f520736b42b10c580f28a46b09238bdf028da283
[lttng-tools.git] / src / common / kernel-consumer / kernel-consumer.cpp
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 "kernel-consumer.hpp"
12
13 #include <common/buffer-view.hpp>
14 #include <common/common.hpp>
15 #include <common/compat/endian.hpp>
16 #include <common/compat/fcntl.hpp>
17 #include <common/consumer/consumer-stream.hpp>
18 #include <common/consumer/consumer-timer.hpp>
19 #include <common/consumer/consumer.hpp>
20 #include <common/consumer/metadata-bucket.hpp>
21 #include <common/index/index.hpp>
22 #include <common/kernel-ctl/kernel-ctl.hpp>
23 #include <common/optional.hpp>
24 #include <common/pipe.hpp>
25 #include <common/relayd/relayd.hpp>
26 #include <common/sessiond-comm/relayd.hpp>
27 #include <common/sessiond-comm/sessiond-comm.hpp>
28 #include <common/utils.hpp>
29
30 #include <bin/lttng-consumerd/health-consumerd.hpp>
31 #include <inttypes.h>
32 #include <poll.h>
33 #include <pthread.h>
34 #include <stdint.h>
35 #include <stdlib.h>
36 #include <string.h>
37 #include <sys/mman.h>
38 #include <sys/socket.h>
39 #include <sys/stat.h>
40 #include <sys/types.h>
41 #include <unistd.h>
42
43 extern struct lttng_consumer_global_data the_consumer_data;
44 extern int consumer_poll_timeout;
45
46 /*
47 * Take a snapshot for a specific fd
48 *
49 * Returns 0 on success, < 0 on error
50 */
51 int lttng_kconsumer_take_snapshot(struct lttng_consumer_stream *stream)
52 {
53 int ret = 0;
54 int infd = stream->wait_fd;
55
56 ret = kernctl_snapshot(infd);
57 /*
58 * -EAGAIN is not an error, it just means that there is no data to
59 * be read.
60 */
61 if (ret != 0 && ret != -EAGAIN) {
62 PERROR("Getting sub-buffer snapshot.");
63 }
64
65 return ret;
66 }
67
68 /*
69 * Sample consumed and produced positions for a specific fd.
70 *
71 * Returns 0 on success, < 0 on error.
72 */
73 int lttng_kconsumer_sample_snapshot_positions(struct lttng_consumer_stream *stream)
74 {
75 LTTNG_ASSERT(stream);
76
77 return kernctl_snapshot_sample_positions(stream->wait_fd);
78 }
79
80 /*
81 * Get the produced position
82 *
83 * Returns 0 on success, < 0 on error
84 */
85 int lttng_kconsumer_get_produced_snapshot(struct lttng_consumer_stream *stream, unsigned long *pos)
86 {
87 int ret;
88 int infd = stream->wait_fd;
89
90 ret = kernctl_snapshot_get_produced(infd, pos);
91 if (ret != 0) {
92 PERROR("kernctl_snapshot_get_produced");
93 }
94
95 return ret;
96 }
97
98 /*
99 * Get the consumerd position
100 *
101 * Returns 0 on success, < 0 on error
102 */
103 int lttng_kconsumer_get_consumed_snapshot(struct lttng_consumer_stream *stream, unsigned long *pos)
104 {
105 int ret;
106 int infd = stream->wait_fd;
107
108 ret = kernctl_snapshot_get_consumed(infd, pos);
109 if (ret != 0) {
110 PERROR("kernctl_snapshot_get_consumed");
111 }
112
113 return ret;
114 }
115
116 static int get_current_subbuf_addr(struct lttng_consumer_stream *stream, const char **addr)
117 {
118 int ret;
119 unsigned long mmap_offset;
120 const char *mmap_base = (const char *) stream->mmap_base;
121
122 ret = kernctl_get_mmap_read_offset(stream->wait_fd, &mmap_offset);
123 if (ret < 0) {
124 PERROR("Failed to get mmap read offset");
125 goto error;
126 }
127
128 *addr = mmap_base + mmap_offset;
129 error:
130 return ret;
131 }
132
133 /*
134 * Take a snapshot of all the stream of a channel
135 * RCU read-side lock must be held across this function to ensure existence of
136 * channel.
137 *
138 * Returns 0 on success, < 0 on error
139 */
140 static int lttng_kconsumer_snapshot_channel(struct lttng_consumer_channel *channel,
141 uint64_t key,
142 char *path,
143 uint64_t relayd_id,
144 uint64_t nb_packets_per_stream)
145 {
146 int ret;
147 struct lttng_consumer_stream *stream;
148
149 DBG("Kernel consumer snapshot channel %" PRIu64, key);
150
151 /* Prevent channel modifications while we perform the snapshot.*/
152 pthread_mutex_lock(&channel->lock);
153
154 rcu_read_lock();
155
156 /* Splice is not supported yet for channel snapshot. */
157 if (channel->output != CONSUMER_CHANNEL_MMAP) {
158 ERR("Unsupported output type for channel \"%s\": mmap output is required to record a snapshot",
159 channel->name);
160 ret = -1;
161 goto end;
162 }
163
164 cds_list_for_each_entry (stream, &channel->streams.head, send_node) {
165 unsigned long consumed_pos, produced_pos;
166
167 health_code_update();
168
169 /*
170 * Lock stream because we are about to change its state.
171 */
172 pthread_mutex_lock(&stream->lock);
173
174 LTTNG_ASSERT(channel->trace_chunk);
175 if (!lttng_trace_chunk_get(channel->trace_chunk)) {
176 /*
177 * Can't happen barring an internal error as the channel
178 * holds a reference to the trace chunk.
179 */
180 ERR("Failed to acquire reference to channel's trace chunk");
181 ret = -1;
182 goto end_unlock;
183 }
184 LTTNG_ASSERT(!stream->trace_chunk);
185 stream->trace_chunk = channel->trace_chunk;
186
187 /*
188 * Assign the received relayd ID so we can use it for streaming. The streams
189 * are not visible to anyone so this is OK to change it.
190 */
191 stream->net_seq_idx = relayd_id;
192 channel->relayd_id = relayd_id;
193 if (relayd_id != (uint64_t) -1ULL) {
194 ret = consumer_send_relayd_stream(stream, path);
195 if (ret < 0) {
196 ERR("sending stream to relayd");
197 goto error_close_stream_output;
198 }
199 } else {
200 ret = consumer_stream_create_output_files(stream, false);
201 if (ret < 0) {
202 goto error_close_stream_output;
203 }
204 DBG("Kernel consumer snapshot stream (%" PRIu64 ")", stream->key);
205 }
206
207 ret = kernctl_buffer_flush_empty(stream->wait_fd);
208 if (ret < 0) {
209 /*
210 * Doing a buffer flush which does not take into
211 * account empty packets. This is not perfect
212 * for stream intersection, but required as a
213 * fall-back when "flush_empty" is not
214 * implemented by lttng-modules.
