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_update_channel_attributes(
545 struct lttng_consumer_stream
*stream
,
546 struct lttng_consumer_channel
*channel
)
548 stream
->channel_read_only_attributes
.tracefile_size
=
549 channel
->tracefile_size
;
550 memcpy(stream
->channel_read_only_attributes
.path
, channel
->pathname
,
551 sizeof(stream
->channel_read_only_attributes
.path
));
554 struct lttng_consumer_stream
*consumer_allocate_stream(uint64_t channel_key
,
556 enum lttng_consumer_stream_state state
,
557 const char *channel_name
,
564 enum consumer_channel_type type
,
565 unsigned int monitor
,
566 uint64_t trace_archive_id
)
569 struct lttng_consumer_stream
*stream
;
571 stream
= zmalloc(sizeof(*stream
));
572 if (stream
== NULL
) {
573 PERROR("malloc struct lttng_consumer_stream");
580 stream
->key
= stream_key
;
582 stream
->out_fd_offset
= 0;
583 stream
->output_written
= 0;
584 stream
->state
= state
;
587 stream
->net_seq_idx
= relayd_id
;
588 stream
->session_id
= session_id
;
589 stream
->monitor
= monitor
;
590 stream
->endpoint_status
= CONSUMER_ENDPOINT_ACTIVE
;
591 stream
->index_file
= NULL
;
592 stream
->last_sequence_number
= -1ULL;
593 stream
->trace_archive_id
= trace_archive_id
;
594 pthread_mutex_init(&stream
->lock
, NULL
);
595 pthread_mutex_init(&stream
->metadata_timer_lock
, NULL
);
597 /* If channel is the metadata, flag this stream as metadata. */
598 if (type
== CONSUMER_CHANNEL_TYPE_METADATA
) {
599 stream
->metadata_flag
= 1;
600 /* Metadata is flat out. */
601 strncpy(stream
->name
, DEFAULT_METADATA_NAME
, sizeof(stream
->name
));
602 /* Live rendez-vous point. */
603 pthread_cond_init(&stream
->metadata_rdv
, NULL
);
604 pthread_mutex_init(&stream
->metadata_rdv_lock
, NULL
);
606 /* Format stream name to <channel_name>_<cpu_number> */
607 ret
= snprintf(stream
->name
, sizeof(stream
->name
), "%s_%d",
610 PERROR("snprintf stream name");
615 /* Key is always the wait_fd for streams. */
616 lttng_ht_node_init_u64(&stream
->node
, stream
->key
);
618 /* Init node per channel id key */
619 lttng_ht_node_init_u64(&stream
->node_channel_id
, channel_key
);
621 /* Init session id node with the stream session id */
622 lttng_ht_node_init_u64(&stream
->node_session_id
, stream
->session_id
);
624 DBG3("Allocated stream %s (key %" PRIu64
", chan_key %" PRIu64
625 " relayd_id %" PRIu64
", session_id %" PRIu64
,
626 stream
->name
, stream
->key
, channel_key
,
627 stream
->net_seq_idx
, stream
->session_id
);
643 * Add a stream to the global list protected by a mutex.
645 void consumer_add_data_stream(struct lttng_consumer_stream
*stream
)
647 struct lttng_ht
*ht
= data_ht
;
652 DBG3("Adding consumer stream %" PRIu64
, stream
->key
);
654 pthread_mutex_lock(&consumer_data
.lock
);
655 pthread_mutex_lock(&stream
->chan
->lock
);
656 pthread_mutex_lock(&stream
->chan
->timer_lock
);
657 pthread_mutex_lock(&stream
->lock
);
660 /* Steal stream identifier to avoid having streams with the same key */
661 steal_stream_key(stream
->key
, ht
);
663 lttng_ht_add_unique_u64(ht
, &stream
->node
);
665 lttng_ht_add_u64(consumer_data
.stream_per_chan_id_ht
,
666 &stream
->node_channel_id
);
669 * Add stream to the stream_list_ht of the consumer data. No need to steal
670 * the key since the HT does not use it and we allow to add redundant keys
673 lttng_ht_add_u64(consumer_data
.stream_list_ht
, &stream
->node_session_id
);
676 * When nb_init_stream_left reaches 0, we don't need to trigger any action
677 * in terms of destroying the associated channel, because the action that
678 * causes the count to become 0 also causes a stream to be added. The
679 * channel deletion will thus be triggered by the following removal of this
682 if (uatomic_read(&stream
->chan
->nb_init_stream_left
) > 0) {
683 /* Increment refcount before decrementing nb_init_stream_left */
685 uatomic_dec(&stream
->chan
->nb_init_stream_left
);
688 /* Update consumer data once the node is inserted. */
689 consumer_data
.stream_count
++;
690 consumer_data
.need_update
= 1;
693 pthread_mutex_unlock(&stream
->lock
);
694 pthread_mutex_unlock(&stream
->chan
->timer_lock
);
695 pthread_mutex_unlock(&stream
->chan
->lock
);
696 pthread_mutex_unlock(&consumer_data
.lock
);
699 void consumer_del_data_stream(struct lttng_consumer_stream
*stream
)
701 consumer_del_stream(stream
, data_ht
);
705 * Add relayd socket to global consumer data hashtable. RCU read side lock MUST
706 * be acquired before calling this.
708 static int add_relayd(struct consumer_relayd_sock_pair
*relayd
)
711 struct lttng_ht_node_u64
*node
;
712 struct lttng_ht_iter iter
;
716 lttng_ht_lookup(consumer_data
.relayd_ht
,
717 &relayd
->net_seq_idx
, &iter
);
718 node
= lttng_ht_iter_get_node_u64(&iter
);
722 lttng_ht_add_unique_u64(consumer_data
.relayd_ht
, &relayd
->node
);
729 * Allocate and return a consumer relayd socket.
731 static struct consumer_relayd_sock_pair
*consumer_allocate_relayd_sock_pair(
732 uint64_t net_seq_idx
)
734 struct consumer_relayd_sock_pair
*obj
= NULL
;
736 /* net sequence index of -1 is a failure */
737 if (net_seq_idx
== (uint64_t) -1ULL) {
741 obj
= zmalloc(sizeof(struct consumer_relayd_sock_pair
));
743 PERROR("zmalloc relayd sock");
747 obj
->net_seq_idx
= net_seq_idx
;
749 obj
->destroy_flag
= 0;
750 obj
->control_sock
.sock
.fd
= -1;
751 obj
->data_sock
.sock
.fd
= -1;
752 lttng_ht_node_init_u64(&obj
->node
, obj
->net_seq_idx
);
753 pthread_mutex_init(&obj
->ctrl_sock_mutex
, NULL
);
760 * Find a relayd socket pair in the global consumer data.
762 * Return the object if found else NULL.
763 * RCU read-side lock must be held across this call and while using the
766 struct consumer_relayd_sock_pair
*consumer_find_relayd(uint64_t key
)
768 struct lttng_ht_iter iter
;
769 struct lttng_ht_node_u64
*node
;
770 struct consumer_relayd_sock_pair
*relayd
= NULL
;
772 /* Negative keys are lookup failures */
773 if (key
== (uint64_t) -1ULL) {
777 lttng_ht_lookup(consumer_data
.relayd_ht
, &key
,
779 node
= lttng_ht_iter_get_node_u64(&iter
);
781 relayd
= caa_container_of(node
, struct consumer_relayd_sock_pair
, node
);
789 * Find a relayd and send the stream
791 * Returns 0 on success, < 0 on error
793 int consumer_send_relayd_stream(struct lttng_consumer_stream
*stream
,
797 struct consumer_relayd_sock_pair
*relayd
;
800 assert(stream
->net_seq_idx
!= -1ULL);
803 /* The stream is not metadata. Get relayd reference if exists. */
805 relayd
= consumer_find_relayd(stream
->net_seq_idx
);
806 if (relayd
!= NULL
) {
807 /* Add stream on the relayd */
808 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
809 ret
= relayd_add_stream(&relayd
->control_sock
, stream
->name
,
810 path
, &stream
->relayd_stream_id
,
811 stream
->chan
->tracefile_size
, stream
->chan
->tracefile_count
);
812 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
817 uatomic_inc(&relayd
->refcount
);
818 stream
->sent_to_relayd
= 1;
820 ERR("Stream %" PRIu64
" relayd ID %" PRIu64
" unknown. Can't send it.",
821 stream
->key
, stream
->net_seq_idx
);
826 DBG("Stream %s with key %" PRIu64
" sent to relayd id %" PRIu64
,
827 stream
->name
, stream
->key
, stream
->net_seq_idx
);
835 * Find a relayd and send the streams sent message
837 * Returns 0 on success, < 0 on error
839 int consumer_send_relayd_streams_sent(uint64_t net_seq_idx
)
842 struct consumer_relayd_sock_pair
*relayd
;
844 assert(net_seq_idx
!= -1ULL);
846 /* The stream is not metadata. Get relayd reference if exists. */
848 relayd
= consumer_find_relayd(net_seq_idx
);
849 if (relayd
!= NULL
) {
850 /* Add stream on the relayd */
851 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
852 ret
= relayd_streams_sent(&relayd
->control_sock
);
853 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
858 ERR("Relayd ID %" PRIu64
" unknown. Can't send streams_sent.",
865 DBG("All streams sent relayd id %" PRIu64
, net_seq_idx
);
873 * Find a relayd and close the stream
875 void close_relayd_stream(struct lttng_consumer_stream
*stream
)
877 struct consumer_relayd_sock_pair
*relayd
;
879 /* The stream is not metadata. Get relayd reference if exists. */
881 relayd
= consumer_find_relayd(stream
->net_seq_idx
);
883 consumer_stream_relayd_close(stream
, relayd
);
889 * Handle stream for relayd transmission if the stream applies for network
890 * streaming where the net sequence index is set.
892 * Return destination file descriptor or negative value on error.
894 static int write_relayd_stream_header(struct lttng_consumer_stream
*stream
,
895 size_t data_size
, unsigned long padding
,
896 struct consumer_relayd_sock_pair
*relayd
)
899 struct lttcomm_relayd_data_hdr data_hdr
;
905 /* Reset data header */
906 memset(&data_hdr
, 0, sizeof(data_hdr
));
908 if (stream
->metadata_flag
) {
909 /* Caller MUST acquire the relayd control socket lock */
910 ret
= relayd_send_metadata(&relayd
->control_sock
, data_size
);
915 /* Metadata are always sent on the control socket. */
916 outfd
= relayd
->control_sock
.sock
.fd
;
918 /* Set header with stream information */
919 data_hdr
.stream_id
= htobe64(stream
->relayd_stream_id
);
920 data_hdr
.data_size
= htobe32(data_size
);
921 data_hdr
.padding_size
= htobe32(padding
);
923 * Note that net_seq_num below is assigned with the *current* value of
924 * next_net_seq_num and only after that the next_net_seq_num will be
925 * increment. This is why when issuing a command on the relayd using
926 * this next value, 1 should always be substracted in order to compare
927 * the last seen sequence number on the relayd side to the last sent.
929 data_hdr
.net_seq_num
= htobe64(stream
->next_net_seq_num
);
930 /* Other fields are zeroed previously */
932 ret
= relayd_send_data_hdr(&relayd
->data_sock
, &data_hdr
,
938 ++stream
->next_net_seq_num
;
940 /* Set to go on data socket */
941 outfd
= relayd
->data_sock
.sock
.fd
;
949 * Allocate and return a new lttng_consumer_channel object using the given key
950 * to initialize the hash table node.
952 * On error, return NULL.
954 struct lttng_consumer_channel
*consumer_allocate_channel(uint64_t key
,
956 const char *pathname
,
961 enum lttng_event_output output
,
962 uint64_t tracefile_size
,
963 uint64_t tracefile_count
,
964 uint64_t session_id_per_pid
,
965 unsigned int monitor
,
966 unsigned int live_timer_interval
,
967 const char *root_shm_path
,
968 const char *shm_path
)
970 struct lttng_consumer_channel
*channel
;
972 channel
= zmalloc(sizeof(*channel
));
973 if (channel
== NULL
) {
974 PERROR("malloc struct lttng_consumer_channel");
979 channel
->refcount
= 0;
980 channel
->session_id
= session_id
;
981 channel
->session_id_per_pid
= session_id_per_pid
;
984 channel
->relayd_id
= relayd_id
;
985 channel
->tracefile_size
= tracefile_size
;
986 channel
->tracefile_count
= tracefile_count
;
987 channel
->monitor
= monitor
;
988 channel
->live_timer_interval
= live_timer_interval
;
989 pthread_mutex_init(&channel
->lock
, NULL
);
990 pthread_mutex_init(&channel
->timer_lock
, NULL
);
993 case LTTNG_EVENT_SPLICE
:
994 channel
->output
= CONSUMER_CHANNEL_SPLICE
;
996 case LTTNG_EVENT_MMAP
:
997 channel
->output
= CONSUMER_CHANNEL_MMAP
;
1007 * In monitor mode, the streams associated with the channel will be put in
1008 * a special list ONLY owned by this channel. So, the refcount is set to 1
1009 * here meaning that the channel itself has streams that are referenced.
1011 * On a channel deletion, once the channel is no longer visible, the
1012 * refcount is decremented and checked for a zero value to delete it. With
1013 * streams in no monitor mode, it will now be safe to destroy the channel.
