wip rotate consumer pipe
[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 * Copyright (C) 2017 - Jérémie Galarneau <jeremie.galarneau@efficios.com>
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License, version 2 only,
8 * as published by the Free Software Foundation.
9 *
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
14 *
15 * You should have received a copy of the GNU General Public License along
16 * with this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
18 */
19
20 #define _LGPL_SOURCE
21 #include <assert.h>
22 #include <poll.h>
23 #include <pthread.h>
24 #include <stdlib.h>
25 #include <string.h>
26 #include <sys/mman.h>
27 #include <sys/socket.h>
28 #include <sys/types.h>
29 #include <inttypes.h>
30 #include <unistd.h>
31 #include <sys/stat.h>
32
33 #include <bin/lttng-consumerd/health-consumerd.h>
34 #include <common/common.h>
35 #include <common/kernel-ctl/kernel-ctl.h>
36 #include <common/sessiond-comm/sessiond-comm.h>
37 #include <common/sessiond-comm/relayd.h>
38 #include <common/compat/fcntl.h>
39 #include <common/compat/endian.h>
40 #include <common/pipe.h>
41 #include <common/relayd/relayd.h>
42 #include <common/utils.h>
43 #include <common/consumer/consumer-stream.h>
44 #include <common/index/index.h>
45 #include <common/consumer/consumer-timer.h>
46
47 #include "kernel-consumer.h"
48
49 extern struct lttng_consumer_global_data consumer_data;
50 extern int consumer_poll_timeout;
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 /*
64 * -EAGAIN is not an error, it just means that there is no data to
65 * be read.
66 */
67 if (ret != 0 && ret != -EAGAIN) {
68 PERROR("Getting sub-buffer snapshot.");
69 }
70
71 return ret;
72 }
73
74 /*
75 * Sample consumed and produced positions for a specific fd.
76 *
77 * Returns 0 on success, < 0 on error.
78 */
79 int lttng_kconsumer_sample_snapshot_positions(
80 struct lttng_consumer_stream *stream)
81 {
82 assert(stream);
83
84 return kernctl_snapshot_sample_positions(stream->wait_fd);
85 }
86
87 /*
88 * Get the produced position
89 *
90 * Returns 0 on success, < 0 on error
91 */
92 int lttng_kconsumer_get_produced_snapshot(struct lttng_consumer_stream *stream,
93 unsigned long *pos)
94 {
95 int ret;
96 int infd = stream->wait_fd;
97
98 ret = kernctl_snapshot_get_produced(infd, pos);
99 if (ret != 0) {
100 PERROR("kernctl_snapshot_get_produced");
101 }
102
103 return ret;
104 }
105
106 /*
107 * Get the consumerd position
108 *
109 * Returns 0 on success, < 0 on error
110 */
111 int lttng_kconsumer_get_consumed_snapshot(struct lttng_consumer_stream *stream,
112 unsigned long *pos)
113 {
114 int ret;
115 int infd = stream->wait_fd;
116
117 ret = kernctl_snapshot_get_consumed(infd, pos);
118 if (ret != 0) {
119 PERROR("kernctl_snapshot_get_consumed");
120 }
121
122 return ret;
123 }
124
125 /*
126 * Take a snapshot of all the stream of a channel
127 *
128 * Returns 0 on success, < 0 on error
129 */
130 int lttng_kconsumer_snapshot_channel(uint64_t key, char *path,
131 uint64_t relayd_id, uint64_t nb_packets_per_stream,
132 struct lttng_consumer_local_data *ctx)
133 {
134 int ret;
135 struct lttng_consumer_channel *channel;
136 struct lttng_consumer_stream *stream;
137
138 DBG("Kernel consumer snapshot channel %" PRIu64, key);
139
140 rcu_read_lock();
141
142 channel = consumer_find_channel(key);
143 if (!channel) {
144 ERR("No channel found for key %" PRIu64, key);
145 ret = -1;
146 goto end;
147 }
148
149 /* Splice is not supported yet for channel snapshot. */
150 if (channel->output != CONSUMER_CHANNEL_MMAP) {
151 ERR("Unsupported output %d", channel->output);
152 ret = -1;
153 goto end;
154 }
155
156 cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
157 unsigned long consumed_pos, produced_pos;
158
159 health_code_update();
160
161 /*
162 * Lock stream because we are about to change its state.
163 */
164 pthread_mutex_lock(&stream->lock);
165
166 /*
167 * Assign the received relayd ID so we can use it for streaming. The streams
168 * are not visible to anyone so this is OK to change it.
169 */
170 stream->net_seq_idx = relayd_id;
171 channel->relayd_id = relayd_id;
172 if (relayd_id != (uint64_t) -1ULL) {
173 ret = consumer_send_relayd_stream(stream, path,
174 LTTNG_DOMAIN_KERNEL);
175 if (ret < 0) {
176 ERR("sending stream to relayd");
177 goto end_unlock;
178 }
179 } else {
180 ret = utils_create_stream_file(path, stream->name,
181 stream->chan->tracefile_size,
182 stream->tracefile_count_current,
183 stream->uid, stream->gid, NULL);
184 if (ret < 0) {
185 ERR("utils_create_stream_file");
186 goto end_unlock;
187 }
188
189 stream->out_fd = ret;
190 stream->tracefile_size_current = 0;
191
192 DBG("Kernel consumer snapshot stream %s/%s (%" PRIu64 ")",
193 path, stream->name, stream->key);
194 }
195
196 ret = kernctl_buffer_flush_empty(stream->wait_fd);
197 if (ret < 0) {
198 /*
199 * Doing a buffer flush which does not take into
200 * account empty packets. This is not perfect
201 * for stream intersection, but required as a
202 * fall-back when "flush_empty" is not
203 * implemented by lttng-modules.
204 */
205 ret = kernctl_buffer_flush(stream->wait_fd);
206 if (ret < 0) {
207 ERR("Failed to flush kernel stream");
208 goto end_unlock;
209 }
210 goto end_unlock;
211 }
212
213 ret = lttng_kconsumer_take_snapshot(stream);
214 if (ret < 0) {
215 ERR("Taking kernel snapshot");
216 goto end_unlock;
217 }
218
219 ret = lttng_kconsumer_get_produced_snapshot(stream, &produced_pos);
220 if (ret < 0) {
221 ERR("Produced kernel snapshot position");
222 goto end_unlock;
223 }
224
225 ret = lttng_kconsumer_get_consumed_snapshot(stream, &consumed_pos);
226 if (ret < 0) {
227 ERR("Consumerd kernel snapshot position");
228 goto end_unlock;
229 }
230
231 if (stream->max_sb_size == 0) {
232 ret = kernctl_get_max_subbuf_size(stream->wait_fd,
233 &stream->max_sb_size);
234 if (ret < 0) {
235 ERR("Getting kernel max_sb_size");
236 goto end_unlock;
237 }
238 }
239
240 consumed_pos = consumer_get_consume_start_pos(consumed_pos,
241 produced_pos, nb_packets_per_stream,
242 stream->max_sb_size);
243
244 while (consumed_pos < produced_pos) {
245 ssize_t read_len;
246 unsigned long len, padded_len;
247
248 health_code_update();
249
250 DBG("Kernel consumer taking snapshot at pos %lu", consumed_pos);
251
252 ret = kernctl_get_subbuf(stream->wait_fd, &consumed_pos);
253 if (ret < 0) {
254 if (ret != -EAGAIN) {
255 PERROR("kernctl_get_subbuf snapshot");
256 goto end_unlock;
257 }
258 DBG("Kernel consumer get subbuf failed. Skipping it.");
259 consumed_pos += stream->max_sb_size;
260 stream->chan->lost_packets++;
261 continue;
262 }
263
264 ret = kernctl_get_subbuf_size(stream->wait_fd, &len);
265 if (ret < 0) {
266 ERR("Snapshot kernctl_get_subbuf_size");
267 goto error_put_subbuf;
268 }
269
270 ret = kernctl_get_padded_subbuf_size(stream->wait_fd, &padded_len);
271 if (ret < 0) {
272 ERR("Snapshot kernctl_get_padded_subbuf_size");
273 goto error_put_subbuf;
274 }
275
276 read_len = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, len,
277 padded_len - len, NULL);
278 /*
279 * We write the padded len in local tracefiles but the data len
280 * when using a relay. Display the error but continue processing
281 * to try to release the subbuffer.
