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 <common/common.h>
34 #include <common/utils.h>
35 #include <common/compat/poll.h>
36 #include <common/index/index.h>
37 #include <common/kernel-ctl/kernel-ctl.h>
38 #include <common/sessiond-comm/relayd.h>
39 #include <common/sessiond-comm/sessiond-comm.h>
40 #include <common/kernel-consumer/kernel-consumer.h>
41 #include <common/relayd/relayd.h>
42 #include <common/ust-consumer/ust-consumer.h>
43 #include <common/consumer-timer.h>
46 #include "consumer-stream.h"
47 #include "../bin/lttng-consumerd/health-consumerd.h"
49 struct lttng_consumer_global_data consumer_data
= {
52 .type
= LTTNG_CONSUMER_UNKNOWN
,
55 enum consumer_channel_action
{
58 CONSUMER_CHANNEL_QUIT
,
61 struct consumer_channel_msg
{
62 enum consumer_channel_action action
;
63 struct lttng_consumer_channel
*chan
; /* add */
64 uint64_t key
; /* del */
68 * Flag to inform the polling thread to quit when all fd hung up. Updated by
69 * the consumer_thread_receive_fds when it notices that all fds has hung up.
70 * Also updated by the signal handler (consumer_should_exit()). Read by the
73 volatile int consumer_quit
;
76 * Global hash table containing respectively metadata and data streams. The
77 * stream element in this ht should only be updated by the metadata poll thread
78 * for the metadata and the data poll thread for the data.
80 static struct lttng_ht
*metadata_ht
;
81 static struct lttng_ht
*data_ht
;
84 * Notify a thread lttng pipe to poll back again. This usually means that some
85 * global state has changed so we just send back the thread in a poll wait
88 static void notify_thread_lttng_pipe(struct lttng_pipe
*pipe
)
90 struct lttng_consumer_stream
*null_stream
= NULL
;
94 (void) lttng_pipe_write(pipe
, &null_stream
, sizeof(null_stream
));
97 static void notify_channel_pipe(struct lttng_consumer_local_data
*ctx
,
98 struct lttng_consumer_channel
*chan
,
100 enum consumer_channel_action action
)
102 struct consumer_channel_msg msg
;
105 memset(&msg
, 0, sizeof(msg
));
111 ret
= write(ctx
->consumer_channel_pipe
[1], &msg
, sizeof(msg
));
112 } while (ret
< 0 && errno
== EINTR
);
115 void notify_thread_del_channel(struct lttng_consumer_local_data
*ctx
,
118 notify_channel_pipe(ctx
, NULL
, key
, CONSUMER_CHANNEL_DEL
);
121 static int read_channel_pipe(struct lttng_consumer_local_data
*ctx
,
122 struct lttng_consumer_channel
**chan
,
124 enum consumer_channel_action
*action
)
126 struct consumer_channel_msg msg
;
130 ret
= read(ctx
->consumer_channel_pipe
[0], &msg
, sizeof(msg
));
131 } while (ret
< 0 && errno
== EINTR
);
133 *action
= msg
.action
;
141 * Find a stream. The consumer_data.lock must be locked during this
144 static struct lttng_consumer_stream
*find_stream(uint64_t key
,
147 struct lttng_ht_iter iter
;
148 struct lttng_ht_node_u64
*node
;
149 struct lttng_consumer_stream
*stream
= NULL
;
153 /* -1ULL keys are lookup failures */
154 if (key
== (uint64_t) -1ULL) {
160 lttng_ht_lookup(ht
, &key
, &iter
);
161 node
= lttng_ht_iter_get_node_u64(&iter
);
163 stream
= caa_container_of(node
, struct lttng_consumer_stream
, node
);
171 static void steal_stream_key(uint64_t key
, struct lttng_ht
*ht
)
173 struct lttng_consumer_stream
*stream
;
176 stream
= find_stream(key
, ht
);
178 stream
->key
= (uint64_t) -1ULL;
180 * We don't want the lookup to match, but we still need
181 * to iterate on this stream when iterating over the hash table. Just
182 * change the node key.
184 stream
->node
.key
= (uint64_t) -1ULL;
190 * Return a channel object for the given key.
192 * RCU read side lock MUST be acquired before calling this function and
193 * protects the channel ptr.
195 struct lttng_consumer_channel
*consumer_find_channel(uint64_t key
)
197 struct lttng_ht_iter iter
;
198 struct lttng_ht_node_u64
*node
;
199 struct lttng_consumer_channel
*channel
= NULL
;
201 /* -1ULL keys are lookup failures */
202 if (key
== (uint64_t) -1ULL) {
206 lttng_ht_lookup(consumer_data
.channel_ht
, &key
, &iter
);
207 node
= lttng_ht_iter_get_node_u64(&iter
);
209 channel
= caa_container_of(node
, struct lttng_consumer_channel
, node
);
215 static void free_stream_rcu(struct rcu_head
*head
)
217 struct lttng_ht_node_u64
*node
=
218 caa_container_of(head
, struct lttng_ht_node_u64
, head
);
219 struct lttng_consumer_stream
*stream
=
220 caa_container_of(node
, struct lttng_consumer_stream
, node
);
225 static void free_channel_rcu(struct rcu_head
*head
)
227 struct lttng_ht_node_u64
*node
=
228 caa_container_of(head
, struct lttng_ht_node_u64
, head
);
229 struct lttng_consumer_channel
*channel
=
230 caa_container_of(node
, struct lttng_consumer_channel
, node
);
236 * RCU protected relayd socket pair free.
238 static void free_relayd_rcu(struct rcu_head
*head
)
240 struct lttng_ht_node_u64
*node
=
241 caa_container_of(head
, struct lttng_ht_node_u64
, head
);
242 struct consumer_relayd_sock_pair
*relayd
=
243 caa_container_of(node
, struct consumer_relayd_sock_pair
, node
);
246 * Close all sockets. This is done in the call RCU since we don't want the
247 * socket fds to be reassigned thus potentially creating bad state of the
250 * We do not have to lock the control socket mutex here since at this stage
251 * there is no one referencing to this relayd object.
253 (void) relayd_close(&relayd
->control_sock
);
254 (void) relayd_close(&relayd
->data_sock
);
260 * Destroy and free relayd socket pair object.
262 void consumer_destroy_relayd(struct consumer_relayd_sock_pair
*relayd
)
265 struct lttng_ht_iter iter
;
267 if (relayd
== NULL
) {
271 DBG("Consumer destroy and close relayd socket pair");
273 iter
.iter
.node
= &relayd
->node
.node
;
274 ret
= lttng_ht_del(consumer_data
.relayd_ht
, &iter
);
276 /* We assume the relayd is being or is destroyed */
280 /* RCU free() call */
281 call_rcu(&relayd
->node
.head
, free_relayd_rcu
);
285 * Remove a channel from the global list protected by a mutex. This function is
286 * also responsible for freeing its data structures.
288 void consumer_del_channel(struct lttng_consumer_channel
*channel
)
291 struct lttng_ht_iter iter
;
292 struct lttng_consumer_stream
*stream
, *stmp
;
294 DBG("Consumer delete channel key %" PRIu64
, channel
->key
);
296 pthread_mutex_lock(&consumer_data
.lock
);
297 pthread_mutex_lock(&channel
->lock
);
299 /* Delete streams that might have been left in the stream list. */
300 cds_list_for_each_entry_safe(stream
, stmp
, &channel
->streams
.head
,
302 cds_list_del(&stream
->send_node
);
304 * Once a stream is added to this list, the buffers were created so
305 * we have a guarantee that this call will succeed.
307 consumer_stream_destroy(stream
, NULL
);
310 if (channel
->live_timer_enabled
== 1) {
311 consumer_timer_live_stop(channel
);
314 switch (consumer_data
.type
) {
315 case LTTNG_CONSUMER_KERNEL
:
317 case LTTNG_CONSUMER32_UST
:
318 case LTTNG_CONSUMER64_UST
:
319 lttng_ustconsumer_del_channel(channel
);
322 ERR("Unknown consumer_data type");
328 iter
.iter
.node
= &channel
->node
.node
;
329 ret
= lttng_ht_del(consumer_data
.channel_ht
, &iter
);
333 call_rcu(&channel
->node
.head
, free_channel_rcu
);
335 pthread_mutex_unlock(&channel
->lock
);
336 pthread_mutex_unlock(&consumer_data
.lock
);
340 * Iterate over the relayd hash table and destroy each element. Finally,
341 * destroy the whole hash table.
343 static void cleanup_relayd_ht(void)
345 struct lttng_ht_iter iter
;
346 struct consumer_relayd_sock_pair
*relayd
;
350 cds_lfht_for_each_entry(consumer_data
.relayd_ht
->ht
, &iter
.iter
, relayd
,
352 consumer_destroy_relayd(relayd
);
357 lttng_ht_destroy(consumer_data
.relayd_ht
);
361 * Update the end point status of all streams having the given network sequence
362 * index (relayd index).
364 * It's atomically set without having the stream mutex locked which is fine
365 * because we handle the write/read race with a pipe wakeup for each thread.
367 static void update_endpoint_status_by_netidx(uint64_t net_seq_idx
,
368 enum consumer_endpoint_status status
)
370 struct lttng_ht_iter iter
;
371 struct lttng_consumer_stream
*stream
;
373 DBG("Consumer set delete flag on stream by idx %" PRIu64
, net_seq_idx
);
377 /* Let's begin with metadata */
378 cds_lfht_for_each_entry(metadata_ht
->ht
, &iter
.iter
, stream
, node
.node
) {
379 if (stream
->net_seq_idx
== net_seq_idx
) {
380 uatomic_set(&stream
->endpoint_status
, status
);
381 DBG("Delete flag set to metadata stream %d", stream
->wait_fd
);
385 /* Follow up by the data streams */
386 cds_lfht_for_each_entry(data_ht
->ht
, &iter
.iter
, stream
, node
.node
) {
387 if (stream
->net_seq_idx
== net_seq_idx
) {
388 uatomic_set(&stream
->endpoint_status
, status
);
389 DBG("Delete flag set to data stream %d", stream
->wait_fd
);
396 * Cleanup a relayd object by flagging every associated streams for deletion,
397 * destroying the object meaning removing it from the relayd hash table,
398 * closing the sockets and freeing the memory in a RCU call.
400 * If a local data context is available, notify the threads that the streams'
401 * state have changed.
403 static void cleanup_relayd(struct consumer_relayd_sock_pair
*relayd
,
404 struct lttng_consumer_local_data
*ctx
)
410 DBG("Cleaning up relayd sockets");
412 /* Save the net sequence index before destroying the object */
413 netidx
= relayd
->net_seq_idx
;
416 * Delete the relayd from the relayd hash table, close the sockets and free
417 * the object in a RCU call.
419 consumer_destroy_relayd(relayd
);
421 /* Set inactive endpoint to all streams */
422 update_endpoint_status_by_netidx(netidx
, CONSUMER_ENDPOINT_INACTIVE
);
425 * With a local data context, notify the threads that the streams' state
426 * have changed. The write() action on the pipe acts as an "implicit"
427 * memory barrier ordering the updates of the end point status from the
428 * read of this status which happens AFTER receiving this notify.
431 notify_thread_lttng_pipe(ctx
->consumer_data_pipe
);
432 notify_thread_lttng_pipe(ctx
->consumer_metadata_pipe
);
437 * Flag a relayd socket pair for destruction. Destroy it if the refcount
440 * RCU read side lock MUST be aquired before calling this function.
442 void consumer_flag_relayd_for_destroy(struct consumer_relayd_sock_pair
*relayd
)
446 /* Set destroy flag for this object */
447 uatomic_set(&relayd
->destroy_flag
, 1);
449 /* Destroy the relayd if refcount is 0 */
450 if (uatomic_read(&relayd
->refcount
) == 0) {
451 consumer_destroy_relayd(relayd
);
456 * Completly destroy stream from every visiable data structure and the given
459 * One this call returns, the stream object is not longer usable nor visible.
461 void consumer_del_stream(struct lttng_consumer_stream
*stream
,
464 consumer_stream_destroy(stream
, ht
);
468 * XXX naming of del vs destroy is all mixed up.
