2 * Copyright (C) 2011 - Julien Desfossez <julien.desfossez@polymtl.ca>
3 * Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
4 * 2012 - David Goulet <dgoulet@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, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
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/utils.h>
36 #include <common/compat/poll.h>
37 #include <common/compat/endian.h>
38 #include <common/index/index.h>
39 #include <common/kernel-ctl/kernel-ctl.h>
40 #include <common/sessiond-comm/relayd.h>
41 #include <common/sessiond-comm/sessiond-comm.h>
42 #include <common/kernel-consumer/kernel-consumer.h>
43 #include <common/relayd/relayd.h>
44 #include <common/ust-consumer/ust-consumer.h>
45 #include <common/consumer/consumer-timer.h>
46 #include <common/consumer/consumer.h>
47 #include <common/consumer/consumer-stream.h>
48 #include <common/consumer/consumer-testpoint.h>
49 #include <common/align.h>
50 #include <common/consumer/consumer-metadata-cache.h>
52 struct lttng_consumer_global_data consumer_data
= {
55 .type
= LTTNG_CONSUMER_UNKNOWN
,
58 enum consumer_channel_action
{
61 CONSUMER_CHANNEL_QUIT
,
64 struct consumer_channel_msg
{
65 enum consumer_channel_action action
;
66 struct lttng_consumer_channel
*chan
; /* add */
67 uint64_t key
; /* del */
70 /* Flag used to temporarily pause data consumption from testpoints. */
71 int data_consumption_paused
;
74 * Flag to inform the polling thread to quit when all fd hung up. Updated by
75 * the consumer_thread_receive_fds when it notices that all fds has hung up.
76 * Also updated by the signal handler (consumer_should_exit()). Read by the
82 * Global hash table containing respectively metadata and data streams. The
83 * stream element in this ht should only be updated by the metadata poll thread
84 * for the metadata and the data poll thread for the data.
86 static struct lttng_ht
*metadata_ht
;
87 static struct lttng_ht
*data_ht
;
90 * Notify a thread lttng pipe to poll back again. This usually means that some
91 * global state has changed so we just send back the thread in a poll wait
94 static void notify_thread_lttng_pipe(struct lttng_pipe
*pipe
)
96 struct lttng_consumer_stream
*null_stream
= NULL
;
100 (void) lttng_pipe_write(pipe
, &null_stream
, sizeof(null_stream
));
103 static void notify_health_quit_pipe(int *pipe
)
107 ret
= lttng_write(pipe
[1], "4", 1);
109 PERROR("write consumer health quit");
113 static void notify_channel_pipe(struct lttng_consumer_local_data
*ctx
,
114 struct lttng_consumer_channel
*chan
,
116 enum consumer_channel_action action
)
118 struct consumer_channel_msg msg
;
121 memset(&msg
, 0, sizeof(msg
));
126 ret
= lttng_write(ctx
->consumer_channel_pipe
[1], &msg
, sizeof(msg
));
127 if (ret
< sizeof(msg
)) {
128 PERROR("notify_channel_pipe write error");
132 void notify_thread_del_channel(struct lttng_consumer_local_data
*ctx
,
135 notify_channel_pipe(ctx
, NULL
, key
, CONSUMER_CHANNEL_DEL
);
138 static int read_channel_pipe(struct lttng_consumer_local_data
*ctx
,
139 struct lttng_consumer_channel
**chan
,
141 enum consumer_channel_action
*action
)
143 struct consumer_channel_msg msg
;
146 ret
= lttng_read(ctx
->consumer_channel_pipe
[0], &msg
, sizeof(msg
));
147 if (ret
< sizeof(msg
)) {
151 *action
= msg
.action
;
159 * Cleanup the stream list of a channel. Those streams are not yet globally
162 static void clean_channel_stream_list(struct lttng_consumer_channel
*channel
)
164 struct lttng_consumer_stream
*stream
, *stmp
;
168 /* Delete streams that might have been left in the stream list. */
169 cds_list_for_each_entry_safe(stream
, stmp
, &channel
->streams
.head
,
171 cds_list_del(&stream
->send_node
);
173 * Once a stream is added to this list, the buffers were created so we
174 * have a guarantee that this call will succeed. Setting the monitor
175 * mode to 0 so we don't lock nor try to delete the stream from the
179 consumer_stream_destroy(stream
, NULL
);
184 * Find a stream. The consumer_data.lock must be locked during this
187 static struct lttng_consumer_stream
*find_stream(uint64_t key
,
190 struct lttng_ht_iter iter
;
191 struct lttng_ht_node_u64
*node
;
192 struct lttng_consumer_stream
*stream
= NULL
;
196 /* -1ULL keys are lookup failures */
197 if (key
== (uint64_t) -1ULL) {
203 lttng_ht_lookup(ht
, &key
, &iter
);
204 node
= lttng_ht_iter_get_node_u64(&iter
);
206 stream
= caa_container_of(node
, struct lttng_consumer_stream
, node
);
214 static void steal_stream_key(uint64_t key
, struct lttng_ht
*ht
)
216 struct lttng_consumer_stream
*stream
;
219 stream
= find_stream(key
, ht
);
221 stream
->key
= (uint64_t) -1ULL;
223 * We don't want the lookup to match, but we still need
224 * to iterate on this stream when iterating over the hash table. Just
225 * change the node key.
227 stream
->node
.key
= (uint64_t) -1ULL;
233 * Return a channel object for the given key.
235 * RCU read side lock MUST be acquired before calling this function and
236 * protects the channel ptr.
238 struct lttng_consumer_channel
*consumer_find_channel(uint64_t key
)
240 struct lttng_ht_iter iter
;
241 struct lttng_ht_node_u64
*node
;
242 struct lttng_consumer_channel
*channel
= NULL
;
244 /* -1ULL keys are lookup failures */
245 if (key
== (uint64_t) -1ULL) {
249 lttng_ht_lookup(consumer_data
.channel_ht
, &key
, &iter
);
250 node
= lttng_ht_iter_get_node_u64(&iter
);
252 channel
= caa_container_of(node
, struct lttng_consumer_channel
, node
);
259 * There is a possibility that the consumer does not have enough time between
260 * the close of the channel on the session daemon and the cleanup in here thus
261 * once we have a channel add with an existing key, we know for sure that this
262 * channel will eventually get cleaned up by all streams being closed.
264 * This function just nullifies the already existing channel key.
266 static void steal_channel_key(uint64_t key
)
268 struct lttng_consumer_channel
*channel
;
271 channel
= consumer_find_channel(key
);
273 channel
->key
= (uint64_t) -1ULL;
275 * We don't want the lookup to match, but we still need to iterate on
276 * this channel when iterating over the hash table. Just change the
279 channel
->node
.key
= (uint64_t) -1ULL;
284 static void free_channel_rcu(struct rcu_head
*head
)
286 struct lttng_ht_node_u64
*node
=
287 caa_container_of(head
, struct lttng_ht_node_u64
, head
);
288 struct lttng_consumer_channel
*channel
=
289 caa_container_of(node
, struct lttng_consumer_channel
, node
);
291 switch (consumer_data
.type
) {
292 case LTTNG_CONSUMER_KERNEL
:
294 case LTTNG_CONSUMER32_UST
:
295 case LTTNG_CONSUMER64_UST
:
296 lttng_ustconsumer_free_channel(channel
);
299 ERR("Unknown consumer_data type");
306 * RCU protected relayd socket pair free.
308 static void free_relayd_rcu(struct rcu_head
*head
)
310 struct lttng_ht_node_u64
*node
=
311 caa_container_of(head
, struct lttng_ht_node_u64
, head
);
312 struct consumer_relayd_sock_pair
*relayd
=
313 caa_container_of(node
, struct consumer_relayd_sock_pair
, node
);
316 * Close all sockets. This is done in the call RCU since we don't want the
317 * socket fds to be reassigned thus potentially creating bad state of the
320 * We do not have to lock the control socket mutex here since at this stage
321 * there is no one referencing to this relayd object.
323 (void) relayd_close(&relayd
->control_sock
);
324 (void) relayd_close(&relayd
->data_sock
);
330 * Destroy and free relayd socket pair object.
332 void consumer_destroy_relayd(struct consumer_relayd_sock_pair
*relayd
)
335 struct lttng_ht_iter iter
;
337 if (relayd
== NULL
) {
341 DBG("Consumer destroy and close relayd socket pair");
343 iter
.iter
.node
= &relayd
->node
.node
;
344 ret
= lttng_ht_del(consumer_data
.relayd_ht
, &iter
);
346 /* We assume the relayd is being or is destroyed */
350 /* RCU free() call */
351 call_rcu(&relayd
->node
.head
, free_relayd_rcu
);
355 * Remove a channel from the global list protected by a mutex. This function is
356 * also responsible for freeing its data structures.
358 void consumer_del_channel(struct lttng_consumer_channel
*channel
)
361 struct lttng_ht_iter iter
;
363 DBG("Consumer delete channel key %" PRIu64
, channel
->key
);
365 pthread_mutex_lock(&consumer_data
.lock
);
366 pthread_mutex_lock(&channel
->lock
);
368 /* Destroy streams that might have been left in the stream list. */
369 clean_channel_stream_list(channel
);
371 if (channel
->live_timer_enabled
== 1) {
372 consumer_timer_live_stop(channel
);
374 if (channel
->monitor_timer_enabled
== 1) {
375 consumer_timer_monitor_stop(channel
);
378 switch (consumer_data
.type
) {
379 case LTTNG_CONSUMER_KERNEL
:
381 case LTTNG_CONSUMER32_UST
:
382 case LTTNG_CONSUMER64_UST
:
383 lttng_ustconsumer_del_channel(channel
);
386 ERR("Unknown consumer_data type");
392 iter
.iter
.node
= &channel
->node
.node
;
393 ret
= lttng_ht_del(consumer_data
.channel_ht
, &iter
);
397 call_rcu(&channel
->node
.head
, free_channel_rcu
);
399 pthread_mutex_unlock(&channel
->lock
);
400 pthread_mutex_unlock(&consumer_data
.lock
);
404 * Iterate over the relayd hash table and destroy each element. Finally,
405 * destroy the whole hash table.
407 static void cleanup_relayd_ht(void)
409 struct lttng_ht_iter iter
;
410 struct consumer_relayd_sock_pair
*relayd
;
414 cds_lfht_for_each_entry(consumer_data
.relayd_ht
->ht
, &iter
.iter
, relayd
,
416 consumer_destroy_relayd(relayd
);
421 lttng_ht_destroy(consumer_data
.relayd_ht
);
425 * Update the end point status of all streams having the given network sequence
426 * index (relayd index).
428 * It's atomically set without having the stream mutex locked which is fine
429 * because we handle the write/read race with a pipe wakeup for each thread.
431 static void update_endpoint_status_by_netidx(uint64_t net_seq_idx
,
432 enum consumer_endpoint_status status
)
434 struct lttng_ht_iter iter
;
435 struct lttng_consumer_stream
*stream
;
437 DBG("Consumer set delete flag on stream by idx %" PRIu64
, net_seq_idx
);
441 /* Let's begin with metadata */
442 cds_lfht_for_each_entry(metadata_ht
->ht
, &iter
.iter
, stream
, node
.node
) {
443 if (stream
->net_seq_idx
== net_seq_idx
) {
444 uatomic_set(&stream
->endpoint_status
, status
);
445 DBG("Delete flag set to metadata stream %d", stream
->wait_fd
);
449 /* Follow up by the data streams */
450 cds_lfht_for_each_entry(data_ht
->ht
, &iter
.iter
, stream
, node
.node
) {
451 if (stream
->net_seq_idx
== net_seq_idx
) {
452 uatomic_set(&stream
->endpoint_status
, status
);
453 DBG("Delete flag set to data stream %d", stream
->wait_fd
);
460 * Cleanup a relayd object by flagging every associated streams for deletion,
461 * destroying the object meaning removing it from the relayd hash table,
462 * closing the sockets and freeing the memory in a RCU call.
464 * If a local data context is available, notify the threads that the streams'
465 * state have changed.
467 static void cleanup_relayd(struct consumer_relayd_sock_pair
*relayd
,
468 struct lttng_consumer_local_data
*ctx
)
474 DBG("Cleaning up relayd sockets");
476 /* Save the net sequence index before destroying the object */
477 netidx
= relayd
->net_seq_idx
;
480 * Delete the relayd from the relayd hash table, close the sockets and free
481 * the object in a RCU call.
483 consumer_destroy_relayd(relayd
);
485 /* Set inactive endpoint to all streams */
486 update_endpoint_status_by_netidx(netidx
, CONSUMER_ENDPOINT_INACTIVE
);
489 * With a local data context, notify the threads that the streams' state
490 * have changed. The write() action on the pipe acts as an "implicit"
491 * memory barrier ordering the updates of the end point status from the
492 * read of this status which happens AFTER receiving this notify.
495 notify_thread_lttng_pipe(ctx
->consumer_data_pipe
);
496 notify_thread_lttng_pipe(ctx
->consumer_metadata_pipe
);
501 * Flag a relayd socket pair for destruction. Destroy it if the refcount
504 * RCU read side lock MUST be aquired before calling this function.
506 void consumer_flag_relayd_for_destroy(struct consumer_relayd_sock_pair
*relayd
)
510 /* Set destroy flag for this object */
511 uatomic_set(&relayd
->destroy_flag
, 1);
513 /* Destroy the relayd if refcount is 0 */
514 if (uatomic_read(&relayd
->refcount
) == 0) {
515 consumer_destroy_relayd(relayd
);
520 * Completly destroy stream from every visiable data structure and the given
523 * One this call returns, the stream object is not longer usable nor visible.
525 void consumer_del_stream(struct lttng_consumer_stream
*stream
,
528 consumer_stream_destroy(stream
, ht
);
532 * XXX naming of del vs destroy is all mixed up.