1015 if (!channel
->monitor
) {
1016 channel
->refcount
= 1;
1019 strncpy(channel
->pathname
, pathname
, sizeof(channel
->pathname
));
1020 channel
->pathname
[sizeof(channel
->pathname
) - 1] = '\0';
1022 strncpy(channel
->name
, name
, sizeof(channel
->name
));
1023 channel
->name
[sizeof(channel
->name
) - 1] = '\0';
1025 if (root_shm_path
) {
1026 strncpy(channel
->root_shm_path
, root_shm_path
, sizeof(channel
->root_shm_path
));
1027 channel
->root_shm_path
[sizeof(channel
->root_shm_path
) - 1] = '\0';
1030 strncpy(channel
->shm_path
, shm_path
, sizeof(channel
->shm_path
));
1031 channel
->shm_path
[sizeof(channel
->shm_path
) - 1] = '\0';
1034 lttng_ht_node_init_u64(&channel
->node
, channel
->key
);
1036 channel
->wait_fd
= -1;
1038 CDS_INIT_LIST_HEAD(&channel
->streams
.head
);
1040 DBG("Allocated channel (key %" PRIu64
")", channel
->key
);
1047 * Add a channel to the global list protected by a mutex.
1049 * Always return 0 indicating success.
1051 int consumer_add_channel(struct lttng_consumer_channel
*channel
,
1052 struct lttng_consumer_local_data
*ctx
)
1054 pthread_mutex_lock(&consumer_data
.lock
);
1055 pthread_mutex_lock(&channel
->lock
);
1056 pthread_mutex_lock(&channel
->timer_lock
);
1059 * This gives us a guarantee that the channel we are about to add to the
1060 * channel hash table will be unique. See this function comment on the why
1061 * we need to steel the channel key at this stage.
1063 steal_channel_key(channel
->key
);
1066 lttng_ht_add_unique_u64(consumer_data
.channel_ht
, &channel
->node
);
1069 pthread_mutex_unlock(&channel
->timer_lock
);
1070 pthread_mutex_unlock(&channel
->lock
);
1071 pthread_mutex_unlock(&consumer_data
.lock
);
1073 if (channel
->wait_fd
!= -1 && channel
->type
== CONSUMER_CHANNEL_TYPE_DATA
) {
1074 notify_channel_pipe(ctx
, channel
, -1, CONSUMER_CHANNEL_ADD
);
1081 * Allocate the pollfd structure and the local view of the out fds to avoid
1082 * doing a lookup in the linked list and concurrency issues when writing is
1083 * needed. Called with consumer_data.lock held.
1085 * Returns the number of fds in the structures.
1087 static int update_poll_array(struct lttng_consumer_local_data
*ctx
,
1088 struct pollfd
**pollfd
, struct lttng_consumer_stream
**local_stream
,
1089 struct lttng_ht
*ht
, int *nb_inactive_fd
)
1092 struct lttng_ht_iter iter
;
1093 struct lttng_consumer_stream
*stream
;
1098 assert(local_stream
);
1100 DBG("Updating poll fd array");
1101 *nb_inactive_fd
= 0;
1103 cds_lfht_for_each_entry(ht
->ht
, &iter
.iter
, stream
, node
.node
) {
1105 * Only active streams with an active end point can be added to the
1106 * poll set and local stream storage of the thread.
1108 * There is a potential race here for endpoint_status to be updated
1109 * just after the check. However, this is OK since the stream(s) will
1110 * be deleted once the thread is notified that the end point state has
1111 * changed where this function will be called back again.
1113 * We track the number of inactive FDs because they still need to be
1114 * closed by the polling thread after a wakeup on the data_pipe or
1117 if (stream
->state
!= LTTNG_CONSUMER_ACTIVE_STREAM
||
1118 stream
->endpoint_status
== CONSUMER_ENDPOINT_INACTIVE
) {
1119 (*nb_inactive_fd
)++;
1123 * This clobbers way too much the debug output. Uncomment that if you
1124 * need it for debugging purposes.
1126 * DBG("Active FD %d", stream->wait_fd);
1128 (*pollfd
)[i
].fd
= stream
->wait_fd
;
1129 (*pollfd
)[i
].events
= POLLIN
| POLLPRI
;
1130 local_stream
[i
] = stream
;
1136 * Insert the consumer_data_pipe at the end of the array and don't
1137 * increment i so nb_fd is the number of real FD.
1139 (*pollfd
)[i
].fd
= lttng_pipe_get_readfd(ctx
->consumer_data_pipe
);
1140 (*pollfd
)[i
].events
= POLLIN
| POLLPRI
;
1142 (*pollfd
)[i
+ 1].fd
= lttng_pipe_get_readfd(ctx
->consumer_wakeup_pipe
);
1143 (*pollfd
)[i
+ 1].events
= POLLIN
| POLLPRI
;
1148 * Poll on the should_quit pipe and the command socket return -1 on
1149 * error, 1 if should exit, 0 if data is available on the command socket
1151 int lttng_consumer_poll_socket(struct pollfd
*consumer_sockpoll
)
1156 num_rdy
= poll(consumer_sockpoll
, 2, -1);
1157 if (num_rdy
== -1) {
1159 * Restart interrupted system call.
1161 if (errno
== EINTR
) {
1164 PERROR("Poll error");
1167 if (consumer_sockpoll
[0].revents
& (POLLIN
| POLLPRI
)) {
1168 DBG("consumer_should_quit wake up");
1175 * Set the error socket.
1177 void lttng_consumer_set_error_sock(struct lttng_consumer_local_data
*ctx
,
1180 ctx
->consumer_error_socket
= sock
;
1184 * Set the command socket path.
1186 void lttng_consumer_set_command_sock_path(
1187 struct lttng_consumer_local_data
*ctx
, char *sock
)
1189 ctx
->consumer_command_sock_path
= sock
;
1193 * Send return code to the session daemon.
1194 * If the socket is not defined, we return 0, it is not a fatal error
1196 int lttng_consumer_send_error(struct lttng_consumer_local_data
*ctx
, int cmd
)
1198 if (ctx
->consumer_error_socket
> 0) {
1199 return lttcomm_send_unix_sock(ctx
->consumer_error_socket
, &cmd
,
1200 sizeof(enum lttcomm_sessiond_command
));
1207 * Close all the tracefiles and stream fds and MUST be called when all
1208 * instances are destroyed i.e. when all threads were joined and are ended.
1210 void lttng_consumer_cleanup(void)
1212 struct lttng_ht_iter iter
;
1213 struct lttng_consumer_channel
*channel
;
1217 cds_lfht_for_each_entry(consumer_data
.channel_ht
->ht
, &iter
.iter
, channel
,
1219 consumer_del_channel(channel
);
1224 lttng_ht_destroy(consumer_data
.channel_ht
);
1226 cleanup_relayd_ht();
1228 lttng_ht_destroy(consumer_data
.stream_per_chan_id_ht
);
1231 * This HT contains streams that are freed by either the metadata thread or
1232 * the data thread so we do *nothing* on the hash table and simply destroy
1235 lttng_ht_destroy(consumer_data
.stream_list_ht
);
1239 * Called from signal handler.
1241 void lttng_consumer_should_exit(struct lttng_consumer_local_data
*ctx
)
1245 CMM_STORE_SHARED(consumer_quit
, 1);
1246 ret
= lttng_write(ctx
->consumer_should_quit
[1], "4", 1);
1248 PERROR("write consumer quit");
1251 DBG("Consumer flag that it should quit");
1256 * Flush pending writes to trace output disk file.
1259 void lttng_consumer_sync_trace_file(struct lttng_consumer_stream
*stream
,
1263 int outfd
= stream
->out_fd
;
1266 * This does a blocking write-and-wait on any page that belongs to the
1267 * subbuffer prior to the one we just wrote.
1268 * Don't care about error values, as these are just hints and ways to
1269 * limit the amount of page cache used.
1271 if (orig_offset
< stream
->max_sb_size
) {
1274 lttng_sync_file_range(outfd
, orig_offset
- stream
->max_sb_size
,
1275 stream
->max_sb_size
,
1276 SYNC_FILE_RANGE_WAIT_BEFORE
1277 | SYNC_FILE_RANGE_WRITE
1278 | SYNC_FILE_RANGE_WAIT_AFTER
);
1280 * Give hints to the kernel about how we access the file:
1281 * POSIX_FADV_DONTNEED : we won't re-access data in a near future after
1284 * We need to call fadvise again after the file grows because the
1285 * kernel does not seem to apply fadvise to non-existing parts of the
1288 * Call fadvise _after_ having waited for the page writeback to
1289 * complete because the dirty page writeback semantic is not well
1290 * defined. So it can be expected to lead to lower throughput in
1293 ret
= posix_fadvise(outfd
, orig_offset
- stream
->max_sb_size
,
1294 stream
->max_sb_size
, POSIX_FADV_DONTNEED
);
1295 if (ret
&& ret
!= -ENOSYS
) {
1297 PERROR("posix_fadvise on fd %i", outfd
);
1302 * Initialise the necessary environnement :
1303 * - create a new context
1304 * - create the poll_pipe
1305 * - create the should_quit pipe (for signal handler)
1306 * - create the thread pipe (for splice)
1308 * Takes a function pointer as argument, this function is called when data is
1309 * available on a buffer. This function is responsible to do the
1310 * kernctl_get_next_subbuf, read the data with mmap or splice depending on the
1311 * buffer configuration and then kernctl_put_next_subbuf at the end.
1313 * Returns a pointer to the new context or NULL on error.
1315 struct lttng_consumer_local_data
*lttng_consumer_create(
1316 enum lttng_consumer_type type
,
1317 ssize_t (*buffer_ready
)(struct lttng_consumer_stream
*stream
,
1318 struct lttng_consumer_local_data
*ctx
),
1319 int (*recv_channel
)(struct lttng_consumer_channel
*channel
),
1320 int (*recv_stream
)(struct lttng_consumer_stream
*stream
),
1321 int (*update_stream
)(uint64_t stream_key
, uint32_t state
))
1324 struct lttng_consumer_local_data
*ctx
;
1326 assert(consumer_data
.type
== LTTNG_CONSUMER_UNKNOWN
||
1327 consumer_data
.type
== type
);
1328 consumer_data
.type
= type
;
1330 ctx
= zmalloc(sizeof(struct lttng_consumer_local_data
));
1332 PERROR("allocating context");
1336 ctx
->consumer_error_socket
= -1;
1337 ctx
->consumer_metadata_socket
= -1;
1338 pthread_mutex_init(&ctx
->metadata_socket_lock
, NULL
);
1339 /* assign the callbacks */
1340 ctx
->on_buffer_ready
= buffer_ready
;
1341 ctx
->on_recv_channel
= recv_channel
;
1342 ctx
->on_recv_stream
= recv_stream
;
1343 ctx
->on_update_stream
= update_stream
;
1345 ctx
->consumer_data_pipe
= lttng_pipe_open(0);
1346 if (!ctx
->consumer_data_pipe
) {
1347 goto error_poll_pipe
;
1350 ctx
->consumer_wakeup_pipe
= lttng_pipe_open(0);
1351 if (!ctx
->consumer_wakeup_pipe
) {
1352 goto error_wakeup_pipe
;
1355 ret
= pipe(ctx
->consumer_should_quit
);
1357 PERROR("Error creating recv pipe");
1358 goto error_quit_pipe
;
1361 ret
= pipe(ctx
->consumer_channel_pipe
);
1363 PERROR("Error creating channel pipe");
1364 goto error_channel_pipe
;
1367 ctx
->consumer_metadata_pipe
= lttng_pipe_open(0);
1368 if (!ctx
->consumer_metadata_pipe
) {
1369 goto error_metadata_pipe
;
1372 ctx
->channel_monitor_pipe
= -1;
1376 error_metadata_pipe
:
1377 utils_close_pipe(ctx
->consumer_channel_pipe
);
1379 utils_close_pipe(ctx
->consumer_should_quit
);
1381 lttng_pipe_destroy(ctx
->consumer_wakeup_pipe
);
1383 lttng_pipe_destroy(ctx
->consumer_data_pipe
);
1391 * Iterate over all streams of the hashtable and free them properly.
1393 static void destroy_data_stream_ht(struct lttng_ht
*ht
)
1395 struct lttng_ht_iter iter
;
1396 struct lttng_consumer_stream
*stream
;
1403 cds_lfht_for_each_entry(ht
->ht
, &iter
.iter
, stream
, node
.node
) {
1405 * Ignore return value since we are currently cleaning up so any error
1408 (void) consumer_del_stream(stream
, ht
);
1412 lttng_ht_destroy(ht
);
1416 * Iterate over all streams of the metadata hashtable and free them
1419 static void destroy_metadata_stream_ht(struct lttng_ht
*ht
)
1421 struct lttng_ht_iter iter
;
1422 struct lttng_consumer_stream
*stream
;
1429 cds_lfht_for_each_entry(ht
->ht
, &iter
.iter
, stream
, node
.node
) {
1431 * Ignore return value since we are currently cleaning up so any error
1434 (void) consumer_del_metadata_stream(stream
, ht
);
1438 lttng_ht_destroy(ht
);
1442 * Close all fds associated with the instance and free the context.
1444 void lttng_consumer_destroy(struct lttng_consumer_local_data
*ctx
)
1448 DBG("Consumer destroying it. Closing everything.");
1454 destroy_data_stream_ht(data_ht
);
1455 destroy_metadata_stream_ht(metadata_ht
);
1457 ret
= close(ctx
->consumer_error_socket
);
1461 ret
= close(ctx
->consumer_metadata_socket
);
1465 utils_close_pipe(ctx
->consumer_channel_pipe
);
1466 lttng_pipe_destroy(ctx
->consumer_data_pipe
);
1467 lttng_pipe_destroy(ctx
->consumer_metadata_pipe
);
1468 lttng_pipe_destroy(ctx
->consumer_wakeup_pipe
);
1469 utils_close_pipe(ctx
->consumer_should_quit
);
1471 unlink(ctx
->consumer_command_sock_path
);
1476 * Write the metadata stream id on the specified file descriptor.
