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/pipe.h>
39 #include <common/relayd/relayd.h>
40 #include <common/utils.h>
41 #include <common/consumer-stream.h>
42 #include <common/index/index.h>
43 #include <common/consumer-timer.h>
45 #include "kernel-consumer.h"
47 extern struct lttng_consumer_global_data consumer_data
;
48 extern int consumer_poll_timeout
;
49 extern volatile int consumer_quit
;
52 * Take a snapshot for a specific fd
54 * Returns 0 on success, < 0 on error
56 int lttng_kconsumer_take_snapshot(struct lttng_consumer_stream
*stream
)
59 int infd
= stream
->wait_fd
;
61 ret
= kernctl_snapshot(infd
);
63 perror("Getting sub-buffer snapshot.");
71 * Get the produced position
73 * Returns 0 on success, < 0 on error
75 int lttng_kconsumer_get_produced_snapshot(struct lttng_consumer_stream
*stream
,
79 int infd
= stream
->wait_fd
;
81 ret
= kernctl_snapshot_get_produced(infd
, pos
);
83 perror("kernctl_snapshot_get_produced");
91 * Get the consumerd position
93 * Returns 0 on success, < 0 on error
95 int lttng_kconsumer_get_consumed_snapshot(struct lttng_consumer_stream
*stream
,
99 int infd
= stream
->wait_fd
;
101 ret
= kernctl_snapshot_get_consumed(infd
, pos
);
103 perror("kernctl_snapshot_get_consumed");
111 * Take a snapshot of all the stream of a channel
113 * Returns 0 on success, < 0 on error
115 int lttng_kconsumer_snapshot_channel(uint64_t key
, char *path
,
116 uint64_t relayd_id
, uint64_t max_stream_size
,
117 struct lttng_consumer_local_data
*ctx
)
120 unsigned long consumed_pos
, produced_pos
;
121 struct lttng_consumer_channel
*channel
;
122 struct lttng_consumer_stream
*stream
;
124 DBG("Kernel consumer snapshot channel %" PRIu64
, key
);
128 channel
= consumer_find_channel(key
);
130 ERR("No channel found for key %" PRIu64
, key
);
135 /* Splice is not supported yet for channel snapshot. */
136 if (channel
->output
!= CONSUMER_CHANNEL_MMAP
) {
137 ERR("Unsupported output %d", channel
->output
);
142 cds_list_for_each_entry(stream
, &channel
->streams
.head
, send_node
) {
144 health_code_update();
147 * Lock stream because we are about to change its state.
149 pthread_mutex_lock(&stream
->lock
);
152 * Assign the received relayd ID so we can use it for streaming. The streams
153 * are not visible to anyone so this is OK to change it.
155 stream
->net_seq_idx
= relayd_id
;
156 channel
->relayd_id
= relayd_id
;
157 if (relayd_id
!= (uint64_t) -1ULL) {
158 ret
= consumer_send_relayd_stream(stream
, path
);
160 ERR("sending stream to relayd");
164 ret
= utils_create_stream_file(path
, stream
->name
,
165 stream
->chan
->tracefile_size
,
166 stream
->tracefile_count_current
,
167 stream
->uid
, stream
->gid
, NULL
);
169 ERR("utils_create_stream_file");
173 stream
->out_fd
= ret
;
174 stream
->tracefile_size_current
= 0;
176 DBG("Kernel consumer snapshot stream %s/%s (%" PRIu64
")",
177 path
, stream
->name
, stream
->key
);
179 if (relayd_id
!= -1ULL) {
180 ret
= consumer_send_relayd_streams_sent(relayd_id
);
182 ERR("sending streams sent to relayd");
187 ret
= kernctl_buffer_flush(stream
->wait_fd
);
189 ERR("Failed to flush kernel stream");
194 ret
= lttng_kconsumer_take_snapshot(stream
);
196 ERR("Taking kernel snapshot");
200 ret
= lttng_kconsumer_get_produced_snapshot(stream
, &produced_pos
);
202 ERR("Produced kernel snapshot position");
206 ret
= lttng_kconsumer_get_consumed_snapshot(stream
, &consumed_pos
);
208 ERR("Consumerd kernel snapshot position");
212 if (stream
->max_sb_size
== 0) {
213 ret
= kernctl_get_max_subbuf_size(stream
->wait_fd
,
214 &stream
->max_sb_size
);
216 ERR("Getting kernel max_sb_size");
223 * The original value is sent back if max stream size is larger than
224 * the possible size of the snapshot. Also, we asume that the session
