projects / lttng-tools.git / blob
? search:


b16a1d053a52e18f53b0c63d535552ffdc81ed59
[lttng-tools.git] / src / common / kernel-consumer / kernel-consumer.c
1 /*
2 * Copyright (C) 2011 EfficiOS Inc.
3 * Copyright (C) 2011 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
4 * Copyright (C) 2017 Jérémie Galarneau <jeremie.galarneau@efficios.com>
5 *
6 * SPDX-License-Identifier: GPL-2.0-only
7 *
8 */
9
10 #define _LGPL_SOURCE
11 #include <assert.h>
12 #include <poll.h>
13 #include <pthread.h>
14 #include <stdlib.h>
15 #include <string.h>
16 #include <sys/mman.h>
17 #include <sys/socket.h>
18 #include <sys/types.h>
19 #include <inttypes.h>
20 #include <unistd.h>
21 #include <sys/stat.h>
22 #include <stdint.h>
23
24 #include <bin/lttng-consumerd/health-consumerd.h>
25 #include <common/common.h>
26 #include <common/kernel-ctl/kernel-ctl.h>
27 #include <common/sessiond-comm/sessiond-comm.h>
28 #include <common/sessiond-comm/relayd.h>
29 #include <common/compat/fcntl.h>
30 #include <common/compat/endian.h>
31 #include <common/pipe.h>
32 #include <common/relayd/relayd.h>
33 #include <common/utils.h>
34 #include <common/consumer/consumer-stream.h>
35 #include <common/index/index.h>
36 #include <common/consumer/consumer-timer.h>
37 #include <common/optional.h>
38 #include <common/buffer-view.h>
39 #include <common/consumer/consumer.h>
40 #include <common/consumer/metadata-bucket.h>
41
42 #include "kernel-consumer.h"
43
44 extern struct lttng_consumer_global_data consumer_data;
45 extern int consumer_poll_timeout;
46
47 /*
48 * Take a snapshot for a specific fd
49 *
50 * Returns 0 on success, < 0 on error
51 */
52 int lttng_kconsumer_take_snapshot(struct lttng_consumer_stream *stream)
53 {
54 int ret = 0;
55 int infd = stream->wait_fd;
56
57 ret = kernctl_snapshot(infd);
58 /*
59 * -EAGAIN is not an error, it just means that there is no data to
60 * be read.
61 */
62 if (ret != 0 && ret != -EAGAIN) {
63 PERROR("Getting sub-buffer snapshot.");
64 }
65
66 return ret;
67 }
68
69 /*
70 * Sample consumed and produced positions for a specific fd.
71 *
72 * Returns 0 on success, < 0 on error.
73 */
74 int lttng_kconsumer_sample_snapshot_positions(
75 struct lttng_consumer_stream *stream)
76 {
77 assert(stream);
78
79 return kernctl_snapshot_sample_positions(stream->wait_fd);
80 }
81
82 /*
83 * Get the produced position
84 *
85 * Returns 0 on success, < 0 on error
86 */
87 int lttng_kconsumer_get_produced_snapshot(struct lttng_consumer_stream *stream,
88 unsigned long *pos)
89 {
90 int ret;
91 int infd = stream->wait_fd;
92
93 ret = kernctl_snapshot_get_produced(infd, pos);
94 if (ret != 0) {
95 PERROR("kernctl_snapshot_get_produced");
96 }
97
98 return ret;
99 }
100
101 /*
102 * Get the consumerd position
103 *
104 * Returns 0 on success, < 0 on error
105 */
106 int lttng_kconsumer_get_consumed_snapshot(struct lttng_consumer_stream *stream,
107 unsigned long *pos)
108 {
109 int ret;
110 int infd = stream->wait_fd;
111
112 ret = kernctl_snapshot_get_consumed(infd, pos);
113 if (ret != 0) {
114 PERROR("kernctl_snapshot_get_consumed");
115 }
116
117 return ret;
118 }
119
120 static
121 int get_current_subbuf_addr(struct lttng_consumer_stream *stream,
122 const char **addr)
123 {
124 int ret;
125 unsigned long mmap_offset;
126 const char *mmap_base = stream->mmap_base;
127
128 ret = kernctl_get_mmap_read_offset(stream->wait_fd, &mmap_offset);
129 if (ret < 0) {
130 PERROR("Failed to get mmap read offset");
131 goto error;
132 }
133
134 *addr = mmap_base + mmap_offset;
135 error:
136 return ret;
137 }
138
139 /*
140 * Take a snapshot of all the stream of a channel
141 * RCU read-side lock must be held across this function to ensure existence of
142 * channel.
143 *
144 * Returns 0 on success, < 0 on error
145 */
146 static int lttng_kconsumer_snapshot_channel(
147 struct lttng_consumer_channel *channel,
148 uint64_t key, char *path, uint64_t relayd_id,
149 uint64_t nb_packets_per_stream,
150 struct lttng_consumer_local_data *ctx)
151 {
152 int ret;
153 struct lttng_consumer_stream *stream;
154
155 DBG("Kernel consumer snapshot channel %" PRIu64, key);
156
157 /* Prevent channel modifications while we perform the snapshot.*/
158 pthread_mutex_lock(&channel->lock);
159
160 rcu_read_lock();
161
162 /* Splice is not supported yet for channel snapshot. */
163 if (channel->output != CONSUMER_CHANNEL_MMAP) {
164 ERR("Unsupported output type for channel \"%s\": mmap output is required to record a snapshot",
165 channel->name);
166 ret = -1;
167 goto end;
168 }
169
170 cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
171 unsigned long consumed_pos, produced_pos;
172
173 health_code_update();
174
175 /*
176 * Lock stream because we are about to change its state.
177 */
178 pthread_mutex_lock(&stream->lock);
179
180 assert(channel->trace_chunk);
181 if (!lttng_trace_chunk_get(channel->trace_chunk)) {
182 /*
183 * Can't happen barring an internal error as the channel
184 * holds a reference to the trace chunk.
185 */
186 ERR("Failed to acquire reference to channel's trace chunk");
187 ret = -1;
188 goto end_unlock;
189 }
190 assert(!stream->trace_chunk);
191 stream->trace_chunk = channel->trace_chunk;
192
193 /*
194 * Assign the received relayd ID so we can use it for streaming. The streams
195 * are not visible to anyone so this is OK to change it.
196 */
197 stream->net_seq_idx = relayd_id;
198 channel->relayd_id = relayd_id;
199 if (relayd_id != (uint64_t) -1ULL) {
200 ret = consumer_send_relayd_stream(stream, path);
201 if (ret < 0) {
202 ERR("sending stream to relayd");
203 goto end_unlock;
204 }
205 } else {
206 ret = consumer_stream_create_output_files(stream,
207 false);
208 if (ret < 0) {
209 goto end_unlock;
210 }
211 DBG("Kernel consumer snapshot stream (%" PRIu64 ")",
212 stream->key);
213 }
214
215 ret = kernctl_buffer_flush_empty(stream->wait_fd);
216 if (ret < 0) {
217 /*
218 * Doing a buffer flush which does not take into
219 * account empty packets. This is not perfect
220 * for stream intersection, but required as a
221 * fall-back when "flush_empty" is not
222 * implemented by lttng-modules.
