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 stream
->trace_archive_id
);
813 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
818 uatomic_inc(&relayd
->refcount
);
819 stream
->sent_to_relayd
= 1;
821 ERR("Stream %" PRIu64
" relayd ID %" PRIu64
" unknown. Can't send it.",
822 stream
->key
, stream
->net_seq_idx
);
827 DBG("Stream %s with key %" PRIu64
" sent to relayd id %" PRIu64
,
828 stream
->name
, stream
->key
, stream
->net_seq_idx
);
836 * Find a relayd and send the streams sent message
838 * Returns 0 on success, < 0 on error
840 int consumer_send_relayd_streams_sent(uint64_t net_seq_idx
)
843 struct consumer_relayd_sock_pair
*relayd
;
845 assert(net_seq_idx
!= -1ULL);
847 /* The stream is not metadata. Get relayd reference if exists. */
849 relayd
= consumer_find_relayd(net_seq_idx
);
850 if (relayd
!= NULL
) {
851 /* Add stream on the relayd */
852 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
853 ret
= relayd_streams_sent(&relayd
->control_sock
);
854 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
859 ERR("Relayd ID %" PRIu64
" unknown. Can't send streams_sent.",
866 DBG("All streams sent relayd id %" PRIu64
, net_seq_idx
);
874 * Find a relayd and close the stream
876 void close_relayd_stream(struct lttng_consumer_stream
*stream
)
878 struct consumer_relayd_sock_pair
*relayd
;
880 /* The stream is not metadata. Get relayd reference if exists. */
882 relayd
= consumer_find_relayd(stream
->net_seq_idx
);
884 consumer_stream_relayd_close(stream
, relayd
);
890 * Handle stream for relayd transmission if the stream applies for network
891 * streaming where the net sequence index is set.
893 * Return destination file descriptor or negative value on error.
895 static int write_relayd_stream_header(struct lttng_consumer_stream
*stream
,
896 size_t data_size
, unsigned long padding
,
897 struct consumer_relayd_sock_pair
*relayd
)
900 struct lttcomm_relayd_data_hdr data_hdr
;
906 /* Reset data header */
907 memset(&data_hdr
, 0, sizeof(data_hdr
));
909 if (stream
->metadata_flag
) {
910 /* Caller MUST acquire the relayd control socket lock */
911 ret
= relayd_send_metadata(&relayd
->control_sock
, data_size
);
916 /* Metadata are always sent on the control socket. */
917 outfd
= relayd
->control_sock
.sock
.fd
;
919 /* Set header with stream information */
920 data_hdr
.stream_id
= htobe64(stream
->relayd_stream_id
);
921 data_hdr
.data_size
= htobe32(data_size
);
922 data_hdr
.padding_size
= htobe32(padding
);
924 * Note that net_seq_num below is assigned with the *current* value of
925 * next_net_seq_num and only after that the next_net_seq_num will be
926 * increment. This is why when issuing a command on the relayd using
927 * this next value, 1 should always be substracted in order to compare
928 * the last seen sequence number on the relayd side to the last sent.
930 data_hdr
.net_seq_num
= htobe64(stream
->next_net_seq_num
);
931 /* Other fields are zeroed previously */
933 ret
= relayd_send_data_hdr(&relayd
->data_sock
, &data_hdr
,
939 ++stream
->next_net_seq_num
;
941 /* Set to go on data socket */
942 outfd
= relayd
->data_sock
.sock
.fd
;
950 * Allocate and return a new lttng_consumer_channel object using the given key
951 * to initialize the hash table node.
953 * On error, return NULL.
955 struct lttng_consumer_channel
*consumer_allocate_channel(uint64_t key
,
957 const char *pathname
,
962 enum lttng_event_output output
,
963 uint64_t tracefile_size
,
964 uint64_t tracefile_count
,
965 uint64_t session_id_per_pid
,
966 unsigned int monitor
,
967 unsigned int live_timer_interval
,
968 const char *root_shm_path
,
969 const char *shm_path
)
971 struct lttng_consumer_channel
*channel
;
973 channel
= zmalloc(sizeof(*channel
));
974 if (channel
== NULL
) {
975 PERROR("malloc struct lttng_consumer_channel");
980 channel
->refcount
= 0;
981 channel
->session_id
= session_id
;
982 channel
->session_id_per_pid
= session_id_per_pid
;
985 channel
->relayd_id
= relayd_id
;
986 channel
->tracefile_size
= tracefile_size
;
987 channel
->tracefile_count
= tracefile_count
;
988 channel
->monitor
= monitor
;
989 channel
->live_timer_interval
= live_timer_interval
;
990 pthread_mutex_init(&channel
->lock
, NULL
);
991 pthread_mutex_init(&channel
->timer_lock
, NULL
);
994 case LTTNG_EVENT_SPLICE
:
995 channel
->output
= CONSUMER_CHANNEL_SPLICE
;
997 case LTTNG_EVENT_MMAP
:
998 channel
->output
= CONSUMER_CHANNEL_MMAP
;
1008 * In monitor mode, the streams associated with the channel will be put in
1009 * a special list ONLY owned by this channel. So, the refcount is set to 1
1010 * here meaning that the channel itself has streams that are referenced.
1012 * On a channel deletion, once the channel is no longer visible, the
1013 * refcount is decremented and checked for a zero value to delete it. With
1014 * streams in no monitor mode, it will now be safe to destroy the channel.
1016 if (!channel
->monitor
) {
1017 channel
->refcount
= 1;
1020 strncpy(channel
->pathname
, pathname
, sizeof(channel
->pathname
));
1021 channel
->pathname
[sizeof(channel
->pathname
) - 1] = '\0';
1023 strncpy(channel
->name
, name
, sizeof(channel
->name
));
1024 channel
->name
[sizeof(channel
->name
) - 1] = '\0';
1026 if (root_shm_path
) {
1027 strncpy(channel
->root_shm_path
, root_shm_path
, sizeof(channel
->root_shm_path
));
1028 channel
->root_shm_path
[sizeof(channel
->root_shm_path
) - 1] = '\0';
1031 strncpy(channel
->shm_path
, shm_path
, sizeof(channel
->shm_path
));
1032 channel
->shm_path
[sizeof(channel
->shm_path
) - 1] = '\0';
1035 lttng_ht_node_init_u64(&channel
->node
, channel
->key
);
1037 channel
->wait_fd
= -1;
1039 CDS_INIT_LIST_HEAD(&channel
->streams
.head
);
1041 DBG("Allocated channel (key %" PRIu64
")", channel
->key
);
1048 * Add a channel to the global list protected by a mutex.
1050 * Always return 0 indicating success.
1052 int consumer_add_channel(struct lttng_consumer_channel
*channel
,
1053 struct lttng_consumer_local_data
*ctx
)
1055 pthread_mutex_lock(&consumer_data
.lock
);
1056 pthread_mutex_lock(&channel
->lock
);
1057 pthread_mutex_lock(&channel
->timer_lock
);
1060 * This gives us a guarantee that the channel we are about to add to the
1061 * channel hash table will be unique. See this function comment on the why
1062 * we need to steel the channel key at this stage.
1064 steal_channel_key(channel
->key
);
1067 lttng_ht_add_unique_u64(consumer_data
.channel_ht
, &channel
->node
);
1070 pthread_mutex_unlock(&channel
->timer_lock
);
1071 pthread_mutex_unlock(&channel
->lock
);
1072 pthread_mutex_unlock(&consumer_data
.lock
);
1074 if (channel
->wait_fd
!= -1 && channel
->type
== CONSUMER_CHANNEL_TYPE_DATA
) {
1075 notify_channel_pipe(ctx
, channel
, -1, CONSUMER_CHANNEL_ADD
);
1082 * Allocate the pollfd structure and the local view of the out fds to avoid
1083 * doing a lookup in the linked list and concurrency issues when writing is
1084 * needed. Called with consumer_data.lock held.
1086 * Returns the number of fds in the structures.
1088 static int update_poll_array(struct lttng_consumer_local_data
*ctx
,
1089 struct pollfd
**pollfd
, struct lttng_consumer_stream
**local_stream
,
1090 struct lttng_ht
*ht
, int *nb_inactive_fd
)
1093 struct lttng_ht_iter iter
;
1094 struct lttng_consumer_stream
*stream
;
1099 assert(local_stream
);
1101 DBG("Updating poll fd array");
1102 *nb_inactive_fd
= 0;
1104 cds_lfht_for_each_entry(ht
->ht
, &iter
.iter
, stream
, node
.node
) {
1106 * Only active streams with an active end point can be added to the
1107 * poll set and local stream storage of the thread.
1109 * There is a potential race here for endpoint_status to be updated
1110 * just after the check. However, this is OK since the stream(s) will
1111 * be deleted once the thread is notified that the end point state has
1112 * changed where this function will be called back again.
1114 * We track the number of inactive FDs because they still need to be
1115 * closed by the polling thread after a wakeup on the data_pipe or
1118 if (stream
->state
!= LTTNG_CONSUMER_ACTIVE_STREAM
||
1119 stream
->endpoint_status
== CONSUMER_ENDPOINT_INACTIVE
) {
1120 (*nb_inactive_fd
)++;
1124 * This clobbers way too much the debug output. Uncomment that if you
1125 * need it for debugging purposes.
1127 * DBG("Active FD %d", stream->wait_fd);
1129 (*pollfd
)[i
].fd
= stream
->wait_fd
;
1130 (*pollfd
)[i
].events
= POLLIN
| POLLPRI
;
1131 local_stream
[i
] = stream
;
1137 * Insert the consumer_data_pipe at the end of the array and don't
1138 * increment i so nb_fd is the number of real FD.
1140 (*pollfd
)[i
].fd
= lttng_pipe_get_readfd(ctx
->consumer_data_pipe
);
1141 (*pollfd
)[i
].events
= POLLIN
| POLLPRI
;
1143 (*pollfd
)[i
+ 1].fd
= lttng_pipe_get_readfd(ctx
->consumer_wakeup_pipe
);
1144 (*pollfd
)[i
+ 1].events
= POLLIN
| POLLPRI
;
1149 * Poll on the should_quit pipe and the command socket return -1 on
1150 * error, 1 if should exit, 0 if data is available on the command socket
1152 int lttng_consumer_poll_socket(struct pollfd
*consumer_sockpoll
)
1157 num_rdy
= poll(consumer_sockpoll
, 2, -1);
1158 if (num_rdy
== -1) {
1160 * Restart interrupted system call.
1162 if (errno
== EINTR
) {
1165 PERROR("Poll error");
1168 if (consumer_sockpoll
[0].revents
& (POLLIN
| POLLPRI
)) {
1169 DBG("consumer_should_quit wake up");
1176 * Set the error socket.
1178 void lttng_consumer_set_error_sock(struct lttng_consumer_local_data
*ctx
,
1181 ctx
->consumer_error_socket
= sock
;
1185 * Set the command socket path.
1187 void lttng_consumer_set_command_sock_path(
1188 struct lttng_consumer_local_data
*ctx
, char *sock
)
1190 ctx
->consumer_command_sock_path
= sock
;
1194 * Send return code to the session daemon.
1195 * If the socket is not defined, we return 0, it is not a fatal error
1197 int lttng_consumer_send_error(struct lttng_consumer_local_data
*ctx
, int cmd
)
1199 if (ctx
->consumer_error_socket
> 0) {
1200 return lttcomm_send_unix_sock(ctx
->consumer_error_socket
, &cmd
,
1201 sizeof(enum lttcomm_sessiond_command
));
1208 * Close all the tracefiles and stream fds and MUST be called when all
1209 * instances are destroyed i.e. when all threads were joined and are ended.
