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>
32 #include <common/common.h>
33 #include <common/utils.h>
34 #include <common/compat/poll.h>
35 #include <common/kernel-ctl/kernel-ctl.h>
36 #include <common/sessiond-comm/relayd.h>
37 #include <common/sessiond-comm/sessiond-comm.h>
38 #include <common/kernel-consumer/kernel-consumer.h>
39 #include <common/relayd/relayd.h>
40 #include <common/ust-consumer/ust-consumer.h>
44 struct lttng_consumer_global_data consumer_data
= {
47 .type
= LTTNG_CONSUMER_UNKNOWN
,
51 * Flag to inform the polling thread to quit when all fd hung up. Updated by
52 * the consumer_thread_receive_fds when it notices that all fds has hung up.
53 * Also updated by the signal handler (consumer_should_exit()). Read by the
56 volatile int consumer_quit
;
59 * Global hash table containing respectively metadata and data streams. The
60 * stream element in this ht should only be updated by the metadata poll thread
61 * for the metadata and the data poll thread for the data.
63 static struct lttng_ht
*metadata_ht
;
64 static struct lttng_ht
*data_ht
;
67 * Notify a thread pipe to poll back again. This usually means that some global
68 * state has changed so we just send back the thread in a poll wait call.
70 static void notify_thread_pipe(int wpipe
)
75 struct lttng_consumer_stream
*null_stream
= NULL
;
77 ret
= write(wpipe
, &null_stream
, sizeof(null_stream
));
78 } while (ret
< 0 && errno
== EINTR
);
82 * Find a stream. The consumer_data.lock must be locked during this
85 static struct lttng_consumer_stream
*consumer_find_stream(int key
,
88 struct lttng_ht_iter iter
;
89 struct lttng_ht_node_ulong
*node
;
90 struct lttng_consumer_stream
*stream
= NULL
;
94 /* Negative keys are lookup failures */
101 lttng_ht_lookup(ht
, (void *)((unsigned long) key
), &iter
);
102 node
= lttng_ht_iter_get_node_ulong(&iter
);
104 stream
= caa_container_of(node
, struct lttng_consumer_stream
, node
);
112 void consumer_steal_stream_key(int key
, struct lttng_ht
*ht
)
114 struct lttng_consumer_stream
*stream
;
117 stream
= consumer_find_stream(key
, ht
);
121 * We don't want the lookup to match, but we still need
122 * to iterate on this stream when iterating over the hash table. Just
123 * change the node key.
125 stream
->node
.key
= -1;
131 * Return a channel object for the given key.
133 * RCU read side lock MUST be acquired before calling this function and
134 * protects the channel ptr.
136 static struct lttng_consumer_channel
*consumer_find_channel(int key
)
138 struct lttng_ht_iter iter
;
139 struct lttng_ht_node_ulong
*node
;
140 struct lttng_consumer_channel
*channel
= NULL
;
142 /* Negative keys are lookup failures */
147 lttng_ht_lookup(consumer_data
.channel_ht
, (void *)((unsigned long) key
),
149 node
= lttng_ht_iter_get_node_ulong(&iter
);
151 channel
= caa_container_of(node
, struct lttng_consumer_channel
, node
);
157 static void consumer_steal_channel_key(int key
)
159 struct lttng_consumer_channel
*channel
;
162 channel
= consumer_find_channel(key
);
166 * We don't want the lookup to match, but we still need
167 * to iterate on this channel when iterating over the hash table. Just
168 * change the node key.
170 channel
->node
.key
= -1;
176 void consumer_free_stream(struct rcu_head
*head
)
178 struct lttng_ht_node_ulong
*node
=
179 caa_container_of(head
, struct lttng_ht_node_ulong
, head
);
180 struct lttng_consumer_stream
*stream
=
181 caa_container_of(node
, struct lttng_consumer_stream
, node
);
187 * RCU protected relayd socket pair free.
189 static void consumer_rcu_free_relayd(struct rcu_head
*head
)
191 struct lttng_ht_node_ulong
*node
=
192 caa_container_of(head
, struct lttng_ht_node_ulong
, head
);
193 struct consumer_relayd_sock_pair
*relayd
=
194 caa_container_of(node
, struct consumer_relayd_sock_pair
, node
);
197 * Close all sockets. This is done in the call RCU since we don't want the
198 * socket fds to be reassigned thus potentially creating bad state of the
201 * We do not have to lock the control socket mutex here since at this stage
202 * there is no one referencing to this relayd object.
204 (void) relayd_close(&relayd
->control_sock
);
205 (void) relayd_close(&relayd
->data_sock
);
211 * Destroy and free relayd socket pair object.
213 * This function MUST be called with the consumer_data lock acquired.
215 static void destroy_relayd(struct consumer_relayd_sock_pair
*relayd
)
218 struct lttng_ht_iter iter
;
220 if (relayd
== NULL
) {
224 DBG("Consumer destroy and close relayd socket pair");
226 iter
.iter
.node
= &relayd
->node
.node
;
227 ret
= lttng_ht_del(consumer_data
.relayd_ht
, &iter
);
229 /* We assume the relayd is being or is destroyed */
233 /* RCU free() call */
234 call_rcu(&relayd
->node
.head
, consumer_rcu_free_relayd
);
238 * Iterate over the relayd hash table and destroy each element. Finally,
239 * destroy the whole hash table.
241 static void cleanup_relayd_ht(void)
243 struct lttng_ht_iter iter
;
244 struct consumer_relayd_sock_pair
*relayd
;
248 cds_lfht_for_each_entry(consumer_data
.relayd_ht
->ht
, &iter
.iter
, relayd
,
250 destroy_relayd(relayd
);
253 lttng_ht_destroy(consumer_data
.relayd_ht
);
259 * Update the end point status of all streams having the given network sequence
260 * index (relayd index).
262 * It's atomically set without having the stream mutex locked which is fine
263 * because we handle the write/read race with a pipe wakeup for each thread.
265 static void update_endpoint_status_by_netidx(int net_seq_idx
,
266 enum consumer_endpoint_status status
)
268 struct lttng_ht_iter iter
;
269 struct lttng_consumer_stream
*stream
;
271 DBG("Consumer set delete flag on stream by idx %d", net_seq_idx
);
275 /* Let's begin with metadata */
276 cds_lfht_for_each_entry(metadata_ht
->ht
, &iter
.iter
, stream
, node
.node
) {
277 if (stream
->net_seq_idx
== net_seq_idx
) {
278 uatomic_set(&stream
->endpoint_status
, status
);
279 DBG("Delete flag set to metadata stream %d", stream
->wait_fd
);
283 /* Follow up by the data streams */
284 cds_lfht_for_each_entry(data_ht
->ht
, &iter
.iter
, stream
, node
.node
) {
285 if (stream
->net_seq_idx
== net_seq_idx
) {
286 uatomic_set(&stream
->endpoint_status
, status
);
287 DBG("Delete flag set to data stream %d", stream
->wait_fd
);
294 * Cleanup a relayd object by flagging every associated streams for deletion,
295 * destroying the object meaning removing it from the relayd hash table,
296 * closing the sockets and freeing the memory in a RCU call.
298 * If a local data context is available, notify the threads that the streams'
299 * state have changed.
301 static void cleanup_relayd(struct consumer_relayd_sock_pair
*relayd
,
302 struct lttng_consumer_local_data
*ctx
)
308 DBG("Cleaning up relayd sockets");
310 /* Save the net sequence index before destroying the object */
311 netidx
= relayd
->net_seq_idx
;
314 * Delete the relayd from the relayd hash table, close the sockets and free
315 * the object in a RCU call.
317 destroy_relayd(relayd
);
319 /* Set inactive endpoint to all streams */
320 update_endpoint_status_by_netidx(netidx
, CONSUMER_ENDPOINT_INACTIVE
);
323 * With a local data context, notify the threads that the streams' state
324 * have changed. The write() action on the pipe acts as an "implicit"
325 * memory barrier ordering the updates of the end point status from the
326 * read of this status which happens AFTER receiving this notify.
329 notify_thread_pipe(ctx
->consumer_data_pipe
[1]);
330 notify_thread_pipe(ctx
->consumer_metadata_pipe
[1]);
335 * Flag a relayd socket pair for destruction. Destroy it if the refcount
338 * RCU read side lock MUST be aquired before calling this function.
340 void consumer_flag_relayd_for_destroy(struct consumer_relayd_sock_pair
*relayd
)
344 /* Set destroy flag for this object */
345 uatomic_set(&relayd
->destroy_flag
, 1);
347 /* Destroy the relayd if refcount is 0 */
348 if (uatomic_read(&relayd
->refcount
) == 0) {
349 destroy_relayd(relayd
);
354 * Remove a stream from the global list protected by a mutex. This
355 * function is also responsible for freeing its data structures.
357 void consumer_del_stream(struct lttng_consumer_stream
*stream
,
361 struct lttng_ht_iter iter
;
362 struct lttng_consumer_channel
*free_chan
= NULL
;
363 struct consumer_relayd_sock_pair
*relayd
;
367 DBG("Consumer del stream %d", stream
->wait_fd
);
370 /* Means the stream was allocated but not successfully added */
374 pthread_mutex_lock(&consumer_data
.lock
);
375 pthread_mutex_lock(&stream
->lock
);
377 switch (consumer_data
.type
) {
378 case LTTNG_CONSUMER_KERNEL
:
379 if (stream
->mmap_base
!= NULL
) {
380 ret
= munmap(stream
->mmap_base
, stream
->mmap_len
);
386 case LTTNG_CONSUMER32_UST
:
387 case LTTNG_CONSUMER64_UST
:
388 lttng_ustconsumer_del_stream(stream
);
391 ERR("Unknown consumer_data type");
397 iter
.iter
.node
= &stream
->node
.node
;
398 ret
= lttng_ht_del(ht
, &iter
);
401 /* Remove node session id from the consumer_data stream ht */
402 iter
.iter
.node
= &stream
->node_session_id
.node
;
403 ret
= lttng_ht_del(consumer_data
.stream_list_ht
, &iter
);
407 assert(consumer_data
.stream_count
> 0);
408 consumer_data
.stream_count
--;
410 if (stream
->out_fd
>= 0) {
411 ret
= close(stream
->out_fd
);
416 if (stream
->wait_fd
>= 0 && !stream
->wait_fd_is_copy
) {
417 ret
= close(stream
->wait_fd
);
422 if (stream
->shm_fd
>= 0 && stream
->wait_fd
!= stream
->shm_fd
) {
423 ret
= close(stream
->shm_fd
);
429 /* Check and cleanup relayd */
431 relayd
= consumer_find_relayd(stream
->net_seq_idx
);
432 if (relayd
!= NULL
) {
433 uatomic_dec(&relayd
->refcount
);
434 assert(uatomic_read(&relayd
->refcount
) >= 0);
436 /* Closing streams requires to lock the control socket. */
437 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
438 ret
= relayd_send_close_stream(&relayd
->control_sock
,
439 stream
->relayd_stream_id
,
440 stream
->next_net_seq_num
- 1);
441 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
443 DBG("Unable to close stream on the relayd. Continuing");
445 * Continue here. There is nothing we can do for the relayd.