225 * daemon should never send a maximum stream size that is lower than
228 consumed_pos
= consumer_get_consumed_maxsize(consumed_pos
,
229 produced_pos
, max_stream_size
);
231 while (consumed_pos
< produced_pos
) {
233 unsigned long len
, padded_len
;
235 health_code_update();
237 DBG("Kernel consumer taking snapshot at pos %lu", consumed_pos
);
239 ret
= kernctl_get_subbuf(stream
->wait_fd
, &consumed_pos
);
241 if (errno
!= EAGAIN
) {
242 PERROR("kernctl_get_subbuf snapshot");
246 DBG("Kernel consumer get subbuf failed. Skipping it.");
247 consumed_pos
+= stream
->max_sb_size
;
251 ret
= kernctl_get_subbuf_size(stream
->wait_fd
, &len
);
253 ERR("Snapshot kernctl_get_subbuf_size");
255 goto error_put_subbuf
;
258 ret
= kernctl_get_padded_subbuf_size(stream
->wait_fd
, &padded_len
);
260 ERR("Snapshot kernctl_get_padded_subbuf_size");
262 goto error_put_subbuf
;
265 read_len
= lttng_consumer_on_read_subbuffer_mmap(ctx
, stream
, len
,
266 padded_len
- len
, NULL
);
268 * We write the padded len in local tracefiles but the data len
269 * when using a relay. Display the error but continue processing
270 * to try to release the subbuffer.
272 if (relayd_id
!= (uint64_t) -1ULL) {
273 if (read_len
!= len
) {
274 ERR("Error sending to the relay (ret: %zd != len: %lu)",
278 if (read_len
!= padded_len
) {
279 ERR("Error writing to tracefile (ret: %zd != len: %lu)",
280 read_len
, padded_len
);
284 ret
= kernctl_put_subbuf(stream
->wait_fd
);
286 ERR("Snapshot kernctl_put_subbuf");
290 consumed_pos
+= stream
->max_sb_size
;
293 if (relayd_id
== (uint64_t) -1ULL) {
294 if (stream
->out_fd
>= 0) {
295 ret
= close(stream
->out_fd
);
297 PERROR("Kernel consumer snapshot close out_fd");
303 close_relayd_stream(stream
);
304 stream
->net_seq_idx
= (uint64_t) -1ULL;
306 pthread_mutex_unlock(&stream
->lock
);
314 ret
= kernctl_put_subbuf(stream
->wait_fd
);
317 ERR("Snapshot kernctl_put_subbuf error path");
320 pthread_mutex_unlock(&stream
->lock
);
327 * Read the whole metadata available for a snapshot.
329 * Returns 0 on success, < 0 on error
331 int lttng_kconsumer_snapshot_metadata(uint64_t key
, char *path
,
332 uint64_t relayd_id
, struct lttng_consumer_local_data
*ctx
)
334 int ret
, use_relayd
= 0;
336 struct lttng_consumer_channel
*metadata_channel
;
337 struct lttng_consumer_stream
*metadata_stream
;
341 DBG("Kernel consumer snapshot metadata with key %" PRIu64
" at path %s",
346 metadata_channel
= consumer_find_channel(key
);
347 if (!metadata_channel
) {
348 ERR("Kernel snapshot metadata not found for key %" PRIu64
, key
);
353 metadata_stream
= metadata_channel
->metadata_stream
;
354 assert(metadata_stream
);
356 /* Flag once that we have a valid relayd for the stream. */
357 if (relayd_id
!= (uint64_t) -1ULL) {
362 ret
= consumer_send_relayd_stream(metadata_stream
, path
);
367 ret
= utils_create_stream_file(path
, metadata_stream
->name
,
368 metadata_stream
->chan
->tracefile_size
,
369 metadata_stream
->tracefile_count_current
,
370 metadata_stream
->uid
, metadata_stream
->gid
, NULL
);
374 metadata_stream
->out_fd
= ret
;
378 health_code_update();
380 ret_read
= lttng_kconsumer_read_subbuffer(metadata_stream
, ctx
);
382 if (ret_read
!= -EAGAIN
) {
383 ERR("Kernel snapshot reading metadata subbuffer (ret: %zd)",
387 /* ret_read is negative at this point so we will exit the loop. */
390 } while (ret_read
>= 0);
393 close_relayd_stream(metadata_stream
);
394 metadata_stream
->net_seq_idx
= (uint64_t) -1ULL;
396 if (metadata_stream
->out_fd
>= 0) {
397 ret
= close(metadata_stream
->out_fd
);
399 PERROR("Kernel consumer snapshot metadata close out_fd");
401 * Don't go on error here since the snapshot was successful at this
402 * point but somehow the close failed.