215 */
216 ret = kernctl_buffer_flush(stream->wait_fd);
217 if (ret < 0) {
218 ERR("Failed to flush kernel stream");
219 goto error_close_stream_output;
220 }
221 goto end_unlock;
222 }
223
224 ret = lttng_kconsumer_take_snapshot(stream);
225 if (ret < 0) {
226 ERR("Taking kernel snapshot");
227 goto error_close_stream_output;
228 }
229
230 ret = lttng_kconsumer_get_produced_snapshot(stream, &produced_pos);
231 if (ret < 0) {
232 ERR("Produced kernel snapshot position");
233 goto error_close_stream_output;
234 }
235
236 ret = lttng_kconsumer_get_consumed_snapshot(stream, &consumed_pos);
237 if (ret < 0) {
238 ERR("Consumerd kernel snapshot position");
239 goto error_close_stream_output;
240 }
241
242 consumed_pos = consumer_get_consume_start_pos(
243 consumed_pos, produced_pos, nb_packets_per_stream, stream->max_sb_size);
244
245 while ((long) (consumed_pos - produced_pos) < 0) {
246 ssize_t read_len;
247 unsigned long len, padded_len;
248 const char *subbuf_addr;
249 struct lttng_buffer_view subbuf_view;
250
251 health_code_update();
252 DBG("Kernel consumer taking snapshot at pos %lu", consumed_pos);
253
254 ret = kernctl_get_subbuf(stream->wait_fd, &consumed_pos);
255 if (ret < 0) {
256 if (ret != -EAGAIN) {
257 PERROR("kernctl_get_subbuf snapshot");
258 goto error_close_stream_output;
259 }
260 DBG("Kernel consumer get subbuf failed. Skipping it.");
261 consumed_pos += stream->max_sb_size;
262 stream->chan->lost_packets++;
263 continue;
264 }
265
266 ret = kernctl_get_subbuf_size(stream->wait_fd, &len);
267 if (ret < 0) {
268 ERR("Snapshot kernctl_get_subbuf_size");
269 goto error_put_subbuf;
270 }
271
272 ret = kernctl_get_padded_subbuf_size(stream->wait_fd, &padded_len);
273 if (ret < 0) {
274 ERR("Snapshot kernctl_get_padded_subbuf_size");
275 goto error_put_subbuf;
276 }
277
278 ret = get_current_subbuf_addr(stream, &subbuf_addr);
279 if (ret) {
280 goto error_put_subbuf;
281 }
282
283 subbuf_view = lttng_buffer_view_init(subbuf_addr, 0, padded_len);
284 read_len = lttng_consumer_on_read_subbuffer_mmap(
285 stream, &subbuf_view, padded_len - len);
286 /*
287 * We write the padded len in local tracefiles but the data len
288 * when using a relay. Display the error but continue processing
289 * to try to release the subbuffer.
290 */
291 if (relayd_id != (uint64_t) -1ULL) {
292 if (read_len != len) {
293 ERR("Error sending to the relay (ret: %zd != len: %lu)",
294 read_len,
295 len);
296 }
297 } else {
298 if (read_len != padded_len) {
299 ERR("Error writing to tracefile (ret: %zd != len: %lu)",
300 read_len,
301 padded_len);
302 }
303 }
304
305 ret = kernctl_put_subbuf(stream->wait_fd);
306 if (ret < 0) {
307 ERR("Snapshot kernctl_put_subbuf");
308 goto error_close_stream_output;
309 }
310 consumed_pos += stream->max_sb_size;
311 }
312
313 consumer_stream_close_output(stream);
314 pthread_mutex_unlock(&stream->lock);
315 }
316
317 /* All good! */
318 ret = 0;
319 goto end;
320
321 error_put_subbuf:
322 ret = kernctl_put_subbuf(stream->wait_fd);
323 if (ret < 0) {
324 ERR("Snapshot kernctl_put_subbuf error path");
325 }
326 error_close_stream_output:
327 consumer_stream_close_output(stream);
328 end_unlock:
329 pthread_mutex_unlock(&stream->lock);
330 end:
331 rcu_read_unlock();
332 pthread_mutex_unlock(&channel->lock);
333 return ret;
334 }
335
336 /*
337 * Read the whole metadata available for a snapshot.
338 * RCU read-side lock must be held across this function to ensure existence of
339 * metadata_channel.
340 *
341 * Returns 0 on success, < 0 on error
342 */
343 static int lttng_kconsumer_snapshot_metadata(struct lttng_consumer_channel *metadata_channel,
344 uint64_t key,
345 char *path,
346 uint64_t relayd_id,
347 struct lttng_consumer_local_data *ctx)
348 {
349 int ret, use_relayd = 0;
350 ssize_t ret_read;
351 struct lttng_consumer_stream *metadata_stream;
352
353 LTTNG_ASSERT(ctx);
354
355 DBG("Kernel consumer snapshot metadata with key %" PRIu64 " at path %s", key, path);
356
357 rcu_read_lock();
358
359 metadata_stream = metadata_channel->metadata_stream;
360 LTTNG_ASSERT(metadata_stream);
361
362 metadata_stream->read_subbuffer_ops.lock(metadata_stream);
363 LTTNG_ASSERT(metadata_channel->trace_chunk);
364 LTTNG_ASSERT(metadata_stream->trace_chunk);
365
366 /* Flag once that we have a valid relayd for the stream. */
367 if (relayd_id != (uint64_t) -1ULL) {
368 use_relayd = 1;
369 }
370
371 if (use_relayd) {
372 ret = consumer_send_relayd_stream(metadata_stream, path);
373 if (ret < 0) {
374 goto error_snapshot;
375 }
376 } else {
377 ret = consumer_stream_create_output_files(metadata_stream, false);
378 if (ret < 0) {
379 goto error_snapshot;
380 }
381 }
382
383 do {
384 health_code_update();
385
386 ret_read = lttng_consumer_read_subbuffer(metadata_stream, ctx, true);
387 if (ret_read < 0) {
388 ERR("Kernel snapshot reading metadata subbuffer (ret: %zd)", ret_read);
389 ret = ret_read;
390 goto error_snapshot;
391 }
392 } while (ret_read > 0);
393
394 if (use_relayd) {
395 close_relayd_stream(metadata_stream);
396 metadata_stream->net_seq_idx = (uint64_t) -1ULL;
397 } else {
398 if (metadata_stream->out_fd >= 0) {
399 ret = close(metadata_stream->out_fd);
400 if (ret < 0) {
401 PERROR("Kernel consumer snapshot metadata close out_fd");
402 /*
403 * Don't go on error here since the snapshot was successful at this
404 * point but somehow the close failed.
405 */
406 }
407 metadata_stream->out_fd = -1;
408 lttng_trace_chunk_put(metadata_stream->trace_chunk);
409 metadata_stream->trace_chunk = nullptr;
410 }
411 }
412
413 ret = 0;
414 error_snapshot:
415 metadata_stream->read_subbuffer_ops.unlock(metadata_stream);
416 consumer_stream_destroy(metadata_stream, nullptr);
417 metadata_channel->metadata_stream = nullptr;
418 rcu_read_unlock();
419 return ret;
420 }
421
422 /*
423 * Receive command from session daemon and process it.
424 *
425 * Return 1 on success else a negative value or 0.