446 * Chances are that the relayd has closed the socket so we just
447 * continue cleaning up.
451 /* Both conditions are met, we destroy the relayd. */
452 if (uatomic_read(&relayd
->refcount
) == 0 &&
453 uatomic_read(&relayd
->destroy_flag
)) {
454 destroy_relayd(relayd
);
459 uatomic_dec(&stream
->chan
->refcount
);
460 if (!uatomic_read(&stream
->chan
->refcount
)
461 && !uatomic_read(&stream
->chan
->nb_init_streams
)) {
462 free_chan
= stream
->chan
;
466 consumer_data
.need_update
= 1;
467 pthread_mutex_unlock(&stream
->lock
);
468 pthread_mutex_unlock(&consumer_data
.lock
);
471 consumer_del_channel(free_chan
);
475 call_rcu(&stream
->node
.head
, consumer_free_stream
);
478 struct lttng_consumer_stream
*consumer_allocate_stream(
479 int channel_key
, int stream_key
,
480 int shm_fd
, int wait_fd
,
481 enum lttng_consumer_stream_state state
,
483 enum lttng_event_output output
,
484 const char *path_name
,
492 struct lttng_consumer_stream
*stream
;
494 stream
= zmalloc(sizeof(*stream
));
495 if (stream
== NULL
) {
496 PERROR("malloc struct lttng_consumer_stream");
497 *alloc_ret
= -ENOMEM
;
504 * Get stream's channel reference. Needed when adding the stream to the
507 stream
->chan
= consumer_find_channel(channel_key
);
509 *alloc_ret
= -ENOENT
;
510 ERR("Unable to find channel for stream %d", stream_key
);
514 stream
->key
= stream_key
;
515 stream
->shm_fd
= shm_fd
;
516 stream
->wait_fd
= wait_fd
;
518 stream
->out_fd_offset
= 0;
519 stream
->state
= state
;
520 stream
->mmap_len
= mmap_len
;
521 stream
->mmap_base
= NULL
;
522 stream
->output
= output
;
525 stream
->net_seq_idx
= net_index
;
526 stream
->metadata_flag
= metadata_flag
;
527 stream
->session_id
= session_id
;
528 strncpy(stream
->path_name
, path_name
, sizeof(stream
->path_name
));
529 stream
->path_name
[sizeof(stream
->path_name
) - 1] = '\0';
530 pthread_mutex_init(&stream
->lock
, NULL
);
533 * Index differently the metadata node because the thread is using an
534 * internal hash table to match streams in the metadata_ht to the epoll set
538 lttng_ht_node_init_ulong(&stream
->node
, stream
->wait_fd
);
540 lttng_ht_node_init_ulong(&stream
->node
, stream
->key
);
543 /* Init session id node with the stream session id */
544 lttng_ht_node_init_ulong(&stream
->node_session_id
, stream
->session_id
);
547 * The cpu number is needed before using any ustctl_* actions. Ignored for
548 * the kernel so the value does not matter.
550 pthread_mutex_lock(&consumer_data
.lock
);
551 stream
->cpu
= stream
->chan
->cpucount
++;
552 pthread_mutex_unlock(&consumer_data
.lock
);
554 DBG3("Allocated stream %s (key %d, shm_fd %d, wait_fd %d, mmap_len %llu,"
555 " out_fd %d, net_seq_idx %d, session_id %" PRIu64
,
556 stream
->path_name
, stream
->key
, stream
->shm_fd
, stream
->wait_fd
,
557 (unsigned long long) stream
->mmap_len
, stream
->out_fd
,
558 stream
->net_seq_idx
, stream
->session_id
);
571 * Add a stream to the global list protected by a mutex.
573 static int consumer_add_stream(struct lttng_consumer_stream
*stream
,
577 struct consumer_relayd_sock_pair
*relayd
;
582 DBG3("Adding consumer stream %d", stream
->key
);
584 pthread_mutex_lock(&consumer_data
.lock
);
585 pthread_mutex_lock(&stream
->lock
);
588 /* Steal stream identifier to avoid having streams with the same key */
589 consumer_steal_stream_key(stream
->key
, ht
);
591 lttng_ht_add_unique_ulong(ht
, &stream
->node
);
594 * Add stream to the stream_list_ht of the consumer data. No need to steal
595 * the key since the HT does not use it and we allow to add redundant keys
598 lttng_ht_add_ulong(consumer_data
.stream_list_ht
, &stream
->node_session_id
);
600 /* Check and cleanup relayd */
601 relayd
= consumer_find_relayd(stream
->net_seq_idx
);
602 if (relayd
!= NULL
) {
603 uatomic_inc(&relayd
->refcount
);
606 /* Update channel refcount once added without error(s). */
607 uatomic_inc(&stream
->chan
->refcount
);
610 * When nb_init_streams reaches 0, we don't need to trigger any action in
611 * terms of destroying the associated channel, because the action that
612 * causes the count to become 0 also causes a stream to be added. The
613 * channel deletion will thus be triggered by the following removal of this
616 if (uatomic_read(&stream
->chan
->nb_init_streams
) > 0) {
617 uatomic_dec(&stream
->chan
->nb_init_streams
);
620 /* Update consumer data once the node is inserted. */
621 consumer_data
.stream_count
++;
622 consumer_data
.need_update
= 1;
625 pthread_mutex_unlock(&stream
->lock
);
626 pthread_mutex_unlock(&consumer_data
.lock
);
632 * Add relayd socket to global consumer data hashtable. RCU read side lock MUST
633 * be acquired before calling this.
635 static int add_relayd(struct consumer_relayd_sock_pair
*relayd
)
638 struct lttng_ht_node_ulong
*node
;
639 struct lttng_ht_iter iter
;
641 if (relayd
== NULL
) {
646 lttng_ht_lookup(consumer_data
.relayd_ht
,
647 (void *)((unsigned long) relayd
->net_seq_idx
), &iter
);
648 node
= lttng_ht_iter_get_node_ulong(&iter
);
650 /* Relayd already exist. Ignore the insertion */
653 lttng_ht_add_unique_ulong(consumer_data
.relayd_ht
, &relayd
->node
);
660 * Allocate and return a consumer relayd socket.
662 struct consumer_relayd_sock_pair
*consumer_allocate_relayd_sock_pair(
665 struct consumer_relayd_sock_pair
*obj
= NULL
;
667 /* Negative net sequence index is a failure */
668 if (net_seq_idx
< 0) {
672 obj
= zmalloc(sizeof(struct consumer_relayd_sock_pair
));
674 PERROR("zmalloc relayd sock");
678 obj
->net_seq_idx
= net_seq_idx
;
680 obj
->destroy_flag
= 0;
681 lttng_ht_node_init_ulong(&obj
->node
, obj
->net_seq_idx
);
682 pthread_mutex_init(&obj
->ctrl_sock_mutex
, NULL
);
689 * Find a relayd socket pair in the global consumer data.
691 * Return the object if found else NULL.
692 * RCU read-side lock must be held across this call and while using the
695 struct consumer_relayd_sock_pair
*consumer_find_relayd(int key
)
697 struct lttng_ht_iter iter
;
698 struct lttng_ht_node_ulong
*node
;
699 struct consumer_relayd_sock_pair
*relayd
= NULL
;
701 /* Negative keys are lookup failures */
706 lttng_ht_lookup(consumer_data
.relayd_ht
, (void *)((unsigned long) key
),
708 node
= lttng_ht_iter_get_node_ulong(&iter
);
710 relayd
= caa_container_of(node
, struct consumer_relayd_sock_pair
, node
);
718 * Handle stream for relayd transmission if the stream applies for network
719 * streaming where the net sequence index is set.
721 * Return destination file descriptor or negative value on error.
723 static int write_relayd_stream_header(struct lttng_consumer_stream
*stream
,
724 size_t data_size
, unsigned long padding
,
725 struct consumer_relayd_sock_pair
*relayd
)
728 struct lttcomm_relayd_data_hdr data_hdr
;
734 /* Reset data header */
735 memset(&data_hdr
, 0, sizeof(data_hdr
));
737 if (stream
->metadata_flag
) {
738 /* Caller MUST acquire the relayd control socket lock */
739 ret
= relayd_send_metadata(&relayd
->control_sock
, data_size
);
744 /* Metadata are always sent on the control socket. */
745 outfd
= relayd
->control_sock
.fd
;
747 /* Set header with stream information */
748 data_hdr
.stream_id
= htobe64(stream
->relayd_stream_id
);
749 data_hdr
.data_size
= htobe32(data_size
);
750 data_hdr
.padding_size
= htobe32(padding
);
752 * Note that net_seq_num below is assigned with the *current* value of
753 * next_net_seq_num and only after that the next_net_seq_num will be
754 * increment. This is why when issuing a command on the relayd using
755 * this next value, 1 should always be substracted in order to compare
756 * the last seen sequence number on the relayd side to the last sent.
758 data_hdr
.net_seq_num
= htobe64(stream
->next_net_seq_num
);
759 /* Other fields are zeroed previously */
761 ret
= relayd_send_data_hdr(&relayd
->data_sock
, &data_hdr
,
767 ++stream
->next_net_seq_num
;
769 /* Set to go on data socket */
770 outfd
= relayd
->data_sock
.fd
;
778 void consumer_free_channel(struct rcu_head
*head
)
780 struct lttng_ht_node_ulong
*node
=
781 caa_container_of(head
, struct lttng_ht_node_ulong
, head
);
782 struct lttng_consumer_channel
*channel
=
783 caa_container_of(node
, struct lttng_consumer_channel
, node
);
789 * Remove a channel from the global list protected by a mutex. This
790 * function is also responsible for freeing its data structures.
