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
);
326 pthread_mutex_destroy(&relayd
->ctrl_sock_mutex
);
331 * Destroy and free relayd socket pair object.
333 void consumer_destroy_relayd(struct consumer_relayd_sock_pair
*relayd
)
336 struct lttng_ht_iter iter
;
338 if (relayd
== NULL
) {
342 DBG("Consumer destroy and close relayd socket pair");
344 iter
.iter
.node
= &relayd
->node
.node
;
345 ret
= lttng_ht_del(consumer_data
.relayd_ht
, &iter
);
347 /* We assume the relayd is being or is destroyed */
351 /* RCU free() call */
352 call_rcu(&relayd
->node
.head
, free_relayd_rcu
);
356 * Remove a channel from the global list protected by a mutex. This function is
357 * also responsible for freeing its data structures.
359 void consumer_del_channel(struct lttng_consumer_channel
*channel
)
362 struct lttng_ht_iter iter
;
364 DBG("Consumer delete channel key %" PRIu64
, channel
->key
);
366 pthread_mutex_lock(&consumer_data
.lock
);
367 pthread_mutex_lock(&channel
->lock
);
369 /* Destroy streams that might have been left in the stream list. */
370 clean_channel_stream_list(channel
);
372 if (channel
->live_timer_enabled
== 1) {
373 consumer_timer_live_stop(channel
);
375 if (channel
->monitor_timer_enabled
== 1) {
376 consumer_timer_monitor_stop(channel
);
379 switch (consumer_data
.type
) {
380 case LTTNG_CONSUMER_KERNEL
:
382 case LTTNG_CONSUMER32_UST
:
383 case LTTNG_CONSUMER64_UST
:
384 lttng_ustconsumer_del_channel(channel
);
387 ERR("Unknown consumer_data type");
393 iter
.iter
.node
= &channel
->node
.node
;
394 ret
= lttng_ht_del(consumer_data
.channel_ht
, &iter
);
398 call_rcu(&channel
->node
.head
, free_channel_rcu
);
400 pthread_mutex_unlock(&channel
->lock
);
401 pthread_mutex_unlock(&consumer_data
.lock
);
405 * Iterate over the relayd hash table and destroy each element. Finally,
406 * destroy the whole hash table.
408 static void cleanup_relayd_ht(void)
410 struct lttng_ht_iter iter
;
411 struct consumer_relayd_sock_pair
*relayd
;
415 cds_lfht_for_each_entry(consumer_data
.relayd_ht
->ht
, &iter
.iter
, relayd
,
417 consumer_destroy_relayd(relayd
);
422 lttng_ht_destroy(consumer_data
.relayd_ht
);
426 * Update the end point status of all streams having the given network sequence
427 * index (relayd index).
429 * It's atomically set without having the stream mutex locked which is fine
430 * because we handle the write/read race with a pipe wakeup for each thread.
432 static void update_endpoint_status_by_netidx(uint64_t net_seq_idx
,
433 enum consumer_endpoint_status status
)
435 struct lttng_ht_iter iter
;
436 struct lttng_consumer_stream
*stream
;
438 DBG("Consumer set delete flag on stream by idx %" PRIu64
, net_seq_idx
);
442 /* Let's begin with metadata */
443 cds_lfht_for_each_entry(metadata_ht
->ht
, &iter
.iter
, stream
, node
.node
) {
444 if (stream
->net_seq_idx
== net_seq_idx
) {
445 uatomic_set(&stream
->endpoint_status
, status
);
446 DBG("Delete flag set to metadata stream %d", stream
->wait_fd
);
450 /* Follow up by the data streams */
451 cds_lfht_for_each_entry(data_ht
->ht
, &iter
.iter
, stream
, node
.node
) {
452 if (stream
->net_seq_idx
== net_seq_idx
) {
453 uatomic_set(&stream
->endpoint_status
, status
);
454 DBG("Delete flag set to data stream %d", stream
->wait_fd
);
461 * Cleanup a relayd object by flagging every associated streams for deletion,
462 * destroying the object meaning removing it from the relayd hash table,
463 * closing the sockets and freeing the memory in a RCU call.
465 * If a local data context is available, notify the threads that the streams'
466 * state have changed.
468 static void cleanup_relayd(struct consumer_relayd_sock_pair
*relayd
,
469 struct lttng_consumer_local_data
*ctx
)
475 DBG("Cleaning up relayd sockets");
477 /* Save the net sequence index before destroying the object */
478 netidx
= relayd
->net_seq_idx
;
481 * Delete the relayd from the relayd hash table, close the sockets and free
482 * the object in a RCU call.
484 consumer_destroy_relayd(relayd
);
486 /* Set inactive endpoint to all streams */
487 update_endpoint_status_by_netidx(netidx
, CONSUMER_ENDPOINT_INACTIVE
);
490 * With a local data context, notify the threads that the streams' state
491 * have changed. The write() action on the pipe acts as an "implicit"
492 * memory barrier ordering the updates of the end point status from the
493 * read of this status which happens AFTER receiving this notify.
496 notify_thread_lttng_pipe(ctx
->consumer_data_pipe
);
497 notify_thread_lttng_pipe(ctx
->consumer_metadata_pipe
);
502 * Flag a relayd socket pair for destruction. Destroy it if the refcount
505 * RCU read side lock MUST be aquired before calling this function.
507 void consumer_flag_relayd_for_destroy(struct consumer_relayd_sock_pair
*relayd
)
511 /* Set destroy flag for this object */
512 uatomic_set(&relayd
->destroy_flag
, 1);
514 /* Destroy the relayd if refcount is 0 */
515 if (uatomic_read(&relayd
->refcount
) == 0) {
516 consumer_destroy_relayd(relayd
);
521 * Completly destroy stream from every visiable data structure and the given
524 * One this call returns, the stream object is not longer usable nor visible.
526 void consumer_del_stream(struct lttng_consumer_stream
*stream
,
529 consumer_stream_destroy(stream
, ht
);
533 * XXX naming of del vs destroy is all mixed up.
535 void consumer_del_stream_for_data(struct lttng_consumer_stream
*stream
)
537 consumer_stream_destroy(stream
, data_ht
);
540 void consumer_del_stream_for_metadata(struct lttng_consumer_stream
*stream
)
542 consumer_stream_destroy(stream
, metadata_ht
);
545 void consumer_stream_update_channel_attributes(
546 struct lttng_consumer_stream
*stream
,
547 struct lttng_consumer_channel
*channel
)
549 stream
->channel_read_only_attributes
.tracefile_size
=
550 channel
->tracefile_size
;
551 memcpy(stream
->channel_read_only_attributes
.path
, channel
->pathname
,
552 sizeof(stream
->channel_read_only_attributes
.path
));
555 struct lttng_consumer_stream
*consumer_allocate_stream(uint64_t channel_key
,
557 enum lttng_consumer_stream_state state
,
558 const char *channel_name
,
565 enum consumer_channel_type type
,
566 unsigned int monitor
,
567 uint64_t trace_archive_id
)
570 struct lttng_consumer_stream
*stream
;
572 stream
= zmalloc(sizeof(*stream
));
573 if (stream
== NULL
) {
574 PERROR("malloc struct lttng_consumer_stream");
581 stream
->key
= stream_key
;
583 stream
->out_fd_offset
= 0;
584 stream
->output_written
= 0;
585 stream
->state
= state
;
588 stream
->net_seq_idx
= relayd_id
;
589 stream
->session_id
= session_id
;
590 stream
->monitor
= monitor
;
591 stream
->endpoint_status
= CONSUMER_ENDPOINT_ACTIVE
;
592 stream
->index_file
= NULL
;
593 stream
->last_sequence_number
= -1ULL;
594 stream
->trace_archive_id
= trace_archive_id
;
595 pthread_mutex_init(&stream
->lock
, NULL
);
596 pthread_mutex_init(&stream
->metadata_timer_lock
, NULL
);
598 /* If channel is the metadata, flag this stream as metadata. */
599 if (type
== CONSUMER_CHANNEL_TYPE_METADATA
) {
600 stream
->metadata_flag
= 1;
601 /* Metadata is flat out. */
602 strncpy(stream
->name
, DEFAULT_METADATA_NAME
, sizeof(stream
->name
));
603 /* Live rendez-vous point. */
604 pthread_cond_init(&stream
->metadata_rdv
, NULL
);
605 pthread_mutex_init(&stream
->metadata_rdv_lock
, NULL
);
607 /* Format stream name to <channel_name>_<cpu_number> */
608 ret
= snprintf(stream
->name
, sizeof(stream
->name
), "%s_%d",
611 PERROR("snprintf stream name");
616 /* Key is always the wait_fd for streams. */
617 lttng_ht_node_init_u64(&stream
->node
, stream
->key
);
619 /* Init node per channel id key */
620 lttng_ht_node_init_u64(&stream
->node_channel_id
, channel_key
);
622 /* Init session id node with the stream session id */
623 lttng_ht_node_init_u64(&stream
->node_session_id
, stream
->session_id
);
625 DBG3("Allocated stream %s (key %" PRIu64
", chan_key %" PRIu64
626 " relayd_id %" PRIu64
", session_id %" PRIu64
,
627 stream
->name
, stream
->key
, channel_key
,
628 stream
->net_seq_idx
, stream
->session_id
);
644 * Add a stream to the global list protected by a mutex.
646 void consumer_add_data_stream(struct lttng_consumer_stream
*stream
)
648 struct lttng_ht
*ht
= data_ht
;
653 DBG3("Adding consumer stream %" PRIu64
, stream
->key
);
655 pthread_mutex_lock(&consumer_data
.lock
);
656 pthread_mutex_lock(&stream
->chan
->lock
);
657 pthread_mutex_lock(&stream
->chan
->timer_lock
);
658 pthread_mutex_lock(&stream
->lock
);
661 /* Steal stream identifier to avoid having streams with the same key */
662 steal_stream_key(stream
->key
, ht
);
664 lttng_ht_add_unique_u64(ht
, &stream
->node
);
666 lttng_ht_add_u64(consumer_data
.stream_per_chan_id_ht
,
667 &stream
->node_channel_id
);
670 * Add stream to the stream_list_ht of the consumer data. No need to steal
671 * the key since the HT does not use it and we allow to add redundant keys
674 lttng_ht_add_u64(consumer_data
.stream_list_ht
, &stream
->node_session_id
);
677 * When nb_init_stream_left reaches 0, we don't need to trigger any action
678 * in terms of destroying the associated channel, because the action that
679 * causes the count to become 0 also causes a stream to be added. The
680 * channel deletion will thus be triggered by the following removal of this
683 if (uatomic_read(&stream
->chan
->nb_init_stream_left
) > 0) {
684 /* Increment refcount before decrementing nb_init_stream_left */
686 uatomic_dec(&stream
->chan
->nb_init_stream_left
);
689 /* Update consumer data once the node is inserted. */
690 consumer_data
.stream_count
++;
691 consumer_data
.need_update
= 1;
694 pthread_mutex_unlock(&stream
->lock
);
695 pthread_mutex_unlock(&stream
->chan
->timer_lock
);
696 pthread_mutex_unlock(&stream
->chan
->lock
);
697 pthread_mutex_unlock(&consumer_data
.lock
);
700 void consumer_del_data_stream(struct lttng_consumer_stream
*stream
)
702 consumer_del_stream(stream
, data_ht
);
706 * Add relayd socket to global consumer data hashtable. RCU read side lock MUST
707 * be acquired before calling this.
709 static int add_relayd(struct consumer_relayd_sock_pair
*relayd
)
712 struct lttng_ht_node_u64
*node
;
713 struct lttng_ht_iter iter
;
717 lttng_ht_lookup(consumer_data
.relayd_ht
,
718 &relayd
->net_seq_idx
, &iter
);
719 node
= lttng_ht_iter_get_node_u64(&iter
);
723 lttng_ht_add_unique_u64(consumer_data
.relayd_ht
, &relayd
->node
);
730 * Allocate and return a consumer relayd socket.
732 static struct consumer_relayd_sock_pair
*consumer_allocate_relayd_sock_pair(
733 uint64_t net_seq_idx
)
735 struct consumer_relayd_sock_pair
*obj
= NULL
;
737 /* net sequence index of -1 is a failure */
738 if (net_seq_idx
== (uint64_t) -1ULL) {
742 obj
= zmalloc(sizeof(struct consumer_relayd_sock_pair
));
744 PERROR("zmalloc relayd sock");
748 obj
->net_seq_idx
= net_seq_idx
;
750 obj
->destroy_flag
= 0;
751 obj
->control_sock
.sock
.fd
= -1;
752 obj
->data_sock
.sock
.fd
= -1;
753 lttng_ht_node_init_u64(&obj
->node
, obj
->net_seq_idx
);
754 pthread_mutex_init(&obj
->ctrl_sock_mutex
, NULL
);
761 * Find a relayd socket pair in the global consumer data.
763 * Return the object if found else NULL.
764 * RCU read-side lock must be held across this call and while using the
767 struct consumer_relayd_sock_pair
*consumer_find_relayd(uint64_t key
)
769 struct lttng_ht_iter iter
;
770 struct lttng_ht_node_u64
*node
;
771 struct consumer_relayd_sock_pair
*relayd
= NULL
;
773 /* Negative keys are lookup failures */
774 if (key
== (uint64_t) -1ULL) {
778 lttng_ht_lookup(consumer_data
.relayd_ht
, &key
,
780 node
= lttng_ht_iter_get_node_u64(&iter
);
782 relayd
= caa_container_of(node
, struct consumer_relayd_sock_pair
, node
);
790 * Find a relayd and send the stream
792 * Returns 0 on success, < 0 on error
794 int consumer_send_relayd_stream(struct lttng_consumer_stream
*stream
,
798 struct consumer_relayd_sock_pair
*relayd
;
801 assert(stream
->net_seq_idx
!= -1ULL);
804 /* The stream is not metadata. Get relayd reference if exists. */
806 relayd
= consumer_find_relayd(stream
->net_seq_idx
);
807 if (relayd
!= NULL
) {
808 /* Add stream on the relayd */
809 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
810 ret
= relayd_add_stream(&relayd
->control_sock
, stream
->name
,
811 path
, &stream
->relayd_stream_id
,
812 stream
->chan
->tracefile_size
, stream
->chan
->tracefile_count
,
813 stream
->trace_archive_id
);
814 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
819 uatomic_inc(&relayd
->refcount
);
820 stream
->sent_to_relayd
= 1;
822 ERR("Stream %" PRIu64
" relayd ID %" PRIu64
" unknown. Can't send it.",
823 stream
->key
, stream
->net_seq_idx
);
828 DBG("Stream %s with key %" PRIu64
" sent to relayd id %" PRIu64
,
829 stream
->name
, stream
->key
, stream
->net_seq_idx
);
837 * Find a relayd and send the streams sent message
839 * Returns 0 on success, < 0 on error
841 int consumer_send_relayd_streams_sent(uint64_t net_seq_idx
)
844 struct consumer_relayd_sock_pair
*relayd
;
846 assert(net_seq_idx
!= -1ULL);
848 /* The stream is not metadata. Get relayd reference if exists. */
850 relayd
= consumer_find_relayd(net_seq_idx
);
851 if (relayd
!= NULL
) {
852 /* Add stream on the relayd */
853 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
854 ret
= relayd_streams_sent(&relayd
->control_sock
);
855 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
860 ERR("Relayd ID %" PRIu64
" unknown. Can't send streams_sent.",
867 DBG("All streams sent relayd id %" PRIu64
, net_seq_idx
);
875 * Find a relayd and close the stream
877 void close_relayd_stream(struct lttng_consumer_stream
*stream
)
879 struct consumer_relayd_sock_pair
*relayd
;
881 /* The stream is not metadata. Get relayd reference if exists. */
883 relayd
= consumer_find_relayd(stream
->net_seq_idx
);
885 consumer_stream_relayd_close(stream
, relayd
);
891 * Handle stream for relayd transmission if the stream applies for network
892 * streaming where the net sequence index is set.