405 metadata_stream
->out_fd
= -1;
411 cds_list_del(&metadata_stream
->send_node
);
412 consumer_stream_destroy(metadata_stream
, NULL
);
413 metadata_channel
->metadata_stream
= NULL
;
420 * Receive command from session daemon and process it.
422 * Return 1 on success else a negative value or 0.
424 int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data
*ctx
,
425 int sock
, struct pollfd
*consumer_sockpoll
)
428 enum lttcomm_return_code ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
429 struct lttcomm_consumer_msg msg
;
431 health_code_update();
433 ret
= lttcomm_recv_unix_sock(sock
, &msg
, sizeof(msg
));
434 if (ret
!= sizeof(msg
)) {
436 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_ERROR_RECV_CMD
);
442 health_code_update();
444 /* Deprecated command */
445 assert(msg
.cmd_type
!= LTTNG_CONSUMER_STOP
);
447 health_code_update();
449 /* relayd needs RCU read-side protection */
452 switch (msg
.cmd_type
) {
453 case LTTNG_CONSUMER_ADD_RELAYD_SOCKET
:
455 /* Session daemon status message are handled in the following call. */
456 ret
= consumer_add_relayd_socket(msg
.u
.relayd_sock
.net_index
,
457 msg
.u
.relayd_sock
.type
, ctx
, sock
, consumer_sockpoll
,
458 &msg
.u
.relayd_sock
.sock
, msg
.u
.relayd_sock
.session_id
,
459 msg
.u
.relayd_sock
.relayd_session_id
);
462 case LTTNG_CONSUMER_ADD_CHANNEL
:
464 struct lttng_consumer_channel
*new_channel
;
467 health_code_update();
469 /* First send a status message before receiving the fds. */
470 ret
= consumer_send_status_msg(sock
, ret_code
);
472 /* Somehow, the session daemon is not responding anymore. */
476 health_code_update();
478 DBG("consumer_add_channel %" PRIu64
, msg
.u
.channel
.channel_key
);
479 new_channel
= consumer_allocate_channel(msg
.u
.channel
.channel_key
,
480 msg
.u
.channel
.session_id
, msg
.u
.channel
.pathname
,
481 msg
.u
.channel
.name
, msg
.u
.channel
.uid
, msg
.u
.channel
.gid
,
482 msg
.u
.channel
.relayd_id
, msg
.u
.channel
.output
,
483 msg
.u
.channel
.tracefile_size
,
484 msg
.u
.channel
.tracefile_count
, 0,
485 msg
.u
.channel
.monitor
,
486 msg
.u
.channel
.live_timer_interval
);
487 if (new_channel
== NULL
) {
488 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_OUTFD_ERROR
);
491 new_channel
->nb_init_stream_left
= msg
.u
.channel
.nb_init_streams
;
492 switch (msg
.u
.channel
.output
) {
493 case LTTNG_EVENT_SPLICE
:
494 new_channel
->output
= CONSUMER_CHANNEL_SPLICE
;
496 case LTTNG_EVENT_MMAP
:
497 new_channel
->output
= CONSUMER_CHANNEL_MMAP
;
500 ERR("Channel output unknown %d", msg
.u
.channel
.output
);
504 /* Translate and save channel type. */
505 switch (msg
.u
.channel
.type
) {
506 case CONSUMER_CHANNEL_TYPE_DATA
:
507 case CONSUMER_CHANNEL_TYPE_METADATA
:
508 new_channel
->type
= msg
.u
.channel
.type
;
515 health_code_update();
517 if (ctx
->on_recv_channel
!= NULL
) {
518 ret_recv
= ctx
->on_recv_channel(new_channel
);
520 ret
= consumer_add_channel(new_channel
, ctx
);
521 } else if (ret_recv
< 0) {
525 ret
= consumer_add_channel(new_channel
, ctx
