
544 goto end_nosignal;
545 }
546 case LTTNG_CONSUMER_ADD_STREAM:
547 {
548 int fd;
549 struct lttng_pipe *stream_pipe;
550 struct lttng_consumer_stream *new_stream;
551 struct lttng_consumer_channel *channel;
552 int alloc_ret = 0;
553
554 /*
555 * Get stream's channel reference. Needed when adding the stream to the
556 * global hash table.
557 */
558 channel = consumer_find_channel(msg.u.stream.channel_key);
559 if (!channel) {
560 /*
561 * We could not find the channel. Can happen if cpu hotplug
562 * happens while tearing down.
563 */
564 ERR("Unable to find channel key %" PRIu64, msg.u.stream.channel_key);
565 ret_code = LTTNG_ERR_KERN_CHAN_NOT_FOUND;
566 }
567
568 health_code_update();
569
570 /* First send a status message before receiving the fds. */
571 ret = consumer_send_status_msg(sock, ret_code);
572 if (ret < 0) {
573 /* Somehow, the session daemon is not responding anymore. */
574 goto error_fatal;
575 }
576
577 health_code_update();
578
579 if (ret_code != LTTCOMM_CONSUMERD_SUCCESS) {
580 /* Channel was not found. */
581 goto end_nosignal;
582 }
583
584 /* Blocking call */
585 health_poll_entry();
586 ret = lttng_consumer_poll_socket(consumer_sockpoll);
587 health_poll_exit();
588 if (ret) {
589 goto error_fatal;
590 }
591
592 health_code_update();
593
594 /* Get stream file descriptor from socket */
595 ret = lttcomm_recv_fds_unix_sock(sock, &fd, 1);
596 if (ret != sizeof(fd)) {
597 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_FD);
598 rcu_read_unlock();
599 return ret;
600 }
601
602 health_code_update();
603
604 /*
605 * Send status code to session daemon only if the recv works. If the
606 * above recv() failed, the session daemon is notified through the
607 * error socket and the teardown is eventually done.
608 */
609 ret = consumer_send_status_msg(sock, ret_code);
610 if (ret < 0) {
611 /* Somehow, the session daemon is not responding anymore. */
612 goto end_nosignal;
613 }
614
615 health_code_update();
616
617 new_stream = consumer_allocate_stream(channel->key,
618 fd,
619 LTTNG_CONSUMER_ACTIVE_STREAM,
620 channel->name,
621 channel->uid,
622 channel->gid,
623 channel->relayd_id,
624 channel->session_id,
625 msg.u.stream.cpu,
626 &alloc_ret,
627 channel->type,
628 channel->monitor);
629 if (new_stream == NULL) {
630 switch (alloc_ret) {
631 case -ENOMEM:
632 case -EINVAL:
633 default:
634 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
635 break;
636 }
637 goto end_nosignal;
638 }
639
640 new_stream->chan = channel;
641 new_stream->wait_fd = fd;
642 switch (channel->output) {
643 case CONSUMER_CHANNEL_SPLICE:
644 new_stream->output = LTTNG_EVENT_SPLICE;
645 break;
646 case CONSUMER_CHANNEL_MMAP:
647 new_stream->output = LTTNG_EVENT_MMAP;
648 break;
649 default:
650 ERR("Stream output unknown %d", channel->output);
651 goto end_nosignal;
652 }
653
654 /*
655 * We've just assigned the channel to the stream so increment the
656 * refcount right now. We don't need to increment the refcount for
657 * streams in no monitor because we handle manually the cleanup of
658 * those. It is very important to make sure there is NO prior
659 * consumer_del_stream() calls or else the refcount will be unbalanced.
660 */
661 if (channel->monitor) {
662 uatomic_inc(&new_stream->chan->refcount);
663 }
664
665 /*
666 * The buffer flush is done on the session daemon side for the kernel
667 * so no need for the stream "hangup_flush_done" variable to be
668 * tracked. This is important for a kernel stream since we don't rely
669 * on the flush state of the stream to read data. It's not the case for
670 * user space tracing.