894 * Return destination file descriptor or negative value on error.
896 static int write_relayd_stream_header(struct lttng_consumer_stream
*stream
,
897 size_t data_size
, unsigned long padding
,
898 struct consumer_relayd_sock_pair
*relayd
)
901 struct lttcomm_relayd_data_hdr data_hdr
;
907 /* Reset data header */
908 memset(&data_hdr
, 0, sizeof(data_hdr
));
910 if (stream
->metadata_flag
) {
911 /* Caller MUST acquire the relayd control socket lock */
912 ret
= relayd_send_metadata(&relayd
->control_sock
, data_size
);
917 /* Metadata are always sent on the control socket. */
918 outfd
= relayd
->control_sock
.sock
.fd
;
920 /* Set header with stream information */
921 data_hdr
.stream_id
= htobe64(stream
->relayd_stream_id
);
922 data_hdr
.data_size
= htobe32(data_size
);
923 data_hdr
.padding_size
= htobe32(padding
);
925 * Note that net_seq_num below is assigned with the *current* value of
926 * next_net_seq_num and only after that the next_net_seq_num will be
927 * increment. This is why when issuing a command on the relayd using
928 * this next value, 1 should always be substracted in order to compare
929 * the last seen sequence number on the relayd side to the last sent.
931 data_hdr
.net_seq_num
= htobe64(stream
->next_net_seq_num
);
932 /* Other fields are zeroed previously */
934 ret
= relayd_send_data_hdr(&relayd
->data_sock
, &data_hdr
,
940 ++stream
->next_net_seq_num
;
942 /* Set to go on data socket */
943 outfd
= relayd
->data_sock
.sock
.fd
;
951 * Allocate and return a new lttng_consumer_channel object using the given key
952 * to initialize the hash table node.
954 * On error, return NULL.
956 struct lttng_consumer_channel
*consumer_allocate_channel(uint64_t key
,
958 const char *pathname
,
963 enum lttng_event_output output
,
964 uint64_t tracefile_size
,
965 uint64_t tracefile_count
,
966 uint64_t session_id_per_pid
,
967 unsigned int monitor
,
968 unsigned int live_timer_interval
,
969 const char *root_shm_path
,
970 const char *shm_path
)
972 struct lttng_consumer_channel
*channel
;
974 channel
= zmalloc(sizeof(*channel
));
975 if (channel
== NULL
) {
976 PERROR("malloc struct lttng_consumer_channel");
981 channel
->refcount
= 0;
982 channel
->session_id
= session_id
;
983 channel
->session_id_per_pid
= session_id_per_pid
;
986 channel
->relayd_id
= relayd_id
;
987 channel
->tracefile_size
= tracefile_size
;
988 channel
->tracefile_count
= tracefile_count
;
989 channel
->monitor
= monitor
;
990 channel
->live_timer_interval
= live_timer_interval
;
991 pthread_mutex_init(&channel
->lock
, NULL
);
992 pthread_mutex_init(&channel
->timer_lock
, NULL
);
995 case LTTNG_EVENT_SPLICE
:
996 channel
->output
= CONSUMER_CHANNEL_SPLICE
;
998 case LTTNG_EVENT_MMAP
:
999 channel
->output
= CONSUMER_CHANNEL_MMAP
;
1009 * In monitor mode, the streams associated with the channel will be put in
1010 * a special list ONLY owned by this channel. So, the refcount is set to 1
1011 * here meaning that the channel itself has streams that are referenced.
1013 * On a channel deletion, once the channel is no longer visible, the
1014 * refcount is decremented and checked for a zero value to delete it. With
1015 * streams in no monitor mode, it will now be safe to destroy the channel.
1017 if (!channel
->monitor
) {
1018 channel
->refcount
= 1;
1021 strncpy(channel
->pathname
, pathname
, sizeof(channel
->pathname
));
1022 channel
->pathname
[sizeof(channel
->pathname
) - 1] = '\0';
1024 strncpy(channel
->name
, name
, sizeof(channel
->name
));
1025 channel
->name
[sizeof(channel
->name
) - 1] = '\0';
1027 if (root_shm_path
) {
1028 strncpy(channel
->root_shm_path
, root_shm_path
, sizeof(channel
->root_shm_path
));
1029 channel
->root_shm_path
[sizeof(channel
->root_shm_path
) - 1] = '\0';
1032 strncpy(channel
->shm_path
, shm_path
, sizeof(channel
->shm_path
));
1033 channel
->shm_path
[sizeof(channel
->shm_path
) - 1] = '\0';
1036 lttng_ht_node_init_u64(&channel
->node
, channel
->key
);
1038 channel
->wait_fd
= -1;
1040 CDS_INIT_LIST_HEAD(&channel
->streams
.head
);
1042 DBG("Allocated channel (key %" PRIu64
")", channel
->key
);
1049 * Add a channel to the global list protected by a mutex.
1051 * Always return 0 indicating success.
1053 int consumer_add_channel(struct lttng_consumer_channel
*channel
,
1054 struct lttng_consumer_local_data
*ctx
)
1056 pthread_mutex_lock(&consumer_data
.lock
);
1057 pthread_mutex_lock(&channel
->lock
);
1058 pthread_mutex_lock(&channel
->timer_lock
);
1061 * This gives us a guarantee that the channel we are about to add to the
1062 * channel hash table will be unique. See this function comment on the why
1063 * we need to steel the channel key at this stage.
1065 steal_channel_key(channel
->key
);
1068 lttng_ht_add_unique_u64(consumer_data
.channel_ht
, &channel
->node
);
1071 pthread_mutex_unlock(&channel
->timer_lock
);
1072 pthread_mutex_unlock(&channel
->lock
);
1073 pthread_mutex_unlock(&consumer_data
.lock
);
1075 if (channel
->wait_fd
!= -1 && channel
->type
== CONSUMER_CHANNEL_TYPE_DATA
) {
1076 notify_channel_pipe(ctx
, channel
, -1, CONSUMER_CHANNEL_ADD
);
1083 * Allocate the pollfd structure and the local view of the out fds to avoid
1084 * doing a lookup in the linked list and concurrency issues when writing is
1085 * needed. Called with consumer_data.lock held.
1087 * Returns the number of fds in the structures.
1089 static int update_poll_array(struct lttng_consumer_local_data
*ctx
,
1090 struct pollfd
**pollfd
, struct lttng_consumer_stream
**local_stream
,
1091 struct lttng_ht
*ht
, int *nb_inactive_fd
)
1094 struct lttng_ht_iter iter
;
1095 struct lttng_consumer_stream
*stream
;
1100 assert(local_stream
);
1102 DBG("Updating poll fd array");
1103 *nb_inactive_fd
= 0;
1105 cds_lfht_for_each_entry(ht
->ht
, &iter
.iter
, stream
, node
.node
) {
1107 * Only active streams with an active end point can be added to the
1108 * poll set and local stream storage of the thread.
1110 * There is a potential race here for endpoint_status to be updated
1111 * just after the check. However, this is OK since the stream(s) will
1112 * be deleted once the thread is notified that the end point state has
1113 * changed where this function will be called back again.
1115 * We track the number of inactive FDs because they still need to be
1116 * closed by the polling thread after a wakeup on the data_pipe or
1119 if (stream
->state
!= LTTNG_CONSUMER_ACTIVE_STREAM
||
1120 stream
->endpoint_status
== CONSUMER_ENDPOINT_INACTIVE
) {
1121 (*nb_inactive_fd
)++;
1125 * This clobbers way too much the debug output. Uncomment that if you
1126 * need it for debugging purposes.
1128 * DBG("Active FD %d", stream->wait_fd);
1130 (*pollfd
)[i
].fd
= stream
->wait_fd
;
1131 (*pollfd
)[i
].events
= POLLIN
| POLLPRI
;
1132 local_stream
[i
] = stream
;
1138 * Insert the consumer_data_pipe at the end of the array and don't
1139 * increment i so nb_fd is the number of real FD.
1141 (*pollfd
)[i
].fd
= lttng_pipe_get_readfd(ctx
->consumer_data_pipe
);
1142 (*pollfd
)[i
].events
= POLLIN
| POLLPRI
;
1144 (*pollfd
)[i
+ 1].fd
= lttng_pipe_get_readfd(ctx
->consumer_wakeup_pipe
);
1145 (*pollfd
)[i
+ 1].events
= POLLIN
| POLLPRI
;
1150 * Poll on the should_quit pipe and the command socket return -1 on
1151 * error, 1 if should exit, 0 if data is available on the command socket
1153 int lttng_consumer_poll_socket(struct pollfd
*consumer_sockpoll
)
1158 num_rdy
= poll(consumer_sockpoll
, 2, -1);
1159 if (num_rdy
== -1) {
1161 * Restart interrupted system call.
1163 if (errno
== EINTR
) {
1166 PERROR("Poll error");
1169 if (consumer_sockpoll
[0].revents
& (POLLIN
| POLLPRI
)) {
1170 DBG("consumer_should_quit wake up");
1177 * Set the error socket.
1179 void lttng_consumer_set_error_sock(struct lttng_consumer_local_data
*ctx
,
1182 ctx
->consumer_error_socket
= sock
;
1186 * Set the command socket path.
1188 void lttng_consumer_set_command_sock_path(
1189 struct lttng_consumer_local_data
*ctx
, char *sock
)
1191 ctx
->consumer_command_sock_path
= sock
;
1195 * Send return code to the session daemon.
1196 * If the socket is not defined, we return 0, it is not a fatal error
1198 int lttng_consumer_send_error(struct lttng_consumer_local_data
*ctx
, int cmd
)
1200 if (ctx
->consumer_error_socket
> 0) {
1201 return lttcomm_send_unix_sock(ctx
->consumer_error_socket
, &cmd
,
1202 sizeof(enum lttcomm_sessiond_command
));
1209 * Close all the tracefiles and stream fds and MUST be called when all
1210 * instances are destroyed i.e. when all threads were joined and are ended.
1212 void lttng_consumer_cleanup(void)
1214 struct lttng_ht_iter iter
;
1215 struct lttng_consumer_channel
*channel
;
1219 cds_lfht_for_each_entry(consumer_data
.channel_ht
->ht
, &iter
.iter
, channel
,
1221 consumer_del_channel(channel
);
1226 lttng_ht_destroy(consumer_data
.channel_ht
);
1228 cleanup_relayd_ht();
1230 lttng_ht_destroy(consumer_data
.stream_per_chan_id_ht
);
1233 * This HT contains streams that are freed by either the metadata thread or
1234 * the data thread so we do *nothing* on the hash table and simply destroy
1237 lttng_ht_destroy(consumer_data
.stream_list_ht
);
1241 * Called from signal handler.
1243 void lttng_consumer_should_exit(struct lttng_consumer_local_data
*ctx
)
1247 CMM_STORE_SHARED(consumer_quit
, 1);
1248 ret
= lttng_write(ctx
->consumer_should_quit
[1], "4", 1);
1250 PERROR("write consumer quit");
1253 DBG("Consumer flag that it should quit");
1258 * Flush pending writes to trace output disk file.
1261 void lttng_consumer_sync_trace_file(struct lttng_consumer_stream
*stream
,
1265 int outfd
= stream
->out_fd
;
1268 * This does a blocking write-and-wait on any page that belongs to the
1269 * subbuffer prior to the one we just wrote.
1270 * Don't care about error values, as these are just hints and ways to
1271 * limit the amount of page cache used.
1273 if (orig_offset
< stream
->max_sb_size
) {
1276 lttng_sync_file_range(outfd
, orig_offset
- stream
->max_sb_size
,
1277 stream
->max_sb_size
,
1278 SYNC_FILE_RANGE_WAIT_BEFORE
1279 | SYNC_FILE_RANGE_WRITE
1280 | SYNC_FILE_RANGE_WAIT_AFTER
);
1282 * Give hints to the kernel about how we access the file:
1283 * POSIX_FADV_DONTNEED : we won't re-access data in a near future after
1286 * We need to call fadvise again after the file grows because the
1287 * kernel does not seem to apply fadvise to non-existing parts of the
1290 * Call fadvise _after_ having waited for the page writeback to
1291 * complete because the dirty page writeback semantic is not well
1292 * defined. So it can be expected to lead to lower throughput in
1295 ret
= posix_fadvise(outfd
, orig_offset
- stream
->max_sb_size
,
1296 stream
->max_sb_size
, POSIX_FADV_DONTNEED
);
1297 if (ret
&& ret
!= -ENOSYS
) {
1299 PERROR("posix_fadvise on fd %i", outfd
);
1304 * Initialise the necessary environnement :
1305 * - create a new context
1306 * - create the poll_pipe
1307 * - create the should_quit pipe (for signal handler)
1308 * - create the thread pipe (for splice)
1310 * Takes a function pointer as argument, this function is called when data is
1311 * available on a buffer. This function is responsible to do the
1312 * kernctl_get_next_subbuf, read the data with mmap or splice depending on the
1313 * buffer configuration and then kernctl_put_next_subbuf at the end.
1315 * Returns a pointer to the new context or NULL on error.