);
527 if (CONSUMER_CHANNEL_TYPE_DATA
) {
528 consumer_timer_live_start(new_channel
,
529 msg
.u
.channel
.live_timer_interval
);
532 health_code_update();
534 /* If we received an error in add_channel, we need to report it. */
536 ret
= consumer_send_status_msg(sock
, ret
);
545 case LTTNG_CONSUMER_ADD_STREAM
:
548 struct lttng_pipe
*stream_pipe
;
549 struct lttng_consumer_stream
*new_stream
;
550 struct lttng_consumer_channel
*channel
;
554 * Get stream's channel reference. Needed when adding the stream to the
557 channel
= consumer_find_channel(msg
.u
.stream
.channel_key
);
560 * We could not find the channel. Can happen if cpu hotplug
561 * happens while tearing down.
563 ERR("Unable to find channel key %" PRIu64
, msg
.u
.stream
.channel_key
);
564 ret_code
= LTTNG_ERR_KERN_CHAN_NOT_FOUND
;
567 health_code_update();
569 /* First send a status message before receiving the fds. */
570 ret
= consumer_send_status_msg(sock
, ret_code
);
572 /* Somehow, the session daemon is not responding anymore. */
576 health_code_update();
578 if (ret_code
!= LTTCOMM_CONSUMERD_SUCCESS
) {
579 /* Channel was not found. */
585 ret
= lttng_consumer_poll_socket(consumer_sockpoll
);
591 health_code_update();
593 /* Get stream file descriptor from socket */
594 ret
= lttcomm_recv_fds_unix_sock(sock
, &fd
, 1);
595 if (ret
!= sizeof(fd
)) {
596 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_ERROR_RECV_FD
);
601 health_code_update();
604 * Send status code to session daemon only if the recv works. If the
605 * above recv() failed, the session daemon is notified through the
606 * error socket and the teardown is eventually done.
608 ret
= consumer_send_status_msg(sock
, ret_code
);
610 /* Somehow, the session daemon is not responding anymore. */
614 health_code_update();
616 new_stream
= consumer_allocate_stream(channel
->key
,
618 LTTNG_CONSUMER_ACTIVE_STREAM
,
628 if (new_stream
== NULL
) {
633 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_OUTFD_ERROR
);
639 new_stream
->chan
= channel
;
640 new_stream
->wait_fd
= fd
;
641 switch (channel
->output
) {
642 case CONSUMER_CHANNEL_SPLICE
:
643 new_stream
->output
= LTTNG_EVENT_SPLICE
;
645 case CONSUMER_CHANNEL_MMAP
:
646 new_stream
->output
= LTTNG_EVENT_MMAP
;
649 ERR("Stream output unknown %d", channel
->output
);
654 * We've just assigned the channel to the stream so increment the
655 * refcount right now. We don't need to increment the refcount for
656 * streams in no monitor because we handle manually the cleanup of
657 * those. It is very important to make sure there is NO prior
658 * consumer_del_stream() calls or else the refcount will be unbalanced.
660 if (channel
->monitor
) {
661 uatomic_inc(&new_stream
->chan
->refcount
);
665 * The buffer flush is done on the session daemon side for the kernel
666 * so no need for the stream "hangup_flush_done" variable to be
667 * tracked. This is important for a kernel stream since we don't rely
668 * on the flush state of the stream to read data. It's not the case for
669 * user space tracing.