223 */
224 ret = kernctl_buffer_flush(stream->wait_fd);
225 if (ret < 0) {
226 ERR("Failed to flush kernel stream");
227 goto end_unlock;
228 }
229 goto end_unlock;
230 }
231
232 ret = lttng_kconsumer_take_snapshot(stream);
233 if (ret < 0) {
234 ERR("Taking kernel snapshot");
235 goto end_unlock;
236 }
237
238 ret = lttng_kconsumer_get_produced_snapshot(stream, &produced_pos);
239 if (ret < 0) {
240 ERR("Produced kernel snapshot position");
241 goto end_unlock;
242 }
243
244 ret = lttng_kconsumer_get_consumed_snapshot(stream, &consumed_pos);
245 if (ret < 0) {
246 ERR("Consumerd kernel snapshot position");
247 goto end_unlock;
248 }
249
250 consumed_pos = consumer_get_consume_start_pos(consumed_pos,
251 produced_pos, nb_packets_per_stream,
252 stream->max_sb_size);
253
254 while ((long) (consumed_pos - produced_pos) < 0) {
255 ssize_t read_len;
256 unsigned long len, padded_len;
257 const char *subbuf_addr;
258 struct lttng_buffer_view subbuf_view;
259
260 health_code_update();
261 DBG("Kernel consumer taking snapshot at pos %lu", consumed_pos);
262
263 ret = kernctl_get_subbuf(stream->wait_fd, &consumed_pos);
264 if (ret < 0) {
265 if (ret != -EAGAIN) {
266 PERROR("kernctl_get_subbuf snapshot");
267 goto end_unlock;
268 }
269 DBG("Kernel consumer get subbuf failed. Skipping it.");
270 consumed_pos += stream->max_sb_size;
271 stream->chan->lost_packets++;
272 continue;
273 }
274
275 ret = kernctl_get_subbuf_size(stream->wait_fd, &len);
276 if (ret < 0) {
277 ERR("Snapshot kernctl_get_subbuf_size");
278 goto error_put_subbuf;
279 }
280
281 ret = kernctl_get_padded_subbuf_size(stream->wait_fd, &padded_len);
282 if (ret < 0) {
283 ERR("Snapshot kernctl_get_padded_subbuf_size");
284 goto error_put_subbuf;
285 }
286
287 ret = get_current_subbuf_addr(stream, &subbuf_addr);
288 if (ret) {
289 goto error_put_subbuf;
290 }
291
292 subbuf_view = lttng_buffer_view_init(
293 subbuf_addr, 0, padded_len);
294 read_len = lttng_consumer_on_read_subbuffer_mmap(
295 stream, &subbuf_view,
296 padded_len - len);
297 /*
298 * We write the padded len in local tracefiles but the data len
299 * when using a relay. Display the error but continue processing
300 * to try to release the subbuffer.
301 */
302 if (relayd_id != (uint64_t) -1ULL) {
303 if (read_len != len) {
304 ERR("Error sending to the relay (ret: %zd != len: %lu)",
305 read_len, len);
306 }
307 } else {
308 if (read_len != padded_len) {
309 ERR("Error writing to tracefile (ret: %zd != len: %lu)",
310 read_len, padded_len);
311 }
312 }
313
314 ret = kernctl_put_subbuf(stream->wait_fd);
315 if (ret < 0) {
316 ERR("Snapshot kernctl_put_subbuf");
317 goto end_unlock;
318 }
319 consumed_pos += stream->max_sb_size;
320 }
321
322 if (relayd_id == (uint64_t) -1ULL) {
323 if (stream->out_fd >= 0) {
324 ret = close(stream->out_fd);
325 if (ret < 0) {
326 PERROR("Kernel consumer snapshot close out_fd");
327 goto end_unlock;
328 }
329 stream->out_fd = -1;
330 }
331 } else {
332 close_relayd_stream(stream);
333 stream->net_seq_idx = (uint64_t) -1ULL;
334 }
335 lttng_trace_chunk_put(stream->trace_chunk);
336 stream->trace_chunk = NULL;
337 pthread_mutex_unlock(&stream->lock);
338 }
339
340 /* All good! */
341 ret = 0;
342 goto end;
343
344 error_put_subbuf:
345 ret = kernctl_put_subbuf(stream->wait_fd);
346 if (ret < 0) {
347 ERR("Snapshot kernctl_put_subbuf error path");
348 }
349 end_unlock:
350 pthread_mutex_unlock(&stream->lock);
351 end:
352 rcu_read_unlock();
353 pthread_mutex_unlock(&channel->lock);
354 return ret;
355 }
356
357 /*
358 * Read the whole metadata available for a snapshot.
359 * RCU read-side lock must be held across this function to ensure existence of
360 * metadata_channel.
361 *
362 * Returns 0 on success, < 0 on error
363 */
364 static int lttng_kconsumer_snapshot_metadata(
365 struct lttng_consumer_channel *metadata_channel,
366 uint64_t key, char *path, uint64_t relayd_id,
367 struct lttng_consumer_local_data *ctx)
368 {
369 int ret, use_relayd = 0;
370 ssize_t ret_read;
371 struct lttng_consumer_stream *metadata_stream;
372
373 assert(ctx);
374
375 DBG("Kernel consumer snapshot metadata with key %" PRIu64 " at path %s",
376 key, path);
377
378 rcu_read_lock();
379
380 metadata_stream = metadata_channel->metadata_stream;
381 assert(metadata_stream);
382
383 /* Take all the appropriate locks hehehe.*/
384 metadata_stream->read_subbuffer_ops.lock(metadata_stream);
385 assert(metadata_channel->trace_chunk);
386 assert(metadata_stream->trace_chunk);
387
388 /* Flag once that we have a valid relayd for the stream. */
389 if (relayd_id != (uint64_t) -1ULL) {
390 use_relayd = 1;
391 }
392
393 if (use_relayd) {
394 ret = consumer_send_relayd_stream(metadata_stream, path);
395 if (ret < 0) {
396 goto error_snapshot;
397 }
398 } else {
399 ret = consumer_stream_create_output_files(metadata_stream,
400 false);
401 if (ret < 0) {
402 goto error_snapshot;
403 }
404 }
405
406 do {
407 health_code_update();
408
409 ret_read = lttng_consumer_read_subbuffer(metadata_stream, ctx, true);
410 if (ret_read < 0) {
411 ERR("Kernel snapshot reading metadata subbuffer (ret: %zd)",
412 ret_read);
413 ret = ret_read;
414 goto error_snapshot;
415 }
416 } while (ret_read > 0);
417
418 if (use_relayd) {
419 close_relayd_stream(metadata_stream);
420 metadata_stream->net_seq_idx = (uint64_t) -1ULL;
421 } else {
422 if (metadata_stream->out_fd >= 0) {
423 ret = close(metadata_stream->out_fd);
424 if (ret < 0) {
425 PERROR("Kernel consumer snapshot metadata close out_fd");
426 /*
427 * Don't go on error here since the snapshot was successful at this
428 * point but somehow the close failed.
429 */
430 }
431 metadata_stream->out_fd = -1;
432 lttng_trace_chunk_put(metadata_stream->trace_chunk);
433 metadata_stream->trace_chunk = NULL;
434 }
435 }
436
437 ret = 0;
438 error_snapshot:
439 metadata_stream->read_subbuffer_ops.unlock(metadata_stream);
440 consumer_stream_destroy(metadata_stream, NULL);
441 metadata_channel->metadata_stream = NULL;
442 rcu_read_unlock();
443 return ret;
444 }
445
446 /*
447 * Receive command from session daemon and process it.
448 *
449 * Return 1 on success else a negative value or 0.