671 */
672 new_stream->hangup_flush_done = 0;
673
674 health_code_update();
675
676 if (ctx->on_recv_stream) {
677 ret = ctx->on_recv_stream(new_stream);
678 if (ret < 0) {
679 consumer_stream_free(new_stream);
680 goto end_nosignal;
681 }
682 }
683
684 health_code_update();
685
686 if (new_stream->metadata_flag) {
687 channel->metadata_stream = new_stream;
688 }
689
690 /* Do not monitor this stream. */
691 if (!channel->monitor) {
692 DBG("Kernel consumer add stream %s in no monitor mode with "
693 "relayd id %" PRIu64, new_stream->name,
694 new_stream->net_seq_idx);
695 cds_list_add(&new_stream->send_node, &channel->streams.head);
696 break;
697 }
698
699 /* Send stream to relayd if the stream has an ID. */
700 if (new_stream->net_seq_idx != (uint64_t) -1ULL) {
701 ret = consumer_send_relayd_stream(new_stream,
702 new_stream->chan->pathname);
703 if (ret < 0) {
704 consumer_stream_free(new_stream);
705 goto end_nosignal;
706 }
707 }
708
709 /* Get the right pipe where the stream will be sent. */
710 if (new_stream->metadata_flag) {
711 ret = consumer_add_metadata_stream(new_stream);
712 if (ret) {
713 ERR("Consumer add metadata stream %" PRIu64 " failed. Continuing",
714 new_stream->key);
715 consumer_stream_free(new_stream);
716 goto end_nosignal;
717 }
718 stream_pipe = ctx->consumer_metadata_pipe;
719 } else {
720 ret = consumer_add_data_stream(new_stream);
721 if (ret) {
722 ERR("Consumer add stream %" PRIu64 " failed. Continuing",
723 new_stream->key);
724 consumer_stream_free(new_stream);
725 goto end_nosignal;
726 }
727 stream_pipe = ctx->consumer_data_pipe;
728 }
729
730 /* Vitible to other threads */
731 new_stream->globally_visible = 1;
732
733 health_code_update();
734
735 ret = lttng_pipe_write(stream_pipe, &new_stream, sizeof(new_stream));
736 if (ret < 0) {
737 ERR("Consumer write %s stream to pipe %d",
738 new_stream->metadata_flag ? "metadata" : "data",
739 lttng_pipe_get_writefd(stream_pipe));
740 if (new_stream->metadata_flag) {
741 consumer_del_stream_for_metadata(new_stream);
742 } else {
743 consumer_del_stream_for_data(new_stream);
744 }
745 goto end_nosignal;
746 }
747
748 DBG("Kernel consumer ADD_STREAM %s (fd: %d) with relayd id %" PRIu64,
749 new_stream->name, fd, new_stream->relayd_stream_id);
750 break;
751 }
752 case LTTNG_CONSUMER_STREAMS_SENT:
753 {
754 struct lttng_consumer_channel *channel;
755
756 /*
757 * Get stream's channel reference. Needed when adding the stream to the
758 * global hash table.
759 */
760 channel = consumer_find_channel(msg.u.sent_streams.channel_key);
761 if (!channel) {
762 /*
763 * We could not find the channel. Can happen if cpu hotplug
764 * happens while tearing down.
765 */
766 ERR("Unable to find channel key %" PRIu64,
767 msg.u.sent_streams.channel_key);
768 ret_code = LTTNG_ERR_KERN_CHAN_NOT_FOUND;
769 }
770
771 health_code_update();
772
773 /*
774 * Send status code to session daemon.
775 */
776 ret = consumer_send_status_msg(sock, ret_code);
777 if (ret < 0) {
778 /* Somehow, the session daemon is not responding anymore. */
779 goto end_nosignal;
780 }
781
782 health_code_update();
783
784 /*
785 * We should not send this message if we don't monitor the
786 * streams in this channel.