671 new_stream
->hangup_flush_done
= 0;
673 health_code_update();
675 if (ctx
->on_recv_stream
) {
676 ret
= ctx
->on_recv_stream(new_stream
);
678 consumer_stream_free(new_stream
);
683 health_code_update();
685 if (new_stream
->metadata_flag
) {
686 channel
->metadata_stream
= new_stream
;
689 /* Do not monitor this stream. */
690 if (!channel
->monitor
) {
691 DBG("Kernel consumer add stream %s in no monitor mode with "
692 "relayd id %" PRIu64
, new_stream
->name
,
693 new_stream
->net_seq_idx
);
694 cds_list_add(&new_stream
->send_node
, &channel
->streams
.head
);
698 /* Send stream to relayd if the stream has an ID. */
699 if (new_stream
->net_seq_idx
!= (uint64_t) -1ULL) {
700 ret
= consumer_send_relayd_stream(new_stream
,
701 new_stream
->chan
->pathname
);
703 consumer_stream_free(new_stream
);
708 /* Get the right pipe where the stream will be sent. */
709 if (new_stream
->metadata_flag
) {
710 ret
= consumer_add_metadata_stream(new_stream
);
712 ERR("Consumer add metadata stream %" PRIu64
" failed. Continuing",
714 consumer_stream_free(new_stream
);
717 stream_pipe
= ctx
->consumer_metadata_pipe
;
719 ret
= consumer_add_data_stream(new_stream
);
721 ERR("Consumer add stream %" PRIu64
" failed. Continuing",
723 consumer_stream_free(new_stream
);
726 stream_pipe
= ctx
->consumer_data_pipe
;
729 /* Vitible to other threads */
730 new_stream
->globally_visible
= 1;
732 health_code_update();
734 ret
= lttng_pipe_write(stream_pipe
, &new_stream
, sizeof(new_stream
));
736 ERR("Consumer write %s stream to pipe %d",
737 new_stream
->metadata_flag
? "metadata" : "data",
738 lttng_pipe_get_writefd(stream_pipe
));
739 if (new_stream
->metadata_flag
) {
740 consumer_del_stream_for_metadata(new_stream
);
742 consumer_del_stream_for_data(new_stream
);
747 DBG("Kernel consumer ADD_STREAM %s (fd: %d) with relayd id %" PRIu64
,
748 new_stream
->name
, fd
, new_stream
->relayd_stream_id
);
751 case LTTNG_CONSUMER_STREAMS_SENT
:
753 struct lttng_consumer_channel
*channel
;
756 * Get stream's channel reference. Needed when adding the stream to the
759 channel
= consumer_find_channel(msg
.u
.sent_streams
.channel_key
);
762 * We could not find the channel. Can happen if cpu hotplug
763 * happens while tearing down.
765 ERR("Unable to find channel key %" PRIu64
,
766 msg
.u
.sent_streams
.channel_key
);
767 ret_code
= LTTNG_ERR_KERN_CHAN_NOT_FOUND
;
770 health_code_update();
773 * Send status code to session daemon.
775 ret
= consumer_send_status_msg(sock
, ret_code
);
777 /* Somehow, the session daemon is not responding anymore. */
781 health_code_update();
784 * We should not send this message if we don't monitor the
785 * streams in this channel.
787 if (!channel
->monitor
) {
791 health_code_update();
792 /* Send stream to relayd if the stream has an ID. */
793 if (msg
.u
.sent_streams
.net_seq_idx
!= (uint64_t) -1ULL) {
794 ret
= consumer_send_relayd_streams_sent(
795 msg
.u
.sent_streams
.net_seq_idx
);
802 case LTTNG_CONSUMER_UPDATE_STREAM
:
807 case LTTNG_CONSUMER_DESTROY_RELAYD
:
809 uint64_t index
= msg
.u
.destroy_relayd
.net_seq_idx
;
810 struct consumer_relayd_sock_pair
*relayd
;
812 DBG("Kernel consumer destroying relayd %" PRIu64
, index
);
814 /* Get relayd reference if exists. */
815 relayd
= consumer_find_relayd(index
);
816 if (relayd
== NULL
) {
817 DBG("Unable to find relayd %" PRIu64
, index
);
818 ret_code
= LTTNG_ERR_NO_CONSUMER
;
822 * Each relayd socket pair has a refcount of stream attached to it
823 * which tells if the relayd is still active or not depending on the
826 * This will set the destroy flag of the relayd object and destroy it
827 * if the refcount reaches zero when called.