1478 static int write_relayd_metadata_id(int fd
,
1479 struct lttng_consumer_stream
*stream
,
1480 unsigned long padding
)
1483 struct lttcomm_relayd_metadata_payload hdr
;
1485 hdr
.stream_id
= htobe64(stream
->relayd_stream_id
);
1486 hdr
.padding_size
= htobe32(padding
);
1487 ret
= lttng_write(fd
, (void *) &hdr
, sizeof(hdr
));
1488 if (ret
< sizeof(hdr
)) {
1490 * This error means that the fd's end is closed so ignore the PERROR
1491 * not to clubber the error output since this can happen in a normal
1494 if (errno
!= EPIPE
) {
1495 PERROR("write metadata stream id");
1497 DBG3("Consumer failed to write relayd metadata id (errno: %d)", errno
);
1499 * Set ret to a negative value because if ret != sizeof(hdr), we don't
1500 * handle writting the missing part so report that as an error and
1501 * don't lie to the caller.
1506 DBG("Metadata stream id %" PRIu64
" with padding %lu written before data",
1507 stream
->relayd_stream_id
, padding
);
1514 * Mmap the ring buffer, read it and write the data to the tracefile. This is a
1515 * core function for writing trace buffers to either the local filesystem or
1518 * It must be called with the stream lock held.
1520 * Careful review MUST be put if any changes occur!
1522 * Returns the number of bytes written
1524 ssize_t
lttng_consumer_on_read_subbuffer_mmap(
1525 struct lttng_consumer_local_data
*ctx
,
1526 struct lttng_consumer_stream
*stream
, unsigned long len
,
1527 unsigned long padding
,
1528 struct ctf_packet_index
*index
)
1530 unsigned long mmap_offset
;
1533 off_t orig_offset
= stream
->out_fd_offset
;
1534 /* Default is on the disk */
1535 int outfd
= stream
->out_fd
;
1536 struct consumer_relayd_sock_pair
*relayd
= NULL
;
1537 unsigned int relayd_hang_up
= 0;
1539 /* RCU lock for the relayd pointer */
1542 /* Flag that the current stream if set for network streaming. */
1543 if (stream
->net_seq_idx
!= (uint64_t) -1ULL) {
1544 relayd
= consumer_find_relayd(stream
->net_seq_idx
);
1545 if (relayd
== NULL
) {
1551 /* get the offset inside the fd to mmap */
1552 switch (consumer_data
.type
) {
1553 case LTTNG_CONSUMER_KERNEL
:
1554 mmap_base
= stream
->mmap_base
;
1555 ret
= kernctl_get_mmap_read_offset(stream
->wait_fd
, &mmap_offset
);
1557 PERROR("tracer ctl get_mmap_read_offset");
1561 case LTTNG_CONSUMER32_UST
:
1562 case LTTNG_CONSUMER64_UST
:
1563 mmap_base
= lttng_ustctl_get_mmap_base(stream
);
1565 ERR("read mmap get mmap base for stream %s", stream
->name
);
1569 ret
= lttng_ustctl_get_mmap_read_offset(stream
, &mmap_offset
);
1571 PERROR("tracer ctl get_mmap_read_offset");
1577 ERR("Unknown consumer_data type");
1581 /* Handle stream on the relayd if the output is on the network */
1583 unsigned long netlen
= len
;
1586 * Lock the control socket for the complete duration of the function
1587 * since from this point on we will use the socket.
1589 if (stream
->metadata_flag
) {
1590 /* Metadata requires the control socket. */
1591 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
1592 if (stream
->reset_metadata_flag
) {
1593 ret
= relayd_reset_metadata(&relayd
->control_sock
,
1594 stream
->relayd_stream_id
,
1595 stream
->metadata_version
);
1600 stream
->reset_metadata_flag
= 0;
1602 netlen
+= sizeof(struct lttcomm_relayd_metadata_payload
);
1605 ret
= write_relayd_stream_header(stream
, netlen
, padding
, relayd
);
1610 /* Use the returned socket. */
1613 /* Write metadata stream id before payload */
1614 if (stream
->metadata_flag
) {
1615 ret
= write_relayd_metadata_id(outfd
, stream
, padding
);
1622 /* No streaming, we have to set the len with the full padding */
1625 if (stream
->metadata_flag
&& stream
->reset_metadata_flag
) {
1626 ret
= utils_truncate_stream_file(stream
->out_fd
, 0);
1628 ERR("Reset metadata file");
1631 stream
->reset_metadata_flag
= 0;
1635 * Check if we need to change the tracefile before writing the packet.
1637 if (stream
->chan
->tracefile_size
> 0 &&
1638 (stream
->tracefile_size_current
+ len
) >
1639 stream
->chan
->tracefile_size
) {
1640 ret
= utils_rotate_stream_file(stream
->chan
->pathname
,
1641 stream
->name
, stream
->chan
->tracefile_size
,
1642 stream
->chan
->tracefile_count
, stream
->uid
, stream
->gid
,
1643 stream
->out_fd
, &(stream
->tracefile_count_current
),
1646 ERR("Rotating output file");
1649 outfd
= stream
->out_fd
;
1651 if (stream
->index_file
) {
1652 lttng_index_file_put(stream
->index_file
);
1653 stream
->index_file
= lttng_index_file_create(stream
->chan
->pathname
,
1654 stream
->name
, stream
->uid
, stream
->gid
,
1655 stream
->chan
->tracefile_size
,
1656 stream
->tracefile_count_current
,
1657 CTF_INDEX_MAJOR
, CTF_INDEX_MINOR
);
1658 if (!stream
->index_file
) {
1663 /* Reset current size because we just perform a rotation. */
1664 stream
->tracefile_size_current
= 0;
1665 stream
->out_fd_offset
= 0;
1668 stream
->tracefile_size_current
+= len
;
1670 index
->offset
= htobe64(stream
->out_fd_offset
);
1675 * This call guarantee that len or less is returned. It's impossible to
1676 * receive a ret value that is bigger than len.
1678 ret
= lttng_write(outfd
, mmap_base
+ mmap_offset
, len
);
1679 DBG("Consumer mmap write() ret %zd (len %lu)", ret
, len
);
1680 if (ret
< 0 || ((size_t) ret
!= len
)) {
1682 * Report error to caller if nothing was written else at least send the
1690 /* Socket operation failed. We consider the relayd dead */
1691 if (errno
== EPIPE
|| errno
== EINVAL
|| errno
== EBADF
) {
1693 * This is possible if the fd is closed on the other side
1694 * (outfd) or any write problem. It can be verbose a bit for a
1695 * normal execution if for instance the relayd is stopped
1696 * abruptly. This can happen so set this to a DBG statement.
1698 DBG("Consumer mmap write detected relayd hang up");
1700 /* Unhandled error, print it and stop function right now. */
1701 PERROR("Error in write mmap (ret %zd != len %lu)", ret
, len
);
1705 stream
->output_written
+= ret
;
1707 /* This call is useless on a socket so better save a syscall. */
1709 /* This won't block, but will start writeout asynchronously */
1710 lttng_sync_file_range(outfd
, stream
->out_fd_offset
, len
,
1711 SYNC_FILE_RANGE_WRITE
);
1712 stream
->out_fd_offset
+= len
;
1713 lttng_consumer_sync_trace_file(stream
, orig_offset
);
1718 * This is a special case that the relayd has closed its socket. Let's
1719 * cleanup the relayd object and all associated streams.
1721 if (relayd
&& relayd_hang_up
) {
1722 cleanup_relayd(relayd
, ctx
);
1726 /* Unlock only if ctrl socket used */
1727 if (relayd
&& stream
->metadata_flag
) {
1728 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
1736 * Splice the data from the ring buffer to the tracefile.
1738 * It must be called with the stream lock held.
1740 * Returns the number of bytes spliced.
1742 ssize_t
lttng_consumer_on_read_subbuffer_splice(
1743 struct lttng_consumer_local_data
*ctx
,
1744 struct lttng_consumer_stream
*stream
, unsigned long len
,
1745 unsigned long padding
,
1746 struct ctf_packet_index
*index
)
1748 ssize_t ret
= 0, written
= 0, ret_splice
= 0;
1750 off_t orig_offset
= stream
->out_fd_offset
;
1751 int fd
= stream
->wait_fd
;
1752 /* Default is on the disk */
1753 int outfd
= stream
->out_fd
;
1754 struct consumer_relayd_sock_pair
*relayd
= NULL
;
1756 unsigned int relayd_hang_up
= 0;
1758 switch (consumer_data
.type
) {
1759 case LTTNG_CONSUMER_KERNEL
:
1761 case LTTNG_CONSUMER32_UST
:
1762 case LTTNG_CONSUMER64_UST
:
1763 /* Not supported for user space tracing */
1766 ERR("Unknown consumer_data type");
1770 /* RCU lock for the relayd pointer */
1773 /* Flag that the current stream if set for network streaming. */
1774 if (stream
->net_seq_idx
!= (uint64_t) -1ULL) {
1775 relayd
= consumer_find_relayd(stream
->net_seq_idx
);
1776 if (relayd
== NULL
) {
1781 splice_pipe
= stream
->splice_pipe
;
1783 /* Write metadata stream id before payload */
1785 unsigned long total_len
= len
;
1787 if (stream
->metadata_flag
) {
1789 * Lock the control socket for the complete duration of the function
1790 * since from this point on we will use the socket.
1792 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
1794 if (stream
->reset_metadata_flag
) {
1795 ret
= relayd_reset_metadata(&relayd
->control_sock
,
1796 stream
->relayd_stream_id
,
1797 stream
->metadata_version
);
1802 stream
->reset_metadata_flag
= 0;
1804 ret
= write_relayd_metadata_id(splice_pipe
[1], stream
,
1812 total_len
+= sizeof(struct lttcomm_relayd_metadata_payload
);
1815 ret
= write_relayd_stream_header(stream
, total_len
, padding
, relayd
);
1821 /* Use the returned socket. */
1824 /* No streaming, we have to set the len with the full padding */
1827 if (stream
->metadata_flag
&& stream
->reset_metadata_flag
) {
1828 ret
= utils_truncate_stream_file(stream
->out_fd
, 0);
1830 ERR("Reset metadata file");
1833 stream
->reset_metadata_flag
= 0;
1836 * Check if we need to change the tracefile before writing the packet.
1838 if (stream
->chan
->tracefile_size
> 0 &&
1839 (stream
->tracefile_size_current
+ len
) >
1840 stream
->chan
->tracefile_size
) {
1841 ret
= utils_rotate_stream_file(stream
->chan
->pathname
,
1842 stream
->name
, stream
->chan
->tracefile_size
,
1843 stream
->chan
->tracefile_count
, stream
->uid
, stream
->gid
,
1844 stream
->out_fd
, &(stream
->tracefile_count_current
),
1848 ERR("Rotating output file");
1851 outfd
= stream
->out_fd
;
1853 if (stream
->index_file
) {
1854 lttng_index_file_put(stream
->index_file
);
1855 stream
->index_file
= lttng_index_file_create(stream
->chan
->pathname
,
1856 stream
->name
, stream
->uid
, stream
->gid
,
1857 stream
->chan
->tracefile_size
,
1858 stream
->tracefile_count_current
,
1859 CTF_INDEX_MAJOR
, CTF_INDEX_MINOR
);
1860 if (!stream
->index_file
) {
1865 /* Reset current size because we just perform a rotation. */
1866 stream
->tracefile_size_current
= 0;
1867 stream
->out_fd_offset
= 0;
1870 stream
->tracefile_size_current
+= len
;
1871 index
->offset
= htobe64(stream
->out_fd_offset
);
1875 DBG("splice chan to pipe offset %lu of len %lu (fd : %d, pipe: %d)",
1876 (unsigned long)offset
, len
, fd
, splice_pipe
[1]);
1877 ret_splice
= splice(fd
, &offset
, splice_pipe
[1], NULL
, len
,
1878 SPLICE_F_MOVE
| SPLICE_F_MORE
);
1879 DBG("splice chan to pipe, ret %zd", ret_splice
);
1880 if (ret_splice
< 0) {
1883 PERROR("Error in relay splice");
1887 /* Handle stream on the relayd if the output is on the network */
1888 if (relayd
&& stream
->metadata_flag
) {
1889 size_t metadata_payload_size
=
1890 sizeof(struct lttcomm_relayd_metadata_payload
);
1892 /* Update counter to fit the spliced data */
1893 ret_splice
+= metadata_payload_size
;
1894 len
+= metadata_payload_size
;
1896 * We do this so the return value can match the len passed as
1897 * argument to this function.
1899 written
-= metadata_payload_size
;
1902 /* Splice data out */
1903 ret_splice
= splice(splice_pipe
[0], NULL
, outfd
, NULL
,
1904 ret_splice
, SPLICE_F_MOVE
| SPLICE_F_MORE
);
1905 DBG("Consumer splice pipe to file (out_fd: %d), ret %zd",
1907 if (ret_splice
< 0) {
1912 } else if (ret_splice
> len
) {
1914 * We don't expect this code path to be executed but you never know
1915 * so this is an extra protection agains a buggy splice().
1918 written
+= ret_splice
;
1919 PERROR("Wrote more data than requested %zd (len: %lu)", ret_splice
,
1923 /* All good, update current len and continue. */
1927 /* This call is useless on a socket so better save a syscall. */
1929 /* This won't block, but will start writeout asynchronously */
1930 lttng_sync_file_range(outfd
, stream
->out_fd_offset
, ret_splice
,
1931 SYNC_FILE_RANGE_WRITE
);
1932 stream
->out_fd_offset
+= ret_splice
;
1934 stream
->output_written
+= ret_splice
;
1935 written
+= ret_splice
;
1938 lttng_consumer_sync_trace_file(stream
, orig_offset
);
1944 * This is a special case that the relayd has closed its socket. Let's
1945 * cleanup the relayd object and all associated streams.