787 */
788 if (!channel->monitor) {
789 break;
790 }
791
792 health_code_update();
793 /* Send stream to relayd if the stream has an ID. */
794 if (msg.u.sent_streams.net_seq_idx != (uint64_t) -1ULL) {
795 ret = consumer_send_relayd_streams_sent(
796 msg.u.sent_streams.net_seq_idx);
797 if (ret < 0) {
798 goto end_nosignal;
799 }
800 }
801 break;
802 }
803 case LTTNG_CONSUMER_UPDATE_STREAM:
804 {
805 rcu_read_unlock();
806 return -ENOSYS;
807 }
808 case LTTNG_CONSUMER_DESTROY_RELAYD:
809 {
810 uint64_t index = msg.u.destroy_relayd.net_seq_idx;
811 struct consumer_relayd_sock_pair *relayd;
812
813 DBG("Kernel consumer destroying relayd %" PRIu64, index);
814
815 /* Get relayd reference if exists. */
816 relayd = consumer_find_relayd(index);
817 if (relayd == NULL) {
818 DBG("Unable to find relayd %" PRIu64, index);
819 ret_code = LTTNG_ERR_NO_CONSUMER;
820 }
821
822 /*
823 * Each relayd socket pair has a refcount of stream attached to it
824 * which tells if the relayd is still active or not depending on the
825 * refcount value.
826 *
827 * This will set the destroy flag of the relayd object and destroy it
828 * if the refcount reaches zero when called.
829 *
830 * The destroy can happen either here or when a stream fd hangs up.
831 */
832 if (relayd) {
833 consumer_flag_relayd_for_destroy(relayd);
834 }
835
836 health_code_update();
837
838 ret = consumer_send_status_msg(sock, ret_code);
839 if (ret < 0) {
840 /* Somehow, the session daemon is not responding anymore. */
841 goto error_fatal;
842 }
843
844 goto end_nosignal;
845 }
846 case LTTNG_CONSUMER_DATA_PENDING:
847 {
848 int32_t ret;
849 uint64_t id = msg.u.data_pending.session_id;
850
851 DBG("Kernel consumer data pending command for id %" PRIu64, id);
852
853 ret = consumer_data_pending(id);
854
855 health_code_update();
856
857 /* Send back returned value to session daemon */
858 ret = lttcomm_send_unix_sock(sock, &ret, sizeof(ret));
859 if (ret < 0) {
860 PERROR("send data pending ret code");
861 goto error_fatal;
862 }
863
864 /*
865 * No need to send back a status message since the data pending
866 * returned value is the response.
867 */
868 break;
869 }
870 case LTTNG_CONSUMER_SNAPSHOT_CHANNEL:
871 {
872 if (msg.u.snapshot_channel.metadata == 1) {
873 ret = lttng_kconsumer_snapshot_metadata(msg.u.snapshot_channel.key,
874 msg.u.snapshot_channel.pathname,
875 msg.u.snapshot_channel.relayd_id, ctx);
876 if (ret < 0) {
877 ERR("Snapshot metadata failed");
878 ret_code = LTTNG_ERR_KERN_META_FAIL;
879 }
880 } else {
881 ret = lttng_kconsumer_snapshot_channel(msg.u.snapshot_channel.key,
882 msg.u.snapshot_channel.pathname,
883 msg.u.snapshot_channel.relayd_id,
884 msg.u.snapshot_channel.max_stream_size,
885 ctx);
886 if (ret < 0) {
887 ERR("Snapshot channel failed");
888 ret_code = LTTNG_ERR_KERN_CHAN_FAIL;
889 }
890 }
891
892 health_code_update();
893
894 ret = consumer_send_status_msg(sock, ret_code);
895 if (ret < 0) {
896 /* Somehow, the session daemon is not responding anymore. */
897 goto end_nosignal;
898 }
899 break;
900 }
901 case LTTNG_CONSUMER_DESTROY_CHANNEL:
902 {
903 uint64_t key = msg.u.destroy_channel.key;
904 struct lttng_consumer_channel *channel;
905
906 channel = consumer_find_channel(key);
907 if (!channel) {
908 ERR("Kernel consumer destroy channel %" PRIu64 " not found", key);
909 ret_code = LTTNG_ERR_KERN_CHAN_NOT_FOUND;
910 }
911
912 health_code_update();
913
914 ret = consumer_send_status_msg(sock, ret_code);
915 if (ret < 0) {
916 /* Somehow, the session daemon is not responding anymore. */
917 goto end_nosignal;
918 }
919
920 health_code_update();
921
922 /* Stop right now if no channel was found. */
923 if (!channel) {
924 goto end_nosignal;
925 }
926
927 /*
928 * This command should ONLY be issued for channel with streams set in
929 * no monitor mode.