450 */
451 int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
452 int sock, struct pollfd *consumer_sockpoll)
453 {
454 ssize_t ret;
455 enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS;
456 struct lttcomm_consumer_msg msg;
457
458 health_code_update();
459
460 ret = lttcomm_recv_unix_sock(sock, &msg, sizeof(msg));
461 if (ret != sizeof(msg)) {
462 if (ret > 0) {
463 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD);
464 ret = -1;
465 }
466 return ret;
467 }
468
469 health_code_update();
470
471 /* Deprecated command */
472 assert(msg.cmd_type != LTTNG_CONSUMER_STOP);
473
474 health_code_update();
475
476 /* relayd needs RCU read-side protection */
477 rcu_read_lock();
478
479 switch (msg.cmd_type) {
480 case LTTNG_CONSUMER_ADD_RELAYD_SOCKET:
481 {
482 uint32_t major = msg.u.relayd_sock.major;
483 uint32_t minor = msg.u.relayd_sock.minor;
484 enum lttcomm_sock_proto protocol =
485 msg.u.relayd_sock.relayd_socket_protocol;
486
487 /* Session daemon status message are handled in the following call. */
488 consumer_add_relayd_socket(msg.u.relayd_sock.net_index,
489 msg.u.relayd_sock.type, ctx, sock,
490 consumer_sockpoll, msg.u.relayd_sock.session_id,
491 msg.u.relayd_sock.relayd_session_id, major,
492 minor, protocol);
493 goto end_nosignal;
494 }
495 case LTTNG_CONSUMER_ADD_CHANNEL:
496 {
497 struct lttng_consumer_channel *new_channel;
498 int ret_recv;
499 const uint64_t chunk_id = msg.u.channel.chunk_id.value;
500
501 health_code_update();
502
503 /* First send a status message before receiving the fds. */
504 ret = consumer_send_status_msg(sock, ret_code);
505 if (ret < 0) {
506 /* Somehow, the session daemon is not responding anymore. */
507 goto error_fatal;
508 }
509
510 health_code_update();
511
512 DBG("consumer_add_channel %" PRIu64, msg.u.channel.channel_key);
513 new_channel = consumer_allocate_channel(msg.u.channel.channel_key,
514 msg.u.channel.session_id,
515 msg.u.channel.chunk_id.is_set ?
516 &chunk_id : NULL,
517 msg.u.channel.pathname,
518 msg.u.channel.name,
519 msg.u.channel.relayd_id, msg.u.channel.output,
520 msg.u.channel.tracefile_size,
521 msg.u.channel.tracefile_count, 0,
522 msg.u.channel.monitor,
523 msg.u.channel.live_timer_interval,
524 msg.u.channel.is_live,
525 NULL, NULL);
526 if (new_channel == NULL) {
527 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
528 goto end_nosignal;
529 }
530 new_channel->nb_init_stream_left = msg.u.channel.nb_init_streams;
531 switch (msg.u.channel.output) {
532 case LTTNG_EVENT_SPLICE:
533 new_channel->output = CONSUMER_CHANNEL_SPLICE;
534 break;
535 case LTTNG_EVENT_MMAP:
536 new_channel->output = CONSUMER_CHANNEL_MMAP;
537 break;
538 default:
539 ERR("Channel output unknown %d", msg.u.channel.output);
540 goto end_nosignal;
541 }
542
543 /* Translate and save channel type. */
544 switch (msg.u.channel.type) {
545 case CONSUMER_CHANNEL_TYPE_DATA:
546 case CONSUMER_CHANNEL_TYPE_METADATA:
547 new_channel->type = msg.u.channel.type;
548 break;
549 default:
550 assert(0);
551 goto end_nosignal;
552 };
553
554 health_code_update();
555
556 if (ctx->on_recv_channel != NULL) {
557 ret_recv = ctx->on_recv_channel(new_channel);
558 if (ret_recv == 0) {
559 ret = consumer_add_channel(new_channel, ctx);
560 } else if (ret_recv < 0) {
561 goto end_nosignal;
562 }
563 } else {
564 ret = consumer_add_channel(new_channel, ctx);
565 }
566 if (msg.u.channel.type == CONSUMER_CHANNEL_TYPE_DATA && !ret) {
567 int monitor_start_ret;
568
569 DBG("Consumer starting monitor timer");
570 consumer_timer_live_start(new_channel,
571 msg.u.channel.live_timer_interval);
572 monitor_start_ret = consumer_timer_monitor_start(
573 new_channel,
574 msg.u.channel.monitor_timer_interval);
575 if (monitor_start_ret < 0) {
576 ERR("Starting channel monitoring timer failed");
577 goto end_nosignal;
578 }
579
580 }
581
582 health_code_update();
583
584 /* If we received an error in add_channel, we need to report it. */
585 if (ret < 0) {
586 ret = consumer_send_status_msg(sock, ret);
587 if (ret < 0) {
588 goto error_fatal;
589 }
590 goto end_nosignal;
591 }
592
593 goto end_nosignal;
594 }
595 case LTTNG_CONSUMER_ADD_STREAM:
596 {
597 int fd;
598 struct lttng_pipe *stream_pipe;
599 struct lttng_consumer_stream *new_stream;
600 struct lttng_consumer_channel *channel;
601 int alloc_ret = 0;
602
603 /*
604 * Get stream's channel reference. Needed when adding the stream to the
605 * global hash table.
606 */
607 channel = consumer_find_channel(msg.u.stream.channel_key);
608 if (!channel) {
609 /*
610 * We could not find the channel. Can happen if cpu hotplug
611 * happens while tearing down.
612 */
613 ERR("Unable to find channel key %" PRIu64, msg.u.stream.channel_key);
614 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
615 }
616
617 health_code_update();
618
619 /* First send a status message before receiving the fds. */
620 ret = consumer_send_status_msg(sock, ret_code);
621 if (ret < 0) {
622 /* Somehow, the session daemon is not responding anymore. */
623 goto error_add_stream_fatal;
624 }
625
626 health_code_update();
627
628 if (ret_code != LTTCOMM_CONSUMERD_SUCCESS) {
629 /* Channel was not found. */
630 goto error_add_stream_nosignal;
631 }
632
633 /* Blocking call */
634 health_poll_entry();
635 ret = lttng_consumer_poll_socket(consumer_sockpoll);
636 health_poll_exit();
637 if (ret) {
638 goto error_add_stream_fatal;
639 }
640
641 health_code_update();
642
643 /* Get stream file descriptor from socket */
644 ret = lttcomm_recv_fds_unix_sock(sock, &fd, 1);
645 if (ret != sizeof(fd)) {
646 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_FD);
647 goto end;
648 }
649
650 health_code_update();
651
652 /*
653 * Send status code to session daemon only if the recv works. If the
654 * above recv() failed, the session daemon is notified through the
655 * error socket and the teardown is eventually done.
656 */
657 ret = consumer_send_status_msg(sock, ret_code);
658 if (ret < 0) {
659 /* Somehow, the session daemon is not responding anymore. */
660 goto error_add_stream_nosignal;
661 }
662
663 health_code_update();
664
665 pthread_mutex_lock(&channel->lock);
666 new_stream = consumer_stream_create(
667 channel,
668 channel->key,
669 fd,
670 channel->name,
671 channel->relayd_id,
672 channel->session_id,
673 channel->trace_chunk,
674 msg.u.stream.cpu,
675 &alloc_ret,
676 channel->type,
677 channel->monitor);
678 if (new_stream == NULL) {
679 switch (alloc_ret) {
680 case -ENOMEM:
681 case -EINVAL:
682 default:
683 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
684 break;
685 }
686 pthread_mutex_unlock(&channel->lock);
687 goto error_add_stream_nosignal;
688 }
689
690 new_stream->wait_fd = fd;
691 ret = kernctl_get_max_subbuf_size(new_stream->wait_fd,
692 &new_stream->max_sb_size);
693 if (ret < 0) {
694 pthread_mutex_unlock(&channel->lock);
695 ERR("Failed to get kernel maximal subbuffer size");
696 goto error_add_stream_nosignal;
697 }
698
699 consumer_stream_update_channel_attributes(new_stream,
700 channel);
701
702 /*
703 * We've just assigned the channel to the stream so increment the
704 * refcount right now. We don't need to increment the refcount for
705 * streams in no monitor because we handle manually the cleanup of
706 * those. It is very important to make sure there is NO prior
707 * consumer_del_stream() calls or else the refcount will be unbalanced.
708 */
709 if (channel->monitor) {
710 uatomic_inc(&new_stream->chan->refcount);
711 }
712
713 /*
714 * The buffer flush is done on the session daemon side for the kernel
715 * so no need for the stream "hangup_flush_done" variable to be
716 * tracked. This is important for a kernel stream since we don't rely
717 * on the flush state of the stream to read data. It's not the case for
718 * user space tracing.
