2 * Copyright (C) 2011 - Julien Desfossez <julien.desfossez@polymtl.ca>
3 * Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License, version 2 only,
7 * as published by the Free Software Foundation.
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License along
15 * with this program; if not, write to the Free Software Foundation, Inc.,
16 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
26 #include <sys/socket.h>
27 #include <sys/types.h>
32 #include <bin/lttng-consumerd/health-consumerd.h>
33 #include <common/common.h>
34 #include <common/kernel-ctl/kernel-ctl.h>
35 #include <common/sessiond-comm/sessiond-comm.h>
36 #include <common/sessiond-comm/relayd.h>
37 #include <common/compat/fcntl.h>
38 #include <common/compat/endian.h>
39 #include <common/pipe.h>
40 #include <common/relayd/relayd.h>
41 #include <common/utils.h>
42 #include <common/consumer/consumer-stream.h>
43 #include <common/index/index.h>
44 #include <common/consumer/consumer-timer.h>
46 #include "kernel-consumer.h"
48 extern struct lttng_consumer_global_data consumer_data
;
49 extern int consumer_poll_timeout
;
50 extern volatile int consumer_quit
;
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 PERROR("Getting sub-buffer snapshot.");
72 * Get the produced position
74 * Returns 0 on success, < 0 on error
76 int lttng_kconsumer_get_produced_snapshot(struct lttng_consumer_stream
*stream
,
80 int infd
= stream
->wait_fd
;
82 ret
= kernctl_snapshot_get_produced(infd
, pos
);
84 PERROR("kernctl_snapshot_get_produced");
92 * Get the consumerd position
94 * Returns 0 on success, < 0 on error
96 int lttng_kconsumer_get_consumed_snapshot(struct lttng_consumer_stream
*stream
,
100 int infd
= stream
->wait_fd
;
102 ret
= kernctl_snapshot_get_consumed(infd
, pos
);
104 PERROR("kernctl_snapshot_get_consumed");
112 * Take a snapshot of all the stream of a channel
114 * Returns 0 on success, < 0 on error
116 int lttng_kconsumer_snapshot_channel(uint64_t key
, char *path
,
117 uint64_t relayd_id
, uint64_t nb_packets_per_stream
,
118 struct lttng_consumer_local_data
*ctx
)
121 unsigned long consumed_pos
, produced_pos
;
122 struct lttng_consumer_channel
*channel
;
123 struct lttng_consumer_stream
*stream
;
125 DBG("Kernel consumer snapshot channel %" PRIu64
, key
);
129 channel
= consumer_find_channel(key
);
131 ERR("No channel found for key %" PRIu64
, key
);
136 /* Splice is not supported yet for channel snapshot. */
137 if (channel
->output
!= CONSUMER_CHANNEL_MMAP
) {
138 ERR("Unsupported output %d", channel
->output
);
143 cds_list_for_each_entry(stream
, &channel
->streams
.head
, send_node
) {
145 health_code_update();
148 * Lock stream because we are about to change its state.
150 pthread_mutex_lock(&stream
->lock
);
153 * Assign the received relayd ID so we can use it for streaming. The streams
154 * are not visible to anyone so this is OK to change it.
156 stream
->net_seq_idx
= relayd_id
;
157 channel
->relayd_id
= relayd_id
;
158 if (relayd_id
!= (uint64_t) -1ULL) {
159 ret
= consumer_send_relayd_stream(stream
, path
);
161 ERR("sending stream to relayd");
165 ret
= utils_create_stream_file(path
, stream
->name
,
166 stream
->chan
->tracefile_size
,
167 stream
->tracefile_count_current
,
168 stream
->uid
, stream
->gid
, NULL
);
170 ERR("utils_create_stream_file");
174 stream
->out_fd
= ret
;
175 stream
->tracefile_size_current
= 0;
177 DBG("Kernel consumer snapshot stream %s/%s (%" PRIu64
")",
178 path
, stream
->name
, stream
->key
);
180 if (relayd_id
!= -1ULL) {
181 ret
= consumer_send_relayd_streams_sent(relayd_id
);
183 ERR("sending streams sent to relayd");
188 ret
= kernctl_buffer_flush(stream
->wait_fd
);
190 ERR("Failed to flush kernel stream");
195 ret
= lttng_kconsumer_take_snapshot(stream
);
197 ERR("Taking kernel snapshot");
201 ret
= lttng_kconsumer_get_produced_snapshot(stream
, &produced_pos
);
203 ERR("Produced kernel snapshot position");
207 ret
= lttng_kconsumer_get_consumed_snapshot(stream