534 void consumer_del_stream_for_data(struct lttng_consumer_stream
*stream
)
536 consumer_stream_destroy(stream
, data_ht
);
539 void consumer_del_stream_for_metadata(struct lttng_consumer_stream
*stream
)
541 consumer_stream_destroy(stream
, metadata_ht
);
544 void consumer_stream_copy_ro_channel_values(struct lttng_consumer_stream
*stream
,
545 struct lttng_consumer_channel
*channel
)
547 stream
->channel_ro_tracefile_size
= channel
->tracefile_size
;
548 memcpy(stream
->channel_ro_pathname
, channel
->pathname
, PATH_MAX
);
551 struct lttng_consumer_stream
*consumer_allocate_stream(uint64_t channel_key
,
553 enum lttng_consumer_stream_state state
,
554 const char *channel_name
,
561 enum consumer_channel_type type
,
562 unsigned int monitor
)
565 struct lttng_consumer_stream
*stream
;
567 stream
= zmalloc(sizeof(*stream
));
568 if (stream
== NULL
) {
569 PERROR("malloc struct lttng_consumer_stream");
576 stream
->key
= stream_key
;
578 stream
->out_fd_offset
= 0;
579 stream
->output_written
= 0;
580 stream
->state
= state
;
583 stream
->net_seq_idx
= relayd_id
;
584 stream
->session_id
= session_id
;
585 stream
->monitor
= monitor
;
586 stream
->endpoint_status
= CONSUMER_ENDPOINT_ACTIVE
;
587 stream
->index_file
= NULL
;
588 stream
->last_sequence_number
= -1ULL;
589 pthread_mutex_init(&stream
->lock
, NULL
);
590 pthread_mutex_init(&stream
->metadata_timer_lock
, NULL
);
592 /* If channel is the metadata, flag this stream as metadata. */
593 if (type
== CONSUMER_CHANNEL_TYPE_METADATA
) {
594 stream
->metadata_flag
= 1;
595 /* Metadata is flat out. */
596 strncpy(stream
->name
, DEFAULT_METADATA_NAME
, sizeof(stream
->name
));
597 /* Live rendez-vous point. */
598 pthread_cond_init(&stream
->metadata_rdv
, NULL
);
599 pthread_mutex_init(&stream
->metadata_rdv_lock
, NULL
);
601 /* Format stream name to <channel_name>_<cpu_number> */
602 ret
= snprintf(stream
->name
, sizeof(stream
->name
), "%s_%d",
605 PERROR("snprintf stream name");
610 /* Key is always the wait_fd for streams. */
611 lttng_ht_node_init_u64(&stream
->node
, stream
->key
);
613 /* Init node per channel id key */
614 lttng_ht_node_init_u64(&stream
->node_channel_id
, channel_key
);
616 /* Init session id node with the stream session id */
617 lttng_ht_node_init_u64(&stream
->node_session_id
, stream
->session_id
);
619 DBG3("Allocated stream %s (key %" PRIu64
", chan_key %" PRIu64
620 " relayd_id %" PRIu64
", session_id %" PRIu64
,
621 stream
->name
, stream
->key
, channel_key
,
622 stream
->net_seq_idx
, stream
->session_id
);
638 * Add a stream to the global list protected by a mutex.
640 int consumer_add_data_stream(struct lttng_consumer_stream
*stream
)
642 struct lttng_ht
*ht
= data_ht
;
648 DBG3("Adding consumer stream %" PRIu64
, stream
->key
);
650 pthread_mutex_lock(&consumer_data
.lock
);
651 pthread_mutex_lock(&stream
->chan
->lock
);
652 pthread_mutex_lock(&stream
->chan
->timer_lock
);
653 pthread_mutex_lock(&stream
->lock
);
656 /* Steal stream identifier to avoid having streams with the same key */
657 steal_stream_key(stream
->key
, ht
);
659 lttng_ht_add_unique_u64(ht
, &stream
->node
);
661 lttng_ht_add_u64(consumer_data
.stream_per_chan_id_ht
,
662 &stream
->node_channel_id
);
665 * Add stream to the stream_list_ht of the consumer data. No need to steal
666 * the key since the HT does not use it and we allow to add redundant keys
669 lttng_ht_add_u64(consumer_data
.stream_list_ht
, &stream
->node_session_id
);
672 * When nb_init_stream_left reaches 0, we don't need to trigger any action
673 * in terms of destroying the associated channel, because the action that
674 * causes the count to become 0 also causes a stream to be added. The
675 * channel deletion will thus be triggered by the following removal of this
678 if (uatomic_read(&stream
->chan
->nb_init_stream_left
) > 0) {
679 /* Increment refcount before decrementing nb_init_stream_left */
681 uatomic_dec(&stream
->chan
->nb_init_stream_left
);
684 /* Update consumer data once the node is inserted. */
685 consumer_data
.stream_count
++;
686 consumer_data
.need_update
= 1;
689 pthread_mutex_unlock(&stream
->lock
);
690 pthread_mutex_unlock(&stream
->chan
->timer_lock
);
691 pthread_mutex_unlock(&stream
->chan
->lock
);
692 pthread_mutex_unlock(&consumer_data
.lock
);
697 void consumer_del_data_stream(struct lttng_consumer_stream
*stream
)
699 consumer_del_stream(stream
, data_ht
);
703 * Add relayd socket to global consumer data hashtable. RCU read side lock MUST
704 * be acquired before calling this.
706 static int add_relayd(struct consumer_relayd_sock_pair
*relayd
)
709 struct lttng_ht_node_u64
*node
;
710 struct lttng_ht_iter iter
;
714 lttng_ht_lookup(consumer_data
.relayd_ht
,
715 &relayd
->net_seq_idx
, &iter
);
716 node
= lttng_ht_iter_get_node_u64(&iter
);
720 lttng_ht_add_unique_u64(consumer_data
.relayd_ht
, &relayd
->node
);
727 * Allocate and return a consumer relayd socket.
729 static struct consumer_relayd_sock_pair
*consumer_allocate_relayd_sock_pair(
730 uint64_t net_seq_idx
)
732 struct consumer_relayd_sock_pair
*obj
= NULL
;
734 /* net sequence index of -1 is a failure */
735 if (net_seq_idx
== (uint64_t) -1ULL) {
739 obj
= zmalloc(sizeof(struct consumer_relayd_sock_pair
));
741 PERROR("zmalloc relayd sock");
745 obj
->net_seq_idx
= net_seq_idx
;
747 obj
->destroy_flag
= 0;
748 obj
->control_sock
.sock
.fd
= -1;
749 obj
->data_sock
.sock
.fd
= -1;
750 lttng_ht_node_init_u64(&obj
->node
, obj
->net_seq_idx
);
751 pthread_mutex_init(&obj
->ctrl_sock_mutex
, NULL
);
758 * Find a relayd socket pair in the global consumer data.
760 * Return the object if found else NULL.
761 * RCU read-side lock must be held across this call and while using the
764 struct consumer_relayd_sock_pair
*consumer_find_relayd(uint64_t key
)
766 struct lttng_ht_iter iter
;
767 struct lttng_ht_node_u64
*node
;
768 struct consumer_relayd_sock_pair
*relayd
= NULL
;
770 /* Negative keys are lookup failures */
771 if (key
== (uint64_t) -1ULL) {
775 lttng_ht_lookup(consumer_data
.relayd_ht
, &key
,
777 node
= lttng_ht_iter_get_node_u64(&iter
);
779 relayd
= caa_container_of(node
, struct consumer_relayd_sock_pair
, node
);
787 * Find a relayd and send the stream
789 * Returns 0 on success, < 0 on error
791 int consumer_send_relayd_stream(struct lttng_consumer_stream
*stream
,
792 char *path
, enum lttng_domain_type domain
)
795 struct consumer_relayd_sock_pair
*relayd
;
798 assert(stream
->net_seq_idx
!= -1ULL);
801 /* The stream is not metadata. Get relayd reference if exists. */
803 relayd
= consumer_find_relayd(stream
->net_seq_idx
);
804 if (relayd
!= NULL
) {
805 /* Add stream on the relayd */
806 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
807 ret
= relayd_add_stream(&relayd
->control_sock
, stream
->name
,
808 path
, &stream
->relayd_stream_id
,
809 stream
->chan
->tracefile_size
, stream
->chan
->tracefile_count
,
811 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
816 uatomic_inc(&relayd
->refcount
);
817 stream
->sent_to_relayd
= 1;
819 ERR("Stream %" PRIu64
" relayd ID %" PRIu64
" unknown. Can't send it.",
820 stream
->key
, stream
->net_seq_idx
);
825 DBG("Stream %s with key %" PRIu64
" sent to relayd id %" PRIu64
,
826 stream
->name
, stream
->key
, stream
->net_seq_idx
);
834 * Find a relayd and send the streams sent message
836 * Returns 0 on success, < 0 on error
838 int consumer_send_relayd_streams_sent(uint64_t net_seq_idx
)
841 struct consumer_relayd_sock_pair
*relayd
;
843 assert(net_seq_idx
!= -1ULL);
845 /* The stream is not metadata. Get relayd reference if exists. */
847 relayd
= consumer_find_relayd(net_seq_idx
);
848 if (relayd
!= NULL
) {
849 /* Add stream on the relayd */
850 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
851 ret
= relayd_streams_sent(&relayd
->control_sock
);
852 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
857 ERR("Relayd ID %" PRIu64
" unknown. Can't send streams_sent.",
864 DBG("All streams sent relayd id %" PRIu64
, net_seq_idx
);
872 * Find a relayd and close the stream
874 void close_relayd_stream(struct lttng_consumer_stream
*stream
)
876 struct consumer_relayd_sock_pair
*relayd
;
878 /* The stream is not metadata. Get relayd reference if exists. */
880 relayd
= consumer_find_relayd(stream
->net_seq_idx
);
882 consumer_stream_relayd_close(stream
, relayd
);
888 * Handle stream for relayd transmission if the stream applies for network
889 * streaming where the net sequence index is set.
891 * Return destination file descriptor or negative value on error.
893 static int write_relayd_stream_header(struct lttng_consumer_stream
*stream
,
894 size_t data_size
, unsigned long padding
,
895 struct consumer_relayd_sock_pair
*relayd
)
898 struct lttcomm_relayd_data_hdr data_hdr
;
904 /* Reset data header */
905 memset(&data_hdr
, 0, sizeof(data_hdr
));
907 if (stream
->metadata_flag
) {
908 /* Caller MUST acquire the relayd control socket lock */
909 ret
= relayd_send_metadata(&relayd
->control_sock
, data_size
);
914 /* Metadata are always sent on the control socket. */
915 outfd
= relayd
->control_sock
.sock
.fd
;
917 /* Set header with stream information */
918 data_hdr
.stream_id
= htobe64(stream
->relayd_stream_id
);
919 data_hdr
.data_size
= htobe32(data_size
);
920 data_hdr
.padding_size
= htobe32(padding
);
922 * Note that net_seq_num below is assigned with the *current* value of
923 * next_net_seq_num and only after that the next_net_seq_num will be
924 * increment. This is why when issuing a command on the relayd using
925 * this next value, 1 should always be substracted in order to compare
926 * the last seen sequence number on the relayd side to the last sent.
928 data_hdr
.net_seq_num
= htobe64(stream
->next_net_seq_num
);
929 /* Other fields are zeroed previously */
931 ret
= relayd_send_data_hdr(&relayd
->data_sock
, &data_hdr
,
937 ++stream
->next_net_seq_num
;
939 /* Set to go on data socket */
940 outfd
= relayd
->data_sock
.sock
.fd
;
948 * Allocate and return a new lttng_consumer_channel object using the given key
949 * to initialize the hash table node.
951 * On error, return NULL.
953 struct lttng_consumer_channel
*consumer_allocate_channel(uint64_t key
,
955 const char *pathname
,
960 enum lttng_event_output output
,
961 uint64_t tracefile_size
,
962 uint64_t tracefile_count
,
963 uint64_t session_id_per_pid
,
964 unsigned int monitor
,
965 unsigned int live_timer_interval
,
966 const char *root_shm_path
,
967 const char *shm_path
)
969 struct lttng_consumer_channel
*channel
;
971 channel
= zmalloc(sizeof(*channel
));
972 if (channel
== NULL
) {
973 PERROR("malloc struct lttng_consumer_channel");
978 channel
->refcount
= 0;
979 channel
->session_id
= session_id
;
980 channel
->session_id_per_pid
= session_id_per_pid
;
983 channel
->relayd_id
= relayd_id
;
984 channel
->tracefile_size
= tracefile_size
;
985 channel
->tracefile_count
= tracefile_count
;
986 channel
->monitor
= monitor
;
987 channel
->live_timer_interval
= live_timer_interval
;
988 pthread_mutex_init(&channel
->lock
, NULL
);
989 pthread_mutex_init(&channel
->timer_lock
, NULL
);
992 case LTTNG_EVENT_SPLICE
:
993 channel
->output
= CONSUMER_CHANNEL_SPLICE
;
995 case LTTNG_EVENT_MMAP
:
996 channel
->output
= CONSUMER_CHANNEL_MMAP
;
1006 * In monitor mode, the streams associated with the channel will be put in
1007 * a special list ONLY owned by this channel. So, the refcount is set to 1
1008 * here meaning that the channel itself has streams that are referenced.
1010 * On a channel deletion, once the channel is no longer visible, the
1011 * refcount is decremented and checked for a zero value to delete it. With
1012 * streams in no monitor mode, it will now be safe to destroy the channel.
1014 if (!channel
->monitor
) {
1015 channel
->refcount
= 1;
1018 strncpy(channel
->pathname
, pathname
, sizeof(channel
->pathname
));
1019 channel
->pathname
[sizeof(channel
->pathname
) - 1] = '\0';
1021 strncpy(channel
->name
, name
, sizeof(channel
->name
));
1022 channel
->name
[sizeof(channel
->name
) - 1] = '\0';
1024 if (root_shm_path
) {
1025 strncpy(channel
->root_shm_path
, root_shm_path
, sizeof(channel
->root_shm_path
));
1026 channel
->root_shm_path
[sizeof(channel
->root_shm_path
) - 1] = '\0';
1029 strncpy(channel
->shm_path
, shm_path
, sizeof(channel
->shm_path
));
1030 channel
->shm_path
[sizeof(channel
->shm_path
) - 1] = '\0';
1033 lttng_ht_node_init_u64(&channel
->node
, channel
->key
);
1035 channel
->wait_fd
= -1;
1037 CDS_INIT_LIST_HEAD(&channel
->streams
.head
);
1039 DBG("Allocated channel (key %" PRIu64
")", channel
->key
);
1046 * Add a channel to the global list protected by a mutex.
1048 * Always return 0 indicating success.
1050 int consumer_add_channel(struct lttng_consumer_channel
*channel
,
1051 struct lttng_consumer_local_data
*ctx
)
1053 pthread_mutex_lock(&consumer_data
.lock
);
1054 pthread_mutex_lock(&channel
->lock
);
1055 pthread_mutex_lock(&channel
->timer_lock
);
1058 * This gives us a guarantee that the channel we are about to add to the
1059 * channel hash table will be unique. See this function comment on the why
1060 * we need to steel the channel key at this stage.
1062 steal_channel_key(channel
->key
);
1065 lttng_ht_add_unique_u64(consumer_data
.channel_ht
, &channel
->node
);
1068 pthread_mutex_unlock(&channel
->timer_lock
);
1069 pthread_mutex_unlock(&channel
->lock
);
1070 pthread_mutex_unlock(&consumer_data
.lock
);
1072 if (channel
->wait_fd
!= -1 && channel
->type
== CONSUMER_CHANNEL_TYPE_DATA
) {
1073 notify_channel_pipe(ctx
, channel
, -1, CONSUMER_CHANNEL_ADD
);
1080 * Allocate the pollfd structure and the local view of the out fds to avoid
1081 * doing a lookup in the linked list and concurrency issues when writing is
1082 * needed. Called with consumer_data.lock held.
1084 * Returns the number of fds in the structures.
1086 static int update_poll_array(struct lttng_consumer_local_data
*ctx
,
1087 struct pollfd
**pollfd
, struct lttng_consumer_stream
**local_stream
,
1088 struct lttng_ht
*ht
)
1091 struct lttng_ht_iter iter
;
1092 struct lttng_consumer_stream
*stream
;
1097 assert(local_stream
);
1099 DBG("Updating poll fd array");
1101 cds_lfht_for_each_entry(ht
->ht
, &iter
.iter
, stream
, node
.node
) {
1103 * Only active streams with an active end point can be added to the
1104 * poll set and local stream storage of the thread.