426 */
427 int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
428 int sock,
429 struct pollfd *consumer_sockpoll)
430 {
431 int ret_func;
432 enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS;
433 struct lttcomm_consumer_msg msg;
434
435 health_code_update();
436
437 {
438 ssize_t ret_recv;
439
440 ret_recv = lttcomm_recv_unix_sock(sock, &msg, sizeof(msg));
441 if (ret_recv != sizeof(msg)) {
442 if (ret_recv > 0) {
443 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD);
444 ret_recv = -1;
445 }
446 return ret_recv;
447 }
448 }
449
450 health_code_update();
451
452 /* Deprecated command */
453 LTTNG_ASSERT(msg.cmd_type != LTTNG_CONSUMER_STOP);
454
455 health_code_update();
456
457 /* relayd needs RCU read-side protection */
458 rcu_read_lock();
459
460 switch (msg.cmd_type) {
461 case LTTNG_CONSUMER_ADD_RELAYD_SOCKET:
462 {
463 uint32_t major = msg.u.relayd_sock.major;
464 uint32_t minor = msg.u.relayd_sock.minor;
465 enum lttcomm_sock_proto protocol =
466 (enum lttcomm_sock_proto) msg.u.relayd_sock.relayd_socket_protocol;
467
468 /* Session daemon status message are handled in the following call. */
469 consumer_add_relayd_socket(msg.u.relayd_sock.net_index,
470 msg.u.relayd_sock.type,
471 ctx,
472 sock,
473 consumer_sockpoll,
474 msg.u.relayd_sock.session_id,
475 msg.u.relayd_sock.relayd_session_id,
476 major,
477 minor,
478 protocol);
479 goto end_nosignal;
480 }
481 case LTTNG_CONSUMER_ADD_CHANNEL:
482 {
483 struct lttng_consumer_channel *new_channel;
484 int ret_send_status, ret_add_channel = 0;
485 const uint64_t chunk_id = msg.u.channel.chunk_id.value;
486
487 health_code_update();
488
489 /* First send a status message before receiving the fds. */
490 ret_send_status = consumer_send_status_msg(sock, ret_code);
491 if (ret_send_status < 0) {
492 /* Somehow, the session daemon is not responding anymore. */
493 goto error_fatal;
494 }
495
496 health_code_update();
497
498 DBG("consumer_add_channel %" PRIu64, msg.u.channel.channel_key);
499 new_channel = consumer_allocate_channel(msg.u.channel.channel_key,
500 msg.u.channel.session_id,
501 msg.u.channel.chunk_id.is_set ? &chunk_id :
502 nullptr,
503 msg.u.channel.pathname,
504 msg.u.channel.name,
505 msg.u.channel.relayd_id,
506 msg.u.channel.output,
507 msg.u.channel.tracefile_size,
508 msg.u.channel.tracefile_count,
509 0,
510 msg.u.channel.monitor,
511 msg.u.channel.live_timer_interval,
512 msg.u.channel.is_live,
513 nullptr,
514 nullptr);
515 if (new_channel == nullptr) {
516 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
517 goto end_nosignal;
518 }
519 new_channel->nb_init_stream_left = msg.u.channel.nb_init_streams;
520 switch (msg.u.channel.output) {
521 case LTTNG_EVENT_SPLICE:
522 new_channel->output = CONSUMER_CHANNEL_SPLICE;
523 break;
524 case LTTNG_EVENT_MMAP:
525 new_channel->output = CONSUMER_CHANNEL_MMAP;
526 break;
527 default:
528 ERR("Channel output unknown %d", msg.u.channel.output);
529 goto end_nosignal;
530 }
531
532 /* Translate and save channel type. */
533 switch (msg.u.channel.type) {
534 case CONSUMER_CHANNEL_TYPE_DATA:
535 case CONSUMER_CHANNEL_TYPE_METADATA:
536 new_channel->type = (consumer_channel_type) msg.u.channel.type;
537 break;
538 default:
539 abort();
540 goto end_nosignal;
541 };
542
543 health_code_update();
544
545 if (ctx->on_recv_channel != nullptr) {
546 int ret_recv_channel = ctx->on_recv_channel(new_channel);
547 if (ret_recv_channel == 0) {
548 ret_add_channel = consumer_add_channel(new_channel, ctx);
549 } else if (ret_recv_channel < 0) {
550 goto end_nosignal;
551 }
552 } else {
553 ret_add_channel = consumer_add_channel(new_channel, ctx);
554 }
555 if (msg.u.channel.type == CONSUMER_CHANNEL_TYPE_DATA && !ret_add_channel) {
556 int monitor_start_ret;
557
558 DBG("Consumer starting monitor timer");
559 consumer_timer_live_start(new_channel, msg.u.channel.live_timer_interval);
560 monitor_start_ret = consumer_timer_monitor_start(
561 new_channel, msg.u.channel.monitor_timer_interval);
562 if (monitor_start_ret < 0) {
563 ERR("Starting channel monitoring timer failed");
564 goto end_nosignal;
565 }
566 }
567
568 health_code_update();
569
570 /* If we received an error in add_channel, we need to report it. */
571 if (ret_add_channel < 0) {
572 ret_send_status = consumer_send_status_msg(sock, ret_add_channel);
573 if (ret_send_status < 0) {
574 goto error_fatal;
575 }
576 goto end_nosignal;
577 }
578
579 goto end_nosignal;
580 }
581 case LTTNG_CONSUMER_ADD_STREAM:
582 {
583 int fd;
584 struct lttng_pipe *stream_pipe;
585 struct lttng_consumer_stream *new_stream;
586 struct lttng_consumer_channel *channel;
587 int alloc_ret = 0;
588 int ret_send_status, ret_poll, ret_get_max_subbuf_size;
589 ssize_t ret_pipe_write, ret_recv;
590
591 /*
592 * Get stream's channel reference. Needed when adding the stream to the
593 * global hash table.
594 */
595 channel = consumer_find_channel(msg.u.stream.channel_key);
596 if (!channel) {
597 /*
598 * We could not find the channel. Can happen if cpu hotplug
599 * happens while tearing down.
600 */
601 ERR("Unable to find channel key %" PRIu64, msg.u.stream.channel_key);
602 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
603 }
604
605 health_code_update();
606
607 /* First send a status message before receiving the fds. */
608 ret_send_status = consumer_send_status_msg(sock, ret_code);
609 if (ret_send_status < 0) {
610 /* Somehow, the session daemon is not responding anymore. */
611 goto error_add_stream_fatal;
612 }
613
614 health_code_update();
615
616 if (ret_code != LTTCOMM_CONSUMERD_SUCCESS) {
617 /* Channel was not found. */
618 goto error_add_stream_nosignal;
619 }
620
621 /* Blocking call */
622 health_poll_entry();
623 ret_poll = lttng_consumer_poll_socket(consumer_sockpoll);
624 health_poll_exit();
625 if (ret_poll) {
626 goto error_add_stream_fatal;
627 }
628
629 health_code_update();
630
631 /* Get stream file descriptor from socket */
632 ret_recv = lttcomm_recv_fds_unix_sock(sock, &fd, 1);
633 if (ret_recv != sizeof(fd)) {
634 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_FD);
635 ret_func = ret_recv;
636 goto end;
637 }
638
639 health_code_update();
640
641 /*
642 * Send status code to session daemon only if the recv works. If the
643 * above recv() failed, the session daemon is notified through the
644 * error socket and the teardown is eventually done.
645 */
646 ret_send_status = consumer_send_status_msg(sock, ret_code);
647 if (ret_send_status < 0) {
648 /* Somehow, the session daemon is not responding anymore. */
649 goto error_add_stream_nosignal;
650 }
651
652 health_code_update();
653
654 pthread_mutex_lock(&channel->lock);
655 new_stream = consumer_stream_create(channel,
656 channel->key,
657 fd,
658 channel->name,
659 channel->relayd_id,
660 channel->session_id,
661 channel->trace_chunk,
662 msg.u.stream.cpu,
663 &alloc_ret,
664 channel->type,
665 channel->monitor);
666 if (new_stream == nullptr) {
667 switch (alloc_ret) {
668 case -ENOMEM:
669 case -EINVAL:
670 default:
671 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
672 break;
673 }
674 pthread_mutex_unlock(&channel->lock);
675 goto error_add_stream_nosignal;
676 }
677
678 new_stream->wait_fd = fd;
679 ret_get_max_subbuf_size =
680 kernctl_get_max_subbuf_size(new_stream->wait_fd, &new_stream->max_sb_size);
681 if (ret_get_max_subbuf_size < 0) {
682 pthread_mutex_unlock(&channel->lock);
683 ERR("Failed to get kernel maximal subbuffer size");
684 goto error_add_stream_nosignal;
685 }
686
687 consumer_stream_update_channel_attributes(new_stream, channel);
688
689 /*
690 * We've just assigned the channel to the stream so increment the
691 * refcount right now. We don't need to increment the refcount for
692 * streams in no monitor because we handle manually the cleanup of
693 * those. It is very important to make sure there is NO prior
694 * consumer_del_stream() calls or else the refcount will be unbalanced.
695 */
696 if (channel->monitor) {
697 uatomic_inc(&new_stream->chan->refcount);
698 }
699
700 /*
701 * The buffer flush is done on the session daemon side for the kernel
702 * so no need for the stream "hangup_flush_done" variable to be
703 * tracked. This is important for a kernel stream since we don't rely
704 * on the flush state of the stream to read data. It's not the case for
705 * user space tracing.