1211 void lttng_consumer_cleanup(void)
1213 struct lttng_ht_iter iter
;
1214 struct lttng_consumer_channel
*channel
;
1218 cds_lfht_for_each_entry(consumer_data
.channel_ht
->ht
, &iter
.iter
, channel
,
1220 consumer_del_channel(channel
);
1225 lttng_ht_destroy(consumer_data
.channel_ht
);
1227 cleanup_relayd_ht();
1229 lttng_ht_destroy(consumer_data
.stream_per_chan_id_ht
);
1232 * This HT contains streams that are freed by either the metadata thread or
1233 * the data thread so we do *nothing* on the hash table and simply destroy
1236 lttng_ht_destroy(consumer_data
.stream_list_ht
);
1240 * Called from signal handler.
1242 void lttng_consumer_should_exit(struct lttng_consumer_local_data
*ctx
)
1246 CMM_STORE_SHARED(consumer_quit
, 1);
1247 ret
= lttng_write(ctx
->consumer_should_quit
[1], "4", 1);
1249 PERROR("write consumer quit");
1252 DBG("Consumer flag that it should quit");
1257 * Flush pending writes to trace output disk file.
1260 void lttng_consumer_sync_trace_file(struct lttng_consumer_stream
*stream
,
1264 int outfd
= stream
->out_fd
;
1267 * This does a blocking write-and-wait on any page that belongs to the
1268 * subbuffer prior to the one we just wrote.
1269 * Don't care about error values, as these are just hints and ways to
1270 * limit the amount of page cache used.
1272 if (orig_offset
< stream
->max_sb_size
) {
1275 lttng_sync_file_range(outfd
, orig_offset
- stream
->max_sb_size
,
1276 stream
->max_sb_size
,
1277 SYNC_FILE_RANGE_WAIT_BEFORE
1278 | SYNC_FILE_RANGE_WRITE
1279 | SYNC_FILE_RANGE_WAIT_AFTER
);
1281 * Give hints to the kernel about how we access the file:
1282 * POSIX_FADV_DONTNEED : we won't re-access data in a near future after
1285 * We need to call fadvise again after the file grows because the
1286 * kernel does not seem to apply fadvise to non-existing parts of the
1289 * Call fadvise _after_ having waited for the page writeback to
1290 * complete because the dirty page writeback semantic is not well
1291 * defined. So it can be expected to lead to lower throughput in
1294 ret
= posix_fadvise(outfd
, orig_offset
- stream
->max_sb_size
,
1295 stream
->max_sb_size
, POSIX_FADV_DONTNEED
);
1296 if (ret
&& ret
!= -ENOSYS
) {
1298 PERROR("posix_fadvise on fd %i", outfd
);
1303 * Initialise the necessary environnement :
1304 * - create a new context
1305 * - create the poll_pipe
1306 * - create the should_quit pipe (for signal handler)
1307 * - create the thread pipe (for splice)
1309 * Takes a function pointer as argument, this function is called when data is
1310 * available on a buffer. This function is responsible to do the
1311 * kernctl_get_next_subbuf, read the data with mmap or splice depending on the
1312 * buffer configuration and then kernctl_put_next_subbuf at the end.
1314 * Returns a pointer to the new context or NULL on error.
1316 struct lttng_consumer_local_data
*lttng_consumer_create(
1317 enum lttng_consumer_type type
,
1318 ssize_t (*buffer_ready
)(struct lttng_consumer_stream
*stream
,
1319 struct lttng_consumer_local_data
*ctx
),
1320 int (*recv_channel
)(struct lttng_consumer_channel
*channel
),
1321 int (*recv_stream
)(struct lttng_consumer_stream
*stream
),
1322 int (*update_stream
)(uint64_t stream_key
, uint32_t state
))
1325 struct lttng_consumer_local_data
*ctx
;
1327 assert(consumer_data
.type
== LTTNG_CONSUMER_UNKNOWN
||
1328 consumer_data
.type
== type
);
1329 consumer_data
.type
= type
;
1331 ctx
= zmalloc(sizeof(struct lttng_consumer_local_data
));
1333 PERROR("allocating context");
1337 ctx
->consumer_error_socket
= -1;
1338 ctx
->consumer_metadata_socket
= -1;
1339 pthread_mutex_init(&ctx
->metadata_socket_lock
, NULL
);
1340 /* assign the callbacks */
1341 ctx
->on_buffer_ready
= buffer_ready
;
1342 ctx
->on_recv_channel
= recv_channel
;
1343 ctx
->on_recv_stream
= recv_stream
;
1344 ctx
->on_update_stream
= update_stream
;
1346 ctx
->consumer_data_pipe
= lttng_pipe_open(0);
1347 if (!ctx
->consumer_data_pipe
) {
1348 goto error_poll_pipe
;
1351 ctx
->consumer_wakeup_pipe
= lttng_pipe_open(0);
1352 if (!ctx
->consumer_wakeup_pipe
) {
1353 goto error_wakeup_pipe
;
1356 ret
= pipe(ctx
->consumer_should_quit
);
1358 PERROR("Error creating recv pipe");
1359 goto error_quit_pipe
;
1362 ret
= pipe(ctx
->consumer_channel_pipe
);
1364 PERROR("Error creating channel pipe");
1365 goto error_channel_pipe
;
1368 ctx
->consumer_metadata_pipe
= lttng_pipe_open(0);
1369 if (!ctx
->consumer_metadata_pipe
) {
1370 goto error_metadata_pipe
;
1373 ctx
->channel_monitor_pipe
= -1;
1377 error_metadata_pipe
:
1378 utils_close_pipe(ctx
->consumer_channel_pipe
);
1380 utils_close_pipe(ctx
->consumer_should_quit
);
1382 lttng_pipe_destroy(ctx
->consumer_wakeup_pipe
);
1384 lttng_pipe_destroy(ctx
->consumer_data_pipe
);
1392 * Iterate over all streams of the hashtable and free them properly.
1394 static void destroy_data_stream_ht(struct lttng_ht
*ht
)
1396 struct lttng_ht_iter iter
;
1397 struct lttng_consumer_stream
*stream
;
1404 cds_lfht_for_each_entry(ht
->ht
, &iter
.iter
, stream
, node
.node
) {
1406 * Ignore return value since we are currently cleaning up so any error
1409 (void) consumer_del_stream(stream
, ht
);
1413 lttng_ht_destroy(ht
);
1417 * Iterate over all streams of the metadata hashtable and free them
1420 static void destroy_metadata_stream_ht(struct lttng_ht
*ht
)
1422 struct lttng_ht_iter iter
;
1423 struct lttng_consumer_stream
*stream
;
1430 cds_lfht_for_each_entry(ht
->ht
, &iter
.iter
, stream
, node
.node
) {
1432 * Ignore return value since we are currently cleaning up so any error
1435 (void) consumer_del_metadata_stream(stream
, ht
);
1439 lttng_ht_destroy(ht
);
1443 * Close all fds associated with the instance and free the context.
1445 void lttng_consumer_destroy(struct lttng_consumer_local_data
*ctx
)
1449 DBG("Consumer destroying it. Closing everything.");
1455 destroy_data_stream_ht(data_ht
);
1456 destroy_metadata_stream_ht(metadata_ht
);
1458 ret
= close(ctx
->consumer_error_socket
);
1462 ret
= close(ctx
->consumer_metadata_socket
);
1466 utils_close_pipe(ctx
->consumer_channel_pipe
);
1467 lttng_pipe_destroy(ctx
->consumer_data_pipe
);
1468 lttng_pipe_destroy(ctx
->consumer_metadata_pipe
);
1469 lttng_pipe_destroy(ctx
->consumer_wakeup_pipe
);
1470 utils_close_pipe(ctx
->consumer_should_quit
);
1472 unlink(ctx
->consumer_command_sock_path
);
1477 * Write the metadata stream id on the specified file descriptor.
1479 static int write_relayd_metadata_id(int fd
,
1480 struct lttng_consumer_stream
*stream
,
1481 unsigned long padding
)
1484 struct lttcomm_relayd_metadata_payload hdr
;
1486 hdr
.stream_id
= htobe64(stream
->relayd_stream_id
);
1487 hdr
.padding_size
= htobe32(padding
);
1488 ret
= lttng_write(fd
, (void *) &hdr
, sizeof(hdr
));
1489 if (ret
< sizeof(hdr
)) {
1491 * This error means that the fd's end is closed so ignore the PERROR
1492 * not to clubber the error output since this can happen in a normal
1495 if (errno
!= EPIPE
) {
1496 PERROR("write metadata stream id");
1498 DBG3("Consumer failed to write relayd metadata id (errno: %d)", errno
);
1500 * Set ret to a negative value because if ret != sizeof(hdr), we don't
1501 * handle writting the missing part so report that as an error and
1502 * don't lie to the caller.
1507 DBG("Metadata stream id %" PRIu64
" with padding %lu written before data",
1508 stream
->relayd_stream_id
, padding
);
1515 * Mmap the ring buffer, read it and write the data to the tracefile. This is a
1516 * core function for writing trace buffers to either the local filesystem or
1519 * It must be called with the stream lock held.
1521 * Careful review MUST be put if any changes occur!
1523 * Returns the number of bytes written
1525 ssize_t
lttng_consumer_on_read_subbuffer_mmap(
1526 struct lttng_consumer_local_data
*ctx
,
1527 struct lttng_consumer_stream
*stream
, unsigned long len
,
1528 unsigned long padding
,
1529 struct ctf_packet_index
*index
)
1531 unsigned long mmap_offset
;
1534 off_t orig_offset
= stream
->out_fd_offset
;
1535 /* Default is on the disk */
1536 int outfd
= stream
->out_fd
;
1537 struct consumer_relayd_sock_pair
*relayd
= NULL
;
1538 unsigned int relayd_hang_up
= 0;
1540 /* RCU lock for the relayd pointer */
1543 /* Flag that the current stream if set for network streaming. */
1544 if (stream
->net_seq_idx
!= (uint64_t) -1ULL) {
1545 relayd
= consumer_find_relayd(stream
->net_seq_idx
);
1546 if (relayd
== NULL
) {
1552 /* get the offset inside the fd to mmap */
1553 switch (consumer_data
.type
) {
1554 case LTTNG_CONSUMER_KERNEL
:
1555 mmap_base
= stream
->mmap_base
;
1556 ret
= kernctl_get_mmap_read_offset(stream
->wait_fd
, &mmap_offset
);
1558 PERROR("tracer ctl get_mmap_read_offset");
1562 case LTTNG_CONSUMER32_UST
:
1563 case LTTNG_CONSUMER64_UST
:
1564 mmap_base
= lttng_ustctl_get_mmap_base(stream
);
1566 ERR("read mmap get mmap base for stream %s", stream
->name
);
1570 ret
= lttng_ustctl_get_mmap_read_offset(stream
, &mmap_offset
);
1572 PERROR("tracer ctl get_mmap_read_offset");
1578 ERR("Unknown consumer_data type");
1582 /* Handle stream on the relayd if the output is on the network */
1584 unsigned long netlen
= len
;
1587 * Lock the control socket for the complete duration of the function
1588 * since from this point on we will use the socket.
1590 if (stream
->metadata_flag
) {
1591 /* Metadata requires the control socket. */
1592 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
1593 if (stream
->reset_metadata_flag
) {
1594 ret
= relayd_reset_metadata(&relayd
->control_sock
,
1595 stream
->relayd_stream_id
,
1596 stream
->metadata_version
);
1601 stream
->reset_metadata_flag
= 0;
1603 netlen
+= sizeof(struct lttcomm_relayd_metadata_payload
);
1606 ret
= write_relayd_stream_header(stream
, netlen
, padding
, relayd
);
1611 /* Use the returned socket. */
1614 /* Write metadata stream id before payload */
1615 if (stream
->metadata_flag
) {
1616 ret
= write_relayd_metadata_id(outfd
, stream
, padding
);
1623 /* No streaming, we have to set the len with the full padding */
1626 if (stream
->metadata_flag
&& stream
->reset_metadata_flag
) {
1627 ret
= utils_truncate_stream_file(stream
->out_fd
, 0);
1629 ERR("Reset metadata file");
1632 stream
->reset_metadata_flag
= 0;
1636 * Check if we need to change the tracefile before writing the packet.
