2 * Copyright (C) 2011 - Julien Desfossez <julien.desfossez@polymtl.ca>
3 * Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
4 * Copyright (C) 2017 - Jérémie Galarneau <jeremie.galarneau@efficios.com>
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License, version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License along
16 * with this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
27 #include <sys/socket.h>
28 #include <sys/types.h>
33 #include <bin/lttng-consumerd/health-consumerd.h>
34 #include <common/common.h>
35 #include <common/kernel-ctl/kernel-ctl.h>
36 #include <common/sessiond-comm/sessiond-comm.h>
37 #include <common/sessiond-comm/relayd.h>
38 #include <common/compat/fcntl.h>
39 #include <common/compat/endian.h>
40 #include <common/pipe.h>
41 #include <common/relayd/relayd.h>
42 #include <common/utils.h>
43 #include <common/consumer/consumer-stream.h>
44 #include <common/index/index.h>
45 #include <common/consumer/consumer-timer.h>
47 #include "kernel-consumer.h"
49 extern struct lttng_consumer_global_data consumer_data
;
50 extern int consumer_poll_timeout
;
53 * Take a snapshot for a specific fd
55 * Returns 0 on success, < 0 on error
57 int lttng_kconsumer_take_snapshot(struct lttng_consumer_stream
*stream
)
60 int infd
= stream
->wait_fd
;
62 ret
= kernctl_snapshot(infd
);
64 * -EAGAIN is not an error, it just means that there is no data to
67 if (ret
!= 0 && ret
!= -EAGAIN
) {
68 PERROR("Getting sub-buffer snapshot.");
75 * Sample consumed and produced positions for a specific fd.
77 * Returns 0 on success, < 0 on error.
79 int lttng_kconsumer_sample_snapshot_positions(
80 struct lttng_consumer_stream
*stream
)
84 return kernctl_snapshot_sample_positions(stream
->wait_fd
);
88 * Get the produced position
90 * Returns 0 on success, < 0 on error
92 int lttng_kconsumer_get_produced_snapshot(struct lttng_consumer_stream
*stream
,
96 int infd
= stream
->wait_fd
;
98 ret
= kernctl_snapshot_get_produced(infd
, pos
);
100 PERROR("kernctl_snapshot_get_produced");
107 * Get the consumerd position
109 * Returns 0 on success, < 0 on error
111 int lttng_kconsumer_get_consumed_snapshot(struct lttng_consumer_stream
*stream
,
115 int infd
= stream
->wait_fd
;
117 ret
= kernctl_snapshot_get_consumed(infd
, pos
);
119 PERROR("kernctl_snapshot_get_consumed");
126 * Take a snapshot of all the stream of a channel
128 * Returns 0 on success, < 0 on error
130 int lttng_kconsumer_snapshot_channel(uint64_t key
, char *path
,
131 uint64_t relayd_id
, uint64_t nb_packets_per_stream
,
132 struct lttng_consumer_local_data
*ctx
)
135 struct lttng_consumer_channel
*channel
;
136 struct lttng_consumer_stream
*stream
;
138 DBG("Kernel consumer snapshot channel %" PRIu64
, key
);
142 channel
= consumer_find_channel(key
);
144 ERR("No channel found for key %" PRIu64
, key
);
149 /* Splice is not supported yet for channel snapshot. */
150 if (channel
->output
!= CONSUMER_CHANNEL_MMAP
) {
151 ERR("Unsupported output %d", channel
->output
);
156 cds_list_for_each_entry(stream
, &channel
->streams
.head
, send_node
) {
157 unsigned long consumed_pos
, produced_pos
;
159 health_code_update();
162 * Lock stream because we are about to change its state.
164 pthread_mutex_lock(&stream
->lock
);
167 * Assign the received relayd ID so we can use it for streaming. The streams
168 * are not visible to anyone so this is OK to change it.
170 stream
->net_seq_idx
= relayd_id
;
171 channel
->relayd_id
= relayd_id
;
172 if (relayd_id
!= (uint64_t) -1ULL) {
173 ret
= consumer_send_relayd_stream(stream
, path
);
175 ERR("sending stream to relayd");
179 ret
= utils_create_stream_file(path
, stream
->name
,
180 stream
->chan
->tracefile_size
,
181 stream
->tracefile_count_current
,
182 stream
->uid
, stream
->gid
, NULL
);
184 ERR("utils_create_stream_file");
188 stream
->out_fd
= ret
;
189 stream
->tracefile_size_current
= 0;
191 DBG("Kernel consumer snapshot stream %s/%s (%" PRIu64
")",
192 path
, stream
->name
, stream
->key
);
195 ret
= kernctl_buffer_flush_empty(stream
->wait_fd
);
198 * Doing a buffer flush which does not take into
199 * account empty packets. This is not perfect
200 * for stream intersection, but required as a
201 * fall-back when "flush_empty" is not
202 * implemented by lttng-modules.