719 */
720 new_stream->hangup_flush_done = 0;
721
722 health_code_update();
723
724 pthread_mutex_lock(&new_stream->lock);
725 if (ctx->on_recv_stream) {
726 ret = ctx->on_recv_stream(new_stream);
727 if (ret < 0) {
728 pthread_mutex_unlock(&new_stream->lock);
729 pthread_mutex_unlock(&channel->lock);
730 consumer_stream_free(new_stream);
731 goto error_add_stream_nosignal;
732 }
733 }
734 health_code_update();
735
736 if (new_stream->metadata_flag) {
737 channel->metadata_stream = new_stream;
738 }
739
740 /* Do not monitor this stream. */
741 if (!channel->monitor) {
742 DBG("Kernel consumer add stream %s in no monitor mode with "
743 "relayd id %" PRIu64, new_stream->name,
744 new_stream->net_seq_idx);
745 cds_list_add(&new_stream->send_node, &channel->streams.head);
746 pthread_mutex_unlock(&new_stream->lock);
747 pthread_mutex_unlock(&channel->lock);
748 goto end_add_stream;
749 }
750
751 /* Send stream to relayd if the stream has an ID. */
752 if (new_stream->net_seq_idx != (uint64_t) -1ULL) {
753 ret = consumer_send_relayd_stream(new_stream,
754 new_stream->chan->pathname);
755 if (ret < 0) {
756 pthread_mutex_unlock(&new_stream->lock);
757 pthread_mutex_unlock(&channel->lock);
758 consumer_stream_free(new_stream);
759 goto error_add_stream_nosignal;
760 }
761
762 /*
763 * If adding an extra stream to an already
764 * existing channel (e.g. cpu hotplug), we need
765 * to send the "streams_sent" command to relayd.
766 */
767 if (channel->streams_sent_to_relayd) {
768 ret = consumer_send_relayd_streams_sent(
769 new_stream->net_seq_idx);
770 if (ret < 0) {
771 pthread_mutex_unlock(&new_stream->lock);
772 pthread_mutex_unlock(&channel->lock);
773 goto error_add_stream_nosignal;
774 }
775 }
776 }
777 pthread_mutex_unlock(&new_stream->lock);
778 pthread_mutex_unlock(&channel->lock);
779
780 /* Get the right pipe where the stream will be sent. */
781 if (new_stream->metadata_flag) {
782 consumer_add_metadata_stream(new_stream);
783 stream_pipe = ctx->consumer_metadata_pipe;
784 } else {
785 consumer_add_data_stream(new_stream);
786 stream_pipe = ctx->consumer_data_pipe;
787 }
788
789 /* Visible to other threads */
790 new_stream->globally_visible = 1;
791
792 health_code_update();
793
794 ret = lttng_pipe_write(stream_pipe, &new_stream, sizeof(new_stream));
795 if (ret < 0) {
796 ERR("Consumer write %s stream to pipe %d",
797 new_stream->metadata_flag ? "metadata" : "data",
798 lttng_pipe_get_writefd(stream_pipe));
799 if (new_stream->metadata_flag) {
800 consumer_del_stream_for_metadata(new_stream);
801 } else {
802 consumer_del_stream_for_data(new_stream);
803 }
804 goto error_add_stream_nosignal;
805 }
806
807 DBG("Kernel consumer ADD_STREAM %s (fd: %d) %s with relayd id %" PRIu64,
808 new_stream->name, fd, new_stream->chan->pathname, new_stream->relayd_stream_id);
809 end_add_stream:
810 break;
811 error_add_stream_nosignal:
812 goto end_nosignal;
813 error_add_stream_fatal:
814 goto error_fatal;
815 }
816 case LTTNG_CONSUMER_STREAMS_SENT:
817 {
818 struct lttng_consumer_channel *channel;
819
820 /*
821 * Get stream's channel reference. Needed when adding the stream to the
822 * global hash table.
823 */
824 channel = consumer_find_channel(msg.u.sent_streams.channel_key);
825 if (!channel) {
826 /*
827 * We could not find the channel. Can happen if cpu hotplug
828 * happens while tearing down.
829 */
830 ERR("Unable to find channel key %" PRIu64,
831 msg.u.sent_streams.channel_key);
832 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
833 }
834
835 health_code_update();
836
837 /*
838 * Send status code to session daemon.
839 */
840 ret = consumer_send_status_msg(sock, ret_code);
841 if (ret < 0 || ret_code != LTTCOMM_CONSUMERD_SUCCESS) {
842 /* Somehow, the session daemon is not responding anymore. */
843 goto error_streams_sent_nosignal;
844 }
845
846 health_code_update();
847
848 /*
849 * We should not send this message if we don't monitor the
850 * streams in this channel.
851 */
852 if (!channel->monitor) {
853 goto end_error_streams_sent;
854 }
855
856 health_code_update();
857 /* Send stream to relayd if the stream has an ID. */
858 if (msg.u.sent_streams.net_seq_idx != (uint64_t) -1ULL) {
859 ret = consumer_send_relayd_streams_sent(
860 msg.u.sent_streams.net_seq_idx);
861 if (ret < 0) {
862 goto error_streams_sent_nosignal;
863 }
864 channel->streams_sent_to_relayd = true;
865 }
866 end_error_streams_sent:
867 break;
868 error_streams_sent_nosignal:
869 goto end_nosignal;
870 }
871 case LTTNG_CONSUMER_UPDATE_STREAM:
872 {
873 rcu_read_unlock();
874 return -ENOSYS;
875 }
876 case LTTNG_CONSUMER_DESTROY_RELAYD:
877 {
878 uint64_t index = msg.u.destroy_relayd.net_seq_idx;
879 struct consumer_relayd_sock_pair *relayd;
880
881 DBG("Kernel consumer destroying relayd %" PRIu64, index);
882
883 /* Get relayd reference if exists. */
884 relayd = consumer_find_relayd(index);
885 if (relayd == NULL) {
886 DBG("Unable to find relayd %" PRIu64, index);
887 ret_code = LTTCOMM_CONSUMERD_RELAYD_FAIL;
888 }
889
890 /*
891 * Each relayd socket pair has a refcount of stream attached to it
892 * which tells if the relayd is still active or not depending on the
893 * refcount value.
894 *
895 * This will set the destroy flag of the relayd object and destroy it
896 * if the refcount reaches zero when called.
897 *
898 * The destroy can happen either here or when a stream fd hangs up.
899 */
900 if (relayd) {
901 consumer_flag_relayd_for_destroy(relayd);
902 }
903
904 health_code_update();
905
906 ret = consumer_send_status_msg(sock, ret_code);
907 if (ret < 0) {
908 /* Somehow, the session daemon is not responding anymore. */
909 goto error_fatal;
910 }
911
912 goto end_nosignal;
913 }
914 case LTTNG_CONSUMER_DATA_PENDING:
915 {
916 int32_t ret;
917 uint64_t id = msg.u.data_pending.session_id;
918
919 DBG("Kernel consumer data pending command for id %" PRIu64, id);
920
921 ret = consumer_data_pending(id);
922
923 health_code_update();
924
925 /* Send back returned value to session daemon */
926 ret = lttcomm_send_unix_sock(sock, &ret, sizeof(ret));
927 if (ret < 0) {
928 PERROR("send data pending ret code");
929 goto error_fatal;
930 }
931
932 /*
933 * No need to send back a status message since the data pending
934 * returned value is the response.