282 */
283 if (relayd_id != (uint64_t) -1ULL) {
284 if (read_len != len) {
285 ERR("Error sending to the relay (ret: %zd != len: %lu)",
286 read_len, len);
287 }
288 } else {
289 if (read_len != padded_len) {
290 ERR("Error writing to tracefile (ret: %zd != len: %lu)",
291 read_len, padded_len);
292 }
293 }
294
295 ret = kernctl_put_subbuf(stream->wait_fd);
296 if (ret < 0) {
297 ERR("Snapshot kernctl_put_subbuf");
298 goto end_unlock;
299 }
300 consumed_pos += stream->max_sb_size;
301 }
302
303 if (relayd_id == (uint64_t) -1ULL) {
304 if (stream->out_fd >= 0) {
305 ret = close(stream->out_fd);
306 if (ret < 0) {
307 PERROR("Kernel consumer snapshot close out_fd");
308 goto end_unlock;
309 }
310 stream->out_fd = -1;
311 }
312 } else {
313 close_relayd_stream(stream);
314 stream->net_seq_idx = (uint64_t) -1ULL;
315 }
316 pthread_mutex_unlock(&stream->lock);
317 }
318
319 /* All good! */
320 ret = 0;
321 goto end;
322
323 error_put_subbuf:
324 ret = kernctl_put_subbuf(stream->wait_fd);
325 if (ret < 0) {
326 ERR("Snapshot kernctl_put_subbuf error path");
327 }
328 end_unlock:
329 pthread_mutex_unlock(&stream->lock);
330 end:
331 rcu_read_unlock();
332 return ret;
333 }
334
335 /*
336 * Read the whole metadata available for a snapshot.
337 *
338 * Returns 0 on success, < 0 on error
339 */
340 int lttng_kconsumer_snapshot_metadata(uint64_t key, char *path,
341 uint64_t relayd_id, struct lttng_consumer_local_data *ctx)
342 {
343 int ret, use_relayd = 0;
344 ssize_t ret_read;
345 struct lttng_consumer_channel *metadata_channel;
346 struct lttng_consumer_stream *metadata_stream;
347
348 assert(ctx);
349
350 DBG("Kernel consumer snapshot metadata with key %" PRIu64 " at path %s",
351 key, path);
352
353 rcu_read_lock();
354
355 metadata_channel = consumer_find_channel(key);
356 if (!metadata_channel) {
357 ERR("Kernel snapshot metadata not found for key %" PRIu64, key);
358 ret = -1;
359 goto error;
360 }
361
362 metadata_stream = metadata_channel->metadata_stream;
363 assert(metadata_stream);
364
365 /* Flag once that we have a valid relayd for the stream. */
366 if (relayd_id != (uint64_t) -1ULL) {
367 use_relayd = 1;
368 }
369
370 if (use_relayd) {
371 ret = consumer_send_relayd_stream(metadata_stream, path,
372 LTTNG_DOMAIN_KERNEL);
373 if (ret < 0) {
374 goto error;
375 }
376 } else {
377 ret = utils_create_stream_file(path, metadata_stream->name,
378 metadata_stream->chan->tracefile_size,
379 metadata_stream->tracefile_count_current,
380 metadata_stream->uid, metadata_stream->gid, NULL);
381 if (ret < 0) {
382 goto error;
383 }
384 metadata_stream->out_fd = ret;
385 }
386
387 do {
388 health_code_update();
389
390 ret_read = lttng_kconsumer_read_subbuffer(metadata_stream, ctx);
391 if (ret_read < 0) {
392 if (ret_read != -EAGAIN) {
393 ERR("Kernel snapshot reading metadata subbuffer (ret: %zd)",
394 ret_read);
395 goto error;
396 }
397 /* ret_read is negative at this point so we will exit the loop. */
398 continue;
399 }
400 } while (ret_read >= 0);
401
402 if (use_relayd) {
403 close_relayd_stream(metadata_stream);
404 metadata_stream->net_seq_idx = (uint64_t) -1ULL;
405 } else {
406 if (metadata_stream->out_fd >= 0) {
407 ret = close(metadata_stream->out_fd);
408 if (ret < 0) {
409 PERROR("Kernel consumer snapshot metadata close out_fd");
410 /*
411 * Don't go on error here since the snapshot was successful at this
412 * point but somehow the close failed.
413 */
414 }
415 metadata_stream->out_fd = -1;
416 }
417 }
418
419 ret = 0;
420
421 cds_list_del(&metadata_stream->send_node);
422 consumer_stream_destroy(metadata_stream, NULL);
423 metadata_channel->metadata_stream = NULL;
424 error:
425 rcu_read_unlock();
426 return ret;
427 }
428
429 /*
430 * Receive command from session daemon and process it.
431 *
432 * Return 1 on success else a negative value or 0.