204 ret
= kernctl_buffer_flush(stream
->wait_fd
);
206 ERR("Failed to flush kernel stream");
212 ret
= lttng_kconsumer_take_snapshot(stream
);
214 ERR("Taking kernel snapshot");
218 ret
= lttng_kconsumer_get_produced_snapshot(stream
, &produced_pos
);
220 ERR("Produced kernel snapshot position");
224 ret
= lttng_kconsumer_get_consumed_snapshot(stream
, &consumed_pos
);
226 ERR("Consumerd kernel snapshot position");
230 if (stream
->max_sb_size
== 0) {
231 ret
= kernctl_get_max_subbuf_size(stream
->wait_fd
,
232 &stream
->max_sb_size
);
234 ERR("Getting kernel max_sb_size");
239 consumed_pos
= consumer_get_consume_start_pos(consumed_pos
,
240 produced_pos
, nb_packets_per_stream
,
241 stream
->max_sb_size
);
243 while ((long) (consumed_pos
- produced_pos
) < 0) {
245 unsigned long len
, padded_len
;
247 health_code_update();
249 DBG("Kernel consumer taking snapshot at pos %lu", consumed_pos
);
251 ret
= kernctl_get_subbuf(stream
->wait_fd
, &consumed_pos
);
253 if (ret
!= -EAGAIN
) {
254 PERROR("kernctl_get_subbuf snapshot");
257 DBG("Kernel consumer get subbuf failed. Skipping it.");
258 consumed_pos
+= stream
->max_sb_size
;
259 stream
->chan
->lost_packets
++;
263 ret
= kernctl_get_subbuf_size(stream
->wait_fd
, &len
);
265 ERR("Snapshot kernctl_get_subbuf_size");
266 goto error_put_subbuf
;
269 ret
= kernctl_get_padded_subbuf_size(stream
->wait_fd
, &padded_len
);
271 ERR("Snapshot kernctl_get_padded_subbuf_size");
272 goto error_put_subbuf
;
275 read_len
= lttng_consumer_on_read_subbuffer_mmap(ctx
, stream
, len
,
276 padded_len
- len
, NULL
);
278 * We write the padded len in local tracefiles but the data len
279 * when using a relay. Display the error but continue processing
280 * to try to release the subbuffer.
282 if (relayd_id
!= (uint64_t) -1ULL) {
283 if (read_len
!= len
) {
284 ERR("Error sending to the relay (ret: %zd != len: %lu)",
288 if (read_len
!= padded_len
) {
289 ERR("Error writing to tracefile (ret: %zd != len: %lu)",
290 read_len
, padded_len
);
294 ret
= kernctl_put_subbuf(stream
->wait_fd
);
296 ERR("Snapshot kernctl_put_subbuf");
299 consumed_pos
+= stream
->max_sb_size
;
302 if (relayd_id
== (uint64_t) -1ULL) {
303 if (stream
->out_fd
>= 0) {
304 ret
= close(stream
->out_fd
);
306 PERROR("Kernel consumer snapshot close out_fd");
312 close_relayd_stream(stream
);
313 stream
->net_seq_idx
= (uint64_t) -1ULL;
315 pthread_mutex_unlock(&stream
->lock
);
323 ret
= kernctl_put_subbuf(stream
->wait_fd
);
325 ERR("Snapshot kernctl_put_subbuf error path");
328 pthread_mutex_unlock(&stream
->lock
);
335 * Read the whole metadata available for a snapshot.
337 * Returns 0 on success, < 0 on error
339 static int lttng_kconsumer_snapshot_metadata(uint64_t key
, char *path
,
340 uint64_t relayd_id
, struct lttng_consumer_local_data
*ctx
)
342 int ret
, use_relayd
= 0;
344 struct lttng_consumer_channel
*metadata_channel
;
345 struct lttng_consumer_stream
*metadata_stream
;
349 DBG("Kernel consumer snapshot metadata with key %" PRIu64
" at path %s",
354 metadata_channel
= consumer_find_channel(key
);
355 if (!metadata_channel
) {
356 ERR("Kernel snapshot metadata not found for key %" PRIu64
, key
);
358 goto error_no_channel
;
361 metadata_stream
= metadata_channel
->metadata_stream
;
362 assert(metadata_stream
);
363 pthread_mutex_lock(&metadata_stream
->lock
);
365 /* Flag once that we have a valid relayd for the stream. */
366 if (relayd_id
!= (uint64_t) -1ULL) {
371 ret
= consumer_send_relayd_stream(metadata_stream
, path
);
376 ret
= utils_create_stream_file(path
, metadata_stream
->name
,
377 metadata_stream
->chan
->tracefile_size
,
378 metadata_stream
->tracefile_count_current
,
379 metadata_stream
->uid
, metadata_stream
->gid
, NULL
);
383 metadata_stream
->out_fd
= ret
;
387 health_code_update();
389 ret_read
= lttng_kconsumer_read_subbuffer(metadata_stream
, ctx
);
391 if (ret_read
!= -EAGAIN
) {
392 ERR("Kernel snapshot reading metadata subbuffer (ret: %zd)",
397 /* ret_read is negative at this point so we will exit the loop. */
400 } while (ret_read
>= 0);
403 close_relayd_stream(metadata_stream
);
404 metadata_stream
->net_seq_idx
= (uint64_t) -1ULL;
406 if (metadata_stream
->out_fd
>= 0) {
407 ret
= close(metadata_stream
->out_fd
);
409 PERROR("Kernel consumer snapshot metadata close out_fd");
411 * Don't go on error here since the snapshot was successful at this
412 * point but somehow the close failed.
415 metadata_stream
->out_fd
= -1;
421 pthread_mutex_unlock(&metadata_stream
->lock
);
422 cds_list_del(&metadata_stream
->send_node
);
423 consumer_stream_destroy(metadata_stream
, NULL
);
424 metadata_channel
->metadata_stream
= NULL
;
431 * Receive command from session daemon and process it.
433 * Return 1 on success else a negative value or 0.