, &consumed_pos
);
209 ERR("Consumerd kernel snapshot position");
213 if (stream
->max_sb_size
== 0) {
214 ret
= kernctl_get_max_subbuf_size(stream
->wait_fd
,
215 &stream
->max_sb_size
);
217 ERR("Getting kernel max_sb_size");
223 consumed_pos
= consumer_get_consume_start_pos(consumed_pos
,
224 produced_pos
, nb_packets_per_stream
,
225 stream
->max_sb_size
);
227 while (consumed_pos
< produced_pos
) {
229 unsigned long len
, padded_len
;
231 health_code_update();
233 DBG("Kernel consumer taking snapshot at pos %lu", consumed_pos
);
235 ret
= kernctl_get_subbuf(stream
->wait_fd
, &consumed_pos
);
237 if (errno
!= EAGAIN
) {
238 PERROR("kernctl_get_subbuf snapshot");
242 DBG("Kernel consumer get subbuf failed. Skipping it.");
243 consumed_pos
+= stream
->max_sb_size
;
247 ret
= kernctl_get_subbuf_size(stream
->wait_fd
, &len
);
249 ERR("Snapshot kernctl_get_subbuf_size");
251 goto error_put_subbuf
;
254 ret
= kernctl_get_padded_subbuf_size(stream
->wait_fd
, &padded_len
);
256 ERR("Snapshot kernctl_get_padded_subbuf_size");
258 goto error_put_subbuf
;
261 read_len
= lttng_consumer_on_read_subbuffer_mmap(ctx
, stream
, len
,
262 padded_len
- len
, NULL
);
264 * We write the padded len in local tracefiles but the data len
265 * when using a relay. Display the error but continue processing
266 * to try to release the subbuffer.
268 if (relayd_id
!= (uint64_t) -1ULL) {
269 if (read_len
!= len
) {
270 ERR("Error sending to the relay (ret: %zd != len: %lu)",
274 if (read_len
!= padded_len
) {
275 ERR("Error writing to tracefile (ret: %zd != len: %lu)",
276 read_len
, padded_len
);
280 ret
= kernctl_put_subbuf(stream
->wait_fd
);
282 ERR("Snapshot kernctl_put_subbuf");
286 consumed_pos
+= stream
->max_sb_size
;
289 if (relayd_id
== (uint64_t) -1ULL) {
290 if (stream
->out_fd
>= 0) {
291 ret
= close(stream
->out_fd
);
293 PERROR("Kernel consumer snapshot close out_fd");
299 close_relayd_stream(stream
);
300 stream
->net_seq_idx
= (uint64_t) -1ULL;
302 pthread_mutex_unlock(&stream
->lock
);
310 ret
= kernctl_put_subbuf(stream
->wait_fd
);
313 ERR("Snapshot kernctl_put_subbuf error path");
316 pthread_mutex_unlock(&stream
->lock
);
323 * Read the whole metadata available for a snapshot.
325 * Returns 0 on success, < 0 on error
327 int lttng_kconsumer_snapshot_metadata(uint64_t key
, char *path
,
328 uint64_t relayd_id
, struct lttng_consumer_local_data
*ctx
)
330 int ret
, use_relayd
= 0;
332 struct lttng_consumer_channel
*metadata_channel
;
333 struct lttng_consumer_stream
*metadata_stream
;
337 DBG("Kernel consumer snapshot metadata with key %" PRIu64
" at path %s",
342 metadata_channel
= consumer_find_channel(key
);
343 if (!metadata_channel
) {
344 ERR("Kernel snapshot metadata not found for key %" PRIu64
, key
);
349 metadata_stream
= metadata_channel
->metadata_stream
;
350 assert(metadata_stream
);
352 /* Flag once that we have a valid relayd for the stream. */
353 if (relayd_id
!= (uint64_t) -1ULL) {
358 ret
= consumer_send_relayd_stream(metadata_stream
, path
);
363 ret
= utils_create_stream_file(path
, metadata_stream
->name
,
364 metadata_stream
->chan
->tracefile_size
,
365 metadata_stream
->tracefile_count_current
,
366 metadata_stream
->uid
, metadata_stream
->gid
, NULL
);
370 metadata_stream
->out_fd
= ret
;
374 health_code_update();
376 ret_read
= lttng_kconsumer_read_subbuffer(metadata_stream
, ctx
);
378 if (ret_read
!= -EAGAIN
) {
379 ERR("Kernel snapshot reading metadata subbuffer (ret: %zd)",
383 /* ret_read is negative at this point so we will exit the loop. */
386 } while (ret_read
>= 0);
389 close_relayd_stream(metadata_stream
);
390 metadata_stream
->net_seq_idx
= (uint64_t) -1ULL;
392 if (metadata_stream
->out_fd
>= 0) {
393 ret
= close(metadata_stream
->out_fd
);
395 PERROR("Kernel consumer snapshot metadata close out_fd");
397 * Don't go on error here since the snapshot was successful at this
398 * point but somehow the close failed.