706 */
707 new_stream->hangup_flush_done = 0;
708
709 health_code_update();
710
711 pthread_mutex_lock(&new_stream->lock);
712 if (ctx->on_recv_stream) {
713 int ret_recv_stream = ctx->on_recv_stream(new_stream);
714 if (ret_recv_stream < 0) {
715 pthread_mutex_unlock(&new_stream->lock);
716 pthread_mutex_unlock(&channel->lock);
717 consumer_stream_free(new_stream);
718 goto error_add_stream_nosignal;
719 }
720 }
721 health_code_update();
722
723 if (new_stream->metadata_flag) {
724 channel->metadata_stream = new_stream;
725 }
726
727 /* Do not monitor this stream. */
728 if (!channel->monitor) {
729 DBG("Kernel consumer add stream %s in no monitor mode with "
730 "relayd id %" PRIu64,
731 new_stream->name,
732 new_stream->net_seq_idx);
733 cds_list_add(&new_stream->send_node, &channel->streams.head);
734 pthread_mutex_unlock(&new_stream->lock);
735 pthread_mutex_unlock(&channel->lock);
736 goto end_add_stream;
737 }
738
739 /* Send stream to relayd if the stream has an ID. */
740 if (new_stream->net_seq_idx != (uint64_t) -1ULL) {
741 int ret_send_relayd_stream;
742
743 ret_send_relayd_stream =
744 consumer_send_relayd_stream(new_stream, new_stream->chan->pathname);
745 if (ret_send_relayd_stream < 0) {
746 pthread_mutex_unlock(&new_stream->lock);
747 pthread_mutex_unlock(&channel->lock);
748 consumer_stream_free(new_stream);
749 goto error_add_stream_nosignal;
750 }
751
752 /*
753 * If adding an extra stream to an already
754 * existing channel (e.g. cpu hotplug), we need
755 * to send the "streams_sent" command to relayd.
756 */
757 if (channel->streams_sent_to_relayd) {
758 int ret_send_relayd_streams_sent;
759
760 ret_send_relayd_streams_sent =
761 consumer_send_relayd_streams_sent(new_stream->net_seq_idx);
762 if (ret_send_relayd_streams_sent < 0) {
763 pthread_mutex_unlock(&new_stream->lock);
764 pthread_mutex_unlock(&channel->lock);
765 goto error_add_stream_nosignal;
766 }
767 }
768 }
769 pthread_mutex_unlock(&new_stream->lock);
770 pthread_mutex_unlock(&channel->lock);
771
772 /* Get the right pipe where the stream will be sent. */
773 if (new_stream->metadata_flag) {
774 consumer_add_metadata_stream(new_stream);
775 stream_pipe = ctx->consumer_metadata_pipe;
776 } else {
777 consumer_add_data_stream(new_stream);
778 stream_pipe = ctx->consumer_data_pipe;
779 }
780
781 /* Visible to other threads */
782 new_stream->globally_visible = 1;
783
784 health_code_update();
785
786 ret_pipe_write =
787 lttng_pipe_write(stream_pipe, &new_stream, sizeof(new_stream)); /* NOLINT
788 sizeof
789 used on a
790 pointer.
791 */
792 if (ret_pipe_write < 0) {
793 ERR("Consumer write %s stream to pipe %d",
794 new_stream->metadata_flag ? "metadata" : "data",
795 lttng_pipe_get_writefd(stream_pipe));
796 if (new_stream->metadata_flag) {
797 consumer_del_stream_for_metadata(new_stream);
798 } else {
799 consumer_del_stream_for_data(new_stream);
800 }
801 goto error_add_stream_nosignal;
802 }
803
804 DBG("Kernel consumer ADD_STREAM %s (fd: %d) %s with relayd id %" PRIu64,
805 new_stream->name,
806 fd,
807 new_stream->chan->pathname,
808 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 int ret_send_status;
820
821 /*
822 * Get stream's channel reference. Needed when adding the stream to the
823 * global hash table.
824 */
825 channel = consumer_find_channel(msg.u.sent_streams.channel_key);
826 if (!channel) {
827 /*
828 * We could not find the channel. Can happen if cpu hotplug
829 * happens while tearing down.
830 */
831 ERR("Unable to find channel key %" PRIu64, 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_send_status = consumer_send_status_msg(sock, ret_code);
841 if (ret_send_status < 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 int ret_send_relay_streams;
860
861 ret_send_relay_streams =
862 consumer_send_relayd_streams_sent(msg.u.sent_streams.net_seq_idx);
863 if (ret_send_relay_streams < 0) {
864 goto error_streams_sent_nosignal;
865 }
866 channel->streams_sent_to_relayd = true;
867 }
868 end_error_streams_sent:
869 break;
870 error_streams_sent_nosignal:
871 goto end_nosignal;
872 }
873 case LTTNG_CONSUMER_UPDATE_STREAM:
874 {
875 rcu_read_unlock();
876 return -ENOSYS;
877 }
878 case LTTNG_CONSUMER_DESTROY_RELAYD:
879 {
880 uint64_t index = msg.u.destroy_relayd.net_seq_idx;
881 struct consumer_relayd_sock_pair *relayd;
882 int ret_send_status;
883
884 DBG("Kernel consumer destroying relayd %" PRIu64, index);
885
886 /* Get relayd reference if exists. */
887 relayd = consumer_find_relayd(index);
888 if (relayd == nullptr) {
889 DBG("Unable to find relayd %" PRIu64, index);
890 ret_code = LTTCOMM_CONSUMERD_RELAYD_FAIL;
891 }
892
893 /*
894 * Each relayd socket pair has a refcount of stream attached to it
895 * which tells if the relayd is still active or not depending on the
896 * refcount value.
897 *
898 * This will set the destroy flag of the relayd object and destroy it
899 * if the refcount reaches zero when called.
900 *
901 * The destroy can happen either here or when a stream fd hangs up.
902 */
903 if (relayd) {
904 consumer_flag_relayd_for_destroy(relayd);
905 }
906
907 health_code_update();
908
909 ret_send_status = consumer_send_status_msg(sock, ret_code);
910 if (ret_send_status < 0) {
911 /* Somehow, the session daemon is not responding anymore. */
912 goto error_fatal;
913 }
914
915 goto end_nosignal;
916 }
917 case LTTNG_CONSUMER_DATA_PENDING:
918 {
919 int32_t ret_data_pending;
920 uint64_t id = msg.u.data_pending.session_id;
921 ssize_t ret_send;
922
923 DBG("Kernel consumer data pending command for id %" PRIu64, id);
924
925 ret_data_pending = consumer_data_pending(id);
926
927 health_code_update();
928
929 /* Send back returned value to session daemon */
930 ret_send =
931 lttcomm_send_unix_sock(sock, &ret_data_pending, sizeof(ret_data_pending));
932 if (ret_send < 0) {
933 PERROR("send data pending ret code");
934 goto error_fatal;
935 }
936
937 /*
938 * No need to send back a status message since the data pending
939 * returned value is the response.
940 */
941 break;
942 }
943 case LTTNG_CONSUMER_SNAPSHOT_CHANNEL:
944 {
945 struct lttng_consumer_channel *channel;
946 uint64_t key = msg.u.snapshot_channel.key;
947 int ret_send_status;
948
949 channel = consumer_find_channel(key);
950 if (!channel) {
951 ERR("Channel %" PRIu64 " not found", key);
952 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
953 } else {
954 if (msg.u.snapshot_channel.metadata == 1) {
955 int ret_snapshot;
956
957 ret_snapshot = lttng_kconsumer_snapshot_metadata(
958 channel,
959 key,
960 msg.u.snapshot_channel.pathname,
961 msg.u.snapshot_channel.relayd_id,
962 ctx);
963 if (ret_snapshot < 0) {
964 ERR("Snapshot metadata failed");
965 ret_code = LTTCOMM_CONSUMERD_SNAPSHOT_FAILED;
966 }
967 } else {
968 int ret_snapshot;
969
970 ret_snapshot = lttng_kconsumer_snapshot_channel(
971 channel,
972 key,
973 msg.u.snapshot_channel.pathname,
974 msg.u.snapshot_channel.relayd_id,
975 msg.u.snapshot_channel.nb_packets_per_stream);
976 if (ret_snapshot < 0) {
977 ERR("Snapshot channel failed");
978 ret_code = LTTCOMM_CONSUMERD_SNAPSHOT_FAILED;
979 }
980 }
981 }
982 health_code_update();
983
984 ret_send_status = consumer_send_status_msg(sock, ret_code);
985 if (ret_send_status < 0) {
986 /* Somehow, the session daemon is not responding anymore. */
987 goto end_nosignal;
988 }
989 break;
990 }
991 case LTTNG_CONSUMER_DESTROY_CHANNEL:
992 {
993 uint64_t key = msg.u.destroy_channel.key;
994 struct lttng_consumer_channel *channel;
995 int ret_send_status;
996
997 channel = consumer_find_channel(key);
998 if (!channel) {
999 ERR("Kernel consumer destroy channel %" PRIu64 " not found", key);
1000 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
1001 }
1002
1003 health_code_update();
1004
1005 ret_send_status = consumer_send_status_msg(sock, ret_code);
1006 if (ret_send_status < 0) {
1007 /* Somehow, the session daemon is not responding anymore. */
1008 goto end_destroy_channel;
1009 }
1010
1011 health_code_update();
1012
1013 /* Stop right now if no channel was found. */
1014 if (!channel) {
1015 goto end_destroy_channel;
1016 }
1017
1018 /*
1019 * This command should ONLY be issued for channel with streams set in
1020 * no monitor mode.