435 int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data
*ctx
,
436 int sock
, struct pollfd
*consumer_sockpoll
)
439 enum lttcomm_return_code ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
440 struct lttcomm_consumer_msg msg
;
442 health_code_update();
444 ret
= lttcomm_recv_unix_sock(sock
, &msg
, sizeof(msg
));
445 if (ret
!= sizeof(msg
)) {
447 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_ERROR_RECV_CMD
);
453 health_code_update();
455 /* Deprecated command */
456 assert(msg
.cmd_type
!= LTTNG_CONSUMER_STOP
);
458 health_code_update();
460 /* relayd needs RCU read-side protection */
463 switch (msg
.cmd_type
) {
464 case LTTNG_CONSUMER_ADD_RELAYD_SOCKET
:
466 /* Session daemon status message are handled in the following call. */
467 consumer_add_relayd_socket(msg
.u
.relayd_sock
.net_index
,
468 msg
.u
.relayd_sock
.type
, ctx
, sock
, consumer_sockpoll
,
469 &msg
.u
.relayd_sock
.sock
, msg
.u
.relayd_sock
.session_id
,
470 msg
.u
.relayd_sock
.relayd_session_id
);
473 case LTTNG_CONSUMER_ADD_CHANNEL
:
475 struct lttng_consumer_channel
*new_channel
;
478 health_code_update();
480 /* First send a status message before receiving the fds. */
481 ret
= consumer_send_status_msg(sock
, ret_code
);
483 /* Somehow, the session daemon is not responding anymore. */
487 health_code_update();
489 DBG("consumer_add_channel %" PRIu64
, msg
.u
.channel
.channel_key
);
490 new_channel
= consumer_allocate_channel(msg
.u
.channel
.channel_key
,
491 msg
.u
.channel
.session_id
, msg
.u
.channel
.pathname
,
492 msg
.u
.channel
.name
, msg
.u
.channel
.uid
, msg
.u
.channel
.gid
,
493 msg
.u
.channel
.relayd_id
, msg
.u
.channel
.output
,
494 msg
.u
.channel
.tracefile_size
,
495 msg
.u
.channel
.tracefile_count
, 0,
496 msg
.u
.channel
.monitor
,
497 msg
.u
.channel
.live_timer_interval
,
499 if (new_channel
== NULL
) {
500 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_OUTFD_ERROR
);
503 new_channel
->nb_init_stream_left
= msg
.u
.channel
.nb_init_streams
;
504 switch (msg
.u
.channel
.output
) {
505 case LTTNG_EVENT_SPLICE
:
506 new_channel
->output
= CONSUMER_CHANNEL_SPLICE
;
508 case LTTNG_EVENT_MMAP
:
509 new_channel
->output
= CONSUMER_CHANNEL_MMAP
;
512 ERR("Channel output unknown %d", msg
.u
.channel
.output
);
516 /* Translate and save channel type. */
517 switch (msg
.u
.channel
.type
) {
518 case CONSUMER_CHANNEL_TYPE_DATA
:
519 case CONSUMER_CHANNEL_TYPE_METADATA
:
520 new_channel
->type
= msg
.u
.channel
.type
;
527 health_code_update();
529 if (ctx
->on_recv_channel
!= NULL
) {
530 ret_recv
= ctx
->on_recv_channel(new_channel
);
532 ret
= consumer_add_channel(new_channel
, ctx
);
533 } else if (ret_recv
< 0) {
537 ret
= consumer_add_channel(new_channel
, ctx
);
539 if (msg
.u
.channel
.type
== CONSUMER_CHANNEL_TYPE_DATA
&& !ret
) {
540 int monitor_start_ret
;
542 DBG("Consumer starting monitor timer");
543 consumer_timer_live_start(new_channel
,
544 msg
.u
.channel
.live_timer_interval
);
545 monitor_start_ret
= consumer_timer_monitor_start(
547 msg
.u
.channel
.monitor_timer_interval
);
548 if (monitor_start_ret
< 0) {
549 ERR("Starting channel monitoring timer failed");
555 health_code_update();
557 /* If we received an error in add_channel, we need to report it. */
559 ret
= consumer_send_status_msg(sock
, ret
);
568 case LTTNG_CONSUMER_ADD_STREAM
:
571 struct lttng_pipe
*stream_pipe
;
572 struct lttng_consumer_stream
*new_stream
;
573 struct lttng_consumer_channel
*channel
;
577 * Get stream's channel reference. Needed when adding the stream to the
580 channel
= consumer_find_channel(msg
.u
.stream
.channel_key
);
583 * We could not find the channel. Can happen if cpu hotplug
584 * happens while tearing down.
586 ERR("Unable to find channel key %" PRIu64
, msg
.u
.stream
.channel_key
);
587 ret_code
= LTTCOMM_CONSUMERD_CHAN_NOT_FOUND
;
590 health_code_update();
592 /* First send a status message before receiving the fds. */
593 ret
= consumer_send_status_msg(sock
, ret_code
);
595 /* Somehow, the session daemon is not responding anymore. */
599 health_code_update();
601 if (ret_code
!= LTTCOMM_CONSUMERD_SUCCESS
) {
602 /* Channel was not found. */
608 ret
= lttng_consumer_poll_socket(consumer_sockpoll
);
614 health_code_update();
616 /* Get stream file descriptor from socket */
617 ret
= lttcomm_recv_fds_unix_sock(sock
, &fd
, 1);
618 if (ret
!= sizeof(fd
)) {
619 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_ERROR_RECV_FD
);
624 health_code_update();
627 * Send status code to session daemon only if the recv works. If the
628 * above recv() failed, the session daemon is notified through the
629 * error socket and the teardown is eventually done.