792 void consumer_del_channel(struct lttng_consumer_channel
*channel
)
795 struct lttng_ht_iter iter
;
797 DBG("Consumer delete channel key %d", channel
->key
);
799 pthread_mutex_lock(&consumer_data
.lock
);
801 switch (consumer_data
.type
) {
802 case LTTNG_CONSUMER_KERNEL
:
804 case LTTNG_CONSUMER32_UST
:
805 case LTTNG_CONSUMER64_UST
:
806 lttng_ustconsumer_del_channel(channel
);
809 ERR("Unknown consumer_data type");
815 iter
.iter
.node
= &channel
->node
.node
;
816 ret
= lttng_ht_del(consumer_data
.channel_ht
, &iter
);
820 if (channel
->mmap_base
!= NULL
) {
821 ret
= munmap(channel
->mmap_base
, channel
->mmap_len
);
826 if (channel
->wait_fd
>= 0 && !channel
->wait_fd_is_copy
) {
827 ret
= close(channel
->wait_fd
);
832 if (channel
->shm_fd
>= 0 && channel
->wait_fd
!= channel
->shm_fd
) {
833 ret
= close(channel
->shm_fd
);
839 call_rcu(&channel
->node
.head
, consumer_free_channel
);
841 pthread_mutex_unlock(&consumer_data
.lock
);
844 struct lttng_consumer_channel
*consumer_allocate_channel(
846 int shm_fd
, int wait_fd
,
848 uint64_t max_sb_size
,
849 unsigned int nb_init_streams
)
851 struct lttng_consumer_channel
*channel
;
854 channel
= zmalloc(sizeof(*channel
));
855 if (channel
== NULL
) {
856 PERROR("malloc struct lttng_consumer_channel");
859 channel
->key
= channel_key
;
860 channel
->shm_fd
= shm_fd
;
861 channel
->wait_fd
= wait_fd
;
862 channel
->mmap_len
= mmap_len
;
863 channel
->max_sb_size
= max_sb_size
;
864 channel
->refcount
= 0;
865 channel
->nb_init_streams
= nb_init_streams
;
866 lttng_ht_node_init_ulong(&channel
->node
, channel
->key
);
868 switch (consumer_data
.type
) {
869 case LTTNG_CONSUMER_KERNEL
:
870 channel
->mmap_base
= NULL
;
871 channel
->mmap_len
= 0;
873 case LTTNG_CONSUMER32_UST
:
874 case LTTNG_CONSUMER64_UST
:
875 ret
= lttng_ustconsumer_allocate_channel(channel
);
882 ERR("Unknown consumer_data type");
886 DBG("Allocated channel (key %d, shm_fd %d, wait_fd %d, mmap_len %llu, max_sb_size %llu)",
887 channel
->key
, channel
->shm_fd
, channel
->wait_fd
,
888 (unsigned long long) channel
->mmap_len
,
889 (unsigned long long) channel
->max_sb_size
);
895 * Add a channel to the global list protected by a mutex.
897 int consumer_add_channel(struct lttng_consumer_channel
*channel
)
899 struct lttng_ht_node_ulong
*node
;
900 struct lttng_ht_iter iter
;
902 pthread_mutex_lock(&consumer_data
.lock
);
903 /* Steal channel identifier, for UST */
904 consumer_steal_channel_key(channel
->key
);
907 lttng_ht_lookup(consumer_data
.channel_ht
,
908 (void *)((unsigned long) channel
->key
), &iter
);
909 node
= lttng_ht_iter_get_node_ulong(&iter
);
911 /* Channel already exist. Ignore the insertion */
915 lttng_ht_add_unique_ulong(consumer_data
.channel_ht
, &channel
->node
);
919 pthread_mutex_unlock(&consumer_data
.lock
);
925 * Allocate the pollfd structure and the local view of the out fds to avoid
926 * doing a lookup in the linked list and concurrency issues when writing is
927 * needed. Called with consumer_data.lock held.
929 * Returns the number of fds in the structures.
931 static int consumer_update_poll_array(
932 struct lttng_consumer_local_data
*ctx
, struct pollfd
**pollfd
,
933 struct lttng_consumer_stream
**local_stream
, struct lttng_ht
*ht
)
936 struct lttng_ht_iter iter
;
937 struct lttng_consumer_stream
*stream
;
939 DBG("Updating poll fd array");
941 cds_lfht_for_each_entry(ht
->ht
, &iter
.iter
, stream
, node
.node
) {
943 * Only active streams with an active end point can be added to the
944 * poll set and local stream storage of the thread.
946 * There is a potential race here for endpoint_status to be updated
947 * just after the check. However, this is OK since the stream(s) will
948 * be deleted once the thread is notified that the end point state has
949 * changed where this function will be called back again.
951 if (stream
->state
!= LTTNG_CONSUMER_ACTIVE_STREAM
||
952 stream
->endpoint_status
== CONSUMER_ENDPOINT_INACTIVE
) {
955 DBG("Active FD %d", stream
->wait_fd
);
956 (*pollfd
)[i
].fd
= stream
->wait_fd
;
957 (*pollfd
)[i
].events
= POLLIN
| POLLPRI
;
958 local_stream
[i
] = stream
;
964 * Insert the consumer_data_pipe at the end of the array and don't
965 * increment i so nb_fd is the number of real FD.
967 (*pollfd
)[i
].fd
= ctx
->consumer_data_pipe
[0];
968 (*pollfd
)[i
].events
= POLLIN
| POLLPRI
;
973 * Poll on the should_quit pipe and the command socket return -1 on error and
974 * should exit, 0 if data is available on the command socket
976 int lttng_consumer_poll_socket(struct pollfd
*consumer_sockpoll
)
981 num_rdy
= poll(consumer_sockpoll
, 2, -1);
984 * Restart interrupted system call.
986 if (errno
== EINTR
) {
989 PERROR("Poll error");
992 if (consumer_sockpoll
[0].revents
& (POLLIN
| POLLPRI
)) {
993 DBG("consumer_should_quit wake up");
1003 * Set the error socket.
1005 void lttng_consumer_set_error_sock(
1006 struct lttng_consumer_local_data
*ctx
, int sock
)
1008 ctx
->consumer_error_socket
= sock
;
1012 * Set the command socket path.
1014 void lttng_consumer_set_command_sock_path(
1015 struct lttng_consumer_local_data
*ctx
, char *sock
)
1017 ctx
->consumer_command_sock_path
= sock
;
1021 * Send return code to the session daemon.
1022 * If the socket is not defined, we return 0, it is not a fatal error
1024 int lttng_consumer_send_error(
1025 struct lttng_consumer_local_data
*ctx
, int cmd
)
1027 if (ctx
->consumer_error_socket
> 0) {
1028 return lttcomm_send_unix_sock(ctx
->consumer_error_socket
, &cmd
,
1029 sizeof(enum lttcomm_sessiond_command
));
1036 * Close all the tracefiles and stream fds and MUST be called when all
1037 * instances are destroyed i.e. when all threads were joined and are ended.
1039 void lttng_consumer_cleanup(void)
1041 struct lttng_ht_iter iter
;
1042 struct lttng_ht_node_ulong
*node
;
1046 cds_lfht_for_each_entry(consumer_data
.channel_ht
->ht
, &iter
.iter
, node
,
1048 struct lttng_consumer_channel
*channel
=
1049 caa_container_of(node
, struct lttng_consumer_channel
, node
);
1050 consumer_del_channel(channel
);
1055 lttng_ht_destroy(consumer_data
.channel_ht
);
1057 cleanup_relayd_ht();
1060 * This HT contains streams that are freed by either the metadata thread or
1061 * the data thread so we do *nothing* on the hash table and simply destroy
1064 lttng_ht_destroy(consumer_data
.stream_list_ht
);
1068 * Called from signal handler.
1070 void lttng_consumer_should_exit(struct lttng_consumer_local_data
*ctx
)
1075 ret
= write(ctx
->consumer_should_quit
[1], "4", 1);
1076 } while (ret
< 0 && errno
== EINTR
);
1077 if (ret
< 0 || ret
!= 1) {
1078 PERROR("write consumer quit");
1081 DBG("Consumer flag that it should quit");
1084 void lttng_consumer_sync_trace_file(struct lttng_consumer_stream
*stream
,
1087 int outfd
= stream
->out_fd
;
1090 * This does a blocking write-and-wait on any page that belongs to the
1091 * subbuffer prior to the one we just wrote.
1092 * Don't care about error values, as these are just hints and ways to
1093 * limit the amount of page cache used.
1095 if (orig_offset
< stream
->chan
->max_sb_size
) {
1098 lttng_sync_file_range(outfd
, orig_offset
- stream
->chan
->max_sb_size
,
1099 stream
->chan
->max_sb_size
,
1100 SYNC_FILE_RANGE_WAIT_BEFORE
1101 | SYNC_FILE_RANGE_WRITE
1102 | SYNC_FILE_RANGE_WAIT_AFTER
);
1104 * Give hints to the kernel about how we access the file:
1105 * POSIX_FADV_DONTNEED : we won't re-access data in a near future after
1108 * We need to call fadvise again after the file grows because the
1109 * kernel does not seem to apply fadvise to non-existing parts of the
1112 * Call fadvise _after_ having waited for the page writeback to
1113 * complete because the dirty page writeback semantic is not well
1114 * defined. So it can be expected to lead to lower throughput in
1117 posix_fadvise(outfd
, orig_offset
- stream
->chan
->max_sb_size
,
1118 stream
->chan
->max_sb_size
, POSIX_FADV_DONTNEED
);
1122 * Initialise the necessary environnement :
1123 * - create a new context
1124 * - create the poll_pipe
1125 * - create the should_quit pipe (for signal handler)
1126 * - create the thread pipe (for splice)
1128 * Takes a function pointer as argument, this function is called when data is
1129 * available on a buffer. This function is responsible to do the
1130 * kernctl_get_next_subbuf, read the data with mmap or splice depending on the
1131 * buffer configuration and then kernctl_put_next_subbuf at the end.
1133 * Returns a pointer to the new context or NULL on error.