1021 */
1022 LTTNG_ASSERT(!channel->monitor);
1023
1024 /*
1025 * The refcount should ALWAYS be 0 in the case of a channel in no
1026 * monitor mode.
1027 */
1028 LTTNG_ASSERT(!uatomic_sub_return(&channel->refcount, 1));
1029
1030 consumer_del_channel(channel);
1031 end_destroy_channel:
1032 goto end_nosignal;
1033 }
1034 case LTTNG_CONSUMER_DISCARDED_EVENTS:
1035 {
1036 ssize_t ret;
1037 uint64_t count;
1038 struct lttng_consumer_channel *channel;
1039 uint64_t id = msg.u.discarded_events.session_id;
1040 uint64_t key = msg.u.discarded_events.channel_key;
1041
1042 DBG("Kernel consumer discarded events command for session id %" PRIu64
1043 ", channel key %" PRIu64,
1044 id,
1045 key);
1046
1047 channel = consumer_find_channel(key);
1048 if (!channel) {
1049 ERR("Kernel consumer discarded events channel %" PRIu64 " not found", key);
1050 count = 0;
1051 } else {
1052 count = channel->discarded_events;
1053 }
1054
1055 health_code_update();
1056
1057 /* Send back returned value to session daemon */
1058 ret = lttcomm_send_unix_sock(sock, &count, sizeof(count));
1059 if (ret < 0) {
1060 PERROR("send discarded events");
1061 goto error_fatal;
1062 }
1063
1064 break;
1065 }
1066 case LTTNG_CONSUMER_LOST_PACKETS:
1067 {
1068 ssize_t ret;
1069 uint64_t count;
1070 struct lttng_consumer_channel *channel;
1071 uint64_t id = msg.u.lost_packets.session_id;
1072 uint64_t key = msg.u.lost_packets.channel_key;
1073
1074 DBG("Kernel consumer lost packets command for session id %" PRIu64
1075 ", channel key %" PRIu64,
1076 id,
1077 key);
1078
1079 channel = consumer_find_channel(key);
1080 if (!channel) {
1081 ERR("Kernel consumer lost packets channel %" PRIu64 " not found", key);
1082 count = 0;
1083 } else {
1084 count = channel->lost_packets;
1085 }
1086
1087 health_code_update();
1088
1089 /* Send back returned value to session daemon */
1090 ret = lttcomm_send_unix_sock(sock, &count, sizeof(count));
1091 if (ret < 0) {
1092 PERROR("send lost packets");
1093 goto error_fatal;
1094 }
1095
1096 break;
1097 }
1098 case LTTNG_CONSUMER_SET_CHANNEL_MONITOR_PIPE:
1099 {
1100 int channel_monitor_pipe;
1101 int ret_send_status, ret_set_channel_monitor_pipe;
1102 ssize_t ret_recv;
1103
1104 ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1105 /* Successfully received the command's type. */
1106 ret_send_status = consumer_send_status_msg(sock, ret_code);
1107 if (ret_send_status < 0) {
1108 goto error_fatal;
1109 }
1110
1111 ret_recv = lttcomm_recv_fds_unix_sock(sock, &channel_monitor_pipe, 1);
1112 if (ret_recv != sizeof(channel_monitor_pipe)) {
1113 ERR("Failed to receive channel monitor pipe");
1114 goto error_fatal;
1115 }
1116
1117 DBG("Received channel monitor pipe (%d)", channel_monitor_pipe);
1118 ret_set_channel_monitor_pipe =
1119 consumer_timer_thread_set_channel_monitor_pipe(channel_monitor_pipe);
1120 if (!ret_set_channel_monitor_pipe) {
1121 int flags;
1122 int ret_fcntl;
1123
1124 ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1125 /* Set the pipe as non-blocking. */
1126 ret_fcntl = fcntl(channel_monitor_pipe, F_GETFL, 0);
1127 if (ret_fcntl == -1) {
1128 PERROR("fcntl get flags of the channel monitoring pipe");
1129 goto error_fatal;
1130 }
1131 flags = ret_fcntl;
1132
1133 ret_fcntl = fcntl(channel_monitor_pipe, F_SETFL, flags | O_NONBLOCK);
1134 if (ret_fcntl == -1) {
1135 PERROR("fcntl set O_NONBLOCK flag of the channel monitoring pipe");
1136 goto error_fatal;
1137 }
1138 DBG("Channel monitor pipe set as non-blocking");
1139 } else {
1140 ret_code = LTTCOMM_CONSUMERD_ALREADY_SET;
1141 }
1142 ret_send_status = consumer_send_status_msg(sock, ret_code);
1143 if (ret_send_status < 0) {
1144 goto error_fatal;
1145 }
1146 break;
1147 }
1148 case LTTNG_CONSUMER_ROTATE_CHANNEL:
1149 {
1150 struct lttng_consumer_channel *channel;
1151 uint64_t key = msg.u.rotate_channel.key;
1152 int ret_send_status;
1153
1154 DBG("Consumer rotate channel %" PRIu64, key);
1155
1156 channel = consumer_find_channel(key);
1157 if (!channel) {
1158 ERR("Channel %" PRIu64 " not found", key);
1159 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
1160 } else {
1161 /*
1162 * Sample the rotate position of all the streams in this channel.
1163 */
1164 int ret_rotate_channel;
1165
1166 ret_rotate_channel = lttng_consumer_rotate_channel(
1167 channel, key, msg.u.rotate_channel.relayd_id);
1168 if (ret_rotate_channel < 0) {
1169 ERR("Rotate channel failed");
1170 ret_code = LTTCOMM_CONSUMERD_ROTATION_FAIL;
1171 }
1172
1173 health_code_update();
1174 }
1175
1176 ret_send_status = consumer_send_status_msg(sock, ret_code);
1177 if (ret_send_status < 0) {
1178 /* Somehow, the session daemon is not responding anymore. */
1179 goto error_rotate_channel;
1180 }
1181 if (channel) {
1182 /* Rotate the streams that are ready right now. */
1183 int ret_rotate;
1184
1185 ret_rotate = lttng_consumer_rotate_ready_streams(channel, key);
1186 if (ret_rotate < 0) {
1187 ERR("Rotate ready streams failed");
1188 }
1189 }
1190 break;
1191 error_rotate_channel:
1192 goto end_nosignal;
1193 }
1194 case LTTNG_CONSUMER_CLEAR_CHANNEL:
1195 {
1196 struct lttng_consumer_channel *channel;
1197 uint64_t key = msg.u.clear_channel.key;
1198 int ret_send_status;
1199
1200 channel = consumer_find_channel(key);
1201 if (!channel) {
1202 DBG("Channel %" PRIu64 " not found", key);
1203 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
1204 } else {
1205 int ret_clear_channel;
1206
1207 ret_clear_channel = lttng_consumer_clear_channel(channel);
1208 if (ret_clear_channel) {
1209 ERR("Clear channel failed");
1210 ret_code = (lttcomm_return_code) ret_clear_channel;
1211 }
1212
1213 health_code_update();
1214 }
1215
1216 ret_send_status = consumer_send_status_msg(sock, ret_code);
1217 if (ret_send_status < 0) {
1218 /* Somehow, the session daemon is not responding anymore. */
1219 goto end_nosignal;
1220 }
1221
1222 break;
1223 }
1224 case LTTNG_CONSUMER_INIT:
1225 {
1226 int ret_send_status;
1227 lttng_uuid sessiond_uuid;
1228
1229 std::copy(std::begin(msg.u.init.sessiond_uuid),
1230 std::end(msg.u.init.sessiond_uuid),
1231 sessiond_uuid.begin());
1232
1233 ret_code = lttng_consumer_init_command(ctx, sessiond_uuid);
1234 health_code_update();
1235 ret_send_status = consumer_send_status_msg(sock, ret_code);
1236 if (ret_send_status < 0) {
1237 /* Somehow, the session daemon is not responding anymore. */
1238 goto end_nosignal;
1239 }
1240 break;
1241 }
1242 case LTTNG_CONSUMER_CREATE_TRACE_CHUNK:
1243 {
1244 const struct lttng_credentials credentials = {
1245 .uid = LTTNG_OPTIONAL_INIT_VALUE(
1246 msg.u.create_trace_chunk.credentials.value.uid),
1247 .gid = LTTNG_OPTIONAL_INIT_VALUE(
1248 msg.u.create_trace_chunk.credentials.value.gid),
1249 };
1250 const bool is_local_trace = !msg.u.create_trace_chunk.relayd_id.is_set;
1251 const uint64_t relayd_id = msg.u.create_trace_chunk.relayd_id.value;
1252 const char *chunk_override_name = *msg.u.create_trace_chunk.override_name ?