930 */
931 assert(!channel->monitor);
932
933 /*
934 * The refcount should ALWAYS be 0 in the case of a channel in no
935 * monitor mode.
936 */
937 assert(!uatomic_sub_return(&channel->refcount, 1));
938
939 consumer_del_channel(channel);
940
941 goto end_nosignal;
942 }
943 default:
944 goto end_nosignal;
945 }
946
947 end_nosignal:
948 rcu_read_unlock();
949
950 /*
951 * Return 1 to indicate success since the 0 value can be a socket
952 * shutdown during the recv() or send() call.
953 */
954 health_code_update();
955 return 1;
956
957 error_fatal:
958 rcu_read_unlock();
959 /* This will issue a consumer stop. */
960 return -1;
961 }
962
963 /*
964 * Populate index values of a kernel stream. Values are set in big endian order.
965 *
966 * Return 0 on success or else a negative value.
967 */
968 static int get_index_values(struct ctf_packet_index *index, int infd)
969 {
970 int ret;
971
972 ret = kernctl_get_timestamp_begin(infd, &index->timestamp_begin);
973 if (ret < 0) {
974 PERROR("kernctl_get_timestamp_begin");
975 goto error;
976 }
977 index->timestamp_begin = htobe64(index->timestamp_begin);
978
979 ret = kernctl_get_timestamp_end(infd, &index->timestamp_end);
980 if (ret < 0) {
981 PERROR("kernctl_get_timestamp_end");
982 goto error;
983 }
984 index->timestamp_end = htobe64(index->timestamp_end);
985
986 ret = kernctl_get_events_discarded(infd, &index->events_discarded);
987 if (ret < 0) {
988 PERROR("kernctl_get_events_discarded");
989 goto error;
990 }
991 index->events_discarded = htobe64(index->events_discarded);
992
993 ret = kernctl_get_content_size(infd, &index->content_size);
994 if (ret < 0) {
995 PERROR("kernctl_get_content_size");
996 goto error;
997 }
998 index->content_size = htobe64(index->content_size);
999
1000 ret = kernctl_get_packet_size(infd, &index->packet_size);
1001 if (ret < 0) {
1002 PERROR("kernctl_get_packet_size");
1003 goto error;
1004 }
1005 index->packet_size = htobe64(index->packet_size);
1006
1007 ret = kernctl_get_stream_id(infd, &index->stream_id);
1008 if (ret < 0) {
1009 PERROR("kernctl_get_stream_id");
1010 goto error;
1011 }
1012 index->stream_id = htobe64(index->stream_id);
1013
1014 error:
1015 return ret;
1016 }
1017 /*
1018 * Sync metadata meaning request them to the session daemon and snapshot to the
1019 * metadata thread can consumer them.
1020 *
1021 * Metadata stream lock MUST be acquired.
1022 *
1023 * Return 0 if new metadatda is available, EAGAIN if the metadata stream
1024 * is empty or a negative value on error.
1025 */
1026 int lttng_kconsumer_sync_metadata(struct lttng_consumer_stream *metadata)
1027 {
1028 int ret;
1029
1030 assert(metadata);
1031
1032 ret = kernctl_buffer_flush(metadata->wait_fd);
1033 if (ret < 0) {
1034 ERR("Failed to flush kernel stream");
1035 goto end;
1036 }
1037
1038 ret = kernctl_snapshot(metadata->wait_fd);
1039 if (ret < 0) {
1040 if (errno != EAGAIN) {
1041 ERR("Sync metadata, taking kernel snapshot failed.");
1042 goto end;
1043 }
1044 DBG("Sync metadata, no new kernel metadata");
1045 /* No new metadata, exit. */
1046 ret = ENODATA;
1047 goto end;
1048 }
1049
1050 end:
1051 return ret;
1052 }
1053
1054 /*
1055 * Consume data on a file descriptor and write it on a trace file.