401 metadata_stream
->out_fd
= -1;
407 cds_list_del(&metadata_stream
->send_node
);
408 consumer_stream_destroy(metadata_stream
, NULL
);
409 metadata_channel
->metadata_stream
= NULL
;
416 * Receive command from session daemon and process it.
418 * Return 1 on success else a negative value or 0.
420 int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data
*ctx
,
421 int sock
, struct pollfd
*consumer_sockpoll
)
424 enum lttcomm_return_code ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
425 struct lttcomm_consumer_msg msg
;
427 health_code_update();
429 ret
= lttcomm_recv_unix_sock(sock
, &msg
, sizeof(msg
));
430 if (ret
!= sizeof(msg
)) {
432 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_ERROR_RECV_CMD
);
438 health_code_update();
440 /* Deprecated command */
441 assert(msg
.cmd_type
!= LTTNG_CONSUMER_STOP
);
443 health_code_update();
445 /* relayd needs RCU read-side protection */
448 switch (msg
.cmd_type
) {
449 case LTTNG_CONSUMER_ADD_RELAYD_SOCKET
:
451 /* Session daemon status message are handled in the following call. */
452 ret
= consumer_add_relayd_socket(msg
.u
.relayd_sock
.net_index
,
453 msg
.u
.relayd_sock
.type
, ctx
, sock
, consumer_sockpoll
,
454 &msg
.u
.relayd_sock
.sock
, msg
.u
.relayd_sock
.session_id
,
455 msg
.u
.relayd_sock
.relayd_session_id
);
458 case LTTNG_CONSUMER_ADD_CHANNEL
:
460 struct lttng_consumer_channel
*new_channel
;
463 health_code_update();
465 /* First send a status message before receiving the fds. */
466 ret
= consumer_send_status_msg(sock
, ret_code
);
468 /* Somehow, the session daemon is not responding anymore. */
472 health_code_update();
474 DBG("consumer_add_channel %" PRIu64
, msg
.u
.channel
.channel_key
);
475 new_channel
= consumer_allocate_channel(msg
.u
.channel
.channel_key
,
476 msg
.u
.channel
.session_id
, msg
.u
.channel
.pathname
,
477 msg
.u
.channel
.name
, msg
.u
.channel
.uid
, msg
.u
.channel
.gid
,
478 msg
.u
.channel
.relayd_id
, msg
.u
.channel
.output
,
479 msg
.u
.channel
.tracefile_size
,
480 msg
.u
.channel
.tracefile_count
, 0,
481 msg
.u
.channel
.monitor
,
482 msg
.u
.channel
.live_timer_interval
,
484 if (new_channel
== NULL
) {
485 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_OUTFD_ERROR
);
488 new_channel
->nb_init_stream_left
= msg
.u
.channel
.nb_init_streams
;
489 switch (msg
.u
.channel
.output
) {
490 case LTTNG_EVENT_SPLICE
:
491 new_channel
->output
= CONSUMER_CHANNEL_SPLICE
;
493 case LTTNG_EVENT_MMAP
:
494 new_channel
->output
= CONSUMER_CHANNEL_MMAP
;
497 ERR("Channel output unknown %d", msg
.u
.channel
.output
);
501 /* Translate and save channel type. */
502 switch (msg
.u
.channel
.type
) {
503 case CONSUMER_CHANNEL_TYPE_DATA
:
504 case CONSUMER_CHANNEL_TYPE_METADATA
:
505 new_channel
->type
= msg
.u
.channel
.type
;
512 health_code_update();
514 if (ctx
->on_recv_channel
!= NULL
) {
515 ret_recv
= ctx
->on_recv_channel(new_channel
);
517 ret
= consumer_add_channel(new_channel
, ctx
);
518 } else if (ret_recv
< 0) {
522 ret
= consumer_add_channel(new_channel
, ctx
);
524 if (CONSUMER_CHANNEL_TYPE_DATA
) {
525 consumer_timer_live_start(new_channel
,
526 msg
.u
.channel
.live_timer_interval
);
529 health_code_update();
531 /* If we received an error in add_channel, we need to report it. */
533 ret
= consumer_send_status_msg(sock
, ret
);
542 case LTTNG_CONSUMER_ADD_STREAM
:
545 struct lttng_pipe
*stream_pipe
;
546 struct lttng_consumer_stream
*new_stream
;
547 struct lttng_consumer_channel
*channel
;
551 * Get stream's channel reference. Needed when adding the stream to the
554 channel
= consumer_find_channel(msg
.u
.stream
.channel_key
);
557 * We could not find the channel. Can happen if cpu hotplug
558 * happens while tearing down.