1638 if (stream
->chan
->tracefile_size
> 0 &&
1639 (stream
->tracefile_size_current
+ len
) >
1640 stream
->chan
->tracefile_size
) {
1641 ret
= utils_rotate_stream_file(stream
->chan
->pathname
,
1642 stream
->name
, stream
->chan
->tracefile_size
,
1643 stream
->chan
->tracefile_count
, stream
->uid
, stream
->gid
,
1644 stream
->out_fd
, &(stream
->tracefile_count_current
),
1647 ERR("Rotating output file");
1650 outfd
= stream
->out_fd
;
1652 if (stream
->index_file
) {
1653 lttng_index_file_put(stream
->index_file
);
1654 stream
->index_file
= lttng_index_file_create(stream
->chan
->pathname
,
1655 stream
->name
, stream
->uid
, stream
->gid
,
1656 stream
->chan
->tracefile_size
,
1657 stream
->tracefile_count_current
,
1658 CTF_INDEX_MAJOR
, CTF_INDEX_MINOR
);
1659 if (!stream
->index_file
) {
1664 /* Reset current size because we just perform a rotation. */
1665 stream
->tracefile_size_current
= 0;
1666 stream
->out_fd_offset
= 0;
1669 stream
->tracefile_size_current
+= len
;
1671 index
->offset
= htobe64(stream
->out_fd_offset
);
1676 * This call guarantee that len or less is returned. It's impossible to
1677 * receive a ret value that is bigger than len.
1679 ret
= lttng_write(outfd
, mmap_base
+ mmap_offset
, len
);
1680 DBG("Consumer mmap write() ret %zd (len %lu)", ret
, len
);
1681 if (ret
< 0 || ((size_t) ret
!= len
)) {
1683 * Report error to caller if nothing was written else at least send the
1691 /* Socket operation failed. We consider the relayd dead */
1692 if (errno
== EPIPE
|| errno
== EINVAL
|| errno
== EBADF
) {
1694 * This is possible if the fd is closed on the other side
1695 * (outfd) or any write problem. It can be verbose a bit for a
1696 * normal execution if for instance the relayd is stopped
1697 * abruptly. This can happen so set this to a DBG statement.
1699 DBG("Consumer mmap write detected relayd hang up");
1701 /* Unhandled error, print it and stop function right now. */
1702 PERROR("Error in write mmap (ret %zd != len %lu)", ret
, len
);
1706 stream
->output_written
+= ret
;
1708 /* This call is useless on a socket so better save a syscall. */
1710 /* This won't block, but will start writeout asynchronously */
1711 lttng_sync_file_range(outfd
, stream
->out_fd_offset
, len
,
1712 SYNC_FILE_RANGE_WRITE
);
1713 stream
->out_fd_offset
+= len
;
1714 lttng_consumer_sync_trace_file(stream
, orig_offset
);
1719 * This is a special case that the relayd has closed its socket. Let's
1720 * cleanup the relayd object and all associated streams.
1722 if (relayd
&& relayd_hang_up
) {
1723 cleanup_relayd(relayd
, ctx
);
1727 /* Unlock only if ctrl socket used */
1728 if (relayd
&& stream
->metadata_flag
) {
1729 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
1737 * Splice the data from the ring buffer to the tracefile.
1739 * It must be called with the stream lock held.
1741 * Returns the number of bytes spliced.
1743 ssize_t
lttng_consumer_on_read_subbuffer_splice(
1744 struct lttng_consumer_local_data
*ctx
,
1745 struct lttng_consumer_stream
*stream
, unsigned long len
,
1746 unsigned long padding
,
1747 struct ctf_packet_index
*index
)
1749 ssize_t ret
= 0, written
= 0, ret_splice
= 0;
1751 off_t orig_offset
= stream
->out_fd_offset
;
1752 int fd
= stream
->wait_fd
;
1753 /* Default is on the disk */
1754 int outfd
= stream
->out_fd
;
1755 struct consumer_relayd_sock_pair
*relayd
= NULL
;
1757 unsigned int relayd_hang_up
= 0;
1759 switch (consumer_data
.type
) {
1760 case LTTNG_CONSUMER_KERNEL
:
1762 case LTTNG_CONSUMER32_UST
:
1763 case LTTNG_CONSUMER64_UST
:
1764 /* Not supported for user space tracing */
1767 ERR("Unknown consumer_data type");
1771 /* RCU lock for the relayd pointer */
1774 /* Flag that the current stream if set for network streaming. */
1775 if (stream
->net_seq_idx
!= (uint64_t) -1ULL) {
1776 relayd
= consumer_find_relayd(stream
->net_seq_idx
);
1777 if (relayd
== NULL
) {
1782 splice_pipe
= stream
->splice_pipe
;
1784 /* Write metadata stream id before payload */
1786 unsigned long total_len
= len
;
1788 if (stream
->metadata_flag
) {
1790 * Lock the control socket for the complete duration of the function
1791 * since from this point on we will use the socket.
1793 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
1795 if (stream
->reset_metadata_flag
) {
1796 ret
= relayd_reset_metadata(&relayd
->control_sock
,
1797 stream
->relayd_stream_id
,
1798 stream
->metadata_version
);
1803 stream
->reset_metadata_flag
= 0;
1805 ret
= write_relayd_metadata_id(splice_pipe
[1], stream
,
1813 total_len
+= sizeof(struct lttcomm_relayd_metadata_payload
);
1816 ret
= write_relayd_stream_header(stream
, total_len
, padding
, relayd
);
1822 /* Use the returned socket. */
1825 /* No streaming, we have to set the len with the full padding */
1828 if (stream
->metadata_flag
&& stream
->reset_metadata_flag
) {
1829 ret
= utils_truncate_stream_file(stream
->out_fd
, 0);
1831 ERR("Reset metadata file");
1834 stream
->reset_metadata_flag
= 0;
1837 * Check if we need to change the tracefile before writing the packet.
1839 if (stream
->chan
->tracefile_size
> 0 &&
1840 (stream
->tracefile_size_current
+ len
) >
1841 stream
->chan
->tracefile_size
) {
1842 ret
= utils_rotate_stream_file(stream
->chan
->pathname
,
1843 stream
->name
, stream
->chan
->tracefile_size
,
1844 stream
->chan
->tracefile_count
, stream
->uid
, stream
->gid
,
1845 stream
->out_fd
, &(stream
->tracefile_count_current
),
1849 ERR("Rotating output file");
1852 outfd
= stream
->out_fd
;
1854 if (stream
->index_file
) {
1855 lttng_index_file_put(stream
->index_file
);
1856 stream
->index_file
= lttng_index_file_create(stream
->chan
->pathname
,
1857 stream
->name
, stream
->uid
, stream
->gid
,
1858 stream
->chan
->tracefile_size
,
1859 stream
->tracefile_count_current
,
1860 CTF_INDEX_MAJOR
, CTF_INDEX_MINOR
);
1861 if (!stream
->index_file
) {
1866 /* Reset current size because we just perform a rotation. */
1867 stream
->tracefile_size_current
= 0;
1868 stream
->out_fd_offset
= 0;
1871 stream
->tracefile_size_current
+= len
;
1872 index
->offset
= htobe64(stream
->out_fd_offset
);
1876 DBG("splice chan to pipe offset %lu of len %lu (fd : %d, pipe: %d)",
1877 (unsigned long)offset
, len
, fd
, splice_pipe
[1]);
1878 ret_splice
= splice(fd
, &offset
, splice_pipe
[1], NULL
, len
,
1879 SPLICE_F_MOVE
| SPLICE_F_MORE
);
1880 DBG("splice chan to pipe, ret %zd", ret_splice
);
1881 if (ret_splice
< 0) {
1884 PERROR("Error in relay splice");
1888 /* Handle stream on the relayd if the output is on the network */
1889 if (relayd
&& stream
->metadata_flag
) {
1890 size_t metadata_payload_size
=
1891 sizeof(struct lttcomm_relayd_metadata_payload
);
1893 /* Update counter to fit the spliced data */
1894 ret_splice
+= metadata_payload_size
;
1895 len
+= metadata_payload_size
;
1897 * We do this so the return value can match the len passed as
1898 * argument to this function.
1900 written
-= metadata_payload_size
;
1903 /* Splice data out */
1904 ret_splice
= splice(splice_pipe
[0], NULL
, outfd
, NULL
,
1905 ret_splice
, SPLICE_F_MOVE
| SPLICE_F_MORE
);
1906 DBG("Consumer splice pipe to file (out_fd: %d), ret %zd",
1908 if (ret_splice
< 0) {
1913 } else if (ret_splice
> len
) {
1915 * We don't expect this code path to be executed but you never know
1916 * so this is an extra protection agains a buggy splice().
1919 written
+= ret_splice
;
1920 PERROR("Wrote more data than requested %zd (len: %lu)", ret_splice
,
1924 /* All good, update current len and continue. */
1928 /* This call is useless on a socket so better save a syscall. */
1930 /* This won't block, but will start writeout asynchronously */
1931 lttng_sync_file_range(outfd
, stream
->out_fd_offset
, ret_splice
,
1932 SYNC_FILE_RANGE_WRITE
);
1933 stream
->out_fd_offset
+= ret_splice
;
1935 stream
->output_written
+= ret_splice
;
1936 written
+= ret_splice
;
1939 lttng_consumer_sync_trace_file(stream
, orig_offset
);
1945 * This is a special case that the relayd has closed its socket. Let's
1946 * cleanup the relayd object and all associated streams.
1948 if (relayd
&& relayd_hang_up
) {
1949 cleanup_relayd(relayd
, ctx
);
1950 /* Skip splice error so the consumer does not fail */
1955 /* send the appropriate error description to sessiond */
1958 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_SPLICE_EINVAL
);
1961 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_SPLICE_ENOMEM
);
1964 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_SPLICE_ESPIPE
);
1969 if (relayd
&& stream
->metadata_flag
) {
1970 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
1978 * Sample the snapshot positions for a specific fd
1980 * Returns 0 on success, < 0 on error
1982 int lttng_consumer_sample_snapshot_positions(struct lttng_consumer_stream
*stream
)
1984 switch (consumer_data
.type
) {
1985 case LTTNG_CONSUMER_KERNEL
:
1986 return lttng_kconsumer_sample_snapshot_positions(stream
);
1987 case LTTNG_CONSUMER32_UST
:
1988 case LTTNG_CONSUMER64_UST
:
1989 return lttng_ustconsumer_sample_snapshot_positions(stream
);
1991 ERR("Unknown consumer_data type");
1997 * Take a snapshot for a specific fd
1999 * Returns 0 on success, < 0 on error
2001 int lttng_consumer_take_snapshot(struct lttng_consumer_stream
*stream
)
2003 switch (consumer_data
.type
) {
2004 case LTTNG_CONSUMER_KERNEL
:
2005 return lttng_kconsumer_take_snapshot(stream
);
2006 case LTTNG_CONSUMER32_UST
:
2007 case LTTNG_CONSUMER64_UST
:
2008 return lttng_ustconsumer_take_snapshot(stream
);
2010 ERR("Unknown consumer_data type");
2017 * Get the produced position
2019 * Returns 0 on success, < 0 on error
2021 int lttng_consumer_get_produced_snapshot(struct lttng_consumer_stream
*stream
,
2024 switch (consumer_data
.type
) {
2025 case LTTNG_CONSUMER_KERNEL
:
2026 return lttng_kconsumer_get_produced_snapshot(stream
, pos
);
2027 case LTTNG_CONSUMER32_UST
:
2028 case LTTNG_CONSUMER64_UST
:
2029 return lttng_ustconsumer_get_produced_snapshot(stream
, pos
);
2031 ERR("Unknown consumer_data type");
2038 * Get the consumed position (free-running counter position in bytes).