1135 struct lttng_consumer_local_data
*lttng_consumer_create(
1136 enum lttng_consumer_type type
,
1137 ssize_t (*buffer_ready
)(struct lttng_consumer_stream
*stream
,
1138 struct lttng_consumer_local_data
*ctx
),
1139 int (*recv_channel
)(struct lttng_consumer_channel
*channel
),
1140 int (*recv_stream
)(struct lttng_consumer_stream
*stream
),
1141 int (*update_stream
)(int stream_key
, uint32_t state
))
1144 struct lttng_consumer_local_data
*ctx
;
1146 assert(consumer_data
.type
== LTTNG_CONSUMER_UNKNOWN
||
1147 consumer_data
.type
== type
);
1148 consumer_data
.type
= type
;
1150 ctx
= zmalloc(sizeof(struct lttng_consumer_local_data
));
1152 PERROR("allocating context");
1156 ctx
->consumer_error_socket
= -1;
1157 /* assign the callbacks */
1158 ctx
->on_buffer_ready
= buffer_ready
;
1159 ctx
->on_recv_channel
= recv_channel
;
1160 ctx
->on_recv_stream
= recv_stream
;
1161 ctx
->on_update_stream
= update_stream
;
1163 ret
= pipe(ctx
->consumer_data_pipe
);
1165 PERROR("Error creating poll pipe");
1166 goto error_poll_pipe
;
1169 /* set read end of the pipe to non-blocking */
1170 ret
= fcntl(ctx
->consumer_data_pipe
[0], F_SETFL
, O_NONBLOCK
);
1172 PERROR("fcntl O_NONBLOCK");
1173 goto error_poll_fcntl
;
1176 /* set write end of the pipe to non-blocking */
1177 ret
= fcntl(ctx
->consumer_data_pipe
[1], F_SETFL
, O_NONBLOCK
);
1179 PERROR("fcntl O_NONBLOCK");
1180 goto error_poll_fcntl
;
1183 ret
= pipe(ctx
->consumer_should_quit
);
1185 PERROR("Error creating recv pipe");
1186 goto error_quit_pipe
;
1189 ret
= pipe(ctx
->consumer_thread_pipe
);
1191 PERROR("Error creating thread pipe");
1192 goto error_thread_pipe
;
1195 ret
= utils_create_pipe(ctx
->consumer_metadata_pipe
);
1197 goto error_metadata_pipe
;
1200 ret
= utils_create_pipe(ctx
->consumer_splice_metadata_pipe
);
1202 goto error_splice_pipe
;
1208 utils_close_pipe(ctx
->consumer_metadata_pipe
);
1209 error_metadata_pipe
:
1210 utils_close_pipe(ctx
->consumer_thread_pipe
);
1212 for (i
= 0; i
< 2; i
++) {
1215 err
= close(ctx
->consumer_should_quit
[i
]);
1222 for (i
= 0; i
< 2; i
++) {
1225 err
= close(ctx
->consumer_data_pipe
[i
]);
1237 * Close all fds associated with the instance and free the context.
1239 void lttng_consumer_destroy(struct lttng_consumer_local_data
*ctx
)
1243 DBG("Consumer destroying it. Closing everything.");
1245 ret
= close(ctx
->consumer_error_socket
);
1249 ret
= close(ctx
->consumer_thread_pipe
[0]);
1253 ret
= close(ctx
->consumer_thread_pipe
[1]);
1257 ret
= close(ctx
->consumer_data_pipe
[0]);
1261 ret
= close(ctx
->consumer_data_pipe
[1]);
1265 ret
= close(ctx
->consumer_should_quit
[0]);
1269 ret
= close(ctx
->consumer_should_quit
[1]);
1273 utils_close_pipe(ctx
->consumer_splice_metadata_pipe
);
1275 unlink(ctx
->consumer_command_sock_path
);
1280 * Write the metadata stream id on the specified file descriptor.
1282 static int write_relayd_metadata_id(int fd
,
1283 struct lttng_consumer_stream
*stream
,
1284 struct consumer_relayd_sock_pair
*relayd
,
1285 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
)
1336 unsigned long mmap_offset
;
1337 ssize_t ret
= 0, written
= 0;
1338 off_t orig_offset
= stream
->out_fd_offset
;
1339 /* Default is on the disk */
1340 int outfd
= stream
->out_fd
;
1341 struct consumer_relayd_sock_pair
*relayd
= NULL
;
1342 unsigned int relayd_hang_up
= 0;
1344 /* RCU lock for the relayd pointer */
1347 /* Flag that the current stream if set for network streaming. */
1348 if (stream
->net_seq_idx
!= -1) {
1349 relayd
= consumer_find_relayd(stream
->net_seq_idx
);
1350 if (relayd
== NULL
) {
1355 /* get the offset inside the fd to mmap */
1356 switch (consumer_data
.type
) {
1357 case LTTNG_CONSUMER_KERNEL
:
1358 ret
= kernctl_get_mmap_read_offset(stream
->wait_fd
, &mmap_offset
);
1360 case LTTNG_CONSUMER32_UST
:
1361 case LTTNG_CONSUMER64_UST
:
1362 ret
= lttng_ustctl_get_mmap_read_offset(stream
->chan
->handle
,
1363 stream
->buf
, &mmap_offset
);
1366 ERR("Unknown consumer_data type");
1371 PERROR("tracer ctl get_mmap_read_offset");
1376 /* Handle stream on the relayd if the output is on the network */
1378 unsigned long netlen
= len
;
1381 * Lock the control socket for the complete duration of the function
1382 * since from this point on we will use the socket.
1384 if (stream
->metadata_flag
) {
1385 /* Metadata requires the control socket. */
1386 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
1387 netlen
+= sizeof(struct lttcomm_relayd_metadata_payload
);
1390 ret
= write_relayd_stream_header(stream
, netlen
, padding
, relayd
);
1392 /* Use the returned socket. */
1395 /* Write metadata stream id before payload */
1396 if (stream
->metadata_flag
) {
1397 ret
= write_relayd_metadata_id(outfd
, stream
, relayd
, padding
);
1400 /* Socket operation failed. We consider the relayd dead */
1401 if (ret
== -EPIPE
|| ret
== -EINVAL
) {
1409 /* Socket operation failed. We consider the relayd dead */
1410 if (ret
== -EPIPE
|| ret
== -EINVAL
) {
1414 /* Else, use the default set before which is the filesystem. */
1417 /* No streaming, we have to set the len with the full padding */
1423 ret
= write(outfd
, stream
->mmap_base
+ mmap_offset
, len
);
1424 } while (ret
< 0 && errno
== EINTR
);
1425 DBG("Consumer mmap write() ret %zd (len %lu)", ret
, len
);
1428 * This is possible if the fd is closed on the other side (outfd)
1429 * or any write problem. It can be verbose a bit for a normal
1430 * execution if for instance the relayd is stopped abruptly. This
1431 * can happen so set this to a DBG statement.
1433 DBG("Error in file write mmap");
1437 /* Socket operation failed. We consider the relayd dead */
1438 if (errno
== EPIPE
|| errno
== EINVAL
) {
1443 } else if (ret
> len
) {
1444 PERROR("Error in file write (ret %zd > len %lu)", ret
, len
);
1452 /* This call is useless on a socket so better save a syscall. */
1454 /* This won't block, but will start writeout asynchronously */
1455 lttng_sync_file_range(outfd
, stream
->out_fd_offset
, ret
,
1456 SYNC_FILE_RANGE_WRITE
);
1457 stream
->out_fd_offset
+= ret
;
1461 lttng_consumer_sync_trace_file(stream
, orig_offset
);
1465 * This is a special case that the relayd has closed its socket. Let's
1466 * cleanup the relayd object and all associated streams.
1468 if (relayd
&& relayd_hang_up
) {
1469 cleanup_relayd(relayd
, ctx
);
1473 /* Unlock only if ctrl socket used */
1474 if (relayd
&& stream
->metadata_flag
) {
1475 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
1483 * Splice the data from the ring buffer to the tracefile.
1485 * It must be called with the stream lock held.
1487 * Returns the number of bytes spliced.
1489 ssize_t
lttng_consumer_on_read_subbuffer_splice(
1490 struct lttng_consumer_local_data
*ctx
,
1491 struct lttng_consumer_stream
*stream
, unsigned long len
,
1492 unsigned long padding
)
1494 ssize_t ret
= 0, written
= 0, ret_splice
= 0;
1496 off_t orig_offset
= stream
->out_fd_offset
;
1497 int fd
= stream
->wait_fd
;
1498 /* Default is on the disk */
1499 int outfd
= stream
->out_fd
;
1500 struct consumer_relayd_sock_pair
*relayd
= NULL
;
1502 unsigned int relayd_hang_up
= 0;
1504 switch (consumer_data
.type
) {
1505 case LTTNG_CONSUMER_KERNEL
:
1507 case LTTNG_CONSUMER32_UST
:
1508 case LTTNG_CONSUMER64_UST
:
1509 /* Not supported for user space tracing */
1512 ERR("Unknown consumer_data type");
1516 /* RCU lock for the relayd pointer */
1519 /* Flag that the current stream if set for network streaming. */
1520 if (stream
->net_seq_idx
!= -1) {
1521 relayd
= consumer_find_relayd(stream
->net_seq_idx
);
1522 if (relayd
== NULL
) {
1528 * Choose right pipe for splice. Metadata and trace data are handled by
1529 * different threads hence the use of two pipes in order not to race or
1530 * corrupt the written data.
1532 if (stream
->metadata_flag
) {
1533 splice_pipe
= ctx
->consumer_splice_metadata_pipe
;
1535 splice_pipe
= ctx
->consumer_thread_pipe
;
1538 /* Write metadata stream id before payload */
1540 int total_len
= len
;
1542 if (stream
->metadata_flag
) {
1544 * Lock the control socket for the complete duration of the function
1545 * since from this point on we will use the socket.
1547 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
1549 ret
= write_relayd_metadata_id(splice_pipe
[1], stream
, relayd
,
1553 /* Socket operation failed. We consider the relayd dead */
1554 if (ret
== -EBADF
) {
1555 WARN("Remote relayd disconnected. Stopping");
1562 total_len
+= sizeof(struct lttcomm_relayd_metadata_payload
);
1565 ret
= write_relayd_stream_header(stream
, total_len
, padding
, relayd
);
1567 /* Use the returned socket. */
1570 /* Socket operation failed. We consider the relayd dead */
1571 if (ret
== -EBADF
) {
1572 WARN("Remote relayd disconnected. Stopping");
1579 /* No streaming, we have to set the len with the full padding */
1584 DBG("splice chan to pipe offset %lu of len %lu (fd : %d, pipe: %d)",
1585 (unsigned long)offset
, len
, fd
, splice_pipe
[1]);
1586 ret_splice
= splice(fd
, &offset
, splice_pipe
[1], NULL
, len
,
1587 SPLICE_F_MOVE
| SPLICE_F_MORE
);
1588 DBG("splice chan to pipe, ret %zd", ret_splice
);
1589 if (ret_splice
< 0) {
1590 PERROR("Error in relay splice");
1592 written
= ret_splice
;
1598 /* Handle stream on the relayd if the output is on the network */
1600 if (stream
->metadata_flag
) {
1601 size_t metadata_payload_size
=
1602 sizeof(struct lttcomm_relayd_metadata_payload
);
1604 /* Update counter to fit the spliced data */
1605 ret_splice
+= metadata_payload_size
;
1606 len
+= metadata_payload_size
;
1608 * We do this so the return value can match the len passed as
1609 * argument to this function.