829 * The destroy can happen either here or when a stream fd hangs up.
832 consumer_flag_relayd_for_destroy(relayd
);
835 health_code_update();
837 ret
= consumer_send_status_msg(sock
, ret_code
);
839 /* Somehow, the session daemon is not responding anymore. */
845 case LTTNG_CONSUMER_DATA_PENDING
:
848 uint64_t id
= msg
.u
.data_pending
.session_id
;
850 DBG("Kernel consumer data pending command for id %" PRIu64
, id
);
852 ret
= consumer_data_pending(id
);
854 health_code_update();
856 /* Send back returned value to session daemon */
857 ret
= lttcomm_send_unix_sock(sock
, &ret
, sizeof(ret
));
859 PERROR("send data pending ret code");
864 * No need to send back a status message since the data pending
865 * returned value is the response.
869 case LTTNG_CONSUMER_SNAPSHOT_CHANNEL
:
871 if (msg
.u
.snapshot_channel
.metadata
== 1) {
872 ret
= lttng_kconsumer_snapshot_metadata(msg
.u
.snapshot_channel
.key
,
873 msg
.u
.snapshot_channel
.pathname
,
874 msg
.u
.snapshot_channel
.relayd_id
, ctx
);
876 ERR("Snapshot metadata failed");
877 ret_code
= LTTNG_ERR_KERN_META_FAIL
;
880 ret
= lttng_kconsumer_snapshot_channel(msg
.u
.snapshot_channel
.key
,
881 msg
.u
.snapshot_channel
.pathname
,
882 msg
.u
.snapshot_channel
.relayd_id
,
883 msg
.u
.snapshot_channel
.max_stream_size
,
886 ERR("Snapshot channel failed");
887 ret_code
= LTTNG_ERR_KERN_CHAN_FAIL
;
891 health_code_update();
893 ret
= consumer_send_status_msg(sock
, ret_code
);
895 /* Somehow, the session daemon is not responding anymore. */
900 case LTTNG_CONSUMER_DESTROY_CHANNEL
:
902 uint64_t key
= msg
.u
.destroy_channel
.key
;
903 struct lttng_consumer_channel
*channel
;
905 channel
= consumer_find_channel(key
);
907 ERR("Kernel consumer destroy channel %" PRIu64
" not found", key
);
908 ret_code
= LTTNG_ERR_KERN_CHAN_NOT_FOUND
;
911 health_code_update();
913 ret
= consumer_send_status_msg(sock
, ret_code
);
915 /* Somehow, the session daemon is not responding anymore. */
919 health_code_update();
922 * This command should ONLY be issued for channel with streams set in
925 assert(!channel
->monitor
);
928 * The refcount should ALWAYS be 0 in the case of a channel in no
931 assert(!uatomic_sub_return(&channel
->refcount
, 1));
933 consumer_del_channel(channel
);
945 * Return 1 to indicate success since the 0 value can be a socket
946 * shutdown during the recv() or send() call.
948 health_code_update();
953 /* This will issue a consumer stop. */
958 * Populate index values of a kernel stream. Values are set in big endian order.
960 * Return 0 on success or else a negative value.