1317 struct lttng_consumer_local_data
*lttng_consumer_create(
1318 enum lttng_consumer_type type
,
1319 ssize_t (*buffer_ready
)(struct lttng_consumer_stream
*stream
,
1320 struct lttng_consumer_local_data
*ctx
),
1321 int (*recv_channel
)(struct lttng_consumer_channel
*channel
),
1322 int (*recv_stream
)(struct lttng_consumer_stream
*stream
),
1323 int (*update_stream
)(uint64_t stream_key
, uint32_t state
))
1326 struct lttng_consumer_local_data
*ctx
;
1328 assert(consumer_data
.type
== LTTNG_CONSUMER_UNKNOWN
||
1329 consumer_data
.type
== type
);
1330 consumer_data
.type
= type
;
1332 ctx
= zmalloc(sizeof(struct lttng_consumer_local_data
));
1334 PERROR("allocating context");
1338 ctx
->consumer_error_socket
= -1;
1339 ctx
->consumer_metadata_socket
= -1;
1340 pthread_mutex_init(&ctx
->metadata_socket_lock
, NULL
);
1341 /* assign the callbacks */
1342 ctx
->on_buffer_ready
= buffer_ready
;
1343 ctx
->on_recv_channel
= recv_channel
;
1344 ctx
->on_recv_stream
= recv_stream
;
1345 ctx
->on_update_stream
= update_stream
;
1347 ctx
->consumer_data_pipe
= lttng_pipe_open(0);
1348 if (!ctx
->consumer_data_pipe
) {
1349 goto error_poll_pipe
;
1352 ctx
->consumer_wakeup_pipe
= lttng_pipe_open(0);
1353 if (!ctx
->consumer_wakeup_pipe
) {
1354 goto error_wakeup_pipe
;
1357 ret
= pipe(ctx
->consumer_should_quit
);
1359 PERROR("Error creating recv pipe");
1360 goto error_quit_pipe
;
1363 ret
= pipe(ctx
->consumer_channel_pipe
);
1365 PERROR("Error creating channel pipe");
1366 goto error_channel_pipe
;
1369 ctx
->consumer_metadata_pipe
= lttng_pipe_open(0);
1370 if (!ctx
->consumer_metadata_pipe
) {
1371 goto error_metadata_pipe
;
1374 ctx
->channel_monitor_pipe
= -1;
1378 error_metadata_pipe
:
1379 utils_close_pipe(ctx
->consumer_channel_pipe
);
1381 utils_close_pipe(ctx
->consumer_should_quit
);
1383 lttng_pipe_destroy(ctx
->consumer_wakeup_pipe
);
1385 lttng_pipe_destroy(ctx
->consumer_data_pipe
);
1393 * Iterate over all streams of the hashtable and free them properly.
1395 static void destroy_data_stream_ht(struct lttng_ht
*ht
)
1397 struct lttng_ht_iter iter
;
1398 struct lttng_consumer_stream
*stream
;
1405 cds_lfht_for_each_entry(ht
->ht
, &iter
.iter
, stream
, node
.node
) {
1407 * Ignore return value since we are currently cleaning up so any error
1410 (void) consumer_del_stream(stream
, ht
);
1414 lttng_ht_destroy(ht
);
1418 * Iterate over all streams of the metadata hashtable and free them
1421 static void destroy_metadata_stream_ht(struct lttng_ht
*ht
)
1423 struct lttng_ht_iter iter
;
1424 struct lttng_consumer_stream
*stream
;
1431 cds_lfht_for_each_entry(ht
->ht
, &iter
.iter
, stream
, node
.node
) {
1433 * Ignore return value since we are currently cleaning up so any error
1436 (void) consumer_del_metadata_stream(stream
, ht
);
1440 lttng_ht_destroy(ht
);
1444 * Close all fds associated with the instance and free the context.
1446 void lttng_consumer_destroy(struct lttng_consumer_local_data
*ctx
)
1450 DBG("Consumer destroying it. Closing everything.");
1456 destroy_data_stream_ht(data_ht
);
1457 destroy_metadata_stream_ht(metadata_ht
);
1459 ret
= close(ctx
->consumer_error_socket
);
1463 ret
= close(ctx
->consumer_metadata_socket
);
1467 utils_close_pipe(ctx
->consumer_channel_pipe
);
1468 lttng_pipe_destroy(ctx
->consumer_data_pipe
);
1469 lttng_pipe_destroy(ctx
->consumer_metadata_pipe
);
1470 lttng_pipe_destroy(ctx
->consumer_wakeup_pipe
);
1471 utils_close_pipe(ctx
->consumer_should_quit
);
1473 unlink(ctx
->consumer_command_sock_path
);
1478 * Write the metadata stream id on the specified file descriptor.
1480 static int write_relayd_metadata_id(int fd
,
1481 struct lttng_consumer_stream
*stream
,
1482 unsigned long padding
)
1485 struct lttcomm_relayd_metadata_payload hdr
;
1487 hdr
.stream_id
= htobe64(stream
->relayd_stream_id
);
1488 hdr
.padding_size
= htobe32(padding
);
1489 ret
= lttng_write(fd
, (void *) &hdr
, sizeof(hdr
));
1490 if (ret
< sizeof(hdr
)) {
1492 * This error means that the fd's end is closed so ignore the PERROR
1493 * not to clubber the error output since this can happen in a normal
1496 if (errno
!= EPIPE
) {
1497 PERROR("write metadata stream id");
1499 DBG3("Consumer failed to write relayd metadata id (errno: %d)", errno
);
1501 * Set ret to a negative value because if ret != sizeof(hdr), we don't
1502 * handle writting the missing part so report that as an error and
1503 * don't lie to the caller.
1508 DBG("Metadata stream id %" PRIu64
" with padding %lu written before data",
1509 stream
->relayd_stream_id
, padding
);
1516 * Mmap the ring buffer, read it and write the data to the tracefile. This is a
1517 * core function for writing trace buffers to either the local filesystem or
1520 * It must be called with the stream lock held.
1522 * Careful review MUST be put if any changes occur!
1524 * Returns the number of bytes written
1526 ssize_t
lttng_consumer_on_read_subbuffer_mmap(
1527 struct lttng_consumer_local_data
*ctx
,
1528 struct lttng_consumer_stream
*stream
, unsigned long len
,
1529 unsigned long padding
,
1530 struct ctf_packet_index
*index
)
1532 unsigned long mmap_offset
;
1535 off_t orig_offset
= stream
->out_fd_offset
;
1536 /* Default is on the disk */
1537 int outfd
= stream
->out_fd
;
1538 struct consumer_relayd_sock_pair
*relayd
= NULL
;
1539 unsigned int relayd_hang_up
= 0;
1541 /* RCU lock for the relayd pointer */
1544 /* Flag that the current stream if set for network streaming. */
1545 if (stream
->net_seq_idx
!= (uint64_t) -1ULL) {
1546 relayd
= consumer_find_relayd(stream
->net_seq_idx
);
1547 if (relayd
== NULL
) {
1553 /* get the offset inside the fd to mmap */
1554 switch (consumer_data
.type
) {
1555 case LTTNG_CONSUMER_KERNEL
:
1556 mmap_base
= stream
->mmap_base
;
1557 ret
= kernctl_get_mmap_read_offset(stream
->wait_fd
, &mmap_offset
);
1559 PERROR("tracer ctl get_mmap_read_offset");
1563 case LTTNG_CONSUMER32_UST
:
1564 case LTTNG_CONSUMER64_UST
:
1565 mmap_base
= lttng_ustctl_get_mmap_base(stream
);
1567 ERR("read mmap get mmap base for stream %s", stream
->name
);
1571 ret
= lttng_ustctl_get_mmap_read_offset(stream
, &mmap_offset
);
1573 PERROR("tracer ctl get_mmap_read_offset");
1579 ERR("Unknown consumer_data type");
1583 /* Handle stream on the relayd if the output is on the network */
1585 unsigned long netlen
= len
;
1588 * Lock the control socket for the complete duration of the function
1589 * since from this point on we will use the socket.
1591 if (stream
->metadata_flag
) {
1592 /* Metadata requires the control socket. */
1593 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
1594 if (stream
->reset_metadata_flag
) {
1595 ret
= relayd_reset_metadata(&relayd
->control_sock
,
1596 stream
->relayd_stream_id
,
1597 stream
->metadata_version
);
1602 stream
->reset_metadata_flag
= 0;
1604 netlen
+= sizeof(struct lttcomm_relayd_metadata_payload
);
1607 ret
= write_relayd_stream_header(stream
, netlen
, padding
, relayd
);
1612 /* Use the returned socket. */
1615 /* Write metadata stream id before payload */
1616 if (stream
->metadata_flag
) {
1617 ret
= write_relayd_metadata_id(outfd
, stream
, padding
);
1624 /* No streaming, we have to set the len with the full padding */
1627 if (stream
->metadata_flag
&& stream
->reset_metadata_flag
) {
1628 ret
= utils_truncate_stream_file(stream
->out_fd
, 0);
1630 ERR("Reset metadata file");
1633 stream
->reset_metadata_flag
= 0;
1637 * Check if we need to change the tracefile before writing the packet.
1639 if (stream
->chan
->tracefile_size
> 0 &&
1640 (stream
->tracefile_size_current
+ len
) >
1641 stream
->chan
->tracefile_size
) {
1642 ret
= utils_rotate_stream_file(stream
->chan
->pathname
,
1643 stream
->name
, stream
->chan
->tracefile_size
,
1644 stream
->chan
->tracefile_count
, stream
->uid
, stream
->gid
,
1645 stream
->out_fd
, &(stream
->tracefile_count_current
),
1648 ERR("Rotating output file");
1651 outfd
= stream
->out_fd
;
1653 if (stream
->index_file
) {
1654 lttng_index_file_put(stream
->index_file
);
1655 stream
->index_file
= lttng_index_file_create(stream
->chan
->pathname
,
1656 stream
->name
, stream
->uid
, stream
->gid
,
1657 stream
->chan
->tracefile_size
,
1658 stream
->tracefile_count_current
,
1659 CTF_INDEX_MAJOR
, CTF_INDEX_MINOR
);
1660 if (!stream
->index_file
) {
1665 /* Reset current size because we just perform a rotation. */
1666 stream
->tracefile_size_current
= 0;
1667 stream
->out_fd_offset
= 0;
1670 stream
->tracefile_size_current
+= len
;
1672 index
->offset
= htobe64(stream
->out_fd_offset
);
1677 * This call guarantee that len or less is returned. It's impossible to
1678 * receive a ret value that is bigger than len.
1680 ret
= lttng_write(outfd
, mmap_base
+ mmap_offset
, len
);
1681 DBG("Consumer mmap write() ret %zd (len %lu)", ret
, len
);
1682 if (ret
< 0 || ((size_t) ret
!= len
)) {
1684 * Report error to caller if nothing was written else at least send the
1692 /* Socket operation failed. We consider the relayd dead */
1693 if (errno
== EPIPE
|| errno
== EINVAL
|| errno
== EBADF
) {
1695 * This is possible if the fd is closed on the other side
1696 * (outfd) or any write problem. It can be verbose a bit for a
1697 * normal execution if for instance the relayd is stopped
1698 * abruptly. This can happen so set this to a DBG statement.
1700 DBG("Consumer mmap write detected relayd hang up");
1702 /* Unhandled error, print it and stop function right now. */
1703 PERROR("Error in write mmap (ret %zd != len %lu)", ret
, len
);
1707 stream
->output_written
+= ret
;
1709 /* This call is useless on a socket so better save a syscall. */
1711 /* This won't block, but will start writeout asynchronously */
1712 lttng_sync_file_range(outfd
, stream
->out_fd_offset
, len
,
1713 SYNC_FILE_RANGE_WRITE
);
1714 stream
->out_fd_offset
+= len
;
1715 lttng_consumer_sync_trace_file(stream
, orig_offset
);
1720 * This is a special case that the relayd has closed its socket. Let's
1721 * cleanup the relayd object and all associated streams.
1723 if (relayd
&& relayd_hang_up
) {
1724 cleanup_relayd(relayd
, ctx
);
1728 /* Unlock only if ctrl socket used */
1729 if (relayd
&& stream
->metadata_flag
) {
1730 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
1738 * Splice the data from the ring buffer to the tracefile.
1740 * It must be called with the stream lock held.
1742 * Returns the number of bytes spliced.
1744 ssize_t
lttng_consumer_on_read_subbuffer_splice(
1745 struct lttng_consumer_local_data
*ctx
,
1746 struct lttng_consumer_stream
*stream
, unsigned long len
,
1747 unsigned long padding
,
1748 struct ctf_packet_index
*index
)
1750 ssize_t ret
= 0, written
= 0, ret_splice
= 0;
1752 off_t orig_offset
= stream
->out_fd_offset
;
1753 int fd
= stream
->wait_fd
;
1754 /* Default is on the disk */
1755 int outfd
= stream
->out_fd
;
1756 struct consumer_relayd_sock_pair
*relayd
= NULL
;
1758 unsigned int relayd_hang_up
= 0;
1760 switch (consumer_data
.type
) {
1761 case LTTNG_CONSUMER_KERNEL
:
1763 case LTTNG_CONSUMER32_UST
:
1764 case LTTNG_CONSUMER64_UST
:
1765 /* Not supported for user space tracing */
1768 ERR("Unknown consumer_data type");
1772 /* RCU lock for the relayd pointer */
1775 /* Flag that the current stream if set for network streaming. */
1776 if (stream
->net_seq_idx
!= (uint64_t) -1ULL) {
1777 relayd
= consumer_find_relayd(stream
->net_seq_idx
);
1778 if (relayd
== NULL
) {
1783 splice_pipe
= stream
->splice_pipe
;
1785 /* Write metadata stream id before payload */
1787 unsigned long total_len
= len
;
1789 if (stream
->metadata_flag
) {
1791 * Lock the control socket for the complete duration of the function
1792 * since from this point on we will use the socket.
1794 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
1796 if (stream
->reset_metadata_flag
) {
1797 ret
= relayd_reset_metadata(&relayd
->control_sock
,
1798 stream
->relayd_stream_id
,
1799 stream
->metadata_version
);
1804 stream
->reset_metadata_flag
= 0;
1806 ret
= write_relayd_metadata_id(splice_pipe
[1], stream
,
1814 total_len
+= sizeof(struct lttcomm_relayd_metadata_payload
);
1817 ret
= write_relayd_stream_header(stream
, total_len
, padding
, relayd
);
1823 /* Use the returned socket. */
1826 /* No streaming, we have to set the len with the full padding */
1829 if (stream
->metadata_flag
&& stream
->reset_metadata_flag
) {
1830 ret
= utils_truncate_stream_file(stream
->out_fd
, 0);
1832 ERR("Reset metadata file");
1835 stream
->reset_metadata_flag
= 0;
1838 * Check if we need to change the tracefile before writing the packet.