1056 */
1057 ssize_t lttng_kconsumer_read_subbuffer(struct lttng_consumer_stream *stream,
1058 struct lttng_consumer_local_data *ctx)
1059 {
1060 unsigned long len, subbuf_size, padding;
1061 int err, write_index = 1;
1062 ssize_t ret = 0;
1063 int infd = stream->wait_fd;
1064 struct ctf_packet_index index;
1065
1066 DBG("In read_subbuffer (infd : %d)", infd);
1067
1068 /* Get the next subbuffer */
1069 err = kernctl_get_next_subbuf(infd);
1070 if (err != 0) {
1071 /*
1072 * This is a debug message even for single-threaded consumer,
1073 * because poll() have more relaxed criterions than get subbuf,
1074 * so get_subbuf may fail for short race windows where poll()
1075 * would issue wakeups.
1076 */
1077 DBG("Reserving sub buffer failed (everything is normal, "
1078 "it is due to concurrency)");
1079 ret = -errno;
1080 goto end;
1081 }
1082
1083 /* Get the full subbuffer size including padding */
1084 err = kernctl_get_padded_subbuf_size(infd, &len);
1085 if (err != 0) {
1086 perror("Getting sub-buffer len failed.");
1087 ret = -errno;
1088 goto end;
1089 }
1090
1091 if (!stream->metadata_flag) {
1092 ret = get_index_values(&index, infd);
1093 if (ret < 0) {
1094 goto end;
1095 }
1096 } else {
1097 write_index = 0;
1098 }
1099
1100 switch (stream->chan->output) {
1101 case CONSUMER_CHANNEL_SPLICE:
1102 /*
1103 * XXX: The lttng-modules splice "actor" does not handle copying
1104 * partial pages hence only using the subbuffer size without the
1105 * padding makes the splice fail.
1106 */
1107 subbuf_size = len;
1108 padding = 0;
1109
1110 /* splice the subbuffer to the tracefile */
1111 ret = lttng_consumer_on_read_subbuffer_splice(ctx, stream, subbuf_size,
1112 padding, &index);
1113 /*
1114 * XXX: Splice does not support network streaming so the return value
1115 * is simply checked against subbuf_size and not like the mmap() op.
1116 */
1117 if (ret != subbuf_size) {
1118 /*
1119 * display the error but continue processing to try
1120 * to release the subbuffer
1121 */
1122 ERR("Error splicing to tracefile (ret: %zd != len: %lu)",
1123 ret, subbuf_size);
1124 write_index = 0;
1125 }
1126 break;
1127 case CONSUMER_CHANNEL_MMAP:
1128 /* Get subbuffer size without padding */
1129 err = kernctl_get_subbuf_size(infd, &subbuf_size);
1130 if (err != 0) {
1131 perror("Getting sub-buffer len failed.");
1132 ret = -errno;
1133 goto end;
1134 }
1135
1136 /* Make sure the tracer is not gone mad on us! */
1137 assert(len >= subbuf_size);
1138
1139 padding = len - subbuf_size;
1140
1141 /* write the subbuffer to the tracefile */
1142 ret = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, subbuf_size,
1143 padding, &index);
1144 /*
1145 * The mmap operation should write subbuf_size amount of data when
1146 * network streaming or the full padding (len) size when we are _not_
1147 * streaming.
1148 */
1149 if ((ret != subbuf_size && stream->net_seq_idx != (uint64_t) -1ULL) ||
1150 (ret != len && stream->net_seq_idx == (uint64_t) -1ULL)) {
1151 /*
1152 * Display the error but continue processing to try to release the
1153 * subbuffer. This is a DBG statement since this is possible to
1154 * happen without being a critical error.
1155 */
1156 DBG("Error writing to tracefile "
1157 "(ret: %zd != len: %lu != subbuf_size: %lu)",
1158 ret, len, subbuf_size);
1159 write_index = 0;
1160 }
1161 break;
1162 default:
1163 ERR("Unknown output method");
1164 ret = -EPERM;
1165 }
1166
1167 err = kernctl_put_next_subbuf(infd);
1168 if (err != 0) {
1169 if (errno == EFAULT) {
1170 perror("Error in unreserving sub buffer\n");
1171 } else if (errno == EIO) {
1172 /* Should never happen with newer LTTng versions */
1173 perror("Reader has been pushed by the writer, last sub-buffer corrupted.");
1174 }
1175 ret = -errno;
1176 goto end;
1177 }
1178
1179 /* Write index if needed. */
1180 if (!write_index) {
1181 goto end;
1182 }
1183
1184 if (stream->chan->live_timer_interval && !stream->metadata_flag) {
1185 /*
1186 * In live, block until all the metadata is sent.