631 ret
= consumer_send_status_msg(sock
, ret_code
);
633 /* Somehow, the session daemon is not responding anymore. */
637 health_code_update();
639 new_stream
= consumer_allocate_stream(channel
->key
,
641 LTTNG_CONSUMER_ACTIVE_STREAM
,
651 if (new_stream
== NULL
) {
656 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_OUTFD_ERROR
);
662 new_stream
->chan
= channel
;
663 new_stream
->wait_fd
= fd
;
664 switch (channel
->output
) {
665 case CONSUMER_CHANNEL_SPLICE
:
666 new_stream
->output
= LTTNG_EVENT_SPLICE
;
667 ret
= utils_create_pipe(new_stream
->splice_pipe
);
672 case CONSUMER_CHANNEL_MMAP
:
673 new_stream
->output
= LTTNG_EVENT_MMAP
;
676 ERR("Stream output unknown %d", channel
->output
);
681 * We've just assigned the channel to the stream so increment the
682 * refcount right now. We don't need to increment the refcount for
683 * streams in no monitor because we handle manually the cleanup of
684 * those. It is very important to make sure there is NO prior
685 * consumer_del_stream() calls or else the refcount will be unbalanced.
687 if (channel
->monitor
) {
688 uatomic_inc(&new_stream
->chan
->refcount
);
692 * The buffer flush is done on the session daemon side for the kernel
693 * so no need for the stream "hangup_flush_done" variable to be
694 * tracked. This is important for a kernel stream since we don't rely
695 * on the flush state of the stream to read data. It's not the case for
696 * user space tracing.
698 new_stream
->hangup_flush_done
= 0;
700 health_code_update();
702 if (ctx
->on_recv_stream
) {
703 ret
= ctx
->on_recv_stream(new_stream
);
705 consumer_stream_free(new_stream
);
710 health_code_update();
712 if (new_stream
->metadata_flag
) {
713 channel
->metadata_stream
= new_stream
;
716 /* Do not monitor this stream. */
717 if (!channel
->monitor
) {
718 DBG("Kernel consumer add stream %s in no monitor mode with "
719 "relayd id %" PRIu64
, new_stream
->name
,
720 new_stream
->net_seq_idx
);
721 cds_list_add(&new_stream
->send_node
, &channel
->streams
.head
);
725 /* Send stream to relayd if the stream has an ID. */
726 if (new_stream
->net_seq_idx
!= (uint64_t) -1ULL) {
727 ret
= consumer_send_relayd_stream(new_stream
,
728 new_stream
->chan
->pathname
);
730 consumer_stream_free(new_stream
);
735 * If adding an extra stream to an already
736 * existing channel (e.g. cpu hotplug), we need
737 * to send the "streams_sent" command to relayd.
739 if (channel
->streams_sent_to_relayd
) {
740 ret
= consumer_send_relayd_streams_sent(
741 new_stream
->net_seq_idx
);
748 /* Get the right pipe where the stream will be sent. */
749 if (new_stream
->metadata_flag
) {
750 ret
= consumer_add_metadata_stream(new_stream
);
752 ERR("Consumer add metadata stream %" PRIu64
" failed. Continuing",
754 consumer_stream_free(new_stream
);
757 stream_pipe
= ctx
->consumer_metadata_pipe
;
759 ret
= consumer_add_data_stream(new_stream
);
761 ERR("Consumer add stream %" PRIu64
" failed. Continuing",
763 consumer_stream_free(new_stream
);
766 stream_pipe
= ctx
->consumer_data_pipe
;
769 /* Vitible to other threads */
770 new_stream
->globally_visible
= 1;
772 health_code_update();
774 ret
= lttng_pipe_write(stream_pipe
, &new_stream
, sizeof(new_stream
));
776 ERR("Consumer write %s stream to pipe %d",
777 new_stream
->metadata_flag
? "metadata" : "data",
778 lttng_pipe_get_writefd(stream_pipe
));
779 if (new_stream
->metadata_flag
) {
780 consumer_del_stream_for_metadata(new_stream
);
782 consumer_del_stream_for_data(new_stream
);
787 DBG("Kernel consumer ADD_STREAM %s (fd: %d) with relayd id %" PRIu64
,
788 new_stream
->name
, fd
, new_stream
->relayd_stream_id
);
791 case LTTNG_CONSUMER_STREAMS_SENT
:
793 struct lttng_consumer_channel
*channel
;
796 * Get stream's channel reference. Needed when adding the stream to the
799 channel
= consumer_find_channel(msg
.u
.sent_streams
.channel_key
);
802 * We could not find the channel. Can happen if cpu hotplug
803 * happens while tearing down.
805 ERR("Unable to find channel key %" PRIu64
,
806 msg
.u
.sent_streams
.channel_key
);
807 ret_code
= LTTCOMM_CONSUMERD_CHAN_NOT_FOUND
;
810 health_code_update();
813 * Send status code to session daemon.
815 ret
= consumer_send_status_msg(sock
, ret_code
);
816 if (ret
< 0 || ret_code
!= LTTCOMM_CONSUMERD_SUCCESS
) {
817 /* Somehow, the session daemon is not responding anymore. */
821 health_code_update();
824 * We should not send this message if we don't monitor the
825 * streams in this channel.