1840 if (stream
->chan
->tracefile_size
> 0 &&
1841 (stream
->tracefile_size_current
+ len
) >
1842 stream
->chan
->tracefile_size
) {
1843 ret
= utils_rotate_stream_file(stream
->chan
->pathname
,
1844 stream
->name
, stream
->chan
->tracefile_size
,
1845 stream
->chan
->tracefile_count
, stream
->uid
, stream
->gid
,
1846 stream
->out_fd
, &(stream
->tracefile_count_current
),
1850 ERR("Rotating output file");
1853 outfd
= stream
->out_fd
;
1855 if (stream
->index_file
) {
1856 lttng_index_file_put(stream
->index_file
);
1857 stream
->index_file
= lttng_index_file_create(stream
->chan
->pathname
,
1858 stream
->name
, stream
->uid
, stream
->gid
,
1859 stream
->chan
->tracefile_size
,
1860 stream
->tracefile_count_current
,
1861 CTF_INDEX_MAJOR
, CTF_INDEX_MINOR
);
1862 if (!stream
->index_file
) {
1867 /* Reset current size because we just perform a rotation. */
1868 stream
->tracefile_size_current
= 0;
1869 stream
->out_fd_offset
= 0;
1872 stream
->tracefile_size_current
+= len
;
1873 index
->offset
= htobe64(stream
->out_fd_offset
);
1877 DBG("splice chan to pipe offset %lu of len %lu (fd : %d, pipe: %d)",
1878 (unsigned long)offset
, len
, fd
, splice_pipe
[1]);
1879 ret_splice
= splice(fd
, &offset
, splice_pipe
[1], NULL
, len
,
1880 SPLICE_F_MOVE
| SPLICE_F_MORE
);
1881 DBG("splice chan to pipe, ret %zd", ret_splice
);
1882 if (ret_splice
< 0) {
1885 PERROR("Error in relay splice");
1889 /* Handle stream on the relayd if the output is on the network */
1890 if (relayd
&& stream
->metadata_flag
) {
1891 size_t metadata_payload_size
=
1892 sizeof(struct lttcomm_relayd_metadata_payload
);
1894 /* Update counter to fit the spliced data */
1895 ret_splice
+= metadata_payload_size
;
1896 len
+= metadata_payload_size
;
1898 * We do this so the return value can match the len passed as
1899 * argument to this function.
1901 written
-= metadata_payload_size
;
1904 /* Splice data out */
1905 ret_splice
= splice(splice_pipe
[0], NULL
, outfd
, NULL
,
1906 ret_splice
, SPLICE_F_MOVE
| SPLICE_F_MORE
);
1907 DBG("Consumer splice pipe to file (out_fd: %d), ret %zd",
1909 if (ret_splice
< 0) {
1914 } else if (ret_splice
> len
) {
1916 * We don't expect this code path to be executed but you never know
1917 * so this is an extra protection agains a buggy splice().
1920 written
+= ret_splice
;
1921 PERROR("Wrote more data than requested %zd (len: %lu)", ret_splice
,
1925 /* All good, update current len and continue. */
1929 /* This call is useless on a socket so better save a syscall. */
1931 /* This won't block, but will start writeout asynchronously */
1932 lttng_sync_file_range(outfd
, stream
->out_fd_offset
, ret_splice
,
1933 SYNC_FILE_RANGE_WRITE
);
1934 stream
->out_fd_offset
+= ret_splice
;
1936 stream
->output_written
+= ret_splice
;
1937 written
+= ret_splice
;
1940 lttng_consumer_sync_trace_file(stream
, orig_offset
);
1946 * This is a special case that the relayd has closed its socket. Let's
1947 * cleanup the relayd object and all associated streams.
1949 if (relayd
&& relayd_hang_up
) {
1950 cleanup_relayd(relayd
, ctx
);
1951 /* Skip splice error so the consumer does not fail */
1956 /* send the appropriate error description to sessiond */
1959 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_SPLICE_EINVAL
);
1962 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_SPLICE_ENOMEM
);
1965 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_SPLICE_ESPIPE
);
1970 if (relayd
&& stream
->metadata_flag
) {
1971 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
1979 * Sample the snapshot positions for a specific fd
1981 * Returns 0 on success, < 0 on error
1983 int lttng_consumer_sample_snapshot_positions(struct lttng_consumer_stream
*stream
)
1985 switch (consumer_data
.type
) {
1986 case LTTNG_CONSUMER_KERNEL
:
1987 return lttng_kconsumer_sample_snapshot_positions(stream
);
1988 case LTTNG_CONSUMER32_UST
:
1989 case LTTNG_CONSUMER64_UST
:
1990 return lttng_ustconsumer_sample_snapshot_positions(stream
);
1992 ERR("Unknown consumer_data type");
1998 * Take a snapshot for a specific fd
2000 * Returns 0 on success, < 0 on error
2002 int lttng_consumer_take_snapshot(struct lttng_consumer_stream
*stream
)
2004 switch (consumer_data
.type
) {
2005 case LTTNG_CONSUMER_KERNEL
:
2006 return lttng_kconsumer_take_snapshot(stream
);
2007 case LTTNG_CONSUMER32_UST
:
2008 case LTTNG_CONSUMER64_UST
:
2009 return lttng_ustconsumer_take_snapshot(stream
);
2011 ERR("Unknown consumer_data type");
2018 * Get the produced position
2020 * Returns 0 on success, < 0 on error
2022 int lttng_consumer_get_produced_snapshot(struct lttng_consumer_stream
*stream
,
2025 switch (consumer_data
.type
) {
2026 case LTTNG_CONSUMER_KERNEL
:
2027 return lttng_kconsumer_get_produced_snapshot(stream
, pos
);
2028 case LTTNG_CONSUMER32_UST
:
2029 case LTTNG_CONSUMER64_UST
:
2030 return lttng_ustconsumer_get_produced_snapshot(stream
, pos
);
2032 ERR("Unknown consumer_data type");
2039 * Get the consumed position (free-running counter position in bytes).
2041 * Returns 0 on success, < 0 on error
2043 int lttng_consumer_get_consumed_snapshot(struct lttng_consumer_stream
*stream
,
2046 switch (consumer_data
.type
) {
2047 case LTTNG_CONSUMER_KERNEL
:
2048 return lttng_kconsumer_get_consumed_snapshot(stream
, pos
);
2049 case LTTNG_CONSUMER32_UST
:
2050 case LTTNG_CONSUMER64_UST
:
2051 return lttng_ustconsumer_get_consumed_snapshot(stream
, pos
);
2053 ERR("Unknown consumer_data type");
2059 int lttng_consumer_recv_cmd(struct lttng_consumer_local_data
*ctx
,
2060 int sock
, struct pollfd
*consumer_sockpoll
)
2062 switch (consumer_data
.type
) {
2063 case LTTNG_CONSUMER_KERNEL
:
2064 return lttng_kconsumer_recv_cmd(ctx
, sock
, consumer_sockpoll
);
2065 case LTTNG_CONSUMER32_UST
:
2066 case LTTNG_CONSUMER64_UST
:
2067 return lttng_ustconsumer_recv_cmd(ctx
, sock
, consumer_sockpoll
);
2069 ERR("Unknown consumer_data type");
2075 void lttng_consumer_close_all_metadata(void)
2077 switch (consumer_data
.type
) {
2078 case LTTNG_CONSUMER_KERNEL
:
2080 * The Kernel consumer has a different metadata scheme so we don't
2081 * close anything because the stream will be closed by the session
2085 case LTTNG_CONSUMER32_UST
:
2086 case LTTNG_CONSUMER64_UST
:
2088 * Close all metadata streams. The metadata hash table is passed and
2089 * this call iterates over it by closing all wakeup fd. This is safe
2090 * because at this point we are sure that the metadata producer is
2091 * either dead or blocked.
2093 lttng_ustconsumer_close_all_metadata(metadata_ht
);
2096 ERR("Unknown consumer_data type");
2102 * Clean up a metadata stream and free its memory.
2104 void consumer_del_metadata_stream(struct lttng_consumer_stream
*stream
,
2105 struct lttng_ht
*ht
)
2107 struct lttng_consumer_channel
*free_chan
= NULL
;
2111 * This call should NEVER receive regular stream. It must always be
2112 * metadata stream and this is crucial for data structure synchronization.
2114 assert(stream
->metadata_flag
);
2116 DBG3("Consumer delete metadata stream %d", stream
->wait_fd
);
2118 pthread_mutex_lock(&consumer_data
.lock
);
2119 pthread_mutex_lock(&stream
->chan
->lock
);
2120 pthread_mutex_lock(&stream
->lock
);
2121 if (stream
->chan
->metadata_cache
) {
2122 /* Only applicable to userspace consumers. */
2123 pthread_mutex_lock(&stream
->chan
->metadata_cache
->lock
);
2126 /* Remove any reference to that stream. */
2127 consumer_stream_delete(stream
, ht
);
2129 /* Close down everything including the relayd if one. */
2130 consumer_stream_close(stream
);
2131 /* Destroy tracer buffers of the stream. */
2132 consumer_stream_destroy_buffers(stream
);
2134 /* Atomically decrement channel refcount since other threads can use it. */
2135 if (!uatomic_sub_return(&stream
->chan
->refcount
, 1)
2136 && !uatomic_read(&stream
->chan
->nb_init_stream_left
)) {
2137 /* Go for channel deletion! */
2138 free_chan
= stream
->chan
;
2142 * Nullify the stream reference so it is not used after deletion. The
2143 * channel lock MUST be acquired before being able to check for a NULL
2146 stream
->chan
->metadata_stream
= NULL
;
2148 if (stream
->chan
->metadata_cache
) {
2149 pthread_mutex_unlock(&stream
->chan
->metadata_cache
->lock
);
2151 pthread_mutex_unlock(&stream
->lock
);
2152 pthread_mutex_unlock(&stream
->chan
->lock
);
2153 pthread_mutex_unlock(&consumer_data
.lock
);
2156 consumer_del_channel(free_chan
);
2159 consumer_stream_free(stream
);
2163 * Action done with the metadata stream when adding it to the consumer internal
2164 * data structures to handle it.
2166 void consumer_add_metadata_stream(struct lttng_consumer_stream
*stream
)
2168 struct lttng_ht
*ht
= metadata_ht
;
2169 struct lttng_ht_iter iter
;
2170 struct lttng_ht_node_u64
*node
;
2175 DBG3("Adding metadata stream %" PRIu64
" to hash table", stream
->key
);
2177 pthread_mutex_lock(&consumer_data
.lock
);
2178 pthread_mutex_lock(&stream
->chan
->lock
);
2179 pthread_mutex_lock(&stream
->chan
->timer_lock
);
2180 pthread_mutex_lock(&stream
->lock
);
2183 * From here, refcounts are updated so be _careful_ when returning an error
2190 * Lookup the stream just to make sure it does not exist in our internal
2191 * state. This should NEVER happen.
2193 lttng_ht_lookup(ht
, &stream
->key
, &iter
);
2194 node
= lttng_ht_iter_get_node_u64(&iter
);
2198 * When nb_init_stream_left reaches 0, we don't need to trigger any action
2199 * in terms of destroying the associated channel, because the action that
2200 * causes the count to become 0 also causes a stream to be added. The
2201 * channel deletion will thus be triggered by the following removal of this
2204 if (uatomic_read(&stream
->chan
->nb_init_stream_left
) > 0) {
2205 /* Increment refcount before decrementing nb_init_stream_left */
2207 uatomic_dec(&stream
->chan
->nb_init_stream_left
);
2210 lttng_ht_add_unique_u64(ht
, &stream
->node
);
2212 lttng_ht_add_u64(consumer_data
.stream_per_chan_id_ht
,
2213 &stream
->node_channel_id
);
2216 * Add stream to the stream_list_ht of the consumer data. No need to steal
2217 * the key since the HT does not use it and we allow to add redundant keys
2220 lttng_ht_add_u64(consumer_data
.stream_list_ht
, &stream
->node_session_id
);
2224 pthread_mutex_unlock(&stream
->lock
);
2225 pthread_mutex_unlock(&stream
->chan
->lock
);
2226 pthread_mutex_unlock(&stream
->chan
->timer_lock
);
2227 pthread_mutex_unlock(&consumer_data
.lock
);
2231 * Delete data stream that are flagged for deletion (endpoint_status).
2233 static void validate_endpoint_status_data_stream(void)
2235 struct lttng_ht_iter iter
;
2236 struct lttng_consumer_stream
*stream
;
2238 DBG("Consumer delete flagged data stream");
2241 cds_lfht_for_each_entry(data_ht
->ht
, &iter
.iter
, stream
, node
.node
) {
2242 /* Validate delete flag of the stream */
2243 if (stream
->endpoint_status
== CONSUMER_ENDPOINT_ACTIVE
) {
2246 /* Delete it right now */
2247 consumer_del_stream(stream
, data_ht
);
2253 * Delete metadata stream that are flagged for deletion (endpoint_status).
2255 static void validate_endpoint_status_metadata_stream(
2256 struct lttng_poll_event
*pollset
)
2258 struct lttng_ht_iter iter
;
2259 struct lttng_consumer_stream
*stream
;
2261 DBG("Consumer delete flagged metadata stream");
2266 cds_lfht_for_each_entry(metadata_ht
->ht
, &iter
.iter
, stream
, node
.node
) {
2267 /* Validate delete flag of the stream */
2268 if (stream
->endpoint_status
== CONSUMER_ENDPOINT_ACTIVE
) {
2272 * Remove from pollset so the metadata thread can continue without
2273 * blocking on a deleted stream.
2275 lttng_poll_del(pollset
, stream
->wait_fd
);
2277 /* Delete it right now */
2278 consumer_del_metadata_stream(stream
, metadata_ht
);
2284 int rotate_notify_sessiond(struct lttng_consumer_local_data
*ctx
,
2290 ret
= write(ctx
->channel_rotate_pipe
, &key
, sizeof(key
));
2291 } while (ret
== -1 && errno
== EINTR
);
2293 PERROR("Failed to write to the channel rotation pipe");
2295 DBG("Sent channel rotation notification for channel key %"
2304 * Perform operations that need to be done after a stream has
2305 * rotated and released the stream lock.