560 ERR("Unable to find channel key %" PRIu64
, msg
.u
.stream
.channel_key
);
561 ret_code
= LTTCOMM_CONSUMERD_CHAN_NOT_FOUND
;
564 health_code_update();
566 /* First send a status message before receiving the fds. */
567 ret
= consumer_send_status_msg(sock
, ret_code
);
569 /* Somehow, the session daemon is not responding anymore. */
573 health_code_update();
575 if (ret_code
!= LTTCOMM_CONSUMERD_SUCCESS
) {
576 /* Channel was not found. */
582 ret
= lttng_consumer_poll_socket(consumer_sockpoll
);
588 health_code_update();
590 /* Get stream file descriptor from socket */
591 ret
= lttcomm_recv_fds_unix_sock(sock
, &fd
, 1);
592 if (ret
!= sizeof(fd
)) {
593 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_ERROR_RECV_FD
);
598 health_code_update();
601 * Send status code to session daemon only if the recv works. If the
602 * above recv() failed, the session daemon is notified through the
603 * error socket and the teardown is eventually done.
605 ret
= consumer_send_status_msg(sock
, ret_code
);
607 /* Somehow, the session daemon is not responding anymore. */
611 health_code_update();
613 new_stream
= consumer_allocate_stream(channel
->key
,
615 LTTNG_CONSUMER_ACTIVE_STREAM
,
625 if (new_stream
== NULL
) {
630 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_OUTFD_ERROR
);
636 new_stream
->chan
= channel
;
637 new_stream
->wait_fd
= fd
;
638 switch (channel
->output
) {
639 case CONSUMER_CHANNEL_SPLICE
:
640 new_stream
->output
= LTTNG_EVENT_SPLICE
;
641 ret
= utils_create_pipe(new_stream
->splice_pipe
);
646 case CONSUMER_CHANNEL_MMAP
:
647 new_stream
->output
= LTTNG_EVENT_MMAP
;
650 ERR("Stream output unknown %d", channel
->output
);
655 * We've just assigned the channel to the stream so increment the
656 * refcount right now. We don't need to increment the refcount for
657 * streams in no monitor because we handle manually the cleanup of
658 * those. It is very important to make sure there is NO prior
659 * consumer_del_stream() calls or else the refcount will be unbalanced.
661 if (channel
->monitor
) {
662 uatomic_inc(&new_stream
->chan
->refcount
);
666 * The buffer flush is done on the session daemon side for the kernel
667 * so no need for the stream "hangup_flush_done" variable to be
668 * tracked. This is important for a kernel stream since we don't rely
669 * on the flush state of the stream to read data. It's not the case for
670 * user space tracing.
672 new_stream
->hangup_flush_done
= 0;
674 health_code_update();
676 if (ctx
->on_recv_stream
) {
677 ret
= ctx
->on_recv_stream(new_stream
);
679 consumer_stream_free(new_stream
);
684 health_code_update();
686 if (new_stream
->metadata_flag
) {
687 channel
->metadata_stream
= new_stream
;
690 /* Do not monitor this stream. */
691 if (!channel
->monitor
) {
692 DBG("Kernel consumer add stream %s in no monitor mode with "
693 "relayd id %" PRIu64
, new_stream
->name
,
694 new_stream
->net_seq_idx
);
695 cds_list_add(&new_stream
->send_node
, &channel
->streams
.head
);
699 /* Send stream to relayd if the stream has an ID. */
700 if (new_stream
->net_seq_idx
!= (uint64_t) -1ULL) {
701 ret
= consumer_send_relayd_stream(new_stream
,
702 new_stream
->chan
->pathname
);
704 consumer_stream_free(new_stream
);
709 /* Get the right pipe where the stream will be sent. */
710 if (new_stream
->metadata_flag
) {
711 ret
= consumer_add_metadata_stream(new_stream
);
713 ERR("Consumer add metadata stream %" PRIu64
" failed. Continuing",
715 consumer_stream_free(new_stream
);
718 stream_pipe
= ctx
->consumer_metadata_pipe
;
720 ret
= consumer_add_data_stream(new_stream
);
722 ERR("Consumer add stream %" PRIu64
" failed. Continuing",
724 consumer_stream_free(new_stream
);
727 stream_pipe
= ctx
->consumer_data_pipe
;
730 /* Vitible to other threads */
731 new_stream
->globally_visible
= 1;
733 health_code_update();
735 ret
= lttng_pipe_write(stream_pipe
, &new_stream
, sizeof(new_stream
));
737 ERR("Consumer write %s stream to pipe %d",
738 new_stream
->metadata_flag
? "metadata" : "data",
739 lttng_pipe_get_writefd(stream_pipe
));
740 if (new_stream
->metadata_flag
) {
741 consumer_del_stream_for_metadata(new_stream
);
743 consumer_del_stream_for_data(new_stream
);
748 DBG("Kernel consumer ADD_STREAM %s (fd: %d) with relayd id %" PRIu64
,
749 new_stream
->name
, fd
, new_stream
->relayd_stream_id
);
752 case LTTNG_CONSUMER_STREAMS_SENT
:
754 struct lttng_consumer_channel
*channel
;
757 * Get stream's channel reference. Needed when adding the stream to the
760 channel
= consumer_find_channel(msg
.u
.sent_streams
.channel_key
);
763 * We could not find the channel. Can happen if cpu hotplug
764 * happens while tearing down.