935 */
936 break;
937 }
938 case LTTNG_CONSUMER_SNAPSHOT_CHANNEL:
939 {
940 struct lttng_consumer_channel *channel;
941 uint64_t key = msg.u.snapshot_channel.key;
942
943 channel = consumer_find_channel(key);
944 if (!channel) {
945 ERR("Channel %" PRIu64 " not found", key);
946 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
947 } else {
948 if (msg.u.snapshot_channel.metadata == 1) {
949 ret = lttng_kconsumer_snapshot_metadata(channel, key,
950 msg.u.snapshot_channel.pathname,
951 msg.u.snapshot_channel.relayd_id, ctx);
952 if (ret < 0) {
953 ERR("Snapshot metadata failed");
954 ret_code = LTTCOMM_CONSUMERD_SNAPSHOT_FAILED;
955 }
956 } else {
957 ret = lttng_kconsumer_snapshot_channel(channel, key,
958 msg.u.snapshot_channel.pathname,
959 msg.u.snapshot_channel.relayd_id,
960 msg.u.snapshot_channel.nb_packets_per_stream,
961 ctx);
962 if (ret < 0) {
963 ERR("Snapshot channel failed");
964 ret_code = LTTCOMM_CONSUMERD_SNAPSHOT_FAILED;
965 }
966 }
967 }
968 health_code_update();
969
970 ret = consumer_send_status_msg(sock, ret_code);
971 if (ret < 0) {
972 /* Somehow, the session daemon is not responding anymore. */
973 goto end_nosignal;
974 }
975 break;
976 }
977 case LTTNG_CONSUMER_DESTROY_CHANNEL:
978 {
979 uint64_t key = msg.u.destroy_channel.key;
980 struct lttng_consumer_channel *channel;
981
982 channel = consumer_find_channel(key);
983 if (!channel) {
984 ERR("Kernel consumer destroy channel %" PRIu64 " not found", key);
985 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
986 }
987
988 health_code_update();
989
990 ret = consumer_send_status_msg(sock, ret_code);
991 if (ret < 0) {
992 /* Somehow, the session daemon is not responding anymore. */
993 goto end_destroy_channel;
994 }
995
996 health_code_update();
997
998 /* Stop right now if no channel was found. */
999 if (!channel) {
1000 goto end_destroy_channel;
1001 }
1002
1003 /*
1004 * This command should ONLY be issued for channel with streams set in
1005 * no monitor mode.
1006 */
1007 assert(!channel->monitor);
1008
1009 /*
1010 * The refcount should ALWAYS be 0 in the case of a channel in no
1011 * monitor mode.
1012 */
1013 assert(!uatomic_sub_return(&channel->refcount, 1));
1014
1015 consumer_del_channel(channel);
1016 end_destroy_channel:
1017 goto end_nosignal;
1018 }
1019 case LTTNG_CONSUMER_DISCARDED_EVENTS:
1020 {
1021 ssize_t ret;
1022 uint64_t count;
1023 struct lttng_consumer_channel *channel;
1024 uint64_t id = msg.u.discarded_events.session_id;
1025 uint64_t key = msg.u.discarded_events.channel_key;
1026
1027 DBG("Kernel consumer discarded events command for session id %"
1028 PRIu64 ", channel key %" PRIu64, id, key);
1029
1030 channel = consumer_find_channel(key);
1031 if (!channel) {
1032 ERR("Kernel consumer discarded events channel %"
1033 PRIu64 " not found", key);
1034 count = 0;
1035 } else {
1036 count = channel->discarded_events;
1037 }
1038
1039 health_code_update();
1040
1041 /* Send back returned value to session daemon */
1042 ret = lttcomm_send_unix_sock(sock, &count, sizeof(count));
1043 if (ret < 0) {
1044 PERROR("send discarded events");
1045 goto error_fatal;
1046 }
1047
1048 break;
1049 }
1050 case LTTNG_CONSUMER_LOST_PACKETS:
1051 {
1052 ssize_t ret;
1053 uint64_t count;
1054 struct lttng_consumer_channel *channel;
1055 uint64_t id = msg.u.lost_packets.session_id;
1056 uint64_t key = msg.u.lost_packets.channel_key;
1057
1058 DBG("Kernel consumer lost packets command for session id %"
1059 PRIu64 ", channel key %" PRIu64, id, key);
1060
1061 channel = consumer_find_channel(key);
1062 if (!channel) {
1063 ERR("Kernel consumer lost packets channel %"
1064 PRIu64 " not found", key);
1065 count = 0;
1066 } else {
1067 count = channel->lost_packets;
1068 }
1069
1070 health_code_update();
1071
1072 /* Send back returned value to session daemon */
1073 ret = lttcomm_send_unix_sock(sock, &count, sizeof(count));
1074 if (ret < 0) {
1075 PERROR("send lost packets");
1076 goto error_fatal;
1077 }
1078
1079 break;
1080 }
1081 case LTTNG_CONSUMER_SET_CHANNEL_MONITOR_PIPE:
1082 {
1083 int channel_monitor_pipe;
1084
1085 ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1086 /* Successfully received the command's type. */
1087 ret = consumer_send_status_msg(sock, ret_code);
1088 if (ret < 0) {
1089 goto error_fatal;
1090 }
1091
1092 ret = lttcomm_recv_fds_unix_sock(sock, &channel_monitor_pipe,
1093 1);
1094 if (ret != sizeof(channel_monitor_pipe)) {
1095 ERR("Failed to receive channel monitor pipe");
1096 goto error_fatal;
1097 }
1098
1099 DBG("Received channel monitor pipe (%d)", channel_monitor_pipe);
1100 ret = consumer_timer_thread_set_channel_monitor_pipe(
1101 channel_monitor_pipe);
1102 if (!ret) {
1103 int flags;
1104
1105 ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1106 /* Set the pipe as non-blocking. */
1107 ret = fcntl(channel_monitor_pipe, F_GETFL, 0);
1108 if (ret == -1) {
1109 PERROR("fcntl get flags of the channel monitoring pipe");
1110 goto error_fatal;
1111 }
1112 flags = ret;
1113
1114 ret = fcntl(channel_monitor_pipe, F_SETFL,
1115 flags | O_NONBLOCK);
1116 if (ret == -1) {
1117 PERROR("fcntl set O_NONBLOCK flag of the channel monitoring pipe");
1118 goto error_fatal;
1119 }
1120 DBG("Channel monitor pipe set as non-blocking");
1121 } else {
1122 ret_code = LTTCOMM_CONSUMERD_ALREADY_SET;
1123 }
1124 ret = consumer_send_status_msg(sock, ret_code);
1125 if (ret < 0) {
1126 goto error_fatal;
1127 }
1128 break;
1129 }
1130 case LTTNG_CONSUMER_ROTATE_CHANNEL:
1131 {
1132 struct lttng_consumer_channel *channel;
1133 uint64_t key = msg.u.rotate_channel.key;
1134
1135 DBG("Consumer rotate channel %" PRIu64, key);
1136
1137 channel = consumer_find_channel(key);
1138 if (!channel) {
1139 ERR("Channel %" PRIu64 " not found", key);
1140 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
1141 } else {
1142 /*
1143 * Sample the rotate position of all the streams in this channel.