1106 * There is a potential race here for endpoint_status to be updated
1107 * just after the check. However, this is OK since the stream(s) will
1108 * be deleted once the thread is notified that the end point state has
1109 * changed where this function will be called back again.
1111 if (stream
->state
!= LTTNG_CONSUMER_ACTIVE_STREAM
||
1112 stream
->endpoint_status
== CONSUMER_ENDPOINT_INACTIVE
) {
1116 * This clobbers way too much the debug output. Uncomment that if you
1117 * need it for debugging purposes.
1119 * DBG("Active FD %d", stream->wait_fd);
1121 (*pollfd
)[i
].fd
= stream
->wait_fd
;
1122 (*pollfd
)[i
].events
= POLLIN
| POLLPRI
;
1123 local_stream
[i
] = stream
;
1129 * Insert the consumer_data_pipe at the end of the array and don't
1130 * increment i so nb_fd is the number of real FD.
1132 (*pollfd
)[i
].fd
= lttng_pipe_get_readfd(ctx
->consumer_data_pipe
);
1133 (*pollfd
)[i
].events
= POLLIN
| POLLPRI
;
1135 (*pollfd
)[i
+ 1].fd
= lttng_pipe_get_readfd(ctx
->consumer_wakeup_pipe
);
1136 (*pollfd
)[i
+ 1].events
= POLLIN
| POLLPRI
;
1141 * Poll on the should_quit pipe and the command socket return -1 on
1142 * error, 1 if should exit, 0 if data is available on the command socket
1144 int lttng_consumer_poll_socket(struct pollfd
*consumer_sockpoll
)
1149 num_rdy
= poll(consumer_sockpoll
, 2, -1);
1150 if (num_rdy
== -1) {
1152 * Restart interrupted system call.
1154 if (errno
== EINTR
) {
1157 PERROR("Poll error");
1160 if (consumer_sockpoll
[0].revents
& (POLLIN
| POLLPRI
)) {
1161 DBG("consumer_should_quit wake up");
1168 * Set the error socket.
1170 void lttng_consumer_set_error_sock(struct lttng_consumer_local_data
*ctx
,
1173 ctx
->consumer_error_socket
= sock
;
1177 * Set the command socket path.
1179 void lttng_consumer_set_command_sock_path(
1180 struct lttng_consumer_local_data
*ctx
, char *sock
)
1182 ctx
->consumer_command_sock_path
= sock
;
1186 * Send return code to the session daemon.
1187 * If the socket is not defined, we return 0, it is not a fatal error
1189 int lttng_consumer_send_error(struct lttng_consumer_local_data
*ctx
, int cmd
)
1191 if (ctx
->consumer_error_socket
> 0) {
1192 return lttcomm_send_unix_sock(ctx
->consumer_error_socket
, &cmd
,
1193 sizeof(enum lttcomm_sessiond_command
));
1200 * Close all the tracefiles and stream fds and MUST be called when all
1201 * instances are destroyed i.e. when all threads were joined and are ended.
1203 void lttng_consumer_cleanup(void)
1205 struct lttng_ht_iter iter
;
1206 struct lttng_consumer_channel
*channel
;
1210 cds_lfht_for_each_entry(consumer_data
.channel_ht
->ht
, &iter
.iter
, channel
,
1212 consumer_del_channel(channel
);
1217 lttng_ht_destroy(consumer_data
.channel_ht
);
1219 cleanup_relayd_ht();
1221 lttng_ht_destroy(consumer_data
.stream_per_chan_id_ht
);
1224 * This HT contains streams that are freed by either the metadata thread or
1225 * the data thread so we do *nothing* on the hash table and simply destroy
1228 lttng_ht_destroy(consumer_data
.stream_list_ht
);
1232 * Called from signal handler.
1234 void lttng_consumer_should_exit(struct lttng_consumer_local_data
*ctx
)
1238 CMM_STORE_SHARED(consumer_quit
, 1);
1239 ret
= lttng_write(ctx
->consumer_should_quit
[1], "4", 1);
1241 PERROR("write consumer quit");
1244 DBG("Consumer flag that it should quit");
1249 * Flush pending writes to trace output disk file.
1252 void lttng_consumer_sync_trace_file(struct lttng_consumer_stream
*stream
,
1256 int outfd
= stream
->out_fd
;
1259 * This does a blocking write-and-wait on any page that belongs to the
1260 * subbuffer prior to the one we just wrote.
1261 * Don't care about error values, as these are just hints and ways to
1262 * limit the amount of page cache used.
1264 if (orig_offset
< stream
->max_sb_size
) {
1267 lttng_sync_file_range(outfd
, orig_offset
- stream
->max_sb_size
,
1268 stream
->max_sb_size
,
1269 SYNC_FILE_RANGE_WAIT_BEFORE
1270 | SYNC_FILE_RANGE_WRITE
1271 | SYNC_FILE_RANGE_WAIT_AFTER
);
1273 * Give hints to the kernel about how we access the file:
1274 * POSIX_FADV_DONTNEED : we won't re-access data in a near future after
1277 * We need to call fadvise again after the file grows because the
1278 * kernel does not seem to apply fadvise to non-existing parts of the
1281 * Call fadvise _after_ having waited for the page writeback to
1282 * complete because the dirty page writeback semantic is not well
1283 * defined. So it can be expected to lead to lower throughput in
1286 ret
= posix_fadvise(outfd
, orig_offset
- stream
->max_sb_size
,
1287 stream
->max_sb_size
, POSIX_FADV_DONTNEED
);
1288 if (ret
&& ret
!= -ENOSYS
) {
1290 PERROR("posix_fadvise on fd %i", outfd
);
1295 * Initialise the necessary environnement :
1296 * - create a new context
1297 * - create the poll_pipe
1298 * - create the should_quit pipe (for signal handler)
1299 * - create the thread pipe (for splice)
1301 * Takes a function pointer as argument, this function is called when data is
1302 * available on a buffer. This function is responsible to do the
1303 * kernctl_get_next_subbuf, read the data with mmap or splice depending on the
1304 * buffer configuration and then kernctl_put_next_subbuf at the end.
1306 * Returns a pointer to the new context or NULL on error.
1308 struct lttng_consumer_local_data
*lttng_consumer_create(
1309 enum lttng_consumer_type type
,
1310 ssize_t (*buffer_ready
)(struct lttng_consumer_stream
*stream
,
1311 struct lttng_consumer_local_data
*ctx
),
1312 int (*recv_channel
)(struct lttng_consumer_channel
*channel
),
1313 int (*recv_stream
)(struct lttng_consumer_stream
*stream
),
1314 int (*update_stream
)(uint64_t stream_key
, uint32_t state
))
1317 struct lttng_consumer_local_data
*ctx
;
1319 assert(consumer_data
.type
== LTTNG_CONSUMER_UNKNOWN
||
1320 consumer_data
.type
== type
);
1321 consumer_data
.type
= type
;
1323 ctx
= zmalloc(sizeof(struct lttng_consumer_local_data
));
1325 PERROR("allocating context");
1329 ctx
->consumer_error_socket
= -1;
1330 ctx
->consumer_metadata_socket
= -1;
1331 pthread_mutex_init(&ctx
->metadata_socket_lock
, NULL
);
1332 /* assign the callbacks */
1333 ctx
->on_buffer_ready
= buffer_ready
;
1334 ctx
->on_recv_channel
= recv_channel
;
1335 ctx
->on_recv_stream
= recv_stream
;
1336 ctx
->on_update_stream
= update_stream
;
1338 ctx
->consumer_data_pipe
= lttng_pipe_open(0);
1339 if (!ctx
->consumer_data_pipe
) {
1340 goto error_poll_pipe
;
1343 ctx
->consumer_wakeup_pipe
= lttng_pipe_open(0);
1344 if (!ctx
->consumer_wakeup_pipe
) {
1345 goto error_wakeup_pipe
;
1348 ret
= pipe(ctx
->consumer_should_quit
);
1350 PERROR("Error creating recv pipe");
1351 goto error_quit_pipe
;
1354 ret
= pipe(ctx
->consumer_channel_pipe
);
1356 PERROR("Error creating channel pipe");
1357 goto error_channel_pipe
;
1360 ctx
->consumer_metadata_pipe
= lttng_pipe_open(0);
1361 if (!ctx
->consumer_metadata_pipe
) {
1362 goto error_metadata_pipe
;
1365 ctx
->channel_monitor_pipe
= -1;
1369 error_metadata_pipe
:
1370 utils_close_pipe(ctx
->consumer_channel_pipe
);
1372 utils_close_pipe(ctx
->consumer_should_quit
);
1374 lttng_pipe_destroy(ctx
->consumer_wakeup_pipe
);
1376 lttng_pipe_destroy(ctx
->consumer_data_pipe
);
1384 * Iterate over all streams of the hashtable and free them properly.
1386 static void destroy_data_stream_ht(struct lttng_ht
*ht
)
1388 struct lttng_ht_iter iter
;
1389 struct lttng_consumer_stream
*stream
;
1396 cds_lfht_for_each_entry(ht
->ht
, &iter
.iter
, stream
, node
.node
) {
1398 * Ignore return value since we are currently cleaning up so any error
1401 (void) consumer_del_stream(stream
, ht
);
1405 lttng_ht_destroy(ht
);
1409 * Iterate over all streams of the metadata hashtable and free them
1412 static void destroy_metadata_stream_ht(struct lttng_ht
*ht
)
1414 struct lttng_ht_iter iter
;
1415 struct lttng_consumer_stream
*stream
;
1422 cds_lfht_for_each_entry(ht
->ht
, &iter
.iter
, stream
, node
.node
) {
1424 * Ignore return value since we are currently cleaning up so any error
1427 (void) consumer_del_metadata_stream(stream
, ht
);
1431 lttng_ht_destroy(ht
);
1435 * Close all fds associated with the instance and free the context.
1437 void lttng_consumer_destroy(struct lttng_consumer_local_data
*ctx
)
1441 DBG("Consumer destroying it. Closing everything.");
1447 destroy_data_stream_ht(data_ht
);
1448 destroy_metadata_stream_ht(metadata_ht
);
1450 ret
= close(ctx
->consumer_error_socket
);
1454 ret
= close(ctx
->consumer_metadata_socket
);
1458 utils_close_pipe(ctx
->consumer_channel_pipe
);
1459 lttng_pipe_destroy(ctx
->consumer_data_pipe
);
1460 lttng_pipe_destroy(ctx
->consumer_metadata_pipe
);
1461 lttng_pipe_destroy(ctx
->consumer_wakeup_pipe
);
1462 utils_close_pipe(ctx
->consumer_should_quit
);
1464 unlink(ctx
->consumer_command_sock_path
);
1469 * Write the metadata stream id on the specified file descriptor.
1471 static int write_relayd_metadata_id(int fd
,
1472 struct lttng_consumer_stream
*stream
,
1473 struct consumer_relayd_sock_pair
*relayd
, unsigned long padding
)
1476 struct lttcomm_relayd_metadata_payload hdr
;
1478 hdr
.stream_id
= htobe64(stream
->relayd_stream_id
);
1479 hdr
.padding_size
= htobe32(padding
);
1480 ret
= lttng_write(fd
, (void *) &hdr
, sizeof(hdr
));
1481 if (ret
< sizeof(hdr
)) {
1483 * This error means that the fd's end is closed so ignore the PERROR
1484 * not to clubber the error output since this can happen in a normal
1487 if (errno
!= EPIPE
) {
1488 PERROR("write metadata stream id");
1490 DBG3("Consumer failed to write relayd metadata id (errno: %d)", errno
);
1492 * Set ret to a negative value because if ret != sizeof(hdr), we don't
1493 * handle writting the missing part so report that as an error and
1494 * don't lie to the caller.
1499 DBG("Metadata stream id %" PRIu64
" with padding %lu written before data",
1500 stream
->relayd_stream_id
, padding
);
1507 * Mmap the ring buffer, read it and write the data to the tracefile. This is a
1508 * core function for writing trace buffers to either the local filesystem or
1511 * It must be called with the stream lock held.
1513 * Careful review MUST be put if any changes occur!
1515 * Returns the number of bytes written
1517 ssize_t
lttng_consumer_on_read_subbuffer_mmap(
1518 struct lttng_consumer_local_data
*ctx
,
1519 struct lttng_consumer_stream
*stream
, unsigned long len
,
1520 unsigned long padding
,
1521 struct ctf_packet_index
*index
)
1523 unsigned long mmap_offset
;
1526 off_t orig_offset
= stream
->out_fd_offset
;
1527 /* Default is on the disk */
1528 int outfd
= stream
->out_fd
;
1529 struct consumer_relayd_sock_pair
*relayd
= NULL
;
1530 unsigned int relayd_hang_up
= 0;
1532 /* RCU lock for the relayd pointer */
1535 /* Flag that the current stream if set for network streaming. */
1536 if (stream
->net_seq_idx
!= (uint64_t) -1ULL) {
1537 relayd
= consumer_find_relayd(stream
->net_seq_idx
);
1538 if (relayd
== NULL
) {
1544 /* get the offset inside the fd to mmap */
1545 switch (consumer_data
.type
) {
1546 case LTTNG_CONSUMER_KERNEL
:
1547 mmap_base
= stream
->mmap_base
;
1548 ret
= kernctl_get_mmap_read_offset(stream
->wait_fd
, &mmap_offset
);
1550 PERROR("tracer ctl get_mmap_read_offset");
1554 case LTTNG_CONSUMER32_UST
:
1555 case LTTNG_CONSUMER64_UST
:
1556 mmap_base
= lttng_ustctl_get_mmap_base(stream
);
1558 ERR("read mmap get mmap base for stream %s", stream
->name
);
1562 ret
= lttng_ustctl_get_mmap_read_offset(stream
, &mmap_offset
);
1564 PERROR("tracer ctl get_mmap_read_offset");
1570 ERR("Unknown consumer_data type");
1574 /* Handle stream on the relayd if the output is on the network */
1576 unsigned long netlen
= len
;
1579 * Lock the control socket for the complete duration of the function
1580 * since from this point on we will use the socket.
1582 if (stream
->metadata_flag
) {
1583 /* Metadata requires the control socket. */
1584 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
1585 if (stream
->reset_metadata_flag
) {
1586 ret
= relayd_reset_metadata(&relayd
->control_sock
,
1587 stream
->relayd_stream_id
,
1588 stream
->metadata_version
);
1593 stream
->reset_metadata_flag
= 0;
1595 netlen
+= sizeof(struct lttcomm_relayd_metadata_payload
);
1598 ret
= write_relayd_stream_header(stream
, netlen
, padding
, relayd
);
1603 /* Use the returned socket. */
1606 /* Write metadata stream id before payload */
1607 if (stream
->metadata_flag
) {
1608 ret
= write_relayd_metadata_id(outfd
, stream
, relayd
, padding
);
1615 /* No streaming, we have to set the len with the full padding */
1618 if (stream
->metadata_flag
&& stream
->reset_metadata_flag
) {
1619 ret
= utils_truncate_stream_file(stream
->out_fd
, 0);
1621 ERR("Reset metadata file");
1624 stream
->reset_metadata_flag
= 0;
1628 * Check if we need to change the tracefile before writing the packet.