766 ERR("Unable to find channel key %" PRIu64
,
767 msg
.u
.sent_streams
.channel_key
);
768 ret_code
= LTTCOMM_CONSUMERD_CHAN_NOT_FOUND
;
771 health_code_update();
774 * Send status code to session daemon.
776 ret
= consumer_send_status_msg(sock
, ret_code
);
777 if (ret
< 0 || ret_code
!= LTTCOMM_CONSUMERD_SUCCESS
) {
778 /* Somehow, the session daemon is not responding anymore. */
782 health_code_update();
785 * We should not send this message if we don't monitor the
786 * streams in this channel.
788 if (!channel
->monitor
) {
792 health_code_update();
793 /* Send stream to relayd if the stream has an ID. */
794 if (msg
.u
.sent_streams
.net_seq_idx
!= (uint64_t) -1ULL) {
795 ret
= consumer_send_relayd_streams_sent(
796 msg
.u
.sent_streams
.net_seq_idx
);
803 case LTTNG_CONSUMER_UPDATE_STREAM
:
808 case LTTNG_CONSUMER_DESTROY_RELAYD
:
810 uint64_t index
= msg
.u
.destroy_relayd
.net_seq_idx
;
811 struct consumer_relayd_sock_pair
*relayd
;
813 DBG("Kernel consumer destroying relayd %" PRIu64
, index
);
815 /* Get relayd reference if exists. */
816 relayd
= consumer_find_relayd(index
);
817 if (relayd
== NULL
) {
818 DBG("Unable to find relayd %" PRIu64
, index
);
819 ret_code
= LTTCOMM_CONSUMERD_RELAYD_FAIL
;
823 * Each relayd socket pair has a refcount of stream attached to it
824 * which tells if the relayd is still active or not depending on the
827 * This will set the destroy flag of the relayd object and destroy it
828 * if the refcount reaches zero when called.
830 * The destroy can happen either here or when a stream fd hangs up.
833 consumer_flag_relayd_for_destroy(relayd
);
836 health_code_update();
838 ret
= consumer_send_status_msg(sock
, ret_code
);
840 /* Somehow, the session daemon is not responding anymore. */
846 case LTTNG_CONSUMER_DATA_PENDING
:
849 uint64_t id
= msg
.u
.data_pending
.session_id
;
851 DBG("Kernel consumer data pending command for id %" PRIu64
, id
);
853 ret
= consumer_data_pending(id
);
855 health_code_update();
857 /* Send back returned value to session daemon */
858 ret
= lttcomm_send_unix_sock(sock
, &ret
, sizeof(ret
));
860 PERROR("send data pending ret code");
865 * No need to send back a status message since the data pending
866 * returned value is the response.
870 case LTTNG_CONSUMER_SNAPSHOT_CHANNEL
:
872 if (msg
.u
.snapshot_channel
.metadata
== 1) {
873 ret
= lttng_kconsumer_snapshot_metadata(msg
.u
.snapshot_channel
.key
,
874 msg
.u
.snapshot_channel
.pathname
,
875 msg
.u
.snapshot_channel
.relayd_id
, ctx
);
877 ERR("Snapshot metadata failed");
878 ret_code
= LTTCOMM_CONSUMERD_ERROR_METADATA
;
881 ret
= lttng_kconsumer_snapshot_channel(msg
.u
.snapshot_channel
.key
,
882 msg
.u
.snapshot_channel
.pathname
,
883 msg
.u
.snapshot_channel
.relayd_id
,
884 msg
.u
.snapshot_channel
.nb_packets_per_stream
,
887 ERR("Snapshot channel failed");
888 ret_code
= LTTCOMM_CONSUMERD_CHAN_NOT_FOUND
;
892 health_code_update();
894 ret
= consumer_send_status_msg(sock
, ret_code
);
896 /* Somehow, the session daemon is not responding anymore. */
901 case LTTNG_CONSUMER_DESTROY_CHANNEL
:
903 uint64_t key
= msg
.u
.destroy_channel
.key
;
904 struct lttng_consumer_channel
*channel
;
906 channel
= consumer_find_channel(key
);
908 ERR("Kernel consumer destroy channel %" PRIu64
" not found", key
);
909 ret_code
= LTTCOMM_CONSUMERD_CHAN_NOT_FOUND
;
912 health_code_update();
914 ret
= consumer_send_status_msg(sock
, ret_code
);
916 /* Somehow, the session daemon is not responding anymore. */
920 health_code_update();
922 /* Stop right now if no channel was found. */
928 * This command should ONLY be issued for channel with streams set in
931 assert(!channel
->monitor
);
934 * The refcount should ALWAYS be 0 in the case of a channel in no
937 assert(!uatomic_sub_return(&channel
->refcount
, 1));
939 consumer_del_channel(channel
);
943 case LTTNG_CONSUMER_DISCARDED_EVENTS
:
946 struct lttng_consumer_channel
*channel
;
947 uint64_t id
= msg
.u
.discarded_events
.session_id
;
948 uint64_t key
= msg
.u
.discarded_events
.channel_key
;
950 channel
= consumer_find_channel(key
);
952 ERR("Kernel consumer discarded events channel %"
953 PRIu64
" not found", key
);
954 ret_code
= LTTCOMM_CONSUMERD_CHAN_NOT_FOUND
;
957 DBG("Kernel consumer discarded events command for session id %"
958 PRIu64
", channel key %" PRIu64
, id
, key
);
960 ret
= channel
->discarded_events
;
962 health_code_update();
964 /* Send back returned value to session daemon */
965 ret
= lttcomm_send_unix_sock(sock
, &ret
, sizeof(ret
));
967 PERROR("send discarded events");
973 case LTTNG_CONSUMER_LOST_PACKETS
:
976 struct lttng_consumer_channel
*channel
;
977 uint64_t id
= msg
.u
.lost_packets
.session_id
;
978 uint64_t key
= msg
.u
.lost_packets
.channel_key
;
980 channel
= consumer_find_channel(key
);
982 ERR("Kernel consumer lost packets channel %"
983 PRIu64
" not found", key
);
984 ret_code
= LTTCOMM_CONSUMERD_CHAN_NOT_FOUND
;
987 DBG("Kernel consumer lost packets command for session id %"
988 PRIu64
", channel key %" PRIu64
, id
, key
);
990 ret
= channel
->lost_packets
;
992 health_code_update();
994 /* Send back returned value to session daemon */
995 ret
= lttcomm_send_unix_sock(sock
, &ret
, sizeof(ret
));
997 PERROR("send lost packets");
1011 * Return 1 to indicate success since the 0 value can be a socket
1012 * shutdown during the recv() or send() call.
1014 health_code_update();
1019 /* This will issue a consumer stop. */
1024 * Populate index values of a kernel stream. Values are set in big endian order.
1026 * Return 0 on success or else a negative value.
1028 static int get_index_values(struct ctf_packet_index
*index
, int infd
)
1032 ret
= kernctl_get_timestamp_begin(infd
, &index
->timestamp_begin
);
1034 PERROR("kernctl_get_timestamp_begin");
1037 index
->timestamp_begin
= htobe64(index
->timestamp_begin
);
1039 ret
= kernctl_get_timestamp_end(infd
, &index
->timestamp_end
);
1041 PERROR("kernctl_get_timestamp_end");
1044 index
->timestamp_end
= htobe64(index
->timestamp_end
);
1046 ret
= kernctl_get_events_discarded(infd
, &index
->events_discarded
);
1048 PERROR("kernctl_get_events_discarded");
1051 index
->events_discarded
= htobe64(index
->events_discarded
);
1053 ret
= kernctl_get_content_size(infd
, &index
->content_size
);
1055 PERROR("kernctl_get_content_size");
1058 index
->content_size
= htobe64(index
->content_size
);
1060 ret
= kernctl_get_packet_size(infd
, &index
->packet_size
);
1062 PERROR("kernctl_get_packet_size");
1065 index
->packet_size
= htobe64(index
->packet_size
);
1067 ret
= kernctl_get_stream_id(infd
, &index
->stream_id
);
1069 PERROR("kernctl_get_stream_id");
1072 index
->stream_id
= htobe64(index
->stream_id
);
1078 * Sync metadata meaning request them to the session daemon and snapshot to the
1079 * metadata thread can consumer them.
1081 * Metadata stream lock MUST be acquired.
1083 * Return 0 if new metadatda is available, EAGAIN if the metadata stream
1084 * is empty or a negative value on error.
1086 int lttng_kconsumer_sync_metadata(struct lttng_consumer_stream
*metadata
)
1092 ret
= kernctl_buffer_flush(metadata
->wait_fd
);
1094 ERR("Failed to flush kernel stream");
1098 ret
= kernctl_snapshot(metadata
->wait_fd
);
1100 if (errno
!= EAGAIN
) {
1101 ERR("Sync metadata, taking kernel snapshot failed.");
1104 DBG("Sync metadata, no new kernel metadata");
1105 /* No new metadata, exit. */
1115 int update_stream_stats(struct lttng_consumer_stream
*stream
)
1118 uint64_t seq
, discarded
;
1120 ret
= kernctl_get_sequence_number(stream
->wait_fd
, &seq
);
1122 PERROR("kernctl_get_sequence_number");
1127 * Start the sequence when we extract the first packet in case we don't
1128 * start at 0 (for example if a consumer is not connected to the
1129 * session immediately after the beginning).