2040 * Returns 0 on success, < 0 on error
2042 int lttng_consumer_get_consumed_snapshot(struct lttng_consumer_stream
*stream
,
2045 switch (consumer_data
.type
) {
2046 case LTTNG_CONSUMER_KERNEL
:
2047 return lttng_kconsumer_get_consumed_snapshot(stream
, pos
);
2048 case LTTNG_CONSUMER32_UST
:
2049 case LTTNG_CONSUMER64_UST
:
2050 return lttng_ustconsumer_get_consumed_snapshot(stream
, pos
);
2052 ERR("Unknown consumer_data type");
2058 int lttng_consumer_recv_cmd(struct lttng_consumer_local_data
*ctx
,
2059 int sock
, struct pollfd
*consumer_sockpoll
)
2061 switch (consumer_data
.type
) {
2062 case LTTNG_CONSUMER_KERNEL
:
2063 return lttng_kconsumer_recv_cmd(ctx
, sock
, consumer_sockpoll
);
2064 case LTTNG_CONSUMER32_UST
:
2065 case LTTNG_CONSUMER64_UST
:
2066 return lttng_ustconsumer_recv_cmd(ctx
, sock
, consumer_sockpoll
);
2068 ERR("Unknown consumer_data type");
2074 void lttng_consumer_close_all_metadata(void)
2076 switch (consumer_data
.type
) {
2077 case LTTNG_CONSUMER_KERNEL
:
2079 * The Kernel consumer has a different metadata scheme so we don't
2080 * close anything because the stream will be closed by the session
2084 case LTTNG_CONSUMER32_UST
:
2085 case LTTNG_CONSUMER64_UST
:
2087 * Close all metadata streams. The metadata hash table is passed and
2088 * this call iterates over it by closing all wakeup fd. This is safe
2089 * because at this point we are sure that the metadata producer is
2090 * either dead or blocked.
2092 lttng_ustconsumer_close_all_metadata(metadata_ht
);
2095 ERR("Unknown consumer_data type");
2101 * Clean up a metadata stream and free its memory.
2103 void consumer_del_metadata_stream(struct lttng_consumer_stream
*stream
,
2104 struct lttng_ht
*ht
)
2106 struct lttng_consumer_channel
*free_chan
= NULL
;
2110 * This call should NEVER receive regular stream. It must always be
2111 * metadata stream and this is crucial for data structure synchronization.
2113 assert(stream
->metadata_flag
);
2115 DBG3("Consumer delete metadata stream %d", stream
->wait_fd
);
2117 pthread_mutex_lock(&consumer_data
.lock
);
2118 pthread_mutex_lock(&stream
->chan
->lock
);
2119 pthread_mutex_lock(&stream
->lock
);
2120 if (stream
->chan
->metadata_cache
) {
2121 /* Only applicable to userspace consumers. */
2122 pthread_mutex_lock(&stream
->chan
->metadata_cache
->lock
);
2125 /* Remove any reference to that stream. */
2126 consumer_stream_delete(stream
, ht
);
2128 /* Close down everything including the relayd if one. */
2129 consumer_stream_close(stream
);
2130 /* Destroy tracer buffers of the stream. */
2131 consumer_stream_destroy_buffers(stream
);
2133 /* Atomically decrement channel refcount since other threads can use it. */
2134 if (!uatomic_sub_return(&stream
->chan
->refcount
, 1)
2135 && !uatomic_read(&stream
->chan
->nb_init_stream_left
)) {
2136 /* Go for channel deletion! */
2137 free_chan
= stream
->chan
;
2141 * Nullify the stream reference so it is not used after deletion. The
2142 * channel lock MUST be acquired before being able to check for a NULL
2145 stream
->chan
->metadata_stream
= NULL
;
2147 if (stream
->chan
->metadata_cache
) {
2148 pthread_mutex_unlock(&stream
->chan
->metadata_cache
->lock
);
2150 pthread_mutex_unlock(&stream
->lock
);
2151 pthread_mutex_unlock(&stream
->chan
->lock
);
2152 pthread_mutex_unlock(&consumer_data
.lock
);
2155 consumer_del_channel(free_chan
);
2158 consumer_stream_free(stream
);
2162 * Action done with the metadata stream when adding it to the consumer internal
2163 * data structures to handle it.
2165 void consumer_add_metadata_stream(struct lttng_consumer_stream
*stream
)
2167 struct lttng_ht
*ht
= metadata_ht
;
2168 struct lttng_ht_iter iter
;
2169 struct lttng_ht_node_u64
*node
;
2174 DBG3("Adding metadata stream %" PRIu64
" to hash table", stream
->key
);
2176 pthread_mutex_lock(&consumer_data
.lock
);
2177 pthread_mutex_lock(&stream
->chan
->lock
);
2178 pthread_mutex_lock(&stream
->chan
->timer_lock
);
2179 pthread_mutex_lock(&stream
->lock
);
2182 * From here, refcounts are updated so be _careful_ when returning an error
2189 * Lookup the stream just to make sure it does not exist in our internal
2190 * state. This should NEVER happen.
2192 lttng_ht_lookup(ht
, &stream
->key
, &iter
);
2193 node
= lttng_ht_iter_get_node_u64(&iter
);
2197 * When nb_init_stream_left reaches 0, we don't need to trigger any action
2198 * in terms of destroying the associated channel, because the action that
2199 * causes the count to become 0 also causes a stream to be added. The
2200 * channel deletion will thus be triggered by the following removal of this
2203 if (uatomic_read(&stream
->chan
->nb_init_stream_left
) > 0) {
2204 /* Increment refcount before decrementing nb_init_stream_left */
2206 uatomic_dec(&stream
->chan
->nb_init_stream_left
);
2209 lttng_ht_add_unique_u64(ht
, &stream
->node
);
2211 lttng_ht_add_u64(consumer_data
.stream_per_chan_id_ht
,
2212 &stream
->node_channel_id
);
2215 * Add stream to the stream_list_ht of the consumer data. No need to steal
2216 * the key since the HT does not use it and we allow to add redundant keys
2219 lttng_ht_add_u64(consumer_data
.stream_list_ht
, &stream
->node_session_id
);
2223 pthread_mutex_unlock(&stream
->lock
);
2224 pthread_mutex_unlock(&stream
->chan
->lock
);
2225 pthread_mutex_unlock(&stream
->chan
->timer_lock
);
2226 pthread_mutex_unlock(&consumer_data
.lock
);
2230 * Delete data stream that are flagged for deletion (endpoint_status).
2232 static void validate_endpoint_status_data_stream(void)
2234 struct lttng_ht_iter iter
;
2235 struct lttng_consumer_stream
*stream
;
2237 DBG("Consumer delete flagged data stream");
2240 cds_lfht_for_each_entry(data_ht
->ht
, &iter
.iter
, stream
, node
.node
) {
2241 /* Validate delete flag of the stream */
2242 if (stream
->endpoint_status
== CONSUMER_ENDPOINT_ACTIVE
) {
2245 /* Delete it right now */
2246 consumer_del_stream(stream
, data_ht
);
2252 * Delete metadata stream that are flagged for deletion (endpoint_status).
2254 static void validate_endpoint_status_metadata_stream(
2255 struct lttng_poll_event
*pollset
)
2257 struct lttng_ht_iter iter
;
2258 struct lttng_consumer_stream
*stream
;
2260 DBG("Consumer delete flagged metadata stream");
2265 cds_lfht_for_each_entry(metadata_ht
->ht
, &iter
.iter
, stream
, node
.node
) {
2266 /* Validate delete flag of the stream */
2267 if (stream
->endpoint_status
== CONSUMER_ENDPOINT_ACTIVE
) {
2271 * Remove from pollset so the metadata thread can continue without
2272 * blocking on a deleted stream.
2274 lttng_poll_del(pollset
, stream
->wait_fd
);
2276 /* Delete it right now */
2277 consumer_del_metadata_stream(stream
, metadata_ht
);
2283 int rotate_notify_sessiond(struct lttng_consumer_local_data
*ctx
,
2289 ret
= write(ctx
->channel_rotate_pipe
, &key
, sizeof(key
));
2290 } while (ret
== -1 && errno
== EINTR
);
2292 PERROR("Failed to write to the channel rotation pipe");
2294 DBG("Sent channel rotation notification for channel key %"
2303 * Perform operations that need to be done after a stream has
2304 * rotated and released the stream lock.
2306 * Multiple rotations cannot occur simultaneously, so we know the state of the
2307 * "rotated" stream flag cannot change.
2309 * This MUST be called WITHOUT the stream lock held.
2312 int consumer_post_rotation(struct lttng_consumer_stream
*stream
,
2313 struct lttng_consumer_local_data
*ctx
)
2317 pthread_mutex_lock(&stream
->chan
->lock
);
2319 switch (consumer_data
.type
) {
2320 case LTTNG_CONSUMER_KERNEL
:
2322 case LTTNG_CONSUMER32_UST
:
2323 case LTTNG_CONSUMER64_UST
:
2325 * The ust_metadata_pushed counter has been reset to 0, so now
2326 * we can wakeup the metadata thread so it dumps the metadata
2327 * cache to the new file.
2329 if (stream
->metadata_flag
) {
2330 consumer_metadata_wakeup_pipe(stream
->chan
);
2334 ERR("Unknown consumer_data type");
2338 if (--stream
->chan
->nr_stream_rotate_pending
== 0) {
2339 DBG("Rotation of channel \"%s\" completed, notifying the session daemon",
2340 stream
->chan
->name
);
2341 ret
= rotate_notify_sessiond(ctx
, stream
->chan
->key
);
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 */
2580 /* 2 for the consumer_data_pipe and wake up pipe */
2581 const int nb_pipes_fd
= 2;
2582 /* Number of FDs with CONSUMER_ENDPOINT_INACTIVE but still open. */
2583 int nb_inactive_fd
= 0;
2584 struct lttng_consumer_local_data
*ctx
= data
;
2587 rcu_register_thread();
2589 health_register(health_consumerd
, HEALTH_CONSUMERD_TYPE_DATA
);
2591 if (testpoint(consumerd_thread_data
)) {
2592 goto error_testpoint
;
2595 health_code_update();
2597 local_stream
= zmalloc(sizeof(struct lttng_consumer_stream
*));
2598 if (local_stream
== NULL
) {
2599 PERROR("local_stream malloc");
2604 health_code_update();
2610 * the fds set has been updated, we need to update our
2611 * local array as well
2613 pthread_mutex_lock(&consumer_data
.lock
);
2614 if (consumer_data
.need_update
) {
2619 local_stream
= NULL
;
2621 /* Allocate for all fds */
2622 pollfd
= zmalloc((consumer_data
.stream_count
+ nb_pipes_fd
) * sizeof(struct pollfd
));
2623 if (pollfd
== NULL
) {
2624 PERROR("pollfd malloc");
2625 pthread_mutex_unlock(&consumer_data
.lock
);
2629 local_stream
= zmalloc((consumer_data
.stream_count
+ nb_pipes_fd
) *
2630 sizeof(struct lttng_consumer_stream
*));
2631 if (local_stream
== NULL
) {
2632 PERROR("local_stream malloc");
2633 pthread_mutex_unlock(&consumer_data
.lock
);
2636 ret
= update_poll_array(ctx
, &pollfd
, local_stream
,
2637 data_ht
, &nb_inactive_fd
);
2639 ERR("Error in allocating pollfd or local_outfds");
2640 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_POLL_ERROR
);
2641 pthread_mutex_unlock(&consumer_data
.lock
);
2645 consumer_data
.need_update
= 0;
2647 pthread_mutex_unlock(&consumer_data
.lock
);
2649 /* No FDs and consumer_quit, consumer_cleanup the thread */
2650 if (nb_fd
== 0 && nb_inactive_fd
== 0 &&
2651 CMM_LOAD_SHARED(consumer_quit
) == 1) {
2652 err
= 0; /* All is OK */
2655 /* poll on the array of fds */
2657 DBG("polling on %d fd", nb_fd
+ nb_pipes_fd
);
2658 if (testpoint(consumerd_thread_data_poll
)) {
2661 health_poll_entry();
2662 num_rdy
= poll(pollfd
, nb_fd
+ nb_pipes_fd
, -1);
2664 DBG("poll num_rdy : %d", num_rdy
);
2665 if (num_rdy
== -1) {
2667 * Restart interrupted system call.