827 if (!channel
->monitor
) {
831 health_code_update();
832 /* Send stream to relayd if the stream has an ID. */
833 if (msg
.u
.sent_streams
.net_seq_idx
!= (uint64_t) -1ULL) {
834 ret
= consumer_send_relayd_streams_sent(
835 msg
.u
.sent_streams
.net_seq_idx
);
839 channel
->streams_sent_to_relayd
= true;
843 case LTTNG_CONSUMER_UPDATE_STREAM
:
848 case LTTNG_CONSUMER_DESTROY_RELAYD
:
850 uint64_t index
= msg
.u
.destroy_relayd
.net_seq_idx
;
851 struct consumer_relayd_sock_pair
*relayd
;
853 DBG("Kernel consumer destroying relayd %" PRIu64
, index
);
855 /* Get relayd reference if exists. */
856 relayd
= consumer_find_relayd(index
);
857 if (relayd
== NULL
) {
858 DBG("Unable to find relayd %" PRIu64
, index
);
859 ret_code
= LTTCOMM_CONSUMERD_RELAYD_FAIL
;
863 * Each relayd socket pair has a refcount of stream attached to it
864 * which tells if the relayd is still active or not depending on the
867 * This will set the destroy flag of the relayd object and destroy it
868 * if the refcount reaches zero when called.
870 * The destroy can happen either here or when a stream fd hangs up.
873 consumer_flag_relayd_for_destroy(relayd
);
876 health_code_update();
878 ret
= consumer_send_status_msg(sock
, ret_code
);
880 /* Somehow, the session daemon is not responding anymore. */
886 case LTTNG_CONSUMER_DATA_PENDING
:
889 uint64_t id
= msg
.u
.data_pending
.session_id
;
891 DBG("Kernel consumer data pending command for id %" PRIu64
, id
);
893 ret
= consumer_data_pending(id
);
895 health_code_update();
897 /* Send back returned value to session daemon */
898 ret
= lttcomm_send_unix_sock(sock
, &ret
, sizeof(ret
));
900 PERROR("send data pending ret code");
905 * No need to send back a status message since the data pending
906 * returned value is the response.
910 case LTTNG_CONSUMER_SNAPSHOT_CHANNEL
:
912 if (msg
.u
.snapshot_channel
.metadata
== 1) {
913 ret
= lttng_kconsumer_snapshot_metadata(msg
.u
.snapshot_channel
.key
,
914 msg
.u
.snapshot_channel
.pathname
,
915 msg
.u
.snapshot_channel
.relayd_id
, ctx
);
917 ERR("Snapshot metadata failed");
918 ret_code
= LTTCOMM_CONSUMERD_ERROR_METADATA
;
921 ret
= lttng_kconsumer_snapshot_channel(msg
.u
.snapshot_channel
.key
,
922 msg
.u
.snapshot_channel
.pathname
,
923 msg
.u
.snapshot_channel
.relayd_id
,
924 msg
.u
.snapshot_channel
.nb_packets_per_stream
,
927 ERR("Snapshot channel failed");
928 ret_code
= LTTCOMM_CONSUMERD_CHAN_NOT_FOUND
;
932 health_code_update();
934 ret
= consumer_send_status_msg(sock
, ret_code
);
936 /* Somehow, the session daemon is not responding anymore. */
941 case LTTNG_CONSUMER_DESTROY_CHANNEL
:
943 uint64_t key
= msg
.u
.destroy_channel
.key
;
944 struct lttng_consumer_channel
*channel
;
946 channel
= consumer_find_channel(key
);
948 ERR("Kernel consumer destroy channel %" PRIu64
" not found", key
);
949 ret_code
= LTTCOMM_CONSUMERD_CHAN_NOT_FOUND
;
952 health_code_update();
954 ret
= consumer_send_status_msg(sock
, ret_code
);
956 /* Somehow, the session daemon is not responding anymore. */
960 health_code_update();
962 /* Stop right now if no channel was found. */
968 * This command should ONLY be issued for channel with streams set in
971 assert(!channel
->monitor
);
974 * The refcount should ALWAYS be 0 in the case of a channel in no
977 assert(!uatomic_sub_return(&channel
->refcount
, 1));
979 consumer_del_channel(channel
);
983 case LTTNG_CONSUMER_DISCARDED_EVENTS
:
987 struct lttng_consumer_channel
*channel
;
988 uint64_t id
= msg
.u
.discarded_events
.session_id
;
989 uint64_t key
= msg
.u
.discarded_events
.channel_key
;
991 DBG("Kernel consumer discarded events command for session id %"
992 PRIu64
", channel key %" PRIu64
, id
, key
);
994 channel
= consumer_find_channel(key
);
996 ERR("Kernel consumer discarded events channel %"
997 PRIu64
" not found", key
);
1000 count
= channel
->discarded_events
;
1003 health_code_update();
1005 /* Send back returned value to session daemon */
1006 ret
= lttcomm_send_unix_sock(sock
, &count
, sizeof(count
));
1008 PERROR("send discarded events");
1014 case LTTNG_CONSUMER_LOST_PACKETS
:
1018 struct lttng_consumer_channel
*channel
;
1019 uint64_t id
= msg
.u
.lost_packets
.session_id
;
1020 uint64_t key
= msg
.u
.lost_packets
.channel_key
;
1022 DBG("Kernel consumer lost packets command for session id %"
1023 PRIu64
", channel key %" PRIu64
, id
, key
);
1025 channel
= consumer_find_channel(key
);
1027 ERR("Kernel consumer lost packets channel %"
1028 PRIu64
" not found", key
);
1031 count
= channel
->lost_packets
;
1034 health_code_update();
1036 /* Send back returned value to session daemon */
1037 ret
= lttcomm_send_unix_sock(sock
, &count
, sizeof(count
));
1039 PERROR("send lost packets");
1045 case LTTNG_CONSUMER_SET_CHANNEL_MONITOR_PIPE
:
1047 int channel_monitor_pipe
;
1049 ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
1050 /* Successfully received the command's type. */
1051 ret
= consumer_send_status_msg(sock
, ret_code
);
1056 ret
= lttcomm_recv_fds_unix_sock(sock
, &channel_monitor_pipe
,
1058 if (ret
!= sizeof(channel_monitor_pipe
)) {
1059 ERR("Failed to receive channel monitor pipe");
1063 DBG("Received channel monitor pipe (%d)", channel_monitor_pipe
);
1064 ret
= consumer_timer_thread_set_channel_monitor_pipe(
1065 channel_monitor_pipe
);
1069 ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
1070 /* Set the pipe as non-blocking. */
1071 ret
= fcntl(channel_monitor_pipe
, F_GETFL
, 0);
1073 PERROR("fcntl get flags of the channel monitoring pipe");
1078 ret
= fcntl(channel_monitor_pipe
, F_SETFL
,
1079 flags
| O_NONBLOCK
);
1081 PERROR("fcntl set O_NONBLOCK flag of the channel monitoring pipe");
1084 DBG("Channel monitor pipe set as non-blocking");
1086 ret_code
= LTTCOMM_CONSUMERD_ALREADY_SET
;
1088 ret
= consumer_send_status_msg(sock
, ret_code
);
1102 * Return 1 to indicate success since the 0 value can be a socket
1103 * shutdown during the recv() or send() call.