433 */
434 int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
435 int sock, struct pollfd *consumer_sockpoll)
436 {
437 ssize_t ret;
438 enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS;
439 struct lttcomm_consumer_msg msg;
440
441 health_code_update();
442
443 ret = lttcomm_recv_unix_sock(sock, &msg, sizeof(msg));
444 if (ret != sizeof(msg)) {
445 if (ret > 0) {
446 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD);
447 ret = -1;
448 }
449 return ret;
450 }
451
452 health_code_update();
453
454 /* Deprecated command */
455 assert(msg.cmd_type != LTTNG_CONSUMER_STOP);
456
457 health_code_update();
458
459 /* relayd needs RCU read-side protection */
460 rcu_read_lock();
461
462 switch (msg.cmd_type) {
463 case LTTNG_CONSUMER_ADD_RELAYD_SOCKET:
464 {
465 /* Session daemon status message are handled in the following call. */
466 consumer_add_relayd_socket(msg.u.relayd_sock.net_index,
467 msg.u.relayd_sock.type, ctx, sock, consumer_sockpoll,
468 &msg.u.relayd_sock.sock, msg.u.relayd_sock.session_id,
469 msg.u.relayd_sock.relayd_session_id);
470 goto end_nosignal;
471 }
472 case LTTNG_CONSUMER_ADD_CHANNEL:
473 {
474 struct lttng_consumer_channel *new_channel;
475 int ret_recv;
476
477 health_code_update();
478
479 /* First send a status message before receiving the fds. */
480 ret = consumer_send_status_msg(sock, ret_code);
481 if (ret < 0) {
482 /* Somehow, the session daemon is not responding anymore. */
483 goto error_fatal;
484 }
485
486 health_code_update();
487
488 DBG("consumer_add_channel %" PRIu64, msg.u.channel.channel_key);
489 new_channel = consumer_allocate_channel(msg.u.channel.channel_key,
490 msg.u.channel.session_id, msg.u.channel.pathname,
491 msg.u.channel.name, msg.u.channel.uid, msg.u.channel.gid,
492 msg.u.channel.relayd_id, msg.u.channel.output,
493 msg.u.channel.tracefile_size,
494 msg.u.channel.tracefile_count, 0,
495 msg.u.channel.monitor,
496 msg.u.channel.live_timer_interval,
497 NULL, NULL);
498 if (new_channel == NULL) {
499 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
500 goto end_nosignal;
501 }
502 new_channel->nb_init_stream_left = msg.u.channel.nb_init_streams;
503 switch (msg.u.channel.output) {
504 case LTTNG_EVENT_SPLICE:
505 new_channel->output = CONSUMER_CHANNEL_SPLICE;
506 break;
507 case LTTNG_EVENT_MMAP:
508 new_channel->output = CONSUMER_CHANNEL_MMAP;
509 break;
510 default:
511 ERR("Channel output unknown %d", msg.u.channel.output);
512 goto end_nosignal;
513 }
514
515 /* Translate and save channel type. */
516 switch (msg.u.channel.type) {
517 case CONSUMER_CHANNEL_TYPE_DATA:
518 case CONSUMER_CHANNEL_TYPE_METADATA:
519 new_channel->type = msg.u.channel.type;
520 break;
521 default:
522 assert(0);
523 goto end_nosignal;
524 };
525
526 health_code_update();
527
528 if (ctx->on_recv_channel != NULL) {
529 ret_recv = ctx->on_recv_channel(new_channel);
530 if (ret_recv == 0) {
531 ret = consumer_add_channel(new_channel, ctx);
532 } else if (ret_recv < 0) {
533 goto end_nosignal;
534 }
535 } else {
536 ret = consumer_add_channel(new_channel, ctx);
537 }
538 if (msg.u.channel.type == CONSUMER_CHANNEL_TYPE_DATA && !ret) {
539 int monitor_start_ret;
540
541 DBG("Consumer starting monitor timer");
542 consumer_timer_live_start(new_channel,
543 msg.u.channel.live_timer_interval);
544 monitor_start_ret = consumer_timer_monitor_start(
545 new_channel,
546 msg.u.channel.monitor_timer_interval);
547 if (monitor_start_ret < 0) {
548 ERR("Starting channel monitoring timer failed");
549 goto end_nosignal;
550 }
551
552 }
553
554 health_code_update();
555
556 /* If we received an error in add_channel, we need to report it. */
557 if (ret < 0) {
558 ret = consumer_send_status_msg(sock, ret);
559 if (ret < 0) {
560 goto error_fatal;
561 }
562 goto end_nosignal;
563 }
564
565 goto end_nosignal;
566 }
567 case LTTNG_CONSUMER_ADD_STREAM:
568 {
569 int fd;
570 struct lttng_pipe *stream_pipe;
571 struct lttng_consumer_stream *new_stream;
572 struct lttng_consumer_channel *channel;
573 int alloc_ret = 0;
574
575 /*
576 * Get stream's channel reference. Needed when adding the stream to the
577 * global hash table.
578 */
579 channel = consumer_find_channel(msg.u.stream.channel_key);
580 if (!channel) {
581 /*
582 * We could not find the channel. Can happen if cpu hotplug
583 * happens while tearing down.
584 */
585 ERR("Unable to find channel key %" PRIu64, msg.u.stream.channel_key);
586 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
587 }
588
589 health_code_update();
590
591 /* First send a status message before receiving the fds. */
592 ret = consumer_send_status_msg(sock, ret_code);
593 if (ret < 0) {
594 /* Somehow, the session daemon is not responding anymore. */
595 goto error_fatal;
596 }
597
598 health_code_update();
599
600 if (ret_code != LTTCOMM_CONSUMERD_SUCCESS) {
601 /* Channel was not found. */
602 goto end_nosignal;
603 }
604
605 /* Blocking call */
606 health_poll_entry();
607 ret = lttng_consumer_poll_socket(consumer_sockpoll);
608 health_poll_exit();
609 if (ret) {
610 goto error_fatal;
611 }
612
613 health_code_update();
614
615 /* Get stream file descriptor from socket */
616 ret = lttcomm_recv_fds_unix_sock(sock, &fd, 1);
617 if (ret != sizeof(fd)) {
618 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_FD);
619 rcu_read_unlock();
620 return ret;
621 }
622
623 health_code_update();
624
625 /*
626 * Send status code to session daemon only if the recv works. If the
627 * above recv() failed, the session daemon is notified through the
628 * error socket and the teardown is eventually done.
629 */
630 ret = consumer_send_status_msg(sock, ret_code);
631 if (ret < 0) {
632 /* Somehow, the session daemon is not responding anymore. */
633 goto end_nosignal;
634 }
635
636 health_code_update();
637
638 new_stream = consumer_allocate_stream(channel->key,
639 fd,
640 LTTNG_CONSUMER_ACTIVE_STREAM,
641 channel->name,
642 channel->uid,
643 channel->gid,
644 channel->relayd_id,
645 channel->session_id,
646 msg.u.stream.cpu,
647 &alloc_ret,
648 channel->type,
649 channel->monitor);
650 if (new_stream == NULL) {
651 switch (alloc_ret) {
652 case -ENOMEM:
653 case -EINVAL:
654 default:
655 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
656 break;
657 }
658 goto end_nosignal;
659 }
660
661 new_stream->chan = channel;
662 new_stream->wait_fd = fd;
663 consumer_stream_copy_ro_channel_values(new_stream, channel);
664 switch (channel->output) {
665 case CONSUMER_CHANNEL_SPLICE:
666 new_stream->output = LTTNG_EVENT_SPLICE;
667 ret = utils_create_pipe(new_stream->splice_pipe);
668 if (ret < 0) {
669 goto end_nosignal;
670 }
671 break;
672 case CONSUMER_CHANNEL_MMAP:
673 new_stream->output = LTTNG_EVENT_MMAP;
674 break;
675 default:
676 ERR("Stream output unknown %d", channel->output);
677 goto end_nosignal;
678 }
679
680 /*
681 * We've just assigned the channel to the stream so increment the
682 * refcount right now. We don't need to increment the refcount for
683 * streams in no monitor because we handle manually the cleanup of
684 * those. It is very important to make sure there is NO prior
685 * consumer_del_stream() calls or else the refcount will be unbalanced.
686 */
687 if (channel->monitor) {
688 uatomic_inc(&new_stream->chan->refcount);
689 }
690
691 /*
692 * The buffer flush is done on the session daemon side for the kernel
693 * so no need for the stream "hangup_flush_done" variable to be
694 * tracked. This is important for a kernel stream since we don't rely
695 * on the flush state of the stream to read data. It's not the case for
696 * user space tracing.
697 */
698 new_stream->hangup_flush_done = 0;
699
700 health_code_update();
701
702 if (ctx->on_recv_stream) {
703 ret = ctx->on_recv_stream(new_stream);
704 if (ret < 0) {
705 consumer_stream_free(new_stream);
706 goto end_nosignal;
707 }
708 }
709
710 health_code_update();
711
712 if (new_stream->metadata_flag) {
713 channel->metadata_stream = new_stream;
714 }
715
716 /* Do not monitor this stream. */
717 if (!channel->monitor) {
718 DBG("Kernel consumer add stream %s in no monitor mode with "
719 "relayd id %" PRIu64, new_stream->name,
720 new_stream->net_seq_idx);
721 cds_list_add(&new_stream->send_node, &channel->streams.head);
722 break;
723 }
724
725 /* Send stream to relayd if the stream has an ID. */
726 if (new_stream->net_seq_idx != (uint64_t) -1ULL) {
727 ret = consumer_send_relayd_stream(new_stream,
728 new_stream->chan->pathname,
729 LTTNG_DOMAIN_KERNEL);
730 if (ret < 0) {
731 consumer_stream_free(new_stream);
732 goto end_nosignal;
733 }
734
735 /*
736 * If adding an extra stream to an already
737 * existing channel (e.g. cpu hotplug), we need
738 * to send the "streams_sent" command to relayd.