2669 if (errno
== EINTR
) {
2672 PERROR("Poll error");
2673 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_POLL_ERROR
);
2675 } else if (num_rdy
== 0) {
2676 DBG("Polling thread timed out");
2680 if (caa_unlikely(data_consumption_paused
)) {
2681 DBG("Data consumption paused, sleeping...");
2687 * If the consumer_data_pipe triggered poll go directly to the
2688 * beginning of the loop to update the array. We want to prioritize
2689 * array update over low-priority reads.
2691 if (pollfd
[nb_fd
].revents
& (POLLIN
| POLLPRI
)) {
2692 ssize_t pipe_readlen
;
2694 DBG("consumer_data_pipe wake up");
2695 pipe_readlen
= lttng_pipe_read(ctx
->consumer_data_pipe
,
2696 &new_stream
, sizeof(new_stream
));
2697 if (pipe_readlen
< sizeof(new_stream
)) {
2698 PERROR("Consumer data pipe");
2699 /* Continue so we can at least handle the current stream(s). */
2704 * If the stream is NULL, just ignore it. It's also possible that
2705 * the sessiond poll thread changed the consumer_quit state and is
2706 * waking us up to test it.
2708 if (new_stream
== NULL
) {
2709 validate_endpoint_status_data_stream();
2713 /* Continue to update the local streams and handle prio ones */
2717 /* Handle wakeup pipe. */
2718 if (pollfd
[nb_fd
+ 1].revents
& (POLLIN
| POLLPRI
)) {
2720 ssize_t pipe_readlen
;
2722 pipe_readlen
= lttng_pipe_read(ctx
->consumer_wakeup_pipe
, &dummy
,
2724 if (pipe_readlen
< 0) {
2725 PERROR("Consumer data wakeup pipe");
2727 /* We've been awakened to handle stream(s). */
2728 ctx
->has_wakeup
= 0;
2731 /* Take care of high priority channels first. */
2732 for (i
= 0; i
< nb_fd
; i
++) {
2733 health_code_update();
2735 if (local_stream
[i
] == NULL
) {
2738 if (pollfd
[i
].revents
& POLLPRI
) {
2739 DBG("Urgent read on fd %d", pollfd
[i
].fd
);
2741 len
= ctx
->on_buffer_ready(local_stream
[i
], ctx
);
2742 /* it's ok to have an unavailable sub-buffer */
2743 if (len
< 0 && len
!= -EAGAIN
&& len
!= -ENODATA
) {
2744 /* Clean the stream and free it. */
2745 consumer_del_stream(local_stream
[i
], data_ht
);
2746 local_stream
[i
] = NULL
;
2747 } else if (len
> 0) {
2748 local_stream
[i
]->data_read
= 1;
2754 * If we read high prio channel in this loop, try again
2755 * for more high prio data.
2761 /* Take care of low priority channels. */
2762 for (i
= 0; i
< nb_fd
; i
++) {
2763 health_code_update();
2765 if (local_stream
[i
] == NULL
) {
2768 if ((pollfd
[i
].revents
& POLLIN
) ||
2769 local_stream
[i
]->hangup_flush_done
||
2770 local_stream
[i
]->has_data
) {
2771 DBG("Normal read on fd %d", pollfd
[i
].fd
);
2772 len
= ctx
->on_buffer_ready(local_stream
[i
], ctx
);
2773 /* it's ok to have an unavailable sub-buffer */
2774 if (len
< 0 && len
!= -EAGAIN
&& len
!= -ENODATA
) {
2775 /* Clean the stream and free it. */
2776 consumer_del_stream(local_stream
[i
], data_ht
);
2777 local_stream
[i
] = NULL
;
2778 } else if (len
> 0) {
2779 local_stream
[i
]->data_read
= 1;
2784 /* Handle hangup and errors */
2785 for (i
= 0; i
< nb_fd
; i
++) {
2786 health_code_update();
2788 if (local_stream
[i
] == NULL
) {
2791 if (!local_stream
[i
]->hangup_flush_done
2792 && (pollfd
[i
].revents
& (POLLHUP
| POLLERR
| POLLNVAL
))
2793 && (consumer_data
.type
== LTTNG_CONSUMER32_UST
2794 || consumer_data
.type
== LTTNG_CONSUMER64_UST
)) {
2795 DBG("fd %d is hup|err|nval. Attempting flush and read.",
2797 lttng_ustconsumer_on_stream_hangup(local_stream
[i
]);
2798 /* Attempt read again, for the data we just flushed. */
2799 local_stream
[i
]->data_read
= 1;
2802 * If the poll flag is HUP/ERR/NVAL and we have
2803 * read no data in this pass, we can remove the
2804 * stream from its hash table.
2806 if ((pollfd
[i
].revents
& POLLHUP
)) {
2807 DBG("Polling fd %d tells it has hung up.", pollfd
[i
].fd
);
2808 if (!local_stream
[i
]->data_read
) {
2809 consumer_del_stream(local_stream
[i
], data_ht
);
2810 local_stream
[i
] = NULL
;
2813 } else if (pollfd
[i
].revents
& POLLERR
) {
2814 ERR("Error returned in polling fd %d.", pollfd
[i
].fd
);
2815 if (!local_stream
[i
]->data_read
) {
2816 consumer_del_stream(local_stream
[i
], data_ht
);
2817 local_stream
[i
] = NULL
;
2820 } else if (pollfd
[i
].revents
& POLLNVAL
) {
2821 ERR("Polling fd %d tells fd is not open.", pollfd
[i
].fd
);
2822 if (!local_stream
[i
]->data_read
) {
2823 consumer_del_stream(local_stream
[i
], data_ht
);
2824 local_stream
[i
] = NULL
;
2828 if (local_stream
[i
] != NULL
) {
2829 local_stream
[i
]->data_read
= 0;
2836 DBG("polling thread exiting");
2841 * Close the write side of the pipe so epoll_wait() in
2842 * consumer_thread_metadata_poll can catch it. The thread is monitoring the
2843 * read side of the pipe. If we close them both, epoll_wait strangely does
2844 * not return and could create a endless wait period if the pipe is the
2845 * only tracked fd in the poll set. The thread will take care of closing
2848 (void) lttng_pipe_write_close(ctx
->consumer_metadata_pipe
);
2853 ERR("Health error occurred in %s", __func__
);
2855 health_unregister(health_consumerd
);
2857 rcu_unregister_thread();
2862 * Close wake-up end of each stream belonging to the channel. This will
2863 * allow the poll() on the stream read-side to detect when the
2864 * write-side (application) finally closes them.
2867 void consumer_close_channel_streams(struct lttng_consumer_channel
*channel
)
2869 struct lttng_ht
*ht
;
2870 struct lttng_consumer_stream
*stream
;
2871 struct lttng_ht_iter iter
;
2873 ht
= consumer_data
.stream_per_chan_id_ht
;
2876 cds_lfht_for_each_entry_duplicate(ht
->ht
,
2877 ht
->hash_fct(&channel
->key
, lttng_ht_seed
),
2878 ht
->match_fct
, &channel
->key
,
2879 &iter
.iter
, stream
, node_channel_id
.node
) {
2881 * Protect against teardown with mutex.
2883 pthread_mutex_lock(&stream
->lock
);
2884 if (cds_lfht_is_node_deleted(&stream
->node
.node
)) {
2887 switch (consumer_data
.type
) {
2888 case LTTNG_CONSUMER_KERNEL
:
2890 case LTTNG_CONSUMER32_UST
:
2891 case LTTNG_CONSUMER64_UST
:
2892 if (stream
->metadata_flag
) {
2893 /* Safe and protected by the stream lock. */
2894 lttng_ustconsumer_close_metadata(stream
->chan
);
2897 * Note: a mutex is taken internally within
2898 * liblttng-ust-ctl to protect timer wakeup_fd
2899 * use from concurrent close.
2901 lttng_ustconsumer_close_stream_wakeup(stream
);
2905 ERR("Unknown consumer_data type");
2909 pthread_mutex_unlock(&stream
->lock
);
2914 static void destroy_channel_ht(struct lttng_ht
*ht
)
2916 struct lttng_ht_iter iter
;
2917 struct lttng_consumer_channel
*channel
;
2925 cds_lfht_for_each_entry(ht
->ht
, &iter
.iter
, channel
, wait_fd_node
.node
) {
2926 ret
= lttng_ht_del(ht
, &iter
);
2931 lttng_ht_destroy(ht
);
2935 * This thread polls the channel fds to detect when they are being
2936 * closed. It closes all related streams if the channel is detected as
2937 * closed. It is currently only used as a shim layer for UST because the
2938 * consumerd needs to keep the per-stream wakeup end of pipes open for
2941 void *consumer_thread_channel_poll(void *data
)
2943 int ret
, i
, pollfd
, err
= -1;
2944 uint32_t revents
, nb_fd
;
2945 struct lttng_consumer_channel
*chan
= NULL
;
2946 struct lttng_ht_iter iter
;
2947 struct lttng_ht_node_u64
*node
;
2948 struct lttng_poll_event events
;
2949 struct lttng_consumer_local_data
*ctx
= data
;
2950 struct lttng_ht
*channel_ht
;
2952 rcu_register_thread();
2954 health_register(health_consumerd
, HEALTH_CONSUMERD_TYPE_CHANNEL
);
2956 if (testpoint(consumerd_thread_channel
)) {
2957 goto error_testpoint
;
2960 health_code_update();
2962 channel_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
2964 /* ENOMEM at this point. Better to bail out. */
2968 DBG("Thread channel poll started");
2970 /* Size is set to 1 for the consumer_channel pipe */
2971 ret
= lttng_poll_create(&events
, 2, LTTNG_CLOEXEC
);
2973 ERR("Poll set creation failed");
2977 ret
= lttng_poll_add(&events
, ctx
->consumer_channel_pipe
[0], LPOLLIN
);
2983 DBG("Channel main loop started");
2987 health_code_update();
2988 DBG("Channel poll wait");
2989 health_poll_entry();
2990 ret
= lttng_poll_wait(&events
, -1);
2991 DBG("Channel poll return from wait with %d fd(s)",
2992 LTTNG_POLL_GETNB(&events
));
2994 DBG("Channel event caught in thread");
2996 if (errno
== EINTR
) {
2997 ERR("Poll EINTR caught");
3000 if (LTTNG_POLL_GETNB(&events
) == 0) {
3001 err
= 0; /* All is OK */
3008 /* From here, the event is a channel wait fd */
3009 for (i
= 0; i
< nb_fd
; i
++) {
3010 health_code_update();
3012 revents
= LTTNG_POLL_GETEV(&events
, i
);
3013 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3016 /* No activity for this FD (poll implementation). */
3020 if (pollfd
== ctx
->consumer_channel_pipe
[0]) {
3021 if (revents
& LPOLLIN
) {
3022 enum consumer_channel_action action
;
3025 ret
= read_channel_pipe(ctx
, &chan
, &key
, &action
);
3028 ERR("Error reading channel pipe");
3030 lttng_poll_del(&events
, ctx
->consumer_channel_pipe
[0]);
3035 case CONSUMER_CHANNEL_ADD
:
3036 DBG("Adding channel %d to poll set",
3039 lttng_ht_node_init_u64(&chan
->wait_fd_node
,
3042 lttng_ht_add_unique_u64(channel_ht
,
3043 &chan
->wait_fd_node
);
3045 /* Add channel to the global poll events list */
3046 lttng_poll_add(&events
, chan
->wait_fd
,
3047 LPOLLERR
| LPOLLHUP
);
3049 case CONSUMER_CHANNEL_DEL
:
3052 * This command should never be called if the channel
3053 * has streams monitored by either the data or metadata
3054 * thread. The consumer only notify this thread with a
3055 * channel del. command if it receives a destroy
3056 * channel command from the session daemon that send it
3057 * if a command prior to the GET_CHANNEL failed.