962 static int get_index_values(struct ctf_packet_index
*index
, int infd
)
966 ret
= kernctl_get_timestamp_begin(infd
, &index
->timestamp_begin
);
968 PERROR("kernctl_get_timestamp_begin");
971 index
->timestamp_begin
= htobe64(index
->timestamp_begin
);
973 ret
= kernctl_get_timestamp_end(infd
, &index
->timestamp_end
);
975 PERROR("kernctl_get_timestamp_end");
978 index
->timestamp_end
= htobe64(index
->timestamp_end
);
980 ret
= kernctl_get_events_discarded(infd
, &index
->events_discarded
);
982 PERROR("kernctl_get_events_discarded");
985 index
->events_discarded
= htobe64(index
->events_discarded
);
987 ret
= kernctl_get_content_size(infd
, &index
->content_size
);
989 PERROR("kernctl_get_content_size");
992 index
->content_size
= htobe64(index
->content_size
);
994 ret
= kernctl_get_packet_size(infd
, &index
->packet_size
);
996 PERROR("kernctl_get_packet_size");
999 index
->packet_size
= htobe64(index
->packet_size
);
1001 ret
= kernctl_get_stream_id(infd
, &index
->stream_id
);
1003 PERROR("kernctl_get_stream_id");
1006 index
->stream_id
= htobe64(index
->stream_id
);
1012 * Sync metadata meaning request them to the session daemon and snapshot to the
1013 * metadata thread can consumer them.
1015 * Metadata stream lock MUST be acquired.
1017 * Return 0 if new metadatda is available, EAGAIN if the metadata stream
1018 * is empty or a negative value on error.
1020 int lttng_kconsumer_sync_metadata(struct lttng_consumer_stream
*metadata
)
1026 ret
= kernctl_buffer_flush(metadata
->wait_fd
);
1028 ERR("Failed to flush kernel stream");
1032 ret
= kernctl_snapshot(metadata
->wait_fd
);
1034 if (errno
!= EAGAIN
) {
1035 ERR("Sync metadata, taking kernel snapshot failed.");
1038 DBG("Sync metadata, no new kernel metadata");
1039 /* No new metadata, exit. */
1049 * Consume data on a file descriptor and write it on a trace file.
1051 ssize_t
lttng_kconsumer_read_subbuffer(struct lttng_consumer_stream
*stream
,
1052 struct lttng_consumer_local_data
*ctx
)
1054 unsigned long len
, subbuf_size
, padding
;
1055 int err
, write_index
= 1;
1057 int infd
= stream
->wait_fd
;
1058 struct ctf_packet_index index
;
1060 DBG("In read_subbuffer (infd : %d)", infd
);
1062 /* Get the next subbuffer */
1063 err
= kernctl_get_next_subbuf(infd
);
1066 * This is a debug message even for single-threaded consumer,
1067 * because poll() have more relaxed criterions than get subbuf,
1068 * so get_subbuf may fail for short race windows where poll()
1069 * would issue wakeups.
1071 DBG("Reserving sub buffer failed (everything is normal, "
1072 "it is due to concurrency)");
1077 /* Get the full subbuffer size including padding */
1078 err
= kernctl_get_padded_subbuf_size(infd
, &len
);
1080 perror("Getting sub-buffer len failed.");
1085 if (!stream
->metadata_flag
) {
1086 ret
= get_index_values(&index
, infd
);
1094 switch (stream
->chan
->output
) {
1095 case CONSUMER_CHANNEL_SPLICE
:
1097 * XXX: The lttng-modules splice "actor" does not handle copying
1098 * partial pages hence only using the subbuffer size without the
1099 * padding makes the splice fail.
1104 /* splice the subbuffer to the tracefile */
1105 ret
= lttng_consumer_on_read_subbuffer_splice(ctx
, stream
, subbuf_size
,
1108 * XXX: Splice does not support network streaming so the return value
1109 * is simply checked against subbuf_size and not like the mmap() op.
1111 if (ret
!= subbuf_size
) {
1113 * display the error but continue processing to try
1114 * to release the subbuffer
1116 ERR("Error splicing to tracefile (ret: %zd != len: %lu)",
1121 case CONSUMER_CHANNEL_MMAP
:
1122 /* Get subbuffer size without padding */
1123 err
= kernctl_get_subbuf_size(infd
, &subbuf_size
);
1125 perror("Getting sub-buffer len failed.");
1130 /* Make sure the tracer is not gone mad on us! */
1131 assert(len
>= subbuf_size
);
1133 padding
= len
- subbuf_size
;
1135 /* write the subbuffer to the tracefile */
1136 ret
= lttng_consumer_on_read_subbuffer_mmap(ctx
, stream
, subbuf_size
,
1139 * The mmap operation should write subbuf_size amount of data when
1140 * network streaming or the full padding (len) size when we are _not_
1143 if ((ret
!= subbuf_size
&& stream
->net_seq_idx
!= (uint64_t) -1ULL) ||
1144 (ret
!= len
&& stream
->net_seq_idx
== (uint64_t) -1ULL)) {
1146 * Display the error but continue processing to try to release the
1147 * subbuffer. This is a DBG statement since this is possible to
1148 * happen without being a critical error.