1630 if (stream
->chan
->tracefile_size
> 0 &&
1631 (stream
->tracefile_size_current
+ len
) >
1632 stream
->chan
->tracefile_size
) {
1633 ret
= utils_rotate_stream_file(stream
->chan
->pathname
,
1634 stream
->name
, stream
->chan
->tracefile_size
,
1635 stream
->chan
->tracefile_count
, stream
->uid
, stream
->gid
,
1636 stream
->out_fd
, &(stream
->tracefile_count_current
),
1639 ERR("Rotating output file");
1642 outfd
= stream
->out_fd
;
1644 if (stream
->index_file
) {
1645 lttng_index_file_put(stream
->index_file
);
1646 stream
->index_file
= lttng_index_file_create(stream
->chan
->pathname
,
1647 stream
->name
, stream
->uid
, stream
->gid
,
1648 stream
->chan
->tracefile_size
,
1649 stream
->tracefile_count_current
,
1650 CTF_INDEX_MAJOR
, CTF_INDEX_MINOR
);
1651 if (!stream
->index_file
) {
1656 /* Reset current size because we just perform a rotation. */
1657 stream
->tracefile_size_current
= 0;
1658 stream
->out_fd_offset
= 0;
1661 stream
->tracefile_size_current
+= len
;
1663 index
->offset
= htobe64(stream
->out_fd_offset
);
1668 * This call guarantee that len or less is returned. It's impossible to
1669 * receive a ret value that is bigger than len.
1671 ret
= lttng_write(outfd
, mmap_base
+ mmap_offset
, len
);
1672 DBG("Consumer mmap write() ret %zd (len %lu)", ret
, len
);
1673 if (ret
< 0 || ((size_t) ret
!= len
)) {
1675 * Report error to caller if nothing was written else at least send the
1683 /* Socket operation failed. We consider the relayd dead */
1684 if (errno
== EPIPE
|| errno
== EINVAL
|| errno
== EBADF
) {
1686 * This is possible if the fd is closed on the other side
1687 * (outfd) or any write problem. It can be verbose a bit for a
1688 * normal execution if for instance the relayd is stopped
1689 * abruptly. This can happen so set this to a DBG statement.
1691 DBG("Consumer mmap write detected relayd hang up");
1693 /* Unhandled error, print it and stop function right now. */
1694 PERROR("Error in write mmap (ret %zd != len %lu)", ret
, len
);
1698 stream
->output_written
+= ret
;
1700 /* This call is useless on a socket so better save a syscall. */
1702 /* This won't block, but will start writeout asynchronously */
1703 lttng_sync_file_range(outfd
, stream
->out_fd_offset
, len
,
1704 SYNC_FILE_RANGE_WRITE
);
1705 stream
->out_fd_offset
+= len
;
1706 lttng_consumer_sync_trace_file(stream
, orig_offset
);
1711 * This is a special case that the relayd has closed its socket. Let's
1712 * cleanup the relayd object and all associated streams.
1714 if (relayd
&& relayd_hang_up
) {
1715 cleanup_relayd(relayd
, ctx
);
1719 /* Unlock only if ctrl socket used */
1720 if (relayd
&& stream
->metadata_flag
) {
1721 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
1729 * Splice the data from the ring buffer to the tracefile.
1731 * It must be called with the stream lock held.
1733 * Returns the number of bytes spliced.
1735 ssize_t
lttng_consumer_on_read_subbuffer_splice(
1736 struct lttng_consumer_local_data
*ctx
,
1737 struct lttng_consumer_stream
*stream
, unsigned long len
,
1738 unsigned long padding
,
1739 struct ctf_packet_index
*index
)
1741 ssize_t ret
= 0, written
= 0, ret_splice
= 0;
1743 off_t orig_offset
= stream
->out_fd_offset
;
1744 int fd
= stream
->wait_fd
;
1745 /* Default is on the disk */
1746 int outfd
= stream
->out_fd
;
1747 struct consumer_relayd_sock_pair
*relayd
= NULL
;
1749 unsigned int relayd_hang_up
= 0;
1751 switch (consumer_data
.type
) {
1752 case LTTNG_CONSUMER_KERNEL
:
1754 case LTTNG_CONSUMER32_UST
:
1755 case LTTNG_CONSUMER64_UST
:
1756 /* Not supported for user space tracing */
1759 ERR("Unknown consumer_data type");
1763 /* RCU lock for the relayd pointer */
1766 /* Flag that the current stream if set for network streaming. */
1767 if (stream
->net_seq_idx
!= (uint64_t) -1ULL) {
1768 relayd
= consumer_find_relayd(stream
->net_seq_idx
);
1769 if (relayd
== NULL
) {
1774 splice_pipe
= stream
->splice_pipe
;
1776 /* Write metadata stream id before payload */
1778 unsigned long total_len
= len
;
1780 if (stream
->metadata_flag
) {
1782 * Lock the control socket for the complete duration of the function
1783 * since from this point on we will use the socket.
1785 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
1787 if (stream
->reset_metadata_flag
) {
1788 ret
= relayd_reset_metadata(&relayd
->control_sock
,
1789 stream
->relayd_stream_id
,
1790 stream
->metadata_version
);
1795 stream
->reset_metadata_flag
= 0;
1797 ret
= write_relayd_metadata_id(splice_pipe
[1], stream
, relayd
,
1805 total_len
+= sizeof(struct lttcomm_relayd_metadata_payload
);
1808 ret
= write_relayd_stream_header(stream
, total_len
, padding
, relayd
);
1814 /* Use the returned socket. */
1817 /* No streaming, we have to set the len with the full padding */
1820 if (stream
->metadata_flag
&& stream
->reset_metadata_flag
) {
1821 ret
= utils_truncate_stream_file(stream
->out_fd
, 0);
1823 ERR("Reset metadata file");
1826 stream
->reset_metadata_flag
= 0;
1829 * Check if we need to change the tracefile before writing the packet.
1831 if (stream
->chan
->tracefile_size
> 0 &&
1832 (stream
->tracefile_size_current
+ len
) >
1833 stream
->chan
->tracefile_size
) {
1834 ret
= utils_rotate_stream_file(stream
->chan
->pathname
,
1835 stream
->name
, stream
->chan
->tracefile_size
,
1836 stream
->chan
->tracefile_count
, stream
->uid
, stream
->gid
,
1837 stream
->out_fd
, &(stream
->tracefile_count_current
),
1841 ERR("Rotating output file");
1844 outfd
= stream
->out_fd
;
1846 if (stream
->index_file
) {
1847 lttng_index_file_put(stream
->index_file
);
1848 stream
->index_file
= lttng_index_file_create(stream
->chan
->pathname
,
1849 stream
->name
, stream
->uid
, stream
->gid
,
1850 stream
->chan
->tracefile_size
,
1851 stream
->tracefile_count_current
,
1852 CTF_INDEX_MAJOR
, CTF_INDEX_MINOR
);
1853 if (!stream
->index_file
) {
1858 /* Reset current size because we just perform a rotation. */
1859 stream
->tracefile_size_current
= 0;
1860 stream
->out_fd_offset
= 0;
1863 stream
->tracefile_size_current
+= len
;
1864 index
->offset
= htobe64(stream
->out_fd_offset
);
1868 DBG("splice chan to pipe offset %lu of len %lu (fd : %d, pipe: %d)",
1869 (unsigned long)offset
, len
, fd
, splice_pipe
[1]);
1870 ret_splice
= splice(fd
, &offset
, splice_pipe
[1], NULL
, len
,
1871 SPLICE_F_MOVE
| SPLICE_F_MORE
);
1872 DBG("splice chan to pipe, ret %zd", ret_splice
);
1873 if (ret_splice
< 0) {
1876 PERROR("Error in relay splice");
1880 /* Handle stream on the relayd if the output is on the network */
1881 if (relayd
&& stream
->metadata_flag
) {
1882 size_t metadata_payload_size
=
1883 sizeof(struct lttcomm_relayd_metadata_payload
);
1885 /* Update counter to fit the spliced data */
1886 ret_splice
+= metadata_payload_size
;
1887 len
+= metadata_payload_size
;
1889 * We do this so the return value can match the len passed as
1890 * argument to this function.
1892 written
-= metadata_payload_size
;
1895 /* Splice data out */
1896 ret_splice
= splice(splice_pipe
[0], NULL
, outfd
, NULL
,
1897 ret_splice
, SPLICE_F_MOVE
| SPLICE_F_MORE
);
1898 DBG("Consumer splice pipe to file (out_fd: %d), ret %zd",
1900 if (ret_splice
< 0) {
1905 } else if (ret_splice
> len
) {
1907 * We don't expect this code path to be executed but you never know
1908 * so this is an extra protection agains a buggy splice().
1911 written
+= ret_splice
;
1912 PERROR("Wrote more data than requested %zd (len: %lu)", ret_splice
,
1916 /* All good, update current len and continue. */
1920 /* This call is useless on a socket so better save a syscall. */
1922 /* This won't block, but will start writeout asynchronously */
1923 lttng_sync_file_range(outfd
, stream
->out_fd_offset
, ret_splice
,
1924 SYNC_FILE_RANGE_WRITE
);
1925 stream
->out_fd_offset
+= ret_splice
;
1927 stream
->output_written
+= ret_splice
;
1928 written
+= ret_splice
;
1931 lttng_consumer_sync_trace_file(stream
, orig_offset
);
1937 * This is a special case that the relayd has closed its socket. Let's
1938 * cleanup the relayd object and all associated streams.
1940 if (relayd
&& relayd_hang_up
) {
1941 cleanup_relayd(relayd
, ctx
);
1942 /* Skip splice error so the consumer does not fail */
1947 /* send the appropriate error description to sessiond */
1950 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_SPLICE_EINVAL
);
1953 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_SPLICE_ENOMEM
);
1956 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_SPLICE_ESPIPE
);
1961 if (relayd
&& stream
->metadata_flag
) {
1962 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
1970 * Sample the snapshot positions for a specific fd
1972 * Returns 0 on success, < 0 on error
1974 int lttng_consumer_sample_snapshot_positions(struct lttng_consumer_stream
*stream
)
1976 switch (consumer_data
.type
) {
1977 case LTTNG_CONSUMER_KERNEL
:
1978 return lttng_kconsumer_sample_snapshot_positions(stream
);
1979 case LTTNG_CONSUMER32_UST
:
1980 case LTTNG_CONSUMER64_UST
:
1981 return lttng_ustconsumer_sample_snapshot_positions(stream
);
1983 ERR("Unknown consumer_data type");
1989 * Take a snapshot for a specific fd
1991 * Returns 0 on success, < 0 on error
1993 int lttng_consumer_take_snapshot(struct lttng_consumer_stream
*stream
)
1995 switch (consumer_data
.type
) {
1996 case LTTNG_CONSUMER_KERNEL
:
1997 return lttng_kconsumer_take_snapshot(stream
);
1998 case LTTNG_CONSUMER32_UST
:
1999 case LTTNG_CONSUMER64_UST
:
2000 return lttng_ustconsumer_take_snapshot(stream
);
2002 ERR("Unknown consumer_data type");
2009 * Get the produced position
2011 * Returns 0 on success, < 0 on error
2013 int lttng_consumer_get_produced_snapshot(struct lttng_consumer_stream
*stream
,
2016 switch (consumer_data
.type
) {
2017 case LTTNG_CONSUMER_KERNEL
:
2018 return lttng_kconsumer_get_produced_snapshot(stream
, pos
);
2019 case LTTNG_CONSUMER32_UST
:
2020 case LTTNG_CONSUMER64_UST
:
2021 return lttng_ustconsumer_get_produced_snapshot(stream
, pos
);
2023 ERR("Unknown consumer_data type");
2030 * Get the consumed position
2032 * Returns 0 on success, < 0 on error
2034 int lttng_consumer_get_consumed_snapshot(struct lttng_consumer_stream
*stream
,
2037 switch (consumer_data
.type
) {
2038 case LTTNG_CONSUMER_KERNEL
:
2039 return lttng_kconsumer_get_consumed_snapshot(stream
, pos
);
2040 case LTTNG_CONSUMER32_UST
:
2041 case LTTNG_CONSUMER64_UST
:
2042 return lttng_ustconsumer_get_consumed_snapshot(stream
, pos
);
2044 ERR("Unknown consumer_data type");
2050 int lttng_consumer_recv_cmd(struct lttng_consumer_local_data
*ctx
,
2051 int sock
, struct pollfd
*consumer_sockpoll
)
2053 switch (consumer_data
.type
) {
2054 case LTTNG_CONSUMER_KERNEL
:
2055 return lttng_kconsumer_recv_cmd(ctx
, sock
, consumer_sockpoll
);
2056 case LTTNG_CONSUMER32_UST
:
2057 case LTTNG_CONSUMER64_UST
:
2058 return lttng_ustconsumer_recv_cmd(ctx
, sock
, consumer_sockpoll
);
2060 ERR("Unknown consumer_data type");
2066 void lttng_consumer_close_all_metadata(void)
2068 switch (consumer_data
.type
) {
2069 case LTTNG_CONSUMER_KERNEL
:
2071 * The Kernel consumer has a different metadata scheme so we don't
2072 * close anything because the stream will be closed by the session
2076 case LTTNG_CONSUMER32_UST
:
2077 case LTTNG_CONSUMER64_UST
:
2079 * Close all metadata streams. The metadata hash table is passed and
2080 * this call iterates over it by closing all wakeup fd. This is safe
2081 * because at this point we are sure that the metadata producer is
2082 * either dead or blocked.
2084 lttng_ustconsumer_close_all_metadata(metadata_ht
);
2087 ERR("Unknown consumer_data type");
2093 * Clean up a metadata stream and free its memory.
2095 void consumer_del_metadata_stream(struct lttng_consumer_stream
*stream
,
2096 struct lttng_ht
*ht
)
2098 struct lttng_consumer_channel
*free_chan
= NULL
;
2102 * This call should NEVER receive regular stream. It must always be
2103 * metadata stream and this is crucial for data structure synchronization.
2105 assert(stream
->metadata_flag
);
2107 DBG3("Consumer delete metadata stream %d", stream
->wait_fd
);
2109 pthread_mutex_lock(&consumer_data
.lock
);
2110 pthread_mutex_lock(&stream
->chan
->lock
);
2111 pthread_mutex_lock(&stream
->lock
);
2112 if (stream
->chan
->metadata_cache
) {
2113 /* Only applicable to userspace consumers. */
2114 pthread_mutex_lock(&stream
->chan
->metadata_cache
->lock
);
2117 /* Remove any reference to that stream. */
2118 consumer_stream_delete(stream
, ht
);
2120 /* Close down everything including the relayd if one. */
2121 consumer_stream_close(stream
);
2122 /* Destroy tracer buffers of the stream. */
2123 consumer_stream_destroy_buffers(stream
);
2125 /* Atomically decrement channel refcount since other threads can use it. */
2126 if (!uatomic_sub_return(&stream
->chan
->refcount
, 1)
2127 && !uatomic_read(&stream
->chan
->nb_init_stream_left
)) {
2128 /* Go for channel deletion! */
2129 free_chan
= stream
->chan
;
2133 * Nullify the stream reference so it is not used after deletion. The
2134 * channel lock MUST be acquired before being able to check for a NULL
2137 stream
->chan
->metadata_stream
= NULL
;
2139 if (stream
->chan
->metadata_cache
) {
2140 pthread_mutex_unlock(&stream
->chan
->metadata_cache
->lock
);
2142 pthread_mutex_unlock(&stream
->lock
);
2143 pthread_mutex_unlock(&stream
->chan
->lock
);
2144 pthread_mutex_unlock(&consumer_data
.lock
);
2147 consumer_del_channel(free_chan
);
2150 consumer_stream_free(stream
);
2154 * Action done with the metadata stream when adding it to the consumer internal
2155 * data structures to handle it.