739 */
740 if (channel->streams_sent_to_relayd) {
741 ret = consumer_send_relayd_streams_sent(
742 new_stream->net_seq_idx);
743 if (ret < 0) {
744 goto end_nosignal;
745 }
746 }
747 }
748
749 /* Get the right pipe where the stream will be sent. */
750 if (new_stream->metadata_flag) {
751 ret = consumer_add_metadata_stream(new_stream);
752 if (ret) {
753 ERR("Consumer add metadata stream %" PRIu64 " failed. Continuing",
754 new_stream->key);
755 consumer_stream_free(new_stream);
756 goto end_nosignal;
757 }
758 stream_pipe = ctx->consumer_metadata_pipe;
759 } else {
760 ret = consumer_add_data_stream(new_stream);
761 if (ret) {
762 ERR("Consumer add stream %" PRIu64 " failed. Continuing",
763 new_stream->key);
764 consumer_stream_free(new_stream);
765 goto end_nosignal;
766 }
767 stream_pipe = ctx->consumer_data_pipe;
768 }
769
770 /* Vitible to other threads */
771 new_stream->globally_visible = 1;
772
773 health_code_update();
774
775 ret = lttng_pipe_write(stream_pipe, &new_stream, sizeof(new_stream));
776 if (ret < 0) {
777 ERR("Consumer write %s stream to pipe %d",
778 new_stream->metadata_flag ? "metadata" : "data",
779 lttng_pipe_get_writefd(stream_pipe));
780 if (new_stream->metadata_flag) {
781 consumer_del_stream_for_metadata(new_stream);
782 } else {
783 consumer_del_stream_for_data(new_stream);
784 }
785 goto end_nosignal;
786 }
787
788 DBG("Kernel consumer ADD_STREAM %s (fd: %d) with relayd id %" PRIu64,
789 new_stream->name, fd, new_stream->relayd_stream_id);
790 break;
791 }
792 case LTTNG_CONSUMER_STREAMS_SENT:
793 {
794 struct lttng_consumer_channel *channel;
795
796 /*
797 * Get stream's channel reference. Needed when adding the stream to the
798 * global hash table.
799 */
800 channel = consumer_find_channel(msg.u.sent_streams.channel_key);
801 if (!channel) {
802 /*
803 * We could not find the channel. Can happen if cpu hotplug
804 * happens while tearing down.
805 */
806 ERR("Unable to find channel key %" PRIu64,
807 msg.u.sent_streams.channel_key);
808 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
809 }
810
811 health_code_update();
812
813 /*
814 * Send status code to session daemon.
815 */
816 ret = consumer_send_status_msg(sock, ret_code);
817 if (ret < 0 || ret_code != LTTCOMM_CONSUMERD_SUCCESS) {
818 /* Somehow, the session daemon is not responding anymore. */
819 goto end_nosignal;
820 }
821
822 health_code_update();
823
824 /*
825 * We should not send this message if we don't monitor the
826 * streams in this channel.
827 */
828 if (!channel->monitor) {
829 break;
830 }
831
832 health_code_update();
833 /* Send stream to relayd if the stream has an ID. */
834 if (msg.u.sent_streams.net_seq_idx != (uint64_t) -1ULL) {
835 ret = consumer_send_relayd_streams_sent(
836 msg.u.sent_streams.net_seq_idx);
837 if (ret < 0) {
838 goto end_nosignal;
839 }
840 channel->streams_sent_to_relayd = true;
841 }
842 break;
843 }
844 case LTTNG_CONSUMER_UPDATE_STREAM:
845 {
846 rcu_read_unlock();
847 return -ENOSYS;
848 }
849 case LTTNG_CONSUMER_DESTROY_RELAYD:
850 {
851 uint64_t index = msg.u.destroy_relayd.net_seq_idx;
852 struct consumer_relayd_sock_pair *relayd;
853
854 DBG("Kernel consumer destroying relayd %" PRIu64, index);
855
856 /* Get relayd reference if exists. */
857 relayd = consumer_find_relayd(index);
858 if (relayd == NULL) {
859 DBG("Unable to find relayd %" PRIu64, index);
860 ret_code = LTTCOMM_CONSUMERD_RELAYD_FAIL;
861 }
862
863 /*
864 * Each relayd socket pair has a refcount of stream attached to it
865 * which tells if the relayd is still active or not depending on the
866 * refcount value.
867 *
868 * This will set the destroy flag of the relayd object and destroy it
869 * if the refcount reaches zero when called.
870 *
871 * The destroy can happen either here or when a stream fd hangs up.
872 */
873 if (relayd) {
874 consumer_flag_relayd_for_destroy(relayd);
875 }
876
877 health_code_update();
878
879 ret = consumer_send_status_msg(sock, ret_code);
880 if (ret < 0) {
881 /* Somehow, the session daemon is not responding anymore. */
882 goto error_fatal;
883 }
884
885 goto end_nosignal;
886 }
887 case LTTNG_CONSUMER_DATA_PENDING:
888 {
889 int32_t ret;
890 uint64_t id = msg.u.data_pending.session_id;
891
892 DBG("Kernel consumer data pending command for id %" PRIu64, id);
893
894 ret = consumer_data_pending(id);
895
896 health_code_update();
897
898 /* Send back returned value to session daemon */
899 ret = lttcomm_send_unix_sock(sock, &ret, sizeof(ret));
900 if (ret < 0) {
901 PERROR("send data pending ret code");
902 goto error_fatal;
903 }
904
905 /*
906 * No need to send back a status message since the data pending
907 * returned value is the response.
908 */
909 break;
910 }
911 case LTTNG_CONSUMER_SNAPSHOT_CHANNEL:
912 {
913 if (msg.u.snapshot_channel.metadata == 1) {
914 ret = lttng_kconsumer_snapshot_metadata(msg.u.snapshot_channel.key,
915 msg.u.snapshot_channel.pathname,
916 msg.u.snapshot_channel.relayd_id, ctx);
917 if (ret < 0) {
918 ERR("Snapshot metadata failed");
919 ret_code = LTTCOMM_CONSUMERD_ERROR_METADATA;
920 }
921 } else {
922 ret = lttng_kconsumer_snapshot_channel(msg.u.snapshot_channel.key,
923 msg.u.snapshot_channel.pathname,
924 msg.u.snapshot_channel.relayd_id,
925 msg.u.snapshot_channel.nb_packets_per_stream,
926 ctx);
927 if (ret < 0) {
928 ERR("Snapshot channel failed");
929 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
930 }
931 }
932
933 health_code_update();
934
935 ret = consumer_send_status_msg(sock, ret_code);
936 if (ret < 0) {
937 /* Somehow, the session daemon is not responding anymore. */
938 goto end_nosignal;
939 }
940 break;
941 }
942 case LTTNG_CONSUMER_DESTROY_CHANNEL:
943 {
944 uint64_t key = msg.u.destroy_channel.key;
945 struct lttng_consumer_channel *channel;
946
947 channel = consumer_find_channel(key);
948 if (!channel) {
949 ERR("Kernel consumer destroy channel %" PRIu64 " not found", key);
950 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
951 }
952
953 health_code_update();
954
955 ret = consumer_send_status_msg(sock, ret_code);
956 if (ret < 0) {
957 /* Somehow, the session daemon is not responding anymore. */
958 goto end_nosignal;
959 }
960
961 health_code_update();
962
963 /* Stop right now if no channel was found. */
964 if (!channel) {
965 goto end_nosignal;
966 }
967
968 /*
969 * This command should ONLY be issued for channel with streams set in
970 * no monitor mode.