470 void consumer_del_stream_for_data(struct lttng_consumer_stream
*stream
)
472 consumer_stream_destroy(stream
, data_ht
);
475 void consumer_del_stream_for_metadata(struct lttng_consumer_stream
*stream
)
477 consumer_stream_destroy(stream
, metadata_ht
);
480 struct lttng_consumer_stream
*consumer_allocate_stream(uint64_t channel_key
,
482 enum lttng_consumer_stream_state state
,
483 const char *channel_name
,
490 enum consumer_channel_type type
,
491 unsigned int monitor
)
494 struct lttng_consumer_stream
*stream
;
496 stream
= zmalloc(sizeof(*stream
));
497 if (stream
== NULL
) {
498 PERROR("malloc struct lttng_consumer_stream");
505 stream
->key
= stream_key
;
507 stream
->out_fd_offset
= 0;
508 stream
->output_written
= 0;
509 stream
->state
= state
;
512 stream
->net_seq_idx
= relayd_id
;
513 stream
->session_id
= session_id
;
514 stream
->monitor
= monitor
;
515 stream
->endpoint_status
= CONSUMER_ENDPOINT_ACTIVE
;
516 stream
->index_fd
= -1;
517 pthread_mutex_init(&stream
->lock
, NULL
);
519 /* If channel is the metadata, flag this stream as metadata. */
520 if (type
== CONSUMER_CHANNEL_TYPE_METADATA
) {
521 stream
->metadata_flag
= 1;
522 /* Metadata is flat out. */
523 strncpy(stream
->name
, DEFAULT_METADATA_NAME
, sizeof(stream
->name
));
524 /* Live rendez-vous point. */
525 pthread_cond_init(&stream
->metadata_rdv
, NULL
);
526 pthread_mutex_init(&stream
->metadata_rdv_lock
, NULL
);
528 /* Format stream name to <channel_name>_<cpu_number> */
529 ret
= snprintf(stream
->name
, sizeof(stream
->name
), "%s_%d",
532 PERROR("snprintf stream name");
537 /* Key is always the wait_fd for streams. */
538 lttng_ht_node_init_u64(&stream
->node
, stream
->key
);
540 /* Init node per channel id key */
541 lttng_ht_node_init_u64(&stream
->node_channel_id
, channel_key
);
543 /* Init session id node with the stream session id */
544 lttng_ht_node_init_u64(&stream
->node_session_id
, stream
->session_id
);
546 DBG3("Allocated stream %s (key %" PRIu64
", chan_key %" PRIu64
547 " relayd_id %" PRIu64
", session_id %" PRIu64
,
548 stream
->name
, stream
->key
, channel_key
,
549 stream
->net_seq_idx
, stream
->session_id
);
565 * Add a stream to the global list protected by a mutex.
567 int consumer_add_data_stream(struct lttng_consumer_stream
*stream
)
569 struct lttng_ht
*ht
= data_ht
;
575 DBG3("Adding consumer stream %" PRIu64
, stream
->key
);
577 pthread_mutex_lock(&consumer_data
.lock
);
578 pthread_mutex_lock(&stream
->chan
->lock
);
579 pthread_mutex_lock(&stream
->chan
->timer_lock
);
580 pthread_mutex_lock(&stream
->lock
);
583 /* Steal stream identifier to avoid having streams with the same key */
584 steal_stream_key(stream
->key
, ht
);
586 lttng_ht_add_unique_u64(ht
, &stream
->node
);
588 lttng_ht_add_u64(consumer_data
.stream_per_chan_id_ht
,
589 &stream
->node_channel_id
);
592 * Add stream to the stream_list_ht of the consumer data. No need to steal
593 * the key since the HT does not use it and we allow to add redundant keys
596 lttng_ht_add_u64(consumer_data
.stream_list_ht
, &stream
->node_session_id
);
599 * When nb_init_stream_left reaches 0, we don't need to trigger any action
600 * in terms of destroying the associated channel, because the action that
601 * causes the count to become 0 also causes a stream to be added. The
602 * channel deletion will thus be triggered by the following removal of this
605 if (uatomic_read(&stream
->chan
->nb_init_stream_left
) > 0) {
606 /* Increment refcount before decrementing nb_init_stream_left */
608 uatomic_dec(&stream
->chan
->nb_init_stream_left
);
611 /* Update consumer data once the node is inserted. */
612 consumer_data
.stream_count
++;
613 consumer_data
.need_update
= 1;
616 pthread_mutex_unlock(&stream
->lock
);
617 pthread_mutex_unlock(&stream
->chan
->timer_lock
);
618 pthread_mutex_unlock(&stream
->chan
->lock
);
619 pthread_mutex_unlock(&consumer_data
.lock
);
624 void consumer_del_data_stream(struct lttng_consumer_stream
*stream
)
626 consumer_del_stream(stream
, data_ht
);
630 * Add relayd socket to global consumer data hashtable. RCU read side lock MUST
631 * be acquired before calling this.
633 static int add_relayd(struct consumer_relayd_sock_pair
*relayd
)
636 struct lttng_ht_node_u64
*node
;
637 struct lttng_ht_iter iter
;
641 lttng_ht_lookup(consumer_data
.relayd_ht
,
642 &relayd
->net_seq_idx
, &iter
);
643 node
= lttng_ht_iter_get_node_u64(&iter
);
647 lttng_ht_add_unique_u64(consumer_data
.relayd_ht
, &relayd
->node
);
654 * Allocate and return a consumer relayd socket.
656 struct consumer_relayd_sock_pair
*consumer_allocate_relayd_sock_pair(
657 uint64_t net_seq_idx
)
659 struct consumer_relayd_sock_pair
*obj
= NULL
;
661 /* net sequence index of -1 is a failure */
662 if (net_seq_idx
== (uint64_t) -1ULL) {
666 obj
= zmalloc(sizeof(struct consumer_relayd_sock_pair
));
668 PERROR("zmalloc relayd sock");
672 obj
->net_seq_idx
= net_seq_idx
;
674 obj
->destroy_flag
= 0;
675 obj
->control_sock
.sock
.fd
= -1;
676 obj
->data_sock
.sock
.fd
= -1;
677 lttng_ht_node_init_u64(&obj
->node
, obj
->net_seq_idx
);
678 pthread_mutex_init(&obj
->ctrl_sock_mutex
, NULL
);
685 * Find a relayd socket pair in the global consumer data.
687 * Return the object if found else NULL.
688 * RCU read-side lock must be held across this call and while using the
691 struct consumer_relayd_sock_pair
*consumer_find_relayd(uint64_t key
)
693 struct lttng_ht_iter iter
;
694 struct lttng_ht_node_u64
*node
;
695 struct consumer_relayd_sock_pair
*relayd
= NULL
;
697 /* Negative keys are lookup failures */
698 if (key
== (uint64_t) -1ULL) {
702 lttng_ht_lookup(consumer_data
.relayd_ht
, &key
,
704 node
= lttng_ht_iter_get_node_u64(&iter
);
706 relayd
= caa_container_of(node
, struct consumer_relayd_sock_pair
, node
);
714 * Find a relayd and send the stream
716 * Returns 0 on success, < 0 on error
718 int consumer_send_relayd_stream(struct lttng_consumer_stream
*stream
,
722 struct consumer_relayd_sock_pair
*relayd
;
725 assert(stream
->net_seq_idx
!= -1ULL);
728 /* The stream is not metadata. Get relayd reference if exists. */
730 relayd
= consumer_find_relayd(stream
->net_seq_idx
);
731 if (relayd
!= NULL
) {
732 /* Add stream on the relayd */
733 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
734 ret
= relayd_add_stream(&relayd
->control_sock
, stream
->name
,
735 path
, &stream
->relayd_stream_id
,
736 stream
->chan
->tracefile_size
, stream
->chan
->tracefile_count
);
737 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
742 uatomic_inc(&relayd
->refcount
);
743 stream
->sent_to_relayd
= 1;
745 ERR("Stream %" PRIu64
" relayd ID %" PRIu64
" unknown. Can't send it.",
746 stream
->key
, stream
->net_seq_idx
);
751 DBG("Stream %s with key %" PRIu64
" sent to relayd id %" PRIu64
,
752 stream
->name
, stream
->key
, stream
->net_seq_idx
);
760 * Find a relayd and close the stream
762 void close_relayd_stream(struct lttng_consumer_stream
*stream
)
764 struct consumer_relayd_sock_pair
*relayd
;
766 /* The stream is not metadata. Get relayd reference if exists. */
768 relayd
= consumer_find_relayd(stream
->net_seq_idx
);
770 consumer_stream_relayd_close(stream
, relayd
);
776 * Handle stream for relayd transmission if the stream applies for network
777 * streaming where the net sequence index is set.
779 * Return destination file descriptor or negative value on error.
781 static int write_relayd_stream_header(struct lttng_consumer_stream
*stream
,
782 size_t data_size
, unsigned long padding
,
783 struct consumer_relayd_sock_pair
*relayd
)
786 struct lttcomm_relayd_data_hdr data_hdr
;
792 /* Reset data header */
793 memset(&data_hdr
, 0, sizeof(data_hdr
));
795 if (stream
->metadata_flag
) {
796 /* Caller MUST acquire the relayd control socket lock */
797 ret
= relayd_send_metadata(&relayd
->control_sock
, data_size
);
802 /* Metadata are always sent on the control socket. */
803 outfd
= relayd
->control_sock
.sock
.fd
;
805 /* Set header with stream information */
806 data_hdr
.stream_id
= htobe64(stream
->relayd_stream_id
);
807 data_hdr
.data_size
= htobe32(data_size
);
808 data_hdr
.padding_size
= htobe32(padding
);
810 * Note that net_seq_num below is assigned with the *current* value of
811 * next_net_seq_num and only after that the next_net_seq_num will be
812 * increment. This is why when issuing a command on the relayd using
813 * this next value, 1 should always be substracted in order to compare
814 * the last seen sequence number on the relayd side to the last sent.
816 data_hdr
.net_seq_num
= htobe64(stream
->next_net_seq_num
);
817 /* Other fields are zeroed previously */
819 ret
= relayd_send_data_hdr(&relayd
->data_sock
, &data_hdr
,
825 ++stream
->next_net_seq_num
;
827 /* Set to go on data socket */
828 outfd
= relayd
->data_sock
.sock
.fd
;
836 * Allocate and return a new lttng_consumer_channel object using the given key
837 * to initialize the hash table node.
839 * On error, return NULL.
841 struct lttng_consumer_channel
*consumer_allocate_channel(uint64_t key
,
843 const char *pathname
,
848 enum lttng_event_output output
,
849 uint64_t tracefile_size
,
850 uint64_t tracefile_count
,
851 uint64_t session_id_per_pid
,
852 unsigned int monitor
,
853 unsigned int live_timer_interval
)
855 struct lttng_consumer_channel
*channel
;
857 channel
= zmalloc(sizeof(*channel
));
858 if (channel
== NULL
) {
859 PERROR("malloc struct lttng_consumer_channel");
864 channel
->refcount
= 0;
865 channel
->session_id
= session_id
;
866 channel
->session_id_per_pid
= session_id_per_pid
;
869 channel
->relayd_id
= relayd_id
;
870 channel
->output
= output
;
871 channel
->tracefile_size
= tracefile_size
;
872 channel
->tracefile_count
= tracefile_count
;
873 channel
->monitor
= monitor
;
874 channel
->live_timer_interval
= live_timer_interval
;
875 pthread_mutex_init(&channel
->lock
, NULL
);
876 pthread_mutex_init(&channel
->timer_lock
, NULL
);
879 * In monitor mode, the streams associated with the channel will be put in
880 * a special list ONLY owned by this channel. So, the refcount is set to 1
881 * here meaning that the channel itself has streams that are referenced.
883 * On a channel deletion, once the channel is no longer visible, the
884 * refcount is decremented and checked for a zero value to delete it. With
885 * streams in no monitor mode, it will now be safe to destroy the channel.
887 if (!channel
->monitor
) {
888 channel
->refcount
= 1;
891 strncpy(channel
->pathname
, pathname
, sizeof(channel
->pathname
));
892 channel
->pathname
[sizeof(channel
->pathname
) - 1] = '\0';
894 strncpy(channel
->name
, name
, sizeof(channel
->name
));
895 channel
->name
[sizeof(channel
->name
) - 1] = '\0';
897 lttng_ht_node_init_u64(&channel
->node
, channel
->key
);
899 channel
->wait_fd
= -1;
901 CDS_INIT_LIST_HEAD(&channel
->streams
.head
);
903 DBG("Allocated channel (key %" PRIu64
")", channel
->key
)
910 * Add a channel to the global list protected by a mutex.
912 * On success 0 is returned else a negative value.
914 int consumer_add_channel(struct lttng_consumer_channel
*channel
,
915 struct lttng_consumer_local_data
*ctx
)
918 struct lttng_ht_node_u64
*node
;
919 struct lttng_ht_iter iter
;
921 pthread_mutex_lock(&consumer_data
.lock
);
922 pthread_mutex_lock(&channel
->lock
);
923 pthread_mutex_lock(&channel
->timer_lock
);
926 lttng_ht_lookup(consumer_data
.channel_ht
, &channel
->key
, &iter
);
927 node
= lttng_ht_iter_get_node_u64(&iter
);
929 /* Channel already exist. Ignore the insertion */
930 ERR("Consumer add channel key %" PRIu64
" already exists!",
936 lttng_ht_add_unique_u64(consumer_data
.channel_ht
, &channel
->node
);
940 pthread_mutex_unlock(&channel
->timer_lock
);
941 pthread_mutex_unlock(&channel
->lock
);
942 pthread_mutex_unlock(&consumer_data
.lock
);
944 if (!ret
&& channel
->wait_fd
!= -1 &&
945 channel
->type
== CONSUMER_CHANNEL_TYPE_DATA
) {
946 notify_channel_pipe(ctx
, channel
, -1, CONSUMER_CHANNEL_ADD
);
952 * Allocate the pollfd structure and the local view of the out fds to avoid
953 * doing a lookup in the linked list and concurrency issues when writing is
954 * needed. Called with consumer_data.lock held.