1131 if (stream
->last_sequence_number
== -1ULL) {
1132 stream
->last_sequence_number
= seq
;
1133 } else if (seq
> stream
->last_sequence_number
) {
1134 stream
->chan
->lost_packets
+= seq
-
1135 stream
->last_sequence_number
- 1;
1137 /* seq <= last_sequence_number */
1138 ERR("Sequence number inconsistent : prev = %" PRIu64
1139 ", current = %" PRIu64
,
1140 stream
->last_sequence_number
, seq
);
1144 stream
->last_sequence_number
= seq
;
1146 ret
= kernctl_get_events_discarded(stream
->wait_fd
, &discarded
);
1148 PERROR("kernctl_get_events_discarded");
1151 if (discarded
< stream
->last_discarded_events
) {
1153 * Overflow has occured. We assume only one wrap-around
1156 stream
->chan
->discarded_events
+= (1ULL << (CAA_BITS_PER_LONG
- 1)) -
1157 stream
->last_discarded_events
+ discarded
;
1159 stream
->chan
->discarded_events
+= discarded
-
1160 stream
->last_discarded_events
;
1162 stream
->last_discarded_events
= discarded
;
1170 * Consume data on a file descriptor and write it on a trace file.
1172 ssize_t
lttng_kconsumer_read_subbuffer(struct lttng_consumer_stream
*stream
,
1173 struct lttng_consumer_local_data
*ctx
)
1175 unsigned long len
, subbuf_size
, padding
;
1176 int err
, write_index
= 1;
1178 int infd
= stream
->wait_fd
;
1179 struct ctf_packet_index index
;
1181 DBG("In read_subbuffer (infd : %d)", infd
);
1183 /* Get the next subbuffer */
1184 err
= kernctl_get_next_subbuf(infd
);
1187 * This is a debug message even for single-threaded consumer,
1188 * because poll() have more relaxed criterions than get subbuf,
1189 * so get_subbuf may fail for short race windows where poll()
1190 * would issue wakeups.
1192 DBG("Reserving sub buffer failed (everything is normal, "
1193 "it is due to concurrency)");
1198 /* Get the full subbuffer size including padding */
1199 err
= kernctl_get_padded_subbuf_size(infd
, &len
);
1201 PERROR("Getting sub-buffer len failed.");
1202 err
= kernctl_put_subbuf(infd
);
1204 if (errno
== EFAULT
) {
1205 PERROR("Error in unreserving sub buffer\n");
1206 } else if (errno
== EIO
) {
1207 /* Should never happen with newer LTTng versions */
1208 PERROR("Reader has been pushed by the writer, last sub-buffer corrupted.");
1217 if (!stream
->metadata_flag
) {
1218 ret
= get_index_values(&index
, infd
);
1220 err
= kernctl_put_subbuf(infd
);
1222 if (errno
== EFAULT
) {
1223 PERROR("Error in unreserving sub buffer\n");
1224 } else if (errno
== EIO
) {
1225 /* Should never happen with newer LTTng versions */
1226 PERROR("Reader has been pushed by the writer, last sub-buffer corrupted.");
1233 ret
= update_stream_stats(stream
);
1241 switch (stream
->chan
->output
) {
1242 case CONSUMER_CHANNEL_SPLICE
:
1244 * XXX: The lttng-modules splice "actor" does not handle copying
1245 * partial pages hence only using the subbuffer size without the
1246 * padding makes the splice fail.
1251 /* splice the subbuffer to the tracefile */
1252 ret
= lttng_consumer_on_read_subbuffer_splice(ctx
, stream
, subbuf_size
,
1255 * XXX: Splice does not support network streaming so the return value
1256 * is simply checked against subbuf_size and not like the mmap() op.
1258 if (ret
!= subbuf_size
) {
1260 * display the error but continue processing to try
1261 * to release the subbuffer
1263 ERR("Error splicing to tracefile (ret: %zd != len: %lu)",
1268 case CONSUMER_CHANNEL_MMAP
:
1269 /* Get subbuffer size without padding */
1270 err
= kernctl_get_subbuf_size(infd
, &subbuf_size
);
1272 PERROR("Getting sub-buffer len failed.");
1273 err
= kernctl_put_subbuf(infd
);
1275 if (errno
== EFAULT
) {
1276 PERROR("Error in unreserving sub buffer\n");
1277 } else if (errno
== EIO
) {
1278 /* Should never happen with newer LTTng versions */
1279 PERROR("Reader has been pushed by the writer, last sub-buffer corrupted.");
1288 /* Make sure the tracer is not gone mad on us! */
1289 assert(len
>= subbuf_size
);
1291 padding
= len
- subbuf_size
;
1293 /* write the subbuffer to the tracefile */
1294 ret
= lttng_consumer_on_read_subbuffer_mmap(ctx
, stream
, subbuf_size
,
1297 * The mmap operation should write subbuf_size amount of data when
1298 * network streaming or the full padding (len) size when we are _not_
1301 if ((ret
!= subbuf_size
&& stream
->net_seq_idx
!= (uint64_t) -1ULL) ||
1302 (ret
!= len
&& stream
->net_seq_idx
== (uint64_t) -1ULL)) {
1304 * Display the error but continue processing to try to release the
1305 * subbuffer. This is a DBG statement since this is possible to
1306 * happen without being a critical error.