971 */
972 assert(!channel->monitor);
973
974 /*
975 * The refcount should ALWAYS be 0 in the case of a channel in no
976 * monitor mode.
977 */
978 assert(!uatomic_sub_return(&channel->refcount, 1));
979
980 consumer_del_channel(channel);
981
982 goto end_nosignal;
983 }
984 case LTTNG_CONSUMER_DISCARDED_EVENTS:
985 {
986 ssize_t ret;
987 uint64_t count;
988 struct lttng_consumer_channel *channel;
989 uint64_t id = msg.u.discarded_events.session_id;
990 uint64_t key = msg.u.discarded_events.channel_key;
991
992 DBG("Kernel consumer discarded events command for session id %"
993 PRIu64 ", channel key %" PRIu64, id, key);
994
995 channel = consumer_find_channel(key);
996 if (!channel) {
997 ERR("Kernel consumer discarded events channel %"
998 PRIu64 " not found", key);
999 count = 0;
1000 } else {
1001 count = channel->discarded_events;
1002 }
1003
1004 health_code_update();
1005
1006 /* Send back returned value to session daemon */
1007 ret = lttcomm_send_unix_sock(sock, &count, sizeof(count));
1008 if (ret < 0) {
1009 PERROR("send discarded events");
1010 goto error_fatal;
1011 }
1012
1013 break;
1014 }
1015 case LTTNG_CONSUMER_LOST_PACKETS:
1016 {
1017 ssize_t ret;
1018 uint64_t count;
1019 struct lttng_consumer_channel *channel;
1020 uint64_t id = msg.u.lost_packets.session_id;
1021 uint64_t key = msg.u.lost_packets.channel_key;
1022
1023 DBG("Kernel consumer lost packets command for session id %"
1024 PRIu64 ", channel key %" PRIu64, id, key);
1025
1026 channel = consumer_find_channel(key);
1027 if (!channel) {
1028 ERR("Kernel consumer lost packets channel %"
1029 PRIu64 " not found", key);
1030 count = 0;
1031 } else {
1032 count = channel->lost_packets;
1033 }
1034
1035 health_code_update();
1036
1037 /* Send back returned value to session daemon */
1038 ret = lttcomm_send_unix_sock(sock, &count, sizeof(count));
1039 if (ret < 0) {
1040 PERROR("send lost packets");
1041 goto error_fatal;
1042 }
1043
1044 break;
1045 }
1046 case LTTNG_CONSUMER_SET_CHANNEL_MONITOR_PIPE:
1047 {
1048 int channel_monitor_pipe;
1049
1050 ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1051 /* Successfully received the command's type. */
1052 ret = consumer_send_status_msg(sock, ret_code);
1053 if (ret < 0) {
1054 goto error_fatal;
1055 }
1056
1057 ret = lttcomm_recv_fds_unix_sock(sock, &channel_monitor_pipe,
1058 1);
1059 if (ret != sizeof(channel_monitor_pipe)) {
1060 ERR("Failed to receive channel monitor pipe");
1061 goto error_fatal;
1062 }
1063
1064 DBG("Received channel monitor pipe (%d)", channel_monitor_pipe);
1065 ret = consumer_timer_thread_set_channel_monitor_pipe(
1066 channel_monitor_pipe);
1067 if (!ret) {
1068 int flags;
1069
1070 ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1071 /* Set the pipe as non-blocking. */
1072 ret = fcntl(channel_monitor_pipe, F_GETFL, 0);
1073 if (ret == -1) {
1074 PERROR("fcntl get flags of the channel monitoring pipe");
1075 goto error_fatal;
1076 }
1077 flags = ret;
1078
1079 ret = fcntl(channel_monitor_pipe, F_SETFL,
1080 flags | O_NONBLOCK);
1081 if (ret == -1) {
1082 PERROR("fcntl set O_NONBLOCK flag of the channel monitoring pipe");
1083 goto error_fatal;
1084 }
1085 DBG("Channel monitor pipe set as non-blocking");
1086 } else {
1087 ret_code = LTTCOMM_CONSUMERD_ALREADY_SET;
1088 }
1089 ret = consumer_send_status_msg(sock, ret_code);
1090 if (ret < 0) {
1091 goto error_fatal;
1092 }
1093 break;
1094 }
1095 case LTTNG_CONSUMER_SET_CHANNEL_ROTATE_PIPE:
1096 {
1097 int channel_rotate_pipe;
1098 int flags;
1099
1100 ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1101 /* Successfully received the command's type. */
1102 ret = consumer_send_status_msg(sock, ret_code);
1103 if (ret < 0) {
1104 goto error_fatal;
1105 }
1106
1107 ret = lttcomm_recv_fds_unix_sock(sock, &channel_rotate_pipe,
1108 1);
1109 if (ret != sizeof(channel_rotate_pipe)) {
1110 ERR("Failed to receive channel rotate pipe");
1111 goto error_fatal;
1112 }
1113
1114 DBG("Received channel rotate pipe (%d)", channel_rotate_pipe);
1115 ctx->channel_rotate_pipe = channel_rotate_pipe;
1116 /* Set the pipe as non-blocking. */
1117 ret = fcntl(channel_rotate_pipe, F_GETFL, 0);
1118 if (ret == -1) {
1119 PERROR("fcntl get flags of the channel rotate pipe");
1120 goto error_fatal;
1121 }
1122 flags = ret;
1123
1124 ret = fcntl(channel_rotate_pipe, F_SETFL,
1125 flags | O_NONBLOCK);
1126 if (ret == -1) {
1127 PERROR("fcntl set O_NONBLOCK flag of the channel rotate pipe");
1128 goto error_fatal;
1129 }
1130 DBG("Channel rotate pipe set as non-blocking");
1131 ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1132 ret = consumer_send_status_msg(sock, ret_code);
1133 if (ret < 0) {
1134 goto error_fatal;
1135 }
1136 break;
1137 }
1138 case LTTNG_CONSUMER_ROTATE_CHANNEL:
1139 {
1140 ret = lttng_consumer_rotate_channel(msg.u.rotate_channel.key,
1141 msg.u.rotate_channel.pathname,
1142 msg.u.rotate_channel.relayd_id,
1143 msg.u.rotate_channel.metadata,
1144 msg.u.rotate_channel.new_chunk_id,
1145 ctx);
1146 if (ret < 0) {
1147 ERR("Rotate channel failed");
1148 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
1149 }
1150
1151 health_code_update();
1152
1153 ret = consumer_send_status_msg(sock, ret_code);
1154 if (ret < 0) {
1155 /* Somehow, the session daemon is not responding anymore. */
1156 goto end_nosignal;
1157 }
1158
1159 /* Rotate the streams that are ready right now. */
1160 ret = lttng_consumer_rotate_ready_streams(
1161 msg.u.rotate_channel.key, ctx);
1162 if (ret < 0) {
1163 ERR("Rotate ready streams failed");
1164 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
1165 }
1166
1167 break;
1168 }
1169 case LTTNG_CONSUMER_ROTATE_RENAME:
1170 {
1171 DBG("Consumer rename session %" PRIu64 " after rotation",
1172 msg.u.rotate_rename.session_id);
1173 ret = lttng_consumer_rotate_rename(msg.u.rotate_rename.current_path,
1174 msg.u.rotate_rename.new_path,
1175 msg.u.rotate_rename.uid,
1176 msg.u.rotate_rename.gid,
1177 msg.u.rotate_rename.relayd_id);
1178 if (ret < 0) {
1179 ERR("Rotate rename failed");
1180 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
1181 }
1182
1183 health_code_update();
1184
1185 ret = consumer_send_status_msg(sock, ret_code);
1186 if (ret < 0) {
1187 /* Somehow, the session daemon is not responding anymore. */
1188 goto end_nosignal;
1189 }
1190 break;
1191 }
1192 case LTTNG_CONSUMER_ROTATE_PENDING_RELAY:
1193 {
1194 uint32_t pending;
1195
1196 DBG("Consumer rotate pending on relay for session %" PRIu64,
1197 msg.u.rotate_pending_relay.session_id);
1198 pending = lttng_consumer_rotate_pending_relay(
1199 msg.u.rotate_pending_relay.session_id,
1200 msg.u.rotate_pending_relay.relayd_id,
1201 msg.u.rotate_pending_relay.chunk_id);
1202 if (ret < 0) {
1203 ERR("Rotate pending relay failed");
1204 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
1205 }
1206
1207 health_code_update();
1208
1209 ret = consumer_send_status_msg(sock, ret_code);
1210 if (ret < 0) {
1211 /* Somehow, the session daemon is not responding anymore. */
1212 goto end_nosignal;
1213 }
1214
1215 /* Send back returned value to session daemon */
1216 ret = lttcomm_send_unix_sock(sock, &pending, sizeof(pending));
1217 if (ret < 0) {
1218 PERROR("send data pending ret code");
1219 goto error_fatal;
1220 }
1221 break;
1222 }
1223 default:
1224 goto end_nosignal;
1225 }
1226
1227 end_nosignal:
1228 rcu_read_unlock();
1229
1230 /*
1231 * Return 1 to indicate success since the 0 value can be a socket
1232 * shutdown during the recv() or send() call.