956 * Returns the number of fds in the structures.
958 static int update_poll_array(struct lttng_consumer_local_data
*ctx
,
959 struct pollfd
**pollfd
, struct lttng_consumer_stream
**local_stream
,
963 struct lttng_ht_iter iter
;
964 struct lttng_consumer_stream
*stream
;
969 assert(local_stream
);
971 DBG("Updating poll fd array");
973 cds_lfht_for_each_entry(ht
->ht
, &iter
.iter
, stream
, node
.node
) {
975 * Only active streams with an active end point can be added to the
976 * poll set and local stream storage of the thread.
978 * There is a potential race here for endpoint_status to be updated
979 * just after the check. However, this is OK since the stream(s) will
980 * be deleted once the thread is notified that the end point state has
981 * changed where this function will be called back again.
983 if (stream
->state
!= LTTNG_CONSUMER_ACTIVE_STREAM
||
984 stream
->endpoint_status
== CONSUMER_ENDPOINT_INACTIVE
) {
988 * This clobbers way too much the debug output. Uncomment that if you
989 * need it for debugging purposes.
991 * DBG("Active FD %d", stream->wait_fd);
993 (*pollfd
)[i
].fd
= stream
->wait_fd
;
994 (*pollfd
)[i
].events
= POLLIN
| POLLPRI
;
995 local_stream
[i
] = stream
;
1001 * Insert the consumer_data_pipe at the end of the array and don't
1002 * increment i so nb_fd is the number of real FD.
1004 (*pollfd
)[i
].fd
= lttng_pipe_get_readfd(ctx
->consumer_data_pipe
);
1005 (*pollfd
)[i
].events
= POLLIN
| POLLPRI
;
1010 * Poll on the should_quit pipe and the command socket return -1 on error and
1011 * should exit, 0 if data is available on the command socket
1013 int lttng_consumer_poll_socket(struct pollfd
*consumer_sockpoll
)
1018 num_rdy
= poll(consumer_sockpoll
, 2, -1);
1019 if (num_rdy
== -1) {
1021 * Restart interrupted system call.
1023 if (errno
== EINTR
) {
1026 PERROR("Poll error");
1029 if (consumer_sockpoll
[0].revents
& (POLLIN
| POLLPRI
)) {
1030 DBG("consumer_should_quit wake up");
1040 * Set the error socket.
1042 void lttng_consumer_set_error_sock(struct lttng_consumer_local_data
*ctx
,
1045 ctx
->consumer_error_socket
= sock
;
1049 * Set the command socket path.
1051 void lttng_consumer_set_command_sock_path(
1052 struct lttng_consumer_local_data
*ctx
, char *sock
)
1054 ctx
->consumer_command_sock_path
= sock
;
1058 * Send return code to the session daemon.
1059 * If the socket is not defined, we return 0, it is not a fatal error
1061 int lttng_consumer_send_error(struct lttng_consumer_local_data
*ctx
, int cmd
)
1063 if (ctx
->consumer_error_socket
> 0) {
1064 return lttcomm_send_unix_sock(ctx
->consumer_error_socket
, &cmd
,
1065 sizeof(enum lttcomm_sessiond_command
));
1072 * Close all the tracefiles and stream fds and MUST be called when all
1073 * instances are destroyed i.e. when all threads were joined and are ended.
1075 void lttng_consumer_cleanup(void)
1077 struct lttng_ht_iter iter
;
1078 struct lttng_consumer_channel
*channel
;
1082 cds_lfht_for_each_entry(consumer_data
.channel_ht
->ht
, &iter
.iter
, channel
,
1084 consumer_del_channel(channel
);
1089 lttng_ht_destroy(consumer_data
.channel_ht
);
1091 cleanup_relayd_ht();
1093 lttng_ht_destroy(consumer_data
.stream_per_chan_id_ht
);
1096 * This HT contains streams that are freed by either the metadata thread or
1097 * the data thread so we do *nothing* on the hash table and simply destroy
1100 lttng_ht_destroy(consumer_data
.stream_list_ht
);
1104 * Called from signal handler.
1106 void lttng_consumer_should_exit(struct lttng_consumer_local_data
*ctx
)
1111 ret
= write(ctx
->consumer_should_quit
[1], "4", 1);
1112 } while (ret
< 0 && errno
== EINTR
);
1113 if (ret
< 0 || ret
!= 1) {
1114 PERROR("write consumer quit");
1117 DBG("Consumer flag that it should quit");
1120 void lttng_consumer_sync_trace_file(struct lttng_consumer_stream
*stream
,
1123 int outfd
= stream
->out_fd
;
1126 * This does a blocking write-and-wait on any page that belongs to the
1127 * subbuffer prior to the one we just wrote.
1128 * Don't care about error values, as these are just hints and ways to
1129 * limit the amount of page cache used.
1131 if (orig_offset
< stream
->max_sb_size
) {
1134 lttng_sync_file_range(outfd
, orig_offset
- stream
->max_sb_size
,
1135 stream
->max_sb_size
,
1136 SYNC_FILE_RANGE_WAIT_BEFORE
1137 | SYNC_FILE_RANGE_WRITE
1138 | SYNC_FILE_RANGE_WAIT_AFTER
);
1140 * Give hints to the kernel about how we access the file:
1141 * POSIX_FADV_DONTNEED : we won't re-access data in a near future after
1144 * We need to call fadvise again after the file grows because the
1145 * kernel does not seem to apply fadvise to non-existing parts of the
1148 * Call fadvise _after_ having waited for the page writeback to
1149 * complete because the dirty page writeback semantic is not well
1150 * defined. So it can be expected to lead to lower throughput in
1153 posix_fadvise(outfd
, orig_offset
- stream
->max_sb_size
,
1154 stream
->max_sb_size
, POSIX_FADV_DONTNEED
);
1158 * Initialise the necessary environnement :
1159 * - create a new context
1160 * - create the poll_pipe
1161 * - create the should_quit pipe (for signal handler)
1162 * - create the thread pipe (for splice)
1164 * Takes a function pointer as argument, this function is called when data is
1165 * available on a buffer. This function is responsible to do the
1166 * kernctl_get_next_subbuf, read the data with mmap or splice depending on the
1167 * buffer configuration and then kernctl_put_next_subbuf at the end.
1169 * Returns a pointer to the new context or NULL on error.
1171 struct lttng_consumer_local_data
*lttng_consumer_create(
1172 enum lttng_consumer_type type
,
1173 ssize_t (*buffer_ready
)(struct lttng_consumer_stream
*stream
,
1174 struct lttng_consumer_local_data
*ctx
),
1175 int (*recv_channel
)(struct lttng_consumer_channel
*channel
),
1176 int (*recv_stream
)(struct lttng_consumer_stream
*stream
),
1177 int (*update_stream
)(uint64_t stream_key
, uint32_t state
))
1180 struct lttng_consumer_local_data
*ctx
;
1182 assert(consumer_data
.type
== LTTNG_CONSUMER_UNKNOWN
||
1183 consumer_data
.type
== type
);
1184 consumer_data
.type
= type
;
1186 ctx
= zmalloc(sizeof(struct lttng_consumer_local_data
));
1188 PERROR("allocating context");
1192 ctx
->consumer_error_socket
= -1;
1193 ctx
->consumer_metadata_socket
= -1;
1194 pthread_mutex_init(&ctx
->metadata_socket_lock
, NULL
);
1195 /* assign the callbacks */
1196 ctx
->on_buffer_ready
= buffer_ready
;
1197 ctx
->on_recv_channel
= recv_channel
;
1198 ctx
->on_recv_stream
= recv_stream
;
1199 ctx
->on_update_stream
= update_stream
;
1201 ctx
->consumer_data_pipe
= lttng_pipe_open(0);
1202 if (!ctx
->consumer_data_pipe
) {
1203 goto error_poll_pipe
;
1206 ret
= pipe(ctx
->consumer_should_quit
);
1208 PERROR("Error creating recv pipe");
1209 goto error_quit_pipe
;
1212 ret
= pipe(ctx
->consumer_thread_pipe
);
1214 PERROR("Error creating thread pipe");
1215 goto error_thread_pipe
;
1218 ret
= pipe(ctx
->consumer_channel_pipe
);
1220 PERROR("Error creating channel pipe");
1221 goto error_channel_pipe
;
1224 ctx
->consumer_metadata_pipe
= lttng_pipe_open(0);
1225 if (!ctx
->consumer_metadata_pipe
) {
1226 goto error_metadata_pipe
;
1229 ret
= utils_create_pipe(ctx
->consumer_splice_metadata_pipe
);
1231 goto error_splice_pipe
;
1237 lttng_pipe_destroy(ctx
->consumer_metadata_pipe
);
1238 error_metadata_pipe
:
1239 utils_close_pipe(ctx
->consumer_channel_pipe
);
1241 utils_close_pipe(ctx
->consumer_thread_pipe
);
1243 utils_close_pipe(ctx
->consumer_should_quit
);
1245 lttng_pipe_destroy(ctx
->consumer_data_pipe
);
1253 * Close all fds associated with the instance and free the context.
1255 void lttng_consumer_destroy(struct lttng_consumer_local_data
*ctx
)
1259 DBG("Consumer destroying it. Closing everything.");
1261 ret
= close(ctx
->consumer_error_socket
);
1265 ret
= close(ctx
->consumer_metadata_socket
);
1269 utils_close_pipe(ctx
->consumer_thread_pipe
);
1270 utils_close_pipe(ctx
->consumer_channel_pipe
);
1271 lttng_pipe_destroy(ctx
->consumer_data_pipe
);
1272 lttng_pipe_destroy(ctx
->consumer_metadata_pipe
);
1273 utils_close_pipe(ctx
->consumer_should_quit
);
1274 utils_close_pipe(ctx
->consumer_splice_metadata_pipe
);
1276 unlink(ctx
->consumer_command_sock_path
);
1281 * Write the metadata stream id on the specified file descriptor.
1283 static int write_relayd_metadata_id(int fd
,
1284 struct lttng_consumer_stream
*stream
,
1285 struct consumer_relayd_sock_pair
*relayd
, unsigned long padding
)
1288 struct lttcomm_relayd_metadata_payload hdr
;
1290 hdr
.stream_id
= htobe64(stream
->relayd_stream_id
);
1291 hdr
.padding_size
= htobe32(padding
);
1293 ret
= write(fd
, (void *) &hdr
, sizeof(hdr
));
1294 } while (ret
< 0 && errno
== EINTR
);
1295 if (ret
< 0 || ret
!= sizeof(hdr
)) {
1297 * This error means that the fd's end is closed so ignore the perror
1298 * not to clubber the error output since this can happen in a normal
1301 if (errno
!= EPIPE
) {
1302 PERROR("write metadata stream id");
1304 DBG3("Consumer failed to write relayd metadata id (errno: %d)", errno
);
1306 * Set ret to a negative value because if ret != sizeof(hdr), we don't
1307 * handle writting the missing part so report that as an error and
1308 * don't lie to the caller.
1313 DBG("Metadata stream id %" PRIu64
" with padding %lu written before data",
1314 stream
->relayd_stream_id
, padding
);
1321 * Mmap the ring buffer, read it and write the data to the tracefile. This is a
1322 * core function for writing trace buffers to either the local filesystem or
1325 * It must be called with the stream lock held.
1327 * Careful review MUST be put if any changes occur!