2307 * Multiple rotations cannot occur simultaneously, so we know the state of the
2308 * "rotated" stream flag cannot change.
2310 * This MUST be called WITHOUT the stream lock held.
2313 int consumer_post_rotation(struct lttng_consumer_stream
*stream
,
2314 struct lttng_consumer_local_data
*ctx
)
2318 pthread_mutex_lock(&stream
->chan
->lock
);
2320 switch (consumer_data
.type
) {
2321 case LTTNG_CONSUMER_KERNEL
:
2323 case LTTNG_CONSUMER32_UST
:
2324 case LTTNG_CONSUMER64_UST
:
2326 * The ust_metadata_pushed counter has been reset to 0, so now
2327 * we can wakeup the metadata thread so it dumps the metadata
2328 * cache to the new file.
2330 if (stream
->metadata_flag
) {
2331 consumer_metadata_wakeup_pipe(stream
->chan
);
2335 ERR("Unknown consumer_data type");
2339 if (--stream
->chan
->nr_stream_rotate_pending
== 0) {
2340 DBG("Rotation of channel \"%s\" completed, notifying the session daemon",
2341 stream
->chan
->name
);
2342 ret
= rotate_notify_sessiond(ctx
, stream
->chan
->key
);
2344 pthread_mutex_unlock(&stream
->chan
->lock
);
2350 * Thread polls on metadata file descriptor and write them on disk or on the
2353 void *consumer_thread_metadata_poll(void *data
)
2355 int ret
, i
, pollfd
, err
= -1;
2356 uint32_t revents
, nb_fd
;
2357 struct lttng_consumer_stream
*stream
= NULL
;
2358 struct lttng_ht_iter iter
;
2359 struct lttng_ht_node_u64
*node
;
2360 struct lttng_poll_event events
;
2361 struct lttng_consumer_local_data
*ctx
= data
;
2364 rcu_register_thread();
2366 health_register(health_consumerd
, HEALTH_CONSUMERD_TYPE_METADATA
);
2368 if (testpoint(consumerd_thread_metadata
)) {
2369 goto error_testpoint
;
2372 health_code_update();
2374 DBG("Thread metadata poll started");
2376 /* Size is set to 1 for the consumer_metadata pipe */
2377 ret
= lttng_poll_create(&events
, 2, LTTNG_CLOEXEC
);
2379 ERR("Poll set creation failed");
2383 ret
= lttng_poll_add(&events
,
2384 lttng_pipe_get_readfd(ctx
->consumer_metadata_pipe
), LPOLLIN
);
2390 DBG("Metadata main loop started");
2394 health_code_update();
2395 health_poll_entry();
2396 DBG("Metadata poll wait");
2397 ret
= lttng_poll_wait(&events
, -1);
2398 DBG("Metadata poll return from wait with %d fd(s)",
2399 LTTNG_POLL_GETNB(&events
));
2401 DBG("Metadata event caught in thread");
2403 if (errno
== EINTR
) {
2404 ERR("Poll EINTR caught");
2407 if (LTTNG_POLL_GETNB(&events
) == 0) {
2408 err
= 0; /* All is OK */
2415 /* From here, the event is a metadata wait fd */
2416 for (i
= 0; i
< nb_fd
; i
++) {
2417 health_code_update();
2419 revents
= LTTNG_POLL_GETEV(&events
, i
);
2420 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2423 /* No activity for this FD (poll implementation). */
2427 if (pollfd
== lttng_pipe_get_readfd(ctx
->consumer_metadata_pipe
)) {
2428 if (revents
& LPOLLIN
) {
2431 pipe_len
= lttng_pipe_read(ctx
->consumer_metadata_pipe
,
2432 &stream
, sizeof(stream
));
2433 if (pipe_len
< sizeof(stream
)) {
2435 PERROR("read metadata stream");
2438 * Remove the pipe from the poll set and continue the loop
2439 * since their might be data to consume.
2441 lttng_poll_del(&events
,
2442 lttng_pipe_get_readfd(ctx
->consumer_metadata_pipe
));
2443 lttng_pipe_read_close(ctx
->consumer_metadata_pipe
);
2447 /* A NULL stream means that the state has changed. */
2448 if (stream
== NULL
) {
2449 /* Check for deleted streams. */
2450 validate_endpoint_status_metadata_stream(&events
);
2454 DBG("Adding metadata stream %d to poll set",
2457 /* Add metadata stream to the global poll events list */
2458 lttng_poll_add(&events
, stream
->wait_fd
,
2459 LPOLLIN
| LPOLLPRI
| LPOLLHUP
);
2460 } else if (revents
& (LPOLLERR
| LPOLLHUP
)) {
2461 DBG("Metadata thread pipe hung up");
2463 * Remove the pipe from the poll set and continue the loop
2464 * since their might be data to consume.
2466 lttng_poll_del(&events
,
2467 lttng_pipe_get_readfd(ctx
->consumer_metadata_pipe
));
2468 lttng_pipe_read_close(ctx
->consumer_metadata_pipe
);
2471 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
2475 /* Handle other stream */
2481 uint64_t tmp_id
= (uint64_t) pollfd
;
2483 lttng_ht_lookup(metadata_ht
, &tmp_id
, &iter
);
2485 node
= lttng_ht_iter_get_node_u64(&iter
);
2488 stream
= caa_container_of(node
, struct lttng_consumer_stream
,
2491 if (revents
& (LPOLLIN
| LPOLLPRI
)) {
2492 /* Get the data out of the metadata file descriptor */
2493 DBG("Metadata available on fd %d", pollfd
);
2494 assert(stream
->wait_fd
== pollfd
);
2497 health_code_update();
2499 len
= ctx
->on_buffer_ready(stream
, ctx
);
2501 * We don't check the return value here since if we get
2502 * a negative len, it means an error occurred thus we
2503 * simply remove it from the poll set and free the
2508 /* It's ok to have an unavailable sub-buffer */
2509 if (len
< 0 && len
!= -EAGAIN
&& len
!= -ENODATA
) {
2510 /* Clean up stream from consumer and free it. */
2511 lttng_poll_del(&events
, stream
->wait_fd
);
2512 consumer_del_metadata_stream(stream
, metadata_ht
);
2514 } else if (revents
& (LPOLLERR
| LPOLLHUP
)) {
2515 DBG("Metadata fd %d is hup|err.", pollfd
);
2516 if (!stream
->hangup_flush_done
2517 && (consumer_data
.type
== LTTNG_CONSUMER32_UST
2518 || consumer_data
.type
== LTTNG_CONSUMER64_UST
)) {
2519 DBG("Attempting to flush and consume the UST buffers");
2520 lttng_ustconsumer_on_stream_hangup(stream
);
2522 /* We just flushed the stream now read it. */
2524 health_code_update();
2526 len
= ctx
->on_buffer_ready(stream
, ctx
);
2528 * We don't check the return value here since if we get
2529 * a negative len, it means an error occurred thus we
2530 * simply remove it from the poll set and free the
2536 lttng_poll_del(&events
, stream
->wait_fd
);
2538 * This call update the channel states, closes file descriptors
2539 * and securely free the stream.
2541 consumer_del_metadata_stream(stream
, metadata_ht
);
2543 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
2547 /* Release RCU lock for the stream looked up */
2555 DBG("Metadata poll thread exiting");
2557 lttng_poll_clean(&events
);
2562 ERR("Health error occurred in %s", __func__
);
2564 health_unregister(health_consumerd
);
2565 rcu_unregister_thread();
2570 * This thread polls the fds in the set to consume the data and write
2571 * it to tracefile if necessary.
2573 void *consumer_thread_data_poll(void *data
)
2575 int num_rdy
, num_hup
, high_prio
, ret
, i
, err
= -1;
2576 struct pollfd
*pollfd
= NULL
;
2577 /* local view of the streams */
2578 struct lttng_consumer_stream
**local_stream
= NULL
, *new_stream
= NULL
;
2579 /* local view of consumer_data.fds_count */
2581 /* 2 for the consumer_data_pipe and wake up pipe */
2582 const int nb_pipes_fd
= 2;
2583 /* Number of FDs with CONSUMER_ENDPOINT_INACTIVE but still open. */
2584 int nb_inactive_fd
= 0;
2585 struct lttng_consumer_local_data
*ctx
= data
;
2588 rcu_register_thread();
2590 health_register(health_consumerd
, HEALTH_CONSUMERD_TYPE_DATA
);
2592 if (testpoint(consumerd_thread_data
)) {
2593 goto error_testpoint
;
2596 health_code_update();
2598 local_stream
= zmalloc(sizeof(struct lttng_consumer_stream
*));
2599 if (local_stream
== NULL
) {
2600 PERROR("local_stream malloc");
2605 health_code_update();
2611 * the fds set has been updated, we need to update our
2612 * local array as well
2614 pthread_mutex_lock(&consumer_data
.lock
);
2615 if (consumer_data
.need_update
) {
2620 local_stream
= NULL
;
2622 /* Allocate for all fds */
2623 pollfd
= zmalloc((consumer_data
.stream_count
+ nb_pipes_fd
) * sizeof(struct pollfd
));
2624 if (pollfd
== NULL
) {
2625 PERROR("pollfd malloc");
2626 pthread_mutex_unlock(&consumer_data
.lock
);
2630 local_stream
= zmalloc((consumer_data
.stream_count
+ nb_pipes_fd
) *
2631 sizeof(struct lttng_consumer_stream
*));
2632 if (local_stream
== NULL
) {
2633 PERROR("local_stream malloc");
2634 pthread_mutex_unlock(&consumer_data
.lock
);
2637 ret
= update_poll_array(ctx
, &pollfd
, local_stream
,
2638 data_ht
, &nb_inactive_fd
);
2640 ERR("Error in allocating pollfd or local_outfds");
2641 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_POLL_ERROR
);
2642 pthread_mutex_unlock(&consumer_data
.lock
);
2646 consumer_data
.need_update
= 0;
2648 pthread_mutex_unlock(&consumer_data
.lock
);
2650 /* No FDs and consumer_quit, consumer_cleanup the thread */
2651 if (nb_fd
== 0 && nb_inactive_fd
== 0 &&
2652 CMM_LOAD_SHARED(consumer_quit
) == 1) {
2653 err
= 0; /* All is OK */
2656 /* poll on the array of fds */
2658 DBG("polling on %d fd", nb_fd
+ nb_pipes_fd
);
2659 if (testpoint(consumerd_thread_data_poll
)) {
2662 health_poll_entry();
2663 num_rdy
= poll(pollfd
, nb_fd
+ nb_pipes_fd
, -1);
2665 DBG("poll num_rdy : %d", num_rdy
);
2666 if (num_rdy
== -1) {
2668 * Restart interrupted system call.
2670 if (errno
== EINTR
) {
2673 PERROR("Poll error");
2674 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_POLL_ERROR
);
2676 } else if (num_rdy
== 0) {
2677 DBG("Polling thread timed out");
2681 if (caa_unlikely(data_consumption_paused
)) {
2682 DBG("Data consumption paused, sleeping...");
2688 * If the consumer_data_pipe triggered poll go directly to the
2689 * beginning of the loop to update the array. We want to prioritize
2690 * array update over low-priority reads.
2692 if (pollfd
[nb_fd
].revents
& (POLLIN
| POLLPRI
)) {
2693 ssize_t pipe_readlen
;
2695 DBG("consumer_data_pipe wake up");
2696 pipe_readlen
= lttng_pipe_read(ctx
->consumer_data_pipe
,
2697 &new_stream
, sizeof(new_stream
));
2698 if (pipe_readlen
< sizeof(new_stream
)) {
2699 PERROR("Consumer data pipe");
2700 /* Continue so we can at least handle the current stream(s). */
2705 * If the stream is NULL, just ignore it. It's also possible that
2706 * the sessiond poll thread changed the consumer_quit state and is
2707 * waking us up to test it.
2709 if (new_stream
== NULL
) {
2710 validate_endpoint_status_data_stream();
2714 /* Continue to update the local streams and handle prio ones */
2718 /* Handle wakeup pipe. */
2719 if (pollfd
[nb_fd
+ 1].revents
& (POLLIN
| POLLPRI
)) {
2721 ssize_t pipe_readlen
;
2723 pipe_readlen
= lttng_pipe_read(ctx
->consumer_wakeup_pipe
, &dummy
,
2725 if (pipe_readlen
< 0) {
2726 PERROR("Consumer data wakeup pipe");
2728 /* We've been awakened to handle stream(s). */
2729 ctx
->has_wakeup
= 0;
2732 /* Take care of high priority channels first. */
2733 for (i
= 0; i
< nb_fd
; i
++) {
2734 health_code_update();
2736 if (local_stream
[i
] == NULL
) {
2739 if (pollfd
[i
].revents
& POLLPRI
) {
2740 DBG("Urgent read on fd %d", pollfd
[i
].fd
);
2742 len
= ctx
->on_buffer_ready(local_stream
[i
], ctx
);
2743 /* it's ok to have an unavailable sub-buffer */
2744 if (len
< 0 && len
!= -EAGAIN
&& len
!= -ENODATA
) {
2745 /* Clean the stream and free it. */
2746 consumer_del_stream(local_stream
[i
], data_ht
);
2747 local_stream
[i
] = NULL
;
2748 } else if (len
> 0) {
2749 local_stream
[i
]->data_read
= 1;
2755 * If we read high prio channel in this loop, try again
2756 * for more high prio data.
2762 /* Take care of low priority channels. */
2763 for (i
= 0; i
< nb_fd
; i
++) {
2764 health_code_update();
2766 if (local_stream
[i
] == NULL
) {
2769 if ((pollfd
[i
].revents
& POLLIN
) ||
2770 local_stream
[i
]->hangup_flush_done
||
2771 local_stream
[i
]->has_data
) {
2772 DBG("Normal read on fd %d", pollfd
[i
].fd
);
2773 len
= ctx
->on_buffer_ready(local_stream
[i
], ctx
);
2774 /* it's ok to have an unavailable sub-buffer */
2775 if (len
< 0 && len
!= -EAGAIN
&& len
!= -ENODATA
) {
2776 /* Clean the stream and free it. */
2777 consumer_del_stream(local_stream
[i
], data_ht
);
2778 local_stream
[i
] = NULL
;
2779 } else if (len
> 0) {
2780 local_stream
[i
]->data_read
= 1;
2785 /* Handle hangup and errors */
2786 for (i
= 0; i
< nb_fd
; i
++) {
2787 health_code_update();
2789 if (local_stream
[i
] == NULL
) {
2792 if (!local_stream
[i
]->hangup_flush_done
2793 && (pollfd
[i
].revents
& (POLLHUP
| POLLERR
| POLLNVAL
))
2794 && (consumer_data
.type
== LTTNG_CONSUMER32_UST
2795 || consumer_data
.type
== LTTNG_CONSUMER64_UST
)) {
2796 DBG("fd %d is hup|err|nval. Attempting flush and read.",
2798 lttng_ustconsumer_on_stream_hangup(local_stream
[i
]);
2799 /* Attempt read again, for the data we just flushed. */
2800 local_stream
[i
]->data_read
= 1;
2803 * If the poll flag is HUP/ERR/NVAL and we have
2804 * read no data in this pass, we can remove the
2805 * stream from its hash table.