1144 */
1145 ret = lttng_consumer_rotate_channel(channel, key,
1146 msg.u.rotate_channel.relayd_id,
1147 msg.u.rotate_channel.metadata,
1148 ctx);
1149 if (ret < 0) {
1150 ERR("Rotate channel failed");
1151 ret_code = LTTCOMM_CONSUMERD_ROTATION_FAIL;
1152 }
1153
1154 health_code_update();
1155 }
1156 ret = consumer_send_status_msg(sock, ret_code);
1157 if (ret < 0) {
1158 /* Somehow, the session daemon is not responding anymore. */
1159 goto error_rotate_channel;
1160 }
1161 if (channel) {
1162 /* Rotate the streams that are ready right now. */
1163 ret = lttng_consumer_rotate_ready_streams(
1164 channel, key, ctx);
1165 if (ret < 0) {
1166 ERR("Rotate ready streams failed");
1167 }
1168 }
1169 break;
1170 error_rotate_channel:
1171 goto end_nosignal;
1172 }
1173 case LTTNG_CONSUMER_CLEAR_CHANNEL:
1174 {
1175 struct lttng_consumer_channel *channel;
1176 uint64_t key = msg.u.clear_channel.key;
1177
1178 channel = consumer_find_channel(key);
1179 if (!channel) {
1180 DBG("Channel %" PRIu64 " not found", key);
1181 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
1182 } else {
1183 ret = lttng_consumer_clear_channel(channel);
1184 if (ret) {
1185 ERR("Clear channel failed");
1186 ret_code = ret;
1187 }
1188
1189 health_code_update();
1190 }
1191 ret = consumer_send_status_msg(sock, ret_code);
1192 if (ret < 0) {
1193 /* Somehow, the session daemon is not responding anymore. */
1194 goto end_nosignal;
1195 }
1196
1197 break;
1198 }
1199 case LTTNG_CONSUMER_INIT:
1200 {
1201 ret_code = lttng_consumer_init_command(ctx,
1202 msg.u.init.sessiond_uuid);
1203 health_code_update();
1204 ret = consumer_send_status_msg(sock, ret_code);
1205 if (ret < 0) {
1206 /* Somehow, the session daemon is not responding anymore. */
1207 goto end_nosignal;
1208 }
1209 break;
1210 }
1211 case LTTNG_CONSUMER_CREATE_TRACE_CHUNK:
1212 {
1213 const struct lttng_credentials credentials = {
1214 .uid = msg.u.create_trace_chunk.credentials.value.uid,
1215 .gid = msg.u.create_trace_chunk.credentials.value.gid,
1216 };
1217 const bool is_local_trace =
1218 !msg.u.create_trace_chunk.relayd_id.is_set;
1219 const uint64_t relayd_id =
1220 msg.u.create_trace_chunk.relayd_id.value;
1221 const char *chunk_override_name =
1222 *msg.u.create_trace_chunk.override_name ?
1223 msg.u.create_trace_chunk.override_name :
1224 NULL;
1225 struct lttng_directory_handle *chunk_directory_handle = NULL;
1226
1227 /*
1228 * The session daemon will only provide a chunk directory file
1229 * descriptor for local traces.
1230 */
1231 if (is_local_trace) {
1232 int chunk_dirfd;
1233
1234 /* Acnowledge the reception of the command. */
1235 ret = consumer_send_status_msg(sock,
1236 LTTCOMM_CONSUMERD_SUCCESS);
1237 if (ret < 0) {
1238 /* Somehow, the session daemon is not responding anymore. */
1239 goto end_nosignal;
1240 }
1241
1242 ret = lttcomm_recv_fds_unix_sock(sock, &chunk_dirfd, 1);
1243 if (ret != sizeof(chunk_dirfd)) {
1244 ERR("Failed to receive trace chunk directory file descriptor");
1245 goto error_fatal;
1246 }
1247
1248 DBG("Received trace chunk directory fd (%d)",
1249 chunk_dirfd);
1250 chunk_directory_handle = lttng_directory_handle_create_from_dirfd(
1251 chunk_dirfd);
1252 if (!chunk_directory_handle) {
1253 ERR("Failed to initialize chunk directory handle from directory file descriptor");
1254 if (close(chunk_dirfd)) {
1255 PERROR("Failed to close chunk directory file descriptor");
1256 }
1257 goto error_fatal;
1258 }
1259 }
1260
1261 ret_code = lttng_consumer_create_trace_chunk(
1262 !is_local_trace ? &relayd_id : NULL,
1263 msg.u.create_trace_chunk.session_id,
1264 msg.u.create_trace_chunk.chunk_id,
1265 (time_t) msg.u.create_trace_chunk
1266 .creation_timestamp,
1267 chunk_override_name,
1268 msg.u.create_trace_chunk.credentials.is_set ?
1269 &credentials :
1270 NULL,
1271 chunk_directory_handle);
1272 lttng_directory_handle_put(chunk_directory_handle);
1273 goto end_msg_sessiond;
1274 }
1275 case LTTNG_CONSUMER_CLOSE_TRACE_CHUNK:
1276 {
1277 enum lttng_trace_chunk_command_type close_command =
1278 msg.u.close_trace_chunk.close_command.value;
1279 const uint64_t relayd_id =
1280 msg.u.close_trace_chunk.relayd_id.value;
1281 struct lttcomm_consumer_close_trace_chunk_reply reply;
1282 char path[LTTNG_PATH_MAX];
1283
1284 ret_code = lttng_consumer_close_trace_chunk(
1285 msg.u.close_trace_chunk.relayd_id.is_set ?
1286 &relayd_id :
1287 NULL,
1288 msg.u.close_trace_chunk.session_id,
1289 msg.u.close_trace_chunk.chunk_id,
1290 (time_t) msg.u.close_trace_chunk.close_timestamp,
1291 msg.u.close_trace_chunk.close_command.is_set ?
1292 &close_command :
1293 NULL, path);
1294 reply.ret_code = ret_code;
1295 reply.path_length = strlen(path) + 1;
1296 ret = lttcomm_send_unix_sock(sock, &reply, sizeof(reply));
1297 if (ret != sizeof(reply)) {
1298 goto error_fatal;
1299 }
1300 ret = lttcomm_send_unix_sock(sock, path, reply.path_length);
1301 if (ret != reply.path_length) {
1302 goto error_fatal;
1303 }
1304 goto end_nosignal;
1305 }
1306 case LTTNG_CONSUMER_TRACE_CHUNK_EXISTS:
1307 {
1308 const uint64_t relayd_id =
1309 msg.u.trace_chunk_exists.relayd_id.value;
1310
1311 ret_code = lttng_consumer_trace_chunk_exists(
1312 msg.u.trace_chunk_exists.relayd_id.is_set ?
1313 &relayd_id : NULL,
1314 msg.u.trace_chunk_exists.session_id,
1315 msg.u.trace_chunk_exists.chunk_id);
1316 goto end_msg_sessiond;
1317 }
1318 case LTTNG_CONSUMER_OPEN_CHANNEL_PACKETS:
1319 {
1320 const uint64_t key = msg.u.open_channel_packets.key;
1321 struct lttng_consumer_channel *channel =
1322 consumer_find_channel(key);
1323
1324 if (channel) {
1325 pthread_mutex_lock(&channel->lock);
1326 ret_code = lttng_consumer_open_channel_packets(channel);
1327 pthread_mutex_unlock(&channel->lock);
1328 } else {
1329 WARN("Channel %" PRIu64 " not found", key);
1330 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
1331 }
1332
1333 health_code_update();
1334 goto end_msg_sessiond;
1335 }
1336 default:
1337 goto end_nosignal;
1338 }
1339
1340 end_nosignal:
1341 /*
1342 * Return 1 to indicate success since the 0 value can be a socket
1343 * shutdown during the recv() or send() call.
1344 */
1345 ret = 1;
1346 goto end;
1347 error_fatal:
1348 /* This will issue a consumer stop. */
1349 ret = -1;
1350 goto end;
1351 end_msg_sessiond:
1352 /*
1353 * The returned value here is not useful since either way we'll return 1 to
1354 * the caller because the session daemon socket management is done
1355 * elsewhere. Returning a negative code or 0 will shutdown the consumer.
1356 */
1357 ret = consumer_send_status_msg(sock, ret_code);
1358 if (ret < 0) {
1359 goto error_fatal;
1360 }
1361 ret = 1;
1362 end:
1363 health_code_update();
1364 rcu_read_unlock();
1365 return ret;
1366 }
1367
1368 /*
1369 * Sync metadata meaning request them to the session daemon and snapshot to the
1370 * metadata thread can consumer them.
1371 *
1372 * Metadata stream lock MUST be acquired.