3061 chan
= consumer_find_channel(key
);
3064 ERR("UST consumer get channel key %" PRIu64
" not found for del channel", key
);
3067 lttng_poll_del(&events
, chan
->wait_fd
);
3068 iter
.iter
.node
= &chan
->wait_fd_node
.node
;
3069 ret
= lttng_ht_del(channel_ht
, &iter
);
3072 switch (consumer_data
.type
) {
3073 case LTTNG_CONSUMER_KERNEL
:
3075 case LTTNG_CONSUMER32_UST
:
3076 case LTTNG_CONSUMER64_UST
:
3077 health_code_update();
3078 /* Destroy streams that might have been left in the stream list. */
3079 clean_channel_stream_list(chan
);
3082 ERR("Unknown consumer_data type");
3087 * Release our own refcount. Force channel deletion even if
3088 * streams were not initialized.
3090 if (!uatomic_sub_return(&chan
->refcount
, 1)) {
3091 consumer_del_channel(chan
);
3096 case CONSUMER_CHANNEL_QUIT
:
3098 * Remove the pipe from the poll set and continue the loop
3099 * since their might be data to consume.
3101 lttng_poll_del(&events
, ctx
->consumer_channel_pipe
[0]);
3104 ERR("Unknown action");
3107 } else if (revents
& (LPOLLERR
| LPOLLHUP
)) {
3108 DBG("Channel thread pipe hung up");
3110 * Remove the pipe from the poll set and continue the loop
3111 * since their might be data to consume.
3113 lttng_poll_del(&events
, ctx
->consumer_channel_pipe
[0]);
3116 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
3120 /* Handle other stream */
3126 uint64_t tmp_id
= (uint64_t) pollfd
;
3128 lttng_ht_lookup(channel_ht
, &tmp_id
, &iter
);
3130 node
= lttng_ht_iter_get_node_u64(&iter
);
3133 chan
= caa_container_of(node
, struct lttng_consumer_channel
,
3136 /* Check for error event */
3137 if (revents
& (LPOLLERR
| LPOLLHUP
)) {
3138 DBG("Channel fd %d is hup|err.", pollfd
);
3140 lttng_poll_del(&events
, chan
->wait_fd
);
3141 ret
= lttng_ht_del(channel_ht
, &iter
);
3145 * This will close the wait fd for each stream associated to
3146 * this channel AND monitored by the data/metadata thread thus
3147 * will be clean by the right thread.
3149 consumer_close_channel_streams(chan
);
3151 /* Release our own refcount */
3152 if (!uatomic_sub_return(&chan
->refcount
, 1)
3153 && !uatomic_read(&chan
->nb_init_stream_left
)) {
3154 consumer_del_channel(chan
);
3157 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
3162 /* Release RCU lock for the channel looked up */
3170 lttng_poll_clean(&events
);
3172 destroy_channel_ht(channel_ht
);
3175 DBG("Channel poll thread exiting");
3178 ERR("Health error occurred in %s", __func__
);
3180 health_unregister(health_consumerd
);
3181 rcu_unregister_thread();
3185 static int set_metadata_socket(struct lttng_consumer_local_data
*ctx
,
3186 struct pollfd
*sockpoll
, int client_socket
)
3193 ret
= lttng_consumer_poll_socket(sockpoll
);
3197 DBG("Metadata connection on client_socket");
3199 /* Blocking call, waiting for transmission */
3200 ctx
->consumer_metadata_socket
= lttcomm_accept_unix_sock(client_socket
);
3201 if (ctx
->consumer_metadata_socket
< 0) {
3202 WARN("On accept metadata");
3213 * This thread listens on the consumerd socket and receives the file
3214 * descriptors from the session daemon.
3216 void *consumer_thread_sessiond_poll(void *data
)
3218 int sock
= -1, client_socket
, ret
, err
= -1;
3220 * structure to poll for incoming data on communication socket avoids
3221 * making blocking sockets.
3223 struct pollfd consumer_sockpoll
[2];
3224 struct lttng_consumer_local_data
*ctx
= data
;
3226 rcu_register_thread();
3228 health_register(health_consumerd
, HEALTH_CONSUMERD_TYPE_SESSIOND
);
3230 if (testpoint(consumerd_thread_sessiond
)) {
3231 goto error_testpoint
;
3234 health_code_update();
3236 DBG("Creating command socket %s", ctx
->consumer_command_sock_path
);
3237 unlink(ctx
->consumer_command_sock_path
);
3238 client_socket
= lttcomm_create_unix_sock(ctx
->consumer_command_sock_path
);
3239 if (client_socket
< 0) {
3240 ERR("Cannot create command socket");
3244 ret
= lttcomm_listen_unix_sock(client_socket
);
3249 DBG("Sending ready command to lttng-sessiond");
3250 ret
= lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_COMMAND_SOCK_READY
);
3251 /* return < 0 on error, but == 0 is not fatal */
3253 ERR("Error sending ready command to lttng-sessiond");
3257 /* prepare the FDs to poll : to client socket and the should_quit pipe */
3258 consumer_sockpoll
[0].fd
= ctx
->consumer_should_quit
[0];
3259 consumer_sockpoll
[0].events
= POLLIN
| POLLPRI
;
3260 consumer_sockpoll
[1].fd
= client_socket
;
3261 consumer_sockpoll
[1].events
= POLLIN
| POLLPRI
;
3263 ret
= lttng_consumer_poll_socket(consumer_sockpoll
);
3271 DBG("Connection on client_socket");
3273 /* Blocking call, waiting for transmission */
3274 sock
= lttcomm_accept_unix_sock(client_socket
);
3281 * Setup metadata socket which is the second socket connection on the
3282 * command unix socket.
3284 ret
= set_metadata_socket(ctx
, consumer_sockpoll
, client_socket
);
3293 /* This socket is not useful anymore. */
3294 ret
= close(client_socket
);
3296 PERROR("close client_socket");
3300 /* update the polling structure to poll on the established socket */
3301 consumer_sockpoll
[1].fd
= sock
;
3302 consumer_sockpoll
[1].events
= POLLIN
| POLLPRI
;
3305 health_code_update();
3307 health_poll_entry();
3308 ret
= lttng_consumer_poll_socket(consumer_sockpoll
);
3317 DBG("Incoming command on sock");
3318 ret
= lttng_consumer_recv_cmd(ctx
, sock
, consumer_sockpoll
);
3321 * This could simply be a session daemon quitting. Don't output
3324 DBG("Communication interrupted on command socket");
3328 if (CMM_LOAD_SHARED(consumer_quit
)) {
3329 DBG("consumer_thread_receive_fds received quit from signal");
3330 err
= 0; /* All is OK */
3333 DBG("received command on sock");
3339 DBG("Consumer thread sessiond poll exiting");
3342 * Close metadata streams since the producer is the session daemon which
3345 * NOTE: for now, this only applies to the UST tracer.
3347 lttng_consumer_close_all_metadata();
3350 * when all fds have hung up, the polling thread
3353 CMM_STORE_SHARED(consumer_quit
, 1);
3356 * Notify the data poll thread to poll back again and test the
3357 * consumer_quit state that we just set so to quit gracefully.
3359 notify_thread_lttng_pipe(ctx
->consumer_data_pipe
);
3361 notify_channel_pipe(ctx
, NULL
, -1, CONSUMER_CHANNEL_QUIT
);
3363 notify_health_quit_pipe(health_quit_pipe
);
3365 /* Cleaning up possibly open sockets. */
3369 PERROR("close sock sessiond poll");
3372 if (client_socket
>= 0) {
3373 ret
= close(client_socket
);
3375 PERROR("close client_socket sessiond poll");
3382 ERR("Health error occurred in %s", __func__
);
3384 health_unregister(health_consumerd
);
3386 rcu_unregister_thread();
3390 ssize_t
lttng_consumer_read_subbuffer(struct lttng_consumer_stream
*stream
,
3391 struct lttng_consumer_local_data
*ctx
)
3395 bool rotated
= false;
3397 pthread_mutex_lock(&stream
->lock
);
3398 if (stream
->metadata_flag
) {
3399 pthread_mutex_lock(&stream
->metadata_rdv_lock
);
3402 switch (consumer_data
.type
) {
3403 case LTTNG_CONSUMER_KERNEL
:
3404 ret
= lttng_kconsumer_read_subbuffer(stream
, ctx
, &rotated
);
3406 case LTTNG_CONSUMER32_UST
:
3407 case LTTNG_CONSUMER64_UST
:
3408 ret
= lttng_ustconsumer_read_subbuffer(stream
, ctx
, &rotated
);
3411 ERR("Unknown consumer_data type");
3417 if (stream
->metadata_flag
) {
3418 pthread_cond_broadcast(&stream
->metadata_rdv
);
3419 pthread_mutex_unlock(&stream
->metadata_rdv_lock
);
3421 pthread_mutex_unlock(&stream
->lock
);
3423 rotate_ret
= consumer_post_rotation(stream
, ctx
);
3424 if (rotate_ret
< 0) {
3425 ERR("Failed after a rotation");
3433 int lttng_consumer_on_recv_stream(struct lttng_consumer_stream
*stream
)
3435 switch (consumer_data
.type
) {
3436 case LTTNG_CONSUMER_KERNEL
:
3437 return lttng_kconsumer_on_recv_stream(stream
);
3438 case LTTNG_CONSUMER32_UST
:
3439 case LTTNG_CONSUMER64_UST
:
3440 return lttng_ustconsumer_on_recv_stream(stream
);
3442 ERR("Unknown consumer_data type");
3449 * Allocate and set consumer data hash tables.
3451 int lttng_consumer_init(void)
3453 consumer_data
.channel_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3454 if (!consumer_data
.channel_ht
) {
3458 consumer_data
.relayd_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3459 if (!consumer_data
.relayd_ht
) {
3463 consumer_data
.stream_list_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3464 if (!consumer_data
.stream_list_ht
) {
3468 consumer_data
.stream_per_chan_id_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3469 if (!consumer_data
.stream_per_chan_id_ht
) {
3473 data_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3478 metadata_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3490 * Process the ADD_RELAYD command receive by a consumer.
3492 * This will create a relayd socket pair and add it to the relayd hash table.
3493 * The caller MUST acquire a RCU read side lock before calling it.
3495 void consumer_add_relayd_socket(uint64_t net_seq_idx
, int sock_type
,
3496 struct lttng_consumer_local_data
*ctx
, int sock
,
3497 struct pollfd
*consumer_sockpoll
,
3498 struct lttcomm_relayd_sock
*relayd_sock
, uint64_t sessiond_id
,
3499 uint64_t relayd_session_id
)
3501 int fd
= -1, ret
= -1, relayd_created
= 0;
3502 enum lttcomm_return_code ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
3503 struct consumer_relayd_sock_pair
*relayd
= NULL
;
3506 assert(relayd_sock
);
3508 DBG("Consumer adding relayd socket (idx: %" PRIu64
")", net_seq_idx
);
3510 /* Get relayd reference if exists. */
3511 relayd
= consumer_find_relayd(net_seq_idx
);
3512 if (relayd
== NULL
) {
3513 assert(sock_type
== LTTNG_STREAM_CONTROL
);
3514 /* Not found. Allocate one. */
3515 relayd
= consumer_allocate_relayd_sock_pair(net_seq_idx
);
3516 if (relayd
== NULL
) {
3517 ret_code
= LTTCOMM_CONSUMERD_ENOMEM
;
3520 relayd
->sessiond_session_id
= sessiond_id
;
3525 * This code path MUST continue to the consumer send status message to
3526 * we can notify the session daemon and continue our work without
3527 * killing everything.
3531 * relayd key should never be found for control socket.