2807 if ((pollfd
[i
].revents
& POLLHUP
)) {
2808 DBG("Polling fd %d tells it has hung up.", pollfd
[i
].fd
);
2809 if (!local_stream
[i
]->data_read
) {
2810 consumer_del_stream(local_stream
[i
], data_ht
);
2811 local_stream
[i
] = NULL
;
2814 } else if (pollfd
[i
].revents
& POLLERR
) {
2815 ERR("Error returned in polling fd %d.", pollfd
[i
].fd
);
2816 if (!local_stream
[i
]->data_read
) {
2817 consumer_del_stream(local_stream
[i
], data_ht
);
2818 local_stream
[i
] = NULL
;
2821 } else if (pollfd
[i
].revents
& POLLNVAL
) {
2822 ERR("Polling fd %d tells fd is not open.", pollfd
[i
].fd
);
2823 if (!local_stream
[i
]->data_read
) {
2824 consumer_del_stream(local_stream
[i
], data_ht
);
2825 local_stream
[i
] = NULL
;
2829 if (local_stream
[i
] != NULL
) {
2830 local_stream
[i
]->data_read
= 0;
2837 DBG("polling thread exiting");
2842 * Close the write side of the pipe so epoll_wait() in
2843 * consumer_thread_metadata_poll can catch it. The thread is monitoring the
2844 * read side of the pipe. If we close them both, epoll_wait strangely does
2845 * not return and could create a endless wait period if the pipe is the
2846 * only tracked fd in the poll set. The thread will take care of closing
2849 (void) lttng_pipe_write_close(ctx
->consumer_metadata_pipe
);
2854 ERR("Health error occurred in %s", __func__
);
2856 health_unregister(health_consumerd
);
2858 rcu_unregister_thread();
2863 * Close wake-up end of each stream belonging to the channel. This will
2864 * allow the poll() on the stream read-side to detect when the
2865 * write-side (application) finally closes them.
2868 void consumer_close_channel_streams(struct lttng_consumer_channel
*channel
)
2870 struct lttng_ht
*ht
;
2871 struct lttng_consumer_stream
*stream
;
2872 struct lttng_ht_iter iter
;
2874 ht
= consumer_data
.stream_per_chan_id_ht
;
2877 cds_lfht_for_each_entry_duplicate(ht
->ht
,
2878 ht
->hash_fct(&channel
->key
, lttng_ht_seed
),
2879 ht
->match_fct
, &channel
->key
,
2880 &iter
.iter
, stream
, node_channel_id
.node
) {
2882 * Protect against teardown with mutex.
2884 pthread_mutex_lock(&stream
->lock
);
2885 if (cds_lfht_is_node_deleted(&stream
->node
.node
)) {
2888 switch (consumer_data
.type
) {
2889 case LTTNG_CONSUMER_KERNEL
:
2891 case LTTNG_CONSUMER32_UST
:
2892 case LTTNG_CONSUMER64_UST
:
2893 if (stream
->metadata_flag
) {
2894 /* Safe and protected by the stream lock. */
2895 lttng_ustconsumer_close_metadata(stream
->chan
);
2898 * Note: a mutex is taken internally within
2899 * liblttng-ust-ctl to protect timer wakeup_fd
2900 * use from concurrent close.
2902 lttng_ustconsumer_close_stream_wakeup(stream
);
2906 ERR("Unknown consumer_data type");
2910 pthread_mutex_unlock(&stream
->lock
);
2915 static void destroy_channel_ht(struct lttng_ht
*ht
)
2917 struct lttng_ht_iter iter
;
2918 struct lttng_consumer_channel
*channel
;
2926 cds_lfht_for_each_entry(ht
->ht
, &iter
.iter
, channel
, wait_fd_node
.node
) {
2927 ret
= lttng_ht_del(ht
, &iter
);
2932 lttng_ht_destroy(ht
);
2936 * This thread polls the channel fds to detect when they are being
2937 * closed. It closes all related streams if the channel is detected as
2938 * closed. It is currently only used as a shim layer for UST because the
2939 * consumerd needs to keep the per-stream wakeup end of pipes open for
2942 void *consumer_thread_channel_poll(void *data
)
2944 int ret
, i
, pollfd
, err
= -1;
2945 uint32_t revents
, nb_fd
;
2946 struct lttng_consumer_channel
*chan
= NULL
;
2947 struct lttng_ht_iter iter
;
2948 struct lttng_ht_node_u64
*node
;
2949 struct lttng_poll_event events
;
2950 struct lttng_consumer_local_data
*ctx
= data
;
2951 struct lttng_ht
*channel_ht
;
2953 rcu_register_thread();
2955 health_register(health_consumerd
, HEALTH_CONSUMERD_TYPE_CHANNEL
);
2957 if (testpoint(consumerd_thread_channel
)) {
2958 goto error_testpoint
;
2961 health_code_update();
2963 channel_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
2965 /* ENOMEM at this point. Better to bail out. */
2969 DBG("Thread channel poll started");
2971 /* Size is set to 1 for the consumer_channel pipe */
2972 ret
= lttng_poll_create(&events
, 2, LTTNG_CLOEXEC
);
2974 ERR("Poll set creation failed");
2978 ret
= lttng_poll_add(&events
, ctx
->consumer_channel_pipe
[0], LPOLLIN
);
2984 DBG("Channel main loop started");
2988 health_code_update();
2989 DBG("Channel poll wait");
2990 health_poll_entry();
2991 ret
= lttng_poll_wait(&events
, -1);
2992 DBG("Channel poll return from wait with %d fd(s)",
2993 LTTNG_POLL_GETNB(&events
));
2995 DBG("Channel event caught in thread");
2997 if (errno
== EINTR
) {
2998 ERR("Poll EINTR caught");
3001 if (LTTNG_POLL_GETNB(&events
) == 0) {
3002 err
= 0; /* All is OK */
3009 /* From here, the event is a channel wait fd */
3010 for (i
= 0; i
< nb_fd
; i
++) {
3011 health_code_update();
3013 revents
= LTTNG_POLL_GETEV(&events
, i
);
3014 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3017 /* No activity for this FD (poll implementation). */
3021 if (pollfd
== ctx
->consumer_channel_pipe
[0]) {
3022 if (revents
& LPOLLIN
) {
3023 enum consumer_channel_action action
;
3026 ret
= read_channel_pipe(ctx
, &chan
, &key
, &action
);
3029 ERR("Error reading channel pipe");
3031 lttng_poll_del(&events
, ctx
->consumer_channel_pipe
[0]);
3036 case CONSUMER_CHANNEL_ADD
:
3037 DBG("Adding channel %d to poll set",
3040 lttng_ht_node_init_u64(&chan
->wait_fd_node
,
3043 lttng_ht_add_unique_u64(channel_ht
,
3044 &chan
->wait_fd_node
);
3046 /* Add channel to the global poll events list */
3047 lttng_poll_add(&events
, chan
->wait_fd
,
3048 LPOLLERR
| LPOLLHUP
);
3050 case CONSUMER_CHANNEL_DEL
:
3053 * This command should never be called if the channel
3054 * has streams monitored by either the data or metadata
3055 * thread. The consumer only notify this thread with a
3056 * channel del. command if it receives a destroy
3057 * channel command from the session daemon that send it
3058 * if a command prior to the GET_CHANNEL failed.
3062 chan
= consumer_find_channel(key
);
3065 ERR("UST consumer get channel key %" PRIu64
" not found for del channel", key
);
3068 lttng_poll_del(&events
, chan
->wait_fd
);
3069 iter
.iter
.node
= &chan
->wait_fd_node
.node
;
3070 ret
= lttng_ht_del(channel_ht
, &iter
);
3073 switch (consumer_data
.type
) {
3074 case LTTNG_CONSUMER_KERNEL
:
3076 case LTTNG_CONSUMER32_UST
:
3077 case LTTNG_CONSUMER64_UST
:
3078 health_code_update();
3079 /* Destroy streams that might have been left in the stream list. */
3080 clean_channel_stream_list(chan
);
3083 ERR("Unknown consumer_data type");
3088 * Release our own refcount. Force channel deletion even if
3089 * streams were not initialized.
3091 if (!uatomic_sub_return(&chan
->refcount
, 1)) {
3092 consumer_del_channel(chan
);
3097 case CONSUMER_CHANNEL_QUIT
:
3099 * Remove the pipe from the poll set and continue the loop
3100 * since their might be data to consume.
3102 lttng_poll_del(&events
, ctx
->consumer_channel_pipe
[0]);
3105 ERR("Unknown action");
3108 } else if (revents
& (LPOLLERR
| LPOLLHUP
)) {
3109 DBG("Channel thread pipe hung up");
3111 * Remove the pipe from the poll set and continue the loop
3112 * since their might be data to consume.
3114 lttng_poll_del(&events
, ctx
->consumer_channel_pipe
[0]);
3117 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
3121 /* Handle other stream */
3127 uint64_t tmp_id
= (uint64_t) pollfd
;
3129 lttng_ht_lookup(channel_ht
, &tmp_id
, &iter
);
3131 node
= lttng_ht_iter_get_node_u64(&iter
);
3134 chan
= caa_container_of(node
, struct lttng_consumer_channel
,
3137 /* Check for error event */
3138 if (revents
& (LPOLLERR
| LPOLLHUP
)) {
3139 DBG("Channel fd %d is hup|err.", pollfd
);
3141 lttng_poll_del(&events
, chan
->wait_fd
);
3142 ret
= lttng_ht_del(channel_ht
, &iter
);
3146 * This will close the wait fd for each stream associated to
3147 * this channel AND monitored by the data/metadata thread thus
3148 * will be clean by the right thread.
3150 consumer_close_channel_streams(chan
);
3152 /* Release our own refcount */
3153 if (!uatomic_sub_return(&chan
->refcount
, 1)
3154 && !uatomic_read(&chan
->nb_init_stream_left
)) {
3155 consumer_del_channel(chan
);
3158 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
3163 /* Release RCU lock for the channel looked up */
3171 lttng_poll_clean(&events
);
3173 destroy_channel_ht(channel_ht
);
3176 DBG("Channel poll thread exiting");
3179 ERR("Health error occurred in %s", __func__
);
3181 health_unregister(health_consumerd
);
3182 rcu_unregister_thread();
3186 static int set_metadata_socket(struct lttng_consumer_local_data
*ctx
,
3187 struct pollfd
*sockpoll
, int client_socket
)
3194 ret
= lttng_consumer_poll_socket(sockpoll
);
3198 DBG("Metadata connection on client_socket");
3200 /* Blocking call, waiting for transmission */
3201 ctx
->consumer_metadata_socket
= lttcomm_accept_unix_sock(client_socket
);
3202 if (ctx
->consumer_metadata_socket
< 0) {
3203 WARN("On accept metadata");
3214 * This thread listens on the consumerd socket and receives the file
3215 * descriptors from the session daemon.
3217 void *consumer_thread_sessiond_poll(void *data
)
3219 int sock
= -1, client_socket
, ret
, err
= -1;
3221 * structure to poll for incoming data on communication socket avoids
3222 * making blocking sockets.
3224 struct pollfd consumer_sockpoll
[2];
3225 struct lttng_consumer_local_data
*ctx
= data
;
3227 rcu_register_thread();
3229 health_register(health_consumerd
, HEALTH_CONSUMERD_TYPE_SESSIOND
);
3231 if (testpoint(consumerd_thread_sessiond
)) {
3232 goto error_testpoint
;
3235 health_code_update();
3237 DBG("Creating command socket %s", ctx
->consumer_command_sock_path
);
3238 unlink(ctx
->consumer_command_sock_path
);
3239 client_socket
= lttcomm_create_unix_sock(ctx
->consumer_command_sock_path
);
3240 if (client_socket
< 0) {
3241 ERR("Cannot create command socket");
3245 ret
= lttcomm_listen_unix_sock(client_socket
);
3250 DBG("Sending ready command to lttng-sessiond");
3251 ret
= lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_COMMAND_SOCK_READY
);
3252 /* return < 0 on error, but == 0 is not fatal */
3254 ERR("Error sending ready command to lttng-sessiond");
3258 /* prepare the FDs to poll : to client socket and the should_quit pipe */
3259 consumer_sockpoll
[0].fd
= ctx
->consumer_should_quit
[0];
3260 consumer_sockpoll
[0].events
= POLLIN
| POLLPRI
;
3261 consumer_sockpoll
[1].fd
= client_socket
;
3262 consumer_sockpoll
[1].events
= POLLIN
| POLLPRI
;
3264 ret
= lttng_consumer_poll_socket(consumer_sockpoll
);
3272 DBG("Connection on client_socket");
3274 /* Blocking call, waiting for transmission */
3275 sock
= lttcomm_accept_unix_sock(client_socket
);
3282 * Setup metadata socket which is the second socket connection on the
3283 * command unix socket.
3285 ret
= set_metadata_socket(ctx
, consumer_sockpoll
, client_socket
);
3294 /* This socket is not useful anymore. */
3295 ret
= close(client_socket
);
3297 PERROR("close client_socket");
3301 /* update the polling structure to poll on the established socket */
3302 consumer_sockpoll
[1].fd
= sock
;
3303 consumer_sockpoll
[1].events
= POLLIN
| POLLPRI
;
3306 health_code_update();
3308 health_poll_entry();
3309 ret
= lttng_consumer_poll_socket(consumer_sockpoll
);
3318 DBG("Incoming command on sock");
3319 ret
= lttng_consumer_recv_cmd(ctx
, sock
, consumer_sockpoll
);
3322 * This could simply be a session daemon quitting. Don't output
3325 DBG("Communication interrupted on command socket");
3329 if (CMM_LOAD_SHARED(consumer_quit
)) {
3330 DBG("consumer_thread_receive_fds received quit from signal");
3331 err
= 0; /* All is OK */
3334 DBG("received command on sock");
3340 DBG("Consumer thread sessiond poll exiting");
3343 * Close metadata streams since the producer is the session daemon which
3346 * NOTE: for now, this only applies to the UST tracer.