2157 int consumer_add_metadata_stream(struct lttng_consumer_stream
*stream
)
2159 struct lttng_ht
*ht
= metadata_ht
;
2161 struct lttng_ht_iter iter
;
2162 struct lttng_ht_node_u64
*node
;
2167 DBG3("Adding metadata stream %" PRIu64
" to hash table", stream
->key
);
2169 pthread_mutex_lock(&consumer_data
.lock
);
2170 pthread_mutex_lock(&stream
->chan
->lock
);
2171 pthread_mutex_lock(&stream
->chan
->timer_lock
);
2172 pthread_mutex_lock(&stream
->lock
);
2175 * From here, refcounts are updated so be _careful_ when returning an error
2182 * Lookup the stream just to make sure it does not exist in our internal
2183 * state. This should NEVER happen.
2185 lttng_ht_lookup(ht
, &stream
->key
, &iter
);
2186 node
= lttng_ht_iter_get_node_u64(&iter
);
2190 * When nb_init_stream_left reaches 0, we don't need to trigger any action
2191 * in terms of destroying the associated channel, because the action that
2192 * causes the count to become 0 also causes a stream to be added. The
2193 * channel deletion will thus be triggered by the following removal of this
2196 if (uatomic_read(&stream
->chan
->nb_init_stream_left
) > 0) {
2197 /* Increment refcount before decrementing nb_init_stream_left */
2199 uatomic_dec(&stream
->chan
->nb_init_stream_left
);
2202 lttng_ht_add_unique_u64(ht
, &stream
->node
);
2204 lttng_ht_add_unique_u64(consumer_data
.stream_per_chan_id_ht
,
2205 &stream
->node_channel_id
);
2208 * Add stream to the stream_list_ht of the consumer data. No need to steal
2209 * the key since the HT does not use it and we allow to add redundant keys
2212 lttng_ht_add_u64(consumer_data
.stream_list_ht
, &stream
->node_session_id
);
2216 pthread_mutex_unlock(&stream
->lock
);
2217 pthread_mutex_unlock(&stream
->chan
->lock
);
2218 pthread_mutex_unlock(&stream
->chan
->timer_lock
);
2219 pthread_mutex_unlock(&consumer_data
.lock
);
2224 * Delete data stream that are flagged for deletion (endpoint_status).
2226 static void validate_endpoint_status_data_stream(void)
2228 struct lttng_ht_iter iter
;
2229 struct lttng_consumer_stream
*stream
;
2231 DBG("Consumer delete flagged data stream");
2234 cds_lfht_for_each_entry(data_ht
->ht
, &iter
.iter
, stream
, node
.node
) {
2235 /* Validate delete flag of the stream */
2236 if (stream
->endpoint_status
== CONSUMER_ENDPOINT_ACTIVE
) {
2239 /* Delete it right now */
2240 consumer_del_stream(stream
, data_ht
);
2246 * Delete metadata stream that are flagged for deletion (endpoint_status).
2248 static void validate_endpoint_status_metadata_stream(
2249 struct lttng_poll_event
*pollset
)
2251 struct lttng_ht_iter iter
;
2252 struct lttng_consumer_stream
*stream
;
2254 DBG("Consumer delete flagged metadata stream");
2259 cds_lfht_for_each_entry(metadata_ht
->ht
, &iter
.iter
, stream
, node
.node
) {
2260 /* Validate delete flag of the stream */
2261 if (stream
->endpoint_status
== CONSUMER_ENDPOINT_ACTIVE
) {
2265 * Remove from pollset so the metadata thread can continue without
2266 * blocking on a deleted stream.
2268 lttng_poll_del(pollset
, stream
->wait_fd
);
2270 /* Delete it right now */
2271 consumer_del_metadata_stream(stream
, metadata_ht
);
2277 * Thread polls on metadata file descriptor and write them on disk or on the
2280 void *consumer_thread_metadata_poll(void *data
)
2282 int ret
, i
, pollfd
, err
= -1;
2283 uint32_t revents
, nb_fd
;
2284 struct lttng_consumer_stream
*stream
= NULL
;
2285 struct lttng_ht_iter iter
;
2286 struct lttng_ht_node_u64
*node
;
2287 struct lttng_poll_event events
;
2288 struct lttng_consumer_local_data
*ctx
= data
;
2291 rcu_register_thread();
2293 health_register(health_consumerd
, HEALTH_CONSUMERD_TYPE_METADATA
);
2295 if (testpoint(consumerd_thread_metadata
)) {
2296 goto error_testpoint
;
2299 health_code_update();
2301 DBG("Thread metadata poll started");
2303 /* Size is set to 1 for the consumer_metadata pipe */
2304 ret
= lttng_poll_create(&events
, 2, LTTNG_CLOEXEC
);
2306 ERR("Poll set creation failed");
2310 ret
= lttng_poll_add(&events
,
2311 lttng_pipe_get_readfd(ctx
->consumer_metadata_pipe
), LPOLLIN
);
2317 DBG("Metadata main loop started");
2321 health_code_update();
2322 health_poll_entry();
2323 DBG("Metadata poll wait");
2324 ret
= lttng_poll_wait(&events
, -1);
2325 DBG("Metadata poll return from wait with %d fd(s)",
2326 LTTNG_POLL_GETNB(&events
));
2328 DBG("Metadata event caught in thread");
2330 if (errno
== EINTR
) {
2331 ERR("Poll EINTR caught");
2334 if (LTTNG_POLL_GETNB(&events
) == 0) {
2335 err
= 0; /* All is OK */
2342 /* From here, the event is a metadata wait fd */
2343 for (i
= 0; i
< nb_fd
; i
++) {
2344 health_code_update();
2346 revents
= LTTNG_POLL_GETEV(&events
, i
);
2347 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2350 /* No activity for this FD (poll implementation). */
2354 if (pollfd
== lttng_pipe_get_readfd(ctx
->consumer_metadata_pipe
)) {
2355 if (revents
& LPOLLIN
) {
2358 pipe_len
= lttng_pipe_read(ctx
->consumer_metadata_pipe
,
2359 &stream
, sizeof(stream
));
2360 if (pipe_len
< sizeof(stream
)) {
2362 PERROR("read metadata stream");
2365 * Remove the pipe from the poll set and continue the loop
2366 * since their might be data to consume.
2368 lttng_poll_del(&events
,
2369 lttng_pipe_get_readfd(ctx
->consumer_metadata_pipe
));
2370 lttng_pipe_read_close(ctx
->consumer_metadata_pipe
);
2374 /* A NULL stream means that the state has changed. */
2375 if (stream
== NULL
) {
2376 /* Check for deleted streams. */
2377 validate_endpoint_status_metadata_stream(&events
);
2381 DBG("Adding metadata stream %d to poll set",
2384 /* Add metadata stream to the global poll events list */
2385 lttng_poll_add(&events
, stream
->wait_fd
,
2386 LPOLLIN
| LPOLLPRI
| LPOLLHUP
);
2387 } else if (revents
& (LPOLLERR
| LPOLLHUP
)) {
2388 DBG("Metadata thread pipe hung up");
2390 * Remove the pipe from the poll set and continue the loop
2391 * since their might be data to consume.
2393 lttng_poll_del(&events
,
2394 lttng_pipe_get_readfd(ctx
->consumer_metadata_pipe
));
2395 lttng_pipe_read_close(ctx
->consumer_metadata_pipe
);
2398 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
2402 /* Handle other stream */
2408 uint64_t tmp_id
= (uint64_t) pollfd
;
2410 lttng_ht_lookup(metadata_ht
, &tmp_id
, &iter
);
2412 node
= lttng_ht_iter_get_node_u64(&iter
);
2415 stream
= caa_container_of(node
, struct lttng_consumer_stream
,
2418 if (revents
& (LPOLLIN
| LPOLLPRI
)) {
2419 /* Get the data out of the metadata file descriptor */
2420 DBG("Metadata available on fd %d", pollfd
);
2421 assert(stream
->wait_fd
== pollfd
);
2424 health_code_update();
2426 len
= ctx
->on_buffer_ready(stream
, ctx
);
2428 * We don't check the return value here since if we get
2429 * a negative len, it means an error occurred thus we
2430 * simply remove it from the poll set and free the
2435 /* It's ok to have an unavailable sub-buffer */
2436 if (len
< 0 && len
!= -EAGAIN
&& len
!= -ENODATA
) {
2437 /* Clean up stream from consumer and free it. */
2438 lttng_poll_del(&events
, stream
->wait_fd
);
2439 consumer_del_metadata_stream(stream
, metadata_ht
);
2441 } else if (revents
& (LPOLLERR
| LPOLLHUP
)) {
2442 DBG("Metadata fd %d is hup|err.", pollfd
);
2443 if (!stream
->hangup_flush_done
2444 && (consumer_data
.type
== LTTNG_CONSUMER32_UST
2445 || consumer_data
.type
== LTTNG_CONSUMER64_UST
)) {
2446 DBG("Attempting to flush and consume the UST buffers");
2447 lttng_ustconsumer_on_stream_hangup(stream
);
2449 /* We just flushed the stream now read it. */
2451 health_code_update();
2453 len
= ctx
->on_buffer_ready(stream
, ctx
);
2455 * We don't check the return value here since if we get
2456 * a negative len, it means an error occurred thus we
2457 * simply remove it from the poll set and free the
2463 lttng_poll_del(&events
, stream
->wait_fd
);
2465 * This call update the channel states, closes file descriptors
2466 * and securely free the stream.
2468 consumer_del_metadata_stream(stream
, metadata_ht
);
2470 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
2474 /* Release RCU lock for the stream looked up */
2482 DBG("Metadata poll thread exiting");
2484 lttng_poll_clean(&events
);
2489 ERR("Health error occurred in %s", __func__
);
2491 health_unregister(health_consumerd
);
2492 rcu_unregister_thread();
2497 * This thread polls the fds in the set to consume the data and write
2498 * it to tracefile if necessary.
2500 void *consumer_thread_data_poll(void *data
)
2502 int num_rdy
, num_hup
, high_prio
, ret
, i
, err
= -1;
2503 struct pollfd
*pollfd
= NULL
;
2504 /* local view of the streams */
2505 struct lttng_consumer_stream
**local_stream
= NULL
, *new_stream
= NULL
;
2506 /* local view of consumer_data.fds_count */
2508 struct lttng_consumer_local_data
*ctx
= data
;
2511 rcu_register_thread();
2513 health_register(health_consumerd
, HEALTH_CONSUMERD_TYPE_DATA
);
2515 if (testpoint(consumerd_thread_data
)) {
2516 goto error_testpoint
;
2519 health_code_update();
2521 local_stream
= zmalloc(sizeof(struct lttng_consumer_stream
*));
2522 if (local_stream
== NULL
) {
2523 PERROR("local_stream malloc");
2528 health_code_update();
2534 * the fds set has been updated, we need to update our
2535 * local array as well
2537 pthread_mutex_lock(&consumer_data
.lock
);
2538 if (consumer_data
.need_update
) {
2543 local_stream
= NULL
;
2546 * Allocate for all fds +1 for the consumer_data_pipe and +1 for
2549 pollfd
= zmalloc((consumer_data
.stream_count
+ 2) * sizeof(struct pollfd
));
2550 if (pollfd
== NULL
) {
2551 PERROR("pollfd malloc");
2552 pthread_mutex_unlock(&consumer_data
.lock
);
2556 local_stream
= zmalloc((consumer_data
.stream_count
+ 2) *
2557 sizeof(struct lttng_consumer_stream
*));
2558 if (local_stream
== NULL
) {
2559 PERROR("local_stream malloc");
2560 pthread_mutex_unlock(&consumer_data
.lock
);
2563 ret
= update_poll_array(ctx
, &pollfd
, local_stream
,
2566 ERR("Error in allocating pollfd or local_outfds");
2567 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_POLL_ERROR
);
2568 pthread_mutex_unlock(&consumer_data
.lock
);
2572 consumer_data
.need_update
= 0;
2574 pthread_mutex_unlock(&consumer_data
.lock
);
2576 /* No FDs and consumer_quit, consumer_cleanup the thread */
2577 if (nb_fd
== 0 && CMM_LOAD_SHARED(consumer_quit
) == 1) {
2578 err
= 0; /* All is OK */
2581 /* poll on the array of fds */
2583 DBG("polling on %d fd", nb_fd
+ 2);
2584 if (testpoint(consumerd_thread_data_poll
)) {
2587 health_poll_entry();
2588 num_rdy
= poll(pollfd
, nb_fd
+ 2, -1);
2590 DBG("poll num_rdy : %d", num_rdy
);
2591 if (num_rdy
== -1) {
2593 * Restart interrupted system call.
2595 if (errno
== EINTR
) {
2598 PERROR("Poll error");
2599 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_POLL_ERROR
);
2601 } else if (num_rdy
== 0) {
2602 DBG("Polling thread timed out");
2606 if (caa_unlikely(data_consumption_paused
)) {
2607 DBG("Data consumption paused, sleeping...");
2613 * If the consumer_data_pipe triggered poll go directly to the
2614 * beginning of the loop to update the array. We want to prioritize
2615 * array update over low-priority reads.
2617 if (pollfd
[nb_fd
].revents
& (POLLIN
| POLLPRI
)) {
2618 ssize_t pipe_readlen
;
2620 DBG("consumer_data_pipe wake up");
2621 pipe_readlen
= lttng_pipe_read(ctx
->consumer_data_pipe
,
2622 &new_stream
, sizeof(new_stream
));
2623 if (pipe_readlen
< sizeof(new_stream
)) {
2624 PERROR("Consumer data pipe");
2625 /* Continue so we can at least handle the current stream(s). */
2630 * If the stream is NULL, just ignore it. It's also possible that
2631 * the sessiond poll thread changed the consumer_quit state and is
2632 * waking us up to test it.
2634 if (new_stream
== NULL
) {
2635 validate_endpoint_status_data_stream();
2639 /* Continue to update the local streams and handle prio ones */
2643 /* Handle wakeup pipe. */
2644 if (pollfd
[nb_fd
+ 1].revents
& (POLLIN
| POLLPRI
)) {
2646 ssize_t pipe_readlen
;
2648 pipe_readlen
= lttng_pipe_read(ctx
->consumer_wakeup_pipe
, &dummy
,
2650 if (pipe_readlen
< 0) {
2651 PERROR("Consumer data wakeup pipe");
2653 /* We've been awakened to handle stream(s). */
2654 ctx
->has_wakeup
= 0;
2657 /* Take care of high priority channels first. */
2658 for (i
= 0; i
< nb_fd
; i
++) {
2659 health_code_update();
2661 if (local_stream
[i
] == NULL
) {
2664 if (pollfd
[i
].revents
& POLLPRI
) {
2665 DBG("Urgent read on fd %d", pollfd
[i
].fd
);
2667 len
= ctx
->on_buffer_ready(local_stream
[i
], ctx
);
2668 /* it's ok to have an unavailable sub-buffer */
2669 if (len
< 0 && len
!= -EAGAIN
&& len
!= -ENODATA
) {
2670 /* Clean the stream and free it. */
2671 consumer_del_stream(local_stream
[i
], data_ht
);
2672 local_stream
[i
] = NULL
;
2673 } else if (len
> 0) {
2674 local_stream
[i
]->data_read
= 1;
2680 * If we read high prio channel in this loop, try again
2681 * for more high prio data.