1947 if (relayd
&& relayd_hang_up
) {
1948 cleanup_relayd(relayd
, ctx
);
1949 /* Skip splice error so the consumer does not fail */
1954 /* send the appropriate error description to sessiond */
1957 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_SPLICE_EINVAL
);
1960 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_SPLICE_ENOMEM
);
1963 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_SPLICE_ESPIPE
);
1968 if (relayd
&& stream
->metadata_flag
) {
1969 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
1977 * Sample the snapshot positions for a specific fd
1979 * Returns 0 on success, < 0 on error
1981 int lttng_consumer_sample_snapshot_positions(struct lttng_consumer_stream
*stream
)
1983 switch (consumer_data
.type
) {
1984 case LTTNG_CONSUMER_KERNEL
:
1985 return lttng_kconsumer_sample_snapshot_positions(stream
);
1986 case LTTNG_CONSUMER32_UST
:
1987 case LTTNG_CONSUMER64_UST
:
1988 return lttng_ustconsumer_sample_snapshot_positions(stream
);
1990 ERR("Unknown consumer_data type");
1996 * Take a snapshot for a specific fd
1998 * Returns 0 on success, < 0 on error
2000 int lttng_consumer_take_snapshot(struct lttng_consumer_stream
*stream
)
2002 switch (consumer_data
.type
) {
2003 case LTTNG_CONSUMER_KERNEL
:
2004 return lttng_kconsumer_take_snapshot(stream
);
2005 case LTTNG_CONSUMER32_UST
:
2006 case LTTNG_CONSUMER64_UST
:
2007 return lttng_ustconsumer_take_snapshot(stream
);
2009 ERR("Unknown consumer_data type");
2016 * Get the produced position
2018 * Returns 0 on success, < 0 on error
2020 int lttng_consumer_get_produced_snapshot(struct lttng_consumer_stream
*stream
,
2023 switch (consumer_data
.type
) {
2024 case LTTNG_CONSUMER_KERNEL
:
2025 return lttng_kconsumer_get_produced_snapshot(stream
, pos
);
2026 case LTTNG_CONSUMER32_UST
:
2027 case LTTNG_CONSUMER64_UST
:
2028 return lttng_ustconsumer_get_produced_snapshot(stream
, pos
);
2030 ERR("Unknown consumer_data type");
2037 * Get the consumed position (free-running counter position in bytes).
2039 * Returns 0 on success, < 0 on error
2041 int lttng_consumer_get_consumed_snapshot(struct lttng_consumer_stream
*stream
,
2044 switch (consumer_data
.type
) {
2045 case LTTNG_CONSUMER_KERNEL
:
2046 return lttng_kconsumer_get_consumed_snapshot(stream
, pos
);
2047 case LTTNG_CONSUMER32_UST
:
2048 case LTTNG_CONSUMER64_UST
:
2049 return lttng_ustconsumer_get_consumed_snapshot(stream
, pos
);
2051 ERR("Unknown consumer_data type");
2057 int lttng_consumer_recv_cmd(struct lttng_consumer_local_data
*ctx
,
2058 int sock
, struct pollfd
*consumer_sockpoll
)
2060 switch (consumer_data
.type
) {
2061 case LTTNG_CONSUMER_KERNEL
:
2062 return lttng_kconsumer_recv_cmd(ctx
, sock
, consumer_sockpoll
);
2063 case LTTNG_CONSUMER32_UST
:
2064 case LTTNG_CONSUMER64_UST
:
2065 return lttng_ustconsumer_recv_cmd(ctx
, sock
, consumer_sockpoll
);
2067 ERR("Unknown consumer_data type");
2073 void lttng_consumer_close_all_metadata(void)
2075 switch (consumer_data
.type
) {
2076 case LTTNG_CONSUMER_KERNEL
:
2078 * The Kernel consumer has a different metadata scheme so we don't
2079 * close anything because the stream will be closed by the session
2083 case LTTNG_CONSUMER32_UST
:
2084 case LTTNG_CONSUMER64_UST
:
2086 * Close all metadata streams. The metadata hash table is passed and
2087 * this call iterates over it by closing all wakeup fd. This is safe
2088 * because at this point we are sure that the metadata producer is
2089 * either dead or blocked.
2091 lttng_ustconsumer_close_all_metadata(metadata_ht
);
2094 ERR("Unknown consumer_data type");
2100 * Clean up a metadata stream and free its memory.
2102 void consumer_del_metadata_stream(struct lttng_consumer_stream
*stream
,
2103 struct lttng_ht
*ht
)
2105 struct lttng_consumer_channel
*free_chan
= NULL
;
2109 * This call should NEVER receive regular stream. It must always be
2110 * metadata stream and this is crucial for data structure synchronization.
2112 assert(stream
->metadata_flag
);
2114 DBG3("Consumer delete metadata stream %d", stream
->wait_fd
);
2116 pthread_mutex_lock(&consumer_data
.lock
);
2117 pthread_mutex_lock(&stream
->chan
->lock
);
2118 pthread_mutex_lock(&stream
->lock
);
2119 if (stream
->chan
->metadata_cache
) {
2120 /* Only applicable to userspace consumers. */
2121 pthread_mutex_lock(&stream
->chan
->metadata_cache
->lock
);
2124 /* Remove any reference to that stream. */
2125 consumer_stream_delete(stream
, ht
);
2127 /* Close down everything including the relayd if one. */
2128 consumer_stream_close(stream
);
2129 /* Destroy tracer buffers of the stream. */
2130 consumer_stream_destroy_buffers(stream
);
2132 /* Atomically decrement channel refcount since other threads can use it. */
2133 if (!uatomic_sub_return(&stream
->chan
->refcount
, 1)
2134 && !uatomic_read(&stream
->chan
->nb_init_stream_left
)) {
2135 /* Go for channel deletion! */
2136 free_chan
= stream
->chan
;
2140 * Nullify the stream reference so it is not used after deletion. The
2141 * channel lock MUST be acquired before being able to check for a NULL
2144 stream
->chan
->metadata_stream
= NULL
;
2146 if (stream
->chan
->metadata_cache
) {
2147 pthread_mutex_unlock(&stream
->chan
->metadata_cache
->lock
);
2149 pthread_mutex_unlock(&stream
->lock
);
2150 pthread_mutex_unlock(&stream
->chan
->lock
);
2151 pthread_mutex_unlock(&consumer_data
.lock
);
2154 consumer_del_channel(free_chan
);
2157 consumer_stream_free(stream
);
2161 * Action done with the metadata stream when adding it to the consumer internal
2162 * data structures to handle it.
2164 void consumer_add_metadata_stream(struct lttng_consumer_stream
*stream
)
2166 struct lttng_ht
*ht
= metadata_ht
;
2167 struct lttng_ht_iter iter
;
2168 struct lttng_ht_node_u64
*node
;
2173 DBG3("Adding metadata stream %" PRIu64
" to hash table", stream
->key
);
2175 pthread_mutex_lock(&consumer_data
.lock
);
2176 pthread_mutex_lock(&stream
->chan
->lock
);
2177 pthread_mutex_lock(&stream
->chan
->timer_lock
);
2178 pthread_mutex_lock(&stream
->lock
);
2181 * From here, refcounts are updated so be _careful_ when returning an error
2188 * Lookup the stream just to make sure it does not exist in our internal
2189 * state. This should NEVER happen.
2191 lttng_ht_lookup(ht
, &stream
->key
, &iter
);
2192 node
= lttng_ht_iter_get_node_u64(&iter
);
2196 * When nb_init_stream_left reaches 0, we don't need to trigger any action
2197 * in terms of destroying the associated channel, because the action that
2198 * causes the count to become 0 also causes a stream to be added. The
2199 * channel deletion will thus be triggered by the following removal of this
2202 if (uatomic_read(&stream
->chan
->nb_init_stream_left
) > 0) {
2203 /* Increment refcount before decrementing nb_init_stream_left */
2205 uatomic_dec(&stream
->chan
->nb_init_stream_left
);
2208 lttng_ht_add_unique_u64(ht
, &stream
->node
);
2210 lttng_ht_add_u64(consumer_data
.stream_per_chan_id_ht
,
2211 &stream
->node_channel_id
);
2214 * Add stream to the stream_list_ht of the consumer data. No need to steal
2215 * the key since the HT does not use it and we allow to add redundant keys
2218 lttng_ht_add_u64(consumer_data
.stream_list_ht
, &stream
->node_session_id
);
2222 pthread_mutex_unlock(&stream
->lock
);
2223 pthread_mutex_unlock(&stream
->chan
->lock
);
2224 pthread_mutex_unlock(&stream
->chan
->timer_lock
);
2225 pthread_mutex_unlock(&consumer_data
.lock
);
2229 * Delete data stream that are flagged for deletion (endpoint_status).
2231 static void validate_endpoint_status_data_stream(void)
2233 struct lttng_ht_iter iter
;
2234 struct lttng_consumer_stream
*stream
;
2236 DBG("Consumer delete flagged data stream");
2239 cds_lfht_for_each_entry(data_ht
->ht
, &iter
.iter
, stream
, node
.node
) {
2240 /* Validate delete flag of the stream */
2241 if (stream
->endpoint_status
== CONSUMER_ENDPOINT_ACTIVE
) {
2244 /* Delete it right now */
2245 consumer_del_stream(stream
, data_ht
);
2251 * Delete metadata stream that are flagged for deletion (endpoint_status).
2253 static void validate_endpoint_status_metadata_stream(
2254 struct lttng_poll_event
*pollset
)
2256 struct lttng_ht_iter iter
;
2257 struct lttng_consumer_stream
*stream
;
2259 DBG("Consumer delete flagged metadata stream");
2264 cds_lfht_for_each_entry(metadata_ht
->ht
, &iter
.iter
, stream
, node
.node
) {
2265 /* Validate delete flag of the stream */
2266 if (stream
->endpoint_status
== CONSUMER_ENDPOINT_ACTIVE
) {
2270 * Remove from pollset so the metadata thread can continue without
2271 * blocking on a deleted stream.
2273 lttng_poll_del(pollset
, stream
->wait_fd
);
2275 /* Delete it right now */
2276 consumer_del_metadata_stream(stream
, metadata_ht
);
2282 int rotate_notify_sessiond(struct lttng_consumer_local_data
*ctx
,
2288 ret
= write(ctx
->channel_rotate_pipe
, &key
, sizeof(key
));
2289 } while (ret
== -1 && errno
== EINTR
);
2291 PERROR("Failed to write to the channel rotation pipe");
2293 DBG("Sent channel rotation notification for channel key %"
2302 * Perform operations that need to be done after a stream has
2303 * rotated and released the stream lock.
2305 * Multiple rotations cannot occur simultaneously, so we know the state of the
2306 * "rotated" stream flag cannot change.
2308 * This MUST be called WITHOUT the stream lock held.
2311 int consumer_post_rotation(struct lttng_consumer_stream
*stream
,
2312 struct lttng_consumer_local_data
*ctx
)
2316 pthread_mutex_lock(&stream
->chan
->lock
);
2318 switch (consumer_data
.type
) {
2319 case LTTNG_CONSUMER_KERNEL
:
2321 case LTTNG_CONSUMER32_UST
:
2322 case LTTNG_CONSUMER64_UST
:
2324 * The ust_metadata_pushed counter has been reset to 0, so now
2325 * we can wakeup the metadata thread so it dumps the metadata
2326 * cache to the new file.
2328 if (stream
->metadata_flag
) {
2329 consumer_metadata_wakeup_pipe(stream
->chan
);
2333 ERR("Unknown consumer_data type");
2337 if (--stream
->chan
->nr_stream_rotate_pending
== 0) {
2338 DBG("Rotation of channel \"%s\" completed, notifying the session daemon",
2339 stream
->chan
->name
);
2340 ret
= rotate_notify_sessiond(ctx
, stream
->chan
->key
);
2342 assert(stream
->chan
->nr_stream_rotate_pending
>= 0);
2343 pthread_mutex_unlock(&stream
->chan
->lock
);
2349 * Thread polls on metadata file descriptor and write them on disk or on the
2352 void *consumer_thread_metadata_poll(void *data
)
2354 int ret
, i
, pollfd
, err
= -1;
2355 uint32_t revents
, nb_fd
;
2356 struct lttng_consumer_stream
*stream
= NULL
;
2357 struct lttng_ht_iter iter
;
2358 struct lttng_ht_node_u64
*node
;
2359 struct lttng_poll_event events
;
2360 struct lttng_consumer_local_data
*ctx
= data
;
2363 rcu_register_thread();
2365 health_register(health_consumerd
, HEALTH_CONSUMERD_TYPE_METADATA
);
2367 if (testpoint(consumerd_thread_metadata
)) {
2368 goto error_testpoint
;
2371 health_code_update();
2373 DBG("Thread metadata poll started");
2375 /* Size is set to 1 for the consumer_metadata pipe */
2376 ret
= lttng_poll_create(&events
, 2, LTTNG_CLOEXEC
);
2378 ERR("Poll set creation failed");
2382 ret
= lttng_poll_add(&events
,
2383 lttng_pipe_get_readfd(ctx
->consumer_metadata_pipe
), LPOLLIN
);
2389 DBG("Metadata main loop started");
2393 health_code_update();
2394 health_poll_entry();
2395 DBG("Metadata poll wait");
2396 ret
= lttng_poll_wait(&events
, -1);
2397 DBG("Metadata poll return from wait with %d fd(s)",
2398 LTTNG_POLL_GETNB(&events
));
2400 DBG("Metadata event caught in thread");
2402 if (errno
== EINTR
) {
2403 ERR("Poll EINTR caught");
2406 if (LTTNG_POLL_GETNB(&events
) == 0) {
2407 err
= 0; /* All is OK */
2414 /* From here, the event is a metadata wait fd */
2415 for (i
= 0; i
< nb_fd
; i
++) {
2416 health_code_update();
2418 revents
= LTTNG_POLL_GETEV(&events
, i
);
2419 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2422 /* No activity for this FD (poll implementation). */
2426 if (pollfd
== lttng_pipe_get_readfd(ctx
->consumer_metadata_pipe
)) {
2427 if (revents
& LPOLLIN
) {
2430 pipe_len
= lttng_pipe_read(ctx
->consumer_metadata_pipe
,
2431 &stream
, sizeof(stream
));
2432 if (pipe_len
< sizeof(stream
)) {
2434 PERROR("read metadata stream");
2437 * Remove the pipe from the poll set and continue the loop
2438 * since their might be data to consume.