1105 health_code_update();
1110 /* This will issue a consumer stop. */
1115 * Populate index values of a kernel stream. Values are set in big endian order.
1117 * Return 0 on success or else a negative value.
1119 static int get_index_values(struct ctf_packet_index
*index
, int infd
)
1122 uint64_t packet_size
, content_size
, timestamp_begin
, timestamp_end
,
1123 events_discarded
, stream_id
, stream_instance_id
,
1126 ret
= kernctl_get_timestamp_begin(infd
, ×tamp_begin
);
1128 PERROR("kernctl_get_timestamp_begin");
1132 ret
= kernctl_get_timestamp_end(infd
, ×tamp_end
);
1134 PERROR("kernctl_get_timestamp_end");
1138 ret
= kernctl_get_events_discarded(infd
, &events_discarded
);
1140 PERROR("kernctl_get_events_discarded");
1144 ret
= kernctl_get_content_size(infd
, &content_size
);
1146 PERROR("kernctl_get_content_size");
1150 ret
= kernctl_get_packet_size(infd
, &packet_size
);
1152 PERROR("kernctl_get_packet_size");
1156 ret
= kernctl_get_stream_id(infd
, &stream_id
);
1158 PERROR("kernctl_get_stream_id");
1162 ret
= kernctl_get_instance_id(infd
, &stream_instance_id
);
1164 if (ret
== -ENOTTY
) {
1165 /* Command not implemented by lttng-modules. */
1166 stream_instance_id
= -1ULL;
1168 PERROR("kernctl_get_instance_id");
1173 ret
= kernctl_get_sequence_number(infd
, &packet_seq_num
);
1175 if (ret
== -ENOTTY
) {
1176 /* Command not implemented by lttng-modules. */
1177 packet_seq_num
= -1ULL;
1180 PERROR("kernctl_get_sequence_number");
1184 index
->packet_seq_num
= htobe64(index
->packet_seq_num
);
1186 *index
= (typeof(*index
)) {
1187 .offset
= index
->offset
,
1188 .packet_size
= htobe64(packet_size
),
1189 .content_size
= htobe64(content_size
),
1190 .timestamp_begin
= htobe64(timestamp_begin
),
1191 .timestamp_end
= htobe64(timestamp_end
),
1192 .events_discarded
= htobe64(events_discarded
),
1193 .stream_id
= htobe64(stream_id
),
1194 .stream_instance_id
= htobe64(stream_instance_id
),
1195 .packet_seq_num
= htobe64(packet_seq_num
),
1202 * Sync metadata meaning request them to the session daemon and snapshot to the
1203 * metadata thread can consumer them.
1205 * Metadata stream lock MUST be acquired.
1207 * Return 0 if new metadatda is available, EAGAIN if the metadata stream
1208 * is empty or a negative value on error.
1210 int lttng_kconsumer_sync_metadata(struct lttng_consumer_stream
*metadata
)
1216 ret
= kernctl_buffer_flush(metadata
->wait_fd
);
1218 ERR("Failed to flush kernel stream");
1222 ret
= kernctl_snapshot(metadata
->wait_fd
);
1224 if (ret
!= -EAGAIN
) {
1225 ERR("Sync metadata, taking kernel snapshot failed.");
1228 DBG("Sync metadata, no new kernel metadata");
1229 /* No new metadata, exit. */
1239 int update_stream_stats(struct lttng_consumer_stream
*stream
)
1242 uint64_t seq
, discarded
;
1244 ret
= kernctl_get_sequence_number(stream
->wait_fd
, &seq
);
1246 if (ret
== -ENOTTY
) {
1247 /* Command not implemented by lttng-modules. */
1250 PERROR("kernctl_get_sequence_number");
1256 * Start the sequence when we extract the first packet in case we don't
1257 * start at 0 (for example if a consumer is not connected to the
1258 * session immediately after the beginning).