3348 lttng_consumer_close_all_metadata();
3351 * when all fds have hung up, the polling thread
3354 CMM_STORE_SHARED(consumer_quit
, 1);
3357 * Notify the data poll thread to poll back again and test the
3358 * consumer_quit state that we just set so to quit gracefully.
3360 notify_thread_lttng_pipe(ctx
->consumer_data_pipe
);
3362 notify_channel_pipe(ctx
, NULL
, -1, CONSUMER_CHANNEL_QUIT
);
3364 notify_health_quit_pipe(health_quit_pipe
);
3366 /* Cleaning up possibly open sockets. */
3370 PERROR("close sock sessiond poll");
3373 if (client_socket
>= 0) {
3374 ret
= close(client_socket
);
3376 PERROR("close client_socket sessiond poll");
3383 ERR("Health error occurred in %s", __func__
);
3385 health_unregister(health_consumerd
);
3387 rcu_unregister_thread();
3391 ssize_t
lttng_consumer_read_subbuffer(struct lttng_consumer_stream
*stream
,
3392 struct lttng_consumer_local_data
*ctx
)
3396 bool rotated
= false;
3398 pthread_mutex_lock(&stream
->lock
);
3399 if (stream
->metadata_flag
) {
3400 pthread_mutex_lock(&stream
->metadata_rdv_lock
);
3403 switch (consumer_data
.type
) {
3404 case LTTNG_CONSUMER_KERNEL
:
3405 ret
= lttng_kconsumer_read_subbuffer(stream
, ctx
, &rotated
);
3407 case LTTNG_CONSUMER32_UST
:
3408 case LTTNG_CONSUMER64_UST
:
3409 ret
= lttng_ustconsumer_read_subbuffer(stream
, ctx
, &rotated
);
3412 ERR("Unknown consumer_data type");
3418 if (stream
->metadata_flag
) {
3419 pthread_cond_broadcast(&stream
->metadata_rdv
);
3420 pthread_mutex_unlock(&stream
->metadata_rdv_lock
);
3422 pthread_mutex_unlock(&stream
->lock
);
3424 rotate_ret
= consumer_post_rotation(stream
, ctx
);
3425 if (rotate_ret
< 0) {
3426 ERR("Failed after a rotation");
3434 int lttng_consumer_on_recv_stream(struct lttng_consumer_stream
*stream
)
3436 switch (consumer_data
.type
) {
3437 case LTTNG_CONSUMER_KERNEL
:
3438 return lttng_kconsumer_on_recv_stream(stream
);
3439 case LTTNG_CONSUMER32_UST
:
3440 case LTTNG_CONSUMER64_UST
:
3441 return lttng_ustconsumer_on_recv_stream(stream
);
3443 ERR("Unknown consumer_data type");
3450 * Allocate and set consumer data hash tables.
3452 int lttng_consumer_init(void)
3454 consumer_data
.channel_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3455 if (!consumer_data
.channel_ht
) {
3459 consumer_data
.relayd_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3460 if (!consumer_data
.relayd_ht
) {
3464 consumer_data
.stream_list_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3465 if (!consumer_data
.stream_list_ht
) {
3469 consumer_data
.stream_per_chan_id_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3470 if (!consumer_data
.stream_per_chan_id_ht
) {
3474 data_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3479 metadata_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3491 * Process the ADD_RELAYD command receive by a consumer.
3493 * This will create a relayd socket pair and add it to the relayd hash table.
3494 * The caller MUST acquire a RCU read side lock before calling it.
3496 void consumer_add_relayd_socket(uint64_t net_seq_idx
, int sock_type
,
3497 struct lttng_consumer_local_data
*ctx
, int sock
,
3498 struct pollfd
*consumer_sockpoll
,
3499 struct lttcomm_relayd_sock
*relayd_sock
, uint64_t sessiond_id
,
3500 uint64_t relayd_session_id
)
3502 int fd
= -1, ret
= -1, relayd_created
= 0;
3503 enum lttcomm_return_code ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
3504 struct consumer_relayd_sock_pair
*relayd
= NULL
;
3507 assert(relayd_sock
);
3509 DBG("Consumer adding relayd socket (idx: %" PRIu64
")", net_seq_idx
);
3511 /* Get relayd reference if exists. */
3512 relayd
= consumer_find_relayd(net_seq_idx
);
3513 if (relayd
== NULL
) {
3514 assert(sock_type
== LTTNG_STREAM_CONTROL
);
3515 /* Not found. Allocate one. */
3516 relayd
= consumer_allocate_relayd_sock_pair(net_seq_idx
);
3517 if (relayd
== NULL
) {
3518 ret_code
= LTTCOMM_CONSUMERD_ENOMEM
;
3521 relayd
->sessiond_session_id
= sessiond_id
;
3526 * This code path MUST continue to the consumer send status message to
3527 * we can notify the session daemon and continue our work without
3528 * killing everything.
3532 * relayd key should never be found for control socket.
3534 assert(sock_type
!= LTTNG_STREAM_CONTROL
);
3537 /* First send a status message before receiving the fds. */
3538 ret
= consumer_send_status_msg(sock
, LTTCOMM_CONSUMERD_SUCCESS
);
3540 /* Somehow, the session daemon is not responding anymore. */
3541 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_FATAL
);
3542 goto error_nosignal
;
3545 /* Poll on consumer socket. */
3546 ret
= lttng_consumer_poll_socket(consumer_sockpoll
);
3548 /* Needing to exit in the middle of a command: error. */
3549 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_POLL_ERROR
);
3550 goto error_nosignal
;
3553 /* Get relayd socket from session daemon */
3554 ret
= lttcomm_recv_fds_unix_sock(sock
, &fd
, 1);
3555 if (ret
!= sizeof(fd
)) {
3556 fd
= -1; /* Just in case it gets set with an invalid value. */
3559 * Failing to receive FDs might indicate a major problem such as
3560 * reaching a fd limit during the receive where the kernel returns a
3561 * MSG_CTRUNC and fails to cleanup the fd in the queue. Any case, we
3562 * don't take any chances and stop everything.
3564 * XXX: Feature request #558 will fix that and avoid this possible
3565 * issue when reaching the fd limit.
3567 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_ERROR_RECV_FD
);
3568 ret_code
= LTTCOMM_CONSUMERD_ERROR_RECV_FD
;
3572 /* Copy socket information and received FD */
3573 switch (sock_type
) {
3574 case LTTNG_STREAM_CONTROL
:
3575 /* Copy received lttcomm socket */
3576 lttcomm_copy_sock(&relayd
->control_sock
.sock
, &relayd_sock
->sock
);
3577 ret
= lttcomm_create_sock(&relayd
->control_sock
.sock
);
3578 /* Handle create_sock error. */
3580 ret_code
= LTTCOMM_CONSUMERD_ENOMEM
;
3584 * Close the socket created internally by
3585 * lttcomm_create_sock, so we can replace it by the one
3586 * received from sessiond.
3588 if (close(relayd
->control_sock
.sock
.fd
)) {
3592 /* Assign new file descriptor */
3593 relayd
->control_sock
.sock
.fd
= fd
;
3594 fd
= -1; /* For error path */
3595 /* Assign version values. */
3596 relayd
->control_sock
.major
= relayd_sock
->major
;
3597 relayd
->control_sock
.minor
= relayd_sock
->minor
;
3599 relayd
->relayd_session_id
= relayd_session_id
;
3602 case LTTNG_STREAM_DATA
:
3603 /* Copy received lttcomm socket */
3604 lttcomm_copy_sock(&relayd
->data_sock
.sock
, &relayd_sock
->sock
);
3605 ret
= lttcomm_create_sock(&relayd
->data_sock
.sock
);
3606 /* Handle create_sock error. */
3608 ret_code
= LTTCOMM_CONSUMERD_ENOMEM
;
3612 * Close the socket created internally by
3613 * lttcomm_create_sock, so we can replace it by the one
3614 * received from sessiond.
3616 if (close(relayd
->data_sock
.sock
.fd
)) {
3620 /* Assign new file descriptor */
3621 relayd
->data_sock
.sock
.fd
= fd
;
3622 fd
= -1; /* for eventual error paths */
3623 /* Assign version values. */
3624 relayd
->data_sock
.major
= relayd_sock
->major
;
3625 relayd
->data_sock
.minor
= relayd_sock
->minor
;
3628 ERR("Unknown relayd socket type (%d)", sock_type
);
3629 ret_code
= LTTCOMM_CONSUMERD_FATAL
;
3633 DBG("Consumer %s socket created successfully with net idx %" PRIu64
" (fd: %d)",
3634 sock_type
== LTTNG_STREAM_CONTROL
? "control" : "data",
3635 relayd
->net_seq_idx
, fd
);
3637 /* We successfully added the socket. Send status back. */
3638 ret
= consumer_send_status_msg(sock
, ret_code
);
3640 /* Somehow, the session daemon is not responding anymore. */
3641 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_FATAL
);
3642 goto error_nosignal
;
3646 * Add relayd socket pair to consumer data hashtable. If object already
3647 * exists or on error, the function gracefully returns.
3655 if (consumer_send_status_msg(sock
, ret_code
) < 0) {
3656 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_FATAL
);
3660 /* Close received socket if valid. */
3663 PERROR("close received socket");
3667 if (relayd_created
) {
3673 * Try to lock the stream mutex.
3675 * On success, 1 is returned else 0 indicating that the mutex is NOT lock.
3677 static int stream_try_lock(struct lttng_consumer_stream
*stream
)
3684 * Try to lock the stream mutex. On failure, we know that the stream is
3685 * being used else where hence there is data still being extracted.
3687 ret
= pthread_mutex_trylock(&stream
->lock
);
3689 /* For both EBUSY and EINVAL error, the mutex is NOT locked. */
3701 * Search for a relayd associated to the session id and return the reference.
3703 * A rcu read side lock MUST be acquire before calling this function and locked
3704 * until the relayd object is no longer necessary.
3706 static struct consumer_relayd_sock_pair
*find_relayd_by_session_id(uint64_t id
)
3708 struct lttng_ht_iter iter
;
3709 struct consumer_relayd_sock_pair
*relayd
= NULL
;
3711 /* Iterate over all relayd since they are indexed by net_seq_idx. */
3712 cds_lfht_for_each_entry(consumer_data
.relayd_ht
->ht
, &iter
.iter
, relayd
,
3715 * Check by sessiond id which is unique here where the relayd session
3716 * id might not be when having multiple relayd.
3718 if (relayd
->sessiond_session_id
== id
) {
3719 /* Found the relayd. There can be only one per id. */
3731 * Check if for a given session id there is still data needed to be extract
3734 * Return 1 if data is pending or else 0 meaning ready to be read.
3736 int consumer_data_pending(uint64_t id
)
3739 struct lttng_ht_iter iter
;
3740 struct lttng_ht
*ht
;
3741 struct lttng_consumer_stream
*stream
;
3742 struct consumer_relayd_sock_pair
*relayd
= NULL
;
3743 int (*data_pending
)(struct lttng_consumer_stream
*);
3745 DBG("Consumer data pending command on session id %" PRIu64
, id
);
3748 pthread_mutex_lock(&consumer_data
.lock
);
3750 switch (consumer_data
.type
) {
3751 case LTTNG_CONSUMER_KERNEL
:
3752 data_pending
= lttng_kconsumer_data_pending
;
3754 case LTTNG_CONSUMER32_UST
:
3755 case LTTNG_CONSUMER64_UST
:
3756 data_pending
= lttng_ustconsumer_data_pending
;
3759 ERR("Unknown consumer data type");
3763 /* Ease our life a bit */
3764 ht
= consumer_data
.stream_list_ht
;
3766 relayd
= find_relayd_by_session_id(id
);
3768 /* Send init command for data pending. */
3769 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
3770 ret
= relayd_begin_data_pending(&relayd
->control_sock
,
3771 relayd
->relayd_session_id
);
3772 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
3774 /* Communication error thus the relayd so no data pending. */
3775 goto data_not_pending
;
3779 cds_lfht_for_each_entry_duplicate(ht
->ht
,
3780 ht
->hash_fct(&id
, lttng_ht_seed
),
3782 &iter
.iter
, stream
, node_session_id
.node
) {
3783 /* If this call fails, the stream is being used hence data pending. */
3784 ret
= stream_try_lock(stream
);
3790 * A removed node from the hash table indicates that the stream has
3791 * been deleted thus having a guarantee that the buffers are closed
3792 * on the consumer side. However, data can still be transmitted
3793 * over the network so don't skip the relayd check.
3795 ret
= cds_lfht_is_node_deleted(&stream
->node
.node
);
3797 /* Check the stream if there is data in the buffers. */
3798 ret
= data_pending(stream
);
3800 pthread_mutex_unlock(&stream
->lock
);
3807 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
3808 if (stream
->metadata_flag
) {
3809 ret
= relayd_quiescent_control(&relayd
->control_sock
,
3810 stream
->relayd_stream_id
);
3812 ret
= relayd_data_pending(&relayd
->control_sock
,
3813 stream
->relayd_stream_id
,
3814 stream
->next_net_seq_num
- 1);
3816 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
3818 pthread_mutex_unlock(&stream
->lock
);
3822 pthread_mutex_unlock(&stream
->lock
);
3826 unsigned int is_data_inflight
= 0;
3828 /* Send init command for data pending. */
3829 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
3830 ret
= relayd_end_data_pending(&relayd
->control_sock
,
3831 relayd
->relayd_session_id
, &is_data_inflight
);
3832 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
3834 goto data_not_pending
;
3836 if (is_data_inflight
) {
3842 * Finding _no_ node in the hash table and no inflight data means that the
3843 * stream(s) have been removed thus data is guaranteed to be available for
3844 * analysis from the trace files.
3848 /* Data is available to be read by a viewer. */
3849 pthread_mutex_unlock(&consumer_data
.lock
);
3854 /* Data is still being extracted from buffers. */
3855 pthread_mutex_unlock(&consumer_data
.lock
);
3861 * Send a ret code status message to the sessiond daemon.
3863 * Return the sendmsg() return value.
3865 int consumer_send_status_msg(int sock
, int ret_code
)
3867 struct lttcomm_consumer_status_msg msg
;
3869 memset(&msg
, 0, sizeof(msg
));
3870 msg
.ret_code
= ret_code
;
3872 return lttcomm_send_unix_sock(sock
, &msg
, sizeof(msg
));
3876 * Send a channel status message to the sessiond daemon.