1233 */
1234 health_code_update();
1235 return 1;
1236
1237 error_fatal:
1238 rcu_read_unlock();
1239 /* This will issue a consumer stop. */
1240 return -1;
1241 }
1242
1243 /*
1244 * Populate index values of a kernel stream. Values are set in big endian order.
1245 *
1246 * Return 0 on success or else a negative value.
1247 */
1248 static int get_index_values(struct ctf_packet_index *index, int infd)
1249 {
1250 int ret;
1251
1252 ret = kernctl_get_timestamp_begin(infd, &index->timestamp_begin);
1253 if (ret < 0) {
1254 PERROR("kernctl_get_timestamp_begin");
1255 goto error;
1256 }
1257 index->timestamp_begin = htobe64(index->timestamp_begin);
1258
1259 ret = kernctl_get_timestamp_end(infd, &index->timestamp_end);
1260 if (ret < 0) {
1261 PERROR("kernctl_get_timestamp_end");
1262 goto error;
1263 }
1264 index->timestamp_end = htobe64(index->timestamp_end);
1265
1266 ret = kernctl_get_events_discarded(infd, &index->events_discarded);
1267 if (ret < 0) {
1268 PERROR("kernctl_get_events_discarded");
1269 goto error;
1270 }
1271 index->events_discarded = htobe64(index->events_discarded);
1272
1273 ret = kernctl_get_content_size(infd, &index->content_size);
1274 if (ret < 0) {
1275 PERROR("kernctl_get_content_size");
1276 goto error;
1277 }
1278 index->content_size = htobe64(index->content_size);
1279
1280 ret = kernctl_get_packet_size(infd, &index->packet_size);
1281 if (ret < 0) {
1282 PERROR("kernctl_get_packet_size");
1283 goto error;
1284 }
1285 index->packet_size = htobe64(index->packet_size);
1286
1287 ret = kernctl_get_stream_id(infd, &index->stream_id);
1288 if (ret < 0) {
1289 PERROR("kernctl_get_stream_id");
1290 goto error;
1291 }
1292 index->stream_id = htobe64(index->stream_id);
1293
1294 ret = kernctl_get_instance_id(infd, &index->stream_instance_id);
1295 if (ret < 0) {
1296 if (ret == -ENOTTY) {
1297 /* Command not implemented by lttng-modules. */
1298 index->stream_instance_id = -1ULL;
1299 } else {
1300 PERROR("kernctl_get_instance_id");
1301 goto error;
1302 }
1303 }
1304 index->stream_instance_id = htobe64(index->stream_instance_id);
1305
1306 ret = kernctl_get_sequence_number(infd, &index->packet_seq_num);
1307 if (ret < 0) {
1308 if (ret == -ENOTTY) {
1309 /* Command not implemented by lttng-modules. */
1310 index->packet_seq_num = -1ULL;
1311 ret = 0;
1312 } else {
1313 PERROR("kernctl_get_sequence_number");
1314 goto error;
1315 }
1316 }
1317 index->packet_seq_num = htobe64(index->packet_seq_num);
1318
1319 error:
1320 return ret;
1321 }
1322 /*
1323 * Sync metadata meaning request them to the session daemon and snapshot to the
1324 * metadata thread can consumer them.
1325 *
1326 * Metadata stream lock MUST be acquired.
1327 *
1328 * Return 0 if new metadatda is available, EAGAIN if the metadata stream
1329 * is empty or a negative value on error.
1330 */
1331 int lttng_kconsumer_sync_metadata(struct lttng_consumer_stream *metadata)
1332 {
1333 int ret;
1334
1335 assert(metadata);
1336
1337 ret = kernctl_buffer_flush(metadata->wait_fd);
1338 if (ret < 0) {
1339 ERR("Failed to flush kernel stream");
1340 goto end;
1341 }
1342
1343 ret = kernctl_snapshot(metadata->wait_fd);
1344 if (ret < 0) {
1345 if (ret != -EAGAIN) {
1346 ERR("Sync metadata, taking kernel snapshot failed.");
1347 goto end;
1348 }
1349 DBG("Sync metadata, no new kernel metadata");
1350 /* No new metadata, exit. */
1351 ret = ENODATA;
1352 goto end;
1353 }
1354
1355 end:
1356 return ret;
1357 }
1358
1359 static
1360 int update_stream_stats(struct lttng_consumer_stream *stream)
1361 {
1362 int ret;
1363 uint64_t seq, discarded;
1364
1365 ret = kernctl_get_sequence_number(stream->wait_fd, &seq);
1366 if (ret < 0) {
1367 if (ret == -ENOTTY) {
1368 /* Command not implemented by lttng-modules. */
1369 seq = -1ULL;
1370 } else {
1371 PERROR("kernctl_get_sequence_number");
1372 goto end;
1373 }
1374 }
1375
1376 /*
1377 * Start the sequence when we extract the first packet in case we don't
1378 * start at 0 (for example if a consumer is not connected to the
1379 * session immediately after the beginning).