1373 */
1374 enum sync_metadata_status lttng_kconsumer_sync_metadata(
1375 struct lttng_consumer_stream *metadata)
1376 {
1377 int ret;
1378 enum sync_metadata_status status;
1379
1380 assert(metadata);
1381
1382 ret = kernctl_buffer_flush(metadata->wait_fd);
1383 if (ret < 0) {
1384 ERR("Failed to flush kernel stream");
1385 status = SYNC_METADATA_STATUS_ERROR;
1386 goto end;
1387 }
1388
1389 ret = kernctl_snapshot(metadata->wait_fd);
1390 if (ret < 0) {
1391 if (errno == EAGAIN) {
1392 /* No new metadata, exit. */
1393 DBG("Sync metadata, no new kernel metadata");
1394 status = SYNC_METADATA_STATUS_NO_DATA;
1395 } else {
1396 ERR("Sync metadata, taking kernel snapshot failed.");
1397 status = SYNC_METADATA_STATUS_ERROR;
1398 }
1399 } else {
1400 status = SYNC_METADATA_STATUS_NEW_DATA;
1401 }
1402
1403 end:
1404 return status;
1405 }
1406
1407 static
1408 int extract_common_subbuffer_info(struct lttng_consumer_stream *stream,
1409 struct stream_subbuffer *subbuf)
1410 {
1411 int ret;
1412
1413 ret = kernctl_get_subbuf_size(
1414 stream->wait_fd, &subbuf->info.data.subbuf_size);
1415 if (ret) {
1416 goto end;
1417 }
1418
1419 ret = kernctl_get_padded_subbuf_size(
1420 stream->wait_fd, &subbuf->info.data.padded_subbuf_size);
1421 if (ret) {
1422 goto end;
1423 }
1424
1425 end:
1426 return ret;
1427 }
1428
1429 static
1430 int extract_metadata_subbuffer_info(struct lttng_consumer_stream *stream,
1431 struct stream_subbuffer *subbuf)
1432 {
1433 int ret;
1434
1435 ret = extract_common_subbuffer_info(stream, subbuf);
1436 if (ret) {
1437 goto end;
1438 }
1439
1440 ret = kernctl_get_metadata_version(
1441 stream->wait_fd, &subbuf->info.metadata.version);
1442 if (ret) {
1443 goto end;
1444 }
1445
1446 end:
1447 return ret;
1448 }
1449
1450 static
1451 int extract_data_subbuffer_info(struct lttng_consumer_stream *stream,
1452 struct stream_subbuffer *subbuf)
1453 {
1454 int ret;
1455
1456 ret = extract_common_subbuffer_info(stream, subbuf);
1457 if (ret) {
1458 goto end;
1459 }
1460
1461 ret = kernctl_get_packet_size(
1462 stream->wait_fd, &subbuf->info.data.packet_size);
1463 if (ret < 0) {
1464 PERROR("Failed to get sub-buffer packet size");
1465 goto end;
1466 }
1467
1468 ret = kernctl_get_content_size(
1469 stream->wait_fd, &subbuf->info.data.content_size);
1470 if (ret < 0) {
1471 PERROR("Failed to get sub-buffer content size");
1472 goto end;
1473 }
1474
1475 ret = kernctl_get_timestamp_begin(
1476 stream->wait_fd, &subbuf->info.data.timestamp_begin);
1477 if (ret < 0) {
1478 PERROR("Failed to get sub-buffer begin timestamp");
1479 goto end;
1480 }
1481
1482 ret = kernctl_get_timestamp_end(
1483 stream->wait_fd, &subbuf->info.data.timestamp_end);
1484 if (ret < 0) {
1485 PERROR("Failed to get sub-buffer end timestamp");
1486 goto end;
1487 }
1488
1489 ret = kernctl_get_events_discarded(
1490 stream->wait_fd, &subbuf->info.data.events_discarded);
1491 if (ret) {
1492 PERROR("Failed to get sub-buffer events discarded count");
1493 goto end;
1494 }
1495
1496 ret = kernctl_get_sequence_number(stream->wait_fd,
1497 &subbuf->info.data.sequence_number.value);
1498 if (ret) {
1499 /* May not be supported by older LTTng-modules. */
1500 if (ret != -ENOTTY) {
1501 PERROR("Failed to get sub-buffer sequence number");
1502 goto end;
1503 }
1504 } else {
1505 subbuf->info.data.sequence_number.is_set = true;
1506 }
1507
1508 ret = kernctl_get_stream_id(
1509 stream->wait_fd, &subbuf->info.data.stream_id);
1510 if (ret < 0) {
1511 PERROR("Failed to get stream id");
1512 goto end;
1513 }
1514
1515 ret = kernctl_get_instance_id(stream->wait_fd,
1516 &subbuf->info.data.stream_instance_id.value);
1517 if (ret) {
1518 /* May not be supported by older LTTng-modules. */
1519 if (ret != -ENOTTY) {
1520 PERROR("Failed to get stream instance id");
1521 goto end;
1522 }
1523 } else {
1524 subbuf->info.data.stream_instance_id.is_set = true;
1525 }
1526 end:
1527 return ret;
1528 }
1529
1530 static
1531 int get_subbuffer_common(struct lttng_consumer_stream *stream,
1532 struct stream_subbuffer *subbuffer)
1533 {
1534 int ret;
1535
1536 ret = kernctl_get_next_subbuf(stream->wait_fd);
1537 if (ret) {
1538 /*
1539 * The caller only expects -ENODATA when there is no data to
1540 * read, but the kernel tracer returns -EAGAIN when there is
1541 * currently no data for a non-finalized stream, and -ENODATA
1542 * when there is no data for a finalized stream. Those can be
1543 * combined into a -ENODATA return value.
1544 */
1545 if (ret == -EAGAIN) {
1546 ret = -ENODATA;
1547 }
1548
1549 goto end;
1550 }
1551
1552 ret = stream->read_subbuffer_ops.extract_subbuffer_info(
1553 stream, subbuffer);
1554 end:
1555 return ret;
1556 }
1557
1558 static
1559 int get_next_subbuffer_splice(struct lttng_consumer_stream *stream,
1560 struct stream_subbuffer *subbuffer)
1561 {
1562 int ret;
1563
1564 ret = get_subbuffer_common(stream, subbuffer);
1565 if (ret) {
1566 goto end;
1567 }
1568
1569 subbuffer->buffer.fd = stream->wait_fd;
1570 end:
1571 return ret;
1572 }
1573
1574 static
1575 int get_next_subbuffer_mmap(struct lttng_consumer_stream *stream,
1576 struct stream_subbuffer *subbuffer)
1577 {
1578 int ret;
1579 const char *addr;
1580
1581 ret = get_subbuffer_common(stream, subbuffer);
1582 if (ret) {
1583 goto end;
1584 }
1585
1586 ret = get_current_subbuf_addr(stream, &addr);
1587 if (ret) {
1588 goto end;
1589 }
1590
1591 subbuffer->buffer.buffer = lttng_buffer_view_init(
1592 addr, 0, subbuffer->info.data.padded_subbuf_size);
1593 end:
1594 return ret;
1595 }
1596
1597 static
1598 int get_next_subbuffer_metadata_check(struct lttng_consumer_stream *stream,
1599 struct stream_subbuffer *subbuffer)
1600 {
1601 int ret;
1602 const char *addr;
1603 bool coherent;
1604
1605 ret = kernctl_get_next_subbuf_metadata_check(stream->wait_fd,
1606 &coherent);
1607 if (ret) {
1608 goto end;
1609 }
1610
1611 ret = stream->read_subbuffer_ops.extract_subbuffer_info(
1612 stream, subbuffer);
1613 if (ret) {
1614 goto end;
1615 }
1616
1617 LTTNG_OPTIONAL_SET(&subbuffer->info.metadata.coherent, coherent);
1618
1619 ret = get_current_subbuf_addr(stream, &addr);
1620 if (ret) {
1621 goto end;
1622 }
1623
1624 subbuffer->buffer.buffer = lttng_buffer_view_init(
1625 addr, 0, subbuffer->info.data.padded_subbuf_size);
1626 DBG("Got metadata packet with padded_subbuf_size = %lu, coherent = %s",
1627 subbuffer->info.metadata.padded_subbuf_size,
1628 coherent ? "true" : "false");
1629 end:
1630 /*
1631 * The caller only expects -ENODATA when there is no data to read, but
1632 * the kernel tracer returns -EAGAIN when there is currently no data
1633 * for a non-finalized stream, and -ENODATA when there is no data for a
1634 * finalized stream. Those can be combined into a -ENODATA return value.