1329 * Returns the number of bytes written
1331 ssize_t
lttng_consumer_on_read_subbuffer_mmap(
1332 struct lttng_consumer_local_data
*ctx
,
1333 struct lttng_consumer_stream
*stream
, unsigned long len
,
1334 unsigned long padding
,
1335 struct lttng_packet_index
*index
)
1337 unsigned long mmap_offset
;
1339 ssize_t ret
= 0, written
= 0;
1340 off_t orig_offset
= stream
->out_fd_offset
;
1341 /* Default is on the disk */
1342 int outfd
= stream
->out_fd
;
1343 struct consumer_relayd_sock_pair
*relayd
= NULL
;
1344 unsigned int relayd_hang_up
= 0;
1346 /* RCU lock for the relayd pointer */
1349 /* Flag that the current stream if set for network streaming. */
1350 if (stream
->net_seq_idx
!= (uint64_t) -1ULL) {
1351 relayd
= consumer_find_relayd(stream
->net_seq_idx
);
1352 if (relayd
== NULL
) {
1358 /* get the offset inside the fd to mmap */
1359 switch (consumer_data
.type
) {
1360 case LTTNG_CONSUMER_KERNEL
:
1361 mmap_base
= stream
->mmap_base
;
1362 ret
= kernctl_get_mmap_read_offset(stream
->wait_fd
, &mmap_offset
);
1364 PERROR("tracer ctl get_mmap_read_offset");
1369 case LTTNG_CONSUMER32_UST
:
1370 case LTTNG_CONSUMER64_UST
:
1371 mmap_base
= lttng_ustctl_get_mmap_base(stream
);
1373 ERR("read mmap get mmap base for stream %s", stream
->name
);
1377 ret
= lttng_ustctl_get_mmap_read_offset(stream
, &mmap_offset
);
1379 PERROR("tracer ctl get_mmap_read_offset");
1385 ERR("Unknown consumer_data type");
1389 /* Handle stream on the relayd if the output is on the network */
1391 unsigned long netlen
= len
;
1394 * Lock the control socket for the complete duration of the function
1395 * since from this point on we will use the socket.
1397 if (stream
->metadata_flag
) {
1398 /* Metadata requires the control socket. */
1399 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
1400 netlen
+= sizeof(struct lttcomm_relayd_metadata_payload
);
1403 ret
= write_relayd_stream_header(stream
, netlen
, padding
, relayd
);
1405 /* Use the returned socket. */
1408 /* Write metadata stream id before payload */
1409 if (stream
->metadata_flag
) {
1410 ret
= write_relayd_metadata_id(outfd
, stream
, relayd
, padding
);
1413 /* Socket operation failed. We consider the relayd dead */
1414 if (ret
== -EPIPE
|| ret
== -EINVAL
) {
1422 /* Socket operation failed. We consider the relayd dead */
1423 if (ret
== -EPIPE
|| ret
== -EINVAL
) {
1427 /* Else, use the default set before which is the filesystem. */
1430 /* No streaming, we have to set the len with the full padding */
1434 * Check if we need to change the tracefile before writing the packet.
1436 if (stream
->chan
->tracefile_size
> 0 &&
1437 (stream
->tracefile_size_current
+ len
) >
1438 stream
->chan
->tracefile_size
) {
1439 ret
= utils_rotate_stream_file(stream
->chan
->pathname
,
1440 stream
->name
, stream
->chan
->tracefile_size
,
1441 stream
->chan
->tracefile_count
, stream
->uid
, stream
->gid
,
1442 stream
->out_fd
, &(stream
->tracefile_count_current
),
1445 ERR("Rotating output file");
1448 outfd
= stream
->out_fd
;
1450 if (stream
->index_fd
>= 0) {
1451 ret
= index_create_file(stream
->chan
->pathname
,
1452 stream
->name
, stream
->uid
, stream
->gid
,
1453 stream
->chan
->tracefile_size
,
1454 stream
->tracefile_count_current
);
1458 stream
->index_fd
= ret
;
1461 /* Reset current size because we just perform a rotation. */
1462 stream
->tracefile_size_current
= 0;
1463 stream
->out_fd_offset
= 0;
1466 stream
->tracefile_size_current
+= len
;
1468 index
->offset
= htobe64(stream
->out_fd_offset
);
1474 ret
= write(outfd
, mmap_base
+ mmap_offset
, len
);
1475 } while (ret
< 0 && errno
== EINTR
);
1476 DBG("Consumer mmap write() ret %zd (len %lu)", ret
, len
);
1479 * This is possible if the fd is closed on the other side (outfd)
1480 * or any write problem. It can be verbose a bit for a normal
1481 * execution if for instance the relayd is stopped abruptly. This
1482 * can happen so set this to a DBG statement.
1484 DBG("Error in file write mmap");
1488 /* Socket operation failed. We consider the relayd dead */
1489 if (errno
== EPIPE
|| errno
== EINVAL
) {
1494 } else if (ret
> len
) {
1495 PERROR("Error in file write (ret %zd > len %lu)", ret
, len
);
1503 /* This call is useless on a socket so better save a syscall. */
1505 /* This won't block, but will start writeout asynchronously */
1506 lttng_sync_file_range(outfd
, stream
->out_fd_offset
, ret
,
1507 SYNC_FILE_RANGE_WRITE
);
1508 stream
->out_fd_offset
+= ret
;
1510 stream
->output_written
+= ret
;
1513 lttng_consumer_sync_trace_file(stream
, orig_offset
);
1517 * This is a special case that the relayd has closed its socket. Let's
1518 * cleanup the relayd object and all associated streams.
1520 if (relayd
&& relayd_hang_up
) {
1521 cleanup_relayd(relayd
, ctx
);
1525 /* Unlock only if ctrl socket used */
1526 if (relayd
&& stream
->metadata_flag
) {
1527 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
1535 * Splice the data from the ring buffer to the tracefile.
1537 * It must be called with the stream lock held.
1539 * Returns the number of bytes spliced.
1541 ssize_t
lttng_consumer_on_read_subbuffer_splice(
1542 struct lttng_consumer_local_data
*ctx
,
1543 struct lttng_consumer_stream
*stream
, unsigned long len
,
1544 unsigned long padding
,
1545 struct lttng_packet_index
*index
)
1547 ssize_t ret
= 0, written
= 0, ret_splice
= 0;
1549 off_t orig_offset
= stream
->out_fd_offset
;
1550 int fd
= stream
->wait_fd
;
1551 /* Default is on the disk */
1552 int outfd
= stream
->out_fd
;
1553 struct consumer_relayd_sock_pair
*relayd
= NULL
;
1555 unsigned int relayd_hang_up
= 0;
1557 switch (consumer_data
.type
) {
1558 case LTTNG_CONSUMER_KERNEL
:
1560 case LTTNG_CONSUMER32_UST
:
1561 case LTTNG_CONSUMER64_UST
:
1562 /* Not supported for user space tracing */
1565 ERR("Unknown consumer_data type");
1569 /* RCU lock for the relayd pointer */
1572 /* Flag that the current stream if set for network streaming. */
1573 if (stream
->net_seq_idx
!= (uint64_t) -1ULL) {
1574 relayd
= consumer_find_relayd(stream
->net_seq_idx
);
1575 if (relayd
== NULL
) {
1582 * Choose right pipe for splice. Metadata and trace data are handled by
1583 * different threads hence the use of two pipes in order not to race or
1584 * corrupt the written data.
1586 if (stream
->metadata_flag
) {
1587 splice_pipe
= ctx
->consumer_splice_metadata_pipe
;
1589 splice_pipe
= ctx
->consumer_thread_pipe
;
1592 /* Write metadata stream id before payload */
1594 int total_len
= len
;
1596 if (stream
->metadata_flag
) {
1598 * Lock the control socket for the complete duration of the function
1599 * since from this point on we will use the socket.
1601 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
1603 ret
= write_relayd_metadata_id(splice_pipe
[1], stream
, relayd
,
1607 /* Socket operation failed. We consider the relayd dead */
1608 if (ret
== -EBADF
) {
1609 WARN("Remote relayd disconnected. Stopping");
1616 total_len
+= sizeof(struct lttcomm_relayd_metadata_payload
);
1619 ret
= write_relayd_stream_header(stream
, total_len
, padding
, relayd
);
1621 /* Use the returned socket. */
1624 /* Socket operation failed. We consider the relayd dead */
1625 if (ret
== -EBADF
) {
1626 WARN("Remote relayd disconnected. Stopping");
1633 /* No streaming, we have to set the len with the full padding */
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_fd
>= 0) {
1654 ret
= index_create_file(stream
->chan
->pathname
,
1655 stream
->name
, stream
->uid
, stream
->gid
,
1656 stream
->chan
->tracefile_size
,
1657 stream
->tracefile_count_current
);
1661 stream
->index_fd
= ret
;
1664 /* Reset current size because we just perform a rotation. */
1665 stream
->tracefile_size_current
= 0;
1666 stream
->out_fd_offset
= 0;
1669 stream
->tracefile_size_current
+= len
;
1670 index
->offset
= htobe64(stream
->out_fd_offset
);
1674 DBG("splice chan to pipe offset %lu of len %lu (fd : %d, pipe: %d)",
1675 (unsigned long)offset
, len
, fd
, splice_pipe
[1]);
1676 ret_splice
= splice(fd
, &offset
, splice_pipe
[1], NULL
, len
,
1677 SPLICE_F_MOVE
| SPLICE_F_MORE
);
1678 DBG("splice chan to pipe, ret %zd", ret_splice
);
1679 if (ret_splice
< 0) {
1680 PERROR("Error in relay splice");
1682 written
= ret_splice
;
1688 /* Handle stream on the relayd if the output is on the network */
1690 if (stream
->metadata_flag
) {
1691 size_t metadata_payload_size
=
1692 sizeof(struct lttcomm_relayd_metadata_payload
);
1694 /* Update counter to fit the spliced data */
1695 ret_splice
+= metadata_payload_size
;
1696 len
+= metadata_payload_size
;
1698 * We do this so the return value can match the len passed as
1699 * argument to this function.
1701 written
-= metadata_payload_size
;
1705 /* Splice data out */
1706 ret_splice
= splice(splice_pipe
[0], NULL
, outfd
, NULL
,
1707 ret_splice
, SPLICE_F_MOVE
| SPLICE_F_MORE
);
1708 DBG("Consumer splice pipe to file, ret %zd", ret_splice
);
1709 if (ret_splice
< 0) {
1710 PERROR("Error in file splice");
1712 written
= ret_splice
;
1714 /* Socket operation failed. We consider the relayd dead */
1715 if (errno
== EBADF
|| errno
== EPIPE
) {
1716 WARN("Remote relayd disconnected. Stopping");
1722 } else if (ret_splice
> len
) {
1724 PERROR("Wrote more data than requested %zd (len: %lu)",
1726 written
+= ret_splice
;
1732 /* This call is useless on a socket so better save a syscall. */
1734 /* This won't block, but will start writeout asynchronously */
1735 lttng_sync_file_range(outfd
, stream
->out_fd_offset
, ret_splice
,
1736 SYNC_FILE_RANGE_WRITE
);
1737 stream
->out_fd_offset
+= ret_splice
;
1739 stream
->output_written
+= ret_splice
;
1740 written
+= ret_splice
;
1742 lttng_consumer_sync_trace_file(stream
, orig_offset
);
1750 * This is a special case that the relayd has closed its socket. Let's
1751 * cleanup the relayd object and all associated streams.
1753 if (relayd
&& relayd_hang_up
) {
1754 cleanup_relayd(relayd
, ctx
);
1755 /* Skip splice error so the consumer does not fail */
1760 /* send the appropriate error description to sessiond */
1763 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_SPLICE_EINVAL
);
1766 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_SPLICE_ENOMEM
);
1769 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_SPLICE_ESPIPE
);
1774 if (relayd
&& stream
->metadata_flag
) {
1775 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
1783 * Take a snapshot for a specific fd
1785 * Returns 0 on success, < 0 on error
1787 int lttng_consumer_take_snapshot(struct lttng_consumer_stream
*stream
)
1789 switch (consumer_data
.type
) {
1790 case LTTNG_CONSUMER_KERNEL
:
1791 return lttng_kconsumer_take_snapshot(stream
);
1792 case LTTNG_CONSUMER32_UST
:
1793 case LTTNG_CONSUMER64_UST
:
1794 return lttng_ustconsumer_take_snapshot(stream
);
1796 ERR("Unknown consumer_data type");
1803 * Get the produced position
1805 * Returns 0 on success, < 0 on error
1807 int lttng_consumer_get_produced_snapshot(struct lttng_consumer_stream
*stream
,
1810 switch (consumer_data
.type
) {
1811 case LTTNG_CONSUMER_KERNEL
:
1812 return lttng_kconsumer_get_produced_snapshot(stream
, pos
);
1813 case LTTNG_CONSUMER32_UST
:
1814 case LTTNG_CONSUMER64_UST
:
1815 return lttng_ustconsumer_get_produced_snapshot(stream
, pos
);
1817 ERR("Unknown consumer_data type");
1823 int lttng_consumer_recv_cmd(struct lttng_consumer_local_data
*ctx
,
1824 int sock
, struct pollfd
*consumer_sockpoll
)
1826 switch (consumer_data
.type
) {
1827 case LTTNG_CONSUMER_KERNEL
:
1828 return lttng_kconsumer_recv_cmd(ctx
, sock
, consumer_sockpoll
);
1829 case LTTNG_CONSUMER32_UST
:
1830 case LTTNG_CONSUMER64_UST
:
1831 return lttng_ustconsumer_recv_cmd(ctx
, sock
, consumer_sockpoll
);
1833 ERR("Unknown consumer_data type");
1840 * Iterate over all streams of the hashtable and free them properly.