1380 */
1381 if (stream->last_sequence_number == -1ULL) {
1382 stream->last_sequence_number = seq;
1383 } else if (seq > stream->last_sequence_number) {
1384 stream->chan->lost_packets += seq -
1385 stream->last_sequence_number - 1;
1386 } else {
1387 /* seq <= last_sequence_number */
1388 ERR("Sequence number inconsistent : prev = %" PRIu64
1389 ", current = %" PRIu64,
1390 stream->last_sequence_number, seq);
1391 ret = -1;
1392 goto end;
1393 }
1394 stream->last_sequence_number = seq;
1395
1396 ret = kernctl_get_events_discarded(stream->wait_fd, &discarded);
1397 if (ret < 0) {
1398 PERROR("kernctl_get_events_discarded");
1399 goto end;
1400 }
1401 if (discarded < stream->last_discarded_events) {
1402 /*
1403 * Overflow has occurred. We assume only one wrap-around
1404 * has occurred.
1405 */
1406 stream->chan->discarded_events += (1ULL << (CAA_BITS_PER_LONG - 1)) -
1407 stream->last_discarded_events + discarded;
1408 } else {
1409 stream->chan->discarded_events += discarded -
1410 stream->last_discarded_events;
1411 }
1412 stream->last_discarded_events = discarded;
1413 ret = 0;
1414
1415 end:
1416 return ret;
1417 }
1418
1419 /*
1420 * Check if the local version of the metadata stream matches with the version
1421 * of the metadata stream in the kernel. If it was updated, set the reset flag
1422 * on the stream.
1423 */
1424 static
1425 int metadata_stream_check_version(int infd, struct lttng_consumer_stream *stream)
1426 {
1427 int ret;
1428 uint64_t cur_version;
1429
1430 ret = kernctl_get_metadata_version(infd, &cur_version);
1431 if (ret < 0) {
1432 if (ret == -ENOTTY) {
1433 /*
1434 * LTTng-modules does not implement this
1435 * command.
1436 */
1437 ret = 0;
1438 goto end;
1439 }
1440 ERR("Failed to get the metadata version");
1441 goto end;
1442 }
1443
1444 if (stream->metadata_version == cur_version) {
1445 ret = 0;
1446 goto end;
1447 }
1448
1449 DBG("New metadata version detected");
1450 stream->metadata_version = cur_version;
1451 stream->reset_metadata_flag = 1;
1452 ret = 0;
1453
1454 end:
1455 return ret;
1456 }
1457
1458 /*
1459 * Consume data on a file descriptor and write it on a trace file.
1460 */
1461 ssize_t lttng_kconsumer_read_subbuffer(struct lttng_consumer_stream *stream,
1462 struct lttng_consumer_local_data *ctx)
1463 {
1464 unsigned long len, subbuf_size, padding;
1465 int err, write_index = 1, rotation_ret, rotate_ready;
1466 ssize_t ret = 0;
1467 int infd = stream->wait_fd;
1468 struct ctf_packet_index index;
1469
1470 DBG("In read_subbuffer (infd : %d)", infd);
1471
1472 /* Get the next subbuffer */
1473 err = kernctl_get_next_subbuf(infd);
1474 if (err != 0) {
1475 /*
1476 * This is a debug message even for single-threaded consumer,
1477 * because poll() have more relaxed criterions than get subbuf,
1478 * so get_subbuf may fail for short race windows where poll()
1479 * would issue wakeups.
1480 */
1481 DBG("Reserving sub buffer failed (everything is normal, "
1482 "it is due to concurrency)");
1483 ret = err;
1484 goto error;
1485 }
1486
1487 /* Get the full subbuffer size including padding */
1488 err = kernctl_get_padded_subbuf_size(infd, &len);
1489 if (err != 0) {
1490 PERROR("Getting sub-buffer len failed.");
1491 err = kernctl_put_subbuf(infd);
1492 if (err != 0) {
1493 if (err == -EFAULT) {
1494 PERROR("Error in unreserving sub buffer\n");
1495 } else if (err == -EIO) {
1496 /* Should never happen with newer LTTng versions */
1497 PERROR("Reader has been pushed by the writer, last sub-buffer corrupted.");
1498 }
1499 ret = err;
1500 goto error;
1501 }
1502 ret = err;
1503 goto error;
1504 }
1505
1506 rotate_ready = lttng_consumer_stream_is_rotate_ready(stream, len);
1507 fprintf(stderr, "consumer read stream %lu, len %lu, ready = %d\n", stream->key,
1508 len, rotate_ready);
1509 if (rotate_ready < 0) {
1510 ERR("Failed to check if stream is ready for rotation");
1511 err = kernctl_put_subbuf(infd);
1512 if (err != 0) {
1513 if (err == -EFAULT) {
1514 PERROR("Error in unreserving sub buffer\n");
1515 } else if (err == -EIO) {
1516 /* Should never happen with newer LTTng versions */
1517 PERROR("Reader has been pushed by the writer, last sub-buffer corrupted.");
1518 }
1519 ret = err;
1520 goto error;
1521 }
1522 ret = -1;
1523 goto error;
1524 }
1525 stream->rotate_ready = rotate_ready;
1526
1527 if (!stream->metadata_flag) {
1528 ret = get_index_values(&index, infd);
1529 if (ret < 0) {
1530 err = kernctl_put_subbuf(infd);
1531 if (err != 0) {
1532 if (err == -EFAULT) {
1533 PERROR("Error in unreserving sub buffer\n");
1534 } else if (err == -EIO) {
1535 /* Should never happen with newer LTTng versions */
1536 PERROR("Reader has been pushed by the writer, last sub-buffer corrupted.");
1537 }
1538 ret = err;
1539 goto error;
1540 }
1541 goto error;
1542 }
1543 ret = update_stream_stats(stream);
1544 if (ret < 0) {
1545 err = kernctl_put_subbuf(infd);
1546 if (err != 0) {
1547 if (err == -EFAULT) {
1548 PERROR("Error in unreserving sub buffer\n");
1549 } else if (err == -EIO) {
1550 /* Should never happen with newer LTTng versions */
1551 PERROR("Reader has been pushed by the writer, last sub-buffer corrupted.");
1552 }
1553 ret = err;
1554 goto error;
1555 }
1556 goto error;
1557 }
1558 } else {
1559 write_index = 0;
1560 ret = metadata_stream_check_version(infd, stream);
1561 if (ret < 0) {
1562 err = kernctl_put_subbuf(infd);
1563 if (err != 0) {
1564 if (err == -EFAULT) {
1565 PERROR("Error in unreserving sub buffer\n");
1566 } else if (err == -EIO) {
1567 /* Should never happen with newer LTTng versions */
1568 PERROR("Reader has been pushed by the writer, last sub-buffer corrupted.");
1569 }
1570 ret = err;
1571 goto error;
1572 }
1573 goto error;
1574 }
1575 }
1576
1577 switch (stream->chan->output) {
1578 case CONSUMER_CHANNEL_SPLICE:
1579 /*
1580 * XXX: The lttng-modules splice "actor" does not handle copying
1581 * partial pages hence only using the subbuffer size without the
1582 * padding makes the splice fail.
1583 */
1584 subbuf_size = len;
1585 padding = 0;
1586
1587 /* splice the subbuffer to the tracefile */
1588 ret = lttng_consumer_on_read_subbuffer_splice(ctx, stream, subbuf_size,
1589 padding, &index);
1590 /*
1591 * XXX: Splice does not support network streaming so the return value
1592 * is simply checked against subbuf_size and not like the mmap() op.