2440 lttng_poll_del(&events
,
2441 lttng_pipe_get_readfd(ctx
->consumer_metadata_pipe
));
2442 lttng_pipe_read_close(ctx
->consumer_metadata_pipe
);
2446 /* A NULL stream means that the state has changed. */
2447 if (stream
== NULL
) {
2448 /* Check for deleted streams. */
2449 validate_endpoint_status_metadata_stream(&events
);
2453 DBG("Adding metadata stream %d to poll set",
2456 /* Add metadata stream to the global poll events list */
2457 lttng_poll_add(&events
, stream
->wait_fd
,
2458 LPOLLIN
| LPOLLPRI
| LPOLLHUP
);
2459 } else if (revents
& (LPOLLERR
| LPOLLHUP
)) {
2460 DBG("Metadata thread pipe hung up");
2462 * Remove the pipe from the poll set and continue the loop
2463 * since their might be data to consume.
2465 lttng_poll_del(&events
,
2466 lttng_pipe_get_readfd(ctx
->consumer_metadata_pipe
));
2467 lttng_pipe_read_close(ctx
->consumer_metadata_pipe
);
2470 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
2474 /* Handle other stream */
2480 uint64_t tmp_id
= (uint64_t) pollfd
;
2482 lttng_ht_lookup(metadata_ht
, &tmp_id
, &iter
);
2484 node
= lttng_ht_iter_get_node_u64(&iter
);
2487 stream
= caa_container_of(node
, struct lttng_consumer_stream
,
2490 if (revents
& (LPOLLIN
| LPOLLPRI
)) {
2491 /* Get the data out of the metadata file descriptor */
2492 DBG("Metadata available on fd %d", pollfd
);
2493 assert(stream
->wait_fd
== pollfd
);
2496 health_code_update();
2498 len
= ctx
->on_buffer_ready(stream
, ctx
);
2500 * We don't check the return value here since if we get
2501 * a negative len, it means an error occurred thus we
2502 * simply remove it from the poll set and free the
2507 /* It's ok to have an unavailable sub-buffer */
2508 if (len
< 0 && len
!= -EAGAIN
&& len
!= -ENODATA
) {
2509 /* Clean up stream from consumer and free it. */
2510 lttng_poll_del(&events
, stream
->wait_fd
);
2511 consumer_del_metadata_stream(stream
, metadata_ht
);
2513 } else if (revents
& (LPOLLERR
| LPOLLHUP
)) {
2514 DBG("Metadata fd %d is hup|err.", pollfd
);
2515 if (!stream
->hangup_flush_done
2516 && (consumer_data
.type
== LTTNG_CONSUMER32_UST
2517 || consumer_data
.type
== LTTNG_CONSUMER64_UST
)) {
2518 DBG("Attempting to flush and consume the UST buffers");
2519 lttng_ustconsumer_on_stream_hangup(stream
);
2521 /* We just flushed the stream now read it. */
2523 health_code_update();
2525 len
= ctx
->on_buffer_ready(stream
, ctx
);
2527 * We don't check the return value here since if we get
2528 * a negative len, it means an error occurred thus we
2529 * simply remove it from the poll set and free the
2535 lttng_poll_del(&events
, stream
->wait_fd
);
2537 * This call update the channel states, closes file descriptors
2538 * and securely free the stream.
2540 consumer_del_metadata_stream(stream
, metadata_ht
);
2542 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
2546 /* Release RCU lock for the stream looked up */
2554 DBG("Metadata poll thread exiting");
2556 lttng_poll_clean(&events
);
2561 ERR("Health error occurred in %s", __func__
);
2563 health_unregister(health_consumerd
);
2564 rcu_unregister_thread();
2569 * This thread polls the fds in the set to consume the data and write
2570 * it to tracefile if necessary.
2572 void *consumer_thread_data_poll(void *data
)
2574 int num_rdy
, num_hup
, high_prio
, ret
, i
, err
= -1;
2575 struct pollfd
*pollfd
= NULL
;
2576 /* local view of the streams */
2577 struct lttng_consumer_stream
**local_stream
= NULL
, *new_stream
= NULL
;
2578 /* local view of consumer_data.fds_count */
2579 int nb_fd
= 0, nb_pipes_fd
;
2580 /* Number of FDs with CONSUMER_ENDPOINT_INACTIVE but still open. */
2581 int nb_inactive_fd
= 0;
2582 struct lttng_consumer_local_data
*ctx
= data
;
2585 rcu_register_thread();
2587 health_register(health_consumerd
, HEALTH_CONSUMERD_TYPE_DATA
);
2589 if (testpoint(consumerd_thread_data
)) {
2590 goto error_testpoint
;
2593 health_code_update();
2595 local_stream
= zmalloc(sizeof(struct lttng_consumer_stream
*));
2596 if (local_stream
== NULL
) {
2597 PERROR("local_stream malloc");
2602 health_code_update();
2608 * the fds set has been updated, we need to update our
2609 * local array as well
2611 pthread_mutex_lock(&consumer_data
.lock
);
2612 if (consumer_data
.need_update
) {
2617 local_stream
= NULL
;
2620 * Allocate for all fds + 2:
2621 * +1 for the consumer_data_pipe
2622 * +1 for wake up pipe
2625 pollfd
= zmalloc((consumer_data
.stream_count
+ nb_pipes_fd
) * sizeof(struct pollfd
));
2626 if (pollfd
== NULL
) {
2627 PERROR("pollfd malloc");
2628 pthread_mutex_unlock(&consumer_data
.lock
);
2632 local_stream
= zmalloc((consumer_data
.stream_count
+ nb_pipes_fd
) *
2633 sizeof(struct lttng_consumer_stream
*));
2634 if (local_stream
== NULL
) {
2635 PERROR("local_stream malloc");
2636 pthread_mutex_unlock(&consumer_data
.lock
);
2639 ret
= update_poll_array(ctx
, &pollfd
, local_stream
,
2640 data_ht
, &nb_inactive_fd
);
2642 ERR("Error in allocating pollfd or local_outfds");
2643 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_POLL_ERROR
);
2644 pthread_mutex_unlock(&consumer_data
.lock
);
2648 consumer_data
.need_update
= 0;
2650 pthread_mutex_unlock(&consumer_data
.lock
);
2652 /* No FDs and consumer_quit, consumer_cleanup the thread */
2653 if (nb_fd
== 0 && nb_inactive_fd
== 0 &&
2654 CMM_LOAD_SHARED(consumer_quit
) == 1) {
2655 err
= 0; /* All is OK */
2658 /* poll on the array of fds */
2660 DBG("polling on %d fd", nb_fd
+ nb_pipes_fd
);
2661 if (testpoint(consumerd_thread_data_poll
)) {
2664 health_poll_entry();
2665 num_rdy
= poll(pollfd
, nb_fd
+ nb_pipes_fd
, -1);
2667 DBG("poll num_rdy : %d", num_rdy
);
2668 if (num_rdy
== -1) {
2670 * Restart interrupted system call.
2672 if (errno
== EINTR
) {
2675 PERROR("Poll error");
2676 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_POLL_ERROR
);
2678 } else if (num_rdy
== 0) {
2679 DBG("Polling thread timed out");
2683 if (caa_unlikely(data_consumption_paused
)) {
2684 DBG("Data consumption paused, sleeping...");
2690 * If the consumer_data_pipe triggered poll go directly to the
2691 * beginning of the loop to update the array. We want to prioritize
2692 * array update over low-priority reads.
2694 if (pollfd
[nb_fd
].revents
& (POLLIN
| POLLPRI
)) {
2695 ssize_t pipe_readlen
;
2697 DBG("consumer_data_pipe wake up");
2698 pipe_readlen
= lttng_pipe_read(ctx
->consumer_data_pipe
,
2699 &new_stream
, sizeof(new_stream
));
2700 if (pipe_readlen
< sizeof(new_stream
)) {
2701 PERROR("Consumer data pipe");
2702 /* Continue so we can at least handle the current stream(s). */
2707 * If the stream is NULL, just ignore it. It's also possible that
2708 * the sessiond poll thread changed the consumer_quit state and is
2709 * waking us up to test it.
2711 if (new_stream
== NULL
) {
2712 validate_endpoint_status_data_stream();
2716 /* Continue to update the local streams and handle prio ones */
2720 /* Handle wakeup pipe. */
2721 if (pollfd
[nb_fd
+ 1].revents
& (POLLIN
| POLLPRI
)) {
2723 ssize_t pipe_readlen
;
2725 pipe_readlen
= lttng_pipe_read(ctx
->consumer_wakeup_pipe
, &dummy
,
2727 if (pipe_readlen
< 0) {
2728 PERROR("Consumer data wakeup pipe");
2730 /* We've been awakened to handle stream(s). */
2731 ctx
->has_wakeup
= 0;
2734 /* Take care of high priority channels first. */
2735 for (i
= 0; i
< nb_fd
; i
++) {
2736 health_code_update();
2738 if (local_stream
[i
] == NULL
) {
2741 if (pollfd
[i
].revents
& POLLPRI
) {
2742 DBG("Urgent read on fd %d", pollfd
[i
].fd
);
2744 len
= ctx
->on_buffer_ready(local_stream
[i
], ctx
);
2745 /* it's ok to have an unavailable sub-buffer */
2746 if (len
< 0 && len
!= -EAGAIN
&& len
!= -ENODATA
) {
2747 /* Clean the stream and free it. */
2748 consumer_del_stream(local_stream
[i
], data_ht
);
2749 local_stream
[i
] = NULL
;
2750 } else if (len
> 0) {
2751 local_stream
[i
]->data_read
= 1;
2757 * If we read high prio channel in this loop, try again
2758 * for more high prio data.
2764 /* Take care of low priority channels. */
2765 for (i
= 0; i
< nb_fd
; i
++) {
2766 health_code_update();
2768 if (local_stream
[i
] == NULL
) {
2771 if ((pollfd
[i
].revents
& POLLIN
) ||
2772 local_stream
[i
]->hangup_flush_done
||
2773 local_stream
[i
]->has_data
) {
2774 DBG("Normal read on fd %d", pollfd
[i
].fd
);
2775 len
= ctx
->on_buffer_ready(local_stream
[i
], ctx
);
2776 /* it's ok to have an unavailable sub-buffer */
2777 if (len
< 0 && len
!= -EAGAIN
&& len
!= -ENODATA
) {
2778 /* Clean the stream and free it. */
2779 consumer_del_stream(local_stream
[i
], data_ht
);
2780 local_stream
[i
] = NULL
;
2781 } else if (len
> 0) {
2782 local_stream
[i
]->data_read
= 1;
2787 /* Handle hangup and errors */
2788 for (i
= 0; i
< nb_fd
; i
++) {
2789 health_code_update();
2791 if (local_stream
[i
] == NULL
) {
2794 if (!local_stream
[i
]->hangup_flush_done
2795 && (pollfd
[i
].revents
& (POLLHUP
| POLLERR
| POLLNVAL
))
2796 && (consumer_data
.type
== LTTNG_CONSUMER32_UST
2797 || consumer_data
.type
== LTTNG_CONSUMER64_UST
)) {
2798 DBG("fd %d is hup|err|nval. Attempting flush and read.",
2800 lttng_ustconsumer_on_stream_hangup(local_stream
[i
]);
2801 /* Attempt read again, for the data we just flushed. */
2802 local_stream
[i
]->data_read
= 1;
2805 * If the poll flag is HUP/ERR/NVAL and we have
2806 * read no data in this pass, we can remove the
2807 * stream from its hash table.
2809 if ((pollfd
[i
].revents
& POLLHUP
)) {
2810 DBG("Polling fd %d tells it has hung up.", pollfd
[i
].fd
);
2811 if (!local_stream
[i
]->data_read
) {
2812 consumer_del_stream(local_stream
[i
], data_ht
);
2813 local_stream
[i
] = NULL
;
2816 } else if (pollfd
[i
].revents
& POLLERR
) {
2817 ERR("Error returned in polling fd %d.", pollfd
[i
].fd
);
2818 if (!local_stream
[i
]->data_read
) {
2819 consumer_del_stream(local_stream
[i
], data_ht
);
2820 local_stream
[i
] = NULL
;
2823 } else if (pollfd
[i
].revents
& POLLNVAL
) {
2824 ERR("Polling fd %d tells fd is not open.", pollfd
[i
].fd
);
2825 if (!local_stream
[i
]->data_read
) {
2826 consumer_del_stream(local_stream
[i
], data_ht
);
2827 local_stream
[i
] = NULL
;
2831 if (local_stream
[i
] != NULL
) {
2832 local_stream
[i
]->data_read
= 0;
2839 DBG("polling thread exiting");
2844 * Close the write side of the pipe so epoll_wait() in
2845 * consumer_thread_metadata_poll can catch it. The thread is monitoring the
2846 * read side of the pipe. If we close them both, epoll_wait strangely does
2847 * not return and could create a endless wait period if the pipe is the
2848 * only tracked fd in the poll set. The thread will take care of closing
2851 (void) lttng_pipe_write_close(ctx
->consumer_metadata_pipe
);
2856 ERR("Health error occurred in %s", __func__
);
2858 health_unregister(health_consumerd
);
2860 rcu_unregister_thread();
2865 * Close wake-up end of each stream belonging to the channel. This will
2866 * allow the poll() on the stream read-side to detect when the
2867 * write-side (application) finally closes them.