1187 */
1188 err = consumer_stream_sync_metadata(ctx, stream->session_id);
1189 if (err < 0) {
1190 goto end;
1191 }
1192 }
1193
1194 err = consumer_stream_write_index(stream, &index);
1195 if (err < 0) {
1196 goto end;
1197 }
1198
1199 end:
1200 return ret;
1201 }
1202
1203 int lttng_kconsumer_on_recv_stream(struct lttng_consumer_stream *stream)
1204 {
1205 int ret;
1206
1207 assert(stream);
1208
1209 /*
1210 * Don't create anything if this is set for streaming or should not be
1211 * monitored.
1212 */
1213 if (stream->net_seq_idx == (uint64_t) -1ULL && stream->chan->monitor) {
1214 ret = utils_create_stream_file(stream->chan->pathname, stream->name,
1215 stream->chan->tracefile_size, stream->tracefile_count_current,
1216 stream->uid, stream->gid, NULL);
1217 if (ret < 0) {
1218 goto error;
1219 }
1220 stream->out_fd = ret;
1221 stream->tracefile_size_current = 0;
1222
1223 if (!stream->metadata_flag) {
1224 ret = index_create_file(stream->chan->pathname,
1225 stream->name, stream->uid, stream->gid,
1226 stream->chan->tracefile_size,
1227 stream->tracefile_count_current);
1228 if (ret < 0) {
1229 goto error;
1230 }
1231 stream->index_fd = ret;
1232 }
1233 }
1234
1235 if (stream->output == LTTNG_EVENT_MMAP) {
1236 /* get the len of the mmap region */
1237 unsigned long mmap_len;
1238
1239 ret = kernctl_get_mmap_len(stream->wait_fd, &mmap_len);
1240 if (ret != 0) {
1241 PERROR("kernctl_get_mmap_len");
1242 ret = -errno;
1243 goto error_close_fd;
1244 }
1245 stream->mmap_len = (size_t) mmap_len;
1246
1247 stream->mmap_base = mmap(NULL, stream->mmap_len, PROT_READ,
1248 MAP_PRIVATE, stream->wait_fd, 0);
1249 if (stream->mmap_base == MAP_FAILED) {
1250 PERROR("Error mmaping");
1251 ret = -1;
1252 goto error_close_fd;
1253 }
1254 }
1255
1256 /* we return 0 to let the library handle the FD internally */
1257 return 0;
1258
1259 error_close_fd:
1260 if (stream->out_fd >= 0) {
1261 int err;
1262
1263 err = close(stream->out_fd);
1264 assert(!err);
1265 stream->out_fd = -1;
1266 }
1267 error:
1268 return ret;
1269 }
1270
1271 /*
1272 * Check if data is still being extracted from the buffers for a specific
1273 * stream. Consumer data lock MUST be acquired before calling this function
1274 * and the stream lock.
1275 *
1276 * Return 1 if the traced data are still getting read else 0 meaning that the
1277 * data is available for trace viewer reading.
1278 */
1279 int lttng_kconsumer_data_pending(struct lttng_consumer_stream *stream)
1280 {
1281 int ret;
1282
1283 assert(stream);
1284
1285 if (stream->endpoint_status != CONSUMER_ENDPOINT_ACTIVE) {
1286 ret = 0;
1287 goto end;
1288 }
1289
1290 ret = kernctl_get_next_subbuf(stream->wait_fd);
1291 if (ret == 0) {
1292 /* There is still data so let's put back this subbuffer. */
1293 ret = kernctl_put_subbuf(stream->wait_fd);
1294 assert(ret == 0);
1295 ret = 1; /* Data is pending */
1296 goto end;
1297 }
1298
1299 /* Data is NOT pending and ready to be read. */
1300 ret = 0;
1301
1302 end:
1303 return ret;
1304 }
This page took 0.05794 seconds and 6 git commands to generate.