1842 * WARNING: *MUST* be used with data stream only.
1844 static void destroy_data_stream_ht(struct lttng_ht
*ht
)
1846 struct lttng_ht_iter iter
;
1847 struct lttng_consumer_stream
*stream
;
1854 cds_lfht_for_each_entry(ht
->ht
, &iter
.iter
, stream
, node
.node
) {
1856 * Ignore return value since we are currently cleaning up so any error
1859 (void) consumer_del_stream(stream
, ht
);
1863 lttng_ht_destroy(ht
);
1867 * Iterate over all streams of the hashtable and free them properly.
1869 * XXX: Should not be only for metadata stream or else use an other name.
1871 static void destroy_stream_ht(struct lttng_ht
*ht
)
1873 struct lttng_ht_iter iter
;
1874 struct lttng_consumer_stream
*stream
;
1881 cds_lfht_for_each_entry(ht
->ht
, &iter
.iter
, stream
, node
.node
) {
1883 * Ignore return value since we are currently cleaning up so any error
1886 (void) consumer_del_metadata_stream(stream
, ht
);
1890 lttng_ht_destroy(ht
);
1893 void lttng_consumer_close_metadata(void)
1895 switch (consumer_data
.type
) {
1896 case LTTNG_CONSUMER_KERNEL
:
1898 * The Kernel consumer has a different metadata scheme so we don't
1899 * close anything because the stream will be closed by the session
1903 case LTTNG_CONSUMER32_UST
:
1904 case LTTNG_CONSUMER64_UST
:
1906 * Close all metadata streams. The metadata hash table is passed and
1907 * this call iterates over it by closing all wakeup fd. This is safe
1908 * because at this point we are sure that the metadata producer is
1909 * either dead or blocked.
1911 lttng_ustconsumer_close_metadata(metadata_ht
);
1914 ERR("Unknown consumer_data type");
1920 * Clean up a metadata stream and free its memory.
1922 void consumer_del_metadata_stream(struct lttng_consumer_stream
*stream
,
1923 struct lttng_ht
*ht
)
1926 struct lttng_ht_iter iter
;
1927 struct lttng_consumer_channel
*free_chan
= NULL
;
1928 struct consumer_relayd_sock_pair
*relayd
;
1932 * This call should NEVER receive regular stream. It must always be
1933 * metadata stream and this is crucial for data structure synchronization.
1935 assert(stream
->metadata_flag
);
1937 DBG3("Consumer delete metadata stream %d", stream
->wait_fd
);
1940 /* Means the stream was allocated but not successfully added */
1941 goto free_stream_rcu
;
1944 pthread_mutex_lock(&consumer_data
.lock
);
1945 pthread_mutex_lock(&stream
->chan
->lock
);
1946 pthread_mutex_lock(&stream
->lock
);
1948 switch (consumer_data
.type
) {
1949 case LTTNG_CONSUMER_KERNEL
:
1950 if (stream
->mmap_base
!= NULL
) {
1951 ret
= munmap(stream
->mmap_base
, stream
->mmap_len
);
1953 PERROR("munmap metadata stream");
1956 if (stream
->wait_fd
>= 0) {
1957 ret
= close(stream
->wait_fd
);
1959 PERROR("close kernel metadata wait_fd");
1963 case LTTNG_CONSUMER32_UST
:
1964 case LTTNG_CONSUMER64_UST
:
1965 if (stream
->monitor
) {
1966 /* close the write-side in close_metadata */
1967 ret
= close(stream
->ust_metadata_poll_pipe
[0]);
1969 PERROR("Close UST metadata read-side poll pipe");
1972 lttng_ustconsumer_del_stream(stream
);
1975 ERR("Unknown consumer_data type");
1981 iter
.iter
.node
= &stream
->node
.node
;
1982 ret
= lttng_ht_del(ht
, &iter
);
1985 iter
.iter
.node
= &stream
->node_channel_id
.node
;
1986 ret
= lttng_ht_del(consumer_data
.stream_per_chan_id_ht
, &iter
);
1989 iter
.iter
.node
= &stream
->node_session_id
.node
;
1990 ret
= lttng_ht_del(consumer_data
.stream_list_ht
, &iter
);
1994 if (stream
->out_fd
>= 0) {
1995 ret
= close(stream
->out_fd
);
2001 /* Check and cleanup relayd */
2003 relayd
= consumer_find_relayd(stream
->net_seq_idx
);
2004 if (relayd
!= NULL
) {
2005 uatomic_dec(&relayd
->refcount
);
2006 assert(uatomic_read(&relayd
->refcount
) >= 0);
2008 /* Closing streams requires to lock the control socket. */
2009 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
2010 ret
= relayd_send_close_stream(&relayd
->control_sock
,
2011 stream
->relayd_stream_id
, stream
->next_net_seq_num
- 1);
2012 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
2014 DBG("Unable to close stream on the relayd. Continuing");
2016 * Continue here. There is nothing we can do for the relayd.
2017 * Chances are that the relayd has closed the socket so we just
2018 * continue cleaning up.
2022 /* Both conditions are met, we destroy the relayd. */
2023 if (uatomic_read(&relayd
->refcount
) == 0 &&
2024 uatomic_read(&relayd
->destroy_flag
)) {
2025 consumer_destroy_relayd(relayd
);
2030 /* Atomically decrement channel refcount since other threads can use it. */
2031 if (!uatomic_sub_return(&stream
->chan
->refcount
, 1)
2032 && !uatomic_read(&stream
->chan
->nb_init_stream_left
)) {
2033 /* Go for channel deletion! */
2034 free_chan
= stream
->chan
;
2039 * Nullify the stream reference so it is not used after deletion. The
2040 * channel lock MUST be acquired before being able to check for
2041 * a NULL pointer value.
2043 stream
->chan
->metadata_stream
= NULL
;
2045 pthread_mutex_unlock(&stream
->lock
);
2046 pthread_mutex_unlock(&stream
->chan
->lock
);
2047 pthread_mutex_unlock(&consumer_data
.lock
);
2050 consumer_del_channel(free_chan
);
2054 call_rcu(&stream
->node
.head
, free_stream_rcu
);
2058 * Action done with the metadata stream when adding it to the consumer internal
2059 * data structures to handle it.
2061 int consumer_add_metadata_stream(struct lttng_consumer_stream
*stream
)
2063 struct lttng_ht
*ht
= metadata_ht
;
2065 struct lttng_ht_iter iter
;
2066 struct lttng_ht_node_u64
*node
;
2071 DBG3("Adding metadata stream %" PRIu64
" to hash table", stream
->key
);
2073 pthread_mutex_lock(&consumer_data
.lock
);
2074 pthread_mutex_lock(&stream
->chan
->lock
);
2075 pthread_mutex_lock(&stream
->chan
->timer_lock
);
2076 pthread_mutex_lock(&stream
->lock
);
2079 * From here, refcounts are updated so be _careful_ when returning an error
2086 * Lookup the stream just to make sure it does not exist in our internal
2087 * state. This should NEVER happen.
2089 lttng_ht_lookup(ht
, &stream
->key
, &iter
);
2090 node
= lttng_ht_iter_get_node_u64(&iter
);
2094 * When nb_init_stream_left reaches 0, we don't need to trigger any action
2095 * in terms of destroying the associated channel, because the action that
2096 * causes the count to become 0 also causes a stream to be added. The
2097 * channel deletion will thus be triggered by the following removal of this
2100 if (uatomic_read(&stream
->chan
->nb_init_stream_left
) > 0) {
2101 /* Increment refcount before decrementing nb_init_stream_left */
2103 uatomic_dec(&stream
->chan
->nb_init_stream_left
);
2106 lttng_ht_add_unique_u64(ht
, &stream
->node
);
2108 lttng_ht_add_unique_u64(consumer_data
.stream_per_chan_id_ht
,
2109 &stream
->node_channel_id
);
2112 * Add stream to the stream_list_ht of the consumer data. No need to steal
2113 * the key since the HT does not use it and we allow to add redundant keys
2116 lttng_ht_add_u64(consumer_data
.stream_list_ht
, &stream
->node_session_id
);
2120 pthread_mutex_unlock(&stream
->lock
);
2121 pthread_mutex_unlock(&stream
->chan
->lock
);
2122 pthread_mutex_unlock(&stream
->chan
->timer_lock
);
2123 pthread_mutex_unlock(&consumer_data
.lock
);
2128 * Delete data stream that are flagged for deletion (endpoint_status).
2130 static void validate_endpoint_status_data_stream(void)
2132 struct lttng_ht_iter iter
;
2133 struct lttng_consumer_stream
*stream
;
2135 DBG("Consumer delete flagged data stream");
2138 cds_lfht_for_each_entry(data_ht
->ht
, &iter
.iter
, stream
, node
.node
) {
2139 /* Validate delete flag of the stream */
2140 if (stream
->endpoint_status
== CONSUMER_ENDPOINT_ACTIVE
) {
2143 /* Delete it right now */
2144 consumer_del_stream(stream
, data_ht
);
2150 * Delete metadata stream that are flagged for deletion (endpoint_status).
2152 static void validate_endpoint_status_metadata_stream(
2153 struct lttng_poll_event
*pollset
)
2155 struct lttng_ht_iter iter
;
2156 struct lttng_consumer_stream
*stream
;
2158 DBG("Consumer delete flagged metadata stream");
2163 cds_lfht_for_each_entry(metadata_ht
->ht
, &iter
.iter
, stream
, node
.node
) {
2164 /* Validate delete flag of the stream */
2165 if (stream
->endpoint_status
== CONSUMER_ENDPOINT_ACTIVE
) {
2169 * Remove from pollset so the metadata thread can continue without
2170 * blocking on a deleted stream.
2172 lttng_poll_del(pollset
, stream
->wait_fd
);
2174 /* Delete it right now */
2175 consumer_del_metadata_stream(stream
, metadata_ht
);
2181 * Thread polls on metadata file descriptor and write them on disk or on the
2184 void *consumer_thread_metadata_poll(void *data
)
2186 int ret
, i
, pollfd
, err
= -1;
2187 uint32_t revents
, nb_fd
;
2188 struct lttng_consumer_stream
*stream
= NULL
;
2189 struct lttng_ht_iter iter
;
2190 struct lttng_ht_node_u64
*node
;
2191 struct lttng_poll_event events
;
2192 struct lttng_consumer_local_data
*ctx
= data
;
2195 rcu_register_thread();
2197 health_register(health_consumerd
, HEALTH_CONSUMERD_TYPE_METADATA
);
2199 health_code_update();
2201 metadata_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
2203 /* ENOMEM at this point. Better to bail out. */
2207 DBG("Thread metadata poll started");
2209 /* Size is set to 1 for the consumer_metadata pipe */
2210 ret
= lttng_poll_create(&events
, 2, LTTNG_CLOEXEC
);
2212 ERR("Poll set creation failed");
2216 ret
= lttng_poll_add(&events
,
2217 lttng_pipe_get_readfd(ctx
->consumer_metadata_pipe
), LPOLLIN
);
2223 DBG("Metadata main loop started");
2226 health_code_update();
2228 /* Only the metadata pipe is set */
2229 if (LTTNG_POLL_GETNB(&events
) == 0 && consumer_quit
== 1) {
2230 err
= 0; /* All is OK */
2235 DBG("Metadata poll wait with %d fd(s)", LTTNG_POLL_GETNB(&events
));
2236 health_poll_entry();
2237 ret
= lttng_poll_wait(&events
, -1);
2239 DBG("Metadata event catched in thread");
2241 if (errno
== EINTR
) {
2242 ERR("Poll EINTR catched");
2250 /* From here, the event is a metadata wait fd */
2251 for (i
= 0; i
< nb_fd
; i
++) {
2252 health_code_update();
2254 revents
= LTTNG_POLL_GETEV(&events
, i
);
2255 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2257 if (pollfd
== lttng_pipe_get_readfd(ctx
->consumer_metadata_pipe
)) {
2258 if (revents
& (LPOLLERR
| LPOLLHUP
)) {
2259 DBG("Metadata thread pipe hung up");
2261 * Remove the pipe from the poll set and continue the loop
2262 * since their might be data to consume.
2264 lttng_poll_del(&events
,
2265 lttng_pipe_get_readfd(ctx
->consumer_metadata_pipe
));
2266 lttng_pipe_read_close(ctx
->consumer_metadata_pipe
);
2268 } else if (revents
& LPOLLIN
) {
2271 pipe_len
= lttng_pipe_read(ctx
->consumer_metadata_pipe
,
2272 &stream
, sizeof(stream
));
2274 ERR("read metadata stream, ret: %zd", pipe_len
);
2276 * Continue here to handle the rest of the streams.