1260 if (stream
->last_sequence_number
== -1ULL) {
1261 stream
->last_sequence_number
= seq
;
1262 } else if (seq
> stream
->last_sequence_number
) {
1263 stream
->chan
->lost_packets
+= seq
-
1264 stream
->last_sequence_number
- 1;
1266 /* seq <= last_sequence_number */
1267 ERR("Sequence number inconsistent : prev = %" PRIu64
1268 ", current = %" PRIu64
,
1269 stream
->last_sequence_number
, seq
);
1273 stream
->last_sequence_number
= seq
;
1275 ret
= kernctl_get_events_discarded(stream
->wait_fd
, &discarded
);
1277 PERROR("kernctl_get_events_discarded");
1280 if (discarded
< stream
->last_discarded_events
) {
1282 * Overflow has occurred. We assume only one wrap-around
1285 stream
->chan
->discarded_events
+= (1ULL << (CAA_BITS_PER_LONG
- 1)) -
1286 stream
->last_discarded_events
+ discarded
;
1288 stream
->chan
->discarded_events
+= discarded
-
1289 stream
->last_discarded_events
;
1291 stream
->last_discarded_events
= discarded
;
1299 * Check if the local version of the metadata stream matches with the version
1300 * of the metadata stream in the kernel. If it was updated, set the reset flag
1304 int metadata_stream_check_version(int infd
, struct lttng_consumer_stream
*stream
)
1307 uint64_t cur_version
;
1309 ret
= kernctl_get_metadata_version(infd
, &cur_version
);
1311 if (ret
== -ENOTTY
) {
1313 * LTTng-modules does not implement this
1319 ERR("Failed to get the metadata version");
1323 if (stream
->metadata_version
== cur_version
) {
1328 DBG("New metadata version detected");
1329 stream
->metadata_version
= cur_version
;
1330 stream
->reset_metadata_flag
= 1;
1338 * Consume data on a file descriptor and write it on a trace file.
1340 ssize_t
lttng_kconsumer_read_subbuffer(struct lttng_consumer_stream
*stream
,
1341 struct lttng_consumer_local_data
*ctx
)
1343 unsigned long len
, subbuf_size
, padding
;
1344 int err
, write_index
= 1;
1346 int infd
= stream
->wait_fd
;
1347 struct ctf_packet_index index
;
1349 DBG("In read_subbuffer (infd : %d)", infd
);
1351 /* Get the next subbuffer */
1352 err
= kernctl_get_next_subbuf(infd
);
1355 * This is a debug message even for single-threaded consumer,
1356 * because poll() have more relaxed criterions than get subbuf,
1357 * so get_subbuf may fail for short race windows where poll()
1358 * would issue wakeups.
1360 DBG("Reserving sub buffer failed (everything is normal, "
1361 "it is due to concurrency)");
1366 /* Get the full subbuffer size including padding */
1367 err
= kernctl_get_padded_subbuf_size(infd
, &len
);
1369 PERROR("Getting sub-buffer len failed.");
1370 err
= kernctl_put_subbuf(infd
);
1372 if (err
== -EFAULT
) {
1373 PERROR("Error in unreserving sub buffer\n");
1374 } else if (err
== -EIO
) {
1375 /* Should never happen with newer LTTng versions */
1376 PERROR("Reader has been pushed by the writer, last sub-buffer corrupted.");
1385 if (!stream
->metadata_flag
) {
1386 ret
= get_index_values(&index
, infd
);
1388 err
= kernctl_put_subbuf(infd
);
1390 if (err
== -EFAULT
) {
1391 PERROR("Error in unreserving sub buffer\n");
1392 } else if (err
== -EIO
) {
1393 /* Should never happen with newer LTTng versions */
1394 PERROR("Reader has been pushed by the writer, last sub-buffer corrupted.");
1401 ret
= update_stream_stats(stream
);
1403 err
= kernctl_put_subbuf(infd
);
1405 if (err
== -EFAULT
) {
1406 PERROR("Error in unreserving sub buffer\n");
1407 } else if (err
== -EIO
) {
1408 /* Should never happen with newer LTTng versions */
1409 PERROR("Reader has been pushed by the writer, last sub-buffer corrupted.");
1418 ret
= metadata_stream_check_version(infd
, stream
);
1420 err
= kernctl_put_subbuf(infd
);
1422 if (err
== -EFAULT
) {
1423 PERROR("Error in unreserving sub buffer\n");
1424 } else if (err
== -EIO
) {
1425 /* Should never happen with newer LTTng versions */
1426 PERROR("Reader has been pushed by the writer, last sub-buffer corrupted.");
1435 switch (stream
->chan
->output
) {
1436 case CONSUMER_CHANNEL_SPLICE
:
1438 * XXX: The lttng-modules splice "actor" does not handle copying
1439 * partial pages hence only using the subbuffer size without the
1440 * padding makes the splice fail.
1445 /* splice the subbuffer to the tracefile */
1446 ret
= lttng_consumer_on_read_subbuffer_splice(ctx
, stream
, subbuf_size
,
1449 * XXX: Splice does not support network streaming so the return value
1450 * is simply checked against subbuf_size and not like the mmap() op.