2687 /* Take care of low priority channels. */
2688 for (i
= 0; i
< nb_fd
; i
++) {
2689 health_code_update();
2691 if (local_stream
[i
] == NULL
) {
2694 if ((pollfd
[i
].revents
& POLLIN
) ||
2695 local_stream
[i
]->hangup_flush_done
||
2696 local_stream
[i
]->has_data
) {
2697 DBG("Normal read on fd %d", pollfd
[i
].fd
);
2698 len
= ctx
->on_buffer_ready(local_stream
[i
], ctx
);
2699 /* it's ok to have an unavailable sub-buffer */
2700 if (len
< 0 && len
!= -EAGAIN
&& len
!= -ENODATA
) {
2701 /* Clean the stream and free it. */
2702 consumer_del_stream(local_stream
[i
], data_ht
);
2703 local_stream
[i
] = NULL
;
2704 } else if (len
> 0) {
2705 local_stream
[i
]->data_read
= 1;
2710 /* Handle hangup and errors */
2711 for (i
= 0; i
< nb_fd
; i
++) {
2712 health_code_update();
2714 if (local_stream
[i
] == NULL
) {
2717 if (!local_stream
[i
]->hangup_flush_done
2718 && (pollfd
[i
].revents
& (POLLHUP
| POLLERR
| POLLNVAL
))
2719 && (consumer_data
.type
== LTTNG_CONSUMER32_UST
2720 || consumer_data
.type
== LTTNG_CONSUMER64_UST
)) {
2721 DBG("fd %d is hup|err|nval. Attempting flush and read.",
2723 lttng_ustconsumer_on_stream_hangup(local_stream
[i
]);
2724 /* Attempt read again, for the data we just flushed. */
2725 local_stream
[i
]->data_read
= 1;
2728 * If the poll flag is HUP/ERR/NVAL and we have
2729 * read no data in this pass, we can remove the
2730 * stream from its hash table.
2732 if ((pollfd
[i
].revents
& POLLHUP
)) {
2733 DBG("Polling fd %d tells it has hung up.", pollfd
[i
].fd
);
2734 if (!local_stream
[i
]->data_read
) {
2735 consumer_del_stream(local_stream
[i
], data_ht
);
2736 local_stream
[i
] = NULL
;
2739 } else if (pollfd
[i
].revents
& POLLERR
) {
2740 ERR("Error returned in polling fd %d.", pollfd
[i
].fd
);
2741 if (!local_stream
[i
]->data_read
) {
2742 consumer_del_stream(local_stream
[i
], data_ht
);
2743 local_stream
[i
] = NULL
;
2746 } else if (pollfd
[i
].revents
& POLLNVAL
) {
2747 ERR("Polling fd %d tells fd is not open.", pollfd
[i
].fd
);
2748 if (!local_stream
[i
]->data_read
) {
2749 consumer_del_stream(local_stream
[i
], data_ht
);
2750 local_stream
[i
] = NULL
;
2754 if (local_stream
[i
] != NULL
) {
2755 local_stream
[i
]->data_read
= 0;
2762 DBG("polling thread exiting");
2767 * Close the write side of the pipe so epoll_wait() in
2768 * consumer_thread_metadata_poll can catch it. The thread is monitoring the
2769 * read side of the pipe. If we close them both, epoll_wait strangely does
2770 * not return and could create a endless wait period if the pipe is the
2771 * only tracked fd in the poll set. The thread will take care of closing
2774 (void) lttng_pipe_write_close(ctx
->consumer_metadata_pipe
);
2779 ERR("Health error occurred in %s", __func__
);
2781 health_unregister(health_consumerd
);
2783 rcu_unregister_thread();
2788 * Close wake-up end of each stream belonging to the channel. This will
2789 * allow the poll() on the stream read-side to detect when the
2790 * write-side (application) finally closes them.
2793 void consumer_close_channel_streams(struct lttng_consumer_channel
*channel
)
2795 struct lttng_ht
*ht
;
2796 struct lttng_consumer_stream
*stream
;
2797 struct lttng_ht_iter iter
;
2799 ht
= consumer_data
.stream_per_chan_id_ht
;
2802 cds_lfht_for_each_entry_duplicate(ht
->ht
,
2803 ht
->hash_fct(&channel
->key
, lttng_ht_seed
),
2804 ht
->match_fct
, &channel
->key
,
2805 &iter
.iter
, stream
, node_channel_id
.node
) {
2807 * Protect against teardown with mutex.
2809 pthread_mutex_lock(&stream
->lock
);
2810 if (cds_lfht_is_node_deleted(&stream
->node
.node
)) {
2813 switch (consumer_data
.type
) {
2814 case LTTNG_CONSUMER_KERNEL
:
2816 case LTTNG_CONSUMER32_UST
:
2817 case LTTNG_CONSUMER64_UST
:
2818 if (stream
->metadata_flag
) {
2819 /* Safe and protected by the stream lock. */
2820 lttng_ustconsumer_close_metadata(stream
->chan
);
2823 * Note: a mutex is taken internally within
2824 * liblttng-ust-ctl to protect timer wakeup_fd
2825 * use from concurrent close.
2827 lttng_ustconsumer_close_stream_wakeup(stream
);
2831 ERR("Unknown consumer_data type");
2835 pthread_mutex_unlock(&stream
->lock
);
2840 static void destroy_channel_ht(struct lttng_ht
*ht
)
2842 struct lttng_ht_iter iter
;
2843 struct lttng_consumer_channel
*channel
;
2851 cds_lfht_for_each_entry(ht
->ht
, &iter
.iter
, channel
, wait_fd_node
.node
) {
2852 ret
= lttng_ht_del(ht
, &iter
);
2857 lttng_ht_destroy(ht
);
2861 * This thread polls the channel fds to detect when they are being
2862 * closed. It closes all related streams if the channel is detected as
2863 * closed. It is currently only used as a shim layer for UST because the
2864 * consumerd needs to keep the per-stream wakeup end of pipes open for
2867 void *consumer_thread_channel_poll(void *data
)
2869 int ret
, i
, pollfd
, err
= -1;
2870 uint32_t revents
, nb_fd
;
2871 struct lttng_consumer_channel
*chan
= NULL
;
2872 struct lttng_ht_iter iter
;
2873 struct lttng_ht_node_u64
*node
;
2874 struct lttng_poll_event events
;
2875 struct lttng_consumer_local_data
*ctx
= data
;
2876 struct lttng_ht
*channel_ht
;
2878 rcu_register_thread();
2880 health_register(health_consumerd
, HEALTH_CONSUMERD_TYPE_CHANNEL
);
2882 if (testpoint(consumerd_thread_channel
)) {
2883 goto error_testpoint
;
2886 health_code_update();
2888 channel_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
2890 /* ENOMEM at this point. Better to bail out. */
2894 DBG("Thread channel poll started");
2896 /* Size is set to 1 for the consumer_channel pipe */
2897 ret
= lttng_poll_create(&events
, 2, LTTNG_CLOEXEC
);
2899 ERR("Poll set creation failed");
2903 ret
= lttng_poll_add(&events
, ctx
->consumer_channel_pipe
[0], LPOLLIN
);
2909 DBG("Channel main loop started");
2913 health_code_update();
2914 DBG("Channel poll wait");
2915 health_poll_entry();
2916 ret
= lttng_poll_wait(&events
, -1);
2917 DBG("Channel poll return from wait with %d fd(s)",
2918 LTTNG_POLL_GETNB(&events
));
2920 DBG("Channel event caught in thread");
2922 if (errno
== EINTR
) {
2923 ERR("Poll EINTR caught");
2926 if (LTTNG_POLL_GETNB(&events
) == 0) {
2927 err
= 0; /* All is OK */
2934 /* From here, the event is a channel wait fd */
2935 for (i
= 0; i
< nb_fd
; i
++) {
2936 health_code_update();
2938 revents
= LTTNG_POLL_GETEV(&events
, i
);
2939 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2942 /* No activity for this FD (poll implementation). */
2946 if (pollfd
== ctx
->consumer_channel_pipe
[0]) {
2947 if (revents
& LPOLLIN
) {
2948 enum consumer_channel_action action
;
2951 ret
= read_channel_pipe(ctx
, &chan
, &key
, &action
);
2954 ERR("Error reading channel pipe");
2956 lttng_poll_del(&events
, ctx
->consumer_channel_pipe
[0]);
2961 case CONSUMER_CHANNEL_ADD
:
2962 DBG("Adding channel %d to poll set",
2965 lttng_ht_node_init_u64(&chan
->wait_fd_node
,
2968 lttng_ht_add_unique_u64(channel_ht
,
2969 &chan
->wait_fd_node
);
2971 /* Add channel to the global poll events list */
2972 lttng_poll_add(&events
, chan
->wait_fd
,
2973 LPOLLERR
| LPOLLHUP
);
2975 case CONSUMER_CHANNEL_DEL
:
2978 * This command should never be called if the channel
2979 * has streams monitored by either the data or metadata
2980 * thread. The consumer only notify this thread with a
2981 * channel del. command if it receives a destroy
2982 * channel command from the session daemon that send it
2983 * if a command prior to the GET_CHANNEL failed.
2987 chan
= consumer_find_channel(key
);
2990 ERR("UST consumer get channel key %" PRIu64
" not found for del channel", key
);
2993 lttng_poll_del(&events
, chan
->wait_fd
);
2994 iter
.iter
.node
= &chan
->wait_fd_node
.node
;
2995 ret
= lttng_ht_del(channel_ht
, &iter
);
2998 switch (consumer_data
.type
) {
2999 case LTTNG_CONSUMER_KERNEL
:
3001 case LTTNG_CONSUMER32_UST
:
3002 case LTTNG_CONSUMER64_UST
:
3003 health_code_update();
3004 /* Destroy streams that might have been left in the stream list. */
3005 clean_channel_stream_list(chan
);
3008 ERR("Unknown consumer_data type");
3013 * Release our own refcount. Force channel deletion even if
3014 * streams were not initialized.
3016 if (!uatomic_sub_return(&chan
->refcount
, 1)) {
3017 consumer_del_channel(chan
);
3022 case CONSUMER_CHANNEL_QUIT
:
3024 * Remove the pipe from the poll set and continue the loop
3025 * since their might be data to consume.
3027 lttng_poll_del(&events
, ctx
->consumer_channel_pipe
[0]);
3030 ERR("Unknown action");
3033 } else if (revents
& (LPOLLERR
| LPOLLHUP
)) {
3034 DBG("Channel thread pipe hung up");
3036 * Remove the pipe from the poll set and continue the loop
3037 * since their might be data to consume.
3039 lttng_poll_del(&events
, ctx
->consumer_channel_pipe
[0]);
3042 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
3046 /* Handle other stream */
3052 uint64_t tmp_id
= (uint64_t) pollfd
;
3054 lttng_ht_lookup(channel_ht
, &tmp_id
, &iter
);
3056 node
= lttng_ht_iter_get_node_u64(&iter
);
3059 chan
= caa_container_of(node
, struct lttng_consumer_channel
,
3062 /* Check for error event */
3063 if (revents
& (LPOLLERR
| LPOLLHUP
)) {
3064 DBG("Channel fd %d is hup|err.", pollfd
);
3066 lttng_poll_del(&events
, chan
->wait_fd
);
3067 ret
= lttng_ht_del(channel_ht
, &iter
);
3071 * This will close the wait fd for each stream associated to
3072 * this channel AND monitored by the data/metadata thread thus
3073 * will be clean by the right thread.
3075 consumer_close_channel_streams(chan
);
3077 /* Release our own refcount */
3078 if (!uatomic_sub_return(&chan
->refcount
, 1)
3079 && !uatomic_read(&chan
->nb_init_stream_left
)) {
3080 consumer_del_channel(chan
);
3083 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
3088 /* Release RCU lock for the channel looked up */
3096 lttng_poll_clean(&events
);
3098 destroy_channel_ht(channel_ht
);
3101 DBG("Channel poll thread exiting");
3104 ERR("Health error occurred in %s", __func__
);
3106 health_unregister(health_consumerd
);
3107 rcu_unregister_thread();
3111 static int set_metadata_socket(struct lttng_consumer_local_data
*ctx
,
3112 struct pollfd
*sockpoll
, int client_socket
)
3119 ret
= lttng_consumer_poll_socket(sockpoll
);
3123 DBG("Metadata connection on client_socket");
3125 /* Blocking call, waiting for transmission */
3126 ctx
->consumer_metadata_socket
= lttcomm_accept_unix_sock(client_socket
);
3127 if (ctx
->consumer_metadata_socket
< 0) {
3128 WARN("On accept metadata");
3139 * This thread listens on the consumerd socket and receives the file
3140 * descriptors from the session daemon.
3142 void *consumer_thread_sessiond_poll(void *data
)
3144 int sock
= -1, client_socket
, ret
, err
= -1;
3146 * structure to poll for incoming data on communication socket avoids
3147 * making blocking sockets.
3149 struct pollfd consumer_sockpoll
[2];
3150 struct lttng_consumer_local_data
*ctx
= data
;
3152 rcu_register_thread();
3154 health_register(health_consumerd
, HEALTH_CONSUMERD_TYPE_SESSIOND
);
3156 if (testpoint(consumerd_thread_sessiond
)) {
3157 goto error_testpoint
;
3160 health_code_update();
3162 DBG("Creating command socket %s", ctx
->consumer_command_sock_path
);
3163 unlink(ctx
->consumer_command_sock_path
);
3164 client_socket
= lttcomm_create_unix_sock(ctx
->consumer_command_sock_path
);
3165 if (client_socket
< 0) {
3166 ERR("Cannot create command socket");
3170 ret
= lttcomm_listen_unix_sock(client_socket
);
3175 DBG("Sending ready command to lttng-sessiond");
3176 ret
= lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_COMMAND_SOCK_READY
);
3177 /* return < 0 on error, but == 0 is not fatal */
3179 ERR("Error sending ready command to lttng-sessiond");
3183 /* prepare the FDs to poll : to client socket and the should_quit pipe */
3184 consumer_sockpoll
[0].fd
= ctx
->consumer_should_quit
[0];
3185 consumer_sockpoll
[0].events
= POLLIN
| POLLPRI
;
3186 consumer_sockpoll
[1].fd
= client_socket
;
3187 consumer_sockpoll
[1].events
= POLLIN
| POLLPRI
;
3189 ret
= lttng_consumer_poll_socket(consumer_sockpoll
);
3197 DBG("Connection on client_socket");
3199 /* Blocking call, waiting for transmission */
3200 sock
= lttcomm_accept_unix_sock(client_socket
);
3207 * Setup metadata socket which is the second socket connection on the
3208 * command unix socket.