1611 written
-= metadata_payload_size
;
1615 /* Splice data out */
1616 ret_splice
= splice(splice_pipe
[0], NULL
, outfd
, NULL
,
1617 ret_splice
, SPLICE_F_MOVE
| SPLICE_F_MORE
);
1618 DBG("Consumer splice pipe to file, ret %zd", ret_splice
);
1619 if (ret_splice
< 0) {
1620 PERROR("Error in file splice");
1622 written
= ret_splice
;
1624 /* Socket operation failed. We consider the relayd dead */
1625 if (errno
== EBADF
|| errno
== EPIPE
) {
1626 WARN("Remote relayd disconnected. Stopping");
1632 } else if (ret_splice
> len
) {
1634 PERROR("Wrote more data than requested %zd (len: %lu)",
1636 written
+= ret_splice
;
1642 /* This call is useless on a socket so better save a syscall. */
1644 /* This won't block, but will start writeout asynchronously */
1645 lttng_sync_file_range(outfd
, stream
->out_fd_offset
, ret_splice
,
1646 SYNC_FILE_RANGE_WRITE
);
1647 stream
->out_fd_offset
+= ret_splice
;
1649 written
+= ret_splice
;
1651 lttng_consumer_sync_trace_file(stream
, orig_offset
);
1659 * This is a special case that the relayd has closed its socket. Let's
1660 * cleanup the relayd object and all associated streams.
1662 if (relayd
&& relayd_hang_up
) {
1663 cleanup_relayd(relayd
, ctx
);
1664 /* Skip splice error so the consumer does not fail */
1669 /* send the appropriate error description to sessiond */
1672 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_SPLICE_EINVAL
);
1675 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_SPLICE_ENOMEM
);
1678 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_SPLICE_ESPIPE
);
1683 if (relayd
&& stream
->metadata_flag
) {
1684 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
1692 * Take a snapshot for a specific fd
1694 * Returns 0 on success, < 0 on error
1696 int lttng_consumer_take_snapshot(struct lttng_consumer_local_data
*ctx
,
1697 struct lttng_consumer_stream
*stream
)
1699 switch (consumer_data
.type
) {
1700 case LTTNG_CONSUMER_KERNEL
:
1701 return lttng_kconsumer_take_snapshot(ctx
, stream
);
1702 case LTTNG_CONSUMER32_UST
:
1703 case LTTNG_CONSUMER64_UST
:
1704 return lttng_ustconsumer_take_snapshot(ctx
, stream
);
1706 ERR("Unknown consumer_data type");
1714 * Get the produced position
1716 * Returns 0 on success, < 0 on error
1718 int lttng_consumer_get_produced_snapshot(
1719 struct lttng_consumer_local_data
*ctx
,
1720 struct lttng_consumer_stream
*stream
,
1723 switch (consumer_data
.type
) {
1724 case LTTNG_CONSUMER_KERNEL
:
1725 return lttng_kconsumer_get_produced_snapshot(ctx
, stream
, pos
);
1726 case LTTNG_CONSUMER32_UST
:
1727 case LTTNG_CONSUMER64_UST
:
1728 return lttng_ustconsumer_get_produced_snapshot(ctx
, stream
, pos
);
1730 ERR("Unknown consumer_data type");
1736 int lttng_consumer_recv_cmd(struct lttng_consumer_local_data
*ctx
,
1737 int sock
, struct pollfd
*consumer_sockpoll
)
1739 switch (consumer_data
.type
) {
1740 case LTTNG_CONSUMER_KERNEL
:
1741 return lttng_kconsumer_recv_cmd(ctx
, sock
, consumer_sockpoll
);
1742 case LTTNG_CONSUMER32_UST
:
1743 case LTTNG_CONSUMER64_UST
:
1744 return lttng_ustconsumer_recv_cmd(ctx
, sock
, consumer_sockpoll
);
1746 ERR("Unknown consumer_data type");
1753 * Iterate over all streams of the hashtable and free them properly.
1755 * WARNING: *MUST* be used with data stream only.
1757 static void destroy_data_stream_ht(struct lttng_ht
*ht
)
1759 struct lttng_ht_iter iter
;
1760 struct lttng_consumer_stream
*stream
;
1767 cds_lfht_for_each_entry(ht
->ht
, &iter
.iter
, stream
, node
.node
) {
1769 * Ignore return value since we are currently cleaning up so any error
1772 (void) consumer_del_stream(stream
, ht
);
1776 lttng_ht_destroy(ht
);
1780 * Iterate over all streams of the hashtable and free them properly.
1782 * XXX: Should not be only for metadata stream or else use an other name.
1784 static void destroy_stream_ht(struct lttng_ht
*ht
)
1786 struct lttng_ht_iter iter
;
1787 struct lttng_consumer_stream
*stream
;
1794 cds_lfht_for_each_entry(ht
->ht
, &iter
.iter
, stream
, node
.node
) {
1796 * Ignore return value since we are currently cleaning up so any error
1799 (void) consumer_del_metadata_stream(stream
, ht
);
1803 lttng_ht_destroy(ht
);
1807 * Clean up a metadata stream and free its memory.
1809 void consumer_del_metadata_stream(struct lttng_consumer_stream
*stream
,
1810 struct lttng_ht
*ht
)
1813 struct lttng_ht_iter iter
;
1814 struct lttng_consumer_channel
*free_chan
= NULL
;
1815 struct consumer_relayd_sock_pair
*relayd
;
1819 * This call should NEVER receive regular stream. It must always be
1820 * metadata stream and this is crucial for data structure synchronization.
1822 assert(stream
->metadata_flag
);
1824 DBG3("Consumer delete metadata stream %d", stream
->wait_fd
);
1827 /* Means the stream was allocated but not successfully added */
1831 pthread_mutex_lock(&consumer_data
.lock
);
1832 pthread_mutex_lock(&stream
->lock
);
1834 switch (consumer_data
.type
) {
1835 case LTTNG_CONSUMER_KERNEL
:
1836 if (stream
->mmap_base
!= NULL
) {
1837 ret
= munmap(stream
->mmap_base
, stream
->mmap_len
);
1839 PERROR("munmap metadata stream");
1843 case LTTNG_CONSUMER32_UST
:
1844 case LTTNG_CONSUMER64_UST
:
1845 lttng_ustconsumer_del_stream(stream
);
1848 ERR("Unknown consumer_data type");
1854 iter
.iter
.node
= &stream
->node
.node
;
1855 ret
= lttng_ht_del(ht
, &iter
);
1858 /* Remove node session id from the consumer_data stream ht */
1859 iter
.iter
.node
= &stream
->node_session_id
.node
;
1860 ret
= lttng_ht_del(consumer_data
.stream_list_ht
, &iter
);
1864 if (stream
->out_fd
>= 0) {
1865 ret
= close(stream
->out_fd
);
1871 if (stream
->wait_fd
>= 0 && !stream
->wait_fd_is_copy
) {
1872 ret
= close(stream
->wait_fd
);
1878 if (stream
->shm_fd
>= 0 && stream
->wait_fd
!= stream
->shm_fd
) {
1879 ret
= close(stream
->shm_fd
);
1885 /* Check and cleanup relayd */
1887 relayd
= consumer_find_relayd(stream
->net_seq_idx
);
1888 if (relayd
!= NULL
) {
1889 uatomic_dec(&relayd
->refcount
);
1890 assert(uatomic_read(&relayd
->refcount
) >= 0);
1892 /* Closing streams requires to lock the control socket. */
1893 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
1894 ret
= relayd_send_close_stream(&relayd
->control_sock
,
1895 stream
->relayd_stream_id
, stream
->next_net_seq_num
- 1);
1896 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
1898 DBG("Unable to close stream on the relayd. Continuing");
1900 * Continue here. There is nothing we can do for the relayd.
1901 * Chances are that the relayd has closed the socket so we just
1902 * continue cleaning up.
1906 /* Both conditions are met, we destroy the relayd. */
1907 if (uatomic_read(&relayd
->refcount
) == 0 &&
1908 uatomic_read(&relayd
->destroy_flag
)) {
1909 destroy_relayd(relayd
);
1914 /* Atomically decrement channel refcount since other threads can use it. */
1915 uatomic_dec(&stream
->chan
->refcount
);
1916 if (!uatomic_read(&stream
->chan
->refcount
)
1917 && !uatomic_read(&stream
->chan
->nb_init_streams
)) {
1918 /* Go for channel deletion! */
1919 free_chan
= stream
->chan
;
1923 pthread_mutex_unlock(&stream
->lock
);
1924 pthread_mutex_unlock(&consumer_data
.lock
);
1927 consumer_del_channel(free_chan
);
1931 call_rcu(&stream
->node
.head
, consumer_free_stream
);
1935 * Action done with the metadata stream when adding it to the consumer internal
1936 * data structures to handle it.
1938 static int consumer_add_metadata_stream(struct lttng_consumer_stream
*stream
,
1939 struct lttng_ht
*ht
)
1942 struct consumer_relayd_sock_pair
*relayd
;
1943 struct lttng_ht_iter iter
;
1944 struct lttng_ht_node_ulong
*node
;
1949 DBG3("Adding metadata stream %d to hash table", stream
->wait_fd
);
1951 pthread_mutex_lock(&consumer_data
.lock
);
1952 pthread_mutex_lock(&stream
->lock
);
1955 * From here, refcounts are updated so be _careful_ when returning an error
1962 * Lookup the stream just to make sure it does not exist in our internal
1963 * state. This should NEVER happen.