2870 void consumer_close_channel_streams(struct lttng_consumer_channel
*channel
)
2872 struct lttng_ht
*ht
;
2873 struct lttng_consumer_stream
*stream
;
2874 struct lttng_ht_iter iter
;
2876 ht
= consumer_data
.stream_per_chan_id_ht
;
2879 cds_lfht_for_each_entry_duplicate(ht
->ht
,
2880 ht
->hash_fct(&channel
->key
, lttng_ht_seed
),
2881 ht
->match_fct
, &channel
->key
,
2882 &iter
.iter
, stream
, node_channel_id
.node
) {
2884 * Protect against teardown with mutex.
2886 pthread_mutex_lock(&stream
->lock
);
2887 if (cds_lfht_is_node_deleted(&stream
->node
.node
)) {
2890 switch (consumer_data
.type
) {
2891 case LTTNG_CONSUMER_KERNEL
:
2893 case LTTNG_CONSUMER32_UST
:
2894 case LTTNG_CONSUMER64_UST
:
2895 if (stream
->metadata_flag
) {
2896 /* Safe and protected by the stream lock. */
2897 lttng_ustconsumer_close_metadata(stream
->chan
);
2900 * Note: a mutex is taken internally within
2901 * liblttng-ust-ctl to protect timer wakeup_fd
2902 * use from concurrent close.
2904 lttng_ustconsumer_close_stream_wakeup(stream
);
2908 ERR("Unknown consumer_data type");
2912 pthread_mutex_unlock(&stream
->lock
);
2917 static void destroy_channel_ht(struct lttng_ht
*ht
)
2919 struct lttng_ht_iter iter
;
2920 struct lttng_consumer_channel
*channel
;
2928 cds_lfht_for_each_entry(ht
->ht
, &iter
.iter
, channel
, wait_fd_node
.node
) {
2929 ret
= lttng_ht_del(ht
, &iter
);
2934 lttng_ht_destroy(ht
);
2938 * This thread polls the channel fds to detect when they are being
2939 * closed. It closes all related streams if the channel is detected as
2940 * closed. It is currently only used as a shim layer for UST because the
2941 * consumerd needs to keep the per-stream wakeup end of pipes open for
2944 void *consumer_thread_channel_poll(void *data
)
2946 int ret
, i
, pollfd
, err
= -1;
2947 uint32_t revents
, nb_fd
;
2948 struct lttng_consumer_channel
*chan
= NULL
;
2949 struct lttng_ht_iter iter
;
2950 struct lttng_ht_node_u64
*node
;
2951 struct lttng_poll_event events
;
2952 struct lttng_consumer_local_data
*ctx
= data
;
2953 struct lttng_ht
*channel_ht
;
2955 rcu_register_thread();
2957 health_register(health_consumerd
, HEALTH_CONSUMERD_TYPE_CHANNEL
);
2959 if (testpoint(consumerd_thread_channel
)) {
2960 goto error_testpoint
;
2963 health_code_update();
2965 channel_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
2967 /* ENOMEM at this point. Better to bail out. */
2971 DBG("Thread channel poll started");
2973 /* Size is set to 1 for the consumer_channel pipe */
2974 ret
= lttng_poll_create(&events
, 2, LTTNG_CLOEXEC
);
2976 ERR("Poll set creation failed");
2980 ret
= lttng_poll_add(&events
, ctx
->consumer_channel_pipe
[0], LPOLLIN
);
2986 DBG("Channel main loop started");
2990 health_code_update();
2991 DBG("Channel poll wait");
2992 health_poll_entry();
2993 ret
= lttng_poll_wait(&events
, -1);
2994 DBG("Channel poll return from wait with %d fd(s)",
2995 LTTNG_POLL_GETNB(&events
));
2997 DBG("Channel event caught in thread");
2999 if (errno
== EINTR
) {
3000 ERR("Poll EINTR caught");
3003 if (LTTNG_POLL_GETNB(&events
) == 0) {
3004 err
= 0; /* All is OK */
3011 /* From here, the event is a channel wait fd */
3012 for (i
= 0; i
< nb_fd
; i
++) {
3013 health_code_update();
3015 revents
= LTTNG_POLL_GETEV(&events
, i
);
3016 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3019 /* No activity for this FD (poll implementation). */
3023 if (pollfd
== ctx
->consumer_channel_pipe
[0]) {
3024 if (revents
& LPOLLIN
) {
3025 enum consumer_channel_action action
;
3028 ret
= read_channel_pipe(ctx
, &chan
, &key
, &action
);
3031 ERR("Error reading channel pipe");
3033 lttng_poll_del(&events
, ctx
->consumer_channel_pipe
[0]);
3038 case CONSUMER_CHANNEL_ADD
:
3039 DBG("Adding channel %d to poll set",
3042 lttng_ht_node_init_u64(&chan
->wait_fd_node
,
3045 lttng_ht_add_unique_u64(channel_ht
,
3046 &chan
->wait_fd_node
);
3048 /* Add channel to the global poll events list */
3049 lttng_poll_add(&events
, chan
->wait_fd
,
3050 LPOLLERR
| LPOLLHUP
);
3052 case CONSUMER_CHANNEL_DEL
:
3055 * This command should never be called if the channel
3056 * has streams monitored by either the data or metadata
3057 * thread. The consumer only notify this thread with a
3058 * channel del. command if it receives a destroy
3059 * channel command from the session daemon that send it
3060 * if a command prior to the GET_CHANNEL failed.
3064 chan
= consumer_find_channel(key
);
3067 ERR("UST consumer get channel key %" PRIu64
" not found for del channel", key
);
3070 lttng_poll_del(&events
, chan
->wait_fd
);
3071 iter
.iter
.node
= &chan
->wait_fd_node
.node
;
3072 ret
= lttng_ht_del(channel_ht
, &iter
);
3075 switch (consumer_data
.type
) {
3076 case LTTNG_CONSUMER_KERNEL
:
3078 case LTTNG_CONSUMER32_UST
:
3079 case LTTNG_CONSUMER64_UST
:
3080 health_code_update();
3081 /* Destroy streams that might have been left in the stream list. */
3082 clean_channel_stream_list(chan
);
3085 ERR("Unknown consumer_data type");
3090 * Release our own refcount. Force channel deletion even if
3091 * streams were not initialized.
3093 if (!uatomic_sub_return(&chan
->refcount
, 1)) {
3094 consumer_del_channel(chan
);
3099 case CONSUMER_CHANNEL_QUIT
:
3101 * Remove the pipe from the poll set and continue the loop
3102 * since their might be data to consume.
3104 lttng_poll_del(&events
, ctx
->consumer_channel_pipe
[0]);
3107 ERR("Unknown action");
3110 } else if (revents
& (LPOLLERR
| LPOLLHUP
)) {
3111 DBG("Channel thread pipe hung up");
3113 * Remove the pipe from the poll set and continue the loop
3114 * since their might be data to consume.
3116 lttng_poll_del(&events
, ctx
->consumer_channel_pipe
[0]);
3119 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
3123 /* Handle other stream */
3129 uint64_t tmp_id
= (uint64_t) pollfd
;
3131 lttng_ht_lookup(channel_ht
, &tmp_id
, &iter
);
3133 node
= lttng_ht_iter_get_node_u64(&iter
);
3136 chan
= caa_container_of(node
, struct lttng_consumer_channel
,
3139 /* Check for error event */
3140 if (revents
& (LPOLLERR
| LPOLLHUP
)) {
3141 DBG("Channel fd %d is hup|err.", pollfd
);
3143 lttng_poll_del(&events
, chan
->wait_fd
);
3144 ret
= lttng_ht_del(channel_ht
, &iter
);
3148 * This will close the wait fd for each stream associated to
3149 * this channel AND monitored by the data/metadata thread thus
3150 * will be clean by the right thread.
3152 consumer_close_channel_streams(chan
);
3154 /* Release our own refcount */
3155 if (!uatomic_sub_return(&chan
->refcount
, 1)
3156 && !uatomic_read(&chan
->nb_init_stream_left
)) {
3157 consumer_del_channel(chan
);
3160 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
3165 /* Release RCU lock for the channel looked up */
3173 lttng_poll_clean(&events
);
3175 destroy_channel_ht(channel_ht
);
3178 DBG("Channel poll thread exiting");
3181 ERR("Health error occurred in %s", __func__
);
3183 health_unregister(health_consumerd
);
3184 rcu_unregister_thread();
3188 static int set_metadata_socket(struct lttng_consumer_local_data
*ctx
,
3189 struct pollfd
*sockpoll
, int client_socket
)
3196 ret
= lttng_consumer_poll_socket(sockpoll
);
3200 DBG("Metadata connection on client_socket");
3202 /* Blocking call, waiting for transmission */
3203 ctx
->consumer_metadata_socket
= lttcomm_accept_unix_sock(client_socket
);
3204 if (ctx
->consumer_metadata_socket
< 0) {
3205 WARN("On accept metadata");
3216 * This thread listens on the consumerd socket and receives the file
3217 * descriptors from the session daemon.
3219 void *consumer_thread_sessiond_poll(void *data
)
3221 int sock
= -1, client_socket
, ret
, err
= -1;
3223 * structure to poll for incoming data on communication socket avoids
3224 * making blocking sockets.
3226 struct pollfd consumer_sockpoll
[2];
3227 struct lttng_consumer_local_data
*ctx
= data
;
3229 rcu_register_thread();
3231 health_register(health_consumerd
, HEALTH_CONSUMERD_TYPE_SESSIOND
);
3233 if (testpoint(consumerd_thread_sessiond
)) {
3234 goto error_testpoint
;
3237 health_code_update();
3239 DBG("Creating command socket %s", ctx
->consumer_command_sock_path
);
3240 unlink(ctx
->consumer_command_sock_path
);
3241 client_socket
= lttcomm_create_unix_sock(ctx
->consumer_command_sock_path
);
3242 if (client_socket
< 0) {
3243 ERR("Cannot create command socket");
3247 ret
= lttcomm_listen_unix_sock(client_socket
);
3252 DBG("Sending ready command to lttng-sessiond");
3253 ret
= lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_COMMAND_SOCK_READY
);
3254 /* return < 0 on error, but == 0 is not fatal */
3256 ERR("Error sending ready command to lttng-sessiond");
3260 /* prepare the FDs to poll : to client socket and the should_quit pipe */
3261 consumer_sockpoll
[0].fd
= ctx
->consumer_should_quit
[0];
3262 consumer_sockpoll
[0].events
= POLLIN
| POLLPRI
;
3263 consumer_sockpoll
[1].fd
= client_socket
;
3264 consumer_sockpoll
[1].events
= POLLIN
| POLLPRI
;
3266 ret
= lttng_consumer_poll_socket(consumer_sockpoll
);
3274 DBG("Connection on client_socket");
3276 /* Blocking call, waiting for transmission */
3277 sock
= lttcomm_accept_unix_sock(client_socket
);
3284 * Setup metadata socket which is the second socket connection on the
3285 * command unix socket.
3287 ret
= set_metadata_socket(ctx
, consumer_sockpoll
, client_socket
);
3296 /* This socket is not useful anymore. */
3297 ret
= close(client_socket
);
3299 PERROR("close client_socket");
3303 /* update the polling structure to poll on the established socket */
3304 consumer_sockpoll
[1].fd
= sock
;
3305 consumer_sockpoll
[1].events
= POLLIN
| POLLPRI
;
3308 health_code_update();
3310 health_poll_entry();
3311 ret
= lttng_consumer_poll_socket(consumer_sockpoll
);
3320 DBG("Incoming command on sock");
3321 ret
= lttng_consumer_recv_cmd(ctx
, sock
, consumer_sockpoll
);
3324 * This could simply be a session daemon quitting. Don't output
3327 DBG("Communication interrupted on command socket");
3331 if (CMM_LOAD_SHARED(consumer_quit
)) {
3332 DBG("consumer_thread_receive_fds received quit from signal");
3333 err
= 0; /* All is OK */
3336 DBG("received command on sock");
3342 DBG("Consumer thread sessiond poll exiting");
3345 * Close metadata streams since the producer is the session daemon which
3348 * NOTE: for now, this only applies to the UST tracer.
3350 lttng_consumer_close_all_metadata();
3353 * when all fds have hung up, the polling thread
3356 CMM_STORE_SHARED(consumer_quit
, 1);
3359 * Notify the data poll thread to poll back again and test the
3360 * consumer_quit state that we just set so to quit gracefully.