2281 /* A NULL stream means that the state has changed. */
2282 if (stream
== NULL
) {
2283 /* Check for deleted streams. */
2284 validate_endpoint_status_metadata_stream(&events
);
2288 DBG("Adding metadata stream %d to poll set",
2291 /* Add metadata stream to the global poll events list */
2292 lttng_poll_add(&events
, stream
->wait_fd
,
2293 LPOLLIN
| LPOLLPRI
);
2296 /* Handle other stream */
2302 uint64_t tmp_id
= (uint64_t) pollfd
;
2304 lttng_ht_lookup(metadata_ht
, &tmp_id
, &iter
);
2306 node
= lttng_ht_iter_get_node_u64(&iter
);
2309 stream
= caa_container_of(node
, struct lttng_consumer_stream
,
2312 /* Check for error event */
2313 if (revents
& (LPOLLERR
| LPOLLHUP
)) {
2314 DBG("Metadata fd %d is hup|err.", pollfd
);
2315 if (!stream
->hangup_flush_done
2316 && (consumer_data
.type
== LTTNG_CONSUMER32_UST
2317 || consumer_data
.type
== LTTNG_CONSUMER64_UST
)) {
2318 DBG("Attempting to flush and consume the UST buffers");
2319 lttng_ustconsumer_on_stream_hangup(stream
);
2321 /* We just flushed the stream now read it. */
2323 health_code_update();
2325 len
= ctx
->on_buffer_ready(stream
, ctx
);
2327 * We don't check the return value here since if we get
2328 * a negative len, it means an error occured thus we
2329 * simply remove it from the poll set and free the
2335 lttng_poll_del(&events
, stream
->wait_fd
);
2337 * This call update the channel states, closes file descriptors
2338 * and securely free the stream.
2340 consumer_del_metadata_stream(stream
, metadata_ht
);
2341 } else if (revents
& (LPOLLIN
| LPOLLPRI
)) {
2342 /* Get the data out of the metadata file descriptor */
2343 DBG("Metadata available on fd %d", pollfd
);
2344 assert(stream
->wait_fd
== pollfd
);
2347 health_code_update();
2349 len
= ctx
->on_buffer_ready(stream
, ctx
);
2351 * We don't check the return value here since if we get
2352 * a negative len, it means an error occured thus we
2353 * simply remove it from the poll set and free the
2358 /* It's ok to have an unavailable sub-buffer */
2359 if (len
< 0 && len
!= -EAGAIN
&& len
!= -ENODATA
) {
2360 /* Clean up stream from consumer and free it. */
2361 lttng_poll_del(&events
, stream
->wait_fd
);
2362 consumer_del_metadata_stream(stream
, metadata_ht
);
2366 /* Release RCU lock for the stream looked up */
2375 DBG("Metadata poll thread exiting");
2377 lttng_poll_clean(&events
);
2379 destroy_stream_ht(metadata_ht
);
2383 ERR("Health error occurred in %s", __func__
);
2385 health_unregister(health_consumerd
);
2386 rcu_unregister_thread();
2391 * This thread polls the fds in the set to consume the data and write
2392 * it to tracefile if necessary.
2394 void *consumer_thread_data_poll(void *data
)
2396 int num_rdy
, num_hup
, high_prio
, ret
, i
, err
= -1;
2397 struct pollfd
*pollfd
= NULL
;
2398 /* local view of the streams */
2399 struct lttng_consumer_stream
**local_stream
= NULL
, *new_stream
= NULL
;
2400 /* local view of consumer_data.fds_count */
2402 struct lttng_consumer_local_data
*ctx
= data
;
2405 rcu_register_thread();
2407 health_register(health_consumerd
, HEALTH_CONSUMERD_TYPE_DATA
);
2409 health_code_update();
2411 data_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
2412 if (data_ht
== NULL
) {
2413 /* ENOMEM at this point. Better to bail out. */
2417 local_stream
= zmalloc(sizeof(struct lttng_consumer_stream
*));
2418 if (local_stream
== NULL
) {
2419 PERROR("local_stream malloc");
2424 health_code_update();
2430 * the fds set has been updated, we need to update our
2431 * local array as well
2433 pthread_mutex_lock(&consumer_data
.lock
);
2434 if (consumer_data
.need_update
) {
2439 local_stream
= NULL
;
2441 /* allocate for all fds + 1 for the consumer_data_pipe */
2442 pollfd
= zmalloc((consumer_data
.stream_count
+ 1) * sizeof(struct pollfd
));
2443 if (pollfd
== NULL
) {
2444 PERROR("pollfd malloc");
2445 pthread_mutex_unlock(&consumer_data
.lock
);
2449 /* allocate for all fds + 1 for the consumer_data_pipe */
2450 local_stream
= zmalloc((consumer_data
.stream_count
+ 1) *
2451 sizeof(struct lttng_consumer_stream
*));
2452 if (local_stream
== NULL
) {
2453 PERROR("local_stream malloc");
2454 pthread_mutex_unlock(&consumer_data
.lock
);
2457 ret
= update_poll_array(ctx
, &pollfd
, local_stream
,
2460 ERR("Error in allocating pollfd or local_outfds");
2461 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_POLL_ERROR
);
2462 pthread_mutex_unlock(&consumer_data
.lock
);
2466 consumer_data
.need_update
= 0;
2468 pthread_mutex_unlock(&consumer_data
.lock
);
2470 /* No FDs and consumer_quit, consumer_cleanup the thread */
2471 if (nb_fd
== 0 && consumer_quit
== 1) {
2472 err
= 0; /* All is OK */
2475 /* poll on the array of fds */
2477 DBG("polling on %d fd", nb_fd
+ 1);
2478 health_poll_entry();
2479 num_rdy
= poll(pollfd
, nb_fd
+ 1, -1);
2481 DBG("poll num_rdy : %d", num_rdy
);
2482 if (num_rdy
== -1) {
2484 * Restart interrupted system call.
2486 if (errno
== EINTR
) {
2489 PERROR("Poll error");
2490 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_POLL_ERROR
);
2492 } else if (num_rdy
== 0) {
2493 DBG("Polling thread timed out");
2498 * If the consumer_data_pipe triggered poll go directly to the
2499 * beginning of the loop to update the array. We want to prioritize
2500 * array update over low-priority reads.
2502 if (pollfd
[nb_fd
].revents
& (POLLIN
| POLLPRI
)) {
2503 ssize_t pipe_readlen
;
2505 DBG("consumer_data_pipe wake up");
2506 pipe_readlen
= lttng_pipe_read(ctx
->consumer_data_pipe
,
2507 &new_stream
, sizeof(new_stream
));
2508 if (pipe_readlen
< 0) {
2509 ERR("Consumer data pipe ret %zd", pipe_readlen
);
2510 /* Continue so we can at least handle the current stream(s). */
2515 * If the stream is NULL, just ignore it. It's also possible that
2516 * the sessiond poll thread changed the consumer_quit state and is
2517 * waking us up to test it.
2519 if (new_stream
== NULL
) {
2520 validate_endpoint_status_data_stream();
2524 /* Continue to update the local streams and handle prio ones */
2528 /* Take care of high priority channels first. */
2529 for (i
= 0; i
< nb_fd
; i
++) {
2530 health_code_update();
2532 if (local_stream
[i
] == NULL
) {
2535 if (pollfd
[i
].revents
& POLLPRI
) {
2536 DBG("Urgent read on fd %d", pollfd
[i
].fd
);
2538 len
= ctx
->on_buffer_ready(local_stream
[i
], ctx
);
2539 /* it's ok to have an unavailable sub-buffer */
2540 if (len
< 0 && len
!= -EAGAIN
&& len
!= -ENODATA
) {
2541 /* Clean the stream and free it. */
2542 consumer_del_stream(local_stream
[i
], data_ht
);
2543 local_stream
[i
] = NULL
;
2544 } else if (len
> 0) {
2545 local_stream
[i
]->data_read
= 1;
2551 * If we read high prio channel in this loop, try again
2552 * for more high prio data.
2558 /* Take care of low priority channels. */
2559 for (i
= 0; i
< nb_fd
; i
++) {
2560 health_code_update();
2562 if (local_stream
[i
] == NULL
) {
2565 if ((pollfd
[i
].revents
& POLLIN
) ||
2566 local_stream
[i
]->hangup_flush_done
) {
2567 DBG("Normal read on fd %d", pollfd
[i
].fd
);
2568 len
= ctx
->on_buffer_ready(local_stream
[i
], ctx
);
2569 /* it's ok to have an unavailable sub-buffer */
2570 if (len
< 0 && len
!= -EAGAIN
&& len
!= -ENODATA
) {
2571 /* Clean the stream and free it. */
2572 consumer_del_stream(local_stream
[i
], data_ht
);
2573 local_stream
[i
] = NULL
;
2574 } else if (len
> 0) {
2575 local_stream
[i
]->data_read
= 1;
2580 /* Handle hangup and errors */
2581 for (i
= 0; i
< nb_fd
; i
++) {
2582 health_code_update();
2584 if (local_stream
[i
] == NULL
) {
2587 if (!local_stream
[i
]->hangup_flush_done
2588 && (pollfd
[i
].revents
& (POLLHUP
| POLLERR
| POLLNVAL
))
2589 && (consumer_data
.type
== LTTNG_CONSUMER32_UST
2590 || consumer_data
.type
== LTTNG_CONSUMER64_UST
)) {
2591 DBG("fd %d is hup|err|nval. Attempting flush and read.",
2593 lttng_ustconsumer_on_stream_hangup(local_stream
[i
]);
2594 /* Attempt read again, for the data we just flushed. */
2595 local_stream
[i
]->data_read
= 1;
2598 * If the poll flag is HUP/ERR/NVAL and we have
2599 * read no data in this pass, we can remove the
2600 * stream from its hash table.
2602 if ((pollfd
[i
].revents
& POLLHUP
)) {
2603 DBG("Polling fd %d tells it has hung up.", pollfd
[i
].fd
);
2604 if (!local_stream
[i
]->data_read
) {
2605 consumer_del_stream(local_stream
[i
], data_ht
);
2606 local_stream
[i
] = NULL
;
2609 } else if (pollfd
[i
].revents
& POLLERR
) {
2610 ERR("Error returned in polling fd %d.", pollfd
[i
].fd
);
2611 if (!local_stream
[i
]->data_read
) {
2612 consumer_del_stream(local_stream
[i
], data_ht
);
2613 local_stream
[i
] = NULL
;
2616 } else if (pollfd
[i
].revents
& POLLNVAL
) {
2617 ERR("Polling fd %d tells fd is not open.", pollfd
[i
].fd
);
2618 if (!local_stream
[i
]->data_read
) {
2619 consumer_del_stream(local_stream
[i
], data_ht
);
2620 local_stream
[i
] = NULL
;
2624 if (local_stream
[i
] != NULL
) {
2625 local_stream
[i
]->data_read
= 0;
2632 DBG("polling thread exiting");
2637 * Close the write side of the pipe so epoll_wait() in
2638 * consumer_thread_metadata_poll can catch it. The thread is monitoring the
2639 * read side of the pipe. If we close them both, epoll_wait strangely does
2640 * not return and could create a endless wait period if the pipe is the
2641 * only tracked fd in the poll set. The thread will take care of closing
2644 (void) lttng_pipe_write_close(ctx
->consumer_metadata_pipe
);
2646 destroy_data_stream_ht(data_ht
);
2650 ERR("Health error occurred in %s", __func__
);
2652 health_unregister(health_consumerd
);
2654 rcu_unregister_thread();
2659 * Close wake-up end of each stream belonging to the channel. This will
2660 * allow the poll() on the stream read-side to detect when the
2661 * write-side (application) finally closes them.
2664 void consumer_close_channel_streams(struct lttng_consumer_channel
*channel
)
2666 struct lttng_ht
*ht
;
2667 struct lttng_consumer_stream
*stream
;
2668 struct lttng_ht_iter iter
;
2670 ht
= consumer_data
.stream_per_chan_id_ht
;
2673 cds_lfht_for_each_entry_duplicate(ht
->ht
,
2674 ht
->hash_fct(&channel
->key
, lttng_ht_seed
),
2675 ht
->match_fct
, &channel
->key
,
2676 &iter
.iter
, stream
, node_channel_id
.node
) {
2678 * Protect against teardown with mutex.
2680 pthread_mutex_lock(&stream
->lock
);
2681 if (cds_lfht_is_node_deleted(&stream
->node
.node
)) {
2684 switch (consumer_data
.type
) {
2685 case LTTNG_CONSUMER_KERNEL
:
2687 case LTTNG_CONSUMER32_UST
:
2688 case LTTNG_CONSUMER64_UST
:
2690 * Note: a mutex is taken internally within
2691 * liblttng-ust-ctl to protect timer wakeup_fd
2692 * use from concurrent close.