1965 lttng_ht_lookup(ht
, (void *)((unsigned long) stream
->wait_fd
), &iter
);
1966 node
= lttng_ht_iter_get_node_ulong(&iter
);
1969 /* Find relayd and, if one is found, increment refcount. */
1970 relayd
= consumer_find_relayd(stream
->net_seq_idx
);
1971 if (relayd
!= NULL
) {
1972 uatomic_inc(&relayd
->refcount
);
1975 /* Update channel refcount once added without error(s). */
1976 uatomic_inc(&stream
->chan
->refcount
);
1979 * When nb_init_streams reaches 0, we don't need to trigger any action in
1980 * terms of destroying the associated channel, because the action that
1981 * causes the count to become 0 also causes a stream to be added. The
1982 * channel deletion will thus be triggered by the following removal of this
1985 if (uatomic_read(&stream
->chan
->nb_init_streams
) > 0) {
1986 uatomic_dec(&stream
->chan
->nb_init_streams
);
1989 lttng_ht_add_unique_ulong(ht
, &stream
->node
);
1992 * Add stream to the stream_list_ht of the consumer data. No need to steal
1993 * the key since the HT does not use it and we allow to add redundant keys
1996 lttng_ht_add_ulong(consumer_data
.stream_list_ht
, &stream
->node_session_id
);
2000 pthread_mutex_unlock(&stream
->lock
);
2001 pthread_mutex_unlock(&consumer_data
.lock
);
2006 * Delete data stream that are flagged for deletion (endpoint_status).
2008 static void validate_endpoint_status_data_stream(void)
2010 struct lttng_ht_iter iter
;
2011 struct lttng_consumer_stream
*stream
;
2013 DBG("Consumer delete flagged data stream");
2016 cds_lfht_for_each_entry(data_ht
->ht
, &iter
.iter
, stream
, node
.node
) {
2017 /* Validate delete flag of the stream */
2018 if (stream
->endpoint_status
== CONSUMER_ENDPOINT_ACTIVE
) {
2021 /* Delete it right now */
2022 consumer_del_stream(stream
, data_ht
);
2028 * Delete metadata stream that are flagged for deletion (endpoint_status).
2030 static void validate_endpoint_status_metadata_stream(
2031 struct lttng_poll_event
*pollset
)
2033 struct lttng_ht_iter iter
;
2034 struct lttng_consumer_stream
*stream
;
2036 DBG("Consumer delete flagged metadata stream");
2041 cds_lfht_for_each_entry(metadata_ht
->ht
, &iter
.iter
, stream
, node
.node
) {
2042 /* Validate delete flag of the stream */
2043 if (stream
->endpoint_status
== CONSUMER_ENDPOINT_ACTIVE
) {
2047 * Remove from pollset so the metadata thread can continue without
2048 * blocking on a deleted stream.
2050 lttng_poll_del(pollset
, stream
->wait_fd
);
2052 /* Delete it right now */
2053 consumer_del_metadata_stream(stream
, metadata_ht
);
2059 * Thread polls on metadata file descriptor and write them on disk or on the
2062 void *consumer_thread_metadata_poll(void *data
)
2065 uint32_t revents
, nb_fd
;
2066 struct lttng_consumer_stream
*stream
= NULL
;
2067 struct lttng_ht_iter iter
;
2068 struct lttng_ht_node_ulong
*node
;
2069 struct lttng_poll_event events
;
2070 struct lttng_consumer_local_data
*ctx
= data
;
2073 rcu_register_thread();
2075 metadata_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_ULONG
);
2077 /* ENOMEM at this point. Better to bail out. */
2081 DBG("Thread metadata poll started");
2083 /* Size is set to 1 for the consumer_metadata pipe */
2084 ret
= lttng_poll_create(&events
, 2, LTTNG_CLOEXEC
);
2086 ERR("Poll set creation failed");
2090 ret
= lttng_poll_add(&events
, ctx
->consumer_metadata_pipe
[0], LPOLLIN
);
2096 DBG("Metadata main loop started");
2099 /* Only the metadata pipe is set */
2100 if (LTTNG_POLL_GETNB(&events
) == 0 && consumer_quit
== 1) {
2105 DBG("Metadata poll wait with %d fd(s)", LTTNG_POLL_GETNB(&events
));
2106 ret
= lttng_poll_wait(&events
, -1);
2107 DBG("Metadata event catched in thread");
2109 if (errno
== EINTR
) {
2110 ERR("Poll EINTR catched");
2118 /* From here, the event is a metadata wait fd */
2119 for (i
= 0; i
< nb_fd
; i
++) {
2120 revents
= LTTNG_POLL_GETEV(&events
, i
);
2121 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2123 /* Just don't waste time if no returned events for the fd */
2128 if (pollfd
== ctx
->consumer_metadata_pipe
[0]) {
2129 if (revents
& (LPOLLERR
| LPOLLHUP
)) {
2130 DBG("Metadata thread pipe hung up");
2132 * Remove the pipe from the poll set and continue the loop
2133 * since their might be data to consume.
2135 lttng_poll_del(&events
, ctx
->consumer_metadata_pipe
[0]);
2136 ret
= close(ctx
->consumer_metadata_pipe
[0]);
2138 PERROR("close metadata pipe");
2141 } else if (revents
& LPOLLIN
) {
2143 /* Get the stream pointer received */
2144 ret
= read(pollfd
, &stream
, sizeof(stream
));
2145 } while (ret
< 0 && errno
== EINTR
);
2147 ret
< sizeof(struct lttng_consumer_stream
*)) {
2148 PERROR("read metadata stream");
2150 * Let's continue here and hope we can still work
2151 * without stopping the consumer. XXX: Should we?
2156 /* A NULL stream means that the state has changed. */
2157 if (stream
== NULL
) {
2158 /* Check for deleted streams. */
2159 validate_endpoint_status_metadata_stream(&events
);
2163 DBG("Adding metadata stream %d to poll set",
2166 ret
= consumer_add_metadata_stream(stream
, metadata_ht
);
2168 ERR("Unable to add metadata stream");
2169 /* Stream was not setup properly. Continuing. */
2170 consumer_del_metadata_stream(stream
, NULL
);
2174 /* Add metadata stream to the global poll events list */
2175 lttng_poll_add(&events
, stream
->wait_fd
,
2176 LPOLLIN
| LPOLLPRI
);
2179 /* Handle other stream */
2184 lttng_ht_lookup(metadata_ht
, (void *)((unsigned long) pollfd
),
2186 node
= lttng_ht_iter_get_node_ulong(&iter
);
2189 stream
= caa_container_of(node
, struct lttng_consumer_stream
,
2192 /* Check for error event */
2193 if (revents
& (LPOLLERR
| LPOLLHUP
)) {
2194 DBG("Metadata fd %d is hup|err.", pollfd
);
2195 if (!stream
->hangup_flush_done
2196 && (consumer_data
.type
== LTTNG_CONSUMER32_UST
2197 || consumer_data
.type
== LTTNG_CONSUMER64_UST
)) {
2198 DBG("Attempting to flush and consume the UST buffers");
2199 lttng_ustconsumer_on_stream_hangup(stream
);
2201 /* We just flushed the stream now read it. */
2203 len
= ctx
->on_buffer_ready(stream
, ctx
);
2205 * We don't check the return value here since if we get
2206 * a negative len, it means an error occured thus we
2207 * simply remove it from the poll set and free the
2213 lttng_poll_del(&events
, stream
->wait_fd
);
2215 * This call update the channel states, closes file descriptors
2216 * and securely free the stream.
2218 consumer_del_metadata_stream(stream
, metadata_ht
);
2219 } else if (revents
& (LPOLLIN
| LPOLLPRI
)) {
2220 /* Get the data out of the metadata file descriptor */
2221 DBG("Metadata available on fd %d", pollfd
);
2222 assert(stream
->wait_fd
== pollfd
);
2224 len
= ctx
->on_buffer_ready(stream
, ctx
);
2225 /* It's ok to have an unavailable sub-buffer */
2226 if (len
< 0 && len
!= -EAGAIN
&& len
!= -ENODATA
) {
2227 /* Clean up stream from consumer and free it. */
2228 lttng_poll_del(&events
, stream
->wait_fd
);
2229 consumer_del_metadata_stream(stream
, metadata_ht
);
2230 } else if (len
> 0) {
2231 stream
->data_read
= 1;
2235 /* Release RCU lock for the stream looked up */
2242 DBG("Metadata poll thread exiting");
2243 lttng_poll_clean(&events
);
2245 destroy_stream_ht(metadata_ht
);
2247 rcu_unregister_thread();
2252 * This thread polls the fds in the set to consume the data and write
2253 * it to tracefile if necessary.
2255 void *consumer_thread_data_poll(void *data
)
2257 int num_rdy
, num_hup
, high_prio
, ret
, i
;
2258 struct pollfd
*pollfd
= NULL
;
2259 /* local view of the streams */
2260 struct lttng_consumer_stream
**local_stream
= NULL
, *new_stream
= NULL
;
2261 /* local view of consumer_data.fds_count */
2263 struct lttng_consumer_local_data
*ctx
= data
;
2266 rcu_register_thread();
2268 data_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_ULONG
);
2269 if (data_ht
== NULL
) {
2270 /* ENOMEM at this point. Better to bail out. */
2274 local_stream
= zmalloc(sizeof(struct lttng_consumer_stream
));
2281 * the fds set has been updated, we need to update our
2282 * local array as well
2284 pthread_mutex_lock(&consumer_data
.lock
);
2285 if (consumer_data
.need_update
) {
2286 if (pollfd
!= NULL
) {
2290 if (local_stream
!= NULL
) {
2292 local_stream
= NULL
;
2295 /* allocate for all fds + 1 for the consumer_data_pipe */
2296 pollfd
= zmalloc((consumer_data
.stream_count
+ 1) * sizeof(struct pollfd
));
2297 if (pollfd
== NULL
) {
2298 PERROR("pollfd malloc");
2299 pthread_mutex_unlock(&consumer_data
.lock
);
2303 /* allocate for all fds + 1 for the consumer_data_pipe */
2304 local_stream
= zmalloc((consumer_data
.stream_count
+ 1) *
2305 sizeof(struct lttng_consumer_stream
));
2306 if (local_stream
== NULL
) {
2307 PERROR("local_stream malloc");
2308 pthread_mutex_unlock(&consumer_data
.lock
);
2311 ret
= consumer_update_poll_array(ctx
, &pollfd
, local_stream
,
2314 ERR("Error in allocating pollfd or local_outfds");
2315 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_POLL_ERROR
);
2316 pthread_mutex_unlock(&consumer_data
.lock
);
2320 consumer_data
.need_update
= 0;
2322 pthread_mutex_unlock(&consumer_data
.lock
);
2324 /* No FDs and consumer_quit, consumer_cleanup the thread */
2325 if (nb_fd
== 0 && consumer_quit
== 1) {
2328 /* poll on the array of fds */
2330 DBG("polling on %d fd", nb_fd
+ 1);
2331 num_rdy
= poll(pollfd
, nb_fd
+ 1, -1);
2332 DBG("poll num_rdy : %d", num_rdy
);
2333 if (num_rdy
== -1) {
2335 * Restart interrupted system call.