3210 ret
= set_metadata_socket(ctx
, consumer_sockpoll
, client_socket
);
3219 /* This socket is not useful anymore. */
3220 ret
= close(client_socket
);
3222 PERROR("close client_socket");
3226 /* update the polling structure to poll on the established socket */
3227 consumer_sockpoll
[1].fd
= sock
;
3228 consumer_sockpoll
[1].events
= POLLIN
| POLLPRI
;
3231 health_code_update();
3233 health_poll_entry();
3234 ret
= lttng_consumer_poll_socket(consumer_sockpoll
);
3243 DBG("Incoming command on sock");
3244 ret
= lttng_consumer_recv_cmd(ctx
, sock
, consumer_sockpoll
);
3247 * This could simply be a session daemon quitting. Don't output
3250 DBG("Communication interrupted on command socket");
3254 if (CMM_LOAD_SHARED(consumer_quit
)) {
3255 DBG("consumer_thread_receive_fds received quit from signal");
3256 err
= 0; /* All is OK */
3259 DBG("received command on sock");
3265 DBG("Consumer thread sessiond poll exiting");
3268 * Close metadata streams since the producer is the session daemon which
3271 * NOTE: for now, this only applies to the UST tracer.
3273 lttng_consumer_close_all_metadata();
3276 * when all fds have hung up, the polling thread
3279 CMM_STORE_SHARED(consumer_quit
, 1);
3282 * Notify the data poll thread to poll back again and test the
3283 * consumer_quit state that we just set so to quit gracefully.
3285 notify_thread_lttng_pipe(ctx
->consumer_data_pipe
);
3287 notify_channel_pipe(ctx
, NULL
, -1, CONSUMER_CHANNEL_QUIT
);
3289 notify_health_quit_pipe(health_quit_pipe
);
3291 /* Cleaning up possibly open sockets. */
3295 PERROR("close sock sessiond poll");
3298 if (client_socket
>= 0) {
3299 ret
= close(client_socket
);
3301 PERROR("close client_socket sessiond poll");
3308 ERR("Health error occurred in %s", __func__
);
3310 health_unregister(health_consumerd
);
3312 rcu_unregister_thread();
3317 int rotate_notify_sessiond(struct lttng_consumer_local_data
*ctx
,
3323 ret
= write(ctx
->channel_rotate_pipe
, &key
, sizeof(key
));
3324 } while (ret
== -1 && errno
== EINTR
);
3326 PERROR("write to the channel rotate pipe");
3328 DBG("Sent channel rotation notification for channel key %"
3336 * Perform operations that need to be done after a stream has
3337 * rotated and released the stream lock.
3339 * Multiple rotations cannot occur simultaneously, so we know the state of the
3340 * "rotated" stream flag cannot change.
3342 * This MUST be called WITHOUT the stream lock held.
3345 int consumer_post_rotation(struct lttng_consumer_stream
*stream
,
3346 struct lttng_consumer_local_data
*ctx
)
3350 if (!stream
->rotated
) {
3354 pthread_mutex_lock(&stream
->chan
->lock
);
3355 switch (consumer_data
.type
) {
3356 case LTTNG_CONSUMER_KERNEL
:
3358 case LTTNG_CONSUMER32_UST
:
3359 case LTTNG_CONSUMER64_UST
:
3361 * The ust_metadata_pushed counter has been reset to 0, so now
3362 * we can wakeup the metadata thread so it dumps the metadata
3363 * cache to the new file.
3365 if (stream
->metadata_flag
) {
3366 consumer_metadata_wakeup_pipe(stream
->chan
);
3370 ERR("Unknown consumer_data type");
3374 if (--stream
->chan
->nr_stream_rotate_pending
== 0) {
3375 ret
= rotate_notify_sessiond(ctx
, stream
->chan
->key
);
3377 pthread_mutex_unlock(&stream
->chan
->lock
);
3378 stream
->rotated
= 0;
3384 ssize_t
lttng_consumer_read_subbuffer(struct lttng_consumer_stream
*stream
,
3385 struct lttng_consumer_local_data
*ctx
)
3390 pthread_mutex_lock(&stream
->lock
);
3391 if (stream
->metadata_flag
) {
3392 pthread_mutex_lock(&stream
->metadata_rdv_lock
);
3395 switch (consumer_data
.type
) {
3396 case LTTNG_CONSUMER_KERNEL
:
3397 ret
= lttng_kconsumer_read_subbuffer(stream
, ctx
);
3399 case LTTNG_CONSUMER32_UST
:
3400 case LTTNG_CONSUMER64_UST
:
3401 ret
= lttng_ustconsumer_read_subbuffer(stream
, ctx
);
3404 ERR("Unknown consumer_data type");
3410 if (stream
->metadata_flag
) {
3411 pthread_cond_broadcast(&stream
->metadata_rdv
);
3412 pthread_mutex_unlock(&stream
->metadata_rdv_lock
);
3414 pthread_mutex_unlock(&stream
->lock
);
3416 rotate_ret
= consumer_post_rotation(stream
, ctx
);
3417 if (rotate_ret
< 0) {
3418 ERR("Failed after a rotation");
3425 int lttng_consumer_on_recv_stream(struct lttng_consumer_stream
*stream
)
3427 switch (consumer_data
.type
) {
3428 case LTTNG_CONSUMER_KERNEL
:
3429 return lttng_kconsumer_on_recv_stream(stream
);
3430 case LTTNG_CONSUMER32_UST
:
3431 case LTTNG_CONSUMER64_UST
:
3432 return lttng_ustconsumer_on_recv_stream(stream
);
3434 ERR("Unknown consumer_data type");
3441 * Allocate and set consumer data hash tables.
3443 int lttng_consumer_init(void)
3445 consumer_data
.channel_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3446 if (!consumer_data
.channel_ht
) {
3450 consumer_data
.relayd_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3451 if (!consumer_data
.relayd_ht
) {
3455 consumer_data
.stream_list_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3456 if (!consumer_data
.stream_list_ht
) {
3460 consumer_data
.stream_per_chan_id_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3461 if (!consumer_data
.stream_per_chan_id_ht
) {
3465 data_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3470 metadata_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3482 * Process the ADD_RELAYD command receive by a consumer.
3484 * This will create a relayd socket pair and add it to the relayd hash table.
3485 * The caller MUST acquire a RCU read side lock before calling it.
3487 void consumer_add_relayd_socket(uint64_t net_seq_idx
, int sock_type
,
3488 struct lttng_consumer_local_data
*ctx
, int sock
,
3489 struct pollfd
*consumer_sockpoll
,
3490 struct lttcomm_relayd_sock
*relayd_sock
, uint64_t sessiond_id
,
3491 uint64_t relayd_session_id
)
3493 int fd
= -1, ret
= -1, relayd_created
= 0;
3494 enum lttcomm_return_code ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
3495 struct consumer_relayd_sock_pair
*relayd
= NULL
;
3498 assert(relayd_sock
);
3500 DBG("Consumer adding relayd socket (idx: %" PRIu64
")", net_seq_idx
);
3502 /* Get relayd reference if exists. */
3503 relayd
= consumer_find_relayd(net_seq_idx
);
3504 if (relayd
== NULL
) {
3505 assert(sock_type
== LTTNG_STREAM_CONTROL
);
3506 /* Not found. Allocate one. */
3507 relayd
= consumer_allocate_relayd_sock_pair(net_seq_idx
);
3508 if (relayd
== NULL
) {
3509 ret_code
= LTTCOMM_CONSUMERD_ENOMEM
;
3512 relayd
->sessiond_session_id
= sessiond_id
;
3517 * This code path MUST continue to the consumer send status message to
3518 * we can notify the session daemon and continue our work without
3519 * killing everything.
3523 * relayd key should never be found for control socket.
3525 assert(sock_type
!= LTTNG_STREAM_CONTROL
);
3528 /* First send a status message before receiving the fds. */
3529 ret
= consumer_send_status_msg(sock
, LTTCOMM_CONSUMERD_SUCCESS
);
3531 /* Somehow, the session daemon is not responding anymore. */
3532 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_FATAL
);
3533 goto error_nosignal
;
3536 /* Poll on consumer socket. */
3537 ret
= lttng_consumer_poll_socket(consumer_sockpoll
);
3539 /* Needing to exit in the middle of a command: error. */
3540 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_POLL_ERROR
);
3541 goto error_nosignal
;
3544 /* Get relayd socket from session daemon */
3545 ret
= lttcomm_recv_fds_unix_sock(sock
, &fd
, 1);
3546 if (ret
!= sizeof(fd
)) {
3547 fd
= -1; /* Just in case it gets set with an invalid value. */
3550 * Failing to receive FDs might indicate a major problem such as
3551 * reaching a fd limit during the receive where the kernel returns a
3552 * MSG_CTRUNC and fails to cleanup the fd in the queue. Any case, we
3553 * don't take any chances and stop everything.
3555 * XXX: Feature request #558 will fix that and avoid this possible
3556 * issue when reaching the fd limit.
3558 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_ERROR_RECV_FD
);
3559 ret_code
= LTTCOMM_CONSUMERD_ERROR_RECV_FD
;
3563 /* Copy socket information and received FD */
3564 switch (sock_type
) {
3565 case LTTNG_STREAM_CONTROL
:
3566 /* Copy received lttcomm socket */
3567 lttcomm_copy_sock(&relayd
->control_sock
.sock
, &relayd_sock
->sock
);
3568 ret
= lttcomm_create_sock(&relayd
->control_sock
.sock
);
3569 /* Handle create_sock error. */
3571 ret_code
= LTTCOMM_CONSUMERD_ENOMEM
;
3575 * Close the socket created internally by
3576 * lttcomm_create_sock, so we can replace it by the one
3577 * received from sessiond.
3579 if (close(relayd
->control_sock
.sock
.fd
)) {
3583 /* Assign new file descriptor */
3584 relayd
->control_sock
.sock
.fd
= fd
;
3585 fd
= -1; /* For error path */
3586 /* Assign version values. */
3587 relayd
->control_sock
.major
= relayd_sock
->major
;
3588 relayd
->control_sock
.minor
= relayd_sock
->minor
;
3590 relayd
->relayd_session_id
= relayd_session_id
;
3593 case LTTNG_STREAM_DATA
:
3594 /* Copy received lttcomm socket */
3595 lttcomm_copy_sock(&relayd
->data_sock
.sock
, &relayd_sock
->sock
);
3596 ret
= lttcomm_create_sock(&relayd
->data_sock
.sock
);
3597 /* Handle create_sock error. */
3599 ret_code
= LTTCOMM_CONSUMERD_ENOMEM
;
3603 * Close the socket created internally by
3604 * lttcomm_create_sock, so we can replace it by the one
3605 * received from sessiond.
3607 if (close(relayd
->data_sock
.sock
.fd
)) {
3611 /* Assign new file descriptor */
3612 relayd
->data_sock
.sock
.fd
= fd
;
3613 fd
= -1; /* for eventual error paths */
3614 /* Assign version values. */
3615 relayd
->data_sock
.major
= relayd_sock
->major
;
3616 relayd
->data_sock
.minor
= relayd_sock
->minor
;
3619 ERR("Unknown relayd socket type (%d)", sock_type
);
3620 ret_code
= LTTCOMM_CONSUMERD_FATAL
;
3624 DBG("Consumer %s socket created successfully with net idx %" PRIu64
" (fd: %d)",
3625 sock_type
== LTTNG_STREAM_CONTROL
? "control" : "data",
3626 relayd
->net_seq_idx
, fd
);
3628 /* We successfully added the socket. Send status back. */
3629 ret
= consumer_send_status_msg(sock
, ret_code
);
3631 /* Somehow, the session daemon is not responding anymore. */
3632 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_FATAL
);
3633 goto error_nosignal
;
3637 * Add relayd socket pair to consumer data hashtable. If object already
3638 * exists or on error, the function gracefully returns.
3646 if (consumer_send_status_msg(sock
, ret_code
) < 0) {
3647 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_FATAL
);
3651 /* Close received socket if valid. */
3654 PERROR("close received socket");
3658 if (relayd_created
) {
3664 * Try to lock the stream mutex.
3666 * On success, 1 is returned else 0 indicating that the mutex is NOT lock.
3668 static int stream_try_lock(struct lttng_consumer_stream
*stream
)
3675 * Try to lock the stream mutex. On failure, we know that the stream is
3676 * being used else where hence there is data still being extracted.
3678 ret
= pthread_mutex_trylock(&stream
->lock
);
3680 /* For both EBUSY and EINVAL error, the mutex is NOT locked. */
3692 * Search for a relayd associated to the session id and return the reference.
3694 * A rcu read side lock MUST be acquire before calling this function and locked
3695 * until the relayd object is no longer necessary.
3697 static struct consumer_relayd_sock_pair
*find_relayd_by_session_id(uint64_t id
)
3699 struct lttng_ht_iter iter
;
3700 struct consumer_relayd_sock_pair
*relayd
= NULL
;
3702 /* Iterate over all relayd since they are indexed by net_seq_idx. */
3703 cds_lfht_for_each_entry(consumer_data
.relayd_ht
->ht
, &iter
.iter
, relayd
,
3706 * Check by sessiond id which is unique here where the relayd session
3707 * id might not be when having multiple relayd.
3709 if (relayd
->sessiond_session_id
== id
) {
3710 /* Found the relayd. There can be only one per id. */
3722 * Check if for a given session id there is still data needed to be extract
3725 * Return 1 if data is pending or else 0 meaning ready to be read.
3727 int consumer_data_pending(uint64_t id
)
3730 struct lttng_ht_iter iter
;
3731 struct lttng_ht
*ht
;
3732 struct lttng_consumer_stream
*stream
;
3733 struct consumer_relayd_sock_pair
*relayd
= NULL
;
3734 int (*data_pending
)(struct lttng_consumer_stream
*);
3736 DBG("Consumer data pending command on session id %" PRIu64
, id
);
3739 pthread_mutex_lock(&consumer_data
.lock
);
3741 switch (consumer_data
.type
) {
3742 case LTTNG_CONSUMER_KERNEL
:
3743 data_pending
= lttng_kconsumer_data_pending
;
3745 case LTTNG_CONSUMER32_UST
:
3746 case LTTNG_CONSUMER64_UST
:
3747 data_pending
= lttng_ustconsumer_data_pending
;
3750 ERR("Unknown consumer data type");
3754 /* Ease our life a bit */
3755 ht
= consumer_data
.stream_list_ht
;
3757 relayd
= find_relayd_by_session_id(id
);
3759 /* Send init command for data pending. */
3760 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
3761 ret
= relayd_begin_data_pending(&relayd
->control_sock
,
3762 relayd
->relayd_session_id
);
3763 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
3765 /* Communication error thus the relayd so no data pending. */
3766 goto data_not_pending
;
3770 cds_lfht_for_each_entry_duplicate(ht
->ht
,
3771 ht
->hash_fct(&id
, lttng_ht_seed
),
3773 &iter
.iter
, stream
, node_session_id
.node
) {
3774 /* If this call fails, the stream is being used hence data pending. */
3775 ret
= stream_try_lock(stream
);
3781 * A removed node from the hash table indicates that the stream has
3782 * been deleted thus having a guarantee that the buffers are closed
3783 * on the consumer side. However, data can still be transmitted
3784 * over the network so don't skip the relayd check.