2694 lttng_ustconsumer_close_stream_wakeup(stream
);
2697 ERR("Unknown consumer_data type");
2701 pthread_mutex_unlock(&stream
->lock
);
2706 static void destroy_channel_ht(struct lttng_ht
*ht
)
2708 struct lttng_ht_iter iter
;
2709 struct lttng_consumer_channel
*channel
;
2717 cds_lfht_for_each_entry(ht
->ht
, &iter
.iter
, channel
, wait_fd_node
.node
) {
2718 ret
= lttng_ht_del(ht
, &iter
);
2723 lttng_ht_destroy(ht
);
2727 * This thread polls the channel fds to detect when they are being
2728 * closed. It closes all related streams if the channel is detected as
2729 * closed. It is currently only used as a shim layer for UST because the
2730 * consumerd needs to keep the per-stream wakeup end of pipes open for
2733 void *consumer_thread_channel_poll(void *data
)
2735 int ret
, i
, pollfd
, err
= -1;
2736 uint32_t revents
, nb_fd
;
2737 struct lttng_consumer_channel
*chan
= NULL
;
2738 struct lttng_ht_iter iter
;
2739 struct lttng_ht_node_u64
*node
;
2740 struct lttng_poll_event events
;
2741 struct lttng_consumer_local_data
*ctx
= data
;
2742 struct lttng_ht
*channel_ht
;
2744 rcu_register_thread();
2746 health_register(health_consumerd
, HEALTH_CONSUMERD_TYPE_CHANNEL
);
2748 health_code_update();
2750 channel_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
2752 /* ENOMEM at this point. Better to bail out. */
2756 DBG("Thread channel poll started");
2758 /* Size is set to 1 for the consumer_channel pipe */
2759 ret
= lttng_poll_create(&events
, 2, LTTNG_CLOEXEC
);
2761 ERR("Poll set creation failed");
2765 ret
= lttng_poll_add(&events
, ctx
->consumer_channel_pipe
[0], LPOLLIN
);
2771 DBG("Channel main loop started");
2774 health_code_update();
2776 /* Only the channel pipe is set */
2777 if (LTTNG_POLL_GETNB(&events
) == 0 && consumer_quit
== 1) {
2778 err
= 0; /* All is OK */
2783 DBG("Channel poll wait with %d fd(s)", LTTNG_POLL_GETNB(&events
));
2784 health_poll_entry();
2785 ret
= lttng_poll_wait(&events
, -1);
2787 DBG("Channel event catched in thread");
2789 if (errno
== EINTR
) {
2790 ERR("Poll EINTR catched");
2798 /* From here, the event is a channel wait fd */
2799 for (i
= 0; i
< nb_fd
; i
++) {
2800 health_code_update();
2802 revents
= LTTNG_POLL_GETEV(&events
, i
);
2803 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2805 /* Just don't waste time if no returned events for the fd */
2809 if (pollfd
== ctx
->consumer_channel_pipe
[0]) {
2810 if (revents
& (LPOLLERR
| LPOLLHUP
)) {
2811 DBG("Channel thread pipe hung up");
2813 * Remove the pipe from the poll set and continue the loop
2814 * since their might be data to consume.
2816 lttng_poll_del(&events
, ctx
->consumer_channel_pipe
[0]);
2818 } else if (revents
& LPOLLIN
) {
2819 enum consumer_channel_action action
;
2822 ret
= read_channel_pipe(ctx
, &chan
, &key
, &action
);
2824 ERR("Error reading channel pipe");
2829 case CONSUMER_CHANNEL_ADD
:
2830 DBG("Adding channel %d to poll set",
2833 lttng_ht_node_init_u64(&chan
->wait_fd_node
,
2836 lttng_ht_add_unique_u64(channel_ht
,
2837 &chan
->wait_fd_node
);
2839 /* Add channel to the global poll events list */
2840 lttng_poll_add(&events
, chan
->wait_fd
,
2841 LPOLLIN
| LPOLLPRI
);
2843 case CONSUMER_CHANNEL_DEL
:
2845 struct lttng_consumer_stream
*stream
, *stmp
;
2848 chan
= consumer_find_channel(key
);
2851 ERR("UST consumer get channel key %" PRIu64
" not found for del channel", key
);
2854 lttng_poll_del(&events
, chan
->wait_fd
);
2855 iter
.iter
.node
= &chan
->wait_fd_node
.node
;
2856 ret
= lttng_ht_del(channel_ht
, &iter
);
2858 consumer_close_channel_streams(chan
);
2860 switch (consumer_data
.type
) {
2861 case LTTNG_CONSUMER_KERNEL
:
2863 case LTTNG_CONSUMER32_UST
:
2864 case LTTNG_CONSUMER64_UST
:
2865 /* Delete streams that might have been left in the stream list. */
2866 cds_list_for_each_entry_safe(stream
, stmp
, &chan
->streams
.head
,
2868 health_code_update();
2870 cds_list_del(&stream
->send_node
);
2871 lttng_ustconsumer_del_stream(stream
);
2872 uatomic_sub(&stream
->chan
->refcount
, 1);
2873 assert(&chan
->refcount
);
2878 ERR("Unknown consumer_data type");
2883 * Release our own refcount. Force channel deletion even if
2884 * streams were not initialized.
2886 if (!uatomic_sub_return(&chan
->refcount
, 1)) {
2887 consumer_del_channel(chan
);
2892 case CONSUMER_CHANNEL_QUIT
:
2894 * Remove the pipe from the poll set and continue the loop
2895 * since their might be data to consume.
2897 lttng_poll_del(&events
, ctx
->consumer_channel_pipe
[0]);
2900 ERR("Unknown action");
2905 /* Handle other stream */
2911 uint64_t tmp_id
= (uint64_t) pollfd
;
2913 lttng_ht_lookup(channel_ht
, &tmp_id
, &iter
);
2915 node
= lttng_ht_iter_get_node_u64(&iter
);
2918 chan
= caa_container_of(node
, struct lttng_consumer_channel
,
2921 /* Check for error event */
2922 if (revents
& (LPOLLERR
| LPOLLHUP
)) {
2923 DBG("Channel fd %d is hup|err.", pollfd
);
2925 lttng_poll_del(&events
, chan
->wait_fd
);
2926 ret
= lttng_ht_del(channel_ht
, &iter
);
2928 consumer_close_channel_streams(chan
);
2930 /* Release our own refcount */
2931 if (!uatomic_sub_return(&chan
->refcount
, 1)
2932 && !uatomic_read(&chan
->nb_init_stream_left
)) {
2933 consumer_del_channel(chan
);
2937 /* Release RCU lock for the channel looked up */
2945 lttng_poll_clean(&events
);
2947 destroy_channel_ht(channel_ht
);
2949 DBG("Channel poll thread exiting");
2952 ERR("Health error occurred in %s", __func__
);
2954 health_unregister(health_consumerd
);
2955 rcu_unregister_thread();
2959 static int set_metadata_socket(struct lttng_consumer_local_data
*ctx
,
2960 struct pollfd
*sockpoll
, int client_socket
)
2967 if (lttng_consumer_poll_socket(sockpoll
) < 0) {
2971 DBG("Metadata connection on client_socket");
2973 /* Blocking call, waiting for transmission */
2974 ctx
->consumer_metadata_socket
= lttcomm_accept_unix_sock(client_socket
);
2975 if (ctx
->consumer_metadata_socket
< 0) {
2976 WARN("On accept metadata");
2987 * This thread listens on the consumerd socket and receives the file
2988 * descriptors from the session daemon.
2990 void *consumer_thread_sessiond_poll(void *data
)
2992 int sock
= -1, client_socket
, ret
, err
= -1;
2994 * structure to poll for incoming data on communication socket avoids
2995 * making blocking sockets.
2997 struct pollfd consumer_sockpoll
[2];
2998 struct lttng_consumer_local_data
*ctx
= data
;
3000 rcu_register_thread();
3002 health_register(health_consumerd
, HEALTH_CONSUMERD_TYPE_SESSIOND
);
3004 health_code_update();
3006 DBG("Creating command socket %s", ctx
->consumer_command_sock_path
);
3007 unlink(ctx
->consumer_command_sock_path
);
3008 client_socket
= lttcomm_create_unix_sock(ctx
->consumer_command_sock_path
);
3009 if (client_socket
< 0) {
3010 ERR("Cannot create command socket");
3014 ret
= lttcomm_listen_unix_sock(client_socket
);
3019 DBG("Sending ready command to lttng-sessiond");
3020 ret
= lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_COMMAND_SOCK_READY
);
3021 /* return < 0 on error, but == 0 is not fatal */
3023 ERR("Error sending ready command to lttng-sessiond");
3027 /* prepare the FDs to poll : to client socket and the should_quit pipe */
3028 consumer_sockpoll
[0].fd
= ctx
->consumer_should_quit
[0];
3029 consumer_sockpoll
[0].events
= POLLIN
| POLLPRI
;
3030 consumer_sockpoll
[1].fd
= client_socket
;
3031 consumer_sockpoll
[1].events
= POLLIN
| POLLPRI
;
3033 if (lttng_consumer_poll_socket(consumer_sockpoll
) < 0) {
3036 DBG("Connection on client_socket");
3038 /* Blocking call, waiting for transmission */
3039 sock
= lttcomm_accept_unix_sock(client_socket
);
3046 * Setup metadata socket which is the second socket connection on the
3047 * command unix socket.
3049 ret
= set_metadata_socket(ctx
, consumer_sockpoll
, client_socket
);
3054 /* This socket is not useful anymore. */
3055 ret
= close(client_socket
);
3057 PERROR("close client_socket");
3061 /* update the polling structure to poll on the established socket */
3062 consumer_sockpoll
[1].fd
= sock
;
3063 consumer_sockpoll
[1].events
= POLLIN
| POLLPRI
;
3066 health_code_update();
3068 health_poll_entry();
3069 ret
= lttng_consumer_poll_socket(consumer_sockpoll
);
3074 DBG("Incoming command on sock");
3075 ret
= lttng_consumer_recv_cmd(ctx
, sock
, consumer_sockpoll
);
3076 if (ret
== -ENOENT
) {
3077 DBG("Received STOP command");
3082 * This could simply be a session daemon quitting. Don't output
3085 DBG("Communication interrupted on command socket");
3089 if (consumer_quit
) {
3090 DBG("consumer_thread_receive_fds received quit from signal");
3091 err
= 0; /* All is OK */
3094 DBG("received command on sock");
3100 DBG("Consumer thread sessiond poll exiting");
3103 * Close metadata streams since the producer is the session daemon which
3106 * NOTE: for now, this only applies to the UST tracer.
3108 lttng_consumer_close_metadata();
3111 * when all fds have hung up, the polling thread
3117 * Notify the data poll thread to poll back again and test the
3118 * consumer_quit state that we just set so to quit gracefully.
3120 notify_thread_lttng_pipe(ctx
->consumer_data_pipe
);
3122 notify_channel_pipe(ctx
, NULL
, -1, CONSUMER_CHANNEL_QUIT
);
3124 /* Cleaning up possibly open sockets. */
3128 PERROR("close sock sessiond poll");
3131 if (client_socket
>= 0) {
3132 ret
= close(client_socket
);
3134 PERROR("close client_socket sessiond poll");
3140 ERR("Health error occurred in %s", __func__
);
3142 health_unregister(health_consumerd
);
3144 rcu_unregister_thread();
3148 ssize_t
lttng_consumer_read_subbuffer(struct lttng_consumer_stream
*stream
,
3149 struct lttng_consumer_local_data
*ctx
)
3153 pthread_mutex_lock(&stream
->lock
);
3154 if (stream
->metadata_flag
) {
3155 pthread_mutex_lock(&stream
->metadata_rdv_lock
);
3158 switch (consumer_data
.type
) {
3159 case LTTNG_CONSUMER_KERNEL
:
3160 ret
= lttng_kconsumer_read_subbuffer(stream
, ctx
);
3162 case LTTNG_CONSUMER32_UST
:
3163 case LTTNG_CONSUMER64_UST
:
3164 ret
= lttng_ustconsumer_read_subbuffer(stream
, ctx
);
3167 ERR("Unknown consumer_data type");
3173 if (stream
->metadata_flag
) {
3174 pthread_cond_broadcast(&stream
->metadata_rdv
);
3175 pthread_mutex_unlock(&stream
->metadata_rdv_lock
);
3177 pthread_mutex_unlock(&stream
->lock
);
3181 int lttng_consumer_on_recv_stream(struct lttng_consumer_stream
*stream
)
3183 switch (consumer_data
.type
) {
3184 case LTTNG_CONSUMER_KERNEL
:
3185 return lttng_kconsumer_on_recv_stream(stream
);
3186 case LTTNG_CONSUMER32_UST
:
3187 case LTTNG_CONSUMER64_UST
:
3188 return lttng_ustconsumer_on_recv_stream(stream
);
3190 ERR("Unknown consumer_data type");
3197 * Allocate and set consumer data hash tables.