3878 * Return the sendmsg() return value.
3880 int consumer_send_status_channel(int sock
,
3881 struct lttng_consumer_channel
*channel
)
3883 struct lttcomm_consumer_status_channel msg
;
3887 memset(&msg
, 0, sizeof(msg
));
3889 msg
.ret_code
= LTTCOMM_CONSUMERD_CHANNEL_FAIL
;
3891 msg
.ret_code
= LTTCOMM_CONSUMERD_SUCCESS
;
3892 msg
.key
= channel
->key
;
3893 msg
.stream_count
= channel
->streams
.count
;
3896 return lttcomm_send_unix_sock(sock
, &msg
, sizeof(msg
));
3899 unsigned long consumer_get_consume_start_pos(unsigned long consumed_pos
,
3900 unsigned long produced_pos
, uint64_t nb_packets_per_stream
,
3901 uint64_t max_sb_size
)
3903 unsigned long start_pos
;
3905 if (!nb_packets_per_stream
) {
3906 return consumed_pos
; /* Grab everything */
3908 start_pos
= produced_pos
- offset_align_floor(produced_pos
, max_sb_size
);
3909 start_pos
-= max_sb_size
* nb_packets_per_stream
;
3910 if ((long) (start_pos
- consumed_pos
) < 0) {
3911 return consumed_pos
; /* Grab everything */
3917 int consumer_flush_buffer(struct lttng_consumer_stream
*stream
, int producer_active
)
3921 switch (consumer_data
.type
) {
3922 case LTTNG_CONSUMER_KERNEL
:
3923 ret
= kernctl_buffer_flush(stream
->wait_fd
);
3925 ERR("Failed to flush kernel stream");
3929 case LTTNG_CONSUMER32_UST
:
3930 case LTTNG_CONSUMER64_UST
:
3931 lttng_ustctl_flush_buffer(stream
, producer_active
);
3934 ERR("Unknown consumer_data type");
3943 * Sample the rotate position for all the streams of a channel. If a stream
3944 * is already at the rotate position (produced == consumed), we flag it as
3945 * ready for rotation. The rotation of ready streams occurs after we have
3946 * replied to the session daemon that we have finished sampling the positions.
3948 * Returns 0 on success, < 0 on error
3950 int lttng_consumer_rotate_channel(uint64_t key
, const char *path
,
3951 uint64_t relayd_id
, uint32_t metadata
, uint64_t new_chunk_id
,
3952 struct lttng_consumer_local_data
*ctx
)
3955 struct lttng_consumer_channel
*channel
;
3956 struct lttng_consumer_stream
*stream
;
3957 struct lttng_ht_iter iter
;
3958 struct lttng_ht
*ht
= consumer_data
.stream_per_chan_id_ht
;
3960 DBG("Consumer sample rotate position for channel %" PRIu64
, key
);
3964 channel
= consumer_find_channel(key
);
3966 ERR("No channel found for key %" PRIu64
, key
);
3971 pthread_mutex_lock(&channel
->lock
);
3972 channel
->current_chunk_id
= new_chunk_id
;
3974 ret
= lttng_strncpy(channel
->pathname
, path
, sizeof(channel
->pathname
));
3976 ERR("Failed to copy new path to channel during channel rotation");
3978 goto end_unlock_channel
;
3981 if (relayd_id
== -1ULL) {
3983 * The domain path (/ust or /kernel) has been created before, we
3984 * now need to create the last part of the path: the application/user
3985 * specific section (uid/1000/64-bit).
3987 ret
= utils_mkdir_recursive(channel
->pathname
, S_IRWXU
| S_IRWXG
,
3988 channel
->uid
, channel
->gid
);
3990 ERR("Failed to create trace directory at %s during rotation",
3993 goto end_unlock_channel
;
3997 cds_lfht_for_each_entry_duplicate(ht
->ht
,
3998 ht
->hash_fct(&channel
->key
, lttng_ht_seed
),
3999 ht
->match_fct
, &channel
->key
, &iter
.iter
,
4000 stream
, node_channel_id
.node
) {
4001 unsigned long consumed_pos
;
4003 health_code_update();
4006 * Lock stream because we are about to change its state.
4008 pthread_mutex_lock(&stream
->lock
);
4010 ret
= lttng_strncpy(stream
->channel_read_only_attributes
.path
,
4012 sizeof(stream
->channel_read_only_attributes
.path
));
4014 ERR("Failed to sample channel path name during channel rotation");
4015 goto end_unlock_stream
;
4017 ret
= lttng_consumer_sample_snapshot_positions(stream
);
4019 ERR("Failed to sample snapshot position during channel rotation");
4020 goto end_unlock_stream
;
4023 ret
= lttng_consumer_get_produced_snapshot(stream
,
4024 &stream
->rotate_position
);
4026 ERR("Failed to sample produced position during channel rotation");
4027 goto end_unlock_stream
;
4030 lttng_consumer_get_consumed_snapshot(stream
,
4032 if (consumed_pos
== stream
->rotate_position
) {
4033 stream
->rotate_ready
= true;
4035 channel
->nr_stream_rotate_pending
++;
4037 ret
= consumer_flush_buffer(stream
, 1);
4039 ERR("Failed to flush stream %" PRIu64
" during channel rotation",
4041 goto end_unlock_stream
;
4044 pthread_mutex_unlock(&stream
->lock
);
4046 pthread_mutex_unlock(&channel
->lock
);
4052 pthread_mutex_unlock(&stream
->lock
);
4054 pthread_mutex_unlock(&channel
->lock
);
4061 * Check if a stream is ready to be rotated after extracting it.
4063 * Return 1 if it is ready for rotation, 0 if it is not, a negative value on
4064 * error. Stream lock must be held.
4066 int lttng_consumer_stream_is_rotate_ready(struct lttng_consumer_stream
*stream
)
4069 unsigned long consumed_pos
;
4071 if (!stream
->rotate_position
&& !stream
->rotate_ready
) {
4076 if (stream
->rotate_ready
) {
4082 * If we don't have the rotate_ready flag, check the consumed position
4083 * to determine if we need to rotate.
4085 ret
= lttng_consumer_sample_snapshot_positions(stream
);
4087 ERR("Taking snapshot positions");
4091 ret
= lttng_consumer_get_consumed_snapshot(stream
, &consumed_pos
);
4093 ERR("Consumed snapshot position");
4097 /* Rotate position not reached yet (with check for overflow). */
4098 if ((long) (consumed_pos
- stream
->rotate_position
) < 0) {
4109 * Reset the state for a stream after a rotation occurred.
4111 void lttng_consumer_reset_stream_rotate_state(struct lttng_consumer_stream
*stream
)
4113 stream
->rotate_position
= 0;
4114 stream
->rotate_ready
= false;
4118 * Perform the rotation a local stream file.
4120 int rotate_local_stream(struct lttng_consumer_local_data
*ctx
,
4121 struct lttng_consumer_stream
*stream
)
4125 DBG("Rotate local stream: stream key %" PRIu64
", channel key %" PRIu64
" at path %s",
4128 stream
->channel_read_only_attributes
.path
);
4130 ret
= close(stream
->out_fd
);
4132 PERROR("Closing trace file (fd %d), stream %" PRIu64
,
4133 stream
->out_fd
, stream
->key
);
4138 ret
= utils_create_stream_file(
4139 stream
->channel_read_only_attributes
.path
,
4141 stream
->channel_read_only_attributes
.tracefile_size
,
4142 stream
->tracefile_count_current
,
4143 stream
->uid
, stream
->gid
, NULL
);
4145 ERR("Rotate create stream file");
4148 stream
->out_fd
= ret
;
4149 stream
->tracefile_size_current
= 0;
4151 if (!stream
->metadata_flag
) {
4152 struct lttng_index_file
*index_file
;
4154 lttng_index_file_put(stream
->index_file
);
4156 index_file
= lttng_index_file_create(
4157 stream
->channel_read_only_attributes
.path
,
4158 stream
->name
, stream
->uid
, stream
->gid
,
4159 stream
->channel_read_only_attributes
.tracefile_size
,
4160 stream
->tracefile_count_current
,
4161 CTF_INDEX_MAJOR
, CTF_INDEX_MINOR
);
4163 ERR("Create index file during rotation");
4166 stream
->index_file
= index_file
;
4167 stream
->out_fd_offset
= 0;
4181 * Perform the rotation a stream file on the relay.
4183 int rotate_relay_stream(struct lttng_consumer_local_data
*ctx
,
4184 struct lttng_consumer_stream
*stream
)
4187 struct consumer_relayd_sock_pair
*relayd
;
4189 DBG("Rotate relay stream");
4190 relayd
= consumer_find_relayd(stream
->net_seq_idx
);
4192 ERR("Failed to find relayd");
4197 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
4198 ret
= relayd_rotate_stream(&relayd
->control_sock
,
4199 stream
->relayd_stream_id
,
4200 stream
->channel_read_only_attributes
.path
,
4201 stream
->chan
->current_chunk_id
,
4202 stream
->last_sequence_number
);
4203 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
4205 ERR("Rotate relay stream");
4213 * Performs the stream rotation for the rotate session feature if needed.
4214 * It must be called with the stream lock held.
4216 * Return 0 on success, a negative number of error.
4218 int lttng_consumer_rotate_stream(struct lttng_consumer_local_data
*ctx
,
4219 struct lttng_consumer_stream
*stream
, bool *rotated
)
4223 DBG("Consumer rotate stream %" PRIu64
, stream
->key
);
4225 if (stream
->net_seq_idx
!= (uint64_t) -1ULL) {
4226 ret
= rotate_relay_stream(ctx
, stream
);
4228 ret
= rotate_local_stream(ctx
, stream
);
4231 ERR("Rotate stream");
4235 if (stream
->metadata_flag
) {
4236 switch (consumer_data
.type
) {
4237 case LTTNG_CONSUMER_KERNEL
:
4239 * Reset the position of what has been read from the metadata
4240 * cache to 0 so we can dump it again.
4242 ret
= kernctl_metadata_cache_dump(stream
->wait_fd
);
4244 ERR("Failed to dump the kernel metadata cache after rotation");
4248 case LTTNG_CONSUMER32_UST
:
4249 case LTTNG_CONSUMER64_UST
:
4251 * Reset the position pushed from the metadata cache so it
4252 * will write from the beginning on the next push.
4254 stream
->ust_metadata_pushed
= 0;
4257 ERR("Unknown consumer_data type");
4261 lttng_consumer_reset_stream_rotate_state(stream
);
4274 * Rotate all the ready streams now.
4276 * This is especially important for low throughput streams that have already
4277 * been consumed, we cannot wait for their next packet to perform the
4280 * Returns 0 on success, < 0 on error
4282 int lttng_consumer_rotate_ready_streams(uint64_t key
,
4283 struct lttng_consumer_local_data
*ctx
)
4286 struct lttng_consumer_channel
*channel
;
4287 struct lttng_consumer_stream
*stream
;
4288 struct lttng_ht_iter iter
;
4289 struct lttng_ht
*ht
= consumer_data
.stream_per_chan_id_ht
;
4293 DBG("Consumer rotate ready streams in channel %" PRIu64
, key
);
4295 channel
= consumer_find_channel(key
);
4297 ERR("No channel found for key %" PRIu64
, key
);
4302 cds_lfht_for_each_entry_duplicate(ht
->ht
,
4303 ht
->hash_fct(&channel
->key
, lttng_ht_seed
),
4304 ht
->match_fct
, &channel
->key
, &iter
.iter
,
4305 stream
, node_channel_id
.node
) {
4306 health_code_update();
4308 pthread_mutex_lock(&stream
->lock
);
4310 if (!stream
->rotate_ready
) {
4311 pthread_mutex_unlock(&stream
->lock
);
4314 DBG("Consumer rotate ready stream %" PRIu64
, stream
->key
);
4316 ret
= lttng_consumer_rotate_stream(ctx
, stream
, NULL
);
4317 pthread_mutex_unlock(&stream
->lock
);
4322 ret
= consumer_post_rotation(stream
, ctx
);
4336 int rotate_rename_local(const char *old_path
, const char *new_path
,
4337 uid_t uid
, gid_t gid
)
4344 ret
= utils_mkdir_recursive(new_path
, S_IRWXU
| S_IRWXG
, uid
, gid
);
4346 ERR("Create directory on rotate");
4350 ret
= rename(old_path
, new_path
);
4351 if (ret
< 0 && errno
!= ENOENT
) {
4352 PERROR("Rename completed rotation chunk");
4362 int rotate_rename_relay(const char *old_path
, const char *new_path
,
4366 struct consumer_relayd_sock_pair
*relayd
;
4368 relayd
= consumer_find_relayd(relayd_id
);
4370 ERR("Failed to find relayd while running rotate_rename_relay command");
4375 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
4376 ret
= relayd_rotate_rename(&relayd
->control_sock
, old_path
, new_path
);
4377 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
4382 int lttng_consumer_rotate_rename(const char *old_path
, const char *new_path
,
4383 uid_t uid
, gid_t gid
, uint64_t relayd_id
)
4385 if (relayd_id
!= -1ULL) {
4386 return rotate_rename_relay(old_path
, new_path
, relayd_id
);
4388 return rotate_rename_local(old_path
, new_path
, uid
, gid
);
4392 int lttng_consumer_rotate_pending_relay(uint64_t session_id
,
4393 uint64_t relayd_id
, uint64_t chunk_id
)
4396 struct consumer_relayd_sock_pair
*relayd
;
4398 relayd
= consumer_find_relayd(relayd_id
);
4400 ERR("Failed to find relayd");
4405 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
4406 ret
= relayd_rotate_pending(&relayd
->control_sock
, chunk_id
);
4407 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
4414 int mkdir_local(const char *path
, uid_t uid
, gid_t gid
)
4418 ret
= utils_mkdir_recursive(path
, S_IRWXU
| S_IRWXG
, uid
, gid
);
4420 /* utils_mkdir_recursive logs an error. */
4430 int mkdir_relay(const char *path
, uint64_t relayd_id
)
4433 struct consumer_relayd_sock_pair
*relayd
;
4435 relayd
= consumer_find_relayd(relayd_id
);
4437 ERR("Failed to find relayd");
4442 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
4443 ret
= relayd_mkdir(&relayd
->control_sock
, path
);
4444 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
4451 int lttng_consumer_mkdir(const char *path
, uid_t uid
, gid_t gid
,
4454 if (relayd_id
!= -1ULL) {
4455 return mkdir_relay(path
, relayd_id
);
4457 return mkdir_local(path
, uid
, gid
);