1635 */
1636 if (ret == -EAGAIN) {
1637 ret = -ENODATA;
1638 }
1639
1640 return ret;
1641 }
1642
1643 static
1644 int put_next_subbuffer(struct lttng_consumer_stream *stream,
1645 struct stream_subbuffer *subbuffer)
1646 {
1647 const int ret = kernctl_put_next_subbuf(stream->wait_fd);
1648
1649 if (ret) {
1650 if (ret == -EFAULT) {
1651 PERROR("Error in unreserving sub buffer");
1652 } else if (ret == -EIO) {
1653 /* Should never happen with newer LTTng versions */
1654 PERROR("Reader has been pushed by the writer, last sub-buffer corrupted");
1655 }
1656 }
1657
1658 return ret;
1659 }
1660
1661 static
1662 bool is_get_next_check_metadata_available(int tracer_fd)
1663 {
1664 const int ret = kernctl_get_next_subbuf_metadata_check(tracer_fd, NULL);
1665 const bool available = ret != -ENOTTY;
1666
1667 if (ret == 0) {
1668 /* get succeeded, make sure to put the subbuffer. */
1669 kernctl_put_subbuf(tracer_fd);
1670 }
1671
1672 return available;
1673 }
1674
1675 static
1676 int signal_metadata(struct lttng_consumer_stream *stream,
1677 struct lttng_consumer_local_data *ctx)
1678 {
1679 ASSERT_LOCKED(stream->metadata_rdv_lock);
1680 return pthread_cond_broadcast(&stream->metadata_rdv) ? -errno : 0;
1681 }
1682
1683 static
1684 int lttng_kconsumer_set_stream_ops(
1685 struct lttng_consumer_stream *stream)
1686 {
1687 int ret = 0;
1688
1689 if (stream->metadata_flag && stream->chan->is_live) {
1690 DBG("Attempting to enable metadata bucketization for live consumers");
1691 if (is_get_next_check_metadata_available(stream->wait_fd)) {
1692 DBG("Kernel tracer supports get_next_subbuffer_metadata_check, metadata will be accumulated until a coherent state is reached");
1693 stream->read_subbuffer_ops.get_next_subbuffer =
1694 get_next_subbuffer_metadata_check;
1695 ret = consumer_stream_enable_metadata_bucketization(
1696 stream);
1697 if (ret) {
1698 goto end;
1699 }
1700 } else {
1701 /*
1702 * The kernel tracer version is too old to indicate
1703 * when the metadata stream has reached a "coherent"
1704 * (parseable) point.
1705 *
1706 * This means that a live viewer may see an incoherent
1707 * sequence of metadata and fail to parse it.
1708 */
1709 WARN("Kernel tracer does not support get_next_subbuffer_metadata_check which may cause live clients to fail to parse the metadata stream");
1710 metadata_bucket_destroy(stream->metadata_bucket);
1711 stream->metadata_bucket = NULL;
1712 }
1713
1714 stream->read_subbuffer_ops.on_sleep = signal_metadata;
1715 }
1716
1717 if (!stream->read_subbuffer_ops.get_next_subbuffer) {
1718 if (stream->chan->output == CONSUMER_CHANNEL_MMAP) {
1719 stream->read_subbuffer_ops.get_next_subbuffer =
1720 get_next_subbuffer_mmap;
1721 } else {
1722 stream->read_subbuffer_ops.get_next_subbuffer =
1723 get_next_subbuffer_splice;
1724 }
1725 }
1726
1727 if (stream->metadata_flag) {
1728 stream->read_subbuffer_ops.extract_subbuffer_info =
1729 extract_metadata_subbuffer_info;
1730 } else {
1731 stream->read_subbuffer_ops.extract_subbuffer_info =
1732 extract_data_subbuffer_info;
1733 if (stream->chan->is_live) {
1734 stream->read_subbuffer_ops.send_live_beacon =
1735 consumer_flush_kernel_index;
1736 }
1737 }
1738
1739 stream->read_subbuffer_ops.put_next_subbuffer = put_next_subbuffer;
1740 end:
1741 return ret;
1742 }
1743
1744 int lttng_kconsumer_on_recv_stream(struct lttng_consumer_stream *stream)
1745 {
1746 int ret;
1747
1748 assert(stream);
1749
1750 /*
1751 * Don't create anything if this is set for streaming or if there is
1752 * no current trace chunk on the parent channel.
1753 */
1754 if (stream->net_seq_idx == (uint64_t) -1ULL && stream->chan->monitor &&
1755 stream->chan->trace_chunk) {
1756 ret = consumer_stream_create_output_files(stream, true);
1757 if (ret) {
1758 goto error;
1759 }
1760 }
1761
1762 if (stream->output == LTTNG_EVENT_MMAP) {
1763 /* get the len of the mmap region */
1764 unsigned long mmap_len;
1765
1766 ret = kernctl_get_mmap_len(stream->wait_fd, &mmap_len);
1767 if (ret != 0) {
1768 PERROR("kernctl_get_mmap_len");
1769 goto error_close_fd;
1770 }
1771 stream->mmap_len = (size_t) mmap_len;
1772
1773 stream->mmap_base = mmap(NULL, stream->mmap_len, PROT_READ,
1774 MAP_PRIVATE, stream->wait_fd, 0);
1775 if (stream->mmap_base == MAP_FAILED) {
1776 PERROR("Error mmaping");
1777 ret = -1;
1778 goto error_close_fd;
1779 }
1780 }
1781
1782 ret = lttng_kconsumer_set_stream_ops(stream);
1783 if (ret) {
1784 goto error_close_fd;
1785 }
1786
1787 /* we return 0 to let the library handle the FD internally */
1788 return 0;
1789
1790 error_close_fd:
1791 if (stream->out_fd >= 0) {
1792 int err;
1793
1794 err = close(stream->out_fd);
1795 assert(!err);
1796 stream->out_fd = -1;
1797 }
1798 error:
1799 return ret;
1800 }
1801
1802 /*
1803 * Check if data is still being extracted from the buffers for a specific
1804 * stream. Consumer data lock MUST be acquired before calling this function
1805 * and the stream lock.
1806 *
1807 * Return 1 if the traced data are still getting read else 0 meaning that the
1808 * data is available for trace viewer reading.
1809 */
1810 int lttng_kconsumer_data_pending(struct lttng_consumer_stream *stream)
1811 {
1812 int ret;
1813
1814 assert(stream);
1815
1816 if (stream->endpoint_status != CONSUMER_ENDPOINT_ACTIVE) {
1817 ret = 0;
1818 goto end;
1819 }
1820
1821 ret = kernctl_get_next_subbuf(stream->wait_fd);
1822 if (ret == 0) {
1823 /* There is still data so let's put back this subbuffer. */
1824 ret = kernctl_put_subbuf(stream->wait_fd);
1825 assert(ret == 0);
1826 ret = 1; /* Data is pending */
1827 goto end;
1828 }
1829
1830 /* Data is NOT pending and ready to be read. */
1831 ret = 0;
1832
1833 end:
1834 return ret;
1835 }
LTTng 2.0 tools and control repository
This page took 0.071651 seconds and 3 git commands to generate.