3533 assert(sock_type
!= LTTNG_STREAM_CONTROL
);
3536 /* First send a status message before receiving the fds. */
3537 ret
= consumer_send_status_msg(sock
, LTTCOMM_CONSUMERD_SUCCESS
);
3539 /* Somehow, the session daemon is not responding anymore. */
3540 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_FATAL
);
3541 goto error_nosignal
;
3544 /* Poll on consumer socket. */
3545 ret
= lttng_consumer_poll_socket(consumer_sockpoll
);
3547 /* Needing to exit in the middle of a command: error. */
3548 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_POLL_ERROR
);
3549 goto error_nosignal
;
3552 /* Get relayd socket from session daemon */
3553 ret
= lttcomm_recv_fds_unix_sock(sock
, &fd
, 1);
3554 if (ret
!= sizeof(fd
)) {
3555 fd
= -1; /* Just in case it gets set with an invalid value. */
3558 * Failing to receive FDs might indicate a major problem such as
3559 * reaching a fd limit during the receive where the kernel returns a
3560 * MSG_CTRUNC and fails to cleanup the fd in the queue. Any case, we
3561 * don't take any chances and stop everything.
3563 * XXX: Feature request #558 will fix that and avoid this possible
3564 * issue when reaching the fd limit.
3566 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_ERROR_RECV_FD
);
3567 ret_code
= LTTCOMM_CONSUMERD_ERROR_RECV_FD
;
3571 /* Copy socket information and received FD */
3572 switch (sock_type
) {
3573 case LTTNG_STREAM_CONTROL
:
3574 /* Copy received lttcomm socket */
3575 lttcomm_copy_sock(&relayd
->control_sock
.sock
, &relayd_sock
->sock
);
3576 ret
= lttcomm_create_sock(&relayd
->control_sock
.sock
);
3577 /* Handle create_sock error. */
3579 ret_code
= LTTCOMM_CONSUMERD_ENOMEM
;
3583 * Close the socket created internally by
3584 * lttcomm_create_sock, so we can replace it by the one
3585 * received from sessiond.
3587 if (close(relayd
->control_sock
.sock
.fd
)) {
3591 /* Assign new file descriptor */
3592 relayd
->control_sock
.sock
.fd
= fd
;
3593 fd
= -1; /* For error path */
3594 /* Assign version values. */
3595 relayd
->control_sock
.major
= relayd_sock
->major
;
3596 relayd
->control_sock
.minor
= relayd_sock
->minor
;
3598 relayd
->relayd_session_id
= relayd_session_id
;
3601 case LTTNG_STREAM_DATA
:
3602 /* Copy received lttcomm socket */
3603 lttcomm_copy_sock(&relayd
->data_sock
.sock
, &relayd_sock
->sock
);
3604 ret
= lttcomm_create_sock(&relayd
->data_sock
.sock
);
3605 /* Handle create_sock error. */
3607 ret_code
= LTTCOMM_CONSUMERD_ENOMEM
;
3611 * Close the socket created internally by
3612 * lttcomm_create_sock, so we can replace it by the one
3613 * received from sessiond.
3615 if (close(relayd
->data_sock
.sock
.fd
)) {
3619 /* Assign new file descriptor */
3620 relayd
->data_sock
.sock
.fd
= fd
;
3621 fd
= -1; /* for eventual error paths */
3622 /* Assign version values. */
3623 relayd
->data_sock
.major
= relayd_sock
->major
;
3624 relayd
->data_sock
.minor
= relayd_sock
->minor
;
3627 ERR("Unknown relayd socket type (%d)", sock_type
);
3628 ret_code
= LTTCOMM_CONSUMERD_FATAL
;
3632 DBG("Consumer %s socket created successfully with net idx %" PRIu64
" (fd: %d)",
3633 sock_type
== LTTNG_STREAM_CONTROL
? "control" : "data",
3634 relayd
->net_seq_idx
, fd
);
3636 /* We successfully added the socket. Send status back. */
3637 ret
= consumer_send_status_msg(sock
, ret_code
);
3639 /* Somehow, the session daemon is not responding anymore. */
3640 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_FATAL
);
3641 goto error_nosignal
;
3645 * Add relayd socket pair to consumer data hashtable. If object already
3646 * exists or on error, the function gracefully returns.
3654 if (consumer_send_status_msg(sock
, ret_code
) < 0) {
3655 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_FATAL
);
3659 /* Close received socket if valid. */
3662 PERROR("close received socket");
3666 if (relayd_created
) {
3672 * Try to lock the stream mutex.
3674 * On success, 1 is returned else 0 indicating that the mutex is NOT lock.
3676 static int stream_try_lock(struct lttng_consumer_stream
*stream
)
3683 * Try to lock the stream mutex. On failure, we know that the stream is
3684 * being used else where hence there is data still being extracted.
3686 ret
= pthread_mutex_trylock(&stream
->lock
);
3688 /* For both EBUSY and EINVAL error, the mutex is NOT locked. */
3700 * Search for a relayd associated to the session id and return the reference.
3702 * A rcu read side lock MUST be acquire before calling this function and locked
3703 * until the relayd object is no longer necessary.
3705 static struct consumer_relayd_sock_pair
*find_relayd_by_session_id(uint64_t id
)
3707 struct lttng_ht_iter iter
;
3708 struct consumer_relayd_sock_pair
*relayd
= NULL
;
3710 /* Iterate over all relayd since they are indexed by net_seq_idx. */
3711 cds_lfht_for_each_entry(consumer_data
.relayd_ht
->ht
, &iter
.iter
, relayd
,
3714 * Check by sessiond id which is unique here where the relayd session
3715 * id might not be when having multiple relayd.
3717 if (relayd
->sessiond_session_id
== id
) {
3718 /* Found the relayd. There can be only one per id. */
3730 * Check if for a given session id there is still data needed to be extract
3733 * Return 1 if data is pending or else 0 meaning ready to be read.
3735 int consumer_data_pending(uint64_t id
)
3738 struct lttng_ht_iter iter
;
3739 struct lttng_ht
*ht
;
3740 struct lttng_consumer_stream
*stream
;
3741 struct consumer_relayd_sock_pair
*relayd
= NULL
;
3742 int (*data_pending
)(struct lttng_consumer_stream
*);
3744 DBG("Consumer data pending command on session id %" PRIu64
, id
);
3747 pthread_mutex_lock(&consumer_data
.lock
);
3749 switch (consumer_data
.type
) {
3750 case LTTNG_CONSUMER_KERNEL
:
3751 data_pending
= lttng_kconsumer_data_pending
;
3753 case LTTNG_CONSUMER32_UST
:
3754 case LTTNG_CONSUMER64_UST
:
3755 data_pending
= lttng_ustconsumer_data_pending
;
3758 ERR("Unknown consumer data type");
3762 /* Ease our life a bit */
3763 ht
= consumer_data
.stream_list_ht
;
3765 relayd
= find_relayd_by_session_id(id
);
3767 /* Send init command for data pending. */
3768 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
3769 ret
= relayd_begin_data_pending(&relayd
->control_sock
,
3770 relayd
->relayd_session_id
);
3771 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
3773 /* Communication error thus the relayd so no data pending. */
3774 goto data_not_pending
;
3778 cds_lfht_for_each_entry_duplicate(ht
->ht
,
3779 ht
->hash_fct(&id
, lttng_ht_seed
),
3781 &iter
.iter
, stream
, node_session_id
.node
) {
3782 /* If this call fails, the stream is being used hence data pending. */
3783 ret
= stream_try_lock(stream
);
3789 * A removed node from the hash table indicates that the stream has
3790 * been deleted thus having a guarantee that the buffers are closed
3791 * on the consumer side. However, data can still be transmitted
3792 * over the network so don't skip the relayd check.
3794 ret
= cds_lfht_is_node_deleted(&stream
->node
.node
);
3796 /* Check the stream if there is data in the buffers. */
3797 ret
= data_pending(stream
);
3799 pthread_mutex_unlock(&stream
->lock
);
3806 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
3807 if (stream
->metadata_flag
) {
3808 ret
= relayd_quiescent_control(&relayd
->control_sock
,
3809 stream
->relayd_stream_id
);
3811 ret
= relayd_data_pending(&relayd
->control_sock
,
3812 stream
->relayd_stream_id
,
3813 stream
->next_net_seq_num
- 1);
3815 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
3817 pthread_mutex_unlock(&stream
->lock
);
3821 pthread_mutex_unlock(&stream
->lock
);
3825 unsigned int is_data_inflight
= 0;
3827 /* Send init command for data pending. */
3828 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
3829 ret
= relayd_end_data_pending(&relayd
->control_sock
,
3830 relayd
->relayd_session_id
, &is_data_inflight
);
3831 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
3833 goto data_not_pending
;
3835 if (is_data_inflight
) {
3841 * Finding _no_ node in the hash table and no inflight data means that the
3842 * stream(s) have been removed thus data is guaranteed to be available for
3843 * analysis from the trace files.
3847 /* Data is available to be read by a viewer. */
3848 pthread_mutex_unlock(&consumer_data
.lock
);
3853 /* Data is still being extracted from buffers. */
3854 pthread_mutex_unlock(&consumer_data
.lock
);
3860 * Send a ret code status message to the sessiond daemon.
3862 * Return the sendmsg() return value.
3864 int consumer_send_status_msg(int sock
, int ret_code
)
3866 struct lttcomm_consumer_status_msg msg
;
3868 memset(&msg
, 0, sizeof(msg
));
3869 msg
.ret_code
= ret_code
;
3871 return lttcomm_send_unix_sock(sock
, &msg
, sizeof(msg
));
3875 * Send a channel status message to the sessiond daemon.
3877 * Return the sendmsg() return value.
3879 int consumer_send_status_channel(int sock
,
3880 struct lttng_consumer_channel
*channel
)
3882 struct lttcomm_consumer_status_channel msg
;
3886 memset(&msg
, 0, sizeof(msg
));
3888 msg
.ret_code
= LTTCOMM_CONSUMERD_CHANNEL_FAIL
;
3890 msg
.ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
3891 msg
.key
= channel
->key
;
3892 msg
.stream_count
= channel
->streams
.count
;
3895 return lttcomm_send_unix_sock(sock
, &msg
, sizeof(msg
));
3898 unsigned long consumer_get_consume_start_pos(unsigned long consumed_pos
,
3899 unsigned long produced_pos
, uint64_t nb_packets_per_stream
,
3900 uint64_t max_sb_size
)
3902 unsigned long start_pos
;
3904 if (!nb_packets_per_stream
) {
3905 return consumed_pos
; /* Grab everything */
3907 start_pos
= produced_pos
- offset_align_floor(produced_pos
, max_sb_size
);
3908 start_pos
-= max_sb_size
* nb_packets_per_stream
;
3909 if ((long) (start_pos
- consumed_pos
) < 0) {
3910 return consumed_pos
; /* Grab everything */
3916 int consumer_flush_buffer(struct lttng_consumer_stream
*stream
, int producer_active
)
3920 switch (consumer_data
.type
) {
3921 case LTTNG_CONSUMER_KERNEL
:
3922 ret
= kernctl_buffer_flush(stream
->wait_fd
);
3924 ERR("Failed to flush kernel stream");
3928 case LTTNG_CONSUMER32_UST
:
3929 case LTTNG_CONSUMER64_UST
:
3930 lttng_ustctl_flush_buffer(stream
, producer_active
);
3933 ERR("Unknown consumer_data type");
3942 * Sample the rotate position for all the streams of a channel. If a stream
3943 * is already at the rotate position (produced == consumed), we flag it as
3944 * ready for rotation. The rotation of ready streams occurs after we have
3945 * replied to the session daemon that we have finished sampling the positions.