1253 msg.u.create_trace_chunk.override_name :
1254 nullptr;
1255 struct lttng_directory_handle *chunk_directory_handle = nullptr;
1256
1257 /*
1258 * The session daemon will only provide a chunk directory file
1259 * descriptor for local traces.
1260 */
1261 if (is_local_trace) {
1262 int chunk_dirfd;
1263 int ret_send_status;
1264 ssize_t ret_recv;
1265
1266 /* Acnowledge the reception of the command. */
1267 ret_send_status = consumer_send_status_msg(sock, LTTCOMM_CONSUMERD_SUCCESS);
1268 if (ret_send_status < 0) {
1269 /* Somehow, the session daemon is not responding anymore. */
1270 goto end_nosignal;
1271 }
1272
1273 ret_recv = lttcomm_recv_fds_unix_sock(sock, &chunk_dirfd, 1);
1274 if (ret_recv != sizeof(chunk_dirfd)) {
1275 ERR("Failed to receive trace chunk directory file descriptor");
1276 goto error_fatal;
1277 }
1278
1279 DBG("Received trace chunk directory fd (%d)", chunk_dirfd);
1280 chunk_directory_handle =
1281 lttng_directory_handle_create_from_dirfd(chunk_dirfd);
1282 if (!chunk_directory_handle) {
1283 ERR("Failed to initialize chunk directory handle from directory file descriptor");
1284 if (close(chunk_dirfd)) {
1285 PERROR("Failed to close chunk directory file descriptor");
1286 }
1287 goto error_fatal;
1288 }
1289 }
1290
1291 ret_code = lttng_consumer_create_trace_chunk(
1292 !is_local_trace ? &relayd_id : nullptr,
1293 msg.u.create_trace_chunk.session_id,
1294 msg.u.create_trace_chunk.chunk_id,
1295 (time_t) msg.u.create_trace_chunk.creation_timestamp,
1296 chunk_override_name,
1297 msg.u.create_trace_chunk.credentials.is_set ? &credentials : nullptr,
1298 chunk_directory_handle);
1299 lttng_directory_handle_put(chunk_directory_handle);
1300 goto end_msg_sessiond;
1301 }
1302 case LTTNG_CONSUMER_CLOSE_TRACE_CHUNK:
1303 {
1304 enum lttng_trace_chunk_command_type close_command =
1305 (lttng_trace_chunk_command_type) msg.u.close_trace_chunk.close_command.value;
1306 const uint64_t relayd_id = msg.u.close_trace_chunk.relayd_id.value;
1307 struct lttcomm_consumer_close_trace_chunk_reply reply;
1308 char path[LTTNG_PATH_MAX];
1309 ssize_t ret_send;
1310
1311 ret_code = lttng_consumer_close_trace_chunk(
1312 msg.u.close_trace_chunk.relayd_id.is_set ? &relayd_id : nullptr,
1313 msg.u.close_trace_chunk.session_id,
1314 msg.u.close_trace_chunk.chunk_id,
1315 (time_t) msg.u.close_trace_chunk.close_timestamp,
1316 msg.u.close_trace_chunk.close_command.is_set ? &close_command : nullptr,
1317 path);
1318 reply.ret_code = ret_code;
1319 reply.path_length = strlen(path) + 1;
1320 ret_send = lttcomm_send_unix_sock(sock, &reply, sizeof(reply));
1321 if (ret_send != sizeof(reply)) {
1322 goto error_fatal;
1323 }
1324 ret_send = lttcomm_send_unix_sock(sock, path, reply.path_length);
1325 if (ret_send != reply.path_length) {
1326 goto error_fatal;
1327 }
1328 goto end_nosignal;
1329 }
1330 case LTTNG_CONSUMER_TRACE_CHUNK_EXISTS:
1331 {
1332 const uint64_t relayd_id = msg.u.trace_chunk_exists.relayd_id.value;
1333
1334 ret_code = lttng_consumer_trace_chunk_exists(
1335 msg.u.trace_chunk_exists.relayd_id.is_set ? &relayd_id : nullptr,
1336 msg.u.trace_chunk_exists.session_id,
1337 msg.u.trace_chunk_exists.chunk_id);
1338 goto end_msg_sessiond;
1339 }
1340 case LTTNG_CONSUMER_OPEN_CHANNEL_PACKETS:
1341 {
1342 const uint64_t key = msg.u.open_channel_packets.key;
1343 struct lttng_consumer_channel *channel = consumer_find_channel(key);
1344
1345 if (channel) {
1346 pthread_mutex_lock(&channel->lock);
1347 ret_code = lttng_consumer_open_channel_packets(channel);
1348 pthread_mutex_unlock(&channel->lock);
1349 } else {
1350 WARN("Channel %" PRIu64 " not found", key);
1351 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
1352 }
1353
1354 health_code_update();
1355 goto end_msg_sessiond;
1356 }
1357 default:
1358 goto end_nosignal;
1359 }
1360
1361 end_nosignal:
1362 /*
1363 * Return 1 to indicate success since the 0 value can be a socket
1364 * shutdown during the recv() or send() call.
1365 */
1366 ret_func = 1;
1367 goto end;
1368 error_fatal:
1369 /* This will issue a consumer stop. */
1370 ret_func = -1;
1371 goto end;
1372 end_msg_sessiond:
1373 /*
1374 * The returned value here is not useful since either way we'll return 1 to
1375 * the caller because the session daemon socket management is done
1376 * elsewhere. Returning a negative code or 0 will shutdown the consumer.
1377 */
1378 {
1379 int ret_send_status;
1380
1381 ret_send_status = consumer_send_status_msg(sock, ret_code);
1382 if (ret_send_status < 0) {
1383 goto error_fatal;
1384 }
1385 }
1386
1387 ret_func = 1;
1388
1389 end:
1390 health_code_update();
1391 rcu_read_unlock();
1392 return ret_func;
1393 }
1394
1395 /*
1396 * Sync metadata meaning request them to the session daemon and snapshot to the
1397 * metadata thread can consumer them.
1398 *
1399 * Metadata stream lock MUST be acquired.