3362 notify_thread_lttng_pipe(ctx
->consumer_data_pipe
);
3364 notify_channel_pipe(ctx
, NULL
, -1, CONSUMER_CHANNEL_QUIT
);
3366 notify_health_quit_pipe(health_quit_pipe
);
3368 /* Cleaning up possibly open sockets. */
3372 PERROR("close sock sessiond poll");
3375 if (client_socket
>= 0) {
3376 ret
= close(client_socket
);
3378 PERROR("close client_socket sessiond poll");
3385 ERR("Health error occurred in %s", __func__
);
3387 health_unregister(health_consumerd
);
3389 rcu_unregister_thread();
3393 ssize_t
lttng_consumer_read_subbuffer(struct lttng_consumer_stream
*stream
,
3394 struct lttng_consumer_local_data
*ctx
)
3398 bool rotated
= false;
3400 pthread_mutex_lock(&stream
->lock
);
3401 if (stream
->metadata_flag
) {
3402 pthread_mutex_lock(&stream
->metadata_rdv_lock
);
3405 switch (consumer_data
.type
) {
3406 case LTTNG_CONSUMER_KERNEL
:
3407 ret
= lttng_kconsumer_read_subbuffer(stream
, ctx
, &rotated
);
3409 case LTTNG_CONSUMER32_UST
:
3410 case LTTNG_CONSUMER64_UST
:
3411 ret
= lttng_ustconsumer_read_subbuffer(stream
, ctx
, &rotated
);
3414 ERR("Unknown consumer_data type");
3420 if (stream
->metadata_flag
) {
3421 pthread_cond_broadcast(&stream
->metadata_rdv
);
3422 pthread_mutex_unlock(&stream
->metadata_rdv_lock
);
3424 pthread_mutex_unlock(&stream
->lock
);
3426 rotate_ret
= consumer_post_rotation(stream
, ctx
);
3427 if (rotate_ret
< 0) {
3428 ERR("Failed after a rotation");
3436 int lttng_consumer_on_recv_stream(struct lttng_consumer_stream
*stream
)
3438 switch (consumer_data
.type
) {
3439 case LTTNG_CONSUMER_KERNEL
:
3440 return lttng_kconsumer_on_recv_stream(stream
);
3441 case LTTNG_CONSUMER32_UST
:
3442 case LTTNG_CONSUMER64_UST
:
3443 return lttng_ustconsumer_on_recv_stream(stream
);
3445 ERR("Unknown consumer_data type");
3452 * Allocate and set consumer data hash tables.
3454 int lttng_consumer_init(void)
3456 consumer_data
.channel_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3457 if (!consumer_data
.channel_ht
) {
3461 consumer_data
.relayd_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3462 if (!consumer_data
.relayd_ht
) {
3466 consumer_data
.stream_list_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3467 if (!consumer_data
.stream_list_ht
) {
3471 consumer_data
.stream_per_chan_id_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3472 if (!consumer_data
.stream_per_chan_id_ht
) {
3476 data_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3481 metadata_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3493 * Process the ADD_RELAYD command receive by a consumer.
3495 * This will create a relayd socket pair and add it to the relayd hash table.
3496 * The caller MUST acquire a RCU read side lock before calling it.
3498 void consumer_add_relayd_socket(uint64_t net_seq_idx
, int sock_type
,
3499 struct lttng_consumer_local_data
*ctx
, int sock
,
3500 struct pollfd
*consumer_sockpoll
,
3501 struct lttcomm_relayd_sock
*relayd_sock
, uint64_t sessiond_id
,
3502 uint64_t relayd_session_id
)
3504 int fd
= -1, ret
= -1, relayd_created
= 0;
3505 enum lttcomm_return_code ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
3506 struct consumer_relayd_sock_pair
*relayd
= NULL
;
3509 assert(relayd_sock
);
3511 DBG("Consumer adding relayd socket (idx: %" PRIu64
")", net_seq_idx
);
3513 /* Get relayd reference if exists. */
3514 relayd
= consumer_find_relayd(net_seq_idx
);
3515 if (relayd
== NULL
) {
3516 assert(sock_type
== LTTNG_STREAM_CONTROL
);
3517 /* Not found. Allocate one. */
3518 relayd
= consumer_allocate_relayd_sock_pair(net_seq_idx
);
3519 if (relayd
== NULL
) {
3520 ret_code
= LTTCOMM_CONSUMERD_ENOMEM
;
3523 relayd
->sessiond_session_id
= sessiond_id
;
3528 * This code path MUST continue to the consumer send status message to
3529 * we can notify the session daemon and continue our work without
3530 * killing everything.
3534 * relayd key should never be found for control socket.
3536 assert(sock_type
!= LTTNG_STREAM_CONTROL
);
3539 /* First send a status message before receiving the fds. */
3540 ret
= consumer_send_status_msg(sock
, LTTCOMM_CONSUMERD_SUCCESS
);
3542 /* Somehow, the session daemon is not responding anymore. */
3543 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_FATAL
);
3544 goto error_nosignal
;
3547 /* Poll on consumer socket. */
3548 ret
= lttng_consumer_poll_socket(consumer_sockpoll
);
3550 /* Needing to exit in the middle of a command: error. */
3551 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_POLL_ERROR
);
3552 goto error_nosignal
;
3555 /* Get relayd socket from session daemon */
3556 ret
= lttcomm_recv_fds_unix_sock(sock
, &fd
, 1);
3557 if (ret
!= sizeof(fd
)) {
3558 fd
= -1; /* Just in case it gets set with an invalid value. */
3561 * Failing to receive FDs might indicate a major problem such as
3562 * reaching a fd limit during the receive where the kernel returns a
3563 * MSG_CTRUNC and fails to cleanup the fd in the queue. Any case, we
3564 * don't take any chances and stop everything.
3566 * XXX: Feature request #558 will fix that and avoid this possible
3567 * issue when reaching the fd limit.
3569 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_ERROR_RECV_FD
);
3570 ret_code
= LTTCOMM_CONSUMERD_ERROR_RECV_FD
;
3574 /* Copy socket information and received FD */
3575 switch (sock_type
) {
3576 case LTTNG_STREAM_CONTROL
:
3577 /* Copy received lttcomm socket */
3578 lttcomm_copy_sock(&relayd
->control_sock
.sock
, &relayd_sock
->sock
);
3579 ret
= lttcomm_create_sock(&relayd
->control_sock
.sock
);
3580 /* Handle create_sock error. */
3582 ret_code
= LTTCOMM_CONSUMERD_ENOMEM
;
3586 * Close the socket created internally by
3587 * lttcomm_create_sock, so we can replace it by the one
3588 * received from sessiond.
3590 if (close(relayd
->control_sock
.sock
.fd
)) {
3594 /* Assign new file descriptor */
3595 relayd
->control_sock
.sock
.fd
= fd
;
3596 fd
= -1; /* For error path */
3597 /* Assign version values. */
3598 relayd
->control_sock
.major
= relayd_sock
->major
;
3599 relayd
->control_sock
.minor
= relayd_sock
->minor
;
3601 relayd
->relayd_session_id
= relayd_session_id
;
3604 case LTTNG_STREAM_DATA
:
3605 /* Copy received lttcomm socket */
3606 lttcomm_copy_sock(&relayd
->data_sock
.sock
, &relayd_sock
->sock
);
3607 ret
= lttcomm_create_sock(&relayd
->data_sock
.sock
);
3608 /* Handle create_sock error. */
3610 ret_code
= LTTCOMM_CONSUMERD_ENOMEM
;
3614 * Close the socket created internally by
3615 * lttcomm_create_sock, so we can replace it by the one
3616 * received from sessiond.
3618 if (close(relayd
->data_sock
.sock
.fd
)) {
3622 /* Assign new file descriptor */
3623 relayd
->data_sock
.sock
.fd
= fd
;
3624 fd
= -1; /* for eventual error paths */
3625 /* Assign version values. */
3626 relayd
->data_sock
.major
= relayd_sock
->major
;
3627 relayd
->data_sock
.minor
= relayd_sock
->minor
;
3630 ERR("Unknown relayd socket type (%d)", sock_type
);
3631 ret_code
= LTTCOMM_CONSUMERD_FATAL
;
3635 DBG("Consumer %s socket created successfully with net idx %" PRIu64
" (fd: %d)",
3636 sock_type
== LTTNG_STREAM_CONTROL
? "control" : "data",
3637 relayd
->net_seq_idx
, fd
);
3639 /* We successfully added the socket. Send status back. */
3640 ret
= consumer_send_status_msg(sock
, ret_code
);
3642 /* Somehow, the session daemon is not responding anymore. */
3643 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_FATAL
);
3644 goto error_nosignal
;
3648 * Add relayd socket pair to consumer data hashtable. If object already
3649 * exists or on error, the function gracefully returns.
3657 if (consumer_send_status_msg(sock
, ret_code
) < 0) {
3658 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_FATAL
);
3662 /* Close received socket if valid. */
3665 PERROR("close received socket");
3669 if (relayd_created
) {
3675 * Try to lock the stream mutex.
3677 * On success, 1 is returned else 0 indicating that the mutex is NOT lock.
3679 static int stream_try_lock(struct lttng_consumer_stream
*stream
)
3686 * Try to lock the stream mutex. On failure, we know that the stream is
3687 * being used else where hence there is data still being extracted.
3689 ret
= pthread_mutex_trylock(&stream
->lock
);
3691 /* For both EBUSY and EINVAL error, the mutex is NOT locked. */
3703 * Search for a relayd associated to the session id and return the reference.
3705 * A rcu read side lock MUST be acquire before calling this function and locked
3706 * until the relayd object is no longer necessary.
3708 static struct consumer_relayd_sock_pair
*find_relayd_by_session_id(uint64_t id
)
3710 struct lttng_ht_iter iter
;
3711 struct consumer_relayd_sock_pair
*relayd
= NULL
;
3713 /* Iterate over all relayd since they are indexed by net_seq_idx. */
3714 cds_lfht_for_each_entry(consumer_data
.relayd_ht
->ht
, &iter
.iter
, relayd
,
3717 * Check by sessiond id which is unique here where the relayd session
3718 * id might not be when having multiple relayd.
3720 if (relayd
->sessiond_session_id
== id
) {
3721 /* Found the relayd. There can be only one per id. */
3733 * Check if for a given session id there is still data needed to be extract
3736 * Return 1 if data is pending or else 0 meaning ready to be read.
3738 int consumer_data_pending(uint64_t id
)
3741 struct lttng_ht_iter iter
;
3742 struct lttng_ht
*ht
;
3743 struct lttng_consumer_stream
*stream
;
3744 struct consumer_relayd_sock_pair
*relayd
= NULL
;
3745 int (*data_pending
)(struct lttng_consumer_stream
*);
3747 DBG("Consumer data pending command on session id %" PRIu64
, id
);
3750 pthread_mutex_lock(&consumer_data
.lock
);
3752 switch (consumer_data
.type
) {
3753 case LTTNG_CONSUMER_KERNEL
:
3754 data_pending
= lttng_kconsumer_data_pending
;
3756 case LTTNG_CONSUMER32_UST
:
3757 case LTTNG_CONSUMER64_UST
:
3758 data_pending
= lttng_ustconsumer_data_pending
;
3761 ERR("Unknown consumer data type");
3765 /* Ease our life a bit */
3766 ht
= consumer_data
.stream_list_ht
;
3768 relayd
= find_relayd_by_session_id(id
);
3770 /* Send init command for data pending. */
3771 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
3772 ret
= relayd_begin_data_pending(&relayd
->control_sock
,
3773 relayd
->relayd_session_id
);
3774 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
3776 /* Communication error thus the relayd so no data pending. */
3777 goto data_not_pending
;
3781 cds_lfht_for_each_entry_duplicate(ht
->ht
,
3782 ht
->hash_fct(&id
, lttng_ht_seed
),
3784 &iter
.iter
, stream
, node_session_id
.node
) {
3785 /* If this call fails, the stream is being used hence data pending. */
3786 ret
= stream_try_lock(stream
);
3792 * A removed node from the hash table indicates that the stream has
3793 * been deleted thus having a guarantee that the buffers are closed
3794 * on the consumer side. However, data can still be transmitted
3795 * over the network so don't skip the relayd check.
3797 ret
= cds_lfht_is_node_deleted(&stream
->node
.node
);
3799 /* Check the stream if there is data in the buffers. */
3800 ret
= data_pending(stream
);
3802 pthread_mutex_unlock(&stream
->lock
);
3809 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
3810 if (stream
->metadata_flag
) {
3811 ret
= relayd_quiescent_control(&relayd
->control_sock
,
3812 stream
->relayd_stream_id
);
3814 ret
= relayd_data_pending(&relayd
->control_sock
,
3815 stream
->relayd_stream_id
,
3816 stream
->next_net_seq_num
- 1);
3818 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
3820 pthread_mutex_unlock(&stream
->lock
);
3824 pthread_mutex_unlock(&stream
->lock
);
3828 unsigned int is_data_inflight
= 0;
3830 /* Send init command for data pending. */
3831 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
3832 ret
= relayd_end_data_pending(&relayd
->control_sock
,
3833 relayd
->relayd_session_id
, &is_data_inflight
);
3834 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
3836 goto data_not_pending
;
3838 if (is_data_inflight
) {
3844 * Finding _no_ node in the hash table and no inflight data means that the
3845 * stream(s) have been removed thus data is guaranteed to be available for
3846 * analysis from the trace files.
3850 /* Data is available to be read by a viewer. */
3851 pthread_mutex_unlock(&consumer_data
.lock
);
3856 /* Data is still being extracted from buffers. */
3857 pthread_mutex_unlock(&consumer_data
.lock
);
3863 * Send a ret code status message to the sessiond daemon.
3865 * Return the sendmsg() return value.
3867 int consumer_send_status_msg(int sock
, int ret_code
)
3869 struct lttcomm_consumer_status_msg msg
;
3871 memset(&msg
, 0, sizeof(msg
));
3872 msg
.ret_code
= ret_code
;
3874 return lttcomm_send_unix_sock(sock
, &msg
, sizeof(msg
));
3878 * Send a channel status message to the sessiond daemon.
3880 * Return the sendmsg() return value.