3199 void lttng_consumer_init(void)
3201 consumer_data
.channel_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3202 consumer_data
.relayd_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3203 consumer_data
.stream_list_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3204 consumer_data
.stream_per_chan_id_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
3208 * Process the ADD_RELAYD command receive by a consumer.
3210 * This will create a relayd socket pair and add it to the relayd hash table.
3211 * The caller MUST acquire a RCU read side lock before calling it.
3213 int consumer_add_relayd_socket(uint64_t net_seq_idx
, int sock_type
,
3214 struct lttng_consumer_local_data
*ctx
, int sock
,
3215 struct pollfd
*consumer_sockpoll
,
3216 struct lttcomm_relayd_sock
*relayd_sock
, uint64_t sessiond_id
,
3217 uint64_t relayd_session_id
)
3219 int fd
= -1, ret
= -1, relayd_created
= 0;
3220 enum lttng_error_code ret_code
= LTTNG_OK
;
3221 struct consumer_relayd_sock_pair
*relayd
= NULL
;
3224 assert(relayd_sock
);
3226 DBG("Consumer adding relayd socket (idx: %" PRIu64
")", net_seq_idx
);
3228 /* Get relayd reference if exists. */
3229 relayd
= consumer_find_relayd(net_seq_idx
);
3230 if (relayd
== NULL
) {
3231 assert(sock_type
== LTTNG_STREAM_CONTROL
);
3232 /* Not found. Allocate one. */
3233 relayd
= consumer_allocate_relayd_sock_pair(net_seq_idx
);
3234 if (relayd
== NULL
) {
3236 ret_code
= LTTCOMM_CONSUMERD_ENOMEM
;
3239 relayd
->sessiond_session_id
= sessiond_id
;
3244 * This code path MUST continue to the consumer send status message to
3245 * we can notify the session daemon and continue our work without
3246 * killing everything.
3250 * relayd key should never be found for control socket.
3252 assert(sock_type
!= LTTNG_STREAM_CONTROL
);
3255 /* First send a status message before receiving the fds. */
3256 ret
= consumer_send_status_msg(sock
, LTTNG_OK
);
3258 /* Somehow, the session daemon is not responding anymore. */
3259 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_FATAL
);
3260 goto error_nosignal
;
3263 /* Poll on consumer socket. */
3264 if (lttng_consumer_poll_socket(consumer_sockpoll
) < 0) {
3265 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_POLL_ERROR
);
3267 goto error_nosignal
;
3270 /* Get relayd socket from session daemon */
3271 ret
= lttcomm_recv_fds_unix_sock(sock
, &fd
, 1);
3272 if (ret
!= sizeof(fd
)) {
3274 fd
= -1; /* Just in case it gets set with an invalid value. */
3277 * Failing to receive FDs might indicate a major problem such as
3278 * reaching a fd limit during the receive where the kernel returns a
3279 * MSG_CTRUNC and fails to cleanup the fd in the queue. Any case, we
3280 * don't take any chances and stop everything.
3282 * XXX: Feature request #558 will fix that and avoid this possible
3283 * issue when reaching the fd limit.
3285 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_ERROR_RECV_FD
);
3286 ret_code
= LTTCOMM_CONSUMERD_ERROR_RECV_FD
;
3290 /* Copy socket information and received FD */
3291 switch (sock_type
) {
3292 case LTTNG_STREAM_CONTROL
:
3293 /* Copy received lttcomm socket */
3294 lttcomm_copy_sock(&relayd
->control_sock
.sock
, &relayd_sock
->sock
);
3295 ret
= lttcomm_create_sock(&relayd
->control_sock
.sock
);
3296 /* Handle create_sock error. */
3298 ret_code
= LTTCOMM_CONSUMERD_ENOMEM
;
3302 * Close the socket created internally by
3303 * lttcomm_create_sock, so we can replace it by the one
3304 * received from sessiond.
3306 if (close(relayd
->control_sock
.sock
.fd
)) {
3310 /* Assign new file descriptor */
3311 relayd
->control_sock
.sock
.fd
= fd
;
3312 fd
= -1; /* For error path */
3313 /* Assign version values. */
3314 relayd
->control_sock
.major
= relayd_sock
->major
;
3315 relayd
->control_sock
.minor
= relayd_sock
->minor
;
3317 relayd
->relayd_session_id
= relayd_session_id
;
3320 case LTTNG_STREAM_DATA
:
3321 /* Copy received lttcomm socket */
3322 lttcomm_copy_sock(&relayd
->data_sock
.sock
, &relayd_sock
->sock
);
3323 ret
= lttcomm_create_sock(&relayd
->data_sock
.sock
);
3324 /* Handle create_sock error. */
3326 ret_code
= LTTCOMM_CONSUMERD_ENOMEM
;
3330 * Close the socket created internally by
3331 * lttcomm_create_sock, so we can replace it by the one
3332 * received from sessiond.
3334 if (close(relayd
->data_sock
.sock
.fd
)) {
3338 /* Assign new file descriptor */
3339 relayd
->data_sock
.sock
.fd
= fd
;
3340 fd
= -1; /* for eventual error paths */
3341 /* Assign version values. */
3342 relayd
->data_sock
.major
= relayd_sock
->major
;
3343 relayd
->data_sock
.minor
= relayd_sock
->minor
;
3346 ERR("Unknown relayd socket type (%d)", sock_type
);
3348 ret_code
= LTTCOMM_CONSUMERD_FATAL
;
3352 DBG("Consumer %s socket created successfully with net idx %" PRIu64
" (fd: %d)",
3353 sock_type
== LTTNG_STREAM_CONTROL
? "control" : "data",
3354 relayd
->net_seq_idx
, fd
);
3356 /* We successfully added the socket. Send status back. */
3357 ret
= consumer_send_status_msg(sock
, ret_code
);
3359 /* Somehow, the session daemon is not responding anymore. */
3360 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_FATAL
);
3361 goto error_nosignal
;
3365 * Add relayd socket pair to consumer data hashtable. If object already
3366 * exists or on error, the function gracefully returns.
3374 if (consumer_send_status_msg(sock
, ret_code
) < 0) {
3375 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_FATAL
);
3379 /* Close received socket if valid. */
3382 PERROR("close received socket");
3386 if (relayd_created
) {
3394 * Try to lock the stream mutex.
3396 * On success, 1 is returned else 0 indicating that the mutex is NOT lock.
3398 static int stream_try_lock(struct lttng_consumer_stream
*stream
)
3405 * Try to lock the stream mutex. On failure, we know that the stream is
3406 * being used else where hence there is data still being extracted.
3408 ret
= pthread_mutex_trylock(&stream
->lock
);
3410 /* For both EBUSY and EINVAL error, the mutex is NOT locked. */
3422 * Search for a relayd associated to the session id and return the reference.
3424 * A rcu read side lock MUST be acquire before calling this function and locked
3425 * until the relayd object is no longer necessary.
3427 static struct consumer_relayd_sock_pair
*find_relayd_by_session_id(uint64_t id
)
3429 struct lttng_ht_iter iter
;
3430 struct consumer_relayd_sock_pair
*relayd
= NULL
;
3432 /* Iterate over all relayd since they are indexed by net_seq_idx. */
3433 cds_lfht_for_each_entry(consumer_data
.relayd_ht
->ht
, &iter
.iter
, relayd
,
3436 * Check by sessiond id which is unique here where the relayd session
3437 * id might not be when having multiple relayd.
3439 if (relayd
->sessiond_session_id
== id
) {
3440 /* Found the relayd. There can be only one per id. */
3452 * Check if for a given session id there is still data needed to be extract
3455 * Return 1 if data is pending or else 0 meaning ready to be read.
3457 int consumer_data_pending(uint64_t id
)
3460 struct lttng_ht_iter iter
;
3461 struct lttng_ht
*ht
;
3462 struct lttng_consumer_stream
*stream
;
3463 struct consumer_relayd_sock_pair
*relayd
= NULL
;
3464 int (*data_pending
)(struct lttng_consumer_stream
*);
3466 DBG("Consumer data pending command on session id %" PRIu64
, id
);
3469 pthread_mutex_lock(&consumer_data
.lock
);
3471 switch (consumer_data
.type
) {
3472 case LTTNG_CONSUMER_KERNEL
:
3473 data_pending
= lttng_kconsumer_data_pending
;
3475 case LTTNG_CONSUMER32_UST
:
3476 case LTTNG_CONSUMER64_UST
:
3477 data_pending
= lttng_ustconsumer_data_pending
;
3480 ERR("Unknown consumer data type");
3484 /* Ease our life a bit */
3485 ht
= consumer_data
.stream_list_ht
;
3487 relayd
= find_relayd_by_session_id(id
);
3489 /* Send init command for data pending. */
3490 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
3491 ret
= relayd_begin_data_pending(&relayd
->control_sock
,
3492 relayd
->relayd_session_id
);
3493 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
3495 /* Communication error thus the relayd so no data pending. */
3496 goto data_not_pending
;
3500 cds_lfht_for_each_entry_duplicate(ht
->ht
,
3501 ht
->hash_fct(&id
, lttng_ht_seed
),
3503 &iter
.iter
, stream
, node_session_id
.node
) {
3504 /* If this call fails, the stream is being used hence data pending. */
3505 ret
= stream_try_lock(stream
);
3511 * A removed node from the hash table indicates that the stream has
3512 * been deleted thus having a guarantee that the buffers are closed
3513 * on the consumer side. However, data can still be transmitted
3514 * over the network so don't skip the relayd check.
3516 ret
= cds_lfht_is_node_deleted(&stream
->node
.node
);
3519 * An empty output file is not valid. We need at least one packet
3520 * generated per stream, even if it contains no event, so it
3521 * contains at least one packet header.
3523 if (stream
->output_written
== 0) {
3524 pthread_mutex_unlock(&stream
->lock
);
3527 /* Check the stream if there is data in the buffers. */
3528 ret
= data_pending(stream
);
3530 pthread_mutex_unlock(&stream
->lock
);
3537 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
3538 if (stream
->metadata_flag
) {
3539 ret
= relayd_quiescent_control(&relayd
->control_sock
,
3540 stream
->relayd_stream_id
);
3542 ret
= relayd_data_pending(&relayd
->control_sock
,
3543 stream
->relayd_stream_id
,
3544 stream
->next_net_seq_num
- 1);
3546 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
3548 pthread_mutex_unlock(&stream
->lock
);
3552 pthread_mutex_unlock(&stream
->lock
);
3556 unsigned int is_data_inflight
= 0;
3558 /* Send init command for data pending. */
3559 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
3560 ret
= relayd_end_data_pending(&relayd
->control_sock
,
3561 relayd
->relayd_session_id
, &is_data_inflight
);
3562 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
3564 goto data_not_pending
;
3566 if (is_data_inflight
) {
3572 * Finding _no_ node in the hash table and no inflight data means that the
3573 * stream(s) have been removed thus data is guaranteed to be available for
3574 * analysis from the trace files.
3578 /* Data is available to be read by a viewer. */
3579 pthread_mutex_unlock(&consumer_data
.lock
);
3584 /* Data is still being extracted from buffers. */
3585 pthread_mutex_unlock(&consumer_data
.lock
);
3591 * Send a ret code status message to the sessiond daemon.
3593 * Return the sendmsg() return value.
3595 int consumer_send_status_msg(int sock
, int ret_code
)
3597 struct lttcomm_consumer_status_msg msg
;
3599 msg
.ret_code
= ret_code
;
3601 return lttcomm_send_unix_sock(sock
, &msg
, sizeof(msg
));
3605 * Send a channel status message to the sessiond daemon.
3607 * Return the sendmsg() return value.
3609 int consumer_send_status_channel(int sock
,
3610 struct lttng_consumer_channel
*channel
)
3612 struct lttcomm_consumer_status_channel msg
;
3617 msg
.ret_code
= -LTTNG_ERR_UST_CHAN_FAIL
;
3619 msg
.ret_code
= LTTNG_OK
;
3620 msg
.key
= channel
->key
;
3621 msg
.stream_count
= channel
->streams
.count
;
3624 return lttcomm_send_unix_sock(sock
, &msg
, sizeof(msg
));
3628 * Using a maximum stream size with the produced and consumed position of a
3629 * stream, computes the new consumed position to be as close as possible to the
3630 * maximum possible stream size.
3632 * If maximum stream size is lower than the possible buffer size (produced -
3633 * consumed), the consumed_pos given is returned untouched else the new value
3636 unsigned long consumer_get_consumed_maxsize(unsigned long consumed_pos
,
3637 unsigned long produced_pos
, uint64_t max_stream_size
)
3639 if (max_stream_size
&& max_stream_size
< (produced_pos
- consumed_pos
)) {
3640 /* Offset from the produced position to get the latest buffers. */
3641 return produced_pos
- max_stream_size
;
3644 return consumed_pos
;