1400 */
1401 enum sync_metadata_status lttng_kconsumer_sync_metadata(struct lttng_consumer_stream *metadata)
1402 {
1403 int ret;
1404 enum sync_metadata_status status;
1405
1406 LTTNG_ASSERT(metadata);
1407
1408 ret = kernctl_buffer_flush(metadata->wait_fd);
1409 if (ret < 0) {
1410 ERR("Failed to flush kernel stream");
1411 status = SYNC_METADATA_STATUS_ERROR;
1412 goto end;
1413 }
1414
1415 ret = kernctl_snapshot(metadata->wait_fd);
1416 if (ret < 0) {
1417 if (errno == EAGAIN) {
1418 /* No new metadata, exit. */
1419 DBG("Sync metadata, no new kernel metadata");
1420 status = SYNC_METADATA_STATUS_NO_DATA;
1421 } else {
1422 ERR("Sync metadata, taking kernel snapshot failed.");
1423 status = SYNC_METADATA_STATUS_ERROR;
1424 }
1425 } else {
1426 status = SYNC_METADATA_STATUS_NEW_DATA;
1427 }
1428
1429 end:
1430 return status;
1431 }
1432
1433 static int extract_common_subbuffer_info(struct lttng_consumer_stream *stream,
1434 struct stream_subbuffer *subbuf)
1435 {
1436 int ret;
1437
1438 ret = kernctl_get_subbuf_size(stream->wait_fd, &subbuf->info.data.subbuf_size);
1439 if (ret) {
1440 goto end;
1441 }
1442
1443 ret = kernctl_get_padded_subbuf_size(stream->wait_fd,
1444 &subbuf->info.data.padded_subbuf_size);
1445 if (ret) {
1446 goto end;
1447 }
1448
1449 end:
1450 return ret;
1451 }
1452
1453 static int extract_metadata_subbuffer_info(struct lttng_consumer_stream *stream,
1454 struct stream_subbuffer *subbuf)
1455 {
1456 int ret;
1457
1458 ret = extract_common_subbuffer_info(stream, subbuf);
1459 if (ret) {
1460 goto end;
1461 }
1462
1463 ret = kernctl_get_metadata_version(stream->wait_fd, &subbuf->info.metadata.version);
1464 if (ret) {
1465 goto end;
1466 }
1467
1468 end:
1469 return ret;
1470 }
1471
1472 static int extract_data_subbuffer_info(struct lttng_consumer_stream *stream,
1473 struct stream_subbuffer *subbuf)
1474 {
1475 int ret;
1476
1477 ret = extract_common_subbuffer_info(stream, subbuf);
1478 if (ret) {
1479 goto end;
1480 }
1481
1482 ret = kernctl_get_packet_size(stream->wait_fd, &subbuf->info.data.packet_size);
1483 if (ret < 0) {
1484 PERROR("Failed to get sub-buffer packet size");
1485 goto end;
1486 }
1487
1488 ret = kernctl_get_content_size(stream->wait_fd, &subbuf->info.data.content_size);
1489 if (ret < 0) {
1490 PERROR("Failed to get sub-buffer content size");
1491 goto end;
1492 }
1493
1494 ret = kernctl_get_timestamp_begin(stream->wait_fd, &subbuf->info.data.timestamp_begin);
1495 if (ret < 0) {
1496 PERROR("Failed to get sub-buffer begin timestamp");
1497 goto end;
1498 }
1499
1500 ret = kernctl_get_timestamp_end(stream->wait_fd, &subbuf->info.data.timestamp_end);
1501 if (ret < 0) {
1502 PERROR("Failed to get sub-buffer end timestamp");
1503 goto end;
1504 }
1505
1506 ret = kernctl_get_events_discarded(stream->wait_fd, &subbuf->info.data.events_discarded);
1507 if (ret) {
1508 PERROR("Failed to get sub-buffer events discarded count");
1509 goto end;
1510 }
1511
1512 ret = kernctl_get_sequence_number(stream->wait_fd,
1513 &subbuf->info.data.sequence_number.value);
1514 if (ret) {
1515 /* May not be supported by older LTTng-modules. */
1516 if (ret != -ENOTTY) {
1517 PERROR("Failed to get sub-buffer sequence number");
1518 goto end;
1519 }
1520 } else {
1521 subbuf->info.data.sequence_number.is_set = true;
1522 }
1523
1524 ret = kernctl_get_stream_id(stream->wait_fd, &subbuf->info.data.stream_id);
1525 if (ret < 0) {
1526 PERROR("Failed to get stream id");
1527 goto end;
1528 }
1529
1530 ret = kernctl_get_instance_id(stream->wait_fd, &subbuf->info.data.stream_instance_id.value);
1531 if (ret) {
1532 /* May not be supported by older LTTng-modules. */
1533 if (ret != -ENOTTY) {
1534 PERROR("Failed to get stream instance id");
1535 goto end;
1536 }
1537 } else {
1538 subbuf->info.data.stream_instance_id.is_set = true;
1539 }
1540 end:
1541 return ret;
1542 }
1543
1544 static enum get_next_subbuffer_status get_subbuffer_common(struct lttng_consumer_stream *stream,
1545 struct stream_subbuffer *subbuffer)
1546 {
1547 int ret;
1548 enum get_next_subbuffer_status status;
1549
1550 ret = kernctl_get_next_subbuf(stream->wait_fd);
1551 switch (ret) {
1552 case 0:
1553 status = GET_NEXT_SUBBUFFER_STATUS_OK;
1554 break;
1555 case -ENODATA:
1556 case -EAGAIN:
1557 /*
1558 * The caller only expects -ENODATA when there is no data to
1559 * read, but the kernel tracer returns -EAGAIN when there is
1560 * currently no data for a non-finalized stream, and -ENODATA
1561 * when there is no data for a finalized stream. Those can be
1562 * combined into a -ENODATA return value.
1563 */
1564 status = GET_NEXT_SUBBUFFER_STATUS_NO_DATA;
1565 goto end;
1566 default:
1567 status = GET_NEXT_SUBBUFFER_STATUS_ERROR;
1568 goto end;
1569 }
1570
1571 ret = stream->read_subbuffer_ops.extract_subbuffer_info(stream, subbuffer);
1572 if (ret) {
1573 status = GET_NEXT_SUBBUFFER_STATUS_ERROR;
1574 }
1575 end:
1576 return status;
1577 }
1578
1579 static enum get_next_subbuffer_status
1580 get_next_subbuffer_splice(struct lttng_consumer_stream *stream, struct stream_subbuffer *subbuffer)
1581 {
1582 const enum get_next_subbuffer_status status = get_subbuffer_common(stream, subbuffer);
1583
1584 if (status != GET_NEXT_SUBBUFFER_STATUS_OK) {
1585 goto end;
1586 }
1587
1588 subbuffer->buffer.fd = stream->wait_fd;
1589 end:
1590 return status;
1591 }
1592
1593 static enum get_next_subbuffer_status get_next_subbuffer_mmap(struct lttng_consumer_stream *stream,
1594 struct stream_subbuffer *subbuffer)
1595 {
1596 int ret;
1597 enum get_next_subbuffer_status status;
1598 const char *addr;
1599
1600 status = get_subbuffer_common(stream, subbuffer);
1601 if (status != GET_NEXT_SUBBUFFER_STATUS_OK) {
1602 goto end;
1603 }
1604
1605 ret = get_current_subbuf_addr(stream, &addr);
1606 if (ret) {
1607 status = GET_NEXT_SUBBUFFER_STATUS_ERROR;
1608 goto end;
1609 }
1610
1611 subbuffer->buffer.buffer =
1612 lttng_buffer_view_init(addr, 0, subbuffer->info.data.padded_subbuf_size);
1613 end:
1614 return status;
1615 }
1616
1617 static enum get_next_subbuffer_status
1618 get_next_subbuffer_metadata_check(struct lttng_consumer_stream *stream,
1619 struct stream_subbuffer *subbuffer)
1620 {
1621 int ret;
1622 const char *addr;
1623 bool coherent;
1624 enum get_next_subbuffer_status status;
1625
1626 ret = kernctl_get_next_subbuf_metadata_check(stream->wait_fd, &coherent);
1627 if (ret) {
1628 goto end;
1629 }
1630
1631 ret = stream->read_subbuffer_ops.extract_subbuffer_info(stream, subbuffer);
1632 if (ret) {
1633 goto end;
1634 }
1635
1636 LTTNG_OPTIONAL_SET(&subbuffer->info.metadata.coherent, coherent);
1637
1638 ret = get_current_subbuf_addr(stream, &addr);
1639 if (ret) {
1640 goto end;
1641 }
1642
1643 subbuffer->buffer.buffer =
1644 lttng_buffer_view_init(addr, 0, subbuffer->info.data.padded_subbuf_size);
1645 DBG("Got metadata packet with padded_subbuf_size = %lu, coherent = %s",
1646 subbuffer->info.metadata.padded_subbuf_size,
1647 coherent ? "true" : "false");
1648 end:
1649 /*
1650 * The caller only expects -ENODATA when there is no data to read, but
1651 * the kernel tracer returns -EAGAIN when there is currently no data
1652 * for a non-finalized stream, and -ENODATA when there is no data for a
1653 * finalized stream. Those can be combined into a -ENODATA return value.