3947 * Returns 0 on success, < 0 on error
3949 int lttng_consumer_rotate_channel(uint64_t key
, const char *path
,
3950 uint64_t relayd_id
, uint32_t metadata
, uint64_t new_chunk_id
,
3951 struct lttng_consumer_local_data
*ctx
)
3954 struct lttng_consumer_channel
*channel
;
3955 struct lttng_consumer_stream
*stream
;
3956 struct lttng_ht_iter iter
;
3957 struct lttng_ht
*ht
= consumer_data
.stream_per_chan_id_ht
;
3959 DBG("Consumer sample rotate position for channel %" PRIu64
, key
);
3963 channel
= consumer_find_channel(key
);
3965 ERR("No channel found for key %" PRIu64
, key
);
3970 pthread_mutex_lock(&channel
->lock
);
3971 channel
->current_chunk_id
= new_chunk_id
;
3973 ret
= lttng_strncpy(channel
->pathname
, path
, sizeof(channel
->pathname
));
3975 ERR("Failed to copy new path to channel during channel rotation");
3977 goto end_unlock_channel
;
3980 if (relayd_id
== -1ULL) {
3982 * The domain path (/ust or /kernel) has been created before, we
3983 * now need to create the last part of the path: the application/user
3984 * specific section (uid/1000/64-bit).
3986 ret
= utils_mkdir_recursive(channel
->pathname
, S_IRWXU
| S_IRWXG
,
3987 channel
->uid
, channel
->gid
);
3989 ERR("Failed to create trace directory at %s during rotation",
3992 goto end_unlock_channel
;
3996 cds_lfht_for_each_entry_duplicate(ht
->ht
,
3997 ht
->hash_fct(&channel
->key
, lttng_ht_seed
),
3998 ht
->match_fct
, &channel
->key
, &iter
.iter
,
3999 stream
, node_channel_id
.node
) {
4000 unsigned long consumed_pos
;
4002 health_code_update();
4005 * Lock stream because we are about to change its state.
4007 pthread_mutex_lock(&stream
->lock
);
4009 ret
= lttng_strncpy(stream
->channel_read_only_attributes
.path
,
4011 sizeof(stream
->channel_read_only_attributes
.path
));
4013 ERR("Failed to sample channel path name during channel rotation");
4014 goto end_unlock_stream
;
4016 ret
= lttng_consumer_sample_snapshot_positions(stream
);
4018 ERR("Failed to sample snapshot position during channel rotation");
4019 goto end_unlock_stream
;
4022 ret
= lttng_consumer_get_produced_snapshot(stream
,
4023 &stream
->rotate_position
);
4025 ERR("Failed to sample produced position during channel rotation");
4026 goto end_unlock_stream
;
4029 lttng_consumer_get_consumed_snapshot(stream
,
4031 if (consumed_pos
== stream
->rotate_position
) {
4032 stream
->rotate_ready
= true;
4034 channel
->nr_stream_rotate_pending
++;
4036 ret
= consumer_flush_buffer(stream
, 1);
4038 ERR("Failed to flush stream %" PRIu64
" during channel rotation",
4040 goto end_unlock_stream
;
4043 pthread_mutex_unlock(&stream
->lock
);
4045 pthread_mutex_unlock(&channel
->lock
);
4051 pthread_mutex_unlock(&stream
->lock
);
4053 pthread_mutex_unlock(&channel
->lock
);
4060 * Check if a stream is ready to be rotated after extracting it.
4062 * Return 1 if it is ready for rotation, 0 if it is not, a negative value on
4063 * error. Stream lock must be held.
4065 int lttng_consumer_stream_is_rotate_ready(struct lttng_consumer_stream
*stream
)
4068 unsigned long consumed_pos
;
4070 if (!stream
->rotate_position
&& !stream
->rotate_ready
) {
4075 if (stream
->rotate_ready
) {
4081 * If we don't have the rotate_ready flag, check the consumed position
4082 * to determine if we need to rotate.
4084 ret
= lttng_consumer_sample_snapshot_positions(stream
);
4086 ERR("Taking snapshot positions");
4090 ret
= lttng_consumer_get_consumed_snapshot(stream
, &consumed_pos
);
4092 ERR("Consumed snapshot position");
4096 /* Rotate position not reached yet (with check for overflow). */
4097 if ((long) (consumed_pos
- stream
->rotate_position
) < 0) {
4108 * Reset the state for a stream after a rotation occurred.
4110 void lttng_consumer_reset_stream_rotate_state(struct lttng_consumer_stream
*stream
)
4112 stream
->rotate_position
= 0;
4113 stream
->rotate_ready
= false;
4117 * Perform the rotation a local stream file.
4119 int rotate_local_stream(struct lttng_consumer_local_data
*ctx
,
4120 struct lttng_consumer_stream
*stream
)
4124 DBG("Rotate local stream: stream key %" PRIu64
", channel key %" PRIu64
" at path %s",
4127 stream
->channel_read_only_attributes
.path
);
4129 ret
= close(stream
->out_fd
);
4131 PERROR("Closing trace file (fd %d), stream %" PRIu64
,
4132 stream
->out_fd
, stream
->key
);
4137 ret
= utils_create_stream_file(
4138 stream
->channel_read_only_attributes
.path
,
4140 stream
->channel_read_only_attributes
.tracefile_size
,
4141 stream
->tracefile_count_current
,
4142 stream
->uid
, stream
->gid
, NULL
);
4144 ERR("Rotate create stream file");
4147 stream
->out_fd
= ret
;
4148 stream
->tracefile_size_current
= 0;
4150 if (!stream
->metadata_flag
) {
4151 struct lttng_index_file
*index_file
;
4153 lttng_index_file_put(stream
->index_file
);
4155 index_file
= lttng_index_file_create(
4156 stream
->channel_read_only_attributes
.path
,
4157 stream
->name
, stream
->uid
, stream
->gid
,
4158 stream
->channel_read_only_attributes
.tracefile_size
,
4159 stream
->tracefile_count_current
,
4160 CTF_INDEX_MAJOR
, CTF_INDEX_MINOR
);
4162 ERR("Create index file during rotation");
4165 stream
->index_file
= index_file
;
4166 stream
->out_fd_offset
= 0;
4180 * Perform the rotation a stream file on the relay.
4182 int rotate_relay_stream(struct lttng_consumer_local_data
*ctx
,
4183 struct lttng_consumer_stream
*stream
)
4186 struct consumer_relayd_sock_pair
*relayd
;
4188 DBG("Rotate relay stream");
4189 relayd
= consumer_find_relayd(stream
->net_seq_idx
);
4191 ERR("Failed to find relayd");
4196 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
4197 ret
= relayd_rotate_stream(&relayd
->control_sock
,
4198 stream
->relayd_stream_id
,
4199 stream
->channel_read_only_attributes
.path
,
4200 stream
->chan
->current_chunk_id
,
4201 stream
->last_sequence_number
);
4202 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
4204 ERR("Rotate relay stream");
4212 * Performs the stream rotation for the rotate session feature if needed.
4213 * It must be called with the stream lock held.
4215 * Return 0 on success, a negative number of error.
4217 int lttng_consumer_rotate_stream(struct lttng_consumer_local_data
*ctx
,
4218 struct lttng_consumer_stream
*stream
, bool *rotated
)
4222 DBG("Consumer rotate stream %" PRIu64
, stream
->key
);
4224 if (stream
->net_seq_idx
!= (uint64_t) -1ULL) {
4225 ret
= rotate_relay_stream(ctx
, stream
);
4227 ret
= rotate_local_stream(ctx
, stream
);
4230 ERR("Rotate stream");
4234 if (stream
->metadata_flag
) {
4235 switch (consumer_data
.type
) {
4236 case LTTNG_CONSUMER_KERNEL
:
4238 * Reset the position of what has been read from the metadata
4239 * cache to 0 so we can dump it again.
4241 ret
= kernctl_metadata_cache_dump(stream
->wait_fd
);
4243 ERR("Failed to dump the kernel metadata cache after rotation");
4247 case LTTNG_CONSUMER32_UST
:
4248 case LTTNG_CONSUMER64_UST
:
4250 * Reset the position pushed from the metadata cache so it
4251 * will write from the beginning on the next push.
4253 stream
->ust_metadata_pushed
= 0;
4256 ERR("Unknown consumer_data type");
4260 lttng_consumer_reset_stream_rotate_state(stream
);
4273 * Rotate all the ready streams now.
4275 * This is especially important for low throughput streams that have already
4276 * been consumed, we cannot wait for their next packet to perform the
4279 * Returns 0 on success, < 0 on error
4281 int lttng_consumer_rotate_ready_streams(uint64_t key
,
4282 struct lttng_consumer_local_data
*ctx
)
4285 struct lttng_consumer_channel
*channel
;
4286 struct lttng_consumer_stream
*stream
;
4287 struct lttng_ht_iter iter
;
4288 struct lttng_ht
*ht
= consumer_data
.stream_per_chan_id_ht
;
4292 DBG("Consumer rotate ready streams in channel %" PRIu64
, key
);
4294 channel
= consumer_find_channel(key
);
4296 ERR("No channel found for key %" PRIu64
, key
);
4301 cds_lfht_for_each_entry_duplicate(ht
->ht
,
4302 ht
->hash_fct(&channel
->key
, lttng_ht_seed
),
4303 ht
->match_fct
, &channel
->key
, &iter
.iter
,
4304 stream
, node_channel_id
.node
) {
4305 health_code_update();
4307 pthread_mutex_lock(&stream
->lock
);
4309 if (!stream
->rotate_ready
) {
4310 pthread_mutex_unlock(&stream
->lock
);
4313 DBG("Consumer rotate ready stream %" PRIu64
, stream
->key
);
4315 ret
= lttng_consumer_rotate_stream(ctx
, stream
, NULL
);
4316 pthread_mutex_unlock(&stream
->lock
);
4321 ret
= consumer_post_rotation(stream
, ctx
);
4335 int rotate_rename_local(const char *old_path
, const char *new_path
,
4336 uid_t uid
, gid_t gid
)
4343 ret
= utils_mkdir_recursive(new_path
, S_IRWXU
| S_IRWXG
, uid
, gid
);
4345 ERR("Create directory on rotate");
4349 ret
= rename(old_path
, new_path
);
4350 if (ret
< 0 && errno
!= ENOENT
) {
4351 PERROR("Rename completed rotation chunk");
4361 int rotate_rename_relay(const char *old_path
, const char *new_path
,
4365 struct consumer_relayd_sock_pair
*relayd
;
4367 relayd
= consumer_find_relayd(relayd_id
);
4369 ERR("Failed to find relayd while running rotate_rename_relay command");
4374 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
4375 ret
= relayd_rotate_rename(&relayd
->control_sock
, old_path
, new_path
);
4376 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
4381 int lttng_consumer_rotate_rename(const char *old_path
, const char *new_path
,
4382 uid_t uid
, gid_t gid
, uint64_t relayd_id
)
4384 if (relayd_id
!= -1ULL) {
4385 return rotate_rename_relay(old_path
, new_path
, relayd_id
);
4387 return rotate_rename_local(old_path
, new_path
, uid
, gid
);
4391 int lttng_consumer_rotate_pending_relay(uint64_t session_id
,
4392 uint64_t relayd_id
, uint64_t chunk_id
)
4395 struct consumer_relayd_sock_pair
*relayd
;
4397 relayd
= consumer_find_relayd(relayd_id
);
4399 ERR("Failed to find relayd");
4404 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
4405 ret
= relayd_rotate_pending(&relayd
->control_sock
, chunk_id
);
4406 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
4413 int mkdir_local(const char *path
, uid_t uid
, gid_t gid
)
4417 ret
= utils_mkdir_recursive(path
, S_IRWXU
| S_IRWXG
, uid
, gid
);
4419 /* utils_mkdir_recursive logs an error. */
4429 int mkdir_relay(const char *path
, uint64_t relayd_id
)
4432 struct consumer_relayd_sock_pair
*relayd
;
4434 relayd
= consumer_find_relayd(relayd_id
);
4436 ERR("Failed to find relayd");
4441 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
4442 ret
= relayd_mkdir(&relayd
->control_sock
, path
);
4443 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
4450 int lttng_consumer_mkdir(const char *path
, uid_t uid
, gid_t gid
,
4453 if (relayd_id
!= -1ULL) {
4454 return mkdir_relay(path
, relayd_id
);
4456 return mkdir_local(path
, uid
, gid
);