1593 */
1594 if (ret != subbuf_size) {
1595 /*
1596 * display the error but continue processing to try
1597 * to release the subbuffer
1598 */
1599 ERR("Error splicing to tracefile (ret: %zd != len: %lu)",
1600 ret, subbuf_size);
1601 write_index = 0;
1602 }
1603 break;
1604 case CONSUMER_CHANNEL_MMAP:
1605 /* Get subbuffer size without padding */
1606 err = kernctl_get_subbuf_size(infd, &subbuf_size);
1607 if (err != 0) {
1608 PERROR("Getting sub-buffer len failed.");
1609 err = kernctl_put_subbuf(infd);
1610 if (err != 0) {
1611 if (err == -EFAULT) {
1612 PERROR("Error in unreserving sub buffer\n");
1613 } else if (err == -EIO) {
1614 /* Should never happen with newer LTTng versions */
1615 PERROR("Reader has been pushed by the writer, last sub-buffer corrupted.");
1616 }
1617 ret = err;
1618 goto error;
1619 }
1620 ret = err;
1621 goto error;
1622 }
1623
1624 /* Make sure the tracer is not gone mad on us! */
1625 assert(len >= subbuf_size);
1626
1627 padding = len - subbuf_size;
1628
1629 /* write the subbuffer to the tracefile */
1630 ret = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, subbuf_size,
1631 padding, &index);
1632 /*
1633 * The mmap operation should write subbuf_size amount of data when
1634 * network streaming or the full padding (len) size when we are _not_
1635 * streaming.
1636 */
1637 if ((ret != subbuf_size && stream->net_seq_idx != (uint64_t) -1ULL) ||
1638 (ret != len && stream->net_seq_idx == (uint64_t) -1ULL)) {
1639 /*
1640 * Display the error but continue processing to try to release the
1641 * subbuffer. This is a DBG statement since this is possible to
1642 * happen without being a critical error.
1643 */
1644 DBG("Error writing to tracefile "
1645 "(ret: %zd != len: %lu != subbuf_size: %lu)",
1646 ret, len, subbuf_size);
1647 write_index = 0;
1648 }
1649 break;
1650 default:
1651 ERR("Unknown output method");
1652 ret = -EPERM;
1653 }
1654
1655 err = kernctl_put_next_subbuf(infd);
1656 if (err != 0) {
1657 if (err == -EFAULT) {
1658 PERROR("Error in unreserving sub buffer\n");
1659 } else if (err == -EIO) {
1660 /* Should never happen with newer LTTng versions */
1661 PERROR("Reader has been pushed by the writer, last sub-buffer corrupted.");
1662 }
1663 ret = err;
1664 goto error;
1665 }
1666 fprintf(stderr, "consumer read stream %lu done\n", stream->key);
1667
1668 /* Write index if needed. */
1669 if (!write_index) {
1670 goto rotate;
1671 }
1672
1673 if (stream->chan->live_timer_interval && !stream->metadata_flag) {
1674 /*
1675 * In live, block until all the metadata is sent.
1676 */
1677 pthread_mutex_lock(&stream->metadata_timer_lock);
1678 assert(!stream->missed_metadata_flush);
1679 stream->waiting_on_metadata = true;
1680 pthread_mutex_unlock(&stream->metadata_timer_lock);
1681
1682 err = consumer_stream_sync_metadata(ctx, stream->session_id);
1683
1684 pthread_mutex_lock(&stream->metadata_timer_lock);
1685 stream->waiting_on_metadata = false;
1686 if (stream->missed_metadata_flush) {
1687 stream->missed_metadata_flush = false;
1688 pthread_mutex_unlock(&stream->metadata_timer_lock);
1689 (void) consumer_flush_kernel_index(stream);
1690 } else {
1691 pthread_mutex_unlock(&stream->metadata_timer_lock);
1692 }
1693 if (err < 0) {
1694 goto error;
1695 }
1696 }
1697
1698 err = consumer_stream_write_index(stream, &index);
1699 if (err < 0) {
1700 goto error;
1701 }
1702
1703 rotate:
1704 if (stream->rotate_ready) {
1705 rotation_ret = lttng_consumer_rotate_stream(ctx, stream);
1706 if (rotation_ret < 0) {
1707 ERR("Stream rotation error");
1708 ret = -1;
1709 goto error;
1710 }
1711 }
1712
1713 error:
1714 return ret;
1715 }
1716
1717 int lttng_kconsumer_on_recv_stream(struct lttng_consumer_stream *stream)
1718 {
1719 int ret;
1720
1721 assert(stream);
1722
1723 /*
1724 * Don't create anything if this is set for streaming or should not be
1725 * monitored.
1726 */
1727 if (stream->net_seq_idx == (uint64_t) -1ULL && stream->chan->monitor) {
1728 ret = utils_create_stream_file(stream->chan->pathname, stream->name,
1729 stream->chan->tracefile_size, stream->tracefile_count_current,
1730 stream->uid, stream->gid, NULL);
1731 if (ret < 0) {
1732 goto error;
1733 }
1734 stream->out_fd = ret;
1735 stream->tracefile_size_current = 0;
1736
1737 if (!stream->metadata_flag) {
1738 struct lttng_index_file *index_file;
1739
1740 index_file = lttng_index_file_create(stream->chan->pathname,
1741 stream->name, stream->uid, stream->gid,
1742 stream->chan->tracefile_size,
1743 stream->tracefile_count_current,
1744 CTF_INDEX_MAJOR, CTF_INDEX_MINOR);
1745 if (!index_file) {
1746 goto error;
1747 }
1748 assert(!stream->index_file);
1749 stream->index_file = index_file;
1750 }
1751 }
1752
1753 if (stream->output == LTTNG_EVENT_MMAP) {
1754 /* get the len of the mmap region */
1755 unsigned long mmap_len;
1756
1757 ret = kernctl_get_mmap_len(stream->wait_fd, &mmap_len);
1758 if (ret != 0) {
1759 PERROR("kernctl_get_mmap_len");
1760 goto error_close_fd;
1761 }
1762 stream->mmap_len = (size_t) mmap_len;
1763
1764 stream->mmap_base = mmap(NULL, stream->mmap_len, PROT_READ,
1765 MAP_PRIVATE, stream->wait_fd, 0);
1766 if (stream->mmap_base == MAP_FAILED) {
1767 PERROR("Error mmaping");
1768 ret = -1;
1769 goto error_close_fd;
1770 }
1771 }
1772
1773 /* we return 0 to let the library handle the FD internally */
1774 return 0;
1775
1776 error_close_fd:
1777 if (stream->out_fd >= 0) {
1778 int err;
1779
1780 err = close(stream->out_fd);
1781 assert(!err);
1782 stream->out_fd = -1;
1783 }
1784 error:
1785 return ret;
1786 }
1787
1788 /*
1789 * Check if data is still being extracted from the buffers for a specific
1790 * stream. Consumer data lock MUST be acquired before calling this function
1791 * and the stream lock.
1792 *
1793 * Return 1 if the traced data are still getting read else 0 meaning that the
1794 * data is available for trace viewer reading.
1795 */
1796 int lttng_kconsumer_data_pending(struct lttng_consumer_stream *stream)
1797 {
1798 int ret;
1799
1800 assert(stream);
1801
1802 if (stream->endpoint_status != CONSUMER_ENDPOINT_ACTIVE) {
1803 ret = 0;
1804 goto end;
1805 }
1806
1807 ret = kernctl_get_next_subbuf(stream->wait_fd);
1808 if (ret == 0) {
1809 /* There is still data so let's put back this subbuffer. */
1810 ret = kernctl_put_subbuf(stream->wait_fd);
1811 assert(ret == 0);
1812 ret = 1; /* Data is pending */
1813 goto end;
1814 }
1815
1816 /* Data is NOT pending and ready to be read. */
1817 ret = 0;
1818
1819 end:
1820 return ret;
1821 }
This page took 0.068811 seconds and 5 git commands to generate.