1150 DBG("Error writing to tracefile "
1151 "(ret: %zd != len: %lu != subbuf_size: %lu)",
1152 ret
, len
, subbuf_size
);
1157 ERR("Unknown output method");
1161 err
= kernctl_put_next_subbuf(infd
);
1163 if (errno
== EFAULT
) {
1164 perror("Error in unreserving sub buffer\n");
1165 } else if (errno
== EIO
) {
1166 /* Should never happen with newer LTTng versions */
1167 perror("Reader has been pushed by the writer, last sub-buffer corrupted.");
1173 /* Write index if needed. */
1178 if (stream
->chan
->live_timer_interval
&& !stream
->metadata_flag
) {
1180 * In live, block until all the metadata is sent.
1182 err
= consumer_stream_sync_metadata(ctx
, stream
->session_id
);
1188 err
= consumer_stream_write_index(stream
, &index
);
1197 int lttng_kconsumer_on_recv_stream(struct lttng_consumer_stream
*stream
)
1204 * Don't create anything if this is set for streaming or should not be
1207 if (stream
->net_seq_idx
== (uint64_t) -1ULL && stream
->chan
->monitor
) {
1208 ret
= utils_create_stream_file(stream
->chan
->pathname
, stream
->name
,
1209 stream
->chan
->tracefile_size
, stream
->tracefile_count_current
,
1210 stream
->uid
, stream
->gid
, NULL
);
1214 stream
->out_fd
= ret
;
1215 stream
->tracefile_size_current
= 0;
1217 if (!stream
->metadata_flag
) {
1218 ret
= index_create_file(stream
->chan
->pathname
,
1219 stream
->name
, stream
->uid
, stream
->gid
,
1220 stream
->chan
->tracefile_size
,
1221 stream
->tracefile_count_current
);
1225 stream
->index_fd
= ret
;
1229 if (stream
->output
== LTTNG_EVENT_MMAP
) {
1230 /* get the len of the mmap region */
1231 unsigned long mmap_len
;
1233 ret
= kernctl_get_mmap_len(stream
->wait_fd
, &mmap_len
);
1235 PERROR("kernctl_get_mmap_len");
1237 goto error_close_fd
;
1239 stream
->mmap_len
= (size_t) mmap_len
;
1241 stream
->mmap_base
= mmap(NULL
, stream
->mmap_len
, PROT_READ
,
1242 MAP_PRIVATE
, stream
->wait_fd
, 0);
1243 if (stream
->mmap_base
== MAP_FAILED
) {
1244 PERROR("Error mmaping");
1246 goto error_close_fd
;
1250 /* we return 0 to let the library handle the FD internally */
1254 if (stream
->out_fd
>= 0) {
1257 err
= close(stream
->out_fd
);
1259 stream
->out_fd
= -1;
1266 * Check if data is still being extracted from the buffers for a specific
1267 * stream. Consumer data lock MUST be acquired before calling this function
1268 * and the stream lock.
1270 * Return 1 if the traced data are still getting read else 0 meaning that the
1271 * data is available for trace viewer reading.
1273 int lttng_kconsumer_data_pending(struct lttng_consumer_stream
*stream
)
1279 if (stream
->endpoint_status
!= CONSUMER_ENDPOINT_ACTIVE
) {
1284 ret
= kernctl_get_next_subbuf(stream
->wait_fd
);
1286 /* There is still data so let's put back this subbuffer. */
1287 ret
= kernctl_put_subbuf(stream
->wait_fd
);
1289 ret
= 1; /* Data is pending */
1293 /* Data is NOT pending and ready to be read. */