1452 if (ret
!= subbuf_size
) {
1454 * display the error but continue processing to try
1455 * to release the subbuffer
1457 ERR("Error splicing to tracefile (ret: %zd != len: %lu)",
1462 case CONSUMER_CHANNEL_MMAP
:
1463 /* Get subbuffer size without padding */
1464 err
= kernctl_get_subbuf_size(infd
, &subbuf_size
);
1466 PERROR("Getting sub-buffer len failed.");
1467 err
= kernctl_put_subbuf(infd
);
1469 if (err
== -EFAULT
) {
1470 PERROR("Error in unreserving sub buffer\n");
1471 } else if (err
== -EIO
) {
1472 /* Should never happen with newer LTTng versions */
1473 PERROR("Reader has been pushed by the writer, last sub-buffer corrupted.");
1482 /* Make sure the tracer is not gone mad on us! */
1483 assert(len
>= subbuf_size
);
1485 padding
= len
- subbuf_size
;
1487 /* write the subbuffer to the tracefile */
1488 ret
= lttng_consumer_on_read_subbuffer_mmap(ctx
, stream
, subbuf_size
,
1491 * The mmap operation should write subbuf_size amount of data when
1492 * network streaming or the full padding (len) size when we are _not_
1495 if ((ret
!= subbuf_size
&& stream
->net_seq_idx
!= (uint64_t) -1ULL) ||
1496 (ret
!= len
&& stream
->net_seq_idx
== (uint64_t) -1ULL)) {
1498 * Display the error but continue processing to try to release the
1499 * subbuffer. This is a DBG statement since this is possible to
1500 * happen without being a critical error.
1502 DBG("Error writing to tracefile "
1503 "(ret: %zd != len: %lu != subbuf_size: %lu)",
1504 ret
, len
, subbuf_size
);
1509 ERR("Unknown output method");
1513 err
= kernctl_put_next_subbuf(infd
);
1515 if (err
== -EFAULT
) {
1516 PERROR("Error in unreserving sub buffer\n");
1517 } else if (err
== -EIO
) {
1518 /* Should never happen with newer LTTng versions */
1519 PERROR("Reader has been pushed by the writer, last sub-buffer corrupted.");
1525 /* Write index if needed. */
1530 if (stream
->chan
->live_timer_interval
&& !stream
->metadata_flag
) {
1532 * In live, block until all the metadata is sent.
1534 pthread_mutex_lock(&stream
->metadata_timer_lock
);
1535 assert(!stream
->missed_metadata_flush
);
1536 stream
->waiting_on_metadata
= true;
1537 pthread_mutex_unlock(&stream
->metadata_timer_lock
);
1539 err
= consumer_stream_sync_metadata(ctx
, stream
->session_id
);
1541 pthread_mutex_lock(&stream
->metadata_timer_lock
);
1542 stream
->waiting_on_metadata
= false;
1543 if (stream
->missed_metadata_flush
) {
1544 stream
->missed_metadata_flush
= false;
1545 pthread_mutex_unlock(&stream
->metadata_timer_lock
);
1546 (void) consumer_flush_kernel_index(stream
);
1548 pthread_mutex_unlock(&stream
->metadata_timer_lock
);
1555 err
= consumer_stream_write_index(stream
, &index
);
1564 int lttng_kconsumer_on_recv_stream(struct lttng_consumer_stream
*stream
)
1571 * Don't create anything if this is set for streaming or should not be
1574 if (stream
->net_seq_idx
== (uint64_t) -1ULL && stream
->chan
->monitor
) {
1575 ret
= utils_create_stream_file(stream
->chan
->pathname
, stream
->name
,
1576 stream
->chan
->tracefile_size
, stream
->tracefile_count_current
,
1577 stream
->uid
, stream
->gid
, NULL
);
1581 stream
->out_fd
= ret
;
1582 stream
->tracefile_size_current
= 0;
1584 if (!stream
->metadata_flag
) {
1585 struct lttng_index_file
*index_file
;
1587 index_file
= lttng_index_file_create(stream
->chan
->pathname
,
1588 stream
->name
, stream
->uid
, stream
->gid
,
1589 stream
->chan
->tracefile_size
,
1590 stream
->tracefile_count_current
,
1591 CTF_INDEX_MAJOR
, CTF_INDEX_MINOR
);
1595 assert(!stream
->index_file
);
1596 stream
->index_file
= index_file
;
1600 if (stream
->output
== LTTNG_EVENT_MMAP
) {
1601 /* get the len of the mmap region */
1602 unsigned long mmap_len
;
1604 ret
= kernctl_get_mmap_len(stream
->wait_fd
, &mmap_len
);
1606 PERROR("kernctl_get_mmap_len");
1607 goto error_close_fd
;
1609 stream
->mmap_len
= (size_t) mmap_len
;
1611 stream
->mmap_base
= mmap(NULL
, stream
->mmap_len
, PROT_READ
,
1612 MAP_PRIVATE
, stream
->wait_fd
, 0);
1613 if (stream
->mmap_base
== MAP_FAILED
) {
1614 PERROR("Error mmaping");
1616 goto error_close_fd
;
1620 /* we return 0 to let the library handle the FD internally */
1624 if (stream
->out_fd
>= 0) {
1627 err
= close(stream
->out_fd
);
1629 stream
->out_fd
= -1;
1636 * Check if data is still being extracted from the buffers for a specific
1637 * stream. Consumer data lock MUST be acquired before calling this function
1638 * and the stream lock.
1640 * Return 1 if the traced data are still getting read else 0 meaning that the
1641 * data is available for trace viewer reading.
1643 int lttng_kconsumer_data_pending(struct lttng_consumer_stream
*stream
)
1649 if (stream
->endpoint_status
!= CONSUMER_ENDPOINT_ACTIVE
) {
1654 ret
= kernctl_get_next_subbuf(stream
->wait_fd
);
1656 /* There is still data so let's put back this subbuffer. */
1657 ret
= kernctl_put_subbuf(stream
->wait_fd
);
1659 ret
= 1; /* Data is pending */
1663 /* Data is NOT pending and ready to be read. */