1308 DBG("Error writing to tracefile "
1309 "(ret: %zd != len: %lu != subbuf_size: %lu)",
1310 ret
, len
, subbuf_size
);
1315 ERR("Unknown output method");
1319 err
= kernctl_put_next_subbuf(infd
);
1321 if (errno
== EFAULT
) {
1322 PERROR("Error in unreserving sub buffer\n");
1323 } else if (errno
== EIO
) {
1324 /* Should never happen with newer LTTng versions */
1325 PERROR("Reader has been pushed by the writer, last sub-buffer corrupted.");
1331 /* Write index if needed. */
1336 if (stream
->chan
->live_timer_interval
&& !stream
->metadata_flag
) {
1338 * In live, block until all the metadata is sent.
1340 pthread_mutex_lock(&stream
->metadata_timer_lock
);
1341 assert(!stream
->missed_metadata_flush
);
1342 stream
->waiting_on_metadata
= true;
1343 pthread_mutex_unlock(&stream
->metadata_timer_lock
);
1345 err
= consumer_stream_sync_metadata(ctx
, stream
->session_id
);
1347 pthread_mutex_lock(&stream
->metadata_timer_lock
);
1348 stream
->waiting_on_metadata
= false;
1349 if (stream
->missed_metadata_flush
) {
1350 stream
->missed_metadata_flush
= false;
1351 pthread_mutex_unlock(&stream
->metadata_timer_lock
);
1352 (void) consumer_flush_kernel_index(stream
);
1354 pthread_mutex_unlock(&stream
->metadata_timer_lock
);
1361 err
= consumer_stream_write_index(stream
, &index
);
1370 int lttng_kconsumer_on_recv_stream(struct lttng_consumer_stream
*stream
)
1377 * Don't create anything if this is set for streaming or should not be
1380 if (stream
->net_seq_idx
== (uint64_t) -1ULL && stream
->chan
->monitor
) {
1381 ret
= utils_create_stream_file(stream
->chan
->pathname
, stream
->name
,
1382 stream
->chan
->tracefile_size
, stream
->tracefile_count_current
,
1383 stream
->uid
, stream
->gid
, NULL
);
1387 stream
->out_fd
= ret
;
1388 stream
->tracefile_size_current
= 0;
1390 if (!stream
->metadata_flag
) {
1391 ret
= index_create_file(stream
->chan
->pathname
,
1392 stream
->name
, stream
->uid
, stream
->gid
,
1393 stream
->chan
->tracefile_size
,
1394 stream
->tracefile_count_current
);
1398 stream
->index_fd
= ret
;
1402 if (stream
->output
== LTTNG_EVENT_MMAP
) {
1403 /* get the len of the mmap region */
1404 unsigned long mmap_len
;
1406 ret
= kernctl_get_mmap_len(stream
->wait_fd
, &mmap_len
);
1408 PERROR("kernctl_get_mmap_len");
1410 goto error_close_fd
;
1412 stream
->mmap_len
= (size_t) mmap_len
;
1414 stream
->mmap_base
= mmap(NULL
, stream
->mmap_len
, PROT_READ
,
1415 MAP_PRIVATE
, stream
->wait_fd
, 0);
1416 if (stream
->mmap_base
== MAP_FAILED
) {
1417 PERROR("Error mmaping");
1419 goto error_close_fd
;
1423 /* we return 0 to let the library handle the FD internally */
1427 if (stream
->out_fd
>= 0) {
1430 err
= close(stream
->out_fd
);
1432 stream
->out_fd
= -1;
1439 * Check if data is still being extracted from the buffers for a specific
1440 * stream. Consumer data lock MUST be acquired before calling this function
1441 * and the stream lock.
1443 * Return 1 if the traced data are still getting read else 0 meaning that the
1444 * data is available for trace viewer reading.
1446 int lttng_kconsumer_data_pending(struct lttng_consumer_stream
*stream
)
1452 if (stream
->endpoint_status
!= CONSUMER_ENDPOINT_ACTIVE
) {
1457 ret
= kernctl_get_next_subbuf(stream
->wait_fd
);
1459 /* There is still data so let's put back this subbuffer. */
1460 ret
= kernctl_put_subbuf(stream
->wait_fd
);
1462 ret
= 1; /* Data is pending */
1466 /* Data is NOT pending and ready to be read. */