3882 int consumer_send_status_channel(int sock
,
3883 struct lttng_consumer_channel
*channel
)
3885 struct lttcomm_consumer_status_channel msg
;
3889 memset(&msg
, 0, sizeof(msg
));
3891 msg
.ret_code
= LTTCOMM_CONSUMERD_CHANNEL_FAIL
;
3893 msg
.ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
3894 msg
.key
= channel
->key
;
3895 msg
.stream_count
= channel
->streams
.count
;
3898 return lttcomm_send_unix_sock(sock
, &msg
, sizeof(msg
));
3901 unsigned long consumer_get_consume_start_pos(unsigned long consumed_pos
,
3902 unsigned long produced_pos
, uint64_t nb_packets_per_stream
,
3903 uint64_t max_sb_size
)
3905 unsigned long start_pos
;
3907 if (!nb_packets_per_stream
) {
3908 return consumed_pos
; /* Grab everything */
3910 start_pos
= produced_pos
- offset_align_floor(produced_pos
, max_sb_size
);
3911 start_pos
-= max_sb_size
* nb_packets_per_stream
;
3912 if ((long) (start_pos
- consumed_pos
) < 0) {
3913 return consumed_pos
; /* Grab everything */
3919 int consumer_flush_buffer(struct lttng_consumer_stream
*stream
, int producer_active
)
3923 switch (consumer_data
.type
) {
3924 case LTTNG_CONSUMER_KERNEL
:
3925 ret
= kernctl_buffer_flush(stream
->wait_fd
);
3927 ERR("Failed to flush kernel stream");
3931 case LTTNG_CONSUMER32_UST
:
3932 case LTTNG_CONSUMER64_UST
:
3933 lttng_ustctl_flush_buffer(stream
, producer_active
);
3936 ERR("Unknown consumer_data type");
3945 * Sample the rotate position for all the streams of a channel. If a stream
3946 * is already at the rotate position (produced == consumed), we flag it as
3947 * ready for rotation. The rotation of ready streams occurs after we have
3948 * replied to the session daemon that we have finished sampling the positions.
3950 * Returns 0 on success, < 0 on error
3952 int lttng_consumer_rotate_channel(uint64_t key
, const char *path
,
3953 uint64_t relayd_id
, uint32_t metadata
, uint64_t new_chunk_id
,
3954 struct lttng_consumer_local_data
*ctx
)
3957 struct lttng_consumer_channel
*channel
;
3958 struct lttng_consumer_stream
*stream
;
3959 struct lttng_ht_iter iter
;
3960 struct lttng_ht
*ht
= consumer_data
.stream_per_chan_id_ht
;
3962 DBG("Consumer sample rotate position for channel %" PRIu64
, key
);
3966 channel
= consumer_find_channel(key
);
3968 ERR("No channel found for key %" PRIu64
, key
);
3973 pthread_mutex_lock(&channel
->lock
);
3974 channel
->current_chunk_id
= new_chunk_id
;
3976 ret
= lttng_strncpy(channel
->pathname
, path
, sizeof(channel
->pathname
));
3978 ERR("Failed to copy new path to channel during channel rotation");
3980 goto end_unlock_channel
;
3983 if (relayd_id
== -1ULL) {
3985 * The domain path (/ust or /kernel) has been created before, we
3986 * now need to create the last part of the path: the application/user
3987 * specific section (uid/1000/64-bit).
3989 ret
= utils_mkdir_recursive(channel
->pathname
, S_IRWXU
| S_IRWXG
,
3990 channel
->uid
, channel
->gid
);
3992 ERR("Failed to create trace directory at %s during rotation",
3995 goto end_unlock_channel
;
3999 cds_lfht_for_each_entry_duplicate(ht
->ht
,
4000 ht
->hash_fct(&channel
->key
, lttng_ht_seed
),
4001 ht
->match_fct
, &channel
->key
, &iter
.iter
,
4002 stream
, node_channel_id
.node
) {
4003 unsigned long consumed_pos
;
4005 health_code_update();
4008 * Lock stream because we are about to change its state.
4010 pthread_mutex_lock(&stream
->lock
);
4012 ret
= lttng_strncpy(stream
->channel_read_only_attributes
.path
,
4014 sizeof(stream
->channel_read_only_attributes
.path
));
4016 ERR("Failed to sample channel path name during channel rotation");
4017 goto end_unlock_stream
;
4019 ret
= lttng_consumer_sample_snapshot_positions(stream
);
4021 ERR("Failed to sample snapshot position during channel rotation");
4022 goto end_unlock_stream
;
4025 ret
= lttng_consumer_get_produced_snapshot(stream
,
4026 &stream
->rotate_position
);
4028 ERR("Failed to sample produced position during channel rotation");
4029 goto end_unlock_stream
;
4032 lttng_consumer_get_consumed_snapshot(stream
,
4034 if (consumed_pos
== stream
->rotate_position
) {
4035 stream
->rotate_ready
= true;
4037 channel
->nr_stream_rotate_pending
++;
4039 ret
= consumer_flush_buffer(stream
, 1);
4041 ERR("Failed to flush stream %" PRIu64
" during channel rotation",
4043 goto end_unlock_stream
;
4046 pthread_mutex_unlock(&stream
->lock
);
4048 pthread_mutex_unlock(&channel
->lock
);
4054 pthread_mutex_unlock(&stream
->lock
);
4056 pthread_mutex_unlock(&channel
->lock
);
4063 * Check if a stream is ready to be rotated after extracting it.
4065 * Return 1 if it is ready for rotation, 0 if it is not, a negative value on
4066 * error. Stream lock must be held.
4068 int lttng_consumer_stream_is_rotate_ready(struct lttng_consumer_stream
*stream
)
4071 unsigned long consumed_pos
;
4073 if (!stream
->rotate_position
&& !stream
->rotate_ready
) {
4078 if (stream
->rotate_ready
) {
4084 * If we don't have the rotate_ready flag, check the consumed position
4085 * to determine if we need to rotate.
4087 ret
= lttng_consumer_sample_snapshot_positions(stream
);
4089 ERR("Taking snapshot positions");
4093 ret
= lttng_consumer_get_consumed_snapshot(stream
, &consumed_pos
);
4095 ERR("Consumed snapshot position");
4099 /* Rotate position not reached yet (with check for overflow). */
4100 if ((long) (consumed_pos
- stream
->rotate_position
) < 0) {
4111 * Reset the state for a stream after a rotation occurred.
4113 void lttng_consumer_reset_stream_rotate_state(struct lttng_consumer_stream
*stream
)
4115 stream
->rotate_position
= 0;
4116 stream
->rotate_ready
= false;
4120 * Perform the rotation a local stream file.
4122 int rotate_local_stream(struct lttng_consumer_local_data
*ctx
,
4123 struct lttng_consumer_stream
*stream
)
4127 DBG("Rotate local stream: stream key %" PRIu64
", channel key %" PRIu64
" at path %s",
4130 stream
->channel_read_only_attributes
.path
);
4132 ret
= close(stream
->out_fd
);
4134 PERROR("Closing trace file (fd %d), stream %" PRIu64
,
4135 stream
->out_fd
, stream
->key
);
4140 ret
= utils_create_stream_file(
4141 stream
->channel_read_only_attributes
.path
,
4143 stream
->channel_read_only_attributes
.tracefile_size
,
4144 stream
->tracefile_count_current
,
4145 stream
->uid
, stream
->gid
, NULL
);
4147 ERR("Rotate create stream file");
4150 stream
->out_fd
= ret
;
4151 stream
->tracefile_size_current
= 0;
4153 if (!stream
->metadata_flag
) {
4154 struct lttng_index_file
*index_file
;
4156 lttng_index_file_put(stream
->index_file
);
4158 index_file
= lttng_index_file_create(
4159 stream
->channel_read_only_attributes
.path
,
4160 stream
->name
, stream
->uid
, stream
->gid
,
4161 stream
->channel_read_only_attributes
.tracefile_size
,
4162 stream
->tracefile_count_current
,
4163 CTF_INDEX_MAJOR
, CTF_INDEX_MINOR
);
4165 ERR("Create index file during rotation");
4168 stream
->index_file
= index_file
;
4169 stream
->out_fd_offset
= 0;
4183 * Perform the rotation a stream file on the relay.
4185 int rotate_relay_stream(struct lttng_consumer_local_data
*ctx
,
4186 struct lttng_consumer_stream
*stream
)
4189 struct consumer_relayd_sock_pair
*relayd
;
4191 DBG("Rotate relay stream");
4192 relayd
= consumer_find_relayd(stream
->net_seq_idx
);
4194 ERR("Failed to find relayd");
4199 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
4200 ret
= relayd_rotate_stream(&relayd
->control_sock
,
4201 stream
->relayd_stream_id
,
4202 stream
->channel_read_only_attributes
.path
,
4203 stream
->chan
->current_chunk_id
,
4204 stream
->last_sequence_number
);
4205 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
4207 ERR("Rotate relay stream");
4215 * Performs the stream rotation for the rotate session feature if needed.
4216 * It must be called with the stream lock held.
4218 * Return 0 on success, a negative number of error.
4220 int lttng_consumer_rotate_stream(struct lttng_consumer_local_data
*ctx
,
4221 struct lttng_consumer_stream
*stream
, bool *rotated
)
4225 DBG("Consumer rotate stream %" PRIu64
, stream
->key
);
4227 if (stream
->net_seq_idx
!= (uint64_t) -1ULL) {
4228 ret
= rotate_relay_stream(ctx
, stream
);
4230 ret
= rotate_local_stream(ctx
, stream
);
4233 ERR("Rotate stream");
4237 if (stream
->metadata_flag
) {
4238 switch (consumer_data
.type
) {
4239 case LTTNG_CONSUMER_KERNEL
:
4241 * Reset the position of what has been read from the metadata
4242 * cache to 0 so we can dump it again.
4244 ret
= kernctl_metadata_cache_dump(stream
->wait_fd
);
4246 ERR("Failed to dump the kernel metadata cache after rotation");
4250 case LTTNG_CONSUMER32_UST
:
4251 case LTTNG_CONSUMER64_UST
:
4253 * Reset the position pushed from the metadata cache so it
4254 * will write from the beginning on the next push.
4256 stream
->ust_metadata_pushed
= 0;
4259 ERR("Unknown consumer_data type");
4263 lttng_consumer_reset_stream_rotate_state(stream
);
4276 * Rotate all the ready streams now.
4278 * This is especially important for low throughput streams that have already
4279 * been consumed, we cannot wait for their next packet to perform the
4282 * Returns 0 on success, < 0 on error
4284 int lttng_consumer_rotate_ready_streams(uint64_t key
,
4285 struct lttng_consumer_local_data
*ctx
)
4288 struct lttng_consumer_channel
*channel
;
4289 struct lttng_consumer_stream
*stream
;
4290 struct lttng_ht_iter iter
;
4291 struct lttng_ht
*ht
= consumer_data
.stream_per_chan_id_ht
;
4295 DBG("Consumer rotate ready streams in channel %" PRIu64
, key
);
4297 channel
= consumer_find_channel(key
);
4299 ERR("No channel found for key %" PRIu64
, key
);
4304 cds_lfht_for_each_entry_duplicate(ht
->ht
,
4305 ht
->hash_fct(&channel
->key
, lttng_ht_seed
),
4306 ht
->match_fct
, &channel
->key
, &iter
.iter
,
4307 stream
, node_channel_id
.node
) {
4308 health_code_update();
4310 pthread_mutex_lock(&stream
->lock
);
4312 if (!stream
->rotate_ready
) {
4313 pthread_mutex_unlock(&stream
->lock
);
4316 DBG("Consumer rotate ready stream %" PRIu64
, stream
->key
);
4318 ret
= lttng_consumer_rotate_stream(ctx
, stream
, NULL
);
4319 pthread_mutex_unlock(&stream
->lock
);
4324 ret
= consumer_post_rotation(stream
, ctx
);
4338 int rotate_rename_local(const char *old_path
, const char *new_path
,
4339 uid_t uid
, gid_t gid
)
4346 ret
= utils_mkdir_recursive(new_path
, S_IRWXU
| S_IRWXG
, uid
, gid
);
4348 ERR("Create directory on rotate");
4352 ret
= rename(old_path
, new_path
);
4353 if (ret
< 0 && errno
!= ENOENT
) {
4354 PERROR("Rename completed rotation chunk");
4364 int rotate_rename_relay(const char *old_path
, const char *new_path
,
4368 struct consumer_relayd_sock_pair
*relayd
;
4370 relayd
= consumer_find_relayd(relayd_id
);
4372 ERR("Failed to find relayd while running rotate_rename_relay command");
4377 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
4378 ret
= relayd_rotate_rename(&relayd
->control_sock
, old_path
, new_path
);
4379 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
4384 int lttng_consumer_rotate_rename(const char *old_path
, const char *new_path
,
4385 uid_t uid
, gid_t gid
, uint64_t relayd_id
)
4387 if (relayd_id
!= -1ULL) {
4388 return rotate_rename_relay(old_path
, new_path
, relayd_id
);
4390 return rotate_rename_local(old_path
, new_path
, uid
, gid
);
4394 int lttng_consumer_rotate_pending_relay(uint64_t session_id
,
4395 uint64_t relayd_id
, uint64_t chunk_id
)
4398 struct consumer_relayd_sock_pair
*relayd
;
4400 relayd
= consumer_find_relayd(relayd_id
);
4402 ERR("Failed to find relayd");
4407 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
4408 ret
= relayd_rotate_pending(&relayd
->control_sock
, chunk_id
);
4409 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
4416 int mkdir_local(const char *path
, uid_t uid
, gid_t gid
)
4420 ret
= utils_mkdir_recursive(path
, S_IRWXU
| S_IRWXG
, uid
, gid
);
4422 /* utils_mkdir_recursive logs an error. */
4432 int mkdir_relay(const char *path
, uint64_t relayd_id
)
4435 struct consumer_relayd_sock_pair
*relayd
;
4437 relayd
= consumer_find_relayd(relayd_id
);
4439 ERR("Failed to find relayd");
4444 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
4445 ret
= relayd_mkdir(&relayd
->control_sock
, path
);
4446 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
4453 int lttng_consumer_mkdir(const char *path
, uid_t uid
, gid_t gid
,
4456 if (relayd_id
!= -1ULL) {
4457 return mkdir_relay(path
, relayd_id
);
4459 return mkdir_local(path
, uid
, gid
);