1654 */
1655 switch (ret) {
1656 case 0:
1657 status = GET_NEXT_SUBBUFFER_STATUS_OK;
1658 break;
1659 case -ENODATA:
1660 case -EAGAIN:
1661 /*
1662 * The caller only expects -ENODATA when there is no data to
1663 * read, but the kernel tracer returns -EAGAIN when there is
1664 * currently no data for a non-finalized stream, and -ENODATA
1665 * when there is no data for a finalized stream. Those can be
1666 * combined into a -ENODATA return value.
1667 */
1668 status = GET_NEXT_SUBBUFFER_STATUS_NO_DATA;
1669 break;
1670 default:
1671 status = GET_NEXT_SUBBUFFER_STATUS_ERROR;
1672 break;
1673 }
1674
1675 return status;
1676 }
1677
1678 static int put_next_subbuffer(struct lttng_consumer_stream *stream,
1679 struct stream_subbuffer *subbuffer __attribute__((unused)))
1680 {
1681 const int ret = kernctl_put_next_subbuf(stream->wait_fd);
1682
1683 if (ret) {
1684 if (ret == -EFAULT) {
1685 PERROR("Error in unreserving sub buffer");
1686 } else if (ret == -EIO) {
1687 /* Should never happen with newer LTTng versions */
1688 PERROR("Reader has been pushed by the writer, last sub-buffer corrupted");
1689 }
1690 }
1691
1692 return ret;
1693 }
1694
1695 static bool is_get_next_check_metadata_available(int tracer_fd)
1696 {
1697 const int ret = kernctl_get_next_subbuf_metadata_check(tracer_fd, nullptr);
1698 const bool available = ret != -ENOTTY;
1699
1700 if (ret == 0) {
1701 /* get succeeded, make sure to put the subbuffer. */
1702 kernctl_put_subbuf(tracer_fd);
1703 }
1704
1705 return available;
1706 }
1707
1708 static int signal_metadata(struct lttng_consumer_stream *stream,
1709 struct lttng_consumer_local_data *ctx __attribute__((unused)))
1710 {
1711 ASSERT_LOCKED(stream->metadata_rdv_lock);
1712 return pthread_cond_broadcast(&stream->metadata_rdv) ? -errno : 0;
1713 }
1714
1715 static int lttng_kconsumer_set_stream_ops(struct lttng_consumer_stream *stream)
1716 {
1717 int ret = 0;
1718
1719 if (stream->metadata_flag && stream->chan->is_live) {
1720 DBG("Attempting to enable metadata bucketization for live consumers");
1721 if (is_get_next_check_metadata_available(stream->wait_fd)) {
1722 DBG("Kernel tracer supports get_next_subbuffer_metadata_check, metadata will be accumulated until a coherent state is reached");
1723 stream->read_subbuffer_ops.get_next_subbuffer =
1724 get_next_subbuffer_metadata_check;
1725 ret = consumer_stream_enable_metadata_bucketization(stream);
1726 if (ret) {
1727 goto end;
1728 }
1729 } else {
1730 /*
1731 * The kernel tracer version is too old to indicate
1732 * when the metadata stream has reached a "coherent"
1733 * (parseable) point.
1734 *
1735 * This means that a live viewer may see an incoherent
1736 * sequence of metadata and fail to parse it.
1737 */
1738 WARN("Kernel tracer does not support get_next_subbuffer_metadata_check which may cause live clients to fail to parse the metadata stream");
1739 metadata_bucket_destroy(stream->metadata_bucket);
1740 stream->metadata_bucket = nullptr;
1741 }
1742
1743 stream->read_subbuffer_ops.on_sleep = signal_metadata;
1744 }
1745
1746 if (!stream->read_subbuffer_ops.get_next_subbuffer) {
1747 if (stream->chan->output == CONSUMER_CHANNEL_MMAP) {
1748 stream->read_subbuffer_ops.get_next_subbuffer = get_next_subbuffer_mmap;
1749 } else {
1750 stream->read_subbuffer_ops.get_next_subbuffer = get_next_subbuffer_splice;
1751 }
1752 }
1753
1754 if (stream->metadata_flag) {
1755 stream->read_subbuffer_ops.extract_subbuffer_info = extract_metadata_subbuffer_info;
1756 } else {
1757 stream->read_subbuffer_ops.extract_subbuffer_info = extract_data_subbuffer_info;
1758 if (stream->chan->is_live) {
1759 stream->read_subbuffer_ops.send_live_beacon = consumer_flush_kernel_index;
1760 }
1761 }
1762
1763 stream->read_subbuffer_ops.put_next_subbuffer = put_next_subbuffer;
1764 end:
1765 return ret;
1766 }
1767
1768 int lttng_kconsumer_on_recv_stream(struct lttng_consumer_stream *stream)
1769 {
1770 int ret;
1771
1772 LTTNG_ASSERT(stream);
1773
1774 /*
1775 * Don't create anything if this is set for streaming or if there is
1776 * no current trace chunk on the parent channel.
1777 */
1778 if (stream->net_seq_idx == (uint64_t) -1ULL && stream->chan->monitor &&
1779 stream->chan->trace_chunk) {
1780 ret = consumer_stream_create_output_files(stream, true);
1781 if (ret) {
1782 goto error;
1783 }
1784 }
1785
1786 if (stream->output == LTTNG_EVENT_MMAP) {
1787 /* get the len of the mmap region */
1788 unsigned long mmap_len;
1789
1790 ret = kernctl_get_mmap_len(stream->wait_fd, &mmap_len);
1791 if (ret != 0) {
1792 PERROR("kernctl_get_mmap_len");
1793 goto error_close_fd;
1794 }
1795 stream->mmap_len = (size_t) mmap_len;
1796
1797 stream->mmap_base =
1798 mmap(nullptr, stream->mmap_len, PROT_READ, MAP_PRIVATE, stream->wait_fd, 0);
1799 if (stream->mmap_base == MAP_FAILED) {
1800 PERROR("Error mmaping");
1801 ret = -1;
1802 goto error_close_fd;
1803 }
1804 }
1805
1806 ret = lttng_kconsumer_set_stream_ops(stream);
1807 if (ret) {
1808 goto error_close_fd;
1809 }
1810
1811 /* we return 0 to let the library handle the FD internally */
1812 return 0;
1813
1814 error_close_fd:
1815 if (stream->out_fd >= 0) {
1816 int err;
1817
1818 err = close(stream->out_fd);
1819 LTTNG_ASSERT(!err);
1820 stream->out_fd = -1;
1821 }
1822 error:
1823 return ret;
1824 }
1825
1826 /*
1827 * Check if data is still being extracted from the buffers for a specific
1828 * stream. Consumer data lock MUST be acquired before calling this function
1829 * and the stream lock.
1830 *
1831 * Return 1 if the traced data are still getting read else 0 meaning that the
1832 * data is available for trace viewer reading.
1833 */
1834 int lttng_kconsumer_data_pending(struct lttng_consumer_stream *stream)
1835 {
1836 int ret;
1837
1838 LTTNG_ASSERT(stream);
1839
1840 if (stream->endpoint_status != CONSUMER_ENDPOINT_ACTIVE) {
1841 ret = 0;
1842 goto end;
1843 }
1844
1845 ret = kernctl_get_next_subbuf(stream->wait_fd);
1846 if (ret == 0) {
1847 /* There is still data so let's put back this subbuffer. */
1848 ret = kernctl_put_subbuf(stream->wait_fd);
1849 LTTNG_ASSERT(ret == 0);
1850 ret = 1; /* Data is pending */
1851 goto end;
1852 }
1853
1854 /* Data is NOT pending and ready to be read. */
1855 ret = 0;
1856
1857 end:
1858 return ret;
1859 }
This page took 0.110552 seconds and 4 git commands to generate.