3786 ret
= cds_lfht_is_node_deleted(&stream
->node
.node
);
3788 /* Check the stream if there is data in the buffers. */
3789 ret
= data_pending(stream
);
3791 pthread_mutex_unlock(&stream
->lock
);
3798 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
3799 if (stream
->metadata_flag
) {
3800 ret
= relayd_quiescent_control(&relayd
->control_sock
,
3801 stream
->relayd_stream_id
);
3803 ret
= relayd_data_pending(&relayd
->control_sock
,
3804 stream
->relayd_stream_id
,
3805 stream
->next_net_seq_num
- 1);
3807 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
3809 pthread_mutex_unlock(&stream
->lock
);
3813 pthread_mutex_unlock(&stream
->lock
);
3817 unsigned int is_data_inflight
= 0;
3819 /* Send init command for data pending. */
3820 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
3821 ret
= relayd_end_data_pending(&relayd
->control_sock
,
3822 relayd
->relayd_session_id
, &is_data_inflight
);
3823 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
3825 goto data_not_pending
;
3827 if (is_data_inflight
) {
3833 * Finding _no_ node in the hash table and no inflight data means that the
3834 * stream(s) have been removed thus data is guaranteed to be available for
3835 * analysis from the trace files.
3839 /* Data is available to be read by a viewer. */
3840 pthread_mutex_unlock(&consumer_data
.lock
);
3845 /* Data is still being extracted from buffers. */
3846 pthread_mutex_unlock(&consumer_data
.lock
);
3852 * Send a ret code status message to the sessiond daemon.
3854 * Return the sendmsg() return value.
3856 int consumer_send_status_msg(int sock
, int ret_code
)
3858 struct lttcomm_consumer_status_msg msg
;
3860 memset(&msg
, 0, sizeof(msg
));
3861 msg
.ret_code
= ret_code
;
3863 return lttcomm_send_unix_sock(sock
, &msg
, sizeof(msg
));
3867 * Send a channel status message to the sessiond daemon.
3869 * Return the sendmsg() return value.
3871 int consumer_send_status_channel(int sock
,
3872 struct lttng_consumer_channel
*channel
)
3874 struct lttcomm_consumer_status_channel msg
;
3878 memset(&msg
, 0, sizeof(msg
));
3880 msg
.ret_code
= LTTCOMM_CONSUMERD_CHANNEL_FAIL
;
3882 msg
.ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
3883 msg
.key
= channel
->key
;
3884 msg
.stream_count
= channel
->streams
.count
;
3887 return lttcomm_send_unix_sock(sock
, &msg
, sizeof(msg
));
3890 unsigned long consumer_get_consume_start_pos(unsigned long consumed_pos
,
3891 unsigned long produced_pos
, uint64_t nb_packets_per_stream
,
3892 uint64_t max_sb_size
)
3894 unsigned long start_pos
;
3896 if (!nb_packets_per_stream
) {
3897 return consumed_pos
; /* Grab everything */
3899 start_pos
= produced_pos
- offset_align_floor(produced_pos
, max_sb_size
);
3900 start_pos
-= max_sb_size
* nb_packets_per_stream
;
3901 if ((long) (start_pos
- consumed_pos
) < 0) {
3902 return consumed_pos
; /* Grab everything */
3908 int consumer_flush_buffer(struct lttng_consumer_stream
*stream
, int producer_active
)
3912 switch (consumer_data
.type
) {
3913 case LTTNG_CONSUMER_KERNEL
:
3914 ret
= kernctl_buffer_flush(stream
->wait_fd
);
3916 ERR("Failed to flush kernel stream");
3920 case LTTNG_CONSUMER32_UST
:
3921 case LTTNG_CONSUMER64_UST
:
3922 lttng_ustctl_flush_buffer(stream
, producer_active
);
3925 ERR("Unknown consumer_data type");
3934 * Sample the rotate position for all the streams of a channel.
3936 * Returns 0 on success, < 0 on error
3938 int lttng_consumer_rotate_channel(uint64_t key
, char *path
,
3939 uint64_t relayd_id
, uint32_t metadata
, uint64_t new_chunk_id
,
3940 struct lttng_consumer_local_data
*ctx
)
3943 struct lttng_consumer_channel
*channel
;
3944 struct lttng_consumer_stream
*stream
;
3945 struct lttng_ht_iter iter
;
3946 struct lttng_ht
*ht
= consumer_data
.stream_per_chan_id_ht
;
3948 DBG("Kernel consumer sample rotate position for channel %" PRIu64
, key
);
3952 channel
= consumer_find_channel(key
);
3954 ERR("No channel found for key %" PRIu64
, key
);
3958 pthread_mutex_lock(&channel
->lock
);
3959 channel
->current_chunk_id
= new_chunk_id
;
3960 snprintf(channel
->pathname
, PATH_MAX
, "%s", path
);
3961 ret
= utils_mkdir_recursive(channel
->pathname
, S_IRWXU
| S_IRWXG
,
3962 channel
->uid
, channel
->gid
);
3964 ERR("Trace directory creation error");
3966 pthread_mutex_unlock(&channel
->lock
);
3969 pthread_mutex_unlock(&channel
->lock
);
3971 cds_lfht_for_each_entry_duplicate(ht
->ht
,
3972 ht
->hash_fct(&channel
->key
, lttng_ht_seed
),
3973 ht
->match_fct
, &channel
->key
, &iter
.iter
,
3974 stream
, node_channel_id
.node
) {
3975 uint64_t consumed_pos
;
3977 health_code_update();
3980 * Lock stream because we are about to change its state.
3982 pthread_mutex_lock(&stream
->lock
);
3984 memcpy(stream
->channel_ro_pathname
, channel
->pathname
, PATH_MAX
);
3985 ret
= lttng_consumer_sample_snapshot_positions(stream
);
3987 ERR("Taking kernel snapshot positions");
3991 ret
= lttng_consumer_get_produced_snapshot(stream
,
3992 &stream
->rotate_position
);
3994 ERR("Produced kernel snapshot position");
3997 fprintf(stderr
, "Stream %lu should rotate after %lu to %s\n",
3998 stream
->key
, stream
->rotate_position
,
4000 lttng_consumer_get_consumed_snapshot(stream
,
4002 fprintf(stderr
, "consumed %lu\n", consumed_pos
);
4003 if (consumed_pos
== stream
->rotate_position
) {
4004 stream
->rotate_ready
= 1;
4005 fprintf(stderr
, "Stream %lu ready to rotate to %s\n",
4006 stream
->key
, channel
->pathname
);
4008 channel
->nr_stream_rotate_pending
++;
4010 ret
= consumer_flush_buffer(stream
, 1);
4012 ERR("Failed to flush stream");
4016 pthread_mutex_unlock(&stream
->lock
);
4023 pthread_mutex_unlock(&stream
->lock
);
4030 * Check if a stream is ready to be rotated after extracting it.
4032 * When we are called between get_next_subbuf and put_next_subbuf, the len
4033 * parameter is the subbuf size of the current subbuffer being extracted. This
4034 * len is with padding, so it is normal to see that the current position is
4035 * farther than the expected rotate position.
4037 * Return 1 if it is ready for rotation, 0 if it is not, a negative value on
4040 int lttng_consumer_stream_is_rotate_ready(struct lttng_consumer_stream
*stream
,
4044 unsigned long consumed_pos
;
4046 if (!stream
->rotate_position
&& !stream
->rotate_ready
) {
4051 if (stream
->rotate_ready
) {
4052 fprintf(stderr
, "Rotate position reached for stream %lu\n",
4059 * If we don't have the rotate_ready flag, check the consumed position
4060 * to determine if we need to rotate.
4062 ret
= lttng_consumer_sample_snapshot_positions(stream
);
4064 ERR("Taking kernel snapshot positions");
4068 ret
= lttng_consumer_get_consumed_snapshot(stream
, &consumed_pos
);
4070 ERR("Produced kernel snapshot position");
4074 fprintf(stderr
, "packet %lu, pos %lu\n", stream
->key
, consumed_pos
);
4075 /* Rotate position not reached yet. */
4076 if ((consumed_pos
+ len
) < stream
->rotate_position
) {
4080 fprintf(stderr
, "Rotate position %lu (expected %lu) reached for stream %lu\n",
4081 consumed_pos
+ len
, stream
->rotate_position
,
4090 * Reset the state for a stream after a rotation occurred.
4092 void lttng_consumer_reset_stream_rotate_state(struct lttng_consumer_stream
*stream
)
4094 stream
->rotate_position
= 0;
4095 stream
->rotate_ready
= 0;
4096 stream
->rotated
= 1;
4100 * Perform the rotation a local stream file.
4102 int rotate_local_stream(struct lttng_consumer_local_data
*ctx
,
4103 struct lttng_consumer_stream
*stream
)
4107 ret
= close(stream
->out_fd
);
4109 PERROR("Closing tracefile");
4113 fprintf(stderr
, "Rotating stream %lu to %s/%s\n", stream
->key
,
4114 stream
->channel_ro_pathname
, stream
->name
);
4115 ret
= utils_create_stream_file(stream
->channel_ro_pathname
, stream
->name
,
4116 stream
->channel_ro_tracefile_size
, stream
->tracefile_count_current
,
4117 stream
->uid
, stream
->gid
, NULL
);
4121 stream
->out_fd
= ret
;
4122 stream
->tracefile_size_current
= 0;
4124 if (!stream
->metadata_flag
) {
4125 struct lttng_index_file
*index_file
;
4127 lttng_index_file_put(stream
->index_file
);
4129 index_file
= lttng_index_file_create(stream
->channel_ro_pathname
,
4130 stream
->name
, stream
->uid
, stream
->gid
,
4131 stream
->channel_ro_tracefile_size
,
4132 stream
->tracefile_count_current
,
4133 CTF_INDEX_MAJOR
, CTF_INDEX_MINOR
);
4137 stream
->index_file
= index_file
;
4138 stream
->out_fd_offset
= 0;
4152 * Perform the rotation a stream file on the relay.
4154 int rotate_relay_stream(struct lttng_consumer_local_data
*ctx
,
4155 struct lttng_consumer_stream
*stream
)
4158 struct consumer_relayd_sock_pair
*relayd
;
4160 relayd
= consumer_find_relayd(stream
->net_seq_idx
);
4162 ERR("Failed to find relayd");
4167 /* FIXME: chan_ro ? */
4168 ret
= relayd_rotate_stream(&relayd
->control_sock
,
4169 stream
->relayd_stream_id
, stream
->channel_ro_pathname
,
4170 stream
->chan
->current_chunk_id
,
4171 stream
->last_sequence_number
);
4178 * Performs the stream rotation for the rotate session feature if needed.
4179 * It must be called with the stream lock held.
4181 * Return 0 on success, a negative number of error.
4183 int lttng_consumer_rotate_stream(struct lttng_consumer_local_data
*ctx
,
4184 struct lttng_consumer_stream
*stream
)
4188 if (stream
->net_seq_idx
!= (uint64_t) -1ULL) {
4189 ret
= rotate_relay_stream(ctx
, stream
);
4191 ret
= rotate_local_stream(ctx
, stream
);
4197 if (stream
->metadata_flag
) {
4198 switch (consumer_data
.type
) {
4199 case LTTNG_CONSUMER_KERNEL
:
4201 * Reset the position of what has been read from the metadata
4202 * cache to 0 so we can dump it again.
4204 ret
= kernctl_metadata_cache_dump(stream
->wait_fd
);
4206 ERR("Failed to dump the kernel metadata cache after rotation");
4210 case LTTNG_CONSUMER32_UST
:
4211 case LTTNG_CONSUMER64_UST
:
4213 * Reset the position pushed from the metadata cache so it
4214 * will write from the beginning on the next push.
4216 stream
->ust_metadata_pushed
= 0;
4219 ERR("Unknown consumer_data type");
4223 lttng_consumer_reset_stream_rotate_state(stream
);
4232 * Rotate all the ready streams.
4234 * This is especially important for low throughput streams that have already
4235 * been consumed, we cannot wait for their next packet to perform the
4238 * Returns 0 on success, < 0 on error
4240 int lttng_consumer_rotate_ready_streams(uint64_t key
,
4241 struct lttng_consumer_local_data
*ctx
)
4244 struct lttng_consumer_channel
*channel
;
4245 struct lttng_consumer_stream
*stream
;
4246 struct lttng_ht_iter iter
;
4247 struct lttng_ht
*ht
= consumer_data
.stream_per_chan_id_ht
;
4251 channel
= consumer_find_channel(key
);
4253 ERR("No channel found for key %" PRIu64
, key
);
4257 cds_lfht_for_each_entry_duplicate(ht
->ht
,
4258 ht
->hash_fct(&channel
->key
, lttng_ht_seed
),
4259 ht
->match_fct
, &channel
->key
, &iter
.iter
,
4260 stream
, node_channel_id
.node
) {
4261 health_code_update();
4264 * Lock stream because we are about to change its state.
4266 pthread_mutex_lock(&stream
->lock
);
4267 if (stream
->rotate_ready
== 0) {
4268 pthread_mutex_unlock(&stream
->lock
);
4271 ret
= lttng_consumer_rotate_stream(ctx
, stream
);
4273 pthread_mutex_unlock(&stream
->lock
);
4274 ERR("Stream rotation error");
4278 pthread_mutex_unlock(&stream
->lock
);
4279 ret
= consumer_post_rotation(stream
, ctx
);
4281 ERR("Failed after a rotation");
4294 int rotate_rename_local(char *current_path
, char *new_path
,
4295 uid_t uid
, gid_t gid
)
4299 ret
= utils_mkdir_recursive(new_path
, S_IRWXU
| S_IRWXG
,
4302 ERR("Create directory on rotate");
4306 ret
= rename(current_path
, new_path
);
4308 * If a domain has not yet created its channel, the domain-specific
4309 * folder might not exist, but this is not an error.
4311 if (ret
< 0 && errno
!= ENOENT
) {
4312 PERROR("Rename completed rotation chunk");
4323 int rotate_rename_relay(char *current_path
, char *new_path
, uint64_t relayd_id
)
4326 struct consumer_relayd_sock_pair
*relayd
;
4328 relayd
= consumer_find_relayd(relayd_id
);
4330 ERR("Failed to find relayd");
4335 ret
= relayd_rotate_rename(&relayd
->control_sock
, current_path
, new_path
);
4342 int lttng_consumer_rotate_rename(char *current_path
, char *new_path
,
4343 uid_t uid
, gid_t gid
, uint64_t relayd_id
)
4345 if (relayd_id
!= (uint64_t) -1ULL) {
4346 return rotate_rename_relay(current_path
, new_path
, relayd_id
);
4348 return rotate_rename_local(current_path
, new_path
, uid
, gid
);
4352 int lttng_consumer_rotate_pending_relay(uint64_t session_id
,
4353 uint64_t relayd_id
, uint64_t chunk_id
)
4356 struct consumer_relayd_sock_pair
*relayd
;
4358 relayd
= consumer_find_relayd(relayd_id
);
4360 ERR("Failed to find relayd");
4365 ret
= relayd_rotate_pending(&relayd
->control_sock
, chunk_id
);