2337 if (errno
== EINTR
) {
2340 PERROR("Poll error");
2341 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_POLL_ERROR
);
2343 } else if (num_rdy
== 0) {
2344 DBG("Polling thread timed out");
2349 * If the consumer_data_pipe triggered poll go directly to the
2350 * beginning of the loop to update the array. We want to prioritize
2351 * array update over low-priority reads.
2353 if (pollfd
[nb_fd
].revents
& (POLLIN
| POLLPRI
)) {
2354 ssize_t pipe_readlen
;
2356 DBG("consumer_data_pipe wake up");
2357 /* Consume 1 byte of pipe data */
2359 pipe_readlen
= read(ctx
->consumer_data_pipe
[0], &new_stream
,
2360 sizeof(new_stream
));
2361 } while (pipe_readlen
== -1 && errno
== EINTR
);
2362 if (pipe_readlen
< 0) {
2363 PERROR("read consumer data pipe");
2364 /* Continue so we can at least handle the current stream(s). */
2369 * If the stream is NULL, just ignore it. It's also possible that
2370 * the sessiond poll thread changed the consumer_quit state and is
2371 * waking us up to test it.
2373 if (new_stream
== NULL
) {
2374 validate_endpoint_status_data_stream();
2378 ret
= consumer_add_stream(new_stream
, data_ht
);
2380 ERR("Consumer add stream %d failed. Continuing",
2383 * At this point, if the add_stream fails, it is not in the
2384 * hash table thus passing the NULL value here.
2386 consumer_del_stream(new_stream
, NULL
);
2389 /* Continue to update the local streams and handle prio ones */
2393 /* Take care of high priority channels first. */
2394 for (i
= 0; i
< nb_fd
; i
++) {
2395 if (local_stream
[i
] == NULL
) {
2398 if (pollfd
[i
].revents
& POLLPRI
) {
2399 DBG("Urgent read on fd %d", pollfd
[i
].fd
);
2401 len
= ctx
->on_buffer_ready(local_stream
[i
], ctx
);
2402 /* it's ok to have an unavailable sub-buffer */
2403 if (len
< 0 && len
!= -EAGAIN
&& len
!= -ENODATA
) {
2404 /* Clean the stream and free it. */
2405 consumer_del_stream(local_stream
[i
], data_ht
);
2406 local_stream
[i
] = NULL
;
2407 } else if (len
> 0) {
2408 local_stream
[i
]->data_read
= 1;
2414 * If we read high prio channel in this loop, try again
2415 * for more high prio data.
2421 /* Take care of low priority channels. */
2422 for (i
= 0; i
< nb_fd
; i
++) {
2423 if (local_stream
[i
] == NULL
) {
2426 if ((pollfd
[i
].revents
& POLLIN
) ||
2427 local_stream
[i
]->hangup_flush_done
) {
2428 DBG("Normal read on fd %d", pollfd
[i
].fd
);
2429 len
= ctx
->on_buffer_ready(local_stream
[i
], ctx
);
2430 /* it's ok to have an unavailable sub-buffer */
2431 if (len
< 0 && len
!= -EAGAIN
&& len
!= -ENODATA
) {
2432 /* Clean the stream and free it. */
2433 consumer_del_stream(local_stream
[i
], data_ht
);
2434 local_stream
[i
] = NULL
;
2435 } else if (len
> 0) {
2436 local_stream
[i
]->data_read
= 1;
2441 /* Handle hangup and errors */
2442 for (i
= 0; i
< nb_fd
; i
++) {
2443 if (local_stream
[i
] == NULL
) {
2446 if (!local_stream
[i
]->hangup_flush_done
2447 && (pollfd
[i
].revents
& (POLLHUP
| POLLERR
| POLLNVAL
))
2448 && (consumer_data
.type
== LTTNG_CONSUMER32_UST
2449 || consumer_data
.type
== LTTNG_CONSUMER64_UST
)) {
2450 DBG("fd %d is hup|err|nval. Attempting flush and read.",
2452 lttng_ustconsumer_on_stream_hangup(local_stream
[i
]);
2453 /* Attempt read again, for the data we just flushed. */
2454 local_stream
[i
]->data_read
= 1;
2457 * If the poll flag is HUP/ERR/NVAL and we have
2458 * read no data in this pass, we can remove the
2459 * stream from its hash table.
2461 if ((pollfd
[i
].revents
& POLLHUP
)) {
2462 DBG("Polling fd %d tells it has hung up.", pollfd
[i
].fd
);
2463 if (!local_stream
[i
]->data_read
) {
2464 consumer_del_stream(local_stream
[i
], data_ht
);
2465 local_stream
[i
] = NULL
;
2468 } else if (pollfd
[i
].revents
& POLLERR
) {
2469 ERR("Error returned in polling fd %d.", pollfd
[i
].fd
);
2470 if (!local_stream
[i
]->data_read
) {
2471 consumer_del_stream(local_stream
[i
], data_ht
);
2472 local_stream
[i
] = NULL
;
2475 } else if (pollfd
[i
].revents
& POLLNVAL
) {
2476 ERR("Polling fd %d tells fd is not open.", pollfd
[i
].fd
);
2477 if (!local_stream
[i
]->data_read
) {
2478 consumer_del_stream(local_stream
[i
], data_ht
);
2479 local_stream
[i
] = NULL
;
2483 if (local_stream
[i
] != NULL
) {
2484 local_stream
[i
]->data_read
= 0;
2489 DBG("polling thread exiting");
2490 if (pollfd
!= NULL
) {
2494 if (local_stream
!= NULL
) {
2496 local_stream
= NULL
;
2500 * Close the write side of the pipe so epoll_wait() in
2501 * consumer_thread_metadata_poll can catch it. The thread is monitoring the
2502 * read side of the pipe. If we close them both, epoll_wait strangely does
2503 * not return and could create a endless wait period if the pipe is the
2504 * only tracked fd in the poll set. The thread will take care of closing
2507 ret
= close(ctx
->consumer_metadata_pipe
[1]);
2509 PERROR("close data pipe");
2512 destroy_data_stream_ht(data_ht
);
2514 rcu_unregister_thread();
2519 * This thread listens on the consumerd socket and receives the file
2520 * descriptors from the session daemon.
2522 void *consumer_thread_sessiond_poll(void *data
)
2524 int sock
= -1, client_socket
, ret
;
2526 * structure to poll for incoming data on communication socket avoids
2527 * making blocking sockets.
2529 struct pollfd consumer_sockpoll
[2];
2530 struct lttng_consumer_local_data
*ctx
= data
;
2532 rcu_register_thread();
2534 DBG("Creating command socket %s", ctx
->consumer_command_sock_path
);
2535 unlink(ctx
->consumer_command_sock_path
);
2536 client_socket
= lttcomm_create_unix_sock(ctx
->consumer_command_sock_path
);
2537 if (client_socket
< 0) {
2538 ERR("Cannot create command socket");
2542 ret
= lttcomm_listen_unix_sock(client_socket
);
2547 DBG("Sending ready command to lttng-sessiond");
2548 ret
= lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_COMMAND_SOCK_READY
);
2549 /* return < 0 on error, but == 0 is not fatal */
2551 ERR("Error sending ready command to lttng-sessiond");
2555 ret
= fcntl(client_socket
, F_SETFL
, O_NONBLOCK
);
2557 PERROR("fcntl O_NONBLOCK");
2561 /* prepare the FDs to poll : to client socket and the should_quit pipe */
2562 consumer_sockpoll
[0].fd
= ctx
->consumer_should_quit
[0];
2563 consumer_sockpoll
[0].events
= POLLIN
| POLLPRI
;
2564 consumer_sockpoll
[1].fd
= client_socket
;
2565 consumer_sockpoll
[1].events
= POLLIN
| POLLPRI
;
2567 if (lttng_consumer_poll_socket(consumer_sockpoll
) < 0) {
2570 DBG("Connection on client_socket");
2572 /* Blocking call, waiting for transmission */
2573 sock
= lttcomm_accept_unix_sock(client_socket
);
2578 ret
= fcntl(sock
, F_SETFL
, O_NONBLOCK
);
2580 PERROR("fcntl O_NONBLOCK");
2584 /* This socket is not useful anymore. */
2585 ret
= close(client_socket
);
2587 PERROR("close client_socket");
2591 /* update the polling structure to poll on the established socket */
2592 consumer_sockpoll
[1].fd
= sock
;
2593 consumer_sockpoll
[1].events
= POLLIN
| POLLPRI
;
2596 if (lttng_consumer_poll_socket(consumer_sockpoll
) < 0) {
2599 DBG("Incoming command on sock");
2600 ret
= lttng_consumer_recv_cmd(ctx
, sock
, consumer_sockpoll
);
2601 if (ret
== -ENOENT
) {
2602 DBG("Received STOP command");
2607 * This could simply be a session daemon quitting. Don't output
2610 DBG("Communication interrupted on command socket");
2613 if (consumer_quit
) {
2614 DBG("consumer_thread_receive_fds received quit from signal");
2617 DBG("received fds on sock");
2620 DBG("consumer_thread_receive_fds exiting");
2623 * when all fds have hung up, the polling thread
2629 * Notify the data poll thread to poll back again and test the
2630 * consumer_quit state that we just set so to quit gracefully.
2632 notify_thread_pipe(ctx
->consumer_data_pipe
[1]);
2634 /* Cleaning up possibly open sockets. */
2638 PERROR("close sock sessiond poll");
2641 if (client_socket
>= 0) {
2644 PERROR("close client_socket sessiond poll");
2648 rcu_unregister_thread();
2652 ssize_t
lttng_consumer_read_subbuffer(struct lttng_consumer_stream
*stream
,
2653 struct lttng_consumer_local_data
*ctx
)
2657 pthread_mutex_lock(&stream
->lock
);
2659 switch (consumer_data
.type
) {
2660 case LTTNG_CONSUMER_KERNEL
:
2661 ret
= lttng_kconsumer_read_subbuffer(stream
, ctx
);
2663 case LTTNG_CONSUMER32_UST
:
2664 case LTTNG_CONSUMER64_UST
:
2665 ret
= lttng_ustconsumer_read_subbuffer(stream
, ctx
);
2668 ERR("Unknown consumer_data type");
2674 pthread_mutex_unlock(&stream
->lock
);
2678 int lttng_consumer_on_recv_stream(struct lttng_consumer_stream
*stream
)
2680 switch (consumer_data
.type
) {
2681 case LTTNG_CONSUMER_KERNEL
:
2682 return lttng_kconsumer_on_recv_stream(stream
);
2683 case LTTNG_CONSUMER32_UST
:
2684 case LTTNG_CONSUMER64_UST
:
2685 return lttng_ustconsumer_on_recv_stream(stream
);
2687 ERR("Unknown consumer_data type");
2694 * Allocate and set consumer data hash tables.
2696 void lttng_consumer_init(void)
2698 consumer_data
.channel_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_ULONG
);
2699 consumer_data
.relayd_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_ULONG
);
2700 consumer_data
.stream_list_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_ULONG
);
2704 * Process the ADD_RELAYD command receive by a consumer.
2706 * This will create a relayd socket pair and add it to the relayd hash table.
2707 * The caller MUST acquire a RCU read side lock before calling it.
2709 int consumer_add_relayd_socket(int net_seq_idx
, int sock_type
,
2710 struct lttng_consumer_local_data
*ctx
, int sock
,
2711 struct pollfd
*consumer_sockpoll
, struct lttcomm_sock
*relayd_sock
,
2712 unsigned int sessiond_id
)
2714 int fd
= -1, ret
= -1, relayd_created
= 0;
2715 enum lttng_error_code ret_code
= LTTNG_OK
;
2716 struct consumer_relayd_sock_pair
*relayd
;
2718 DBG("Consumer adding relayd socket (idx: %d)", net_seq_idx
);
2720 /* First send a status message before receiving the fds. */
2721 ret
= consumer_send_status_msg(sock
, ret_code
);
2723 /* Somehow, the session daemon is not responding anymore. */
2727 /* Get relayd reference if exists. */
2728 relayd
= consumer_find_relayd(net_seq_idx
);
2729 if (relayd
== NULL
) {
2730 /* Not found. Allocate one. */
2731 relayd
= consumer_allocate_relayd_sock_pair(net_seq_idx
);
2732 if (relayd
== NULL
) {
2733 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_OUTFD_ERROR
);
2737 relayd
->sessiond_session_id
= (uint64_t) sessiond_id
;
2741 /* Poll on consumer socket. */
2742 if (lttng_consumer_poll_socket(consumer_sockpoll
) < 0) {
2747 /* Get relayd socket from session daemon */
2748 ret
= lttcomm_recv_fds_unix_sock(sock
, &fd
, 1);
2749 if (ret
!= sizeof(fd
)) {
2750 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_ERROR_RECV_FD
);
2752 fd
= -1; /* Just in case it gets set with an invalid value. */
2756 /* We have the fds without error. Send status back. */
2757 ret
= consumer_send_status_msg(sock
, ret_code
);
2759 /* Somehow, the session daemon is not responding anymore. */
2763 /* Copy socket information and received FD */
2764 switch (sock_type
) {
2765 case LTTNG_STREAM_CONTROL
:
2766 /* Copy received lttcomm socket */
2767 lttcomm_copy_sock(&relayd
->control_sock
, relayd_sock
);
2768 ret
= lttcomm_create_sock(&relayd
->control_sock
);
2769 /* Immediately try to close the created socket if valid. */
2770 if (relayd
->control_sock
.fd
>= 0) {
2771 if (close(relayd
->control_sock
.fd
)) {
2772 PERROR("close relayd control socket");
2775 /* Handle create_sock error. */
2780 /* Assign new file descriptor */
2781 relayd
->control_sock
.fd
= fd
;
2784 * Create a session on the relayd and store the returned id. Lock the
2785 * control socket mutex if the relayd was NOT created before.
2787 if (!relayd_created
) {
2788 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
2790 ret
= relayd_create_session(&relayd
->control_sock
,
2791 &relayd
->relayd_session_id
);
2792 if (!relayd_created
) {
2793 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
2800 case LTTNG_STREAM_DATA
:
2801 /* Copy received lttcomm socket */
2802 lttcomm_copy_sock(&relayd
->data_sock
, relayd_sock
);
2803 ret
= lttcomm_create_sock(&relayd
->data_sock
);
2804 /* Immediately try to close the created socket if valid. */
2805 if (relayd
->data_sock
.fd
>= 0) {
2806 if (close(relayd
->data_sock
.fd
)) {
2807 PERROR("close relayd data socket");
2810 /* Handle create_sock error. */
2815 /* Assign new file descriptor */
2816 relayd
->data_sock
.fd
= fd
;
2819 ERR("Unknown relayd socket type (%d)", sock_type
);
2824 DBG("Consumer %s socket created successfully with net idx %d (fd: %d)",
2825 sock_type
== LTTNG_STREAM_CONTROL
? "control" : "data",
2826 relayd
->net_seq_idx
, fd
);
2829 * Add relayd socket pair to consumer data hashtable. If object already
2830 * exists or on error, the function gracefully returns.
2838 /* Close received socket if valid. */
2841 PERROR("close received socket");
2845 if (relayd_created
) {
2846 /* We just want to cleanup. Ignore ret value. */
2847 (void) relayd_close(&relayd
->control_sock
);
2848 (void) relayd_close(&relayd
->data_sock
);
2856 * Try to lock the stream mutex.
2858 * On success, 1 is returned else 0 indicating that the mutex is NOT lock.
2860 static int stream_try_lock(struct lttng_consumer_stream
*stream
)
2867 * Try to lock the stream mutex. On failure, we know that the stream is
2868 * being used else where hence there is data still being extracted.
2870 ret
= pthread_mutex_trylock(&stream
->lock
);
2872 /* For both EBUSY and EINVAL error, the mutex is NOT locked. */
2884 * Search for a relayd associated to the session id and return the reference.
2886 * A rcu read side lock MUST be acquire before calling this function and locked
2887 * until the relayd object is no longer necessary.
2889 static struct consumer_relayd_sock_pair
*find_relayd_by_session_id(uint64_t id
)
2891 struct lttng_ht_iter iter
;
2892 struct consumer_relayd_sock_pair
*relayd
= NULL
;
2894 /* Iterate over all relayd since they are indexed by net_seq_idx. */
2895 cds_lfht_for_each_entry(consumer_data
.relayd_ht
->ht
, &iter
.iter
, relayd
,
2898 * Check by sessiond id which is unique here where the relayd session
2899 * id might not be when having multiple relayd.
2901 if (relayd
->sessiond_session_id
== id
) {
2902 /* Found the relayd. There can be only one per id. */
2914 * Check if for a given session id there is still data needed to be extract
2917 * Return 1 if data is pending or else 0 meaning ready to be read.
2919 int consumer_data_pending(uint64_t id
)
2922 struct lttng_ht_iter iter
;
2923 struct lttng_ht
*ht
;
2924 struct lttng_consumer_stream
*stream
;
2925 struct consumer_relayd_sock_pair
*relayd
= NULL
;
2926 int (*data_pending
)(struct lttng_consumer_stream
*);
2928 DBG("Consumer data pending command on session id %" PRIu64
, id
);
2931 pthread_mutex_lock(&consumer_data
.lock
);
2933 switch (consumer_data
.type
) {
2934 case LTTNG_CONSUMER_KERNEL
:
2935 data_pending
= lttng_kconsumer_data_pending
;
2937 case LTTNG_CONSUMER32_UST
:
2938 case LTTNG_CONSUMER64_UST
:
2939 data_pending
= lttng_ustconsumer_data_pending
;
2942 ERR("Unknown consumer data type");
2946 /* Ease our life a bit */
2947 ht
= consumer_data
.stream_list_ht
;
2949 relayd
= find_relayd_by_session_id(id
);
2951 /* Send init command for data pending. */
2952 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
2953 ret
= relayd_begin_data_pending(&relayd
->control_sock
,
2954 relayd
->relayd_session_id
);
2955 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
2957 /* Communication error thus the relayd so no data pending. */
2958 goto data_not_pending
;
2962 cds_lfht_for_each_entry_duplicate(ht
->ht
,
2963 ht
->hash_fct((void *)((unsigned long) id
), lttng_ht_seed
),
2964 ht
->match_fct
, (void *)((unsigned long) id
),
2965 &iter
.iter
, stream
, node_session_id
.node
) {
2966 /* If this call fails, the stream is being used hence data pending. */
2967 ret
= stream_try_lock(stream
);
2973 * A removed node from the hash table indicates that the stream has
2974 * been deleted thus having a guarantee that the buffers are closed
2975 * on the consumer side. However, data can still be transmitted
2976 * over the network so don't skip the relayd check.
2978 ret
= cds_lfht_is_node_deleted(&stream
->node
.node
);
2980 /* Check the stream if there is data in the buffers. */
2981 ret
= data_pending(stream
);
2983 pthread_mutex_unlock(&stream
->lock
);
2990 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
2991 if (stream
->metadata_flag
) {
2992 ret
= relayd_quiescent_control(&relayd
->control_sock
,
2993 stream
->relayd_stream_id
);
2995 ret
= relayd_data_pending(&relayd
->control_sock
,
2996 stream
->relayd_stream_id
,
2997 stream
->next_net_seq_num
- 1);
2999 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
3001 pthread_mutex_unlock(&stream
->lock
);
3005 pthread_mutex_unlock(&stream
->lock
);
3009 unsigned int is_data_inflight
= 0;
3011 /* Send init command for data pending. */
3012 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
3013 ret
= relayd_end_data_pending(&relayd
->control_sock
,
3014 relayd
->relayd_session_id
, &is_data_inflight
);
3015 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
3017 goto data_not_pending
;
3019 if (is_data_inflight
) {
3025 * Finding _no_ node in the hash table and no inflight data means that the
3026 * stream(s) have been removed thus data is guaranteed to be available for
3027 * analysis from the trace files.
3031 /* Data is available to be read by a viewer. */
3032 pthread_mutex_unlock(&consumer_data
.lock
);
3037 /* Data is still being extracted from buffers. */
3038 pthread_mutex_unlock(&consumer_data
.lock
);
3044 * Send a ret code status message to the sessiond daemon.
3046 * Return the sendmsg() return value.
3048 int consumer_send_status_msg(int sock
, int ret_code
)
3050 struct lttcomm_consumer_status_msg msg
;
3052 msg
.ret_code
= ret_code
;
3054 return lttcomm_send_unix_sock(sock
, &msg
, sizeof(msg
));